Analytical Chemistry Chemical Cytometry Quantitates Superoxide

Anal. Chem. 2010, 82, 6745–6750
Letters to Analytical Chemistry
Chemical Cytometry Quantitates Superoxide
Levels in the Mitochondrial Matrix of Single
Myoblasts
Xin Xu and Edgar A. Arriaga*
Department of Chemistry, University of Minnesota, Minneapolis, Minnesota 55455
Triphenylphosphonium hydroethidine (TPP-HE) is a membrane-permeable
probe that reacts with superoxide and
forms hydroxytriphenylphosphonium ethidium (OH-TPP-
E + ), a fluorescent product that has been previously
used in qualitative measurements of superoxide production.
In order to develop quantitative methods to
measure superoxide, it is necessary to take into
consideration the principles that drive TPP-HE accumulation
into various subcellular compartments. In
the mitochondria matrix, TPP-HE accumulation depends
on the mitochondrial membrane potential, which
varies from cell to cell. Here we address this issue by
including rhodamine 123 (R123) as an internal mitochondrial
membrane potential calibrant in chemical
cytometry experiments. After loading with TPP-HE and
R123, a single cell is lysed within a separation capillary
and its contents are separated and detected by micellar
electrokinetic capillary chromatography with laserinduced
fluorescence detection (MEKC-LIF). Using
theoretical arguments, we show that the ratio [OH-TPP-
E + ]/[R123] is adequate to obtain a relative quantitation
of mitochondrial matrix superoxide levels for each
analyzed cell. We applied this method to single skeletal
muscle myoblasts and determined that the steady state
superoxide levels in the mitochondrial matrix is ∼(0.29
( 0.10) × 10 -12 M. The development of this quantitative
method is a critical step toward establishing the
importance of reactive oxygen species in biological
systems, including those relevant to aging and disease.
Superoxide is a reactive oxygen species (ROS) that plays a
decisive role in the generation of secondary reactive radicals, 1,2
which could result in oxidative damage associated with multiple
diseases and aging. 2-4 As a probe to detect superoxide, 5 triph-
* Corresponding author. Phone: 1-612-624-8024. Fax: 1-612-626-7541. E-mail:
arriaga@umn.edu.
(1) St.-Pierre, J.; Buckingham, J. A.; Roebuck, S. J.; Brand, M. D. J. Biol. Chem.
2002, 277, 44784.
(2) Orrenius, S.; Gogvadze, V.; Zhivotovsky, B. Annu. Rev. Pharmacol. Toxicol.
2007, 47, 143.
(3) Ashok, B. T.; Ali, R. Exp. Gerontol. 1999, 34, 293.
(4) Beckman, K. B.; Ames, B. N. Physiol. Rev. 1998, 78, 547.
enylphosphonium hydroethidine (TPP-HE, also known as (aka)
MitoSOX Red) has several advantages over commonly used
probes such as dichlorodihydrofluorescein (H2DCF) and hydroethidine.
TPP-HE is more selective than H2DCF 6 and forms
the product hydroxytriphenylphosphonium ethidium (OH-TPP-
E + ) upon reaction with superoxide. 5 Other reactive species also
react with TPP-HE, but relative to the reaction with an
equivalent level of superoxide the OH-TPP-E + formed is a small
fraction (i.e., ∼0.4%, ∼6%, and ∼4% for the reactions with
hydrogen peroxide, hydroxyl radical, and hypochlorite, respectively).
7 In comparison with hydroethidine, 8 TPP-HE accumulates
in the mitochondrial matrix of cells according to the
mitochondrial membrane potential, 9 making it possible to
detect superoxide in the matrix even in the presence of
competing superoxide dismutase (SOD). 5,7 Previous studies
based on the use of TPP-HE are, however, only qualitative because
the dependence of TPP-HE accumulation on the mitochondrial
membrane potential was not taken into account. 5,7,10-12 Variations
in mitochondrial membrane potential between cells could cause
different TPP-HE distributions and bias the measured ROS
levels. 7,13
Here we demonstrate that chemical cytometry can be used to
quantitate superoxide levels in the mitochondrial matrix of single
myoblasts. Single myoblasts were simultaneously treated with
both probes, TPP-HE and rhodamine 123 (R123), before the
(5) Robinson, K. M.; Janes, M. S.; Pehar, M.; Monette, J. S.; Ross, M. F.; Hagen,
T. M.; Murphy, M. P.; Beckman, J. S. Proc. Natl. Acad. Sci. U.S.A. 2006,
103, 15038.
(6) McArdle, F.; Pattwell, D. M.; Vasilaki, A.; McArdle, A.; Jackson, M. J. Free
Radical Biol. Med. 2005, 39, 651.
(7) Xu, X.; Arriaga, E. A. Free Radical Biol. Med. 2009, 46, 905.
(8) Zhao, H.; Joseph, J.; Fales, H. M.; Sokoloski, E. A.; Levine, R. L.; Vasquez-
Vivar, J.; Kalyanaraman, B. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 5727.
(9) Ross, M. F.; Kelso, G. F.; Blaikie, F. H.; James, A. M.; Cocheme, H. M.;
Filipovska, A.; Da Ros, T.; Hurd, T. R.; Smith, R. A. J.; Murphy, M. P.
Biochemistry (Moscow) 2005, 70, 222.
(10) Mukhopadhyay, P.; Rajesh, M.; Yoshihiro, K.; Hasko, G.; Pacher, P.
Biochem. Biophys. Res. Commun. 2007, 358, 203.
(11) Han, Z.; Varadharaj, S.; Giedt, R. J.; Zweier, J. L.; Szeto, H. H.; Alevriadou,
B. R. J. Pharmacol. Exp. Ther. 2009, 329, 94.
(12) Watanabe, N.; Zmijewski, J. W.; Takabe, W.; Umezu-Goto, M.; Le Goffe,
C.; Sekine, A.; Landar, A.; Watanabe, A.; Aoki, J.; Arai, H.; Kodama, T.;
Murphy, M. P.; Kalyanaraman, R.; Darley-Usmar, V. M.; Noguchi, N. Am. J.
Pathol. 2006, 168, 1737.
(13) Zielonka, J.; Kalyanaraman, B. Free Radical Biol. Med. 2010, 48, 983.
10.1021/ac101509d © 2010 American Chemical Society 6745
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/20/2010

micellar electrokinetic capillary chromatography with laserinduced
fluorescence detection (MEKC-LIF) analysis. An intact
cell was then introduced into a narrow-bore capillary and lysed
therein. Subsequently, the R123 and OH-TPP-E + contents released
from each cell were analyzed by MEKC-LIF. 7 Since both
R123 and TPP-HE accumulate into mitochondria according to
the mitochondrial membrane potential, the mole ratio of OH-
TPP-E + /R123 is approximately proportional to the steady state
superoxide concentration in the mitochondrial matrix of
individual cells with polarized mitochondria. Thus, determination
of this ratio make it possible to quantitate superoxide in
single cells. In the future, this method could be used for
investigating the role of superoxide in xenobiotic toxicity, aging,
and disease.
THEORY
When cells are incubated with TPP-HE and R123, both probes
accumulate in their mitochondria according to the mitochondrial
membrane potential (∆ψ). 5,14 The relevant Nernst equations are
Thus
∆ψ )- RT
nF ln [TPP-HE] inside
[TPP-HE] outside
∆ψ )- RT
nF ln [R123] inside
[R123] outside
[TPP-HE] inside ) [R123] inside
[TPP-HE] outside
[R123] outside
where [TPP-HE]inside and [R123]inside, and [TPP-HE]outside and
[R123]outside are the concentrations of TPP-HE and R123 in the
mitochondrial matrix and outside the mitochondria, respectively.
In this study, [TPP-HE]outside is 10 µM and [R123]outside
is 50 nM.
The reaction of TPP-HE with superoxide and its relevant
kinetics equations are
k
-
TPP-HE + O2 f
d[OH-TPP-E + ]
dt
OH-TPP-E +
[OH-TPP-E + ] T ) k[TPP-HE]∫ 0
(1)
(2)
(3)
-
) k[TPP-HE][O2 ] (4)
T
-
[O2 ]dt (5)
where k is the kinetic rate of the reaction of TPP-HE with
superoxide (∼4 × 10 6 M -1 s -1 ), 5 T is the incubation time of the
cells with TPP-HE (60 min), and ∫0 T [O2 - ]dt is the integral of
steady state superoxide concentration over the entire incubation
time. If [O2 - ] is constant during the incubation time,
T
- - ∫ [O
0 2 ]dt ) [O2 ]averageT (6)
(14) Scaduto, R. C.; Grotyohann, L. W. Biophys. J. 1999, 76, 469.
6746 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
where [O2 - ]average is the average steady state superoxide
concentration over the length of the incubation time.
When a cell is lysed and analyzed by MEKC-LIF, the detected
OH-TPP-E + is the total from both the matrix and outside the
mitochondria of the cell. 7 Thus, the amount of OH-TPP-E + in
each individual cell is
m OH-TPP-E+ ) m OH-TPP-E+, inside + m OH-TPP-E+, outside
where mOH-TPP-E + ,inside and mOH-TPP-E + ,outside are the amounts of OH-
TPP-E + in the mitochondrial matrix and outside the mitochondria
of a given cell, respectively. On the basis of the results
from bulk analysis (Supporting Information, Part I; Figure S-1),
the mOH-TPP-E + ,outside is a small fraction of the mOH-TPP-E + ,inside under
basal conditions and upon treatments with rotenone and
antimycin A. To simplify eq 7, the amount of OH-TPP-E + in one
cell can be expressed as
m OH-TPP-E+ ) (1 + x)m OH-TPP-E+, inside
where x is the fraction of mOH-TPP-E + ,outside relative to mOH-TPP-E + ,inside.
This fraction “x” represent the bias of the measurement due
to the superoxide found outside the mitochondria. On the basis
of bulk measurements, x is ∼0.048 under basal conditions,
∼0.063 upon treatment with rotenone, and ∼0.071 upon
treatment with antimycin A (Supporting Information, Part I).
Note that eq 8 is not applicable when mitochondria are depolarized
by carbonyl cyanide m-chlorophenylhydrazone (CCCP) treatment
(Supporting Information, Part II).
With the combination of eqs 5, 6, and 8,
m ) OH-TPP-E+ (1 + x)k[TPP-HE] outside
-
mR123 [O2 ]average,insideT [R123] outside
(9)
where [O2 - ]average,inside is the average of steady state superoxide
concentration in the mitochondrial matrix of single cells over
the incubation time.
On the basis of the MEKC-LIF calibration curves of OH-TPP-
E + and R123, the following equations are yielded
and
(7)
(8)
m OH-TPP-E+ ) aA OH-TPP-E+ + m (10)
m R123 ) bA R123 + n (11)
where the respective slopes are a and b, and the respective
intercepts are m and n, and the respective peak areas in the
electropherograms are AOH-TPP-E + and AR123.
Thus,
-
[O2 ]average,inside ) K
1 (aA + m)
OH-TPP-E+
(1 + x) (bAR123 + n)
(12)
where K ) ([R123]outside)/(k[TPP-HE]outsideT) ≈ 3.47 × 10 -13 M.
EXPERIMENTAL SECTION
Chemicals. Rhodamine 123 (R123), TPP-HE (aka MitoSOX
Red), and tetramethylrhodamine methyl ester (TMRM) were

purchased from Invitrogen-Molecular Probes (Eugene, OR). OH-
TPP-E + was synthesized and purified following a previously
described procedure. 15 Its purity was verified by MEKC-LIF.
Its concentration was determined spectrophotometrically using
the reported molar extinction coefficient at 478 nm. 5,15 OH-
TPP-E + is stable for at least 1 month when stored at -20 °C. 15
All the other reagents were obtained from Sigma-Aldrich (St.
Louis, MO). The MEKC running buffer contained 10 mM
sodium borate and 1 mM cetyltrimethylammonium bromide
(CTAB) (pH 9.3). All buffers were made with Milli-Q deionized
water and filtered through 0.22 µm filters before use.
Cell Culture. Rat muscle L6 myoblast cell line was obtained
from the American Tissue Culture Collection (Manassas, VA). The
cells were cultured in Dulbecco’s modified eagle medium (DMEM)
containing 10% (v/v) calf serum and 10 µg/mL gentamicin at 37
°C and 5% CO2. The cells were maintained by splitting them
every 3-4 days.
Cell Treatments. The cultured myoblasts were incubated with
10 µM TPP-HE and 50 nM R123 in DMEM at 37 °C for 1 h. When
needed, before incubation in the presence of TPP-HE and R123,
cells were treated with either 1 mM tiron, 50 µM carbonyl cyanide
m-chlorophenylhydrazone (CCCP), 5 µM rotenone, or 5 µM
antimycin A. After treatments, the cells were washed two times
with DMEM, trypsinized, and resuspended in PBS for either whole
cell lysate or single cell analysis.
MEKC-LIF. To do the whole cell lysate analysis, ∼2 × 10 6
cells in PBS were centrifuged down at 600g for 5 min. The cell
pellet was then dissolved in running buffer, treated with 2 mg/
mL proteinase K and 400U/mL DNase 1, and analyzed by
MEKC-LIF following previously described procedures. 7 Briefly,
separations were carried out at -400 V/cm in MEKC running
buffer. 7 The 488 nm line (12 mW) of an argon-ion laser (Melles
Griot, Irvine, CA) was used for excitation, and fluorescence
was selected with a bandpass filter transmitting in the 607-662
nm range (Omega Optical, Brattleboro, VT).
To do single-cell analysis, a cell was injected into a 150 µm
o.d., 30 µm i.d. fused silica capillary (Polymicro Technologies,
Phoenix, AZ) following previously reported procedures. 16,17 After
the cell lysis in the capillary, the MEKC-LIF procedure was done
as described for bulk analysis. Upon completion of the separation,
the capillary was washed for 2 min with 0.1 M NaOH and 5 min
with MEKC running buffer between runs.
Separate calibration curves for OH-TPP-E + and R123, (area
in electropherogram versus amount injected) were built by
injecting standards of these compounds. The resulting calibration
curves for OH-TPP-E + and R123 were
m ) (122.4 ( 0.4)A + (1.0 ( 0.6);
OH-TPP-E+ OH-TPP-E+
(R 2 ) 0.99) (13)
mR123 ) (12.9 ( 0.0)AR123 + (0.3 ( 0.1); (R 2 ) 0.99)
(14)
where m is given in attomoles, A is the peak area in the
electropherogram, and the errors are standard deviations. These
(15) Zielonka, J.; Vasquez-Vivar, J.; Kalyanaraman, B. Nat. Protoc. 2008, 3, 8.
(16) Krylov, S. N.; Starke, D. A.; Arriaga, E. A.; Zhang, Z.; Chan, N. W.; Palcic,
M. M.; Dovichi, N. J. Anal. Chem. 2000, 72, 872.
(17) Anderson, A. B.; Gergen, J.; Arriaga, E. A. J. Chromatogr., B: Anal. Technol
Biomed. Life Sci. 2002, 769, 97.
Figure 1. Electropherograms of TPP-HE oxidation products and
R123 in the whole cell lysate (∼2 × 10 6 myoblasts) and the blank
buffer. Separations were performed in a 42 cm-long capillary at -400
V/cm in MEKC running buffer. The 488 nm line of an argon-ion laser
was used for excitation, and a 607-663 nm bandpass filter was used
for detection. Traces have been offset vertically for clarity. The
respective electrophoretic mobilities of OH-TPP-E + and R123 were
-(6.89 ( 0.11) × 10 -4 and -(6.41 ( 0.09) × 10 -4 cm 2 V -1 s -1 (n )
3). Separation efficiencies were calculated to be ∼260 000 for OH-
TPP-E + and ∼31 000 for R123.
equations were used to carry out calculations described by the
eqs 10 and 11. The limits of detection (LOD at signal/noise ) 3)
were ∼2 amol for OH-TPP-E + and ∼0.2 amol for R123. It would
be possible to obtain a better LOD (e.g., zeptomole levels) for
R123 by using a different bandpass filter (e.g., centered at 530
nm). This alternative was not pursued here.
Data Analysis. The MEKC electropherograms were analyzed
using Igor Pro 5.0 software (Wavemetrics, Lake Oswego, OR).
Superoxide concentrations in single myoblasts under each condition
were reported as mean ± standard deviation (SD). A student’s
t test was used to determine statistical significance of the data,
with p values of

of OH-TPP-E + in the whole cell lysate (cf. eq 13), on average
each cell contains ∼7.7 amol. However, this bulk analysis is biased
because it ignores that cells do not necessarily have the same
membrane potential. 18-20
It is important to bear in mind that superoxide dismutases
(SODs) are present in the matrix (Mn-SOD) and outside the
mitochondria (Cu, Zn-SOD). 5 In general SODs transform superoxide
into hydrogen peroxide (k ∼ 2 × 10 9 M -1 s -1 ), 21-23
outcompeting the relatively slow reaction of superoxide with the
TPP-HE probe (i.e., k ∼ 4 × 10 6 M -1 s -1 ). 5 However, in the
mitochondrial matrix, the mitochondrial membrane potential
increases the TPP-HE concentration ∼100-fold relative to the
cytosol 7 that partially compensates for its relative slow reaction
rate constant with superoxide. The concentration of Mn-SOD
in the matrix has been reported to be ∼20 µM. 24 On the basis
of this value and the kinetic rate constants of TPP-HE and Mn-
SOD with superoxide (and assuming that there are no other
reactions consuming superoxide), the estimated rate of OH-
TPP-E + accumulation is ∼10% of the formation rate of H2O2
by Mn-SOD. This simplified kinetic analysis argues that, at this
relative rate of formation, the detected amounts of OH-TPP-
E + are adequate to calculate [O2 - ]average,inside (eq 12), even in
the presence of competing Mn-SOD.
On the other hand, outside the mitochondria there is no
enhanced accumulation of TPP-HE. Given a concentration of Cu,
Zn-SOD ∼6 µM, 25 the rate of OH-TPP-E + accumulation outside
the mitochondria is only ∼0.3% of the rate of H2O2 formed by
Cu, Zn-SOD. This calculation further confirms that OH-TPP-
E + produced outside the mitochondria does not contribute
significantly to the total OH-TPP-E + detected (i.e., “x” ineq8
is small; cf. Supporting Information, Part I).
Single Cell Superoxide Quantitative Analysis. We used
chemical cytometry to analyze mitochondrial matrix superoxide
levels in single cells (Figure 2). In order to correct for variations
in mitochondrial membrane potentials observed among individual
cells, cells were treated with TPP-HE and R123. The MEKC-LIF
separation and detection of OH-TPP-E + and R123 in single
myoblasts provides the measurements needed to determine
superoxide at the single-cell level (cf. eq 12).
The detection of mitochondrial matrix superoxide at the single
cell level was further confirmed by treatment with tiron and CCCP.
Tiron is an SOD mimetic that scavenges intracellular superoxide
in cultured cells. 6,26 As expected, OH-TPP-E + was not detected
in single myoblasts treated with tiron (Figure 2). The R123 peak
was still detected showing that tiron treatment does not compro-
(18) Johnson, L. V.; Walsh, M. L.; Bockus, B. J.; Chen, L. B. J. Cell Biol. 1981,
88, 526.
(19) Ludovico, P.; Sansonetty, F.; Corte-Real, M. Microbiology 2001, 147, 3335.
(20) Heerdt, B. G.; Houston, M. A.; Augenlicht, L. H. Cancer Res. 2005, 65,
9861.
(21) Klug, D.; Rabani, J.; Fridovich, I. J. Biol. Chem. 1972, 247, 4839.
(22) Pick, M.; Rabani, J.; Yost, F.; Fridovich, I. J. Am. Chem. Soc. 1974, 96,
7329.
(23) Behar, D.; Czapski, G.; Rabani, J.; Dorfman, L. M.; Schwarz, H. A. J. Phys.
Chem. 1970, 74, 3209.
(24) Quijano, C.; Hernandez-Saavedra, D.; Castro, L.; McCord, J. M.; Freeman,
B. A.; Radi, R. J. Biol. Chem. 2001, 276, 11631.
(25) Chang, L. Y.; Slot, J. W.; Geuze, H. J.; Crapo, J. D. J. Cell Biol. 1988, 107,
2169.
(26) Yamada, J.; Yoshimura, S.; Yamakawa, H.; Sawada, M.; Nakagawa, M.; Hara,
S.; Kaku, Y.; Iwama, T.; Naganawa, T.; Banno, Y.; Nakashima, S.; Sakai, N.
Neurosci. Res. 2003, 45, 1.
6748 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 2. Electropherograms of TPP-HE oxidation products and
R123 in individual myoblasts under basal conditions and upon
treatments with tiron and CCCP. Separations and detection conditions
were same to those described for Figure 1.
mise the mitochondrial membrane potential. CCCP is a proton
ionophore that dissipates the mitochondrial membrane potential. 27,28
Upon CCCP treatment both R123 and TPP-HE concentrations in
the matrix decrease ∼100-fold relative to untreated cells (Supporting
Information, Parts II and III). The expected R123 amount
in a CCCP-treated cell is ∼0.2 amol of R123 (Supporting Information,
Part III), which is undetectable in the MEKC-LIF system
used in this study (i.e., limit of detection ∼0.2 amol of R123).
Furthermore, the low TPP-HE concentration in CCCP-treated cells
significantly decreases the rate of formation of OH-TPP-E + inside
the mitochondria (cf. eq 5; Supporting Information, Part II). The
expected amount of OH-TPP-E + formed in a CCCP-treated
myoblast is ∼0.4 amol, which is undetectable in the MEKC-
LIF that was used in this study (i.e., limit of detection ∼2 amol
of OH-TPP-E + ). Thus, the concomitant disappearance of the
R123 and the OH-TPP-E + peaks in CCCP-treated myoblasts
(Figure 2) demonstrates that the mitochondrial membrane
potential is vital to the accumulation of TPP-HE and formation of
OH-TPP-E + in the mitochondrial matrix. 5
On the basis of fluorescence microscopy, individual myoblasts
display ∼39% variation in mitochondrial membrane potential
(Supporting Information, Part V; Figure S-3). In MEKC-LIF, the
ratio of the areas of OH-TPP-E + and R123 corrects for variations
in mitochondrial membrane potential between individual cells
(cf. eq 12). This calculation is based on the assumptions that both
R123 and TPP-HE accumulate into the mitochondria according
to a Nernstian behavior (cf. eqs 1 and 2) and that the amount of
superoxide product outside the mitochondria is only a small
fraction of that in the mitochondrial matrix (cf. eqs 7 and 8;
Supporting Information, Part I). On the basis of MEKC-LIF, the
steady state superoxide concentration in the mitochondrial matrix
of single myoblasts under basal conditions was ∼(0.29 ± 0.10) ×
10 -12 M with a RSD of 35% (n ) 12) (Figure 4). This variance
is statistically different from that obtained if the signal of OH-
(27) Heytler, P. G.; Prichard, W. W. Biochem. Biophys. Res. Commun. 1962, 7,
272.
(28) Lim, M. L.; Minamikawa, T.; Nagley, P. FEBS Lett. 2001, 503, 69.

Figure 3. Electropherograms of TPP-HE oxidation products and
R123 in individual myoblasts under basal conditions and upon
treatments with rotenone and antimycin. Separations and detection
conditions were same to those described for Figure 1.
TPP-E + is not corrected by membrane potential (RSD ∼69%)
(F-test: p < 0.05). In summary, the simultaneous monitoring
of OH-TPP-E + and R123 provides a correction for variations in
membrane potential and provides the means to quantitate
mitochondrial matrix superoxide levels in single cells.
Effects of Mitochondrial Respiratory Inhibitors. The mitochondria
is considered a major source of superoxide in cells,
where superoxide is mainly released by complex I and III in the
mitochondrial electron transport chain (ETC). 1,2 Rotenone and
antimycin A are two respiratory inhibitors that block electron
transfer through complex I and III, respectively, thereby stimulating
superoxide production. 29,30 Although recent studies have
utilized the TPP-HE probe to investigate the effects of rotenone
and antimycin A on intracellular superoxide release in single cells
by fluorescent microscopy and/or flow cytometry, 5,10-12 these
studies are qualitative because they have not considered the effect
of mitochondrial membrane potential.
Here we demonstrated that chemical cytometry is suitable
to quantify mitochondrial matrix superoxide levels in individual
myoblasts upon their treatment with either rotenone or antimycin
A (Figure 3). Some studies have reported such inhibitors
at elevated concentrations may alter the membrane potential
of isolated mitochondria. 31-34 In this report, bulk analyses
showed that the treatments with rotenone and antimycin A did
appear to alter the mitochondria membrane potentials because
there were no significant changes in the ratio of
mOH-TPP-E + ,outside to mOH-TPP-E + ,inside compared to basal conditions
(29) Grivennikova, V. G.; Vinogradov, A. D. Biochim. Biophys. Acta 2006, 1757,
553.
(30) Muller, F. L.; Liu, Y. H.; Van Remmen, H. J. Biol. Chem. 2004, 279, 49064.
(31) Petit, P. X.; O’Connor, J. E.; Grunwald, D.; Brown, S. C. Eur. J. Biochem.
1990, 194, 389.
(32) Kataoka, M.; Fukura, Y.; Shinohara, Y.; Baba, Y. Electrophoresis 2005, 26,
3025.
(33) Tzung, S. P.; Kim, K. M.; Basanez, G.; Giedt, C. D.; Simon, J.; Zimmerberg,
J.; Zhang, K. Y.; Hockenbery, D. M. Nat. Cell Biol. 2001, 3, 183.
(34) De Bari, L.; Atlante, A.; Guaragnella, N.; Principato, G.; Passarella, S.
Biochem. J. 2002, 365, 391.
Figure 4. Effects of rotenone and antimycin A on the superoxide
levels detected in the mitochondrial matrix of single myoblasts. Data
of individual cells as well as means ( SD are presented for each
condition (n ) 12 cells for basal, 10 for rotenone, and 12 for antimycin
A). * stands for p < 0.05 vs basal.
(Supporting Information, Part I). Thus, eq 12 is adequate to
determine mitochondrial matrix superoxide levels following
inhibitor treatments.
In single myoblasts treated with rotenone or antimycin A, the
steady state superoxide concentrations in the mitochondrial matrix
were (0.97 ± 0.61) × 10 -12 M(n ) 10) and (2.15 ± 1.20) × 10 -12
M(n ) 12), respectively (Figure 4). These values are ∼3.5- and
8.2-fold higher compared to basal levels. The magnitude of such
changes are comparable with the ∼2 to 3-fold change observed
in aged cells or cells associated with diabetic hyperglycemia
relative to their respective controls. 35-37 Therefore, the use of
inhibitors demonstrates that chemical cytometry is a suitable
approach to quantitate mitochondrial superoxide levels expected
in aging and disease related studies.
CONCLUDING REMARKS
Here we report the quantitation of mitochondrial matrix
superoxide levels in single cells using a MEKC-based chemical
cytometry method that is not biased by the mitochondrial
membrane potential. On the basis of eq 12, the ratio of OH-
TPP-E + and R123 signals corrects for variation in membrane
potential of mitochondria among individual cells. According
to this method, the mitochondrial matrix superoxide levels
of single myoblasts are in the picomolar range. The method
could also be extended to investigate other cellular systems
such as cultured single skeletal muscle fibers, 38,39 whose
superoxide levels were analyzed qualitatively in our earlier
(35) Chen, H.; Cangello, D.; Benson, S.; Folmer, J.; Zhu, H.; Trush, M. A.; Zirkin,
B. R. Exp. Gerontol. 2001, 36, 1361.
(36) Klamt, F.; Gottfried, C.; Tramontina, F.; Dal-Pizzol, F.; Da Frota, M. L., Jr.;
Moreira, J. C.; Dias, R. D.; Moriguchi, E.; Wofchuk, S.; Souza, D. O.
Neuroreport 2002, 13, 1515.
(37) Nishikawa, T.; Edelstein, D.; Du, X. L.; Yamagishi, S.-i.; Matsumura, T.;
Kaneda, Y.; Yorek, M. A.; Beebe, D.; Oates, P. J.; Hammes, H.-P.; Giardino,
I.; Brownlee, M. Nature 2000, 404, 787.
(38) Pye, D.; Palomero, J.; Kabayo, T.; Jackson, M. J. J. Physiol. 2007, 581,
309.
(39) Palomero, J.; Pye, D.; Kabayo, T.; Spiller, D. G.; Jackson, M. J. Antioxid.
Redox Signaling 2008, 10, 1463.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6749

study. 40 The method may also be modified to monitor superoxide
released outside the mitochondria, which would be
particularly useful to investigate cell lines derived from Cu, Zn-
SOD-deficient animal models. 41,42 Long-term application of
chemical cytometry to quantify superoxide levels may find wide
applications to the fields of toxicology, aging, and oxidativestress
related disease. 4,43,44
ACKNOWLEDGMENT
The National Institutes of Health supported this work through
Grant R01-AG-20866. We thank Dr. Margaret Donoghue for
providing comments on the manuscript.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
(40) Xu, X.; Thompson, L. V.; Navratil, M.; Arriaga, E. A. Anal. Chem. 2010,
82, 4570–4576.
(41) Reaume, A. G.; Elliott, J. L.; Hoffman, E. K.; Kowall, N. W.; Ferrante, R. J.;
Siwek, D. F.; Wilcox, H. M.; Flood, D. G.; Beal, M. F.; Brown, R. H., Jr.; Received for review June 7, 2010. Accepted July 4, 2010.
Scott, R. W.; Snider, W. D. Nat. Genet. 1996, 13, 43.
AC101509D
(42) Veerareddy, S.; Cooke, C. L.; Baker, P. N.; Davidge, S. T. Am. J. Physiol.
Heart Circ. Physiol. 2004, 287, H40.
(43) Thompson, L. V. Exp. Gerontol. 2009, 44, 106. (44) Amacher, D. E. Curr. Med. Chem. 2005, 12, 1829.
6750 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010

Anal. Chem. 2010, 82, 6751–6755
Resolving Disulfide Structural Isoforms of IgG2
Monoclonal Antibodies by Ion Mobility Mass
Spectrometry
Dhanashri Bagal, † John F. Valliere-Douglass, ‡ Alain Balland,* ,‡ and Paul D. Schnier* ,†
Molecular Structure, Amgen, Thousand Oaks, California 91320, and Process and Product Development, Amgen,
Seattle, Washington 98119
Recombinant monoclonal antibodies are an important
class of therapeutic agents that have found widespread
use for the treatment of many human diseases. Here, we
have examined the utility of ion mobility mass spectrometry
(IMMS) for the rapid characterization of disulfide
variants in intact IgG2 monoclonal antibodies. It is shown
that IMMS reveals 2 to 3 gas-phase conformer populations
for IgG2s. In contrast, a single gas-phase conformer is
revealed using IMMS for both an IgG1 antibody and a Cys-
232 f Ser mutant IgG2, both of which are homogeneous
with respect to disulfide bonding. This provides strong
evidence that the observed IgG2 gas-phase conformers
are related to disulfide bond heterogeneity. Additionally,
IMMS analysis of redox enriched disulfide isoforms allows
assignment of the mobility peaks to established disulfide
bonding patterns. These data clearly illustrate how IMMS
can be used to quickly provide information on the higher
order structure of antibody therapeutics.
The overall structure of the immunoglobulin G (IgG) family
is organized in 12 subdomains, each closed by an intrachain
disulfide bond. 1 Heavy chain (HC) and light chain (LC) are
connected by interchain disulfide bonds to form a covalent
complex of the form (HC-LC)2. IgG1, IgG2, and IgG4 isotypes
share greater than 90% sequence homology in their constant
domains but differ significantly in the hinge region. IgG1 and IgG4
hinge core sequences are very similar with two cysteines on each
heavy chain involved in interheavy chain connection, whereas
IgG2 is unique in presenting four cysteine residues in the hinge
region, notably two consecutive residues, Cys-232 and Cys-233
(amino acid numbering of Kabat et al.), 2 that have no equivalent
in any other immunoglobulin subclass. Researchers at Amgen
recently reported that these residues confer distinctive structural
features to the human IgG2 isotype resulting in the formation of
disulfide-related structural isoforms. 3,4
Three distinct structural isoforms (IgG2-A, IgG2-B, and
IgG2-A/B) 3,4 specific to human IgG2s were revealed by chro-
* To whom correspondence should be addressed. E-mail: ballanda@amgen.com
(A.B.); pschnier@amgen.com (P.D.S.).
† Molecular Structure.
‡ Process and Product Development.
(1) Padlan, E. A. Adv. Protein Chem. 1996, 49, 57–133.
(2) Kabat, E. A.; Wu, T. T.; Perry, H. M.; Gottesman, K. S.; Foeller, C. Sequences
of Proteins of Immunological Interest, 5th ed.; U.S. Public Health Service,
NIH: Washington, DC., 1991.
matographic and electrophoretic methods including capillary
electrophoresis with sodium dodecyl sulfate (CE-SDS), 3,5 reversedphase
high performance liquid chromatography (RP-HPLC), 4 and
cation exchange chromatography (CEX). 3,6 IgG2-A corresponds
to the classical model with independent Fab and Fc domains
connected by the hinge (Figure 1a). 7 IgG2-B is a symmetrical form
with HC and LC covalently linked to the hinge by disulfide bridges
(Figure 1b). IgG2-A/B is an asymmetrical form intermediate
between A and B. Detailed analysis of each IgG2 kappa structural
isoforms showed that the different interchain disulfide bond
arrangements involved only four residues: the cysteine in constant
region one of the heavy chain (CH1), Cys-127 (Kabat numbering),
the cysteine at the C-terminus of the light chain, Cys-214, and
two cysteines in the upper hinge region, specific to the IgG2
subclass, Cys-232 and Cys-233. The precise cysteine connectivity
of each structural form was established by partial reductionalkylation
and mass spectrometry (MS) 8 and Edman sequencing
MS. 9
Modeling of the IgG2 sequence based on the three-dimensional
antibody structure places the four cysteines in close spatial
proximity, supporting the concept that a variable arrangement of
these residues could generate IgG2 structural isoforms. Two
specific cysteine-to-serine mutants were designed at positions 232
and 233 to disrupt potential disulfide rearrangements. 10 These
mutants both exhibited no significant difference in expression and
potency characteristics when compared to wild type IgG2 but
proved structurally homogeneous with respect to the disulfide
bonding of the IgG2-A type. 10
(3) Wypych, J.; Li, M.; Guo, A.; Zhang, Z.; Martinez, T.; Allen, M. J.; Fodor, S.;
Kelner, D. N.; Flynn, G. C.; Liu, Y. D.; Bondarenko, P. V.; Ricci, M. S.;
Dillon, T. M.; Balland, A. J. Biol. Chem. 2008, 283, 16194–16205.
(4) Dillon, T. M.; Ricci, M. S.; Vezina, C.; Flynn, G. C.; Liu, Y. D.; Rehder,
D. S.; Plant, M.; Henkle, B.; Li, Y.; Deechongkit, S.; Varnum, B.; Wypych,
J.; Balland, A.; Bondarenko, P. V. J. Biol. Chem. 2008, 283, 16206–16215.
(5) Guo, A.; Han, M.; Martinez, T.; Ketchem, R. R.; Novick, S.; Jochheim, C.;
Balland, A. Electrophoresis 2008, 29, 2550–2556.
(6) Zhang, Y.; G. A.; Novick, S.; Jochheim, C.; Boyce, J. M.; Gerhart, M.; Qin,
X.; Gombotz, W. I. Bioprocessing J. 2003, (Nov-Dec), 37–43.
(7) Milstein, C.; Frangione, B. Biochem. J. 1971, 121, 217–225.
(8) Martinez, T.; Guo, A.; Allen, M. J.; Han, M.; Pace, D.; Jones, J.; Gillespie,
R.; Ketchem, R. R.; Zhang, Y.; Balland, A. Biochemistry 2008, 47, 7496–
7508.
(9) Zhang, B.; Harder, A. G.; Connelly, H. M.; Maheu, L. L.; Cockrill, S. L.
Anal. Chem. 2009, 82, 1090–1099.
(10) Allen, M. J.; Guo, A.; Martinez, T.; Han, M.; Flynn, G. C.; Wypych, J.; Liu,
Y. D.; Shen, W. D.; Dillon, T. M.; Vezina, C.; Balland, A. Biochemistry 2009,
48, 3755–3766.
10.1021/ac1013139 © 2010 American Chemical Society 6751
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/21/2010

Figure 1. Illustration of the hinge region disulfide bonding pattern
of human (a) IgG2-A and (b) IgG2-B antibodies.
The ability to rapidly detect and characterize IgG2 isoforms is
of great interest, as it may help to facilitate the transition of new
IgG2 molecules from discovery into development and ultimately
commercialization. In recent years, mass spectrometry has played
an increasingly important role in the analytical characterization
of IgG therapeutics. Mass spectrometry is now widely used to
confirm the intact molecular weight of IgGs, 11,12 establish their
glycosylation profile, 13,14 and confirm 15 or establish 16 the primary
(11) Gadgil, H. S.; Pipes, G. D.; Dillon, T. M.; Treuheit, M. J.; Bondarenko, P. V.
J. Am. Soc. Mass Spectrom. 2006, 17, 867–872.
(12) Brady, L. J.; Valliere-Douglass, J.; Martinez, T.; Balland, A. J. Am. Soc. Mass
Spectrom. 2008, 19, 502–509.
(13) Damen, C. W. N.; Chen, W.; Chakraborty, A. B.; van Oosterhout, M.;
Mazzeo, J. R.; Gebler, J. C.; Schellens, J. H. M.; Rosing, H.; Beijnen, J. H.
J. Am. Soc. Mass Spectrom. 2009, 20, 2021–2033.
(14) Olivova, P.; Chen, W.; Chakraborty, A. B.; Gebler, J. C. Rapid Commun.
Mass Spectrom. 2008, 22, 29–40.
(15) Ren, D.; Pipes, G. D.; Hambly, D.; Bondarenko, P. V.; Treuheit, M. J.; Gadgil,
H. S. Anal. Biochem. 2009, 384, 42–48.
(16) Bandeira, N.; Pham, V.; Pevzner, P.; Arnott, D.; Lill, J. R. Nat. Biotechnol.
2008, 26, 1336–1338.
6752 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
structure with a high degree of detail. Although MS is not
routinely used to characterize higher order structural elements
in IgGs, MS coupled with hydrogen/deuterium exchange was
recently demonstrated as a method to characterize the conformational
dynamics of IgG1 antibodies in solution. 17
Ion mobility mass spectrometry (IMMS) has shown great
promise as an intact protein separation and analysis methodology
to probe higher order structural elements including the overall
size/shapeofbiopolymersandlargemacromolecularassemblies. 18-28
Recently, Waters Corporation commercialized an ion mobility
mass spectrometer (Synapt) based on traveling waves (T-Wave). 29
In the T-Wave implementation of ion mobility, ion separation
occurs when a sequence of dc pulses push ions through the
mobility cell in the presence of an inert gas at relatively high
pressure. 29,30 The ability of an ion to “surf” the T-wave depends
on its collision cross section (CCS). Ions with compact structures
are pushed through the mobility cell faster than ions with more
elongated structures. In this work, we present evidence that
T-Wave IMMS can be used to separate disulfide variants of intact
IgG2 antibodies. Attractive features of the method include high
sensitivity (µg sample consumption), minimal sample preparation,
and fast analysis time (minutes).
EXPERIMENTAL SECTION
Human monoclonal antibodies mAb#1 (IgG2), mAb#2 (IgG1),
and mAb#3 (IgG2) were produced recombinantly in Chinese
hamster ovary (CHO) cells and purified at Amgen. All additional
reagents were purchased from Sigma-Aldrich (St. Louis, MO)
unless otherwise specified. For control experiments, constant
region two (CH2) domain N-glycans were removed by adding
1500 U of PNGase F (New England Biolabs, Ipswich, MA) per
100 µg of protein and incubating at 37 °C for 16 h.
Disulfide isoforms IgG2-A and IgG2-B were selectively enriched
using methods established by Dillon et al. 4 Briefly, to
enrich isoform B, IgG2s were incubated in 200 mM Tris buffer
(17) Houde, D.; Arndt, J.; Domeier, W.; Berkowitz, S.; Engen, J. R. Anal. Chem.
2009, 81, 2644–2651.
(18) Clemmer, D. E.; Hudgins, R. R.; Jarrold, M. F. J. Am. Chem. Soc. 1995,
117, 10141–10142.
(19) Bohrer, B. C.; Merenbloom, S. I.; Koeniger, S. L.; Hilderbrand, A. E.;
Clemmer, D. E. Annu. Rev. Anal. Chem. 2008, 1, 293–327.
(20) Ruotolo, B. T.; Giles, K.; Campuzano, I.; Sandercock, A. M.; Bateman, R. H.;
Robinson, C. V. Science 2005, 310, 1658–1661.
(21) Kaddis, C. S.; Loo, J. A. Anal. Chem. 2007, 79, 1778–1784.
(22) Leary, J. A.; Schenauer, M. R.; Stefanescu, R.; Andaya, A.; Ruotolo, B. T.;
Robinson, C. V.; Thalassinos, K.; Scrivens, J. H.; Sokabe, M.; Hershey,
J. W. B. J. Am. Soc. Mass Spectrom. 2009, 20, 1699–1706.
(23) Kim, H. I.; Kim, H.; Pang, E. S.; Ryu, E. K.; Beegle, L. W.; Loo, J. A.;
Goddard, W. A.; Kanik, I. Anal. Chem. 2009, 81, 8289–8297.
(24) Atmanene, C. D.; Wagner-Rousset, E.; Malissard, M.; Chol, B.; Robert, A.;
Corvaïa, N.; Dorsselaer, A. V.; Beck, A.; Sanglier-Cianferani, S. Anal. Chem.
2009, 81, 6364–6373.
(25) Ruotolo, B. T.; Hyung, S.-J.; Robinson, P. M.; Giles, K.; Bateman, R. H.;
Robinson, C. V. Angew. Chem., Int. Ed. 2007, 46, 8001–8004.
(26) Hilton, G. R.; Thalassinos, K.; Grabenauer, M.; Sanghera, N.; Slade, S. E.;
Wyttenbach, T.; Robinson, P. J.; Pinheiro, T. J. T.; Bowers, M. T.; Scrivens,
J. H. J. Am. Soc. Mass Spectrom. 2010, 21, 845–854.
(27) Schenauer, M. R.; Meissen, J. K.; Seo, Y.; Ames, J. B.; Leary, J. A. Anal.
Chem. 2009, 81, 10179–10185.
(28) Thalassinos, K.; Grabenauer, M.; Slade, S. E.; Hilton, G. R.; Bowers, M. T.;
Scrivens, J. H. Anal. Chem. 2008, 81, 248–254.
(29) Pringle, S. D.; Giles, K.; Wildgoose, J. L.; Williams, J. P.; Slade, S. E.;
Thalassinos, K.; Bateman, R. H.; Bowers, M. T.; Scrivens, J. H. Int. J. Mass
Spectrom. 2007, 261, 1–12.
(30) Shvartsburg, A. A.; Smith, R. D. Anal. Chem. 2008, 80, 9689–9699.

(pH 8.0) with cysteine and cystamine at concentrations of 6 and
1 mM, respectively. Isoform-A was enriched by incubating the
IgG2 under the same conditions but with the addition of 1.0 M
guanidinium chloride (GuHCl) to the buffer. The samples were
protected from light at 2-8 °C for approximately 48 h.
The IgG2 mAb#3 was subjected to Cysf Ser mutagenesis at
position 232 as described by Allen et al. 10 Site-directed mutagenesis
was performed using the QuickChange XL site-directed mutagenesis
kit (Stratagene, La Jolla, CA). The mutant was stably transfected into
a serum-free suspension-adapted CHO cell line. 31 Following production,
the mAb was purified by protein A affinity chromatography.
Incorporation of the expected mutation was verified on the purified
molecule by size exclusion chromatography (SEC) MS. 12
For IMMS analysis, IgG samples were buffer exchanged and
concentrated using Vivaspin 30 kDa molecular weight cutoff filters
(GE Healthcare, Buckinghamshire, UK). For experiments performed
under native conditions, the IgGs were diluted using 160 mM
ammonium acetate (pH not adjusted) to a final working concentration
of ∼3 µM.
Mass spectrometry experiments were performed using a
hybrid ion mobility quadrupole time-of-flight MS (Synapt, Waters
Inc., Milford, MA) equipped with a nanoelectrospray ionization
(ESI) source using metal coated borosilicate glass capillaries
(nanoflow probe tips, long thin walled, Waters Corporation).
Solution flow rates of ∼75 nL/min and an ESI capillary voltage of
∼1.3 kV were used for all experiments. The source temperature
was 50 °C, and the pressure of the vacuum/backing region was
3.5 mbar. Each ion mobility mass spectrum was acquired from
m/z 4000-8000 every 2 s; approximately 12 counts per scan were
observed. The signal was typically averaged for approximately 10
min. Gentle source conditions were used to minimize gas phase
unfolding of the protein (sample cone: 40 V, trap voltage: 10 V,
transfer lens: 12 V, bias: 25; cone gas: 30 L/Hr; P TRAP(Ar): 0.0175
mbar). Nitrogen was used as the mobility carrier gas, and the
following parameters were found to give optimal ion mobility
separation (PIMS: 0.5 mbar, wave velocity: 300 m/s, wave height:
9.8 V). The instrument was mass calibrated using a 50 µg/µL
CsI solution. Waters’ raw data files were translated to Matlab
binary files using software developed in-house. Data processing
and plotting were performed in Matlab (The MathWorks Inc.
Natick, MA) and Igor Pro (WaveMetrics Inc., Lake Oswego, OR).
RESULTS AND DISCUSSION
Antibodies belonging to the IgG2 subclass exist as a group of
distinct isoforms with different disulfide connectivities between
the Fab domain and the hinge region of the molecule. The goal
of this study was to investigate the ability of IMMS to successfully
separate these discrete IgG2 isoforms.
A nano-ESI ion mobility mass spectrum for a solution of 3 µM
mAb#1 (theoretical MW for the most abundant glycoform (G1F/
G1F): 149821.50 Da) in 160 mM ammonium acetate is shown
Figure 2a. A narrow charge state distribution centered on the 24+
ion is observed. Ion mobility spectra for all MAbs analyzed were
acquired with minimal acceleration voltages (sample cone: 40 V,
trap: 10 V, transfer: 12 V). These gentle tuning conditions were
found to provide optimal resolution in the ion mobility dimension
(31) Rasmussen, B.; Davis, R.; Thomas, J.; Reddy, P. Cytotechnology 1998, 28,
31–42.
Figure 2. Ion mobility TOF mass spectra of (a) mAb#1 and (c) mAb#2.
The normalized ion mobility intensities are graphed as contour plots with
9 levels from 4% to 100% intensity. Extracted arrival time distributions
for the 26+ charge states of (b) mAb#1 and (d) mAb#2.
(vide infra). However, with these conditions, the mass spectral
peaks are relatively broad making it impossible to observe the
individual glycoforms of the IgG. The glycoforms for these IgG
molecules can, however, be readily resolved using higher source
voltages (data not shown). Robinson and co-workers have previously
observed that optimal conditions for the mass and mobility
dimensions are often not compatible on the Synapt instrument
when analyzing large protein assemblies. 32
Ion mobility separation of mAb#1 (Figure 2a) reveals two distinct
conformer populations. For example, the mobility spectrum for the
26+ charge state (Figure 2b) shows two distinct peaks with a 1.1
ms difference in drift times (9.3 and 10.4 ms, respectively). Similar
distributions are observed for each charge state. However, with the
tuning conditions employed here, the best mobility resolution was
achieved for the 27+, 26+, and 25+ charge states. To elucidate if
the heterogeneity of the sugar chains on the antibody influences the
observed gas-phase conformers, deglycosylated mAb#1 was analyzed.
The same distribution of gas-phase conformers is revealed in the
ion mobility mass spectrum of mAb#1 with the sugars released
(Figure S1a, Supporting Information). This clearly demonstrates that
these multiple conformers are not caused by sugar heterogeneity
but instead are related to conformational differences in the protein
chains. The arrival time distributions for each charge state virtually
overlaps with that of the glycosylated protein. For example, the major
peaks in the arrival time distribution plot of the 26+ charge state of
deglycosylated mAb#1 are only shifted by -0.13 ms compared to
(32) Ruotolo, B. T.; Benesch, J. L. P.; Sandercock, A. M.; Hyung, S.-J.; Robinson,
C. V. Nat. Protoc. 2008, 3, 1139–1152.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6753

that of the glycosylated IgG (Figure S1b, Supporting Information).
This indicates that the glycosylated and deglycosylated antibodies
have nearly identical gas-phase collision cross sections.
The relative abundance of each gas-phase conformer can be
roughly estimated by fitting the arrival time distribution to a sum
of log-normal functions and integrating the relative area of each
peak. The log-normal function is an asymmetric Gaussian function
whose logarithm is normally distributed. 33 The log-normal function
was used because it was empirically found to best fit the
experimental peak shape of this ion mobility data. For the 26+
charge state, the normalized area of peaks 1 and 2 from the best
fits are 42% and 58%, respectively. Similar areas are observed for
other charge states. These abundances are similar to the relative
peak areas observed in electropherograms of IgG2s separated with
capillary electrophoresis. We and others have shown CE-SDS to
be a resolving technique for the separation of IgG2 structural
isoforms. 3,5,35 However, this correlation alone does not signify that
the resolved gas-phase conformers are definitively due to disulfide
variants.
To elucidate whether the gas-phase conformers observed for
IgG2 molecules are related to their disulfide connectivity, we
analyzed an IgG1 antibody, mAb#2 (theoretical MW for the most
abundant glycoform (G1F/G0F):148408.0 Da), as a control (Figure
2c). The most significant difference between human IgG1 and
IgG2 subclasses is the primary structure of the hinge region,
resulting in the absence of disulfide related isoforms in the IgG1. 3,4
In contrast to the IgG2 mobility data, for each charge state of
mAb#2, the arrival time profile is relatively narrow and consists
of a single uniform distribution (Figure 2c). For example, the
arrival time profile for the 26+ charge state of mAb#2 (Figure
2d) shows a single peak at 8.7 ms. This arrival time profile is
representative of the distribution observed for each of the charge
states. This suggests that the multiple conformers observed of
mAb#1 are due to disulfide variants in the antibody.
To further demonstrate that the observed gas-phase conformers
are indeed IgG2 disulfide variants, individual disulfide isoforms
were selectively enriched in a refolding experiment using redox
chemistry employing cysteine/cystamine. Dillon et al. previously
demonstrated that isoform IgG2-A and IgG2-B can be redoxenriched
by refolding with and without 1 M GuHCl, respectively. 4
Under redox conditions in buffer alone, IgG2-A is refolded to IgG2-
B. Liu et al. demonstrated that a slow conversion of IgG2-A to
IgG2-B also occurs in vivo. 34 Isoform conversion toward IgG2-A
requires in vitro refolding in presence of low levels of chaotropic
reagents. 4 Figure 3a,b shows the arrival time distributions for the
+26 charge state for redox enriched mAb#1 in the presence of
guanidine (isoform A) and redox enriched mAb#1 in the absence
of guanidine (isoform B), respectively. In contrast to IMMS for
the untreated Mab#1 antibody (Figure 3, dashed line), a single
abundant conformer is observed for each of these enriched
isoforms. As a control, the IgG1 antibody, mAb#2, was also
subjected to the same redox refolding protocol with or without
guanidine. All treated IgG1 samples have identical arrival time
(33) Brown, R. Personal Eng. Instrum. News 1991, 8, 51–54.
(34) Liu, Y. D.; Chen, X.; Enk, J. Z.; Plant, M.; Dillon, T. M.; Flynn, G. C. J. Biol.
Chem. 2008, 283, 29266–29272.
(35) Lacher, N. A.; Wang, Q.; Roberts, R. K.; Holovics, H. J.; Aykent, S.; Schlittler,
M. R.; Thompson, M. R.; Demarest, C. W. Electrophoresis 2010, 31, 448–
458.
6754 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. Arrival time distributions for the 26+ charge states of IgGs
treated with redox reagents (cystamine, cysteine). (a,b) mAb#1 and
(c,d) mAb#2 (control). Samples represented in (a) and (c) had 1 M
GuHCl added to the buffer. The dashed line shows the arrival time
distribution for the 26 + charge state of untreated mAb#1.
distribution profiles as the untreated IgG1 molecule (Figure 3c,d),
demonstrating that this refolding protocol does not affect the
overall tertiary structure of the antibody. Comparing the untreated
IgG2 IMMS trace with the enriched isoform distributions (Figure
3a,b) identifies peaks 1 (9.3 ms) and 2 (10.4 ms) in the mobility
spectra as isoform A and isoform B, respectively. The relatively
late arrival time of isoform B indicates that this form of the
antibody has a larger gas-phase collision cross section compared
to isoform A. The IMMS resolution of the isoforms correlates with
the previously observed capillary electrophoresis separation. 5,8,35
With CE, IgG2 isoforms were resolved into two peaks, with IgG2-A
migrating more rapidly than the IgG2-B isoform. The intermediate
IgG2-A/B forms were split between the two peaks, IgG2-A/B1
migrating with -A and IgG2-A/B2 with -B. 8
While the disulfide connectivities of IgG2 isoforms have been
well characterized, 3,4,8,9 only limited information is available
describing the overall tertiary structure of human IgGs. To
investigate if the relative ordering of gas-phase collision cross
sections (IgG1 ≈ IgG2-A < IgG2-B) correlates with IgG solution
structures, collision cross sections were calculated for two IgG
structures using Waters’ CCS software. 36 The calculated cross
section of an IgG antibody, with a disulfide bonding pattern
consistent with the B isoform, is 8385 Å 2 (unpublished results).
This value is 3% smaller than the calculated cross section of
an IgG1 antibody (protein data bank code 1HZH, 8653 Å 2 ). 37
(36) Williams, J. P.; Lough, J. A.; Campuzano, I.; Richardson, K.; Sadle, P. J.
Rapid Commun. Mass Spectrom. 2009, 23, 3563.
(37) Saphire, E. O.; Parren, P. W.; Pantophlet, R.; Zwick, M. B.; Morris, G. M.;
Rudd, P. M.; Dwek, R. A.; Stanfield, R. L.; Burton, D. R.; Wilson, I. A. Science
2001, 293, 1155–1159.

Figure 4. Ion mobility mass spectra of (a) mAb#3 and (b) Cys f Ser Mab#3 mutant. The extracted arrival time distributions for the 25+ charge
states of these molecules are shown in (c) and (d), respectively.
These limited calculations using individual static structures do
not explain the relatively late arrival time of the IgG2-B isoform
and suggest that the overall three-dimensional structure of
these gas-phase ions may be quite different than the solution
structure. While the IMMS results demonstrate that ion
mobility separates covalent disulfide mediated IgG2 structural
isoforms, the overall three-dimensional structure of these gasphase
ions is presently not clear.
To investigate the generality of these ion mobility results,
measurements were also performed on a different IgG2 antibody
(mAb#3) which also exists as an ensemble of disulfide mediated
isoforms (Figure 4a). Ion mobility separation of mAb#3 reveals
two abundant gas-phase conformer populations and one minor
conformer, for each charge state. These three populations are
readily apparent in the arrival time distribution for the 25+ charge
state (Figure 4c) which shows three distinct peaks with drift times
of 9.5, 10.3, and 11.9 ms. The relative abundance is determined
by fitting the arrival time distribution to a sum of log-normal
functions of each conformer and is roughly estimated to be 46%,
49%, and 5%. Figure 4b, shows the ion mobility mass spectrum
for a mutant form of this antibody with a single point mutation
introduced at amino acid 232 (Cys f Ser) by site-directed
mutagenesis. A single narrow peak is observed for the arrival time
distribution of each charge state. The arrival time profile of the
25+ charge state is shown in Figure 4d as an illustrative example
demonstrating a homogeneous population, with a single peak at
9.3 ms. This peak corresponds to the gas-phase conformer with
the more compact structure (isoform A, vide supra). This is
consistent with previous studies demonstrating that this mutant
is homogeneous with respect to disulfide bonding and of the
IgG2-A type. 10,38 This result also unambiguously confirms our
assertion that the multiple IMMS peaks observed in this study of
selected IgG2s can be correlated with the disulfide bonding
patterns in these molecules.
(38) Lightle, S.; Aykent, S.; Lacher, N.; Mitaksov, V.; Wells, K.; Zobel, J.; Oliphant,
T. Protein Sci. 2010, 19, 753–762.
CONCLUSIONS
The results of the present study demonstrate that ion mobility
as a shape-selective separation methodology can be used to detect
disulfide heterogeneity in large (150 kDa) intact IgG2 antibodies.
Two to three gas-phase conformers are observed by ion mobility
for IgG2 antibodies. These gas-phase conformers were maintained
with deglycosylated IgG2s. Analysis of redox refolded IgG2s as
well as an IgG2 with a Cys f Ser single point mutation clearly
demonstrates that the observed gas-phase conformers are related
to disulfide variants. Ion mobility is fast (millisecond measurements),
sensitive (nanomole), and amenable to high throughput
automation. IMMS is a powerful new methodology for the
characterization of intact antibodies and may be useful to routinely
fingerprint higher order structure of these protein biopharmaceuticals
in the near future. We are currently extending these
measurements to investigate the utility of IMMS for the analysis
of IgG2s containing lambda light chains as well as to directly
characterize the binding of antigen targets to individual disulfide
isoforms of IgG2 antibodies.
ACKNOWLEDGMENT
We are grateful to Allen Sickmier and Leszek Poppe for
insightful discussions, Keith Richardson (Waters Inc.) for providing
the Waters’ CCS software, Mike Berke, Rick Stanton, and
Mikhail Toupikov for help with the data analysis software, and
Mike Treuheit, Dean Pettit, Peter Grandsard, and Philip Tagari
for championing and supporting this work. We also thank our
Amgen colleagues, whose names are listed in the references, for
their expert contributions to the collective knowledge built
recently on IgG2 isoforms.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 19, 2010. Accepted July 6, 2010.
AC1013139
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6755

Anal. Chem. 2010, 82, 6756–6763
An Enzymatic Microreactor Based on Chaotic
Micromixing for Enhanced Amperometric
Detection in a Continuous Glucose Monitoring
Application
Byeong-Ui Moon, †,‡ Sander Koster, § Klaas J. C. Wientjes, † Radosław M. Kwapiszewski, |
Adelbert J. M. Schoonen, † Ben H. C. Westerink, † and Elisabeth Verpoorte* ,‡
Biomonitoring and Sensoring, Pharmaceutical Analysis, Department of Pharmacy, University of Groningen, Antonius
Deusinglaan 1, P.O. Box 196, 9700 AD Groningen, The Netherlands, TNO Quality of Life, Utrechtseweg 48,
3700 AJ Zeist, The Netherlands, and Department of Microbioanalytics, Faculty of Chemistry, Warsaw University of
Technology, Noakowskiego 3, Warsaw, 00-664, Poland
The development of continuous glucose monitoring systems
is a major trend in diabetes-related research. Small,
easy-to-wear systems which are robust enough to function
over many days without maintenance are the goal. We
present a new sensing system for continuous glucose
monitoring based on a microreactor incorporating chaotic
mixing channels. Two different types of chaotic mixing
channels with arrays of either slanted or herringbone
grooves were fabricated in poly(dimethylsiloxane) (PDMS)
and compared to channels containing no grooves. Mixing
in channels with slanted grooves was characterized using
a fluorescence method as a function of distance and at
different flow rates, and compared to the mixing behavior
observed in channels with no grooves. For electrochemical
detection, a thin-film Pt electrode was positioned at the
end of the fluidic channel as an on-chip detector of the
reaction product, H 2O2. Glucose determination was
performed by rapidly mixing glucose and glucose
oxidase (GOx) in solution at a flow rate of 0.5 µL/min
and 1.5 µL/min, respectively. A 150 U/mL GOx
solution was selected as the optimum concentration
of enzyme. In order to investigate the dependence of
device response on flow rate, experiments with a
premixed solution of glucose and GOx were compared
to experiments in which glucose and GOx were reacted
on-chip. Calibration curves for glucose (0-20 mM, in
the clinical range of interest) were obtained in channels
with and without grooves, using amperometric detection
and a 150 U/mL GOx solution for in-chip reaction.
1. INTRODUCTION
Diabetes mellitus is a widespread disease causing heart
disease, weight loss, blurry vision, neurological disorders, and
* Corresponding author. E-mail: e.m.j.verpoorte@rug.nl, Telephone: +31-50-
363-3337; fax: +31-50-363-7582.
† Biomonitoring and Sensoring, University of Groningen.
‡ Department of Pharmacy, University of Groningen.
§ TNO Quality of Life.
| Warsaw University of Technology.
6756 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
even death. 1 Proper management of blood glucose is thus of
crucial importance for diabetic patients. The conventional way
blood glucose determinations are carried out involves the
finger-prick method. Usually, diabetic patients measure their
own blood glucose several times per day by applying a drop of
blood to a portable device. However, this intermittent monitoring
does not yield the full range of information necessary to
effectively control glucose levels 24 h a day. The current
research trend is thus toward real-time in vivo monitoring over
longer periods (days). 2-5 Type 1 diabetic patients in particular
could benefit from a portable glucose sensor to keep their
glucose values within a reasonable range. Theoretically, a
portable glucose sensor could include an insulin pump in a
feedback loop to serve as an artificial pancreas.
Electrochemical detection is commonly used for glucose
analysis. 6 The advantages of electrochemical detection include
sensitivity and ease of interfacing with detection electronics.
Several reports incorporating electrochemical detection on a
microchip have been published for the detection of glucose. 7-9
The principle of the enzymatic reaction involved is shown in
eqs 1 and 2. Glucose, O2 and glucose oxidase (GOx) react to
generate hydrogen peroxide (H2O2) at concentrations proportional
to the original glucose concentrations. The H2O2
is then oxidized under applied potential into O2 and H + , with
the resulting electrons being recorded as an electrical
current.
(1) Patient Education Association. http://www.nlm.nih.gov/medlineplus/
tutorials/diabetesintroduction/htm/index.htm. (Accessed July 12, 2010).
(2) Klonoff, D. C. Diabetes Care 2005, 28, 1231–1239.
(3) Feldman, B.; Brazg, R.; Schwartz, S.; Weinstein, R. Diabetes Technol. Ther.
2003, 5, 769–779.
(4) Lutgers, H. L.; Hullegie, L. M.; Hoogenberg, K.; Sluiter, W. J.; Dullaart,
R. P. F.; Wientjes, K. J.; Schoonen, A. J. M. Neth. J. Med. 2000, 57, 7–12.
(5) Tamada, J. A.; Garg, S.; Jovanovic, L.; Pitzer, K. R.; Fermi, S.; Potts, R. O.
JAMA, J. Am. Med. Assoc. 1999, 282, 1839–1844.
(6) Updike, S. J.; Hicks, G. P. Nature 1967, 214, 986–988.
(7) Wang, J.; Chatrathi, M. P.; Tian, B.; Polsky, R. Anal. Chem. 2000, 72, 2514–
2518.
(8) Wilke, R.; Büttgenbach, S. Biosens. Bioelectron. 2003, 19, 149–153.
(9) Yamaguchi, A.; Jin, P.; Tsuchiyama, H.; Masuda, T.; Sun, K.; Matsuo, S.;
Misawa, H. Anal. Chim. Acta 2002, 468, 143–152.
10.1021/ac1000509 © 2010 American Chemical Society
Published on Web 07/21/2010

GOx
glucose + O298 gluconic acid +
0.7 V
H2O298 H 2 O 2 (GOx: glucose oxidase) (1)
2H + + O 2 + 2e -
The reaction may be carried out by mixing the reagents in solution
to produce H2O2 which is then subsequently detected by
integrated electrodes, 7,8 or by immobilizing GOx directly onto
the detection electrode surface to react with glucose in solution. 9
The electrochemical detectors in these microchips displayed fast
response times and low detection limits. Immobilizing GOx
certainly reduces its consumption significantly compared to
devices in which GOx is added in solution. However, the major
disadvantage of this approach is that the enzyme layer is prone
to biofouling in in vivo applications, which might cause a decrease
in activity over time and poor performance as a result. 10,11
Over the past decade, a number of glucose sensor designs
have been developed and become available for continuous
monitoring purposes. 12-16 Mastrototaro et al. 14 reported an in
vivo needle-type continuous glucose monitoring system (CGMS),
which is currently being used by consumers as a kit (MiniMed).
The detector in this system is based on the reaction of immobilized
glucose oxidase on the electrode with glucose. Schoemaker et
al. 15 and Wientjes et al. 17 have both reported a subcutaneous
continuous glucose monitoring system (SCGM) based on a
microdialysis probe and using the solution-based enzymatic
reaction with glucose. They reported reliable measurements in
diabetes patients over periods of four days 18 up to two weeks. 19,20
However, due to their bulky size and the need for connecting
tubes between system components, these systems exhibited
relatively long physical lag times of 30 min or more. Pickup et
al. 13 reviewed the clinical use of glucose monitoring systems and
introduced new technology for noninvasive glucose sensing, based
on near-infrared spectroscopy and fluorescence. Wentholt et al. 21
(10) Gerritsen, M.; Jansen, J. A.; Lutterman, J. A. Neth. J. Med. 1999, 54, 167–
179.
(11) Wisniewski, N.; Moussy, F.; Reichert, W. M. Fresenius’ J. Anal. Chem. 2000,
366, 611–621.
(12) Maran, A.; Crepaldi, C.; Tiengo, A.; Grassi, G.; Vitali, E.; Pagano, G.; Bistoni,
S.; Calabrese, G.; Santeusanio, F.; Leonetti, F.; Ribaudo, M.; Di Mario, U.;
Annuzzi, G.; Genovese, S.; Riccardi, G.; Previti, M.; Cucinotta, D.; Giorgino,
F.; Bellomo, A.; Giorgino, R.; Poscia, A.; Varalli, M. Diabetes Care 2002,
25, 347–352.
(13) Pickup, J. C.; Hussain, F.; Evans, N. D.; Sachedina, N. Biosens. Bioelectron.
2005, 20, 1897–1902.
(14) Mastrototaro, J. J. Diabetes Technol. Ther. 2000, 2, 13–18.
(15) Schoemaker, M.; Andreis, E.; Roper, J.; Kotulla, R.; Lodwig, V.; Obermaier,
K.; Stephan, P.; Reuschling, W.; Rutschmann, M.; Schwaninger, R.;
Wittmann, U.; Rinne, H.; Kontschieder, H.; Strohmeier, W. Diabetes Technol.
Ther. 2003, 5, 599–608.
(16) Suzuki, H.; Tokuda, T.; Miyagishi, T.; Yoshida, H.; Honda, N. Sens. Actuators,
B 2004, 97, 90–97.
(17) Wientjes, K. J. C.; Grob, U.; Hattemer, A.; Hoogenberg, K.; Jungheim, K.;
Kapitza, C.; Schoonen, A. J. M. Diabetes Technol. Ther. 2003, 5, 615–620.
(18) Kapitza, C.; Lodwig, V.; Obermaier, K.; Wientjes, K. J. C.; Hoogenberg, K.;
Jungheim, K.; Heinemann, L. Diabetes Technol. Ther. 2003, 5, 609–614.
(19) Wientjes, K. J.; Vonk, P.; Vonk-van Klei, Y.; Schoonen, A. J.; Kossen, N. W.
Diabetes Care 1998, 21, 1481–1488.
(20) Schoonen, A. J. M.; Schmidt, F. J.; Hasper, H.; Verbrugge, D. A.; Tiessen,
R. G.; Lerk, C. F. Biosens. Bioelectron. 1990, 5, 37–46.
(21) Wentholt, I. M. E.; Hoekstra, J. B. L.; DeVries, J. H. Diabetes Technol. Ther.
2007, 9, 399–409.
(2)
reported an overview of the current applications and clinically
relevant aspects of continuous glucose monitors (CGMs), with
emphasis on the calibration procedure, interpretation of continuous
glucose data, and some important limitations. Overall, these
authors concluded that improved accuracy, reliability for longer
periods of time, miniaturization, and cost-effectiveness are the
main issues which need to be considered in the further development
of continuous glucose monitoring systems. 21
Chip-based microfluidic technologies are a good alternative to
improve on conventional monitoring approaches such as the
SCGM described above. Microfluidic systems, also termed
“miniaturized total analysis systems (µTAS)” 22 or “lab-on-a-chip”,
are now widely used in analytical chemistry and biological
applications. Advantages of these systems include dramatically
reduced consumption of chemical reagents, faster reaction times
and cost-effectiveness. Micro SCGM systems can thus be envisaged
which exploit the solution-based reaction of GOx with
glucose in nL volumes in micrometer-sized channels for glucose
sensing. The speed of this analysis is determined by the efficiency
of mixing reagents in the microfluidic channels. Mixing at the
micrometer scale is a challenge, as flows are generally extremely
well-defined and laminar. Mixing of two solution streams in a
straight microchannel is possible only through means of diffusion,
a passive molecular transport process which is very slow. There
has therefore been an enormous amount of research done in the
past decade or so on how to implement efficient mixing at the nL
scale. So-called passive micromixers are generally preferred for
many micro analytical flow systems, since these elements do not
require the application of an external force to achieve mixing. A
large number of passive micromixers have been reported, including
a planar laminar flow mixer, 23 a cross-shaped micromixer 24
and a droplet mixer. 25 The approach chosen for our work is one
based on chaotic mixing, first described by Stroock et al. 26,27
Mixing is achieved through the incorporation of an array of
microgrooves into a microchannel. Flow over the groove array
assumes a helical or corkscrew pattern, in which the contact area
between two adjacent solutions is increased dramatically to
facilitate mixing by diffusion.
The long-term goal of the present project is to realize an
autonomous, portable sensing system for continuous in vivo
glucose monitoring, based on the reaction in solution of GOx with
glucose to produce H 2O2. To accomplish this, we have designed
a miniaturized glucose sensing system based on microdialysis
sampling and lab-on-a-chip technology. In this system, nL
amounts of sample and enzyme rapidly mix and react. As highrecovery
microdialysis requires flow rates in tissue less than 1
µL/min, 17 we have adapted the microreactor dimensions
according to these conditions. In this paper, we describe a new
application for chaotic mixing, that is, the efficient and fast
mixing of GOx and glucose for reaction. To that end, either
(22) Reyes, D. R.; Iossifidis, D.; Auroux, P.-A.; Manz, A. Anal. Chem. 2002, 74,
2623–2636.
(23) Melin, J.; Gimenez, G.; Roxhed, N.; van der Wijngaart, W.; Stemme, G.
Lab Chip 2004, 4, 214–219.
(24) Wong, S. H.; Bryant, P.; Ward, M.; Wharton, C. Sens. Actuators, B 2003,
95, 414–424.
(25) Paik, P.; Pamula, V. K.; Fair, R. B. Lab Chip 2003, 3, 253–259.
(26) Stroock, A. D.; Dertinger, S. K. W.; Ajdari, A.; Mezic, I.; Stone, H. A.;
Whitesides, G. M. Science 2002, 295, 647–651.
(27) Stroock, A. D.; Dertinger, S. K.; Whitesides, G. M.; Ajdari, A. Anal. Chem.
2002, 74, 5306–5312.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6757

slanted or herringbone grooves were fabricated in poly(dimethylsiloxane)
(PDMS) chips, and the mixing characteristics
of these channels compared to channels without grooves. A
thin-film Pt electrode at the end of the fluidic channel served
as electrochemical detector of the reaction product, H2O2.
2. MATERIALS AND METHODS
2.1. Chemicals and Reagents. All chemicals were analytical
reagent-grade. Fluorescein and potassium iodide (KI) were
purchased from Sigma-Aldrich (Germany) and used to prepare
0.002 M fluorescein solution and 0.2 M KI solution, respectively.
Potassium ferricyanide (K3Fe(CN)6) and potassium ferrocyanide
(K4Fe(CN) 6) were supplied by Sigma-Aldrich (Germany) and
used to prepare a1mMsolution in a 20 mM phosphate buffer
(pH 7.2). pH was measured using a pH meter (inoLab pH 720,
WTW, Germany). The phosphate buffer was prepared by
mixing solutions of NaH2PO4 · H2O (63 mL) and Na2HPO4 · 2H2O
(100 mL) containing 0.1 M potassium nitrate (KNO3) asa
supporting electrolyte. H2O2 (30%) was supplied by VWR (The
Netherlands) and prepared freshly every day at various
concentrations in phosphate buffer. D-glucose and GOx were
supplied from Merck (Germany) and used to prepare 100 mM
and 5000 U/mL stock solutions, respectively. D-glucose was
stored in a refrigerator at 2 °C and used for 1 month. GOx
was stored at -20 °C and used for 3 months. All solutions were
prepared with 18 MΩ ultrapure water purified in an Arium 611
(Sartorius Stedim Biotech, Germany).
2.2. Microfluidic Chip Fabrication. The first microchannels
were constructed by standard microfabrication and replicated in
the silicone rubber, PDMS (Sylgard 184, Dow Corning, U.S.). The
chip layout and design were drawn using the program, Clewin
(Wieweb software, Hengelo, The Netherlands). The structure on
the silicon master, which served as a mold, was processed with
two steps of standard photolithography (Figure 1, right column). 28
A 4 in. p-type (100) silicon wafer (Si-Mat, Germany) (525 µm
thickness) was employed as a substrate. The silicon wafer was
first cleaned sequentially with acetone, isopropyl alcohol, and
deionized water, and dried with N2 gas. The wafer was then
treated with hexamethyldisilazane (HMDS) (Sigma-Aldrich,
Germany) in a vacuum desiccator for 30 min to improve
adhesion of the photoresist (PR). A thick positive PR, AZ4562
(Microchemicals GmbH, Germany), was coated on the silicon
wafer using a spin-coater at 1000 rpm for 3 s and baked at 100
°C for 30 min. The resulting PR layer was 35 µm thick. After
rehydration at ambient temperature for 3 h, the coated wafer
was exposed to ultraviolet (UV) light (365 nm, 10 mW/cm 2 )
using a photomask printed on a high-resolution transparency
to pattern the microchannels (resolution 3,810 dpi; Pro-Art BV,
Groningen, The Netherlands). The exposed PR was then
removed by dipping in a developer solution (AZ351B: deionized
water (DI) ) 1:3, Microchemicals GmbH, Germany) and
agitating the beaker for 20 min, leaving behind PR ridges for
microchannel replication. The wafer was then rinsed in DI
water and dried with N 2 gas. Contrary to usual protocol, no
postbake step was undertaken for this layer. Rather, the
substrate was again exposed to UV light for 10 s using a second
transparency photomask to pattern the groove arrays. The
(28) Kim, J.; Heo, J.; Crooks, R. M. Langmuir 2006, 22, 10130–10134.
6758 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 1. Fabrication process of microfluidic chip. See text for
details.
patterned PR was again immersed in a developer (AZ351B: DI
) 1:4) solution for 6 min to form groove structures on top of
the microchannel structures.
PDMS resin and a curing agent were used for microchannel
replication in a PDMS slab. The two liquids were mixed at a
weight ratio of 10:1 PDMS: curing agent, and the solution left to
stand at ambient conditions for 20 min to remove air bubbles.
Afterward, the mixture was poured over the prepared silicon
master. A polycarbonate alignment piece was used together with
fused-silica capillaries to form access holes to the microchannels
during the replication process. Holes (300 µm in diameter) were
drilled in the polycarbonate using a CNC machine (Sherline, U.S.)
at points corresponding to the locations of the required inlets/
outlets when this piece was aligned with the silicon master. Fusedsilica
capillaries were then inserted through the holes and aligned
on the microchannel structures on the silicon master. The
prepolymer mixture was allowed to cure against the master at
50 °C for4h.
2.3. Thin-Film Electrode Formation. Platinum (Pt) sensing
electrodes were formed on a glass wafer by a standard photolithography
and lift-off process (Figure 1, left column). A 4 in., 525-
µm-thick borofloat glass wafer (Telic, U.S.) was used as a substrate
for the electrode patterning. The glass wafer was first cleaned
using a piranha solution (96% H 2SO4: 30% H2O2 ratio of 3:1), then
thoroughly rinsed in DI water and dried with N2 gas. (WARN-
ING: Piranha solution must be handled with extreme caution,
as it is highly oxidizing and reacts explosively when it comes
into contact with easily oxidized organic solvents. This solution
must be used in a well-ventilated fumehood with absolutely no
organic solvents in the vicinity. Explosive reactions also occur
in the presence of trace amounts of metals such as Pt, Ag, and
Mn. Metal tweezers must not be used when cleaning wafers
in piranha.) The wafer was then treated with HMDS in a
vacuum desiccator for 30 min. An image reversal PR 5214E
(Microchemicals GmbH, Germany) was coated using a spin-

coater at 2000 rpm for 20 s, and the substrate was placed on a
hot plate at 105 °C for 50 s to soft-bake the resist and remove
solvents. Subsequently, the substrate was exposed to UV light
(365 nm, 10 mW/cm 2 ) for 2 s using a photomask with electrode
pattern. After reversal baking of the substrate at 115 °C for 2
min, a flood exposure was performed for 20 s without mask to
solubilize unexposed areas, resulting in a negative image of
the mask. Finally, the substrate was immersed in a developer
solution (AZ351B: DI water, ratio 1:4) to remove soluble PR
and expose the glass surface where the electrodes were to be
formed.
A lift-off process was performed to obtain Pt sensing electrodes
on the patterned glass wafer. A Ti thin-film layer (20 nm) was
deposited over the entire wafer by E-beam evaporation (Temescal,
U.S.) under a high vacuum of 9 × 10 -6 Torr. Ti was deposited
first, as it adheres better to glass than Pt, and in turn, Pt
adheres well to Ti. Subsequently, a Pt thin-film layer (150 nm)
was grown by E-beam evaporation at 2 × 10 -5 Torr. The glass
substrate was then immersed in acetone to dissolve the
remaining PR layer, thereby simultaneously lifting off the thin
Pt/Ti film on top of it. The metal remained behind in the
predefined electrode regions. Groups of three Pt electrodes
were patterned; the patterned electrode dimensions were 100
× 1500 µm, and the distance between two electrodes was 30
µm.
2.4. Chip Integration. The dimensions of the channel and
grooves on the structured silicon master were determined using
a stylus profilometer (Veeco Instrument BV, The Netherlands)
before casting PDMS. There are two inlets and one outlet (waste)
to the microreactor channel. The two inlet channels are half as
wide as the microreactor channel after the Y-junction, to ensure
a constant linear flow velocity throughout the device upon
introduction of two solutions at the same flow rate. A 1:1 dilution
of solutions will also result under these conditions of operation.
The inlet channels are both 100 µm wide, 35 µm deep, and 5 mm
in length. A ruler is located along the channel to show the distance
from the Y-junction. The total length of channel from the Y-junction
is 22 mm, while the width and depth of the channel are 200 and
35 µm, respectively, and groove widths are 50 µm. The 122
grooves in the mixing channel were 6-8 µm deep and were
patterned at an angle of 45° with respect to the central axis of the
channel. The total volume of the mixing/reaction channel is about
150 nL.
In order to bond the PDMS slab to the glass chip with the
electrodes, the PDMS was smoothly peeled off from the master
and cut. The two devices were first treated with UV-generated
ozone for 15 min to oxidize the PDMS surface, which creates
hydrophilic properties and improves bonding strength. 29 Subsequently,
the PDMS slab and glass were immediately aligned under
a microscope and brought into contact with each other. The
assembled chip was then placed on a hot plate at 140 °C and
allowed to cool down to room temperature, after which the chip
was irreversibly bonded. The whole bonding procedure was
carried out in the cleanroom. Figure 2 shows a fabricated
microfluidic chip and examples of the three different types of
PDMS channels with integrated electrodes investigated in this
(29) Berdichevsky, Y.; Khandurina, J.; Guttman, A.; Lo, Y. H. Sens. Actuators, B
2004, 97, 402–408.
Figure 2. (a) Photograph of the detection region of a fabricated
microfluidic chip. (b) Three different types of PDMS channels with
integrated electrodes: from top to bottom, channels with no grooves,
slanted grooves, and herringbone grooves. (c) Schematic diagram
of grooves in a channel structure. Note that the grooves were 6-8
µm deep.
study, containing either no grooves, slanted grooves, or herringbone
grooves.
2.5. Experimental Setup for Fluorescence. In order to
characterize mixing, fluorescence detection was used to gain
quantitative information about the degree of mixing along the
length of the channel. A fluorescence microscope, model “DMIL”
(Leica Microsystems, The Netherlands), was equipped with a 10×
objective, a mercury arc lamp, and a CCD camera. The mixing
process was visualized using a fluorescein filter, which is made
for 488 nm excitation and 518 nm emission. Fluorescence
quenching was carried out for quantitative analysis using 2 mM
fluorescein and 200 mM potassium iodide (KI). KI quenches the
fluorescence produced by fluorescein when it is excited at 488
nm. Fluorescein and KI were introduced separately through the
inlets into the Y-junction at various flow rates, and the image was
captured with a CCD camera using a 20.1 ms exposure time, a
gamma setting of 1.0, and a gain of 1.0. The images were acquired
at different distances from the Y-junction and analyzed by
determining the standard deviation (SD) of the intensity distribution
across the entire channel using Lx95P image analysis
software. Flow rates were controlled by syringe pumps (ProSense,
The Netherlands) connected with silica capillaries to the chip
inlets using 350-µm-o.d. and 250-µm-i.d. silica capillaries.
2.6. Experimental Setup for Electrochemical Detection.
Cyclic voltammetry and chronoamperometry were carried out
using a potentiostat, “Electrochemical Analyzer” (CH Instrument,
U.S.), interfaced to a computer. Two thin-film Pt electrodes were
used, one as a working electrode (WE), the other as a counter
electrode (CE). The Pt electrodes were located 21 mm downstream
from the Y-junction. The active area of the Pt WE was
0.02 mm 2 (200-µm-wide channel ×100-µm-wide electrode). A
Ag/AgCl wire was used as a reference electrode (RE) and
positioned in the reservoir at the end of the device. The RE
was prepared by dipping a short piece of Ag wire (250-µmdiameter)
for 5 s into a crucible containing melted silver
chloride (Sigma-Aldrich, Germany). For the glucose analysis
experiments, a 1:3 flow rate ratio of glucose sample to GOx
solution was introduced through the two inlets, and measurements
were carried out under continuous flow conditions
controlled by the syringe pumps. (For microdialysis experi-
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6759

Figure 3. Electrochemical detection experimental setup.
ments, the microdialysis probe outlet will be connected to the
glucose sample inlet of the device using a simple, low-deadvolume
tubing connection.) For the comparison of system
response to premixed glucose-GOx solutions and direct reaction
of glucose and GOx in the channel, premixed solution was
prepared using a 1:3 volume ratio of 5.6 mM glucose and 150
U/mL GOx in a vial. After adding the two solutions, the vial
was gently shaken 10 times and introduced in the reactor
channel through both inlets approximately 5 min later after
shaking. A two-position actuator switching valve (Valco Instrument
Co., Inc.) was adapted to alternatively introduce buffer
solution and glucose sample solution to measure background
and sensing signals, respectively. Once the flow rate had been
set each time, the valve was switched from buffer solution to
sample solution. The same solution was used for measurements
at each flow rate, and current measurements were made
continuously as flow rate was varied. All experiments were
carried out at room temperature, and data were saved directly
on a computer. A schematic diagram of the electrochemical
detection experimental setup is shown in Figure 3.
3. RESULTS AND DISCUSSION
3.1. Characterization of the Micromixers. A fabricated
micromixer was characterized by capturing images at different
distances from the Y-junction using a fluorescence microscope
and determining the variation in fluorescence intensity across the
width of the channel. The standard deviation (SD) was calculated
using the following eq 3: 30
SD ) � 1
N
N ∑ i)1
(x i - x¯) 2
where xi is the gray-scale intensity value of pixel i, and x¯ is the
mean intensity value of pixels across the entire channel.
Quantitative values of the SD of mixing efficiency varied
between 0.5 for completely unmixed solutions (variation of
intensity from 0 for KI solution to 100% fluorescence for
fluorescein) and 0 for completely mixed solutions (intensity
equal across the channel).
Figure 4 shows the SD of intensity versus distance as a function
of flow rate, comparing channels with slanted grooves and no
(30) Xia, H. M.; Wan, S. Y. M.; Shu, C.; Chew, Y. T. Lab Chip 2005, 5, 748–
755.
6760 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(3)
Figure 4. Comparison of microfluidic mixers having no grooves and
slanted grooves as a function of distance from the Y-junction and at
different flow rates (one device, n ) 3). The flow rate in each case is
the total flow rate in the mixing channel. The flow rates of fluorescein
and KI at the inlets are each half of the total flow rate.
grooves. Flow rates that fall in the range required to achieve high
glucose recoveries in microdialysis sampling of subcutaneous
tissue were considered. The results showed that grooved channels
have a higher mixing efficiency compared to channels without
grooves. In fact, mixing in channels with slanted grooves is almost
complete at a distance of 1 cm along the channel at total flow
rates ranging from 0.4 to 2 µL/min (Reynolds number ) 0.08 at
1 µL/min). This result confirms that grooves formed in the
channel ceiling at an oblique angle with respect to flow direction
enhance the mixing rate of two adjacent solution streams.
Differences in measured standard deviations of fluorescence
intensity at shorter distances along the channel are most likely
due to the reproducibility of the experiment itself. The depth and
width of the grooves were designed based on Stroock’s geometric
parameters (relative groove height to channel height, R < 0.3;
channel height, h E width of the channel, w) and have values of
6-8 and 50 µm, respectively, as shown in Figure 2 (c). 27 These
initial results of mixing characterization were used as important
information for further mixing and reaction experiments.
3.2. Cyclic Voltammograms of Three Different Types of
Microfluidic Channel. Cyclic voltammograms were carried out
to check the electrochemical properties of the microfluidic chip.
A ferri/ferrocyanide couple provides an ideal electrochemical
model for a preliminary chip test with integrated electrodes. The
redox couple ferri/ferrocyanide reaction is as follows (eq 4):
3- - 4-
Fe(CN) 6 + e a Fe(CN)6 E° ) 0.356 V (4)
First, the redox curve was examined with a stationary flow when
1mM(K3Fe(CN)6) was introduced in the channel at a scan
range of -0.3 to 0.7 V. The oxidation and reduction peaks were
observed at 0.21 V (35 nA current) and 0.10 V (56 nA current)
respectively, at a scan rate of 100 mV/s (data not shown). For
cyclic voltammetry under continuous flow conditions, 1 mM
Fe(CN)6 3- /1 mM Fe(CN)6 4- was prepared in 20 mM phosphatebuffered
solution (pH 7.2) containing 0.1 M potassium nitrate
as a supporting electrolyte. Experiments were performed at a

Figure 5. Comparison of cyclic voltammograms for the three different
types of microfluidic channel. A 20 mM phosphate-buffered solution
(pH 7.2) containing 1 mM Fe(CN)6 3- /Fe(CN)6 4- in 0.1 M potassium
nitrate was used at a flow rate of 1 µL/min and a scan rate of 20
mV/s.
flow rate of 1 µL/min and a scan rate of 20 mV/s. The cyclic
voltammograms recorded over an applied potential range of
0-0.6 V versus Ag/AgCl are shown in Figure 5. In these early
devices, the array of mixing grooves extended over the entire
length of the channel, including the integrated electrodes. Higher
oxidation and reduction currents were observed for channels with
slanted or herringbone grooves than for unstructured channels.
This was probably caused by higher local velocities over the
electrode surfaces in the chaotic mixers, leading to thinner
diffusion layers at these surfaces and thus improved analyte
delivery and higher currents. Because of the observed, rather
uncontrolled local flow effect on detector response, grooves were
not structured over the electrodes in later devices used in the
rest of this study. Instead, the groove patterns ended more than
1 mm upstream from the electrodes to reduce the flow effect on
the electrode surface, as shown Figure 2(b). Comparing structured
channels, the herringbone grooves did not yield a significant
improvement in electrochemical signal with respect to the slanted
grooves in the channel. It was therefore decided to focus further
studies on comparing channels with no grooves to channels
containing slanted grooves. This was because channels with
slanted grooves were easier to fabricate than channels with
herringbone grooves; aligning the photomask for slanted grooves
in the second photolithographic step was simpler than aligning
the mask for herringbone grooves.
A linear sweep voltammogram was recorded by introducing 6
mM H 2O2 in 0.1 M KNO3 solution, in order to decide on an
applied potential for glucose measurement. The onset of
oxidation current was observed at 0.3 V and showed a plateau
around 0.7 V, at a total flow rate of 1 µL/min and a scan rate
of 100 mV/s (data not shown). Therefore, this applied potential
was selected for subsequent glucose experiments.
3.3. Optimization of GOx Concentration. Prior to the
optimization of GOx concentration, electrode variation from one
chip to the next was characterized using a premixed solution of
20 mM glucose and GOx solution. One example of each mixer
type (grooved and ungrooved) was selected, and signal variation
was tested three times with each device. The results of these
Figure 6. Amperometric response arising from reaction of 20 mM
glucose with GOx at different concentrations in channels without
grooves and with slanted grooves (one device, n ) 3). The flow rates
of glucose and GOx were 0.5 and 1.5 µL/min, respectively; an
example of raw data showing background signal and sensing signal
is shown in the inset.
experiments showed that although current values varied from one
device to the next, current values fell within the same range. Two
devices, one with a grooved micromixer, the other an ungrooved
micromixer, were selected for subsequent experiments to compare
the effect of no grooves versus slanted grooves in the channel.
Since the concentration of GOx plays an important role in the
reaction, various concentrations of GOx solutions were tested in
the two channel types. To date, only a few papers have reported
on-chip glucose sensing by mixing glucose sample and free
enzyme in solution rather than using immobilized enzyme on
electrode surfaces. 8,31-33 Pijanowska et al. 32 developed a flowthrough
amperometric sensor implemented in a silicon-glass
sandwich construction having a volume of 5 µL. The glucose
sample was reacted in a reaction cell using only two different
enzyme activities, 22 and 132 U/mL GOx, with a higher sensitivity
being obtained with the latter concentration. Wang et al. 7 proposed
a microchip capillary electrophoresis (MCE) approach for glucose
sensing, in which GOx solution was injected together with glucose
sample from separate reservoirs into the separation channel. A
75 U/mL GOx solution proved sufficient for these researchers to
obtain a calibration curve.
Figure 6 shows the result of amperometric response arising
from reaction of 20 mM glucose with GOx at different concentrations
in channels with no grooves and slanted grooves. The flow
rates were set at 0.5 µL/min glucose and 1.5 µL/min GOx (1:3
mixing ratio) in order to avoid the oxygen depletion effect. 34 This
is an effect which has been observed at higher concentrations of
glucose for both electrochemical sensors utilizing immobilized
GOx and devices based on the reaction of GOx and glucose in
solution. The sensitivity of sensor response decreases because
(31) Wang, J.; Chatrathi, M. P.; Collins, G. E. Anal. Chim. Acta 2007, 585,
11–16.
(32) Pijanowska, D. G.; Sprenkels, A. J.; Olthuis, W.; Bergveld, P. Sens. Actuators,
B 2003, 91, 98–102.
(33) Böhm, S.; Pijanowska, D.; Olthuis, W.; Bergveld, P. Biosens. Bioelectron.
2001, 16, 391–397.
(34) Wientjes, K. J. C. Ph.D. Dissertation, University of Groningen, Groningen,
The Netherlands, 2000.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6761

the amount of oxygen available in the solution is not sufficient
for full conversion of glucose conversion to H2O2 with enzyme
(see also eq 1). In effect, oxygen is depleted from the solution
by the enzyme reaction. To prevent the effect of oxygen depletion
in measurements employing immobilized enzymes on electrodes,
electrodes have been prepared with a semipermeable coating to
reduce the amount of glucose diffusing to the electrode. Other
approaches have involved the use of mediators, molecules which
take on the role of oxygen in the reaction. 3,35 In early experiments,
we also observed that the signal was dramatically decreased and
calibration curves were not linear at higher concentrations of
glucose when the mixing ratio between glucose and GOx was
1:1; instead, the signal tended to level off. It was decided to
increase the flow rate of GOx with respect to glucose sample to
a ratio of 3:1, diluting the effective glucose concentration in the
channel by a factor of 4 to prevent oxygen depletion.
All experiments were performed three times and a repeatable
output signal was observed (as shown in the inset of Figure 6).
As the concentrations of GOx solution increased, oxidation current
gradually increased from approximately 20 nA (10 U/mL) to reach
a maximum of 100 and 120 nA (150 U/mL) in channels with no
grooves and slanted grooves, respectively. At higher GOx concentrations,
the current leveled off and then decreased in both
channels with no grooves and slanted grooves. These results
indicated that increasing concentrations of enzyme provided better
conversion to H2O2 and product within a reaction time of
approximately 4 s. However, at concentrations of GOx greater
than 150 U/mL, the system exhibited less stable signal, as
reported elsewhere; 8 this could be due to the interaction of
GOx and H2O2 produced. Therefore, a 150 U/mL concentration
of GOx solution was chosen as an optimum condition for
further experiments. Compared with slanted grooves in the
channel, the absence of grooves in the channel resulted in
lower current values. It can be concluded that the chaotic flow
pattern results in more efficient mixing caused by transverse
flow in the channel.
3.4. Comparison of System Response for Premixed
Glucose-GOx Solutions and Reaction of Glucose and GOx
in the Channel. In order to investigate dependence on flow rate,
the response measured for premixed glucose-GOx solutions was
compared with that measured for the reaction of glucose and GOx
in the mixing/reaction channel as a function of flow velocity. For
the latter experiments, a 5.6 mM glucose sample and a 150 U/mL
GOx solution were introduced at a 1:3 mixing ratio in a slanted
grooves channel and the flow velocity was varied from 0.95 mm/s
(0.4 µL/min) to 14.3 mm/s (6 µL/min). The amperometric
responses for premixed solutions and reaction in the mixing
channel are presented in Figure 7. There are two dominant factors
in this continuous flow-through reaction system which affect
detector response, namely flow velocity and reaction time. The
effect of flow velocity was measured by introducing premixed
solution (5.6 mM glucose sample and 150 U/mL GOx solution
had already reacted to produce H 2O2) in both inlets and varying
the flow rate. The results showed that as the flow velocity
increased, the recorded current values also increased gradually.
It was also observed that current values increased as a function
(35) Dixon, B. M.; Lowry, J. P.; O’Neill, R. D. J. Neurosci. Methods 2002, 119,
135–142.
6762 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 7. Amperometric response in a channel with slanted grooves,
arising from a 1:3 reaction ratio of 5.6 mM glucose and 150 U/mL
GOx as a function of flow velocity (from 0.4 to 6 µL/min). The response
due to premixed solution (5.6 mM glucose and 150 U/mL GOx) was
compared with the glucose-GOx reaction in the mixing channel (a
single device was used for both curves, n ) 3 for each point).
of flow velocity when experimenting with 0.5 mM H2O2
solutions (data not shown). This is very different from the
decreasing signal observed for GOx immobilized on electrodes
at higher flow rates, which Lowry et al. ascribed to the removal
of H2O2 from the electrode surface. 36 On the other hand, Wu
et al. 37 observed currents which increased as a function of flow
rate at two closely spaced oxygen sensors in a microchannel.
This latter report supports our hypothesis that the observed
increase in current values for premixed solutions is due to the
thinner diffusion layers at the electrode surface which result
at higher flow rates. These layers are stagnant, and molecules
must diffuse through them to reach the electrode surface. As
diffusion layers decrease in thickness, the time required for
diffusion to the electrode surface also decreases, resulting in
the delivery of analyte to that surface to be less limited by
diffusion. Higher rates of electron exchange and thus higher
recorded currents are the result.
In the case of glucose reaction with GOx in the mixing channel,
a decreasing signal as a function of flow velocity was observed.
This decrease in current is primarily due to lower H2O2 concentrations,
the result of shorter reaction times. The maximum
reaction of glucose and GOx was observed at very low flow
rates (lower than 0.4 µL/min). Thus, based on reaction times
of 22 and 4 s (0.4 and 2 µL/min, respectively) from the
Y-junction to the electrodes, the most efficient conversion of
glucose with GOx can be achieved either at very low flow rates
and/or in longer reaction channels.
3.5. Calibration Curve. A calibration curve was obtained
when a 150 U/mL GOx solution was reacted with glucose sample
in a channel at a total flow rate of 2 µL/min, with a 1:3 flow rate
(and thus 1:4 dilution) ratio of glucose and GOx. Figure 8 shows
the amperometric response as a function of glucose concentration
(36) Lowry, J. P.; McAteer, K.; El Atrash, S. S.; Duff, A.; O’Neill, R. D. Anal.
Chem. 1994, 66, 1754–1761.
(37) Wu, J.; Ye, J. Lab Chip 2005, 5, 1344–1347.

in channels with either no grooves or slanted grooves. The data
exhibit a linear relationship between current and glucose concentrations
of 0-20 mM, with a sensitivity of 5.3 and 6.9 nA/mM
in channels with no grooves and slanted grooves, respectively.
The current density was calculated by dividing sensitivity by the
active area of the working electrode, yielding 26.5 and 34.5 mA/
M · cm2 for channels with no grooves and slanted grooves,
respectively. The linearity of the curves was tested by plotting
observed current values versus predicted current values and
examining the distribution of points around the resulting
diagonal line (data not shown). In both cases, points were very
symmetrically distributed around the line. The slope of the plot
for the no-groove case was 0.9992, with an R2 of 0.9983. In the
groove case, the slope was 1.0032, with an R2 Figure 8. Calibration curves obtained using 150 U/mL GOx solution
at a total flow rate of 2 µL/min (0.5 µL/min glucose: 1.5 µL/min GOx)
in two channel types, without grooves and with slanted grooves (one
device, n ) 3). Glucose concentrations were in the clinical range of
interest.
of 0.9989. In both
cases, excellent linearity was thus observed.
In our approach, the flow velocity and time for reaction of
glucose sample and GOx enzyme strongly affect the sensitivity
(38) Ghosal, S. Anal. Chem. 2002, 74, 771–775.
(39) Mei, Q.; Xia, Z.; Xu, F.; Soper, S. A.; Fan, Z. H. Anal. Chem. 2008, 80,
6045–6050.
of the sensor signal, allowing flow conditions to be tuned to obtain
optimal results. One of the big advantages of this microfluidic
reactor approach is that by varying flow rates of glucose and GOx,
it is possible not only to avoid the oxygen depletion effect without
any electrode treatment but to tune the sensitivity for the
application. However, continuous perfusion of glucose and GOx
to the device does lead to a slightly decreased signal over time.
This might be caused by GOx adsorbing either onto the microchannel
or electrode surface as described elsewhere. 38 This effect
will be further investigated as part of the development of systems
for long-term measurement of glucose in in vivo applications.
4. CONCLUSION
We have successfully demonstrated an enzymatic glucose
reactor based on chaotic mixing in a microfluidic channel network
for continuous glucose monitoring. Together with another recent
report describing improved lucerifase detection in a chaotic
mixer, 39 our example of glucose detection is one of the first
examples of this type of mixer being applied to the enhancement
of a biochemical reaction at the nL scale. A linear calibration curve
was obtained in both microchannels with slanted grooves and no
grooves, using a 150 U/mL GOx enzyme solution. Higher
sensitivity was obtained with slanted groove arrays compared to
micromixers with no grooves, due to enhanced mixing. The
factors which determine sensitivity are flow velocity and extent
of reaction. Though there is a loss of GOx due to the continuous
flow to the outlet, this disadvantage is alleviated by the use of
micofluidics for nL liquid handling and the application of low flow
rates (2 µL/min for the optimized system). The low flow rates
used are also compatible with microdialysis sampling and are
required to achieve a high recovery of glucose from the subcutaneous
tissue and reduce solution consumption. The possible
influence of interfering substances such as ascorbic acid, uric acid,
and acetaminophen on glucose determination are now under
investigation, prior to testing the system in vivo in rats.
Received for review January 8, 2010. Accepted June 24,
2010.
AC1000509
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6763

Anal. Chem. 2010, 82, 6764–6769
Development of a Compound-Specific Carbon
Isotope Analysis Method for 2-Methyltetrols,
Biomarkers for Secondary Organic Aerosols from
Atmospheric Isoprene
Qiang Li, § Wu Wang,* ,†,‡ Hong-Wei Zhang, ‡ Yang-Jun Wang,* ,† Bing Wang, | Li Li, † Huai-Jian Li, †
Bang-Jin Wang, † Jie Zhan, † Mei Wu, ⊥ and Xin-Hui Bi @
Institute of Environmental Pollution and Health, Shanghai University, 200444 Shanghai, China, Southern Medical
University, 510515 Guangzhou, China, School of Stomatology, Tongji University, 200072 Shanghai, China, Hefei
Artillery Academy, 230002 Hefei, China, Laboratory of Human Micromorphology, Qingdao University Medical College,
266071 Qingdao, China, and State Key Laboratory of Organic Geochemistry, Guangzhou Institute of Geochemistry,
Chinese Academy of Sciences, 510640 Guangzhou, China
The stable carbon isotope compositions of 2-methyltetrols,
biomarker compounds for secondary organic
aerosols formed from isoprene in the atmosphere, have
been determined by gas chromatography/combustion/
isotope ratio mass spectrometry (GC/C/IRMS). In this
work, isoprene with various δ 13 C values was used to
produce 2-methyltetrols via an oxidation reaction
with hydrogen peroxide in sulfuric acid under direct
sunlight. The target compounds with different stable
carbon isotope compositions were then derivatized
by methylboronic acid with a known δ 13 C value and
measured by GC/C/IRMS. With δ 13 C values of 2-methyltetrols
and methylboronic acid predetermined,
isotopic fractionation is evaluated for the derivatization
process. Through reduplicate δ 13 C measurements,
the carbon isotope analysis achieved excellent
reproducibility and high accuracy with an average
error of

indicators of atmospheric processing of volatile organic compounds;
9 for example, carbon isotope ratio measurements allow
the calculation of the extent of photochemical processing of
isoprene in the atmosphere. 10 Furthermore, Rudolph et al. 11
developed a gas chromatography/combustion/isotope ratio mass
spectrometry (GC/C/IRMS) technique to measure the stable
carbon isotope ratios of isoprene and its gas-phase oxidation
products, MACR and MVK, and to gain insight into the atmospheric
oxidation of isoprene. Very recently, this group has
reported studies on the stable carbon kinetic isotope effects (KIE)
of the reactions of isoprene, MACR, and MVK with OH radicals,
as well as with ozone in the gas phase. 12-14 These data are
valuable for obtaining insight into the role of loss processes in
determining the atmospheric mixing ratios. However, there are
no isotopic studies of isoprene SOA products in the aerosol phase.
Methylboronic acid (MBA) has been reported to determine
natural 13 C abundances of monosaccharides by derivatizing
adjacent hydroxyl groups of monosaccharides followed by N,Obis(trimethylsilyl)trifluoroacetamide
(BSTFA) derivatization of
the remaining single OH groups. 15,16 Recently, Boschker et al. 17
reported a versatile method for stable carbon isotope analysis of
carbohydrates by high-performance liquid chromatography/
isotope ratio mass spectrometry. To develop a compound-specific
isotope analysis method for 2-methyltetrols, marker compounds
of photooxidation products of isoprene, a technique from our
previous work was adapted. 18 MBA was used as the derivatization
reagent prior to gas chromatography/combustion/isotope ratio
mass spectrometry (GC/C/IRMS). The derivatizing C from the
reagent accounts for 29% of the analyte in the boronates, but 70%
in the trimethylsilyl (TMS) derivatives. Therefore, the sensitivity
of the MBA method should be high. The oxidation reaction of
isoprene, atmospheric sampling, accuracy, and reproducibility of
the method will be discussed in detail, and the stable carbon
isotope effects during the procedure will be evaluated. δ 13 C data
for atmospheric 2-methyltetrols will also be presented to
demonstrate the practical utility of this method.
EXPERIMENTAL SECTION
Materials. Isoprene was obtained from three suppliers: Fluka,
Sigma-Aldrich (>98% pure, M1); Alfa-Aesar (Lancaster, England)
(99% pure, M2); and Toyo Kasei Kogyo Co. (Osaka, Japan) (99%
pure, M3). Hydrogen peroxide (30% in water) was purchased from
Guoyao (Shanghai, China). Methylboronic acid (MBA) was
(9) Rudolph, J.; Czuba, E. Geophys. Res. Lett. 2000, 27, 3865–3868.
(10) Rudolph, J.; Anderson, R. S.; Czapiewski, K. V.; Czuba, E.; Ernst, D.;
Gillespie, T.; Huang, L.; Rigby, C.; Thompson, A. E. J. Atmos. Chem. 2003,
44, 39–55.
(11) Iannone, R.; Koppmann, R.; Rudolph, J. J. Atmos. Chem. 2007, 58, 181–
202.
(12) Rudolph, J.; Czuba, E.; Huang, L. J. Geophys. Res. 2000, 105, 29323–29346.
(13) Iannone, R.; Koppmann, R.; Rudolph, J. Atmos. Environ. 2008, 38, 4093–
4098.
(14) Iannone, R.; Koppmann, R.; Rudolph, J. Atmos. Environ. 2009, 43, 3103–
3110.
(15) van Dongen, B.; Schouten, S.; Damste, J. Rapid Commun. Mass Spectrom.
2001, 15, 496–500.
(16) Gross, S.; Glaser, B. Rapid Commun. Mass Spectrom. 2004, 18, 2753–2764.
(17) Boschker, H. T. S.; Moerdijk-Poortvliet, T. C. W.; van Breugel, P.;
Houtekamer, M.; Middelburg, J. J. Rapid Commun. Mass Spectrom. 2008,
22, 3902–3908.
(18) Wang, W.; Li, L.; Li, H.; Zhang, D.; Wen, S.; Jia, W.; Wang, B.; Sheng, G.;
Fu, J. Rapid Commun. Mass Spectrom. 2009, 23, 2675–2678.
purchased from ABCR GmbH and Co. KG (Karlsruhe, Germany)
(97% pure) and recrystallized three times from a benzene/acetone
mixture (3:1). MBA of the same lot number was used for all
derivatizations. Anhydrous pyridine (99% pure) was supplied by
Acros Organics (Geel, Belgium). BSTFA [N,O-bis(trimethylsilyl)
trifluoroacetamide] was purchased from Pierce (Rockford, IL). All
solvents employed were HPLC grade.
Preparation of Standard 2-Methyltetrols. 2-Methyltetrols
were made by photooxidation of isoprene, which we conducted
by exposing a 30 mL flask with a mixture of 5 mL of 30% H2O2
and 5 mL of isoprene (0.05 mol, 3.4 g) to sunlight. A few drops
of sulfuric acid (0.1 M) was added until the pH of reaction
mixture was between 1 and 2. The resulting mixture was then
maintained with continuous and vigorous stirring in sunlight
for4h; 19 15 mg of barium carbonate was added to 1 mL of the
reacted solution for neutralization. After centrifugation, the
supernatant was dried, and a slightly yellow oil (2.5 g, 36% yield)
was obtained. It worth noting that exposure to sunlight is
crucial for the production of 2-methyltetrols.
The purification of crude 2-methyltetrols was performed
according to a procedure reported by Wang et al. 20 The purity of
2-methyltetrols was verified by gas chromatography/mass spectrometry
(GC/MS) after they had been derivatized with BSTFA,
and the δ 13 C value was determined by elemental analyzer/
isotope ratio mass spectrometry (EA/IRMS).
Derivatization of 2-Methyltetrols. The derivatization technique
of 2-methyltetrols with methylboronic acid was adapted from
Wang et al.; 18 100 µL of a solution of 2-methyltetrols (approximately
1 mg/mL in methanol) was dried under a gentle
nitrogen flow, and then 5 mL of a solution of 1 mg of methylboronic
acid in 10 mL of anhydrous pyridine was added. The molar
ratio of MBA to 2-methyltetrols was ∼10:1. The mixture was
allowed to react at 60 °C for 60 min. It should be noted that the
pretreatment of pyridine with 4 Å molecular sieves in excess is
crucial for a successful MBA derivatization. The δ 13 C value of
methylboronic derivatives was determined by gas chromatography/combustion/isotopic
ratio mass spectrometry (GC/C/
IRMS).
Measurements of the δ 13 C Value of Standard Isoprene.
The method for determining the δ 13 C value of isoprene was as
follows. 21 Isoprene (1 mL) was sealed in a2mLglass vial with
an open screw cap containing a Teflon-lined silica septum. After
∼1 h for equilibrium, 15 µL gas samples from the glass bottle
were injected into the split/splitless injection port of the gas
chromatograph using a Hamilton gastight locking syringe.
Aerosol Sampling. Samples were collected in boreal-temperate
Changbai Mountain Forest Ecosystem Research Station in Jilin
Province and subtropical Dinghu Mountain Nature Reserve in
Guangdong Province. The details for sampling sites have been
described previously. 6 The sampling occurred during the summer
when the meteorological conditions and the maximum solar
radiation, as well as high temperatures, were favorable for the
photooxidation of isoprene. A high-volume PM2.5 air sampler
(19) Santos, L.; Dalmazio, L.; Eberlin, M.; Claeys, M.; Augusti, R. Rapid Commun.
Mass Spectrom. 2006, 20, 2104–2108.
(20) Wang, W.; Vas, G.; Dommisse, R.; Loones, K.; Claeys, M. Rapid Commun.
Mass Spectrom. 2004, 18, 1787–1797.
(21) Yu, Y.; Wen, S.; Feng, Y.; Bi, X.; Wang, X.; Peng, P.; Sheng, G.; Fu, J. Anal.
Chem. 2006, 78, 1206–1211.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6765

(Graseby-Andersen) was operated at a flow rate of 1.13 m3 /
min, and Whatman quartz fiber filters (20.3 cm × 25.4 cm) were
used; 24 h samplers were collected. All filters were baked at
550 °Cfor4htoremove organic contaminants. After collection,
the filters were stored at -20 °C until they were analyzed. Part
of the filters was extracted with methanol. The solvent was
then concentrated, filtered, and finally dried completely. All
samples were derivatized using the same procedure as described
above and measured by GC/C/IRMS.
Analytical Systems. The gas chromatography/mass spectrometry
(GC/MS) instrument consisted of a Hewlett-Packard
(Fullerton, CA) model 6890 gas chromatograph equipped with a
DP-5MS device (30 m × 0.25 mm inside diameter, 0.25 µm film
thickness), coupled to a Hewlett-Packard model 5975MSD quadrupole
analyzer. Data were acquired and processed with Chem-
Station (Hewlett-Packard). The temperature program was as
follows: initial temperature at 100 °C held for 5 min, a gradient of
3 °C/min up to 200 °C, a gradient of 30 °C/min up to 290 °C,
held for 2 min. The mass spectrometer was operated in the
electron ionization mode at 70 eV and an ion source temperature
of 150 °C. Full scan mode was used in the mass range of m/z
50-420.
Two kinds of gas chromatography/combustion/isotopic ratio
mass spectrometry (GC/C/IRMS) systems were used in this
study. The analysis of MBA derivatives was performed on an HP
6890 GC system (Agilent, Santa Clara, CA) equipped with a DP-
5MS device (30 m × 0.25 mm × 0.25 µm), connected to an isotope
ratio mass spectrometer (Isoprime, GV instrument, Manchester,
U.K.). The oven temperature program was as follows: initially at
60 °C, a gradient of 4 °C/min up to 110 °C, held for 2 min, a
gradient of 50 °C/min up to 290 °C, held for 2 min. The injector
was set at 250 °C in splitless mode. Helium was used as the carrier
gas at 1.5 mL/min. CO2 with a known δ13C value (-26.65‰)
was used as the external reference gas. The combustion
furnace containing the CuO catalyst and the reduction oven
containing the Cu catalyst were kept at 940 and 650 °C,
respectively. The temperature of the interface between the GC
and combustion furnace was set at 290 °C. The reproducibility
and accuracy of carbon isotopic analyses were evaluated
routinely every day using 10 laboratory isotopic standards (C12,
C14, C16, C18, C20, C22, C25, C28, C30, and C32 n-alkanes supplied
by Indiana University, Bloomington, IN) with known isotopic
values (-31.89, -30.67, -30.53, -31.02, -32.24, -32.77,
-28.49, -32.11, -33.05, and -29.41‰, respectively). 2-Methyltetrol
methylboronates prepared in the laboratory were
analyzed 10 times and used as laboratory standards. For δ13C analysis of isoprene, the same GC/C/IRMS system described
above and a CP-PoraPLOT Q column (25 m × 0.32 mm ×
10 µm, Varian) were used. The GC was run in a split ratio of
∼40:1, and the injector was set at 150 °C. The initial oven
temperature was held at 120 °C for 1 min, followed by a
gradient of 20 °C/min up to 195 °C, at which point the
temperature was held for 10 min. Standard CO2 (δ13C )
-26.65‰) was used as the external reference gas. CH4 with a
known δ13C value (-36.30‰) was used as the laboratory
isotopic standard to check the reproducibility and accuracy of
6766 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
the carbon isotopic analysis routinely. 22 For both analysis
systems, the sextuple analysis of laboratory isotopic standards
indicated excellent accuracy and reproducibility of carbon
isotopic analysis [the corresponding standard deviation ranged
from 0.13 to 0.30‰, and the deviation between the measured
data and the predetermined data ranged from -0.02 to 0.13‰
(Table 1S of the Supporting Information)].
Elemental analyzer/isotope ratio mass spectrometry (EA/
IRMS) was performed as follows. Standard 2-methyltetrols and
recrystallized methylboronic acid were put into cleaned tin
capsules and weighed, repectively. Capsules containing weighed
samples were placed in the CE (Wigan, U.K.) EA1112 C/N/S
analyzer and burned at 960 °C in an O2 atmosphere in a
combustion tube. Combustion gases were swept through a
reduction oven and entered a GC column where CO2 was
separated from other gases. Then the CO2 passed through a
Conflo III interface (Finnigan, Waltham, MA) and entered a
DELTA plus XL mass spectrometer (Thermo Finnigan MAT,
Bremen, Germany) where it was compared to the reference
CO2 with a known δ 13 C value (-29.10‰, calibrated against the
NBS-22 reference material with a δ 13 C value of -29.70‰).
During every batch of analyses, an empty tin capsule was
analyzed as a blank to check the background, and the carbon
black sample with a known δ 13 C value (-36.91‰) was used to
evaluate the reproducibility and accuracy. The standard deviation
of analysis and the deviation between the measured data
and the predetermined data were less than 0.3‰ (Table 2S of
the Supporting Information).
All 13 C: 12 C ratios are expressed in conventional delta (δ)
notation, which is the per mil (‰) deviation from the standard
Vienna Pee Dee Belemnite (VPDB) reference point.
RESULTS AND DISCUSSION
δ 13 C Analysis of Standard 2-Methyltetrols. Figure 1
shows a typical total ion current (TIC) GC/MS chromatogram
from isoprene oxidation products derivatized by BSTFA. Identification
of these two major compounds was made by comparison
of the retention time and mass spectra with those published in
the literature. 1,6,20 Figure 2 presents the TIC GC/MS chromatogram
of 2-methyltetrol-MBA derivatives. The mass spectra of
these three compounds are very similar, as could be expected
for diastereoisomers. The mass spectrum of compound 2 reveals
a tiny molecular ion (m/z 184). Characteristic ions at m/z 99 and
85 correspond to the cleavage of the C-C bond at the C2 position.
Other ions at m/z 69, 57, and 43 can be explained by the further
fragment of the ion at m/z 85 or 99.
The δ 13 C values of methylboronic acid and standard 2-methyltetrols
were determined by EA/IRMS to be -24.96 ± 0.06‰
(eight measurements) and -32.40 ± 0.14‰ (n ) 6, from M1),
-29.84 ± 0.12‰ (n ) 6, from M2), and -32.36 ± 0.18‰ (n )
6, from M3) (Table 1).
δ 13 C Analysis of MBA Derivatives. In derivatization, the
preparation of 2-methyltetrol-MBA derivatives alters the original
stable isotope compositions of the 2-methyltetrols. To correct for
the introduction of carbon during derivatization, it is necessary
to assess the isotopic reproducibility of the derivatization method.
(22) Wen, S.; Feng, Y.; Yu, Y.; Bi, X.; Wang, X.; Sheng, G.; Fu, J. Environ. Sci.
Technol. 2005, 39, 6202–6027.

Figure 1. GC/MS total ion chromatograms obtained for the reaction mixtures of isoprene with H2O2 and sulfuric acid derivatized by BSTFA and
spectra for products (insets) 1 (2-methylthreitol) and 2 (2-methylerythritol).
Figure 2. GC/MS total ion chromatogram obtained for 2-methyltetrols derivatized by MBA and mass spectra for the products (insets). Compounds
1 and 2 correspond to 2-methylerythritol-MBA derivatives, and compound 3 corresponds to a 2-methylthreitol-MBA derivative.
Table 1. Stable Carbon Isotopic Compositions of Isoprene, 2-Methyltetrols (2-MT), and MBA Derivatives Measured
and Predicted in the Derivatization Reaction
supplier measured isoprene b,c
measured 2-MT b,d
The reproducibility of the carbon isotope composition for three
2-methyltetrols (with different δ 13 C values) was evaluated. Their
δ 13 C values and those of the corresponding MBA derivatives
are listed in Table 1. The analytical errors (standard deviation)
obtained for six EA/IRMS analyses of 2-methyltetrols produced
from isoprene from the same supplier ranged from 0.12 to 0.18‰,
with an average of 0.15 ± 0.03‰, while for GC/C/IRMS analyses
of MBA derivatives, the analytical errors were from 0.15 to 0.22‰,
with an average of 0.17 ± 0.04‰. The reproducibility compares
well with those obtained in the derivatization of fatty acids. 23
Isotopic Effects of the Method. According to eq 1, the
theoretical δ 13 C values of methylboronates can be calculated
(23) Abrajano, T. A.; Murphy, D. E., Jr.; Fang, J.; Comet, P. A.; Brooks, J. M.
Org. Geochem. 1994, 21, 611–617.
δ 13 C a
measured methylboronates b,c,e
calculated methylboronates b,f
M1 -24.54 ± 0.18 (n ) 7) -32.40 ± 0.14 (n ) 6) -29.98 ± 0.22 (n ) 6) -30.27 0.29
M2 -23.72 ± 0.23 (n ) 7) -29.84 ± 0.12 (n ) 6) -28.08 ± 0.15 (n ) 6) -28.24 0.16
M3 -25.19 ± 0.19 (n ) 7) -32.36 ± 0.18 (n ) 6) -30.35 ± 0.16 (n ) 6) -30.25 -0.10
a Stable carbon isotopic compositions reported in per mil relative to PDB. b Arithmetic means and standard deviations. c δ 13 C values determined
by GC/C/IRMS. d δ 13 C values determined by EA/IRMS. e Derivatized by methylboronic acid with a δ 13 C value of -24.96 ± 0.06‰ (eight
measurements) determined by EA/IRMS. f On the basis of mass balance relationship eq 1. g Calculated δ 13 C - measured δ 13 C.
according to stoichiometric mass balance and reflect the
relative contributions of carbon from 2-methyltetrols, methylboronic
acid, and their respective δ 13 C values:
δ 13 C methylboronate ) f MBA δ 13 C MBA + f 2-methyltetrol δ 13 C 2-methyltetrols
(1)
where fMBA and f2-methyltetrol are the molar fractions of carbon
in methylboronate arising from underivatized 2-methyltetrol
and MBA reagent, respectively. Here, fMBA ) 2 /7, and
f2-methyltetrols ) 5 /7. The stable carbon isotopic compositions
obtained for 2-methyltetrols and derivatives are listed in Table
1. The measured data for methylboronates were compared with
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
∆ g
6767

Figure 3. GC/C/IRMS chromatogram of 2-methyltetrol-MBA derivatives: (a) standard derivative and (b) extract of the fine size fraction of a
24 h Hi-Vol aerosol sample collected at Dinghu, China, on August 2, 2006. Compounds 1 and 2 are 2-methylerythritol-MBA derivatives, and
compound 3 is the 2-methylthreitol-MBA derivative.
those predicted by eq 1. The predicted and measured δ 13 C values
of methylboronates agreed well with each other.
According to Rieley’s discussion of the kinetic isotope effect, 24
when a bond containing the carbon atom under consideration is
changed in the rate-determining step, the primary isotopic effect
is the most significant. If no carbon bond changed in the ratedetermining
reaction or if no carbon-containing bond is involved
in this step, there should not be a primary isotope effect on the
δ 13 C value. In this work, four hydroxyl groups from two
molecules of MBA react with four hydroxyl groups of 2-methyltetrol
and eliminate four molecules of water. No other carboncontaining
bonds, except C-O bonds of 2-methyltetrols, are
involved in the reaction. The difference between measured and
calculated δ 13 C values of methylboronic derivatives ranged from
-0.10 to 0.29‰ (Table 1). Accuracy was well within the isotope
technical specification (±0.5‰). It should be pointed out that the
analytical error of the calculated data for underivatized 2-methyltetrol
(usually expressed as the standard deviation, S) could be
calculated by the following equation:
2
S2-methyltetrol ) (1/f2-methyltetrol ) 2 2
Smethylboronate +
(fMBA /f2-methyltetrol ) 2 2
SMBA where fMBA and f2-methyltetrol are the same as in eq 1 and S2-methyltetrol,
SMBA, and Smethylboronate are the analytical standard deviations of
(24) Rieley, G. Analyst 1994, 119, 915–919.
6768 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(2)
2-methyltetrols, MBA, and methylboronates, respectively. In
this work, SMBA was 0.06‰ and Smethylboronate was 0.17 ± 0.04‰
(0.15-0.22‰). According to eq 2, the standard deviation of the
measured δ 13 C value of 2-methyltetrol ranged from 0.21 to
0.31‰. The results imply that the method is promising and
introduces no isotopic fractionation in the derivatization process,
as confirmed by Table 1. The stable carbon isotopic
determination exhibited high precision and good reproducibility.
On the other hand, precursor isoprene was approximately
6.12-7.86‰ more enriched in 13 C than 2-methyltetrols (Table
1). This might reflect isotope fractionation during the photochemical
oxidation reaction. Rudolph et al. 12 reported that the stable
carbon isotope fractionation for the reaction of isoprene with OH
radicals in the gas phase was 6.94 ± 0.80‰, very similar to the
value in this work.
To fully evaluate the usefulness of relating precursor isoprene
δ 13 C values to those measured in aerosol, one must investigate
the potential for the primary kinetic isotope effect (KIE)
involving the carbon-carbon bond cleavage products (such as
methylglyceric acid, glyoxal, and methylglyoxal) to influence
the δ 13 C value of 2-methyltetrols. This is especially true since
the reaction pathways for formation of 2-methyltetrols remain
unclear. Gas-phase and mixed-phase mechanisms involving the
aqueous aerosol phase have been considered. 1-3
Measurements of Atmospheric Samples. The target compounds
in samples were well-separated (Figure 3), while the
2-methylerythritol derivative gave rise to double peaks, due to the

Table 2. Concentrations and Stable Carbon Isotopic Compositions of 2-Methyltetrols at Two Forested Sites
sampling site sampling date concentration of 2-methyltetrols (ng/m 3 ) measured methylboronates b
formation of both E and Z isomers during the reaction. From our
previous work, 18 the erythro form of tetrols, erythritol, eluted
earlier than the threo form, threitol. Under the same conditions,
peaks 1 and 2 were identified as (Z)- and (E)-2-methylerythritol-MBA
derivatives, respectively, while peak 3 was the
2-methylthreitol-MBA derivative. The latter derivative gives rise
to only one tiny peak, due to the stereo hindrance of the 2-methyl
group. The δ 13 C value of the 2-methyltetrol-MBA derivatives
was obtained by defining these three peaks as one using the
integral tools of IRMS.
It could be seen from Table 2 that δ 13 C values of 2-methyltetrols
were distinctly different for the two sites. It has been
documented that there were significant differences in carbon
isotope ratios among C3, C4, and CAM plants, e.g., -35 to
-23‰ for C3 with an average of -26‰, -14 to -10‰ for C4
with an average of -13‰, and intermediate carbon isotope
compositions for CAM. 25 The δ 13 C of 2-methyltetrols might
reveal the relation of atmospheric aerosol with the vegetation
types and be used as an indicator of proportions of C3 to C4
plants in vegetation.
CONCLUSIONS
The stable carbon isotope compositions of 2-methyltetrols,
biomarker compounds of isoprene in atmospheric aerosols, were
determined by GC/C/IRMS. MBA was used as the derivatization
reagent which allows only 29% contribution of the analyzed C.
The δ 13 C values of 2-methyltetrols are calculated on the basis
of the stoichiometric mass balance equation among 2-methyltetrols,
methylboronic acid, and methylboronate derivatives.
Excellent reproducibility and accuracy are achieved without
carbon isotopic fractionation. Moreover, photosynthesis of
2-methyltetrols through photochemical oxidation of isoprene
(25) Cerling, T. E.; Wang, Y.; Quade, J. Nature 1993, 361, 344–345.
is introduced. The δ 13 C values of atmospheric 2-methyltetrols
were determined for two forest aerosols. Considering the
complex formation pathways of 2-methyltetrols, this technique
will be helpful for full field application and will have potential
in differentiation between homogeneous and heterogeneous
oxidation processes. The application of this method may
provide additional information about the sources and sinks of
atmospheric isoprene. Further experiments on photochemical
oxidation in a smog chamber are warranted.
ACKNOWLEDGMENT
This work was funded by the Knowledge Innovation Program
of the Chinese Academy of Sciences (Grant kzcx2-yw-139), the
Natural Science Foundation of China (Grants 20677036, 20877051,
and 40873073), the Innovation Program of Shanghai Municipal
Education Commission (10YZ08), the Shanghai Municipal Health
Bureau (054088), Shanghai Leading Academic Disciplines (S30109),
and the Scientific Research Foundation for Returned Overseas
Chinese Scholars, State Education Ministry, China. We thank the
anonymous referee for insightful comments. We also thank Dr.
Jia Wanglu (State Key Laboratory of Organic Geochemistry,
Guangzhou Institute of Geochemistry, Chinese Academy of
Sciences) for his technical assistance.
SUPPORTING INFORMATION AVAILABLE
δ 13 C values (‰) of the Indiana reference n-alkanes measured
by GC/C/IRMS (Isoprime) (Table 1S) and δ 13 C values of the
reference materials of CH4 and carbon black (Table 2S). This
material is available free of charge via the Internet at
http://pubs.acs.org.
Received for review January 25, 2010. Accepted April 19,
2010.
AC100214P
δ 13 C a
calculated 2-MT c
Changbai July 27, 2007 76.29 -24.27 ± 0.13 (n ) 3) -24.00 ± 0.18
July 28, 2007 108.8 -24.79 ± 0.61 (n ) 3) -24.73 ± 0.86
Dinghu August 1, 2006 65.30 -27.29 ± 0.10 (n ) 3) -28.22 ± 0.14
August 2, 2006 83.54 -26.93 ± 0.46 (n ) 3) -26.99 ± 0.64
a Stable carbon isotopic compositions reported in per mil relative to PDB. b Arithmetic means and standard deviations. Derivatized by methylboronic
acid with a δ 13 C value of -24.96 ± 0.06‰ (eight measurements) determined by EA/IRMS. δ 13 C values determined by GC/C/IRMS. c Arithmetic
means of calculated δ 13 C values based on mass balance relationship eq 1 and standard deviations (S) calculated according to eq 2.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6769

Anal. Chem. 2010, 82, 6770–6774
Surface-Enhanced Raman Spectroscopy as a Tool
for Detecting Ca 2+ Mobilizing Second Messengers
in Cell Extracts
Elina A. Vitol, † Eugen Brailoiu, ‡ Zulfiya Orynbayeva, § Nae J. Dun, ‡ Gary Friedman, † and
Yury Gogotsi* ,|
Department of Electrical and Computer Engineering, and Department of Materials Science and Engineering, Drexel
University, Philadelphia, Pennsylvania 19104, and Department of Pharmacology, Temple University, Philadelphia,
Pennsylvania 19140
Understanding of calcium signaling pathways in cells is
essential for elucidating the mechanisms of both normal
cell function and cancer development. Calcium messengers
play the crucial role for intracellular Ca 2+ release.
We propose a new approach to detecting the calcium
second messenger nicotinic acid adenine dinucleotide
phosphate (NAADP) in cell extracts using surfaceenhanced
Raman spectroscopy (SERS). Currently available
radioreceptor binding and enzymatic assays require
extensive sample preparation and take more than
12 h. With a SERS sensor, NAADP can be detected in
less than 1 min without any special sample preparation.
To the best of our knowledge, this is the first
demonstration of using SERS for calcium signaling
applications.
Calcium signaling is one of the fundamental cellular processes
involved in any cell metabolic and physiologic activity. 1 Calcium
signals convey information from the cell plasma membrane to
intracellular targets. The mechanism of calcium concentration
modulations is a complex problem associated with calcium influx
from the extracellular matrix and release from intracellular stores
mobilized by calcium messengers. Calcium signaling pathways
of two calcium messengers, inositol trisphosphate (IP3) 2,3 and
cyclic adenine dinucleotide ribose (cADPR), 4 have been studied
extensively in different types of cells. Nicotinic acid adenine
dinucleotide phosphate (NAADP) has a unique physiological role
in cells 5 in the release of Ca 2+ from acid-filled calcium stores
* To whom correspondence should be addressed. E-mail: gogotsi@drexel.edu.
† Department of Electrical and Computer Engineering, Drexel University.
‡ Temple University.
§ School of Biomedical Engineering, Science and Health System, Drexel
University.
| Department of Materials Science and Engineering, Drexel University.
(1) Patel, S.; Churchill, G. C.; Galione, A. Biochem. J. 2000, 352, 725–729.
(2) Taylor, C. W.; Thorn, P. Curr. Biol. 2001, 11, R352–R355.
(3) Cancela, J. M.; Gerasimenko, O. V.; Gerasimenko, J. V.; Tepikin, A. V.;
Petersen, O. H. EMBO J. 2000, 19, 2549–2557.
(4) Guse, A. H.; da Silva, C. P.; Berg, I.; Skapenko, A. L.; Weber, K.; Heyer, P.;
Hohenegger, M.; Ashamu, G. A.; Schulze-Koops, H.; Potter, B. V. L.; Mayr,
G. W. Nature 1999, 398, 70–73.
(5) Lee, H. C.; Aarhus, R. J. Biol. Chem. 1995, 270, 2152–2157.
6770 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
through two-pore channels 1, 2, and 3. 6,7 NAADP is the least
investigated Ca 2+ mobilizing second messenger, because of the
lack of widely accessible and efficient techniques for detecting
and quantifying its concentration in cells. Enzymatic bioassays
and radioreceptor binding assays are the primary methods
which have been used for detecting NAADP in cell extracts. 8,9
The enzymatic assay 9 requires NAADP to be first converted to
nicotinamide adenine dinucleotide phosphate (NADP) using ADPribosyl
cyclase, which is followed by two enzymatic cycling
reactions of oxidation/reoxidation of NADP. 10 Diaphorase, the
enzyme for reoxidation of NADP to nicotinamide adenine dinucleotide
phosphate (NADPH), also serves as a catalyst for conversion
of the reaction indicator resazurin to a highly fluorescent resorufin.
The latter is then used for fluorimetric assessment of the NAADP
concentration. Importantly, the described assay requires a very
high purity of all of the components and takes more than 12 h. 10
The radioreceptor binding assay is less time-consuming and can
be conducted without extensive sample purification, 8 but due to
the need for unique specialized equipment, the availability of this
method is extremely limited.
Here we present an alternative, label-free technique for the
detection of NAADP enabled by surface-enhanced Raman spectroscopy
(SERS). 11-13 SERS enhances Raman scattering due to
the amplification of the electric field around metal nanostructures.
14,15 Solutions of metal colloids have been used for SERS, 16,17
but in some cases they show relatively poor data reproducibility
resulting from uncontrollable aggregation of colloidal particles. 17–19
For this reason, SERS sensors with metal nanostructures fixed
(6) Brailoiu, E.; Churamani, D.; Cai, X. J.; Schrlau, M. G.; Brailoiu, G. C.; Gao,
X.; Hooper, R.; Boulware, M. J.; Dun, N. J.; Marchant, J. S.; Patel, S. J. Cell
Biol. 2009, 186, 201–209.
(7) Calcraft, P. J.; Ruas, M.; Pan, Z.; Cheng, X. T.; Arredouani, A.; Hao, X. M.;
Tang, J. S.; Rietdorf, K.; Teboul, L.; Chuang, K. T.; Lin, P. H.; Xiao, R.;
Wang, C. B.; Zhu, Y. M.; Lin, Y. K.; Wyatt, C. N.; Parrington, J.; Ma, J. J.;
Evans, A. M.; Galione, A.; Zhu, M. X. Nature 2009, 459, 596–U130.
(8) Lewis, A. M.; Masgrau, R.; Vasudevan, S. R.; Yarnasaki, M.; O’Neill, J. S.;
Garnham, C.; James, K.; Macdonald, A.; Ziegler, M.; Galione, A.; Churchill,
G. C. Anal. Biochem. 2007, 371, 26–36.
(9) Graeff, R.; Lee, H. C. Biochem. J. 2002, 367, 163–168.
(10) Gasser, A.; Bruhn, S.; Guse, A. H. J. Biol. Chem. 2006, 281, 16906–16913.
(11) Kneipp, K.; Wang, Y.; Kneipp, H.; Perelman, L. T.; Itzkan, I.; Dasari, R.;
Feld, M. S. Phys. Rev. Lett. 1997, 78, 1667–1670.
(12) Moskovits, M. Rev. Mod. Phys. 1985, 57, 783–826.
(13) Nabiev, I. R.; Morjani, H.; Manfait, M. Eur. Biophys. J. 1991, 19, 311–316.
(14) Vitol, E. A.; Orynbayeva, Z.; Bouchard, M. J.; Azizkhan-Clifford, J.; Friedman,
G.; Gogotsi, Y. ACS Nano 2009, 3, 3529–3536.
10.1021/ac100563t © 2010 American Chemical Society
Published on Web 07/26/2010

on a substrate are preferred. 20-24 The SERS sensor employed in
this work is comprised of a glass substrate coated with gold
nanoparticles. Similar sensors fabricated in the form of glass
nanopipets have been recently demonstrated for minimally invasive
in situ intracellular SERS measurements. 14 In the future, the
results of this study could be potentially extended to NAADP
detection inside cells using the SERS-based approach enabled by
SERS-active nanopipets.
EXPERIMENTAL SECTION
Cell Culture. Breast cancer SkBr3 cells were grown in
McCoy’s 5A modified medium, supplemented with 10% fetal
serum, streptomycin, and penicillin. Cells were purchased from
ATCC.
Acid Extraction of NAADP. For the extraction of NAADP
we used the protocol reported by Lewis et al. 8 All chemicals were
purchased from Sigma-Aldrich. Briefly, SkBr3 cells were treated
with trypsin and suspended in the cell medium. Before treatment
with the agonists, cells were preincubated for 30 min with BAPTA-
AM, 1,2-bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid tetrakis(acetoxymethyl
ester). Then in the presence of BAPTA-AM
cells were stimulated for 20 min with the agonists, according to
the technique described in ref 8. This technique has been shown
to trigger the prolonged NAADP synthesis for at least 20 min
during the agonist stimulation. This results in highly amplified
NAADP production. Here we used histamine, adenosine triphosphate
(ATP), and acetylcholine, all at a 5 µM concentration. The
reaction was stopped by adding 0.75 M ice-cold HClO4. Next the
cells were disrupted by sonication and then kept on ice for 10
min. The disrupted cells were centrifuged at 9000g for 10 min.
Supernatant was neutralized with 1 M KHCO3 and vortexed.
The resulting KClO4 precipitate was removed by centrifugation
at 9000g for 10 min. Samples were stored at -80 °C for later
analysis.
Fabrication of the SERS Sensor. Microscope glass slides
were cut into 1 cm × 1 cm pieces and sonicated in a mixture of
NaOH and ethanol. After being washed with plenty of 15 MΩ
deionized water, the slides were dried at room temperature. Next
the slides were dip coated with 0.001% poly-L-lysine, dried at room
temperature for 24 h, and then coated with gold nanoparticles by
dipping them in the gold colloid for 3 h. Poly-L-lysine promotes
the adhesion of the gold nanoparticles to the glass surface. The
mechanism of nanoparticle attachment is based on the electrostatic
interaction between the negatively charged particles and
(15) Vo-Dinh, T.; Yan, F.; Wabuyele, M. B. J. Raman Spectrosc. 2005, 36, 640–
647.
(16) Ivleva, N. P.; Wagner, M.; Horn, H.; Niessner, R.; Haisch, C. Anal. Chem.
2008, 80, 8538–8544.
(17) Kneipp, J.; Kneipp, H.; McLaughlin, M.; Brown, D.; Kneipp, K. Nano Lett.
2006, 6, 2225–2231.
(18) Chourpa, I.; Lei, F. H.; Dubois, P.; Manfait, M.; Sockalingum, G. D. Chem.
Soc. Rev. 2008, 37, 993–1000.
(19) Willets, K. A. Anal. Bioanal. Chem. 2009, 394, 85–94.
(20) McFarland, A. D.; Van Duyne, R. P. Nano Lett. 2003, 3, 1057–1062.
(21) Hartschuh, A.; Qian, H.; Meixner, A. J.; Anderson, N.; Novotny, L. Surf.
Interface Anal. 2006, 38, 1472–1480.
(22) Haynes, C. L.; Van Duyne, R. P. J. Phys. Chem. B 2003, 107, 7426–7433.
(23) Shoute, L. C. T.; Bergren, A. J.; Mahmoud, A. M.; Harris, K. D.; McCreery,
R. L. Appl. Spectrosc. 2009, 63, 133–140.
(24) Deckert, V.; Zeisel, D.; Zenobi, R.; Vo-Dinh, T. Anal. Chem. 1998, 70, 2646–
2650.
Figure 1. (a) Scanning electron micrograph of the SERS sensor,
(b) close-up view of the gold nanoparticles on the SERS sensor, (c)
extinction spectrum of the SERS sensor, and (d) SERS spectrum of
NAADP and a background spectrum from the substrate.
positively charged NH2 functional groups of poly-L-lysine. 14,25,26
After fabrication, the substrates were imaged with a scanning
electron microscope to confirm the nanoparticle distribution on
the surface. SEM images were collected with a field emission Zeiss
Supra 50VP scanning electron microscope at a low accelerating
voltage (0.7-2 kV) without any conductive coating. In addition,
the UV-vis extinction spectra of the substrates were measured
using a home-built setup employing a fiber-optic spectrometer,
HR-4000, Ocean Optics.
SERS Measurements. Raman spectroscopy was performed
using a micro-Raman spectrometer (Renishaw, RM 1000) equipped
with a 632.8 nm HeNe laser (1800 lines/mm grating) and a diode
InGaAs laser operating at 785 nm wavelength (1200 lines/mm
grating). The lasers are manufactured by Renishaw Inc., U.K. The
laser source was focused on the sample through a long working
distance 50× objective to a spot size of approximately 2 µm. The
typical sample volume was 1 µL. The acquisition time for all
spectra was 10 s. Data analysis was performed using the Renishaw
Wire 2.0 software. Experimental data were analyzed using principal
component analysis 27 in the Matlab environment.
RESULTS AND DISCUSSION
Testing the SERS Sensor for Distinguishing between
Different Secondary Ca 2+ Mobilizing Messengers: NAADP,
cADPR, and IP3. Figure 1a shows the SEM image of the SERS
sensor, with the close-up view presented in panel b. The average
diameter of the nanoparticles is on the order of 50 nm. Assembly
of the SERS sensor is based on the wet chemistry two-step
protocol. First, the glass substrates are coated with a positively
charged polymer (poly-L-lysine) layer. The functionalized substrates
are then coated with a monolayer of negatively charged
gold nanoparticles through electrostatic binding from the gold
(25) Freeman, R. G.; Grabar, K. C.; Allison, K. J.; Bright, R. M.; Davis, J. A.;
Guthrie, A. P.; Hommer, M. B.; Jackson, M. A.; Smith, P. C.; Walter, D. G.;
Natan, M. J. Science 1995, 267, 1629–1632.
(26) Nie, S.; Emory, S. R. Science 1997, 275, 1102–1106.
(27) Jackson, J. E. A User’s Guide to Principal Components; John Wiley: New
York, 1991.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6771

Figure 2. SERS spectra of aqueous solutions of NAADP, cADPR,
and IP3. The bottom spectrum was collected from the mixture of all
three Ca 2+ mobilizing second messengers (3 µM concentration of
each). The excitation laser wavelength was 633 nm. The data
acquisition time was 10 s. The data were collected immediately after
1 µL of the sample was placed on the SERS sensor. The spectra
are normalized and offset for clarity.
colloid. The average interparticle distance is controlled by the time
that the substrates are exposed to the gold colloid. Here, the
average distance is approximately 75 nm. The UV-vis spectrum
of the SERS-enabled substrate, with the maximum extinction at
around 540 nm, is shown in Figure 1c.
The selectivity of the SERS sensor for calcium messengers
was studied with three different samples with 10 µM concentration:
NAADP, IP3, and cADPR. The mixture of all three
messengers was also analyzed. Figure 1 shows that each
messenger has its characteristic SERS spectrum. Moreover, the
analysis of the mixture of all three messengers (bottom spectrum
in Figure 2) shows that it is possible to distinguish the features
of each component.
For example, the adenine moiety of NAADP represents itself
in the spectrum of the mixture with the 733 cm -1 peak, similar
to that observed in the spectrum of the control NAADP
solution. The contribution from cADPR and IP3 to the mixture
spectrum is confirmed by the presence of 898 cm -1 , 1257 cm -1
(amide II), and 1416 cm -1 (C-H stretch) peaks, which are
present in the spectra collected from pure solutions. The 898
cm -1 peak in the spectrum of cADPR can be attributed to
ribose. 17,28 Interestingly, although cADPR contains adenine, it
shows only as a weak signal at 733 cm -1 . This is likely due to
the circular structure of the cADPR molecule, where adenine
is located between two ribose groups. Further analysis of the
Raman spectra is beyond the scope of this work. The key result
demonstrated above is that detection of a specific analyte
(NAADP in our case) is clearly possible using spectral
signatures as a whole obtained from SERS-enabled substrates.
It is important to note that, for the enzymatic cycling assay,
samples must be purified from NADP, which interferes with
(28) Kneipp, J.; Kneipp, H.; Rice, W. L.; Kneipp, K. Anal. Chem. 2005, 77, 2381–
2385.
6772 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. SERS spectra of aqueous solutions of NAD, NADP, and
NAADP at a 100 µM concentration. The data were collected using
the 785 nm excitation laser; the signal acquisition time was 10 s.
NAADP detection. SERS, however, makes it possible to distinguish
between NAADP and its metabolites, such as nicotinamide
adenine dinucleotide (NAD), NADP, and cADPR. Such specificity
to the molecular structure is unattainable, as far as we know, by
any other technique previously applied for NAADP detection.
Figure 3 illustrates the difference between the SERS spectra of
NAD, NADP, and NAADP at a 100 µM concentration. Furthermore,
SERS substrates employed in this work can be utilized in
a wide range of excitation wavelengths. Figures 2 and 3 clearly
illustrate that SERS signals obtained with 633 and 785 nm
excitations have good signal-to-noise ratios. Therefore, the substrates
can be used in conjunction with different lasers, depending
on availability and on the demands of a particular application.
Multiwavelength spectroscopy can also be employed to further
improve differentiation of NAADP spectral signatures using
appropriate pattern recognition.
SERS Detection of an Agonist-Induced Change of the
NAADP Concentration in Cells. Next we studied an agonistinduced
change of the NAADP concentration in breast cancer
SkBr3 cells using the SERS sensor. An increase of the NAADP
concentration was triggered by treating cells with three different
agonists with a 5 µM concentration: ATP, acetylcholine, and
histamine. The protocol for inducing NAADP concentration
modulation results in a final NAADP concentration that is
significantly increased as compared to its basal level. 8,10 The latter
has been estimated to be on the order of tens of nanomolar. The
acid extraction protocol established in ref 8 was used to obtain
NAADP samples. Figure 4a shows the SERS spectrum collected
from the untreated cell extracts, denoted as the control, and those
of the treated cells (Figure 4b-d). Multiple spectra were acquired
from each sample for further analysis. Importantly, the data
collected with the SERS sensor show a good repeatability, as can
be seen in Figure 4. This is expected given the fixed configuration
of the gold nanoparticles on the sensor’s surface. 29,30
(29) Hering, K.; Cialla, D.; Ackermann, K.; Dorfer, T.; Moller, R.; Schneidewind,
H.; Mattheis, R.; Fritzsche, W.; Rosch, P.; Popp, J. Anal. Bioanal. Chem.
2008, 390, 113–124.

Figure 4. SERS analysis of NAADP concentration modulation in cell extracts. Each graph contains 12 spectra collected at different locations
on the SERS sensor for each sample to demonstrate the data reproducibility. (a) SERS spectrum of the untreated cells, marked as the control.
(b-d) SERS spectra of cell extracts with induced NAADP release by treating cells with (b) histamine, (c) ATP, and (d) acetylcholine. All three
agonists had a concentration of 5 µM. The sample volume used in this experiment was on the order of 2 µL. The data acquisition time was 10 s.
A 785 nm excitation laser was used. The spectra are normalized and offset for clarity.
Figure 5. (a) Concentration-dependent SERS spectra of an aqueous solution of NAADP. The bottom spectrum represents the background
signal from the SERS sensor. (b) Principal component analysis of SERS data collected on cells treated with acetylcholine, an aqueous solution
of 100 µM NAADP, and untreated control cells. Each point in the principal component space represents an SERS spectrum with the distance
between data points proportional to the degree of similarity between the spectra. (c) Pareto chart showing the amount of information about data
variability explained by the first two principal components. The results demonstrate that there is a strong correlation between the SERS spectra
of cells with the modulated NAADP concentration and that of the pure 100 µM NAADP solution.
In order to quantify the NAADP concentration in treated cells,
we compare their SERS spectra with the reference NAADP spectra
collected from the pure NAADP aqueous solution. Concentrationdependent
SERS spectra of NAADP are presented in Figure 5a.
(30) Zou, S. L.; Schatz, G. C. Chem. Phys. Lett. 2005, 403, 62–67.
As is often the case in SERS, the spectral signature changes with
concentration. 31 The 733 cm -1 peak of adenine, for example,
dominates the spectrum at higher concentrations. The reduc-
(31) Kim, S. K.; Joo, T. H.; Suh, S. W.; Kim, M. S. J. Raman Spectrosc. 1986,
17, 381–386.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6773

tion in the relative peak intensity for lower concentrations can
be caused by reorientation of NAADP molecules with respect
to the gold nanoparticles’ surfaces in the SERS sensor. 31,32 The
correlation between the SERS spectra of cells treated with
acetylcholine and that of pure NAADP was conducted using
principal component analysis (PCA). 33 Principal component analysis
is a technique which minimizes the dimensionality of the
analyzed data array and permits assessment of the degree of
correlation between large data sets. It is one of the most widely
used methods in chemometrics, and it has been demonstrated to
be efficient for analyzing SERS data. 34-36
According to the PCA results (Figure 5b), the control data
acquired on untreated cells form a cluster which is denoted as
“A”, away from the data of the treated cells, denoted as “B”. This
confirms that the SERS sensor employed here distinguishes
between the cells producing different amounts of NAADP.
Furthermore, there is a clear correlation between the SERS
spectra of the treated cells and that of the aqueous solution of
100 µM NAADP. This concentration is within the range of the
expected induced NAADP concentration increase, according to
the protocol that was used in this work. While this concentration
(32) Barhoumi, A.; Zhang, D. M.; Halas, N. J. J. Am. Chem. Soc. 2008, 130,
14040–14041.
(33) Jolliffe, I. T. Principal Component Analysis, 2nd ed.; Springer: New York,
2002.
(34) Eliasson, C.; Loren, A.; Engelbrektsson, J.; Josefson, M.; Abrahamsson, J.;
Abrahamsson, K. Spectrochim. Acta, Part A 2005, 61, 755–760.
(35) Pearman, W. F.; Fountain, A. W. Appl. Spectrosc. 2006, 60, 356–365.
(36) Hedegaard, M.; Krafft, C.; Ditzel, H. J.; Johansen, L. E.; Hassing, S.; Popp,
J. R. Anal. Chem. 2010, 82, 2797–2802.
6774 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
is higher than the one which can be detected with enzymatic and
radioreceptor binding assays, there is a significant advantage of
time efficiency and accessibility of the SERS-based method.
CONCLUSION
Label-free NAADP detection and quantification in cell extracts
is enabled by SERS, which permits the rapid detection of NAADP
with a 100 µM concentration without any special sample purification
or labeling. Importantly, this concentration does not represent
a limit for SERS sensing of second calcium messengers. We were
able to successfully detect 10 nM concentrations of NAADP in
aqueous solution, which is on the order of basal levels of NAADP
in cells, suggesting that intracellular SERS detection of the calcium
messengers is possible.
ACKNOWLEDGMENT
This work was supported by a W. M. Keck Foundation grant
to establish the W. M. Keck Institute for Attofluidic Nanotube-
Based Probes at Drexel University, by the Pennsylvania Nanotechnology
Institute (NTI) through Ben Franklin Technology
Partners of Southeastern Pennsylvania, and by NIH Grants HL
90804 and HL 90804-01A2S1 to E.B. Raman spectroscopy analysis
and scanning electron microscopy were conducted at the Centralized
Research Facilities (CRF) at Drexel University.
Received for review March 2, 2010. Accepted July 2, 2010.
AC100563T

Anal. Chem. 2010, 82, 6775–6781
Highly Sensitive Fluorescent Method for the
Detection of Cholesterol Aldehydes Formed by
Ozone and Singlet Molecular Oxygen
Fernando V. Mansano, Rafaella M. A. Kazaoka, Graziella E. Ronsein, Fernanda M. Prado,
Thiago C. Genaro-Mattos, Miriam Uemi, Paolo Di Mascio, and Sayuri Miyamoto*
Departamento de Bioquímica, Instituto de Química, Universidade de São Paulo,
CP26077, CEP 05513-970, São Paulo, SP, Brazil
Cholesterol oxidation gives rise to a mixture of oxidized
products. Different types of products are generated according
to the reactive species being involved. Recently,
attention has been focused on two cholesterol aldehydes,
3�-hydroxy-5�-hydroxy-B-norcholestane-6�-carboxyaldehyde
(1a) and 3�-hydroxy-5-oxo-5,6-secocholestan-6-al
(1b). These aldehydes can be generated by ozone-, as well
as by singlet molecular oxygen-mediated cholesterol oxidation.
It has been suggested that 1b is preferentially
formed by ozone and 1a is preferentially formed by singlet
molecular oxygen. In this study we describe the use of
1-pyrenebutyric hydrazine (PBH) as a fluorescent probe
for the detection of cholesterol aldehydes. The formation
of the fluorescent adduct between 1a with PBH was
confirmed by HPLC-MS/MS. The fluorescence spectra of
PBH did not change upon binding to the aldehyde.
Moreover, the derivatization was also effective in the
absence of an acidified medium, which is critical to avoid
the formation of cholesterol aldehydes through Hock
cleavage of 5r-hydroperoxycholesterol. In conclusion,
PBH can be used as an efficient fluorescent probe for the
detection/quantification of cholesterol aldehydes in biological
samples. Its analysis by HPLC coupled to a
fluorescent detector provides a sensitive and specific way
to quantify cholesterol aldehydes in the low femtomol
range.
Cholesterol (cholest-5-en-3�-ol) is a neutral lipid found in the
cellular membranes of mammals. 1 Cholesterol is susceptible to
oxidation mediated by enzymatic and nonenzymatic mechanisms. 2-5
The nonenzymatic oxidation can be mediated by reactive oxygen
species. 2 Several oxidized products of cholesterol have been characterized
including hydroperoxides, epoxides, and aldehydes. 2,6,7
These oxysterols have been detected in biological tissues and their
formation has been associated to neurodegenerative and cardio-
* To whom correspondence should be addressed. Phone: (55) (11) 3091-
3810 (x261). Fax: (55) (11) 3815-5579. E-mail: miyamoto@iq.usp.br.
(1) Maxfield, F. R.; Tabas, I. Nature 2005, 438, 612–621.
(2) Brown, A. J.; Jessup, W. Atherosclerosis 1998, 142, 1–28.
(3) Björkhem, I.; Diczfalusy, U. Arterioscler. Thromb. Vasc. Biol. 2002, 22,
734–742.
(4) Garenc, C.; Julien, P.; Levy, E. Free Radical Res. 2010, 44, 47–73.
(5) Schroepfer, G. J., Jr. Physiol. Rev. 2000, 80, 361–554.
(6) Smith, L. L. Lipids 1996, 31, 453–487.
(7) Girotti, A. W. J. Photochem. Photobiol. B 1992, 13, 105–118.
vascular diseases. 2-5 Recently, attention has been focused on
cholesterol aldehydes that can be formed by the oxidation of
cholesterol by ozone 8 and singlet molecular oxygen. 9,10
The ozonation of cholesterol produces several oxidized products,
in special two aldehydes (Figure 1), the cholesterol 5,6secosterols,3�-hydroxy-5�-hydroxy-B-norcholestane-6�-carboxyaldehyde
(1a) and 3�-hydroxy-5-oxo-5,6-secocholestan-6-al (1b). 11,12
Wentworth and co-workers showed the presence of 1a and 1b
in atherosclerotic plaques 8 and LDL oxidized with several oxidants.
13 These aldehydes have been also detected in neurodegenerative
diseases, like Lewy body dementia 14 and Alzheimer
disease. 15 The role of 1a and 1b in the pathogenesis of
cardiovascular and neurodegenerative diseases has been investigated.
In vitro studies have shown that cholesterol aldehydes
can covalently modify proteins, as well as, accelerate their
aggregation, as in the case of amyloid �-peptide formation 15-17
and R-synuclein 14 . Further studies have shown that covalent
modification of apo-B by 1b causes this protein to misfold,
rendering the LDL particle more susceptible to macrophage
uptake. 18 Moreover, some reports have also shown that
cholesterol aldehydes can induce apoptosis in macrophages
and cardiomyoblasts. 19,20 The induction of apoptosis in cardi-
(8) Wentworth, P., Jr.; Nieva, J.; Takeuchi, C.; Galve, R.; Wentworth, A. D.;
Dilley, R. B.; DeLaria, G. A.; Saven, A.; Babior, B. M.; Janda, K. D.;
Eschenmoser, A.; Lerner, R. A. Science 2003, 302, 1053–1056.
(9) Brinkhorst, J.; Nara, S. J.; Pratt, D. A. J. Am. Chem. Soc. 2008, 130, 12224–
12225.
(10) Uemi, M.; Ronsein, G. E.; Miyamoto, S.; Medeiros, M. H. G.; Di Mascio,
P. Chem. Res. Toxicol. 2009, 22, 875–884.
(11) Gumulka, J.; Smith, L. L. J. Am. Chem. Soc. 1983, 105, 1972–1979.
(12) Jaworski, K.; Smith, L. L. J. Org. Chem. 1988, 53, 545–554.
(13) Wentworth, A. D.; Song, B. D.; Nieva, J.; Shafton, A.; Tripurenani, S.;
Wentworth, P. Chem. Commun. 2009, 3098–3100.
(14) Bosco, D. A.; Fowler, D. M.; Zhang, Q.; Nieva, J.; Powers, E. T.; Wentworth,
P.; Lerner, R. A.; Kelly, J. W. Nat. Chem. Biol. 2006, 2, 249–253.
(15) Zhang, Q.; Powers, E. T.; Nieva, J.; Huff, M. E.; Dendle, M. A.; Bieschke,
J.; Glabe, C. G.; Eschenmoser, A.; Wentworth, P., Jr.; Lerner, R. A.; Kelly,
J. W. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 4752–4757.
(16) Scheinost, J. C.; Wang, H.; Boldt, G. E.; Offer, J.; Wentworth, P. Alzheimer
Dis. 2008, 47, 3919–3922.
(17) Usui, K.; Hulleman, J. D.; Paulsson, J. F.; Siegel, S. J.; Powers, E. T.; Kelly,
J. W. Proc. Natl. Acad. Sci. U.S.A. 2009, 106, 18563–18568.
(18) Takeuchi, C.; Galve, R.; Nieva, J.; Witter, D. P.; Wentworth, A. D.; Troseth,
R. P.; Lerner, R. A.; Wentworth, P. Biochemistry 2006, 45, 7162–7170.
(19) Sathishkumar, K.; Gao, X.; Raghavamenon, A. C.; Parinandi, N.; Pryor, W. A.;
Uppu, R. M. Free Radical Biol. Med. 2009, 47, 548–558.
(20) Gao, X.; Raghavamenon, A. C.; D’Auvergne, O.; Uppu, R. M. Biochem.
Biophys. Res. Commun. 2009, 389, 382–387.
10.1021/ac1006427 © 2010 American Chemical Society 6775
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/21/2010

Figure 1. Structures of cholesterol carboxyaldehyde (1a), cholesterol secocholestanal (1b) and their corresponding fluorescent adducts formed
upon derivatization with 1-pyrenebutyric hydrazide (2a and 2b).
omyoblasts was associated with the induction of ROS generation
and the activation of extrinsic and intrinsic pathway. 19
Besides ozone, recent studies have evidenced the generation
of cholesterol aldehydes in the reaction of cholesterol with singlet
molecular oxygen. The proposed mechanisms involve Hockcleavage
of cholesterol 5R-hydroperoxide (5R-OOH) in the presence
of acids 9 or cholesterol dioxetane decomposition. 10 In both
cases, 1a was detected as the major product. Supporting these
studies, Tomono et al. 21 have also detected 1a as the major
product in the incubation of cholesterol with human myeloperoxidase
in the presence H2O2 and chloride ions, a system that
is suggested to generate singlet molecular oxygen in activated
neutrophils. 22
Several methods have been used for the determination of
cholesterol aldehydes, including mass spectrometry, 10 UVvisible,
8,13,23 and fluorescence detection, 14,21 most of them coupled
to liquid chromatography. Although mass spectrometry based
methods have shown a good sensitivity, this technique requires
specialized people and expensive equipment. UV-visible detection
methods using 2,4-dinitrophenylhydrazine (DNPH) have been
widely used to detect and quantify aldehydes derived from lipid
peroxidation. However, this method is usually carried out in a
strong acidic media, which is able to induce the cleavage of
cholesterol 5R-OOH to form 1a and 1b, 9 therefore leading to an
overestimation of the real aldehyde concentration in biological
samples. An alternative method used to detect aldehydes involves
their reaction with fluorescent probes and the detection of the
fluorescent adducts formed. The fluorescent methods have been
used to increase detection limits and exclude interferences. The
fluorescent probe described in the literature for cholesterol 5,6secosterols
detection is dansyl hydrazine. Bosco and co-workers
used this probe in acidic media to detect cholesterol aldehydes
in brain. 14
(21) Tomono, S.; Miyoshi, N.; Sato, K.; Ohba, Y.; Ohshima, H. Biochem. Biophys.
Res. Commun. 2009, 383, 222–227.
(22) Kiryu, C.; Makiuchi, M.; Miyazaki, J.; Fujinaga, T.; Kakinuma, K. FEBS
Lett. 1999, 443, 154–158.
(23) Wang, K. Y.; Bermudez, E.; Pryor, W. A. Steroids 1993, 58, 225–229.
6776 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
This manuscript describes a new fluorescence-based method
to detect and quantify cholesterol aldehydes using the probe
1-pyrenebytiric hydrazine (PBH) (Figure 1). This method proved
to be highly sensitive and has the advantage of minimizing the
effect of interfering compounds and not requiring acidic media.
The use of acid conditions showed to cause an overestimation of
1a in a sample containing cholesterol 5R-OOH. Using PBH as
the fluorescent probe we could detect 1a in the low femtomolar
range by HPLC coupled to fluorescence detector. Moreover, this
methodology allowed detecting and calculating the ratio of 1a and
1b formed by the oxidation of cholesterol exposed to singlet
molecular oxygen and ozone, respectively.
EXPERIMENTAL SECTION
Reagents. Cholesterol, silica gel (200-400 mesh, 60 Å),
1-pyrenebutyric hydrazide (PBH), deuterated chloroform (CDCl3),
sodium phosphate monobasic and dibasic were purchased from
Sigma (St. Louis, MO). Methylene blue was purchased from
Merck (Rio de Janeiro, Brazil). All the other solvents were of
HPLC grade and were acquired from Mallinckrodt Baker
(Phillipsburg, NJ). The water used in the experiments was
treated with the Nanopure Water System (Barnstead, Dubuque,
IA).
Synthesis of 3�-Hydroxy-5�-hydroxy-B-norcholestane-6�carboxyaldehyde
(1a). 1a was synthesized by photooxidation
of cholesterol in the presence of methylene blue as described
previously. 10 Briefly, 200 mg of cholesterol dissolved in 20 mL of
chloroform was mixed with 250 µL of methylene blue (10 mM in
methanol). This solution was cooled at 4 °C and irradiated under
continuous agitation using two tungsten lamps (500 W) during
2.5 h. The formation of cholesterol oxidation products were
checked by thin-layer chromatography using ethyl acetate and
isooctane (1:1, v/v) as the eluent. 1a was purified from the
reaction mixture by flash column chromatography, using silica
gel and a gradient of hexane and ethyl eter. The purified 1a was
analyzed and quantified in CDCl 3 by NMR spectroscopy using
DRX500 instrument, AVANCE series (Bruker-Biospin, Rheinstetten,
Germany) as described previously. 10 For the quantifica-

tion, 2 µL of proprionaldehyde were added as the internal
standard into 750 µLofthe1a solution in CDCl3. The 1 H signal
corresponding to the aldehyde group of the internal standard
and 1a appeared at 9.73 and 9.62 ppm, respectively. The relative
signals of 1 H of aldehydes were integrated by Mestre C
software (Figure S1 in the Supporting Information (SI)).
Synthesis of 3�-Hydroxy-5-oxo-5,6-secocholestan-6-al (1b).
1b was prepared by ozonation of cholesterol as described by
Wentworth et al. 8 and Wang et al. 24 Ozone was produced at a
rate of 100 mg/h by passing pure oxygen through AquaZone
PLUS 200 instrument (Red Sea Fish Pharm. Ltd., Houston, TX).
Oxygen flow rate was set to 10 mL/min. A solution of 10 mg/mL
of cholesterol in chloroform was cooled at dry ice temperature
and oxidized by bubbling ozone for 5 min. After oxidation,
chloroform was evaporated with nitrogen gas and the residue was
stirred for 2hatroom temperature with Zn powder (19.4 mg) in
water-acetic acid (1:19 v/v; 3.1 mL). Dichloromethane (15 mL)
was added and the mixture was washed five times with deionized
water (5 × 15 mL). The organic phase was evaporated to dryness
in vacuo. The residue was dissolved in isopropyl alcohol and kept
at -80 °C for further analysis. The characterization of 1b was
performed by NMR spectroscopy using a DRX500 instrument,
AVANCE series (Bruker-Biospin, Rheinstetten, Germany) operating
at 11.7 T. Cholesterol ozonation followed by Zn reduction in
acetic acid gave 1b as a major aldehyde, as confirmed by 1 H NMR
analysis (δ, ppm, CDCl3): δ 9.607 (s, J ) 0.5 Hz, 1H, CHO),
4.466 (s, 1H, H-3), 3.097 (dd, J ) 13.5, 4.0 Hz, 1H, H-4e), 1.010
(s, 3H, CH3-19), 0.671 (s, 3H, CH3-18) (Figure S2 in the SI).
Derivatization of Cholesterol Aldehydes with PBH. The
optimal conditions for the derivatization of cholesterol aldehydes
were established using the purified 1a. Time-course of PBH
adduct formation with 1a was monitored by incubating 100 µM
1a with 2 mM PBH in isopropanol at 37 °C under continuous
agitation for up to 8 h. Aliquots of 1 µL of the reaction mixture
were taken at specified times and analyzed by HPLC coupled to
fluorescence detector. The ideal concentration of the probe was
determined by incubating 1 µM of1a in the presence of 1-1000
µM of PBH in isopropanol at 37 °C under continuous agitation
for 6 h. The effect of the pH in the formation of 2a was analyzed
by incubating 1 µM of1a with 50 µM PBH in a reaction system
consisted of isopropanol:10 mM phosphate buffer at pH 5.7, 7.4,
and 8.0 (90:10, vol/vol). Quantitative analysis of 1a was done by
incubating an aliquot of sample with 600 µM of the probe in
isopropanol containing 1 mM phosphate buffer at pH 7.4 (neutral
condition) or 0.1 mM HCl (acid condition) at 37 °C for6h.
Analysis of the Fluorescent Adduct by HPLC Coupled
with Fluorescence Detector. HPLC analysis was carried out on
a Shimadzu Prominence system (Tokyo, Japan), consisted of LC-
20AT pumps, SIL-20AV autosampler, RF-10Axl fluorescence detector,
and a CBM-20A controller. One microliter of the sample was
injected into a reversed-phase column Synergi C18 (50 × 4.6 mm,
2.5 µm particle size, Phenomenex, Torrance, CA). The HPLC
mobile phase consisted of water (A) and methanol (B), and the
flow rate was 1 mL/min. The separation of the fluorescent adducts
was done using the following condition: 86% B for 5 min, 86-92%
in 1 min, 92% B for 15 min, and 92-86% B in 1 min. The excitation
and emission wavelengths were fixed at 339 and 380 nm,
respectively. 25 For the analysis 1 µL was injected through the
autosampler. Data were acquired at high sensitivity and gain of
1× in the fluorescence detector. Fluorescence spectra were
acquired by the fluorescence detector RF-551 (Shimadzu, Japan).
HPLC data were processed using the LC solution software
(Shimadzu, Japan).
Analysis of the Fluorescent Adducts by HPLC-MS/MS.
The PBH fluorescent adducts formed with 1a and 1b were
analyzed by a HPLC system connected to a UV-visible detector
(SPD 10 AVVP, Shimadzu, Kyoto, Japan) and a Quattro II triple
quadrupole tandem mass spectrometer (Micromass, Manchester,
UK) (HPLC-MS/MS). HPLC separation was carried out as
described above. The eluent from the column was monitored at
342 nm and 10% of the HPLC flow rate was directed into the mass
spectrometer. The fluorescent adducts were detected using
electrospray ionization (ESI) in the positive ion mode. The source
and desolvation temperature of the mass spectrometer were set
at 100 and 200 °C, respectively. The cone voltage was set to 50 V,
the extractor cone voltage was set to 5 V and the capillary and
the electrode potentials were set at 4.5 and 0.5 kV, respectively.
Collision energy was set at 20 eV for 2a and 30 eV for 2b. Fullscan
data was acquired over a mass range of 100-900 m/z. Data
was processed by means of the MassLynx NT software.
Cholesterol Oxidation Induced by Photooxidation, Ozone
and HOCl/H 2O2. Cholesterol (10 mg/mL in chloroform) was
photooxidized in the presence of methylene blue for 2.5 h. An
aliquot of this sample was taken for cholesterol aldehyde determination.
The ozonation of cholesterol (10 mg/mL in chloroform)
was conducted essentially as described for the synthesis of 1b.
After 5 min ozonation, chloroform was evaporated by nitrogen
gas and the residue was dissolved in isopropanol. An aliquot of
this sample was diluted and used for cholesterol aldehyde
determination. The oxidation of cholesterol by HOCl/H 2O2
system was conducted by incubating a 10 mM cholesterol
solution containing 1% ethanol in 10 mM phosphate buffer at
pH 7.4 with 1 mM HOCl and H2O2, under continuous agitation
at 37 °C for 5 min.
Method Validation. The limit of quantification (LOQ) was
established as the amount of adduct formed that generated a signal
6-fold higher than baseline. The limit of detection (LOD) was
established as the amount of adduct formed that generated a signal
3-fold higher than baseline. The reproducibility was checked by
intra- and interday analysis. Interday analyses were conducted on
three consecutive days.
RESULTS
Synthesis and Characterization of 1a. Cholesterol carboxyaldehyde
(1a) was synthesized by photooxidation of cholesterol. 10
After the reaction, the oxidized products were purified by flash
column chromatography and 1a was isolated. The identity of 1a
was confirmed by NMR spectroscopy as described by Uemi et
al. 10 The 1 H NMR analysis showed a doublet peak at 9.62 ppm
consistent with the presence of the aldehyde group in 1a
structure (Figure S1 in the SI). 1a was quantified by 1 H NMR
analysis using proprionaldehyde as the internal standard as
described in the experimental section.
Characterization of 2a Fluorescent Adduct. Using the
purified 1a sample, several experiments were conducted to
(24) Wang, K. Y.; Bermudez, E.; Pryor, W. A. Steroids 1993, 58, 225–229. (25) Morsel, J. T.; Schmiedl, D. Fresenius´ J. Anal. Chem. 1994, 349, 538–541.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6777

Figure 2. Analysis of 2a fluorescent adduct, formed in the reaction
of 1a with PBH by HPLC coupled to fluorescence detector. HPLC/
fluorescence chromatogram obtained for the analysis of 2a using
excitation at 339 nm and emission at 380 nm. For the derivatization,
5 pmol of 1a was incubated with PBH 100 µM in a final volume of
100 µL isopropanol at 37 °C for 6 h under agitation. Analysis was
done by injecting 1 µL of the reaction mixture into the HPLC. The
inset shows the time dependent formation of 2a during incubation of
100 µM of1a in the presence of 2 mM of PBH for up to 8 h.
establish the optimal conditions for the reaction with PBH.
Incubations of 1a with PBH were conducted in isopropanol at 37
°C under continuous agitation. HPLC analysis using fluorescence
detection showed the appearance of an intense peak at 12.5 min,
corresponding to the fluorescent adduct 2a (Figure 2). The pH
effect on 2a adduct formation was analyzed. Incubations conducted
at pH 5.7, 7.4, and 8.5 showed that this adduct was
preferentially formed at pH 5.7 (Figure S3 in the SI), which is in
accordance with favorable formation of Schiff base adduct at
slightly acidic conditions.
A time-course analysis of 2a showed a time-dependent increase
of the peak area up to 4 h, reaching a plateau after this time (inset,
Figure 2). Based on this, the incubation time for the derivatization
was fixed to 6 h. The ideal concentration of PBH for the reaction
was established by incubating 1a (1 µM) in the presence of
1-1000 µM of PBH. The fluorescent adduct formation reached
its maximum with PBH concentration higher than 50 µM.
The fluorescence of 2a was characterized to establish the
optimal excitation and emission wavelengths. Fluorescence analysis
of 2a showed a spectrum similar to the original probe with
excitation and emission wavelengths maximum at 339 and 380
nm, respectively (Figure 3). Thus, indicating that adduct formation
does not alter the fluorescence spectra of the probe.
The formation of 2a was also confirmed by HPLC-MS/MS.
The mass spectrum of 2a acquired by ESI in the positive ion mode
showed peaks corresponding to 2a molecular ion ([M+H] + ) and
its sodium adducts ([M+Na] + )atm/z 703 and 725, respectively
(Figure 4A and 4B). Two other peaks at m/z 685 and 667 were
also detected, corresponding to the loss of one ([M+H-H2O] + )
and two water molecules ([M+H-2H2O] + ), respectively. These
ions were also observed in the MS/MS spectrum of m/z 703
(Figure 4C). The MS/MS spectrum also showed other fragment
ions at m/z 431, 398, 365, and 303. The first three, correspond to
the ions formed by the loss of pyrene butyric (PB) group
6778 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. Fluorescence excitation and emission spectra of PBH and
2a. Both compounds showed the same excitation and emission
maximum wavelengths at 339 and 380 nm, respectively.
([M+H-PB] + , m/z 431), PBH and two protons ([M+H-PBH-
2H] + , m/z 398) and PBH and two water molecules
([M+H-PBH-2H2O] + , m/z 365), respectively. The fragment
ion at m/z 303 corresponds to the positively charged PBH ion.
Reproducibility and Stability. The intra- and interday reproducibility
of the assay procedure was determined by evaluating
three individual reactions containing 1a (1 pmol; 1 µM final
concentration) and PBH (50 µM). The relative standard deviation
for the intra- and interday reproducibility at this concentration was
1.7% and 7.6%, respectively.
The stability of 2a adduct was evaluated in triplicate reactions
of 1a (1 pmol; 1 µM final concentration) with PBH (50 µM). After
derivatization, these samples were kept in vials inside the
autosampler at 4 °C and the 2a peak was monitored up to 6 h.
There was no significant change in the peak area of the
fluorescence signal during this period.
Standard Curve, Limit of Detection and Limit of Quantification.
A calibration curve was constructed using 5-100 fmol
(50-1000 nM) of 1a in triplicate reaction in a final volume of 100
µL. Peak area of 2a increased linearly over this concentration
range with R 2 value of 0.9962 and Y ) 6.64 × 10 3 X + 1.65 ×
10 4 . The limit of detection (LOD) and quantification (LOQ) of
1a was 10 fmol (10 nM) and 20 fmol (20 nM), respectively.
Application of PBH for Cholesterol Aldehyde Detection
in Oxidized Cholesterol Samples. The newly developed fluorescence-based
method was applied for the detection and quantification
of 1a in cholesterol samples oxidized by photooxidation
or ozone. Cholesterol photooxidation gave rise to a major peak at
12.5 min (Figure 5A). On the other hand, cholesterol ozonation
yielded an intense peak at 10.3 min as well as some other smaller
peaks, including the same peak at 12.5 min observed for the
photooxidation (Figure 5B). Peak assignment was done by
comparison of the retention times of the peaks with those obtained
for the standard samples of 1a (Figure 5C) and 1b (Figure 5D).
In this way, the peaks observed at 10.3 and 12.5 min were assigned
to the fluorescent adducts 2b and 2a, respectively. Additionally,
the identity of the peaks was confirmed by HPLC-MS/MS analysis
(Figure S4 in the SI). It should be noted that the overall retention
times for the analysis by HPLC-MS/MS were increased by almost
2 min compared to the analysis by HPLC/fluorescence detection.
This retention time delay was due to the differences in room

Figure 4. Analysis of 2a by HPLC-MS/MS using ESI in the positive ion mode. The cone voltage and the collision energy were set to 50 V and
20 eV, respectively. Selected ion chromatogram of the ion at m/z 703 (A). Spectrum of the peak corresponding to 2a at 14 min (B). Fragment
ion spectrum of m/z 703 (C).
Figure 5. Analysis of cholesterol aldehydes formed in the photooxidation
(A) and ozonation (B) of cholesterol using PBH. An aliquot
of the sample was reacted with PBH (600 µM) for 6 h and analyzed
by HPLC coupled to fluorescence detector. For peak assignment
standard samples containing 1a (C) or 1b (D) were also reacted with
PBH and analyzed.
temperatures, which was about 5 °C lower in the case of HPLC-
MS/MS analysis.
The major fluorescent adduct detected in cholesterol ozonation
was further characterized by HPLC-MS/MS analysis (Figure 6).
Selected ion mass chromatogram of the ion at m/z 703 showed a
single peak at 12 min. (Figure 6A). Mass spectrum for this peak
showed two major ions, one at m/z 703 and other at m/z 725
(Figure 6B), which corresponds to the molecular ion ([M+H] + )
and the sodium adduct ([M+Na] + )of2b (Figure 6B). Collision
induced dissociation of the ion at m/z 703, showed the same
fragment ions observed for 2a, suggesting that the peak corresponds
to its isomer, 2b (Figure 6C). Fragments ions observed
at m/z 685 and 667 are formed by the loss of one and two water
molecules from 2b, respectively. The ions at m/z 431, 398, 383,
and 365 are formed by the loss of PB, PBH and two protons, PBH
and one water, and PBH and two water molecules from 2b,
respectively (Figure 6C).
Aiming to get a quantitative data on cholesterol aldehyde
formation, a calibration curve constructed for 1a was used to
determine its concentration. For the quantification, oxidized
cholesterol samples were diluted to get a final cholesterol
concentration of 219 µM. The amounts of 1a detected in the
samples by PBH method at neutral pH were 19.1 ± 4.0 µM and
0.7 ± 0.3 µM for the photooxidation and ozonation, respectively.
This corresponds to a 1a yield of 8.7% for the photooxidation and
0.3% for the ozonation. For comparison, 1a quantification was also
done using dansyl hydrazine method in the presence of acid (0.1
mM HCl) (method in the SI). The concentrations of 1a determined
by this method were 35.2 ± 0.7 µM and 1.0 ± 0.1 µMinthe
photooxidation and ozonation, respectively. This provides 1a
yields of 16% for the photooxidation and 0.4% for the ozonation.
As can be clearly noticed, the 1a yield determined by the dansyl
hydrazine method was overestimated by almost 2-folds for the
photooxidation. The same result was obtained when PBH derivatization
was conducted in the presence of acid. Under this
condition, the 1a concentration was 32.6 ± 0.7 µM, which
corresponds to a 1a yield of 15% in the photooxidation. These
results are consistent with the fact that photooxidized cholesterol
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6779

Figure 6. Analysis of 2b by HPLC-MS/MS using ESI in the positive ion mode. The cone voltage was set to 50 V and the collision energy was
set to 30 eV. Selected ion chromatogram for the ion at m/z 703 (A). Mass spectrum of the peak corresponding to 2b at 12 min (B). Fragment
ion spectrum of m/z 703 (C).
samples contain cholesterol 5R-OOH, which in the presence of
acids is easily converted to 1a. 9
Moreover, the sensitivity of PBH and dansyl hydrazine
methods were also compared (Figure S5 in the SI). PBH method
conducted at neutral pH was almost 10 times more sensitive than
dansyl hydrazine method. This difference was even larger when
PBH reaction was carried out with acid, reaching a 90 times higher
sensitivity.
PBH was also used for the quantification of ChAld formed
during cholesterol oxidation promoted by the HOCl/H 2O2 reaction
system. This is a biologically relevant reaction system that
is known to generate stoichiometric amounts of singlet molecular
oxygen. The formation of 2a was analyzed by HPLC
coupled with fluorescence detector (Figure 7). The complete
reaction system containing cholesterol and HOCl/H2O2 showed
the appearance of an intense fluorescent peak corresponding
to 2a. The fluorescent adduct was quantified using the
calibration curve and a value of 0.2 µM of1a was found. This
corresponds to a yield of 0.2%.
DISCUSSION
Cholesterol is oxidized in the presence of reactive oxygen
species generating aldehydes, hydroperoxides and epoxides. 6,7
Recently, two cholesterol aldehydes have attracted attention, the
cholesterol carboxyaldehyde (1a) and cholesterol secoaldehyde
(1b). The formation of 1a and 1b was first described in the
oxidation of cholesterol with ozone. 11,12 Wentworth and co-workers
reported the presence of 1a and 1b in atherosclerotic plaques
and related the detection of these aldehydes as an evidence for
ozone generation in human tissues. 8 On the other hand, two recent
studies identified the generation of 1a in the reaction of cholesterol
with singlet molecular oxygen. 9,10 Indeed, Brinkhorst and
6780 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 7. Analysis of cholesterol aldehydes generated by the
reaction of cholesterol with H2O2 and HOCl using PBH. HPLC/
fluorescence chromatograms obtained for cholesterol (A), and cholesterol
incubated with H2O2 (B), HOCl (C), and H2O2 + HOCl (D).
Reactions were carried out with 100 µM cholesterol in the presence
of 1 mM HOCl, 1 mM H2O2 and 100 µM PBHfor6hat37°C under
continuous agitation.
co-workers reported the formation of 1a from cholesterol 5Rhydroperoxide
by Hock cleavage in acid media 9 and Uemi and
co-workers reported the formation of 1a in the reaction of
cholesterol with singlet molecular oxygen generated either by
photooxidation or by the thermodecomposition of endoperoxides.
10
In this study, we have developed a new detection method for
cholesterol carboxyaldehyde, 1a, using the fluorescent probe
PBH, which contains the highly fluorescent pyrene group. The

eaction of 1a with PBH can be conducted in the absence of acids,
which is important to avoid the formation of 1a from acid catalyzed
cholesterol 5R-OOH decomposition. 9 Bosco and co-workers 14 used
dansyl hydrazine as a fluorescent label to detected 1a and 1b in
a brain tissue sample from Lewy body disease. However, the
derivatization of cholesterol aldehydes with dansyl hydrazine is
conducted in the presence of strong acids, such as, 5 mM sulfuric
acid. 14 In fact, we have done a comparative analysis of the two
fluorescent probes (PBH and dansyl hydrazine) to detect 1a in
vitro and observed that the method using dansyl hydrazine can
overestimate the basal level of 1a, especially in samples containing
cholesterol 5R-OOH. The 1a yields in photooxidized cholesterol
samples estimated by PBH (neutral pH) and dansyl hydrazine
were 8.7% and 16.0%, respectively. This data indicates that the
derivatization in the absence of acid is a critical point that should
be considered when estimating 1a basal level in samples,
especially when they were oxidized by singlet molecular oxygen.
When comparing dansyl hydrazine and PBH methods, the latter
was approximately 10 times more sensitive for 1a detection
(Figure S5 in the SI). This difference was even larger when PBH
derivatization was carried out at acid condition, clearly evidencing
the superiority of PBH method compared to dansyl hydrazine in
terms of sensitivity.
A number of studies has detected the formation of cholesterol
aldehydes in vitro, 9,10,21 as well as in vivo. 8,14-16 In this study, we
have used PBH to analyze the formation of 1a during cholesterol
oxidation mediated by ozone and singlet molecular oxygen. The
reaction of cholesterol with ozone yielded 1a and 1b and also
other minor unidentified products. We could not quantify exactly
the amount of 1b due to the lack of an appropriate pure standard
for it. Nonetheless, the relative peak intensities suggest that
cholesterol ozonation yields 1b as a major product and a small
amount of its aldolization product (Figure 5). The estimated ratio
of 1a:1b in the ozonation was 1:10. Similar results were described
for the ozonation of human LDL, where a relative yield of 1:4 was
found. 13 On the other hand, cholesterol oxidation promoted by
singlet molecular oxygen yielded 1a as the major product
consistent with the data described by Brinkhorst et al. 9 and Uemi
et al. 10 The estimated yield of 1a in the photooxidation of
cholesterol was approximately 8.7% and the ratio of 1a:1b was
165:1. These values are in agreement with those reported for the
photooxidation of human LDL using hematoporphyrin IX for 14 h
where only 1a was detected. 13 The incubation of cholesterol in
(26) Khan, A. U.; Kasha, M. J. Am. Chem. Soc. 1970, 92, 3293–3300.
the presence of H2O2 and HOCl also gave 1a as the major
product, consistent with the oxidation of cholesterol promoted
by singlet molecular oxygen. It is known that the reaction of
H2O2 with HOCl generate stoichiometric amounts of singlet
molecular oxygen 26 and this reaction is suggested to play an
important role during inflammatory conditions. 22 The predominant
formation of 1a over 1b by this reaction was also reported
by Tomono and co-workers. 21 They incubated cholesterol (100
µM) in the presence of H2O2 (100 µM) and NaOCl (100 µM)
and detected five times more 1a than 1b using dansyl
hydrazine as a fluorescent probe. 21
CONCLUSIONS
We have developed a new sensitive method for the detection
of cholesterol aldehydes using PBH as a fluorescent probe. This
new methodology allows detecting and quantifying the relative
amounts of 1a and 1b with high sensitivity, which is important
to assess the relative contributions of ozone and singlet molecular
oxygen to the oxidation of cholesterol in biological systems.
Derivatization of cholesterol aldehydes using PBH can be conducted
without the addition of strong acids, which is important
for the determination of the basal level of cholesterol aldehydes
present in biological tissues from different sources. In conclusion,
the easy and reliable method developed in this study can help to
investigate the formation cholesterol aldehydes in biological
system, which is critical to clarify their relevance in disease
progression.
ACKNOWLEDGMENT
This work was supported by the Brazilian research funding
institutions FAPESP (Fundação de Amparo à Pesquisa do Estado
de São Paulo), CNPq (Conselho Nacional para o Desenvolvimento
Científico e Tecnológico), INCT de Processos Redox em Biomedicina
- Redoxoma, and Pró-Reitoria de Pesquisa-USP.
SUPPORTING INFORMATION AVAILABLE
NMR analysis of 1a and 1b. HPLC-MS/MS detection of 2a
and 2b. HPLC-fluorescence data showing the pH effect on 2a
formation and the sensitivity of dansyl hydrazine and PBH
methods. Dansyl hydrazine derivatization method. This material
is available free of charge via the Internet at http://pubs.acs.org.
Received for review March 10, 2010. Accepted July 8,
2010.
AC1006427
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6781

Anal. Chem. 2010, 82, 6782–6789
Magnetic Nanoparticle Enhanced Surface Plasmon
Resonance Sensing and Its Application for the
Ultrasensitive Detection of Magnetic
Nanoparticle-Enriched Small Molecules
Jianlong Wang, Ahsan Munir, Zanzan Zhu, and H. Susan Zhou*
Department of Chemical Engineering, Worcester Polytechnic Institute, 100 Institute Road, Worcester,
Massachusetts 01609
Magnetic nanoparticles (MNPs) have been frequently
used in bioseparation, but their applicability in bioassays
is limited due to their extremely small size so that
sensitive detection is difficult to achieve using a general
technique. Here, we present an amplification technique
using MNPs for an enhanced surface plasmon resonance
(SPR) bioassay. The amplification effect of carboxyl group
modified Fe3O4 MNPs of two sizes on SPR spectroscopy
is first demonstrated by assembling MNPs on amino
group modified SPR gold substrate. To further evaluate
the feasibility of the use of Fe3O4 MNPs in enhancing
a SPR bioassay, a novel SPR sensor based on an
indirect competitive inhibition assay (ICIA) is developed
for detecting adenosine by employing Fe3O4
MNP-antiadenosine aptamer conjugates as the amplification
reagent. The results confirm that Fe3O4
MNPs can be used as a powerful amplification agent
to provide a sensitive approach to detect adenosine by
SPR within the range of 10-10 000 nM, which is
much superior to the detection result obtained by a
general SPR sensor. Importantly, the present detection
methodology could be easily extended to detect other
biomolecules of interest by changing the corresponding
aptamer in Fe3O4 MNP-aptamer conjugates. This
novel technique not only explores the possibility of the
use of SPR spectroscopy in a highly sensitive detection
of an MNP-based separation product but also offers a
new direction in the use of Fe3O4 MNPs as an amplification
agent to design high performance SPR biosensors.
Over the past few decades, magnetic nanoparticles (MNPs)
have been receiving increasing attention due to their unprecedented
advantages such as higher surface-to-volume ratio for
chemical binding, minimum disturbance to attached biomolecules,
faster binding rates, higher miscibility, and higher specificity. 1,2
These characteristics of MNPs render them easier labeling by
biomolecules, as well as easier binding with its target analytes.
* Corresponding author. Tel.: 508-831-5275. Fax: 508-831-5936. E-mail:
szhou@wpi.edu.
(1) Pankhurst, Q. A.; Connoliy, J.; Johns, S. K.; Dobson, J. J. Appl. Phys. 2003,
36, 167–181.
(2) Gijs, M. A. M. Microfluid. Nanofluid. 2004, 1, 22–40.
6782 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Up to now, all kinds of biomolecules including DNA and RNA, 3
protein and peptides, 4 cell and virus 5-7 have been separated and
concentrated under an external magnet by the use of MNPs as
carriers. Compared with the springing up of MNPs in bioseparation,
the application of MNPs in bioassays is limited because their
size is too small to be detected by general techniques developed
for detecting magnetic microbeads. In order to extend the
application of MNPs in bioassays, some novel techniques had been
used to detect MNPs, such as the electrochemical method, 8 IR
spectroscopy, 9 fluorescence spectroscopy, 10 magnetic atomic force
microscopy (AFM), 11 magnetic resonance imaging (MRI), 12
bio-bar-code, 13 etc. However, these methods either need labeling
MNPs by electroactive probes or fluorescence molecules or need
expensive experiment setups, thereby limiting them to be used
on a benchtop scale and cannot be used for simple, in situ, and
cost-effective detection of real samples.
Here, we investigate the application of surface plasmon
resonance (SPR) spectroscopy for fast, ultrasensitive, and in situ
detection of the MNP-enriched biomolecules. SPR being a surfacesensitive
characterization method not only can be used for
analyzing the kinetic data including the equilibrium constant and
the association and dissociation parameters between biomolecules
by simulating SPR kinetic curves but also can be used in situ to
detect the concentrations of biomolecules with high sensitivity
and selectivity. 14,15 The surface plasmon used in SPR spectroscopy
is highly sensitive to changes in the effective refractive index or
the thickness of the test medium in the vicinity of the metal
surface, especially for the molecules with high mass change.
(3) Obata, K.; Tajima, H.; Yohda, M.; Matsunaga, T. Pharmacogenomics 2002,
3, 697–708.
(4) Safarik, I.; Safarikova, M. BioMag. Res. Technol. 2004, 2, 7.
(5) Pamme, N. Lab Chip 2006, 6, 24–38.
(6) Zakhireh, J.; Gomez, R.; Esserman, L. Eur. J. Cancer 2008, 44, 2742–2752.
(7) Safarik, I.; Safarikova, M. J. Chromatogr., B 1999, 722, 33–53.
(8) Hsing, I. M.; Xu, Y.; Zhao, W. T. Electroanalysis 2007, 19, 755–768.
(9) Ravindranath, S. P.; Mauer, L.; DebRoy, C.; Irudayaraj, J. Anal. Chem. 2009,
81, 2840–2846.
(10) Song, Y. J.; Zhao, C.; Ren, J. S.; Qu, X. G. Chem. Commun. 2009, 1975–
1977.
(11) Arakaki, A.; Hideshima, S.; Nakagawa, T.; Niwa, D.; Tanaka, T.; Matsunaga,
T. Biotechnol. Bioeng. 2004, 88, 543–546.
(12) Perez, J. M.; Josephson, L.; O’Loughlin, T.; Hogemann, D.; Weissleder, R.
Nat. Biotechnol. 2002, 20, 816–820.
(13) Li, Y.; Hong Cu, Y. T.; Luo, D. Nat. Biotechnol. 2005, 23, 885–889.
(14) Li, X.; Wei, X. L.; Husson, S. M. Biomacromolecules 2004, 5, 869–876.
(15) Li, X.; Husson, S. M. Biosens. Bioelectron. 2006, 22, 336–348.
10.1021/ac100812c © 2010 American Chemical Society
Published on Web 07/20/2010

Numerous references had demonstrated nanoparticles (NPs)
could greatly enhance the sensitivity of SPR spectroscopy due to
the large molecular weight of nanoparticles in spite of the fact
that the size of nanoparticles is so small that it could not be
observed by other techniques. Several kinds of NPs including Au
NPs, 16-20 SiO2 NPs, 21 Pd NPs, 22 and Pt NPs 23 had been applied
to increase the SPR sensitivity for detecting all kinds of
biomolecules. However, the application of MNPs in the SPR
field is still limited. Considering the high refractive index and
the high molecular weight of MNPs, 24 it is possible to design
an excellent SPR biosensor using MNPs as an amplification
reagent. Once the amplifying effect of MNPs for a SPR signal
is demonstrated, it can then be proved that SPR will be a
powerful candidate for detecting MNP-based separation products.
There are very few works that have been done to study the
SPR response of MNPs, and most of them focus on utilizing
commercial strepavidin-conjugated MNPs for signal amplification.
25,26 Obviously, biotin needs to be attached on a SPR substrate
surface for the further binding of strepavidin-conjugated MNPs,
which limits the extensive application of MNPs in the SPR field.
To further understand the SPR response of MNPs and extend
the application of SPR in detecting MNP labeled biomolecules
and their separation product, in this work, we study the SPR
response of the carboxyl group modified Fe3O4 MNPs by
nonspecifically adsorbing the Fe3O4 MNPs on amino group
modified SPR gold substrate. The carboxyl groups on Fe3O4
MNPs allow the MNPs to be easily functionalized by all kinds
of biomolecules for extensive applications. Our results demonstrate
that the monolayer adsorption of Fe3O4 MNPs could
result in a big SPR angle shift with a low optical loss. On the
basis of the amplification effect of Fe3O4 MNPs, we further
demonstrate SPR spectroscopy can be used to sensitively detect
Fe3O4 MNP-enriched small molecules by an indirect competitive
inhibition assay (ICIA). In this case, Fe3O4 MNPs labeled
by antiadenosine aptamer are used both as the enrichment
reagent of adenosine and the amplification reagent of SPR
spectroscopy.
EXPERIMENTAL SECTION
Materials. Adenosine, uridine, cytidine, guanosine, ethanolamine,
6-mercaptohexan-1-ol (MCH), 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide
hydrochloride (EDC), N-hydroxysuccinimide
(NHS), thrombin, FeO(OH), oleic acid, 1-octadecene, acetone,
(16) Golub, E.; Pelossof, G.; Freeman, R.; Zhang, H.; Willner, I. Anal. Chem.
2009, 81, 9291–9298.
(17) Riskin, M.; Tel-Vered, R.; Lioubashevski, O.; Willner, I. J. Am. Chem. Soc.
2009, 131, 7368–7378.
(18) Lioubashevski, O.; Chegel, V. I.; Patolsky, F.; Katz, E.; Willner, I. J. Am.
Chem. Soc. 2004, 126, 7133–7143.
(19) Zayats, M.; Pogorelova, S. P.; Kharitonov, A. B.; Lioubashevski, O.; Katz,
E.; Willner, I. Chem.sEur. J. 2003, 9, 6108–6114.
(20) Wang, J. L.; Munir, A.; Zhou, H. S. Talanta 2009, 79, 72–76.
(21) Luckarift, H. R.; Balasubramanian, S.; Paliwal, S.; Johnson, G. R.; Simonian,
A. L. Colloids Surf., B 2007, 58, 28–33.
(22) Lin, K. Q.; Lu, Y. H.; Chen, J. X.; Zheng, R. S.; Wang, P.; Ming, H. Opt.
Express 2008, 16, 18599–18604.
(23) Beccati, D.; Halkes, K. M.; Batema, G. D.; Guillena, G.; de Souza, A. C.;
van Koten, G.; Kamerling, J. P. ChemBioChem 2005, 6, 1196–1203.
(24) Grigoriev, D.; Gorin, D.; Sukhorukov, G. B.; Yashchenok, A.; Maltseva, E.;
Moehwald, H. Langmuir 2007, 23, 12388–12396.
(25) Teramura, Y.; Arima, Y.; Iwata, H. Anal. Biochem. 2006, 357, 208–215.
(26) Soelberg, S. D.; Stevens, R. C.; Limaye, A. P.; Furlong, C. E. Anal. Chem.
2009, 81, 2357–2363.
chloroform, poly(maleic anhydride-alt-1-octadecene) (molecular
weight: 30 000-50 000) and 2-(2-aminoethoxy)-ethanol were purchased
from Sigma and used as received. Sodium hydrogen
phosphate heptahydrate, potassium dihydrogen phosphate, and
sodium chloride were ordered from Alfa Aesar. All DNA molecules
were obtained from Integrated DNA Technologies (IDT). The
sequence of the adenosine-binding aptamer was 5′-NH 2-C6-AGA
GAA CCT GGG GGA GTA TTG CGG AGG AAG GT-3′
(aptamer), the sequence of its partial complementary strand
was 5′-SH-C6-ACC TTC CTC CGC-3′ (ss-DNA). DNA solutions
were prepared by dissolving DNA in 50 mM, pH 8.0 Tris-HCl
buffer including 138 mM NaCl. Different concentrations of
adenosine and 1 mM uridine, cytidine, and guanosine were all
prepared in the Tris-HCl buffer. All glassware used in the
experiment was cleaned in a bath of freshly prepared 3:1 HCl/
HNO3 (aqua regia) and rinsed thoroughly in H2O prior to use.
(Caution: Aqua regia solution is dangerous and should be
handled with care.)
Synthesis of Monodisperse Fe3O4 MNPs. Monodisperse
Fe3O4 MNPs were synthesized by the pyrolysis of iron carboxylate
in the organic phase. 27 In brief, a mixture of FeO(OH),
oleic acid, and 1-octadecene was refluxed at 320 °C for1h
under a nitrogen atmosphere. During this process, the solution
changed its color from turbid black to black. The resulting
MNPs were precipitated with acetone and collected by centrifuge
at 4000g. After that, Fe3O4 MNPs were further purified
by repeated extraction of the precipitate with CHCl3/acetone
(1:10) until a powder of Fe3O4 MNPs was obtained. The powder
of Fe3O4 MNPs was stored at room temperature for further
application.
Forming Soluble Fe3O4 MNPs by Phase Transfer. Fe3O4
MNPs were transferred to a PBS solution according to Yu’s
work with minor modifications. 28 Carboxy group modified
amphiphilic polymers was first prepared by mixing poly(maleic
anhydride-alt-1-octadecene) with 2-(2-aminoethoxy)-ethanol (molar
ratio 1:120) in chloroform overnight. Then, the monodisperse
Fe3O4 MNPs (purified and dispersed in chloroform) were
dispersed in the carboxy group modified amphiphilic polymer
solution, and the mixture was stirred overnight at room
temperature (molar ratio of Fe3O4/polymer was 1:10). After
that, PBS buffer (pH 8.0, 10 mM) was added to the chloroform
solution of the complexes with at least a 1/1 volume ratio;
chloroform was then gradually removed by rotary evaporation
at 35 °C and water-soluble carboxy group modified Fe3O4
MNPs were obtained in a clear and dark purple solution. This
transfer process had a 100% efficiency, and no residue was
observed. The original concentrations of ∼14.51 and ∼32.82
nm soluble Fe3O4 MNPs analyzed by atomic absorption
spectroscopy are 205.4 and 16.2 nM, respectively, which will
be used to prepare other concentrations of Fe3O4 MNP
solutions by dilution.
Synthesis of Fe3O4 MNP-Aptamer Conjugates. The
monodisperse and soluble Fe3O4 MNPs with ∼32.82 nm were
diluted into pH 8.0 PBS buffer with a final concentration of 1.6
nM. Then, 1 mg of EDC and 1 mg of NHS were added to 5
(27) Yu, W. W.; Falkner, J. C.; Yavuz, C. T.; Colvin, V. L. Chem. Commun. 2004,
20, 2306–2307.
(28) Yu, W. W.; Chang, E.; Sayes, C. M.; Drezek, R.; Colvin, V. L. Nanotechnology
2006, 17, 4483–4487.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6783

mL of Fe3O4 MNP solution under stirring for 0.5 h to activate
the carboxyl group on the surface of Fe3O4 MNPs. After that,
100 µL of 37.7 µM amino-modified antiadenosine aptamer was
added in this solution, and they were allowed to react for 2 h
to immobilize aptamer on the surface of Fe3O4 MNPs. After
that, 1 M ethanolamine was added for 1 h to block the
unreacted carboxyl groups. Then, this solution was centrifuged
at 14 000g at room temperature for 25 min twice to remove
the free amino-aptamer. At last, the Fe3O4 MNPs were
dispersed in 5 mL of pH 8.0 PBS buffer and stored at 4 °C.
In Situ SPR Measurement. The SPR experiments were done
using Eco Chemie Autolab SPR systems (Brinkmann Instruments,
New York). 29,30 It works with a laser diode fixed at a wavelength
of 670 nm, using a vibrating mirror to modulate the angle of
incidence of the p-polarized light beam on the SPR substrate. The
instrument was equipped with a cuvette. A gold sensor disk (25
mm in diameter) was mounted on the hemicylindrical lens (with
index-matching oil) to form the base of the cuvette. The cuvette
could contain sample with adjustable volume from 10 to 1000 µL.
An O-ring (3 mm inner diameter) between the cuvette and disk
prevents leakage. An autosampler (Eco Chemie) with a controllable
aspirating-dispensing-mixing pipet was used to add
samples into the cuvette and provide constant mixture by aspiration
and dispensing during measurements. This experimental
arrangement maintains a homogeneous solution and reproducible
hydrodynamic conditions. The injection rate and mixing rate for
all samples were 10 and 40 µL/s, respectively, with the total
volume for all samples dispensed in the SPR cell equal to 40 µL.
This setup allows us to measure the SPR angle shift in millidegrees
(m°) as a response unit to quantify the binding amount of
macromolecules to the sensor surface.
Details of the experiment are as follows. The SPR gold film
was initially immersed into the thiolated ss-DNA solution for 12 h
in order to assemble the monolayer of ss-DNA. Then, the modified
gold film was thoroughly rinsed with 50 mM Tris-HCl buffer
and water to remove the weakly adsorbed ss-DNA. Then, the ss-
DNA modified SPR gold film was immersed in 100 µM 6-mercaptohexanol
for 1htoblock the uncovered gold surface. This gold
film was used as a sensing surface to detect the amount of aptamer
possessing ss-DNA structure on Fe 3O4 MNP-aptamer conjugates,
which can be adjusted by the concentration of adenosine
added to the Fe3O4 MNP-aptamer conjugate solution. The
detection procedure is made up of two steps. First, Fe3O4
MNP-aptamer conjugate solution was mixed with different
concentrations of adenosine for 30 min. After that, adenosine
bound with Fe3O4 MNP-aptamer conjugates were separated
and enriched by centrifuging at 14 000g for 1 h twice. The
precipitation was dispersed in PBS again. The resulting solution
was injected into the SPR cell, and the SPR angle-time curve
was recorded. In order to reduce the disturbance of DNA
denaturation that resulted from the regenerating process for
the detecting results, we change a new substrate after each
detection. The modification of each gold substrate is carried
out under the same experimental condition. To confirm the
reproducibility of the detection result, different concentrations
(29) Wang, J. L.; Zhou, H. S. Anal. Chem. 2008, 80, 7174–7178.
(30) Wang, J. L.; Munir, A.; Li, Z. H.; Zhou, H. S. Biosens. Bioelectron. 2009,
25, 124–129.
6784 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
of the MNPs and adenosine are repeatedly detected for three
times.
RESULTS AND DISCUSSION
Characterization of Fe3O4 MNPs. Although soluble Fe3O4
MNPs could be easily synthesized by coprecipitation of aqueous
Fe 2+ /Fe 3+ salt solutions with the addition of a base under
an inert atmosphere at room temperature or at elevated
temperature, the size, shape, and composition of the MNPs
very much depends on the type of salts used (e.g., chlorides,
sulfates, nitrates), the Fe 2+ /Fe 3+ ratio, the reaction temperature,
the pH value, and ionic strength of the media. 31-33 Furthermore,
the Fe3O4 MNPs are also easy to aggregate. Inspired by the
synthesis of high-quality semiconductor nanocrystals and
oxides in nonaqueous media by thermal decomposition, 34-36
monodisperse Fe3O4 MNPs with controlled size has essentially
been synthesized through thermal decomposition of ion
compounds in high-boiling organic solvents. 37,38 Here, we
synthesize Fe3O4 MNPs by the pyrolysis of iron carboxylate in
an organic phase. By changing the ratio of FeO(OH) and oleic
acid, two kinds of Fe3O4 MNPs are prepared. Figure 1A,B
provides the TEM images of the prepared Fe3O4 MNPs and their
size distribution. The average size of Fe3O4 MNPs derived from
Figure 1A,B are 14.51 nm (n ) 300 particles) and 32.82 nm
(n ) 138 particles), respectively. Importantly, both kinds of Fe3O4
MNP size distributions are narrow, which indicates the
prepared Fe3O4 MNPs are monodisperse. After transferring
Fe3O4 MNPs from organic reagent to water solution by the
amphiphilic polymer, Fe3O4 MNPs show a good stability in both
PBS and Tris buffer due to the large hydrodynamic size of
polymer (data not shown), which are all beneficial for the
acquisition of accurate and repeated SPR analytical results.
SPR Response and Concentration Dependence of Fe3O4
MNPs. The basis of a particle-enhanced bioassay is that biomolecular
interaction events lead to particle immobilization, i.e., more
immobilized proteins yield higher particle coverage. 39 Currently,
most SPR instruments are able to quantitatively detect the
concentration of biomolecules through calculating the SPR angle
shift enhanced by the binding of nanoparticle. Considering our
synthesized Fe3O4 MNPs are protected by negative polymer,
2-mercaptoethyamine is used on SPR Au substrates for the
adsorption of Fe3O4 MNPs. The thiol group binds to the Au
surface, leaving the amine group free to bind with carboxyl
group in soluble Fe3O4 MNPs by electrostatic interaction.
(31) Lu, A. H.; Salabas, E. L.; Schuth, F. Angew. Chem., Int. Ed. 2007, 46, 1222–
1244.
(32) Kang, Y. S.; Risbud, S.; Rabolt, J. F.; Stroeve, P. Chem. Mater. 1996, 8,
2209–2211.
(33) Massart, R. IEEE Trans. Magn. 1981, 17, 1247–1248.
(34) Murray, C. B.; Norris, D. J.; Bawendi, M. G. J. Am. Chem. Soc. 1993, 115,
8706–8715.
(35) Peng, X.; Wickham, J.; Alivisatos, A. P. J. Am. Chem. Soc. 1998, 120, 5343–
5344.
(36) O’Brien, S.; Brus, L.; Murray, C. B. J. Am. Chem. Soc. 2001, 123, 12085–
12086.
(37) Sun, S.; Zeng, H.; Robinson, D. B.; Raoux, S.; Rice, P. M.; Wang, S. X.; Li,
G. J. Am. Chem. Soc. 2004, 126, 273–279.
(38) Redl, F. X.; Black, C. T.; Papaefthymiou, G. C.; Sandstrom, R. L.; Yin, M.;
Zeng, H.; Murray, C. B.; O’Brien, S. P. J. Am. Chem. Soc. 2004, 126, 14583–
14599.
(39) Lyon, L. A.; Pena, D. J.; Natan, M. J. J. Phys. Chem. B 1999, 103, 5826–
5831.

Figure 1. TEM images of the prepared Fe3O4 MNPs: (A) 14.51 nm and (B) 32.82 nm.
Figure 2. Variation of SPR angle-time curves with the concentration of Fe3O4 MNPs: (A) 14.51 nm; (B) 32.82 nm. (C) Comparison of the
variation of SPR angle shift with the concentration of Fe3O4 MNPs of 14.51 and 32.82 nm, respectively.
Figure 2 illustrates the changes that occur in the SPR angle shift
as a function of the concentration for 14.51 nm (Figure 2A) and
32.82 nm (Figure 2B) Fe3O4 MNPs. In the case of 14.51 nm
Fe3O4 MNPs (Figure 2A), the SPR angle shifts gradually increase
from 6.4 to 1111.06 m° with the increase of Fe3O4 MNP
concentrations from 0.016 to 1.6 nM. It is evident that a higher
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6785

Figure 3. AFM images of 14.51 nm Fe3O4 MNP (A) and 32.82 nm Fe3O4 MNP (B) modified SPR gold substrate. The concentration of Fe3O4
MNPs is 1.6 nM, and the assembly time is 10 min.
coverage of Fe3O4 MNPs on SPR gold film could be reached
at a higher concentration due to the rapid diffusion adsorption.
It should be pointed out that the SPR angle shift resulting from
the adsorption of Fe3O4 MNPs is much higher than that of the
value resulting from the adsorption of most biomolecules under
the same concentration. 40,41 It means Fe3O4 MNPs greatly
enhance the signal of SPR spectroscopy. Importantly, the SPR
angle shift could be further increased when 32.82 nm Fe3O4
MNPs are used. Figure 2B shows the SPR angle shift curves
resulting from the adsorption of 32.82 nm Fe3O4 MNPs with the
same concentration sequence as 14.51 nm Fe3O4 MNPs. With
the increase of Fe3O4 MNP concentrations from 0.016 to 1.6
nM, the SPR angle shifts increase from 79.22 to 2479.79 m°.
To further understand the size effect of Fe3O4 MNPs on the
SPR response, the relation between SPR angle shifts and the
concentrations of Fe3O4 MNPs with the two sizes are compared
in Figure 2C. Obviously, much larger angle shifts are observed
for 32.82 nm Fe3O4 MNPs because of its larger moleculer
weight. Besides that, the affinity constants and surface coverage
for both kinds of Fe3O4 MNPs could also be obtained from
Figure 2C. From the data of Figure 2C, we can see that the
plasmon resonance shifts with the increasing Fe3O4 MNP
concentration and a simple Langmuir isotherm 42 could be used
to fit the data. The Langmuir equation used to fit the data is
given by
KA [C] Fe3O4 ∆λ ) ∆λmax1 + KA [C] Fe3O4 Where ∆λ is the angle shift caused by the adsorption of Fe3O4
MNPs, ∆λmax is the angle shift which will be observed at
saturation, KA is the apparent equilibrium affinity constant, and
[C]Fe3O4 is the concentration of Fe3O4 MNPs. Using the above
equation, KA values for 14.51 and 32.82 nm Fe3O4 MNPs are
(40) Mullett, W. M.; Lai, E. P. C.; Yeung, J. M. Methods 2000, 22, 77–91.
(41) Hoa, X. D.; Kirk, A. G.; Tabrizian, M. Biosens. Bioelectron. 2007, 23, 151–
160.
(42) Lee, H. J.; Wark, A. W.; Corn, R. M. Langmuir 2006, 22, 5241–5250.
6786 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(1)
calculated to be around 7.14 × 10 8 and 2.63 × 10 9 M -1 ,
respectively, which are similar to the value of NPs as found in
literature. 43 The saturation responses ∆λmax of 14.51 and 32.82
nm are calculated to be 2165 and 3134 0 m, respectively. It can
be seen that 32.82 nm Fe3O4 MNPs have slightly higher affinity
for adsorption on the sensing surface due to their larger
molecular weight; therefore, they show larger angle shifts as
well as higher saturation response (∆λmax) compared to 14.51
nm Fe3O4 MNPs. In our following experiments, we, therefore,
use 32.82 nm Fe3O4 MNPs. The ratio (∆λ)/(∆λmax) gives the
fraction of surface coverage, and if the bulk concentration of
[C]Fe3O4 is equal to (1)/(KA), half of the surface sites will be
occupied ((∆λ)/(∆λmax) ) 0.5). On the basis of the experimental
results and the fitting parameters, we calculate that 80%
of the coverage ((∆λ)/(∆λmax) ) 0.8) will be obtained when
32.82 nm Fe3O4 MNPs with a concentration of 1.6 nM are used.
This concentration will be used in our following experiments
because the changes in SPR angle shift (∆λ) will be very small
even if we further increase the concentration of Fe3O4 MNPs.
To demonstrate that a dense monolayer of MNPs could be
formed on a SPR substrate within 10 min using a 1.6 nM Fe3O4
MNP solution as assembly solution, AFM is used to evaluate
the surface morphology of MNP modified SPR gold substrate.
Figure 3A,B shows the AFM images of 14.51 nm Fe3O4 MNPs
and 32.82 nm Fe3O4 MNPs on SPR gold substrate, respectively.
As seen from the AFM figures, the nanoislands are formed
after MNPs are assembled. The diameter of nanoislands in
Figure 3A is smaller than that of the value in Figure 3B due to
different diameter of MNPs. However, the dense layer of MNPs
on SPR gold substrate has been observed from Figure 3 for both
kinds of Fe3O4 MNPs at the present experimental condition,
which indicates the adsorption of Fe3O4 MNPs on SPR gold
substrate is fast and the present experimental condition is
suitable for further experiments.
(43) Liao, W. S.; Chen, X.; Yang, T. L.; Castellana, E. T.; Chen, J. X.; Cremer,
P. S. Biointerphases 2009, 4, 80–85.

Figure 4. Schematic representation of the SPR biosensor for the detection of the small molecules.
Fe3O4 MNP Enhanced SPR Sensing for the Detection
of Small Molecules. To further demonstrate the practicability
of the amplification effect of Fe3O4 MNPs in an enhancing SPRbased
bioassay, a novel SPR sensor based on indirect competitive
inhibition assay (ICIA) for the detection of adenosine is
constructed. The principle of this SPR sensor is shown in
Figure 4. The partial complementary thiolated ss-DNA of antiadenosine
aptamer is first immobilized on SPR gold film as a
sensing surface. When Fe3O4 MNP-antiadenosine aptamer
conjugate solution is added to the SPR cell in the absence of
adenosine, Fe3O4 MNP-antiadenosine aptamer conjugates will
be adsorbed to the SPR sensor by the DNA hybridization
reaction and result in a huge change of SPR signal due to the
amplification effect of Fe3O4 MNPs. However, the change of
SPR signal will decrease after Fe3O4 MNP-antiadenosine
aptamer conjugates bind with adenosine. This is because
adenosine reacts with antiadenosine aptamer in Fe3O4 MNPantiadenosine
aptamer conjugates and changes its structure
from ss-DNA to tertiary structure, which cannot hybridize with
its partial complementary ss-DNA immobilized on the SPR gold
surface. Thus, the change of SPR signal will decrease with the
increase of the number of Fe3O4 MNP-antiadenosine aptamer
conjugates possessing tertiary structure, which is proportional
to the concentration of adenosine.
The essential prerequisite of this detection is to prepare Fe3O4
MNP-antiadenosine aptamer conjugates. For this purpose,
EDC and NHS are used as a carboxyl activating agent for the
coupling of primary amines in aptamer and carboxyl groups
in polymer coated MNPs. To demonstrate that MNPs have
been labeled by an aptamer successfully, FT-IR spectroscopy
is used to evaluate the surface modification of MNPs before
and after aptamer labeling. Figure 5 shows the IR absorbance
spectra of Fe3O4 MNPs before and after the aptamer label. For
the polymer coated Fe3O4 MNPs, the absorbance peak near
1710 cm -1 is assigned to υ (CdO) (stretch vibration of
carbonyl) and amide I, peaks at 1550 cm -1 are assigned to
amide II, and the peaks at 2854 and 2925 cm -1 are assigned to
υs (C-H2) and υas (C-H2) (symmetric and asymmetric stretch
vibrations of C-H2), respectively. In the region of 3200-3570
Figure 5. FT-IR of polymer coated Fe3O4 MNPs before and after
aptamer labeling.
cm -1 , a peak due to the O-H stretch is expected. Several new
peaks are observed after MNPs are labeled by the aptamer.
The peaks at 1048 and 1211 cm -1 are attributed to stretching
vibrations of the PO 2- in the aptamer. It should be pointed out
that we are not able to confirm the presence of DNA-related
peaks at 1550 cm -1 because the peak overlaps with amide II.
After Fe3O4 MNP-antiadenosine aptamer conjugates are
prepared successfully, the SPR detection is carried out
according to the principle described in Figure 4. Figure 6A
shows the SPR angle-time curves of the separation products
obtained after Fe3O4 MNP-antiadenosine aptamer conjugates
are reacted with different concentrations of adenosine for
30 min. In the absence of adenosine, Fe3O4MNP-antiadenosine
aptamer conjugates in solution directly hybridize with ss-
DNA immobilized on SPR gold film, resulting in the largest
SPR angle shift (∼1082.94 m°). This angle shift is much
larger than that of the angle shift resulting from the binding
of ss-DNA or most of the protein. The SPR angle shift
resulting from the binding of Fe3O4 MNP-antiadenosine
aptamer conjugates decreases (∼820.14 m°) after Fe3O4
MNP-antiadenosine aptamer conjugates are reacted with
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6787

Figure 6. (A) SPR angle-time curves of the separation products obtained after Fe3O4 MNP-antiadenosine aptamer conjugates are reacted
with different concentrations of adenosine for 30 min. Inset: The linear relationship between the logarithms of adenosine concentrations and the
SPR angle shift resulting from the binding of Fe3O4 MNP-antiadenosine aptamer conjugates. (B) SPR angle-time curves of 1 µM antiadenosine
aptamer without adenosine (red line) and after react with 1 mM adenosine for 30 min (black line).
10 nM adenosine because parts of aptamers on Fe3O4 MNPs
react with adenosine and form its tertiary structure, which
cannot hybridize with its complementary ss-DNA. With the
further increase of the concentration of adenosine added to
the Fe3O4 MNP-antiadenosine aptamer conjugate solution,
the SPR angle shift continuously decreases until most of the
antiadenosine aptamer binds with adenosine. By analyzing
the change of SPR angle shift with the concentrations of
adenosine, a good linear relationship between the logarithms
of adenosine concentrations and the SPR angle shift is
obtained with a range of 10-1 × 10 4 nM (shown in the insert
of Figure 6A). This detection result is comparable with most
other aptasensors 44 and is a little lower than that of detection
results from SPR aptasensor which utilize Au NPs as an
amplification reagent 20,29 but much superior to the detection
results obtained by a general SPR sensor based on a molecularly
imprinted technique. 45,46 It should be pointed out that the
SPR angle shift only decreases to ∼107.33 m° even when the
aptamer reacts with 1 mM adenosine. This nonspecific adsorption
may come from the stereohindrance effect of aptamer
possessing different structures. After adenosine is added to
Fe3O4 MNP-aptamer conjugates, most free-coiled aptamers
react with adenosine and form its tertiary structure. However,
the formation of a tertiary structure of an aptamer may
inhibit the further reaction between its adjacent free-coiled
aptamer and adenosine. To clearly show the amplification
effect of Fe3O4 MNPs, as a comparison, the SPR response
resulting from the binding of aptamer without MNP is also
studied. The black line in Figure 6B is the SPR angle shift
resulting from the binding of 1 µM antiadenosine aptamer after
antiadenosine aptamer reactes with 1 mM adenosine for 0.5 h.
Only a 24.6 m° SPR angle shift is observed. Although the SPR
angle shift resulting from the binding of antiadenosine aptamer
could increase to 65.7 m° (Red line in Figure 6B) in the absence
of adenosine, this value is still much lower than that of the
(44) Wang, Y. L.; Wei, H.; Li, B. L.; Ren, W.; Guo, S. J.; Dong, S. J.; Wang, E. K.
Chem. Commun. 2007, 5220–5222.
(45) Taniwaki, K.; Hyakutake, A.; Aoki, T.; Yoshikawa, M.; Guiver, M. D.;
Robertson, G. P. Anal. Chim. Acta 2003, 489, 191–198.
(46) Yoshikawa, M.; Guiver, M. D.; Robertson, G. P. J. Mol. Struct. 2005, 739,
41–46.
6788 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 7. SPR angle-time curves of the separation products
obtained after Fe3O4 MNP-antiadenosine aptamer conjugates are
reacted with different analytes for 30 min.
value resulting from the binding of Fe3O4 MNP-antiadenosine
aptamer conjugates (1082.94 m°). Besides sensitivity, the
specification of aptamer promises the selectivity of the
present SPR sensor for adenosine. Figure 7 exhibits SPR
angle-time curves of the separation products obtained after
Fe3O4 MNP-antiadenosine aptamer conjugates are reacted
with different analytes for 30 min. It could be easily seen
that the addition of Fe3O4 MNP-antiadenosine aptamer
conjugates which are reacted with 1 × 10 6 nM cytidine,
guanosine, and uridine result in big SPR angle shifts.
However, The SPR angle shift only increases a little after
Fe3O4 MNP-antiadenosine aptamer conjugates are reacted
with 1 × 10 6 nM adenosine. These results not only demonstrate
that Fe3O4 MNPs could greatly enhance the SPR signal
but also, more importantly, give us an important indication
that SPR spectroscopy could be an excellent candidate for
detecting an MNP-based separation product.
CONCLUSION
In summary, the SPR response of the carboxyl group
modified Fe3O4 MNPs of two different sizes onto an amino
group modified SPR gold substrate has been studied. The
results show the monolayer adsorption of Fe3O4 MNPs could

esult in a big SPR signal change with a low optical loss. To
evaluate the practicability of the use of Fe3O4 MNPs in
enhancing the SPR signal for biosensing, a novel SPR sensor
based on ICIA for the detection of adenosine is constructed
using Fe3O4 MNP-antiadenosine aptamer conjugates as the
amplification reagents. The experimental results demonstrate
that the SPR sensor possesses a good sensitivity and a high
selectivity for adenosine. Importantly, by changing the kind
of biomolecules labeled by MNPs, the present detection
method will be able to explore new applications of SPR
spectroscopy for the detection of a large variety of MNPbased
separation products. At the same time, the technique
demonstrated in this work could also offer a new direction
in designing high performance SPR biosensors for sensitive
and selective detection of small molecules by the amplification
effect of MNPs.
ACKNOWLEDGMENT
This work is supported by National Science Foundation (EEC-
0823974). Authors thank Prof. Camesano Terri and her graduate
student Sena Ada for assistance with AFM characterization.
Received for review March 10, 2010. Accepted July 8,
2010.
AC100812C
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6789

Anal. Chem. 2010, 82, 6790–6796
Improved Sensitivity Mass Spectrometric
Detection of Eicosanoids by Charge Reversal
Derivatization
James G. Bollinger, † Wallace Thompson, † Ying Lai, ‡ Rob C. Oslund, † Teal S. Hallstrand, ‡
Martin Sadilek, † Frantisek Turecek, † and Michael H. Gelb* ,†,§
Departments of Chemistry, Medicine, and Biochemistry, University of Washington, Seattle, Washington 98195
Combined liquid chromatography-electrospray ionization-tandem
mass spectrometry (LC-ESI-MS/MS) is a
powerful method for the analysis of oxygenated metabolites
of polyunsaturated fatty acids including eicosanoids.
Here we describe the synthesis of a new derivatization
reagent N-(4-aminomethylphenyl)pyridinium (AMPP) that
can be coupled to eicosanoids via an amide linkage in
quantitative yield. Conversion of the carboxylic acid of
eicosanoids to a cationic AMPP amide improves sensitivity
of detection by 10- to 20-fold compared to negative
mode electrospray ionization detection of underivatized
analytes. This charge reversal derivatization allows detection
of cations rather than anions in the electrospray
ionization mass spectrometer, which enhances sensitivity.
Another factor is that AMPP amides undergo considerable
collision-induced dissociation in the analyte portion rather
than exclusively in the cationic tag portion, which allows
isobaric derivatives to be distinguished by tandem mass
spectrometry, and this further enhances sensitivity and
specificity. This simple derivatization method allows prostaglandins,
thromboxane B2, leukotriene B4, hydroxyeicosatetraenoic
acid isomers, and arachidonic acid to
be quantified in complex biological samples with limits
of quantification in the 200-900 fg range. One can
anticipate that the AMPP derivatization method can be
extended to other carboxylic acid analytes for enhanced
sensitivity detection.
Liquid chromatography coupled to electrospray ionization
tandem mass spectrometry (LC-ESI-MS/MS) has emerged as a
powerful method to detect oxygenated derivatives of fatty acids
including eicosanoids (for example, refs 1-3). With these methods
it is possible to analyze a large collection of eicosanoids in a single
LC-ESI-MS/MS run. These lipid mediators are detected by single
* To whom correspondence should be addressed. Michael H. Gelb, Depts.
of Chemistry and Biochemistry, Campus Box 35100, University of Washington,
Seattle, WA 98195. Phone: 206 525-8405. Fax: 206 685-8665. E-mail:
gelb@chem.washington.edu.
† Department of Chemistry.
‡ Department of Medicine.
§ Department of Biochemistry.
(1) Dickinson, J. S.; Murphy, R. C. J. Am. Soc. Mass Spectrom. 2002, 13, 1227.
(2) Kita, Y.; Takahashi, T.; Uozumi, N.; Nallan, L.; Gelb, M. H.; Shimizu, T.
Biochem. Biophys. Res. Commun. 2005, 330, 898.
(3) Buczynski, M. W.; Stephens, D. L.; Bowers-Gentry, R. C.; Grkovich, A.;
Deems, R. A.; Dennis, E. A. J. Biol. Chem. 2007, 282, 22834.
6790 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
reaction monitoring (SRM) in which precursor ions are isolated
in the first stage of the mass spectrometer followed by collisioninduced
dissociation to give fragment ions, which are detected
after an additional stage of mass spectrometer isolation. The
current limit of quantification for these analytes is in the ∼10-20
pg range. This sensitivity level is appropriate for studies with
cultured cells in vitro or with relatively large tissue samples, but
it is not sufficient for studies with smaller volume samples such
as joint synovial fluid or bronchoalveolar lavage fluid from
experimental rodents. Given the importance of oxygenated fatty
acid derivatives in numerous medically important processes such
as inflammation and resolution of inflammation, we sought to
improve the LC-ESI-MS/MS sensitivity of detection of these lipid
mediators using a widely available analytical platform.
For reasons that are not well understood, cations generally
form gaseous ions better than anions in the electrospray ionization
source of the mass spectrometer. Additionally, for underivatized
carboxylic acids it is required to add a weak organic acid to the
chromatographic mobile phase, i.e., formic acid, so that the
carboxylic acid is kept in its protonated state, which allows it to
be retained on the reverse-phase column to ensure chromatographic
separation. However, the presence of the weak acid offsets
the formation of carboxylate anions in the electrospray source
because the weak acid carries most of the anionic charge in the
electrospray droplets, and thus formation of analyte anions is
suppressed. We reasoned that conversion of the carboxylic acid
to a fixed-charge cationic derivative would lead to improved
detection sensitivity by ESI-MS/MS. Charge-reversal derivatization
of carboxylic acids with quaternary amines has been explored in
previous work (for example, refs 4-6). However, these reagents
utilize organic cations that tend to fragment by collision-induced
dissociation near the cationic site. Fragmentation in the derivatization
tag is not desirable because analytes that form isobaric
precursor ions and that comigrate on the LC column will not be
distinguished in the mass spectrometer if they give rise to the
same detected fragment ion. This loss of analytical specificity is
a serious problem when analyzing complex biological samples.
Fragmentation in the analyte portion rather than in the tag portion
(4) Lamos, S. M.; Shortreed, M. R.; Frey, B. L.; Belshaw, P. J.; Smith, L. M.
Anal. Chem. 2007, 79, 5143.
(5) Yang, W.; Adamec, J.; Regnier, F. E. Anal. Chem. 2007, 79, 5150.
(6) Pettinella, C.; Lee, S. H.; Cipollone, F.; Blair, I. A. J. Chromatogr., B: Anal.
Technol. Biomed. Life Sci. 2007, 850, 168.
10.1021/ac100720p © 2010 American Chemical Society
Published on Web 07/22/2010

also reduces chemical noise, which also enhances sensitivity of
detection.
In this study we report the design and synthesis of a new
cationic tag and show that it can be quantitatively attached via an
amide linkage to the carboxyl group of eicosanoids by a simple
derivatization procedure. We then show that the derivatized
eicosanoids can be analyzed by LC-ESI-MS/MS with limits of
quantification that are well below those reported for underivatized
eicosanoids.
EXPERIMENTAL METHODS
Synthesis of AMPP. Pyridine (40 mmol, 3.2 mL) was
dissolved in 46 mL of absolute ethanol followed by the addition
of 1-chloro-2,4-dinitrobenzene (40 mmol, 8.2 g, Aldrich). The
mixture was heated with a reflux condenser at 98 °C for 16 h
under nitrogen. After cooling, ethanol was removed by rotary
evaporation, and the crude product was recrystallized by dissolving
in a minimal amount of hot ethanol and allowing the solution to
slowly cool. The product N-2,4 dinitrophenyl pyridinium chloride
was isolated as a yellow solid in 62% yield, and its identity was
confirmed by melting point analysis (189-191 °C observed,
189-190 °C reported 7 ). N-2,4-Dinitrophenyl pyridinium chloride
(16.8 mmol, 4.76 g) was dissolved in 70 mL of ethanol-pyridine
(3:1). 4-[(N-Boc) -amino-methyl] aniline (33.6 mmol, 7.56 g,
Aldrich) was added, and the reaction mixture was heated under
a reflux condenser at 98 °C under nitrogen for 3 h. After cooling,
700 mL of water was added to precipitate 2,4-dinitroaniline. After
filtration, the filtrate was concentrated to dryness by rotary
evaporation, and the product was isolated as a brown oil. This oil
was treated with 112 mL of 25% (v/v) trifluoroacetic acid in
dichloromethane for 30 min at room temperature. The mixture
was concentrated by rotary evaporation, and the solid was
triturated twice with benzene to remove excess trifluoroacetic acid.
The mixture was again concentrated by rotary evaporation. The
residue was dissolved in a minimal amount of heated ethanol, the
solution was allowed to cool for ∼5 min, and then diethyl ether
was added with swirling until the solution started to cloud up.
The mixture was transferred to the freezer (-20 °C) and left
overnight. The mixture was allowed to warm to room temperature
and then decanted. The obtained mother liquor was treated with
additional diethyl ether as above to give additional AMPP solid.
The solids were combined and triturated with diethyl ether. The
solid was dried under vacuum to give 3.30 g of AMPP as a brown
solid, 59% yield. 1 H NMR (300 MHz, D6-DMSO) 9.34 (d, 2H),
8.81 (t, 1H), 8.53 (broad, 3H), 8.33 (t, 2H), 7.95 (d, 2H), 7.80
(d,2H), 4.22 (s, 2H) ( 1 H NMR spectra are shown in the
Supporting Information, estimated purity of AMPP is >95%).
Preparation of Eicosanoid Stock Solutions. The following
eicosanoid standards from Cayman Chemicals were used (PGE2,
PGD2, PGF2R, 6-keto-PGF1R, TxB2, 5(S)-HETE, 8(S)-HETE,
11(S)-HETE, 12(S)-HETE, 15(S)-HETE, LTB4, arachidonic acid,
D4-PGE2, D4-PGD2, D4-PGF2R, D4-6-keto-PGF1R, D4-TXB2, D8-
5(S)-HETE, D4-LTB4, D8-arachidonic acid). Stock solutions of
eicosanoids were prepared at a concentration of 100 pg/µL in
absolute ethanol and stored at -80 °C under Ar in Teflon
septum, screw cap vials. Serial dilutions of the stock solutions
were made in absolute ethanol for standard curve and extrac-
(7) Park, K. K.; Lee, J.; Han, D. Bull. Korean Chem. Soc. 1985, 6, 141.
tion analysis. Internal standards were diluted to a working stock
of 5 pg/µL in absolute ethanol.
Preparation of Samples Prior to Derivatization with
AMPP. Standard Curves. Each sample contained 50 pg of each
internal standard and various amounts of nonisotopic eicosanoids
(added from the stock solutions described above) transferred to
a glass autosampler vial insert (Agilent catalog no. 5183-2085).
Solvent was removed with a stream of nitrogen, and the residue
was derivatized with AMPP as described below.
Mouse Serum. Analysis of endogenous eicosanoids in serum
was carried out with commercial mouse serum (Atlantic Biologicals
catalog no. S18110). A volume of 1, 5, or 10 µL of serum was
placed in a glass autosampler vial insert. Two volumes of methanol
(LC/MS, JT Baker catalog no. 9863-01) containing 50 pg of each
internal standard were added. The vial insert was mixed on a
vortex mixer for ∼10 s. The concentration of methanol was
lowered to 10% (v/v) by addition of purified water (Milli-Q,
Millipore Corp.), and the samples were loaded via a glass Pasteur
pipet onto a solid phase extraction cartridge (10 mg Oasis-HLB,
Waters catalog no. 186000383). The cartridges were previously
washed with 1 mL of methanol and then 2 × 0.75 mL of 95:5
water-methanol. After sample loading, the sample tube was rinsed
with 200 µL of purified water-methanol (95:5, v/v), and this was
added to the cartridge. The cartridge was washed with 2 × 1mL
of water-methanol (95:5, v/v). Additional solvent was forced out
of the cartridge solid phase by applying medium pressure N2
(house N2 passed through a 0.2 µm cartridge filter) for a few
seconds. Column eluant receiver vials (Waters Total Recovery
autosampler vials, Waters catalog no. 186002805) were placed
under the cartridges. The cartridges were then eluted with
methanol (1 mL). All cartridge steps were carried out using a
vacuum manifold (Waters catalog no. WAT200606) attached
to a water aspirator. Solvent was removed by placing the
receiver vials in a centrifugal evaporator (Speed-Vac). These
processed samples were derivatized with AMPP (see below)
without storage.
Lung Epithelial Cells. The University of Washington Institutional
Review Board approved the studies involving human
subjects, and written informed consent was obtained from all
participants. Primary bronchial epithelial cells were isolated from
a volunteer with asthma during a bronchoscopy using a nylon
cytology brush of cells from subsegmental airways. To establish
primary culture, the epithelial cells were seeded into a culture
vessel coated with type 1 collagen in bronchial epithelial basal
media (BEBM, Lonza, Allendale, NJ) supplemented with bovine
pituitary extract, insulin, hydrocortisone, gentamicin, amphotericin
B, fluconazole, retinoic acid, transferrin, triiodothyronine, epinephrine,
and human recombinant epidermal growth factor
(serum-free BEGM) and maintained at 37 °C in a humidified
incubator. After expansion in vitro, passage 2 epithelial cells were
grown to >90% confluence on a 12-well plate. The medium was
changed to 200 µL of Hanks balanced salt solution, and the cells
were treated with either calcium ionophore (A23187, 10 µM in
DMSO) or a DMSO-containing control solution for 20 min at 37
°C. The synthesis of eicosanoids was stopped by the addition of
4 volumes of ice-cold methanol with 0.2% formic acid, and samples
were stored at -80 °C until processed. The number of epithelial
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6791

cells was 5.0 × 10 5 cells per well. The studies were performed
in duplicate.
Frozen samples were thawed on ice, 50 pg of each internal
standard was added, and Milli-Q water was added to bring the
methanol to 10% (v/v). The samples were processed by solidphase
extraction (Oasis-HLB) as described for mouse serum
samples.
Rat 3Y1 Cells. 3Y1 cells were a gift from Dr. H. Kuwata
(Department of Health Chemistry, School of Pharmaceutical
Sciences, Showa University). 3Y1 cells were maintained in
complete medium consisting of Dulbecco’s Modified Eagle
Medium (DMEM) with low glucose (Invitrogen catalog no. 11885-
084) containing 10% heat inactivated, fetal bovine serum (FBS)
(Invitrogen catalog no. 16140), 100 units/mL penicillin, and 100
µg/mL streptomycin (Invitrogen catalog no. 15140) in plastic
tissue culture dishes (Nunc catalog no. 172958) in a humidified
atmosphere of 5% CO 2 at 37 °C. Passage of cells was performed
using 0.25% trypsin/EDTA (Invitrogen catalog no. 25200-056).
For PGE2 analysis, 3Y1 cells were plated at 5 × 10 4 cells/
well in a 24-well plate (Nunc catalog no. 142475) in 1 mL of
complete medium and incubated overnight. The medium was
then replaced with DMEM containing 2% FBS. After 24 h
incubation, the medium was removed from each well and
replaced with 1 mL DMEM containing 2% FBS and the
cytosolic phospholipase A2-R inhibitor Wyeth-2 (compound 10
from ref 9) or DMSO vehicle control (final concentration in
each well did not exceed 0.1% (v/v)). Cells were incubated for
20 min at room temperature. Mouse interleukin-1� (1 ng/well)
and human tumor necrosis factor-R (1 ng/well) (R & D Systems
catalog no. 401-ML and 210-TA) were both added to each well
in 10 µL of DMEM containing 2% FBS (negative control wells
received only 10 µL of DMEM containing 2% FBS). Cells were
incubated for 48 h at 37 °C in a humidified atmosphere of 5%
CO 2. The supernatant was then carefully removed from each
well and transferred into a glass test tube and placed on ice
for immediate analysis. PGE2 levels were measured by an
enzyme immunoassay kit (Cayman Chemical catalog no.
500141) according to the manufacturer’s instructions using 50
µL of medium above the cells. A sample of medium (50 µL)
was also analyzed by LC-ESI-MS/MS as follows. To the
medium was added 2 volumes of methanol containing 50 pg
of each internal standard, methanol was reduced to 10% v/v
by addition of Milli-Q water, and samples were submitted to
solid-phase extraction using Oasis-HLB cartridges as described
for mouse serum samples.
Derivatization with AMPP. To the residue in the Waters
Total Recovery autosampler vial was added 10 µL of ice-cold
acetonitrile (JT Baker catalog no. 9017-03)-N,N-dimethylformamide
(Sigma catalog no. 227056) (4:1, v:v). Then 10 µL of icecold
640 mM (3-(dimethylamino)propyl)ethyl carbodiimide hydrochloride
(TCI America catalog no. D1601) in purified water
was added. The vial was briefly mixed on a vortex mixer and
placed on ice while other samples were processed as above. To
each vial was added 20 µL of 5 mM N-hydroxybenzotriazole
(8) Kita, Y.; Takahashi, T.; Uozumi, N.; Shimizu, T. Anal. Biochem. 2005, 342,
134.
(9) McKew, J. C.; Lee, K. L.; Chen, L.; Vargas, R.; Clark, J. D.; Williams, C.;
Clerin, V.; Marusic, S; Pong, K. Inhibitors of Cytosolic Phospholipase A2.
U.S. Patent 7,557,135 B2, July 7, 2009.
6792 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(Pierce catalog no. 24460)-15 mM AMPP in acetonitrile. The vials
were mixed briefly on a vortex mixer, capped, and placed in a 60
°C incubator for 30 min. The cap was replaced with a split-septum
screw cap (Agilent catalog no. 5185-5824) for autoinjection onto
the LC-ESI-MS/MS. Samples were analyzed on the same day.
Samples were kept in the autosampler rack at 4 °C while queued
for injection.
LC-ESI-MS/MS Analysis. Some studies were carried out
with a Waters Quattro Micro triple quadrupole mass spectrometer,
a 2795 Alliance HT LC/autosampler system, and the QuanLynx
software package. Chromatography was carried out with a C18
reverse-phase column (Ascentis Express C18, 2.1 mm × 150 mm,
2.7 µm, Supelco catalog no. 53825-U). Solvent A is 95% H 2O/5%
CH3CN/1% acetic acid, and solvent B is CH3CN/1% acetic acid.
The solvent program is (linear gradients) 0-1.0 min, 95-78%
A; 1.0-7.0 min, 78-74% A; 7.0-7.1 min, 74-55% A; 7.1-12.1
min, 55-40% A; 12.1-13.0 min, 40-0% A; 13.0-15.0 min, 0%
A; 15.0-15.1 min, 0-95% A; 15.1-20.1 min, 95% A. The flow
rate is 0.25 mL/min.
Similarly, some studies were carried out with a Waters Quattro
Premier triple quadrupole mass spectrometer interfaced to an
Acquity UPLC. Solvent A is 100% water (Fisher Optima grade
catalog no. L-13780)-0.1% formic acid (Fluka catalog no. 94318),
and solvent B is CH3CN (Fisher Optima grade catalog no.
L-14338)-0.1% formic acid. The same LC column was used but
with a modified solvent program (linear gradients): 0-1.0 min,
95% A; 1.0-2.0 min, 95%-85% A; 2.0-2.1 min, 85%-74% A;
2.1-6.0 min, 74%-71% A; 6.0-6.1 min, 71%-56% A; 6.1-10.0
min, 56% A; 10.0-14.0 min, 56%-0% A; 14.0-14.1 min, 0%-95%
A; 14.1-18.0 min, 95% A. Supplemental Material Tables 1 and
2 in the Supporting Information give the autosampler and ESI-
MS/MS data collection parameters for the Waters Quattro Micro
triple quadrupole and the Waters Quattro Premier triple quadrupole
mass spectrometers, respectively.
For comparison purposes, we also carried out LC-ESI-MS/
MS analysis of underivatized eicosanoids in the negative ion mode.
The same LC column, solvents A and B, and flow rate were used
as for the AMPP amides. The solvent program was slightly
modified as follows: For the Waters Quattro Micro, we used 0-1.0
min, 95-63% A; 1.0-7.0 min, 63%A; 7.0-7.1 min, 63-38% A;
7.1-12.1 min, 38-23% A; 12.1-13.0 min, 23-0% A; 13.0-15.0 min,
0% A; 15.0-15.1 min, 0-95% A; 15.1-20.1 min, 95% A. For the
Waters Quattro Premier, we used 0-1.0 min, 95% A; 1.0-2.0 min,
95%-85% A; 2.0-2.1 min, 85%-59% A; 2.1-6.0 min, 59% A; 6.0-6.1
min, 59%-38% A; 6.1-11 min, 38%-23% A; 11-11.1 min, 23%-0%
A; 11.1-14 min, 0% A; 14.0-14.1 min, 0%-95% A; 14.1-19.0 min,
95% A. Values of m/z for precursor and fragment ions were as
published, 8 and cone voltages and collision energies were optimized
for each instrument and for each analyte in the usual way
(values not shown but similar to those published 8 ).
Eicosanoid Recovery Studies. We measured the recovery
of eicosanoids following the sample workup procedure given
above. Recovery studies were done using either 30 mg Strata-X
(Phenomenex Cat. 8B-S100-TAK-S) or 10 mg Oasis-HLB (Waters
Cat. 186000383) cartridges with 50 or 5 pg of each eicosanoid in
phosphate buffered saline. For these recovery studies, 50 pg of
each internal standard was added to samples just prior to
derivatization with AMPP (i.e., post-solid phase extraction).

Table 1. Extraction Yields, Liquid Chromatography Retention Times, and Tandem Mass Spectrometry Parameters
for Eicosanoid AMPP Amide Molecular Species
eicosanoid
extraction
yield (%) a
LC retention
time (min) b
limit of
quantitation in
positive mode
(pg) (CV) c
Recoveries were obtained by comparing the LC-ESI-MS/MS peak
integrals to those obtained from a sample of eicosanoids that were
derivatized with AMPP and injected directly onto the LC column
without sample processing. Recovery yields are given in Table 1.
The peak areas for the internal standards were similar in all
samples studied (not shown) showing that the eicosanoid recoveries
were similar regardless of sample matrix.
RESULTS AND DISCUSSION
Design and Preparation of AMPP Amides. Our goal was
to develop a simple derivatization procedure to convert the
carboxylic acid terminus of eicosanoids to a cationic group so that
mass spectrometry could be carried out in positive mode. The
tag should be designed so that fragmentation occurs in the analyte
portion rather than in the tag portion so that the power of tandem
mass spectrometry can be used for analytical specificity and
sensitivity. Our computational analysis of substituted N-pyridinium
methylamino derivatives 10 indicated that these had only moderate
dissociation energies for loss of pyridine and other single bond
cleavages in the linker and thus might not be suitable for our
purposes, which require fragmentation in the fatty acyl chain.
Therefore, the charge tag was redesigned to strengthen the
pyridinium N-C bond by inserting a phenyl ring as a linker. The
limit of
quantitation in negative
mode (pg) d precursorion e (m/z)
fragment
ion e (m/z)
cone
voltage f (V)
collision
energy f (eV)
6-keto-PGF1R 102/87 5.4/4.1 5/0.3 (6.8) 110/10 537 239 65/80 55/55
PGF2R 64/100 6.7/5.2 5/0.5 (9.1) 110/7 521 239 60/75 43/52
PGE 2 67/74 6.8/5.3 5/0.3 (5.5) 100/7 519 239 60/75 40/45
PGD 2 63/67 7.2/5.7 5/0.6 (7.5) 120/7 519 307 60/75 40/48
TxB2 86/86 6.2/4.6 5/0.2 (2.3) 90/4 537 337 65/70 43/42
LTB 4 42/63 9.8/7.7 5/0.5 (10.3) 80/7 503 323 50/60 35/35
5(S)-HETE 42/65 10.4/9.7 5/0.4 (10.4) 120/10 487 283 55/65 37/34
8(S)-HETE 40/75 10.3/9.1 5/0.2 (6.7) 120/10 487 295 55/65 37/40
11(S)-HETE 31/66 10.2/8.8 5/0.3 (8.9) 80/5 487 335 55/65 32/30
12(S)-HETE 69/62 10.2/8.9 10/0.9 (6.6) 140/10 487 347 55/65 35/35
15(S)-HETE 35/53 10.1/8.5 5/0.4 (11.1) 200/10 487 387 55/65 33/34
arachidonic acid 73/113 11.7/12.4 10/1 (5.6) no data 471 239 55/65 40/50
D 4-6-keto-PGF1R 5.4/4.1 541 241 65/80 55/55
D 4-PGF 2R 6.7/5.2 525 241 60/75 43/52
D 4-PGE2 6.8/5.3 523 241 60/75 40/45
D 4-PGD2 7.1/5.6 523 311 60/75 40/48
D 4-TxB 2 6.2/4.6 541 341 65/70 43/42
D 4-LTB4 9.8/7.7 507 325 50/60 35/35
D 8-5(S)-HETE 10.4/9.6 495 284 55/65 37/34
D 8-arachidonic acid 11.7/12.4 479 239 55/65 40/50
a The first number is for the Waters Oasis HLB cartridge, and the second number is for the Phenomenex Strata-X cartridge. b The first number
is for the Waters Quattro Micro, and the second number is for the Waters Quattro Premier. c LOQ values are for eicosanoid AMPP amides in
positive mode. The first number is for the Waters Quattro Micro, and the second number is for the Waters Quattro Premier. The values in
parentheses are the CVs based on 6 independent 15 µL injections of 0.075 pg/µL on the Waters Quattro Premier (see the main text). d LOQ values
are for underivatized eicosanoids analyzed by LC-ESI-MS/MS in negative mode. The first number is for the Waters Quattro Micro, and the
second number is for the Waters Quattro Premier. e m/z values listed are calculated monoisotopic values. The actual values used are derived
from instrument tuning, which is instrument dependent. f Cone voltages and collision energies were optimized for each analyte. These numbers
are instrument dependent. The first number is for the Waters Quattro Micro instrument, and the second number is for the Waters Quattro Premier
instrument.
Figure 1. Structure of AMPP and an AMPP amide along with the reagents used for derivatization.
phenyl ring also serves to enhance the derivative’s interaction with
the LC reverse phase column, which is necessarily used to
minimize any matrix effects during ESI-MS/MS analysis. Density
functional theory calculations (B3LYP/6-31G*) indicated that
homolytic cleavage of the benzylic CH 2-N bond in the charge
tag was a high-energy process that required >350 kJ mol -1 of
dissociation energy and was deemed not to out compete
fragmentations in the fatty acyl chain (Supplementary Material
Figure 1 in the Supporting Information). A side reaction that
could not be evaluated with the model system shown in Supplementary
Material Figure 1 in the Supporting Information is
elimination of 4-(N-pyridyl)benzylamine, which gives rise to the
m/z 183 marker fragment ion. An amide linkage of the charge
tag to the analytic carboxylic acid was preferred over an ester
since the former are generally more resistant to fragmentation in
the mass spectrometer. 11 The final feature of the design of AMPP
is its ease of synthesis in pure form.
We developed a simple synthesis of highly pure N-(4-aminomethylphenyl)-pyridinium
(AMPP, Figure 1) and used it as a new
derivatization reagent. We developed a simple method to convert
(10) Chung, T. W.; Turecek, F. Int. J. Mass Spectrom. 2008, 276, 127.
(11) McLafferty, F. W.; Turecek, F. Interpretation of Mass Spectra, 4th ed.;
University Science Books: Mill Valley, CA, 1993.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6793

Figure 2. LC-ESI-MS/MS traces of eicosanoids after workup from a solution in phosphate buffered saline: (A) early eluting eicosanoids and
(B) latter eluting eicosanoids.
the carboxyl group of lipid mediators to the AMPP amide using
a well-known carbodiimide coupling reagent. Coupling conditions
were optimized to give near quantitative formation of the AMPP
amide. The was shown by converting the fluorescent fatty acid
10-pyrene-decanoic acid into its AMPP amide and examining the
reaction mixture by fluorimetric HPLC and ESI-MS. No remaining
10-pyrenedecanoic acid was detected, and the AMPP amide was
formed in >95% yield with

scanning of AMPP amides of the analytes are shown in the
Supporting Information Supplementary Material Figure 6A-D.
In all the spectra, some cleavage of the AMPP tag occurs giving
rise to the peaks at m/z ) 169 and 183. To best ensure analytical
specificity, we quantified the eicosanoids using a fragment in the
analyte portion rather than in the AMPP tag. For PGD2, we used
the m/z ) 307 fragment, which is likely due to cross-ring
cleavage of the cyclopentanone ring and represents the most
abundant non-tag fragment ion. For PGE2 and PGF2R, we used
m/z ) 239 due to cleavage between C3 and C4. For AA and
6-keto-PGF1R, we used m/z ) 239, again due to C3-C4
cleavage. The major nontag fragment for TxB2 is m/z ) 337
due to cross-ring cleavage. For all HETE species, we could
identify a unique fragment ion (see Table 1) to avoid cross
contamination of MS/MS signals due to partially unresolved LC
peaks. Finally, for LTB4, we chose m/z ) 323, which is the most
abundant non-tag fragment ion but has a nonobvious origin.
Next we studied a mixture of eicosanoids in phosphate buffered
saline and optimized pre-ESI-MS/MS sample workup and LC-ESI-
MS/MS conditions. Critical for a successful method with ultrasensitive
analyte detection is high yield recovery of analytes from
the sample prior to LC-ESI-MS/MS analysis. We found that solid
phase extraction with a rapidly wettable matrix (Oasis HLB
cartridges, Waters Inc. or Strata-X cartridges, Phenomenex)
combined with analyte elution with methanol gave high yield
recovery of all eicosanoids analyzed (Table 1). The data in Table
1 is based on 50 pg of each analyte submitted to recovery studies.
When the same recovery studies were carried out with 5 pg of
eicosanoid, results were essentially identical (data not shown).
Inferior recoveries were obtained using elution of the solid phase
cartridges with acidified methanol or if the sample was subjected
to liquid-liquid extraction or protein precipitation using various
solvents (Supporting Information). Figure 2A,B shows a typical
LC-ESI-MS/MS run of an eicosanoid mixture. All of the analytes
examined elute from the LC column in less than 11 min. All
HETEs except 11(S)-HETE and 12(S)-HETE are baseline resolved.
The partial overlap of the 11(S)- and 12(S)-HETEs is not a concern
because completely selective fragment ions are being monitored.
The LC elution profile and peak shape for both the derivatized
and underivatized analytes (not shown) are similar. This shows
that the presence of the quaternary ammonium group of the
AMPP tag has no negative effect on the LC. TxB 2 gives rise to
the typical broad peak shape due to interconversion between
the two hemiacetal isomers. Also similar to underivatized
analysis, the fragments used to monitor PGD2 and PGE2
(isobars of each other) are not completely unique to each
species, but baseline LC resolution of these two eicosanoids
resolves the issue. The same is true for the isobaric pair TxB2
and 6-keto-PGF1R.
To ensure high yield conversion of eicosanoids to AMPP
amides with minimal formation of N-acyl-ureas, we used a large
excess of derivatization reagents. The AMPP tag and the EDCI
coupling reagent elute in the void volume (not shown), and the
HOBt elutes in the large time window between PGD2 and LTB4
(not shown). Thus removal of excess derivatization reagents
is not necessary prior to LC-ESI-MS/MS.
Standard curve analysis was performed starting below the limit
of quantification up to 2 orders of magnitude above this limit. In
Table 2. Coefficient of Variation (%) for LC-ESI-MS/MS
Analysis of Eicosanoid AMPP Amides
intrasample a
intersample
(10 µL PBS) b
intersample
(10 µL serum) c
6-keto-PGF1〈 4.9 16.9 not detected
TxB2 2.4 3.0 3.7
PGF2 〈 8.1 6.5 6.4
PGE2 8.0 9.2 not detected
PGD2 4.5 8.3 not detected
LTB4 7.3 17.2 7.3
5(S)-HETE 3.3 2.8 6.4
8(S)-HETE 6.7 20.3 13.7
11(S)-HETE 5.5 12.2 8.8
12(S)-HETE 5.2 10.7 6.3
15(S)-HETE 4.0 5.8 12.4
arachidonic acid 12.7 28.2 7.9
a Coefficient of variation for the analysis of the same sample of 1.25
pg/µLofeicosanoidstandardsinjectedthreetimesontotheLC-ESI-MS/MS.
b Coefficient of variation for the analysis of three independent samples
of 1.25 pg/µL of eicosanoid standards (50 pg of standards spiked into
phosphate buffered saline and worked up as described in the
Experimental Methods section for LC-ESI-MS/MS.). c Coefficient of
variation for three separate serum samples spiked with internal
standards only.
all cases, the mass spectrometry response was linear from the
limit of quantification up to the highest amount analyzed (19 pg
on-column for the Waters Quattro Micro and 5-10 pg on-column
fortheWatersQuattroPremier)(SupplementalMaterialFigure2A,B
in the Supporting Information). We define limit of quantification
as the amount of eicosanoid AMPP amide needed to give an ESI-
MS/MS signal that is 10-fold the noise (as determined using the
Waters MassLynx software). The limits of quantification for all
eicosanoid AMPP amides are listed in Table 1. The values using
the Waters Quattro Micro are 5 pg for all eicosanoids except 12(S)-
HETE (10 pg) and AA (10 pg). With the Waters Quattro Premier,
the values are 0.2-0.5 pg for all eicosanoids except 12(S)-HETE
(0.9 pg) and AA (1 pg). We also determined limit of quantification
of underivatized eicosanoids analyzed by LC-ESI-MS/MS in
negative ion mode. The values are shown in Table 1 for the Waters
Quattro Micro and the Waters Quattro Premier. The new method
using AMPP amides is found to be 10- to 20-fold more sensitive
on both instruments. This constitutes a significant improvement
in detection sensitivity of eicosanoids.
To validate the above stated estimates of the limit of quantification,
we injected eicosanoids standard mixtures six times on the
Waters Quattro Premier, where each analyte was injected at
roughly twice the estimated limit of quantification. Values of CV
for analyte to internal standard peak ratios for each analyte are
listed in Table 1. These CVs validate our estimates for the limits
of quantification by demonstrating that eicosanoid analysis can
be performed with reasonable reproducibility proximal to these
stated values.
Next, we more thoroughly investigated the reproducibility of
the AMPP derivatization method for eicosanoid analysis. The
results are summarized in Table 2. Three repetitive LC-ESI-MS/
MS analyses of an identical sample (19 pg on-column levels of
each eicosanoid) gave coefficients of variation in the range of the
2.4-12.7% range. We also prepared three independent mixtures
of eicosanoids (50 pg each) in phosphate buffered saline, submitted
each to sample workup, and analyzed each by LC-ESI-MS/
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6795

Table 3. Eicosanoid Levels (pg/µL) in Mouse Serum and Bronchial Epithelial Cells a
serum (µL) TxB2 PGE2 PGD2 PGF2〈 LTB4 5(S)-HETE 8(S)-HETE 11(S)-HETE 12(S)-HETE 15(S)-HETE
Arachi-donic
Acid
1 46 15 198 748 393 437 11 345 240 28 623
1 49 11 340 1075 545 646 17 840 703 23 246
5 43 13 174 792 490 350 14 116 617 14 900
5 43 11 169 986 469 350 16 512 672 16 631
10 51 9 255 971 586 536 15 108 749 12 200
10 39 13 248 958 593 416 14 487 732 15 476
bronchial
epithelial
cellsb 4 (7) 105 (195) 25 (22) 170 (300) 5 (19) 880 (1210)
a Eicosanoids not listed in the table were not detected. b The first number is for unstimulated cells, and the number in parentheses is for
stimulated cells. Values are for 0.47 million cells per sample stimulated in 200 µL of Hank’s balanced salt solution.
MS (19 pg on-column) (Table 2). Coefficients of variation ranged
from 2.8 to 28.2%. We also analyzed three independent serum
aliquots (10 µL), and the coefficient of variation ranged from 3.7
to 13.7%.
We then analyzed the set of endogenous eicosanoids present
in mouse serum in order to test the AMPP derivatization method
on a complex biological sample. Representative LC-ESI-MS/MS
traces are shown in the Supporting Information Supplementary
Material Figure 7 for a relatively high abundant serum eicosanoid
(TxB 2) and a relatively low abundant eicosanoid (PGF2R).
Traces for the full set of eicosanoids and internal standards
are shown in the Supporting Information Supplementary Material
Figure 3. Eicosanoids levels are shown in Table 3 for multiple
runs and with different volumes of serum analyzed.
As another example of low-level eicosanoid analysis in a
complex biological matrix, we analyzed eicosanoid formation in
calcium ionophore stimulated primary bronchial epithelial cells.
Results are summarized in Table 3, and selected ion traces for
the full set of eicosanoids are shown as Supplementary Material
Figure 4 in the Supporting Information.
Finally, we cross-validated the AMPP method with a commercially
available antibody-based assay kit. We analyzed PGE 2
production by Rat 3Y1 cells and compared the results obtained
via LC-ESI-MS/MS analysis of AMPP amides with those
obtained from the EIA kit. Results are summarized in the
Supporting Information Supplementary Material Figure 8.
There is strong agreement between the two methods.
CONCLUSIONS
We developed a simple derivatization procedure to convert
carboxylic acids to AMPP amides and showed that this significantly
improves the sensitivity for detection of eicosanoids by
LC-ESI-MS/MS. Antibody-based quantifications of eicosanoids
6796 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
have a limit of quantification around 1 pg, and the method requires
that each analyte be quantified in a single assay well. Thus, if 12
eicosanoids are to be analyzed, one requires a sample containing
∼12 pg of each species. The AMPP amide method disclosed in
the current study can detect all 12 eicosanoid species in a single
sample containing ∼0.3-1 pg of each eicosanoid and is thus more
than an order of magnitude more sensitive than antibody based
detection. Previously reported LC-ESI-MS/MS detection of
underivatized eicosanoids in negative ion mode provide a limit of
quantification in the 10-20 pg range 1–3 (see also our data in Table
1), and thus our method is more than an order of magnitude more
sensitive than these methods. We are currently studying the use
of AMPP to improve the detection sensitivity for other oxygenated
fatty acids including cysteinyl-leukotrienes, lipoxins, resolvins, and
protectans. Analysis of the full spectrum of fatty acids should be
feasible based on the results with AA reported in this study. This
new method based on AMPP amides should find widespread use
in the quantification of lipid mediator levels where sample supply
is limited.
ACKNOWLEDGMENT
This work was supported by grants from the National Institutes
of Health (Grant HL50040 to M.H.G. and Grant HL089215 to
T.S.H.) and by grants from the National Science Foundation
(Grants CHE-0750048 and CHE-0349595 to F.T.).
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review March 19, 2010. Accepted July 8,
2010.
AC100720P

Anal. Chem. 2010, 82, 6797–6806
δ 13 C Stable Isotope Analysis of Atmospheric
Oxygenated Volatile Organic Compounds by Gas
Chromatography-Isotope Ratio Mass Spectrometry
Brian M. Giebel,* Peter K. Swart, and Daniel D. Riemer
University of Miami, Rosenstiel School of Marine and Atmospheric Science, 4600 Rickenbacker Causeway,
Miami, Florida 33149
We present a new method for analyzing the δ 13 C isotopic
composition of several oxygenated volatile organic
compounds (OVOCs) from direct sources and ambient
atmospheric samples. Guided by the requirements for
analysis of trace components in air, a gas chromatograph
isotope ratio mass spectrometer (GC-IRMS)
system was developed with the goal of increasing
sensitivity, reducing dead-volume and peak band broadening,
optimizing combustion and water removal, and
decreasing the split ratio to the isotope ratio mass
spectrometer (IRMS). The technique relies on a twostage
preconcentration system, a low-volume capillary
reactor and water trap, and a balanced reference gas
delivery system. The instrument’s measurement precision
is 0.6 to 2.9‰ (1σ), and results indicate that
negligible sample fractionation occurs during gas sampling.
Measured δ 13 C values have a minor dependence
on sample size; linearity for acetone was 0.06‰ ng C -1
and was best over 1-10 ng C. Sensitivity is ∼10 times
greater than similar instrumentation designs, incorporates
the use of a diluted working reference gas
(0.1% CO2), and requires collection of >0.7 ng C to
produce accurate and precise results. With this detection
limit, a 1.0 L sample of ambient air provides
sufficient carbon for isotopic analysis. Emissions from
vegetation and vehicle exhaust are compared and show
clear differences in isotopic signatures. Ambient
samples collected in metropolitan Miami and the
Everglades National Park can be differentiated and
reflect multiple sources and sinks affecting a single
sampling location. Vehicle exhaust emissions of ethanol,
and those collected in metropolitan Miami, have
anomalously enriched δ 13 C values ranging from -5.0
to -17.2‰; we attribute this result to ethanol’s origin
from corn and use as an additive in automotive fuels.
Oxygenated volatile organic compounds (OVOCs) such as
methanol, ethanol, acetaldehyde, and acetone are gases found
throughout the troposphere that influence atmospheric chemistry
in many ways. These compounds act as a source of radicals and
a sink for the hydroxyl radical (OH), participate in tropospheric
* Corresponding author. Fax: 305-421-4689. E-mail: bgiebel@rsmas.miami.edu.
ozone formation, and are precursors to formaldehyde and CO. 1
Mixing ratios for OVOCs are typically at the low parts per billion
by volume (ppbv) level and depend on sampling location and
season. 2-5 Most atmospheric OVOC measurements have reported
information on ambient levels, source emission strengths, and flux
rates, 2,3,5-9 with two individual OVOCs, methanol and acetone,
receiving the majority of focus thus far.
Methanol is the second most abundant organic gas in the
atmosphere after methane and its global budget has been studied
extensively. 3,10-12 Emissions from vegetation are the single largest
source to the atmosphere and are estimated between 75-312 Tg
year -1 . Other sources of methanol exist, including fossil fuel
combustion, biomass burning, plant decay, and in situ atmospheric
production via oxidation of methane. Combined, these
sources are estimated at

estimated source strength of 40-95 Tg year -1 , primary biogenic
emissions comprise 22%-40% of acetone’s presence in the
atmosphere. 15 Whereas secondary sources, such as the photooxidation
of nonmethane hydrocarbons (NMHCs), contribute
the largest fraction and are estimated to account for 24%-96%
of acetone’s total source strength. 10,14,15 Sinks for acetone consist
of an oxidation mechanism by OH and direct photolysis, and
acetone’s estimated atmospheric lifetime against each of these
loss pathways is 60 days. 13,14
The value of stable isotope measurements in the unique
isotopic compositions which exist for individual sources and in
the fractionation associated with atmospheric removal. The
distinctions arise from isotope fractionation effects which occur
during chemical and physical processes. There are two types of
fractionation effects, kinetic effects and equilibrium effects. Both
kinetic and equilibrium effects influence the isotopic composition
of gas phase products produced during photosynthesis, fermentation,
and secondary metabolic processes within vegetation. Kinetic
effects, however, are principally responsible for isotopic compositions
resulting from unidirectional removal processes where
equilibrium is not achieved, as is the case of in situ chemical
oxidation by OH or photolysis. The use of stable isotopes of
carbon, oxygen, and hydrogen is a valuable tool to enhance our
understanding of global patterns for OVOCs, a unique class of
hydrocarbons that originate from not only primary biogenic and
anthropogenic sources but also secondary sources. Collectively,
the measured average isotopic composition of an OVOC in the
atmosphere, coupled with isotopic characterization of its major
sources and sinks, can be used as a means to elucidate the
chemistry of these compounds within the atmosphere, their
formation pathways, and reduce uncertainty in their budgets. An
excellent review by Goldstein and Shaw contains more details on
the use of stable isotopic data and its application to atmospheric
budgets of VOCs. 16
Herein, we restrict the discussion to variations in the stable
isotopes of carbon, 13 C and 12 C. The stable isotopic composition
of a sample is expressed as a ratio (R) of 13 Cto 12 C. During
analysis, the ratio of a sample is compared to that of Vienna
Pee Dee Belemnite (V-PDB) by way of a working reference
gas using an isotope ratio mass spectrometer (IRMS). The ratio
of the sample is reported in delta (δ) notation as a per mil (‰)
difference of the sample compared to the reference: δ 13 C (‰)
) ((RSAMPLE/RSTANDARD) - 1) × 10 3 . If a compound has a
negative value, then it contains less 13 C than the standard and
is said to be isotopically light (or depleted); if a compound has
a positive value then it is isotopically heavy (or enriched).
Application of stable isotopic techniques in atmospheric
chemistry has focused on more abundant components of the
atmosphere, notably, CO2, CH4, and CO. The use of isotopes
led to an increased understanding of sources, sinks, and
seasonal cycling of these compounds. 17-24 Rudolph and his
(15) Mao, H.; Talbot, R.; Nielsen, C.; Sive, B. Geophys. Res. Lett. 2006, 33.
(16) Goldstein, A. H.; Shaw, S. L. Chem. Rev. 2003, 103, 5025–5048.
(17) Allison, C. E.; Francey, R. J. J. Geophys. Res. 2007, 112, DOI: 10.1029/
2006JD007345.
(18) Mak, J. E.; Manning, M. R.; Lowe, D. C. J. Geophys. Res. 2000, 105, 1329–
1336.
(19) Mak, J. E.; Kra, G. Chemosphere 1999, 1, 205–218.
(20) Mak, J. E.; Kra, G.; Sandomenico, T.; Bergamaschi, P. J. Geophys. Res. 2003,
101, 14415–14420.
6798 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
colleagues are credited as the first to use gas chromatographyisotope
ratio mass spectrometry (GC-IRMS) to analyze NMHCs
in atmospheric gases. 25 These compounds are difficult to measure
because their mixing ratios are low; they require separation from
the atmospheric matrix and need to be combusted to CO2 before
transfer to the IRMS. For this reason, analyses are performed
using a gas chromatograph and combustion oven coupled to
an IRMS operating under continuous flow conditions. Rudolph
et al. sampled a variety of locales, ranging from pristine
noncontaminated air to heavily contaminated air dominated by
automobile emissions. 27 Results from their work showed clear
distinctions between the δ 13 C signatures of NMHCs sampled
at each location. The signatures reflected the local sources,
meteorological conditions, and fractionation mechanisms acting
at each site. Additional studies of NMHCs include studying
specific sources, oxidation processes, and other locations. 26-28
Applications of GC-IRMS techniques for analysis of OVOCs
are restricted in scope and focus on one or two compounds. 29-39
This is due to additional difficulties associated with sampling and
analyzing OVOCs, including low mixing ratios, tendency for
sample to become lost on sampling surfaces caused by OVOCs’
polar nature, difficulty in selectively removing water and CO2 from
the sample matrix without perturbing the OVOCs, and optimizing
chromatographic separation in the presence of associated
NMHCs. To circumvent some of these difficulties, derivatization
has been used as a technique to make the polar compounds
amenable to analysis. 29-31,36,37 Four studies of OVOCs
using direct GC-IRMS are available, two for formaldehyde 34,35 and
two for acetaldehyde. 32,39
The technique presented here allows for direct conversion of
several OVOCs to CO2 for measurement of δ 13 C from small
sample volumes and a range of sample types, including
(21) Quay, P.; Stutsman, J.; Wilbur, D.; Snover, A.; Dlugokencky, E.; Brown, T.
Global Biogeochem. Cycles 1999, 13, 445–461.
(22) Rice, A. L.; Gotoh, A. A.; Ajie, H. O.; Tyler, S. C. Anal. Chem. 2001, 73,
4104–4110.
(23) Tyler, S. C. In Stable Isotopes in Ecological Research; Rundel, P. W.,
Ehleringer, J. R., Nagy, K. A., Eds.; Springer-Verlag: New York, 1989.
(24) Tyler, S. C.; Rice, A. L.; Ajie, H. O. J. Geophys. Res. 2007, 112.
(25) Rudolph, J.; Lowe, D. C.; Martin, R. J.; Clarkson, T. S. Geophys. Res. Lett.
1997, 24, 659–662.
(27) Tsunogai, U.; Yoshida, N.; Gamo, T. J. Geophys. Res. 1999, 104, 16033–
16040.
(26) McCauley, S. E.; Goldstein, A. H.; DePaolo, D. J. Proc. Natl. Acad. Sci.
U.S.A. 1999, 96, 10006–10009.
(28) Rudolph, J.; Czuba, E.; Huang, L. Geophys. Res. Lett. 2000, 27, 3865–3868.
(29) Guo, S.; Wen, S.; Wang, X.; Sheng, G.; Fu, J.; Hu, P.; Yu, Y. Atmos. Environ.
2009, 43, 3489–3495.
(30) Guo, S.; Wen, S.; Wang, X.; Sheng, G.; Fu, J.; Jia, W.; Yu, Y.; Lu, H. Rapid
Commun. Mass Spectrom. 2007, 21, 1809–1812.
(31) Guo, S. J.; Wen, S.; Zu, G. W.; Wang, X. M.; Sheng, G. Y.; Fu, J. M. Chin.
J. Anal. Chem. 2008, 36, 19–23.
(32) Jardine, K.; Karl, T.; Lerdau, M.; Harley, P.; Guenther, A.; Mak, J. E. Plant
Biol. 2008, 9999.
(33) Keppler, F.; Kalin, R. M.; Harper, D. B.; McRoberts, W. C.; Hamilton, J. T. G.
Biogeosciences 2004, 1, 123–131.
(34) Rice, A. L.; Quay, P. Environ. Sci. Technol. 2009, 43, 8752–8758.
(35) Rice, A. L.; Quay, P. D. Anal. Chem. 2006, 78, 6320–6326.
(36) Wen, S.; Feng, Y. L.; Yu, Y. X.; Bi, X. H.; Wang, X. M.; Sheng, G. Y.; Fu,
J. M.; Peng, P. A. Environ. Sci. Technol. 2005, 39, 6202–6207.
(37) Wen, S.; Yu, Y. X.; Guo, S. J.; Feng, Y. L.; Sheng, G. Y.; Wang, X. M.; Bi,
X. H.; Fu, J. M.; Jia, W. L. Rapid Commun. Mass Spectrom. 2006, 20, 1322–
1326.
(38) Yamada, K.; Hattori, R.; Ito, Y.; Shibata, H.; Yoshida, N. Geophys. Res. Lett.
2009, 36.
(39) Jardine, K.; Harley, P.; Karl, T.; Guenther, A.; Lerdau, M.; Mak, J. E.
Biogeosciences 2008, 5, 1559–1572.

Figure 1. Schematic view of the preconcentration system and GC-IRMS. The GC-IRMS relied on a low-volume capillary reactor, water trap,
and a balanced reference gas delivery system (WRG, working reference gas).
emissions from vegetation and automobiles and rural and urban
ambient air. The system consists of a two-stage trapping
procedure using a carbon sorbent followed by cryofocusing. A
low-volume capillary reactor, water trap, and open split combined
with high-speed chromatography enhance sensitivity and
improve peak resolution.
EXPERIMENTAL SECTION
Reference and Calibrant Gases. A working CO2 reference
gas calibrated against the V-PDB carbonate standard was used
for determination of the 13 C/ 12 C ratio of OVOCs using IRMS.
To achieve proper instrument response, a 0.1% CO2 working
reference was prepared and calibrated. A cylinder of research
grade CO2 was used to create a gas phase subsample, from
which, a dilution of 0.1% CO2 was made in helium. The isotopic
values for pure CO2, its subsample, and the 0.1% CO2 dilution
were measured using a Finnigan MAT 251 IRMS with conventional
dual inlet and a Europa Scientific 20-20 IRMS linked to
an automated nitrogen carbon analyzer (ANCA) under continuous
flow. Henceforth, the 0.1% CO2 gas will be referred to as
working reference gas.
The primary calibrant gas mixture used for testing and
development was prepared gravimetrically. 40 Liquid compounds
with purity g98% were obtained from Sigma-Aldrich and included
methanol, ethanol, propanal, acetone, methyl ethyl ketone, 2-pentanone,
and 3-pentanone. Microliter volumes of the compounds
were discharged directly into a clean, evacuated aluminum
cylinder (Luxfer Gas Cylinders, Riverside, CA) with a stainless
steel valve (Ceoedux, Mount Pleasant, PA). The cylinder was
pressurized with high-purity BIP Technology N2 gas (all compressed
gases utilized in this work were obtained from Airgas
South, Keenesaw, GA) to achieve desired mixing ratios for all
components. The mixing ratio and purity of the mixture were
(40) Apel, E. C.; Calvert, J. G.; Greenberg, J. P.; Riemer, D.; Zika, R.; Kleindienst,
T. E.; Lonneman, W. A.; Fung, K.; Fujita, E. J. Geophys. Res. 1998, 103,
22281–22294.
verified via GC-flame ionization detector (FID) and GC/mass
spectrometry (MS). Final mixing ratios for each individual
component fell between 365 ppbv (2-pentanone) and 931 ppbv
(methanol) with an associated error of 5% for each component.
The calibration gas mixture was diluted further by dynamic
dilution into moist zero air when needed. 41
Low-pressure single and multicomponent gases were made
similarly to the calibrant gas mixture except an evacuated
electropolished stainless steel canister (Entech Instruments, Simi
Valley, CA) was used and pressurized with 55 pounds per square
inch absolute (psia) UHP helium for dilution. Mixing ratios for
the single mixed gases were selected to equal that of the working
reference gas.
Elemental Analyzer. A Europa Scientific 20-20 IRMS linked
to an automated nitrogen carbon analyzer (ANCA) was used to
obtain the δ 13 C signature of each raw liquid compound used to
prepare the calibrant gas mixture. Microliter volumes of liquid
reagents were loaded on a sorbent (Chromosorb W, Advanced
Minerals, Goleta, CA) contained within a small tin capsule. The
capsule was folded over itself to prevent sample loss and
quickly dropped into the furnace to begin the analysis. Raw
compounds were compared against a working lab standard of
pure CO2 calibrated against V-PDB.
Preconcentration System. A custom preconcentration system
was installed on the inlet of the gas chromatograph and included
two manual stainless steel six-port rotary valves (1/16 in. fitting,
0.40 mm i.d.; all rotary valves in this work were obtained from
Vici Valco Instruments, Houston, TX) connected in series. Fused
silica-lined stainless steel tubing joined the entire precononcentration
system (1.59 mm o.d., 0.51 mm i.d.; Silcosteel tubing, all
Silcosteel tubing in this work were obtained from Restek, Bellefonte,
PA). A schematic of the complete analytical design appears
in Figure 1.
(41) Apel, E. C.; Hills, A. J.; Lueb, R.; Zindel, S.; Eisele, S.; Riemer, D. D. J.
Geophys. Res. 2003, 108.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6799

The first rotary valve (RV1) used an injection loop (14.3 cm ×
1.6 mm o.d. × 1.02 mm i.d.; Silcosteel tubing) yielding a volume
of 0.12 cm 3 . Rotary valve 1 allowed for analysis of working
reference gas for leak check purposes and single-component
gases in helium to test the system with individual OVOCs.
Sample flow rates through RV1 were set at 3 cm 3 min -1 and
regulated upstream by a 0-100 standard cubic centimeters per
minute (sccm) mass flow controller (MFC, no. 1) (all MFCs
used in this work were obtained from Unit/Celerity, San Jose,
CA).
The second valve (RV2) allowed for loading of the multicomponent
calibrant gas during development tests and samples for
analysis. The calibrant gas or samples connected to a simple valve
system giving the operator ease of on/off control and an additional
helium input used to purge the sample loop. Flow rates through
RV2 were 50 cm 3 min -1 and controlled by an oil pump (Alcatel
Vacuum Products, Smyrna, GA) with a liquid nitrogen vapor
trap and a 0-200 sccm MFC (no. 2) located downstream. The
sampling loop of RV2 included a carbon adsorbent trap which
is nonretentive for the major components of air (N2, O2, CO2,
and H2O) while trapping OVOCs. The carbon adsorbent trap
consisted of a section of Silcosteel (30.5 cm × 3.2 mm o.d. ×
2.1 mm i.d) packed with 24.6 cm of Carboxen 1016 (Supleco,
St. Louis, MO). Carboxen 1016 retained all the analytes of
interest, completely desorbed them when heated, and contributed
negligible amounts of CO2 during the desorption process.
Typically, 1.0 L volumes were sampled over 20 min. Helium
purged the sampling loop and adsorbent trap (50 cm 3 min -1 ) for
5 min to remove unwanted gases and water that may remain
after loading. After the purge procedure, RV2 was manually
switched allowing carrier gas (0.7 cm 3 min -1 ) from the GC to
back flush the adsorbent trap for 10 min. During this step, the
adsorbent trap was desorbed with resistive heating at 200 °C
using high temperature heater tape, thermocouple, and a
CN76000 temperature controller (Omega Engineering, Stamford,
CT).
Desorbed material was cryo-focused as it passed through an
open loop trap (46 cm, 0.56 mm o.d., 0.28 mm i.d., Siltek/Sulfinert
tubing, Restek, Bellefonte, PA) contained in liquid nitrogen. Before
injection, RV2 was switched to remove the carbon adsorbent trap
from the carrier gas flow path. The cryo-focuser was heated with
boiling water to inject the compounds onto the GC column.
GC-IRMS System. An Agilent 6890 gas chromatograph was
fitted with a DB-624 capillary chromatography column (20 m ×
0.18 mm i.d. × 1 µm film thickness, 6% cyanopropylphenyl/94%
dimethyl polysiloxane, Agilent Technologies, Santa Clara, CA).
Ultrahigh purity helium entering the GC and IRMS passed
through liquid nitrogen as a final measure to scrub it of possible
contaminants, including trace CO2. The GC operated in constant
flow mode with a nominal carrier flow rate set to 0.7 cm 3 min -1 .
Carrier flow was measured at the open split located immediately
upstream of the IRMS ion source. Temperature programming
for the GC oven started at 30 °C for 3 min followed by a ramp
of 10 °C min -1 and held at a final temperature of 125 °C for 7.5
min. The total GC run time was 23 min.
A capillary reactor used for combusting OVOCs to CO2 was
constructed of fused silica tubing and contained two catalytic
wires. One was an alloy blend consisting of Cu/Mn/Ni (84%/
6800 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
12%/4%, 0.1 mm diameter, Alfa-Aesar, Ward Hill, MA), the other
is Pt (99.99%, 0.1 mm diameter, Sigma-Aldrich, St. Louis, MO).
Details about the fabrication and conditioning of the reactor
are provided in the Supporting Information.
Trace amounts of H2O, created with CO2 during the combustion
process, can interfere by contributing signal at m/z -45
and -46 due to protonation of CO2 within the ion source. A
continuous length of deactivated fused silica (∼1.5 m × 0.36
mm o.d. × 0.25 mm i.d., Supelco, St. Louis, MO) was connected
to the downstream end of the capillary reactor and served as
a transfer line, water trap, and link to the open split. At the
midpoint of the transfer line a single loop (∼10-15 cm in
length) created a water trap when immersed in a dry ice-ethanol
slurry (-55 to -65 °C) and prevented water from reaching
the ion source. Sample derived CO 2 continued to the outlet
located in the open split. The looped trap was removed from
the slurry after every run, and condensed water was driven off
with helium carrier gas and by warming the loop to 100 °C for
several minutes.
A custom designed open split (Figure 1) was positioned
adjacent to the ion source of a Europa Scientific GEO 20-20
IRMS. 42 The open split resided within a chamber that housed four
distinct outlets: two primary outlets for (1) the GC effluent and
(2) the helium blank/0.1% CO2 reference gas; two secondary
outlets for (1) tuning the IRMS ion beams and (2) supplying
the chamber assembly with an inert atmosphere of helium
(MFC no. 4, 10 cm 3 min -1 ). Details of the open split and
working reference gas delivery system are contained in the
Supporting Information. A length of deactivated fused silica (∼3
m × 0.36 mm o.d., 0.10 mm i.d. SGE, Austin, TX) produced a
capillary leak to the ion source and was manually moved between
the various outlets located in the open split as needed.
Europa Scientific GEO 20-20 IRMS. The IRMS was tuned
daily with working reference gas and operated with an accelerating
voltage of 2.5 kV. The ion trap and filament emission currents
were set to approximately 600 and 1300 µA, respectively. The
vacuum system of the IRMS created a 0.2 cm 3 min -1 flow rate
through the restriction capillary leading to the ion source while
maintaining an operating source pressure of 4.5 × 10 -6 Torr.
When combined, the carrier flow rate (0.7 cm 3 min -1 ) exiting
the GC outlet and the ion source flow (0.2 cm 3 min -1 ) resulting
from the capillary leak produced a split ratio of 3.5 and reduced
the amount of carbon transferred to the IRMS by approximately
70%.
Data Collection and Analysis. Data for time (t) and mass to
charge ratios (m/z) 44, 45, and 46 were recorded at 1.0 Hz to a
PC using acquisition software provided by the IRMS manufacturer.
Three separate third party software packages were used (i.e.,
Thermo Grams AI/8.0, Origin Lab Origin Pro 8, and Microsoft
Excel 2007) for data analysis. Details of the data manipulation are
described in the Supporting Information.
Source and Ambient Sampling. Two types of sources were
investigated in this study, natural emissions released from five
tropical plants and exhaust from an automobile without a catalytic
converter. Plant emissions were studied using static enclosure
methods with (1) clipped and (2) intact vegetation. For the (1)
clipped vegetation experiments, a branch was physically removed
(42) Sacks, G. L.; Zhang, Y.; Brenna, J. T. Anal. Chem. 2007, 79, 6348–6358.

from the specimen and enclosed in a Teflon sampling bag with a
Nupro valve fitting (SS-4H, Swagelok Co., Solon, OH). Hydrocarbonfree
zero-air flushed the enclosed branch before being zip-tied
shut and returned to direct sunlight for a period of 20-60 min.
After this period, the bag and branch were brought into the lab
and connected to the preconcentration system via the gas
manifold. Sample volumes between 100-200 cm 3 were loaded
using the RV2 loop and carbon sorbent. For (2) intact branch
experiments, a single branch on the specimen was enclosed
with the Teflon bag assembly and flushed with zero-air before
being zip-tied shut and left in ambient light for a period 20-60
min. Instead of clipping the branch and returning the bag to
the laboratory, an evacuated canister was connected to the
bag’s valve and its contents were removed. As with the clipped
branch, sample volumes between 100-200 cm 3 were loaded
for analysis.
Automobile emissions from the right exhaust bank of a 1972
International Scout were collected. The Scout lacks a catalytic
converter and was operated at constant cruising and load conditions
(∼2000 rpm, ∼80 kph) during sample collection. A piece of
stainless steel tubing (∼1.5 m length) was secured to the inside
of the passenger side exhaust and brought to the inside of the
passenger compartment for connection to an evacuated canister.
Samples were at atmospheric pressure after collection.
Ambient whole air samples were collected at three locations
in South Florida. Everglades National Park was chosen to
represent a rural atmosphere dominated more by biogenic sources
than anthropogenic sources. A partially enclosed lower roadway
of Miami International Airport was sampled to ascertain a
collective measure of concentrated and fresh vehicular emissions;
Miami’s Financial District represented a more semiurban environment.
Samples from the Everglades were pressurized to approximately
30 pounds per square inch gauge (psig) by a DC
powered metal bellows pump (MB-302, Senior Flexonics, Sharon,
MA) into an evacuated canister; airport and financial district
samples were collected at atmospheric pressure. Sample volumes
for analysis ranged between 100-200 cm 3 for the airport and
financial district to 1.0 L for Everglades National Park.
RESULTS AND DISCUSSION
Guided by the requirements for analysis of trace components
in air, a GC-IRMS system was developed with the goal of analyzing
OVOCs in ambient air. This included developing a preconcentration
stage, increasing sensitivity, reducing dead-volume and peak
band broadening, optimizing combustion and water removal, and
decreasing the split ratio to the IRMS. An extended discussion
focusing on these developments, including images of the working
reference gas delivery system and open split, are available in the
Supporting Information. These developments allowed for analysis
of several OVOCs in ambient air at mixing ratios representative
of a large portion of the troposphere ranging from urban to
remote. Figure 2 shows the range of mixing ratios present in the
atmosphere for methanol, ethanol, acetone, and MEK and the
corresponding nanograms of C transmitted to the ion source with
this analytical technique. 3,6,9,12,14,43 All but the most remote
portions of the atmosphere can be investigated using this method.
Reference Gas Analyses. Working Reference Gas. The diluted
working reference gas was prepared carefully in a series of steps
in order to monitor possible fractionation during the dilution
Figure 2. Range of atmospheric OVOC mixing ratios expected in
large portions of the troposphere and their relation to carbon
transmitted to the IRMS ion source. The shaded area corresponds
to the method detection limit.
process. Results and further discussion of the preparation procedure,
including an interlab comparison performed with Stony
Brook University, can be found in the Supporting Information
(Working Reference Gas Analyses and Table S-1).
System Diagnostic Leak Assessments. Rotary Valve 1 (RV1) was
used for various diagnostics tests. One test evaluated the leak
tightness of the complete sample path through the preconcentration
system, gas chromatograph, reactor assembly, water trap,
and open split delivery to the IRMS. We compared working
reference gas injected at the preconcentration system to that of
its normal injection site at the open split. In the absence of
detectable leaks in the sample path, daily offsets of 0.3-0.5‰ were
routinely observed. On occasions when a microcrack was suspected
of breaching the capillary reactor, offsets >2‰ were
observed. This test was performed daily to evaluate the integrity
of the instrumentation. If the offsets were 0.3-0.5‰, the concurrent
data for the day were corrected by the appropriate amount.
If the offset exceeded 0.5‰, the capillary reactor was replaced.
Dynamic Range and Linearity. A 6 L electropolished stainless
steel bulb with a dip tube assembly served as an exponential
dilution flask to test the dynamic range and linearity of the method
and IRMS in the absence of combustion. The bulb contained a
1% CO 2 mixture in helium made from the same subsampled
CO2 used in the production of working reference gas. A diluant
flow of helium entered the steel bulb through the dip tube at
a rate of ∼125 cm 3 min -1 and the outflow was plumbed to RV1.
The total analysis occurred over a 5.5 h period broken into 9
segments with the introduction of working reference gas. The
amount of carbon reaching the ion source was ∼0.1-80 ng.
The δ 13 C values over this range are expressed as a difference
of the measured (and corrected) exponentially diluted CO2 from
the working reference gas’ accepted value and are displayed
in Figure 3a. Of particular interest is the appearance of a positive
offset, ∼0.46‰, from zero. Efforts were made to minimize
fractionations during the gas transfers, and the gases were made
from the same stock. Despite this effort, the offset still persisted.
(43) Apel, E. C.; Emmons, L. K.; Karl, T.; Flocke, F.; Hills, A. J.; Madronich, S.;
Lee-Taylor, J.; Fried, A.; Weibring, P.; Walega, J.; Richter, D.; Tie, X.;
Mauldin, L.; Campos, T.; Weinheimer, A.; Knapp, D.; Sive, B.; Kleinman,
L.; Springston, S.; Zaveri, R.; Ortega, J.; Voss, P.; Blake, D.; Baker, A.;
Warneke, C.; Welsh-Bon, D.; de Gouw, J.; Zheng, J.; Zhang, R.; Rudolph,
J.; Junkermann, W.; Riemer, D. D. Atmos. Chem. Phys. 2010, 10, 2353–
2375.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6801

Figure 3. (A,B) Results for exponentially diluted (A) CO2 and (B) acetone samples. δ 13 C values are expressed as a difference of the diluted
CO2 and acetone from the accepted value of the 0.1% working reference gas and acetone value obtained on the elemental analyzer. An offset,
opposite in sign but almost equal magnitude, exists for CO2 (∼0.5‰) and acetone (∼-0.6‰). For CO2, this is thought to be the result of an
ambient leak whereby atmospheric CO2 enters the system. For acetone, incomplete combustion within the capillary reactor may contribute to
the observed negative offset.
Table 1. Tabulated δ 13 C Values for OVOCs Used in This Work a
elemental analyzer
liquid compounds
mean
±1σ
(‰)
GC-IRMS
single-component gas
mean
±1σ
(‰)
95%
confidence
(±‰)
%
error
The positive offset for CO2 was likely a result of small
amounts of ambient CO2 becoming entrained in the system
and thereby enriching the measured δ 13 C value. For large
sample sizes this appeared to have a minimal effect. However,
the effect became magnified for sample sizes below 1 ng C,
where deviations up to 2‰ are observed. The linearity over
this range, determined by ordinary linear regression, was
0.01‰ ng C -1 . This suggests that variation of the δ 13 C signature
with sample size is negligible and requires 10 ng C to induce
a change of 0.1‰. Accuracy and precision were best over the
range of ∼0.2 to ∼20 ng C. Samples containing more than 20
ng C and less that 0.2 ng C were enriched in 13 C.
Liquid Compounds and Single-Component Gases. Lowpressure
single-component gas samples entered RV1 at a rate of
3cm 3 min -1 and were loop injected onto the GC-IRMS with no
cryo-focusing before the chromatographic column. Ten injec-
GC-IRMS
low-pressure 7-component
mean
±1σ
(‰)
95%
confidence
(±‰)
%
error
GC-IRMS
high-pressure 7-component pooled GC-IRMS
mean
±1σ
(‰)
95%
confidence
(±‰)
%
error
mean
±1σ
(‰)
95%
confidence
(±‰)
methanol –35.3 ± 0.1 –34.1 ± 0.1 0.1 3.5 –33.0 ± 0.1 0.1 6.5 –34.4 ± 2.8 2.1 2.5 –34.0 ± 1.6 0.7 3.7
ethanol –29.2 ± 0.2 –27.8 ± 0.8 0.6 4.8 –26.2 ± 0.8 1.3 10.2 –26.5 ± 0.5 0.4 9.2 –26.6 ± 1.4 0.6 8.7
propanal –32.8 ± 0.1 –31.9 ± 0.2 0.1 2.7 –35.0 ± 0.6 1.0 –6.9 –27.4 ± 0.8 0.6 16.5 –30.5 ± 2.9 1.2 6.9
acetone –27.5 ± 0.2 –28.5 ± 0.7 0.5 –3.7 –27.9 ± 0.2 0.3 –1.4 –27.6 ± 0.2 0.1 –0.4 –28.0 ± 0.6 0.2 –1.7
MEK –23.2 ± 0.2 –23.5 ± 0.9 0.6 –1.3 –25.5 ± 0.7 1.1 –10.1 –22.5 ± 0.3 0.2 –3.0 –23.5 ± 1.2 0.5 –1.2
2-pentanone –25.0 ± 0.3 –26.1 ± 0.2 0.1 –4.3 –33.7 ± 1.1 1.8 –34.8 –28.4 ± 0.6 0.5 –13.6 –28.6 ± 2.8 1.1 –14.3
3-pentanone –30.7 ± 0.2 –30.7 ± 0.6 0.4 –0.1 –34.3 ± 0.3 0.5 –11.6 –32.7 ± 0.7 0.5 –6.5 –32.3 ± 1.5 0.6 –5.1
sample number n ) 3 n ) 10 n ) 4 n ) 9 n ) 23
diluent He
Instrument Variables
He N2 sample loop RV1 RV1 RV2
cryo-focus no yes yes
carbon sorbent no no yes
zero air dilution no no yes
a Accuracy and precision is traced from the elemental analyzer through the final design of the GC-IRMS system. Different variables tested
during each phase are listed. The system’s total precision was calculated between 0.6 and 2.9‰ when compared to the values obtained on the
elemental analyzer.
6802 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
%
error
tions of each gas were compared to six injections of working
reference gas. Single-component gases tested the GC-IRMS
instrumentation by comparison to the isotopic values obtained
for the raw liquids on the elemental analyzer (ANCA). The
values for the raw liquids served as the basis of all our
comparisons. The calculated percent difference between the
two measurements ranged between -0.1 and 4.8%, and the
results are listed in Table 1.
Acetone was suitable to test the dynamic range and linear
response of the IRMS with the added step of combustion. Acetone
was chosen as the analyte because it showed consistent and
excellent reproducibility across all aspects of this study. The
experimental design was identical to that of the CO2 experiment
described previously but with one exception, the acetone
mixture had a starting carbon equivalent of 0.1% before the
diluent flow of helium was added. A 1% mixture of acetone was

Figure 4. (A) IRMS mass-44 chromatographic response for the high-pressure, seven-component gas. (B) An enlarged plot shows reasonable
separation for all seven components. The first peak observed in part B is a combination of CO2 collected during the zero-air dilution of the
calibrant gas and CO2 created by Carboxen 1016 during the desorption process. The peak immediately preceding methanol was identified as
the OVOC acetaldehyde and is an artifact background from the adsorption trap.
avoided for two reasons. First, ambient samples are not
expected to be greater than 0.1%, and second, there is more
concern for what happens to measured isotopic signatures as
smaller sample concentrations are approached. The amount
of carbon reaching the ion source was determined similarly to
the CO2 test and ranged between ∼0.8 and 12 ng. The δ 13 C
values over this range are expressed as a difference of the
measured and corrected exponentially diluted acetone from
the value obtained on the elemental analyzer (Figure 3b).
The linearity over this range, determined by ordinary linear
regression, was 0.06 ‰ ng C -1 . For acetone, this indicates
that sample size can influence measured δ 13 C and that a
change between 1 and 10 ng C can induce a noticeable shift
of 0.6‰ in measured δ 13 C. Accuracy and precision were best
over the range of 0.2-10 ng C.
Also worth noting is the apparent negative offset for acetone,
∼-0.56‰, compared to the positive offset for CO2, ∼0.46‰. Raw
data for both experiments were corrected by 0.5‰ and 0.4‰
for acetone and CO2, respectively. However, the offsets still
exist. Entrainment of ambient CO2 did not appear to affect
acetone because of its separation on the chromatographic
column. The negative offset for acetone was likely related to
incomplete combustion within the capillary reactor. Thermodynamic
principles support 12 C being combusted before 13 C;
thus, if combustion was incomplete, we would observe a lighter
δ 13 C value.
Calibrant Gas Analyses. Low-Pressure Seven-Component Gas
Mixture. A low-pressure seven-component gas mixture in helium
was used preliminarily to test chromatographic conditions in the
absence of the carbon sorbent by using the RV1 loop (Table 1).
This was a logical step between the use of single-component gases
and a gravimetrically prepared, high-pressure, seven-component
calibration gas in nitrogen. The low-pressure seven-component
gas mixture flowed through the RV1 loop for 5 min prior to
starting the analysis. The flow rate (3 cm 3 min -1 ) was maintained
by MFC (no. 1) upstream of RV1. After the initial 5 min purge
period, RV1 was manually switched and the gas within the
injection loop was diverted through RV2 and cryogenically
focused in liquid nitrogen for an additional 5 min before
injection into the chromatographic column. Of particular note
are the values obtained for 2- and 3-pentanone, which are
depleted in 13 C compared to both the liquid compounds and
the single component gas mixtures. This may indicate an
unknown effect resulting from the analytical column. The
percent error between this measurement technique and that
performed on the elemental analyzer for the pure liquid
compounds ranges between 1.4 and 35%.
Gravimetric Seven-Component Gas Mixture. One of the main
goals of this work was to develop a GC-IRMS system capable of
measuring OVOCs over the dynamic range found in the atmosphere.
To mimic ambient levels of these compounds in the
atmosphere, the high-pressure calibrant gas was diluted into moist
zero-air using a dynamic dilution system. Dilution produced
mixing ratios between ∼18.6 ppbv (methanol) and 7.3 ppbv (2pentanone)
for all components. The diluted calibrant was connected
directly to the gas manifold (Figure 1). Using the range
of mixing ratios produced after the high-pressure calibrant gas
was diluted in zero-air (7.3-18.6 ppbv), the volume of air
concentrated (1.0 L), and the open split dilution (∼30%), we
calculated ∼2.5-5 ng C were delivered to the ion source for all
components. Results for nine replicate analyses are presented in
Table 1, and an example of the chromatographic response appears
in Figure 4. Reasonable agreement exists for all seven components
compared to the liquid reagents analyzed on the elemental
analyzer; the margin of error between these two measurements
ranged between 0.4 and 16.5%. The components with the two
largest errors were propanal (16.5%) and 2-pentanone (13.6%). Both
of these peaks are the leading peak in a pair (propanal/acetone
and 2-penatnone/3-penatanone), and perhaps the later eluting
compounds influence the measured δ 13 C values of the earlier
compounds. This is supported by the observation that analysis
of the single-component gases for the same compounds on the
GC-IRMS had a lower error (

Table 2. δ 13 C Values for Compounds Emitted from Various Tropical Plants and a Fossil Fuel Combustion Source a
sand live oak
Quercus geminata
orange
Citrus sinensis
lemon
Citrus limon
plant type
Students’ t test, and the results show that differences between
the raw material on the elemental analyzer and the gases on the
GC-IRMS system are significant. However, the percent difference
between the two procedures is small; less than 9% error exists
between the pooled results and those of the raw liquid compounds
for all compounds except 2-pentanone, which has an associated
error of 14.3%.
Source Measurement Results. The δ 13 C results from the
automotive exhaust and tropical plants are presented in Table
2. A representative chromatogram from the automotive exhaust
sample is shown in the Supporting Information (Figure S-1) and
serves as a good example of a complex sample matrix and the
system’s peak resolution. We have also incorporated, in some
instances, for both source and ambient samples, results for the
OVOC acetaldehyde even though analysis of this compound is
influenced by a variable artifact background. 9,41 Data for NMHCs,
isoprene, benzene, and toluene are made known because they
were easily identifiable in the chromatograms and offer a
comparison to previously measured NMHCs δ 13 C values. 44-46
In general, a clear distinction exists between plant and fossil fuel
emissions, and as would be expected based on the range of δ 13 C
values for oils and fuels worldwide, 16,46 fossil fuel derived OVOC
emissions were more enriched in 13 C compared to the plant
results.
Of particular note are the substantially enriched δ 13 C values
(-5.0 ± 0.4‰) for ethanol, compared to the other compounds
in the fossil fuel combustion experiments. Ethanol, derived
from corn, is currently being blended into gasoline in mixtures
g10% (v/v). It has been reported that biofuels consume >25%
of U.S. corn production and that ethanol constitutes 99% of all
biofuels in the United States. 47,48 Utilizing C4 photosynthesis,
which discriminates less against 13 C, corn and other C4 plants
are generally enriched in the isotope compared to C3 plants.
(44) Iannone, R.; Koppmann, R.; Rudolph, J. J. Atmos. Chem. 2007, 58, 181–
202.
(45) Rudolph, J.; Anderson, R. S.; Czapiewski, K. V.; Czuba, E.; Ernst, D.;
Gillespie, T.; Huang, L.; Rigby, C.; Thompson, A. E. J. Atmos. Chem. 2003,
44, 39–55.
(46) Rudolph, J.; Czuba, E.; Norman, A. L.; Huang, L.; Ernst, D. Atmos. Environ.
2002, 36, 1173–1181.
(47) Barnett, M. O. Environ. Sci. Technol. 2010, 44, 5330-5331.
(48) Farrell, A. E.; Plevin, R. J.; Turner, B. T.; Jones, A. D.; O’Hare, M.; Kammen,
D. M. Science 2006, 311, 506–508.
philodendron
Philodendron selloum
sea grape
Coccoloba uvifera Keppler et al. 33
fossil fuel
combustion
acetaldehyde -29.9 (2.3) -25.7 (0.1) -22.4 (1.4)* -17.5 (0.5) -30.7 (1.1)* -24.9 (2.2) -20.9 (0.4)
methanol -41.9 (3.1) -59.7 (2.9) -37.8 (2.6)* -27.5 (0.5) -30.7 (1.0)* -68.2 (11.2) -16.9 (1.3)
ethanol -41.5 (0.8) -37.5 (0.3) -30.6 (0.2) -36.5 (0.2)* -29.4 (2.6) -5.0 (0.4)
propanal -25.6 (2.7)
isoprene off scale b -26.9 (3.7) -35.2 (3.5) -33.8 (2.6) -23.0 (2.6)* -16.7 (1.2) -32.6 (0.9)*
acetone -35.7 (4.1) -37.4 (2.4) -32.8 (1.2) -38.8 (1.1) -29.3 (1.5)* -33.8 (0.8) -31.3 (0.8)* -28.1 (2.5) -25.6 (0.5)
2-pentanone -35.2 (1.4)
benzene -26.9 (0.3)
toluene -27.5 (0.6)
a Also included are values for prepped and incubated biogenic samples from Keppler et al.. 33 All values are reported as the average (standard
deviation). All samples n ) 5, except the fossil fuel source where n ) 3. All biogenic samples are wounded/clipped branches, except where noted
(*), which represents an intact branch on the sample specimen. The fossil fuel source was collected from a 1972 Scout International with no
catalytic converter at a constant cruise. b Isoprene was present; however, it saturated the detectors and the signal response was off scale and the
δ 13 C value could not be calculated.
6804 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Bulk carbohydrate analyses between the two plant types show
an enrichment of ∼15‰ in carbohydrates extracted from C4
plant material. 49 Investigations of industrially produced ethanol
originating from corn have been shown to have δ 13 C values of
-10.71 ± 0.31‰. 49 The values we observed in the Scout samples
are ∼5‰ heavier and, considering the widespread use of
ethanol (7.5 billion gallons are expected to be used in fuel by
2012 48 ), may serve as a tracer for transportation related sources
to the atmosphere.
Some biogenic samples in this study, such as sand live oak
and orange citrus, had substantially depleted values for methanol
and agree with incubated emissions from various deciduous trees
and grasses made by Keppler et al. 33 (Table 2). However, this
observation is not consistent across all samples and suggest that
variations in δ 13 C values may result from interspecies differences,
microbe interaction on the leaf’s surface, prey/injury
response, the potential presence of a methanol utilization
pathway which oxidizes methanol to formaldehyde and formic
acid/formate, 50,51 and other lesser known metabolic, formation,
and loss pathways within plants. 52 Finally, a wound response may
be observed between the clipped and intact philodendron and sea
grape samples. In one distinct case, acetaldehyde emitted from
clipped sea grape specimens were enriched by ∼4‰ compared
to the fossil fuel emissions.
Ambient Measurement Results. Considerable differences
in δ 13 C are observed between ambient sampling locations
(Table 3). Results from Miami International Airport are reflective
of an averaged value for fresh vehicular sources. The measured
δ 13 C range for airport samples is between -12.3 ± 3.7‰
(ethanol) to -35.3 ± 1.7‰ (3-pentanone). With the exception
of ethanol, which has a δ 13 C value consistent with its C4 plant
source, and 2- and 3-pentanone, the measured range at the
airport agrees with that established for NMHCs from transportation-related
sources by Rudolph, namely, -21.9 to
-31.3‰. 25 Our acetaldehyde value is consistent with the range
(49) Ishida-Fujii, K.; Goto, S.; Uemura, R.; Yamada, K.; Sato, M.; Yoshida, N.
Biosci., Biotechnol., Biochem. 2005, 69, 2193–2199.
(50) Cossins, E. A. Can. J. Biochem. 1964, 42, 1793–1802.
(51) Gout, E.; Aubert, S.; Bligny, R.; Rebeille, F.; Nonomura, A. R.; Benson, A. A.;
Douce, R. Plant Physiol. 2000, 123, 287–296.
(52) Fall, R. In Reactive Hydrocarbons in the Atmosphere; Hewitt, C. N., Ed.;
Academic Press: San Diego, CA, 1999; pp 43-97.

Table 3. Ambient Measurement Results for Samples
Collected from Metropolitan Miami and Everglades
National Park a
Miami
International
Airport
δ 13 C ± 1σ (‰)
Miami
Financial
District
Everglades
National
Park
acetaldehyde -26.7 ± 0.7 -26.8 ± 1.2 -19.0 ± 2.7
methanol -36.3 ± 3.7
ethanol -12.3 ± 3.7 -17.2 ± 4.1
isoprene -30.3 ± 2.1
propanal -28.4 ± 1.5 -26.2 ± 2.4
acetone -31.0 ± 3.5 -26.6 ± 0.4 -23.7 ± 0.4
MEK -28.3 ± 2.1 -25.9 ± 1.9
2-pentanone -34.8 ± 6.5 -29.4 ± 0.1
3-pentanone -35.3 ± 1.7 -37.8 ± 1.8
toluene -33.7 ± 2.0
a Miami International Airport, n ) 5; Miami financial district, n )
4; Everglades National Park, n ) 3.
presented by Wen et al. who measured values via a derivatization
procedure of ∼-21.0‰ and ∼-29.2‰ for samples
collected at a bus station and petrochemical refinery, respectively.
36
Some observations at Miami’s Financial District are between
2.2 and 4.4‰ enriched in 13 C compared to the same compounds
at Miami International Airport, and again we observe an
anomalously enriched value for ethanol (-17.2 ± 4.1‰).
Samples from the airport are general δ 13 C values we can expect
for OVOCs from transportation related sources without addition
from other sources and losses caused by solar radiation and
reaction with OH. Miami’s Financial District is located within
0.1 mile of Biscayne Bay and 1 mile of the Port of Miami and
was dominated by an onshore breeze during the sample
collection. Therefore, we can expect values from the financial
district to be enriched since the δ 13 C signature for each
compound will reflect a combination of vehicular, biogenic, and
possibly marine sources and, additionally, losses attributable
to reactivity with OH and photolysis. Isotopic values for samples
from the financial district are bound within the reported range
of -15.8 to -37.4‰ for NMHCs sampled at a moderately
polluted waterfront in Wellington, New Zealand. 25
In comparison with automobile exhaust (Table 2), the mean
values observed at Miami International Airport and Miami’s
financial district are generally depleted in 13 C. The two most
obvious differences among these samples that may influence
the observations are the fuel source and the presence of a
catalytic converter. Emissions collected at the airport are a mix
of refined petroleum and diesel, whereas the Scout International
was fueled by unleaded gasoline. Furthermore, vehicle
emissions at the airport are assumed to be produced by engines
having a catalytic converter. However, the Scout lacked a
converter, and the speeds of the engines producing the
emissions were very different. Traffic through the airport’s
lower roadway moved at an idle pace and rarely exceeded 15
mph. The Scout samples were obtained with the engine under
significant load and at a constant revolution per minute (2000
rpm) and cruise speed (80 kph). To our knowledge, no studies
exist showing how the presence of a catalytic converter or
engine speed may influence the δ 13 C of emitted hydrocarbons.
Samples from Everglades National Park spanned a large range
from -19.0 to -36.3‰. Measured methanol from within the
National Park was -36.3 ± 3.7‰, considerably depleted and
consistent with other values obtained in the tropical plant
enclosure studies (i.e., sand live oak δ13CMethanol )-41.9 ± 3.1‰)
and with the results presented earlier from Keppler et al. 33
Similarly, δ13C values for isoprene released from C3 plants
range from -26 to -29‰. 45 Isoprene values at the National
Park are lighter (-30.3 ± 2.1‰) than the range presented by
Rudolph et al. However, when the precision of the measurement
is considered, the isoprene values measured from the
Everglades’ samples overlap the range observed with that
previous work. Acetone and acetaldehyde values from within
the National Park are more enriched than anticipated. The
mean δ13C values for these compounds are -23.7‰ and
-19.0‰, respectively. Each are enriched approximately 7.5‰
compared to samples collected at Miami International Airport
and are fairly consistent with samples from Miami’s financial
district and fossil fuel combustion.
When estimated atmospheric lifetimes (τ) are considered for
these compounds in the troposphere for losses caused by
OH OH reactivity with OH (τacetone ) 66 days; τacetaldehyde ) 11 h) and
hν hν photolysis (τacetone ) 38 days; τacetaldehyde ) 5 days), these
observations can be explained, especially for the enrichment
of acetaldehyde over acetone (∼5‰). Few studies of ambient
δ13C for acetaldehyde exist, 29,36 and only one exists for acetone. 37
For samples collected within a biosphere reserve in China, Guo
et al. measured acetaldehyde values between -31.6 and -34.9‰.
These values are depleted in 13C compared to our measurements.
However, they report weak photolytic loss of formaldehyde in
the same study, and considering formaldehyde’s lifetime
against photolysis is shorter (4 h) compared to acetaldehyde
(5 days), we assume this to be true for acetaldehyde at the
same location.
Guo et al. used a derivatization method to calculate δ13C values
for acetone collected at a forested site (∼-31‰) and at the
top of a 10 m building influenced by vehicle emissions
(∼-26‰). The acetone values from Everglades National Park
are enriched by 2-7‰ compared to the values presented by
Guo et al. Isotopic values obtained from the forest may reflect
the signature of fresh acetone emissions from biomass, while
values for Everglades National Park samples may be more
strongly influenced by photochemistry. The measured values
for acetone and acetaldehyde from within the Everglades may
also indicate contributions from in situ atmospheric production
via oxidation and photolysis of higher order hydrocarbons. An
exact assessment to separate direct emissions from photochemical
production and loss is not possible at this time since
fractionations associated with these pathways are not known.
CONCLUSIONS
A new method for measuring δ 13 C values of low-molecular
weight OVOCs from direct sources and ambient samples was
developed. The method incorporated a carbon sorbent, a lowvolume
capillary reactor, water trap, and balanced working
reference gas delivery system. The method’s total precision
ranged between 0.6 and 2.9‰, and negligible sample fractionation
occurred while sampling and trapping gases. Further
testing showed that measured δ 13 C values had little dependence
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6805

on sample size (0.06 ‰ ng C 1- ), and linearity was best over
the range of 1-10 ng C. The method was sensitive, requiring
>0.2 ng C into the ion source to produce accurate and precise
results. The analysis of ambient samples required small sample
volumes, with ∼1.0 L of gas providing sufficient carbon for
analysis.
Clear distinctions in δ 13 C were observed between emissions
released from plants and automobiles. In particular, ethanol
emissions from automotive exhaust and metropolitan Miami
were significantly enriched in 13 C. This is related to ethanol’s
C4 plant origin and use as a fuel additive. Ambient samples
can be differentiated, but the variation in δ 13 C values was not
as great as for the source samples. Ambient samples suffer
from additional complexity with multiple sources and sinks
affecting single sampling locations. Clearly, more studies of
sources and ambient sampling are required to define and
characterize OVOCs in the troposphere along with laboratory
studies to determine the kinetic isotope effects associated with
OVOCs’ in situ production and loss from reaction with OH and
photolysis. As it stands now, this technique can be used to
differentiate OVOC sources and to assess the carbon isotopic
6806 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
values for OVOCs in ambient air. It should serve as a useful
way to investigate transformations of organic gases in the
atmosphere.
ACKNOWLEDGMENT
We thank Tom Brenna and Herbert Tobias for helpful discussions
in developing this method and Rich Iannone for providing
a template for raw data calculations. We acknowledge John Mak
and Zhihui Wang for the working reference gas interlab comparison.
We appreciate the efforts of Kevin Polk and his 1972
International Scout. Finally, we gratefully acknowledge the helpful
comments made by two anonymous reviewers and support
provided by NSF Grant No. 0450939.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review March 23, 2010. Accepted July 6,
2010.
AC1007442

Anal. Chem. 2010, 82, 6807–6813
Direct Voltammetric Analysis of DNA Modified with
Enzymatically Incorporated 7-Deazapurines
Hana Pivoňková, † Petra Horáková, †,‡ Miloslava Fojtová, †,§ and Miroslav Fojta* ,†
Institute of Biophysics, v.v.i., Academy of Sciences of the Czech Republic, Královopolská 135,
CZ-612 65 Brno, Czech Republic, Department of Analytical Chemistry, Faculty of Chemical Technology, University of
Pardubice, Studentská 573, CZ-532 10 Pardubice, Czech Republic, and Department of Functional Genomics and
Proteomics, Institute of Experimental Biology, Faculty of Science, Masaryk University, Kotlárˇská 2,
CZ-611 37 Brno, Czech Republic
Nucleic acids studies use 7-deazaguanine (G*) and 7-deazaadenine
(A*) as analogues of natural purine bases
incapable of forming Hoogsteen base pairs, which prevents
them from being involved in DNA triplexes and
tetraplexes. Reduced propensity of the G*- and/or A*modified
DNA to form alternative DNA structures is
utilized, for example, in PCR amplification of guanine-rich
sequences. Both G* and A* exhibit significantly lower
potentials of their oxidation, compared to the respective
natural nucleobases. At carbon electrodes, A* yields an
oxidation peak which is by about 200-250 mV less
positive than the peak due to adenine, but coincides with
oxidation peak produced by natural guanine residues. On
the other hand, oxidation signal of G* occurs at a potential
by about 300 mV less positive than the peak due to
guanine, being well separated from electrochemical signals
of any natural DNA component. We show that
enzymatic incorporation of G* and A* can easily be
monitored by simple ex situ voltammetric analysis of the
modified DNA at carbon electrodes. Particularly G* is
shown as an attractive electroactive marker for DNA,
efficiently incorporable by PCR. While densely G*-modified
DNA fragments exhibit strong quenching of fluorescence
of SYBR dyes, commonly used as fluorescent
indicators in both gel staining and real time PCR applications,
the electrochemical detection provides G*-specific
signal suitable for the quantitation of the amplified DNA
as well as for the determination of the DNA modification
extent. Determination of DNA amplicons based on the
measurement of peak G* ox is not affected by signals
produced by residual oligonucleotide primers or primary
templates containing natural purines.
Electrochemical techniques are increasingly applied in the area
of nucleic acids sensing (reviewed in refs 1-4). Nucleic acids
* To whom correspondence should be addressed. E-mail: fojta@ibp.cz .
† Academy of Sciences of the Czech Republic.
‡ University of Pardubice.
§ Masaryk University.
(1) Fojta, M. In Electrochemistry of Nucleic Acids and Proteins. Towards
Electrochemical Sensors for Genomics and Proteomics; Palecek, E., Scheller,
F., Wang, J., Eds.; Elsevier: Amsterdam, 2005, pp 386-431.
(2) Fojta, M.; Jelen, F.; Havran, L.; Palecek, E. Curr Anal Chem 2008, 4, 250–
262.
possess intrinsic electroactivity due to the presence of electrochemically
oxidizable or reducible nucleobases, 2,4 making it
possible to analyze them electrochemically without any labeling.
Indeed, various label-free electrochemical techniques were proposed
for the detection of DNA damage 1 or DNA hybridization. 3
In spite of these efforts, application of various redox indicators
and labels proved useful particularly in sequence-specific DNA
sensing requiring reliable discrimination between two complementary
strands (e.g., target DNA and hybridization probe 5,6 ),
specific determination of newly synthesized pieces or fragments
of DNA (in primer extension or PCR-based assays 7-9 ) or even
identification of a single nucleobase incorporated at a specific
position (in SNP typing 9-11 ). Generally, introducing electroactive
tags producing “new” specific electrochemical responses (not
yielded by natural DNA components) increases specificity of the
assays considerably. Electrochemically active moieties can be
incorporated into nucleic acids during chemical oligonucleotide
synthesis, postsynthetically by chemical modification of “natural”
nucleic acids 5,6 or using modified nucleoside triphosphates
(dNTPs) and DNA polymerases. 7-12 The latter approach represents
a versatile way to facile construction of labeled or otherwise
functionalized nucleic acids and to efficient sequence-specific DNA
sensing. 12 A critical prerequisite for these applications is the ability
of a DNA polymerase to use modified dNTPs as substrates for
efficient incorporation without losing sequence-specificity. C7substituted
7-deazapurines and C5-substituted pyrimidines are
usually acceptable substrates for (at least some) DNA polymerases
(3) Palecek, E.; Fojta, M. Talanta 2007, 74, 276–290.
(4) Palecek, E.; Jelen, F. In Electrochemistry of Nucleic Acids and Proteins.
Towards Electrochemical Sensors for Genomics and Proteomics.; Palecek, E.,
Scheller, F., Wang, J., Eds.; Elsevier: Amsterdam, 2005, pp 74-174.
(5) Flechsig, G. U.; Reske, T. Anal. Chem. 2007, 79, 2125–2130.
(6) Fojta, M.; Kostecka, P.; Trefulka, M.; Havran, L.; Palecek, E. Anal. Chem.
2007, 79, 1022–1029.
(7) Brazdilova, P.; Vrabel, M.; Pohl, R.; Pivonkova, H.; Havran, L.; Hocek, M.;
Fojta, M. Chem.-Eur. J. 2007, 13, 9527–9533.
(8) Patolsky, F.; Weizmann, Y.; Willner, I. J. Am. Chem. Soc. 2002, 124, 770–
772.
(9) Vrabel, M.; Horakova, P.; Pivonkova, H.; Kalachova, L.; Cernocka, H.;
Cahova, H.; Pohl, R.; Sebest, P.; Havran, L.; Hocek, M.; Fojta, M. Chem.-
Eur. J. 2009, 15, 1144–1154.
(10) Cahova, H.; Havran, L.; Brazdilova, P.; Pivonkova, H.; Pohl, R.; Fojta, M.;
Hocek, M. Angew. Chem., Int. Ed. 2008, 47, 2059–2062.
(11) Horakova, P.; Simkova, E.; Vychodilova, Z.; Brazdova, M.; Fojta, M.
Electroanalysis 2009, 21, 1723–1729.
(12) Hocek, M.; Fojta, M. Org. Biomol. Chem. 2008, 6, 2233–2241.
10.1021/ac100757v © 2010 American Chemical Society 6807
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/23/2010

Scheme 1. Top: Formulas of Purine Nucleobases (A,
G) and Their 7-Deaza Analogues (A*, G*) a
in primer extension (PEX) experiments. 12,13 Nevertheless, many
of the nucleobase conjugates have displayed less facile incorporation
at adjacent positions 7,9,10 and only some of them have been
efficiently used in PCR. 13,14
The 7-deazapurines, 7-deazaguanine (G*), or 7-deazaadenine
(A*) (Scheme 1) are substitutes of standard purine nucleobases
incorporable into DNA by PCR. Although the rate of G* and A*
incorporation by DNA polymerases has been reported 15 to be
lower, compared to the “parent” purines, both of these nucleobase
analogues allow efficient sequence-specific DNA amplification by
PCR and it has been possible to prepare PCR products with a
high density of the corresponding modification. The 7-deazapurines
are able to form Watson-Crick base pairs, maintaining
pairing specificity of the respective natural nucleobases, but cannot
form Hoogsteen pairs due to absence of the N7 atom (which is
substituted by CH group, Scheme 1). Thus, the 7-deazapurines
cannot be involved in triplex and tetraplex DNA structures. 16
Reduced tendency of the “deaza-modified” DNA to adopting
multistranded conformations has been utilized to improve PCR
amplification of G-rich sequences such as (CGG)n repeat, 17 the
length of which is analyzed during molecular diagnostics of
fragile X syndrome. Similarly, 7-deaza-8-azaguanine has been
used to create G-rich DNA probes with reduced propensity to
aggregate and improved specificity. 18 Absence of the nitrogen
atom at the 7-position in G* was also utilized in a study of
sequence-specificity of DNA modification with cisplatin 19 and
in studies of echinomycin-DNA interaction modes. 20 Moreover,
DNA substituted with G* or A* has been shown to resist
cleavage with certain endonucleases. 15
a
Atoms at 7-position are highlighted in red. Bottom: Watson-Crick
and Hoogsteen base pairing. N7 is involved in the Hoogsteen pairing
of natural purines.
(13) Cahova, H.; Pohl, R.; Bednarova, L.; Novakova, K.; Cvacka, J.; Hocek, M.
Org. Biomol. Chem. 2008, 6, 3657–3660.
(14) Raindlova, V.; Pohl, R.; Sanda, M.; Hocek, M. Angew. Chem., Int. Ed. 2010,
49, 1064–1066.
(15) Seela, F.; Roling, A. Nucleic Acids Res. 1992, 20, 55–61.
(16) Palecek, E. Crit. Rev. Biochem. Mol. Biol. 1991, 26, 151–226.
(17) Cao, J.; Tarleton, J.; Barberio, D.; Davidow, L. S. Mol. Cell. Probes 1994,
8, 177–180.
(18) Kutyavin, I. V.; Lokhov, S. G.; Afonina, I. A.; Dempcy, R.; Gall, A. A.; Gorn,
V. V.; Lukhtanov, E.; Metcalf, M.; Mills, A.; Reed, M. W.; Sanders, S.;
Shishkina, I.; Vermeulen, N. M. J. Nucleic Acids Res. 2002, 30, 4952–4959.
(19) Cairns, M. J.; Murray, V. Biochim. Biophys. Acta 1994, 1218, 315–321.
(20) Sayers, E. W.; Waring, M. J. Biochemistry 1993, 32, 9094–9107.
6808 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Table 1. Oligonucleotides Used in This Work a
prim rnd 5′-CATGGGCGGCATGGG-3′
prim noG 5′-TACTCATCATATCAA-3′
temp rnd16 5′-CTAGCATGAGCTCAGTCCCATGCCGCCCATG-3′
temp noG 5′-AATATAAATATATTGATATGATGAGTA-3′
p53-for 5′-GAGGTTGTGAGGCGCTGCCC-3′
p53-rev 5′-TCCTCTGTGCGCCGGTCTCT-3′
a The template strands (temp rnd16 , temp noG ) used in PEX experiments
were 5′end-biotinylated.
Compared to the standard purines, 7-deazapurines exhibit
significantly lower potentials of their oxidation. 9,21,22 G* is thus
easily photooxidized by intercalated ethidium and this photooxidation
has been interrogated as a function of distance, nucleotide
sequence and integrity of π-stacking in studies of DNA-mediated
charge transfer. 21 The G* ability of being selectively oxidized by
a redox mediator with relatively low redox potential, such as
Ru(dmb)3 3+/2+ (dmb )4,4′-dimethyl-2,2′-bipyridine), compared
to A* together with natural G requiring stronger oxidants, such
as Ru(bpy)3 3+/2+ (bpy )2,2′-bipyridine), has been utilized in a
PCR-coupled electrochemical technique proposed for parallel
detection of two genes. 22 Mediated electrooxidation of G* (or
G) was also applied in an indirect electrochemical real time
monitoring of PCR via measuring consumption of the respective
dNTPs. 23 To our best knowledge, direct electrochemical
analysis of DNA with incorporated G* or A* as electrochemically
oxidizable tags has not been reported to date.
In this paper we report on adsorptive transfer stripping
voltammetric analysis of DNA with enzymatically incorporated G*
or A* residues at a carbon electrode. We show that, particularly
G* producing a specific signal separated from those yielded by
natural DNA components, can be utilized as an excellent electroactive
label suitable for monitoring of DNA amplification by PCR.
MATERIALS AND METHODS
Material. Synthetic ODNs (Table 1) were purchased from
VBC genomics (Austria). Templates used in experiments involving
the magnetoseparation procedure were biotinylated at their 5′
ends. Plasmid pT77 bearing wild type p53 cDNA insert 24 (used
as primary template for PCR amplification of the 347-bp fragment)
was isolated from E. coli cells using Qiagen Plasmid Purification
Kit and linearized with EcoR I restrictase (Takara). Streptavidincoated
magnetic beads (MBstv) were purchased from Novagen
(Germany), DyNAzyme II DNA Polymerase from Finnzymes
(Finland), Pfu DNA Polymerase from Promega (U.S.), unmodified
nucleoside triphosphates (dATP, dTTP, dCTP and dGTP),
SYBR Green I, SYBR Gold and Stains-All reagent from Sigma,
7-deaza-dGTP and 7-deaza-dATP from Jena Bioscience. Other
chemicals were of analytical grade.
Primer Extension (PEX). The primer (0.7 µM) was mixed
with corresponding template ODN (0.7 µM), dNTPs (100 µM
each; composition of the dNTP is specified in the text and Figure
legends for individual experiments) and the DyNAzyme II DNA
(21) Kelley, S. O.; Barton, J. K. Chem. Biol. 1998, 5, 413–425.
(22) Yang, I. V.; Ropp, P. A.; Thorp, H. H. Anal. Chem. 2002, 74, 347–354.
(23) Defever, T.; Druet, M.; Rochelet-Dequaire, M.; Joannes, M.; Grossiord, C.;
Limoges, B.; Marchal, D. J. Am. Chem. Soc. 2009, 131, 11433–11441.
(24) Hupp, T. R.; Meek, D. W.; Midgley, C. A.; Lane, D. P. Cell 1992, 71, 875–
886.

Polymerase (1 U per sample). Reactions were carried out at 60 °C
for 30 min.
MBstv Magnetoseparation Procedure. The PEX products
were captured at MBstv via biotin tags. Then, 50 µL aliquots of
the PEX reaction mixtures were added to the MBstv (25 µL of
the stock suspension washed twice with 100 µL of 0.3 M NaCl,
10 mM Tris-HCl, pH 7.4, buffer H). The mixture was incubated
on a shaker for 30 min at 20 °C. Then the beads were washed
three times with 100 µL of PBS (0.14 M NaCl, 3 mM KCl, 4
mM sodium phosphate, pH 7.4) containing 0.01% Tween20,
three times with 100 µL of the buffer H and resuspended in
deionized water (50 µL). The extended primers were released
by heating at 75 °C for 2 min. Prior to the electrochemical
measurements, NaCl in total concentration of 0.2 M was added
to the samples.
Preparative PCR. Amplification of the DNA fragment: 500
ng of the pT77 template was mixed with p53-for and p53-rev
primers (0.5 µM each), Pfu DNA Polymerase (3 U) and mix of
dNTPs (125 µM each) in total volume of 100 µL. The PCR involved
30 cycles if not stated otherwise (denaturation 94 °C/90 s,
annealing 60 °C/120 s, polymerization 72 °C/180 s) and was run
using C1000 Thermal Cycler (BioRad). The PCR products were
purified using QIAquick PCR Purification Kit (Qiagen) and their
concentrations were determined spectrophotometrically using
NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies,
U.S.).
Real Time PCR. Reactions were prepared in duplicates in 20
µL handmade PCR mix consisting of 30 ng of pT77 template, p53for
and p53-rev primers at concentrations 0.5 µM, 1 × DyNAzyme
II buffer, 1.25 mM MgCl 2,20000× diluted SYBR Green I, each
dNTP (or sum of dGTP+dG*TP) 125 µM,1UofDyNAzyme
II DNA polymerase. PCR involved 20 cycles (denaturation
94 °C/30 s, annealing 56 °C/30 s, polymerization 72 °C/30 s,
fluorescence measurement in SYBR Green I channel 15 s/78 °C,
wavelength of the source 470 nm, wavelength of the detection
filter 585 nm; final extension 72 °C/3 min) and was run and
evaluated using the RotorGene-3000 (Qiagen, Germany).
Native PAGE. The PCR products (1 µL each) were mixed
with loading buffer (0.1% SDS, 5% glycerol, 5 mg mL -1 bromophenol
blue) and subjected to electrophoresis in 5% native gel
containing 1 × TBE buffer (pH 8) at 150 V, 4 °C for 35 min.
Gels stained with ethidium bromide or SYBR dyes were
visualized using LAS-3000 (FUJIFILM Corporation), those
stained with the Stains-All reagent were scanned.
Electrochemical Analysis. The PEX and PCR products were
analyzed by using ex situ (adsorptive transfer stripping, AdTS)
square-wave voltammetry (SWV). The DNA was accumulated at
the basal-plane pyrolytic graphite electrode (PGE; prepared and
pretreated as described in ref 25) surface from 5 µL aliquots
containing 0.3 M NaCl for 60 s. Then the electrode was rinsed by
deionized water and was placed into the electrochemical cell. SWV
settings: initial potential -1.0 V, final potential +1.5 V, pulse
amplitude 25 mV, frequency 200 Hz, potential step 5 mV. The
measurements were performed at ambient temperature in 0.2 M
acetate buffer pH 5 by using CHI440 Electrochemical Workstation
(CH Instruments, Inc., U.S.) in a three-electrode setup (with the
PGE as working electrode, Ag/AgCl/3 M KCl as reference, and
(25) Fojta, M.; Havran, L.; Kizek, R.; Billova, S. Talanta 2002, 56, 867–874.
platinum wire as counter electrode). Baseline correction of the
voltammograms was performed by means of a moving average
algorithm (GPES 4 software, EcoChemie).
RESULTS AND DISCUSSION
Primer extension (PEX) has recently been used for the
introduction of labeled nucleobases into oligodeoxynucleotides
(ODNs). 7-13 Using nucleobase-modified deoxynucleotide triphosphates
(dNTPs) and DNA polymerases, we prepared ODNs
bearing various tags producing analytically useful electrochemical
responses. Since 7-deazapurines are electrochemically oxidized
at considerably less positive potentials, compared to the respective
natural purine nucleobases, 9,21,22 it is in principle possible to apply
these nucleobase analogues themselves (without introducing any
extra label groups) as electroactive markers of the in vitro synthesized
DNA.
We performed PEX with primrnd primer and 5′-end-biotinylated
temprnd16 template, the nucleotide sequence of which was
designed to accommodate all four DNA bases (four-times each)
in the synthesized DNA stretch (Figure 1A). The doublestranded
PEX products were pulled-down from the reaction
mixture using streptavidin coated magnetic beads (MBstv) and,
after magnetic separation, the captured duplex ODN was
thermally denatured to release the extended primer strand. The
latter was finally analyzed using adsorptive transfer stripping
square wave voltammetry (AdTS SWV). As shown in Figure
1B,C, voltammetric responses of the pexrnd16 products reflected
the composition of dNTP mix used. For a standard dNTP mix
(dATP, dGTP, dCTP, and dTTP), we observed two anodic
signals corresponding to electrochemical oxidation of guanine
(peak Gox at 1.15 V) and adenine (peak Aox at 1.35 V, Figure
1B,C). When dGTP was replaced by 7-deaza-dGTP (dG*TP), peak
Gox was decreased and a new anodic peak appeared at 0.80 V
(peak G* ox , Figure 1B,C) which has been ascribed to electrochemical
oxidation of G* introduced into the extended ODN
stretch. Lower intensity of the peak Gox corresponded to lower
number of G residues in the G*-modified PEX product,
compared to the PEX product composed of standard nucleotides
(note that after PEX in the presence of dG*TP, there
are still natural guanines present in the primer stretch, but not
in the extended stretch, decreasing the total number of G
residues from 12 to 8). When dATP was replaced by 7-deazadATP
(dA*TP), we observed a decrease of the peak Aox intensity (due to lowering of the total number of adenines in
the entire extended primer strand) and increase of the peak
at 1.15 V. The potential of electrochemical oxidation of A* was
shown 9 to coincide with the potential of peak Gox , and thus
currents due to A* oxidation were responsible for the increase
of the apparent signal intensity. To verify this hypothesis, we
prepared another PEX product using primnoG primer and
tempnoG template (Table 1). Neither of these ODNs contains G
residues and no G is incorporated during PEX, which is reflected
in the absence of peak Gox at the voltammogram of unmodified
pexnoG product, showing only peak Aox (black curve in Figure
1C, inset). After performing PEX with the same primer-template
pair but with dATP replaced by dA*TP, the peak Aox was
decreased and another signal appeared at 1.15 V, corresponding
to the A* oxidation (peak A* ox ; green curve in Figure 1C, inset).
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6809

Figure 1. Electrochemical responses of oligonucleotides containing 7-deazapurines incorporated by primer extension (PEX). (A) Scheme of
the experiment. The PEX was performed using a 5′-terminally biotinylated template. After extension of the primer using either standard dNTP
mix, a mixture with dGTP replaced by dG*TP, or a mixture with dATP replaced by dA*TP, the duplex ODN was captured at magnetic beads
covered with streptavidin, the beads were washed and the modified strand released by thermal denaturation, followed by AdTS SWV measurements
at the PGE (shown for temp rnd16 and prim rnd sequences; blue and red letters are used to highlight A and G positions, respectively). (B) SWV
voltammograms of pex rnd16 : standard nucleobases (black); A* instead of A (blue); G* instead of G (red); background electrolyte (dotted); (C)
baseline-corrected curves taken from (B). Inset: PEX products obtained with temp noG and prim noG for dTTP+dATP mix (black) or dTTP+dA*TP
mix (green).
In next experiment, we prepared a 347-bp DNA fragment by
the polymerase chain reaction (PCR) using the pT77 template and
various dNTP mixtures. The PCR products were isolated from
the reaction mixture using Qiagen columns and analyzed by AdTS
SWV as above. During thermal cycling in the PCR, forward and
backward primers (Table 1) are elongated on templates of both
DNA strands, resulting in exponential amplification of the fragment
delimited by the two primers (see Figure 2A). Thus, each
strand of the double-stranded PCR product begins with the primer
at its 5′-terminus, followed by newly synthesized stretch, the
nucleotide composition of which (i.e., presence or absence of the
7-deazapurines) is dictated by the composition of the dNTP mix
(Figure 2A). Accordingly, when using mix of four standard dNTPs
(in Figure 2B denoted as G+A), we observed peak G ox and peak
A ox corresponding to natural purine bases. Replacement of
dGTP with dG*TP resulted in appearance of an intense peak
G* ox (due to G* incorporated into the polymerase-synthesized
strands) and strong decrease of the intensity of peak G ox which,
however, never reached zero. This peak G ox was due to G
residues in the primer stretches of the G*-modified amplicons
(Figure 2A). In addition, certain amount of unconsumed primers
and primary templates remaining in the samples after the
6810 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
purification step might contribute to the peak G ox intensity, as
indicated by negative PCR control experiment (PCR mixture
not subjected to thermal cycling, dotted curve in Figure 2B).
(It should be however noted that contributions to the peak G ox
from ODN primers in the negative PCR control, as apparent
from Figure 2B, and those from primer stretches in the G*modified
amplicon are not simply additive, because concentration
of residual free primers after the PCR cycling is much lower than
their initial concentration). When dATP was replaced with dA*TP,
we observed increase of the signal close to 1.15 V (peak G ox +
peak A* ox ) and decrease of peak A ox (again, the residual peak
A ox was yielded by A residues in the primer stretches).
Further, we focused our attention on analysis of DNA amplicons
containing G* as an independently detectable marker and
performed PCR experiments using dNTP mixes containing both
dGTP and dG*TP at different ratios (while keeping the total dGTP
+ dG*TP concentration constant). The resulting PCR products
were analyzed electrochemically as above. As shown in Figure 3,
changes in the relative abundances of G and G* in the PCR
product (given by the molar fraction of the respective dNTP) were
reflected in changes of corresponding electrochemical signals
(peak G ox and peak G* ox ). Peak A ox due to adenine, the content

of which was constant for all amplicons analyzed in this
experiment, did not show any significant trend, suggesting
approximately constant yields of the PCR products obtained
for any dGTP/dG*TP ratio after the 30 PCR cycles (which was
also confirmed by UV-vis spectrophotometry, not shown).
Dependences of the heights of peak Gox (yielded by the
unmodified 347-bp PCR product) and peak G* ox (yielded by
the same DNA fragment amplified with 100% of dG*TP) on
concentrations of the amplicons followed similar trends (Figure
4A), showing approximately linear regions below 10 µgmL-1and sublinear trends, suggesting saturation of the electrode surface,
at higher DNA concentrations. The lowest detectable DNA
concentrations were around 1 µg mL-1 in both cases. Electrochemical
determination of the modified and unmodified 347bp
amplicons was compared to other commonly used techniques,
based on gel electrophoresis followed by DNA staining
(Figure 4B). The amplicons were separated in native 5% polyacrylamide
gel (1 µL of the reaction mixture, containing about 30
µg mL-1 Figure 2. Electrochemical responses of a PCR-amplified 347-bp
DNA fragment modified with 7-deazapurines. (A) Scheme of the PCR.
Primers at the 5′-ends of both strands of the PCR product (black)
contain only standard nucleobases, regardless of the dNTP composition.
The synthesized stretches contain modified nucleobases depending
on the dNTP mix. (B) Baseline-corrected voltammograms
obtained for DNA fragment resulting from PCR in the presence of
standard dNTP mix (black), for a mix with G* instead of G (red) and
for a mix with A* instead of A (blue). The PCR was conducted in 30
cycles and the products were purified using Qiagen PCR Purification
Kit. Dotted curve corresponds to control PCR mixture (with G*+A)
which was not subjected to thermal cycling.
of the amplicon, and three binary dilutions). After
electrophoresis, the gels were stained with ethidium bromide,
Figure 3. Dependence of the heights of peak G* ox (red), peak G ox
(black) and peak A ox (empty triangles) on the [dG*TP]/[dGTP]+[dG*TP]
ratio in the PCR reaction used for amplification of the 347-bp DNA
fragment. Other conditions as in Figure 2.
SYBR Green I or Stains-All reagent (Figure 4B). Ethidiumstained
bands of the G*-modified PCR products were considerably
weaker than those of the unmodified amplicon (even when only
50% of Gs were substituted by G*, see Figure 4B). Such
observation was in agreement with literature data 26 showing that
ethidium fluorescence is quenched when the dye is intercalated
next to G*. Notably, we observed even stronger quenching effect
of G* on the fluorescence of SYBR Green I (Figure 4B) and SYBR
Gold (not shown) dyes. Thus, results of the fluorescent DNA
staining might be misinterpreted, for the densely G*-modified
DNA, in terms of (strongly) decreasing amount of the PCR
products with increasing G/G* ratio. On the other hand, staining
of the polyacrylamide gel with the Stains-All reagent (Figure 4B)
did not reveal significant differences in the amounts of the
unmodified and G*-substituted amplicons, in agreement with the
electrochemical data.
Despite the approximately same yields of the PCR products
after 30 cycles, we were interested whether we are able to follow
electrochemically differences in the kinetics of the PCR reactions
through analysis of the amplicons after lower number of amplification
cycles. Previous data 15 revealed less efficient DNA amplification
by PCR in the presence of dG*TPs, compared to PCR with
standard dNTPs only. We followed electrochemical signals of the
unmodified and fully G*-substituted PCR products after 5, 10, 20,
and 30 cycles (Figure 5A). For five cycles, peak G ox produced
by the unmodified amplicon was considerably higher than peak
G* ox of the G*-modified PCR product. Even after subtraction
of signal intensity produced by the control PCR mix not
subjected to the thermal cycling (containing initial concentrations
of ODN primers and the primary template), the peak G ox
was at least twice higher than peak G* ox produced by the G*modified
amplicon after the same number of cycles. Large
differences between the signal intensities were also observed
after 10 amplification cycles, while after 20 and 30 cycles both
peaks reached their limiting values. Hence, more cycles were
required to reach the limiting amount of the G*-modified PCR
(26) Latimer, L. J. P.; Lee, J. S. J. Biol. Chem. 1991, 266, 13849–13851.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6811

Figure 4. (A) Dependence of the intensities of peak G ox (measured
for unmodified 347-bp amplicon, black) and peak G* ox (measured for
the same DNA fragment amplified in the presence of dG*TP instead
of dGTP, red) on DNA concentration. Other conditions as in Figure
2. (B) Staining of the 347-bp amplicon in polyacryalmide gels with
(from top to bottom) ethidium bromide, SYBR Green I and Stains-
All. The PCR reaction contained either four standard dNTPs, G/G*
) 1, or dGTP fully replaced with dG*TP as indicated on the top.
Loading of the PCR products: 1 µL of the undiluted reaction mixture
(about 30 ng of the amplicon), followed by binary dilutions as
indicated.
product, compared to PCR with standard dNTP mix. To support
this conclusion, we performed a real-time PCR experiment
using standard and dG*TP-containing dNTP mixes. Figure 5B
shows increase of the fluorescence of SYBR Green I (which was
used here as fluorescent dye indicating progression of DNA
amplification) as a function of the number of PCR cycles for dNTP
mixes containing 0, 10, 20, 30, 50, and 100% of dG*TP (of
dGTP+dG*TP total). The steepest increase of the signal was
observed for the unmodified amplicon and the amplification rate
decreased with the G* content, showing clear difference even for
10% G*. For 100% of G* the apparent amplification rate was
remarkably depressed. However, it should be taken into consideration
that SYBR Green I fluorescence is quenched by G*
incorporated (see above), and thus the weak signal detected for
the G*-substituted amplicon can have reflected the dense DNA
modification rather than the amount of the PCR product. We
therefore focused on the evaluation of the shapes of the amplification
curves rather than the absolute signal magnitudes. RotorGene-
3000 software enables to compare reactions parameters in the
6812 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. (A) Effects of the number of PCR cycles on intensities of
peak G ox (measured for unmodified 347-bp amplicon, gray columns)
and peak G* ox (measured for the same DNA fragment amplified in
the presence of dG*TP instead of dGTP, red columns). Other
conditions as in Figure 2. (B) Quantitative (real time) PCR amplification
of the 347-bp DNA fragment: standard dNTP mix (no G*) (1);
10% G* (2); 20% G*(3); 30% G* (4); 50% G (5); G fully replaced with
G* (6); negative “no template” control for standard dNTP mix (7). The
graph shows fluorescence of SYBR Green I complexes with the DNA
amplicons as a function of the number of PCR cycles. Inset, takeoff
graphs derived from the real time PCR data (colors correspond to
the raw data plot). All samples are plotted in duplicate.
comparative quantification mode. The takeoff points for each
reaction (sample) are calculated from the second derivatives of
raw data (Figure 5B, inset). Generally, the takeoff value is not
possible to determine exactly and it is taken as 80% below the
peak of second derivative (i.e., below the point where the
amplification curve is increasing most steeply). In the same mode,
reaction efficiencies of particular reactions are reporting the
amplificability of the template under the given conditions. For
example, amplification factors of 1.72, 1.65, and 1.61 obtained for
reactions with 100% G, G/G* ) 1 and 100% G*, respectively, clearly
show less efficient amplification in the dG*TP-substituted reactions.
Similar effects of G* on the PCR kinetics were obtained
using Taqman hydrolytic probes which release fluorescent indicators
into solution as the PCR progresses, and thus the fluorescence
intensity was not affected by the incorporated G* as it was in the
case of the intercalative dyes (not shown). Hence, results of the
real time PCR experiment were in a qualitative accordance with
the above electrochemical data indicating lower amplification rate
for the reaction with dG*TP.
CONCLUSIONS
7-deazapurines G* and A* enzymatically incorporated into DNA
are electrochemically oxidizable at carbon electrodes, producing

analytically useful signals at less positive potentials, compared to
the corresponding natural purine nucleobases. Peak A* ox due to
electrooxidation of A* occurs at the same potential as the peak
G ox due to oxidation of natural G, preventing qualitative
discrimination of A* incorporated into DNA containing guanine.
On the other hand, total or partial substitution of G with G*
results in appearance of a new anodic signal, peak G* ox ,at
potential less positive than potentials of oxidation of any natural
component of DNA, allowing independent determination of G*
incorporated. G* can thus be utilized as an inexpensive,
commercially available electroactive label for easy monitoring
of primer extension or polymerase chain reactions via simple
direct electrochemistry using cheap, widely accessible carbon
electrodes. In turn, considering applications of 7-deazaguanine
as DNA modifications preventing formation of multistranded
alternative structures in PCR analysis of G-rich sequences 17
and problems with quenching of fluorescence of ethidium 26
or “SYBR” (to our knowledge, for the first time reported in
this paper) dyes, electrochemical analysis appears an attractive
complementary approach providing modification-specific signal
suitable for quantitation of the fully modified amplified DNA
as well as for the determination of the DNA modification extent.
Specifically for the G*-modified DNA, the voltammetric analysis
represents a simple and direct way to differentiate between the
(27) Fojta, M.; Brazdilova, P.; Cahova, K.; Pecinka, P. Electroanalysis 2006, 18,
141–151.
(28) Fojta, M.; Havran, L.; Vojtísˇková, M.; Palecek, E. J. Am. Chem. Soc. 2004,
126, 6532–6533.
natural G and G* residues and to determine relative content
of both. In contrast to using peak G ox or peak A ox , produced
by natural purines, determination of DNA amplicons based on
the measurement of peak G* ox is not affected by signals
produced by residual ODN primers and/or the primary
template. It has to be naturally taken into consideration that
peak G* ox intensity must depend on the G + C content within
the amplified region; alternatively, this feature may potentially
be useful for the determination of the G + C content or
estimation of the length of the amplified DNA fragment (e.g.,
triplet repeat expansion 27,28 ) provided that a proper normalization
of the signal intensity (such as fragment ends “counting” through
the intensity of natural G in primers) is used. Besides the PCR
applications, the G* electroactive is also potentially useful for taillabeling
of DNA probes for electrochemical hybridization assays
and other bioanalytical applications which are metter of our
ongoing research and will be reported elsewhere.
ACKNOWLEDGMENT
This work was supported by Grant Agency of the ASCR (grant
IAA400040901), by the ASCR (AV0Z50040507 and AV0Z50040702)
and by the MEYS CR (LC06035, MSM0021622415). H.P. and P.H.
contributed equally to this work.
Received for review March 24, 2010. Accepted July 10,
2010.
AC100757V
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6813

Anal. Chem. 2010, 82, 6814–6820
Screening Assay of Very Long Chain Fatty Acids in
Human Plasma with Multiwalled Carbon
Nanotube-Based Surface-Assisted Laser
Desorption/Ionization Mass Spectrometry
Wei-Yi Hsu, † Wei-De Lin, †,‡ Wuh-Liang Hwu, § Chien-Chen Lai,* ,‡,| and Fuu-Jen Tsai* ,†,|,&
Department of Medical Research, China Medical University Hospital, Taichung, Taiwan, Institute of Molecular Biology,
National Chung Hsing University, Taichung, Taiwan, Department of Medical Genetics, National Taiwan University
Hospital and National Taiwan University College of Medicine, Taipei, Taiwan, Graduate Institute of Chinese Medical
Science, China Medical University, Taichung, Taiwan, and Department of Health and Nutrition Biotechnology, Asia
University, Taichung, Taiwan
Peroxisomal disorders are characterized biochemically by
elevated levels of very long chain fatty acids (VLCFAs) in
serum. Herein, we describe a novel approach for quantification
of VLCFAs in serum, namely, eicosanoic acid
(C20:0), docosanoic acid (C22:0), tetracosanoic acid
(C24:0), and hexacosanoic acid (C26:0). The methodology
is based on (i) enrichment of VLCFA derivatives using
multiwalled carbon nanotubes (MWCNTs); (ii) quantification
using stable isotope-labeled internal standards; and
(iii) direct detection using MWCNT-based surface-assisted
laser desorption/ionization-time-of-flight mass spectrometry
(SALDI-TOFMS). Four kinds of MWCNTs (Aldrich
636843, 636495, 636509, and 636819) of different
lengths and diameters were tested using the developed
technique. The data show that 636843, the MWCNT with
the largest outer diameter (o.d.), the widest wall thickness,
and shortest length, had the best limit of detection
(0.5-1 µg/mL) We also found that there was no significant
difference in enrichment efficiency of VLCFAs between
the four MWCNTs, which suggests that the size of
the MWCNT may contribute to desorption/ionization
efficiency. To our knowledge, this is the first study to test
the enrichment of VLCFAs using MWCNTs of different
sizes. We have shown that the VLCFAs adsorbed by
MWCNTs can be analyzed by SALDI-TOFMS. In addition,
this method does not require liquid/gas chromatography
separation, thereby allowing for high-throughput screening
of VLCFAs in peroxisomal disorders.
Carbon nanotubes (CNTs) are basically elongated fullerenes
consisting of rolled graphite sheets with diameters in the nanom-
* To whom correspondence should be addressed. (C.-C.L.) Phone: (886)
4-22840485ext. 235. Fax: (886) 4-22858163. E-Mail: lailai@dragon.nchu.edu.tw.
(F.-J.T.) Phone: (886) 4-22062121ext. 7076. Fax: (886) 4-22033295. E-Mail:
d0704@mail.cmuh.org.tw.
† China Medical University Hospital.
‡ National Chung Hsing University.
§ National Taiwan University Hospital and National Taiwan University College
of Medicine.
| China Medical University.
& Asia University.
6814 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
eter range and lengths in the micrometer range. CNTs were first
discovered in the cathode deposit of arc discharge between
graphite rods. 1 CNTs are classified based on the number of carbon
atom layers into multiwalled carbon nanotubes (MWCNTs) and
single-walled carbon nanotubes (SWCNTs). 2 CNTs have unique
chemical, physical, mechanical, and electrical properties, and
therefore, they have potential applications in a variety of fields. 3,4
The hydrophobic features and large adsorption surface areas of
CNTs make them effective solid-phase extraction adsorbents of
a wide range of chemical compounds including endocrine disruptors,
5 chlorophenols, 6 polycyclic aromatic hydrocarbons, 7 benzodiazepine
residues, 8 barbiturates, 9 herbicides, 10 tetracyclines, 11
and polybrominated diphenyl ethers. 12 It has been shown that
functional groups (i.e., hydroxyl or carboxyl groups) introduced
onto the surface of CNTs can modulate their affinity for and
selectivity of hydrophilic solutes or compounds with polar functional
groups. 13 From a structural point of view, buckminster
fullerene (C60) has also been shown to be an effective solid-phase
extraction adsorbent. Munoz and Valcarcel showed that MWCNTs
as well as C60 and C70 fullerenes were superior to graphitized
(1) Iijima, S. Nature 1991, 354, 56–58.
(2) Iijima, S.; Ichihashi, T. Nature 1993, 363, 603–605.
(3) Saito, N.; Usui, Y.; Aoki, K.; Narita, N.; Shimizu, M.; Hara, K.; Ogiwara, N.;
Nakamura, K.; Ishigaki, N.; Kato, H.; Taruta, S.; Endo, M. Chem. Soc. Rev.
2009, 38, 1897–1903.
(4) Valcarcel, M.; Cardenas, S.; Simonet, B. M. Anal. Chem. 2007, 79, 4788–
4797.
(5) Cai, Y.; Jiang, G.; Liu, J.; Zhou, Q. Anal. Chem. 2003, 75, 2517–2521.
(6) Cai, Y. Q.; Cai, Y. E.; Mou, S. F.; Lu, Y. Q. J. Chromatogr., A 2005, 1081,
245–247.
(7) Wang, W. D.; Huang, Y. M.; Shu, W. Q.; Cao, J. J. Chromatogr., A 2007,
1173, 27–36.
(8) Wang, L.; Zhao, H.; Qiu, Y.; Zhou, Z. J. Chromatogr., A 2006, 1136, 99–
105.
(9) Zhao, H.; Wang, L.; Qiu, Y.; Zhou, Z.; Zhong, W.; Li, X. Anal. Chim. Acta
2007, 586, 399–406.
(10) Zhou, Q.; Xiao, J.; Wang, W.; Liu, G.; Shi, Q.; Wang, J. Talanta 2006, 68,
1309–1315.
(11) Suarez, B.; Santos, B.; Simonet, B. M.; Cardenas, S.; Valcarcel, M.
J. Chromatogr., A 2007, 1175, 127–132.
(12) Wang, J. X.; Jiang, D. Q.; Gu, Z. Y.; Yan, X. P. J. Chromatogr., A 2006,
1137, 8–14.
(13) Sae-Khow, O.; Mitra, S. J. Chromatogr., A 2009, 1216, 2270–2274.
10.1021/ac100772j © 2010 American Chemical Society
Published on Web 07/22/2010

carbon black and RP-C18 for the extraction of the organometallic
compounds. 14
Surface-assisted laser desorption/ionization (SALDI), a matrixfree
method for laser desorption/ionization mass spectrometry, 15
has overcome the limitations of matrix-assisted laser desorption/
ionization (MALDI). The lower matrix interference in the low mass
spectrum (molecular mass < 500 Da) makes SALDI more
appropriate for quantifying small molecules. There are four major
types of SALDI substrates: silicon (DIOS), 16,17 sol-gels, 18 carbonbased
substrates, and metal particle-based substrates. 19-22 Carbonbased
SALDI, which utilizes graphite with sizes in the micrometer
range, was first used as a desorption/ionization substrate for
detection of peptides and proteins in 1995. 20 CNTs have also been
used as SALDI substrates to analyze small peptides (

Scheme 1. (A) Procedure of Derivatizating VLCFAs to Quaternary Ammonium Salt and a Neutral Loss in the
Postsource Decay (PSD) and (B) Multiple Functions of MWCNTs for Analyte Enrichment and as the SALDI Substrate
was added to dissolve the residue at room temperature for 10
min. The residue was removed in a stream of nitrogen at 40 °C
and was reconstituted in 50% (v/v) methanol immediately before
analysis.
Enrichment of VLCFAs and Analysis with SALDI-TOFMS
by MWCNTs. The procedure was performed according to the
procedure reported by Pan et al. with minor modifications 32
(Scheme 1B). MWCNTs (10 mg) were first suspended in 1 mL
of 50% (v/v) methanol and then sonicated for 3 min.
Then, 10 µL of the suspension was pipetted immediately into
a centrifuge tube containing the analyte solution. The tube was
sonicated for less than 5 s, after which the MWCNTs with analytes
solution were homogeneously spread in the solution and the
analytes were adsorbed from the liquid phase to the surface of
MWCNTs within 10 min. The analyte-adsorbed MWCNTs were
precipitated by centrifugation at 10 000 rpm for 10 min. Then, the
supernatant was removed, and 4 µL of dispersant solution of 50%
methanol (v/v) with 5% glycerol (v/v) and 1% sucrose (w/w) was
added to the centrifuge tube to resuspend the MWCNTs. Finally,
about 2 µL of the MWCNT solution was pipetted onto the SALDI-
TOFMS sample plate. The sample plate was left at room temperature
for 10-15 min to allow for the evaporation of the solvent
before analysis by SALDI-TOFMS.
(32) Pan, C.; Xu, S.; Zou, H.; Guo, Z.; Zhang, Y.; Guo, B. J. Am. Soc. Mass
Spectrom. 2005, 16, 263–270.
6816 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
SALDI-TOFMS analyses were performed on an AXIMA-CFR
plus (Shimazu/KRATOS, Manchester, UK) instrument equipped
with a nitrogen laser (wavelength at 337 nm) and delayed
extraction. Ions were accelerated by energy of 20 kV before
entering the TOF mass spectrometer. All spectra were acquired
in the linear mode. In the SALDI-TOF/PSD MS experiments,
spectra were acquired in the reflectron mode. The mass spectra
of 50 different profiles of a sample spot were averaged. Each
profile included five laser shots. The mass calibration was
achieved using low-mass standards (MassPREP Calibration Mix-
DIOS, Waters) in the mass range 20-1500 Da. Quantitative
analysis of VLCFAs was carried out by measuring the ion peak
intensity of the individual mass peaks.
Limit of Detection (LOD) of Derivatized VLCFAs with
MWCNT-Based SALDI-MS and MALDI-MS. Serial concentrations
of the standard solutions of four VLCFAs (60 µL) were
prepared: 50, 25, 10, 5, 1, and 0.5 µg/mL. Following the derivatization
of VLCFAs to quaternary ammonium salt derivatives, the
products were reconstituted in 60 µL of 50% (v/v) methanol. The
derivatized products (10 µL) were enriched by MWCNTs according
to the procedure mentioned above. The amounts of VLCFAs
analyzed per spot were 250, 125, 50, 25, 5, and 2.5 ng.
The sensitivity of MALDI-TOFMS to detect VLCFAs was
tested using CHCA (10 mg/mL in 80% acetone) as the matrix.
The reconstituted solution (10 µL) was mixed with CHCA

solution in an equal volume, and 1 µL of the mixture was
pipetted onto the MALDI-MS sample plate without the enrichment
procedure.
Adsorption Efficiency of VLCFAs by MWCNTs in the
Enrichment Procedure. Adsorption efficiency was determined
using standard mixtures of four VLCFAs (40 µg/mL each).
Standard mixtures were derivatized first and then subjected to
the enrichment procedure with MWCNTs as mentioned above.
After the standard mixture had adsorbed to the surface of the
MWCNTs, the supernatants were analyzed by HPLC/ESI-MS.
HPLC analysis was performed on a 3 µm C18 column (AtlantisdC18,
2.1 mm i.d × 50 mm, Waters) with an isocratic program
(mobile phases of acetonitrile/methanol/formic acid ) 80/20/
0.1). The autosampler was a Finnigan Surveyor autosampler fitted
with a 10 µL loop. The HPLC and autosampler systems were all
synchronized via Xcalibur software (Xcalibur, Finnigan Corp.). A
Finnigan LCQ DECA XP PLUS quadrupole ion trap mass spectrometer
(Finnigan Corp., San Jose, CA) equipped with a
pneumatically assisted electrospray ionization source was used.
The mass spectrometer was operated in positive ion mode by
applying a voltage of 4.5 kV to the ESI needle. The temperature
of the heated capillary in the ESI source was set at 260 °C.
The flow rate of the sheath gas (nitrogen) was set at 25
(arbitrary units). Helium was used as the damping gas at a
pressure of 10 -3 Torr. Voltages across the capillary and the
octapole lenses were tuned by an automated procedure to
maximize the signal of the ion of interest. We used the
following SIM (selected ion monitoring) transitions for the
analysis: C20:0 (center mass 398.2), C22:0 (center mass 426.2),
C24:0 (center mass 454.2), and C26:0 (center mass 482.2). The
isolation width was set as 1.5 Da. The experimental programs
and data analyses were performed with the software package
Xcalibur (Xcalibur, Finnigan Corp.).
Preparation of Standard Solutions and Calibration Curves.
Stock solutions of internal standard mixtures of three stable
isotope-labeled VLCFAs (C20:0-d3, C22:0-d3, and C26:0-d4) and
standard mixtures of four VLCFAs (C20:0, C22:0, C24:0, C26:
0) were prepared at a concentration of 50 µg/mL in methanol
and kept in the dark at -20 °C when not in use. For the
calibration curve, the concentrations of the calibration solutions
of standard mixture were 5, 10, 15, and 20 µg/mL, and the
concentration of the calibration solution of the internal standard
mixture was 10 µg/mL. The calibration solutions (50 µL) were
evaporated in a stream of nitrogen prior to the derivatization
procedure and reconstituted in 10 µL of 50% (v/v) methanol
for the enrichment procedure.
Hydrolysis of Lipids and Extraction of Total Fatty Acids
from Plasma. For hydrolysis of lipids, a solution of acetonitrile
(720 µL) and 5N hydrochloric acid (80 µL) was added to 50 µL of
plasma in a1mLglass tube and heated at 80 °C for 1 h. After
that, 50 µL of internal standard mixture (C20:0-d3, C22:0-d3, and
C26:0-d4) at a concentration of 10 µg/mL was added before
extraction of fatty acids with 2 mL of hexane. The hexane
extraction step was repeated three times. The hexane extracts
were combined and evaporated in a stream of nitrogen prior
to the following derivatization procedure and reconstituted in
10 µL of 50% (v/v) methanol for the enrichment procedure.
Figure 1. Enrichment and SALDI-MS analysis of VLCFAs (1 µg/
mL) derivatized to quaternary ammonium salt (TMAE-VLCFAs)
(A-D); and MALDI-MS analysis of VLCFAs with CHCA as matrix (E).
(* indicated the background peaks).
RESULTS AND DISCUSSION
Preparation of Quaternary Ammonium Salt Derivatives
of VLCFAs. We found that VLCFAs (stock concentration 50 µg/
mL) could not been detected in MALDI-MS or SALDI-MS ether
in the positive or negative ion mode. Using ESI-MS, Johnson et
al. 30 reported that trimethyaminoethyl-VLCFAs (TMAE-VLCFAs)
afforded 8- to 12-fold greater signal intensity than the corresponding
dimethylaminoethyl-VLCFAs (DMAE-VLCFAs). In order to
enhance the detection in SALDI-MS, the same derivatization
strategy was used in this study. The carboxylic acid groups of
VLCFAs were first derivatized to DMAE-VLCFAs. The limit of
detection (LOD) in MALDI-MS with CHCA as the matrix ranged
from 10 to 50 µg/mL (Figure S-1B in the Supporting Information).
Subsequently, DMAE-VLCFAs were reactived with methyl iodide
to form TMAE-VLCFAs. The LOD of TMAE-VLCFAs in MALDI-
MS (15 µg/mL) were about 10 times greater than the LOD of
DMAE-VLCFAs (Figure S-1A in the Supporting Information). The
DMAE-VLCFAs were not detected in the SALDI analysis. TMAE-
VLCFAs are quaternary ammonium salts with permanent positive
charges. The mode of ionization of quaternary ammonium salts
under SALDI conditions is by dissociation of the salts in the ion
source, leading to the detection of the positively charged moieties.
So, the quaternary TMAE-VLCFAs iodide is expected to lead to
higher SALDI-TOFMS sensitivity due to the dissociation of the
iodide ion. Besides, in the process of SALDI, no additional acids
were added. This ensured that the permanent positively charged
VLCFAs have high priority in ionization over the uncharged or
neutral compounds.
Detection of Derivatized VLCFAs with Various MWCNTs
for Enrichment and as SALDI Substrates (Scheme 1).
MWCNTs come in a variety of diameters and lengths, depending
on the growth process. In this study, MWCNTs of different
diameters and lengths were chosen from commercial products
with the same producing method (CVD method), carbon content,
melting point, and density. The sizes of the four MWCNTs listed
in Table 1 were provided by the manufacturer. The structural
information was confirmed by SEM and TEM and is provided in
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6817

Figure 2. SALDI-PSD MS spectra of TMAE-VLCFAs.
the Supporting Information (Figures S-2 and S-3, respectively).
The SALDI-TOFMS spectra for the enrichment and SALDI
analysis of TMAE-VLCFAs with various MWCNTs are depicted
in Figure 1A-D and compared with the MALDI-MS with CHCA
as the matrix (Figure 1E). The TMAE-VLCFAs (1 µg/mL) ions
were only detected in MWCNT 636843 (Figure 1A). The TMAE-
VLCFAs ions were [C20:0-TMAE] + at m/z 398.4, [C22:0-TMAE] +
at m/z 426.4, [C24:0-TMAE] + at m/z 454.5, and [C26:0-TMAE] +
at m/z 482.5. SALDI-MS spectra showed LODs of 0.5-1 µg/
mL for MWCNT 636843, 1-5 µg/mL for MWCNT 636495 and
MWCNT 636509, and 5-10 µg/mL for MWCNT 636819
(Figure 1 and S-4 in the Supporting Information). The MALDI-
MS analysis with CHCA as the matrix showed an LOD of 1-5
µg/mL. In addition, more background peaks were observed in
MALDI-MS spectra compared to MWCNT-based SALDI-TOFMS
spectra. There were at least two factors that contributed to the
analytical sensitivity in this study: the adsorption efficiency in the
enrichment procedure and the desorption/ionization efficiency in
SALDI. The adsorption efficiency of the MWCNTs was analyzed
by HPLC/ESI-MS. The LC/MS chromatograms of TMAE-VLCFAs
are shown in Figure S-5 in the Supporting Information. The
calibration and validation data of the LC/MS method are given
in Figure S-6 and Table S-1 in the Supporting Information. The
adsorption efficiencies of the four MWCNTs for VLCFAs were
>99.8%. There were no significant differences (p > 0.05) between
the four MWCNTs. We attribute the varied desorption/ionization
efficiency in SALDI to the different sizes of MWCNTs. The
MWCNT 636843 had a much larger outer diameter (40-70 nm),
a wider wall thickness (0-65 nm), and a shorter length (0.5-2
µm) than the MWCNT 636819, which had a much smaller outer
diameter (

Table 2. Concentrations and Ratios of VLCFAs in Plasma from Patients with ALD and from Normal Controls as
Measured by SALDI-TOFMS
Postsource Decay (PSD) of TMAE-VLCFAs. The PSD MS
spectra of the derivatized VLCFAs in SALDI-TOFMS, which
detected a neutral loss of trimethylamine (59 Da), were similar
to those in LC-MS/MS. 30 SALDI-PSD MS spectra of TMAE-
VLCFAs are recorded in Figure 2. Fragment ions were detected
at m/z 339.3 for C20:0, m/z 367.4 for C22:0, m/z 395.4 for C24:0,
and m/z 423.4 for C26:0. Those data demonstrate that derivatized
VLCFAs not only increased the ionization efficiency in SALDI but
also made the PSD easier in TOFMS. Furthermore, the neutral
loss fragmentations helped confirm the presence of VLCFAs in
the complex plasma matrix.
Quantitative Analysis of the TMAE-VLCFAs. CNT-based
SALDI has been shown to have excellent reproducibility of
spectrum signals, making it an acceptable modality for quantitative
analysis. 22,19,32 In addition, MALDI-MS with and without the stable
isotope-labeled method has been shown to be suitable for
quantifiation of small compounds in biologic samples. 38-40 In this
study, we used stable isotope-labeled VLCFAs (C26:0-d4, C22:0d3,
C20:0-d3) as the internal standards. The mass difference of
3-4 Da between VLCFAs and isotope-labeled VLCFAs provided
sufficient m/z separation space to avoid interference with
isotopologues of the ions. Figure 3 shows the SALDI-TOF mass
spectra of the four derivative VLCFAs. Three of the derivative
isotope-labeled VLCFAs had a molar ratio of 2:1 and a peak
intensity ratio of 2:1. The isotope-labeled C22:0-d3 and C26:0-d4
were both evaluated in the quantification of C24:0 as the
internal standards, with C26:0-d4 showing superior linearity (R 2
regression coefficient varied from 0.9876 to 0.9969). The
accuracy of the method was measured by determining the
mean concentration at various concentrations of analyte and
was calculated as percentage error of theoretical versus
measured concentrations. Both the intra- and interday accuracy
and precision of the method were determined by triplicate
analysis of standard samples containing VLCFAs at the concentrations
used to construct the calibration curves. Precision
was estimated as the coefficient of variance (CV) of the
analyses. The interday accuracy ranged from -5.41 to 3.23%,
and the intraday accuracy ranged from -3.95 to 4.27% (Table
C24:0/C22:0
(ion peaks
intensity ratio)
C24:0/C22:0
(conc ratio)
C20:0 (µg/mL) C22:0 (µg/mL) C24:0 (µg/mL)
1 16.23 15.21
patients with ALD
17.58 1.02 1.16
2 17.25 11.81 22.21 1.15 1.88
3 14.53 9.10 15.30 1.49 1.68
4 17.89 11.47 14.70 1.19 1.28
mean ± SD 16.47 ± 1.46 11.90 ± 2.5 17.45 ± 3.41 1.21 ± 0.20 1.50 ± 0.34
GC/MS median a
(5-90% range)
20.8 (11.8-29.5) 13.7 (7.0-21.1) 1.50 (1.18-1.74)
1 22.52 28.32
control group
19.52 0.64 0.69
2 20.28 26.30 20.95 0.76 0.80
3 18.86 22.13 18.25 0.76 0.82
4 18.16 18.45 14.96 0.74 0.81
mean ± SD 19.95 ± 1.92 23.80 ± 4.40 18.42 ± 2.55 0.72 ± 0.06 0.78 ± 0.06
GC/MS median a
(5-90% range)
25.3 (10.5-51.0) 17.4 (8.5-35.7) 0.73 (0.48-0.89)
a Value reported by Schutgens. 27
S-2 in the Supporting Information). The interassay precision
ranged from 1.05 to 7.52% and the intra-assay precision ranged
from 0.89 to 8.85% (Table S-2 in the Supporting Information). It
was found that the quantification of VLCFAs by MWCNT-based
SALDI, using isotope-labeled VLCFAs as internal standards, was
of acceptable accuracy and precision.
Patient Plasma Analysis of VLCFAs. In order to test the
clinical applicability of our assay, plasma samples collected from
four subjects with X-linked adrenoleukodystrophy (ALD) and four
normal controls were analyzed by MWCNT-based enrichment and
SALDI-TOFMS. The SALDI-TOFMS spectra of VLCFAs in ALD
patients and in normal controls are shown in Figure 4. The
intensity ratio of C24:0/C22:0 was significantly higher in the ALD
patients (1.02-1.49) than in the normal controls (0.64-0.76). The
peaks at m/z 440.4 and 452.4 were assumed to be C23:0 and
C24:1. 25,27,30,41 Table 2 summarizes the quantification data for
VLCFAs concentrations and ratios in this study and compares
them with the reference range reported by Schutgens. 27 In their
study, samples from 109 healthy subjects and 41 patients with
X-ALD were analyzed by GC-MS following derivatization by methyl
ester and purification by thin-layer chromatography. The quantification
results in our study were within the reported range.
Usually, concentration ratios of C24:0/C22:0 are used to diagnose
ALD. Stable-isotope quantification is important in the MALDI-
TOFMS method because of the suppression effect of matrix
interference peaks. In the MWCNT-based SALDI-TOFMS method
described herein, the relative quantification can also be directly
determined by measuring the ion peak intensities without the
isotope-labeled internal standards calibration. Although the sample
numbers are small, the concentration ratios of C24:0/C22:0
correlated well with the ion peak intensity ratios of C24:0/C22:0
(Pearson Correlation coefficient ) 0.88) (Figure S-8 in the
(38) Koulman, A.; Petras, D.; Narayana, V. K.; Wang, L.; Volmer, D. A. Anal.
Chem. 2009, 81, 7544–7551.
(39) Sleno, L.; Volmer, D. A. Rapid Commun. Mass Spectrom. 2005, 19, 1928–
1936.
(40) Hsu, W. Y.; Lo, W. Y.; Lai, C. C.; Tsai, F. J.; Tsai, C. H.; Tsai, Y.; Lin, W. D.;
Chao, M. C. Rapid Commun. Mass Spectrom. 2007, 21, 1915–1919.
(41) Aveldano, M. I.; Donnari, D. Clin. Chem. 1996, 42, 454–461.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6819

Supporting Information). That finding suggests that ALD can be
screened for by directly comparing the ion peak intensity ratios
of C24:0/C22:0 when stable isotope-labeled internal standards are
not available in the laboratory.
Although GC-MS is a full quantitative method, quantification
of VLCFAs in serum or plasma is typically time-consuming. 25,27
For example, esterification of fatty acids with methylating reagent
can take up to 16 h. Ultrasound-assisted transmethylation of fatty
acids has been reported 28,29 to simplify the sample preparation
procedure and improve the efficiency of transmethylation of fatty
acids. However, the VLCFAs are not all detected in these reports
(especially the specific markers C22:0 and C24:0 for the screening
of peroxisomal disorder).
CONCLUSION
The described method has a number of advantages that satisfy
the increased demand for selective and sensitive high-throughput
techniques. Derivative VLCFAs make desorption/ionization in the
SALDI process more efficient, and the neutral loss in PSD allows
6820 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
for the specific identification of VLCFAs in plasma matrix.
Enrichment of VLCFAs by MWCNT without eluting and evaporation
procedures is an easy method for prepurification and
concentration of samples. Use of VLCFAs-adsorbed MWCNTs as
the SALDI substrate works well for detection in the low mass
range without background peak interference and suppression.
ACKNOWLEDGMENT
The study was funded by a grant from the National Research
Council of the Republic of China and the China Medical University
Hospital (DMR 93-018).
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review March 25, 2010. Accepted July 2,
2010.
AC100772J

Anal. Chem. 2010, 82, 6821–6829
MacroSEQUEST: Efficient Candidate-Centric
Searching and High-Resolution Correlation
Analysis for Large-Scale Proteomics Data Sets
Brendan K. Faherty † and Scott A. Gerber* ,†,‡
Department of Genetics, Dartmouth Medical School, Lebanon, New Hampshire 03756, and Norris Cotton Cancer
Center, Lebanon, New Hampshire 03756
Modern mass spectrometers are now capable of producing
tens of thousands of tandem mass (MS/MS) spectra
per hour of operation, resulting in an ever-increasing
burden on the computational tools required to translate
these raw MS/MS spectra into peptide sequences. In the
present work, we describe our efforts to improve the
performance of one of the earliest and most commonly
used algorithms, SEQUEST, through a wholesale redesign
of its processing architecture. We call this new program
MacroSEQUEST, which exhibits a dramatic improvement
in processing speed by transiently indexing the array of
MS/MS spectra prior to searching FASTA databases. We
demonstrate the performance of MacroSEQUEST relative
to a suite of other programs commonly encountered in
proteomics research. We also extend the capability of
SEQUEST by implementing a parameter in MacroSE-
QUEST that allows for scalable sparse arrays of experimental
and theoretical spectra to be implemented for highresolution
correlation analysis and demonstrate the
advantages of high-resolution MS/MS searching to the
sensitivity of large-scale proteomics data sets.
Mass spectrometry (MS) coupled with computer-assisted
database spectral matching has evolved into a cornerstone
technology that drives research for the field of proteomics. 1 Recent
developments in MS instrumentation have resulted in commercial
mass spectrometers capable of generating tens of thousands of
tandem mass spectra (MS/MS) per penultimate online reversephase
liquid chromatography (LC) separation. 2-4 When coupled
with biochemical prefractionation methods, a complete data set
for a proteomics experiment (including technical and biological
replicates) can consist of millions of MS/MS spectra per biological
* To whom correspondence should be addressed. E-mail: scott.a.gerber@
dartmouth.edu.
† Dartmouth Medical School.
‡ Norris Cotton Cancer Center.
(1) Aebersold, R.; Mann, M. Nature 2003, 422, 198–207.
(2) Makarov, A.; Denisov, E.; Kholomeev, A.; Balschun, W.; Lange, O.; Strupat,
K.; Horning, S. Anal. Chem. 2006, 78, 2113–2120.
(3) Olsen, J. V.; Schwartz, J. C.; Griep-Raming, J.; Nielsen, M. L.; Damoc, E.;
Denisov, E.; Lange, O.; Remes, P.; Taylor, D.; Splendore, M.; Wouters, E. R.;
Senko, M.; Makarov, A.; Mann, M.; Horning, S. Mol. Cell. Proteomics 2009,
8, 2759–2769.
(4) Second, T. P.; Blethrow, J. D.; Schwartz, J. C.; Merrihew, G. E.; MacCoss,
M. J.; Swaney, D. L.; Russell, J. D.; Coon, J. J.; Zabrouskov, V. Anal. Chem.
2009, 81, 7757–7765.
sample. Importantly, although the absolute sensitivity of new
instruments to detect a single peptide species has improved, it
has been suggested that much of the credit for increased depth
of peptide and protein coverage, and improved detection of
substoichiometric species belongs to the increased rate of MS/
MS spectral acquisition of these instruments, which allows them
to penetrate transient rasters of precursor ions to greater ion peak
depth per chromatographic unit time. 5 Indeed, the number of
candidate precursor ions (MS1 features) is often at least an order
of magnitude greater than the number of MS/MS sequencing
events in a typical LC-MS analysis. 6 Given the current level of
success with this strategy, it is only reasonable to predict that
this trend of increasing MS/MS bandwidth to improve overall
peptide identification rates per sample will continue, which places
an additional burden on the computational tools required to search
these larger data sets.
A single raw MS/MS spectrum is translated into a peptide
spectral match (PSM) through the use of algorithms that first
search translated genomic databases from an organism of interest
for candidate peptides by a defined enzyme specificity (based on
the protease used for digestion) and precursor mass (based on
the MS1 feature mass from which the MS/MS spectrum was
derived and a desired mass precision). 7 Other parameters (fixed
protein modifications, variable post-translational modifications,
number of missed enzyme cleavage loci, maximum and/or
minimum peptide size, etc.) may also be included. This search
results in a list of candidate peptides that may contain the correct
peptide sequence from which the MS/MS spectrum was derived.
These candidate peptides are then evaluated for correctness by
comparing the observed MS/MS spectrum with a dynamically
generated theoretical spectrum for each candidate peptide and
given a score that reflects the quality of their match. These scores
are then ranked and, in some cases, further evaluated 8 before
reporting the “best” candidate peptide match (PSM) for the MS/
MS spectrum. In a collection of MS/MS spectra from a single
LC-MS run, a series of LC-MS runs for a given sample or from
(5) Haas, W.; Faherty, B. K.; Gerber, S. A.; Elias, J. E.; Beausoleil, S. A.;
Bakalarski, C. E.; Li, X.; Villen, J.; Gygi, S. P. Mol. Cell. Proteomics 2006,
5, 1326–1337.
(6) Hoopmann, M. R.; Finney, G. L.; MacCoss, M. J. Anal. Chem. 2007, 79,
5620–5632.
(7) Nesvizhskii, A. I. Methods Mol. Biol. 2007, 367, 87–119.
(8) Kall, L.; Canterbury, J. D.; Weston, J.; Noble, W. S.; MacCoss, M. J. Nat.
Methods 2007, 4, 923–925.
10.1021/ac100783x © 2010 American Chemical Society 6821
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/28/2010

a series of different samples, the general process is iterated
thousands, if not millions, of times.
Today, the modern proteomics researcher can choose from
several computational tools to translate MS/MS spectra into
PSMs, including SEQUEST, 9 Mascot, 10 X!Tandem, 11,12 and
OMSSA, 13 among others. A number of studies have been
performed that evaluate the analytical performance of these
algorithms, with regards to precision and sensitivity of the PSM
collections they generate. 14-16 While it appears that certain
algorithms perform slightly better or worse than others based on
the nature of the sample (e.g., phosphorylation, enzyme digest),
the type of mass spectrometer used to generate these MS/MS
data, and the mechanism of peptide fragmentation, etc., they are
in general more similar than they are different, at least, in terms
of the nature of the PSM collections they produce. However, given
the recent trend toward larger and larger numbers of MS/MS
spectra per experiment, the relative processing speed of these
algorithms has become an important practical consideration, as
some of these algorithms perform significantly slower than others.
In particular, SEQUEST has lagged significantly behind, although
limited efforts to improve performance by parallelization have been
reported. 17 In general, the most basic solution to any large
discrepancies in performance (or “productivity”, defined as the
number of MS/MS spectra processed per unit computational time)
has been to index the protein database by predigesting the protein
sequences to peptides, then indexing each candidate peptide by
precursor mass. While these peptide indices do result in significant
performance gains that in general level the playing field
among algorithms, there are drawbacks to using indexed databases.
For example, the use of an indexed database as opposed
to the use of a FASTA database results in a significant expansion
of disk memory, requires separate indices for each enzyme used
for digestion, and is problematic when considering post-translational
modifications, all of which is repeated and exacerbated for
each organism and/or release version of the organism-specific
genome sequence. A recent report attempted to reconcile the
issues of both performance and indexed database files by reading
the target FASTA database once for the first MS/MS spectrum
to be analyzed, indexing “on-the-fly” and storing this peptide index
in fast CPU memory for use in finding candidates for subsequent
MS/MS spectra. 18
An alternative approach that addresses the performance and
convenience issues associated with both indexed database files
(9) Eng, J. K.; Mccormack, A. L.; Yates, J. R. J. Am. Soc. Mass Spectrom. 1994,
5, 976–989.
(10) Perkins, D. N.; Pappin, D. J.; Creasy, D. M.; Cottrell, J. S. Electrophoresis
1999, 20, 3551–3567.
(11) Craig, R.; Beavis, R. C. Bioinformatics 2004, 20, 1466–1467.
(12) Craig, R.; Beavis, R. C. Rapid Commun. Mass Spectrom. 2003, 17, 2310–
2316.
(13) Geer, L. Y.; Markey, S. P.; Kowalak, J. A.; Wagner, L.; Xu, M.; Maynard,
D. M.; Yang, X.; Shi, W.; Bryant, S. H. J. Proteome Res. 2004, 3, 958–964.
(14) Bakalarski, C. E.; Haas, W.; Dephoure, N. E.; Gygi, S. P. Anal. Bioanal.
Chem. 2007, 389, 1409–1419.
(15) Elias, J. E.; Haas, W.; Faherty, B. K.; Gygi, S. P. Nat. Methods 2005, 2,
667–675.
(16) Kapp, E. A.; Schutz, F.; Connolly, L. M.; Chakel, J. A.; Meza, J. E.; Miller,
C. A.; Fenyo, D.; Eng, J. K.; Adkins, J. N.; Omenn, G. S.; Simpson, R. J.
Proteomics 2005, 5, 3475–3490.
(17) Sadygov, R. G.; Eng, J.; Durr, E.; Saraf, A.; McDonald, H.; MacCoss, M. J.;
Yates, J. R. J. Proteome Res. 2002, 1, 211–215.
(18) Park, C. Y.; Klammer, A. A.; Kall, L.; MacCoss, M. J.; Noble, W. S. J.
Proteome Res. 2008, 7, 3022–3027.
6822 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
as well as indexing the protein database in CPU memory is to
consider an entire set of MS/MS spectra to be searched (e.g., a
complete LC-MS/MS run) as an array of precursor masses and
spectra and to transiently index these MS/MS spectra by precursor
mass at the beginning of a search. This allows for efficient
matching of candidate peptides during a single in silico digestion
pass through a FASTA database. Because the number of MS/
MS spectra in an LC-MS/MS run is small relative to the number
of candidate peptides in a typical organism database, this indexing
step is extremely fast and can easily be implemented at the launch
of a search. Furthermore, additional performance gains can be
realized through elimination of redundant candidate peptide
spectral processing steps that occur when single MS/MS spectra
are considered in the absence of other MS/MS spectra that have
overlapping candidate peptide search spaces. Indeed, the benefits
of this “candidate-centric” (as opposed to “spectrum-centric”)
searching approach have been described explicitly, first in
principle by Edwards and Lippert 19 and later in practice by Tabb
and co-workers. 20
In the present work, we demonstrate the general utility of this
“candidate-centric” approach by modifying the commercial version
of SEQUEST to perform searches in a candidate-centric fashion,
resulting in a program we call MacroSEQUEST. MacroSEQUEST
reads FASTA-formatted protein databases and returns PSMs for a
collection of MS/MS spectra in a fraction of the time it takes legacy
SEQUEST. We compare the performance of MacroSEQUEST to a
suite of other commonly used database search engines. Finally, we
demonstrate a useful application for the performance gains associated
with MacroSEQUEST by leveraging this increase in speed to perform
high-resolution MS/MS correlation analysis and describe the benefits
to classical SEQUEST scores associated with high-mass precision
fragment ion searching.
MATERIALS AND METHODS
Sample Preparation. Peptide synthesis was performed by
New England Peptide (Gardner, MA). Yeast cells were harvested
during logarithmic growth, pelleted, and lysed by addition of SDS-
PAGE pH 8.1 sample buffer (3× volume buffer/pellet weight) and
bead beating at 4 °C. HeLa cell lysate was prepared by harvesting
a confluent 15 cm dish of HeLa cells by trypsinization, washing
2× in PBS, and addition of 4 mL of lysis buffer (0.5% Triton X-100,
50 mM Tris pH 8.1, 150 mM NaCl, 1 mM MgCl2, Roche Minicomplete
protease inhibitors), followed by sonication and
clarification of the lysate in a centrifuge (14 000g) for 10 min
at 4 °C. Both protein preps were reduced by addition of DTT
to 5 mM and incubation in a water bath at 55 °C for 20 min,
followed by alkylation of cysteines in 12.5 mM iodoacetamide
at room temperature for 45 min. After the alkylation reaction
was quenched (addition of 2.5 mM DTT), the lysates were
aliquotted, snap frozen in liquid nitrogen, and stored at -80
°C until use. To prepare the protein digests, an aliquot of cell
lysate was warmed rapidly under warm water and mixed with
SDS-PAGE sample buffer to a protein concentration of ∼0.25
(19) Edwards, N.; Lippert, R. In Algorithms in Bioinformatics: Second International
Workshop, WABI 2002, Rome, Italy, September 17-21, 2002 Proceedings;
Guigo, R., Gusfield, D., Eds.; Springer: Berlin, Germany, 2002; Vol. 2452,
pp 68-81.
(20) Tabb, D. L.; Narasimhan, C.; Strader, M. B.; Hettich, R. L. Anal. Chem.
2005, 77, 2464–2474.

mg/mL protein, followed by separation on 2-well, 4-12%
NOVEX minigels (for 2D separations) with a protein marker
in the narrow lane. Proteins were visualized with Coomassie
blue, and the region between 80 and 125 kDa was excised,
destained, and digested with trypsin. Peptide samples were
analyzed on an LTQ Orbitrap (ThermoFisher Scientific, Bremen,
Germany) per established procedures. 5 LTQ-Orbitrap
.RAW files were converted to .mzXML files using ReAdW.exe
(version 4.0, http://sourceforge.net/projects/sashimi/files/).
Peptide precursor mass assignments were adjusted postacquisition
with in-house software that (i) updates MS2 scan headers
with high-mass accuracy precursor information from MS1 scans
and (ii) averages multiple observations of these precursor
values across each peptide chromatographic elution profile.
MacroSEQUEST. MacroSEQUEST was written from scratch
in ANSI C using standard libraries and compiled with GCC version
4.1.2 on a Unix platform using the SEQUEST2.8 source code as
a guide. Input is provided as command line arguments. Experimental
data is read in .DTA file format, and candidate peptides
are read from FASTA-formatted protein databases. Macro outputs
SEQUEST-like .out files for each input spectrum.. To measure
the efficiency of the “candidate-centric” method, we created a
version of Macro (MacroParser) in which the Xcorr scoring
function was replaced with a random number generator.
Database Searching. Searches were performed using the
latest database builds from the yeast proteome (Saccharomyces
Genome Database, http://www.yeastgenome.org/) or the human
proteome (UniProtKB, http://www.uniprot.org/). Target-decoy
databases were generated using in-house scripts that reverse each
protein sequence, label decoys proteins with a specific identifier,
and append the decoy sequences to the end of a forward
database. 21 Unless otherwise stated, all searches were performed
with a 1.1 Da precursor mass tolerance and filtered to ±1.5 ppm
precursor mass measurement accuracy (Figure S-1 in the Supporting
Information); Xcorr and dCn (delta-correlation) cutoff
values were adjusted to achieve a false discovery rate (FDR) of
less than 1% (Table S-1 in the Supporting Information). Only the
yeast searches were conducted with semitryptic enzyme specificity;
all other searches were performed with full trypsin specificity.
The maximum number of missed cleavages for all searches was
set to three. All searches were also performed with acetamidemodified
Cys as a static modification (+57.021 461 Da) and with
oxidized Met as a variable modification (±15.991 915 Da); for
phosphorylation searches, Ser, Thr, and Tyr were allowed to vary
by ±79.966 331 Da. Variable modifications were limited to a
maximum of 3 per peptide, where applicable. Refinement or
iterative searches were not permitted when using X!Tandem and
OMSSA. No multithreading was used for any search algorithm.
RESULTS
Candidate-Centric Database Spectral Matching. Historically,
database spectral matching algorithms such as SEQUEST
dealt with very small numbers of MS/MS spectra acquired per
experiment, from as few as tens to at most one hundred spectra
in a typical analysis. 9 Individual MS/MS spectra were then
matched with candidate peptides by searching through a target
database. In general, the focus for algorithm development has
(21) Elias, J. E.; Gygi, S. P. Nat. Methods 2007, 4, 207–214.
been on the steps involved in this single iteration and on improving
the sensitivity, accuracy, and productivity of a single analysis, with
the expectation that this “optimized” core process is iterated until
PSMs have been generated for all MS/MS spectra in an analysis
which, while computationally inefficient, was adequate given the
historical context of the problem space. A generalized scheme
describing this classical workflow for SEQUEST is depicted in
Figure 1a. However, this workflow fails to recognize potential
elements of relatedness between MS/MS spectra that are collected
together. Clearly, all of these spectra require matching to
candidate peptides in the same database; thus, presenting the
entire collection of spectra to the database (or vice versa) as a
“single analysis” has the potential to reduce areas of overhead
associated with repeating a single process, such as digesting a
database, thousands of times (Figure 1b). Our analysis of the work
distribution for SEQUEST when searching a target-decoy human
UniProtKB protein database (∼150 000 proteins) against 10 000
spectra reveals that only a very small portion of the actual search
time is spent on scoring candidate PSMs (∼1%), while the
remainder is spent parsing the database (Figure 1c). Although
the database must be read once in order to generate a PSM for
a single spectrum, it does not need to be reread to search other
MS/MS spectra under the same set of parameters if those
additional spectra are considered simultaneously. In this way,
legacy SEQUEST wastes a significant amount of search time when
large numbers of MS/MS spectra are searched against large
databases.
We sought to address these issues through a wholesale
rearchitecting of the SEQUEST scoring components in order to
conduct them in a candidate-centric fashion and named the
resulting program MacroSEQUEST or “Macro” for short. The
Macro workflow is described in Figure 1b. Similar to the candidate-centric
algorithm DBDigger, 20 a single Macro process at
launch is fed a parameter set that includes a path to a collection
of MS/MS spectra. This collection of spectra is first read into
memory and indexed by precursor ion mass to create a “spectral
data array” that includes preprocessing for each experimental
spectrum and memory mapping for candidate peptide sequences,
protein references, search scores, etc. After this data structure is
created, Macro begins parsing the FASTA database of interest
by digesting it based on a desired enzyme cleavage specificity.
For each newly digested peptide, Macro calculates its mass and
checks it against the spectral array, including a given mass
tolerance; if a peptide falls outside these boundaries, Macro
discards it and proceeds to the next logical peptide in the database.
If, however, a peptide falls into one or more spectral “bins” in the
spectral array, Macro enters a scoring loop in which a reimplementation
of a fast SEQUEST cross correlation algorithm 22,23 is
performed on the candidate peptide for each MS/MS spectrum
that falls within the peptide’s desired precursor mass tolerance,
and plugs these scores and associated peptide/protein information
into the spectral array. The scoring loop then closes by returning
to database parsing and to the next logical peptide in the database;
Macro continues this cycle of peptide generation, spectral array
scanning, and scoring until it reaches the end of the FASTA
(22) Eng, J. K.; Fischer, B.; Grossmann, J.; Maccoss, M. J. J. Proteome Res. 2008,
7, 4598–4602.
(23) Venable, J. D.; Xu, T.; Cociorva, D.; Yates, J. R. Anal. Chem. 2006, 78,
1921–1929.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6823

Figure 1. Comparison of legacy SEQUEST and MacroSEQUEST (Macro) workflows. (a) Scheme of the serial nature of SEQUEST. To analyze
n number of spectra, SEQUEST must be executed n number of times, parsing the database with each instance and using both preliminary and
XCorr scores. (b) Scheme of the Macro workflow. Macro considers all spectra in a single analysis and therefore requires only one pass through
a target database during a search. (c) Analysis of the time distribution for SEQUEST and Macro searches of 10 000 spectra with a target-decoy
human (UniProt) database. SEQUEST spends approximately 1% of the search time on scoring functions while Macro spends only 13% of the
search time parsing the database. The inset shows an expanded view of the Macro time distribution for the first 20 min of the search time.
database. Macro then concludes by calculating final score differences,
etc. for the top-ranked candidate PSMs and writing this
information to result files. Because Macro was developed from
the original SEQUEST source code, the primary scoring metric
Xcorr in Macro is identical to those created by legacy SEQUEST,
a major difference between the two is that Macro no longer
performs Sp scoring as a preliminary scoring step but instead
calculates Xcorr for all candidate PSMs, owing to the computationally
fast, non-FFT correlation algorithm. Macro spends significantly
less time parsing the target-decoy human database than
SEQUEST during a search of the same 10 000 spectra (Figure
1c, inset), and almost 87% of the total run time performing scoring
functions.
Because such a small fraction of the actual SEQUEST process
is spent scoring candidate PSMs and because this cycle is repeated
in its entirety for each spectrum to be searched, SEQUEST’s
productivity (number of spectra searched/unit time spent searching)
is relatively flat as a function of the number of spectra
searched (Figure 2). Macro, however, benefits from having more
spectra to search by distributing the fixed time cost associated
with a single digestion and parsing pass through the database
across many spectra until the time spent parsing the database is
small relative to the time spent scoring candidate PSMs, at which
point Macro’s search productivity flattens out. This improvement
in productivity as a function of number of spectra searched is
clearly depicted in Figure 2, where searches of a target-decoy yeast
(Figure 2a), human (Figure 2b), and human databases with
dynamic phosphorylation on serine, threonine, and tyrosine
(Figure 2c) using Macro all outperform legacy SEQUEST by 102×,
107×, and 42×, respectively, when searching 10 000 MS/MS
6824 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
spectra. Note that the increase in parsing logic associated with
determining dynamic protein post-translational modifications, such
as phosphorylation, substantially compresses and flattens the
Macro productivity curve and extends the point in numbers of
spectra at which Macro achieves maximum productivity to well
beyond the 10 000 mark.
In order to assess the level of spectral array processing
overhead in Macro, we generated a “MacroParser” version of the
program that replaces the XCorr scoring functions with a simple
random number generator. This allowed us to subtract the time
it takes Macro to parse the database and scan the array of spectra
from the total search time, which generated an ideal power fit as
a function of the number of spectra searched (Figure 2a-c).
Comparison of MacroSEQUEST to Other Contemporary
Tools. Although it was among the earliest algorithms to be
adopted into widespread use, SEQUEST is now not the only
program available to retrieve candidate PSMs from FASTA
databases. Other commercial (Mascot) and noncommercial
(OMSSA, X!Tandem) programs are also designed to execute
database searches, although the core components and matching
algorithms differ substantially. To establish Macro’s performance
rank among programs that are commonly encountered in proteomics
research laboratories and protein identification core
facilities, we generated productivity curves for searches against
our target-decoy human database for Macro, SEQUEST, Mascot,
OMSSA, and X!Tandem. We also noted that all three of these
programs have been written to take advantage of multiple physical
and logical cores now increasingly common in modern CPU
architectures, including the Intel Conroe chip used in this
comparison. In order to cleanly reconcile differences in perfor-

Figure 2. Productivity of SEQUST and Macro under typical experimental conditions. We created three biological samples to use in testing the
performance of Macro (green circles) versus legacy SEQUEST (orange circles) in common applications: (a) yeast 80-130 kDa protein digest,
(b) human 80-130 kDa protein digest, and (c) human phosphorylation samples, all analyzed by a 90 min gradient LC-Orbitrap-MS/MS. Variable,
random portions of each full collection of MS/MS spectra were searched to define the productivity (number of MS/MS spectra searched/minute
of search time) of each search type as shown in the left panel. A version of Macro that does not produce scores and is useful as a measurement
of database parsing work, MacroParser, was used to calculate the time spent on scoring by difference. In the right panel, the Macro, MacroParser,
and calculated Macro Scoring productivity distributions are plotted as the log for the variable, random portion search sizes versus search time.
A power function was fitted to the Macro Scoring distribution using a nonlinear least-squares method in R. Note that Macro is 102×, 107×, and
42× faster than SEQUEST under each search condition, respectively.
mance due to multithreading, we disabled one of the two cores
on our Conroe prior to performing the comparison, the results of
which are depicted in Figure 3. At peak productivity, Macro
performs equivalent to OMSSA, about 20% faster than Mascot,
and twice as fast X!Tandem, with legacy SEQUEST lagging far
behind the others.
High-Resolution MS/MS Correlation Analysis. In light of
the significant performance improvement of Macro over legacy
SEQUEST, we reasoned that this additional speed could be leveraged
to also produce gains in the quality of spectral matches by creating
sparse arrays of MS/MS fragment ions from high-resolution, highmass
accuracy instruments such as the LTQ Orbitrap 2 and, in
particular, the LTQ Orbitrap Velos 3 and by performing full correlation
analyses on them to produce Xcorr scores, a feature not currently
available in SEQUEST. Given the novelty of Xcorr and dCn values
derived from high-resolution MS/MS spectra, we were also interested
in evaluating the qualitative impact that variable search “resolution”
might have on these scores.
To do this, we created a parameter for Macro that defines the
bin width of the arrays (in m/z) used for correlation analysis and
used this factor to scale the number of bins that are generated
within the range of observed fragment ions during MS/MS
spectrum preprocessing steps. Macro then populates these bins
with normalized ion intensity values using logic consistent with
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6825

Figure 3. Comparison of Macro performance with contemporary
database searching algorithms. The same data set in Figure 2b was
researched with other commonly used database searching algorithms,
including Mascot, OMSSA, and X!Tandem. All algorithms were run
in single-threaded mode to provide an accurate comparison.
legacy SEQUEST and performs the fast Xcorr preprocessing math
on the resultant sparse array. Given the mass precision and
resolution of ion traps, MS/MS data from such instruments
normally defines these bin widths as 1 m/z wide, and the spectral
preprocessing logic of SEQUEST determines which ions fall into
which m/z bin(s). Figure 4a depicts an MS/MS spectrum
collected in an LTQ Orbitrap with a resolution setting of 15 000.
Figure 4b describes this spectrum preprocessed for fast Xcorr
analysis by Macro with a bin width of 1 m/z, while Figure 4c
depicts an array from the same spectrum but with bin widths of
0.01 m/z. Note that, in the insets, binning logic in Macro reduced
the number of available ions for matching across a 2 m/z-wide
window in the original spectrum from 6 to 2 when 1 m/z-wide
bins were used versus 0.01 m/z-wide bins. Clearly, the higher
“resolution” of the array in Figure 4c allows for more accurate
ion-to-bin assignments and consequently a more robust discrimination
between ion fragments of similar m/z. This is also apparent
in the significant reduction in the average magnitude of negative
values in the high-resolution preprocessed array (Figure 4c),
which spreads these anticorrelations out over a much larger bin
space. Similar to the preprocessing steps that were performed
for experimental spectra, theoretical spectra for each candidate
peptide were created using the same bin width scaling factor, and
Xcorr values were then calculated from the dot product of the
two arrays.
We then tested the utility of these scalable MS/MS arrays by
collecting data from our LTQ Orbitrap at relatively high resolution
and mass accuracy (R ) 15 000). We analyzed our HeLa cell
80-120 kDa protein lysate digest over a 90 min LTQ Orbitrap
gradient, during which a total of 5 725 MS/MS spectra were
collected. We then used Macro to iteratively search this run with
fragment ion bin widths of 1.0, 0.75, 0.5, 0.25, 0.1, 0.075, 0.05, 0.025,
and 0.01 m/z. Because the size of the arrays scales inversely with
bin width, there is a modest but significant speed penalty
6826 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
associated with high-resolution scoring. On our Intel Conroe test
platform fitted with 8Gb of RAM, we observed a 3× increase in
search time using bin widths of 0.025 m/z relative to unit
resolution and a 6× increase in search time when using bin widths
of 0.01 m/z. However, we also observed a significant increase in
the total number of true positive (TP) PSMs at a 1% false discovery
rate (FDR). Figure 5a depicts the total number of TP PSMs for
this run as a function of search bin width. We observed a 24%
increase in the number of TP PSMs when using bin widths of
0.025 m/z versus 1.0 m/z bin widths. We considered the possibility
that simply limiting the number of candidate PSMs by using a
very narrow precursor ion tolerance may allow these “rescuable”
PSMs to rise in rank. However, consistent with previous reports, 24,25
we only observed a very slight increase in TP assignments in a
search limited to ±10 ppm precursor ion mass measurement
accuracy (Figure 5b). This suggests that the sensitivity gains we
observe by high-resolution fragment ion searching represent
PSMs that are not accessible by simply limiting the number of
candidate PSMs in a search. Although these mediocre yet
”rescuable” MS/MS spectra are being considered in the context
of fewer competing candidates during a narrow window precursor
ion search, their inherently poor correlation characteristics do not
sufficiently enhance their differential Xcorr rank to allow for their
rescue in this search mode.
This increase in true positive assignments can be ascribed to
a combination of features of merit associated with the higher
resolution versions of Xcorr and dCn. Although we did not see an
appreciable change in Xcorr for TP PSMs when searching at
higher fragment ion resolution, we noted that false positive (FP)
PSMs exhibited a decrease in their median Xcorr values (from
1.6 to 1.2 for 1.0 and 0.025 m/z-wide bins, respectively; Figure
5c). This allows for significantly lower Xcorr cutoff values to be
used when establishing a 1% FDR. From Figure 5c, clearly many
of the “rescued” TP PSMs come from low Xcorr scores that are
otherwise inaccessible due to the extension of the FP PSM
distribution into that score space. Equally dramatic was the change
in TP PSM dCn scores, whose median values increased from 0.25
to 0.42 across the entire data set (Figure 5d); for those “rescued”
TP PSMs, the median dCn value jumped from 0.06 to 0.33. As
dCn is a direct measure of the relative Xcorr score separation
between the highest-ranked PSM and its nearest neighbor (de
facto a FP PSM), our observations of overall reduced FP Xcorr
scores and increased TP dCn scores are consistent with this
relationship.
To further describe the nature of some of these “rescued” TP
PSMs, we examined their scores and individual PSM rankings.
Figure 6a depicts the top two ranked PSMs for a particular MS/
MS spectrum that yielded an excellent Xcorr score (4.3), but a
very low dCn score of 0.005 when scored with a 1.0 m/z bin width.
Note that these PSMs differ only by a K/Q (mass difference )
0.036 Da) in the middle of the peptide sequences. Although
distinguishable by Orbitrap precursor ion mass precision, for
correlation analysis at unit resolution fragment ion correlation,
this difference is transparent and results in almost identical Xcorr
(24) Beausoleil, S. A.; Jedrychowski, M.; Schwartz, D.; Elias, J. E.; Villen, J.; Li,
J.; Cohn, M. A.; Cantley, L. C.; Gygi, S. P. Proc. Natl. Acad. Sci. U.S.A.
2004, 101, 12130–12135.
(25) Hsieh, E. J.; Hoopmann, M. R.; Maclean, B.; Maccoss, M. J. J. Proteome
Res. 2010, 9, 1138–1143.

Figure 4. High-resolution MS/MS spectra and Macro. (a) A typical, raw high-mass accuracy MS/MS spectrum generated using an LTQ-
Orbitrap (R ) 15 000). (b) The raw spectrum in part a preprocessed for fast Xcorr using 1 m/z-wide bins to define the spectrum array. (c) The
same spectrum preprocessed for fast Xcorr using 0.01 m/z-wide bins. Note in the inset that multiple ion features persist in a 2 m/z-wide space
in the higher resolution array, while they are compressed into two features when searches are performed at unit resolution.
scores for these two sequences. However, when searched by
Macro using 0.025 m/z bin widths, this K/Q difference is readily
distinguished by dCn, likely for all singly and doubly charged
fragment ions from this quadruply charged precursor that contain
the Lys residue. Figure 6b describes the general behavior of all
candidate PSMs for this MS/MS spectrum at the two fragment
ion tolerances. It is worth noting again in this plot that the
discriminating power of these high-resolution correlations lies
predominantly in positive overlap between the theoretical and
experimental arrays and not in the “signal-to-noise” of this overlap
relative to neighboring offsets (±75 m/z equivalents), an important
aspect of the unit resolution Xcorr. This can be observed in the
large decrease in the number of negative Xcorr values for lowranked
PSMs between the high- and low-resolution searches.
Indeed, of the 819 “rescued” PSMs, 26% of them were for
peptide sequences that were correctly assigned (top-ranked) in
the lower resolution correlation analysis but with extremely low
dCn/Xcorr combinations such that they precluded a definitive
assignment when score cutoffs were applied to achieve a 1% FDR.
The remaining rescued peptide matches were reranked from
lower ranks to the top-ranked candidate PSM. Figure 6c depicts
the distribution across the entire run of original candidate PSM
ranks when searches were performed with 1.0 m/z-wide fragment
ion bins that are “rescued” when the search is done at 0.025 m/z
bin widths. Of those PSMs that are reranked between the two
searches, the median rank in the low-resolution search is 9; the
lowest reranked candidate PSM moved from the 4 517th position
to the top-ranked spot when searched at high resolution. Although
this MS/MS spectrum is assigned a PSM (AASVHTVGEDTEET-
PHR, M + 4H + , MMA ) 0.1 ppm) with an Xcorr of only 0.67,
the dCn is 0.08 (dCn = 0.001 at unit fragment ion resolution).
This peptide is from the protein hPOP1 with a molecular weight
of 115 kDa, which is in the range of molecular weights (80-125
kDa) of SDS-PAGE-fractionated human cell lysate from which
the sample was derived. Figure 6d displays a reciprocal plot of
the endogenous, rescued MS/MS spectrum from the lysate
with the matched ions noted and a mirrored spectrum from a
synthetic peptide that we analyzed under identical LC-MS/MS
conditions, confirming this rescued sequence as being a correct
match. Additionally, there were two other peptides from the same
protein that were identified in both the high- and low-resolution
correlation analyses with scores and mass measurement accuracy
assignments that pass cutoffs (KTHQPSDEVGTSIEHPR, M +
4H + , Xcorr ) 3.78, dCn ) 0.43, MMA ) -0.2 ppm and
IPILLIQQPGK, M + 2H + , Xcorr ) 3.31, dCn ) 0.51, MMA )
0.1 ppm), supporting the likelihood that this PSM is a valid
match.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6827

Figure 5. Behavior of primary scoring metrics for high-resolution correlation analysis. (a) Histogram depicting the sensitivity of an LC-Orbitrap-
MS/MS analysis as a function of fragment ion bin widths at FDR < 1%. (b) High-resolution correlation analysis rescues false negatives that are
inaccessible by high-mass accuracy precursor ion searching alone. (c) ROC plot using XCorr filtering during 1.0, 0.25, and 0.025 m/z-wide bin
searches. Insets show histograms of the behavior of Xcorr during 1.0 and 0.025 m/z-wide bin searches. The 1.0 m/z TP distribution is blue,
while the 1.0 m/z FP distribution is red. The 0.025 m/z TP distribution is green, while the 0.025 m/z FP distribution is orange. Note that while
the primary score for TPs remains largely unchanged, FP scores are significantly reduced. (d) ROC plot using dCn filtering during 1.0, 0.25, and
0.025 m/z-wide bin searches. Insets show histograms of the behavior of dCn during 1.0 and 0.025 m/z-wide bin searches; the distributions are
colored the same as in part c.
DISCUSSION
Soon, proteomics researchers will be swimming in a sea of
mass spectra. During the later stages of development of Macro,
we were asked to search a5hLTQVelos run with Macro and
were surprised to find that it consisted of ∼170 000 tandem mass
spectra. Clearly, the relative processing speed of computer-assisted
spectral matching algorithms will be a critical feature for these
algorithms in the coming years. Our intent with developing Macro
was to refresh a scoring metric (Xcorr) that adds significant value
to proteomics research by reorganizing the way it conducted
searches. The fact that Macro is also very competitive with other
search engines reinforces our view that candidate-centric searching
represents a practical approach to search algorithm design in
general. The core Macro process is also scalable: the MS/MS
spectral array may be accessed in main CPU memory by multiple
independent processing cores, each of which could digest a
different portion of the FASTA database or perform scoring
functions, or both; design modifications such as this are planned
for future versions of Macro.
Although the SEQUEST-specific version of Macro contains
proprietary information that precludes its pubic release, we offer
the source code for MacroParser to the proteomics community
as a general framework for candidate-centric database parsing into
which any peptide scorer could be inserted. Details regarding the
Macro project can be found on our lab Web server at http://
proteomics.dartmouth.edu.
We feel that the performance gains associated with Macro can
allow for a host of features and functions to be added to the
classical scoring function Xcorr, including post-translational modi-
6828 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
fication scanning and, as we show here, high-resolution correlation
analysis. Although a parameter does exist in legacy SEQUEST
that appears to allow users to scale fragment ion tolerances, closer
inspection of the SEQUEST source code and search output when
using this parameter reveals that it is not performing these
searches as the user likely intends. Our inclusion of a variable
fragment ion bin width parameter that accurately bins fragment
ions into a scalable sparse array enables users with a mass
spectrometer capable of better than unit resolution MS/MS
spectra to search their data with the corresponding fragment ion
mass tolerance. This has the added benefit of improved overall
search accuracy relative to unit resolution searching. Although
this does come as a speed penalty, we note that even when
running at 0.025 m/z bin width, Macro is still 20× faster than
SEQUEST on our platform. We anticipate that these highresolution
correlation searches will result in a significant decrease
in the minimum Xcorr threshold required to achieve a desired
FDR: we observed the median Xcorr value of rescued PSMs,
across all charge states, to be 1.52, with the lowest rescued PSM
yielding an Xcorr of 0.67, albeit with a dCn of 0.08. Legacy
SEQUEST users will likely need some adjusting to these unusually
low score results, but with careful application of strategies to
estimate data set false discovery rates 21 and calculate PSM
posterior probability assignments, 26 we hope they will also
ultimately appreciate the benefit of high mass precision to the
improved accuracy of their data sets.
(26) Kall, L.; Storey, J. D.; MacCoss, M. J.; Noble, W. S. J. Proteome Res. 2008,
7, 29–34.

Figure 6. Characterization of PSMs “rescued” by high-resolution correlation analysis. (a) Top-ranked PSMs and scores for a given MS/MS
search result using 1.0 and 0.025 m/z-wide bins. Note that these sequences are identical, except for a K/Q substitution in the middle of the
peptide, marked by a red box. While the 1.0 m/z-wide bin search rank 1 PSM satisfies a precursor ion tolerance and Xcorr cutoff, the dCn
(0.005) precludes unambiguous TP assignment. The top-ranked PSMs and scores from the 0.025 m/z-wide bin search are shown below. The
resolving power of much narrower fragment ion bins clearly allows for discrimination of the K-containing ion fragments, resulting in a dCn of
0.36 and an unambiguous TP PSM assignment. (b) Graphical representation of the distribution of candidate PSMs by Xcorr for the 1.0 (blue)
and 0.025 (red) m/z-wide bin searches of the PSM from part a. (c) Histogram depicting the relative ranks of rescued PSMs in the lower resolution
search. The lowest ranked PSM was rescued from the 4 715th position. (d) Reciprocal MS/MS plots, sequence, and matched fragment ions for
the PSM rescued from the 4 715th rank position. The upper MS/MS spectrum is from the endogenous peptide from the HeLa cell lysate, and
the lower MS/MS spectrum is from the synthetic peptide “AASVHTVGEDTEETPHR”.
ACKNOWLEDGMENT
We thank Mike Senko, Justin Blethrow, and colleagues at
ThermoFisher Scientific for facilitating transfer of the SEQUST
source code, additional platform testing, and general discussions.
We thank Arminja Kettenbach for preparing the HeLa cell lysate
digest and phosphopeptide-enriched sample and also thank Jason
Gilmore for critical reading of the manuscript. This work was
supported by the National Institutes of Health Grant P20-RR018787
from the IDeA Program of the National Center for Research
Resources (to S.A.G.) and a predoctoral fellowship from the
National Institute of General Medical Sciences Grant T32-
GM008704 (to B.K.F.).
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review March 26, 2010. Accepted July 8,
2010.
AC100783X
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6829

Anal. Chem. 2010, 82, 6830–6837
Colorimetric Sensing of Silver(I) and Mercury(II)
Ions Based on an Assembly of Tween 20-Stabilized
Gold Nanoparticles
Cheng-Yan Lin, † Cheng-Ju Yu, † Yen-Hsiu Lin, † and Wei-Lung Tseng* ,†,‡
Department of Chemistry, National Sun Yat-sen University, Taiwan, and National Sun Yat-sen University-Kaohsiung
Medical University Joint Research Center, Kaohsiung, Taiwan
We have developed a rapid and homogeneous method for
the highly selective detection of Hg 2+ and Ag + using
Tween 20-modified gold nanoparticles (AuNPs). Citrate
ions were found to still be adsorbed on the Au
surface when citrate-capped AuNPs were modified with
Tween 20, which stabilizes the citrate-capped AuNPs
against conditions of high ionic strength. When citrate
ions had reduced Hg 2+ and Ag + to form Hg-Au alloys
and Ag on the surface of the AuNPs, Tween 20 was
removed from the NP surface. As a result, the AuNPs
were unstable under a high-ionic-strength solution,
resulting in NP aggregation. The formation of Hg-Au
alloys or Ag on the surface of the AuNPs was demonstrated
by means of inductively coupled plasma mass
spectroscopy and energy-dispersive X-ray spectroscopy.
Tween 20-AuNPs could selectively detect Hg 2+ and
Ag + at concentrations as low as 0.1 and 0.1 µM inthe
presence of NaCl and EDTA, respectively. Moreover,
the probe enables the analysis of AgNPs with a minimum
detectable concentration that corresponds to 1
pM. This probe was successfully applied to detect Hg 2+
in drinking water and seawater, Ag + in drinking water,
and AgNPs in drinking water.
Interest in monitoring toxic metal ions in aquatic ecosystems
continues because these contaminants adversely affect the environment
and have serious medical effects. 1 Silver and mercury
are two of the most hazardous metal pollutants, and they are
widely distributed in ambient air, water, soil, and even food. 2,3
For example, silver can inactivate sulfhydryl enzymes and accumulate
in the body, 4 and mercury exposure can damage a
variety of organs and the immune system. 5 Current approaches
to detecting these two metal ions include inductively coupled
* To whom correspondence should be addressed. Fax: 011-886-7-3684046.
E-mail: tsengwl@mail.nsysu.edu.tw.
† Department of Chemistry, National Sun Yat-sen University.
‡ National Sun Yat-sen University-Kaohsiung Medical University Joint Research
Center.
(1) Campbell, L.; Dixon, D. G.; Hecky, R. E. J. Toxicol. Environ. Health, Part B
2003, 6, 325–356.
(2) Wood, C. M.; McDonald, M. D.; Walker, P.; Grosell, M.; Barimo, J. F.;
Playle, R. C.; Walsh, P. J. Aquat. Toxicol. 2004, 70, 137–157.
(3) Boening, D. W. Chemosphere. 2000, 40, 1335–1351.
(4) Ratte, H. T. Environ. Toxicol. Chem. 1999, 18, 89–108.
(5) Holmes, P.; James, K. A.; Levy, L. S. Sci. Total Environ. 2009, 408, 171–
182.
6830 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
plasma mass spectrometry (ICP-MS), 6,7 atomic absorption
spectrometry, 8,9 and stripping voltammetry. 10,11 Although these
methods offer excellent sensitivity and multielement analysis, they
are rather costly, time-consuming, complex, and nonportable.
In response to these shortcomings, various sensors using small
organic molecules, 12,13 oligonucleotides, 14,15 DNAzymes, 16,17 and
semiconductor quantum dots 18,19 have been investigated for
the selective detection of Ag + or Hg 2+ in aqueous solutions.
Unfortunately, most of these methods suffer from low water
solubility, a complex synthesis procedure, and time-consuming
DNA probe preparation. Recently, gold nanoparticles (AuNPs)
have become another emerging material for sensing Hg 2+ or
Ag + , because they have a high extinction coefficient in the
visible region and behavior that depends on the interparticle
distance. When the distances between the AuNPs become less
than the average particle diameter, the color of the AuNPs
changes from red to purple. Because of the coordination
between the carboxyl groups of thiols and Hg 2+ , thiol-capped
AuNPs have been used for colorimetric sensing of Hg 2+ . 20-22
Also, Hg 2+ can selectively coordinate thymine (T) bases and
forms stable T-Hg 2+ -T complexes. The melting temperature
of cDNA containing T-Hg 2+ -T complexes is higher than that
(6) Karunasagar, D.; Arunachalam, J.; Gangadharan, S. J. Anal. At. Spectrom.
1998, 13, 679–682.
(7) Barriada, J. L.; Tappin, A. D.; Evans, E. H.; Achterberg, E. P. TrAC, Trends
Anal. Chem. 2007, 26, 809–817.
(8) Li, Y.; Chen, C.; Li, B.; Sun, J.; Wang, J.; Gao, Y.; Zhao, Y.; Chai, Z. J. Anal.
At. Spectrom. 2006, 21, 94–96.
(9) Chamsaz, M.; Arbab-Zavar, M. H.; Akhondzadeh, J. Anal. Sci. 2008, 24,
799–801.
(10) Kim, H. J.; Park, D. S.; Hyun, M. H.; Shim, Y. B. Electroanalysis 1998, 10,
303–306.
(11) Mikelova, R.; Baloun, J.; Petrlova, J.; Adam, V.; Havel, L.; Petrek, J.; Horna,
A.; Kizek, R. Bioelectrochemistry 2007, 70, 508–518.
(12) Chatterjee, A.; Santra, M.; Won, N.; Kim, S.; Kim, J. K.; Kim, S. B.; Ahn,
K. H. J. Am. Chem. Soc. 2009, 131, 2040–2041.
(13) Zhan, X. Q.; Qian, Z. H.; Zheng, H.; Su, B. Y.; Lan, Z.; Xu, J. G. Chem.
Commun. 2008, 1859–1861.
(14) Lin, Y.-H.; Tseng, W.-L. Chem. Commun. 2009, 6619–6621.
(15) Wang, J.; Liu, B. Chem. Commun. 2008, 4759–4761.
(16) Li, T.; Shi, L.; Wang, E.; Dong, S. Chemistry 2009, 15, 3347–3350.
(17) Hollenstein, M.; Hipolito, C.; Lam, C.; Dietrich, D.; Perrin, D. M. Angew.
Chem., Int. Ed. 2008, 47, 4346–4350.
(18) Koneswaran, M.; Narayanaswamy, R. Sens. Actuators, B 2009, 139, 91–
96.
(19) Chen, J.-L.; Zhu, C.-Q. Anal. Chim. Acta 2005, 546, 147–153.
(20) Huang, C.-C.; Chang, H.-T. Anal. Chem. 2006, 78, 8332–8338.
(21) Yu, C.-J.; Tseng, W.-L. Langmuir 2008, 24, 12717–12722.
(22) Darbha, G. K.; Singh, A. K.; Rai, U. S.; Yu, E.; Yu, H.; Chandra Ray, P.
J. Am. Chem. Soc. 2008, 130, 8038–8043.
10.1021/ac1007909 © 2010 American Chemical Society
Published on Web 07/16/2010

containing T-T mismatches. 23 Based on this concept, two
strands of DNA, which are designed to be complementary
except for a single T-T mismatch, are used to modify the
surface of the AuNPs. The resulting two types of DNAfunctionalized
AuNPs are selectively aggregated in the presence
of Hg 2+ based on T-Hg 2+ -T coordination and temperature
control. 24 Similarly, Hg 2+ was selectively detected using two
types of T-rich DNA-modified AuNPs and a T-rich DNA linker
at room temperature. 25 Citrate-capped AuNPs interacting with
single strands of T-rich oligonucleotide were found to be stable
in a high-salt solution. 26-28 When Hg 2+ causes the conformation
of T-rich DNA into a folded structure, this folded DNA cannot
be adsorbed onto the AuNP surface. Salt-induced NP aggregation
occurs because T-rich DNA is removed. In addition, it is
well-known that a Au surface exhibits a strong affinity for
Hg 2+ . 29-32 Thus, after the reduction of Hg 2+ with NaBH4, the
Hg(0) thus generated is strongly bonded onto the surface of
Au-based nanomaterials to form a solid amalgam-like structure.
The surface plasmon resonance (SPR) band of Au nanorods
and NPs in an excess of NaBH4 has been found to undergo a
blue shift and a decrease in intensity after Hg 2+ was added. 29,31
The only work on the detection of Ag + reported that AuNPs
functionalized with cytosine-(C)-rich oligonucleotide selectively
aggregated in the presence of Ag + based on the formation of
C-Ag + -C complexes. 33 Although these methods all show good
sensitivity and selectivity to Hg 2+ or Ag + , analysis of these two
metal ions using a single type of AuNPs remains a challenge.
In this study, we present a label-free, rapid, and homogeneous
method for sensing both Hg 2+ and Ag + using Tween 20stabilized
AuNPs (Tween 20-AuNPs). Because the surfaces of
Tween 20-AuNPs still had citrate ions, the reduction of Hg 2+
or Ag + with citrate resulted in the formation of Hg-Au alloy
and Ag on the Au surface. When the Tween 20 was removed,
NP aggregation occurred. We also investigated the effect of
masking agents on the selectivity of this probe. To demonstrate
its practicality, the present method was further applied to the
determination of Hg 2+ ,Ag + , and AgNPs in complex matrices.
EXPERIMENTAL SECTION
Chemicals. Hydrogen tetrachloroaurate (III) dehydrate,
Na2HPO4, and Na3PO4 were purchased from Alfa Aesar (Ward
Hill, MA). Trisodium citrate, ethylenediaminetetraacetic acid
(EDTA), Tween 20, Tween 40, Tween 60, Tween 80, NaBH4,
(23) Tanaka, Y.; Oda, S.; Yamaguchi, H.; Kondo, Y.; Kojima, C.; Ono, A. J. Am.
Chem. Soc. 2007, 129, 244–245.
(24) Lee, J.-S.; Han, M.-S.; Mirkin, C. A. Angew. Chem., Int. Ed. 2007, 46, 4093–
4096.
(25) Xue, X.; Wang, F.; Liu, X. J. Am. Chem. Soc. 2008, 130, 3244–3245.
(26) Yu, C.-J.; Cheng, T.-L.; Tseng, W.-L. Biosens. Bioelectron. 2009, 25, 204–
210.
(27) Li, D.; Wieckowska, A.; Willner, I. Angew. Chem., Int. Ed. 2008, 47, 3927–
3931.
(28) Liu, C.-W.; Hsieh, Y.-T.; Huang, C.-C.; Lin, Z.-H.; Chang, H.-T. Chem.
Commun. 2008, 2242–2244.
(29) Rex, M.; Hernandez, F. E.; Campiglia, A. D. Anal. Chem. 2006, 78, 445–
451.
(30) Leopold, K.; Foulkes, M.; Worsfold, P. J. Anal. Chem. 2009, 81, 3421–
3428.
(31) Lisha, K. P.; Anshup; Pradeep, T. Gold Bull. 2009, 42, 144–152.
(32) Barrosse-Antle, L. E.; Xiao, L.; Wildgoose, G. G.; Baron, R.; Salter, C. J.;
Crossley, A.; Compton, R. G. New J. Chem. 2007, 31, 2071–2075.
(33) Li, B.; Du, Y.; Dong, S. Anal. Chim. Acta 2009, 644, 78–82.
ascorbic acid, and NaCl were ordered from Sigma-Aldrich
(Louis, MO). LiCl, KCl, MgCl2, CaCl2, SrCl2, BaCl2, CrCl3,
MnCl2, FeCl2, FeCl3, CoCl2, NiCl2, CuCl2, ZnCl2, Cd(ClO4)2,
AlCl3, Pb(NO3)2, HgCl2, and AgNO3 were purchased from Acros
(Geel, Belgium). Water used in all experiments was doubly
distilled and purified by a Milli-Q system (Millipore, Milford,
MA).
Characterization of the AuNPs. Extinction spectra of the
AuNPs were measured using a double-beam UV-visible spectrophotometer
(Cintra 10e; GBC, Victoria, Australia). High-resolution
transmission electron microscopy (HRTEM, FEI Tecnai G2 F20
S-Twin working at 200 kV) was used to collect HRTEM images
of dispersed and aggregated AuNPs. Energy-dispersive X-ray
(EDX) spectra were obtained using a HRTEM microscope. The
zeta potential and size distribution of the AuNPs were measured
using Delsa nano zeta potential and submicrometer particle size
analyzer (Beckman Coulter Inc., U.S.). The hydrodynamic size
of the AuNPs was measured using dynamic light scattering (DLS)
(N5 Submicrometer Particle Size Analyzer, Beckman Coulter Inc.,
U.S.).
To understand the sensing mechanism, we equilibrated aliquots
(1.0 mL) of 0.48 nM Tween 20-AuNPs in the presence of
Hg 2+ (0-10 µM) or Ag + (0-10 µM) for 5 min at ambient
temperature. The resulting mixture was subjected to centrifugation
at 17 000 rpm for 10 min. Following removal of the
supernatants, the precipitates were washed with water. After
five centrifugation/washing cycles, the pellets were resuspended
in water. A portion of the samples (∼200 µL) was
diluted to 50-fold and then measured by ICP-MS (Perkin-Elmer-
SCIEX, Thornhill, ON, Canada). Additionally, the composition
of the obtained pellets was analyzed by EDX spectroscopy.
For surface-assisted laser desorption/ionization time-of-flight
ionization mass spectrometry (SALDI-TOF MS) (Autoflex, Bruker)
measurements, citrate-capped AuNPs and Tween 20-AuNPs were
separately pipetted into a stainless steel 384-well target (Bruker
Daltonics) and dried under ambient temperature. Desorption/
ionization was obtained by using a 337-nm-diameter nitrogen laser
witha3nspulse width. MS experiments were performed in the
positive-ion mode on a reflectron-type TOF MS equipped with a
3 m flight tube. To obtain good resolution and signal-to-noise
ratios, the laser power was adjusted to slightly above the threshold,
and each mass spectrum was generated by averaging 500 laser
pulses.
Nanoparticle Synthesis. We prepared citrate-capped AuNPs
by means of the chemical reduction of a metal salt precursor
(hydrogen tetrachloroaurate, HAuCl4) in the liquid phase. To
achieve this, we rapidly added HAuCl4 (0.35 M, 54 µL) to a
solution of sodium citrate (2.55 mM, 60 mL) that was heated
under reflux. This heating continued for an additional 15 min,
during which time the color of the solution changed to a deep
red. The size of citrate-capped AuNPs determined by TEM
images was 13 ± 1 nm. The SPR wavelength of citrate-capped
AuNPs located at 520 nm. The particle concentration of the
AuNP solution was estimated to be 4.8 nM by Beer’s law; the
extinction coefficient of 13 nm AuNPs at 520 nm is 2.7 × 10 8
M -1 cm -1 . Tween 20-AuNPs were synthesized by adding Tween
20 (10% v/v, 240 µL) to a solution of citrate-capped AuNPs (4.8
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6831

Scheme 1. Illustration of the Mechanism of Tween 20-AuNPs for Sensing Hg 2+ and Ag +
nM, 60 mL). 34 To investigate the effect of the kind of the
surfactant on the sensing of Hg 2+ and Ag + , Tween 20 was
replaced by Tween 40, Tween 60, and Tween 80, once at a time.
Additionally, to test the effect of the type of the AuNPs on the
detection of Hg 2+ and Ag + , we replaced citrate-capped AuNPs
with bare AuNPs in the synthesis of Tween 20-AuNPs. The
preparation of bare AuNPs (3.7 nM; 12 nm) is described in
the Supporting Information (SI).
Sample Preparation. Tween 20-AuNPs were prepared in
100-1000 mM sodium phosphate solution at pH 12.0. For Hg 2+
sensing, metal ions (800 µL, 125-1250 nM) were added to a
solution containing Tween 20-AuNPs (100 µL, 0.96-14.4 nM)
and NaCl (100 µL, 1 M). For Ag + sensing, metal ions (800 µL,
125-1250 nM) were added to a solution containing Tween 20-
AuNPs (100 µL, 0.96-14.4 nM) and EDTA (100 µL, 0.1 M).
We equilibrated the resulting solutions at ambient temperature
for the optimum incubation time, and then recorded the
extinction spectra of the solutions.
Analysis of Real Samples. Samples of drinking water and
seawater (pH 7.9) were collected from National Sun Yat-sen
University campus. We then prepared a series of samples by
“spiking” them with standard solutions of Hg 2+ (125-1250 nM)
or Ag + (375-1250 nM). These spiked samples (800 µL) were
added either to a solution containing 100 µL of 2.4 nM Tween
20-AuNPs and 100 µL of 1 M NaCl or to a solution containing
100 of 4.8 nM Tween 20-AuNPs and 100 µL of 0.1 M EDTA.
We incubated the resulting solutions for 5 min before measuring
their extinction spectra.
On the other hand, this proposed method was utilized to detect
10 nm AgNPs in drinking water. The preparation of citrate-capped
AgNPs 35 is described in the SI. Different concentrations of AgNPs
(1-10 pM) present in drinking water were oxidized to Ag + with
(34) Huang, C.-C.; Tseng, W.-L. Analyst 2009, 134, 1699–1705.
(35) Wei, H.; Chen, C.; Han, B.; Wang, E. Anal. Chem. 2008, 80, 7051–7055.
6832 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
a solution of 1 µM H2O2 and 1 µM H3PO4. After 10 min, the
resulting solution (800 µL) was added to a solution containing
100 of 4.8 nM Tween 20-AuNPs and 100 µL of 0.1 M EDTA.
The extinction spectra of the resulting solutions were recorded
after 5 min incubation.
RESULTS AND DISCUSSION
Sensing Mechanism. SALDI-TOF MS technique is applied
to the detection of small molecules that are adsorbed on the
surface of the AuNPs when AuNPs are used as SALDI matrices. 36,37
Thus, we utilized this technique to determine whether citrate ions
were adsorbed on the surface of Tween 20-AuNPs. SI Figure S1A
shows the SALDI spectrum of citrate-capped AuNPs. The peak
detected at m/z 258.00 corresponded to [citrate +3Na] + . This
peak was also observed in the SALDI spectrum of Tween 20-
AuNPs (SI Figure S1B), indicating that citrate ions are capped
on the surface of Tween 20-AuNPs. Moreover, the zeta potential
of Tween 20-AuNPs was found to be -14.0 ± 0.8 mV in 10 mM
phosphate at pH 12.0. These results suggest that a neutral Tween
20 coating only shields the citrate ions (no displacement occurs).
Accordingly, we reasoned that citrate ions adsorbed onto the NP
surface can act as a reducing agent for Hg 2+ and Ag + when
citrate-capped AuNPs have been modified with Tween 20.
Scheme 1 shows the mechanism by which Tween 20-AuNPs
sense Hg 2+ and Ag + . Because of the high affinity between Au
and Hg, 29-32 the reduced Hg(0) is directly deposited onto the
Au surface through the formation of Hg-Au alloys; meanwhile,
Tween 20 molecules are desorbed from the AuNPs. The removal
of the stabilizer (Tween 20) causes the AuNPs to aggregate under
conditions of high ionic strength. Also, citrate ions can be used
(36) Su, C.-L.; Tseng, W.-L. Anal. Chem. 2007, 79, 1626–1633.
(37) Wu, H.-P.; Yu, C.-J.; Lin, C.-Y.; Lin, Y.-H.; Tseng, W.-L. J. Am. Soc. Mass
Spectrom. 2009, 20, 875–882.

Figure 1. Extinction spectra of solutions of (A) 0.48 nM Tween 20-
AuNPs and (B) 0.37 nM Tween 20-modified bare AuNPs (a) before
and (b, c) after the addition of (b) 1 µMHg 2+ and (c) 1 µMAg + . Tween
20-AuNPs are prepared in 20 mM phosphate at pH 12.0. The
incubation time is 5 min.
to reduce Ag + onto the Au surface. 38,39 The formation of Agcoated
AuNPs enables Tween 20 to be removed from the NP
surface, thereby inducing aggregation of the AuNPs in a highionic-strength
solution.
Evidence for the Formation of Hg-Au Alloys and Ag
Shells. To test the hypothesis mentioned above, we monitored
the extinction spectra of Tween 20-AuNPs in the absence and
presence of Hg 2+ and Ag + . Curve a in Figure 1A shows that the
SPR wavelength of Tween 20-AuNPs appears at 520 nm, indicating
that they are well dispersed in 20 mM phosphate solution at pH
12.0. In other words, Tween 20 molecules can indeed protect
citrate-capped AuNPs against a high-ionic-strength solution. 40 After
separately adding 1.0 µMHg 2+ (curve b) and 1.0 µMAg + (curve
c), we observed a decrease in the strength of SPR band at 520
nm and the formation of a new red-shift band. These changes
were characteristic of AuNP aggregation. The Hg 2+ - and Ag + -
(38) Xie, W.; Su, L.; Donfack, P.; Shen, A.; Zhou, X.; Sackmann, M.; Materny,
A.; Hu, J. Chem. Commun. 2009, 5263–5265.
(39) Xia, H.; Bai, S.; Hartmann, J.; Wang, D. Langmuir 2009.
(40) Shen, C.-C.; Tseng, W.-L.; Hsieh, M.-M. J. Chromatogr., A 2009, 1216,
288–293.
induced NP aggregation was nearly complete after 5 min (SI
Figure S2). Obviously, the deposition of Hg and Ag onto the Au
surface enables Tween 20 to be removed, thereby driving NP
aggregation. We ruled out the possibility that coordination
between Tween 20 and metal ions induces the NP aggregation
because Tween 20 does not contain any functional group to
interact with Hg 2+ and Ag + . To ensure the role of citrate ions
in the reduction of Hg 2+ and Ag + , the extinction spectra of bare
AuNPs modified with Tween 20 were examined under the same
conditions. The addition of Hg 2+ and Ag + to this type of AuNP
resulted in a rare shift in the SPR wavelength (Figure 1B),
clearly indicating that citrate ions are indispensable for detecting
Hg 2+ and Ag + using Tween 20-AuNPs. On the other hand, we
prepared Hg-Au alloy- and Ag-coated AuNPs and modified
them with Tween 20. Compared to Tween 20-AuNPs, a blue
shift in SPR and a decrease in SPR intensity were observed
for Tween 20-modified Hg-Au alloy-coated AuNPs (SI Figure
S3). This is attributed to the formation of Hg-Au alloy on the
surface of the AuNPs. A similar phenomenon was reported when
citrate-capped AuNPs were exposed in the presence of Hg vapor. 41
Moreover, this kind of NPs was found to be unstable in a highionic-strength
solution. This result reflects that Tween 20 molecules
were not attached to the surface of Hg-Au alloy, thereby
incapable of protecting NPs against a high-ionic strength solution.
SI Figure S4 shows that the deposition of Ag on the surface of
the AuNPs resulted in an increase SPR intensity relative to Tween
20-AuNPs. 42,43 Similarly, Tween 20-modified Ag-coated AuNPs
were aggregated in a high-ionic-strength solution because Tween
20 molecules were not adsorbed on the surface of Ag shell.
To provide further evidence for the formation of Hg-Au alloys
and Ag shells, we used ICP-MS to quantitatively determine the
composition of the precipitates, which were obtained by five cycles
of centrifugation of a solution of Tween 20-AuNPs and metal ions.
When a series of concentrations (0-1 µM) of Hg 2+ were present
in a solution of 0.48 nM Tween 20-AuNPs, the molar ratio of
Hg to Au in the precipitates gradually increased with increasing
Hg 2+ concentration (Figure 2A). A similar phenomenon was seen
in the case of Ag + (Figure 2B). If Hg-Au alloys and Ag shells
did not form on the Au surface, the molar ratios of Hg to Au and
Ag to Au should remain constant under these conditions. Under
identical treatment conditions (five centrifugation/washing cycles),
the composition of precipitates was also determined by EDX
analysis. The Hg content in the precipitates increased with
increasing Hg 2+ concentration (Figure 2C). We observed a
similar phenomenon when different concentrations of Ag + were
present in a solution of Tween 20-AuNPs (Figure 2D). These
results are in agreement with those obtained by ICP-MS. These
findings strongly support the idea that Hg 2+ - and Ag + -induced
aggregation of Tween 20-AuNPs is indeed the result of the
formation of Hg-Au alloys and Ag on the Au surface.
Effect of Surfactant Chain Length, NP Concentration, and
Ionic Strength. We next explored the effect of surfactant chain
length on the metal-ion-induced aggregation of the AuNPs. The
(41) Morris, T.; Copeland, H.; McLinden, E.; Wilson, S.; Szulczewski, G.
Langmuir 2002, 18, 7261–7264.
(42) Anandan, S.; Grieser, F.; Ashokkumar, M. J. Phys. Chem. C 2008, 112,
15102–15105.
(43) Guerrini, L.; Garcia-Ramos, J. V.; Domingo, C.; Sanchez-Cortes, S. J. Raman
Spectrosc. 2010, 41, 508–515.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6833

Figure 2. (A, B) Effect of the concentration of (A) Hg 2+ and (B) Ag + on the composition ratio of (A) Au to Ag and (B) Au to Hg of the precipitates.
A series of concentration of (A) 0-1000 nM Hg 2+ and (B) 0-1000 nM Ag + was added to 1.0 mL of 0.48 nM Tween 20-AuNPs. The precipitates
were obtained by five cycles of centrifugation of the resulting solutions. (C, D) EDX spectra of the precipitates obtained after the addition of (a)
0.1, (b) 1, and (c) 10 µM (C) Hg 2+ and (D) Ag + to a solution of 0.48 nM Tween 20-AuNPs. Tween 20-AuNPs are prepared in 20 mM phosphate
at pH 12.0. The incubation time is 5 min.
extinction values of the solution at 650 and 520 nm corresponded
to the quantities of dispersed and aggregated AuNPs, respectively.
Thus, the molar ratio of dispersed to aggregated AuNPs can be
expressed by the ratio of the extinction value Ex at 650 nm to
that at 520 nm (Ex650 nm/Ex520 nm). As shown in Figure 3A, the
addition of both Hg2+ and Ag + to a solution of Tween 20-AuNPs
resulted in a high value of Ex650 nm/Ex520 nm. However, when
we replaced Tween 20 with Tween 40, the value of Ex650 nm/
Ex520 nm became small. This suggests a relatively small amount
of Hg-Au alloys or Ag on the surface of Tween 40-modified
AuNPs, resulting in a small degree of NP aggregation. Similar
phenomena were observed in the case of Tween 60- and Tween
80-modifed AuNPs. To further confirm our hypothesis, ICP-
MS was used to determine the composition of the NPs. After
adding 1 µM Hg2+ to different kinds of AuNPs, the concentrations
of Hg in Tween 20-, 40-, 60-, and 80-modified AuNPs were
148, 64, 58, and 51 ppb, respectively. Similarly, upon the
addition of 1 µM Ag + , the concentrations of Ag in Tween 20-,
40-, 60-, and 80-modified AuNPs were 65, 19, 24, and 25 ppb,
respectively. On the basis of these results, Tween 20 was
selected for the following studies.
Previous studies have shown that the sensitivity of a AuNPbased
sensor is highly dependent on the concentration of
6834 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
AuNPs. 44 At relatively high concentrations of Tween 20-AuNPs,
the deposition of Hg2+ or Ag + on the surface of a single particle
decreased, reducing the degree of NP aggregation. Moreover,
the aggregation rate of NPs increased with increasing NP
concentration. It can be seen that the optimum concentrations
of Tween 20-AuNPs for sensing Hg2+ and Ag + were 0.24 and
0.48 nM, respectively, when the incubation time was fixed at 5
min (Figure 3B). Moreover, we investigated the effect of Na3PO4
concentration on the colorimetric sensitivity of Tween 20-
AuNPs to Hg2+ and Ag + .SI Figure S5 shows that the zeta
potential of Tween 20-AuNPs reduced with increasing Na3PO4
concentration, implying that electrostatic repulsion between
Tween 20-AuNPs decreased with increasing ionic strength of
the solution. Thus, under conditions of high ionic strength, the
slight electrostatic repulsion between Tween 20-AuNPs provided
a low barrier for metal-ion-induced NP aggregation. As
expected, the difference in Ex650 nm/Ex520 nm for the cases with
and without 1.0 µM Hg2+ gradually increased with increasing
Na3PO4 concentration and reached a plateau at 80 mM Na3PO4
(Figure 3C). A similar effect was found in the analysis of Ag +
(Figure 3D). Consequently, 80 mM Na3PO4 was chosen for the
following studies.
(44) Huang, C.-C.; Tseng, W.-L. Anal. Chem. 2008, 80, 6345–6350.

Figure 3. (A) The value of Ex650 nm/Ex520 nm of Tween 20- 40-, 60-, and 80-modified AuNPs (0.48 nM) after the addition of (a) 1 µM Hg 2+ and
(b) 1 µM Ag + . (B) Effect of the concentration of Tween 20-AuNPs on the ratio Ex650 nm/Ex520 nm in the presence of (a) 1 µM Hg 2+ and (b) 1 µM
Ag + . (C) Effect of phosphate concentration on the ratio Ex650 nm/Ex520 nm of 0.24 nM Tween 20-AuNPs in the (a) absence and (b) presence of 1
µM Hg 2+ . (D) Effect of phosphate concentration on the ratio Ex650 nm/Ex520 nm of 0.48 nM Tween 20-AuNPs in the (a) absence and (b) presence
of 1 µM Ag + . (A, B) Tween 20-AuNPs are prepared in 20 mM phosphate at pH 12.0. (A-D) The incubation time is 5 min.
Selectivity, Sensitivity, and Application. To realize the
selectivity of 0.24 and 0.48 nM Tween 20-AuNPs toward Hg 2+ and
Ag + , respectively, other metal ionssincluding Li + ,Na + ,K + ,
Mg 2+ ,Ca 2+ ,Sr 2+ ,Ba 2+ ,Cr 3+ ,Mn 2+ ,Fe 2+ ,Fe 3+ ,Co 2+ ,Ni 2+ ,Cu 2+ ,
Zn 2+ ,Cd 2+ ,Al 3+ ,Pb 2+ ,Hg 2+ , and Ag + swere examined under
identical conditions. Only Hg 2+ and Ag + caused the aggregation
of both 0.24 and 0.48 nM Tween 20-AuNPs (SI Figure S6),
revealing that this probe is selective for Hg 2+ and Ag + . To
circumvent this problem, we tested the effect of masking
agents, including NaCl and EDTA. It is well-known that the
solubility product of AgCl is 1.8 × 10 -10 , whereas HgCl2 is
soluble in water (70 g/L). Accordingly, we tested the ability of
NaCl to mask the interfering metal ions in our sensing system.
In the presence of 0.1 M NaCl, the addition of 1 µM Hg 2+ to a
solution of 0.24 nM Tween 20-AuNPs resulted in an apparent
change in Ex650 nm/Ex520 nm, whereas the remaining metal ions
(100 µM) had negligible effects on the same system (Figure
4A). The selectivity of this probe is more than 100-fold for Hg 2+
over all other tested metal ions. Additionally, to improve the
selectivity of 0.48 nM Tween 20-AuNPs for Ag + , we chose
EDTA as a masking agent, since it forms a more stable complex
with Hg 2+ than with Ag + . As expected, the presence of 0.01 M
EDTA masked Tween 20-AuNPs toward Hg 2+ (Figure 4B). As
a result, Tween 20-AuNPs provided high selectivity (>100-fold)
toward Ag + over all other tested metal ions.
Under optimum conditions (for sensing Hg 2+ : 0.24 nM
Tween 20-AuNPs, 80 mM Na3PO4, and 0.1 M NaCl; for
sensing Ag + : 0.48 nM Tween 20-AuNPs, 80 mM Na3PO4, and
0.01 M EDTA), we evaluated the sensitivity of this probe
toward Hg 2+ and Ag + . When the concentrations of Hg 2+
varied from 0 to 1000 nM, the extinction spectra of 0.24 nM
Tween 20-AuNPs showed a gradual increase in extinction
at 650 nm and their color gradually changed from red to
purple (Figure 5A). A solution of 0.48 nM Tween 20-AuNPs
showed a similar response to Ag + (Figure 5B). These spectra
showed clear isosbestic points at 556 and 549 nm upon the
addition of Hg 2+ and Ag + , respectively. This observation
reveals that the aggregation of Tween 20-AuNPs is directly
related to the concentration of Hg 2+ or Ag + . This result was
further confirmed by HRTEM images and DLS measure-
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6835

Figure 4. The value of Ex650 nm/Ex520 nm of a solution of 80 mM Na3PO4 containing (A) 0.24 nM Tween 20-AuNPs and 0.1 M NaCl and (B) 0.48
nM Tween 20-AuNPs and 0.01 M EDTA upon the addition of (A) 1 µM Hg 2+ and 100 µM other metal ions and (B) 1 µM Ag + and 100 µM other
metal ions. The incubation time is 5 min.
ments. SI Figure S7 shows the concentration-dependent TEM
images of the aggregated AuNPs after adding different concentrations
(0, 0.1, 1, and 10 µM) of Hg 2+ or Ag + to a solution
of Tween 20-AuNPs. Moreover, the hydrodynamic size of
the aggregated AuNPs increased with an increase in the
concentration of Hg 2+ or Ag + (SI Figure S8). These results
provide clear evidence that the aggregation degree of Tween
20-AuNPs is highly dependent on the concentration of Hg 2+
or Ag + . We observed that the ratio Ex650 nm/Ex520 nm
increased linearly with increasing Hg 2+ and Ag + concentration
over the range of 200 to 800 nM (R 2 ) 0.9943) and 400
to 1000 nM (R 2 ) 0.9935), respectively (Figure 5C and D).
A difference in linearity between two metal ions could be due
to that the sensing mechanism of Tween 20-AuNPs for Hg 2+
is different from that for Ag + . This probe could detect Hg 2+
and Ag + at concentrations as low as 100 and 100 nM,
respectively. The result is useful for detecting Ag + in
drinking water, because the maximum level of silver in
drinking water permitted by the United States Environmental
Protection Agency (EPA) is 50 µg/L (∼460 nM).
To test the practicality of the present approach, a solution of
0.24 nM Tween 20-AuNPs was used to analyze Hg 2+ in drinking
water and seawater. As shown in SI Figures S9 and S10,
Ex650 nm/Ex520 nm increased linearly upon increasing the spiked
concentration of Hg 2+ in drinking water over the range of
200-600 nM (R 2 ) 0.9944) and in seawater over the range of
300-1000 nM (R 2 ) 0.9977). Evidence of the Hg 2+ -induced
aggregation of Tween 20-AuNPs in seawater can be seen in
the HRTEM images (SI Figure S11). The lowest detectable
concentrations of Hg 2+ in drinking water and seawater were
6836 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
estimated to be 200 and 100 nM, respectively. Although the
sensitivity of this probe is insufficient to detect the maximum
level of mercury (2 ppb) in drinking water permitted by the
U.S. EPA, we suggest that Tween 20-AuNPs can be used as
probes for solid-phase preconcentration of mercury in complex
matrices prior to ICP-MS analysis. 30
We also evaluated the feasibility of this approach for sensing
of Ag + and AgNPs in drinking water. We obtained a linear
correlation (R 2 ) 0.9963) between the ratio Ex650 nm/Ex520 nm
and the concentration of Ag + spiked into the drinking water
over the range of 400-1000 nM (SI Figure S12), which includes
the maximum permissible limit of silver in drinking water.
Moreover, since hazardous AgNPs may pose threats to human
health or the environment, 45 this approach was further used to
monitor AgNPs in drinking water. Under acidic conditions (1.0
µM H3PO4), AgNPs (10 ± 2 nm) are oxidized to Ag + ions with
1.0 mM H2O2. 12 The oxidation of AgNPs to Ag + was complete
after 10 min. The generated Ag + was directly detected by 0.48
nM Tween 20-AuNPs. The degree of aggregation of Tween 20-
AuNPs increased when the concentration of AgNPs was
increased from 1 to 10 pM (SI Figure S13). The correlation
coefficient (R 2 ) for the determination of AgNPs in the range
1-6 pM was 0.9988. These results suggest that this probe will
be suitable for routine assays of AgNPs in consumer products
such as cosmetics and fabrics. 46 Table 1 shows the quantitative
measurements of Hg 2+ ,Ag + , and AgNPs in different matrices
based on the use of Tween 20-AuNPs.
(45) Lubick, N. Environ. Sci. Technol. 2008, 42, 8617.
(46) Benn, T. M.; Westerhoff, P. Environ. Sci. Technol. 2008, 42, 4133–4139.

Figure 5. Extinction spectra and color changes of solutions containing (A) 0.24 nM Tween 20-AuNPs and 0.1 M NaCl and (B) 0.48 nM Tween
20-AuNPs and 0.01 M EDTA upon the addition of (A) 0-1000 nM Hg 2+ and (B) 0-1000 nM Ag + . The arrows indicate the signal changes with
increases in analyte concentrations (A: 0, 100, 200, 300, 400, 500, 600, 700, 800, and 1000 nM; B: 0, 100, 200, 300, 400, 500, 600, 700, 800,
900, and 1000 nM). A plot of Ex650 nm/Ex520 nm versus the concentration of (C) Hg 2+ and (D) Ag + . The incubation time is 5 min. The error bars
represent standard deviations based on three independent measurements.
Table 1. Quantification of Hg 2+ ,Ag + and AgNPs in the
Different Sample Matrix Based on the Use of Tween
20-AuNPs
analyte matrix linear rang (M) R 2 MDC (M) a
Hg 2+ deionized water 2 × 10 -7 to 8 × 10 -7 0.9943 1 × 10 -7
Hg 2+ drinking water 2 × 10 -7 to 6 × 10 -7 0.9944 2 × 10 -7
Hg 2+ seawater 3 × 10 -7 to 1 × 10 -6 0.9977 1 × 10 -7
Ag + deionized water 4 × 10 -7 to 1 × 10 -6 0.9935 1 × 10 -7
Ag + drinking water 4 × 10 -7 to 1 × 10 -6 0.9963 3 × 10 -7
AgNPs drinking water 6 × 10 -12 to 1 × 10 -11 0.9988 1 × 10 -12
a MDC, minimum detectable concentration.
CONCLUSIONS
This study reports a new assay for the selective detection of
Hg 2+ and Ag + using Tween 20-AuNPs. This probe can be
further applied to detecting hazardous AgNPs. We demonstrated
that the aggregation of Tween 20-AuNPs results from
the formation of Hg-Au alloy or Ag on the surface of the
AuNPs. Thus, in the opinion of the authors, Tween 20-AuNPs
should be used as a selective probe for extracting a large
volume of Hg 2+ and Ag + prior to ICP-MS analysis. Moreover,
we believe that the present approach holds great potential for
monitoring Ag + and AgNPs in environmental samples.
ACKNOWLEDGMENT
We thank National Science Council (NSC 98-2113-M-110-009-
MY3) and National Sun Yat-sen University-Kaohsiung Medical
University Joint Research Center for the financial support of this
study.
SUPPORTING INFORMATION AVAILABLE
Experimental details, additional references, and Figures S1-13.
This material is available free of charge via the Internet at
http://pubs.acs.org.
Received for review March 27, 2010. Accepted June 25,
2010.
AC1007909
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6837

Anal. Chem. 2010, 82, 6838–6846
Analysis of Inorganic Polyphosphates by Capillary
Gel Electrophoresis
Andrew Lee and George M. Whitesides*
Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford Street, Cambridge, MA
This paper describes the development of a method that
uses capillary gel electrophoresis (CGE) to analyze mixtures
of inorganic polyphosphate ((Pi)n). Resolution of
(Pi)n on the basis of n, the number of residues of
dehydrated phosphate, is accomplished by CGE using
capillaries filled with solutions of poly(N,N-dimethylacrylamide)
(PDMA) and indirect detection by the UV
absorbance of a chromophore, terephthalate, added to
the running buffer. The method is capable of resolving
peaks representing (P i)n with n up to ∼70; preparation
and use of authentic standards enables the identification
of peaks for (Pi)n with n ) 1-10. The main
advantages of this method over previously reported
methods for analyzing mixtures of (Pi)n (e.g., gel
electrophoresis, CGE using polyacrylamide-filled capillaries)
are its resolution, convenience, and reproducibility;
gel-filled capillaries are easily regenerated by
pumping in fresh, low-viscosity solutions of PDMA. The
resolution is comparable to that of ion-exchange chromatography
and detection of (P i)n by suppressed
conductivity. The method is useful for analyzing (Pi)n
generated by the dehydration of Pi at low temperature
(125-140 °C) with urea, in a reaction that may have
been important in prebiotic chemistry. The method
should also be useful for characterizing mixtures of
other anionic, oligomeric, or polymeric species without
an intrinsic chromophore (e.g., sulfated polysaccharides,
oligomeric phospho-diesters).
This paper describes a technique for analyzing mixtures of
inorganic oligo- and polyphosphates, (Pi)n, by capillary gel
electrophoresis. Samples of condensed inorganic phosphate are
typically mixtures of (Pi)n with different chain length, n.
Mixtures of (Pi)n can have a wide range in n; (Pi)n with
estimated values of n as high as 1000 have been reported. 1
The analytical method developed here characterizes samples
of (Pi)n at high resolution by separating and detecting each
species in a mixture. Our primary motivation to develop a new
method of analysis was to explore the synthesis and reactivity
of (Pi)n relevant to the chemical origins of life (i.e., the prebiotic
chemistry leading to self-replicating systems in a “pre-RNA”
or “RNA world”). 2-7
* To whom correspondence should be addressed. E-mail: gwhitesides@
gmwgroup.harvard.edu.
(1) Clark, J. E.; Wood, H. G. Anal. Biochem. 1987, 161, 280–290.
(2) Joyce, G. F. Nature 1989, 338, 217–224.
6838 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Species of (Pi)n in aqueous solution are anionic and differ
from each other in the number of residues of condensed
phosphate and in net negative electrostatic charge. Capillary
electrophoresis (CE), in its most straightforward mode of
operation (capillary zone electrophoresis (CZE), that is electrophoresis
of analytes through free solution, combined with
optical detection of chromophoric analytes), cannot resolve and
detect (Pi)n. We developed a method that addresses the two
problems of (i) separating (Pi)n in mixtures and (ii) detecting
and quantifying each (Pi)n. Capillary gel electrophoresis (CGE),
using capillaries filled with aqueous solutions of poly(N,Ndimethylacrylamide)
(PDMA), resolved cyclic and linear (Pi)n
in order of their electrophoretic mobility. Addition of the
chromophoric anion, terephthalate (1,4-(CO2 - )2C6H4, abbreviated
as TP 2- ), to the running buffer enabled the detection of
separated (Pi)n by indirect UV absorbance.
We demonstrated the resolution of mixtures of (Pi)n (with n
up to ∼70) and the identification of components in mixtures
with the use of authentic standards. The areas of peaks in
electropherograms, determined by indirect detection, allowed
us to quantify the relative concentration of each species in a
mixture of (Pi)n. In addition to analyzing commercially available
samples of (Pi)n, prepared by the thermal dehydration of Pi
(>220 °C), we analyzed (Pi)n generated by dehydration reactions
that might have occurred on the prebiotic earth and, thus,
might have been involved in the chemical origins of life. 5,6,8
MOTIVATION
(Pi)n and adenosine triphosphate (ATP) have an essential
functional group in common: residues of dehydrated phosphate
connected by phosphoanhydride bonds. (Pi)n is simpler in
composition than ATP but, in principle, may provide the same
chemical function (e.g., activation of -OH groups) and has led
to the suggestion that (Pi)n is a molecular fossil, a species
important in the origin of life, and a precursor to ATP. 9-13 We
(3) Joyce, G. F. Nature 2002, 418, 214–221.
(4) Gilbert, W. Nature 1986, 319, 618–618.
(5) Keefe, A. D.; Miller, S. L. Origins Life Evol. Biospheres 1996, 26, 15–25.
(6) Keefe, A. D.; Miller, S. L. J. Mol. Evol. 1995, 41, 693–702.
(7) Pasek, M. A.; Kee, T. P.; Bryant, D. E.; Pavlov, A. A.; Lunine, J. I. Angew.
Chem., Int. Ed. Engl. 2008, 47, 7918–7920.
(8) Osterberg, R.; Orgel, L. E.; Lohrmann, R. J. Mol. Evol. 1973, 2, 231–234.
(9) Rao, N. N.; Gomez-Garcia, M. R.; Kornberg, A. Annu. Rev. Biochem. 2009,
78, 605–647.
(10) Kornberg, A. J. Bacteriol. 1995, 177, 491–496.
(11) Kornberg, A.; Rao, N. N.; Ault-Riche, D. Annu. Rev. Biochem. 1999, 68,
89–125.
(12) Orgel, L. E.; Lohrmann, R. Acc. Chem. Res. 1974, 7, 368–377.
(13) Lohrmann, R.; Orgel, L. E. Nature 1973, 244, 418–420.
10.1021/ac1008018 © 2010 American Chemical Society
Published on Web 07/27/2010

wished to explore the dehydration of Pi in detail and needed a
method that was more convenient and reproducible than those
reported so far for the resolution of mixtures of (Pi)n varying
in chain length. The method we have developed should also
be useful for investigating the biochemistry of (Pi)n 9-11 and for
developing (Pi)n as a reagent in chemical synthesis. In addition,
the method may be useful in applications for quality control:
(Pi)n is a component of many commercial materials (e.g.,
fertilizers, food products, detergent formulations, building
materials). 14
PREVIOUSLY REPORTED METHODS FOR
SEPARATING MIXTURES OF (PI)N
Polyacrylamide gel electrophoresis (PAGE) can resolve mixtures
of (Pi)n, with n ∼ 2-450. Analysis by PAGE involves gel
electrophoresis (typical runs require g3 h) and the detection
of (Pi)n by staining gels with the cationic dye toluidine blue O
(TBO); 1,15,16 autoradiography can detect species of (Pi)n synthesized
from 32 P-ATP. 17 The disadvantages to PAGE are the
difficulty of the experiments and the time required for analysis.
We found it difficult to cast 20% gels that are homogeneous
and provide reproducible resolution; to run a gel requires
several hours (for separation, staining, and destaining).
Anion-exchange chromatography is useful both for analyzing
mixtures of (Pi)n and for preparing samples of purified (Pi)n.
The best results for the chromatographic resolution of (Pi)n
are chromatograms showing up to ∼50 peaks, in runs of less
than 30 min; 18,19 this method requires HPLC instrumentation
with a suppressed conductivity detector and an online KOH
gradient generator (for minimizing the amount of CO2/CO3 2adsorbed
from the atmosphere). Distinguishing samples containing
(Pi)n with n > 45 is difficult by this method, and the
resolution of species with n ∼ 100 has not been demonstrated.
Other qualitative or semiquantitative methods used to analyze
mixtures of (Pi)n include paper chromatography, 20,21 31 P
NMR, 22-27 ESI-MS, 28 and the analysis of terminal phosphate
groups with phosphoglucokinase. 20
(14) Phosphoric Acid and Phosphates. Kirk-Othmer Encyclopedia of Chemical
Technology, 4th ed.; John Wiley & Sons: New York, 1991; Vol. 18, pp 669-
718.
(15) Ogawa, N.; DeRisi, J.; Brown, P. O. Mol. Biol. Cell 2000, 11, 4309–4321.
(16) Robinson, N. A.; Wood, H. G. J. Biol. Chem. 1986, 261, 4481–4485.
(17) Gomez-Garcia, M. R.; Kornberg, A. Proc. Natl. Acad. Sci. U.S.A. 2004,
101, 15876–15880.
(18) Baluyot, E. S.; Hartford, C. G. J. Chromatogr., A 1996, 739, 217–222.
(19) Sekiguchi, Y.; Matsunaga, A.; Yamamoto, A.; Inoue, Y. J. Chromatogr., A
2000, 881, 639–644.
(20) Greenfield, S.; Clift, M. Analytical Chemistry of the Condensed Phosphates; 1
ed.; Pergamon Press: New York, 1975.
(21) Osterberg, R.; Orgel, L. E. J. Mol. Evol. 1972, 1, 241–250.
(22) Rao, N. N.; Roberts, M. F.; Torriani, A. J. Bacteriol. 1985, 162, 242–247.
(23) Gard, D. R.; Burquin, J. C.; Gard, J. K. Anal. Chem. 1992, 64, 557–561.
(24) Crutchfield, M. M.; Callis, C. F.; Irani, R. R.; Roth, G. C. Inorg. Chem. 1962,
1, 813–817.
(25) Teleman, A.; Richard, P.; Toivari, M.; Penttilla, M. Anal. Biochem. 1999,
272, 71–79.
(26) Moreno, B.; Urbina, J. A.; Oldfield, E.; Bailey, B. N.; Rodrigues, C. O.;
Docampo, R. J. Biol. Chem. 2000, 275, 28356–28362.
(27) Glonek, T.; Lunde, M.; Mudgett, M.; Myers, T. C. Arch. Biochem. Biophys.
1971, 142, 508–513.
(28) Choi, B. K.; Hercules, D. M.; Houalla, M. Anal. Chem. 2000, 72, 5087–
5091.
CAPILLARY ELECTROPHORESIS OF (PI)N AND
DNA
Capillary electrophoresis separates and resolves analytes based
on differences in electrophoretic mobility. Since both the amount
of negative charge and hydrodynamic drag of (Pi)n increase with
n, in a way similar to that of single-stranded DNA, we expected
poor resolution in the analysis of (Pi)n in free-solution capillary
electrophoresis (i.e., CZE). 29 Capillaries filled with a sieving
matrix, however, are capable of resolving DNA in order of chain
length. 30-32 The use of replaceable solutions of entangled polymer
(such as linear acrylamide or PDMA) in CGE enabled the automated
and massively parallel analysis of samples of DNA and was
essential to the completion of the Human Genome Project. 33,34
Previous reports of methods for analyzing (Pi)n by CGE used
capillaries coated and filled with linear polyacrylamide (capillaries
were prepared by filling capillaries with aqueous acrylamide
and polymerizing in situ). 35,36 We, and others, found
that capillaries prepared this way had short lifetimes; 37-40 these
capillaries could not be reused, required time-consuming preparation
of new capillaries, and limited the reproducibility of the
method. Stover 36,41 and Wang 29,35 detected (Pi)n by indirect UV
absorbance in CGE experiments (chromate or pyromelltic acid
were the chromophores added to the running buffer). These
demonstrations did not, however, identify peaks for (Pi)n
beyond P3 nor did they quantify resolved (Pi)n.
EXPERIMENTAL DESIGN
CZE: Separation of (Pi)n in Free Solution. We used the
results of CZE experiments to guide the development of a CGE
technique. Although the resolution of mixtures of (Pi)n in free
solution is poor, CZE experiments are easier to run than CGE
experiments and allowed us to (i) test different chromophores
for indirect detection and identify terephthalate (TP 2- )asan
optimal choice; (ii) measure µ for analytical standards of (Pi)n
in free solution and determine the influence of pH and net
electrostatic charge on the mobility of (Pi)n (pH in the ranges
of 6.8-7.1 and 8.4-8.7); (iii) optimize the composition of the
running buffer. Table 1 gives literature values of pKa for (Pi)n.
Analysis by CZE required the migration of (Pi)n to the anode
and the use of capillaries with suppressed electro-osmotic flow.
Capillaries covalently modified by reaction of the fused silica
surface with a copolymer of N,N-dimethylacrylamide and
3-methacryloxy-propyltrimethoxysilane 42 had an electro-osmotic
flow of

Table 1. Values of pKa of Inorganic Polyphosphates
abbreviation name pKa
P1
P2
P3
cyclo-P3
(Pi)n
orthophosphate a
pyrophosphate a
tri(poly)phosphate b
trimetaphosphate a
polyphosphate d
2.15, 7.20, 12.35
0.8, 2.2, 6.7, 9.4
0.5, 1.0, 2.4, 6.5, 9.4
2.05 c
∼1-2, 7.2-8.2 d
a Values at infinite dilution and 25 °C, taken from ref 57. b Values at
infinite dilution and 25 °C, taken from ref 58. c Results of titration of
cyclo-P3 cannot distinguish the pKa of the three ionizable groups of
cyclo-P3; each of the three groups has a value of pKa of approximately
2.05. d Values taken from ref 14; ranges are inferred from titration
curves for long chain polyphosphates and describe the strongly acidic
hydrogen at each residue of phosphate (pKa of 1 to 2) and two weakly
acidic hydrogens at the ends of the chain (pKa of 7.2-8.2).
CGE: Resolution of (Pi)n in Order of n. To achieve a
combination of high resolution, speed, and convenience in the
analysis of (Pi)n, we used CGE with solutions of PDMA (average
molecular weight of 57 kDa, 9.1% w/v) as the sieving medium.
A key advantage of the use of a solution of PDMA 39,43-47 is
its low viscosity ( 6.8; composition of the buffer
discussed below). TP 2- carries current in the capillary and
migrates to the anode during electrophoresis. Migration of TP 2-
(43) Heller, C. Electrophoresis 1999, 20, 1978–1986.
(44) Wang, Y. M.; Liang, D. H.; Ying, O. C.; Chu, B. Electrophoresis 2005, 26,
126–136.
(45) Rosenblum, B. B.; Oaks, F.; Menchen, S.; Johnson, B. Nucleic Acids Res.
1997, 25, 3925–3929.
(46) He, H.; Buchholz, B. A.; Kotler, L.; Miller, A. W.; Barron, A. E.; Karger,
B. L. Electrophoresis 2002, 23, 1421–1428.
(47) Barbier, V.; Viovy, J. L. Curr. Opin. Biotechnol. 2003, 14, 51–57.
(48) Zhou, H. H.; Miller, A. W.; Sosic, Z.; Buchholz, B.; Barron, A. E.; Kotler,
L.; Karger, B. L. Anal. Chem. 2000, 72, 1045–1052.
(49) Kotler, L.; He, H.; Miller, A. W.; Karger, B. L. Electrophoresis 2002, 23,
3062–3070.
(50) Gao, Q. F.; Yeung, E. S. Anal. Chem. 1998, 70, 1382–1388.
(51) Giovannoli, C.; Anfossi, L.; Tozzi, C.; Giraudi, G.; Vanni, A. J. Sep. Sci. 2004,
27, 1551–1556.
(52) Fung, E. N.; Pang, H. M.; Yeung, E. S. J. Chromatogr., A 1998, 806, 157–
164.
(53) Huang, M. F.; Huang, C. C.; Chang, H. T. Electrophoresis 2003, 24, 2896–
2902.
(54) Shamsi, S. A.; Danielson, N. D. Anal. Chem. 1995, 67, 1845–1852.
(55) Wang, P. G.; Giesel, R. W. Anal. Biochem. 1995, 230, 329–332.
(56) Wang, P. G.; Giese, R. W. Anal. Chem. 1993, 65, 3518–3520.
(57) Smith, R. M.; Martell, A. E. Critical Stability Constants; Plenum Press: New
York, 1989.
(58) Smith, R. M.; Martel, A. E. NIST Critically Selected Stability Constants of
Metal Complexes; National Institute of Standards and Technology: Gaithersburg,
MD, 1997.
6840 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
by the detector produces a steady-state signal in UV absorbance,
generating the baseline of electropherograms. Sample
zones containing (Pi)n also migrate toward the anode. Within
a sample zone, the current is carried in part by (Pi)n, rather
than TP 2- . Zones containing (Pi)n are detected by a decrease
in UV absorbance, resulting from a decrease in the concentration
of TP 2- .
The sensitivity of detection of (Pi)n depends primarily on the
extinction coefficient of TP 2- and the mobilities of TP 2- and
(Pi)n. Terephthalate, TP 2- , absorbs at λ ) 254 nm (ε ) 8.2 · 10 3
M -1 cm -1 ) and has mobility (µ ∼ 28 cm 2 kV -1 min -1 ) close to
that of (Pi)n (26-34 cm 2 kV -1 min -1 in free solution, pH ) 7.0).
The similarity in µ for TP 2- and (Pi)n is important for limiting
dispersion in electromigration and destacking, which result in
the asymmetry and broadening of peaks. 59 The estimated limit
of detection of residues of Pi is ∼0.2 µM.
Ions in the Running Buffer. The running buffer, by design,
contains only one type of anion: TP 2- . This characteristic avoids
complications in the indirect detection of (Pi)n by the absorbance
of TP 2- . Anions in addition to TP 2- would lead to system
zones (that show up as negative peaks unrelated to (Pi)n),
artifacts in peak shape, and complications in the analysis of
peak areas. 59-63 We, therefore, used buffers prepared by adding
terephthalic acid to solutions of 18.0-24.0 mM bis-tris (pKa )
6.46) or tris (pKa ) 8.06). The resulting buffers contained TP 2-
(3.0 mM), protonated amine (6.0 mM bis-tris-H + or tris-H + ),
and free amine (12.0-18.0 mM of bis-tris or tris).
pH. Running buffers made with bis-tris or tris allowed us to
compare the resolution of (Pi)n at two ranges of pH: 6.8-7.2
(bis-tris) and 8.4-8.7 (tris). These values of pH are well
beyond the pKa of the first ionizable group of phosphate
residues (∼2) but are near the pKa of the second ionizable
groups of the terminal residues of phosphate of (Pi)n (∼6.3-7.2)
(Table 1). Values of mobility for (Pi)n are sensitive to the pH of
the running buffer, particularly for n < 5.
Addition of Polyethylene Glycol (PEG) to Reservoirs of
Running Buffer. In CGE, capillaries are filled with a solution of
PDMA (density ∼1.06 g/mL); the open ends of the capillary are
immersed into reservoirs of running buffer that do not contain
PDMA (∼1.01 g/mL). As a result, solutions of PDMA leak out of
the capillary into the buffer reservoirs, under gravity. The
reproducibility of this experiment was, therefore, poor; the current
decreased by >10% within 2 h, and retention times increased by
>5% in each subsequent run.
The solution to this problem was to add polyethylene glycol
(PEG) to the reservoirs of running buffer, generating solutions
isodense with the solution inside the capillary. PEG is water
soluble and uncharged; it does not migrate during electrophoresis.
CGE experiments using reservoirs of running buffer with 9.0%
PEG (w/v) showed stable currents and improved run-to-run
reproducibility. This procedure was essential for maintaining a
(59) Poppe, H.; Xu, X. In High Performance Capillary Electrophoresis: Theory,
Techniques, and Applications, 1 ed.; Khaledi, M., Ed.; Wiley & Sons: New
York, 1998; Vol. 146.
(60) Gas, B.; Kenndler, E. Electrophoresis 2004, 25, 3901–3912.
(61) Beckers, J. L.; Bocek, P. Electrophoresis 2003, 24, 518–535.
(62) Bruin, G. J. M.; Vanasten, A. C.; Xu, X. M.; Poppe, H. J. Chromatogr. 1992,
608, 97–107.
(63) Macka, M.; Haddad, P. R.; Gebauer, P.; Bocek, P. Electrophoresis 1997,
18, 1998–2007.

constant medium for separation; we routinely collected data for
120 min of electrophoresis for each preparation of a filled capillary
(enough for 5 to 6 typical analyses of (Pi)n). 64
Loading Samples by Electrokinetic Injection. During electrokinetic
injection, an applied field forces anions in the sample
to enter the capillary and migrate toward the anode. Rather than
transferring plugs of solution into the capillary, electrokinetic
injection only transfers anions. This procedure avoids the creation
of discontinuities in the capillary and improves the reproducibility
of separation medium.
The total number of ions injected is determined by the
electrical field, conductivity of the running buffer, duration of the
injection, and to a smaller extent, the conductivity of the sample
(further discussion in the Supporting Information). Typical injections
(175 V cm -1 for 2.0 s) correspond to the loading of ∼0.1
nmol of charge. The amount of each ion injected depends on
both the concentration and mobility of the ion, as well as the
concentration and mobility of all other ions in the sample. Our
quantitative treatment of the data accounts for the bias in
sampling caused by electrokinetic injection (discussed in the
Results and Discussion section and Supporting Information). 65,66
Sample Preparation. The procedure described above results
in the injection and detection of all anions in a sample, not just
(Pi)n. 67 Analysis of (Pi)n by indirect detection works best on
samples that are free of salts besides (Pi)n. The following steps
were useful for preparing samples free of additional anions,
originating from either synthesis or preparative separation: (i)
adsorption to anion-exchange resin; (ii) elution of anions other
than (Pi)n (e.g., Cl - ) with 0.1 M Na2CO3; (iii) elution of (Pi)n
with concentrated 2.0 M NH4HCO3; (iv) removal of NH4HCO3
under vacuum. 68
Analytes: Authentic Standards of (Pi)n (n ) 1-10).
Commercially available oligophosphates of a single chain length
are limited to P1,P2,P3, and cyclo-P3. Preparative-scale separation
of a mixture of (Pi)n (117% polyphosphoric acid 69 ), by anionexchange
chromatography, generated samples of purified (Pi)n,
n ) 4-10, that served as analytical standards. By analyzing
standards added to mixtures of (Pi)n, we identified peaks
representing (Pi)n with n ) 1-10.
Analytes: Mixtures of (Pi)n. We demonstrated the resolution
of the method by analyzing commercially available samples of
higher (Pi)n, covering a range of average length (n¯): 117%
polyphosphoric acid, n¯ ) 17, n¯ ) 21, n¯ ) 48, and n¯ ) 65. In
addition, we prepared samples of (Pi)n generated by the
dehydration of NH4H2PO4 in mixtures with urea, using conditions
reported by Orgel et.al., which were presumed to be
plausible in prebiotic chemistry. 8,21
Internal standard(s). We added an internal standard,
K + CH3SO3 - , to each sample prior to analysis by CE. The peak
(64) We used PEG instead of PDMA in buffer reservoirs because commercial
PDMA (∼$100 per g) is much more expensive than PEG (∼$0.1 per g);
each set of experiments requires ∼1 g of polymer. PDMA can be obtained
inexpensively, however, by preparing it from N,N-dimethylacrylamide.
(65) Huang, X. H.; Gordon, M. J.; Zare, R. N. Anal. Chem. 1988, 60, 375–377.
(66) Rose, D. J.; Jorgenson, J. W. Anal. Chem. 1988, 60, 642–648.
(67) Satow, T.; Machida, A.; Funakushi, K.; Palmieri, R. J. High Resolut.
Chromatogr. 1991, 14, 276–279.
(68) Cohn, W. E.; Bollum, F. J. Biochim. Biophys. Acta 1961, 48, 588–590.
(69) The nomenclature (i.e., 117%) is based on comparing the ratio of phosphorus
to oxygen in samples of dehydrated phosphate to the ratio of a standard
solution of 85% phosphoric acid.
observed for the internal standard allowed us to (i) monitor
the reproducibility of the method; (ii) define the mobilities of
(Pi)n relative to that of a standard (CH3SO3 - )(µP,rel in eq 1);
(iii) determine the relative and absolute concentrations of resolved
(Pi)n, by comparing the areas of peaks for CH3SO3 - and (Pi)n.
In eq 1, tCH3SO3 - and tP are the retention times for CH3SO3 - and
(Pi)n, V is the voltage, L1 is the length of the capillary, and L2
is the distance between the inlet and detector.
µ P,rel ) µ P - µ CH3SO - )
3 L1L2 V (
1
t CH3 SO 3 -
- 1
t P)
The electrophoretic mobility of CH3SO3 - in free solution was
near that of (Pi)n, but peaks for CH3SO3 - and (Pi)n did not
overlap (µCH3SO3 - ) 26.7 cm 2 kV -1 min -1 ). For experiments
requiring additional internal standards, we also used Cl - ,
CF3CO2 - , and CH3-C6H4-SO3 - .
RESULTS AND DISCUSSION
Resolution and Detection of Lower Oligophosphates by
CZE. We analyzed mixtures of P1, P2, P3, and cyclo-P3 by CZE,
using coated capillaries and running buffer composed of 3.0
mM TP 2- and 18.0 mM bis-tris (pH ) 6.9) or 24.0 mM tris
(pH ) 8.4); 70 data from these experiments are shown in the
Supporting Information. The results demonstrated (i) the indirect
detection of anions, enabled by the steady-state absorbance signal
of TP 2- at 254 nm and (ii) mobilities in the order cyclo-P3 > P3
> P2 > P1 (at pH ) 6.9) that correspond to the number of
ionizable groups, values of pKa, and structures of these (Pi)n
(i.e., the radius of cyclo-P3 is constrained in a way that P3 is
not).
Resolution of Mixtures of (Pi)n by CGE. The resolution of
mixtures of (Pi)n with n > 5 required the use of a sieving gel.
The best results were obtained with capillaries filled with
solutions of 9.1% PDMA (w/v; average molecular weight 58.9
kDa) in running buffer (24.0 mM tris, 3.0 mM terephthalic acid,
pH ) 8.4). We filled capillaries (100 µm internal diameter and
57 cm in length) by pumping in solutions of PDMA with
positive pressure (30 psi of ultrahigh purity N2 applied to the
inlet) for 10 min. During electrophoresis, the ends of the
capillary were immersed into reservoirs of running buffer
containing 9% PEG (w/v; average molecular weight of 1.5 kDa).
The upper trace in Figure 1A shows the separation of (Pi)n in
a commercially available mixture of sodium polyphosphate
(reported chain length of n¯ ∼ 17). The sample had a concentration
of 19.0 mM (in phosphate residues); P3 (2.0 mM) and
CH3SO3 - (2.0 mM) were added to serve as internal standards.
For comparison, the lower trace in Figure 1A shows the analysis
of the same sample by CZE.
We identified peaks for P1, P2, and P3 (marked with a dotted
line in Figure 1B) by comparing the traces in Figure 1A to those
collected for samples containing standards added to the mixture.
In CGE experiments, mobilities of (Pi)n with n > 3 are lower than
that of CH3SO3 - . In CZE, the mobilities of all (Pi)n are greater
than that of CH3SO3 - . The contrast in the order of mobility
(70) Capillaries modified by the method of Cretich 42 were used over the course
of several months in more than 100 runs, without observable change in
the retention times for analytical standards.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(1)
6841

Figure 1. Resolution of (Pi)n by CGE. (A) Analysis of sodium polyphosphate (average chain length 17) in capillaries filled with 9.1% PDMA
(w/v) gel (upper trace) or running buffer alone (24.0 mM tris, 3.0 mM terephthalic acid, pH ) 8.4, 25 °C) (lower trace). Electrophoresis was
performed by applying 14.6 kV across capillaries having a length of 57 cm (50 cm from the inlet to detector). (B,C) Expanded views of the upper
and lower traces in A, respectively.
indicates size-sieving during CGE and the separation of (Pi)n
in order of n. The series of peaks detected by CGE (shown in
Figure 1B) suggests species as large as ∼P35 in the sample.
6842 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Preparation of (Pi)n, n ) 4-10. Identification of peaks
observed for broad distributions of (Pi)n required oligophosphate
standards that are not commercially available. We separated a

Figure 2. Standards of (Pi)n, n ) 1-10. CZE traces analyzing a mixture of commercially available P1, P2, P3, cyclo-P3, and CH3SO3 - (top
trace) and analytical standards P4-P10 purified by anion-exchange chromatography (no internal standard added). CZE was performed by applying
14.6 kV across coated capillaries (57 cm in length, 50 cm from the inlet to detector) filled with running buffer composed of 18.0 mM bis-tris and
3.0 mM terephthalic acid (pH ) 6.9, 25 °C).
Figure 3. Identification of P3-P10 in mixtures of (Pi)n by CGE. Traces are for the analysis of mixtures of (Pi)n (19 mM in total phosphate, pH
∼ 8) with the addition of analytical standards for (Pi)n (2 mM). Mixtures were resolved by CGE at 14.6 kV using capillaries filled with 9.1% PDMA
(w/v) and running buffer (24.0 mM tris, 3.0 mM TP 2- ,pH) 8.4). Dotted lines mark peaks for species identified by added standards. Traces for
samples with added P7 or P9 are available in the Supporting Information.
mixture of (Pi)n (n ∼ 1-10, derived from 117% polyphosphoric
acid 69 ) on an anion-exchange chromatography column (Cl -
form) by elution with a gradient of KCl(aq). A second application
of anion-exchange chromatography removed KCl from fractions
containing (Pi)n (HCO3 - form; elution with NH4HCO3). Removal
of NH4HCO3 by vacuum generated oligophosphate standards
as the ammonium salt.
CZE traces analyzing purified (Pi)n, n ) 4-10, show one
major peak in each sample (Figure 2). Small peaks next to the
major peak and a small peak for P1 (top trace in Figure 2 for
reference) suggest small amounts of (Pi)(n+1) or (Pi)(n-1), and P1.
The relatively clean traces suggest that the conditions used in
the preparation of samples do not cause extensive hydrolysis
of (Pi)n or equilibration in chain length. Characterization of
oligophosphate standards by 31 P NMR is available in the
Supporting Information.
Identification of Peaks for n ) 3-10 in Mixtures of (Pi)n.
Figure 3 shows CGE data analyzing mixtures of (Pi)n (19.0 mM
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6843

Figure 4. Resolution of commercially available mixtures of (Pi)n. Samples consisting of commercially available mixtures of polyphosphate
glass (20 mM in total phosphate, Na + salt) and CH3SO3 - (2 mM) were analyzed by CGE. After electrokinetic injection (4.0 s at 10 kV), (Pi)n were
separated by electrophoresis at 14.6 kV in coated capillaries (57 cm in length, 50 cm between inlet and detector) filled with solutions of 9.1%
PDMA (w/v) and running buffer (24.0 mM tris, 3.0 mM TP 2- ,pH) 8.4). Peaks in trace C, for n ∼ 10-40, appear less sharp than the peaks in
other runs. The reason for this difference is unclear; one possibility is that the peaks are affected by irregularities in pH within capillaries during
separation, caused by a small mismatch in pH between samples (∼8) and the running buffer (pH ) 8.4). 72
in phosphate residues; n¯ ∼ 17) and added P3, P4, P6, P8, orP10
(2.0 mM). The x-axis of traces in Figure 3 are in units of mobility
relative to CH3SO3 - µp,rel (eq 1). The traces show the alignment
of peaks from run to run and allowed us to assign peaks with
increased area to specific (Pi)n (dotted lines in Figure 3). The
dotted line for n ) 20 in Figure 3B is based on the reasonable
assumption that the peaks continue in order of n. 71
Mixtures of (Pi)n with n up to ∼70. The CGE data in Figure
4 characterizes commercially available mixtures of (Pi)n with
different distributions in chain length. Samples analyzed in
traces A-D are in order of increasing average chain length.
The results show that peaks with n up to ∼70 can be resolved
in a single run; longer (Pi)n may potentially be resolved using
lower concentrations of PDMA.
Shapes of Peaks. The asymmetric peaks observed in both
CZE and CGE experiments are typical for electropherograms
collected by indirect UV absorbance. (Pi)n with mobility greater
than that of TP 2- shows up as peaks broadened to the left,
while (Pi)n with mobility less than that of TP 2- show up as peaks
broadened to the right. These shapes are the results of
dispersion by electromigration, originating from (i) differences
in mobility between analytes and TP 2- and (ii) nonuniform
electric fields inside sample zones. Discussion of the origin of
the shapes of peaks is available in the Supporting Information,
as well as in ref 59.
Quantitative Analysis of (Pi)n: Areas of Peaks. Three
contributions determine the area of a peak: (i) the response of
[TP 2- ]to(Pi)n; (ii) the amount of (Pi)n transferred from the
sample to the capillary by electrokinetic injection; (iii) residence
time of the analyte passing the detector. Analysis of the areas
of peaks for (Pi)n and CH3SO3 - enables the quantification of
resolved (Pi)n. To account for the effect of (iii), areas of peaks
are adjusted by the factor (1/ti), where t is the retention time
of analyte i. 73 In eq 2, the ratio of adjusted areas for analyte i
and CH3SO3 - (Ai and ACH3SO3 -) is related to the concentrations
[i]S and [CH3SO3 - ] of the sample: zi and zCH3SO3 - are the
electrostatic charge of i and CH3SO3 - ; µCH3SO3 - and µi are the
mobilities of i and CH3SO3 - ; µC + is the mobility of the cation
in the sample zone (bis-tris-H + or tris-H + ).
A i
A CH3 SO 3 -
[i] S
)
[CH3SO3 - ] · [
z i
z CH3 SO 3 -]
· µ + µ C+ i [ µ + µ C+ CH3SO -] 3
(2)
The Supporting Information contains our derivation of eq 2 and a
discussion of the quantitative aspects of electrokinetic injection
and indirect UV absorbance.
Equation 2 reveals the advantage of analyzing electropherograms
by comparing peaks for (Pi)n and CH3SO3 - . The ratio (Ai/
ACH3SO3 -) depends on the ratio ([i]S/[CH3SO3 - ]) but does not
depend on the concentration of other ions in the sample or
the voltage and duration of electrokinetic injection.
Calibration of Peak Areas to Concentrations of P1, P2,
P3, and cyclo-P3. Analysis of samples of containing P1, P2, P3,
or cyclo-P3 and added CH3SO3 - by CZE demonstrated the
relationship between peak areas and concentrations. Figure 5
shows values of (Acyclo-P3 /ACH3SO3 -) determined from the analysis
of samples containing cyclo-P3 and CH3SO3 - . Samples had ratios
(71) The alignment is not perfect; small differences from run to run are possibly
due to dispersion in electromigration, caused by differences between
samples in pH or ionic strength.
(72) Another possibility is that electrokinetic injections may result in concentrations
of (Pi)n at the inlet that can lead to precipitation. (73) Hilser, V. J.; Freire, E. Anal. Biochem. 1995, 224, 465–485.
6844 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010

Figure 5. Quantitative calibration of cyclo-P3. The plot shows values
for Acyclo-P3 /ACH3SO3 - determined in the analysis of cyclo-P3 and
CH3SO3 - by CZE, using coated capillaries (57 cm in length, 50 cm
between inlet and detector) filled with running buffer (18.0 mM
bis-tris, 3.0 mM TP2- ,pH) 6.9). Points for ([cyclo-P3]/[CH3SO3 - ])
) 0.1, 1.0, or 10.0 are the average taken from eight experiments.
Standard deviations for (Acyclo-P3
/ACH3SO3 -) are 40 and distinguished the composition of
mixtures generated at 125 and 140 °C. Cyclo-P3 is the most
abundant species in mixtures prepared at 140 °C. In contrast,
the mixture prepared at 125 °C does not contain cyclo-P3, despite
containing linear (Pi)n with n > 40. The reason for the
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6845

preferential formation of cyclo-P3 over linear (Pi)n at 140 °C is
not clear; the synthesis of cyclo-P3 is, however, potentially
important for prebiotic chemistry. Reactions of (Pi)n with -OH
groups, leading to polyphosphorylated compounds, likely
depend on whether (Pi)n are cyclic or linear. Reactions of linear
(Pi)n potentially transfer phosphate residues from either terminal
or middle positions of the chain, while reactions of cyclo-
P3 are ring-opening and can generate a triphosphate group
similar to that of ATP. The resolution and convenience of CGE
should enable a broad survey of conditions for the synthesis
of (Pi)n and for reactions of (Pi)n with -OH groups. The results
should be helpful in refining hypotheses for the importance of
dehydrated phosphate in the chemical origins of life.
CONCLUSION
Previously reported methods using electrophoresis (slab gels
or CGE) to analyze (Pi)n successfully resolved species on the
basis of their size (n). These methods provided a way to
qualitatively characterize mixtures of (Pi)n at high resolution
but involved difficult and time-consuming experiments with
limited reproducibility. The method we have demonstrated in
this paper exploits the sieving properties of low-viscosity
solutions of PDMA and is capable of high resolution and
quantitative analysis. The advantages of CGE using solutions
of PDMA as the separation medium are convenient preparation,
rapid analysis, and reproducibility (enabled by refilling capillaries
with fresh solutions of gel). Our identification of P 1-P10
and resolution up to P70 validates a method that will be useful
6846 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
in studies of (Pi)n relevant to prebioitic chemistry and
biochemistry.
In addition to characterizing the composition of samples of
(Pi)n, the method we have developed is potentially useful for
analyzing other anionic, oligomeric species without a sensitive
chromophore. Examples relevant to prebiotic chemistry are
oligomers of phosphate condensed with organic compounds
(e.g., structures with formula (PO3 - -RO)n or (P2O6 2- -RO)n)).
Examples of biological polymers include teichoic acid, hyaluronic
acid, and sulfated polysaccharides such as heparin and
chondroitin sulfate.
ACKNOWLEDGMENT
We acknowledge the Harvard University Origins of Life Initiative
for research support. We thank Douglas B. Weibel, Katherine L.
Gudiksen, Paul J. Bracher, and Dosil Pereira de Jesus for technical
assistance and helpful discussion. A.L. acknowledges salary support
from NIH GM051559.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in the text, as well as
information about experimental procedures and sources of
chemicals. This material is available free of charge via the
Internet at http://pubs.acs.org.
Received for review March 28, 2010. Accepted July 10,
2010.
AC1008018

Anal. Chem. 2010, 82, 6847–6853
Method to Determine 226 Ra in Small Sediment
Samples by Ultralow Background Liquid
Scintillation
Joan-Albert Sanchez-Cabeza,* ,†,‡ Laval Liong Wee Kwong, § and Maria Betti §
Institute of Environmental Science and Technology, and Physics Department, Autonomous University of Barcelona,
ES-08193 Bellaterra, Spain, Departamento de Medio Ambiente, CIEMAT, 28040 Madrid, Spain, and Environment
Laboratories, International Atomic Energy Agency, MC-98000 Monaco
210 Pb dating of sediment cores is a widely used tool to
reconstruct ecosystem evolution and historical pollution
during the last century. Although 226 Ra can be
determined by γ spectrometry, this method shows
severe limitations which are, among others, sample
size requirements and counting times. In this work,
we propose a new strategy based on the analysis of
210 Pb through 210 Po in equilibrium by r spectrometry,
followed by the determination of 226 Ra (base or supported
210 Pb) without any further chemical purification
by liquid scintillation and with a higher sample throughput.
Although γ spectrometry might still be required
to determine 137 Cs as an independent tracer, the effort
can then be focused only on those sections dated
around 1963, when maximum activities are expected.
In this work, we optimized the counting conditions,
calibrated the system for changing quenching, and
described the new method to determine 226 Ra in small
sediment samples, after 210 Po determination, allowing
a more precise determination of excess 210 Pb ( 210 Pbex).
The method was validated with reference materials
IAEA-384, IAEA-385, and IAEA-313.
The natural radionuclide 210 Pb is used as an environmental
tracer of numerous biogeochemical processes in the aquatic,
soil, and atmospheric sciences. Among these, possibly the most
common application is its use in the reconstruction of recent
environmental changes through 210 Pb dating, as described in
some seminal papers. 1-5 210 Pb (half-life ) 22.23 ± 0.12 years) 6
* To whom correspondence should be addressed. Phone: +34 93 581 1915.
E-mail: joanalbert.sanchez@uab.cat.
† Autonomous University of Barcelona.
‡ CIEMAT.
§ International Atomic Energy Agency.
(1) Goldberg, E. D. In Radioactive Dating; International Atomic Energy Agency:
Vienna, Austria, 1963; pp 121-131.
(2) Crozaz, G.; Picciotto, E.; de Breuck, W. J. Geophys. Res. 1964, 69, 2597–
2604.
(3) Krishnaswamy, S.; Lal, D.; Martin, J.; Meybeck, M. Earth Planet. Sci. Lett.
1971, 11, 407–414.
(4) Robbins, J. A. In Biochemistry of Lead; Nriagu, J. O., Ed.; Elsevier:
Amsterdam, The Netherlands, 1998; pp 285-393.
(5) Appleby, P. G.; Oldfield, F. Catena 1978, 5, 1–8.
(6) All half-lives used in this work are from Decay Data Evaluation Project,
http://www.nucleide.org/DDEP.htm, updated by the “Laboratoire National
Henri Becquerel” on January 22, 2010.
is a natural radionuclide of the 238 U radioactive chain. In closed
systems, 226 Ra (T1/2 ) 1600 ± 7 years) decays, through various
daughter radionuclides, to 210 Pb, named base or supported
210 Pb, which should be in equilibrium with 226 Ra (base 210 Pb )
226 Ra). On the other hand, some 210 Pb can reach bottom aquatic
sediments from either the atmosphere (after 222 Rn exhalation
from soils) or the water column (in situ production). This is
called the excess or unsupported fraction ( 210 Pbex) and is the
basis of all 210 Pb dating models. 7 Therefore, bottom sediments
contain a mixture of base and excess 210 Pb, and 210 Pbex is
determined by the difference between the total 210 Pb and 226 Ra
concentration for each sediment section ( 210 Pbex ) 210 Pb -
226 Ra). This is commonly determined in sections (typically 1
cm width) of undisturbed sediment cores collected from areas
of interest.
High-resolution γ spectrometry with HPGe (high-purity Ge)
detectors is commonly used to simultaneously determine both
210 Pb and 226 Ra in sediment samples. 8 Some of the reasons for
its success are (i) 137 Cs is also determined, which may be used
to validate the 210 Pb chronology, 9 and (ii) it is nondestructive,
as it does not require chemical separation. However, it has
important drawbacks for 210 Pb dating: (i) calibration in the low
γ emission energy region is difficult and requires tedious selfabsorption
corrections, thus affecting method accuracy, (ii)
precision is limited by the unavoidable presence of a relevant
background due to Compton scattered electrons in the detector,
(iii) counting times are long, thus sample throughput is small,
and (iv) most importantly for 210 Pb dating applications, sample
size requirement is usually large, typically more than 5 g dry
weight (dw), although some laboratories with HPGe well-type
detectors and/or special low-level counting conditions might
use a sample size as low as 1-2 g dw. Sample size is in many
cases a limitation for 210 Pb dating as commonly a large number
of other analyses need to be carried out. Although all samples
could be measured by γ spectrometry and saved for further
analysis, special care is needed during sample manipulation
(e.g., if the sample cannot be ground) and in order to avoid
(7) Appleby, P. G. Chronostratigraphic techniques in recent sediments. In
Tracking Environmental Change Using Lake Sediments: Basin Analysis,
Coring, and Chronological Techniques, Last, W. M., Smol, J. P., Eds.; Kluwer
Academic, 2001; Vol. 1, pp 171-203.
(8) Schelske, C. L.; Peplow, A.; Brenner, M.; Spencer, C. N. J. Paleolimnol.
1994, 10, 115–128.
(9) Smith, J. N. J. Environ. Radioact. 2001, 55, 121–123.
10.1021/ac1008332 © 2010 American Chemical Society 6847
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/16/2010

contamination of the substances of interest (e.g., trace metals).
When this strategy is used, the time needed to complete γ
spectrometry before further analyses can be performed also
needs to be taken into consideration.
In many laboratories, the total 210 Pb concentration is determined
by R spectrometry, through its daughter radionuclide
210 Po in equilibrium. The technique is simple, rapid, reliable
(recoveries >95%), and shows good precision. 10,11 Furthermore,
although the technique is destructive, only ca. 250 mg of dry
sediment is needed, and because laboratories usually have several
R detectors (they are much cheaper than Ge detectors), sample
throughput is much larger than with γ spectrometry. Typical
counting times range from 1 to 7 days, depending on sample
activity and desired uncertainty. An experimented analyst might
produce a full total 210 Pb profile within a month after sample
receipt (ca. 40 samples). When R spectrometry is used, the
base 210 Pb can only be indirectly determined by assuming that
the 226 Ra concentration is constant along the profile and
therefore estimated as the average of concentrations in the
profile bottom sections, as equilibrium should have been
reached. 12 However, as 226 Ra may vary along the profile, this
may lead to inaccurate 210 Pbex values. It is not uncommon that,
once the total 210 Pb profile is known, γ spectrometry is carried
out in selected sections, thus optimizing the use of this limiting
resource, although because of sample size requirements this
might be impracticable for many laboratories. As γ spectrometry
is carried out on the untreated sample and 210 Po on a
digestate, these techniques sometimes yield results which are
not fully consistent.
Liquid scintillation counting (LSC), mostly used for the
determination of � emitters, 13,14 has also been extensively used
to quantify R emitters, such as 226 Ra, in environmental samples
(mainly waters). 15,16 Some of the reported methods include direct
R LSC after water sample concentration, 17 226 Ra extraction from
water samples with specific scintillation cocktails such as
Radaex, 18 or generation of a radium-barium sulfate coprecipitate
that is transformed into a soluble chloride or nitrate. 19,20
Villa et al. 21 opted for this approach for sediment: the sediment
is digested, and after the elimination of actinides as hydroxides,
radium is recovered as Ra-Ba-SO4, dissolved in EDTA 0.2 M
ammonia solution, and counted. However, most of these
methods cannot be directly used with sediment digestates and/
or are excessively resource-consuming.
(10) Sanchez-Cabeza, J. A.; Masqué, P.; Ani-Ragolta, I. J. Radioanal. Nucl. Chem.
1998, 227, 19–22.
(11) Vesterbacka, P.; Ikaheimonen, T. K. Anal. Chim. Acta 2005, 545, 252–
261.
(12) Binford, M. W. J. Paleolimnol. 1990, 3, 253–268.
(13) Pujol, L.; Sanchez-Cabeza, J. A. J. Radioanal. Nucl. Chem. 1999, 2, 391–
398.
(14) Liong Wee Kwong, L.; LaRosa, J. J.; Lee, S. H.; Povinec, P. P. J. Radioanal.
Nucl. Chem. 2000, 248, 751–755.
(15) Salonen, L. Sci. Total Environ. 1993, 130-131, 23–35.
(16) Salonen, L.; Hukkanen, H. J. Radioanal. Nucl. Chem. 1997, 226, 67–74.
(17) Sanchez-Cabeza, J. A.; Pujol, L. Analyst 1998, 123, 399–403.
(18) Aupiais, J. Anal. Chim. Acta 2005, 532, 199–207.
(19) Repinc, U.; Benedik, L. J. Radioanal. Nucl. Chem. 2002, 254, 181–185.
(20) Galan-Lopez, M.; Martin-Sanchez, A.; Tosheva, Z.; Kies, A. In LSC 2005
Advances in Liquid Scintillation Spectrometry; Chalupnik, S., Schoenhofer,
F., Noakes, J., Eds.; Radiocarbon: Tucson, AZ, 2006; pp 165-170.
(21) Villa, M.; Moreno, H. P.; Manjón, G. Radiat. Meas. 2005, 39, 543–550.
6848 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
After 210 Po analysis for R spectrometry, 226 Ra remains in
solution. In this work, we propose, develop, and validate a new
method to determine 226 Ra by 222 Rn emanation to a scintillation
cocktail, which eliminates the need to perform any further
purification, and counting with an ultralow background liquid
scintillation system.
EXPERIMENTAL SECTION
Equipment. An ultralow background liquid scintillation system,
Quantulus 1220 TM (Wallac, Turku, Finland), was used to
carry out this work. In this system, background is reduced by
an optimized combination of active and passive shields. A pulseshape
analysis (PSA) circuit permits the discrimination of
pulses produced by R and � radiation by comparing the area
of the pulse tail after 50 ns from the start with its total area.
Pulse-shape discrimination is accomplished using a software
adjustable parameter (PSA parameter) which can vary between
1 and 256. 16,22 Quenching (sample extinction) was quantified with
the standard quenching parameter (SQP(E)) which is used to
determine the counting efficiency for each sample through
calibration curves. 23,24 Counting was performed with Wallac
OptiScint HiSafe III, a diisopropyl naphthalene based aqueous
immiscible cocktail, and low-diffusion PE counting vials (Packard
BioScience).
Counting Solutions. The tracer solutions were prepared by
gravimetrically spiking 226 Ra/2 M HNO3 (NIST, SRM4967,
U.S.A.) into known amounts of deionized water contained in
20 mL low-diffusion PE counting vials. OptiScint HiSafe was
then added to reach a total admixture volume of 20 mL. These
were stored for 3 weeks in a dark temperature-controlled area
to allow in-growth and equilibrium of the radioactive progenies.
The background solutions, used for calibration purposes, were
prepared with 10 mL of deionized water, acidified to match
the standard solutions, to which 10 mL of the scintillation
cocktail was added. In all cases, quenching was changed by
adding different amounts of CCl 4, ranging from 0 to 200 µL.
All reagents used in the experiments were of analytical grade
(Fisher Scientific).
RESULTS
When counting a 226 Ra aqueous solution with an immiscible
scintillant (such as OptiScint Hisafe), the R emitter 226 Ra decays
to the R emitter 222 Rn (T1/2 ) 3.8332 ± 0.0008 days). Radon is
highly soluble in oil-based scintillators and is selectively
extracted in the cocktail, suffering some decay while this
process takes place. Once solubilized in the organic phase,
222 Rn decays to the R emitter 218 Po (T1/2 ) 3.094 ± 0.006 min),
this mainly decays to the � emitter 214 Pb (T1/2 ) 26.8 ± 9 min),
which decays to the � emitter 214 Bi (T1/2 ) 19.9 ± 0.4 min),
which mainly decays to the R emitter 214 Po (T1/2 ) 162.3 ± 1.2
µs), and this one to 210 Pb (T1/2 ) 22.23 ± 0.12 years). Therefore,
in the scintillant, and after an appropriate equilibration time
(usually set to about 3 weeks), the R emitters 222 Rn, 218 Pb, and
214 Pb are in secular equilibrium (Figure 1). As the probability
of these R decays is in all cases close to one, the maximum
(22) Kaihola, L. J. Radioanal. Nucl. Chem. 2000, 243, 313–317.
(23) Villa, M.; Manjon, G.; Garcia-Leon, M. Nucl. Instrum. Methods Phys. Res.,
Sect. A 2003, 496 (2-3), 413–424.
(24) Sanchez-Cabeza, J. A.; Pujol, L. Health Phys. 1995, 68 (5), 674–82.

Figure 1. 226 Ra spectrum by liquid scintillation (only 6 × 10 -4 Bq).
Vertical dashed lines indicate the approximate counting windows.
efficiency reached when counting all R events should be 300%.
This is another advantage of R versus γ spectrometry, where final
counting efficiencies for 226 Ra, depending on the counting
configuration, normally do not usually exceed a few percent.
The magnitudes used for optimization were those related to
counting precision.
• Counting efficiency E: this is the ratio between the observed
number of counts and the 226 Ra decays. The maximum
observed efficiency is close to 300% (Figure 1) because with
this method we simultaneously count R particles from 222 Rn,
218 Pb, and 214 Pb in equilibrium.
• Minimum detectable activity (MDA): 25 we used the following
expression
2.71 + 4.65√Bt
MDA)
tVE
where B is the background count rate, t is the counting time,
and V is the volume of solution used (in the case of sediment
analysis, V was substituted by m, the mass of the aliquot
analyzed).
• Figure of merit: this is a magnitude commonly used to
compare methods, which emphasizes counting efficiency (and
therefore sample throughput). FM was calculated as FM )
E 2 /B.
In this section, we describe the results of the optimization
process for the relevant counting parameters.
Optimal Admixture Composition. Once the scintillation vial
and total volume were fixed, the optimal “scintillator-to-water” ratio
was optimized. We prepared and counted several composition
mixtures with tracer solutions and background solutions, by using
1, 5, 10, 15, and 19 mL of scintillant (Figure 2).
Not unexpectedly, the highest efficiency was observed for the
maximum volume of scintillant, as this maximizes radiation
interaction with the detector (the scintillant). However, as the
scintillant itself is an important source of background, this also
increases the MDA value. This behavior is well-captured with the
FM, which shows a maximum value for a 10:10 mL admixture.
Therefore, we used this proportion in all experiments.
Optimal PSA Parameter. The PSA circuit of Quantulus 1220
sends the signals to one of the two multichannel analyzers
Figure 2. Determination of the optimal admixture composition.
Figure 3. Change of background and 226 Ra counting efficiency with
PSA.
depending on the result of the PSA analysis. However, this method
is not error-free and shows interference as (i) R and � events can
be wrongly assigned and (ii) the efficiency of this method depends
on the energy of the incident radiation particles. 26 In order to
minimize the interference, tracer solutions of 226 Ra activity 5.34
Bq and background solutions, with a 10:10 mL admixture
composition, were counted by LSC during 12 h and with the
PSA parameter ranging from PSA ) 0 (no R-� discrimination)
to PSA ) 256 (maximum R-� discrimination) with a step
increment of 5 (Figure 3).
The background signals assigned to the R spectrum have a
varied nature and may include γ interactions with the scintillant,
� events wrongly assigned to the R spectrum, R emission from
impurities in the vial, and a variety of cosmic-ray originated signals.
Therefore, the background spectrum does not show prominent
peaks. The background count rate versus PSA plot (Figure 3)
shows an almost constant value until PSA ∼ 90, as all events are
recorded in the R spectrum, and then its value decays smoothly
to a value close to 0 cpm, when all background events are recorded
in the � spectrum.
On the other hand, the 226 Ra spectrum shows well-resolved
peaks within a quite narrow energy window (Figure 1), and
the effect of the R particle energy on the interference is clearly
shown in the efficiency versus PSA curve (Figure 3). Until PSA
∼ 110, the total efficiency exceeds the maximum value of 300%
because of the misclassification of background and � events in
the R spectrum. In the PSA range of 120-160, most background
(25) Currie, L. A. Anal. Chem. 1968, 40, 586–593. (26) Pujol, L.; Sanchez-Cabeza, J. A. Analyst 1997, 122, 383–385.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6849

Figure 4. Variation of MDA and FM with respect to PSA for 226 Ra
standard solutions.
and � events are correctly discriminated and a counting plateau,
of almost constant efficiency (ca. 265%), is observed.
In order to choose the optimum value within the efficiency
plateau, both MDA and FM were plotted (Figure 4). In broad
terms, MDA followed the efficiency curve, suggesting that, from
a limit of detection perspective and within the plateau, best results
would be achieved at higher PSA, as the background slowly
decreases with the PSA parameter and stabilizes for the PSA
ranging from 145 to 190. It is worth noticing that, as from PSA )
135 the background rate is very low, the MDA is correspondingly
low. Higher PSA values are discarded from the discussion as the
efficiency there tends rapidly to zero. On the other hand, the FM
shows a clear peak at PSA ) 145, and shows the optimum
combination of efficiency and background, taking advantage of a
high efficiency (and therefore sample throughput) and providing
the lowest possible MDA.
Counting Window. Due to the presence of many substances
in the digestate, which may be partially soluble in the scintillator,
the sample may show different levels of quenching, which also
affect the position of the peaks. In order to compensate for the
peak shift, we opted to define counting windows with a fixed width
but a variable position in the spectrum, manually set to comprise
the peaks of interest.
The first strategy was to define a wide counting window of
150 channels. This window allowed to include, irrespective of the
quenching level (quantified through the SQP(E) parameter), the
three R peaks (Figure 1). The main advantage of this strategy is
that counting efficiency is large, up to the maximum theoretical
level of 300%, and above 200% for the real samples commonly
analyzed. However, as the spectrum resolution does not allow us
to distinguish other R impurities present in the sample and some
� interferences, this method might lead to slight activity
overestimation.
In order to minimize the effect of background from impurities
and interferences, the second strategy was to define a 50 channel
wide counting window to only comprise the higher energy R peak
( 214 Po), which stands in an area of much smaller background
and interference, 15 and is easily identified. However, in this
case the maximum theoretical efficiency is only 100%.
Calibration. Quenching is the only parameter affecting the
counting configuration that is sample-dependent and cannot be
easily controlled. Therefore, several 226 Ra standard solutions were
quenched with CCl4 to values covering the expected SQP(E)
6850 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. Effect of quenching (SQP(E)) on R counting efficiency.
Figure 6. Time stability test for sample quenching (SQP(E)).
value in environmental samples (Figure 5). For both counting
windows, efficiency is close to zero for SQP(E) < 575. Therefore,
the calibration curves, obtained by regression analysis between
the counting efficiency (y) and the SQP(E) parameter (x), were
150 channels:
y ) (0.74 ( 0.06)x - (435 ( 51) and r ) 0.95
50 channels:
y ) (0.21 ( 0.02)x - (119 ( 19) and r ) 0.91
Usually, LSC techniques involve the measurement of a batch
of samples in cycles. As the typical activities expected in this type
of work are low (of the order of 10 mBq), long-term counting, of
the order of days for a single batch, is required. We successfully
checked the stability of the scintillator through the monitoring of
SQP(E) for a quenched tracer with an activity of 10 mBq during
4 days (Figure 6).
Detection Limits. Each counting configuration (including
counting window and quenching) shows different detection limits.
In Table 1, we show the counting properties for a standard solution
quenched to SQP(E) ) 850, which is an average value of sediment
samples analyzed in our laboratory. The combination of the
indirect detection of 226 Ra by 222 Rn emanation, the high resulting
efficiency, and the ultralow background of Quantulus 1220
result in an extremely low MDA (0.29 Bq kg -1 ) for samples as
low as 250 mg, a typical amount used when analyzing 210 Pb by
R spectrometry. 10 Although both MDA are very low, and
therefore any of the windows can be chosen, the larger
efficiency (and therefore precision of measurements for the
same counting time) favors the use of the largest window,
which is the recommended value in this work. From a

Table 1. Counting Properties and Detection Limits for a 226 Ra Standard Solution a
window (channels) efficiency (%) background (cpm) b
spectrometric point of view, a smaller window and higher R
energy minimize the possibility of interferences with other
emissions and should be favored when expected sample activity
or desired analytical throughput are not compromised.
DISCUSSION
Although this technique is well-suited to easily determine 226 Ra
in small samples from most environmental matrixes, the system
has been designed and optimized for small sediment samples,
subsequent to 210 Pb determination through 210 Po in equilibrium.
Proposed Method. We recommend operating in batches of
about 12 samples, where both a reference material and a blank
are processed and measured at the same time. The solution from
which the Po source was deposited (about 15 mL) is recovered
and, in order to destroy ascorbic acid and its degradation
compounds, heated under reflux with 5 mL of concentrated HNO3
for half an hour or until a clear solution is obtained. This
solution is then evaporated to incipient dryness, and small
amounts of 0.5 M HCl are added to dissolve the residue. This
operation is carried out two more times to ensure the elimination
of HNO3, and the total volume is then adjusted to 10 mL
in a 20 mL low-diffusion PE counting vial. Then 10 mL of
OptiScint HiSafe cocktail is carefully added by avoiding disturbing
the interface formed by the two immiscible liquids in the
vial (Figure 7).
The mixture is kept for 3 weeks in a dark temperaturecontrolled
area, in order to wait for radioactive equilibrium and
to minimize chemiluminescence, and is counted by liquid scintillation
by using the following conditions:
• Counting time: all samples of the batch are sequentially
counted for 1 h each, during a minimum of 30 cycles.
• Counting windows are approximately set to channels 750-900
and 850-900. However, best results are obtained by summing
all spectra and visually adjusting the windows manually to their
optimum value.
• Counting mode is set to discriminate R and other pulses,
with PSA ) 145.
Counting conditions and calibration are equipment-sensitive,
and therefore, the quenching calibration must be performed for
each instrument. We also recommend carrying out some tests
with 226 Ra tracer solutions to confirm that the chosen PSA value
is correct for each instrument.
The counting information needed to calculate the activity are
sample count rate, background count rate, and quenching parameter
(such as SQP(E) in Quantulus 1220). From the quenching
value the efficiency can be obtained (Figure 5), and the sample
activity finally calculated.
Low-Activity Test. We tested the linearity and the low-activity
response of the detector in the chosen configuration by measuring
a set of 226 Ra unquenched tracer solutions (0.5 M HCl) with
Ld (cpm) MDA (Bq kg -1 ) c
710-860 173 ± 12 0.0070 ± 0.0020 0.0076 0.29
810-860 54 ± 4 0.0024 ± 0.0012 0.26 0.59
a Activity ) 0.01 Bq. Quenched to SQP(E) ) 850. b Counting time of the background was 55 h. c MDA was calculated by assuming a typical
sediment mass of 250 mg.
Figure 7. Proposed procedure to measure 210 Po and 226 Ra in
sediment samples.
Table 2. Count Rates When Measuring 226 Ra
Unquenched Standard Solutions by LSC
sample activity (mBq) net count rate (cpm)
0.017
0.12

activities ranging over 2 orders of magnitude, all yielding satisfactory
results (Table 3).
The IAEA-384, a Fangataufa Lagoon sediment, is a reference
material designed for the determination of anthropogenic and
natural radionuclides in sediment. 27 This material is certified for
10 radionuclides, including 210Pb ( 210Po in equilibrium), and
information values are given for 10 radionuclides, including
226 226 -1 Ra. The Ra activity ranges from 2.0 to 2.9 Bq kg (95%
confidence interval) with a median value of 2.4 Bq kg-1 , which
is the low range of 226Ra activities in sediments. Dry sediment
aliquots of 500 mg were analyzed by using the proposed
method (N ) 15). Spectra were analyzed by using both a wide
and a narrow counting window, and results are shown in Figure
9. The median of both distributions is 2.2 Bq kg-1 , and means
were 2.3 Bq kg-1 (50 channels) and 2.5 Bq kg-1 (150 channels),
very close to the reference material reported value and within
its 95% confidence interval.
The IAEA-385, a certified reference material for radionuclides
in an Irish Sea sediment, is intended to be used, among other
purposes, for the development and validation of radiometric and
mass spectrometry analytical methods. 28 Its certified median value,
22.7 Bq kg-1 , is typical of most soils and sediments. The median
value obtained with the proposed method (N ) 15) is 23.5 Bq
kg-1 Figure 8. Calibration curve of unquenched
, which, although slightly higher than the upper limit of
the certified confidence interval, represents a deviation of only
7.3%.
226Ra. (27) Povinec, P. P.; Pham, M. K.; Sanchez-Cabeza, J. A.; Barci-Funel, G.;
Bojanawski, R.; Boshkova, T.; Burnett, W.; Carvalho, F.; Chapeyron, B.;
Cunha, I. L.; Dahlgaard, H.; Galabov, N.; Fifield, L.; Gaustaud, J.; Geering,
J.-J.; Gomez, I. F.; Green, N.; Hamilton, T.; Ibanez, F. L.; Ibn Majah, M.;
John, M.; Kanisch, G.; Kenna, T. C.; Kloster, M.; Korun, M.; Liong Wee
Kwong, L.; La Rosa, J.; Lee, S.-H.; Levy-Plaomo, I.; Malatova, M.; Maruo,
Y.; Michell, P.; Murciano, I. V.; Nelson, R.; Nouredine, A.; Oh, J.-S.; Oregioni,
B.; Petit, G.; Pettersson, H. B. L.; Reineking, A.; Smedley, P. A.; Suckow,
A.; Struijs, T.; Voors, P. I.; Yoshimiza, K.; Wyse, E. J. Radioanal. Nucl. Chem.
2007, 273, 383–393.
(28) Pham, M. K.; Sanchez-Cabeza, J. A.; Povinec, P. P.; Andor, K.; Arnold, D.;
Benmansour, M.; Bikit, I.; Carvalho, F. P.; Dimitrova, K.; Edrev, Z. H.;
Engeler, C.; Fouche, F. J.; Garcia-Orellana, J.; Gascó, C.; Gastaud, J.; Gudelis,
A.; Hancock, G.; Holm, E.; Legarda, F.; Ikäheimonen, T. K.; Ilchmann, C.;
Jenkinson, A. V.; Kanisch, G.; Kis-Benedek, G.; Kleinschmidt, R.; Koukouliou,
V.; Kuhar, B.; LaRosa, J.; Lee, S.-H.; LePetit, G.; Levy-Palomo, I.;
Liong Wee Kwong, L.; Llauradó, M.; Maringer, F. J.; Meyer, M.; Michalik,
B.; Michel, H.; Nies, H.; Nour, S.; Oh, J.-S.; Oregioni, B.; Palomares, J.;
Pantelic, G.; Pfitzner, J.; Pilviok, R.; Puskeiler, L.; Satake, H.; Schikowski,
J.; Vitorovic, G.; Woodhead, D.; Wyse, E. Appl. Radiat. Isot. 2008, 66, 1711–
1717.
6852 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
The IAEA-313, a stream sediment from Indonesia, 29 is a
reference material designed for the analysis of 226Ra, U, and Th
in geological samples. Its median 226Ra activity, 343 Bq kg-1 ,
is about 1 order of magnitude higher than typical sediment
activities. The median value (N ) 5) is within the reported 95%
confidence interval.
The observed deviations are typical of the precision of
environmental materials, and in overall, we concluded that the
proposed method is adequate to analyze 226Ra in sediment
samples.
Calibration Stability. With each sample batch, we have
analyzed the IAEA-384 reference material for quality control
purposes. Results of all analyses are shown in Figure 10. The mean
value, 2.3 Bq kg-1 , lies within the reference material 95%
confidence interval (Table 3), and the time stability of the activity
shows that the system’s calibration is stable for times spanning
at least during the measurement period, namely, one year.
Although this needs to be checked for each instrument in a
continuous manner, we do not anticipate the need of recalibration
of Quantulus 1220 for periods shorter than a year.
210 226 Pb and Ra in a Sediment Core. The method is
designed to provide fast and reliable results of 226Ra when 210Pb dating sediment cores through the analysis of 210Po in equilibrium.
The advantage of LSC over γ spectrometry is that
results for a large number of samples, typical of 210Pb dating
experiments, can be obtained within a month after sample
digestion. We provide here an example of the proposed
strategy.
The DYFAMED station (Ligurian Sea, 43°25′ N; 7°52′ E) has
been the subject of multidisciplinary research since 1987, including
the study of atmospheric deposition of metals and their association
with marine particles in the water column. Martín et al. 30 studied
the concentrations and fluxes of trace metals in a sediment core
collected from 2300 m water depth at the sea floor beneath the
DYFAMED site. In Figure 11, we show the 210Pb and 226 Figure 9. Box-and-whisker plots of
Ra
profiles obtained by R spectrometry and LSC, showing that
226Ra activities in the reference
material IAEA-384 (N ) 15). The two dots in the 50 channels counting
window were statistically identified as outliers.
(29) Strachnov, V.; Valkovic, V.; Zeisler, R.; Dekner, R. Report on the Worldwide
Intercomparison Exercise IAEA-314: 226 Ra, Th and U in Stream Sediment;
International Atomic Energy Agency (IAEA/AL/038): Vienna, Austria, 1991.
(30) Martin, J.; Sanchez-Cabeza, J. A.; Eriksson, M.; Miquel, J. C. Mar. Pollut.
Bull. 2009, 59, 146–153.

Table 3. Analysis of 226 Ra in Sediment Reference Materials (RM)
reference material no. of analyses method median (Bq kg-1 ) RM median (Bq kg-1 ) RM 95% confidence interval deviation
IAEA-384 15 2.2 2.4 2.0–2.9 -8.3
IAEA-385 15 23.5 21.9 21.6–22.4 +7.3
IAEA-313 5 372 343 307–379 +8.5
Figure 10. Time stability test for 226 Ra concentrations in the IAEA-
384 reference material.
226Ra reaches equilibrium in the core bottom sections. An
important finding was that, although a constant 226Ra is
commonly estimated from core bottom sections, where 210Pb and 226Ra are assumed to be in equilibrium, this is not the case
in the DYFAMED core, where 226Ra ranged from 22.6 to 41.7
Bq kg-1 . This reinforces the need for paired 210Pb/ 226 Figure 11.
Ra
210Pb and 226Ra profiles in a sediment core from the
DYFAMED site (Ligurian Sea).
measurements and methodologies to accomplish it accurately,
as does the technique described in this paper. This may have
important implications on the dating results for this particular
sediment core, which will be discussed elsewhere.
CONCLUSIONS
The accurate 210 Pb dating of sediment cores requires the
determination of 226 Ra in all sections. Although this can be done
by γ spectrometry, the large sample size required (>5 g dry
weight in well detectors, >20 g in coaxial detectors) might be
impossible to obtain when sediments are used for the analysis
of multiple magnitudes (such as grain size, elemental composition,
trace metals, organic substances, biomarkers, ...) and
counting times (typically >2 days per sample) might be a
limiting factor for many laboratories. The proposed strategy
in this work is the determination of 210 Pb through 210 Po in
equilibrium by R spectrometry 10 followed by 226 Ra by LSC
without any further radiochemical processing.
We optimized the counting parameters for an ultralow background
scintillation system with R-� separation capabilities
(Quantulus 1220, Wallac) and propose the use of a PSA parameter
of 145. The system was calibrated with a series of quenched 226 Ra
standard solutions. For a typical sediment sample quenching
(SQP(E) ) 850), the efficiency was (173 ± 12)% in the wide
energy counting window (150 channels) due to the simultaneous
counting of three radionuclides in equilibrium ( 222 Rn,
218 Po, and 214 Po). When analyzing 250 mg dw sediment samples,
the MDA was as low as 0.29 Bq kg -1 , which is about 2 orders
of magnitude lower than typical sediment concentrations,
showing the usefulness of the technique for many other
environmental applications. The method was validated with
three reference materials spanning 3 orders of magnitude of
concentration. The proposed method can greatly improve the
reliability of 210 Pb chronologies of sediment cores, and can also
be tested for 226 Ra/ 210 Pb dating of carbonates such as corals
and speleothems.
ACKNOWLEDGMENT
The authors thank Mr. Jacobo Martín (IAEA) for providing
samples of a sediment core from the DYFAMED site and helpful
comments. The IAEA is grateful for the support provided to its
Marine Environment Laboratories by the Government of the
Principality of Monaco.
Received for review March 31, 2010. Accepted June 30,
2010.
AC1008332
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6853

Anal. Chem. 2010, 82, 6854–6861
Improved Sensitivity of DNA Microarrays Using
Photonic Crystal Enhanced Fluorescence
Patrick C. Mathias, †,‡ Sarah I. Jones, § Hsin-Yu Wu, ‡,| Fuchyi Yang, ‡,| Nikhil Ganesh, ‡,⊥
Delkin O. Gonzalez, § German Bollero, § Lila O. Vodkin, § and Brian T. Cunningham* ,†,‡,|
Department of Bioengineering, 1304 W. Springfield Ave., Micro and Nanotechnology Laboratory, 208 N. Wright St.,
Department of Crop Sciences, 1102 S. Goodwin Ave., Department of Electrical and Computer Engineering,
1406 W. Green St., and Department of Materials Science and Engineering, 1304 W. Green St., University of Illinois at
Urbana-Champaign, Urbana, Illinois 61801
DNA microarrays are used to profile changes in gene
expression between samples in a high-throughput manner,
but measurements of genes with low expression
levels can be problematic with standard microarray substrates.
In this work, we expand the detection capabilities
of a standard microarray experiment using a photonic
crystal (PC) surface that enhances fluorescence observed
from microarray spots. This PC is inexpensively and
uniformly fabricated using a nanoreplica molding technique,
with very little variation in its optical properties
within- and between-devices. By using standard protocols
to process glass microarray substrates in parallel with
PCs, we evaluated the impact of this substrate on a onecolor
microarray experiment comparing gene expression
in two developmental stages of Glycine max. The PCs
enhanced the signal-to-noise ratio observed from microarray
spots by 1 order of magnitude, significantly increasing
the number of genes detected above substrate fluorescence
noise. PC substrates more than double the number
of genes classified as differentially expressed, detecting
changes in expression even for low expression genes. This
approach increases the dynamic range of a surface-bound
fluorescence-based assay to reliably quantify small quantities
of DNA that would be impossible with standard
substrates.
The DNA microarray is a valuable tool for high-throughput
quantification of gene expression, allowing a large number of
candidate genes to be examined for differential expression
simultaneously without extensive prior knowledge of gene
functions. Eukaryotic gene expression is typically characterized
by a large number of genes expressed at very low levels and a
decreasing number of genes expressed at high levels. 1,2 Often
the noise present in DNA microarray experiments is high
enough that only a small fraction of genes assayed can be
* To whom correspondence should be addressed. Phone: 217-265-6291.
E-mail: bcunning@illinois.edu.
† Department of Bioengineering.
‡ Micro and Nanotechnology Laboratory.
§ Department of Crop Sciences.
| Department of Electrical and Computer Engineering.
⊥ Department of Materials Science and Engineering.
(1) Kuznetsov, V. A.; Knott, G. D.; Bonner, R. F. Genetics 2002, 161, 1321–
1332.
6854 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
detected by fluorescence measurements. While sample variation
and nonspecific binding play an important role in this experimental
noise, microarray substrate fluorescence can contribute
to noise as well. This may explain the poor performance of
microarrays in detecting genes with low expression levels
relative to other methods. 3,4 To overcome the difficulties of
quantifying the abundance of low expression genes, substrates
that enhance the fluorescence observed from microarray spots
can be used to achieve better assay performance.
Researchers have utilized various nanostructured metal substratres
to achieve increased intensity from common microarray
dyes, with signal enhancements ranging from 1 to 2 orders of
magnitude. These methods include the growth of metal island
films on a substrate 5,6 or the deposition of nanoparticles fabricated
by spray pyrolysis 7 to produce optical resonances to which
microarray dye excitation and/or emission can couple. However,
the practical impact of these enhancement methods is unclear,
because previous work has not characterized the signal-to-noise
ratio (SNR) enhancement for a large number of spots over many
substrates in a conventional gene expression microarray experiment.
One potential obstacle in achieving this result is the need
for an inexpensive nanoscale fabrication method achieving highthroughput
and good uniformity over large areas; previous reports
of microarray dye enhancement have not utilized photolithography
to generate consistent patterns and thus are subject to a random
arrangement of structures. Another potential obstacle is integration
of these substrates with the existing commercial equipment
used to fabricate and scan DNA microarrays, as previous literature
has not explicitly demonstrated nanostructures over areas as large
as conventional microscope slides. We addressed these issues by
designing a nanostructured photonic crystal (PC) substrate
capable of enhancing Cyanine-5 (Cy-5) fluorescence in a com-
(2) Ueda, H. R.; Hayashi, S.; Matsuyama, S.; Yomo, T.; Hashimoto, S.; Kay,
S. A.; Hogenesch, J. B.; Iino, M. Proc. Natl. Acad. Sci. U.S.A. 2004, 101,
3765–3769.
(3) Kuo, W. P.; Liu, F.; Trimarchi, J.; Punzo, C.; Lombardi, M.; et al. Nat.
Biotechnol. 2006, 24, 832–840.
(4) t Hoen, P. A. C.; Ariyurek, Y.; Thygesen, H. H.; Vreugdenhil, E.; Vossen,
R. H. A. M.; de Menezes, R. X.; Boer, J. M.; van Ommen, G.-J. B.; den
Dunnen, J. T. Nucleic Acids Res. 2008, 36, e141.
(5) Sabanyagam, C. R.; Lakowicz, J. R. Nucleic Acids Res. 2007, 35, e13.
(6) Moal, E. L.; Leveque-Fort, S.; Potier, M.-C.; Fort, E. Nanotechnology 2009,
20, 225502.
(7) Guo, S.-H.; Tsai, S.-J.; Kan, H.-C.; Tsai, D.-H.; Zachariah, M. R.; Phaneuf,
R. J. Adv. Mater. 2008, 20, 1424–1428.
10.1021/ac100841d © 2010 American Chemical Society
Published on Web 07/16/2010

mercial microarray scanner and fabricating it by nanoreplica
molding to fit standard microscope slides.
The PC substrates are composed of a subwavelength, periodic
SiO2 surface structure coated with a high refractive index
dielectric layer of TiO2, creating a periodic modulation in
refractive index along the device surface. The periodic modulation
gives rise to optical resonances 8 that can be used to achieve
fluorescence enhancement. These resonances can be used to
generate strong optical near-fields at the device surface when
spectrally aligned to the excitation wavelength 9 and to spatially
alter the fluorescence emission pattern to maximize light
collection. 10 The overall effect of these phenomena is to amplify
the fluorescent signal from molecules within approximately 100
nm of the PC surface. While the first demonstrations of PC
enhanced fluorescence for microarrays required expensive
lithographic procedures for each device and yielded modest
enhancement factors of approximately 6× signal enhancement
in a commercial scanner, 11,12 inexpensive and uniform fabrication
over large areas in a nanoreplica molding process currently used
to make commercial label-free biosensors 13 has since been
employed to make these structures.
Recently, PCs have been engineered by our group to enhance
the common microarray dye Cyanine-5 (Cy-5) by more than 1
order of magnitude when scanned in a commercial microarray
scanner. 14 This work details the application of this PC design to
a microarray experiment assessing differential expression between
Glycine max cotyledons and trifoliates, which represent tissues
from two distinct developmental stages in the soybean plant.
Multiple PCs exhibiting highly uniform optical characteristics over
the area of entire microscope slides were fabricated by nanoreplica
molding. These PCs were processed using published protocols
in parallel with commercial microarray substrates. By enhancing
fluorescence, a larger number of genes can be detected above
noise on the PC compared with glass substrates. This effect more
than doubles the number of genes identified as differentially
expressed between the trifoliate and cotyledon tissues, demonstrating
that enhanced fluorescence offers practical benefit to a
DNA microarray experiment.
MATERIALS AND METHODS
Photonic Crystal Fabrication and Characterization. The
PCs used for this work were designed by simulation software
employing Rigorous Coupled-Wave Analysis (DiffractMOD, RSoft
Design Group, Inc.) to align optical resonances to the excitation
(632.8 nm) and emission wavelengths (670-710 nm) of Cy-5. As
described in previous work, 14 the period of the structure, grating
depth, and thickness of the high refractive index TiO2 layer were
(8) Rosenblatt, D.; Sharon, A.; Friesem, A. A. IEEE J. Quantum Electron. 1997,
33, 2038–2059.
(9) Ganesh, N.; Mathias, P. C.; Zhang, W.; Cunningham, B. T. J. Appl. Phys.
2008, 103, 083104.
(10) Ganesh, N.; Block, I. D.; Mathias, P. C.; Zhang, W.; Chow, E.; Malyarchuk,
V.; Cunningham, B. T. Opt. Express 2008, 16, 21626–21640.
(11) Neuschafer, D.; Budach, W.; Wanke, C.; Chibout, S.-D. Biosensors Bioelectron.
2003, 18, 489–497.
(12) Budach, W.; Neuschafer, D.; Wanke, C.; Chibout, S.-D. Anal. Chem. 2003,
75, 2571–2577.
(13) Cunningham, B.; Lin, B.; Qiu, J.; Li, P.; Pepper, J.; Hugh, B. Sensors Actuators
B 2002, 85, 219–226.
(14) Mathias, P. C.; Wu, H.-Y.; Cunningham, B. T. Appl. Phys. Lett. 2009, 95,
021111.
manipulated given the known refractive indices of PC materials
to achieve optical resonances overlapping both excitation and
emission wavelength ranges. These PCs were then fabricated
by nanoreplica molding 13 to create six distinct devices for
microarray experiments. The silicon “master” for the molding
process consisted of a 360 nm period one-dimensional grating
structure with a 60 nm grating depth and 50% duty cycle,
patterned on an 8 in. silicon wafer by deep-UV lithography.
After immersion in 2% dichlorodimethylsilane (PlusOne Repel-
Silane ES, GE Healthcare) to promote clean release, a UVcurable
liquid polymer (Gelest, Inc.) with index of refraction
npolymer ) 1.46 was dispensed on a sheet of polyethylene
terephthalate (PET), and the grating pattern was transferred
with a roller. After curing the polymer under a high-intensity
ultraviolet lamp (Xenon) for 90 s through the transparent PET
sheet, 300 nm of SiO2 (nSiO2 ) 1.46) and 160 nm of TiO2 (nTiO2
) 2.35) were added to the grating structure by sputter coating.
The completed PCs were cut into 1 in. × 3 in. sections and
adhered to glass microscope slides with an optically clear
adhesive (3M).
The PCs were initially profiled for surface characteristics by
atomic force microscopy (Dimension 3000, Digital Instruments)
to compare the actual dimensions with the PC design. The PCs
were then optically characterized by passing broadband light in
the visible spectrum from a tungsten halogen lamp (Ocean Optics)
through a polarizer and collimator before transmission through
the device, which was aligned on a rotational stage to be
perpendicular to the direction of incident light. 15 PCs were
illuminated with both transverse magnetic (incident electric field
perpendicular to grating lines) and transverse electric (incident
electric field parallel to grating lines) polarizations to characterize
the two distinct resonances. Light transmitted through the PCs
was collected by an optical fiber and measured using a UV-visible
light spectrometer (Ocean Optics).
Slide Preparation. PCs were cleaned by O2 plasma treatment
and incubated with 3-glycidoxypropyltrimethoxysilane at 185
mTorr overnight for surface functionalization. The control slides
for microarray experiments were commercially available silanized
glass slides (Corning GAPS II). Oligonucleotides were
printed on the slides using a Genetix QArray 2 robot. A set of
previously annotated 192 70-mer oligonucleotides derived from
publicly available soybean EST and mRNA sequences was
spotted on the slides with 40 repeats per sequence per slide.
These 192 oligonucleotides are a subset of a larger 19 200 set
of oligonucleotides detailed in previous work. 16 Each of six
spotted PCs was matched with a spotted glass control slide to
receive identical experimental treatments.
Microarray Sample Preparation and Hybridization. Sample
RNA was extracted using previously published protocols. 16
Cotyledon RNA was extracted from freeze-dried soybean cultivar
Williams seeds with fresh weight between 100-200 mg. Cotyledon
RNA was purified using a Qiagen RNeasy kit. Trifoliate RNA was
extracted from freeze-dried rolled-up trifoliates of soybean cultivar
Williams from leaves between 0.5 and 1.5 in. in length. Sample
RNA was labeled with Cy-5 by reverse transcription. Three
replicate slides were hybridized for each of the two tissue samples,
(15) Yang, F.; Yen, G.; Cunningham, B. T. Opt. Express 2010, 18, 11846–11858.
(16) Gonzalez, D. O.; Vodkin, L. O. BMC Genomics 2007, 8, 468.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6855

with an identical number of glass slides processed in parallel with
the PCs. Slides were blocked prior to hybridization with bovine
serum albumin to prevent nonspecific binding, hybridized at 42
°C overnight, and washed as described previously. 16
Data Collection. Slides were scanned with a Tecan LS
Reloaded scanner with a transverse magnetic polarized laser (λ
) 632.8 nm) at normal incidence and an emission filter spanning
670-710 nm. All slides were scanned at 10 µm resolution. Initial
scans were at equal photomultiplier tube (PMT) gain to compare
fluorescence intensities at equal measurement conditions, but
afterward, gains were adjusted for each slide such that spots with
the largest fluorescence intensities did not saturate the scanner
PMT. Fluorescence images were analyzed using GenePix Pro 6.0
to compute spot and local background intensities as well as their
standard deviations for each spot.
Signal-to-Noise Ratio Analysis. Fluorescence data was
analyzed to calculate signal-to-noise ratios (SNRs) for each spot
at identical scan conditions, where the SNR is the local backgroundsubtracted
spot intensity divided by the standard deviation of the
local background pixels. For each of the six PCs and six glass
slides, within-slide repeats (40 per probe sequence) were averaged
to generate a SNR value for each of the 192 sequences probed in
the experiment. A SNR enhancement factor was calculated by
dividing the PC SNR value for each gene by the glass SNR value
for the same gene. The proportion of detected genes was
determined by calculating the percentage of genes on each slide
with a SNR > 3.
Differential Expression Analysis. Expression data was
analyzed using the Linear Models for Microarray Data (LIMMA)
package in R. Data was background corrected using the normalized
plus exponential convolution model with an offset of one. 17
Quantile normalization was used to normalize between arrays. Logtransformed
Cyanine-5 intensities were condensed by averaging
within-slide repeats and then fit to a linear model. Empirical Bayes
moderated t-statistics were calculated to assess differential expression
between the trifoliate and cotyledon samples, with p-values
adjusted by the Benjamini and Hochberg method to control the
false discovery rate. 18 The significance level for testing was set
to R)0.05.
High-Throughput RNA Sequencing and Analysis. The
mRNAs were also subjected to high-throughput sequencing (RNAseq)
performed at the Keck Center of the University of Illinois
using the Illumina Genome Analyzer II resulting in 10-18 million
total reads for leaf trifoliate and the immature cotyledon mRNA
samples, respectively. After processing, the RNA-seq reads are
all 70 bases in length. The sequence reads were aligned using
Bowtie 19 to the approximately 78 700 predicted Glyma gene
models available at Phytozome 5.0 (http://www.phytozome.net)
for the recently sequenced soybean genome. 20 The Bowtie
parameters allowed three mismatches to each Glyma model and
allowed up to 25 gene model matches to detect repetitive gene
models. Normalization of RNA-seq data as RPKM (reads per
(17) Ritchie, M. E.; Silver, J.; Oshlack, A.; Holmes, M.; Diyagama, D.; Holloway,
A.; Smyth, G. K. Bioinformatics 2007, 23, 2700–2707.
(18) Benjamini, Y.; Hochberg, Y. J. R. Stat. Soc. Series B (Methodological) 1995,
57, 289–300.
(19) Langmead, B.; Trapnell, C.; Pop, M.; Salzberg, S. L. Genome Biol. 2009,
10, R25.
(20) Schmutz, J.; Cannon, S. B.; Schlueter, J.; Ma, J.; Mitros, T.; et al. Nature
2010, 463, 178–183.
6856 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 1. (a) Schematic of PC design dimensions. (b) Atomic force
micrograph of completed PC structure (after TiO2 deposition), with a
measured period of 366 nm and height of 50 nm.
kilobase of gene model per million mapped reads) was calculated
as shown in previous work. 21 Assignment of the 70-mer oligos to
Glyma gene models was also performed using Bowtie with the
same parameters. Most of the 70-mer oligos on the arrays matched
only one or a few Glyma models.
RESULTS
Characterization of Photonic Crystal Substrates. A schematic
of the nanoreplica-molded one-dimensional PC design
optimized in previous work 14 to enhance fluorescence from Cy-5
(with period Λ ) 360 nm and grating step height h ) 60 nm)
appears in Figure 1a. A representative atomic force micrograph
detailing the surface structure appears in Figure 1b, with a
measured period of Λ ) 366 nm and a measured height of h )
50 nm showing good agreement with the expected dimensions.
Enhancement of Cy-5 was achieved by aligning a narrow PC
resonance (full-width at half-maximum, fwhm ) 4 nm) with the
laser excitation wavelength of 632.8 nm and engineering a second
broad resonance (fwhm ) 20 nm) to overlap the emission filter
wavelengths of 670-710 nm (Figure 2). A more narrow excitation
resonance increases the magnitude of the enhanced optical near
(21) Mortzazvi, A.; Williams, B. A.; McCue, K.; Schaeffer, L.; Wold, B. Nat.
Methods 2008, 5, 621–628.

Figure 2. Optical transmission measurements from all six PCs (each
represented by a colored solid line) used in this study, obtained by
illuminating the devices with polarized, collimated white light. Resonances
with narrow spectral features are excited when the PCs are
illuminated with transverse magnetic polarized light and overlap the
excitation wavelength of 632.8 nm (dotted line). Resonances with
broader spectral features are excited when the PCs are illuminated
with transverse electric polarized light and overlap the emission filter
wavelengths of 670-710 nm (dotted box).
fields at the device surface, 22 while the broad extraction resonance
maximizes the spectrum of emitted light redirected toward the
detection optics. 10 Good spectral uniformity of the narrow excitation
resonance over large areas of the PC is required to ensure
precise overlap of the resonance with a narrowband excitation
source regardless of the location of a microarray spot on the PC.
The nanoreplica molding fabrication process achieves this uniformity
throughout individual microscope slide-sized PCs, with a
maximum observed resonance wavelength standard deviation of
σwithin-PC ) 0.239 nm over 6 distinct PCs. Figure 2 illustrates
that excellent between-device uniformity is achieved as well; the
standard deviation in mean resonance wavelength between the 6
PCs used in this work is σbetween-PC ) 0.691 nm, making both
the within-device variation and between-device variation in
resonance wavelength significantly smaller than the spectral
width of the resonance.
Signal-to-Noise Ratio Analysis. After pairing each PC with
a control glass slide and printing a set of 192 oligonucleotides
and negative controls on the slides, 16 a one-color DNA microarray
protocol was simultaneously run on each glass-PC pair. Three
pairs of slides were hybridized with Cy-5 labeled cotyledon RNA
(extracted from seeds), and three pairs were hybridized with Cy-5
labeled trifoliate RNA (extracted from leaves) from Glycine max
cultivar Williams, representing two distinct tissues and stages of
development. After averaging the 40 duplicates of the 192 genes
on each slide, a ratio of averaged PC SNR to averaged glass SNR
was generated for each PC-glass slide pair, resulting in a median
SNR enhancement across all slides of 10.6×. The effect of this
SNR enhancement on the raw fluorescence data is observed in
Figure 3, which shows line profiles of identical probes for a
microarray grid on both a PC and its control. Considerable
enhancement is observed for spots of varying expression levels
(Figure 3c, d), with low expression genes being much easier to
discriminate from noise on the PC.
To explore the relevance of this SNR enhancement on gene
expression measurements, additional fluorescence scans of PCs
and glass slides were performed after gain adjustment to utilize
the full dynamic range of the scanner photomultiplier tube (PMT).
Because the PCs demonstrate fluorescence enhancement, they
were scanned at lower PMT gain values, resulting in lower noise
levels. A detection threshold of SNR ) 3 was applied to determine
the proportion of spots that could be detected on each slide.
Across all cotyledon samples, 25.0% of spots could be detected
on the glass slides, while 46.3% of the spots were detected on the
Figure 3. Fluorescence images at identical gains of a single identical microarray grid on glass (a) and PC (b), with brightness and
contrast adjustment to make the maximum number of spots visible on both images. For comparison of spot intensities, line profiles of
identical locations on the grid for glass and PC are illustrated on the same plots. Lower expression genes appear in (c) and higher
expression genes appear in (d).
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6857

Figure 4. Logarithmic plots of duplicate-averaged SNR values for the 192 probed genes on selected glass-PC slide pairs for each tissue.
Genes are organized in decreasing expression order for each chip, and an SNR detection threshold of 3 appears as the cutoff line in each
graph. SNR expression profiles for cotyledon RNA appear in (a) and (c) for a glass slide and its paired PC, respectively. Trifoliate RNA expression
profiles are plotted in (b) and (d) for a glass slide and its paired PC, respectively. Negative control spots on all slides appeared below the
detection threshold.
PCs. A more dramatic increase in the number of detected spots
was observed across all trifoliate samples, with 14.7% of spots on
the glass slides being detected as compared to 49.0% of the spots
on the PC. The number of genes that could be detected above
noise thus almost doubled for the cotyledon sample and more
than tripled for the trifoliate sample, as illustrated in plots of SNR
for each gene for representative slides in Figure 4. SNR values
(averaged across duplicate spots) are graphed for each gene in
decreasing expression order for a single slide pair hybridized to
a cotyledon sample (Figure 4a, c) and a single slide pair hybridized
to a trifoliate sample (Figure 4b, d). As expected, negative control
spots on both the PCs and the glass slides had SNRs below the
detection threshold.
Differential Expression Analysis. Background-corrected, 17
normalized, log-transformed spot intensities were fit to a linear
model using the LIMMA package in R, and empirical Bayes
moderated t-statistics were calculated to assess differential expression
in the trifoliate sample relative to the cotyledon sample. The
analysis was carried out for glass slides and PCs separately to
allow for comparison between the two substrates. Volcano plots
(simultaneously illustrating the fold change and the adjusted
p-value for each gene across glass slides or PCs) appear in Figure
5. Ideally, the plot should be a v-shape, since the p-value should
decrease as the fold-change increases. However, this relationship
is distorted by variation, since variation in fold-change measurements
smears the curve horizontally and variation between
samples leads to lower p-values (smearing the curve vertically).
Figure 5a and b plot all 7680 spots without averaging of withinslide
repeat spots in order to illustrate more clearly the effect of
the PC on the experimental data. Red circles denote genes that
have an adjusted p-value of less than 0.05 and a greater than 2-fold
change. 1431 spots on the PCs fulfill these criteria, compared with
865 spots on the glass slides. The PC data more tightly conforms
to the expected v-shape, suggesting there is less variation between
within-slide repeats in fold-change values and p-values, particularly
6858 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
for spots with low fold-change value, compared to the glass slide.
This lowered variation is expected to allow for discrimination of
smaller changes in expression during statistical testing.
A similar analysis was carried out after averaging within-slide
repeats, reducing the data set into 192 genes, with Figure 5c and
d illustrating volcano plots for the averaged data set. On the glass
slide, 27 genes fulfilled the criteria of statistically significant
changes at an adjusted p-value less than 0.05 and a greater than
2-fold change, while on the PC, 68 genes fulfilled these criteria.
Importantly, all 27 of the genes fulfilling these criteria on the glass
slide were also identified as differentially expressed on the PCs
(Table S-1 in the Supporting Information). The average measurement
for these 27 genes are similar on both substrates, with
measurements of 2350 counts (of fluorescence intensity) on the
glass slides and 2880 counts on the PCs. However, an additional
41 genes were identified as differentially expressed on the PCs,
suggesting that statistically significant changes in expression that
were overwhelmed by noise on the glass slide could be identified
on the PCs. Genes with a greater than 2-fold change as measured
on the PCs (p < 0.05) but not the glass slides (Table S-2 in the
Supporting Information) had an average expression level of 180
counts on the PCs, demonstrating lower expression levels than
those genes classified as differentially expressed on the glass
slides.
Validation of the microarray data by an independent method
was obtained by high-throughput sequencing with the Illumina
platform yielding 10-18 million total reads for leaf trifoliate and
the immature cotyledon mRNA samples, respectively. High quality
reads of 70 bases in length were mapped to 78 700 soybean gene
models to obtain a quantitative view gene expression. Fold change
values for 5 randomly selected genes from the 41 genes found to
be differentially expressed on the PC but not the glass substrates
(22) Mathias, P. C.; Ganesh, N.; Zhang, W.; Cunningham, B. T. J. Appl. Phys.
2008, 103, 094320.

Figure 5. Volcano plots detailing the relationship between fold-change and inverse p-value to assess differential expression between the
trifoliate and cotyledon samples, with positive fold changes indicating increased trifoliate expression and negative fold changes indicating increased
cotyledon expression. Green vertical lines represent the 2-fold change cutoff, and the yellow horizontal line denotes a p-value cutoff of 0.05.
Genes meeting both thresholds are indicated by red spots. Unaveraged data representing all 7680 spots across all experimental slides (3
replicates per tissue) appear in (a) for the glass slides and (b) for the PCs, with 865 spots differentially expressed on glass slides and 1431
spots on the PCs. Averaging within-slide repeats condensed the data to 192 distinct genes and controls, which appear in (c) for the glass slides
and (d) for the PCs. Of the 192 genes probed, 27 were classified as differentially expressed on the glass slides, while 68 met this classification
on the PCs.
(Table S-2 in the Supporting Information) are plotted in Figure 6,
with a comparison of glass microarray, PC microarray, and
sequencing data. The direction of differential expression of the
genes represented by a majority of probes on the arrays was
confirmed by the transcriptome sequencing data, although the
absolute fold changes vary due to the fundamental differences
between techniques. For example, Table S-3 in the Supporting
Information shows agreement for 22 of 26 genes (one sequence
was highly repetitive in the genome and thus could not be
quantified) found to be differentially expressed on both PC and
glass microarrays and Table S-4 in the Supporting Information
shows agreement for 39 of 41 genes found to be differentially
expressed only on PC microarrays. It is also apparent that the
PC arrays detect genes with lower average RPKM values (average
RPKM of 65.0 for both samples in Table S-4 in the Supporting
Information) than detected reliably on the glass slides alone
(average RPKM of 2160 for both samples in Table S-3 in the
Supporting Information).
As shown in Table S-1 in the Supporting Information (which
lists genes with significant differences in expression as detected
on both glass and PCs), a number of the genes found to be
overexpressed in the seed cotyledons (with negative fold changes)
include those that encode well-known soybean storage proteins
(i.e., glycinin, lectin, Kunitz trypsin inhibitor, and the Bowman-
Birk proteinase inhibitor) whose mRNA transcripts are abundant
during seed embryogenesis. 23,24 For example, RNA-seq transcriptomics
data confirms some of these with very large RPKM values
of up to 17 340 in the seed and no detectable transcripts in the
leaves (Table S-3 in the Supporting Information). On the other
hand, those genes encoding photosynthetic proteins as the
Rubisco small chain precursor and chlorophyll a/b binding protein
are overexpressed in the trifoliate leaves, as expected. As shown
in Table S-2 in the Supporting Information, the additional genes
detected as differentially expressed with significant p-values on
the PCs represent various enzymes and transcription factors found
to be expressed at lower levels by RNA transcriptome sequencing
(Table S-4 in the Supporting Information), demonstrating the
usefulness of the PCs to detect low expression transcripts.
DISCUSSION
By engineering PC resonances for compatibility with a commercial
laser scanner, the benefits of enhanced fluorescence can
be applied to a standard microarray experiment with no changes
(23) Thibaud-Nissen, F.; Shealy, R. T.; Khanna, A.; Vodkin, L. O. Plant Physiol.
2003, 132, 118–136.
(24) Jones, S. I.; Gonzalez, D. O.; Vodkin, L. O. BMC Genomics 2010, 11, 136.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6859

Figure 6. Logarithmic plot of fold change comparisons between
glass microarray, PC microarray, and sequencing results for five
genes randomly selected from the list of genes found to be differentially
expressed on the PCs but not on the glass slides. Fold change
values for glass and PC microarrays were calculated by determining
the ratio between average microarray expression level for trifoliate
samples to average microarray expression level for cotyledon
samples. A similar ratio was calculated using number of reads for
sequencing data. The PC microarray results show similar directions
and magnitudes of change compared to sequencing data for all 5
genes, while glass microarray data for CHP089 and CHP004 does
not agree with the sequencing data.
to the experimental protocol. While the initial photolithography
process needed to fabricate the silicon mold of the grating has
high costs, a single round of photolithography on silicon can be
translated into thousands of devices that are fabricated uniformly
over large areas. By fabricating the mold on an 8 in. wafer, there
is a large degree of flexibility in fitting PCs to preferred labware
formats such as microscope slides and microtiter plates. Because
PCs were cut to fit standard microscope slides, they could be
processed with existing protocols and scanned with commercially
available equipment, allowing for convenient adoption of PC
substrates in a standardized experiment. Not only does the
nanoreplica molding process provide a convenient form factor for
the substrates, it also enables the excellent level of optical
uniformity required to ensure that every spot on the microarray
experiences the same level of enhancement. This is key to ensure
that the data obtained from PCs does not have a higher level of
variation than the data obtained from glass slides.
The signal enhancement factor primarily used in previous work
in this field is defined as spot intensity subtracted by the local
background observed on the enhancement substrate divided by
the same value observed on the glass slide or control substrate.
The signal enhancement factor observed from Cy-5 spots with
high expression genes in this microarray experiment was approximately
60×, which is identical to the enhancement demonstrated
in previous work with this substrate. 14 Thus, the PC signal
enhancement compares favorably to experiments with metal island
films that have yielded signal enhancement factors of 10-40×. 5,6
However, the signal enhancement factor is not an ideal measurement
to assess the practical utility of the substrate. This work
has focused on SNR enhancement rather than signal enhancement
because microarray data analysis programs use SNR values to
classify spots as detected or not detected. It is possible to achieve
6860 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
good signal enhancement without achieving similar SNR enhancement
if a substrate enhances fluorescence but has a large noise
value thus voiding any advantages of fluorescence enhancement.
Without knowledge of a substrate’s impact on the SNR observed
from spots, it is difficult to ascertain whether a substrate will
benefit a target assay. The PC not only attains a large signal
enhancement but it also achieves an SNR enhancement of
approximately 10× (measured over all spots in the experiment),
suggesting that the array can detect hybridization at concentrations
10× lower than can be detected on glass substrates.
The noise in a DNA microarray experiment arises primarily
from the following sources: sample variation, nonspecific binding,
instrumentation, and substrate fluorescence. Variation in the
amount of nucleic acid sample captured is accounted for by
hybridizing multiple arrays and figures prominently into tests of
significance for differential expression experiments. Nonspecific
binding is controlled largely by blocking and hybridization
conditions and is assessed by evaluating negative control spots.
The noise observed from instrumentation can be characterized
by measurements of dark noise, but in this experiment, this
represents only

data as well. The 41 genes in Tables S-4 and S-2 in the Supporting
Information corresponding to genes detected as differentially
expressed on the PC slides had an average RPKM value of 56 in
the leaf sample and 67 in the seed sample, whereas the 26 genes
detected as differentially expressed on both PC and glass slides
had much higher average RPKMs in both the leaf (341 PRKM)
and seed (3965 RPKM) samples, respectively. Thus, both microarray
and sequencing data suggest the PC can reliably quantify
genes with expression levels at least 1 order of magnitude lower
than measured with conventional glass microarrays. By expanding
the dynamic range of the microarray experiment, the number of
genes for which statistically significant changes in expression
could be observed improved from 27 to 68 genes, or from 13 to
34% of the genes probed in the experiment. Because the gene
expression follows a power law distribution, modest enhancements
in the performance of the assay can dramatically increase the
number of genes researchers are able to probe in this microarray
format. This data thus suggests that the detection capabilities of
microarray protocols currently used today can be greatly expanded
by substitution of conventional substrates with enhanced fluorescence
substrates such as PCs.
The increased SNRs provided by PCs may allow researchers
to perform experiments that are currently problematic on glass
slides. Because lower amounts of bound sample can be detected
with the PC, sample sizes may be reduced to volumes that would
be difficult to probe using normal glass substrates. This may be
particularly helpful for profiling gene expression in smaller tissue
samples or small populations of rare cells such as stem cells.
Alternately, the reduction in experimental variation afforded by
this substrate may allow researchers to confidently identify
differentially expressed genes with fewer replicates, which may
also prove useful with small sample sizes or rare cells. This
approach is not limited to conventional DNA microarray experiments.
Any surface-bound biomolecular assay can be performed
on these PCs for improved performance, as is illustrated in
previous work with immunoassays. 25 This substrate can also
potentially be adapted to improve reliability of novel technologies
such as next-generation genomic sequencing platforms, since
these instruments make extensive use of fluorescent molecules.
CONCLUSION
We have demonstrated that enhanced fluorescence is capable
of significantly improving a DNA microarray that probes changes
(25) Mathias, P. C.; Ganesh, N.; Cunningham, B. T. Anal. Chem. 2008, 80,
9013–9020.
in gene expression between samples. By using a PC substrate
with uniform optical characteristics over microscope slide-sized
areas, the SNR from microarray spots was increased by an order
of magnitude compared to commercial glass substrates. This SNR
enhancement translated into a greater number of genes detected
above the noise level and allowed for the detection of statistically
signficant changes in low expression genes. After evaluating
differential expression in soybean trifoliate tissue versus cotyledon
tissue, more than twice as many genes were characterized as
differentially expressed on the PCs compared to the glass slides,
and many of these were validated by high-throughput mRNA
sequencing data. Using a PC substrate for microarray experiments
thus opens the possibility to interrogate the roles of genes that
previously could not be reliably quantified in a high-throughput
fashion.
ACKNOWLEDGMENT
This work was supported by the National Institutes of Health
(grant no. GM086382A), the National Science Foundation (grant
no. CBET 07-54122), SRU Biosystems, and the Illinois Soybean
Association. Any opinions, findings, conclusions, or recommendations
expressed in this material are those of the authors and do
not necessarily reflect the views of the National Institutes of Health
or the National Science Foundation. The authors thank Stephen
Schulz, Brenda Hugh, Frank Jackson, and Kurt Albertson at SRU
Biosystems for attaching photonic crystal substrates to microscope
slides. The authors thank the staff at the Micro and Nanotechnology
Laboratory at the University of Illinois at Urbana-Champaign.
The authors also thank Sean Bloomfield for bioinformatics
assistance and the staff of the Keck Center at the Biotechnology
Center at University of Illinois for Illumina sequencing.
SUPPORTING INFORMATION AVAILABLE
Tables S-1 and S-2 list genes found to have statistically
significant changes in expression on glass and photonic crystal
slides and Tables S-3 and S-4 show corresponding RNA transcript
levels. This material is available free of charge via the Internet at
http://pubs.acs.org.
Received for review March 31, 2010. Accepted July 2,
2010.
AC100841D
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6861

Anal. Chem. 2010, 82, 6862–6869
Gas Chromatography-Combustion-Mass
Spectrometry with Postcolumn Isotope Dilution for
Compound-Independent Quantification:
Its Potential to Assess HS-SPME Procedures
Sergio Cueto Díaz, Jorge Ruiz Encinar,* Alfredo Sanz-Medel, and J. Ignacio García Alonso*
Department of Physical and Analytical Chemistry, University of Oviedo, Julián Clavería 8, 33006 Oviedo, Spain
A quadrupole GC-MS instrument with an electron ionization
(EI) source has been modified to enable application
of postcolumn isotope dilution analysis for the standardless
quantification of organic compounds injected in the
gas chromatograph. Instrumental modifications included
the quantitative conversion of the separated compounds
into CO 2, using a postcolumn combustion furnace, and
the subsequent mixing of the gas with a constant flow
of 13 CO2 diluted in helium. The online measurement
of the 12 CO2/ 13 CO2 (44/45) ratio in the EI-MS allowed
us to obtain quantitative data without resorting to
compound-specific standards. Validation of the procedure
involved the analysis of standard solutions
containing different families of organic compounds
(C 9-C20 linear hydrocarbons, BTEX and esters) obtaining
satisfactory results in all cases in terms of
absolute errors (

corresponding compound if its chemical formula is known. Until
now such methodology has been exclusively applied to the
quantitative determination of compounds containing ICPMS
detectable elements for trace-elemental speciation of metals 9 or
semimetals. 10,11 Unfortunately, the use of the ICPMS for the
detection of carbon, hydrogen, nitrogen, or oxygen is seriously
hampered by the low ionization yields of these elements in the
ICP and the high carbon background under normal ICP operating
conditions (atmospheric pressure). In fact, there are only two
reports so far describing the use of HPLC and the postcolumn
addition of 13 C-labeled species for the quantification of organic
compounds by ICP-MS. 12,13 In both cases, the observed detection
limits were not satisfactory.
For the MS detection of organic compounds, previously
separated by GC, electron ionization (EI) is the most common
ionization source employed. It provides both structural and
quantitative information of any volatile organic compound injected
in the gas chromatograph. Unfortunately, as pointed out before,
it requires specific analytical standards as its response is structurespecific
for each single molecule subjected to analysis. 14 Recently,
we have introduced a new quantitative detection concept in gas
chromatography, based on the postcolumn addition of 13 CO2 for
carbon isotope dilution analysis using EI. 15 This concept
constitutes a patented procedure in which organic compounds
separated by liquid or gas chromatography are converted
quantitatively into carbon dioxide, by an oxidation or combustion
reaction, and then are mixed with a postcolumn flow of
enriched 13 CO2. 16 This procedure should provide quantitative
information of every single compound previously separated in
the chromatograph without the need for individual standards.
We selected electron ionization because it operates under high
vacuum conditions providing much better sensitivity and lower
background for carbon detection than the above-mentioned ICP
atmospheric source. For the correct application of isotope
dilution analysis conversion of all carbon containing compounds
to a unique chemical species is required (isotope equilibration).
The isotopic equilibrium between the 13 C-containing species
continuously added postcolumn ( 13 CO2) and the separated
organic compounds is reached by their quantitative conversion
into CO2, after the chromatographic separation, in a combustion
furnace. 17 As the only species to be finally measured by EI is
CO2, compound independent response and isotopic equilibration
is finally obtained. This approach can become a universal
quantitative detector in GC-MS analysis, without the classical
need for specific standards and, what is more, it is compatible
with the exceptional structural identification capabilities provided
by current GC-EI-MS instruments. Thus, in this work
(9) Sariego-Muñiz, C.; Marchante-Gayón, J. M.; García-Alonso, J. I.; Sanz-Medel,
A. J. Anal. At. Spectrom. 2001, 16, 587–592.
(10) Giusti, P.; Schaumlöffel, D.; Ruiz-Encinar, J.; Szpunar, J. J. Anal. At. Spectrom.
2005, 20, 1101–1107.
(11) Heilmann, J.; Heumann, K. G. Anal. Chem. 2008, 80, 1952–1961.
(12) Vogl, J.; Heumann, K. G. Anal. Chem. 1998, 70, 2038–2043.
(13) Smith, C.; Jensen, B. P.; Wilson, I. D.; Abou-Shakra, F.; Crowther, D. Rapid
Commun. Mass Spectrom. 2004, 18, 1487–1492.
(14) Mark, T. D. Int. J. Mass Spectrom. Ion Phys. 1982, 45, 125–145.
(15) Cueto-Díaz, S.; Ruiz-Encinar, J.; Sanz-Medel, A.; García-Alonso, J. I. Angew.
Chem., Int. Ed. 2009, 48, 2561–2564.
(16) Ruiz-Encinar, J.; García-Alonso, J. I World Intellectual Property Organization,
Spain; International Patent WO 2007/042597 A1, 2007.
(17) Merritt, D. A.; Freeman, K. H.; Ricci, M. P.; Studley, S. A.; Hayes, J. M.
Anal. Chem. 1995, 67, 2461–2473.
we describe the instrumental developments and its analytical
features for integral characterization and determination of
organic compounds in detail.
Furthermore, the compound-independent quantification provided
by the developed approach can be an inestimable tool in
the optimization and quantitative assessment of sample extraction
and preconcentration procedures applied prior to GC-MS analysis.
For instance, solid-phase microextraction (SPME) is today a
widely used preconcentration technique in Gas Chromatography. 18
Fiber absorption and recovery are common parameters used when
validating a given SPME procedure. However, those two parameters
are almost impossible to compute using conventional
approaches. 19 In fact, it is necessary to inject specific liquid
standards and to assume that the transfer efficiency in the GC
injector for every compound under analysis is the same when
using conventional injection and the thermal desorption. Thus,
the important topic of absolute absorption yields will be evaluated
in this work, as exemplified for the HS-SPME analysis of BTEX
(Benzene, Toluene, Ethylbenzene, and o,m,p-Xylenes) in different
water samples, using the proposed IDA-EI-MS quantification
procedure.
EXPERIMENTAL SECTION
Reagents and Materials. Solid enriched Na2CO3 (99% 13 C
enrichment) was obtained as a highly pure chemical reagent
(purity >98%) from Cambridge Isotopes Laboratories (Andover,
Massachusetts, USA). A standard mixture of n-alkanes (40 µg/
mL each in n-hexane) was purchased from Fluka (Seelze,
Germany). Phosphoric acid (99% puriss.) was purchased from
Sigma-Aldrich (St. Louis, USA). Individual compounds (undecane,
dodecane, tridecane, pentadecane, butyl butyrate, and
hexyl butanoate) with certified purities ranging from 99 to 99,7%
were used as analytical standards (Fluka, Seelze, Germany or
Sigma-Aldrich, St. Louis, USA). A solution of 1,2,4-trimethylbenzene
in methanol (5000 µg/mL) and a BTEX standard
mixture (2000 µg/mL each in methanol) were both from
Supelco (Bellefonte, USA). Ultrapure water (18.2 MΩcm) was
obtained with a Milli-Q system (Millipore, Bedford, MA).
n-Hexane for organic trace analysis grade was purchased from
Merck (Darmstadt, Germany). SPME holder and fibers were
purchased from Supelco (Bellefonte, USA).
Instrumentation. GC-MS. A Konik-Tech (Sant Cugat del
Vallés, Spain) 4000-B Gas Chromatograph coupled to a Konik-
Tech MS-Q12 quadrupole mass spectrometer with an electron
ionization source was used. This instrument uses a specially
designed injector which keeps the septum at relatively low
temperature, does not require a septum purge flow, and minimizes
losses of organic compounds in the injector port. The analytical
column was a 30 m long (0.32 mm internal diameter, 0.25 µm
stationary phase) DB-XLB column (Agilent J&W Scientific, Santa
Clara, USA). The sample volume injected in the GC was always 1
µL using manual injection. Sample injection was performed in the
splitless (1 min)/split mode. Carrier gas flow was set at 1 mL/
min for all samples and conditions.
Six-Way Valve. A manually actuated 0.25 mm bore stainlesssteel
six-way two-position valve (VICI AG International, Schenkon,
(18) Vas, G.; Vekey, K. J. Mass Spectrom. 2004, 39, 233–254.
(19) Langelfeld, J. J.; Hawthorne, S. B.; Miller, D. J. Anal. Chem. 1996, 68,
144–155.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6863

Switzerland), was placed inside the chromatographic oven to
bypass the combustion furnace when required. The valve has an
extension between the body and the actuator which allows the
valve to be mounted in the heated zone while the handle remains
outside at ambient temperature. This valve prevented the solvent
from entering the combustion unit and therefore enlarged the
catalytic activity of the Cu and Pt wires. Additionally, the valve
allowed the direct connection of the column with the EI source
for conventional GC-MS work in order to identify the different
species under analysis by their fragmentation pattern. Moreover,
such qualitative analysis is essential to assess peak purity for each
compound. For postcolumn isotope dilution analysis the valve was
initially in the “load” position (combustion oven bypassed) and
the valve was manually switched to the “inject” position at the
same time that the EI filament switched on. All connections
between the valve and the rest of the components of the system
were performed by means of 0.32 mm i.d. deactivated fused silica
capillaries and fixed to the valve with appropriate polyimide coated
fused silica adapters (VICI AG International, Schenkon, Switzerland).
Combustion Furnace. The laboratory-made combustion furnace
consisted of a 60 cm long ceramic tube (3 mm O.D., 0.5 mm I.D.)
(Elemental microanalysis, Devon, U. K.) filled with copper and
platinum wires and heated by a Nichrome wire. Proper thermal
isolation of the combustion furnace was provided with glass wool.
The combustion furnace was set vertically on top of the GC with
the lower end of the ceramic tube inside the chromatographic
oven to avoid cold spots after the separation. High temperatures
were accurately controlled using a temperature sensor and an
external controller, allowing temperature settings inside the tube
to be within ±1 °C over a wide temperature range (50-1200 °C).
The copper wires were previously oxidized by passing an oxygen
flow (1 mL/min) at 450 °C during 4-5 h and this procedure was
performed on a weekly basis to preserve its oxidizing capabilities.
Fused silica capillaries (0,32 mm i.d.) were used to connect the
furnace to the valve and to the ion source respectively. For this
purpose a length (∼1 cm) at one end of each capillary was
uncoated, to prevent the polyimide coating to be burnt, and
introduced into the ceramic tube. Reducing unions 1/8′′ to 1/16′′
(SGE, Victoria, Australia) and appropriate graphite ferrules (0.5
mm i.d.) were used to hold the capillaries in place at both sides
of the ceramic tube. For oxidation and combustion, the furnace
was operated at 850 °C.
Gas Cylinder and Mass Flow Controller. The 13 CO2 container
was a dual inlet 5 L high pressure stainless steel gas cylinder
(Iberfluid, Barcelona, Spain). The cylinder was equipped on
one end with, in this order, an opening valve, a Swagelok tee
where helium could be introduced from a high pressure
cylinder, and a second opening valve connected to the other
end of the tee. This second valve was connected to a septum
for the manual injection of gases into the cylinder. Before the
cylinder was pressurized, 13 CO2 was injected into the container
by means of a gastight syringe (Hamilton, Reno, U. S. A.)
through this valve. After the injection of the tracer, the valve
was closed and the container was pressurized up to 6 bar with
helium using the connection in the Swagelok tee. When the
set pressure was reached the filling valve was also closed. At
the other end of the cylinder a pressure gauge, an opening
6864 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
valve and a mass-flow controller (Bronkhorst, Ruurlo, Netherlands)
calibrated for He were coupled for the accurate control
of the tracer flow. The opening valve was closed during the
filling of the cylinder, remaining open the rest of the time. The
flow rate for the postcolumn spike was set at 0.5 mL/min.
Effluent and spike flows were mixed after combustion by means
of a 0.25 mm bore stainless steel microvolume “Y” connector
(VICI AG International, Schenkon, Switzerland).
The whole instrumental setup is shown in Figure 1. As can be
observed, the instrument can be operated in the standard GC-
MS configuration (qualitative) or in the combustion-postcolumn
configuration (quantitative) depending on the position of the
switching valve.
Procedures. Preparation of 13 CO2. The spike was prepared
from 13 C enriched Na2CO3 (99%). An accurate weighed amount
(∼200 mg) was placed in a 25 mL three-necked round-bottomed
flask, previously purged with He to avoid natural abundances
CO2 contamination from ambient air. A small quantity (300 µL)
of concentrated H3PO4, was injected into the flask through a
septum cap. After the acid-base reaction, 4 mL of the gaseous
phase, containing 13 CO2 diluted in He, were removed using a
gastight syringe and injected into the container shown in
Figure 1.
Calibration of the 13 CO2 Postcolumn Flow. The flow rate of
the spike could be accurately controlled by the mass flow
controller between 0.1 and 5 mL min -1 . In our experiments, it
was set at 0.5 mL min -1 . The exact mass flow (ng of 13 CO2 per
min) being mixed with the natural CO2 coming from the
column and combustion furnace was determined by adding
internal standards of known concentration spiked to the sample
as described before. 11,15
Quantification using Postcolumn Isotope Dilution. In our case,
the isotope ratio 12 C/ 13 C was measured as the signal ratio at
masses 44 and 45 (I44/I45) corresponding to the continuous
blend of natural abundance 12 CO2 present in the chromatographic
eluent and the enriched 13 CO2, added postcolumn.
Selected Ion Monitoring (SIM) at masses 44.0 and 45.0 was
performed for the duration of the chromatogram with 70 ms
integration time per mass. The mass window was ∼0.1 mass
units. Then, the isotope ratio in the blend, Rb ) I44/I45, was
calculated to build the isotope ratio chromatogram (Rb vs time).
The postcolumn isotope dilution equation, 7 shown as equation 1
below, was then applied to every point in the chromatogram
to obtain the mass-flow chromatogram (ng of C/min vs time).
The integration of the mass flow chromatogram directly
provided the amount of carbon (in ng) eluted in each chromatographic
peak.
MF n ) MF t
AW n
13
At AWt An 12( Rb - Rt 1 - RbRn) In this equation MFn corresponds to the mass flow of carbon
from the natural abundance sample injected, whereas MFt
corresponds to the mass flow of carbon from the postcolumn
spike or tracer. AWn and AWt correspond to the atomic weight
of carbon in the sample and tracer, respectively. The isotope
abundances At 13 and An 12 correspond to the isotopic composition
of 13 C in the tracer and 12 C in the sample. Finally, Rt is the
(1)

Figure 1. Schematic illustration of the setup showing the modification made in the GC-MS instrument. The two possible valve configurations
are shown in the inset: Left, for structural identification by conventional GC-EI-MS analysis, and right, for absolute quantification using GC-
Combustion-MS and postcolumn addition of 13 CO2. Adapted from reference. 15
isotope ratio (12/13) in the tracer and Rn is the natural isotope
ratio (13/12) in the sample, whereas Rb is the measured isotope
ratio in the blend.
Additionally, eq 1 can be used to quantify the mass flow of
enriched carbon dioxide when an internal standard of known
concentration is spiked to the sample.
Head-Space SPME Procedure. Test solutions (∼10 ng/g) were
prepared by diluting appropriately the BTEX standards in Milli
Q water. Samples of ∼7 g were pipetted into 10 mL glass vials
with open-top phenolic closures and PTFE/silicone septa (Supelco,
Bellefonte, USA). Three fiber coatings, 100 µm polydimethylsiloxane
(PDMS), 70/30 µm polydimethylsiloxane/divinylbenzene
(PDMS/DVB), and 50/30 µm carboxen/polydimethylsiloxane/
divinylbenzene (CAR/PDMS/DVB) were tested for comparative
purposes. Samples were saturated in all cases with NaCl to
improve extraction efficiency and stirred at 900 rpm at 30 °C
during 30 min to allow equilibration between the liquid and gas
phases. Then the fibers were inserted in the headspace and
extracted during 15 min to ensure equilibrium conditions between
the headspace and the fiber. After extraction, the fibers were
removed and thermal desorption was carried out in the injection
port of the GC at 250 °C.
RESULTS AND DISCUSSION
Online Continuous Measurement of the 12 C/ 13 C Isotope
Ratios. The optimum conditions for the continuous measurement
of 12 C/ 13 C isotope ratios were established by connecting directly
the reservoir that contained the 13 CO2 spike to the mass
spectrometer. 15 The mass spectrum obtained for the mass
range 43-47 when such postcolumn flow was set at 1 mL/
min is provided in the Supporting Information (Figure SI-1).
The 13 C isotope enrichment in the postcolumn spike (∼97%)
can be clearly seen from the spectrum. Resolution was
optimized while still preserving a good sensitivity. The mass
resolution finally obtained allowed baseline separation even for
isotope ratios very far from 1. The experiments performed for
the selection of the optimum integration time are given in the
Supporting Information (Figure SI-2). An integration time of 70
ms was finally selected considering both the precision in the ratio
measurement and the number of points per chromatographic peak
(∼15) to obtain well-defined peak profiles.
Spike Enrichment and Stability of the 12 CO2/ 13 CO2
Isotope Ratio. Again, to estimate the 13 C enrichment obtained
after the preparation of the 13 CO2 spike, its reservoir was
connected directly to the mass spectrometer. A constant flow
of1mLmin -1 was set and the signals at m/z ratios 44 and 45
were continuously monitored during 2 min for each independent
isotope ratio measurement under the measurement
conditions described in the procedures. The isotope ratio was
calculated obtaining an average value of 0.034 ± 0.0003, which
corresponds to a 13 C abundance of 96.70 ± 0.02% (n ) 5) (see
the mass spectrum shown in the Figure SI-1). The slight
difference between the experimentally obtained enrichment and
that provided for the precursor Na2 13 CO3 by the supplier (99%),
can be ascribed to small contaminations by natural abundance
CO2 from the ambient air during the generation of the tracer
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6865

Figure 2. Intensity chromatograms for the n-alkanes mixture
obtained in the SIM mode: (a) at m/z ) 71 operating the setup in the
qualitative configuration; (b) at m/z 44 and 45 operating the setup in
the quantitative configuration.
and the filling of the reservoir, traces of this compound possibly
present in the Helium gas used to dilute de 13 CO2 and traces
of air still present in the EI source. The high enrichment
obtained provides a very wide range of optimum analyte/spike
ratios and then a sample containing compounds in a large range
of concentrations could be quantified in the same analysis
without affecting the accuracy of the results. 3
The short-term (30 min) and long-term (9 h) stability of the
12 C/ 13 C isotope ratio is given in the Supporting Information
(Figures SI-3 and SI-4). In both cases the measured isotope ratios,
0.0339 ± 0.0002 and 0.034 ± 0.001 (n ) 8), respectively, were
remarkably stable. Similar results were obtained in different days
indicating that the setup was robust enough for routine analysis.
Evaluation of the GC-Combustion-IDMS System. The
overall performance of the developed instrumentation was evaluated
here by analyzing a standard solution of a mixture of
n-alkanes (C9-C20). To study peak broadening, 1 µL of a
solution containing approximately 5 µg/g of these compounds
in n-hexane was injected in the chromatograph both in the
qualitative (combustion oven bypassed) and quantitative (through
the oven) modes. The Selected Ion Monitoring chromatogram
obtained at m/z ) 71 (fragment characteristic of n-alkanes) is
shown in Figure 2a, whereas the chromatogram obtained at
masses 44 and 45 after combustion and postcolumn isotope
dilution analysis is shown in Figure 2b. As can be observed, no
significant peak broadening due to the combustion unit or to the
different connections was observed, being the peak width measured
at the half height equal to that found when operating the
6866 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. Mass flow chromatograms obtained for the n-akanes
mixture.
GC-MS in the conventional way (0.02 min, Figure 2a). Only a small
increase in retention times was observed (∼20 s) because of the
combustion furnace. The mass flow chromatogram obtained after
the application of the online isotope dilution equation (eq 1) is
shown in Figure 3. As can be observed, the sensitivity is roughly
constant for all compounds which is in clear contrast with the
results shown in Figure 2a for mass 71 without combustion (SIM
detection). The peak areas in the mass flow chromatogram were
linear with the amount of carbon injected over 2 orders of
magnitude (the maximum range assayed). The detection limit
obtained for tetradecane was 9 ppb (ng/g) based on three times
the standard deviation of the baseline. Taking into account that
the injection volume was 0.5 µL, this value corresponds to an
absolute limit of detection of 3 pg of tetradecane injected. If we
take into account the peak width at the baseline (4s approximately
in this particular case), then we can make our detection limit
independent of the column employed and the carrier gas flowrate.
The value found was 0.8 pg C s -1 , being in the same order
that those provided by a mass spectrometer in full scan mode
(1 pg C s -1 ) and clearly better than those of a flame ionization
detector (10 pg C s -1 ). This detection limit is mainly limited
by the background level observed at m/z ) 44, and explains
the difference in comparison with the detection limit achievable
using the mass spectrometer in SIM mode bypassing the oven
(typically 0.1 pg C s -1 ) in spite of the fact that both sensitivities
were very similar. In contrast, it should be mentioned that the
EI detection limit observed is 3 orders of magnitude lower than
that obtained for carbon using 13 C postcolumn isotope dilution
and ICP-MS detection (0.7 ng C s -1 ). 12
For the quantitative analysis of the standard mixture of
n-alkanes (C9-C20) we needed first to quantify the mass flow
of postcolumn 13 CO2. Out of the two possibilities of using
internal standard (IS) or external standard, we selected an IS
because it provided the advantage of compensation for small
variations in the sample volume injected (we were using manual
injection). Moreover, as recently pointed out by Heilmann and
Heumann, 11 an additional advantage of using an internal
standard is that the mass flow of enriched carbon dioxide does
not need to be determined. The peak areas of both the analyte
and the internal standard in the mass flow chromatogram are
related to the unknown mass flow of spike (eq 1). However,
the ratio of peak areas analyte/IS will be equal to the actual ratio
of concentrations in the injected sample and independent of the
mass flow of spike used in the calculations. Thus, quantification

Table 1. Absolute Quantification Results for the
Alkane Model Mixture a
compound added (µg/g) found (µg/g)
nonane 8.0 8.3 ± 0.5
decane 8.0 8.3 ± 0.2
undecane 8.0 8.4 ± 0.1
dodecane 8.0 8.3 ± 0.3
tridecane 8.0 8.3 ± 0.3
tetradecane 8.0 I.S.
pentadecane 8.0 7.7 ± 0.1
hexadecane 8.0 7.7 ± 0.1
heptadecane 8.0 7.4 ± 0.3
octadecane 8.0 7.9 ± 0.1
nonadecane 8.0 7.8 ± 0.1
eicosane 8.0 8.3 ± 0.2
a Tetradecane was used as an internal standard. The uncertainty is
expressed as the standard deviation for n ) 3 injections.
of the amount of carbon under each peak can be determined just
by calculating the areas of the internal standard and those
corresponding to the other organic compounds. Therefore, the
knowledge of the spike mass flow is not needed. The quantitative
results obtained from the data shown in Figure 3, using tetradecane
as internal standard, are given in Table 1. As can be observed,
this methodology allowed us to quantify a series of organic
compounds containing from 9 to 20 atoms of carbon and with a
range of boiling points from 128 to 343 °C, with acceptable
precision (

Figure 4. Mass flow chromatogram obtained for the mixture of BTEX
and 1,2,4-trimethylnenzene in water using HS-SPME-GC-Combustion-MS
and postcolumn addition of 13 CO2.
nately, the CAR/PDMS/DVB fiber generated broader peaks and
tailing, resulting in lower signal-to-noise ratio and poorer precision
in comparison with the PDMS/DVB fiber, which was finally
selected for further experiments. In addition, the fiber/sample
distribution constants (Kfs) were calculated for a more reliable
comparison between the extraction efficiency of the fibers as
it is a parameter related to the phase volume 23 (see Table SI-2
in the Supporting Information). The Kfs values obtained for the
PDMS fiber were of the same order of those cited in the
literature. 19 Interestingly, Kfs values obtained for the PDMS/
DVB fiber were significantly higher than those obtained for
the CAR/PDMS/DVB fiber, in spite that both fibers provided
very similar absolute recoveries (see Table 2). These results
seem to indicate that the PDMS/DVB stationary phase showed
higher affinity for the BTEX compounds. Notably, only nine
analysis (three replicates per fiber tested) were necessary to obtain
this valuable information in fundamental studies, critical to
optimize new procedures of SPME.
As an example, Figure 4 shows the mass-flow chromatogram
obtained using the PDMS/DVB fiber. In this case, together with
the impressive preconcentration factor typically observed by
SPME, the carbon background was significantly decreased (two
or three times lower). The detection limits obtained using HS-
SPME were in the low ng/L level (

ment. Therefore, the long-time dreamed of chromatographic
detection allowing generic quantification and complete characterization
of organic compounds in a single instrument is
realized. In addition, the proposed method turned out to be
very cost-effective because expenses associated with the use
of isotopically labeled 13 CO2 are negligible and there is no need
for external calibration, resulting in considerable savings in time
and money (e.g., cost of analytical standards). Beyond its
potential use for quantitative quality control in a wide range of
standard laboratories (e.g., environmental ones), a powerful
application of our approach can be foreseen in oil-spill fingerprinting
and in pharmaceutical analysis, for which the number
of target organic analytes is increasing exponentially (so it is
virtually impossible to have standards for each compound, even
if they exist).
In addition, the unique compound-independent calibration
capabilities of the approach proposed could be advantageously
exploited for the assessment of sample introduction procedures
in GC. In the present work, we have selected HS-SPME to
illustrate this promising ability of our approach. Quantitative data
obtained for BTEX compounds in different water samples have
demonstrated that quantitative assessment of HS-SPME procedures
in terms of absolute absorption yields is successful (the
analytical features of different fiber coatings could be evaluated
and critically compared in this way), opening its use to assess
the performance of many other reported sample preparation and
preconcentration methods.
ACKNOWLEDGMENT
Financial support was provided by the Spanish Ministry of
Education (CTQ2006-05722) and FICYT (PC06-016) and technical
support by KONIK-TECH. J.R.E. acknowledges the MEC (European
Social Fund) for a Ramon y Cajal contract, and S.C.D.
acknowledges the FICYT for a PhD grant.
SUPPORTING INFORMATION AVAILABLE
The mass spectrum obtained for the mass range 43-47 when
1 mL/min of the spike flow was introduced directly to the ion
source (Figure SI-1). The optimization of the integration time for
the 44/45 isotope ratio measurement (Figure SI-2). The short term
(30 min) and long term (9 h) stability of the 12 CO2/ 13 CO2 isotope
ratio (Figures SI-3 and SI-4, respectively). The quantification
results of a mixture of different families of compounds containing
saturated, insaturated and functionalized compounds (Table SI-
1). The Kfs values obtained for reliable comparison between the
different fibers used (Table SI-2). The multiple headspace
experiment carried out for internal validation of the approach,
comparing the sum of the amount obtained in each single
extraction to the total amount initially present in the spiked
solution (Table SI-3). This material is available free of charge via
the Internet at http://pubs.acs.org.
Received for review April 12, 2010. Accepted July 2, 2010.
AC101103N
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6869

Anal. Chem. 2010, 82, 6870–6876
Autonomous Microfluidic Control by Chemically
Actuated Micropumps and Its Application to
Chemical Analyses
Atsushi Takashima, Kenichi Kojima, and Hiroaki Suzuki*
Graduate School of Pure and Applied Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba,
Ibaraki 305-8573, Japan
Autonomous control of microfluidic transport was realized
through the use of chemically actuated diaphragm micropumps
connected to a network of controlling flow channels.
A hydrogen peroxide (H 2O2) solution was transported
in the controlling flow channel by capillary
action. Upon the solution’s arrival at the lower compartment
of a micropump filled with manganese dioxide
(MnO 2) powder, a volume change that accompanied
the production of oxygen caused by the catalytic
decomposition of H 2O2 induced inflation of the diaphragm.
This in turn caused the movement of a
solution in another network of flow channels formed
in the upper layer. Micropumps that only exert pressure
were also fabricated. By positioning the micropumps
at appropriate locations in conjunction with
additional flow-delaying components, the ejection of
solutions from the reservoir of each micropump could
be initiated at coordinated times. Furthermore, the
solutions could be transported by the application of
pressure from other micropumps. In other words, the
information for switching from one micropump to
another could be described on the chip in the form of
a network of flow channels. This autonomous processing
of solutions was demonstrated for enzymatic analyses
of H 2O2, glucose, and lactate.
With the progress now being made in microfluidic technologies,
innovative devices that make possible the complicated
manipulation of solutions have been proposed for various
applications. 1-5 However, as far as microfluidic transport is
concerned, many of these previous devices have relied on external
instruments such as microsyringe pumps or power sources to
produce pressure-driven flows or to generate electroosmotic flows.
Such bulky instruments, however, are obstacles to the increased
integration of components and miniaturization of the entire system.
* To whom correspondence should be addressed. Phone: +81-29-853-5598.
Fax: +81-29-853-4490. E-mail: hsuzuki@ims.tsukuba.ac.jp.
(1) Thorsen, T.; Maerkl, S. J.; Quake, S. R. Science 2002, 298, 580–584.
(2) Balagaddé, F. K.; You, L.; Hansen, C. L.; Arnold, F. H.; Quake, S. R. Science
2005, 309, 137–140.
(3) Shiu, J.-Y.; Chen, P. Adv. Mater. 2005, 17, 1866–1869.
(4) Wang, C.-H.; Lee, G.-B. Biosens. Bioelectron. 2005, 21, 419–425.
(5) Satoh, W.; Hosono, H.; Yokomaku, H.; Morimoto, K.; Upadhyay, S.; Suzuki,
H. Sensors 2008, 8, 1111–1127.
6870 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
To address this problem, trials have been performed to integrate
active microfluidic components on a single chip. 6-10
The long history of the development of micropumps and
microvalves has produced a variety of devices that are based on
various principles. 11-13 In the reported devices, actuation of the
components has usually been based on switching by electrical
signals that are programmed in a number of ways. In this
approach, however, specially designed electronic circuits and
software are needed to realize cooperative operation of the
components. In addition, for disposable devices, it would be
preferable for the microfluidic system to function autonomously.
To resolve this problem, chemical actuators that are based upon
the spontaneous volume change of a hydrogel have been
reported. 14-16 Capillary action has also been used to realize a
variety of devices for the autonomous transport of solutions and
various other applications. 17-23 These previous approaches,
particularly the latter one, suggest a direction for the realization
of more sophisticated devices in the next generation. The
manipulation of solutions in devices has been based on programmed
instructions described on the chip as a structural
(6) Choi, J.-W.; Oh, K. W.; Han, A.; Okulan, N.; Wijayawardhana, C. A.; Lannes,
C.; Bhansali, S.; Schlueter, K. T.; Heineman, W. R.; Halsall, H. B.; Nevin,
J. H.; Helmicki, A. J.; Henderson, H. T.; Ahn, C. H. Biomed. Microdevices
2001, 3, 191–200.
(7) Srinivasan, V.; Pamula, V. K.; Fair, R. B. Lab Chip 2004, 4, 310–315.
(8) Satoh, W.; Hosono, H.; Suzuki, H. Anal. Chem. 2005, 77, 6857–6863.
(9) Nashida, N.; Satoh, W.; Fukuda, J.; Suzuki, H. Biosens. Bioelectron. 2007,
22, 3167–3173.
(10) Abdelgawad, M.; Wheeler, A. Adv. Mater. 2009, 21, 920–925.
(11) Gravesen, P.; Branebjerg, J.; Jensen, O. S. J. Micromech. Microeng. 1993,
3, 168–182.
(12) Shoji, S.; Esashi, M. J. Micromech. Microeng. 1994, 4, 157–171.
(13) Laser, D. J.; Santiago, J. G. J. Micromech. Microeng. 2004, 14, R35–R64.
(14) Beebe, D. J.; Moore, J. S.; Yu, Q.; Liu, R. H.; Kraft, M. L.; Jo, B.-H.; Devadoss,
C. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 13488–13493.
(15) Suzuki, H.; Kumagai, A.; Ogawa, K.; Kokufuta, E. Biomacromolecules 2004,
5, 486–491.
(16) Suzuki, H.; Tokuda, T.; Kobayashi, K. Sens. Actuators, B 2002, 83, 53–59.
(17) Zhao, B.; Moore, J. S.; Beebe, D. J. Science 2001, 291, 1023–1026.
(18) Ahn, C. H.; Choi, J.-W.; Beaucage, G.; Nevin, J. H.; Lee, J.-B.; Puntambekar,
A.; Lee, J. Y. Proc. IEEE 2004, 92, 154–173.
(19) Bouaidat, S.; Hansen, O.; Bruus, H.; Berendsen, C.; Bau-Madsen, N. K.;
Thomsen, P.; Wolff, A.; Jonsmann, J. Lab Chip 2005, 5, 827–836.
(20) Delamarche, E.; Juncker, D.; Schmid, H. Adv. Mater. 2005, 17, 2911–
2933.
(21) Chung, K. H.; Hong, J. W.; Lee, D.-S.; Yoon, H. C. Anal. Chim. Acta 2007,
585, 1–10.
(22) Zimmermann, M.; Hunziker, P.; Delamarche, E. Microfluid. Nanofluid.
2008, 5, 395–402.
(23) Swickrath, M. J.; Burns, S. D.; Wnek, G. E. Sens. Actuators, B 2009, 140,
656–662.
10.1021/ac1009657 © 2010 American Chemical Society
Published on Web 07/29/2010

Figure 1. Chemically actuated micropumps with flow channels. (A) Exploded view of the micropumps and the flow channels. (B) Operation of
the micropump. Cross-sections are shown that include the flow channel for transport, the diaphragm, and the lower compartment for the H2O2
solution. First, the reservoir of the micropump is filled with a solution to be transported (top). When a H2O2 solution is transported in the controlling
flow channel and reaches the lower compartment of the micropump, bubbles are produced, the diaphragm inflates, and the solution in the upper
reservoir is injected into the upper flow channel (bottom).
arrangement of components, including the flow channel network. 18
Although there have been limitations in manipulation that is
performed only through capillary action, even the complicated
manipulation of solutions may be realized by coupling a programmed
microfluidic network with chemically actuated microfluidic
components.
In a number of previous studies, gas bubbles produced by the
electrolysis of water were used to produce a volume change that
would mobilize a solution in a microflow channel. 24-29 This
principle of operation is attractive for the realization of a chemically
actuated micropump, because gas production is accompanied by
many chemical reactions. In creating our device, we used the
volume change of oxygen bubbles produced by the catalytic
decomposition of H2O2. 26 To trigger the pumping action, a H2O2
solution was transported and supplied to the micropumps by
capillary action in a controlling flow channel. A network of
controlling flow channels described on a chip could be used
as a program to operate many micropumps cooperatively. In
other words, the timing of the switching among pumps could
be adjusted by changing the relative positions of the micropumps
and the length or other dimensional parameters of the
flow channels. In this paper we present the basic concept for
a chemically actuated micropump and its programming and
characterize the performance of the device.
(24) Böhm, S.; Timmer, B.; Olthuis, W.; Bergveld, P. J. Micromech. Microeng.
2000, 10, 498–504.
(25) Suzuki, H.; Yoneyama, R. Sens. Actuators, B 2003, 96, 38–45.
(26) Choi, Y. H.; Son, S. U.; Lee, S. S. Sens. Actuators, A 2004, 111, 8–13.
(27) Satoh, W.; Shimizu, Y.; Kaneto, T.; Suzuki, H. Sens. Actuators, B 2007,
123, 1153–1160.
(28) Shimizu, Y.; Takashima, A.; Satoh, W.; Sassa, F.; Fukuda, J.; Suzuki, H.
Sens. Actuators, B 2009, 140, 649–655.
(29) Blanco-Gomez, G.; Glidle, A.; Flendrig, L. M.; Cooper, J. M. Anal. Chem.
2009, 81, 1365–1370.
EXPERIMENTAL SECTION
Materials and Reagents. A thick-film photoresist (SU-8) was
purchased from MicroChem, Newton, MA. A precursor solution
of poly(dimethylsiloxane) (PDMS) (KE-1300T) was purchased
from Shin-Etsu Chemical, Tokyo, Japan. A precursor solution of
PVA-SbQ, SPP-H-13, was purchased from Toyo Gosei Kogyo,
Chiba, Japan. H2O2, manganese dioxide, and poly(oxyethylene)
sorbitan monolaurate (Tween 20) were purchased from Wako
Pure Chemical Industries, Osaka, Japan. The enzymes and
related reagents were obtained from the following commercial
sources: horseradish peroxidase (HRP; 100 U/mg), lactate
oxidase (LOD; 38 U/mg), and bovine serum albumin (BSA)
from Wako Pure Chemical Industries, Osaka, Japan; glucose
oxidase (GOD; 151 U/mg) and 25% glutaraldehyde (GA)
solution from Sigma-Aldrich, St. Louis, MO; N-acetyl-3,7dihydroxyphenoxazine
(Amplex Red) from AnaSpec, San Jose,
CA.
Basic Structure and Fabrication of the Microfluidic Devices.
The devices were constructed by stacking two PDMS
substrates on a glass substrate (Figure 1). Flow channels were
formed with PDMS using a template formed with a thick-film
photoresist (SU-8). The compartments for the pumps and solutions
to be transported were formed in the lower and upper PDMS
layers by punching.
A critical part of each micropump was a circular compartment
(diameter 2.5 mm) with a diaphragm. The diaphragm was formed
by intercalating a 50 µm thick PDMS sheet between the two
PDMS substrates. The lower part of the compartment was
connected to a controlling flow channel for the transport of a H2O2
solution. To form a MnO2 layer in the vicinity of the diaphragm,
a droplet of water containing a suspension of MnO2 powder
was put into the compartment that was then placed upside
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6871

down until the water evaporated. The amount of the powder
that remained was between 0.75 and 1.35 mg, depending on
the size of the micropump. To prevent leakage of the powder
and more effectively produce oxygen bubbles and exert
pressure upon the diaphragm, the compartment beneath the
diaphragm was stuffed with a plug of a PVA-SbQ gel, leaving
a space below it that could be filled with the H2O2 solution
introduced from the controlling flow channel. In forming the
plug, that space was filled with a precursor solution of PVA-
SbQ, which was then cured under a UV light. Air vents were
formed at appropriate locations to release pressure and facilitate
the transport and filling of the H2O2 solution. A circular
reservoir (diameter 1.0 mm) for a solution to be transported
was also formed with PDMS on the diaphragm layer and was
connected to a flow channel. The flow channels extending from
the reservoirs of several micropumps formed an appropriate
network that reflected their different purposes. An inlet was
formed on the reservoir to be filled with a transported solution.
Micropumps that were used only to apply pressure were
formed in a similar manner. In this case, an inlet for the
transported solution was not formed. For all devices, the
heights of the controlling flow channels and of the flow
channels in which solutions would be transported were 75 and
150 µm, respectively. The width of the flow channel for
transportation was 500 µm. After the reservoirs in the upper
PDMS substrate were filled with the necessary solutions, the
entire structure was inserted between two poly(methyl methacrylate)
(PMMA) plates and then fixed in place with bolts
and nuts.
Procedure and Principle of Operation. A change in the
volume in the upper solution reservoir is caused by the deformation
of the diaphragm that follows the production of oxygen
bubbles produced by the catalytic decomposition of H2O2. First,
aH2O2 solution is transported in the controlling flow channel
by capillary action and is injected into the lower compartment
of the micropump (Figure 1B, top). When the solution reaches
the MnO2 powder below the diaphragm, oxygen bubbles are
produced by the catalytic decomposition of H2O2:
MnO2 2H2O298 2H 2 O + O 2
The diaphragm then inflates and exerts pressure upon the solution
in the reservoir. As a result, the solution is pushed forward in the
flow channel and is transported to the lower stream (Figure 1B,
bottom). The structure can also be modified so that the reservoir
is filled with only air, instead of a solution to be transported. In
this case, only a change in pressure is generated to move a liquid
column that may be present in the lower stream of the extending
flow channel. Several micropumps can be connected to the
controlling flow channel and the flow channel for the transport of
necessary solutions. Each micropump can be switched on individually
and sequentially according to a predetermined schedule
that is programmed in the shape of the network of controlling
flow channels.
Construction and Operation of Analysis Systems. To
demonstrate the sequential manipulation of solutions for chemical
6872 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
analyses, two devices were fabricated. In one of them (Figure 5),
a solution containing 50 U/mL HRP and another solution containing
H 2O2 as an analyte and Amplex Red (5 mM) were filled in
the reservoir of two micropumps. The solutions were prepared
with a 50 mM Tris-HCl buffer solution (pH 7.4). After the
solutions were ejected from the pumps and merged at the
T-junction according to the programmed pumping, their
intensity of fluorescence was measured using a fluorescence
microscope (VB-G25, Keyence, Tokyo, Japan) equipped with
a CCD detection system (Keyence VB-7000/7010). In a more
complicated device (Figure 6), enzymes (GOD and LOD) were
immobilized in two of three injection ports. HRP was immobilized
in a reaction chamber formed in the lower stream. In the
immobilization of the enzymes, an enzyme solution, a 0.1 wt %
BSA solution, and a 0.1 wt % GA solution were mixed in a 1:1:1
ratio. The mixed solution was then dropped into the corresponding
reservoirs or the reaction chamber (5 µL for the GOD and LOD
solutions and 1 µL for the HRP solution), and a cross-linking
reaction was allowed to proceed. Following this, the enzymeimmobilized
layers were immersed in a 0.1 M glycine solution
for 60 min. The activity of the immobilized enzymes was 1.3 U
for GOD, 0.17 U for LOD, and 3.3 × 10 -2 U for HRP. After the
injection ports were filled with solutions containing either
glucose or lactate or both of them, along with the immobilized
enzymes, H2O2 was produced by the enzymatic reactions. The
solutions were then transported to the reaction chamber. The
enzymatic reactions of HRP were accompanied by the generation
of fluorescence, whose intensity was measured. Values for
time, flow velocity, and fluorescence intensity were obtained
in five measurements, whose averages are used in the
following discussion.
RESULTS AND DISCUSSION
Movement of a Solution in the Controlling Flow Channel.
The controlling flow channel consisted of three walls of PDMS
and a bottom of glass. Although PDMS is hydrophobic (contact
angle 110°), the hydrophilic glass bottom alone (contact angle
15°) was able to generate a sufficient driving force to produce
capillary action. The velocity of the column of H2O2 changed
depending on device parameters such as the width, height, and
wettability of the flow channel. The velocity of a liquid column,
x˙, in a straight flow channel is expressed as follows: 30-33
x˙ ) γLV 8ηx( hw
h + w) 2
[
2 cos θPDMS w
+ cos θPDMS + cos θglass h ]
Here, γLV is the interfacial tension between the solution and
the air, η is the viscosity of the solution, x is the distance from
the inlet of the capillary to the meniscus of the moving liquid
column, h and w are the height and width of the flow channel,
and θPDMS and θglass are the contact angles on PDMS and glass,
respectively.
As could be anticipated from the equation, the movement of
the column slowed as it moved forward in the fabricated flow
(30) Satoh, W.; Yokomaku, H.; Hosono, H.; Ohnishi, N.; Suzuki, H. J. Appl. Phys.
2008, 103, 034903.
(31) Janshoff, A.; Künneke, S. Eur. Biophys. J. 2000, 29, 549–554.
(32) Delamarche, E.; Bernard, A.; Schmid, H.; Bietsch, A.; Michel, B.; Biebuyck,
H. J. Am. Chem. Soc. 1998, 120, 500–508.
(33) Kim, E.; Xia, Y.; Whitesides, G. M. Science 1995, 376, 581–584.

Figure 2. Influence of a surfactant (Tween 20) on microfluidic
transport. The distance indicates that of the meniscus of the liquid
column from the reservoir for the H2O2 solution in a straight flow
channel of 500 µm × 75 µm in cross-section. Concentration of the
surfactant: [, 0wt%;×, 0.005 wt %; 2, 0.01 wt %; b, 0.05 wt %; 9,
0.1 wt %.
channels. In addition, the movement occasionally became irregular,
possibly due to the morphological or chemical nonuniformity
of the channel. In the worst case, the column stopped
midway in its journey and did not reach the lower compartment
of the micropump. This problem was solved by adding a surfactant
to the solution, which facilitated smooth movement. Figure 2
shows the dependence of the movement of the solution on the
concentration of the surfactant (Tween 20) that was added. The
influence of the surfactant was dramatic, and the solution’s
movement became smoother and faster with increasing concentration
of surfactant. At concentrations higher than 0.01 wt %, the
flow velocity almost leveled off. In the following experiments, the
concentration was therefore fixed at 0.01 wt %.
In biochemical analyses in microsystems, the length of time
required for a reaction is often on the order of seconds or minutes.
In view of this, an additional requirement in such cases is the
presence of structures that can slow the flow velocity of solutions.
For this reason, we then examined how the velocity of the column
changed with changes in the width of the flow channel. In Figure
3A, flow channels 2-4 are straight and have widths of 250 µm,
500 µm, and 1.0 mm, respectively. With a change made only in
the width, a marked difference in flow velocity was observed that
demonstrated accelerated movement of liquid plugs in wider flow
channels. The presence of compartments positioned along the flow
channel exerts an additional similar influence. 34,35 Therefore,
rectangular compartments with dimensions of 1.5 mm × 880 µm
and 6.0 mm × 3.5 mm were attached to the 250 µm wide and 1.0
mm wide flow channels (flow channels 1 and 5, respectively). A
portion of the solution, however, also penetrated into the extending
controlling flow channel while the solution filled each compartment.
As a result, the movement of the column was not
significantly different from the case in which there were no
compartments. This result indicated that branched compartments
are not effective for this purpose.
We then tried a sequential arrangement. Figure 3B shows 500
µm wide flow channels. Flow channel 1 is straight, and flow
channels 2 and 3 have compartments of different sizes (2.0 mm ×
2.0 mm and 3.5 mm × 3.5 mm, respectively). By locating the exit
at an appropriate position in the compartment, the transport in
the flow channel was resumed after the compartment was filled
completely, and the effect of the structures was more significant
than that in Figure 3A. There was a tendency for bubbles to
remain in the corners of the square compartments. Although these
bubbles were small and had no adverse effect on the transport of
solutions, circular or elliptic compartments might be better, both
to avoid this potential problem and to realize a more accurate
adjustment of timing.
The movement of the column could also be delayed by the
introduction of constrictions. The width of the flow channels in
Figure 3C is 500 µm. For flow channels 2 and 3, the constrictions
were positioned near the inlet and had widths of 300 and 200 µm,
respectively. The constrictions also had an effect, and the movement
of the column was slowed with narrower constrictions.
Although we used only simple delaying structures because of the
limited space, microfluidic transport can be delayed further
through the use of a more complicated network of flow channels. 22
The movement of a solution in a flow channel with compartments
and/or constrictions can be understood using numerical
Figure 3. Movement of solutions in the flow channels with various delaying structures. (A) Effect of changing the width of the flow channel and
attaching rectangular compartments on the sides. (B) Effect of adding rectangular compartments in series. (C) Effect of creating constrictions.
The images were taken 1, 3, and 6 s (from left to right) after the introduction of the solution from the left. Scale bars correspond to 2 mm.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6873

simulations, 34,35 which can be helpful in the design of controlling
flow channels that punctually provide H2O2 solution to micropumps
according to a predetermined schedule.
Autonomous Sequential Switching of the Micropumps.
We next studied the function of the micropump. After the lower
compartment of the pump was filled with H2O2 solution, the
diaphragm inflated and the solution in the upper reservoir was
ejected and then mobilized in the extending flow channel.
During this step, the H2O2 solution moved forward in the
controlling flow channel and filled the lower compartment of
the next pump. From that point forward, the same step was
repeated. Needless to say, the concentration of H2O2 injected
into the controlling flow channel affects the flow velocity of
the solution in the upper flow channel. Although the flow
velocity definitely depends on the H2O2 concentration, significant
leakage and destruction of the gel layer was observed over
a certain threshold, which also depended on the size of the
pump. Considering these dynamics, 1.6 or 3.2 M H2O2 was used
in the following experiments, depending on the size of the
pump.
For the handling of many solutions, several micropumps can
be connected to the controlling flow channel and to the upper
flow channels for solutions to be transported. To adjust the timing
and order of the injection of solutions, the distance between pumps
can be adjusted in a network of flow channels. Compartments and
constrictions can also be introduced, as was discussed earlier. In
the device shown in parts A and B of Figure 4, micropumps are
connected with one another by rectangular compartments at their
sides and centers, respectively. In the device shown in Figure
4B, the compartments are filled with solution after the compartment
of a neighboring micropump in the upper stream is filled
with solution. With the small compartments in the crowded layout,
however, the extension of the liquid columns from the upper
reservoirs was not so significant compared with that of the device
shown in Figure 4A. The device shown in Figure 4C has additional
small compartments to delay the movement of the solution. Note
the difference in the length of the liquid columns, which shows
that the actuation of the micropump can be distinctly delayed in
the lower stream of the controlling flow channel, unlike the device
shown in Figure 4A and 4B. This result demonstrates that
additional delaying structures can in fact be used to adjust the
timing to trigger the actuation of the micropumps.
Coordinated Operation of Micropumps for Chemical
Analyses. By properly designing a network of flow channels with
micropumps located in appropriate positions, microfluidic devices
can be constructed for various purposes, including chemical
analyses that require the specific processing of solutions. In the
device shown in Figure 5, there are two micropumps to eject
solutions and two pumps to apply pressure. Here, H 2O2 was also
used as an analyte. A solution containing H2O2 and Amplex
Red (5 mM) and another solution containing 50 U/mL HRP,
both prepared with a 50 mM Tris-HCl buffer solution (pH 7.4),
were used to fill the reservoirs. After another H2O2 solution
was injected into the controlling flow channel, it first filled the
compartments of the micropumps. Following this, the analyte
H2O2 solution in the reservoir was injected into the upper flow
(34) Erickson, D.; Li, D.; Park, C. B. J. Colloid Interface Sci. 2002, 250, 422–
430.
(35) Young, W.-B. Colloids Surf., A 2004, 234, 123–128.
6874 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 4. Arrays of micropumps with controlling flow channels with
different structures. Rectangular compartments are connected with
the controlling flow channels at the edges (A), at the center (B), and
via smaller delaying compartments (C).
channels for transport after 9 s (Figure 5B, panels 1 and 2).
After a time delay (117 s), the larger pumps that apply pressure
were switched on and the solutions in the upper flow channel
were pinched off from the rest of the solutions, transported to
the center, and merged in the mixing channel (Figure 5B, panels
3 and 4). The average flow velocity in the mixing channel was 94
µm/s. Solutions containing the enzyme and the substrates were
transported and mixed in the flow channel. The enzymatic reaction
by HRP produced highly fluorescent resorufin, which generated
red fluorescence under a fluorescence microscope. Figure 5C
shows the dependence of the fluorescence intensity on the
concentration. The fluorescence intensity was measured 3 min
after mixing. In the graph plotted on the semilog scale, the
dependence of the fluorescence intensity on the concentration
could be clearly observed.

Figure 5. Device to eject and mix solution plugs. (A) Layout of
the controlling flow channels (dashed line, shaded) and the flow
channels for transport (solid line). (B) Movement of a H2O2 solution
in the controlling flow channels (dashed arrows) and an analyte
H2O2 solution in flow channels for the transport of solutions (solid
arrows). (1) The reservoirs of the micropumps were filled with
solutions. (2) The solutions were injected into the main channel.
(3, 4) The solutions were transported by the exertion of pressure
from the leftmost and rightmost pumps and were merged at the
T-junction. (C) Dependence of the fluorescence intensity on the
concentration of H2O2 ejected from a micropump. Five runs were
performed, and the averages and standard deviations are shown.
The dashed lines show the average +3σ of the background
fluorescence. The concentration of the H2O2 solution injected into
the controlling flow channel was 1.6 M. The dimensions of the chip
were 25 mm × 16 mm.
More complicated manipulation of solutions could also be
carried out. The device shown in Figure 6 has three injection
pumps and three micropumps to exert pressure upon the ejected
plugs. Three lines of controlling flow channels were used. In each
channel, a pump to eject a necessary solution and a pump to apply
pressure were connected. The lower compartment of each
injection pump was used to delay the arrival of the H2O2 solution
to the pumps to apply pressure located in the lower stream.
GOD and LOD were immobilized at the bottom of the
reservoirs of pumps A and C. First, we used solutions containing
either glucose or lactate. The reservoir of pump A was filled
with a phosphate buffer solution containing glucose and Amplex
Red (5 mM), and the reservoir of pump C was filled with
another phosphate buffer solution containing lactate and
Amplex Red (5 mM). The reservoir of pump B was filled with
a phosphate buffer solution to wash the reaction chamber. The
enzymatic reactions by the oxidases produced H2O2. In view
of the volume of the solutions and the activity of the immobilized
enzymes, it could be assumed that the enzymatic
conversion was virtually completed in the examined ranges of
concentration of glucose and lactate during this preparatory
period. Three minutes after the filling with solutions, the H2O2
solution was introduced into the controlling flow channel.
Following this, a row of plugs was injected into the main flow
channel from the reservoirs of the pumps located in the lower
stream. In accordance with the flow channel design, pumps
A-F were switched on 28, 85, 174, 187, 259, and 336 s after
the injection of the H2O2 solution into the controlling flow
channel. The velocity of a plug passing through the reaction
chamber was 35 µm/s, the same velocity at which the three
pumps would apply pressure. When the first plug containing
Amplex Red and H2O2 produced by GOD arrived at the reaction
chamber, fluorescence was generated in the chamber, as it was
in the previously mentioned device, and its intensity was
measured 90 s after the solution reached the chamber. After
the reaction chamber was rinsed with a rinsing plug that was
ejected from pump B, the last plug containing Amplex Red and
H2O2 produced by LOD was introduced into the reaction
chamber, and the fluorescence intensity was measured as
before. The same experiment was carried out using the same
solution containing glucose, lactate, and Amplex Red to fill
pumps A and C. Figure 7 shows the calibration plots obtained
using solutions containing only glucose or lactate and the data
points obtained for solutions containing both of them. For the
limited ranges of concentration that were employed, the plots were
apparently linear. The values obtained for solutions containing
both glucose and lactate in different combinations of concentration
came close to the calibration plots, which demonstrated that the
device can be used for the analysis of different analytes that coexist
in the same solution.
We note again that no wires or tubes were deployed around
the chip to apply electrical signals or pressure. The necessary
reagent solutions can be stored in the injection ports beforehand.
The only thing that must be done is to introduce a sample solution
and initiate the flow of the H2O2 solution in the controlling flow
channel. The H2O2 solution can also be stored in a separate
reservoir and can then be injected by applying a small degree
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6875

Figure 6. Enzyme analysis that accompanies the manipulation of
three plugs. (A) Layout of the controlling flow channels (dashed line,
shaded) and flow channels for transport (solid line). For purposes of
clarity, the two flow channel networks are drawn separately in the
lower figure. (B) Fluorescence images showing the movement of the
H2O2 solution in the network of controlling flow channels (dashed
arrows) and solution plugs in the network of upper flow channels (solid
arrows). (1) The first solution was transported to the reaction chamber.
(2) After flushing of the first solution, the reaction chamber was
washed with the second plug. (3) After flushing of the solution, the
third solution was transported to the reaction chamber. Although the
flows in the controlling flow channels are described separately, they
began to flow in the flow channels simultaneously. The concentration
of the H2O2 solution injected into the controlling flow channel was
3.2 M. The dimensions of the chip were 33 mm × 27 mm.
6876 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 7. On-chip detection of glucose and lactate using the device
shown in Figure 6. Dependence of the fluorescence intensity on the
concentration of glucose (O, 0) and lactate (b, 9). O and b indicate
data obtained in experiments using solutions that contained only
glucose or lactate. Five runs were performed, and the averages and
standard deviations are shown. 0 and 9 indicate data obtained by
filling the injection ports of pumps A and C with solutions containing
both glucose and lactate. The inset shows the plot at a lower
concentration range near the detection limits. The dashed lines show
the average +3σ of the background fluorescence.
of pressure that enables it to pass through a hydrophobic valve
set at the entrance.
CONCLUSIONS
Chemically actuated micropumps can be realized by making
use of the volume change produced by the catalytic decomposition
of H2O2. The pumps are located along a controlling flow channel
that transports a H2O2 solution via capillary action. The row of
pumps can be switched on sequentially following the introduction
of the H2O2 solution into the controlling flow channel. The
structure of the pump itself can also be used to exert pressure
upon a solution in a flow channel. The timing of the switching
among pumps can be adjusted by locating them at appropriate
positions in a network of flow channels or by employing
additional structures such as compartments and/or constrictions.
In other words, the information for switching among
pumps is directly described on the chip as a program.
As demonstrated, one potential application for our autonomous
devices is that of portable analysis systems. Although on-chip
biochemical analyses for molecules such as proteins have already
been carried out using integrated microfluidic components, 6,27,28
this technique will simplify the construction of the entire system.
Moreover, such an autonomous device can be useful for a variety
of purposes, such as micromixing, 23 tissue culturing, 19 and
gas-liquid reactions, 17 since it can minimize the burden involved
in the handling of solutions.
Received for review April 12, 2010. Accepted July 8, 2010.
AC1009657

Anal. Chem. 2010, 82, 6877–6886
Immunoaffinity Purification Using Anti-PEG
Antibody Followed by Two-Dimensional Liquid
Chromatography/Tandem Mass Spectrometry for
the Quantification of a PEGylated Therapeutic
Peptide in Human Plasma
Yang Xu,* John T. Mehl, † Ray Bakhtiar, † and Eric J. Woolf
Department of Drug Metabolism and Pharmacokinetics, Regulated Bioanalysis, Merck Research Laboratories,
WP75B-300, West Point, Pennsylvania 19486
Quantification of a PEGylated peptide in human plasma
using LC-MS/MS to support clinical studies presented
challenges in terms of assay sensitivity, selectivity, and
ruggedness. To ensure specific recognition of PEGylated
species, an immunoaffinity purification method (IAP)
using anti-PEG antibody followed by two-dimensional
(2D) LC-MS/MS was developed for MK-2662, an investigational
peptide containing 38 amino acids with a 40
kDa branched PEG [poly(ethylene glycol)] at C-terminus.
Biotinylated anti-PEG antibody, bound to streptavidincoated
magnetic beads, was used to capture MK-2662
and its stable-isotope-labeled internal standard from human
plasma. After on-bead digestion with trypsin, the
supernatant was injected on a 2D high-performance liquid
chromatography (HPLC) system constructed with strong
cation-exchange and reversed-phase columns, followed by
MS/MS detection of the surrogate N 1-12-mer of MK-
2662 on an API5000. The assay ruggedness was
improved by optimizing the trypsin digestion and
sample storage conditions. The intraday validation,
conducted in parallel with protein precipitation (PPT)
assay, demonstrated 94.8-105.8% accuracy with

Figure 1. Chemical structures of MK-2662 and its internal standard,
[ 13 C18, 15 N2] MK-2662 (ISTD). The underlined region indicates the
N-terminal tryptic peptide used as surrogate for quantification.
is time-consuming; therefore, reagent availability could become
a rate-limiting step. In the case of MK-2662 analysis, ELISA was
abandoned because a drug-specific antibody was not available at
the time of assay development.
Bioassay is a cell-based or enzyme-based assay that measures
the biologic activity of a specific biological process. 7 Since bioassay
measures the total activity of the sample irrespective of the
chemical structures, the assay specificity is usually less than that
of ELISA. It could provide a measure of pharmacodynamic (PD)
properties 7 but may not be appropriate for pharmacokinetic (PK)
assessment.
Liquid chromatography-mass spectrometry (LC-MS) has
been increasingly used to quantify peptides and proteins in
biological matrixes 9-11 because of its selectivity. Despite commonly
used sample cleanup techniques, such as protein precipitation
(PPT), 9 solid-phase extraction (SPE), 12-14 two-dimensional
(2D) SPE, 15 and 2D high-performance liquid chromatography
(HPLC), 16,17 the immunoaffinity purification (IAP) coupled with
LC-MS/MS represents an emerging strategy for peptide and
protein bioanalysis. The IAP strategies include immunoaffinity
depletion that can be used to remove abundant proteins from
biological matrixes 18,19 and immunoaffinity capture that utilizes
a single antibody to isolate and enrich the target peptides or
(9) van den Broek, I.; Sparidans, R. W.; Schellens, J. H. M.; Beijnen, J. H.
J. Chromatogr., B 2008, 872, 1–22.
(10) Careri, M.; Mangia, A. J. Chromatogr., A 2003, 1000, 609–635.
(11) John, H.; Walden, M.; Schafer, S.; Genz, S.; Forssanmm, W. G. Anal.
Bioanal.Chem. 2004, 378, 883–897.
(12) van den Broek, I.; Sparidans, R. W.; Huitema, A. D. R.; Schellens, J. H. M.;
Beijnen, J. H. J. Chromatogr., B 2006, 837, 49–58.
(13) van den Broek, I.; Sparidans, R. W.; Huitema, A. D. R.; Schellens, J. H. M.;
Beijnen, J. H. J. Chromatogr., B 2007, 854, 245–259.
(14) Heudi, O.; Barteau, S.; Zimmer, D.; Schmidt, J.; Bill, K.; Lehmann, N.; Bauer,
C.; Kretz, O. Anal. Chem. 2008, 80, 4200–4207.
(15) Yang, Z.; Hayes, M.; Fang, X.; Daley, M. P.; Ettenberg, S.; Francis, L. S. T.
Anal. Chem. 2007, 79, 9294–9301.
(16) Motoyama, A.; Xu, T.; Ruse, C. I.; Wohlschlegel, J. A.; Yates, J. R. Anal.
Chem. 2007, 79, 3623–3634.
(17) Linke, T.; Ross, A. C.; Harrison, E. H. J. Chromatogr A. 2006, 1123, 160–
169.
6878 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
proteins from biological samples. Although many applications
using online immunoaffinity columns have been reported, 20-24
offline affinity purification using different carriers, such as
macroporous polymeric beads, agarose or sepharose beads, 25,26
and magnetic beads, 27-29 etc., for immobilization of a variety of
enzyme and/or capture antibodies enables more flexibility in
terms of selection of carriers, antibodies, assay formats, and
experimental conditions as compared to the online approach.
This report presents a sensitive and reproducible bioanalytical
approach for quantification of a PEGylated peptide in plasma. To
support human clinical studies for MK-2662, a 2D HPLC-MS/
MS method using a turbo ion spray (TIS) interface, monitoring
the surrogate N-terminal peptide (HAibDGTFTSDYSK) of MK-
2662 after tryptic digestion, has been developed and validated to
quantify MK-2662 in human plasma. A stable-isotope-labeled
internal standard, [ 13 C18, 15 N2] MK-2662, was employed. Two
sample preparation methods, PPT versus IAP using an anti-
PEG antibody, followed by trypsin digestion in a 96-well format
using 0.2 mL plasma sample, have been evaluated and applied
to clinical sample analysis. The potential caveats of the protein
precipitation approach and the benefit of immunoaffinity
purification in regard to assay specificity are discussed. To our
best knowledge, this is the first report describing use of anti-
PEG antibody to capture a PEGylated peptide in combination
with LC-MS/MS analysis. This approach is potentially applicable
to analysis of other PEGylated peptides or proteins.
EXPERIMENTAL SECTION
Chemicals and Reagents. MK-2662 and stable-isotope-labeled
internal standard (ISTD, Figure 1) were synthesized at the Merck
Research Laboratories, Merck & Co. (Rahway, NJ). HPLC grade
acetonitrile, HPLC grade methanol, laboratory grade formic acid
(88%), certified ammonium formate, and certified ammonium
bicarbonate were obtained from Fisher Scientific (Pittsburgh, PA).
Human control plasma (K2EDTA as anticoagulant) was purchased
from Biological Specialty Co. (Colmar, PA). Water was
purified by a Milli-Q ultrapure water system from Millipore
(Bedford, MA). Bovine serum albumin (BSA) at 22% in 0.85%
(18) Hagman, C.; Ricke, D.; Ewert, S.; Bek, S.; Falchetto, R.; Bitsch, F. Anal.
Chem. 2008, 80, 1290–1296.
(19) Dekker, L. J.; Bosman, J.; Burgers, P. C.; van Rijswijk, A.; Freije., R.; Luider,
T.; Bischoff, R. J. Chromatogr., B 2007, 847, 65–69.
(20) Radabaugh, M. R.; Nemirovskiy, O. V.; Misko, T. P.; Aggarwal, P.; Rodney
Mathews, W. Anal. Biochem. 2008, 380, 68–76.
(21) Li, W. W.; Nemirovskiy, O.; Fountain, S.; Rodney Mathews, W.; Szekely-
Klepser, G. Anal. Biochem. 2007, 369, 41–53.
(22) Berna, M.; Schmalz, C.; Duffin, K.; Mitchell, P.; Chambers, M.; Ackermann,
B. Anal. Biochem. 2006, 356, 235–243.
(23) Zhang, X.; Martens, D.; Kramer, P. M.; Kettrup, A. A.; Liang, X. J. Chromatogr.,
A 2006, 1133, 112–118.
(24) Edinboro, L. E.; Karnes, H. T. J. Chromatogr., A 2005, 1083, 127–132.
(25) Huang, L.; Harvie, G.; Feitelson, J. S.; Gramatikoff, K.; Herold, D. A.; Allen,
D. L.; Amunngama, R.; Hagler, R. A.; Pisano, M. R.; Zhang, W. W.; Fang,
X. Proteomics 2005, 5, 3314–3328.
(26) Hoofnagle., A. N.; Becker, J. O.; Wener, M. H.; Heinecke, J. W. Clin. Chem.
2008, 54, 1796–1804.
(27) Lee, Y. C.; Block, G.; Chen, H.; Folch-Puy, E.; Foronjy, R.; Jalili, R.;
Jendresen, C. B.; Kimura, M.; Kraft, E.; Lindemose, S.; Lu, J.; McLain, T.;
Nutt, L.; Ramon-Garcia, S.; Smith, J.; Spivak, A.; Wang, M. L.; Zanic, M.;
Lin, S. H. Protein Expression Purif. 2008, 62, 223–229.
(28) Dubois, M.; Becher, F.; Herbet, A.; Ezan, E. Rapid Commun. Mass Spectrom.
2007, 21, 352–358.
(29) Berna, M. J.; Zhen, Y.; Watson, D. E.; Hale, J. E.; Ackermann, B. L. Anal.
Chem. 2007, 79, 4199–4205.

NaCl, phosphate buffer saline (PBS), phosphate buffer saline
with 0.05% Tween-20 pH 7.4 (PBST), and Triton-X 100 were
purchased from Sigma-Aldrich (Milwaukee, WI). Linco DPP-
IV inhibitor was purchased from Millipore Corporation (St.
Charles, MO). BD P700 blood collection tubes that contain
proprietary DPP-IV inhibitors were purchased from BD Bioscience
(Franklin Lakes, NJ). Lyophilized sequence grade
modified trypsin (TRSEQZ at 1 mg/vial) was purchased from
Worthington Biochemical Corporation (Lakewood, NJ). Dynabeads
M-280-streptavidin was purchased from Invitrogen Corporation
(Carlsbad, CA). Biotinylated anti-PEG rabbit monoclonal
antibody, anti-PEG-biotin (or PEG-B-47-bio), was purchased
from Epitomics, Inc. (Burlingame, CA).
Instrumentation. Dynal MPC-L and MPC-96 (Invitrogen
Corporation, Carlsbad, CA) were used to capture magnetic beads
in the vial and 96-well plate, respectively. A TomTec Quadra 96
workstation, model 320 (TomTec Corporation, Hamden, CT) was
used to perform automated liquid transfer for protein precipitation.
A MaxQ mini 4450 benchtop incubating orbital shaker (Thermo
Scientific, Waltham, MA) and an SPE-Dry 96 micropate sample
concentrator (Jones Chromatography, Lakewood, CO) were used
for sample preparation. A Cohesive Aria 2300 system (Cohesive
Technologies Inc., Franklin, MA), which included two quaternary
Flux pumps, a valve module, and a CTC HTS autosampler, was
used for 2D HPLC. A Sciex API 5000 triple-quadrupole mass
spectrometer with a Sciex turbo ion spray interface (Sciex,
Toronto, Canada) was used as a detector. The data were collected
and processed through Analyst 1.4 software.
Preparation of Standards and Quality Control (QC)
Samples. MK-2662 stock solutions at about 40 µM were prepared
from two separate weighings and dissolved in acetonitrile/water
(30/70, v/v). One set of analyte stock solutions was used to
prepare calibration standards, and the other set was used to make
QC samples. Working standards, containing MK-2662 at the
different concentration levels, were prepared by serial dilutions
of analyte stock solution with 1% BSA, aliquoted to 1.5 mL
polypropylene microcentrifuge tubes and stored at -70 °C. A 500
nM ISTD working solution was obtained by dilution of an ∼40
µM ISTD stock solution in 1% BSA. Plasma calibration standards
were prepared daily by adding 40 µL of working standard and 40
µL of 500 nM IS into 200 µL of control plasma to provide final
concentrations of MK-2662 ranging from 1.0 to 1000 nM for the
PPT process and 2-200 nM for the IAP process.
The plasma QC samples were prepared in human control
plasma that contained Linco DPP-IV inhibitor (20 µL inhibitor
solution per milliliter of plasma) at 3, 50, 800, and 10 000 nM (for
testing dilution integrity) MK-2662 for the PPT assay and 3, 6,
50, and 150 nM for IAP. All QC aliquots were stored in a -70 °C
freezer.
Sample Preparation: Protein Precipitation Followed by
Trypsin Digestion. Clinical samples and QCs were thawed at
room temperature, mixed, and centrifuged at 4000 rpm (∼1300 g
RCF), at 10 °C for 10 min. A 200 µL aliquot of sample was then
mixed well with 40 µL of 1% BSA (to match the volume of
standards), 40 µL of 500 nM ISTD, and 500 µL of 0.1% formic
acid in acetonitrile/methanol (90/10, v/v) ina2mL96-well deepwell
plate. After centrifuging at 2000 RCF for 5 min, 600 µL of
supernatant was transferred into a clean 1.2 mL 96-well plate using
a TomTec automated liquid handling system and dried on an SPE
Dry-96 under a stream of nitrogen at 40 °C. After resuspending
the residue in with 140 µL of 50 mM ammonium bicarbonate (pH
8.0) containing 67 µg/mL trypsin, the samples were incubated at
37 °C for3hinaMaxQmini shaker to allow trypsin digestion,
and then the reaction was quenched by adding 10 µL of 44% formic
acid to each well at the end of incubation. After mixing and
centrifugation, 20 µL of sample was injected into the LC/LC-MS/
MS for analysis.
Sample Preparation: Immunoaffinity Capture Using Anti-
PEG Antibody Followed by On-Bead Trypsin Digestion.
DynaBeads M280-steptavidin (magnetic beads) were washed with
PBST (pH 7.4) three times and resuspended into the same volume
of PBST. Biotinylated anti-PEG rabbit monoclonal antibody, anti-
PEG-biotin (or PEG-B-47-bio), was added to the bead suspension
to reach the concentration of 20 µg/mL and then gently shaken
at room temperature for 1htoallow binding of biotinylated
antibody to the steptavidin-coated bead. The beads with antibody
were washed and resuspended with PBST three times. A 50 µL
final suspension was added into a sample mix that contained 200
µL of plasma sample, 40 µL of 1% BSA (to match the volume for
calibration standards), and 40 µL of 500 nM ISTD in a 96-well
plate. The sample mix was shaken at 700 rpm at 30 °C for2hto
allow antibody capture of MK-2662 from the plasma sample. After
washing the beads with PBS (pH 7.4) that contained 0.1% Triton-X
100 and then with ammonium bicarbonate (pH 8), an aliquot of
140 µL of67µg/mL trypsin in 50 mM ammonium bicarbonate
(pH 8) was added into each well, incubated at 37 °C with shaking
for about 3 h. The supernatant was transferred into a 96-well plate
with a 600 µL insert in each well, where 10 µL of 44% formic acid
was added to quench the reaction. After mixing and centrifugation,
20 µL of sample was injected on LC/LC-MS/MS for analysis.
Two-Dimensional HPLC Conditions. The 2D HPLC was
conducted with three basic steps: loading, eluting, and separating,
using a dual-column quick-elution mode on a Cohesive Aria
system. A cation-exchange column, Thermo BioBasic SCX (50 mm
× 2.1 mm, 5 µm), was used as the first dimension with a salt
gradient of mobile phases A and B [A, 50 mM ammonium formate
(pH 3) in 10% acetonitrile; B, 200 mM ammonium formate (pH 3)
in 10% acetonitrile]. A reversed-phase column, Thermo-Hypersil
Gold PFP (50 mm × 2.1 mm, 5 µm), was used as the second
dimension with an organic solvent gradient of mobile phases C
and D [C, 10 mM ammonium formate in 0.1% formic acid; D, 10
mM ammonium formate in 90% acetonitrile with 0.1% formic acid].
The LC method on the Cohesive system is listed in the Supporting
Information Table S1. The compartment of the autosampler was
set at 5 °C, and the reversed-phase column was controlled at 40
°C.
Mass Spectrometry Detection and Calculation. A PE Sciex
API 5000 triple-quadrupole mass spectrometer (MS/MS) with a
turbo ion spray interface ionization source operated in a positive
ion mode was used to quantitate MK-2662. The ion pairs (precursor
ion f product ion) m/z 672.1 f 223.1 for MK-2662 and m/z
676.9 f 223.1 for ISTD were selected for multiple reaction
monitoring (MRM). The instrument setting was adjusted to
maximize the response for the analyte and IS, respectively. The
turbo gas temperature was 650 °C. The flow settings of nebulizing
gas (nitrogen), collision gas (nitrogen), and curtain gas (nitrogen)
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6879

were 50, 6, and 30 L/min. The optimized declustering potential
(DP), collision energy (CE), collision cell exit potential (CXP),
and entrance potential (EP) were 140, 36, 15, and 10 V. The dwell
time was 100 ms for MK-2662 and ISTD. Both Q1 and Q3
quadrupoles were set at unit resolution. The total run time for
each injection was 4.5 min, while the MS data acquisition window
was started at 100 s after injection and kept open for 50 s. Peak
area ratios were calculated using Analyst software version 1.4. A
calibration curve was obtained by weighed (1/x 2 ) least-squares
linear regression of the peak area ratio of the analyte to the IS
versus the nominal concentration (x) of analyte.
Method Validation. The selectivity of the assay was confirmed
by processing control plasma from six different lots. Intraday
precision and accuracy were determined by analyzing six sets of
standard curve samples, each prepared in a different lot of control
plasma. Assay accuracy was calculated from a least-squares
regression curve constructed using all six replicate values at each
concentration, and the intraday precision (% CV) was calculated
from the peak area ratio of MK-2662 versus ISTD for each
concentration used to construct the standard curve. QC samples
were analyzed after first freezing and thawing, and the measured
concentrations were considered as the initial values. Freeze-thaw
stability was evaluated using QC samples that went through three
cycles of freezing and thawing, with at least one day of storage at
-70 °C between each thawing. Benchtop QC stability was tested
following 5hatroom temperature and comparing the measured
concentrations with their initial values. The stability of processed
samples in the autosampler was assessed by comparing the results
of QC samples analyzed at the end of a 14 h run with those
analyzed at the beginning of the run. The reinjection stability was
assessed by freezing the processed sample plate at -70 °C and
reinjected after 8 days of storage. In order to examine the dilution
integrity, five replicates of 10-fold of upper limit of quantification
(ULOQ) were diluted by 20-fold with the corresponding control
matrix during sample preparation and analyzed on LC/LC-MS/
MS.
RESULTS AND DISCUSSION
Mass Spectrometry Detection of Surrogate Peptide of MK-
2662. Tandem mass spectrometry (MS/MS) has been increasingly
used for peptide quantification in biological matrixes because
of its specificity. However, in the case of MK-2662, the PEGylated
peptide exhibited a complex mass spectrum obtained using turbo
ion spray (TIS), due to the large molecular weight (40 kDa) and
oligomeric dispersity of the conjugated PEG polymer. DePEGylation
of MK-2662 could be achieved by base hydrolysis and/or
enzymatic hydrolysis. For MK-2662, base hydrolysis liberated the
peptide via cleavage of the linker moiety. However, the MRM
sensitivity of the dePEGylated peptide was low due to weak
collision-induced dissociation (CID) of highly charged precursor
ions. Additionally, base hydrolysis could add a risk of chemical
degradation such as deamidation or hydrolysis to the peptide.
Because of these limitations of base hydrolysis, enzymatic hydrolysis
was selected to remove the PEG portion from MK-2662
prior to MS/MS analysis.
The N-terminus of MK-2662 is required for activity, and more
importantly, it contains the non-native amino acid (Aib) substitution
at the second amino acid in the peptide sequence, providing
mass selectivity against possible low levels of endogenous oxyn-
6880 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
tomodulin. Digestion of MK-2662 with trypsin produces a 12 amino
acid N-terminal peptide (N 1-12-mer), HAibDGTFTSDYSK, and
it was selected as a surrogate for quantification of MK-2662.
In comparison to the other tryptic cleavage sites in the MK-
2662 sequence, the single cleavage reaction at the 12th amino
acid (to generate N1-12-mer) demonstrated more rapid and
reproducible digestion. An internal standard was synthesized
by preparing a stable-isotope-labeled MK-2662 (PEGylated) that
contained 13 C/ 15 N-labeled phenylalanine at the sixth amino acid
in the sequence. The use of a chemically matched IS resulted
in improved assay precision by compensating for variation
during the sample preparation and detection process.
Precursor ions for the surrogate N1-12-mers of MK-2662 and
its IS were monitored as the doubly charged molecular weight
(MW) ions, [M + 2H] 2+ ,atm/z 672.1 (MW ) 1342) and 676.9
(MW ) 1352), respectively. After optimizing CID, CE, and DP
values, the product ion spectra of the N1-12-mers were collected,
showing singly charged b- and y-ions (Figure 2). The b2 ion at
223 m/z was selected based on its ion intensity and specificity,
where a non-natural amino acid (Aib) in b2 could allow MS
differentiation of MK-2662 from the endogenous OXM fragments.
Therefore, the MRM at m/z 672 f 223 (MK-2662) and
m/z 677 f 223 (ISTD) was used for quantification of MK-2662.
Two-Dimensional Chromatographic Conditions. One major
challenge for supporting clinical studies was the assay sensitivity
requirement. Our initial effort on developing a traditional onedimensional
(1D) reversed-phase HPLC-MS/MS assay allowed
us to successfully achieve a lower limit of quantification (LLOQ)
of 5 nM using protein precipitation followed by tryptic digestion
(Figure 3A). Since the clinical assay required a LLOQ of 1 nM,
addition assay development efforts were focused on reducing
background and increasing signal-to-noise ratio.
An attempt to use SPE to extract full length of MK-2662 from
plasma was not successful (very low recovery), mostly likely due
to the fact that the large PEG portion blocked or interfered with
the interaction of the peptide with the SPE sorbent. Other
approaches, such as optimizing the sample recovery and digestions
conditions or evaluating different CID transitions, were
unsuccessful toward obtaining an LLOQ below 5 nM, due to high
background signal in the LC-MS/MS chromatogram.
Multidimensional protein identification technology
(MudPIT) 30-33 has been widely used in the proteomics area.
Applying the same concept, a 2D LC-MS/MS was developed to
separate a surrogate peptide of MK-2662 (after trypsin digestion)
from the background using two different separating chromatographic
mechanisms. The 2D HPLC was performed on a Cohesive
Flux 2300 under quick elution mode with built-in switching valves.
A strong cation-exchange column, BioBasic SCX (50 mm × 2.1
mm, 5 µm), was used as the first dimension. An ∼60 s salt gradient
from 50 to 200 mM ammonium formate was used at a flow rate of
0.5 mL/mL to maintain a net positive charge on the N-terminal
tryptic peptide and separate N1-12-mer from other charged
components in the digest mix. A small percentage of organic
(30) Lohrig, K.; Wolters, D. Methods Mol. Biol. 2009, 564, 143–153.
(31) Florens, L.; Washburn, M. P. Methods Mol. Biol. 2006, 328, 159–175.
(32) Cagney, G.; Park, S.; Chung, C.; Tong, B.; O’Dushlaine, C.; Shields, D. S.;
Emili, A. J. Proteome Res. 2005, 4, 1757–1767.
(33) Krapfenbauer, K.; Fountoulakis, M. Methods Mol. Biol. 2009, 566, 165–
180.

Figure 2. Product ion mass spectra of tryptic peptides (A) HAibDGTFTSDYSK from MK-2662, [M + 2H] 2+ , and (B) HAibDGTF[ 13 C9, 15 N]TSDYSK
from [ 13 C18, 15 N2] MK-2662 (ISTD), [M + 2H] 2+ .
solvent (10% acetonitrile, v/v) was necessary to maintain good
chromatographic peak shape and high recovery from the SCX
column. Around the retention time of the N1-12-mer on the SCX
column, a narrow peak window (about 0.3-0.4 min) was
transferred from the SCX column to a reversed-phase (RP)
columnsthe second dimensionsfollowed by continued organic
gradient mobile phase to further separate the surrogate peptide
from background noise, and therefore, to provide a clean
chromatogram (Figure 3B). During method development, several
reversed-phase columns were tested for the second dimension,
and a Thermo Scientific Hyperil Gold, PFP column (50 mm × 2.1
mm, 5 µm) was found to provide the best retention. The analyte
focusing was achieved by introducing 10 mM ammonium formate
(pH 3) into 20% acetonitrile as a mobile phase. The chromatographic
conditions, including mobile phase selection, gradient,
and the timing for column switching (SCX to RP and switching
back), were optimized so as to provide a good separation within
a reasonable run time. Figure S1 (see the Supporting Information)
is a schematic of the column and switching valve arrangement
for online 2D LC, and Table S1 (see the Supporting Information)
details the chromatographic conditions including the gradients,
flow rates, and event timing. A RP column heater was set at 40
°C to produce a symmetric and well-resolved chromatographic
peak. An SCX guard column was used to protect the SCX column
which resulted in over 2000 injections per set of SCX/RP columns.
The acquisition window on the MS was set for 50 s so that the
MS ion source was kept clean and well-maintained. The overall
run time was 4.5 min per sample.
Pretreatment of Clinical Plasma Samples Using DPP-IV
Inhibitors as Stabilizer. Clinical samples were collected in blood
collection tubes (BD-P700) containing proprietary DPP-IV inhibitors,
however the composition and quantity of the enzyme inhibitor
was unknown. Since BD-supplied inhibitor can only be obtained
by purchasing tubes from the vendor, it presented a practicality
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6881

Figure 3. Representative chromatograms of the N1-12-mer [m/z 672 f 223] from MK-2662 plasma sample analyzed using (A) 1D LC-MS/MS
or (B) 2D LC-MS/MS. The left panel of part B is the chromatogram from the SCX column (the first dimension), where T18-30 and T19-30 were
the peptides obtained from trypsin digestion of MK-2662. The right panel of part B is from the RP column (the second dimension). The total run
time for 2D LC-MS/MS was 4.5 min per sample.
problem for preparing control plasma in the analysis laboratories.
Therefore, a test was conducted to determine equivalence of BD-
P700 inhibitor tubes to Linco brand DPP-IV inhibitor that was
available in easy to use solution form. The results demonstrated
that the stabilizers from BD tubes (1.5 mL of plasma per tube)
and Linco inhibitor (20 µL/mL plasma) provided comparable
effects in terms of stability (including freeze-thaw, benchtop,
autosampler, and reinjection stabilities), matrix effect, and recovery
in human plasma.
Optimization of Trypsin Digestion Conditions. A surrogate
proteolytic peptide approach using trypsin digestion was selected
for the MS/MS-based quantification of MK-2662 in human plasma.
Following protein precipitation, the supernatant was brought to
dryness and then reconstituted in NH 4HCO3 buffer that contained
trypsin for digestion. This step needed to be optimized to
ensure rapid, complete, and reproducible digest within a
reasonable time frame.
The digestion yield was tested by varying the amount of
trypsin from 2 to 10 µg, the volume of plasma in the BD-P700
tube, and different digestion time at 2, 3, 4, and 16 h in both
Linco DPP-IV-containing plasma and the BD-P700-treated
plasma, where the latter appeared to need more trypsin due to
the inhibitory effect of proprietary enzyme inhibitors. The
reason for testing plasma volume was that, at the clinical site,
although 3 mL of blood per sample was required in BD tube,
there was a chance that less volume was collected. Since the
BD tubes contain a fixed amount of protease inhibitor, the lower
the blood volume, the higher the inhibitor concentration in the
tube, which may lead to the more difficult trypsin digestion.
6882 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
The tested plasma volume at 1 and 1.5 mL covered a range of
2-3 mL of blood collection.
The test results are shown in Figure S2 (see the Supporting
Information), where the experiment was conducted at 10 nM MK-
2662 in all samples, and the yield was calculated assuming 100%
digestion in a neat MK-2662 solution incubated with trypsin for
3 h. The spiked plasma without inhibitor appeared to give higher
yield compared to the neat sample, presumably due to surface
absorption loss of the neat sample. The data indicated that a
digestion with 7 µg of trypsin was very sensitive to the change of
blood volumes1 mL of plasma (equivalent to 2 mL of blood
collection) required much longer digestion time to achieve >80%
digestion. In contrast, a digestion with 10 µg of trypsin demonstrated
an improved tolerability, yielding nearly complete digestion
within 3-4 h for both 1 and 1.5 mL of plasma (equivalent to 2-3
mL of blood collected in the BD-P700 tube). Balancing the benefit
and risk by considering digestion yield and sample processing
time, the digestion conditions10 µg of trypsin for 3 hswas
selected. The digestion outcome under these conditions was
robust and highly reproducible. Consequently, the clinical site was
instructed to collect 3 mL of blood (1.5 mL plasma equivalent)
and to record significant deviation from this volume.
Solvent for Working Standard Solutions. It is well-known
that peptides are prone to absorption loss to container surfaces. 9
Initially, 30% acetonitrile was used for preparation of both primary
stock and working standard solutions. Following storage at 4 °C
for 3 weeks, the primary stock was stable; however, more than
30% of MK-2662 loss was observed in working stock solutions.
Switching the working stock solvent to 1% BSA and storing at

-70 °C resolved the problem, giving

Figure 4. Representative chromatograms of MK-2662 in human plasma using LC/LC-MS/MS analysis: (A) single blank from protein purification,
(B) LLOQ at 1 nM from protein purification, (C) single blank from immunoaffinity purification, and (D) LLOQ at 2 nM from immunoaffinity purification
(left panel, MK-2662; right panel, ISTD).
samples to the peak areas of neat standards prepared in the same
solvent and injected directly. The results (Table S2 in the
Supporting Information) show that recoveries ranged from 100.07%
to 119.36%, and matrix effects ranged between 91.32% and 97.2%
across the tested concentrations. On the basis of the intraday
precision results obtained using six different lots of control plasma
(Table 1), an absolute matrix effect should not have any impact
on assay precision.
Validation of the IAP 2D LC-MS/MS Method. The accuracy
and precision of the IAP 2D LC-MS/MS was accessed
and compared with the PPT 2D LC-MS/MS method. The IAP
6884 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
intraday validation data obtained from three standard curves
constructed in three lots of human plasma over the range of 2-200
nM are presented in Table 1, and the QC precision and accuracy
data from three sets of QCs at each of 3, 6, 50, and 150 nM MK-
2662 are shown in Table 2. The chromatograms showed that there
was no interference in the blank and an adequate single-to-noise
ratio for the analyte peak at the LLOQ (Figure 4, parts C and D).
Both methods demonstrated good precision and accuracy, and
met the acceptance criteria for handling small molecules, specified
in FDA guidelines. 34 In comparison to the PPT assay, the IAP
method has a 2-fold higher LLOQ. This was due to differences in

Table 1. Intraday Precision and Accuracy for the
Determination of MK-2662 in Different Lots of Human
Plasma
method
nominal
concn
(nM)
mean
measured
concn (nM)
(n ) 6) a
accuracy
(%) b
precision
(%) c
PPT (n ) 6) 1.00 1.01 101.00 2.97
2.00 1.98 99.00 3.03
10.00 9.93 99.30 2.92
50.00 49.86 99.72 3.43
200.00 201.30 100.65 2.93
500.00 503.99 100.80 2.92
800.00 801.50 100.19 3.05
1000.00 998.47 99.85 2.93
IAP (n ) 3) 2.00 1.97 98.6 9.76
5.00 5.29 105.8 2.56
10.00 9.48 94.8 3.44
50.00 51.22 102.4 3.17
100.00 96.73 96.7 4.24
160.00 162.42 101.5 4.12
200.00 200.14 100.1 1.98
a Back-calculated by interpolation from the calibration curve, constructed
by weighted (1/x 2 ) linear regression of peak area ratios
(analyte over ISTD) vs concentrations of the calibration standards in
plasma. b Expressed as [((mean observed concentration)/(nominal
concentration)) × 100] (%). c Coefficient of variation of back-calculated
concentration.
the analytical recovery of the two methods, determined to be
greater than 90% for the PPT method and approximately 30% for
the IAP method (data not shown). The low recovery of IAP might
be the result of relatively low binding affinity of anti-PEG antibody,
which consequently leads to incomplete analyte capture during
the binding step and partial analyte loss during the washing steps.
However, since the stable-isotope-labeled PEGylated MK-2662 was
used as the IS, the low recovery of IAP, as expected, did not have
significant impact on the assay precision and accuracy.
Application of PPT 2D LC-MS/MS and IAP 2D LC-MS/
MS to Clinical Sample Analysis. The protein precipitation
followed by 2D LC-MS/MS was initially applied to analysis of
the clinical plasma samples to support the first in-human study
after subcutaneous (sc) administration of MK-2662. The interday
QC data collected during seven daily sample analysis runs showed
accuracy ranging from 104.72% to 106.33% with

Table 3. MK-2662 QC (with DPP-IV Inhibitor) Stability in Human Plasma
nominal concn (nM) 3 F/T [% CV] a
room temp for 5h[%CV] a
PEG antibody, it provides an alternative approach for PEGylated
peptide analysis.
CONCLUSIONS
PPT 2D LC-MS/MS- and IAP 2D LC-MS/MS-based methods
for the quantitative analysis of a PEGylated peptide, MK-2662, in
human plasma were developed and validated in support of clinical
studies. Challenges associated with surrogate peptide selection,
assay sensitivity, selectivity, and ruggedness were addressed.
Protein precipitation represented a simple sample preparation
method. With the use of 2D HPLC-MS/MS, the background
noise from PPT extract was significantly reduced, leading to
increased assay sensitivity. The trypsin digestion conditions were
optimized to overcome enzyme inhibition due to protease inhibitors
used to stabilize clinical samples and, therefore, improved
(35) Alley, S. C.; Benjamin, D. R.; Jeffery, S. C.; Okeley, N. M.; Meyer, D. L.;
Sanderson, R. J.; Senter, P. D. Bioconjugate Chem. 2008, 19, 759–765.
assay ruggedness. To address assay specificity, an IAP method
was developed to selectively capture the PEGylated peptide from
human plasma samples using anti-PEG antibody followed by
surrogate peptide analysis. Comparison of two sample preparation
methods (PPT and IAP) applied to clinical sample analysis showed
that protein precipitation may overestimate the PEGylated peptide
concentrations. This is consistent with the concern that the PPT
2D LC-MS/MS method may detect intact surrogate peptide not
only originating from the PEGylated but also from dePEGylated
or other truncated metabolite species presented in the clinical
samples.
It is worth mentioning that the immunoaffinity purification
coupled with MS/MS is a special assay formatssimilar to ELISA,
the peptide is captured using a capture antibody; different from
ELISA, the peptide-specific measurement uses tandem mass
spectrometry rather than an antibody-based detection. LC-MS/
MS detection combined with IAP not only provides high specificity
but also overcomes a rate-limiting step of ELISAsproduction of
peptide-specific antibody. Potentially, this new assay format could
be generally applicable to other PEGylated peptides.
ACKNOWLEDGMENT
The authors gratefully acknowledge Maria Bednarek, Andrew
Zhang, and Matt Braun for providing the stable-isotope-labeled
internal standard to make this work possible. In addition, we
express our gratitude to Tae Han, Vijay B. G. Reddy, and Carmen
Fernandez-Metzler for project-related discussions and to Albert
S. Yuan, Kuo-Chang Yin, and Thorsten Verch for antibody-related
discussions.
NOTE ADDED AFTER ASAP PUBLICATION
This paper was published on July 16, 2010 with a minor text
error. The revised version was published on July 21, 2010.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review April 14, 2010. Accepted June 27,
2010.
AC1009832
autosampler for 14 h [% CV] a
reinjection [% CV] a
3 105.86 [1.17] 103.26 [2.55] 101.30 [1.53] 98.70 [1.82]
50 103.20 [0.49] 101.71 [1.20] 99.75 [0.76] 97.00 [0.97]
800 102.46 [0.29] 100.99 [0.31] 98.74 [1.21] 97.25 [0.53]
a [((mean concentration after storage)/(initial mean concentration)) × 100] (%).
Figure 5. Mean plasma concentration-time profiles following a
single 6 mg subcutaneous administration of MK-2662 to healthy young
men: (b) the measured mean peptide concentrations from individual
samples (n ) 6) analyzed using protein precipitation (PPT) coupled
with LC/LC-MS/MS; (9) the measured concentration from the pooled
samples obtained from the same individual subjects using PPT LC/
LC-MS/MS; (1) the measured concentration from the pooled
samples obtained from the same individuals using immunoaffinity
purification (IAP) coupled with LC/LC-MS/MS.
6886 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010

Anal. Chem. 2010, 82, 6887–6894
Ferrocene Bound Poly(vinyl chloride) as Ion to
Electron Transducer in Electrochemical Ion
Sensors
Marcin Pawlak, Ewa Grygolowicz-Pawlak, and Eric Bakker* ,†
Nanochemistry Research Institute, Department of Chemistry, Curtin University of Technology, Perth,
Western Australia 6845, Australia
We report here on the synthesis of poly(vinyl chloride)
(PVC) covalently modified with ferrocene groups (FcPVC)
and the electrochemical behavior of the resulting polymeric
membranes in view of designing all solid state
voltammetric ion sensors. The Huisgen cycloaddition
(“click chemistry”) was found to be a simple and efficient
method for ferrocene attachment. A degree of PVC modification
with ferrocene groups between 1.9 and 6.1 mol
% was achieved. The chemical modification of the PVC
backbone does not significantly affect the ion-selective
properties (selectivity, mobility, and solvent casting ability)
of potentiometric sensing membranes applying this
polymer. Importantly, the presence of such ferrocene
groups may eliminate the need for an additional redoxactive
layer between the membrane and the inner electric
contact in all solid state sensor designs. Electrochemical
doping of this system was studied in a symmetrical
sandwich configuration: glassy carbon electrode |FcPVC|
glassy carbon electrode. Prior electrochemical doping
from aqueous solution, resulting in a partial oxidation of
the ferrocene groups, was confirmed to be necessary for
the sandwich configuration to pass current effectively. The
results suggest that only ∼2.3 mol % of the ferrocene
groups are electrochemically accessible, likely due to
surface confined electrochemical behavior in the polymer.
Indeed, cyclic voltammetry of aqueous hexacyanoferrate
(III) remains featureless at cathodic potentials (down to
-0.5 V). This indicates that the modified membrane is
not responsive to redox-active species in the sample
solution, making it possible to apply this polymer as a
traditional, single membrane. Yet, the redox capacity of
the electrode modified with this type of membrane was
more than 520 µC considering a 20 mm 2 active electrode
area, which appears to be sufficient for numerous
practical ion voltammetric applications. The electrode
was observed to operate reproducibly, with 1% standard
deviation, when applying pulsed amperometric
techniques.
Poly(vinyl chloride) is a polymeric matrix commonly used for
electrochemical electrode preparation, and is popular in classical
* To whom correspondence should be addressed.
† Current address: Department of Inorganic, Analytical and Applied Chemistry,
University of Geneva, CH-1211 Geneva, Switzerland.
aqueous inner contact 1 and solid state electrode (SSE) configurations
2-10 because of its attractive mechanical properties and
compatibility with electro-active components. With SSEs, an
effective ion-to-electron transduction is necessary, 2,3 which is
usually applied as a separate layer between the ion-selective
membrane (ISM) and the inner metallic contact 4-8 or as an
additive to the membrane. 9 The most popular redox-active materials
used for this purpose are conductive polymers. 4-6,9 Many types
of such polymers have been successfully applied in potentiometric
electrodes, e.g., poly(pyrrole) (PPy), 4,5,11 poly(octyl thiophene)
(POT), 9,10,12 poly(3,4-ethylenedioxythiophene) (PEDOT), 6,13,14 and
poly(aniline) (PANI), 9 and the suppression of a water layer
between the conducting polymer and underlying electrode was
found to be a key characteristic to guarantee stable potentiometric
behavior. 15-17 In recent years, ion-selective electrodes interrogated
by voltammetric techniques have increased in importance. 18 This
class of sensors borrows materials and techniques from the fields
of ion-selective electrodes and the electrochemistry at the interface
(1) Bakker, E.; Buhlmann, P.; Pretsch, E. Chem. Rev. 1997, 97, 3083–3132.
(2) Cattrall, R. W.; Drew, D. M.; Hamilton, I. C. Anal. Chim. Acta 1975, 76,
269–277.
(3) Hulanicki, A.; Trojanowicz, M. Anal. Chim. Acta 1976, 87, 411–417.
(4) Cadogan, A.; Gao, Z.; Lewenstam, A.; Ivaska, A. Anal. Chem. 1992, 64,
2496–2501.
(5) Michalska, A.; Hulanicki, A.; Lewenstam, A. Microchem. J. 1997, 57, 59–
64.
(6) Vazquez, M.; Bobacka, J.; Ivaska, A.; Lewenstam, A. Sens. Actuators, B
2002, 82, 7–13.
(7) Cosofret, V. V.; Erdosy, M.; Johnson, T. A.; Buck, R. P. Anal. Chem. 1995,
67, 1647–1653.
(8) Fibbioli, M.; Badyopadhyay, K.; Liu, S. G.; Echegoyen, L.; Enger, O.;
Diederich, F.; Buhlmann, P.; Pretsch, E. Chem. Commun. 2000, 5, 339–
340.
(9) Bobacka, J.; Lindfors, T.; McCarrick, M.; Ivaska, A.; Lewenstam, A. Anal.
Chem. 1995, 67, 3819–3823.
(10) Song, F.; Ha, J.; Park, B.; Kwak, T. H.; Kim, I. T.; Nam, H.; Cha, G. S.
Talanta 2002, 57, 263–270.
(11) Sutter, J.; Lindner, E.; Gyurcsanyi, R. E.; Pretsch, E. Anal. Bioanal. Chem.
2004, 380, 7–14.
(12) Sutter, J.; Radu, A.; Peper, S.; Bakker, E.; Pretsch, E. Anal. Chim. Acta
2004, 523, 53–59.
(13) Bobacka, J. Anal. Chem. 1999, 71, 4932–4937.
(14) Vazquez, M.; Danielson, P.; Bobacka, J.; Lewenstam, A.; Ivaska, A. Sens.
Actuators, B 2004, 97, 182–189.
(15) Fibbioli, M.; Morf, W. E.; Badertscher, M.; de Rooij, N. F.; Pretsch, E.
Electroanalysis 2000, 12, 1286–1292.
(16) Gyurcsanyi, R. E.; Rangisetty, N.; Clifton, S.; Pendley, B. D.; Lindner, E.
Talanta 2004, 63, 89–99.
(17) De Marco, R.; Veder, J.-P.; Clarke, G.; Nelson, A.; Prince, K.; Pretsch, E.;
Bakker, E. Phys. Chem. Chem. Phys. 2008, 10, 73–76.
(18) Samec, Z.; Samcova, E.; Girault, H. H. Talanta 2004, 63, 21–32.
10.1021/ac1010662 © 2010 American Chemical Society 6887
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/28/2010

of two immiscible electrolyte solutions (ITIES) and has allowed
researchers to design reversibly operating polyion sensors, 19-21
flash chronopotentiometric electrodes for rapid determination of
kinetically labile components, 22 ion channel mimetic systems, 23
thin layer coulometric sensors, 24 thin layer electrodes for stripping
voltammetry, 25 as well as controlled current coulometric sensors. 26
The requirements for the realization of all solid state voltammetric
ion sensors are more difficult to satisfy than the potentiometric
sensors mentioned above, since the required redox capacity
must be sufficiently high to sustain the ion transfer current.
Shvarev showed that PEDOT is a promising conducting polymer
underlayer for this purpose, 27 Amemiya 25,28 and our group 29
studied conducting polymers for thin layer voltammetric applications.
With these materials, the redox properties are typically
conducive to anion extraction from the sample solution. To allow
for cation extraction, the polymer may sometimes be doped to
obtain its stable oxidized form. 29
An alternative strategy was suggested by Samec et al. 30 by the
incorporation of lipophilic ferrocene derivatives. This appears to
be an attractive alternative to conducting polymer based systems,
owing to the well-defined reversible redox couple and a potentially
high redox capacity of the resulting membranes. Freely dissolving
such redox active species may, however, result in eventual
leaching from the membrane (especially in its oxidized, positively
charged form) and in interferences from redox active species in
the sample.
To address these limitations and to introduce an alternative
material, the chemical attachment of ferrocene groups to polymer
chains is explored here. From a chemical point of view, poly(vinyl
chloride) is easily modifiable due to the susceptibility for nucleophilic
substitution of the chlorine atoms. 31,32 The most commonly
used method for introducing new functional groups into the PVC
chain is the reaction with sodium azide. 33,34 This modification was
used for example in the cross-linking of the polymer. 35 Recently,
the Huisgen cycloaddition, known also as “click chemistry”, has
become a very popular method of transforming azide groups. 36
This type of reaction is attractive because of its high selectivity,
mild reaction conditions, very high yields, and lack of byproducts.
It was successfully applied for modification of a wide range of
(19) Shvarev, A.; Bakker, E. J. Am. Chem. Soc. 2003, 125, 11192–11193.
(20) Samec, Z.; Trojanek, A.; Langmaier, J.; Samcova, E. Electrochem. Commun.
2003, 5, 867–870.
(21) Yuan, Y.; Amemiya, S. Anal. Chem. 2004, 76, 6877–6886.
(22) Gemene, K. L.; Bakker, E. Anal. Chem. 2008, 80, 3743–3750.
(23) Xu, Y.; Bakker, E. Langmuir 2009, 26, 568–573.
(24) Yoshizumi, A.; Uehara, A.; Kasuno, M.; Kitatsuhi, Y.; Yoshida, Z.; Kihara,
S. J. Electroanal. Chem. 2005, 581, 275.
(25) Kim, Y.; Amemiya, S. Anal. Chem. 2008, 80, 6056–6065.
(26) Bhakthavatsalam, V.; Shvarev, A.; Bakker, E. Analyst 2006, 131, 895–900.
(27) Perera, H.; Fordyce, K.; Shvarev, A. Anal. Chem. 2007, 79, 4564–4573.
(28) Guo, J.; Amemiya, S. Anal. Chem. 2006, 78, 6893–6902.
(29) Si, P.; Bakker, E. Chem. Commun. 2009, 35, 5260–5262.
(30) Langmaier, J.; Olsak, J.; Samcova, E.; Samec, Z.; Trojanek, A. Electroanalysis
2006, 18, 1329–1338.
(31) Kameda, T.; Ono, M.; Grause, G.; Mizoguchi, T.; Yoshioka, T. Polym.
Degrad. Stab. 2009, 94, 107–112.
(32) Kameda, T.; Fukuda, Y.; Grause, G.; Yoshioka, T. J. Appl. Polym. Sci. 2010,
116, 36–44.
(33) Sacristan, J.; Mijangos, C.; Reinecke, H.; Spells, S.; Yarwood, J. Macromolecules
2000, 33, 6134–6139.
(34) Martinez, G. J. Polym. Sci., Part A: Polym. Chem. 2006, 44, 2476–2486.
(35) Jayakrishnan, A.; Sunny, M. C. Polymer 1996, 37, 5213–5218.
(36) Rostovtsev, V. V.; Green, L. G.; Fokin, V. V.; Sharpless, K. B. Angew. Chem.,
Int. Ed. 2002, 41, 2596–2599.
6888 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
different polymers, 37 but it has not yet become popular for the
modification of PVC. Considering that the modified polymer is
intended to be applied for electrochemical electrode preparation,
any redox activity is expected to be related to the ferrocene groups
only. The redox active properties of a triazol linker obtained by
applying the Huisgen cycloaddition has been studied by Kirrs and
co-workers, 38 revealing that it is indeed electrochemically inert.
In this paper we present studies on properties of membranes
containing poly(vinyl chloride) modified with ferrocene groups
via the click-chemistry reaction.
EXPERIMENTAL SECTION
General Methods and Materials. High molecular weight
poly(vinyl chloride) (PVC), sodium azide, ethynylferrocene, Lascorbic
acid, copper sulfate pentahydrate, tetradodecylammonium
tetrakis(4-chlorophenyl)borate (ETH 500), bis(2-ethylhexyl) sebacate
(DOS), potassium hexacyanoferrate (III), potassium chloride,
sodium perchlorate, sodium nitrate, sodium thiocyanate,
anhydrous tetrahydrofuran (THF), and dimethylformamide (DMF)
were purchased from Sigma Aldrich and used without further
purification. Aqueous solutions were prepared by dissolving the
appropriate salts in Milli-Q-purified distilled water.
IR spectra were taken using a Perkin-Elmer Spectrum 100 FT-
IR spectrometer with attenuated total reflectance (ATR) sampling
accessory and ATR corrected. UV-vis spectra were recorded in
the 900-350 nm wavelength range on a GBC UV-vis 916
spectrophotometer using 1 cm thick quartz cuvettes. The UV-vis
calibration curve was obtained for three samples containing 2, 6,
and 12 µmol/mL of ferrocene dissolved in a matrix containing 33
mg of PVC per mL of THF. The first order derivative of the
spectrum was used to determine the content of ferrocene attached
to the PVC. Glassy carbon rods of 5 mm diameter were purchased
from SPI Supplies/Structure Probe, Inc. (West Chester, PA) and
assembled into an epoxy resin body (EpoFix Kit, Struers Australia,
Milton QLD, Australia).
Synthesis. The substitution reaction of PVC with NaN 3 was
carried out using 1 g (16 mmol based on monomeric unit) of
PVC and 1.04 g (16 mmol) of NaN3 in 42 mL of DMF-water
mixture (5:1 volume ratio) and heated at 50 °C under nitrogen
for various times from 24 to 168 h. When the desired degree
of modification was achieved, the reaction mixture was cooled
to room temperature, the product was filtered, washed with
distilled water and methanol, and subsequently dried under
reduced pressure. In the click-chemistry reaction between
azide-modified PVC and ethynylferrocene, 300 mg of N3PVC,
60 mg (0.29 mmol) of ethynylferrocene, 140 mg (0.56 mmol)
of CuSO4 · 5H2O, and 500 mg (2.84 mmol) of ascorbic acid were
added to 28 mL of a THF-water mixture (6:1 volume ratio).
After 4hofreaction time, the product was precipitated with
water. Ferrocene modified PVC was filtered, washed thoroughly
with methanol, and then dissolved in 20 mL of THF.
The THF solution was then filtered again to remove insoluble
impurities, and the product was precipitated with methanol,
filtered, and dried under reduced pressure.
Composition of Membrane Cocktails. Membrane cocktail
was prepared by dissolving 7.5 mg of FcPVC, 37.5 mg of DOS,
(37) Lutz, J.-F. Angew. Chem., Int. Ed. 2007, 46, 1018–1025.
(38) Verschoor-Kirss, M.; Kreisz, J.; Feighery, W.; Reiff, W. M.; Frommen, C. M.;
Kirss, R. U. Organomet. Chem. 2009, 694, 3262–3269.

Figure 1. Infrared spectra of (A) unmodified PVC, (B) azide-modified PVC, and (C) ferrocene-modified PVC. Insets: structures of azide modified
PVC and FcPVC.
and 5 mg of ETH 500 in 1 mL of THF. DOS is a plasticizer
commonly used as a solvent in PVC-based (liquid) membranes.
The lipophilic electrolyte ETH 500 is added to provide counterions
for the extracted analyte in the membrane and is also used to
keep the electrical resistance of the membrane low. Electrodes
were prepared by drop casting 40 µL of cocktail and letting the
THF evaporate overnight under ambient conditions.
Electrochemical Experiments. All cyclic voltammetry experiments
were performed at a 50 mV s -1 scan rate. Normal pulse
voltammograms were performed in a -0.6 to +0.8 V potential
range using 10 mM solutions of NaCl, NaSCN, NaNO3, and
NaClO4. Potential pulses (1 s) with a 20 mV potential increase
were followed by a5sregeneration pulse (baseline potential)
at open circuit potential (as measured before each voltammetric
experiment). Membrane doping was performed with a single
pulse of 10 min duration in a potential range of 400-800 mV
in 10 mM NaClO4. A time of 10 min was chosen experimentally
as the shortest time allowing the observed current to sufficiently
decay. In the sandwiched electrode experiment, a three
pulse procedure was applied. The first 30 s pulse at 0 mV was
to confirm the steady state of the system before a 300 s long
pulse at 200 mV was applied to induce electrochemical
processes, while the last, carried out for 300 s at 0 mV, was to
allow the system to return to its original state.
Cyclic voltammetry in the presence of hexacyanoferrate(III)
ion was performed in 1 mM K3Fe(CN)6 solution with 0.1 M KCl
as a background, and the potential was scanned between a 0.8
and -0.5 V potential range starting from 0.4 V and going to
negative potentials first. The stability of the signal generated
by the FcPVC modified electrode was examined by application
ofa1spulse at 0.5 V and followed by 20 s regeneration at the
applied open circuit potential, using a three electrode configuration
and 10 mM sodium perchlorate. The procedure was
repeated 40 times. The redox capacity of the FcPVC was tested
in the same solution by applying a 0.8 V potential for 30, 60,
120, 300, and 600 s, each time followed by a discharge pulse
at 0 V for 200 s.
Table 1. Reaction Time of Azidation and Resulting
Degree of Ferrocene Modification of PVC, Estimated
by Spectrophotometry
reaction time [h] degree of modification [mol %]
24 1.9
60 3.2
120 5.3
168 6.1
RESULTS AND DISCUSSION
In an initial step, PVC was modified with azide groups using
NaN3 according to Sacristan et al. 33 The efficiency of the
reaction was confirmed by comparing the IR spectrum for
commercial, not modified PVC, and the same polymer after
modification with azide groups. As shown in Figure 1A,B, the
distinctive azide band at 2100 cm -1 , not visible for unmodified
PVC, gives a well developed peak after reaction, confirming
the effective partial modification of the polymeric backbone
with azide groups.
Further modification of azide-modified PVC with ferrocene
groups required the use of a large excess of catalyst (200 mol %
relative to the ethynylferrocene) to allow a high yield of reaction.
As seen in Figure 1C, the IR spectrum for the reaction product
did not reveal any signal from azide groups, indicating full
conversion of the substrate. Since the ferrocene groups absorb
strongly at 440 nm, the content of the ferrocene groups in the
obtained FcPVC could be conveniently estimated from a calibration
curve prepared for mixtures of nonmodified PVC with various
ethynylferrocene contents dissolved in THF. Since the baselines
for a THF solution of the ferrocene-modified PVC were significantly
different, the ferrocene content was calculated based on
spectral derivatives. On the basis of the spectrophotometric results
and considering complete conversion of azide to ferrocene in the
polymer backbone (as indicated by FT-IR) it may be concluded
that the content of the ferrocene groups in PVC can be easily
controlled by adjusting the time of azidation reaction. As shown
in Table 1 for the polymer that has been modified with azide
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6889

Figure 2. Chronoamperograms obtained for a pair of sandwiched electrodes modified with a FcPVC-DOS membrane containing 10 wt % of
ETH 500, anodically doped with perchlorate from a 10 mM NaClO4 solution at the indicated potentials. Inset: charge calculated from the
chronoamperograms as a function of the doping potential.
groups in the first step reaction for only 24 h, the content of
ferrocene groups attached during the second reaction step was
1.9 mol %. The degree of modification significantly increased to
3.2 mol % when the reaction time for the first step was increased
to 60 h. After 120 h of the azidation, the reaction rate started to
slow down while after 168 h the efficiency reached 6.1 mol % and
its further increase was negligible.
Unfortunately, with higher conversion rate the solubility of the
modified PVC in THF drops significantly. For example, the highest
concentration that could be obtained for a membrane cocktail
containing FcPVC with 3.2 mol % of ferrocene groups is 50 mg/
mL of THF (which is half of the typical cocktail concentration
when using unmodified PVC), while only 17 mg of the cocktail
containing polymer with 6 mol % of ferrocene groups could be
dissolved in the same volume. This property is most likely related
to the presence of aromatic linker between the ferrocene group
and polymer backbone, which may result in a loose cross-linking
due to π-stacking. Therefore, for practical reasons FcPVC containing
3.2 mol % of ferrocene (461 mmol/kg) was chosen for further
studies as a compromise between electrochemical activity and
solubility.
In initial studies, the redox properties of FcPVC in the ionselective
membrane were studied independently of an aqueous
electrolyte. The redox process at the inner side of the membrane
is otherwise difficult to distinguish from the outer ion transfer
process, which may be rate limiting owing to the low mobility of
ion-selective membrane materials. 39,40 Removal of the aqueous
electrolyte was achieved here in a fully symmetrical cell, sandwiching
the ion-selective ferrocene functionalized membrane
between two glassy carbon (GC) electrodes. Note that a fully
reduced ferrocene based membrane is not expected to pass
current effectively, since any oxidation at one electrode must be
accompanied by a concurrent reduction at the second electrode.
Ideally, therefore, the membrane must be composed of a reduced
(39) Long, R.; Bakker, E. Electroanalysis 2003, 15, 1261–1269.
(40) Long, R.; Bakker, E. Anal. Chim. Acta 2004, 511, 91–95.
6890 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
and oxidized form in a one to one molar ratio for the following
reaction to occur efficiently:
GC anode| Fc0 Fc +| Fc+ Fc 0| GC cathode
This was confirmed here by an electrochemical doping step from
sodium perchlorate solution prior to assembling the sandwich
configuration. Two identical glassy carbon electrodes modified
with a FcPVC/DOS/ETH 500 membrane were placed in 10 mM
NaClO4 solution and held at 400 mV for 600 s to partially oxidize
ferrocene groups by incorporation of perchlorate ions into the
membrane. Immediately after the doping process, the electrodes
were sandwiched together. A potential of 200 mV was
applied between the two electrodes for 300 s and the current
recorded (see Figure 2). The observed current very soon decayed
to near zero values, indicating a low ferrocenium concentration
in the membrane. The total number of coulombs calculated from
the integral of the chronoamperogram was 16.45 µC. After
electrode separation, they were again immersed into the doping
solution and held at 500 mV for the same time period. The
sandwich procedure was repeated and higher currents were
observed, resulting in a charge of 102.88 µC. This experimental
sequence was repeated for increasing doping potentials. A plot of
calculated charge as a function of doping potential is presented
in the inset of Figure 2. Doping potentials higher than 600 mV
gave a decreasing charge, consistent with ferrocene as a limiting
reagent in reaction 1 above. Note that more positive potentials
may also result in the oxidation of the tetraphenylborate anion, 28
a component of the lipophilic electrolyte, and hence may explain
the observed distortion of the expected bell-shaped curve (number
of coulombs vs doping potential) at larger potentials.
From the maximum charge found for the optimally doped
electrodes it can be calculated that the amount of ferrocenium/
ferrocene groups participating in the process was 3.16 nmol per
20 mm 2 of electrode area, or about 10 mmol kg -1 (of which
(1)

Figure 3. Cyclic voltammograms obtained for a pair of sandwiched electrodes modified with FcPVC membrane containing 461 mmol/kg of
ferrocene groups and 10 wt % of ETH 500 anodically doped from 10 mM NaClO4 solution at 600 mV carried out for the indicated times after the
doping.
half can be oxidized and half reduced). This is just about 2.3
mol % of what was calculated based on the spectrophotometric
calibration curve. Considering that the mobility of ferrocene/
ferrocenium groups attached to a long chain of the polymer is
limited, only the molecules being at a close distance from the
glassy carbon electrode may participate in the electrochemical
processes within the experimental time frame. This, however,
may be an advantage considering the method presented here,
since it suggests that the material is insensitive to redox active
species in the contacting aqueous solution (see below).
When cyclic voltammetry was performed on a freshly doped
sandwiched system, a redox couple was observed, with oxidation
and reduction peaks comparable to the theoretical shape obtained
for ferrocene dissolved in acetonitrile on a gold electrode. 41 The
peak heights for the oxidation and reduction processes were
almost identical indicating that the system is fully reversible. The
peaks were separated by 320 mV which is more than in the case
of ferrocene dissolved in acetonitrile (56 mV). This is not
unexpected, given the fact that ferrocene is attached to the PVC
backbone (and not diffusion limited as in the theoretical case)
and the second electrode is acting as both the counter and
reference electrode. However, for successive scans important
changes of the maximum and minimum current values were
observed. A series of 10 scans was repeated 40, 80, 120, and 160
min after doping and showed that the maximum peak currents
successively diminished with time (Figure 3). On the basis of our
previous observations that within the time frame of the doping
experiment (600 s) only about 2% of redox-active groups undergo
oxidation, the decrease of observed currents may be explained
by the gradual diffusion of oxidized ferrocene away from the
electrode surface into the surrounding bulk membrane. Functionalities
covalently attached to the PVC backbone are indeed
not strictly immobile, as evidenced by earlier work demonstrating
(41) Bard, A. J.; Faulkner, L. R. Electrochemical Methods; Wiley: New York, 2001.
equilibration of solvent cast plasticized PVC films with covalently
attached chromophore functionalities throughout the bulk of the
material. 42
Subsequent work focused on three electrode systems in order
to characterize the solid state membrane in contact with aqueous
electrolyte. The ion-selective properties of the new polymer were
studied in 10 mM solutions with normal pulse voltammetry, which
has been established as a robust and gentle method of ionselective
membrane characterization. 43 The results are presented
in Figure 4. The potentials at which the ion transfer of ClO4 - ,
SCN - ,NO3 - , and Cl - ions occur at a current of 1 µA were
found at about 420, 460, 580, and 700 mV, respectively. The
order and the magnitude of the potential separation are
consistent with the theoretical selectivity expected for lipophilicity-based
extraction according to the Hofmeister series. 43
This confirms that the ferrocene modification has a negligible
effect on the anion extraction properties of the polymer. When
scanning the electrode to a negative potential range, a cation
extraction wave should normally occur with aqueous inner
contact systems. 44 The inset of Figure 4 suggests for the sodium
chloride sample that the voltammogram is very quiet at cathodic
potentials, suggesting no cationic response. As mentioned above,
the ferrocene groups present in the polymeric phase are predominately
in their reduced state, suppressing cathodic response. It is
expected that doping the membrane with lipophilic cationexchanger
according to a procedure recently reported for poly-
(octyl thiophene) (POT) 29 will allow one to yield cation transfer
voltammetric characteristics, but this was not pursued here.
Ferrocene groups are randomly distributed in the membrane
phase and hence some of them are believed to be present on the
membrane surface. Exposure of the conductive groups to the
sample solution may cause undesired electrode sensitivity to
redox-active solutes. Therefore, the studied electrodes were tested
(42) Rosatzin, T.; Holy, P.; Seiler, K.; Rusterholz, B.; Simon, W. Anal. Chem.
1992, 64, 2029–2035.
(43) Jadhav, S.; Meir, A. J.; Bakker, E. Electroanalysis 2000, 12, 1251–1257.
(44) Jadhav, S.; Bakker, E. Anal. Chem. 2001, 73, 80–90.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6891

Figure 4. Normal pulse voltammograms for an electrode modified with FcPVC-DOS with 10 wt % of ETH 500 performed in 10 mM solutions
of the indicated anions. Inset: Response behavior to sodium chloride at more negative potentials, demonstrating the suppression of cation
transfer processes.
Figure 5. Cyclic voltammogram for an FcPVC modified electrode in 1 mM solution of K3Fe(CN)6 and 0.1 M KCl.
for redox sensitivity in the presence of potassium hexacyanoferrate(III),
which is known to undergo extraction into the organic
phase at positive potentials, 45 giving anodic currents, while the
undesired reduction of the Fe(III) species mediated by FcPVC
should result in cathodic currents at negative potentials. To
evaluate this, electrodes were scanned from -0.5 to 0.8 V in
aqueous hexacyanoferrate(III) solutions (Figure 5). The cyclic
voltammograms revealed indeed an anodic anion extraction wave
at positive potentials, together with its subsequent stripping on
the return scan. On the other hand, the cyclic voltammogram in
the 0.4 to -0.5 V potential range remained featureless, suggesting
that the Fe(III) cannot easily be reduced at the membrane surface.
This is consistent with the observation above that only a small
fraction of total membrane-bound ferrocene is electrochemically
accessible. This suggests that the material may be used as a
singular membrane matrix without the need for a more complex
two-layer system that is necessary with the conducting polymer
based ion to electron transducing layers.
(45) Chen, S.-M.; Chzo, W.-Y.; Thangamuthu, R. J. Solid State Electrochem. 2008,
12, 1487–1495.
6892 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
One disadvantage of the conducting polymer layer POT as an
ion-to-electron transducer appears to be its limited redox capacity.
As shown by Kim et al., a layer of this polymer obtained during
electro-deposition on a 5 mm diameter gold electrode and modified
with a thin, spin-coated PVC membrane showed a capacity limit
of about 3 µC. 25 This value was found for a stripping voltammetry
analysis applying 30, 60, 120, 300, and 600 s doping times. An
analogous experiment was performed here for FcPVC-based
electrode applying the same doping times at 0.8 V. Since the
stripping process was expected to be slower for a thicker
membrane (about 100 µm), more suitable procedure for FcPVC
redox-capacity study was to calculate the charge of a doping
process applying chronoamperometry. The doping was performed
at 0.8 V for 30, 60, 120, 300, and 600 s. The charge calculated
based on integrals of obtained chronoamperograms is presented
as a function of doping time in Figure 6. Contrary to results
obtained by Kim et al. where the POT based electrode reached
its maximum capacity after 2 min to a value of less than 3 µC, 25
the doping charge was still growing for the FcPVC-based electrode
after 600 s. It was found that after 600 s of FcPVC doping, the

Figure 6. Calculated charge from chronoamperometric perchlorate ion uptake (0.1 M NaClO4) into the ferrocene modified plasticized PVC
membrane electrode, performed at 0.8 V vs Ag/AgCl and carried out for the indicated times.
Figure 7. Normal pulse amperometry of a FcPVC-based membrane electrode immersed in 10 mM NaClO4 for 40 anodic perchlorate ion
transfer pulses for 1sat0.5V,each followed by 20 s stripping pulses at 0.3 V (open circuit potential, found before the experiment). Shown here
for clarity are the first and last two of the recorded pulses.
calculated doping charge was more than 520 µC, which is 2 orders
of magnitude larger than for POT based electrodes. The value is
somewhat larger than for the sandwiched system described above
(see Figure 2), likely because of the applied electrochemical
doping times were not long enough to effect conversion of the
entire membrane material. Moreover, more than 50 µC conversion
was observed for FcPVC in the first 30 s, significantly more than
the 2.5 µC for POT-based membranes in the same time period, 25
which is advantageous for ion voltammetry sensor development.
When operating the membrane by normal pulse amperometry
or normal pulse voltammetry, the observed currents are often
limited by ion transport processes in the aqueous or membrane
phase. 43 If this holds true, changes in charge transfer kinetics of
the underlying redox couple may not substantially influence the
observed amperometric response if the potential of the redox
couple remains constant in the course of the experiment. Figure
7 demonstrates normal pulse amperometry in a 10 mM NaClO4
solution, using a1sexcitation pulse at 0.5 V followed by a
20 s baseline pulses at the open circuit potential. Indeed, the
current response generated during the 1 s long pulse was very
reproducible (1% relative SD) in the course of a 40 pulse
sequence. This suggests that the ferrocene modified PVC
introduced here is a promising material for voltammetric ion
sensor design. While we have not tested the behavior of freely
dissolved lipophilic ferrocene species in this work, we note that
the group of Samec has reported on adequate stability with
membranes containing freely dissolved 1,1′-dimethylferrocene
as a redox active additive, although possible interference by
redox active species in the sample were not yet studied there. 30
CONCLUSIONS
This paper presented a modification of PVC with ferrocene
groups via an attractive “click chemistry” approach. The new
material (FcPVC) was characterized in view of membrane electrode
applications, specifically with voltammetric transduction
principles. A new method applying two identical electrodes at the
same doping state sandwiched together allowed one to confirm
the redox properties of the FcPVC, independent of the processes
occurring at the sample side of the membrane. This is important
because the resulting sensors are often operationally limited by
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6893

ion transfer and diffusion processes, which are not the focus of
this research. The redox capacity of the system was found to
correspond to several nanomolar of oxidizable ferrocene groups
in the membrane, which is only a fraction of nominal ferrocene
concentration. This suggests that the ferrocene groups are
sufficiently dilute and immobile in the membrane and that it is
the surface confined functionalities that are predominantly electrochemically
accessible. Nonetheless, the redox capacity appears
to be adequate for most voltammetric ion transfer applications.
The membranes were characterized in contact with aqueous
electrolytes, and interrogation with normal pulse voltammetry
suggests excellent reproducibility and an anion selectivity pattern
that corresponds to the expected Hofmeister selectivity sequence.
The cathodic potential window remained featureless, supporting
the notion of a membrane containing predominantly ferrocene
moieties that can be oxidized by the concomitant extraction of
anions from the sample. It is expected that adequate chemical
doping of the membrane with cation-exchanger species will make
the cationic response region at cathodic potentials analytically
6894 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
accessible, but this was not yet explored here. Experiments in
the presence of potassium hexacyanoferrate(III) showed that
possible presence of ferrocene groups at the surface of the ionselective
membrane do not cause redox sensitivity of the electrode.
Moreover, the high reproducibility of the sampled anion transfer
currents (standard deviation below 1%) as well as the high redox
capacity of the material (about 520 µC) makes ferrocene bound
PVC an attractive material for electrochemical all solid state ionselective
electrode fabrication.
ACKNOWLEDGMENT
The authors wish to thank the Australian Research Council
(Grant DP0987851) and the National Institutes of Health (Grant
EB002189) for financial support of this research.
Received for review April 23, 2010. Accepted July 10,
2010.
AC1010662

Anal. Chem. 2010, 82, 6895–6903
Electrophoretic Analysis of Biomarkers using
Capillary Modification with Gold Nanoparticles
Embedded in a Polycation and Boron Doped
Diamond Electrode
Lin Zhou, † Jeremy D. Glennon, † and John H. T. Luong* ,†,‡
Innovative Chromatography Group, Irish Separation Science Cluster (ISSC), Department of Chemistry & the ABCRF,
University College Cork, Cork, Ireland and Biotechnology Research Institute, National Research Council Canada,
Montreal, Quebec, Canada H4P 2R2
Field-amplified sample stacking using a fused silica
capillary coated with gold nanoparticles (AuNPs) embedded
in poly(diallyl dimethylammonium) chloride (PDDA)
has been investigated for the electrophoretic separation
of indoxyl sulfate, homovanillic acid (HVA), and vanillylmandelic
acid (VMA). AuNPs (27 nm) exhibit ionic and
hydrophobic interactions, as well as hydrogen bonding
with the PDDA network to form a stable layer on the
internal wall of the capillary. This approach reverses
electro-osmotic flow allowing for fast migration of the
analytes while retarding other endogenous compounds
including ascorbic acid, uric acid, catecholamines, and
indoleamines. Notably, the two closely related biomarkers
of clinical significance, HVA and VMA, displayed differential
interaction with PDDA-AuNPs which enabled the separation
of this pair. The detection limit of the three analytes obtained
by using a boron doped diamond electrode was ∼75 nM,
which was significantly below their normal physiological
levels in biological fluids. This combined separation and
detection scheme was applied to the direct analysis of these
analytes and other interfering chemicals including uric and
ascorbic acids in urine samples without off-line sample
treatment or preconcentration.
Indoxyl sulfate (IXS), a metabolite of tryptophan (TRP) and a
dietary protein, is derived from intestinal metabolism and liver
conjugation 1 and excreted in urine in high concentration. It is also
an endogenous compound in mammals, in mouse plasma and
brain samples, as detected by liquid chromatography/tandem
mass spectrometry. 2 This circulating protein-bound uremic toxin
stimulates glomerular sclerosis, interstitial fibrosis, and the
progression rate of renal failure. IXS induces endothelial dysfunction
by inhibiting endothelial proliferation and migration in vitro
* To whom correspondence should be addressed.
† Innovative Chromatography Group, Irish Separation Science Cluster (ISSC),
Department of Chemistry & the ABCRF, University College Cork.
‡ Biotechnology Research Institute, National Research Council Canada.
(1) Dealler, S. F.; Hawkey, P. M.; Millar, M. R. J. Clin. Microbiol. 1988, 26
(10), 2152–2156.
(2) Wang, G.-F.; Korfmacher, W. A. Rapid Commun. Mass Spectrom. 2008,
23 (13), 2061–2069.
and its role in oxidative stress is implicated. 3 Homovanillic acid
(HVA) is a major catecholamine metabolite, associated with the
brain dopamine level. As a biomarker of metabolic stress of
2-deoxy-D-glucose, the HVA level in the brain and the cerebrospinal
fluid is indicative of pheochromocytoma and neuoroblastoma. 4
Metanephrine, one of the hormones produced by the adrenal
glands, breaks down to normetanephrine and vanillylmandelic acid
(VMA) via the intermediate 4-hydroxy-3-methoxy-phenylglycol.
Thus, VMA is always detected in the urine together with HVA
and other catecholamine metabolites from pheochromocytoma or
catecholamine-secreting chromaffin tumor cells. 5,6 Catecholamine
levels can be found in blood samples; however, the urine test
reflects the production rate of the catecholamines over the
collection period. Even at abnormal levels, the elevated quantities
of these catecholamines are still very low, i.e., highly selective
and sensitive analytical methods are needed for such important
biomarkers. The urinary VMA and HVA values are most useful
and can be correlated with the stage of disease, management, any
maturation of tumor, and prognosis from children with neuroblastoma.
4 The urine HVA/VMA ratio could be a screening tool
to support earlier detection of Menkes disease, a disorder that
affects copper levels in the body, leading to copper deficiency. 7a
The mechanism that removes HVA from the brain is still poorly
understood, however, the efflux transport of HVA from the brain
plays an important role in controlling the HVA level in the brain.
This HVA efflux transport system is inhibited by several organic
anions including IXS, and metabolites of monoamine neurotransmitters
but not neurotransmitters per se. 7b Thus, it is of clinical
importance to develop a rapid and sensitive method for simultaneous
analysis of HVA, VMA, and IXS in urine and other biological
samples.
(3) Tumur, Z.; Niwa, T. Am. J. Nephrol. 2009, 29, 551–557.
(4) Liebner, E. J.; Rosenthal, I. M. Cancer 2006, 32 (3), 623–633.
(5) Eisenhofer, G.; Lenders, J. W. M.; Linehan, W. M.; Walther, M. M.;
Goldstein, D. S.; Keiser, H. R. New Engl. J. Med. 1999, 340, 1872–1879.
(6) Lenders, J. W. M.; Keiser, H. R.; Goldstein, D. S.; Willemsen, J. J.; Friberg,
P.; Jacobs, M.-C.; Kloppenborg, P. W. C.; Thien, T.; Eisenhofer, G. Ann.
Intern. Med. 1995, 123, 101–109.
(7) (a) Menkes, J. H.; Alter, M.; Steigleder, G. K.; Weakley, D. R.; Sung, J. H.
Pediatrics 1962, 29, 764–779. (b) Mori, S.; Takanaga, H.; Ohtsuki, S.;
Deguchi, T.; Kang, Y.-S.; Hosoya, K.-I.; Terasaki, T. J. Cereb. Blood Flow
Metab. 2003, 23, 432–440.
(8) (a) Issaq, H. J.; Delviks, K.; Janini, G. M.; Muschik, G. M. J. Liquid
Chromatogr. Relat. Technol. 1992, 15/18, 3193–3201.
10.1021/ac101105q © 2010 American Chemical Society 6895
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/16/2010

HVA and VMA in infant urine has been analyzed by capillary
electrophoresis (CE) with UV detection with a detection limit
(LOD) of 3.5 × 10 -4 M for HVA and 1.8 × 10 -4 M for VMA. 8a A
simultaneous determination of VMA, HVA, creatinine, and uric
acid uses capillary electrophoresis and a 30 mM phosphate
buffer (pH 7.0) containing 150 mM sodium dodecyl sulfate
(SDS). The detection is by UV absorbance at 245 nm and the
run is rather lengthy (15 min). 9 Although the authors claimed
lower LODs for both HVA and VMA (5.5 × 10 -5 M and 5.0 ×
10 -5 M), such LODs are still significantly above the normal
levels found in healthy subjects (8.2 to 41 µM for HVA and
11.6 to 28.7 µM for VMA). 10 Besides high LOD, UV detection
is also problematic as biological samples often consist of several
compounds with strong absorption at low wavelengths. The
neurotransmitters can be conjugated with a strong fluorophore
and analyzed by MEKC with fluorescence detection to achieve
high sensitivity. 11a This approach might be feasible for routine
analysis of a few compounds but becomes impractical and timeconsuming
for multiple analytes. The use of CE-MS for analysis
of VMA, HVA, and other biomarkers for metabolic disorders
in newborns is available elsewhere. 11b CE equipped electrochemical
detection (ECD) using a bare silica capillary can be
used to detect HVA and VMA after electrophoretic separation
at pH 5. 12a,b However, HVA and VMA are not baseline separated
and other catecholamines including IXS are not included. 12a HVA
and VMA also become less electroactive at pH > 5, the minimal
required pH for electrophoretic separation of HVA and VMA. In
general, bare fused silica capillaries at high pH are used in such
studies to resolve IXS, HVA, and VMA. The separation is lengthy
since negatively charged IXS, HVA, and VMA migrate to the
anode, i.e., opposite flow direction to the electroosmotic flow
(EOF). In addition, HPLC with isocratic elution and spectrophotometric
detection are often used for analysis of tryptophan
metabolites including IXS. 12c HPLC is also coupled with mass
spectrometry to detect compounds associated with purple urinary
bag syndrome (PUBS) and IXS is one of these toxic
compounds. 12d Micellar electrokinetic chromatography (MEKC)
using a bare fused silica capillary and laser induced fluorescence
detection (a KrF excimer laser, λ ) 248 nm) has been used to
detect HVA, VMA, and IXS. Under the best condition with this
expensive instrumentation, the detection limit for HVA and VMA
is 170 nM and 150 nM, respectively. 12e
This work describes a novel scheme for the analysis of IXS,
HVA, and VMA in the presence of tryptophan and other important
catecholamines and indoleamines (Table 1). The fused silica
capillary is coated with a thin layer of poly(diallyl dimethylammonium)
chloride (PDDA) or gold nanoparticles (AuNPs) embed-
(9) Shirao, M. K.; Suzuki, S.; Kobayashi, J.; Nakazawa, H.; Mochizuki, E.
J. Chromatogr. B 1997, 693, 463–467.
(10) Garcia, A.; Heinanen, M.; Jimenez, L. M.; Barbas, C. J. Chromatogr. A 2000,
871, 341–350.
(11) (a) Caslavska, J.; Gassmann, E.; Thormann, W. J. Chromatogr. A 1995,
709, 147–156. (b) Senk, P.; Kozak, L.; Foret, F. Electrophoresis 2004, 25,
1447–1456.
(12) (a) Li, X. J.; Jin, W. R. Chin. Chem. Lett. 2002, 13 (9), 874–876. (b) Li,
X. J.; Jin, W. R.; Weng, Q. F. Anal. Chim. Acta 2002, 461 (1), 123–130. (c)
Marklova, E.; Makovickova, I.; Krakorova, I. J. Chromatogr. A 2000, 870,
289–293. (d) Bar-Or, D.; Rael, L. T.; Bar-Or, R.; Craun, M. L.; Statz, J.;
Garrett, R. E. Clin. Chim. Acta 2007, 378, 216–218. (e) Paquette, D. M.;
Sing, R.; Banks, P. R.; Waldron, K. C. J. Chromatogr. B: Biomed. Sci. Appl.
1998, 714 (1), 47–57.
6896 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
ded in PDDA to reverse the EOF, allowing fast migration of IXS,
VMA, HVA, and tryptophan. In contrast, catecholamines and
indoleamines migrate against the EOF and emerge very late in
the electropherogram, i.e., they do not interfere with the analysis
of IXS, VMA, HVA, and tryptophan. The presence of AuNPs plays
an important role in baseline separation of several compounds
and a mechanism is given to decipher the interaction between
AuNPs and the analytes. Although the composite consisting of
AuNPs embedded in PDDA has been reported by Chen et al., 13
this is the first systematic application of an AuNP-PDDA coated
capillary for simultaneous analysis of IXS, HMA, and VMA in urine
samples. Together with sample stacking, the applicability of this
approach for analysis of such important biomarkers in urine
samples with improved detection sensitivity is also presented and
discussed in detail.
EXPERIMENTAL SECTION
Chemicals. Poly(diallyl dimethylammonium) chloride (PDDA,
MW ) 200 000-350 000, 20 wt % in water), hydrogen tetrachloroaurate
tetrahydrate (HAuCl4 · 4H2O), Tris (hydroxymethyl)aminomethane,
phosphoric acid (H3PO4), and other chemicals
were purchased from Sigma (Dublin, Ireland). Unless otherwise
stated, a 50 mM H3PO4 solution was adjusted to pH 3.0
with 0.5 M Tris buffer and used as the separation buffer. The
standard stock solutions (5.0 mM) of the analytes were
prepared daily in deionized water. All solutions were prepared
in Milli-Q ultrapure water and filtered through a 0.22 µm pore
size membrane followed by sonication for 5 min prior to use.
Synthesis of PDDA-Gold Nanoparticle (AuNPs) Composite.
The PDDA-AuNP composite was prepared as described by
Chen et al. 13 PDDA (250 µL, 4% wt. in H2O), 40 mL of H2O, 200
µL of 0.5 M NaOH, and 300 µL of HAuCl4 (10 mg/mL) were
thoroughly mixed in a beaker, covered with an inverted culture
dish for 2 min. The mixture was then maintained at 100 °C for
30 min, resulting in a ruby red solution. The UV-visible
spectrum of the AuNP colloid was recorded on a HP 8453
UV-visible spectrophotometer in a1cmoptical path quartz
cuvette. The size distribution of AuNPs in PDDA was measured
using a zetasizer Nano ZS system (Malvern Instruments, MA)
which is based on a dynamic light scattering (DLS) technique.
TEM micrographs were obtained by a Delong LVEM (Soquelec,
Montreal, QC, Canada) low-voltage TEM at 5 kV. A
small amount of PDDA-AuNPs was sonicated to disperse the
material. A 20 µL sample of well dispersed suspension was then
dried on a Formvar-carbon coated grid and analyzed.
Preparation of Coated Capillaries. A fused-silica capillary
(50 µm id and 365 µm od) purchased from Polymicro Technologies
(Phoenix, AZ, USA) was cut to 45 cm as the effective capillary
length. In order to expose the maximum number of silanol groups
on the silica surface, the fused-silica capillary was rinsed with 1.0
M NaOH and deionized water for 15 min each. The preconditioned
capillary was then rinsed with the PDDA or the PDDA-AuNP
(13) Chen, H.-J.; Wang, Y.-L.; Wang, Y.-H.; Dong, S.-J.; Wang, E. Polymer 2006,
47 (2), 763–766.
(14) (a) Luong, J. H. T.; Male, K. B.; Glennon, J. D. Analyst 2009, 134 (10),
1965–1979. (b) Kraft, A. Int. J. Electrochem. Sci. 2007, 2, 355–385. (c) Swain,
G. M.; Ramesham, R. Anal. Chem. 1993, 65, 345–351. (d) Xi, J.; Granger,
M.; Chen, Q.; Strojek, K.; Lister, T.; Swain, G. Anal. Chem. 1997, 69, 591A–
597A. (e) Tenne, R.; Patek, K.; Hashimoto, K.; Fujishima, A. J. Electroanal.
Chem. 1993, 347, 409–415.

Table 1. Chemical Structure, pKa Values and Aqueous Solubilities (A, g/L) of Various Analytes
solution for 15 min followed by an incubation period of 15 min.
The coated capillary was gently rinsed with deionized water for 3
min to flush out unadsorbed coating materials. Before the first
run, the coated capillary was equilibrated with the running buffer
for 15 min but only for 3 min between the runs. All of these
procedures were performed at 25 °C. For overnight or prolonged
storage, the capillary was rinsed with deionized water for 15 min
and then stored with the capillary ends dipped in deionized water.
Capillary Electrophoresis with Electrochemical Detection.
The capillary outlet was epoxy sealed into a pipet tip so that only
∼1 cm protruded. The pipet tip was firmly attached vertically into
a micromanipulator (HS6, World Precision Instruments, Sarasota,
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6897

FL, USA) with three-dimensional adjustment capabilities. A
cylindrical cathodic/detection reservoir (2 cm diameter ×1 cm
height) contained Pt wires (1 mm in diameter, 99.9% purity),
serving as the counter electrode for amperometric detection and
the cathode for electrophoresis. An Ag/AgCl (3 M NaCl) reference
electrode was placed vertically into the reservoir, whereas the
BDD electrode was inserted upward from the reservoir’s bottom
and sealed with epoxy (the working reservoir volume was ∼3 mL).
The micromanipulator and a laboratory jack (to which the
reservoir was solidly mounted) were attached to a solid breadboard
to prevent movement during alignment. The capillary outlet
was aligned to the detecting electrode using the micromanipulator
with the aid of a surgical microscope (World Precision Instruments).
The capillary outlet was adjusted until it touched the
electrode surface (evident by a slight bend in the capillary
observed by microscopic inspection) and it was then backed off
25-30 µm using the micromanipulator’s z-control. The BDD
electrode was connected to an electrochemical workstation
(CHI660C, CH Instruments, Austin, TX, USA) consisting also of
a platinum wire (1 mm in diameter) as counter electrode and an
Ag/AgCl (3 M NaCl) electrode as reference electrode. The BDD
electrode (3 mm in diameter, 0.1% doped diamond) was purchased
from Windsor Scientific (Slough, Berkshire, U. K.). The analytes
are detected by a boron doped diamond (BDD) electrode which
is positioned close to the capillary outlet. The BDD electrode is
advocated in this work due to its stable low background current
and a wide applied potential window. BDD is also resistant to
fouling due to the hydrogen surface termination and sp 3 carbon
bonding (no extended pi-electron system). 14a,b BDD can be
considered as general-purpose working electrodes with a broad
array of applications for use with HPLC and CE. Pioneering work
in the early 1990s was conducted by Swain and co-workers 14c,d
and the Fujishima group. 14e
The electrophoretic separation was conducted at -10 kV
(reversed polarity) unless otherwise stated. A plastic cap with a
central hole of ∼1 mm was firmly attached to the surface of the
BDD electrode to reduce the active sensing area. The analyte
sample was injected electrokinetically for 5sat-10 kV. Peak
identification was based on the migration time of a single standard
with that of unknown peaks. However, if the resolution between
any peak pair was low, then peak identification was performed
by spiking both solutes individually. The pKa values and aqueous
solubility of the analytes were obtained using the ACD/
Structure Designer software (Advanced Chemistry Development,
Toronto, ON, Canada). The degree of ionization was
estimated as pH ) pKa+ log [A - /AH] for acid and pH )
pKb+log [BH + /B] for base (the Henderson-Hasselbalch
equation).
RESULTS AND DISCUSSION
Performance of the PDDA Coated Capillary. PDDA was
firmly adsorbed on the inner walls of the capillary via ionic
interactions between the negatively charged SiO - of fused silica
and the quaternary ammonium groups of the polymer. Indeed,
immersion of a substrate (glass, quartz, silica wafer, gold, silver,
and even Teflon) into an aqueous 1% solution of this positively
charged polymer results in the strong adsorption of a mono-
6898 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 1. Electropherograms obtained using a PDDA coating
capillary (50 µm id and 45 cm effective length) for the separation of
20 µM IXS, 25 µM VMA, 25 µM HVA, 25 µM TRP, 100 µM
isoproterenol (ISP), 100 µM normetanephrine (NMN), 100 µM
epinephrine (EP), 50 µM 5-hydroxytryptamine (5-HT), 125 µM 4-hydroxy-3-methoxybenzylamine
(HMBA), and 75 µM tryptamine (TA).
The running buffer consisted of 50 mM H3PO4-Tris, (a) pH 3, (b) pH
4, and (c) pH 5. The separation voltage was applied at -10 kV with
an injection time of 5s at -10 kV. BDD at +1.0 V vs Ag/AgCl, 3 M
NaCl.
layer (1.6 nm) of PDDA on the substrate. 15,16 The adsorption
of a PDDA thin film on a glass substrate was also reported
elsewhere. 17 Notice also that the charge of PDDA is not pH
dependent, as reflected by constant EOF in the range pH 2-8
provided the capillary is coated with high-molecular weight
PDDA. 18 Thus, PDDA with MW of 200 000-350 000 was used in
this study for coating the capillary.
At -10 kV, except for the epinephrine (EP) and normetanephrine
(NMN) pair, all analytes were baseline resolved when 50 mM
H3PO4-Tris pH 3.0 was used as the running buffer (Figure 1,
curve a). On the basis of the calculated pKa values for the analytes
(Table 1), fully deprotonated and highly negatively charged IXS
exhibited high electrophoretic mobility and migrated concomitantly
with EOF as the first peak in the electropherogram. The
carboxylate/carboxyl ratio, estimated as 10 (pH-pK a) ,is∼0.16 and
0.04 for VMA and HVA, respectively. Thus, VMA should
emerge before HVA and slightly ahead of EOF. At pH 3, the
neutral form of TRP should be predominant (neutral TRP/TRP +
) 4.2);therefore, it should migrate very closely to EOF and
trail behind both VMA and HMA. The EP-NMN pair was
slightly split further by conducting the separation at pH 4 with
improved detection sensitivity but the running time was also
slightly longer and VMA emerged very close to IXS (Figure 1,
curve b). The run was lengthier at pH 5, with only 4 discernible
peaks emerging in the electropherogram as HVA comigrated with
VMA and TRP became more neutral at this pH and emerged far
behind the HVA peak. The remaining analytes acquired more
negative charges and interacted strongly with positively charged
PDDA and were not eluted after 1200 s into the experiment. The
detection sensitivity was also greatly compromised at this running
(15) Kotov, N. A.; Harazsti, T.; Turi, L.; Zavala, G.; Geer, R. E.; Dekany, I.;
Fendler, J. H. J. Am. Chem. Soc. 1997, 119, 6821–6832.
(16) Moriguchi, I.; Teraoka, Y.; Kagawa, S.; Fendler, J. H. Chem. Mater. 1999,
1, 1603–1608.
(17) Hrapovic, S.; Liu, Y.; Enright, G.; Bensabaa, F.; Luong, J. H. T. Langmuir
2003, 19, 3958–3965.
(18) Wang, Y.; Dubin, P. L. Anal. Chem. 1999, 71, 3463–3468.

pH with a tilted baseline (Figure 1, curve c). It is of interest to
compare the migration order of these peaks with MEKC performed
at normal electrophoretic separation conditions by Caslavska
et al. 11a In such a study, a buffer composed of 75 mM sodium
dodecyl sulfate (SDS), 6 mM Na2B4O7, and 10 mM Na2HPO4
(pH 9.2) is used with the separation potential set at +20 kV.
The migration sequence is TRP, HVA, VMA, and IXS, which
is opposite to the migration order shown in Figure 1. The
remaining 7 catecholamines and indoleamines with a positive
charge from the ammonium ion (NH3 + ) and/or the secondary
amino group (-NH + -) were completely ionized at pH 3 and
migrated to the cathode with high mobilities, i.e., countercurrent
to the EOF. However, they were eventually driven out the
capillary by EOF with higher mobility.
Notice that a buffer composed of 200 mM boric acid, 100 mM
potassium hydroxide, and 0.1% hydroxyethylcellulose, pH 9.2, has
been used to separate VMA from various indoleamines under
normal separation. 19 In this case, VMA trailed far behind other
indoleamines, as expected from its negative charge at this pH.
PDDA can also be added as a buffer additive as described by
Tseng et al., 20 resulting in high and reversed EOF. The mobility
of indolamines and catecholamines decreases as the PDDA
concentration increases. The separation of 14 analytes including
indolamines, catecholamines, and metanephrines is achieved
within 33 min under optimal separation conditions (1.2% PDDA
and 5 mM formic acid at pH 4.0). Indeed, PDDA has been used
to coat fused silica capillary to form a thin film for the subsequent
absorption of carbon nanotubes. 21 Besides adsorption, PDDA can
be chemically bonded onto the interior capillary wall with an
anodal EOF independent of pH ranging from 2.2 to 8.8. 22 The
lifetimes of both the bonded and physically coated capillaries
exceeded 40 h of continuous use at 240 V/cm at pH 4. Our
experimental data confirmed that the PDDA could be reused for
several repeated runs and the capillary was easily reconditioned
as described earlier. The coated capillary also exhibited good
tolerance to methanol and 0.1 HCl.
Separation on Capillary Coated with PDDA-Gold Nanoparticles.
Our next strategy was to form gold nanoparticles
(AuNPs) on the capillary wall, since they have been known to
interact with several compounds with amino, hydroxyl, and
carboxylic groups. 23 In principle, AuNPs can be prepared separately
and adsorbed on the PDDA layer. This concept has been
used to coat the glass microchip channel with citrate-stabilized
AuNPs for the separation of phenols. 24 Another approach is to
form AuNPs by electroless plating via hydroxylamine-mediated
reduction. 17 However, the charge of the PDDA-AuNP layer returns
to negative and the separation must be performed under normal
separation. Indeed, covalent attachment can be used to immobilize
AuNPs onto the capillary wall, for instance, the preparation of
dodecanethiol AuNPs on prederivatized 3-aminopropyl-trimethox-
(19) Stocking, C. J.; Slater, J. M.; Simpson, C. F. Exp. Nephrol. 1998, 6, 415–
420.
(20) Tseng, W.-L.; Chen, S.-M.; Hsu, C.-Y.; Hsieh, M.-M. Anal. Chim. Acta 2008,
613 (1), 108–115.
(21) Luong, J. H.T.; Bouvrette, P.; Liu, Y.; Yang, D.-Q.; Sacher, E. J. Chromatogr.
A 2005, 1 (2), 187–194.
(22) Liu, Q.; Lin, F.; Hartwick, R. A. J. Chromatogr. Sci. 1997, 35 (3), 126–130.
(23) Zhong, Z. Y.; Patskovskyy, S.; Bouvrette, P.; Luong, J. H. T.; Gendanken,
A. J. Phys. Chem. B 2004, 108 (13), 4046–4052.
(24) Pumera, M.; Wang, J.; Grushka, E.; Polsky, R. Anal. Chem. 2001, 73, 5625–
5628.
Figure 2. Electropherograms obtained using a PDDA-AuNPs coated
capillary at pH 3, 4, and 5. Other conditions were same as Figure 1.
ysilane or 3-mercaptopropyl-trimethoxysilane fused-silica capillaries.
25 Again; several steps are required for the preparation of such
modified capillaries.
In order to maintain the positive charge for the coating layer,
a one step procedure described by Chen et al. 13 was used to
prepare the PDDA-AuNP composite since PDDA acts as both
reducing and stabilizing agents for AuNPs. The PDDA-gold colloid
exhibited a surface plasmon resonance (SPR) around 527-528
nm, which could be related to AuNPs with the mean diameter of
27-28 nm, in agreement with the report on the SPR peak obtained
for AuNPs with an average diameter of 27 nm. 26 Dynamic light
scattering confirmed that AuNPs should have a mean diameter
of 28 nm with narrow particle size distribution. A TEM micrograph
of PDDA-AuNPs indicated that AuNPs exhibited an average size
of 25 nm (figure not shown). Notice that the excellent stability of
AuNPs arises from the electrosteric effect of PDDA, in agreement
with Mayer et al. 27 where PDDA is simply used as a protecting
and stabilizing agent for AuNPs. Under the same electrophoretic
and detection condition for the PDDA coated capillary, the
electropherogram obtained for the 10 analytes using the PDDA-
AuNP coated capillary is shown in Figure 2. At pH 3, EP was
satisfactorily separated from NMN and the peaks were significantly
sharper with low background current, resulting in significantly
improved detection sensitivity. For the microchip channel
with coated PDDA/AuNPs layer-by-layer, improved resolution and
detection sensitivity has been reported for the three aminophenols,
although the rationale behind such behavior is not known. 24 Our
experimental data also confirmed that the run was longer at pH
4 without any improvement in the separation of the NMN-EP pair
with slightly reduced detection sensitivity. At pH 5, only four peaks
(IXS, VMA, HVA and TRP) emerged in the electropherogram with
the TRP peak far behind the HVA peak.
The circumstances determining the effect of the PDDA-AuNP
coated capillary are complex. The viscosity of the PDDA-AuNP
solution is about 90% of the PDDA solution, i.e., it is easier to
pump the PDDA-AuNP solution through the capillary to form a
(25) (a) O’Mahony, T.; Owens, V. P.; Murrihy, J. P.; Guihen, E.; Holmes, J. D.;
Glennon, J. D J. Chromatogr. A 2003, 1004 (1-2), 181–193. (b) Yang, L.;
Guihen, E.; Holmes, J. D.; Loughran, M.; O’Sullivan, G. P.; Glennon, J. D.
Anal. Chem. 2005, 77 (6), 1840–1846.
(26) Nasir, S. M.; Nur, H. J. Fundam. Sci. 2008, 4, 245–252.
(27) Mayer, A. B. R.; Hausner, S. H.; Mark, J. E. Polym. J. 2000, 32, 15–22.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6899

Figure 3. Electropherograms obtained using a PDDA-AuNP coated capillary with the BDD electrode for the study of (A) buffer concentration
effect: 25, 50, 75, and 100 mM H3PO4-Tris, pH 3, (B) injection time effect (3, 5, 7, and 10 s), (C) effect of detection potential (+0.6, +0.8, +1.0,+
and 1.2 V vs Ag/AgCl, 3 M NaCl), (D) effect of separation voltage (-5, -10, -15, and -20 kV). The sample mixture consisted of 20 µM IXS,
25 µM (each) VMA, HVA, TRP, 100 µM ISP, 100 µM NMN, 100 µM EP, 50 µM 5-HT, 125 µM HMBA, 125 µM DHBA, and 75 µM TA.
homogeneous layer. 28 The incorporation of AuNPs in the polymer
network could be attributed to the structural change in pristine
PDDA after the synthesis of AuNPs. It has been shown that CdC
and C-N are produced after PDDA reacts with HAuCl4 by
(28) Zhang, Z.-X.; Yan, B.; Liu, K.; Liao, Y.-P.; Liu, H.-W. Electrophoresis 2009,
30, 379–387.
6900 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
comparing the FTIR spectra of PDDA and PDDA-AuNPs. 13 For
the latter, a new peak appears at 1578 cm -1 , which can be
assigned for CdC and CdN in-plane vibration. In aqueous
milieu, gold nanoparticles are negatively charged and the
absolute value of the �-potential decreases as the pH decreases,
whereas above pH 5.8, the absolute value of the �-potential

Table 2. Linearity and Detection Limit of the Analytes a,b
analytes linear range (µM) calibration equation R2 detection limit (µM)
indoxyl sulfate (IXS) 1.00-40 I ) 0.4839X+0.1490 0.9998 0.3
vanillylmandelic acid (VMA) 1.25-50 I ) 0.3635X+0.3036 0.9998 0.4
homovanillic acid (HVA) 1.25-50 I ) 0.3034X+0.5667 0.9998 0.4
tryptophan (TRP) 1.25-50 I ) 0.3025X+1.0870 0.9998 0.4
isoproterenol (ISP) 5.00-200 I ) 0.0871X+0.7481 0.9995 1.7
normetanephrine (NMN) 5.00-200 I ) 0.0865X+0.6429 0.9995 1.7
epinephrine (EP) 5.00-200 I ) 0.0945X+0.3850 0.9995 1.7
5-hydroxytryptamine (5-HT) 2.50-100 I ) 0.1686X+0.9364 0.9995 0.8
4-hydroxy-3-methoxybenzylamine (HMBA) 6.25-250 I ) 0.0822X+0.0446 0.9991 2.1
3,4- dihydroxybenzylamine (DHBA) 6.25-250 I ) 0.0866X+0.3842 0.9991 2.1
tryptamine (TA) 3.75-150 I ) 0.1476X+0.5307 0.9995 1.3
a I ) peak current (nA), X ) analyte concentration (µM). b Electrophoretic separation was carried out using a PDDA-AuNP coated capillary (50
µm id, 365 µm od) with an effective length of 45 cm, running buffer, 50 mM H3PO4-Tris, pH 3.0, separation voltage, -10 kV, injection for 5s at -10
kV. Detection: BDD electrode at +1.0 V vs. Ag/AgCl, 3M NaCl.
Figure 4. The comparison electropherograms for (a) nonsample
stacking using a running buffer of 50 mM H3PO4-Tris, pH 4 with an
injection time of 5 s (b) sample stacking, injection buffer: 10 mM
H3PO4-Tris, pH 2, and the running buffer of 50 mM H3PO4-Tris, pH 4
with an injection time of 10 s. (c) sample stacking, injection buffer:
10 mM H3PO4-Tris, pH 2, and the running buffer was similar to the
injection buffer with an injection time of 10 s. Concentration of
analytes: 50 µM (each) IXS, VMA, AA, HVA, UA, TRP. Separation
voltage: -10 kV. BDD at +1.0 V vs Ag/AgCl, 3 M NaCl.
becomes almost constant (∼-45 to -50 mV). The negative
charge of the gold surface may be explained by the fact that
the Au-OH present on the gold surface can lose protons as
the pH increases to form Au-O-groups on the nanoparticle
surface. 29 AuNPs have a partially hydroxylated surface with a
pKa value of 3.2 (Au-OH) in the presence of water. 29 At pH 3,
around the pKa value of AuNPs, -OH groups are predominant
(-OH groups/O - groups ) 1.6). The rest of the gold surface
should be metallic, i.e., essentially hydrophobic. Therefore,
AuNPs would display ionic and hydrophobic interaction, as well
as hydrogen bonding with the PDDA network. The synthesis
and incorporation of AuNPs in the polymer network would
affect the PDDA structural change, which in turn increased
the charge density and coverage efficiency of the coating. At
pH4or5,Au-O - groups were predominant, resulting in a
(29) Sylvestre, J.-P.; Kabashin, A. V.; Sacher, E.; Meunier, M.; Luong, J. H. T.
J. Am. Chem. Soc. 2004, 126, 7176–7177.
decrease of the �-potential of the PDDA-AuNPs composite, i.e.,
the EOF is reduced. Notice that TRP also becomes more
neutral at pH 5, whereas the charges of the catechoamines and
indolamines were highly negative and such analytes were
expected to interact strongly with PDDA. Indeed, all catecholamines
and indoleamines were not eluted after 24 min into the
experiment. In contrast, Zhang et al. 28 reported that the
separation using the PDDA coated capillary is significantly
longer than the one modified with PDDA-AuNPs for analysis
of heroin and impurities. In such a study, the separation is
performed at pH 5.2 with addition of 3% methanol into the
running buffer consisting of 120 mM ammonium acetate.
Therefore, it was somewhat difficult to compare the result
obtained in this work with that of Zhang et al. 28
Other Optimal Conditions and Sample Stacking. The 50
mM phosphoric-Tris buffer pH 3 appeared to provide the highest
detection sensitivity with good separation resolution. Indeed,
4-hydroxy-3-methoxybenzylamine comigrated with 3,4-dihydroxybenzylamine
when the buffer strength increased to 100 mM and
the run was longer with lower detection sensitivity (Figure 3A).
This pair was coeluted as one single peak if the separation was
carried out with the PDDA coated capillary. Thus, a running buffer
consisting of 50 mM phosphoric-Tris, pH 3 was used to optimize
the separation potential and the detection potential of the BDD
electrode. With respect also to detection sensitivity and separation
efficiency (N/cm), the electrokinetic injection of the sample at
-10 kV for 5 s was optimal compared to the results obtained at
shorter (3 s) or longer (7 and 10 s) injection times (Figure 3B).
Beyond 5 s, the resulting peaks were very broad with compromised
detection sensitivity and separation efficiency. In contrast,
3 s was not sufficient, as reflected by considerably lower peaks.
The BDD detection potential, poised at +1 V vs 3 M Ag/AgCl
provided highest detection sensitivity compared to the detection
performed at lower or higher applied potentials (Figure 3C). Both
resolution and detection sensitivity were severely deteriorated
when the separation was conducted at -15 kV and -20 kV,
compared to the run at -10 kV (Figure 3D). Linearity and LOD
obtained for 11 analytes are summarized in Table 2. The four fast
migrating analytes, IXS, VMA, HVA, and tryptophan exhibited a
LOD of 0.3 µM, whereas a higher LOD, ∼1 µM was obtained for
the slower migrating group.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6901

Figure 5. Analysis of urine sample using a running buffer consisting
of 50 mM H3PO4-Tris, pH 4. Separation voltage: -10 kV with an
injection time of 5sat-10 kV. Detection: BDD at +1.0 V vs Ag/
AgCl, 3 M NaCl. The urine samples were filtered using a 0.22 µm
Millipore filter and diluted in the injection buffer with different dilution.
(A) 10-fold dilution, pristine and spiked urine samples were presented
as the lower and upper curve, respectively. Peak identification: (1)
IXS; (2) VMA; (3) AA; (4) HVA; (5) UA; and (6) TRP. The diluted
urine sample was spiked with 20 µM IXS, 20 µM VMA, 100 µM AA,
20 µM HVA, 100 µM UA, and 20 µM TRP. (B) 15-fold dilution, pristine
and spiked urine samples were presented as the lower and upper
curve, respectively. Peak identification: (1) IXS; (2) VMA; (3) AA; (4)
HVA; (5) UA; and (6) TRP. The diluted urine sample was spiked with
20 µM IXS, 20 µM VMA, 100 µM AA, 20 µM HVA, 100 µM UA, and
20 µM TRP.
For further LOD improvement, a series of experiments based
on field-amplified sample stacking (FASS) 30 was performed by
exploiting the conductivity difference between the sample zone
and the running buffer to effect preconcentration. A standard
solution of 6 analytes including ascorbic and uric acids, two
endogenous urinary compounds, was prepared in 10 mM H3PO4
pH 2 and injected at -10 kVfor 10 s with the separation
performed at -10 kV using 50 mM H3PO4 pH 4 as the running
buffer. It should be noted that at pH 3, ascorbic acid migrated
very closely with HVA, whereas uric acid was not baseline
separated from TRP. In principle, the amount of stacking is
proportional to the conductivity difference between the running
(30) Weng, Q.-F.; Xu, G.-W.; Yuan, K.-L; Tang, P J. Chromatogr. B 2006, 835,
55–61.
6902 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 6. Analysis of an urine sample using a running buffer
consisting of 50 mM H3PO4-Tris, pH 4. Separation voltage: -10 kV
with an injection time of 5sat-10 kV. Detection: BDD at +1.0Vvs
Ag/AgCl, 3 M NaCl. The urine sample was filtered using a 0.22 µm
Millipore filter and diluted 8-fold in the injection buffer. (A) with sample
stacking and (B) without sample stacking.
buffer and the sample solution. A preconcentration factor of 4
was observed for IXS, AA, and HVA compared to 3 for VMA
and UA and only 1.25 for TRP (Figure 4, curve b). With sample
stacking, the detection limit of IXS, AA, and HVA was 75 nM
compared with nonstacking (300 nM, Figure 4, curve a), significantly
lower than the physiological levels of these biomarkers in
urine as discussed later. Notice that peak-width was broadened
when the injection time was greater than 10 s (data not shown).
In another attempt, the capillary was first filled with 50 mM H3PO4
pH 4 followed by a sample injection as described above. The
cathodic running buffer was changed to 10 mM H3PO4, pH2
instead of 50 mM H3PO4, pH 4. In this case, the separation
window was narrower, resulting in the comigration of three
pairs: IXS/VMA, AA/HVA, and UA/TRP (Figure 4, curve c).
No further improvement in separation resolution was attained by
adding several organic modifiers including 10% methanol, 10%
acetonitrile, 5 mM methyl �-cyclodextrin, and 5 mM cetyl
trimethylammonium bromide in the running buffer.
Analysis of Biomarkers in Urine. The electropherogram of
an authentic urine sample obtained from a healthy female shows
5 low peaks and one very high peak. All biomarkers were detected
even if the urine sample was diluted to 10- and 15-fold, respectively
in the injection buffer (Figure 5). Standard IXS, HVA, VMA,

ascorbic acid (AA), uric acid (UA), and TRP were spiked into the
urine sample for peak identification. Dilution of the urine sample
also improved peak to peak separation and peak identification.
All peaks were identified and the sequence was assigned as IXS,
VMA, AA, HVA, UA, and TRP. The highest peak was attributed
to UA as expected since its normal concentration in urine ranges
from 1.23 to 3.7 mM (1.48-4.43 mmol/day with an average urine
volume of 1200 mL). 31 The UA peak was confirmed by matching
the migration time and spiking the urine sample with 30 µM uric
acid. Peak deconvolution and peak area were obtained using
WinPLOTR (Version: May, 2009, http://www.cdifx.univ-rennes1.fr/
winplotr/winplotr.htm). The concentration for each analyte was
estimated by comparing the peak areas of the authentic and spiked
urine samples. The average concentration of IXS, VMA, HVA, AA,
and UA for two different urine samples obtained from the same
healthy female was determined to be 170, 55, 87, 142, and 1,075
µM, respectively. Normal levels of such analytes in urine are
100-1000 µM for IXS, 32 50-1000 µM for VMA, 10 14-125 µM for
HVA, 10 and 150-200 µM for AA, 33 i.e., the values estimated by
CE-ECD fall in the normal range for healthy subjects. Therefore,
sample staking was not mandatory for the simultaneous analysis
of IXS, HVA, and VMA in the presence of UA, AA, and TRP since
the LOD of such analytes was significantly lower than their normal
physiological levels.
Sample stacking of the urine samples at low dilution (below
5-fold) did not improve the detection limits (figure not shown),
very likely due to a high level of salts in the urine sample. With
an 8-fold dilution, sample stacking provided a significant improvement
in the LOD with a sharper corresponding peak for each
analyte as shown in Figure 6A. In particular, sample stacking also
allowed for the detection of TRP, the last peak that migrated very
(31) Tietz, N. W. Fundamentals of Clinical Chemistry: W.B. Saunders Co.:
Philadelphia, 1987.
(32) Harlit, H. J. Biol. Chem. 1933, 537–545.
(33) Ridi, E.; Moubasher, M. S. R.; Hassan, Z. F. Biochem. J. 1951, 49, 246–
251.
(34) Matsuo, M.; Tasaki, R.; Kodama, H.; Hamasaki, Y. J. Inherit. Metab. Dis.
2005, 28 (1), 89–93.
(35) Luo, D.; Wu, L.; Zhi, J. ACS Nano 2009, 3 (8), 2121–2128.
closely to the uric acid peak. In addition, both HVA and VMA
were positively detected and identified compared with nonsample
stacking (Figure 6B). Thus, the method became useful for
estimation of the urine HVA/VMA ratio, a useful screening
method for Menkes disease. This ratio could range from 4.1 to
69.7 owing to impaired activity of dopamine �-hydroxylase, a
copper-dependent enzyme. 34
CONCLUSIONS
In brief, a novel scheme was described for electrophoretic
separation and detection of several important biomarkers in urine.
A fused silica capillary with a coating layer of PDDA and AuNPs
reversed the electroosmotic flow and served as a stable layer for
resolving the analytes. No electrode fouling was observed during
repeated analysis and even with the urine sample. The detection
limit obtained for IXS, HVA, and VMA was considerably below
their normal physiological levels in biological samples. The
method was simple and capable of measuring several important
biomarkers in urine samples without sample pretreatment with
excellent selectivity and detection sensitivity. Sample stacking
could be easily performed to improve detection limits of these
analytes in urine samples provided such samples were diluted
properly to reduce the level of uric acid and salts. Furthermore,
many other basic neurotransmitters and their acidic metabolites
could also be detected. Notice also that a BDD nanoforest
electrode (BDDNF) can be fabricated by hot filament chemical
vapor deposition. 35 This type of electrode exhibits improved
detection sensitivity compared to conventional planar BDD
electrodes. Integration of BDDSNF with capillary electrophoresis
is a subject of future endeavor.
ACKNOWLEDGMENT
The authors thank the Science Foundation Ireland (SFI) for
an SFI Walton Fellowship (JHTL), an IRCSET Embark Award
(LZ), and an SFI-SRC Grant for the Irish Separation Science
Cluster (ISSC).
Received for review April 27, 2010. Accepted July 5, 2010.
AC101105Q
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6903

Anal. Chem. 2010, 82, 6904–6910
Human Plasma Copper Proteins Speciation by Size
Exclusion Chromatography Coupled to Inductively
Coupled Plasma Mass Spectrometry. Solutions for
Columns Calibration by Sulfur Detection
Souleiman El Balkhi, †,§,| Joël Poupon,* ,† Jean-Marc Trocello, ‡,§ France Massicot, |
France Woimant, ‡,§ and Olivier Laprévote †,|
Laboratoire de toxicologie biologique, Service de Neurologie, Centre de Référence Pour la Maladie De Wilson,
AP-HP, Hôpital Lariboisière, 2, rue Ambroise Paré, 75475 Paris cedex 10, France, and ChimiesToxicologie
Analytique et Celullaire, EA 4463, Faculté de Pharmacie, Université Paris Descartes, 4, avenue de l’Observatoire,
75006 Paris, France
Among the hyphenated techniques used to probe and
identify metalloproteins, size exclusion chromatography
coupled to inductively coupled plasma mass spectrometry
(SEC-ICP-MS) has shown to have a central place. However,
the calibration of SEC columns reveals to be tedious and
always involves UV detection prior to ICP-MS. The presence
of sulfur in 98% of proteins allows their detection by quadrupole
ICP-MS, despite the isobaric interference ( 16 O 16 O) on
S, by monitoring 32 S 16 O at mass to charge ratio (m/z) 48.
The formation of SO occurs spontaneously in the argon
plasma but can be optimized by the introduction of oxygen
gas into a reaction cell (RC) to achieve nM levels. In this
article, sulfur detection was discussed upon instrumental
conditions and S detection was then optimized by applying
O 2 as a reaction gas. SO formation was used to calibrate
SEC columns without UV detection. This simple SEC-ICP-
MS method was used for plasma copper proteins in plasma
healthy subjects (HS) and an untreated Wilson disease
(WD) patient. Copper proteins identified in healthy subjects
were transcuprein, ceruloplasmin (Cp) and albumin. The
method led to results in good agreement with other methods
of determination. Copper bound to Cp in the WD
patient was lowered with regard to the HS, and the exchangeable
Cu was highly increased.
In the postgenomics era, proteomics has become a central branch
in life sciences. Information from proteomics (and peptidomics)
studies may reveal alterations in gene expression due to a disease
and facilitate the understanding of the metabolic pathways such as
copper metabolism in Wilson and Menkes diseases. 1 Since one-third
* Author for correspondence: Dr Joël Poupon, Laboratoire de Toxicologie
biologique, Hôpital Lariboisière, 2 rue Ambroise Paré, 75475 Paris cedex 10,
France, E-Mail: joel.poupon@lrb.aphp.fr, Phone: 33 1 49 95 66 00, Fax: 33 1 49
95 65 71.
† Laboratoire de toxicologie biologique.
‡ Service de Neurologie.
§ Centre de Référence Pour la Maladie De Wilson.
| Université Paris Descartes.
6904 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
of all proteins are metalloproteins, a new field recently defined by
Haraguchi, namely “metallomics”, has been introduced. This topic
attracted a fast growing interest during the past decade. It integrates
different analytical approaches that focus on metalloproteins and
metal-containing biomolecules. Metal ions in metalloproteins have a
regulatory and structural role allowing them to fulfill their different
functions. The biocatalysis of some specific enzymatic reactions
including gene DNA synthesis, metabolism, antioxidation, and
detoxification are examples of these functions. 2,3
Since the separation, identification, and quantification of
metalloproteins are mandatory in our understanding of their
complex role in biological systems, a growing number of studies
are covering these fields. Usually, authors employ hyphenated
techniques combining a separation method followed by mass
spectrometry analysis. Matrix-assisted laser desorption/ionization
(MALDI) MS, electrospray ionization (ESI) MS, two-dimensional
polyacrylamide gel electrophoresis (2D-PAGE) laser ablation (LA)
ICP-MS and high performance liquid chromatography (HPLC)
ICP-MS with different modes of separation have been proposed.
More recently, capillary electrophoresis (CE) ICP-MS and Fourier
transform ion cyclotron resonance mass spectrometry (FT-ICR-
MS) have been used for this purpose. 3,4
The overwhelming majority of recent applications concerning
probing and quantification of metalloproteins were developed
using ICP-MS coupled to size-exclusion chromatography (SEC). 5-9
ICP-MS is the most widely used mode of detection because of its
extremely low limits of detection (LOD), a wide dynamic range,
(1) Wang, M.; Feng, W. Y.; Zhao, Y. L.; Chai, Z. F. Mass Spectrom. Rev. 2009,
29, 326–348.
(2) Haraguchi, H. J. Anal. At. Spectrom. 2004, 19, 5–14.
(3) Shi, W.; Chance, M. R. Cell. Mol. Life Sci. 2008, 65, 3040–3048.
(4) Szpunar, J. Analyst 2005, 130, 442–465.
(5) Cabrera, A.; Alonzo, E.; Sauble, E.; Chu, Y. L.; Nguyen, D.; Linder, M. C.;
Sato, D. S.; Mason, A. Z. Biometals 2008, 21, 525–543.
(6) Franca Maltez, H.; Villanueva Tagle, M.; Fernandez De La Campa Maria
Del, R.; Sanz Medel, A. Anal. Chim. Acta 2009, 650, 234–240.
(7) Gonzalez-Fernandez, M.; Garcia-Barrera, T.; Arias-Borrego, A.; Bonilla-
Valverde, D.; Lopez-Barea, J.; Pueyo, C.; Gomez-Ariza, J. L. Anal. Bioanal.
Chem. 2008, 390, 17–28.
(8) Lopez-Avila, V.; Sharpe, O.; Robinson, W. H. Anal. Bioanal. Chem. 2006,
386, 180–187.
(9) Sviridov, D.; Meilinger, B.; Drake, S. K.; Hoehn, G. T.; Hortin, G. L. Clin.
Chem. 2006, 52, 389–397.
10.1021/ac101128x © 2010 American Chemical Society
Published on Web 07/19/2010

the multielement capabilities, the continuous spectra signals
survey and a high sample throughput. 10 In addition, SEC has the
advantage, in theory, to preserve metalloproteins from solid
phase/analytes interactions. Maintaining the integrity of the
probed proteins is actually of a critical importance when metal
ions simply coordinate to organic ligands.
Nevertheless, the calibration of SEC columns presents some
drawbacks rending SEC-ICP coupling laborious because of (i) a
tedious installation of a UV detector in parallel to the ICP-MS,
(ii) the use of an additional software for the integration of
chromatographic peaks, and (iii) the increase of dead volume.
This way of SEC calibration involves the injection of high
concentrations of proteins. These are expensive and can irreversibly
bound to columns leading to columns clogging. It is
noteworthy that other detection techniques have been proposed
to calibrate SEC columns such as IR, NMR, MS, evaporative light
scattering, and viscometers 11 but their coupling to ICP has not
yet been reported.
The capability of ICP-MS to detect in a single run the metal
ions of interest and hetero nonmetallic elements present in
proteins such as S and P offers great opportunities. It allows a
specific detection, molecular characterization, and even quantification
of some metalloproteins. For instance, when the amino acid
(AA) sequence is known for a given protein, and hence the
number of AA containing S, the determination of S provides an
accurate value of molar protein concentration. 12,13
Unfortunately, these heteroelements (i.e., S and P) have high
ionization energies resulting in low sensitivities in ICP-MS.
Moreover, both S and P are extensively interfered by isobaric
polyatomic ions generated in the atmospheric pressure argon
plasma ( 16 O 16 O for 32 S and 15 N 16 O for 31 P). The use of high
resolution mass spectrometry systems such as sector field ICP-
MS with a resolution higher than 1 800 m/∆m 14-17 allows
distinguishing of these isobaric species. Alternatively, the detection
of S remains feasible with lower resolution (m/∆m 400) quadrupole
ICP-MS by using a reaction cell. This device is a quadrupole
filled with a reaction gas (oxygen or xenon) and placed between
the ion lenses and the quadrupole mass filter of the ICP-MS
instrument. Oxygen used as reacting gas in the reaction cell
converts S ions into 32 S 16 O. Sulfur is then measured as SO at
m/z 48 which shows less interferences. This technique was
previously used for the determination of some sulfured amino
acids. 18,19 In the same way, the measurement of phosphorylation
degree of some peptides 20 and the determination of metal-sulfur
ratios in some metalloproteins 12 were obtained by detecting S and
(10) Lobinski, R.; Moulin, C.; Ortega, R. Biochimie 2006, 88, 1591–1604.
(11) Striegel, A. M. Anal. Chem. 2005, 77, 104 A113 A.
(12) Hann, S.; Koellensperger, G.; Obinger, C.; Furtmuller, P. G.; Stingeder, G.
J. Anal. At. Spectrom. 2004, 19, 74–79.
(13) Profrock, D.; Leonhard, P.; Prange, A. Anal. Bioanal. Chem. 2003, 377,
132–139.
(14) Bettmer, J.; Montes Bayon, M.; Encinar, J. R.; Fernandez Sanchez, M. L.;
Fernandez de la Campa Mdel, R.; Sanz Medel, A. J. Proteomics 2009, 72,
989–1005.
(15) Rappel, C.; Schaumloffel, D. Anal. Bioanal. Chem. 2008, 390, 605–615.
(16) Wind, M.; Wesch, H.; Lehmann, W. D. Anal. Chem. 2001, 73, 3006–3010.
(17) Zinn, N.; Kruger, R.; Leonhard, P.; Bettmer, J. Anal Bioanal. Chem. 2008.
(18) Schaumloffel, D.; Giusti, P.; Preud’Homme, H.; Szpunar, J.; Lobinski, R.
Anal. Chem. 2007, 79, 2859–2868.
(19) Yeh, C. F.; Jiang, S. J.; Hsi, T. S. Anal. Chim. Acta 2004, 502, 57–63.
(20) Bandura, D. R.; Baranov, V. I.; Tanner, S. D. Anal. Chem. 2002, 74, 1497–
1502.
Pas 32 S 16 O + and 33 P 16 O + species (m/z 48 and 49, respectively)
together with metals of interest. More recently, Wang used
SEC-ICP-MS for a quantitative analysis of proteins via S
determination. 21
Interestingly, by using argon plasma, under atmospheric
condition, oxides are formed spontaneously at percentages corresponding
to ≈3% of measured elements. 22,23 These percentages
are influenced by instrumental parameters of ICP-MS. 24,25 Consequently,
in applications involving high concentrations of proteins,
such as undiluted or weakly diluted plasma, the amount of
produced SO could be sufficient for its detection. Moreover,
recommended concentrations of proteins in SEC column calibration
kits are around 1-10 g · L -1 (i.e., S concentration higher
than 50 µM) which is enough to monitor SO without any
addition of O2. In other words, the calibration of SEC columns
should be possible by using SEC-ICP-MS in standard mode
and without any prior UV detection.
In this study, we present a simple SEC-ICP-MS method for
plasma copper proteins speciation. This method was applied to
healthy subjects (HS) and an untreated Wilson disease (WD)
patient. The distribution of copper in the blood is still an important
subject of research. The knowledge of copper distribution contributes
to better understand copper metabolism and to better
diagnose and monitor related diseases. In WD, a mutation in the
gene ATP7B leads to a dysfunction of ceruloplasmin (Cp) which
is the major protein binding Cu. This binding is mediated by the
protein ATP7B that lacks in WD. Clinically, serum Cp concentration
diminishes and the so-called “free Cu” increases and becomes
toxic due to Cu deposits in target organs (liver, brain, kidney,
and eyes). If not treated, irreversible damages can occur. 26-28
For the speciation purpose, we demonstrate that SEC column
calibration can be conducted with a simple SEC-ICP-MS at
moderate m/z resolution and without any UV detection. ICP-MS
operational parameters influence on oxides formations in standard
mode (without O2) was studied. The method was then optimized
by the application of O2 as reaction gas in the dynamic reaction
cell (DRC). Excellent limits of detection (LODs) for S and Cu
were obtained after optimization. Copper proteins identification,
stability and quantification are explored.
EXPERIMENTAL SECTION
Reagents and Chemicals. Mobile phase (NH4NO3 200 mM)
was prepared daily by dissolving 16 g of NH4NO3 (Sigma
Chemical Co, St-Quentin Fallavier, France) in 1Lofultrapure
water Milli-Q (Millipore, Molsheim, France) and degassed by
vacuum filtration on a 0.22 µm Millipore filters.
For SEC column calibration, two SEC marker kits were used.
Kit 1 (obtained from Sigma) included: Blue dextran (2000 kDa),
thyroglobulin (670 kDa), apoferritin (443 kDa), �-amylase (200
(21) Wang, M.; Feng, W.; Lu, W.; Li, B.; Wang, B.; Zhu, M.; Wang, Y.; Yuan,
H.; Zhao, Y.; Chai, Z. Anal. Chem. 2007, 79, 9128–9134.
(22) Divjak, B.; Goessler, W. J. Chromatogr., A 1999, 844, 161–169.
(23) Du, Z.; Houk, R. S. J. Anal. At. Spectrom. 2000, 15, 383–388.
(24) Gray, A. L.; Williams, J. G. J. Anal. At. Spectrom. 1987, 2, 599–606.
(25) Vaughan, M. A.; Horlick, G. Appl. Spectrosc. 1986, 40, 434–445.
(26) Das, S. K.; Ray, K. Nat. Clin. Pract. Neurol. 2006, 2, 482–493.
(27) El Balkhi, S.; Poupon, J.; Trocello, J. M.; Leyendecker, A.; Massicot, F.;
Galliot-Guilley, M.; Woimant, F. Anal. Bioanal. Chem. 2009, 394, 1477–
1484.
(28) Linder, M. C.; Hazegh-Azam, M. Am. J. Clin. Nutr. 1996, 63, 797S–811S.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6905

Table 1. ICP-MS Parameters
nebulizer
glass SeaSpray concentric
(nominal flow:1 mL · min-1 )
spray chamber glass cyclonic baffled (50 mL)
nebulizer gas flow 0.92 mL min-1 auxiliary gas flow 1.20 mL min-1 plasma gas flow 15 L min-1 ICP RF power 1125 W
O2flow rate (DRC mode) 0.60 (0.1 - 1mL· min-1 )
RPQ 0.25 (0.05-0.7)
pulse stage voltage 800 V
analog stage voltage -1612.5 V
measured m/z
32 16 34 16 63 65 70 45 151 S O, S O, Cu, Cu, Ga Rh, Eu
scan mode peak hopping
dwell time per isotope 250 ms
kDa), alcohol dehydrogenase (150 kDa), bovine serum albumin
(BSA) (66 kDa), carbonic anhydrase (29 kDa). Human serum
albumin (HSA) (69 kDa) was needed for void volume (V0)
determination (see results). A second ready-to-use marker kit
(Kit 2) (Column Performance Check Std, Aqueous SEC 1,
Phenomenex Inc., Le Pecq, France) was used to control SEC
calibration containing bovine thyroglobulin (670 kDa), human
gamma globulin (150 kDa), ovalbumin (44 kDa), and human
myoglobin (17 kDa).
D-penicillamine (D-pen) (purity >97%), chosen as a source of
S, was obtained from Sigma. Cu, Ga, Eu, and Rh 1 g · L -1 standard
solutions (Inorganic Ventures, distributed by Analab, Bischeim,
France) were used to prepare working solutions by appropriate
dilution in the mobile phase. Ga, Rh, and Eu were tested as
internal standards (IS). All reagents were tested for copper
contamination before use. Milli-Q water was used to make
intermediate dilutions.
Instrumentation. Separation of proteins was performed via
SEC using a BioSep-SEC-S 2000 and Biosep-SEC-S 3000 (300 ×
7.8 mm both) columns from Phenomenex. A Series 200 Perkin-
Elmer chromatography system, composed of an injector and a
high pressure pump equipped of PEEK tubing, (Perkin-Elmer,
Courtaboeuf, France) was operating at a flow rate of 1 mL · min -1 .
The injection volume was 100 µL. The column effluent was
directly coupled to a1mL· min -1 glass SeaSpray concentric
nebulizer (Perkin-Elmer) and a 50 mL cyclonic Baffled spray
chamber from Perkin-Elmer. Oxygen (purity N48, Air Liquide
Santé, Puteaux, France) was used as the reaction gas. Detection
and quantification of elements was made by an ICP-MS Elan-
DRCe (Perkin-Elmer) and the Chromera software (Perkin-
Elmer) was used for the HPLC signal monitoring. Signal
intensity was given as counts per seconde (Cps). The ICP-MS
operation parameters are given in Table 1.
The two isotopes of Cu were used for copper detection and
quantification in all experiments because of the polyatomic
interference 40 Ar 23 Na that was observed on 63 Cu. The ion peak
at m/z 63 was attributed to pure Cu when ( 63 Cu/ 65 Cu) isotopes
ratio was 0.45 ± 5%.
DRC and Reaction Gas Flow Optimization. A10µg · L -1
solution of D-pen (source of S) was used for the optimization
of the oxygen reaction gas flow in the DRC. The optimal
flow was obtained by varying the O2 flow (from 0.1 to 1.0
mL · min -1 ) as so to attain the highest signal to background
ratio (S/BKG). However, the effect of O2 on the Cu signal
6906 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(Cps) had to be estimated. Therefore, a solution of D-pen
and Cu at 200 and 0.5 µM, respectively, was used to measure
SO and Cu signals by varying the oxygen gas flow in the
DRC.
SEC Column Calibration. Proteins were dissolved in the
mobile phase and 100 µL of the proteins solutions of Kit 1 were
injected on columns: apoferritin, �-amylase, alcohol dehydrogenase,
BSA and carbonic anhydrase at 10, 5, 4, 5, and 10 g · L -1 ,
respectively. These concentrations are recommended by the
manufacturer and are used for a UV detection to give A280 nm
≈ 1ona90× 1.6 cm column. In our study, the proteins were
detected by their sulfur content (i.e., 32 S 16 O + referred to as 48 SO
hereafter) by ICP-MS without any reaction gas in the DRC. A
calibration curve was calculated between of the known molecular
masses of proteins and their respective volume of elution
divided by the dead volume (Ve/V0). To determine the void
volume, a1g· L -1 solution of HSA was injected separately,
vortex mixed with blue dextran (Bdx at 3 g · L -1 ) and injected
again (see results for explanation).
After a 100 fold dilution of the previous proteins solutions in the
mobile phase, the same volume was injected while the optimal flow
of oxygen gas in the DRC was applied. The calibration curve was
calculated again with the new data. At last, to confirm the proteins
separation, two mixtures were prepared and injected: (1) the readyto-use
mixture and (2) a mixture of apoferritin, alcohol dehydrogenase,
BSA and carbonic anhydrase at 1, 15, 6, 15 g · L -1 , respectively.
Copper and Sulfur Calibration Curves. External calibrations
for Cu and S were performed by using five solutions containing
increasing amounts of Cu and D-pen. Concentrations were chosen
on the basis of expected amounts of metalloproteins in the plasma
samples (0.1-5.0 µM for Cu and 100-5000 µM for S). Ga was
added as IS (100 µg · L -1 ) to all calibration solutions. External
calibrations were, first, carried out by flow injection (FI) without
any column connected to the HPLC system. The calibrations
were operated in both standard (STD) and DRC modes. The
peaks’ areas of measured elements reported to IS areas were
tested for linearity. The same experiments were conducted
under HPLC condition by injecting the calibration points on
the Precolumn Biosep 2000.
Species Unspecific Calibration. For Cu and S quantification
in metalloproteins, the external calibration previously obtained by
FI was used for the quantification of all Cu and S peaks in the
chromatograms by using the peak areas. This is referred to as
the “species unspecific” mode of calibration. 12,13
Limits of Detection (LODs). LODs were calculated for each
measured elements in both modes (STD and DRC) using FI first
and the precolumn afterward. Sulfur and Cu LODs were calculated
as 3-fold the standard deviation (SD) of the baseline variations
for each elements on three blank injections.
Sample Preparation. Blood samples were collected on a Liheparin
Vacutainer tubes (ref. 365952, Becton- Dickinson, Le Pont
de Claix, France) centrifuged for 10 min at 2000g, diluted (1:3) in
a mobile phase solution containing 100 µg · L -1 of IS and injected
on the column within 15 min. For exchangeable Cu study,
determined and discussed in a previous work, 27 plasmas were
diluted (1:1) in a 3 g· L -1 EDTA solution, kept at room
temperature for one hour precisely and injected on the column
after a final dilution (1:2) in the mobile phase.

RESULTS AND DISCUSSION
Sulfur Detection by SO Signal in Standard Mode. Sulfur
is present in two naturally occurring amino acids, cysteine, and
methionine, together exhibiting a global abundance of about 5%
of all amino acids in eukaryotic proteins. Therefore sulfur is
present in the vast majority of proteins (>98%) and its direct
detection by ICP-MS constitutes an almost general screening
method for peptides and proteins. 14,15
The high first ionization potential (10.36 eV) of S 15 makes the
ionization efficiency in argon plasma to be less than 15% leading
to low S sensitivity in ICP-MS. However, when Sulfur is present
at concentrations sufficiently elevated its survey as 48 SO in
standard mode is still possible due to a spontaneous oxide
formation corresponding to ≈3% of the total element (i.e., SO
formation in normal atmospheric pressure argon plasma).
Attempts to increase SO signal in the standard mode by varying
RF power or nebulizer gas flow (not shown) in order to facilitate
oxides formation did not lead to satisfying results. This is in
disagreement with Divjak et al. who’s results found that cold
plasma increases CeO formation and expected that S would
behave the same. 22 Indeed, cold plasma and high nebulizer gas
flow were favorable for the formation of CeO in our experiments
as well. However, increasing gas flow and decreasing RF was
deleterious for both Cu and SO signals. The reaction between Ce
and O2 is highly exothermic and was selected by some
manufacturers as an example of a bimolecular reaction whose
rate is inversely proportional to temperature. 29 Unfortunately,
this is not the case of the reaction between S and O2 that seems
to be endothermic under atmospheric pressure.
SEC Column Calibration. Standard Mode. The separation of
proteins on a SEC column and their detection by 48 SO offer the
possibility to calibrate the column on the basis of their apparent
molar mass (Mr) vs their volume of elution. This strategy
permits to overcome the tedious installation of a UV detector
in parallel to the ICP-MS and to use only one data integration
software instead of two needed in general.
The proteins of the marker kit 1 at the recommended
concentrations were used to calibrate the Biosep 3000 column.
Respective Ve of reference proteins are shown in Figure 1. To
calculate SEC Calibration curve V0 is needed to plot molar
masses vs Ve/V0. However, V0 could not be determined by Bdx
alone because it does not have any measurable element by ICP-
MS. This obstacle could be sidestepped as follows: giving that
HSA easily binds to Bdx 30,31 we tried to produce what manufacturer
advice against (i.e., mixing Bdx with other proteins of the marker
kit). In fact, the injection of 100 µL ofBdxat2g· L -1 did not differ
from baseline for all measured elements even for 12 C. On the other
hand, HSA eluted at 11.3 min (namely Ve ) 11.3 mL as shown in
Figure 2). HSA was then dissolved directly ina3g· L -1 Bdx solution
(HSA ) 1g· L -1 ), incubated for one hour and injected on the
column. We noticed that HSA 48 SO peak disappeared while a new
peak at 7.31 min (namely Ve ) 7.31 mL) appeared. This peak
results from the binding of HSA to Bdx and allows measuring
the void volume V0. In such a way, all data needed to calculate
the calibration curve are obtained (Figure 3).
(29) Baranov, V. I.; Tanner, S. D. J. Anal. At. Spectrom. 1999, 14, 1133–1142.
(30) Antoni, G.; Casagli, M. C.; Bigio, M.; Borri, G.; Neri, P. Ital. J. Biochem.
1982, 31, 100–106.
(31) Ponder, E.; Ponder, R. V. J. Gen. Physiol. 1960, 43, 753–758.
Figure 1. 48 SO chromatograms of proteins injected separately.
Elution volume of proteins: (1) apoferritin (10 g/L) Ve ) 10.45 mL, (2)
�-amylase (4 g/L) Ve ) 11.32, (3) alcohol dehydrogenase (5 g/L) Ve
) 11.64, (4) bovine albumin (10 g/L) 11.94 mL, (5) carbonic
anhydrase (30 g/L) Ve ) 12.9 mL, (6) cytochrome C (2 g/L) Ve )
13.71 mL. 1, 2, and 4 are on the left scale; 3, 5, and 6 are on the
right scale.
Figure 2. Void volume determination by interaction of Human
albumin and Blue dextran. 48 SO signal in STD mode for human
albumin (1 g/L) (blue line), and Human albumin bound to blue dextran
(1 g/L and 3 g/L, respectively) (red line).
The separation of proteins from Kits 1 and 2 was satisfying
(not shown). However, It is noteworthy that myoglobin peak is
hardly distinguishable from the baseline because of a poor
sulfured AA content (only two methionines and one cysteine) and
its coelution with low Mr sulfured residues present in all proteins.
Signals Optimization for SO Detection in DRC Mode. The
O2 flow offering the highest S/BKG for SO was found to be
0.6 mL · min -1 . At this optimal flow, Cu signal for both isotopes
decreases by about 20-30% compared to STD mode. Above
0.6 mL · min -1 O2 all signals decreased.
Sulfured Proteins in DRC Mode. By using O2 as a reacting
gas in the DRC, S detection was markedly improved to achieve
nanomolar levels (see after LOD). This implies the possibility
(1) to inject lower concentrations of proteins on the column
and so preserving it from irreversible proteins adsorption, (2)
to quantify precisely proteins even those with a low sulfur
content, (3) to use minor isotope 34 S by following 34 S 16 O( 50 SO)
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6907

Figure 3. SEC calibration curve between the known molar mass of
each protein and their respective volume of elution divided by the
dead volume (Ve/V0).
and finally, (4) to use S/Cu ratios as a specific tool to Cu
proteins detection since these ratios are theoretically known
(Table 2 adapted from refs 32 and 33). Actually, when S/Cu ratio
for a detected peak is higher than the theoretical value, proteins
coelution should be suspected.
Proteins injected separately could be identified by their
calculated molar masses on the calibration curve confirming the
validity of the column calibration. Figure 4 shows an example of
chromatograms of cytochrom C (2 g · L -1 ) and B-amylase (4
g · L -1 ). 48 SO signals are 1000 fold higher after O2 introduction
and 50 SO becomes readily measurable offering a very wide
dynamic range for S measurement.
Copper and Sulfur Calibration Curves. By flow injection
as well as under HPLC conditions, linearity was found for both
Cu isotopes and in both modes (STD and DRC). 48 SO in STD
mode was linear up to 10 000 µM and only to 2000 µM inDRC
mode because of signal saturation. 50 SO linearity test was only
possible in the DRC mode because it was not detectable in
standard mode. The possibility to measure both isotopes
permits to choose one or other depending on S concentrations
in measured samples. All coefficients of variations (CV) of three
injections of each of the calibration points in the same day were
less than 3% and the coefficients of correlation (r 2 ) were higher
than 0.99 for calibrated elements. In ICP techniques, internal
standard is used to compensate all kind of instrumental drifts.
The use of Ga as internal standard improves r 2 to reach 0.999
and higher giving equal slopes for both Cu isotopes in STD
and DRC modes. Rh was abandoned as internal standard
because it binds to plasma proteins and Eu revealed to be less
sensitive to the instrumental drift than Cu and S. None of the
tested proteins could be used as source of S for calibration
because of the sulfured residues of small molar mass detected
in all protein samples. In addition, manufacturer did not
mention the purity degree of the used proteins. This explains
the lack of accuracy that we observed when we tried to quantify
them by the obtained calibration curves.
(32) Manley, S. A.; Byrns, S.; Lyon, A. W.; Brown, P.; Gailer, J. J Biol. Inorg.
Chem. 2009, 14, 61–74.
(33) Michalski, W. P. J Chromatogr., B 1996, 684, 59–75.
6908 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Limits of Detection. The introduction of O2 in the DRC and
the use of a chromatographic support are sensitive parameters
influencing LODs of Cu and S and had to be estimated. LODs
of 65 Cu, 63 Cu, 48 SO, and 50 SO are given in Table 3. 48 SO LOD
turned out to be about 27 000 nM in standard mode and about
264 nM in DRC mode. This confirms that the detection of
48 SO in standard mode is still possible when S concentrations
are sufficiently high. On the other hand, when 48 SO saturates
in DRC mode (as it is the case of weakly diluted plasmas) 50 SO
could be used to monitor and measure sulfur despite its high
LOD.
Plasma Copper Proteins Analysis. Detection. Three healthy
subjects plasmas were injected pure or diluted (1:3) on the
columns. Neither Biosep 2000 nor Biosep 3000 were able to
separate ceruloplasmin copper (Cp-Cu) from albumin copper
(Alb-Cu) peaks. Only three peaks were observed in these
plasmas: the first one corresponded to transcuprein copper
(TC-Cu) (270 kDa), the second to (Cp-Cu) (132 KDa) coeluted
with (Alb-Cu) and the last one corresponded to low molar mass
Cu containing molecules (LMr). Therefore, the two columns were
connected in series to improve the separation. The connected
columns were then calibrated as previously described (in DRC
mode by monitoring 48 SO signal of different proteins at concentrations
10-100 fold lower than recommended). Excellent
separation between the Cp-Cu and the Alb-Cu peaks was
obtained (Figure 5A). The following peaks were detected:
TC-Cu, a major peak of Cu-Cp, Alb-Cu, and one LMr peak. It
should be noted that the obtained molar masses are relative to
the SEC column calibration and further analysis are needed to
clearly identified the detected proteins. Transcuprein, for instance,
was recently sequenced and its molar mass determined, 34 but one
should be aware that the retention volume is not sufficient to
identify this protein.
Manley et al. 32 have been able to detect Cu coagulation factor
Va (FVa-Cu), TC-Cu, Cp-Cu, Alb-Cu, and LMr-Cu in freshly
sampled rabbit plasma by injecting a large quantity (500 µL) of
undiluted plasma. The concentrations of FVa-Cu and Alb-Cu
they found were surprisingly abnormal as noted by authors. In
order to detect FVa-Cu in human plasma, we sampled blood on
a Li heparin tube (LiH 65 UI) to which 50 UI of extra LiH were
added (in order to prevent any possible coagulation to which FVa
participates). One hundred microliters of pure plasma were then
injected immediately after centrifugation (within 10 min). No peak
corresponding to the FVa-Cu was detected despite the immediate
injection. This is not surprising because of the very low theoretical
concentration in human plasma (FVa-Cu ) 30 nM approximately).
Conversely, in the majority of published studies working
on Cu proteins speciation, tris-hydroxyl-methyl-aminomethane
(Tris) or acetate buffers were used. We noted that these mobile
phases either precipitate ionic Cu or prevent it to be eluted form
SEC columns. For instance, injection of CuNO3 or CuCl2 solutions
(1-5 µM) on the tested columns was not reproducible at all
when Tris or acetate buffers were used. In addition, the
injection of EDTA 3 g/L after the injection of plasma or a simple
aqueous Cu solution provided an important and variable peak
of Cu confirming that SEC columns are able to bound Cu when
inappropriate mobile phases are used. This is in concordance
with Inagaki et al. and Meng et al. observations. 21,35 To avoid

Table 2. Molecular Properties and Concentrations of Cu Proteins and Stoichiometric S/Cu Ratio (Adapted from Refs
31 and 32
protein
molar mass
(kDa)
number of Cu bound
per protein
this, Inagaki choose to pretreat plasmas on a chelating resin before
SEC analysis to be sure that loosely bound Cu is captured.
Plasma from the untreated Wilson disease patient was also
tested. Despite a retention time shift that was observed, due
to columns loss of efficiency, three major peaks were detected:
Cp-Cu, Alb-Cu, and low molar mass Cu complexes (Figure
5B). The loss of efficiency, is a well-known phenomena that
occurs when SEC column performance deteriorates after a large
number of injections or after injection of highly concentrated
samples.
Copper Proteins Quantification. In ICP-MS, a simple
inorganic compound of an element can calibrate the element in
all kind of proteins because the response is independent of the
element environment. Therefore, the conventional calibration
methods (internal and external standardization and standard
additions) can be used for protein quantification but only if the
analytes have been identified. 1,4 This is referred to as the species
unspecific mode of calibration. 12,13
Koellensperger et al demonstrated that isotope dilution remains
the method of choice for species unspecific quantification in
hyphenated ICP-MS techniques although requiring mathematical
approximations. 36 They explained, however, that flow injection
yielded acceptable uncertainties and could be used for quantifica-
(34) Liu, N.; Lo, L. S.; Askary, S. H.; Jones, L.; Kidane, T. Z.; Trang, T.; Nguyen,
M.; Goforth, J.; Chu, Y. H.; Vivas, E.; Tsai, M.; Westbrook, T.; Linder, M. C.
J. Nutr. Biochem. 2007, 18, 597–608.
(35) Inagaki, K.; Mikuriya, N.; Morita, S.; Haraguchi, H.; Nakahara, Y.; Hattori,
M.; Kinosita, T.; Saito, H. Analyst 2000, 125, 197–203.
number of sulfur
containing AA
S/Cu
ratio
plasma proteins
concentration
maximum Cu
concentration (nM)
blood coagulation factor V 330 1 17 17 10 mg/L 30.3
transcuprein 190 0.5 56 112 180 µg/L 0.47
ceruloplasmin 132 6 40 6.7 0.2-0.6 g/L 9090-27 272
albumin 66 0-1 42 42 36-53.6 g/L
Cu, Zn-SOD 32 0-2 0.014-0.021 mg/L
Figure 4. 48 SO and 50 SO chromatograms of separate injections of
�-amylase (4 g/L) and cytochrome C (2 g/L) in DRC mode. (1) 48 SO
�-amylase; (2) 50 SO �-amylase; (3) 48 SO cytochrome C; (4) 50 SO
cytochrome C; (5) sulfured residues.
tion. In our study, no difference was observed between the
calibration curves obtained by flow injection and those obtained
by injection on the precolumn, allowing the quantification of all
species by using the daily flow injection calibrations.
Ceruloplasmin coeluted with other sulfured proteins making
its quantification impossible for healthy subjects and WD. Nevertheless,
Alb was well separated and could be measured by its
50 SO peak and found at 47.3 g · L -1 (716 µM) to be compared
with 49.7 g · L -1 obtained by a routine immunochemical analysis.
Calculated concentrations of Cu species in a healthy subject
showed that Transcuprein, Ceruloplasmin, Alb and LMr bound
0.4%, 87.5% 6.0%, and 6.0% of total Cu, respectively, (namely 0.08,
14.7, 1.0, and 1.0 µM) giving a total Cu in good agreement with
the concentrations found by atomic absorption spectrometry
(AAS) (16.8 vs 15.5 µM). For the WD patient, total Cu was at 4.4
µM by HPLC-ICP-MS vs 3.6 µM by AAS and included: 14.9%
Cp-Cu, 12.4% Alb-Cu, and 72.7% of LMr-Cu (namely 0.66, 0.55,
and 3.2 µM, respectively).
Unexpectedly, Cu-LMr represents 6% or more of total Cu in
the healthy subject plasma. In our previous study, 27 normal values
of ultrafilterable Cu (CuUF) were established (obtained by the
ultrafiltration of 44 healthy subjects plasmas on a 30 kDa cutoff
filter) and we know that this fraction does not exceed 0.15 µM in
healthy subjects (less than 1% of total Cu). This means that only
0.15 µM of Cu do not bind proteins having molar mass greater
than 30 kDa. We have described in this previous work that loosely
bound Cu is the copper fraction complexed to Alb and/or other
proteins that can be mobilized in the presence of high copper
affinity chelators, such as EDTA. To quantify this exchangeable
Cu (CuEXC) fraction the most useful ETDA concentration was 3
g/L with incubation time of one hour (time found necessary to
reach exchange equilibration). Quantification of CuEXC in healthy
subjects offered normal values ranging from 0.6 to 1.1 µM.
Interestingly, the LMr fraction of the healthy subject (1 µM Cu)
lies within this reference range and is likely to correspond to
CuEXC. Similarly, the Cu-LMr fraction found in the WD patient
(3.2 µmol/L) is in good agreement with the CuEXC fraction
determined by EDTA-ultrafiltration (2.8 µmol/L).
To confirm this, we injected the plasmas after one hour
incubation with EDTA 3 g/L (1:1) in order to probe the CuEXC.
No difference was observed between the chromatograms before
and after incubation confirming that loosely bound Cu is released
on the column after injection and further elutes in the low Mr
range. This loosely bound Cu is originally Alb copper and the
(36) Koellensperger, G.; Hann, S.; Nurmi, J.; Prohaska, T.; Stingeder, G. J. Anal.
At. Spectrom. 2003, 18, 1047–1055.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6909

Table 3. LODs in nM for Measured Elements in Different Operating Conditions (NM: Not Measurable)
flow injection flow injection precolumn precolumn
measured masses STD mode DRC mode SEC STD mode SEC DRC mode
63Cu 4.8 5.8 13.0 20.0
65Cu 6.8 7.6 17.0 27.0
48SO 27 077 264 55 760 747
50SO NM 1816 NM 5241
Figure 5. (A) chromatogram of diluted (1:3) healthy subject plasma
and (B) Wilson disease untreated patient plasma by using Biosep
3000 and Biosep 2000 connected in series. Cu peak (blue line): (1)
TC-Cu; (2) Cp-Cu; (3) Alb-Cu; (4) LMWM-Cu. 50 SO peak (red
line): (5) unknown protein(s); (6) albumin.
type 1 copper from ceruloplasmin reported to be as labile
copper. 37
(37) Musci, G.; Fraterrigo, T. Z.; Calabrese, L.; McMillin, D. R. J. Biol. Inorg.
Chem. 1999, 4, 441–446.
6910 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Even if SEC columns are known to be neutral toward the
analytes/solid phase interactions, our results show that weakly
bound Cu by unspecific coordination with proteins such as
albumin and the type 1 Cu of ceruloplasmin can be partially
released. SEC columns are neutral toward proteins but not toward
metals of the proteins if it is not structurally protected. The copper
atoms in our case only coordinate to amino acids of the protein
and can readily be captured by remaining silanol groups in this
silica based column. Upon the profile obtained from the WD we
better understand the deposits of Cu in target organs which is
unlikely to happen in healthy subjects because their loosely bound
Cu is much lower. We believe that this fraction represents the
copper that proteins can deliver and that its quantification is useful
in WD.
CONCLUSION
This method is designated to be applied to a larger number of
WD patients in order to use it for diagnostic purpose. Moreover,
a possible correlation between the loosely bound Cu and the
clinical state should be explored in WD. The same approach could
be undertaken for other metalloproteins where metal ions are
bound by coordination such as zinc proteins. At last, the sulfur
monitoring improved tremendously our proteins analysis method.
It allowed the SEC column calibration and offered a wide dynamic
range for S determination, and hence proteins quantification, by
offering the possibility to use both S isotopes.
ACKNOWLEDGMENT
We thank the Association Bernard PEPIN pour la Maladie de
Wilson for financial support.
Received for review April 29, 2010. Accepted July 7, 2010.
AC101128X

Anal. Chem. 2010, 82, 6911–6918
Direct Quantification of Single-Molecules of
MicroRNA by Total Internal Reflection
Fluorescence Microscopy
Ho-Man Chan, † Lai-Sheung Chan, ‡ Ricky Ngok-Shun Wong, ‡ and Hung-Wing Li* ,†
Department of Chemistry, Hong Kong Baptist University, Kowloon Tong, Hong Kong, P.R. China, and Department of
Biology, Hong Kong Baptist University, Kowloon Tong, Hong Kong, P.R. China
MicroRNAs (miRNAs) express differently in normal and
cancerous tissues and thus are regarded as potent cancer
biomarkers for early diagnosis. However, the short length
and low abundance of miRNAs have brought challenges
to the established detection assay in terms of sensitivity
and selectivity. In this work, we present a novel miRNA
detection assay in single-molecule level with total internal
reflection fluorescence microscopy (TIRFM). It is a solution-based
hybridization detection system that does not
require pretreatment steps such as sample enrichment
or signal amplification. The hsa-miR-21 (miR-21) is
chosen as target miRNA for its significant elevated content
in a variety of cancers as reported previously. Herein,
probes of complementary single-stranded oligonucleotide
were hybridized in solution to miR-21 and labeled with
fluorescent dye YOYO-1. The fluorescent hybrids were
imaged by an electron-multiplying charge-coupled device
(EMCCD) coupled TIRFM system and quantified by
single-molecule counting. This single molecule detection
(SMD) assay shows a good correlation between the
number of molecules detected and the factual concentration
of miRNA. The detection assay is applied to quantify
the miR-21 in extracted total RNA samples of cancerous
MCF-7 cells, HepG2 cells, and normal HUVEC cells,
respectively. The results agreed very well with those from
the prevalent real-time polymerase chain reaction (qRT-
PCR) analysis. This assay is of high potential for applications
in miRNA expression profiling and early cancer
diagnosis.
MicroRNAs (miRNAs) are categorized as a class of small
noncoding RNAs with approximately 19-23 nucleotides. It serves
as the gene expression and cell development regulator in animals,
plants, and viruses by interfering protein synthesis. 1,2 Evidences
indicated that miRNAs play important roles in cell proliferation,
differentiation, and apoptosis. 2-4 Recent researches have also
explored the differential expression levels of miRNAs in cancerous
* Corresponding author: (phone) +852-3411-7065; (fax) +852-3411-7348;
(e-mail) hwli@hkbu.edu.hk.
† Department of Chemistry, Hong Kong Baptist University.
‡ Department of Biology, Hong Kong Baptist University.
(1) Ambros, V. Cell 2003, 113, 673–676.
(2) Ambros, V. Nature 2004, 431, 350–355.
(3) Carthew, R. W. Curr. Opin. Genet. Dev. 2006, 16, 203–208.
(4) Nakahara, K.; Carthew, R. W. Curr. Opin. Cell Biol. 2004, 16, 127–133.
and noncancerous tissues and displayed its roles as tumor
suppressors or oncogenes. 5-7 Tumor formation caused by miR-
NAs is notable since it has been proven that a single sequence of
miRNA can regulate multiple gene targets. In the meanwhile, a
single gene target can also be regulated by multiple miRNAs.
Therefore, on account of the connections between miRNA and
cancer development, profiling of miRNA expression levels is
proposed as a vital tool for the preliminary diagnosis and prognosis
of cancer. 8
A majority of the miRNA detection methods are based on the
approach of hybridization, in which target molecules of interest
arecapturedorhybridizedwiththecomplementaryoligonucleotides. 8,9
Hereafter, the frequency of hybridization events are presented in
form of measurable signals for the quantification of miRNA.
However, the detection of the small size and trace amount of
miRNAs is always challenging. Northern blotting is the most
prevalent and accredited method for the quantification of miR-
NAs. 10 The technique allows multiplex detection, but it is laborious
and sample intensive. Another convincing technique for the
determination of miRNA is the real-time polymerase chain reaction
(qRT-PCR). 11-13 With PCR, copies of specific miRNAs and their
corresponding signals are amplified and intensified, respectively.
Nonetheless, the short length of target miRNA and intricate design
of primer limited the reliability of amplification and labeling. 8
Other amplification strategies have also been proposed in recent
years aiming at higher detection sensitivity, specificity, and lower
consumption of starting materials. For instance, nanoparticles such
as OsO2, 14 quantum dots, 15 and gold 15,16 are chemically labeled
on detecting probes as signal transducers and amplifiers. Besides,
(5) Calin, G. A.; Croce, C. M. Nat. Rev. Cancer 2006, 6, 857–866.
(6) Esquela-Kerscher, A.; Slack, F. J. Nat. Rev. Cancer 2006, 6, 259–269.
(7) Volinia, S.; Calin, G. A.; Liu, C. G.; Ambs, S.; Cimmino, A.; Petrocca, F.;
Visone, R.; Iorio, M.; Roldo, C.; Ferracin, M.; Prueitt, R. L.; Yanaihara, N.;
Lanza, G.; Scarpa, A.; Vecchione, A.; Negrini, M.; Harris, C. C.; Croce, C. M.
Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 2257–2261.
(8) Wark, A. W.; Lee, H. J.; Corn, R. M. Angew. Chem., Int. Ed. 2008, 47,
644–652.
(9) Cissell, K. A.; Deo, S. K. Anal. Bioanal. Chem. 2009, 394, 1109–1116.
(10) Lagos-Quintana, M.; Rauhut, R.; Lendeckel, W.; Tuschl, T. Science 2001,
294, 853–858.
(11) Chen, C. F.; Ridzon, D. A.; Broomer, A. J.; Zhou, Z. H.; Lee, D. H.; Nguyen,
J. T.; Barbisin, M.; Xu, N. L.; Mahuvakar, V. R.; Andersen, M. R.; Lao, K. Q.;
Livak, K. J.; Guegler, K. J. Nucleic Acids Res. 2005, 33, 9.
(12) Raymond, C. K.; Roberts, B. S.; Garrett-Engele, P.; Lim, L. P.; Johnson,
J. M. RNA 2005, 11, 1737–1744.
(13) Schmittgen, T. D.; Jiang, J. M.; Liu, Q.; Yang, L. Q. Nucleic Acids Res. 2004,
32, e43.
(14) Gao, Z. Q.; Yang, Z. C. Anal. Chem. 2006, 78, 1470–1477.
10.1021/ac101133x © 2010 American Chemical Society 6911
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/23/2010

enzymatic amplification techniques such as RAKE assay, 17 bioluminescence-enzyme
labeling, 18 and rolling-circle amplification
(RCA) 19 are also reported for sensitivity enhancement. These
reported detection methods bring great benefit and improvement
to the sensitivity of miRNA profiling to certain extends. Nevertheless,
sample pretreatment and the modifications may result in loss
of samples throughout the multiple pretreatment steps such as
sample enrichment, labeling, and purification. The detection
sensitivity and reliability are hence hindered. Since cellular miRNA
concentration can be as low as 1000 molecules per cell, 20
deficiencies in sensitivity may result in unsuccessful quantification
of low-abundance miRNAs and, thus, false diagnostic result.
Therefore, an ultrasensitive miRNA-profiling assay without the
need of pretreatment is demanded.
Single-molecule detection (SMD) technique has been widely
applied in the behavioral study of individual biomolecules. 21,22
Scientific issues such as monitoring enzymatic kinetics, 23,24 DNA
adsorption/desorption behavior, 25,26 verification of nucleic acid
hybridization, 27 DNA mismatch discrimination, 28 protein and DNA
conformation dynamics, 29,30 and DNA mapping 31,32 have successfully
been accomplished with SMD. Besides, several groups have
also demonstrated the competence of SMD in the quantitation of
biomolecules such as proteins 33 and viral DNA 34,35 by singlemolecule
counting. Recently, quantitation of miRNA in singlemolecule
level was displayed by Neely and co-workers. 36 The
novel miRNA detection assay is free of sample enrichment and
(15) Liang, R. Q.; Li, W.; Li, Y.; Tan, C. Y.; Li, J. X.; Jin, Y. X.; Ruan, K. C. Nucleic
Acids Res. 2005, 33.
(16) Yang, W. J.; Li, X. B.; Li, Y. Y.; Zhao, L. F.; He, W. L.; Gao, Y. Q.; Wan,
Y. J.; Xia, W.; Chen, T.; Zheng, H.; Li, M.; Xu, S. Q. Anal. Biochem. 2008,
376, 183–188.
(17) Nelson, P. T.; Baldwin, D. A.; Scearce, L. M.; Oberholtzer, J. C.; Tobias,
J. W.; Mourelatos, Z. Nat. Methods 2004, 1, 155–161.
(18) Cissell, K. A.; Rahimi, Y.; Shrestha, S.; Hunt, E. A.; Deo, S. K. Anal. Chem.
2008, 80, 2319–2325.
(19) Cheng, Y.; Zhang, X.; Li, Z.; Jiao, X.; Wang, Y.; Zhang, Y. Angew. Chem.,
Int. Ed. 2009, 48, 3268–3272.
(20) Lim, L. P.; Lau, N. C.; Weinstein, E. G.; Abdelhakim, A.; Yekta, S.; Rhoades,
M. W.; Burge, C. B.; Bartel, D. P. Genes Dev. 2003, 17, 991–1008.
(21) Joo, C.; Balci, H.; Ishitsuka, Y.; Buranachai, C.; Ha, T. Annu. Rev. Biochem.
2008, 77, 51–76.
(22) Weiss, S. Science 1999, 283, 1676–1683.
(23) Li, H. W.; Yeung, E. S. Anal. Chem. 2005, 77, 4374–4377.
(24) Li, J. W.; Yeung, E. S. Anal. Chem. 2008, 80, 8509–8513.
(25) Kang, S. H.; Shortreed, M. R.; Yeung, E. S. Anal. Chem. 2001, 73, 1091–
1099.
(26) Li, H. W.; Park, H. Y.; Porter, M. D.; Yeung, E. S. Anal. Chem. 2005, 77,
3256–3260.
(27) Kang, S. H.; Kim, Y. J.; Yeung, E. S. Anal. Bioanal. Chem. 2007, 387,
2663–2671.
(28) Gunnarsson, A.; Jonsson, P.; Marie, R.; Tegenfeldt, J. O.; Hook, F. Nano
Lett. 2008, 8, 183–188.
(29) Cohen, A. E.; Moerner, W. E. Proc. Natl. Acad. Sci. U.S.A. 2007, 104,
12622–12627.
(30) Funatsu, T.; Harada, Y.; Tokunaga, M.; Saito, K.; Yanagida, T. Nature 1995,
374, 555–559.
(31) Chan, E. Y.; Goncalves, N. M.; Haeusler, R. A.; Hatch, A. J.; Larson, J. W.;
Maletta, A. M.; Yantz, G. R.; Carstea, E. D.; Fuchs, M.; Wong, G. G.; Gullans,
S. R.; Gilmanshin, R. Genome Res. 2004, 14, 1137–1146.
(32) Xiao, M.; Phong, A.; Ha, C.; Chan, T. F.; Cai, D. M.; Leung, L.; Wan, E.;
Kistler, A. L.; DeRisi, J. L.; Selvin, P. R.; Kwok, P. Y. Nucleic Acids Res.
2007, 35, e16.
(33) Tessler, L. A.; Reifenberger, J. G.; Mitra, R. D. Anal. Chem. 2009, 81, 7141–
7148.
(34) Lee, J. Y.; Li, J. W.; Yeung, E. S. Anal. Chem. 2007, 79, 8083–8089.
(35) Li, J. W.; Lee, J. Y.; Yeung, E. S. Anal. Chem. 2006, 78, 6490–6496.
(36) Neely, L. A.; Patel, S.; Garver, J.; Gallo, M.; Hackett, M.; McLaughlin, S.;
Nadel, M.; Harris, J.; Gullans, S.; Rooke, J. Nat. Methods 2006, 3, 41–46.
6912 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
amplification, but continuous sample flow is needed for improvement
in sensitivity. This microfluidic-assisted fluorescence correlation
spectroscopy platform is also comparatively sophisticated.
In this article, we present a quantitative single-molecule
detection of miRNAs using total internal reflection fluorescence
microscopy (TIRFM). For the proof of concept, we have chosen
hsa-miR-21 (miR-21) as our detection target. The miR-21 is known
as one of the most significant miRNAs elevated in at least six types
of cancers including breast, colon, lung, pancreas, prostate, and
stomach cancers. 7 Studies have indicated the miR-21 mediated
tumor growth by serving as an oncogene 37 and targeting the
tumor suppressor genes such as TPM1 and PDCD 4 in invasion
and metastasis. 38-40 Compared to the conventional methods, the
developed assay here is straightforward because no pretreatment
steps are involved. Both the probe and target oligonucleotides
are free of chemical modifications. The YOYO-1 labeled miRNA
hybrids diffuse freely on unmodified coverslips and are monitored
by electron-multiplying charge-coupled device (EMCCD) under
TIRFM. TIRFM is a highly sensitive microscopic technique that
has been used for SMD in solution. The total internal reflection
(TIR) generates evanescent field layer that has a penetration depth
of about 100-300 nm depending on the incident angle of the
excitation laser beam. The excitation of fluorophores is confined
within the evanescent field layer such that background signal from
the bulk is greatly suppressed. Herein, the diffusing hybrids are
observed as single fluorescent spots when they enter the excitation
volume and are excited. Image of fluorescent molecules are
acquired for single-molecule counting. The counted number is
found to be proportional to the quantity of miRNAs in bulk
solution. The developed assay was also employed for the determination
of miR-21 in normal and cancerous cell lines and the
results were validated with that of qRT-PCR detection.
EXPERIMENTAL SECTION
Slide Pretreatment. All coverslips were prewashed prior to
experiments. Briefly, No. 1 22-mm square cover glasses (Gold
Seal, Electron Microscopy System, Hatfield, PA) were sequentially
sonicated for 30 min in household detergent, 30 min in acetone
(AR grade, Labscan), and 30 min in absolute ethanol. The slides
were then successively soaked for 30 min in Piranha solution
(H2SO4/30% H2O2) (v/v 1:1), rinsed with distilled water
extensively, sonicated for 30 min in HCl/30% H2O2/H2O (v/
v/v 1:1:1) solution, sonicated in distilled water for 15 min,
further sonicated for 30 min in Piranha solution, and finally
sonicated for 15 min in distilled water twice. The slides were
stored in distilled water and blow-dried with nitrogen before
use.
Preparation of Hybridization Buffers. A1× Tris-NaCl-EDTA
(TNE) buffer containing 20 mM pH 8.0 Tris-HCl (Invitrogen,
Carlsbad, CA), 1 mM EDTA, and various concentration (0, 50,
150, 250, and 500 mM) of sodium chloride was prepared with
DEPC-treated water (Ambion, Austin, TX) accordingly as the
(37) Si, M. L.; Zhu, S.; Wu, H.; Lu, Z.; Wu, F.; Mo, Y. Y. Oncogene 2007, 26,
2799–2803.
(38) Lu, Z.; Liu, M.; Stribinskis, V.; Klinge, C. M.; Ramos, K. S.; Colburn, N. H.;
Li, Y. Oncogene 2008, 27, 4373–4379.
(39) Zhu, S. M.; Si, M. L.; Wu, H. L.; Mo, Y. Y. J. Biol. Chem. 2007, 282, 14328–
14336.
(40) Zhu, S. M.; Wu, H. L.; Wu, F. T.; Nie, D. T.; Sheng, S. J.; Mo, Y. Y. Cell
Res. 2008, 18, 350–359.

dilution and hybridization buffer. The pH of the TNE buffers was
adjusted to 7.4 by addition of 1 M HCl dropwisely. The buffer
solution was then filtered through a 0.22 µm nylon membrane
filter, autoclaved, and photobleached with UV-C lamp overnight
prior to use.
Preparation of Probe and Target MicroRNA Oligonucleotides.
Commercial available LNA-modified oligonucleotide probe
(miRCURYLNAmicroRNADetectionProbe,Productno:38102-00)
(3′-TCAACATCAGTCTGATAAGCTA-5′) specific to hsa-miR-21
was purchased from Exiqon (Denmark). Two HPLC-purified
synthetic DNA oligonucleotides having the complementary sequence
to hsa-miR-21 (3′-TCAACATCAGTCTGATAAGCTA-5′)
and the mature sequence of hsa-miR-214 (3′-ACAGCAGGCACA-
GACAGGCAGU-5′) were custom-designed and obtained from
Invitrogen (Hong Kong) as the DNA probe and the negative
control of the experiments. Two Anti-miR miRNA inhibitors of
miR-21 (AM17000, Product ID: AM10206, 3′-CAACACCAGUC-
GAUGGGCUGU-5′), and anti-miR-21 (AM17000, Product ID:
AM12979, 3′-UAGCUUAUCAGACUGAUGUUGA-5′), were purchased
from Ambion, acting as the RNA probe and target miR-21
strands, respectively. All oligonucleotides were suspended in 1
µL of DEPC-treated water (Ambion) and further diluted into
appropriate concentration with 1× TNE buffer. The melting
temperatures (Tm) of the oligonucleotides were predicted using
the Probe Tm Predictor accessible online (www.exiqon.com)
based on the thermodynamic nearest neighbor model.
Optimization of Hybridization Conditions. Ionic Strength.
For the determination of optimum hybridization ionic strength,
sodium chloride concentrations from 0 to 500 mM in the TNE
buffer were used throughout the oligonucleotides dilution and
hybridization mixture preparation.
Selection of Probe. For the selection of optimal probe with the
best hybridization affinity toward miRNA, probes of DNA, RNA
and LNA were prepared and hybridized with target miR-21 as
described below. In addition, two control experiments including
probes only and negative control miR-214 were also performed.
Hybridization Time. For the determination of optimum hybridization
time, hybrids of same concentration and ionic strength
(250 mM NaCl) were incubated for 15 min, 30 min, 1 and 3 h
respectively.
Hybridization and Labeling of MicroRNA. The hybridization
and fluorescence labeling of miRNAs were performed according
to the following procedures. LNA probe, DNA probe, RNA
probe, miRNA target, and DNA negative control oligonucleotides
were diluted with 1× TNE buffers (pH 7.4) to 300 pM in
concentration, respectively. The hybridization cocktail contained
15 µL of 300 pM probe strand, 15 µL of appropriate concentration
of target miR-21 strand, and 14 µL of TNE buffer. The cocktail
was incubated in form of free-diffusing solutions in dry bath
(AccuBlock Digital Dry Bath D1100, Labnet, NJ) for 1 h. The
hybridization temperature was set to be 20 °C below the T m of
the probe, that is, 52, 47, and 36 °C for LNA, DNA, and RNA
probe, respectively. After incubation, 1 µL of 100 nM YOYO-1
Iodide (YOYO) (Invitrogen) was added subsequently to label
the hybrids. YOYO labels the hybrids in a dye to base pair
(dye/bp) ratio of 1:1. The mixture was kept for 5 min in order
to achieve equilibrium and 10 µL of solution was pipetted to
the precleaned coverslips for TIRF imaging.
External Calibration of MicroRNA. A calibration curve was
established to correlate number of detected single fluorescent
spots and concentration of miRNAs. Synthetic miRNA with final
concentration of 100, 75, 50, 25, 10, 5, and 1 pM target miR-21
strand was hybridized with probe strand of final concentration of
100 pM under optimal conditions as mentioned above.
Cell Culture. Human umbilical vein endothelial cells (HU-
VEC) were purchased from Lonza (Walkersville, MD). M199
medium, heparin, gelatin, and endothelial cell growth supplement
(ECGS) were purchased from Sigma. Fetal bovine serum (FBS)
and penicillin-streptomycin (PS) were purchased from Invitrogen.
Dulbecco’s modified Eagle medium (DMEM) and Roswell Park
Memorial Institute medium (RPMI) were purchased from Gibco
(Grand Island, NY). HUVEC were grown in M199 medium
supplemented with 20% heat-inactivated FBS, 20 µg/mL ECGS,
90 µg/mL heparin, 1% PS in 75 cm 2 culture flasks coated with
0.1% gelatin. HUVEC from passage 2-6 were used in this study.
In addition, human hepatocellular carcinoma (HepG2) was
cultured in DMEM supplemented with 10% FBS and 1% PS,
and invasive breast ductal carcinoma (MCF-7) was cultured in
RPMI supplemented with 10% FBS and 0.5% PS. All cells were
maintained in a humidified incubator at 37 °C with 5% CO2 and
80% relative humidity.
Total RNA Isolation. Total RNA of HUVEC, HepG2, and
MCF-7 were extracted using TRIzol Reagent (Invitrogen) according
to the manufacturer’s protocol. Briefly, the sample was lysed
and homogenized with TRIzol Reagent. Phase separation was
followed by addition of chloroform and centrifugation. RNA in the
aqueous phase was then recovered by precipitation with isopropyl
alcohol. The RNA pellet was washed with 75% ethanol and finally
redissolved in RNase-free water. The RNA quantity was determined
by measuring optical density at 260 nm using the NanoDrop
ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington,
DE), and the RNA quality was assessed by performing
agarose gel electrophoresis.
Quantification of miR-21 in Cells by Quantitative Real-
Time PCR (qRT-PCR). Complementary DNA (cDNA) was
generated from 10 ng of total RNA per 5 µL of gene specific
reverse transcription (RT) reaction by using reagents from
TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems,
Foster City, CA) and mature hsa-miR-21-specific RT primer
(5×) from TaqMan MicroRNA Assays (P/N: 4373090, Applied
Biosystems). Each RT reaction contained 1 mM dNTPs, 50 U
MultiScribe reverse transcriptase, 1× reverse transcription buffer,
and 3.8 U RNase inhibitor as well as 1× specific RT primer. The
reaction mixture was incubated in PTC-100 Programmable Thermal
Controller (MJ Research, Inc., Waltham, MA) for 30 min at
16 °C, 30 min at 40 °C, 5 min at 85 °C, and then held at 4 °C. The
cDNA sample was then amplified by PCR using TaqMan 2×
Universal PCR Master Mix (No AmpErase UNG) (Applied
Biosystems) and TaqMan Assay (20×) from TaqMan MicroRNA
Assays (P/N: 4373090, Applied Biosystems). Each PCR reaction
included 0.67 µL of RT product, 5 µL of TaqMan 2× Universal
PCR Master Mix, No AmpErase UNG, and 0.5 µLof20× TaqMan
Assay (a mix preformulated miRNA-specific forward PCR primer,
specific reverse PCR primer and miRNA-specific TaqMan MGB
probe). The reaction mixture was then run at 95 °C for 10 min,
followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min in
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6913

Figure 1. Schematic illustration of the hybridization-based TIRFM assay for the detection of single miRNA molecules in solution.
Bio-Rad iCycler, Version 4.006 (Bio-Rad Laboratories, Inc., Hercules,
CA). The PCR reactions were run along with no-template
control and RT-minus control. Data were analyzed by iCycler iQ
Optical System Software, Version 3.0a (Bio-Rad Laboratories), and
the miRNA expression level was measured using threshold cycle
(Ct) which is the cycle number at which the fluorescence
generated within a reaction crosses the threshold. The Ct values
were then converted to absolute amount using a standard curve
of mature miR-21. Six independent experiments were performed,
and each experiment was run in duplicate.
Establishment of miR-21 Calibration by qRT-PCR. A
mixture of synthetic RNA oligonucleotides from mirVana miRNA
Reference Panel v9.1 (Applied Biosystems) was used to generate
a standard curve for miR-21. The RNA oligonucleotides representing
mature miR-21 ranging from 10 -6 to 1 fmol were reverse
transcribed, also using TaqMan MicroRNA Reverse Transcription
Kit (Applied Biosystems) and mature miR-21-specific
TaqMan MicroRNA Assays (P/N: 4373090, Applied Biosystems).
The PCR amplification of the cDNA was then performed
using same materials as mentioned above. Three independent
experiments were performed, and each experiment was run
in duplicate. Calibration curve was shown in Supporting
Information.
Quantification of miR-21 in Cells by TIRFM. Total RNA
of HUVEC, HepG2, and MCF-7 was diluted to 150 ng/µL with
TNE buffer, respectively. Consequently, 7.5 µL of diluted total
RNAs were spiked into a mixture standard solution of miR-21 of
a final concentration of 0, 1, 2, 5, 10, and 15 pM and a LNA probe
with a final concentration of 100 pM, and finally diluted with TNE
buffer to a volume of 44 µL. The solution was incubated for 1hat
52 °C and followed by addition of 1 µL of YOYO dye as mentioned
previously. After equilibrium for 5 min, 10 µL (with 250 ng of total
RNA) of solution was pipetted to coverslips for TIRF imaging and
quantitation. The contents of miR-21 in the three cell lines were
estimated by standard addition method and the value of miR-21
quantified by the TIRFM platform was compared with the outcome
of the qRT-PCR method.
Imaging System. An inverted Olympus IX-71 microscope
(Olympus, Tokyo, Japan) was equipped with a high-numerical
aperture 60× oil-immersion objective (1.45 NA, PlanApo, Olympus)
as shown in Figure S1A. The sample coverslip was located under
the fused-silica Isosceles Brewster Prism (CVI Melles Griot,
Carksvadm, CA) and above the 60× objectives with immersion
oil (η ) 1.52, Nonfluorescence, Olympus) in between. A 488 nm
cyan laser (50 mW, CMA1-01983, Newport, NJ) was used as the
excitation source. The laser beam was first filtered with neutral
6914 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
density filter (FSR-OD 60, Newport), focused by cylindrical lens
(focal length, 150 mm; CVI Melles Griot), and eliminated with
the aid of pinholes before its entry to the prism with an incident
angle of approximately 66°. Evanescent field was generated by
the TIR of laser beam occurring at prism and was used to excite
the fluorescently labeled hybrid molecules. A band-pass emission
filter (D535/40 m Chroma Technology) was placed between the
objective lens and the EMCCD to collect emitted photons. A
uniphase mechanical shutter (model LS272, Vincent Associates,
Rochester, NY) and a driver (model VMM-T1, Vincent Associates)
were synchronized with the PhotonMax: 512B electron-multiplied
CCD camera (EMCCD, Princeton Instruments, Princeton, NJ) in
external synchronization mode and frame-transfer mode. The
mechanical shutter blocked laser beam when the camera was off
in order to reduce photobleaching. The ADC rate of the camera
was 10 Hz, exposure time was 50 ms, and the multiplication gain
was set at 4000, with the shutter driver set to 100 ms exposure
and 100 ms delay. Typically, an image series of 10 sequential
frames on 10 locations were acquired from a single slide. Images
were obtained with the WinSpec/32 software (Version 2.5.22.0,
Downingtown, PA) provided by Princeton Instruments.
Data Analysis. All captured images were analyzed with a
public-domain image-processing program Image J (version 1.43i,
NIH, Bethesda, MD). A region of interest (ROI) with 200 pixelssquare
at the center of the light spot with relatively even laser
intensity was selected for single molecule counting. Intensity of
images may also suffer while the shutter was triggered on or off.
Five subframes (frame 3 to frame 7) of the image series were
selected for analysis. The threshold for image acquirement was
chosen at a value of three times the standard deviation of the mean
intensity of the image. The image was then further processed with
the Analyze Particles function in Image J to determine the number
of single fluorescence particles computationally. The size of
particles was set at 2-10 pixels to reduce false positive signals
generated from noises. Number of spots in five frames was
counted separately and summed up (i.e., accumulation of spots
imaged within 250 ms). Consequently, the sum of spots from 10
image series of a single slide was averaged. All experiments were
done in triplicate and the error bars of charts shown in the Results
and Discussion refer to the standard error of mean of the triplicate
experiments unless specified.
RESULTS AND DISCUSSION
In this study, the miRNA detection is based on the hybridization
approach as illustrated in Figure 1. Complementary probes
of oligonucleotides were hybridized with the target miR-21 in

solution under appropriate incubation conditions. Compared with
surface-based hybridization that involves washing of excess
reagents, solution-based hybridization offers the advantage of
higher hybridization efficiency. The hybridized duplexes were
labeled with fluorescent dye YOYO-1 iodide (YOYO) for the singlemolecule
fluorescence detection. YOYO is an intercalating fluorescent
dye which electrostatically binds to the backbone of
oligonucleotide. 41-43 The binding affinity of YOYO dyes on
double-stranded oligonucleotide is very high (∼6 × 10 8 M -1 ) and
there is a fluorescence intensity enhancement of approximately
400-fold upon binding to DNA. 42 Since the binding mechanism
is based on geometrical insertion, neither dye molecules nor
the oligonucleotides have to be chemically modified during the
labeling process. The detection of miRNA is, thus, straightforward.
To improve the fluorescence signal intensity, the hybrids
(∼20 bp) were labeled with YOYO in the ratio of 1 dye
molecule/1 bp. After direct labeling of hybrids with YOYO,
microliters of sample solution were sandwiched in a pair of
precleaned coverslips with a solution depth of approximately
20 µm, and observed under EMCCD-TIRF microscope. The
cleanness of the coverslips was found to be very significant in
the assay as scattering and autofluorescence of any dirt and
stains on glass surface may result in false positive signals. It is
crucial to clean coverslips extensively before use.
YOYO-labeled miRNA hybrids were visualized by a singlemolecule
TIRFM imaging system (Figure S1A). A 488 nm laser
was used to excite the bound YOYO dyes. The laser beam with
an incident angle of approximately 66° was total-internal-reflected
at the glass/solution interface. The thickness of evanescent field
layer (EFL) generated by total internal reflection was calculated
to be ∼190 nm by d ) λ/(2π(η2 2 sin 2 θ - η1 2 ) 1/2 ), where d is the
penetration depth of the field, λ is the wavelength of the
excitation light in vacuum, η1 and η2 are the refraction indices
of the solution and glass slides, and θ is the angle of incidence.
For more homogeneous exciting laser intensity, a central
region of 200 × 200 pixels (53 × 53 µm 2 ) of the EMCCD image
was selected as the sampling area, and thus, the probe volume
is estimated to be 0.54 pL. When 100 pM of miRNA hybrids
was loaded on the coverslip, the theoretical number of observed
hybrid molecules existing in the sampling region was 100 pM
× (53 µm × 53 µm) × 190 nm × 6.02 × 10 23 molecules/mol )
33. 44 Figure S1B shows a typical TIRFM image of miRNA hybrids
acquired in single-molecule level. It is noted that molecules
undergo random diffusional motion in a nonimmobilized system. 45
The diffusion coefficient of the miRNA hybrids was calculated as
76 µm 2 s -1 in bulk solution. 46 However, it was showed that
molecular diffusion rate is much slower at the glass/solution
interface compared to the bulk because of the electrostatic
interaction between molecules and macroscopic glass surface
in microsized domain. 45,47-49 The fluorescence signal generated
(41) Cosa, G.; Focsaneanu, K. S.; McLean, J. R. N.; McNamee, J. P.; Scaiano,
J. C. Photochem. Photobiol. 2001, 73, 585–599.
(42) Gurrieri, S.; Wells, K. S.; Johnson, I. D.; Bustamante, C. Anal. Biochem.
1997, 249, 44–53.
(43) Rye, H. S.; Yue, S.; Wemmer, D. E.; Quesada, M. A.; Haugland, R. P.;
Mathies, R. A.; Glazer, A. N. 1992, 20, 2803–2812.
(44) He, Y.; Li, H. W.; Yeung, E. S. J. Phys. Chem. B 2005, 109, 8820–8832.
(45) Xu, X. H.; Yeung, E. S. Science 1997, 275, 1106–1109.
(46) Zhdanov, V. P. Mol. BioSyst. 2009, 5, 638–643.
(47) Xu, X. H. N.; Yeung, E. S. Science 1998, 281, 1650–1653.
Figure 2. Correlation between the number of observed miR-21
molecules and expected number of miR-21 in the sampling volume
(0.54 pL). The number of observed miR-21 was corrected as the net
number of hybrids (number of observed molecules - number of
observed in blank). The slope of the correlation curve is 0.80, which
indicates the assay has a hybridization efficiency of approximately
80%.
from the single molecules is spreaded as they diffuse and the size
of the fluorescence spots in the image is larger than the physical
size of molecules of interest (see also movie file in Supporting
Information). 45 Although molecules interact with the glass surface,
nonspecific adsorption of hybrid molecules on the surface of glass
slide was insignificant as both the oligonucleotides and the glass
surface are highly negative-charged at pH 7.4. Figure S2 shows
the histogram of residence time for each miRNA hybrids in 8
consecutive frames. Among the 114 molecules detected in the 8
frames, ∼ 80% of the molecules appeared and then disappeared
in a single frame; 11%, 4%, and 4% of the molecules stayed at the
same position for 2, 3, and 4 frames, respectively, and less than
1% retained for 7 frames and eventually desorbed from the surface.
In the TIRFM image, each fluorescent spot was regarded as a
single molecule as a linear correlation on the number of counted
fluorescence spots and the expected number of miR-21 calculated
from the corresponding concentration was established (R 2 )
0.991). Herein, hybridization is the main factor attributed to
the differences in observed and expected number of molecules
and its efficiency was approximated to be 80% from the slope
of the plot as shown in Figure 2.
Optimization of Hybridization Conditions. The stringency
of hybridization governs the detection sensitivity and selectivity.
It is crucial to optimize the hybridization conditions before
performing further detection. The effects of ionic strength,
selection of probes, and incubation time on the hybridization
efficiency were evaluated.
First, the effect of buffer ionic strength on hybridization
efficiency was studied (Figure 3A). In general, hybridization
affinity is improved at higher ionic strength as the electrostatic
repulsions between the negatively charged oligonucleotides can
be effectively shielded in the presence of salt. However, high ionic
strength may also result in aggregation of molecules. The
aggregates will be misinterpreted as an individual in SMD, and
(48) Lyon, W. A.; Nie, S. M. Anal. Chem. 1997, 69, 3400–3405.
(49) Isailovic, S.; Li, H. W.; Yeung, E. S. J. Chromatogr., A 2007, 1150, 259–
266.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6915

Figure 3. Optimization of miRNA hybridization conditions. (A) Effect
of ionic strength adjusted by NaCl on hybridization efficiency (number
of hybrid counts). (B) Performance of LNA, DNA, and RNA probe on
hybridization with complementary miR-21 and negative control of miR-
214. (C) Effect of incubation time on hybridization efficiency (number
of hybrid counts). The data depict the averages of three experiments,
and the error bars are the standard error of mean of the three trials.
as a consequence, the number of counts drops. Here, we adjusted
the ionic strength by varying the concentration of NaCl in 1× Tris-
NaCl-EDTA (TNE) buffer. Figure 3A shows the molecule counts
as a function of the concentration of NaCl in the hybridization
buffers for probes of locked nucleic acid (LNA), DNA, and RNA,
6916 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
respectively, with the same concentration of the target miR-21.
The experimental condition was considered optimum when the
highest number of molecule counts was obtained; indicating that
the largest number of individual hybrids was formed. A gradual
increase in the number of hybrids was observed from buffer
containing 0-250 mM NaCl and a decrease after the salt
concentration exceeds. We analyzed the pixel size of each
fluorescent spots of LNA/miRNA hybrids prepared in TE buffer
with 250 and 500 mM NaCl, respectively, as shown in Figure S3.
It is obvious that higher percentages of molecules with larger pixel
sizes (size >4 pixels) were found in the series of 500 mM NaCl
compared to that of 250 mM. This implies that increase in ionic
strength would induce aggregation among hybrids and cause
underestimation on the count of molecules in single molecule
level. Thus, it is concluded that the optimal NaCl concentration
was 250, 150, and 150 mM with respect to probes of LNA, DNA,
and RNA.
Oligonucleotide probe is commonly used as a reporter in
hybridization-based nucleic acid detection. Conventional miRNA
detection assays use DNA oligonucleotides as the capturing probe,
although the stability of DNA-miRNA duplex is not high.
Recently, several groups demonstrated the potential of LNAs as
an alternative to DNA probes. 8,9,50,51 LNA is a nucleic acid
analogue known as a mimic of RNA that has high binding affinity
to RNA molecules. The thermostability of the LNA-miRNA
duplex is significantly higher than that of unmodified DNA
probes. 52-55 Consequently, both the detection sensitivity and
hybridization discrimination efficiency are enhanced due to the
increase in binding affinity and stability of LNA-miRNA complex.
To assess the performance of different nucleic acid analogues of
probes in the detection of miRNA, the hybridization affinities of
LNA, DNA, and RNA probes toward (i) complementary miR-21
and (ii) negative control miR-214 (sequence showed in Experimental
Section) in free solution were investigated. As shown in
Figure 3B, LNA probe-based hybridization yields a 1.5-fold
enhancement in the molecule counts compared with those of DNA
and RNA probes, while the binding of LNA probes to negative
control of miR-214 was maintained at low level. This indicates that
LNA probe not only promotes higher capturing efficiency, but also
displays a good mismatch discrimination capability.
Incubation time also plays a role on the overall hybridization
efficiency. Counts of hybrids obtained after incubation for 15, 30,
60, and 180 min were shown in Figure 3C. It was observed that
the number of counts increased and reached maximum in the
first 60 min incubation and then reduced after 3hofincubation.
The reduction in counts after prolonged incubation may be due
to the structural conformation change of LNA strands as proven
by MALDI-TOF mass spectrometry analysis (see Supporting
(50) Castoldi, M.; Schmidt, S.; Benes, V.; Noerholm, M.; Kulozik, A. E.; Hentze,
M. W.; Muckenthaler, M. U. RNA 2006, 12, 913–920.
(51) Kloosterman, W. P.; Wienholds, E.; de Bruijn, E.; Kauppinen, S.; Plasterk,
R. H. A. Nat. Methods 2006, 3, 27–29.
(52) Bondensgaard, K.; Petersen, M.; Singh, S. K.; Rajwanshi, V. K.; Kumar, R.;
Wengel, J.; Jacobsen, J. P. Chem.sEur. J. 2000, 6, 2687–2695.
(53) Braasch, D. A.; Corey, D. R. Chem. Biol. 2001, 8, 1–7.
(54) Nielsen, K. E.; Rasmussen, J.; Kumar, R.; Wengel, J.; Jacobsen, J. P.;
Petersen, M. Bioconjugate Chem. 2004, 15, 449–457.
(55) Petersen, M.; Nielsen, C. B.; Nielsen, K. E.; Jensen, G. A.; Bondensgaard,
K.; Singh, S. K.; Rajwanshi, V. K.; Koshkin, A. A.; Dahl, B. M.; Wengel, J.;
Jacobsen, J. P. J. Mol. Recognit. 2000, 13, 44–53.

Figure 4. Standardization curve for the quantification of miR-21.
Different concentrations of miR-21 were hybridized with 100 pM LNA
probe in solution at 52 °C for 1 h. The data depict the averages of
three experiments, and the error bars are the standard error of mean
of the three trials.
Information, Figure S4). As a result, the hybridization time was
set to 60 min for later experiments in this study.
Quantification of Synthetic miR-21. Under the optimal
hybridization conditions of miRNAs as discussed above, a calibration
plot of molecules counts as a function of the target miR-21
concentration was constructed. Synthetic miRNA of 0-100 pM
was hybridized with 100 pM LNA probes in solution and labeled
with YOYO. By single-molecule counting on the series of images
acquired, a plot of the number of molecule counts as a function
of the concentration of synthetic miR-21 was obtained with a
coefficient of determination of 0.991 (Figure 4). The detection limit
of the assay was estimated to be 5 pM (i.e., 50 amol in 10 µL of
sample). The ultimate theoretic limit of single molecule detection
is to detect a sole molecule in the bulk sample solution sandwiched
between the glass slides. Ideally, the theoretic limit is calculated
to be 170 zM, with a single target molecule in 10-µL sample
solution. To achieve such an ultimate theoretic limit requires
sufficiently long sampling time until the single target molecule
enters the probe volume (0.54 pL) by chance and gets excited
and detected. Otherwise, regardless of the sampling time, the
concentration of sample solution should be ∼3 pM, so that a single
molecule always locates in the probe volume. To improve the
detection limit, several factors including the probe volume,
viscosity of sample solution, sampling time, photostability, and
Figure 5. Quantification of miR-21 contents in total RNA of (A) MCF-7, (B) HepG2, and (C) HUVEC cells by standard addition methods with
TIRFM. Synthetic miR-21 was spiked into matrix of total RNA and LNA probe. The data depict the averages of three experiments, and the error
bars are the standard error of mean of the three trials. (D) Comparison between the miR-21 contents in total RNA of HUVEC, HepG2, and
MCF-7 cells determined by qRT-PCR assay and single-molecule TIRFM assay.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6917

quantum efficiency of fluorescence dyes may be optimized as
reported by Nie and co-workers. 56
Quantification of miR-21 in Normal and Cancerous Cell
Lines. MCF-7 (Invasive breast ductal carcinoma cell line) and
HepG2 (human hepatocarcinoma cell line) are regarded as a
common breast and liver cancer cell line model, respectively, while
HUVEC (human umbilical vein endothelial cells) is regarded as
normal cell line. Herein, we employed the single-molecule detection
assay to quantify differential-expressing miRNAs in normal
and cancerous cell lines. Briefly, total RNA of each cell line was
extracted prior to the direct miRNA detection. The YOYO dye
has no selectivity toward oligonucleotides, mRNAs, tRNAs, and
other small RNAs in such environment. To eliminate the signals
drawn from the complex matrixes without supplementary pretreatments
on the cell samples, standard addition method is
adopted in the manner that synthetic target miR-21 is spiked to
the mixture of total RNA sample and probes. The original
concentration of miR-21 in the sample of total RNA of each cell
lines can eventually be obtained by extrapolation of the calibration
curve. Three independent standardization curves were prepared
for the three cell lines and the amounts of miR-21 in each of the
cell line were determined. All three calibration curves showed
strong correlation between the numbers of counted molecules and
the miRNA concentration with the coefficients of determination
for all three of them greater than 0.997 (Figure 5A-C). Since the
hybridization efficiency was ∼80% as mentioned previously, the
concentration of miR-21 in each cell line estimated by standard
addition method was multiplied by the efficiency conversion factor
of 1.25 such that the actual miR-21 contents in total RNA can be
obtained. The contents of miR-21 in total RNA of each cell line
were found to be 0.92, 0.43, and 0.32 amol/ng for MCF-7, HepG2,
and HUVEC, respectively.
For the purpose of result validation, we quantified the content
of miR-21 in the three cell lines using the same batch of cells by
qRT-PCR method. Quantitative RT-PCR is a technique commonly
adopted as a standard method for miRNA profiling. It is superior
to other detection assays for its high specificity and minute
amounts of starting materials used in the detection. Amplification
steps however are involved and it takes approximately 5 h for the
whole process. The output of qRT-PCR is usually expressed in
terms of fold-change and so the raw data is semiquantitative.
Additional calibration curve has to be established for quantitation
purpose. To compare, our SMD detection assay is relatively rapid
as it takes only 1hofsample incubation and promptly followed
by microscopic detection. The result is quantitative and obtained
by applying standard addition methods. The standardization curve
(56) Nie, S. M.; Chiu, D. T.; Zare, R. N. Anal. Chem. 1995, 67, 2849–2857.
6918 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
by qRT-PCR is shown in the Supporting Information (Figure S5).
Figure 5D displays the contents of miR-21 in MCF-7, HepG2, and
HUVEC cells determined by the SMD assays and those by qRT-
PCR with calibration standardization. The SMD result agrees very
well with the outcome of qRT-PCR. The high correlation with the
accredited qRT-PCR methods demonstrated that the pretreatmentfree
SMD system developed here is of high potential in profiling
expression of miRNAs in different cell lines and thus applicable
in early cancer diagnosis.
CONCLUSIONS
We developed a direct and amplification-free quantitative assay
of single miRNA molecules using solution-based hybridization
approach and fluorescence-based detection with TIRFM. This
assay is straightforward, rapid, and highly sensitive. The fluorescent
hybrids diffuse randomly in the refined detection volume.
Improvement on the limit of detection could be achieved by
increasing sampling time or immobilizing target fluorescent
hybrids onto the coverslips. As a proof of concept, the content of
miR-21 were determined in cancerous MCF-7 and HepG2 and
noncancerous HUVEC cell lines and the result agreed very well
with that of conventional qRT-PCR. Both the success in discriminating
differentially expressing miR-21 in different cell lines and
the high correlation with the conventional detection method
justified the potential of the system in application for future early
cancer diagnosis.
ACKNOWLEDGMENT
This work was fully supported by the Faculty Research Grant
of Hong Kong Baptist University (FRG/07-08/II-68) and grant
from the University Grants Council of the Hong Kong Special
Administrative Region, China (HKBU I/06C). We thank Dr.
C. K. C. Wong from the Department of Biology of HKBU for
providing the MCF-7 cells.
SUPPORTING INFORMATION AVAILABLE
Additional information includes video of miRNA diffusing in
probe volume, adsorption time analysis of single hybrid molecules,
pixel size analysis of hybrids prepared in TE buffer containing
250 and 500 mM NaCl, MALDI-TOF mass spectrum of LNA
strands, calibration curve of qRT-PCR. This material is available
free of charge via the Internet at http://pubs.acs.org.
Received for review April 30, 2010. Accepted July 14,
2010.
AC101133X

Anal. Chem. 2010, 82, 6919–6925
Electrochemical Modulation for Signal
Discrimination in Surface Enhanced Raman
Scattering (SERS)
Emiliano Cortés,* ,† Pablo G. Etchegoin,* ,‡ Eric C. Le Ru, ‡ Alejandro Fainstein, § María E. Vela, †
and Roberto C. Salvarezza †
Instituto de Investigaciones Fisicoquímicas Teóricas y Aplicadas (INIFTA), Universidad Nacional de La
Plata-CONICET, Sucursal 4 Casilla de Correo 16 (1900), La Plata, Argentina, The MacDiarmid Institute for Advanced
Materials and Nanotechnology, School of Chemical and Physical Sciences, Victoria University of Wellington,
PO Box 600, Wellington, New Zealand, Centro Atómico Bariloche and Instituto Balseiro, Comisión Nacional de
Energía Atómica and Universidad Nacional de Cuyo, (8400) San Carlos de Bariloche, Río Negro, Argentina
Electrochemical modulation to induce controlled fluctuations
in SERS signals is introduced as a method to
discriminate and isolate different contributions to the
spectra. The modulationswhich can be changed in potential
range, amplitude, and frequencysacts as a controllable
“switch” to turn on, off, or change specific Raman
signals which can then be correlated within the spectra
by different fluctuation analysis techniques. Principal
component analysis (PCA), either by itself or assisted by
fast fourier transform (FFT) prefiltering, are shown to
provide viable tools to isolate the different components
of the spectra. Electrochemical modulation provides,
therefore, a technique to study complex cases of coadsorption,
and resolve problems of spectral congestion in
SERS signals.
The links between surface-enhanced raman scattering (SERS) 1,2
and electrochemistry go all the way back to the discovery of the
effect in the 1970s. 3-5 Since then, electrochemical studies have
provided invaluable information on the details of the electronic
interaction of molecules with the underlying metal substrate
responsible for the SERS enhancement. Particularly important
have been studies where the origin of the so-called “chemical
enhancement” 6,7 in SERS can be discerned. From a more modern
point of view, SERS and electrochemistry have diversified into a
myriad of different areas that include the studies of coadsorption
* To whom correspondence should be addressed. E-mail: emilianocll@gmail.com
(E.C.), pablo.etchegoin@vuw.ac.nz (P.G.E.).
† Universidad Nacional de La Plata-CONICET.
‡ Victoria University of Wellington.
§ Comisión Nacional de Energía Atómica and Universidad Nacional de Cuyo.
(1) Le Ru, E. C.; Etchegoin, P. G. Principles of Surface Enhanced Raman
Spectroscopy and Related Plasmonic Effects; Elsevier: Amsterdam, 2009.
(2) Aroca, R. F. Surface-Enhanced Vibrational Spectroscopy; John Wiley & Sons:
Chichester, 2006.
(3) Fleischmann, M.; Hendra, P. J.; McQuillan, A. J. Chem. Phys. Lett. 1974,
26, 163–166.
(4) Jeanmaire, D. L.; Van Duyne, R. P. J. Electroanal. Chem. 1977, 84, 1–20.
(5) Albrecht, M. G.; Creighton, J. A. J. Am. Chem. Soc. 1977, 99, 5215–5217.
(6) Otto, A. Surface-Enhanced Raman Scattering: Classical and Chemical Origins;
Springer-Verlag: Berlin, 1984.
(7) Tian, Z. Q. Faraday Discuss. 2006, 132, 309.
(multiple species), biological redox system, 8,9 corrosion, surface
science, 10 fuel cells, electrocatalysis, electronic models for chargetransfer
processes on surfaces, 11-16 and single-molecule or single
hot-spot SERS, 17,18 to name only a few. 19,20 The simultaneous
presence of the laser field with electric potentials and/or currents
in the many different experimental configurations of substrates
with simple, multiple, or tip-like electrodes, together with the
variety of environmental variables (like pH or the exact nature of
the electrolyte) produce the vast diversity of experimental situations
found in the modern applications of SERS to electrochemistry.
Tian and co-workers 20 provide a lucid summary of recent
advances in SERS for electrochemical applications. It is argued
in ref 20 in fact, that electrochemical systems are among the most
complex one can study in SERS. The combination of electrochemical
processes at interfaces with SERS make these systems
particularly challenging, but undoubtedly very relevant to understand
fundamental aspects in real analytical and bioanalytical
applications. Last, but not least, there are several techniques under
development in SERS that combine the basic elements of
electrochemical experiments, even though they would not be
strictly speaking considered as such. Examples of the latter are
recent attempts to combine SERS and molecule transport proper-
(8) Murgida, D. H.; Hildebrandt, P. Chem. Soc. Rev. 2008, 37, 937–945.
(9) Hildebrandt, P.; Murgida, D. H. Bioelectrochemistry 2002, 55, 139.
(10) Cai, W. B.; Ren, B.; Li, X. Q.; She, C. X.; Liu, F. M.; Cai, X. W.; Tian, Z. Q.
Surf. Sci. 1998, 406, 9–22.
(11) Gersten, J. I.; Birke, R. L.; Lombardi, J. R. Phys. Rev. Lett. 1979, 43, 147–
150.
(12) Lombardi, J. R.; Birke, R. L.; Lu, T.; Xu, J. J. Chem. Phys. 1986, 84, 4174–
4180.
(13) Xie, Y.; Wu, D. Y.; Liu, G. K.; Huang, Z. F.; Ren, B.; Yan, J. W.; Yang, Z. L.;
Tian, Z. Q. J. Electroanal. Chem. 2003, 554, 417–425.
(14) Lombardi, J. R.; Birke, R. L. Acc. Chem. Res. 2009, 42, 734–742.
(15) Tognalli, N.; Fainstein, A.; Bonazzola, C.; Calvo, E. J. J. Chem. Phys. 2004,
120, 1905.
(16) Tognalli, N.; Scodeller, P.; Flexer, V.; Szamocki, R.; Ricci, A.; Tagliazucchi,
M.; Calvo, E. J.; Fainstein, A. Phys. Chem. Chem. Phys. 2009, 11, 7412.
(17) dos Santos, D. P.; Andrade, G. F. S.; Temperini, M. L. A.; Brolo, A. G. J.
Phys. Chem. C 2009, 113, 17737–17744.
(18) Shegai, T.; Vaskevich, A.; Rubinstein, I.; Haran, G. J. Am. Chem. Soc. 2009,
131, 14390.
(19) Tian, Z. Q.; Ren, B. Annu. Rev. Phys. Chem. 2004, 55, 197–229.
(20) Wu, D. Y.; Li, J. F.; Ren, B.; Tian, Z. Q. Chem. Soc. Rev. 2008, 37, 1025–
1041.
10.1021/ac101152t © 2010 American Chemical Society 6919
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/26/2010

ties, 21 or the use of electrostatic fields to control the adsorption/
desorption of molecules on SERS substrates according to their
cationic/anionic nature in solution. 17,22
It is perhaps due to the complexity of the systems under study
that despite all the accumulated work over the last decades the
combination of SERS and electrochemistry is still undergoing
methodological and analytical advances as a technique. There is
a genuine need to develop both new and more advanced methods
and analysis tools to attack the ever increasing complexity of the
problems that are being investigated (in particular in bioelectrochemical
areas). An early example of the latter is the technique
of potential averaged SERS developed by Tian and co-workers 23
to address the problem of fast electrochemical processes and
SERS monitoring of unstable species on the electrodes. Detection
of intermediate species in electrochemical reactions by timeresolved
SERS has also been tried by Lombardi et al. 24 A more
recent example, on the other hand, is the introduction of local
imaging of the electrochemical current by surface-plasmon
resonances in ref 25. Many of these new techniques are not
directly linked or applied to SERS, 26,27 even though they share
the same basic elements; that is, the combination of electrochemistry
with a detection technique based on surface plasmon
resonances. 1
This paper aims at a methodological development in the
combination of SERS and electrochemistry. As such we shall show
some basic examples of the technique we propose. To this end,
we borrow ideas from well established techniques of optical
modulation 28 by using the ability of electrochemistry to turn “on”
and “off” Raman signals (by changing the resonance conditions
of the reduced or oxidized states), as well as to introduce other
(more subtle) spectral changes as a function of the applied
potential. Variations in SERS signals with potential have been
extensively studied in the literature. A common trait is the
observation of changes in the overall intensity of the spectra (as
the main consequence of the applied potential) when switching
from oxidized to reduced species. As a result, changes in the
intensity of the SERS spectra due to the oxidation state have been
used as a detection parameter in nanobiosensing, 29 to check the
permeability of phospholipid membranes 30 or thiol self-assembled
monolayers, 31 and to evaluate integrity and redox behavior in
proteins; 32 among others.
SERS provides a tool to monitor electrochemical phenomena
at concentration levels (∼nanomolars, and below) that cannot be
followed either in the voltamperogram or with other techniques.
(21) Ward, D. R.; Halas, N. J.; Ciszek, J. W.; Tour, J. M.; Wu, Y.; Nordlander,
P.; Natelson, D. Nano Lett. 2008, 8, 919–924.
(22) Lacharmoise, P. D.; Etchegoin, P. G.; Le Ru, E. C. ACS Nano 2009, 3,
66–72.
(23) Tian, Z. Q.; Li, W. H.; Mao, B. W.; Zou, S. Z.; Gao, J. S. Appl. Spectrosc.
1996, 50, 1569.
(24) Shi, C.; Zhang, W.; Birke, R. L.; Lombardi, J. R. J. Phys. Chem. 1990, 94,
4766–4769.
(25) Shan, X.; Patel, U.; Wang, S.; Iglesias, R.; Tao, N. Science 2010, 327, 1363–
1366.
(26) Huang, B.; Yu, F.; Zare, R. N. Anal. Chem. 2007, 79, 2979–2983.
(27) Wang, S.; Huang, X.; Shan, X.; Foley, K. J.; Tao, N. Anal. Chem. 2010, 82,
935–941.
(28) Yu, P. Y.; Cardona, M. Fundamentals of Semiconductors: Physics and
Materials Properties; Springer: Berlin, 2004.
(29) Scodeller, P.; Flexer, V.; Szamocki, R.; Calvo, E. J.; Tognalli, N.; Troiani,
H.; Fainstein, A. J. Am. Chem. Soc. 2008, 130, 12690–12697.
(30) Daza Millone, M. A.; Vela, M. E.; Salvarezza, R. C.; Creczynski-Pasa, T. B.;
Tognalli, N. G.; Fainstein, A. Chem. Phys. Chem. 2009, 10, 1927–1933.
6920 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Its sensitivity can, in fact, go routinely all the way down to single
molecule levels. 33 But spectral congestion is very often cited as a
common problem for its use 34,35 as was recently reported (for
example) to separate resonances of a target molecule from a
surrounding lipid matrix. 36 In the technique developed here,
electrochemistry provides an external “switch” wherefrom variations
in specific SERS signals can be introduced at will (and with
a given periodicity). Fluctuation analysis with methods like
principal component analysis (PCA)seither by itself or assisted
by Fourier “lock-in” prefiltering (vide infra)scan easily follow from
here, thus providing a systematic method to isolate SERS signals
from different species along the electrochemical cycle (in the
background of many possible, sometimes undesirable, contributions).
Electrochemical modulation and signal discrimination could
become a very important tool in areas like bioelectrochemistry,
to isolate the signals from different species in redox active sites,
in samples where we cannot choose at will the purity or spectral
characteristics of all the components. The situation of overlapping
peakssor much larger signals from spurious moleculesscan
prevent the isolation of the interesting spectra displaying an
oxidation/reduction cycle. Therefore, the technique is aimed at
applications in this latter case, and it applies particularly well to
address cases of coadsorption; which is a classic case applicable
to the electrochemistry of complex multicomponent systems (like
biological systems).
EXPERIMENTAL SECTION
A three-electrode cell with a Ag/AgCl (1 M Cl - ) electrode
and a high-area platinum foil as reference and counter electrodes,
respectively, was used. The working electrode (see
below for further details) is immersed in the electrolyte solution
(phosphate buffer, pH 6) and this is placed inside an open
electrochemical cell that allows focusing on the substrate with
long working distance objectives (×10, ×20, ×50, or ×100)
through a water/air interface (see the Supporting Information
(SI) for further details). All the potentials reported here are
referenced to the Ag/AgCl (1 M Cl - ) electrode. Cyclic voltammetry
is performed with a potentiostat with digital data
acquisition. The whole assembly is placed on top of a motorized
x-y stage for microscopy (to allow exploration and mapping
of the electrode) and is used on a BX41 Olympus microscope
attached to a Jobin-Yvon LabRam spectrometer. All the experiments
performed in this paper are done with the 633 nm line
of a HeNe laser with 3 mW at the sample. For samples with
“high” concentrations of dyes (for SERS standards) like
∼20-40 nM, we are mainly interested in average signals over
the substrate. Therefore, we typically use for these experiments
(31) Tognalli, N. G.; Fainstein, A.; Vericat, C.; Vela, M. E.; Salvarezza, R. C. J.
Phys. Chem. C 2008, 112, 3741–3746.
(32) Kranich, A.; Naumann, H.; Molina-Heredia, F. P.; Moore, H. J.; Lee, T. R.;
Lecomte, S.; de la Rosa, M. A.; Hildebrandt, P.; Murgida, D. H. Phys. Chem.
Chem. Phys. 2009, 11, 7390–7397.
(33) Etchegoin, P. G.; Van Duyne, R. P. Phys. Chem. Chem. Phys. 2008, 10,
6079.
(34) Casadio, F.; Leona, M.; Lombardi, J. R.; Van Duyne, R. P. Acc. Chem. Res.
2010, 43, 782–791.
(35) Golightly, R. S.; Doering, W. E.; Natan, M. J. ACS nano 2009, 3, 2859–
2869.
(36) Nguyen, T. T.; Rembert, K.; Conboy, J. C. J. Am. Chem. Soc. 2009, 131,
1401–1403.

the ×10 objective, to have a large spot diameter ∼10 µm, and
minimize photobleaching effects as much as possible.
Rhodamine 6G (RH6G), nile blue (NB), and crystal violet (CV)
were obtained from commercial sources (Aldrich) and mixed at
the appropriate concentrations with borohydride-reduced Ag
colloids 37 (to avoid a citrate capping layer) and with 20 mM KCl;
to induce a slight destabilization and the formation of clusters. 38
For each case reported here, the details about the specific
concentrations being used are specified in the captions. RH6G,
NB, and CV adsorb to the negatively-charged colloids in this case
through electrostatic interactions. The colloidal solution is subsequently
drop-casted and dried under a mild heat on a clean Ag
foil. Once dried, the colloids stick to the Ag foil by van der Waals
forces, and remain attached to it upon reimmersion in the
phosphate buffer solution. Several regions with multiple dried
clusters of colloids can be easily distinguished in the microscope
image on the Ag surface after drying. Hence, the Ag foil with the
colloids and the dyes is our working electrode, and also provides
the ideal means whereupon SERS enhancements (to observe low
dye concentrations) can coexists with the ability to perform
electrochemistry on the attached molecules. We checked at much
higher concentrations (∼1 mM) that, as far the voltamperogram
is concerned, the electrochemistry of the dyes on the colloids is
indistinguishable from the one where the dyes are directly
attached to the Ag foil itself. Note also that, in all of the
experiments performed in this work, the dyes were adsorbed
beforehand on the colloids (i.e., the dyes are not dissolved in the
electrolyte solution itself).
ELECTROCHEMICAL MODULATION AND SIGNAL
ISOLATION
Mixture of NB and RH6G. To fix ideas, we develop the
method through an example. Consider the case of a mixture of
two classic SERS dyes: RH6G and NB in concentrations of 40 and
20 nM, respectively. We perform cyclic voltammetric runs varying
the scan rate (i.e., the period), typically in the range ∼20-100 s
per cycle, while monitoring the SERS signal on the electrode
simultaneously with a much smaller integration time to follow the
dynamics. This gives us the possibility to choose the best scan
rate that allows us to follow the dynamics of the system. According
also to the choice of the potential window, the different coadsorbed
species will be modulated differently. This adds then a second
degree of freedom (besides the period of the modulation) to be
chosen by the experimentalist: that is, the potential range of the
modulation. In the case of Figure 1 we modulated the potential
in the range between -50 and -550 mV at a fix scan rate of 50
mV · sec -1 . NB will experience oxidation/reduction cycles
between these two values, while RH6G will only start to be
reduced at a potential of ∼-400 mV and lower. 17 Another
important detail is that the electrochemical modulation does
not cross at any point the potential of zero charge (pzc )-0.9
V) of the electrode, 39 meaning that the dyes (which are
positively charged in solution) remain all the time on a
(37) Creighton, J. A.; Blatchford, C. G.; Albrecht, M. G. J. Chem. Soc., Faraday
Trans. 2 1979, 75, 790–798.
(38) Meyer, M.; Le Ru, E. C.; Etchegoin, P. G. J. Phys. Chem. B 2006, 110,
6040.
(39) Niaura, G.; Gaigalas, A. K.; Vilker, V. L. J. Phys. Chem. B 1997, 101, 9250–
9262.
Figure 1. (a) Average spectrum for a sample with 20 nM of NB and
40 nM of RH6G which is modulated electrochemically in the potential
range -50 to -550 mV at a scan rate of 50 mV · sec -1 . Both signals
of NB and RH6G are clearly visible, and both dyes are modulated in
different electrochemical potential ranges. The “fingerprint” region
containing the 590 cm -1 mode of NB and the 610 cm -1 one of RH6G
(dashed box in (a)) is analyzed by PCA, resulting in the two dominant
eigenvectors shown in (b). The spectra in (b) contains a linear
combination of the two independent spectra of NB and RH6G, which
can be obtained by a suitable change of basis (described in full detail
in ref 40). The new eigenvectors (obtained from a linear combination
of the PCA ones) that describe the two physically independent
components (NB and RH6G) are shown in (c). With these eigenvectors
we can make a decomposition of the spectra to obtain the two
coefficients that describe (for each event) the intensity as a linear
combination of v1 and v2. These coefficients are shown in (d), with a
blown-out region in (e) where the dephasing of the electrochemical
modulation of both signals can be observed.
positively charged electrode in the entire modulation range.
None of the phenomena reported in this paper can be ascribed
to adsorption/desorption produced by a change in polarity of
the electrode across the pzc point, but rather to oxidation/
reduction of the species. Note also that the scan rates used in
our experiences are fast enough compared to those reported
for “electrostatic guiding”. 17,22
In order to optimize the presentation of the material here
without dwelling too much into collateral issues, we present the
main details of the basic electrochemistry of NB and RH6G in
the SI, as well as the basic aspects of the analysis with principal
component analysis (PCA), and PCA with FFT prefiltering. 41,42
As explained in the SI, both analysis techniques (PCA and FFT)
are quite widespread, and we shall therefore assume that the
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6921

eader has a basic understanding of them. Hereafter, we shall go
directly to the presentation of the main results.
Figure 1(a) shows the average signal over 400 spectra taken
with integration time of 1 s, with an electrochemical modulation
period of 20 s. The average has clear fingerprint peaks of both
dyes; for example the 590 and 610 cm -1 modes of NB and RH6G,
respectively. The intensity of these peaks is modulated differently
in the potential window selected, that is, their redox
potentials are different. This is a case nevertheless where both
signals of the coadsorbed species are clearly visible, and show
in some cases some degree of overlap (like the 1645 and 1650
cm -1 modes of NB and RH6G, respectively). We first show an
analysis in Figure 1 based on a spectral region that contains
fingerprint modes of both dyes, like the region around ∼600 cm -1
shown in a dashed box in Figure 1(a). A PCA analysis of this
region 40 immediately identifies the presence of two main modulated
components in the spectra. As explained in the SI though,
being a linear decomposition technique, PCA always provides a
linear combination of the right answer in the form of principal
“eigenvector” spectra (in this case the first two of them, shown
in Figure 1(b)). In order to go from this two PCA eigenvectors to
the real spectra that represent the individual contributions of NB
and RH6G, a linear transformation (which we shall call “rotation”)
of these eigenvectors is required. An entirely equivalent situation
occurs with the application of PCA to the analysis of fluctuations
from single molecule spectra; as explained in full length in ref
40. The “rotated” eigenvectors obtained in this case are shown in
Figure 1(c), and represent the isolated contributions of NB and
RH6G to each individual spectrum. Once these two eigenvector
spectra (v 1 and v2 in Figure 1(c)) are known, a standard linear
decomposition can be performed on each spectra Ii in the time
series (i ) 1, 2, ..., 400), to represent it as a linear combination
(with two coefficients) of the rotated PCA eigenvectors; that
is, Ii ) c1v1 + c2v2. The values of the coefficients, therefore,
represent the particular contribution of NB and RH6G to an
individual spectrum, and this is shown explicitly in Figure 1(d).
Figure 1(e), in addition, shows a blown-out region of the time
evolution of the coefficients where it can be appreciated that the
two signals are being modulated differently at different times,
accounting for the different intrinsic electrochemical properties
of both dyes as a function of the sweeping potential. Note that
time evolution can be converted to potential, since we know
exactly the period of electrochemical modulation applied externally.
An example of this is shown in the SI. The coefficients in
Figure 1(d) and (e) are obviously in arbitrary units, since the
Raman intensity itself is in arbitrary units too. What matters is
not the absolute units, but rather the relative values of the
coefficients representing the particular contributions to the total
coming from the different electrochemical species.
Therefore, Figure 1 demonstrates clearly that the combination
of electrochemical modulation with PCA analysis can (i) resolve
the problem of spectral congestion, and (ii) identify in addition
the ranges of electrochemical potentials where the different
species experience their modulation. This is emphasized in Figure
(40) Etchegoin, P. G.; Meyer, M.; Blackie, E.; Le Ru, E. C. Anal. Chem. 2007,
79, 8411.
(41) Jolliffe, I. T. Principal Component Analysis; Springer Verlag:Berlin, 2002.
(42) Hyvrinen, A.; Karhunen, J.; Oja, E. Independent Component Analysis; John
Wiley & Sons: New York, 2001.
6922 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
1(e), where the “dephasing” of the coefficients c 1 and c2 representing
the relative contributions of NB and RH6G, respectively,
can be easily seen. The fact that RH6G only starts to be
reduced toward the lowest negative range of the potential (E
< -400 mV) appears as a “dephasing” of the coefficients in
time, while the successive electrochemical cycles evolve. Figure
1(d) also shows a slight reduction in the amplitude of the
coefficients produced by photobleaching at the observation point.
It may also have partial contribution from natural long-term
desorption of the dyes into the buffer. Photobleaching, in general,
tends to be one of the major limiting factors to obtain unlimited
sampling of the signal.
Moving on from the simplest case in Figure 1, if we want to
carry out the analysis over the entire spectral range (rather than
a small window with fingerprint modes, like in Figure 1), special
care must be taken with background contributions. Backgrounds
are in general problematic in SERS 43 and, unlike the Raman peaks
themselves, they can have a variety of origins (including molecules
that are not in contact with the electrode). Furthermore, there
will be in general a correlation between the background and the
Raman signal created by the dispersion of the plasmon resonances
producing SERS. 43-45 Here, we want to concentrate in a first
approximation only on the Raman signals themselves and,
therefore, the background will be removed 46 to avoid its presence
in the PCA analysis. This should be done with care and the results
should be assessed on a case-by-case basis. Backgrounds in SERS
can be particularly problematic in the single molecule limit, where
the dispersion of an individual plasmon resonance at a hot-spot
can be revealed. 43,47 Measuring an average over many molecules
is in a way an advantage, for differing types of backgrounds tend
to average out and we are left with a case of background
contribution that is easier to subtract. The background of each
individual spectrum is subtracted with a wavelet transform that
is fully described in ref 46 (including the program which is freely
available 46 ). The success of the background subtraction procedure
to deconvolute the spectra is judged here by the ability of the
PCA analysis to distinguish the two main contributions to the
signal in the full spectral range (which are known in this case).
Effectively, Figure 2 shows the equivalent analysis of Figure
1 in the full spectral range. Once the backgrounds are subtracted,
and the two main PCA eigenvector spectra rotated, we recover
the independent spectra of NB and RH6G from the mixture.
Except for a few minor imperfections, the separation of the spectral
components is almost perfect, and can be also distinguished in
terms of the respective electrochemical modulation ranges for the
potential. As in Figures 1(d) and (e), the dephasing of the
coefficients from the contributions of NB and RH6G, according
to the different values of the redox potentials over the cyclic
voltammetry runs, can be clearly observed again in Figure 2(d).
It is particularly worth emphasizing that the spectral deconvolution
(43) Buchanan, S.; Le Ru, E. C.; Etchegoin, P. G. Phys. Chem. Chem. Phys. 2009,
11, 7406.
(44) Le Ru, E. C.; Etchegoin, P. G.; Grand, J.; Felidj, N.; Aubard, J.; Levi, G.;
Hohenau, A.; Krenn, J. R. Curr. Appl. Phys. 2008, 8, 467.
(45) Le Ru, E. C.; Grand, J.; Felidj, N.; Aubard, J.; Levi, G.; Hohenau, A.; Krenn,
J. R.; Blackie, E.; Etchegoin, P. G. J. Phys. Chem. C 2008, 112, 8117.
(46) Galloway, C.; Le Ru, E. C.; Etchegoin, P. G. Appl. Spectrosc. 2009, 63,
1371.
(47) Itoh, T.; Yoshida, K.; Biju, V.; Kikkawa, Y.; Ishikawa, M.; Ozaki, Y. Phys.
Rev. B 2007, 76, 085405.

Figure 2. With a suitable removal of the background 46 for all events,
the full spectra of NB and RH6G can be deconvoluted again from
the modulated time series through PCA. The region analyzed now
is the full spectral window shown in (a) (dashed line). The two main
eigenvectors and the “rotated” ones 40 (representing the independent
contributions of NB and RH6G) are shown in (b) and (c), respectively.
The spectra of two compounds can be deconvoluted, and their time
(or potential) dependence followed through the time evolution of the
two coefficients in (d), that describe each event as a linear combination
of v1 and v2 in (c). The dephasing of the electrochemical
modulation of the two compounds can again be appreciated in (d).
through the electrochemical modulation works very well even in
regions where a substantial overlap of peaks occurs. An example
of the latter is shown in Figure 3, for the region around ∼1650
cm -1 , which has contributions from breathing modes of both
NB and RH6G (but at slightly different frequencies). As can
be appreciated from Figure 3, the different modulation induced
through the electrochemical cycle on both compounds is good
enough to “deconvolute” closely laying peaks within their natural
widths.
Mixture of NB and CV. All in all, the case presented in the
previous section represents a relative “easy” case of spectral
deconvolution for PCA with electrochemical modulation; starting
from the fact that both compounds have comparable contributions
to the intensity of the spectra and can be easily identified in terms
of their contributions to fluctuations (which is the whole basis of
PCA 41,42 ). A substantially more challenging case is to try recover
a small signal from a background of other contributions to the
spectra (that might be unavoidable in many real cases). We treat
this case here also with an example.
Figure 3. The deconvolution of the spectra works even in regions
where there is a substantial overlap of peak intensities for the different
compounds. Here we show explicitly the region ∼1650 cm -1 , where
the RH6G peak is deconvoluted from the equivalent ring-breathing
mode in NB, which is at a slightly lower frequency.
Consider the case of a mixture of 80 and 20 nM of CV and
NB, respectively, as shown in Figure 4(a). The mixture is prepared
purposely so that the signal of NB is barely visible in the average
spectrum (Figure 1(a)). This is shown explicitly, for example, with
the arrows pointing at the ∼590 and ∼1650 cm -1 fingerprint
peaks of NB in Figure 4(a), which are almost buried in the
immediate surrounding of much larger peaks from CV. Both NB
and CV experience (different) electrochemical modulations in the
potential range between -50 and -550 mV (the electrochemistry
of CV is also summarized in the SI). We can basically follow in a
first approximation the analysis performed in the previous section.
With an unlimited amount of sampling, PCA has all what it
needs to distinguish the different components in the spectra. This
is because the eigenvalues and eigenvectors of the covariance
matrix (which is the basis of PCA 41,42 ) will eventually distinguish
what is correlated and what is random in the signal. For the same
reason that unlimited integration time should in principle always
lift a signal above the noise level, sufficient sampling will always
provide PCA with all the information required to distinguish the
different contributions to the spectra. Nevertheless, when big
signals are mixed with very small ones from other independent
contributions, it is the comparative (with respect to other larger
signals in the spectrum) size of the small signals that fix the
amount of sampling we need to obtain the principal components
from the spectra. This amount of sampling can be, however,
prohibitively large. In the case of SERS, for example, photobleaching
and other long-term stability problems of the signal puts
serious limitation to how much sampling we can obtain. We are
left here basically with a conundrum: random fluctuations in large
signals with limited sampling might be comparablesas far as
contributions to the covariance matrix of PCA is concernedsto
the small signals we are trying to modulate and detect on purpose.
An example of this is explicitly shown in Figure 4; the region
encircled in a box in Figure 4(a) contains a contribution from NB
and a much larger peak of CV in the same range. This region is
PCA-analyzed and the first four eigenvectors are shown in Figure
4(b). We can see that there is no clear signature of the NB peak
in the PCA analysis. The first eigenvector is clearly dominated
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6923

Figure 4. With limited sampling, PCA can have problems to
distinguish small components in the spectra. An example is shown
here with a mixture of 80 and 20 nM of CV and NB, respectively. In
(a) we show the average (over 400 spectra), with arrows showing
the positions where the fingerprint modes of NB at 590 cm -1 and
1645 cm -1 should appear. The signal is dominated by CV throughout
the entire spectral range, with many overlap regions. In (b) we show
a PCA analysis of the ∼1630 cm -1 region, where both the 1620 and
1645 cm -1 peaks of CV and NB are expected, respectively. But with
a difference of factor of ∼20 between the intensities of CV and NB in
this range, fluctuations larger than ∼5% in the CV signal have a larger
weight for the covariance matrix than the induced (electrochemical)
variations of NB. With limited sampling, PCA cannot clearly distinguish
the NB signal. We show here the first four PCA eigenvectors as an
example. The limited sampling problem can be partially compensated
by a prefiltering of the data with a FFT-transform, filtered at the
appropriate electrochemical modulation frequency (see Figure 5).
by the intensity fluctuations of the much larger CV peak, whereas
the second one has the typical characteristic of an eigenvector
accounting for frequency shifts of the CV peak. 40 From the third
eigenvector onward, the covariance matrix in PCA is trying to
find “correlations in the noise”. With enough sampling, these
correlations will be zero and the peak of NB will appear as a third
eigenvector. But in this case, the limited sampling establishes a
competition between the random fluctuations of the much larger
peak and the signal we are trying to observe (from NB).
It is in situations like this one that a prefiltering of the data
can help. The basic idea of Fourier prefiltering is explained in
full in the SI with further examples, but here we shall give a brief
version of it. From a FFT-transform of the total integrated intensity
as a function of time we can easily reveal the main modulation
frequency in the FFT-spectrum, as can be seen in Figure 5(a).
This comes primarily from the modulation of the CV signal, but
that is irrelevant at this stage: the FFT-transform of the total
integrated intensity is used to identify the modulation in FFTfrequency
units. 48-50 We can now perform a wavelength-bywavelength
(i.e., pixel-by-pixel) FFT filtering of the data at this
6924 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. (a) A FFT of the total integrated intensity of the spectra
(which varies mainly here due to the electrochemical modulation of
CV), reveals clearly the main frequency of the modulation (i.e., scan
rate of 50 mV. sec -1 ). We can filter each pixel (wavelength) at this
frequency and then transformed back the spectra into the time
domain. The average of the filtered data (at the electrochemical
modulation frequency) is shown in (b). Now the signal of NB is more
clearly visible in the average and its relative intensity with respect to
the CV peak is larger (because only the part of the CV signal that is
modulated survives the filtering). A PCA analysis of the prefiltered
signal shows a first eigenvector representing the CV intensity, and a
second eigenvector which now contains the NB peak at ∼1645 cm -1 ,
and a small fraction of frequency shifts of the CV peak. The vertical
intensity scale for both curves in (c) is the same. Green lines are
guides to the eye.
particular frequency. This is simply done by an FFT-transform to
the time evolution of each pixel, multiplication by an appropriate
filter centered at the spikes of Figure 5(a), and followed by an
inverse FFT-transform to go back to the time domain. In these
data, therefore, we have enhanced the importance of only those
fluctuations that happen at the known electrochemical modulation
frequency, with respect to any other random fluctuation.
Figure 5(b) and (c) show the resulting PCA analysis after FFTfiltering.
It is worth noting that the average of the filtered signal
has now a much better ratio of intensities between the CV and
NB signals; this is because only the fraction of the CV signal that
is being modulated survives the filtering. Accordingly, this puts
(48) Press, W. H.; Flannery, B. P.; Teukolsky, S. A.; Vetterling, W. T. Numerical
Recipes in FORTRAN: The Art of Scientific Computing; Cambridge University
Press: Cambridge, 1989.
(49) Kreyszig, E. Advanced Engineering Mathematics, 9th ed.; John Wiley & Sons:
New York, 2006.
(50) Kauppinen, J.; Partanen, J. Fourier Transforms in Spectroscopy; Wiley-VCH
Verlag: Berlin, 2001.

the NB signal in a much better footing to be identified among
the first PCA eigenvectors. Effectively, we can see now in Figure
5(c) that the first eigenvector is still dominated by the CV signal,
but the second one contains both the NB signal and some
contributions to frequency shifts of the CV peak. What we have
achieved by FFT-prefiltering is to put the NB signal at the same
level of importance of frequency shifts of the CV peak (according
to the covariance matrix) and this therefore “lifts” a very small
signal from accidental correlations in the noise. As explained
further in the SI, this is nothing but the well established principle
of “lock-in” amplification; applied here as a prefiltering for PCA.
In general, in any other experimental situation, if we have access
to unlimited sampling then time lock-in “amplification” might not
be needed; for the signal will always increase linearly with time
while the noise increases like the square root of it. Therefore,
the signal-to-noise ratio will always improve with integration time
and eventually a small signal can be resolved. But we can achieve
a much faster result in a shorter time (with less sampling) with
a lock-in. Like in the case at hand here, unlimited amount of
sampling is sometimes not possible in other experimental situations
because the long-term stability of the experiment is compromised
or it is simply too long. The situation depicted here is
the exact analog of this latter situation but for the problem of
spectral deconvolution with PCA.
The combination of FFT-prefiltering and PCA might not be
necessary in many situations and whenever possible, the simplest
analysis should prevail. But the covariance matrix of PCA is
basically “blind” to the time sequence of the electrochemical cycle.
If we scrambled all the spectra (in time) we would still have the
same covariance matrix and the same PCA eigenvectors. However,
we do know the frequency (or period) with which we are inducing
the electrochemical modulation. What we are doing with FFTprefiltering
is, accordingly, to “pick” the appropriate fluctuations
that happen at the frequency where we know the real physical
effect is happening. In that manner, we bias the PCA analysis
toward a physically relevant set of fluctuations, and this helps to
improve the ability of PCA to distinguish what is correlated and
what is not, in a situation where limited sampling is unavoidable.
In the language of lock-in amplification, we have improved the
signal-to-noise ratio by discarding all fluctuations except those that
happen at the right frequency. The example chosen here is tailormade
to show the concept and, as such, we have different options
at hand. We could have done, for example, a PCA analysis is a
much narrower window around the NB peak (to try to minimize
the influence of the CV peak nearby). However, in real applications,
we might not even know where to expect the peaks, and
considerable spectral overlap is always possible. FFT-prefiltering
is one additional tool to consider in these situation, and it could
-in many cases- be the difference between being able to isolate a
particular spectral feature of the weaker signal or not.
CONCLUSIONS
We have shown a combination of electrochemical modulation
with fluctuation analysis to discriminate and isolate different
species in cases where there is spectral congestion and coadsorption
of molecules. Needless to say, the technique is not limited
to two species, and can be applied to multicomponent systems in
general. PCA contains, in principle, all the elements it needs to
distinguish all spectral components given enough sampling. The
restriction of limited sampling to “lift” small signals from other
contributions has also been shown through the addition of FFTprefiltering.
Prefiltering is not strange in PCA analysis. The most
common type of prefiltering used in the PCA literature is
“pre-whitening”. 41,42 In this latter case, this is done for a different
purpose: to ensure that the samples are as unbiased as possible
before the PCA analysis is carried out. Here we use prefiltering
in a different way: to bias the study of correlations in the data
that happen at a prescribed frequency, imposed externally by the
electrochemical modulation. We believe that techniques to solve
spectral congestion of coadsorbed species will be of great interest,
as SERS moves into areas (like bioelectrochemistry) where the
purity and characteristics of the sample cannot be chosen at will.
We hope the concepts developed here in our paper will contribute
to that endeavor.
ACKNOWLEDGMENT
E.C. acknowledges the financial support of UNLP, ANPCyT
(Argentina) and the MacDiarmid Institute (New Zealand) for a
research/exchange program between Argentina and New Zealand.
Thanks are also given to the host institution, Victoria University
of Wellington, where the work was carried out. We acknowledge
financial support from ANPCyT (Argentina, PICT06-621, PAE
22711, PICT06-01061, PICT-CNPQ 08-019). E.C., A.F., and RCS
are also at CONICET. MEV is a member of the research career
of CIC BsAs. R.C.S and A.F. are Guggenheim Foundation Fellows.
PGE and ECLR are indebted to the Royal Society of New Zealand
for additional financial support under a Marsden Grant.
SUPPORTING INFORMATION AVAILABLE
Additional information and figures. This material is available
free of charge via the Internet at http://pubs.acs.org.
Received for review May 3, 2010. Accepted July 10, 2010.
AC101152T
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6925

Anal. Chem. 2010, 82, 6926–6932
Study of Highly Selective and Efficient Thiol
Derivatization Using Selenium Reagents by
Mass Spectrometry
Kehua Xu, †,‡ Yun Zhang, † Bo Tang,* ,‡ Julia Laskin, § Patrick J. Roach, § and Hao Chen* ,†
Center for Intelligent Chemical Instrumentation, Department of Chemistry and Biochemistry, Ohio University, Athens,
Ohio 45701, Key Laboratory of Molecular and Nano Probes, Ministry of Education, College of Chemistry, Chemical
Engineering and Materials Science, Shandong Normal University, Jinan, China, 250014, and Chemical and Materials
Sciences Division, Pacific Northwest National Laboratory, P.O. Box 999, MSIN K8-88, Richland, Washington 99352
This paper reports a systemic mass spectrometry (MS)
investigation of a novel strategy for labeling biological
thiols, involving the cleavage of the Se-N bond by thiol
to form a new Se-S bond. Our data show that the reaction
is highly selective, rapid, reversible, and efficient. Among
20 amino acids, only cysteine is reactive toward Se-N
containing reagents and the reaction occurs in seconds.
With the addition of dithiothreitol, peptides derivatized
by selenium reagents can be recovered. The high reaction
selectivity and reversibility provide potential in both
selective identification and isolation of thiols from mixtures.
Also, with dependence on the selenium reagent
used, derivatized peptide ions exhibit tunable dissociation
behaviors (either facile cleavage or preservation of the
formed Se-S bond upon collision-induced dissociation),
a feature that is useful in proteomics studies. Equally
importantly, the thiol derivatization yield is striking, as
reflected by 100% conversion of protein �-lactoglobulin
A using ebselen within 30 s. In addition, preliminary
applications such as rapid screening of thiol peptides from
mixtures and identification of the number of protein free
and bound thiols have been demonstrated. The unique
selenium chemistry uncovered in this study would be
valuable in the MS analysis of thiols and disulfide bonds
of proteins/peptides.
Biological thiols, such as glutathione (GSH) and thiol proteins,
are critical physiological components found in animal tissues and
fluids and involved in a plethora of important cellular functions. 1,2
In the past decades chemical characterization of biological thiols
has attracted significant attention. Various measurement
methods have been developed including UV, 3 fluorescence (FL)
* To whom correspondence should be addressed. Hao Chen: phone, 740-
593-0719; fax, 740-597-3157; e-mail, chenh2@ohio.edu. Bo Tang: phone, (86)531-
86180010; e-mail, tangb@sdnu.edu.cn.
† Ohio University.
‡ Shandong Normal University.
§ Pacific Northwest National Laboratory.
(1) Basford, R. E.; Huennekens, F. M. J. Am. Chem. Soc. 1955, 77, 3873–
3877.
(2) Rahman, I.; MacNee, W. Free Radical Biol. Med. 2000, 28, 1405–1420.
(3) Kusmierek, K.; Chwatko, G.; Glowacki, R.; Bald, E. J. Chromatogr., B 2009,
877, 3300–3308.
6926 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
spectroscopy, 4-6 and mass spectrometry (MS), 7-12 in which the
derivatization of thiol groups with a suitable chemical reagent is
necessary for increasing thiol stability and improving detection
selectivity.
While spectroscopic methods such as FL are very useful for
detection and imaging of thiol compounds, MS can provide
molecular weight and structural information. The inherent sensitivity
and chemical specificity offered by MS 13-16 is essential in
further identifying biological thiols and investigating their physiological
functions on a molecular level. Also, a number of excellent
MS studies of thiols and disulfides of proteins/peptides based on
the novel ion chemistry have been reported. 10,11,17-20 At present,
MS labeling strategies are mainly based on three types of chemical
reactions. Nucleophilic substitution such as using heptafluorobutyl
chloroformate 21 and iodoacetamide is commonly used for the
analysis of protein tryptic digest; 22 however, these reagents are
not specific for thiol compounds, that is, they can also couple with
amino/hydroxyl groups. The second approach involves Michael-
(4) Lee, J. H.; Lim, C. S.; Tian, Y. S.; Han, J. H.; Cho, B. R. J. Am. Chem. Soc.
2010, 132, 1216–1217.
(5) Pullela, P. K.; Chiku, T.; Carvan, M. J.; Sem, D. S. Anal. Biochem. 2006,
352, 265–273.
(6) Tang, B.; Yin, L.; Wang, X.; Chen, Z.; Tong, L.; Xu, K. Chem. Commun.
2009, 5293–5295.
(7) Gorman, J. J.; Wallis, T. P.; Pitt, J. J. Mass Spectrom. Rev. 2002, 21, 183–
216.
(8) Bilusich, D.; Bowie, J. H. Mass Spectrom. Rev. 2009, 28, 20–34.
(9) Chrisman, P. A.; Pitteri, S. J.; Hogan, J. M.; McLuckey, S. A. J. Am. Soc.
Mass Spectrom. 2005, 16, 1020–1030.
(10) Diedrich, J. K.; Julian, R. R. J. Am. Chem. Soc. 2008, 130, 12212–12213.
(11) Gunawardena, H. P.; O’Hair, R. A. J.; McLuckey, S. A. J. Proteome Res. 2006,
5, 2087–2092.
(12) Seiwert, B.; Karst, U. Anal. Chem. 2007, 79, 7131–7138.
(13) Wang, Y.; Vivekananda, S.; Zhang, K. Anal. Chem. 2002, 74, 4505–4512.
(14) Pingitore, F.; Wesdemiotis, C. Anal. Chem. 2005, 77, 1796–1806.
(15) Blanksby, S. J.; Bierbaum, V. M.; Ellison, B. G.; Kato, S. Angew. Chem.,
Int. Ed. 2007, 46, 4948–4950.
(16) Gronert, S. Chem. Rev. 2001, 101, 329–360.
(17) Kim, H. I.; Beauchamp, J. L. J. Am. Soc. Mass Spectrom. 2009, 20, 157–
166.
(18) Chrisman, P. A.; McLuckey, S. A. J. Proteome Res. 2002, 1, 549–557.
(19) Li, J.; Shefcheck, K.; Callahan, J.; Fenselau, C. Int. J. Mass Spectrom. 2008,
278, 109–133.
(20) Qiao, L.; Bi, H.; Busnel, J.-M.; Liu, B.; Girault, H. H. Chem. Commun. 2008,
47, 6357–6359.
(21) Simek, P.; Husek, P.; Zahradnickova, H. Anal. Chem. 2008, 80, 5776–
5782.
(22) Williams, D. K., Jr.; Meadows, C. W.; Bori, I. D.; Hawkridge, A. M.; Comins,
D. L.; Muddiman, D. C. J. Am. Chem. Soc. 2008, 130, 2122–2123.
10.1021/ac1011602 © 2010 American Chemical Society
Published on Web 07/16/2010

Scheme 1. Reactions of Selenium Reagents 1 and 2 with Thiols of Proteins/Peptides
addition of thiols onto unsaturated CdC bonds such as using
acrylate 23 or maleimide derivatives as common labeling agents. 24
These reagents have selectivity toward thiols and have been often
used in practical thiol analysis. However, the Michael-addition
product is irreversible, which cannot allow enriching and purifying
analyte compounds from complex matrices. 25 The third approach
involves thiol exchange reaction such as using 5,5′-dithiobis(2nitrobenzoic
acid) known as Ellman’s reagent; 26,27 a large excess
amount of the disulfide reagent is necessary to ensure complete
derivatization of all thiols in the sample. As a result, the newly
formed disulfides in the reaction can possibly further react with
residual thiols of the target molecule to form undesirable disulfides.
For example, Udgaonkar et al. used Ellman’s reagent in a
100-fold molar excess to label the thiol protein based on the
exchange reaction in the study of the cooperativity of a fast protein
folding reaction by MS. 27 Besides the problems mentioned above,
these derivatization reactions have other limitations, including a
long reaction time, low conversion yield, or more than one possible
site for tagging. Therefore, new derivatization chemistry suitable
for MS detection of thiol, particularly with high thiol selectivity,
fast reaction speed, good reaction reversibility, and high conversion
yield, is still needed for the structural analysis of peptides
and proteins in biological samples.
Because selenium is an essential element in vivo 28 and a key
component of selenoproteins known as antioxidant enzymes,
selenium chemistry has recently attracted increasing attention.
Ebselen, 2-phenyl-1,2-benzisoselenazol-3(2H)-one (reagent 1,
Scheme 1), has been considered as a mimetic of glutathione
peroxidase (GPx) 29-32 and an anti-inflammatory drug since being
found. 33,34 The mechanism for ebselen-catalyzed thiol oxidation
(23) Wang, G.; Hsieh, Y.; Wang, L.; Prelusky, D.; Korfmacher, W. A.; Morrison,
R. Anal. Chim. Acta 2003, 492, 215–221.
(24) Seiwert, B.; Hayen, H.; Karsta, U. J. Am. Soc. Mass Spectrom. 2008, 19,
1–7.
(25) Jalili, P. R.; Ball, H. L. J. Am. Soc. Mass Spectrom. 2008, 19, 741–750.
(26) Sevcikova, P.; Glatz, Z.; Tomandl, J. J. Chromatogr., A 2003, 990, 197–
204.
(27) Kumar Jha, S.; Udgaonkar, J. B. J. Biol. Chem. 2007, 282, 37479–37491.
(28) Yoneda, S. J.; Kazuo, T. S. Toxicol. Appl. Pharmacol. 1997, 143, 274–280.
(29) Forstrom, J. W.; Zakowski, J. J.; Tappel, A. L. Biochemistry 1978, 17, 2639–
2644.
(30) Birringer, M.; Pilawa, S.; Flohé, L. Nat. Prod. Rep. 2002, 19, 693–718.
(31) Jacob, C.; Giles, G. I.; Giles, N. M.; Sies, H. Angew. Chem, Int. Ed. 2003,
42, 4742–4758.
(32) Haenen, G. R.; De Rooij, B. M.; Vermeulen, N. P.; Bast, A. Am. Soc.
Pharmacol. Exp. Therapeut. 1990, 37, 412–422.
has been proposed: 35,36 the Se-N bond of ebselen is cleaved by
thiol RSH to produce the corresponding selenenyl sulfide Se-S,
which further reacts with excess thiol RSH to produce selenol
and disulfide compound RSSR. However, Mugesh et al. recently
reported that ebselen as a GPx-like redox catalyst is indeed
inactive, and the reaction of ebselen with PhSH only produces
the Se-S product, not disulfide PhSSPh, even with an excess
amount of PhSH. 36 Very recently, based on this novel selenium
chemistry, we synthesized fluorescent probes carrying Se-N
bonds and used them for detecting and imaging thiols in living
cells 6,37 by FL. It was shown that the FL probes capture cellular
thiols selectively in the presence of diverse species such as
inorganic metal ions, dopamine, histamine, L-adrenaline, etc.
Following that, Zhang et al. used piazselenole containing Se-N
bond to probe physiological thiols based on electrochemical
reactions. 38 Highly specific derivatization of thiols using compounds
containing Se-N bonds makes them promising candidates
to derivatize peptides and proteins for subsequent MS characterization.
Although the analysis of protein thiols and disulfide bonds
is important in proteomics applications, the selenium chemistry
received surprisingly limited attention. 39 Here we present a
systematic study of derivatization of amino acids, peptides, and
proteins using Se-N containing reagents and examine the utility
of this chemistry for analytical applications involving mass
spectrometry.
In this study, we carried out a series of reactions of two Se-N
containing reagents, ebselen and N-(phenylseleno)phthalimide
(reagent 2, Scheme 1), with various biological thiol compounds
such as cysteine, reduced peptide glutathione GSH, and �-lactoglobulin
A protein. As shown in Scheme 1, the Se-N bonds in
reagents 1 and 2 are cleaved by thiols to form new Se-S bonds
as shown in products 3 and 4, respectively (Scheme 1). Our
experimental results showed that this thiol derivatization reaction
is highly selective, rapid, reversible, and efficient (quantitative in
(33) Müller, A.; Cadenas, E.; Graf, P.; Sies, H. Biochem. Pharmacol. 1984, 33,
3235–3239.
(34) Sies, H.; Masumoto, H. Adv. Pharmacol. 1997, 38, 229–246.
(35) Mugesh, G.; du Mont, W.-W.; Sies, H. Chem. Rev. 2001, 101, 2125–2179.
(36) Sarma, B. K.; Mugesh, G. J. Am. Chem. Soc. 2005, 127, 11477–11485.
(37) Tang, B.; Xing, Y.; Li, P.; Zhang, N.; Yu, F.; Yang, G. J. Am. Chem. Soc.
2007, 129, 11666–11667.
(38) Wang, W.; Li, L.; Liu, S.; Ma, C.; Zhang, S. J. Am. Chem. Soc. 2008, 130,
10846–10847.
(39) Sakurai, T.; Kanayama, M.; Shibata, T.; Itoh, K.; Kobayashi, A.; Yamamoto,
M.; Uchida, K. Chem. Res. Toxicol. 2006, 19, 1196–1204.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6927

Figure 1. ESI-MS spectrum showing the selective reaction of ebselen with cysteine in the presence of methionine, tryptophan, and valine. The
insets show the agreement of the simulated product peak isotopic distribution (top) with the recorded reaction product ion of m/z 397 (bottom).
some cases) at room temperature. On the basis of these revealed
features of the reaction, associated analytical applications were
explored, such as the selective identification of thiol-containing
peptides from mixtures and measurement of the number of
cysteine residues of proteins. In addition, the dissociation behaviors
of the derivatized protein/peptide ions were also examined.
EXPERIMENTAL SECTION
Electrospray ionization (ESI) of samples was performed using
a Thermo Finnigan LCQ DECA Mass Spectrometer (San Jose,
CA) or a hybrid triple-quadrupole-linear ion trap mass spectrometer
(Q-trap 2000; Applied Biosystems/MDS SCIEX, Concord,
Canada). The sample injection flow rate was 10 µL/min. A high
voltage of +5 kV was applied to the spray probe for the positive
ion mode. For the Thermo Finnigan LCQ DECA mass spectrometer,
the optimized heated transfer capillary tube temperature was
150 °C. Collision-induced dissociation (CID) was used for further
structural confirmation of the product ions. Data acquisition was
performed using Xcalibur (rev. 2.0.7, Thermo Scientific, San Jose,
CA). For the Q-trap 2000 mass spectrometer, the mass spectrometer
curtain gas (N 2) was kept as 20 (manufacturer’s units) and
the declustering potential was set at 10 V. Collision induced
dissociation (CID) was carried out to provide ion structural
information using enhanced product ion scan mode. Precursor
ion scanning (PIS) was used to select reaction products by
monitoring the characteristic fragment ions, and N2 was used
as the collision gas. Data acquisition was performed using the
Analyst software (version 1.4.2, Applied Biosystems/MDS
SCIEX, Concord, Canada). Deconvolution of mass spectra was
carried out using Mag-Tran 1.03b2 software (Amgen Inc.,
Thousand Oaks, CA) written based on the ZScore algorithm. 40
High-Resolution MS. High-resolution LTQ-Orbitrap mass
spectrometer (Thermo Electron, Bremen, Germany) with a
(40) Zhang, Z.; Marshall, A. J. Am. Soc. Mass Spectrom. 1998, 9, 225–233.
6928 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
modified nanoelectrospray ionization (nano-ESI) was used for
collecting high-resolution data. The system was operated in the
positive ion mode with a resolving power of 60 000 at m/z 400.
The Molecular Weight Calculator (http://ncrr.pnl.gov/software/)
was used to simulate the isotope distribution for thiol-derivatized
products.
Chemicals. The 20 amino acids, angiotensin II human (FW
1046 Da), bradykinin acetate (FW 1060 Da), MRFA (Met-Arg-
Phe-Ala) acetate salt (FW 523 Da), �-lactoglobulin A from bovine
milk, tris(2-carboxyethyl)phos-phine hydrochloride (TCEP), DLdithiothreitol
(DTT), TPCK-treated trypsin from bovine pancreas
(MW ∼23.8 KDa), ammonium bicarbonate, 1,4-benzoquinone, and
N-(phenylseleno)phthalimide were purchased from Sigma-Aldrich
(St. Louis, MO). Glutathione (GSH) (reduced form, MW 307 Da)
and ebselen were obtained from TCI America (Tokyo, Japan) and
Calbiochem (Cincinnati, OH), respectively. HPLC-grade methanol
and acetonitrile from GFS Chemicals (Columbus, OH) and Sigma-
Aldrich (St. Louis, MO) were used, and acetic acid was purchased
from Fisher Scientific (Pittsburgh, PA). The deionized water used
for sample preparation was obtained using a Nanopure Diamond
Barnstead purification system (Barnstead International, Dubuque,
IA).
Protein Reduction. A volume of 175 µL of 0.1 mM �-lactoglobulin
A in methanol/water (1:1 by volume) containing 2% acetic
acid and 17.5 µL of 50 mM TCEP in 20 mM ammonium
bicarbonate aqueous solution were mixed resulting in the molar
ratio of 1:50 (protein/TCEP). The protein was reduced by TCEP
for 3.5 h at room temperature. Then Millipore-ZipTip Pipette Tips
were used to remove TCEP and ammonium bicarbonate via
desalting.
RESULTS AND DISCUSSION
Reaction Selectivity. In this experiment, the selectivity of the
derivatization reaction was examined first using amino acids as

Figure 2. (a) ESI-MS spectrum showing the reaction of GSH with ebselen; (b) CID MS 2 spectrum of ebselen derivatized GSH product ion (m/z
583); (c) ESI-MS spectrum showing the reaction of GSH with reagent 2; (d) CID MS 2 spectrum of the product ion of GSH derivatized by reagent
2 (m/z 464); (e) ESI-MS spectrum showing the selective reaction of ebselen with GSH in the presence of angiotensin II, bradykinin, and MRFA;
(f) precursor ion scanning based on the monitoring of the characteristic fragment ion of m/z 276 showing the selective detection of GSH.
reaction substrates and using ebselen as a reagent. We found that
ebselen reacts exclusively with side chains of cysteine residues
that contain a free thiol. As illustrated in Figure 1, the ESI-MS
spectrum displays the reaction of ebselen (10 µM) with cysteine
(5 µM) in the presence of methionine, tryptophan, and valine (5
µM for each) in methanol/water (1:1 by volume) containing 0.5%
acetic acid. A dominant peak at m/z 397 corresponds to the
reaction product between ebselen and cysteine. The inset shows
the zoom-in area for the reaction product ion C 16H17O3N2SSe (m/z
397) (bottom) in comparison with the calculated product
isotope distribution (top). Exact match is observed, confirming
the peak assignment. The characteristic isotope distribution
helps to identify products of the derivatization reaction. Upon
CID, the ion at m/z 397 fragments into m/z 379, 353, and 276
by losses of one water molecule, one carbon dioxide, and one
cysteine, respectively (data not shown), confirming the product
ion structure and the covalent nature of the newly formed bond.
In addition, peaks at m/z 118, 122, 150, 205, and 276 are seen
in the Figure 1, corresponding to the protonated valine, cysteine,
methionine, tryptophan, and ebselen, respectively. We also tested
the reaction of ebselen with cysteine in the presence of 19 other
natural amino acids (Figure 1-S in the Supporting Information).
Again, only the reaction product of ebselen and cysteine (m/z
397) were observed, demonstrating that ebselen has exclusive
specificity toward cysteine derivatization. Similarly, it was also
found that another selenium reagent 2 has similar selectivity for
cysteine in the presence of other amino acids. Furthermore, in
our observation, the derivatization reaction takes place rapidly and
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6929

Figure 3. ESI-MS spectrum showing the recovery of unmodified
GSH 30 min after adding DTT to the reaction mixture of GSH and
ebselen.
is completed in seconds at room temperature. 41,42 This is an
advantage for shortening analysis time in real-world applications,
particularly for high-throughput proteomics analysis.
Reactions with Peptide Thiols. In addition to reacting with
amino acid cysteine, both ebselen and the selenium reagent 2
were also tested with thiol peptides. Figure 2a displays the ESI-
MS spectrum showing the reaction of GSH (0.1 mM, the structure
of this tripeptide is shown in the figure inset) with ebselen (0.2
mM) in acetonitrile/water (1:1 by volume) containing 1% acetic
acid. As expected, the reaction product resulting from the reaction
of ebselen with GSH (m/z 583) was observed. As shown in the
CID spectrum of m/z 583 (Figure 2b), water loss (the formation
of m/z 565), backbone cleavages (the formation of m/z 508 and
m/z 454), and the protonated GSH (m/z 308) and ebselen (m/z
276) were seen, confirming the product ion structure and the
addition of peptide thiol onto ebselen (step a of eq 1 in Scheme
1). It can be seen that, besides backbone cleavages, Se-S bond
cleavage was also observed leading to the formation of the
fragment ion of m/z 276 and 308, which may be driven by the
reformation of the five-membered ring of ebselen as assisted via
the nucleophilic attack of selenium by the adjacent amide nitrogen
of the ebselen tag (shown in the inset of Figure 2b). On the basis
of this hypothesis, the CID behavior of the derivatized peptide
ions could be tuned, simply by removing the adjacent amide group
of the derivatizing selenium reagent. We thus tested the reagent
2. Figure 2c illustrates ESI-MS spectrum showing the reaction of
GSH (0.1 mM) with reagent 2 (0.5 mM) in acetonitrile/water (1:1
by volume) containing 1% acetic acid, in which the reaction product
(m/z 464) is seen. As shown in eq 2 in Scheme 1, the derivatized
thiol tag does not contain amide group as the reagent 2 is split
during the reaction. Indeed, the CID of the product ion at m/z
464 (Figure 2d) does not lead to the cleavage of the Se-S bond.
Instead, it shows the loss of one water molecule (the formation
of m/z 446), the backbone cleavages (the formation of m/z 389,
335, and 318), and the cleavage of the S-C bond (the formation
of m/z 189). In this case, the selenium tag is stable, surviving the
CID process, which would be valuable in top-down proteomics
(41) Cotgreave, I. A.; Mogenstern, R.; Engman, L.; Ahokas, J. Chem.-Biol.
Interact. 1992, 84, 69–76.
(42) Haenen, G. A. M. M.; Rooij, B. M. D.; Vermeulen, N. P. E.; Bast, A. Mol.
Pharm. 1990, 37, 412–422.
6930 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
studies (e.g., for pinpointing thiol location). In Figure 2c, one peak
at m/z 191 corresponding to the protonated PhSe(dO)OH
probably arose from the impurity of the reagent 2 used, which
was technical grade.
In addition, it appears that ebselen is more reactive than
reagent 2, probably due to the tension of its five-membered ring.
The calculated Se-N bond length of ebselen of 1.916 Å is longer
than that of reagent 2 (1.855 Å, Figure 2-S in the Supporting
Information), so the Se-N bond of ebselen is easier to break by
thiol than that of reagent 2. Further support for this conclusion
comes from the fact that nearly all of the GSH was reacted with
ebselen in the mixing ratio of 1:2 (GSH to ebselen, Figure 2a),
while a small amount of GSH remained when the ratio of GSH to
reagent 2 used was 1:5 (Figure 2c).
In the case of labeling by ebselen, the facile cleavage of Se-S
could be useful in selective detection of cysteine containing
peptides from mixtures, based on the resulting characteristic
fragment ion of m/z 276. In this experiment, selective reaction of
ebselen (0.1 mM) with GSH (50 µM) in the presence of other
peptides of angiotensin II, bradykinin, and MRFA (50 µM for each)
in methanol/water (1:1 by volume) containing 1% acetic acid was
carried out. As illustrated in Figure 2e, the ESI-MS spectrum
displays that the reaction of ebselen with GSH is highly selective,
only a product (m/z 583) resulting from the reaction of ebselen
with GSH was detected. However, the spectrum has multiple
peaks and appears complicated. Precursor ion scanning (PIS) was
applied to ions generated by ESI of the same reaction mixture
and a much clearer spectrum containing only m/z 276 (the
protonated ebselen) and 583 (the protonated ion of ebselenderivatized
GSH) are seen (Figure 2f), in which cysteinecontaining
peptide GSH can be rapidly identified.
Reversibility of a derivatization reaction is very important as it
allows enrichment and purification of analyte compounds from
complex matrices. 43 However, as already mentioned above,
commonly used thiol derivatization reactions such as nucleophilic
substitution and Michael-addition are irreversible, limiting their
analytical utilities. In this study, the reversibility of the thiol
derivatization reaction was investigated by using dithiothreitol
(DTT) as a reductant. As shown in Figure 3, the ESI-MS spectrum
reveals the recovery of GSH (m/z 308) 30 min after adding DTT
(0.5 mM) to the reaction mixture containing GSH (5 µM) and
ebselen (10 µM). In the spectrum, one can see that m/z 583, the
protonated molecule of the ebselen-derivatized GSH, disappears
upon reaction with DTT. In addition, m/z 553 arose, probably
indicating that the resulting reduction product, ebselen selenol
(compound 5, Scheme 1), was air sensitive and readily oxidized
by O 2 in the ambient environment into a more stable diselenide
Se-Se product. 41,44 The CID spectrum of m/z 553 shows the
loss of PhNH2 and the formation of the protonated ebselen (m/z
276, Figure 3-Sa in the Supporting Information), in agreement
with the peak assignment. The m/z 575 corresponds to the
sodiated molecule of the diselenide (see its CID in Figure 3-Sb in
the Supporting Information). These results clearly demonstrate
the excellent reversibility of the thiol derivatization. In combination
with the extraordinary selectivity of the reaction, this reactivity
(43) Smith, M. E. B.; Schumacher, F. F.; Ryan, C. P.; Tedaldi, L. M.; Papaioannou,
D.; Waksman, G.; Caddick, S.; Baker, J. R. J. Am. Chem. Soc. 2010, 132,
1960–1965.
(44) Engman, L.; Hallberg, A. J. Org. Chem. 1989, 54, 2964–2966.

Figure 4. (a) ESI-MS spectrum showing �-lactoglobulin A; (b) ESI-MS spectrum showing the �-lactoglobulin A fully derivatized by ebselen
(charge numbers are labeled with primes). The insets show the corresponding deconvoluted spectra; (c) high-resolution Orbitrap MS spectrum
showing the exact match of the isotopic peak distribution between the simulated (shown in blue discrete line) and detected peak (shown in black
solid line) of 16+ derivatized �-lactoglobulin A ions; (d) ESI-MS spectrum showing the partially derivatized �-lactoglobulin A (charge numbers
are labeled with double primes) 30 s after mixing the protein (5 µM) with 1,4-benzoquinone (10 µM); in this spectrum, the major peaks correspond
to intact protein ions.
can be used in a variety of applications focused on enrichment
and purification of biological thiols from cells and tissues.
Reaction with Protein Thiols. Selenium derivatizing reagents
were used for identification of free cysteine residues in proteins
due to the high selectivity for thiol groups. �-Lactoglobulin A used
as a model system in this study (162 amino acid residues) contains
two disulfide bridges (Cys 66 -Cys 160 and Cys 106 -Cys 119 ) and one
free Cys 121 . 45 Figure 4a shows the ESI-MS spectrum of �-lactoglobulin
A (5 µM) in methanol/water (1:1 by volume) containing
1% acetic acid. Multiply charged ions of �-lactoglobulin A are
detected, and deconvolution of the mass spectrum provides the
protein mass of 18 364 Da (shown in the inset of Figure 4a). In
Figure 4b, the ESI-MS spectrum shows the reaction of �-lactoglobulin
A (5 µM) and ebselen (10 µM) in methanol/water (1:1
by volume) containing 1% acetic acid. Deconvolution of the mass
spectrum of ebselen-derivatized �-lactoglobulin A provides a mass
of 18 639 Da (shown in the inset of Figure 4b). After derivatization
of �-lactoglobulin A by ebselen, a mass gain of 275 Da was
observed, indicating addition of one ebselen (MW 275 Da) to the
sole free cysteine residue of the protein. Furthermore, only the
ions of the ebselen-derivatized �-lactoglobulin A were seen in the
spectrum shown in Figure 4b, suggesting that all of the protein
was reacted with ebselen and a quantitative conversion yield was
obtained. High-resolution mass analysis using Orbitrap MS was
used for unambiguous identification of the derivatization product.
(45) Surroca, Y.; Haverkamp, J.; Heck, A. J. R. J. Chromatogr., A 2002, 970,
275–285.
Figure 4c displays the Orbitrap MS spectrum showing the exact
match between the simulated (shown in blue discrete line) and
experimentally observed (shown in black solid line) isotope
distribution of the 16+ charge state of the derivatized �-lactoglobulin
A ions. Furthermore, we performed tandem MS analysis
using the LTQ-Orbitrap. High-resolution Orbitrap CID MS 2
spectrum of the 16+ derivatized �-lactoglobulin A ions (m/z
1166, in Figure 4-S in the Supporting Information) contains
fragment ions y′139 12+ and y′130 11+ that carry the selenium tags.
This suggests that the derivatization site is located on the last
130 amino acid residues of the protein, in agreement with the
position of the free cysteine residue in �-lactoglobulin A
(Cys 121 ). These results indicate the potential use of the selenium
derivatization strategy in top-down proteomics studies.
In the case of �-lactoglobulin A derivatization, ebselen was
compared with one commonly used thiol tagging reagent, 1,4benzoquinone.
It was found by ESI-MS that the protein can be
fully reacted with ebselen 30 s after mixing (The spectrum is the
same as shown in Figure 4b). By contrast, under the same
conditions (e.g., concentrations and solvents used for the reaction
were kept the same), only ∼30% protein was derivatized with 1,4benzoquinone
30 s after mixing the protein and the reagent. This
indicates that the labeling using selenium reagents is much faster
and more efficient than using the Michael-addition reagents such
as 1,4-benzoquinone. The result suggests that the selenium
chemistry would be quite useful in the high-throughput analysis
of thiol containing proteins.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6931

Figure 5. ESI spectrum showing the derivatized protein resulting
from reacting the reduced �-lactoglobulin A with ebselen. The
superscript in the charge number shows the number of selenium tags
added to the protein after derivatization.
High selectivity and efficiency of the selenium chemistry
investigated in this study makes it useful in identification of the
number of free and bound thiol groups in proteins, which is of
importance in the protein structural analysis. The reaction of intact
protein �-lactoglobulin A with ebselen as described above shows
that the protein has only one free cysteine residue. We further
examined the derivatization reaction with reduced protein. In the
experiment, the �-lactoglobulin A was first reduced by TCEP,
which is known to be more stable and effective to reduce disulfide
bonds than DTT. 46 After reduction and removal of the excess
amount of TECP, the reagent ebselen was added to the protein
solution for thiol derivatization. As shown in Figure 5, the reduced
�-lactoglobulin A containing three selenium tags has a high
relative abundance. It is likely that the reduction of disulfide bond
of Cys 66 -Cys 160 is easier than the other Cys 106 -Cys 119 bond
leading to reduced protein mainly with three free thiols. 47
Another possible reason is that the reduced protein is partially
folded so that it is not easy for the relatively large ebselen
reagent to access all of the free thiols. Nevertheless, in Figure
5, the multiply charged ions of fully modified proteins with five
selenium tags were clearly observed, suggesting that the reduced
protein has a maximum of five free cysteine residues. This
indicates that the four additional free cysteine residues result from
reduction and the protein has two disulfide bonds prior to
reduction, which is exactly in agreement with the known structure
(46) Han, J. C.; Han, G. Y. Anal. Biochem. 1994, 220, 5–10.
(47) Sakai, K.; Sakurai, K.; Sakai, M.; Hoshino, M.; Goto, Y. Protein Sci. 2000,
9, 1719–1729.
6932 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
of �-lactoglobulin A as mentioned before. Thus, it is possible to
identify the number of free cysteine, total cysteine residues, and
disulfide linkages in proteins using a simple approach based on
the selenium chemistry.
CONCLUSIONS
In summary, ebselen and N-(phenylseleno)phthalimide, as
Se-N bond containing compounds, are excellent labeling reagents
for characterization of thiol-containing compounds by mass
spectrometry. In this study we examined a series of reactions of
these two selenium reagents with amino acids, peptides, and
proteins. Our study reveals that the thiol derivatization reaction
is highly selective, rapid, reversible, and efficient (quantitative in
the case of �-lactoglobulin A derivatization by ebselen), which is
of high value in proteomics research. In comparison to the wellknown
Michael-addition reactions used for thiol tagging, selenium
reagents appear to be more efficient and faster. Analytical
applications stemming from this investigation include (i) fast
screening of peptides/proteins containing free cysteine residues
from complex mixtures and (ii) identification of the number of
free and bound thiols of proteins and their locations using MS/
MS experiments. Given the significance of thiols in life and the
important reaction features uncovered, it is expected that there
will be many novel MS applications based on the powerful
selenium chemistry reported in this study.
ACKNOWLEDGMENT
This work was supported by U.S. NSF (Grant CHE-0911160),
National Basic Research Program of China (973 Program, Grant
2007CB936000), National Natural Science Funds for Distinguished
Young Scholar (Grant No. 20725518), National Natural Science
Foundation of China (Grant No. 20875057), and Natural Science
Foundation of Shandong Province in China (Grant No. Y2007B02).
Part of the research described in this manuscript was performed
at the W. R. Wiley Environmental Molecular Sciences Laboratory
(EMSL), a national scientific user facility sponsored by the U.S.
Department of Energy’s Office of Biological and Environmental
Research and located at Pacific Northwest National Laboratory
(PNNL). PNNL is operated by Battelle for the U.S. Department
of Energy. We also thank Mr. Xiaoyong Lu for his help.
SUPPORTING INFORMATION AVAILABLE
Additional supporting mass spectra. This material is available
free of charge via the Internet at http://pubs.acs.org.
Received for review May 4, 2010. Accepted July 6, 2010.
AC1011602

Anal. Chem. 2010, 82, 6933–6939
Difference between Ultramicroelectrodes and
Microelectrodes: Influence of Natural Convection
Christian Amatore,* Cécile Pebay, Laurent Thouin,* Aifang Wang, and J-S. Warkocz
Ecole Normale Supérieure, Département de Chimie, UMR CNRS-ENS-UPMC 8640 “Pasteur”, 24 rue Lhomond,
F-75231 Paris Cedex 05, France
Natural convection in macroscopically immobile solutions
may still alter electrochemical experiments performed
with electrodes of micrometric dimensions. A model
accounting for the influence of natural convection allowed
delineating conditions under which it interferes with mass
transport. Several electrochemical behaviors may be
observed according to the time scale of the experiment,
electrode dimensions, and intensity of natural convection.
The range of parameters in which ultramicrelectrodes
behave under a true diffusional steady state was identified.
Mapping of concentration profiles was performed experimentally
by scanning electrochemical microscopy in the
vicinity of microelectrodes of various radii. The results
validated remarkably the predictions of the model, evidencing
in particular the alteration of the diffusional
steady state by natural convection.
Microelectrodes are versatile tools for the study of electrochemical
processes of mechanistic and/or analytical interest. Their
advantageous properties stem from their small size. Microelectrodes
may be used in highly resistive environments and in very
small sample volumes. They enable the detection of very small
amounts of material and allow short time responses. 1-9 However,
the definition of a microelectrode is still nowadays ambiguous.
Actually, the notion of a microelectrode differs greatly according
to the particular origin of electrochemists, i.e., electroanalytical
chemists or molecular electrochemists. The term microelectrode
may then encompass electrodes of either millimetric or micrometric
dimensions. Electrodes of smaller sizes are referred to as
ultramicroelectrodes. Such definitions, based mainly on historical
* To whom correspondence should be addressed. E-mail: christian.amatore@
ens.fr (C.A.); laurent.thouin@ens.fr.
(1) Fleischmann, M.; Pons, S.; Rolison, D. R. Ultramicroelectrodes; Datatech
Systems, Inc.: Morgantown, NC, 1987.
(2) Bond, A. M.; Oldham, K. B.; Zoski, C. G. Anal. Chim. Acta 1989, 216,
177–230.
(3) Wightman, R. M.; Wipf, D. O. Electroanalytical Chemistry; Marcel Dekker:
New York, 1989; Vol. 15, pp 267-353.
(4) Montenegro, M. I.; Queiros, M. A.; Daschbach, J. L. Microelectrodes: Theory
and Applications; Kluwer Academic Press: Dordrecht, The Netherlands,
1991; Vol. 197.
(5) Aoki, K. Electroanalysis 1993, 5, 627–639.
(6) Heinze, J. Angew. Chem., Int. Ed. 1993, 32, 1268–1288.
(7) Amatore, C. Electrochemistry at ultramicroelectrodes. In Physical Electrochemistry;
Rubinstein, I., Ed.; Marcel Dekker: New York, 1995.
(8) Stulik, K.; Amatore, C.; Holub, K.; Marecek, V.; Kutner, W. Pure Appl. Chem.
2000, 72, 1483–1492.
(9) Forster, R. J. Encyclopedia of Electrochemistry; John Wiley & Sons: New
York, 2003; Vol. 3, pp 160-195.
criteria, may appear useless since they better define the origin of
the users than the object itself. A better classification of these
electrodes would be obtained if it were based on their particular
properties. Since electrochemical reactions are interfacial reactions,
mass transport is one of the key processes to consider. 10
In a liquid, elementary contributions in the mass transport are
diffusion, migration, and convection. Under most circumstances,
migration is suppressed by adding a large excess of dissociated
inert salt or supporting electrolyte. Convection is often neglected
at electrodes of micrometric dimensions in macroscopically still
solutions. Indeed, convection originates from movement of fluid
packets of micrometric size. 11 It necessarily vanishes close to the
electrode interface over distances where concentrations differ
significantly from their bulk values. 12,13 In such a case, only
diffusion is assumed to govern the final approach of an electroactive
molecule toward the electrode. However, according to the
size of these electrodes and time scale of the experiments,
convective fluxes due to natural convection may still compete with
diffusional fluxes in motionless solutions. This may occur even
in the absence of any density gradients 14 or effects induced by a
magnetic field. 15 These situations arise as soon as the thickness
of the diffusion layer becomes comparable to the thickness of the
convection-free domain. 7 Under such conditions, the responses
do not follow the classical relationships given for currents in
dynamic and steady-state regimes. Therefore, under given experimental
conditions, it is of importance to decide the largest
size of an electrode for eliminating any influence of natural
convection. 16,17 Such a criterion may then allow distinguishing
properties of ultramicroelectrodes from those of other electrodes
of micrometric sizes.
To assess the conditions of convection-free regimes at electrodes
of micrometric dimensions, we investigated in some
previous studies the current responses of micrometric disk
(10) Bard, A. J.; Faulkner, L. R. Electrochemical Methods, 2nd ed.; John Wiley &
Sons: New York, 2001.
(11) Moreau, M.; Turq, P. Chemical Reactivity in Liquids: Fundamental Aspects;
Kluwer Academic/Plenum Press: New York, 1988; pp 561-606.
(12) Levich, V. G. Physicochemical Hydrodynamics; Prentice Hall: Englewoods
Cliffs, NJ, 1962.
(13) Davies, J. E. Turbulence Phenomena; Academic Press: New York, 1972.
(14) Li, Q. G.; White, H. S. Anal. Chem. 1995, 67, 561–569.
(15) Grant, K. M.; Hemmert, J. W.; White, H. S. J. Electroanal. Chem. 2001,
500, 95–99.
(16) Hapiot, P.; Lagrost, C. Chem. Rev. 2008, 108, 2238–2264.
(17) Molina, A.; Gonzalez, J.; Martinez-Ortiz, F.; Compton, R. G. J. Phys. Chem.
C 2010, 114, 4093–4099.
10.1021/ac101210r © 2010 American Chemical Society 6933
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/26/2010

electrodes in various conditions, in chronoamperometry 18 and
cyclic voltammetry. 19 The mapping of concentration profiles was
alsoperformedintheirvicinityusingamethodalreadydescribed. 18-23
At the same time, we proposed a theoretical model to evaluate
the influence of natural convection on mass transport in still
media. 18 The good agreement observed between theory and
experiments demonstrated the validity of this model over a wide
range of experimental conditions. 18,22-26 The purpose of this study
is then to take the benefit of this model to delineate the
experimental conditions that allow convection-free regimes to be
observed in dynamic and steady-state regimes. These conditions
are better presented in a zone diagram showing the influence of
all the parameters: time scale of the experiment, electrode radius,
and thickness of the convection-free domain. Comparison with
experimental data will also serve to illustrate the relative contributions
of convection and diffusion at electrodes of different sizes
performing under the steady-state regime.
MODEL OF NATURAL CONVECTION
The influence of convection on the electrode responses can
be quantified from deviations of their diffusive currents or from
alteration of their diffusion layers. Under pure diffusional conditions,
the concentration profile of a species at a disk electrode is
given by integration of Fick’s second law: 10
∂c(r, z, t)
∂t
) D( ∂2c(r, z, t)
∂r 2
+ ∂2c(r, z, t)
∂z 2
+ 1 ∂c(r, z, t)
r ∂r )
(1)
where r describes the radial position normal to the axis of
symmetry at r ) 0 and z describes the linear displacement normal
to the plane of the electrode at z ) 0. D is the diffusion coefficient.
For a chronoamperometric experiment, the pertinent boundary
conditions are
t < 0; r, z g 0; c(r, z, t) ) c° (2)
t g 0; r e r 0 ; c(r,0,t) ) 0 (3)
r, z f ∞; c(r, z, t) ) c° (4)
The current is readily obtained from integration of the concentration
gradients at the electrode surface with
(18) Amatore, C.; Szunerits, S.; Thouin, L.; Warkocz, J. S. J. Electroanal. Chem.
2001, 500, 62–70.
(19) Amatore, C.; Pebay, C.; Thouin, L.; Wang, A. F. Electrochem. Commun.
2009, 11, 1269–1272.
(20) Amatore, C.; Pebay, C.; Scialdone, O.; Szunerits, S.; Thouin, L. Chem.sEur.
J. 2001, 7, 2933–2939.
(21) Amatore, C.; Szunerits, S.; Thouin, L.; Warkocz, J. S. Electroanalysis 2001,
13, 646–652.
(22) Amatore, C.; Knobloch, K.; Thouin, L. Electrochem. Commun. 2004, 6, 887–
891.
(23) Baltes, N.; Thouin, L.; Amatore, C.; Heinze, J. Angew. Chem., Int. Ed. 2004,
43, 1431–1435.
(24) Rudd, N. C.; Cannan, S.; Bitziou, E.; Ciani, L.; Whitworth, A. L.; Unwin,
P. R. Anal. Chem. 2005, 77, 6205–6217.
(25) Amatore, C.; Sella, C.; Thouin, L. J. Electroanal. Chem. 2006, 593, 194–
202.
(26) Amatore, C.; Knobloch, K.; Thouin, L. J. Electroanal. Chem. 2007, 601,
17–28.
6934 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
r0 ∂c(r, z, t)
i )(2πnFD∫ r ∂r (5)
0 ∂z
In still solutions, natural convection operates perpendicularly
to the electrode surface. It is based on microscopic motions of
the solution except in the very near vicinity of the electrodes,
where it vanishes. Experimentally, the resulting velocity field is
extremely difficult to estimate. Moreover, it is almost impossible
to master mathematically since it depends on many parameters
which are not easy to control (vibrations, temperature gradients,
movement of the cell atmosphere, etc.). However, beyond these
difficulties, we demonstrated successfully that, for electroactive
species, the influence of natural convection can be assimilated to
that of an apparent diffusion coefficient depending on the
orthogonal distance z from the electrode plane. 18 Moreover, since
the electrochemical perturbation affects only the viscous sublayer
adjacent to the electrode, we showed that Dapp could be evaluated
by
z
Dapp ) D( 1 + 1.522( δconv) 4
)
where δconv is the thickness of the convection-free layer. This
is the only parameter introduced into the model to account for
the effects of natural convection. It is possible to evaluate its
influence on the electrode response by replacing D by Dapp in
eq 1 and solving numerically the new mass transport equation in
association with the same boundary conditions (eqs 2-4).
EXPERIMENTAL SECTION
All the solutions were prepared in purified water (Milli-Q,
Millipore). A 10 mM concentration of K4Fe(CN)6 (Acros) was
dissolved in 1 M KCl (Aldrich), which was used as the
supporting electrolyte. Reciprocally 2 mM FcCH2OH (Acros)
was prepared in 0.1 M KNO3 (Fluka). The diffusion coefficients
were DFe ) (6.0 ± 0.5) × 10 -6 cm 2 s -1 27 for Fe(CN)6 4- /
Fe(CN)6 3- and DFc ) (7.6 ± 0.5) × 10 -6 cm 2 s -1 for FcCH2OH/
FcCH2OH + .
The working electrodes were Pt disk electrodes of 12.5, 25,
62.5, 125, 250, and 500 µm radii. They were obtained from the
cross section of Pt wires (Goodfellow) sealed into soft glass. The
reference electrode was a Ag/AgCl electrode, and the counter
electrode was a platinum coil. A scanning electrochemical microscope
(910B CH Instruments) was used to establish the concentration
profiles. The amperometric probe was a Pt disk electrode
of submicrometric dimension (∼500 nm radius). Its fabrication
and the related procedure to map the concentrations have already
been reported. 23 For Fe(CN)6 4- /Fe(CN)6 3- experiments, the
working electrode was biased at +0.6 V/ref on the oxidation
plateau of Fe(CN)6 4- . The probe was biased at +0.6 V/ref to
collect Fe(CN)6 4- or -0.1 V/ref to collect Fe(CN)6 3- . For
FcCH2OH/FcCH2OH + experiments, the working electrode was
biased at +0.25 V/ref. In this case, the probe was biased at
+0.25 V/ref to collect FcCH2OH and -0.1 V/ref to collect
FcCH2OH + .
The mass transport equation was solved numerically in the
conformal space adapted to the geometry of a microdisk elec-
(27) Amatore, C.; Szunerits, S.; Thouin, L.; Warkocz, J.-S. Electrochem. Commun.
2000, 2, 353–358.
(6)

trode 28 by a finite element using Comsol Multiphysics software.
The thickness of the convection-free layer, δconv, was determined
before each experiment by chronoamperometry as described
previously. 18
RESULTS AND DISCUSSION
An important property of disk ultramicroelectrodes is that their
diffusion layers develop with time until reaching a steady-state
limit imposed by hemispherical-type diffusion. However, natural
convection may interfere with the mass transport as soon as the
thickness of the expanding diffusion layer becomes comparable
to δconv. In such a case, a steady-state regime is still achieved
but is then controlled by the respective contributions of
diffusion and natural convection. Depending on the electrode
radius, r0, and the thickness of the convection-free layer, δconv,
two situations may be encountered, whether the diffusion at
the electrode surface is planar or not. Indeed, at short time
scales, the thickness of the diffusion layer is considerably
smaller than the electrode radius. The electrodes then behave
as electrodes of infinite dimensions, and planar diffusion
operates. In that particular situation, the Cottrell equation
applies with
i planar )( nFADc°
√πDt
for electrodes of surface area A. Using the Nernst formulation,
eq 7 is similar to that given in hydrodynamic electrochemical
methods: 10
i )( nFADc°
δ
where δ ) (πDt) 1/2 . Therefore, natural convection interferes
significantly with the mass transport as soon as δconv ≈ (πDt) 1/2 .
Whenever this condition is not met, the diffusion layer may
develop, possibly reaching a hemispherical behavior until being
eventually limited by δconv.
However, since diffusion and natural convection occur together
in steady-state regimes, their contributions in mass transport
remain difficult to comprehend. To solve this problem, one needs
first to investigate the concentration profiles established under
the pure diffusional steady-state regime, i.e., without any influence
of natural convection. In such a case, solution of eq 1 shows that
most of the concentration gradients operate over a distance
comparable to the electrode radius (Figure 1A). Concentration
along the z axis (Figure 1B) varies according to
c 2 z
) arctan( c° π r0) The diffusion layer presents a hemispherical shape, and the
steady-state current is given by:
(7)
(8)
(9)
i hemisph )(4nFr 0 Dc° (10)
(28) Amatore, C.; Fosset, B. J. Electroanal. Chem. 1992, 328, 21.
Figure 1. Steady-state concentration profile simulated at a disk
electrode without considering the influence of natural convection: (A)
2D concentration profile with isoconcentration lines ranging from c/c°
) 0.1 to c/c° ) 0.9. (B) Concentration profile along the vertical axis
of symmetry.
Using the Nernst formulation again, comparison between eqs 8
and 10 leads to an equivalent diffusion layer thickness, δ ) πr0/
4. This thickness differs slightly from δz, which may be obtained
by extrapolating the concentration gradient at z ) 0, r ) 0 along
the z axis. Indeed, when z f 0, the concentration profile at r
) 0 tends to (Figure 1B)
c 2 z
)
c° π r0 (11)
which gives δz ) πr0/2. The difference in the δ and δz values
results from the nonradial distribution of diffusion fields at disk
electrodes. Concentration gradients are higher at the electrode
edges than at the center (Figure 1A). Since δ is evaluated from
the integration of concentration gradients over the whole electrode
surface (eq 5), it is necessarily smaller than δz. In the following,
the variation of these two parameters will provide an accurate
estimation of the influence of natural convection, either from
the current (i.e., through δ) or from the alteration of concentration
profiles in the z direction (i.e., through δz), where natural
convection prevails.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6935

Figure 2. Steady-state concentration profiles and steady-state
currents simulated at disk electrodes of different radii under the
influence of natural convection. (A) Isoconcentration lines c/c° ) 0.9
for disk electrodes of various radii. From left to right, r0/δconv ) 0.05,
0.5, 1.0, 1.5, 2.0, 4.2, and 6.0. (B, C) Variations of the diffusion layer
thicknesses δ and δz as a function of the electrode radius, with (solid
curves) and without (dashed curves) the influence of natural convection.
(D) Error on steady-state currents due to the influence of natural
convection as a function of r0/δconv.
Figure 2A displays isoconcentration lines c/c° ) 0.9 calculated
under the influence of natural convection for electrodes of various
radii. They were obtained from combination of eqs 1-6 under
the steady-state regime. To provide a more general representation
of the problem, the spatial coordinates r and z were normalized
by δconv. When the electrode radii are small enough (i.e., r0/
δconv < 0.5), one observes that the diffusion layers retain their
quasi-hemispherical shapes, like those previously simulated in
the absence of natural convection (Figure 1A). For larger
electrode radii, the concentration profiles become flattened, their
expansion along the z axis being restricted by the boundary at
δconv. Therefore, when convection operates, it has two major
effects depending on the ratio r0/δconv. On one hand, concentration
gradients (∂c/∂z)z)0 become more uniform over the central
area of the electrodes than when natural convection is absent.
On the other hand, the development of the diffusion layers still
operates laterally along the r axis. This can be easily observed
in parts B and C of Figure 2, where the variations of δ/δconv and
δz/δconv, respectively, are reported as a function of r0/δconv. In
particular, one observes that when r0/δconv ) 4, δz has reached
6936 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. Zone diagrams describing the influence of natural convection
on planar and hemispherical diffusion at disk electrodes. (A) Zone
diagram established from two boundary conditions, δ ) δconv (solid curve)
and r0/(πDt) 1/2 ) 4/π (vertical straight line). Note that this diagram is
independent of the model of convection. (B) Zone diagram established
on the basis of the present model of natural convection with four
boundary conditions: |δ - δdiff|/δ or |i - idiff|/|idiff| ) 0.1 (curve 1), |δ -
δconv|/δ or |i - iconv|/|iconv| ) 0.1 (curve 2), |δ - δplanar|/δ or |i - iplanar|/
|iplanar| ) 0.1 (curve 3), and |δ - δhemisph|/δ or |i - ihemisph|/|ihemisph| ) 0.1
(curve 4). The black symbols correspond to the experimental conditions
considered in Figure 4: from left to right, r0 ) 12.5, 25, 62.5, 125, 250,
and 500 µm, with δconv ) 200-250 µm.
its limit δconv whereas δ is only equal to 0.8δconv. Accordingly,
the convolution of these two effects on the development of
diffusion layers leads to a deviation of the currents from eq
10. This alteration may be drastic since the relative error |i -
ihemisph|/|ihemisph| increases almost linearly with r0/δconv (Figure
2D). In particular, |i - ihemisph|/|ihemisph| ≈ 0.65 when δconv ≈ r0.
A first attempt to summarize all these situations is to establish
a zone diagram describing the boundary condition imposed by
δconv on δ, whether the diffusion is planar or hemispherical.
As previously mentioned, two other parameters have to be
considered: the electrode radius, r0, and the diffusion length,
(πDt) 1/2 . The transition between planar and quasi-hemispherical
diffusion depends on the ratio r0/(πDt) 1/2 , whereas the influence
of convection is fixed by r0/δconv. Therefore, a diagram
with two coordinates, r0/(πDt) 1/2 and r0/δconv, allows plotting
the limit, which differentiates the domains where convection
or diffusion prevails independently. This limit is then δ ) δconv.

Figure 4. Comparison between simulated (curves) and experimental (symbols) steady-state concentration profiles along the vertical axis of
symmetry at disk electrodes of different radii when the electrode potential is poised on the oxidation plateau of Fe(CN)6 4- . Concentration profiles
simulated without (dashed curves) or with (solid curves) natural convection (δconv ) 200-250 µm). Experimental concentration profiles of the
substrate Fe(CN)6 4- (0) and product Fe(CN)6 3- (O). t ) 60 s. [Fe(CN)6 4- ] ) 10 mM in 1 M KCl.
The diagram is reported in Figure 3A, where the zones above
and below this limit correspond to the control of convection and
diffusion, respectively. In the lower zone, the equality between
eqs 7 and 10 discriminates by a vertical line located at r0/(πDt) 1/2
) 4/π two other domains where planar diffusion (i.e., r0/
(πDt) 1/2 > 4/π) and quasi-hemispherical diffusion (i.e., r0/
(πDt) 1/2 < 4/π) dominate.
One must note that this diagram is independent of the model
of natural convection since δ was calculated without considering
eq 6 and by only assuming δ ) δconv. In this context, the model
enables the transitions between the three regimes of the
diagram to be determined. For this purpose, the model is used
to evaluate δ and to compare its value to a given reference
thickness, δref, predicted by considering only one specific
regime: (1) δref ) δdiff for pure diffusion control without any
influence of convection, (2) δref ) δconv, (3) δref ) δplanar for
planar diffusion with δplanar ) (πDt) 1/2 , and (4) δref ) δhemisph
for hemispherical diffusion with δhemisph ) πr0/4. The transition
from one of these specific regimes to the others may then be
estimated by setting a relative threshold on δ such as
Note that eq 12 is equivalent to
|δ - δref |
) 0.1 (12)
δ
|
i - iref i | ) 0.1 (13)
ref
where iref is the reference current obtained from eq 8 with
i ref )( nFADc°
δ ref
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(14)
6937

The zone diagram built from eq 12 or 13 is reported in Figure
3B. Thus, three curves (curves 2-4) delineate a new domain
corresponding to a mixed regime between the limiting ones
previously identified (Figure 3A). In particular, the transitions from
hemispherical diffusion to convection (i.e., vertical displacement
on the diagram) and from hemispherical diffusion to planar
diffusion (i.e., horizontal displacement) are relatively broad since
they occur approximately over 2 orders of magnitude on the r0/
δconv and r0/(πDt) 1/2 scales, respectively. Only the transition
from planar diffusion to convection is very sharp. Indeed, as
soon as δplanar ≈ δconv, the diffusion layer reaches its steadystate
limit and mass transport is fully controlled by convection.
In contrast, when the condition δz ≈ δconv is met for hemispherical-type
diffusion, the layer may still expand laterally
along the r axis until reaching its steady-state limit (see Figure
2A-C). This latter condition corresponds to curve 1 in Figure
3B when |δ - δdiff|/δ ) 0.1 or |i - idiff|/|idiff| ) 0.1. It allows
delineating the upper zone of the diagram where convection
starts to interfere in the mass transport.
A chronoamperometric experiment can be represented on the
diagram by a horizontal straight line described from the right to
the left when the time duration increases. According to the size
of the electrode, r0, and thickness, δconv, the nature of the steadystate
regime reached at longer time may be different. On the
one hand, if log(r0/δconv) > 0.95, a sharp transition from planar
diffusion to convection occurs. On the other hand, if log(r0/
δconv) < -0.7, a broad transition with a mixed regime from
planar diffusion to quasi-hemispherical diffusion operates
without any influence of natural convection. When log(r0/
(πDt) 1/2 ) < -0.75, a steady-state regime is always observed
though its nature (diffusional or convective) only depends on
the ratio r0/δconv.
Under given experimental conditions (i.e., the same position
of the electrode in the cell, temperature, viscosity of the electrolyte,
environment, etc.), δconv is approximately constant so that the
mass transport regime under steady state depends only on the
electrode dimension. This was checked experimentally by
mapping diffusion layers in the vicinity of electrodes of various
radii. Figure 4 shows the concentration profiles along the vertical
axis of symmetry of the electrodes for both the reactant and
product. δconv was evaluated independently by chronoamperometry
at a large electrode 18 and was found to range from 200
to 250 µm. It was thus possible to compare the experimental
data with concentration profiles predicted with or without
natural convection. A very good agreement was observed in
Figure 4 whatever the size of the electrodes between experimental
data and predictions issued from the model when natural convection
was taken into account. Alterations on the concentration
profiles due to convection were apparent as soon as r0 ) 25 µm.
The experimental conditions pertaining to each concentration
profile in Figure 4 are reported as symbols in the zone diagram
of Figure 3B. According to the threshold previously defined with
|δ - δhemisph|/δ ) 0.1 or |i - ihemisph|/|ihemisph| ) 0.1, the results
show that a hemispherical diffusion regime was reached for r0
) 12.5 and 25 µm while a mixed regime was achieved for the
other radii (r0 ) 62.5-500 µm).
These experimental data validate the predictions of the present
model, yet they involved only the effect of natural convection along
6938 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. Comparison between simulated (curves) and experimental
(symbols) steady-state concentration profiles at a disk electrode of
radius r0 ) 25 µm when the electrode potential is poised onto the
oxidation plateau of FeCH2OH. (A) Experimental concentration profiles
of the product FcCH2OH + along the vertical axis of symmetry (circles).
(B) Experimental concentration profiles of FcCH2OH + along the r axis
at various vertical distances z: z ) 6(O), 16 (0), 26 (]), 36 (×), 46
(+), and 56 µm (∆). The black area indicates the extent of the
electrode coordinates along the r axis. The concentration profiles are
simulated without (dashed curves) and with (solid curves) consideration
of the influence of natural convection (δconv ) 200 µm).
[FeCH2OH] ) 2 mM in 0.1 KNO3.
the axis of symmetry of the electrodes. Conversely, we showed
above (see Figure 2) that this effect is also effective along lateral
directions due to the compensation of transport between vertical
and lateral fluxes. In the following, we investigated this latter issue
experimentally by performing 2D imaging. Figure 5 reports the
mapping of concentration profiles established in the steady-state
regime along the z axis and r axis when r0 ) 25 µm. As in Figure
4, the concentration profiles were compared to the predictions
established with and without the influence of convection. Apart
from the good agreement obtained between the data and predictions,
these results clearly illustrate the fact that convection may
still alter the diffusion layers even when quasi-hemispherical
diffusion is expected to prevail (Figure 3B). In the present case,
the concentration profiles are distorted over distances z equivalent
to 10 times the electrode radius, r0. Simultaneously, the relative

Figure 6. Comparison between simulated (curves) and experimental
(symbols) thicknesses of the diffusion layer at disk electrodes of
various radii: δz/δconf (dashed lines, O) and δ/δconf (solid curve, 0).
error in the current obtained by the model is |i - ihemisph|/
|ihemisph| ) 0.07.
Finally, variation of δz issued from the mapping of concentration
profiles in Figure 4 is reported in Figure 6 as a function of
the electrode size and then compared to the predicted one. The
diffusion layer thicknesses, δ, estimated from the experimental
steady-state currents (through eq 8) are also plotted. As observed,
all these data show that the model applies satisfactorily under the
steady-state regime to predict the influence of natural convection
on current responses or concentration profiles.
CONCLUSION
The model elaborated in this work predicts within a very
good accuracy the relative contributions of diffusion and natural
convection to the mass transport at disk electrodes. The
electrochemical behaviors of the electrodes not only are related
to their dimensions but also depend on the time scale of the
experiment and thickness of the convection-free layer (i.e.,
δconv). These results stress once more the futility of trying
to propose an absolute definition of ultramicroelectrodes
based on the objects themselves. Indeed, the same electrode
may behave as a microelectrode or an ultramicroelectrode,
depending on these parameters. Our model allowed us to
clearly delineate the situations where natural convection
alters both the dynamic and steady-state regimes at disk
electrodes. The properties of ultramicroelectrodes are mainly
achieved when r0/δconv < 0.2. This condition has practical
consequences if one needs, for example, to exploit the
characteristics of ultramicroelectrodes to detect or measure
concentrations in restricted volumes, without any alteration
of natural convection on the measurements.
ACKNOWLEDGMENT
This work has been supported in part by the CNRS (Grant
UMR8640), Ecole Normale Superieure, UPMC, and French
Ministry of Research.
Received for review May 7, 2010. Accepted July 9, 2010.
AC101210R
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6939

Anal. Chem. 2010, 82, 6940–6946
Determination of Double Bond Location in Fatty
Acids by Manganese Adduction and Electron
Induced Dissociation
Hyun Ju Yoo and Kristina Håkansson*
Department of Chemistry, University of Michigan, 930 North University Avenue, Ann Arbor, Michigan 48109-1055
Double bond locations in fatty acids can be determined
from characteristic charge-remote fragmentation patterns
of alkali metal-adducted fatty acids following high energy
collision activated dissociation (CAD). With low energy
CAD, several chemical derivatization methods, including
ozonization, epoxidation, and hydroxylation, have been
used to generate characteristic fragments. However, high
energy CAD is not universally available and involves a high
degree of scattering, causing product ion loss. Further,
derivatization reactions involve side reactions and sample
loss. Here, we analyzed metal-adducted fatty acids to
investigate the utility of electron induced dissociation
(EID) for determining double bond location. EID has been
proposed to involve both electronic excitation, similar to
high energy CAD, and vibrational exciation. Various
metals (Li, Zn, Co, Ni, Mg, Ca, Fe, and Mn) were
investigated to fix one charge at the carboxylate end of
fatty acids to promote charge-remote fragmentation. EID
of Mn(II)-adducted fatty acids allowed determination of
all double bond locations of arachidonic acid, linolenic
acid, oleic acid, and stearic acid. For Mn(II)-adducted
fatty acids, reduced characteristic charge-remote product
ion abundances at the double bond positions are indicative
of double bond locations. However, other metal
adducts did not generally provide characteristic product
ion abundances at all double bond locations.
Polyunsaturated fatty acids are essential for cell membrane
functioning because many membrane properties, such as
fluidity and permeability, are closely related to the level of
unsaturation. 1,2 Lipid peroxidation results in loss of membrane
polyunsaturated fatty acids. 1 Also, when an oil or fat becomes
oxidized, health concern is due to the potential production of
free radicals, which can be highly carcinogenic. 3-6 Double bond
sites in unsaturated fatty acids and lipids are plausibly oxidized
* To whom correspondence should be addressed. E-mail: kicki@umich.edu.
Tel: (734) 615-0570. Fax: (734) 647 4865.
(1) Farooqui, A. A.; Horrocks, L. A. Cell. Mol. Neurobiol. 1998, 18, 599–608.
(2) Mitchell, T. W.; Pham, H.; Thomas, M. C.; Blanksby, S. J. J. Chromatogr.,
B 2009, 877, 2722–2735.
(3) Yao, D.; Shi, W.; Gou, Y.; Zhou, X.; Yee Aw, T.; Zhou, Y.; Liu, Z. Free Radical
Biol. Med. 2005, 39, 1385–1398.
(4) North, J. A.; Spector, A. A.; Buettner, G. R. Am. J. Physiol. 1994, 267, C177–
188.
(5) Pandey, M.; Sharma, L. B.; Singh, S.; Shukla, V. K. World J. Surg. Oncol.
2003, 1, 5.
(6) Black, H. S. Integr. Cancer Ther. 2004, 3, 279–293.
6940 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
and form free radicals, which can cause tissue damage and
alterations in cell membranes. 1,3,4,7,8 Thus, the identification
of double bond locations in fatty acids can be beneficial for
understanding lipid biology and also its related disease states. 1,2
Double bond locations in aliphatic compounds, including fatty
acids, can be obtained from mass spectrometry (MS). 9-14 Such
information can be determined by charge-remote fragmentation
processes of alkali metal-adducted fatty acids in high energy collisional
activation with fast atom bombardment (FAB) ionization. 9,15
However, FAB desorption of fatty acid mixtures can result in
preferential desorption of some ions, chemical noise, and low
sensitivity. 17,19 In addition, sector-type instruments for high energy
collision activated dissociation (CAD) are not universally available,
and high energy CAD involves a high degree of scattering, causing
product ion loss. 20 Recently, low energy CAD of Cu(II)-adducted
fatty acids in an ion trap instrument was used to provide diagnostic
product ions to aid the identification of double bond locations in
unsaturated fatty acids; 32 however, double bond localization
remains challenging for monounsaturated fatty acids.
Charge remote fragmentation occurs remote from a charge
site and appears to readily occur in high energy CAD of fatty acids,
lipids, steroids, and other compounds containing long alkyl
chains. 15,17,33 Charge remote fragmentation has been used to
provide double bond positions of fatty acids from FAB-MS/MS
(high energy CAD) of lithiated fatty acids using sector-type mass
spectrometry by Gross and others. 17,33,34 More recently, tandem
time-of-flight (TOF/TOF) MS was applied to determine double
bond locations of lithiated fatty acids by McEwen and co-workers
using solvent-free matrix-assisted laser desorption ionization
(MALDI). 10 Charge-remote bond cleavages in fatty acids are
(7) de Kok, T. M.; ten Vaarwerk, F.; Zwingman, I.; van Maanen, J. M.; Kleinjans,
J. C. Carcinogenesis 1994, 15, 1399–1404.
(8) Montine, T. J.; Morrow, J. D. Am. J. Pathol. 2005, 166, 1283–1289.
(9) Adams, J.; Gross, M. L. Anal. Chem. 1987, 59, 1576–1582.
(10) Trimpin, S.; Clemmer, D. E.; McEwen, C. N. J. Am. Soc. Mass Spectrom.
2007, 18, 1967–1972.
(11) Van Pelt, C. K.; Brenna, J. T. Anal. Chem. 1999, 71, 1981–1989.
(12) Buser, H. R.; Arn, H.; Cuerin, P.; Rauscher, S. Anal. Chem. 1983, 55, 818–
822.
(13) Schneider, B.; Budzikiewicz, H. Rapid Commun. Mass Spectrom. 1990, 4,
550–551.
(14) Malosse, C.; Kerhoas, L.; Einhorn, J. J. Chromatogr. 1998, 803, 203–209.
(15) Cheng, C.; Gross, M. L. Mass Spectrom. Rev. 2000, 19, 398–420.
(16) Hayes, R. N.; Gross, M. L. Methods Enzymol. 1990, 193, 237–263.
(17) Adams, J. Mass Spectrom. Rev. 1990, 9, 141–186.
(18) Jensen, N. J.; Tomer, K. B.; Gross, M. L. J. Am. Chem. Soc. 1985, 107,
1863–1868.
(19) Adams, J.; Gross, M. L. Org. Mass Spectrom. 1988, 23, 307–316.
(20) Fallick, A. E. Int. J. Mass Spectrom. Ion Phys. 1983, 46, 59–62.
10.1021/ac101217x © 2010 American Chemical Society
Published on Web 07/27/2010

educed in the vicinity of double bonds, thus providing information
on double bond locations. 16 Fixed charge sites are important for
predominant charge-remote fragmentation processes because
charge-driven processes compete with charge-remote processes.
In charge-driven fragmentation, charge migration results in
rearrangement of chemical structure, making identification of
double bond positions impossible. 15 High energy CAD is known
to involve electronic excitation as a dominant process for generating
charge-remote product ions, and vibrational/rotational excitation
is considerably less efficient. 16
Electron induced dissociation (EID) involves interactions
between singly charged analyte ions and free electrons. This
concept was first shown with 3-9 eV electrons in 1979 by Cody
and Freiser for radical cations. 21 EID does not require multiply
charged precursor ions, thereby differing from other ion-electron
interactions such as electron capture dissociation (ECD) 22-25 and
electron detachment dissociation (EDD). 26-28 Consequently, EID
is compatible with smaller biomolecules (such as fatty acids) for
which formation of gas-phase multiply charged ions is energetically
unfavorable. Zubarev and co-workers applied EID (10-13
eV electron irradiation) to singly charged oligosaccharide cations,
29 and we have applied EID for structural characterization of
phosphate-containing metabolites. 30 EID of phosphate-containing
metabolites provided complementary structural information compared
to CAD and infrared multiphoton dissociation (IRMPD) and
generally generated more extensive fragmentation than the latter
two techniques. 44 O’Hair and co-workers have proposed that
EID occurs via electronic and vibrational excitation, based on
similarities between the types of product ions observed from
EID, ultraviolet photodissociation, and electron ionization (EI)
mass spectra. 31 Thus, EID may be an alternative technique to
high-energy CAD for revealing information on double bond
locations in fatty acids. To our knowledge, EID has not
previously been applied toward fatty acid analysis. We used
several metals (Li, Zn, Co, Ni, Mg, Ca, Fe, and Mn) to fix a
positive charge at the end of fatty acids. Mn(II) adduction
consistently generated charge-remote fragmentation in EID.
The charge-remote product ion abundances at each carbon
location were compared to deduce double bond positions. EID
of Mn(II)-adducted arachidonic acid was compared to IRMPD
of the same species to illustrate that EID involves both
electronic and vibrational excitation in contrast to IRMPD
which only involves the latter.
(21) Cody, R. B.; Freiser, B. S. Anal. Chem. 1979, 51, 541–551.
(22) Zubarev, R. A.; Horn, D. M.; Fridriksson, E. K.; Kelleher, N. L.; Kruger,
N. A.; Lewis, M. A.; Carpenter, B. K.; McLafferty, F. W. Anal. Chem. 2000,
72, 563–573.
(23) Zubarev, R. A.; Kelleher, N. L.; McLafferty, F. W. J. Am. Chem. Soc. 1998,
120, 3265–3266.
(24) Zubarev, R. A. Curr. Opin. Biotechnol. 2004, 15, 12–16.
(25) Cooper, H. J.; Hakansson, K.; Marshall, A. G. Mass Spectrom. Rev. 2005,
24, 201–222.
(26) Budnik, B. A.; Haselmann, K. F.; Zubarev, R. A. Chem. Phys. Lett. 2001,
299–302.
(27) Yang, J.; Mo, J.; Adamson, J. T.; Hakansson, K. Anal. Chem. 2005, 77,
1876–1882.
(28) Wolff, J. J.; Chi, L. L.; Linhardt, R. J.; Amster, I. J. Anal. Chem. 2007, 79,
2015–2022.
(29) Budnik, B. A.; Haselmann, K. F.; Elkin, Y. N.; Gorbach, V. I.; Zubarev, R. A.
Anal. Chem. 2003, 75, 5994–6001.
(30) Yoo, H. J.; Liu, H.; Hakansson, K. Anal. Chem. 2007, 20, 7858–7866.
(31) Lioe, H.; O’Hair, R. A. Anal. Bioanal. Chem. 2007, 389, 1429–1437.
EXPERIMENTAL SECTION
Sample Preparation. Fatty acids used in this work include
stearic acid, oleic acid, linolenic acid, and arachidonic acid. Fatty acids
and metal salts, including MnCl2, CoBr2, and NiBr2, were purchased
from Sigma-Aldrich (St. Louis, MO). Fatty acid (70-200 µM) was
mixed with 200-600 µM metal salt in methanol/water (80/20
v/v). Sample solutions of metal (Met)-adducted fatty acids were
freshly made 10-30 min prior to MS analysis.
Fourier Transform Ion Cyclotron Resonance Mass Spectrometry.
Singly charged metal-adducted fatty acids, [M + Met
- H] + ([M + 2Met - H] + for Li), were generated by external
electrospray ionization (ESI) at 70 µL/h (Apollo II dual stage
ion funnel ion source, Bruker Daltonics, Billerica, MA). All
experiments were performed with a 7 T quadrupole (Q)-FTICR
mass spectrometer (APEX-Q, Bruker Daltonics) as previously
described. 28 All data were obtained in positive ion mode.
Briefly, ions produced by ESI were mass-selectively externally
accumulated 32,33 in a hexapole for 0.1-2 s, transferred via high
voltage ion optics, and captured in the ICR cell by dynamic
trapping. This accumulation sequence was looped three times to
improve precursor ion abundance. In MS/MS experiments, mass
selective external accumulation of [M + Met - H] + ([M + 2Met
- H] + for Li) was employed. In some cases, mass selective
external accumulation was followed by further isolation via
correlated harmonic excitation fields (CHEF) 34 inside the ICR
cell to eliminate unwanted peaks caused by impurities and
byproducts from adduct forming reactions. An indirectly heated
hollow dispenser cathode was used for electron generation. 35
A heating current of 1.8 A was applied to a heater element
located behind the cathode. For EID, performed inside the ICR
cell, the cathode bias voltage was pulsed to 25-50 eV for
50-500 ms. IRMPD was performed inside the ICR cell with a
25 W, 10.6 µm, CO2 laser (Synrad, Mukilteo, WA). The laser
beam was deflected by two mirrors for alignment through the
hollow dispenser cathode to the center of the ICR cell. The
beam entered the vacuum system through a BaF2 window.
Photon irradiation was performed for 300-600 ms at 8.75-10
W laser power. All mass spectra were acquired with XMASS
software (version 6.1, Bruker Daltonics) in broadband mode
from m/z 21 to 1000 with 256 K data points and summed over
10-30 scans. Data processing was performed with the MIDAS
analysis software. 36 Calculated masses of precursor ions, [M
+ Met - H] + ([M + 2Met - H] + for Li), and one of the most
abundant product ions were used for internal calibration.
RESULTS
Charge-Remote Fragmentation in EID of Mn(II)-Adducted
Fatty Acids. Figure 1 shows EID spectra of Mn(II)-adducted
arachidonic acid, linolenic acid, oleic acid, and stearic acid, where
(32) Belov, M. E.; Nikolaev, E. N.; Anderson, G. A.; Udseth, H. R.; Conrads,
T. P.; Veenstra, T. D.; Masselon, C. D.; Gorshkov, M. V.; Smith, R. D. Anal.
Chem. 2001, 73, 253–261.
(33) Hendrickson, C. L.; Quinn, J. P.; Emmett, M. R.; Marshall, A. G. 49th ASMS
Conference on Mass Spectrometry and Allied Topics, Chicago, IL, 2001; CD-
ROM.
(34) de Koning, L. J.; Nibbering, N. M. M.; van Orden, S. L.; Laukien, F. H. Int.
J. Mass Spectrom. 1997, 165, 209–219.
(35) Tsybin, Y. O.; Witt, M.; Baykut, G.; Kjeldsen, F.; Hakansson, P. Rapid
Commun. Mass Spectrom. 2003, 17, 1759–1768.
(36) Senko, M. W.; Canterbury, J. D.; Guan, S.; Marshall, A. G. Rapid Commun.
Mass Spectrom. 1996, 10, 1839–1844.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6941

Figure 1. EID of Mn(II)-adducted fatty acids: (a) arachidonic acid (d ) 4), (b) linolenic acid (d ) 3), (c) oleic acid (d ) 1), and (d) stearic acid (d )
0), where d indicates the number of double bonds in each fatty acid. Only charge-remote product ion peaks, [CxHyO2 + Mn] + , are labeled. The
precursor ion peaks, [M + Mn - H] + , are outside the displayed m/z range. The Y-axis is zoomed 50 or 200 times as indicated by “×50” or “×200”.
6942 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010

Figure 2. Normalized product ion abundances of [CxHyO2 + Met] + vs Cn (n denotes carbon position from the carboxylate end of a fatty acid)
in EID of Mn(II)-adducted fatty acids: (a) arachidonic acid (d ) 4), (b) linolenic acid (d ) 3), (c) oleic acid (d ) 1), and (d) stearic acid (d ) 0),
where d indicates the number of double bonds in each fatty acid. Three observations were made for each Mn(II)-adducted fatty acid.
each fatty acid has 4, 3, 1, and 0 double bonds, respectively. In
this figure, only characteristic charge-remote fragments, [CxHyO2
+ Mn] + , are labeled whereas unlabeled peaks are mostly due
to charge-driven fragmentation, including [CxHy + Mn] + .
Charge-remote product ion abundances at each carbon-carbon
cleavage site were calculated by adding all of the [CxHyO2 +
Met] + -type ion abundances at each carbon position and
normalizing to the total [CxHyO2 + Met] + -type ion abundances
in the EID spectrum. In EID of Mn(II)-adducted fatty acids,
even and odd electron species were observed, generated
by heterolytic (resulting in product ions of, e.g., the types
[CnH2n-1O2 + Mn] + , [CnH2n-3O2 + Mn] + , and [CnH2n-5O2 +
Mn] + ) and homolytic bond cleavages (producing, e.g., the
product ion types [CnH2n-2O2 + Mn] + ,[CnH2n-4O2 + Mn] + , and
[CnH2n-6O2 + Mn] + ), respectively. Similar behavior was noted
in high energy CAD of ESI-generated precursor ions, while
mostly heterolytic bond cleavages were observed in FAB-high
energy CAD. 37
For clarity, Supplementary Table 1 (Supporting Information)
shows peak assignments for charge-remote product ions from EID
of Mn(II)-adducted ararchidonic acid as well as an example (C5
position) of how normalized charge-remote product ion abundances
were calculated. Double bond locations can be identified
from normalized product ion abundances at each carbon site,
as shown in Figure 2. EID experiments for each fatty acid were
(37) Cheng, C.; Pittenauer, E.; Gross, M. L. J. Am. Soc. Mass Spectrom. 1998,
9, 840–844.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6943

Figure 3. Normalized product ion abundances of [CxHyO2 + Met] + vs Cn (n denotes carbon position from the carboxylate end of a fatty acid)
in EID of Ni(II)- and Mg(II)-adducted arachidonic acid. The structure of arachidonic acid is shown with double bond locations.
repeated three times on three different days to verify the reliability
of EID as a method for double bond localization in fatty acids. As
shown in Figure 2, EID spectra of Mn(II)-adducted fatty acids
provided highly reproducible structural information regarding
double bond positions for each fatty acid. Lower product ion
abundances at the C4 position, observed in EID of Mn(II)adducted
linolenic acid, oleic acid, and stearic acid, indicated
that those fatty acids do not have double bonds between C5
and C6. In contrast, the lower product ion abundance at C5
compared to C4 was indicative of the existence of a double bond
between C5 and C6 in arachidonic acid. Other double bond
locations in each fatty acid could be obtained from valley
positions in the graph of normalized charge-remote product
ion abundances vs carbon position (Cn), as shown in Figure 2.
Gas-phase ion fragmentation reactions are charge remote when
there is no important interaction between the charge and the
cleavage sites. However, hydrocarbon chains are flexible, and such
interactions can, therefore, occur even in the presence of a
terminal fixed charge, particularly at less remote sites. 15 Gross
and co-workers have reported that some product ions formed by
cleavage near the charge are stabilized by a cyclic conformation. 15
6944 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
For example, C4-C5 bond cleavage in deprotonated palmitic acid
was enhanced due to formation of a cyclic structure. 18 Similarly,
enhanced cleavage at the C3 position occurred in EID of Mn(II)adducted
linolenic, oleic, and stearic acid (Figure 2b-d),
possibly due to ring formation.
Mn(II) is expected to bind tightly with the carboxylate anion
end of fatty acids because both Mn(II) and the carboxylate anion
have hard Lewis acid and base properties, respectively. Binding
energies of divalent metal ions and H2O/OH - increase sharply
at the transition between d 0 (Ca(II)) and d n (Sc(II)) due to
electron occupation in d orbitals. 38 These chemical properties
may explain why Mn(II) appears to be more efficient than other
divalent metals examined (see below) for “fixing” a charge at
the carboxylate end of fatty acids.
Charge-Remote Fragmentation in EID of Arachidonic Acid
Adducted with Metals Other than Mn(II). In addition to Mn(II),
which yielded successful charge remote fragmentation in EID (see
above), Li(I) and other divalent metals (Zn(II), Co(II), Ni(II),
Mg(II), Ca(II), and Fe(II)) were examined as adducts in EID of
(38) Magnusson, E.; Moriarty, N. W. Inorg. Chem. 1996, 35, 5711–5719.

Figure 4. EID (a) and IRMPD (b) of Mn(II)-adducted arachidonic acid. ν2 and ν3 denote the second and third harmonic peaks. [CxHyO2 + Mn] +
product ion peaks are labeled with “b”, and [CxHy] + -type peaks are labeled with “4”. Figure 4a shows the same spectrum as in Figure 1a, but
here, all peaks are labeled for comparison with the IRMPD spectrum. Y-axis is zoomed 200 or 5 times, indicated by “×200” or “×5”.
arachidonic acid. Lithium was our first metal of choice because
doubly Li(I)-adducted fatty acids, [M + 2Li - H] + , have been
shown to yield charge-remote fragmentation in high energy
CAD. 9,10,16,18 Electron irradiation (∼26 eV electrons) of the [M
+ 2Li - H] + form of arachidonic acid did not yield any product
ions (Supplementary Figure 1a, Supporting Information). One
explanation for this discrepancy may be the different means of
electronic excitation in EID vs high energy CAD: the latter
involves interactions between ions and neutrals whereas the
former involves ion-electron interactions. Thus, physical parameters
such as polarizibility (which is low for lithium) should be
more important in EID compared to high energy CAD.
EID (∼20 eV electrons) of Zn(II)- and Co(II)-adducted arachidonic
acid yielded very limited charge-remote fragmentation
(Supplementary Figure 1b,c, Supporting Information). In contrast,
EID of Ni(II)-, Mg(II)-, Ca(II)-, and Fe(II)-adducted arachidonic
acid provided extensive charge-remote product ions of the type
[C xHyO2 + Met] + . (As an example, EID of Ni(II)-adducted
arachidonic acid is shown in Supplementary Figure 1d, Supporting
Information.)
Figure 3 shows charge-remote product ion abundances as a
function of alkyl chain carbon position from EID spectra of Ni(II)and
Mg(II)-adducted arachidonic acid. Similar to the Mn(II) data in
Figure 2, [CxHyO2 + Met] + -type product ions were used to
generate this plot. EID of Ca(II)- and Fe(II)-adducted arachidonic
acid provided very similar results; thus, the corresponding data
are not shown. EID of these metal-adducted arachidonic acid did
not provide characteristic ion abundance patterns at all double
bond positions. The double bond (C14-C15) far from the carboxylate
end could not be identified from EID of Ni(II)-adducted
arachidonic acid (Figure 3a). Further, the double bond (C5-C6)
close to the carboxylate end could not be determined from EID
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6945

of Ni(II)- nor Mg(II)-adducted arachidonic acid (Figure 3a,b). We
hypothesize that Ni(II) or Mg(II) adduction may yield a “less fixed”
charge compared to Mn(II) adduction, thereby failing to provide all
double bond locations in EID.
EID of Mn(II)-adducted arachidonic acid was compared to
IRMPD of the same species. In contrast to EID (believed to occur
via both electronic and vibrational activation), dissociation in
IRMPD occurs solely via vibrational excitation. The mixture of
mostly [C xHy + Mn] + and [CxHyO2 + Mn] + species observed
in EID (Figure 4a) implies competition between charge-driven
and charge-remote processes. However, IRMPD of the same
precursor ions mainly yielded product ions from charge-driven
fragmentation, [CxHy + Mn] + , as expected from vibrational
activation (Figure 4b). In contrast, high energy CAD (known to
involve electronic excitation) provides mainly charge-remote
product ions. 9 The internal energy required for charge-remote
fragmentation is estimated to be ∼1.4-2.9 eV for protonated fatty
acids. 39 In 70 eV EI, molecular ions of small alkenes were found
to be isomerized to a mixture of interconverting structures. 40 We
propose that the 25-50 eV electron energies used in EID are
sufficient to promote charge-remote fragmentation but not high
enough to cause isomerization of Mn(II)-adducted fatty acids,
which would result in double bond migration.
CONCLUSION
Mn(II)-adducted fatty acids were analyzed to investigate the
utility of EID for determining double bond locations. Charge-
(39) Deterding, L.; Gross, M. L. Org. Mass. Spectrom. 1988, 23, 169–177.
(40) Borchers, F.; Levsen, K.; Schwartz, H.; Wesdemiotis, C.; Winkler, H. C.
J. Am. Chem. Soc. 1977, 99, 6359–6365.
6946 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
remote product ion abundances of [CxHyO2 + Mn] + -type fragments
generated by EID are significantly reduced at double
bond positions. Analysis of [CxHyO2 + Mn] + -type product ion
abundances from EID of Mn(II)-adducted fatty acids allowed
determination of all double bond positions. However, other
metal adducts did not generally provide characteristic product
ion abundances at all double bond locations. The resulting
structural information on double bond locations for Mn(II)adducted
fatty acids may be explained by dominant electronic
excitation processes in EID and efficient generation of a fixed
charge at the carboxylate end due to strong interaction between
Mn(II) cation and carboxylate anion. EID of Mn(II)-adducted
arachidonic acid was compared with IRMPD of the same
species. As expected, mostly charge-driven fragmentation was
observed in IRMPD whereas both charge-remote and chargedriven
product ions were observed in EID. In contrast, high
energy CAD is known to occur mainly via electronic excitation
and results in dominant charge-remote product ions.
ACKNOWLEDGMENT
This work was supported by the University of Michigan and
an NSF CAREER award to K.H. (CHE-05-47699).
SUPPORTING INFORMATION AVAILABLE
Peak assignments and calculations for Figures 1a and 2a. EID
spectra of Li-, Zn-, Co-, and Ni-adducted arachidonic acid. This material
is available free of charge via the Internet at http://pubs.acs.org.
Received for review May 8, 2010. Accepted July 14, 2010.
AC101217X

Anal. Chem. 2010, 82, 6947–6957
Identification of Metallothionein Subisoforms in
HPLC Using Accurate Mass and Online Sequencing
by Electrospray Hybrid Linear Ion Trap-Orbital Ion
Trap Mass Spectrometry
Sandra Mounicou,* ,† Laurent Ouerdane, † BéatriceL’Azou, ‡ Isabelle Passagne, ‡
CélineOhayon-Courtès, ‡ Joanna Szpunar, † and Ryszard Lobinski †
CNRS/UPPA, Laboratoire de Chimie Analytique Bio-Inorganique et Environnement, UMR 5254, 2, av. Pr. Angot,
64053 Pau, France, and EA 3672 Santé-Travail-Environnement, Université Victor Segalen, 146, rue Léo Saignat,
33076 Bordeaux
A comprehensive approach to the characterization of
metallothionein (MT) isoforms based on microbore HPLC
with multimodal detection was developed. MTs were
separated as Cd7 complexes, detected by ICP MS and
tentatively identified by molecular mass measured with
1-2 ppm accuracy using Orbital ion trap mass spectrometry.
The identification was validated by accurate
mass of the corresponding apo-MTs after postcolumn
acidification and by their sequences acquired online
by higher-energy collision dissociation MS/MS. The
detection limits down to 10 fmol and 45 fmol could
be obtained by ESI MS for apo- and Cd7-isoforms,
respectively, and were lower than those obtained by
ICP MS (100 fmol). The individual MT isoforms could
be sequenced at levels as low as 200 fmol with the
sequence coverage exceeding 90%. The approach was
successfully applied to the identification of MT isoforms
induced in a pig kidney cell line (LLC-PK1)
exposed to CdS nanoparticles.
Mammalian metallothioneins (MT) are low molecular weight
proteins (60-62 amino acid with molecular mass of ca. 6000-7000
Da) characterized by high cysteine content (up to 30% residues)
enabling to bind a wide range of transition and heavy metal
ions. 1-3 They are involved in a variety of biochemical processes
essential for life, such as cellular growth, stress response, copper
and zinc homeostasis, and detoxification of heavy metals. Hence,
MTs can be valuable biomarkers of stress conditions and several
pathologies which require the development of methods for the
determination and identification of the individual MT isoforms and
products of their post-translational modifications. 4-7 The primary
* To whom correspondence should be addressed.
† Laboratoire de Chimie Analytique Bio-Inorganique et Environnement.
‡ Université Victor Segalen.
(1) Hamer, D. H. Annu. Rev. Biochem. 1986, 55, 913–951.
(2) Kagi, J. H. R. Methods Enzymol. 1991, 205, 613–626.
(3) Metallothioneins; Stillman, M. J.; Shaw, C. F. I.; Suzuki, K. T., Eds.; VCh:
New York, 1992.
(4) Kagi, J. H. R.; Schaffer, A. Biochemistry 1988, 27, 8509–8515.
(5) Cousins, R. J. Physiol. Rev. 1985, 65, 238–309.
(6) Klaassen, C. D.; Liu, J.; Choudhuri, S. Ann. Rev. Pharmacol. Toxicol. 1999,
39, 267–294.
structure of metallothioneins exhibits modifications up to 15 amino
acids leading to the occurrence of several subisoforms. 1-3
Electrospray ionization (ESI) MS was first proposed by
Fenselau group for in vitro studies of the complexation of metalions
by MTs. 8,9 These seminal works have been followed by a
large number of studies using ESI MS to study the stoichiometry
of metal binding to MTs and its domains, 10-14 reactivity and
kinetics of metal exchange, 15-18 and reaction of MT with metallodrugs
19 which have been competently reviewed. 15,20,21 Recombinant
and highly purified MTs were used in these studies because
of the very low purity of the commercial MT which are mixtures
of different isoforms.
Electrospray MS, coupled with a high-resolution separation
techniques, such as reversed-phase chromatography 22 or capillary
electrophoresis, 23-25 was shown to be an attractive technique to
(7) Coyle, P.; Philcox, J. C.; Carey, L. C.; Rofe, A. M. Cell. Mol. Life Sci. 2002,
59, 627–647.
(8) Yu, X.; Wojciechowski, M.; Fenselau, C. Anal. Chem. 1993, 65, 1355–
1359.
(9) Afonso, C.; Hathout, Y.; Fenselau, C. J. Mass Spectrom. 2002, 37, 755–
759.
(10) Gehrig, P. M.; You, C.; Dallinger, R.; Gruber, C.; Brouwer, M.; Kôgi,
J. H. R.; Hunziker, P. E. Protein Sci. 2000, 9, 395–402.
(11) Merrifield, M. E.; Huang, Z.; Kille, P.; Stillman, M. J. J. Inorg. Biochem.
2002, 88, 153–172.
(12) Ngu, T. T.; Krecisz, S.; Stillman, M. J. Biochem. Biophys. Res. Commun.
2010, 396, 206–212.
(13) Orihuela, R.; Domènech, J.; Bofill, R.; You, C.; Mackay, E. A.; Kägi, J. H. R.;
Capdevila, M.; Atrian, S. J. Biol. Inorg. Chem. 2008, 13, 801–812.
(14) Palumaa, P.; Eriste, E.; Kruusel, K.; Kangur, L.; Jörnvall, H.; Sillard, R. Cell.
Mol. Biol. (Noisy-le-Grand, Fr.) 2003, 49, 763–768.
(15) Duncan, K. E. R.; Ngu, T. T.; Chan, J.; Salgado, M. T.; Merrifield, M. E.;
Stillman, M. J. Exper. Biol. Med. 2006, 231, 1488–1499.
(16) Palumaa, P.; Eriste, E.; Njunkova, O.; Pokras, L.; Jörnvall, H.; Sillard, R.
Biochemistry 2002, 41, 6158–6163.
(17) Vaher, M.; Romero-Isart, N.; Vasak, M.; Palumaa, P. J. Inorg. Biochem.
2001, 83, 1–6.
(18) Zeitoun-Ghandour, S.; Charnock, J. M.; Hodson, M. E.; Leszczyszyn, O. I.;
Blindauer, C. A.; Stürzenbaum, S. R. FEBS J. 2010, 277, 2531–2542.
(19) Karotki, A. V.; Vasak, M. J. Biol. Inorg. Chem. 2009, 14, 1129–1138.
(20) Chan, J.; Huang, Z.; Merrifield, M. E.; Salgado, M. T.; Stillman, M. J. Coord.
Chem. Rev. 2002, 233-234, 319–339.
(21) Ngu, T. T.; Stillman, M. J. Dalton Trans. 2009, 5425–5433.
(22) Chassaigne, H.; Lobinski, R. J. Chromatogr., A 1998, 829, 127–136.
(23) Mounicou, S.; Polec, K.; Chassaigne, H.; Potin-Gautier, M.; Lobinski, R. J.
Anal. At. Spectrom. 2000, 15, 635–642.
(24) Andon, B.; Barbosa, J.; Sanz-Nebot, V. Electrophoresis 2006, 27, 3661–
3670.
10.1021/ac101245h © 2010 American Chemical Society 6947
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/29/2010

separate and identify the individual isoforms but the hitherto
studies have been plagues by poor sensitivity preventing the direct
analysis of biological cytosols. Indeed, most of the reports referred
to the analysis of a1mg· mL -1 solution by capillary electrophoresis
and 100 µg · mL -1 by HPLC. Such high MT concentrations
are largely superior to those encountered in real-world
biological systems and can be achieved only after a large-scale
purification. Moreover, the ESI MS sensitivity is severely
degraded by the coelution of other biomolecules, especially in
ESI TOF MS. Consequently, the most widely applied MS
technique for the screening of biological cytosols for MTs has
been inductively coupled plasma mass spectrometry (ICP MS)
used in combination with HPLC or capillary electrophoresis. 26,27
ICP MS is a convenient technique for the detection of the
individual isoforms and the determination of the stoichiometry of
the metal-complexes formed, indeed, but does not allow their
identification.
The few successful examples of the identification of MTs in
biological cytosols by ESI MS were carried out on the basis of
the molecular mass determined using quadrupole 28,29 and TOF
mass spectrometers 30 with limited mass accuracy. The acquisition
of sequence information by conventional bottom-up proteomics
approaches is hampered by resistance of MT to tryptic digestion
and frequent miscleavages in the presence of residual metals. 31
Also, as MT isoforms exhibit a significant sequence homology
(70-90%), 32 many tryptic peptides are common for many isoforms
thus preventing de novo identification. Indeed, examples of
successful identification of MTs on the basis of MS/MS of tryptic
peptides have been scarce. 30,33
The above drawbacks and the tediousness of off-line bottomup
identification procedures can be alleviated by top-down MS
allowing one to obtain structural information from intact proteins. 34
Although most of data have been obtained with high magnetic
field strength FT ICR MS using infrared multiple photon dissociation
(IRMPD) or electron capture detection (ECD), 35 the efficiency
of direct fragmentation of intact protein in quadrupole collision
cells is increasingly explored. 36,37 Particularly interesting is the
combination of hybrid linear quadrupole ion trap with an Orbitrap
mass spectrometer which offers resolution exceeding 60 000 and
often sub-ppm mass accuracy. 38-42
(25) Benavente, F.; Andon, B.; Gimenez, E.; Olivieri, A. C.; Barbosa, J.; Sanz-
Nebot, V. Electrophoresis 2008, 29, 4355–4367.
(26) Prange, A.; Schaumloffel, D. Anal. Bioanal. Chem. 2002, 373, 441–453.
(27) Szpunar, J. Analyst 2005, 130, 442–465.
(28) Infante, H. G.; Cuyckens, F.; Van Campenhout, K.; Blust, R.; Claeys, M.;
Van Vaeck, L.; Adams, F. C. J. Anal. At. Spectrom. 2004, 19, 159–166.
(29) Polec, K.; Perez-Calvo, M.; Garcia-Arribas, O.; Szpunar, J.; Ribas-Ozonas,
B.; Lobinski, R. J. Inorg. Biochem. 2002, 88, 197–206.
(30) Wang, R.; Sens, D. A.; Albrecht, A.; Garrett, S.; Somji, S.; Sens, M. A.; Lu,
X. Anal. Chem. 2007, 79, 4433–4441.
(31) Wang, R.; Sens, D. A.; Garrett, S.; Somjii, S.; Sens, M. A.; Lu, X.
Electrophoresis 2007, 28, 2942–2952.
(32) Sewell, A. K.; Yokoya, F.; Yu, W.; Miyagawa, T.; Murayama, T.; Winge,
D. R. J. Biol. Chem. 1995, 270, 25079–25086.
(33) Feng, W.; Benz, F. W.; Cai, J.; Pierce, W. M.; Kang, Y. J. J. Biol. Chem.
2006, 281, 681–687.
(34) Kelleher, N. L.; Lin, H. Y.; Valaskovic, G. A.; Aaserud, D. J.; Fridriksson,
E. K.; McLafferty, F. W. J. Am. Chem. Soc. 1999, 121, 806–812.
(35) Tolmachev, A. V.; Robinson, E. W.; Wu, S.; Pasa-Tolic, L.; Smith, R. D. Int.
J. Mass Spectrom. 2009, 287, 32–38.
(36) Mandal, R.; Li, X. F. Rapid Commun. Mass Spectrom. 2006, 20, 48–52.
(37) Moreno-Gordaliza, E.; Canas, B.; Palacios, M. A.; Gomez-Gomez, M. M.
Anal. Chem. 2009, 81, 3507–3516.
6948 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
The goal of this study was the description on the molecular
level of the chemical response of a cell line exposed to the stress
of CdS nanoparticles, increasingly used in diverse areas from
electronics to targeted drug delivery, in view of the understanding
of the mechanisms of their toxicity. As the preliminary experiments
indicated MT induction at picomole levels, analytical
development focused on the detection of traces of metal-complexes
with individual MT isoforms and the unambiguous identification
of the latter using both the accurate mass and topdown sequencing
information, directly in the MT fraction, without extensive
purification.
EXPERIMENTAL SECTION
Reagent and Standards. Analytical reagent grade chemicals
purchased from Sigma-Aldrich (Saint-Quentin Fallavier, France)
and water (18 MΩ cm) obtained from a Milli-Q system (Millipore,
Bedford, MA) were used throughout unless stated otherwise. The
rabbit liver metallothionein-2 isoform standard (purity: 95%) was
purchased from Enzo Life Sciences (Villeurbanne, France). It was
reported by the manufacturer to be a mixture of isoforms (major:
MT-2a, minor: MT-2b and MT-2c) and to contain 67% of protein
and 9% of metals (Cd and Zn). The stock solution (1 mg · mL -1 )
was prepared by dissolving 1 mg of metallothionein in 1 mL of
water, subdivided in 10 µL aliquots to avoid multiple thawing
and freezing, and frozen at -20 °C. Working solutions were
prepared daily by dilution with water at 4 °C.
Instrumentation. Microbore reversed-phase HPLC (µRP
HPLC) separations were carried out using an Agilent 1100
capillary HPLC system (Agilent, Tokyo, Japan) equipped with a
100 µL · min -1 splitter module. ICP MS detection was achieved
using a model 7500cs instrument (Agilent) fitted with platinum
cones, 1 mm i.d injector torch and a T-connector allowing the
introduction of 5% O2. The µRP HPLC-ICP MS coupling was
done via an Isomist interface (Glass Expansion, Melbourne,
Vic, Australia) consisting of a 20 mL model Cinnabar spray
chamber cooled at 2 °C and fitted with a 50 µL · min -1
Micromist nebulizer.
For µRP HPLC-ESI MS experiments, the µRP HPLC system
was connected to a LTQ Orbitrap Velos mass spectrometer
(ThermoFisher Scientific, Bremen, Germany). The coupling was
achieved via a heated electrospray ionization source (H-ESI II)
(ThermoFisher Scientific). The postcolumn acidification manifold,
described in detail elsewhere, 43 consisted of a zero dead volume
PEEK T-piece allowing the mixing of the chromatographic effluent
with a formic acid:MeOH (30/70%, v/v) solution delivered by
means of a syringe pump (Pump 33 model, Harvard Apparatus,
South Natick, MA) and a mixing PEEK tubing coil (250 µm i.d. ×
250 mm) connected to the inlet of the electrospray ion source.
Only PEEK tubing and connectors were used to avoid metal
contamination.
(38) Bondarenko, P. V.; Second, T. P.; Zabrouskov, V.; Makarov, A. A.; Zhang,
Z. J. Am. Soc. Mass Spectrom. 2009, 20, 1415–1424.
(39) Scigelova, M.; Makarov, A. Proteomics 2006, 1, 16–21.
(40) Macek, B.; Waanders, L. F.; Olsen, J. V.; Mann, M. Mol. Cell. Proteomics
2006, 5, 949–958.
(41) Wynne, C.; Fenselau, C.; Demirev, P. A.; Edwards, N. Anal. Chem. 2009,
81, 9633–9642.
(42) Erales, J.; Gontero, B.; Whitelegge, J.; Halgand, F. Biochem. J. 2009, 419,
75–82.
(43) Chassaigne, H.; ?obin?ski, R. J. Chromatogr., A 1998, 829, 127–136.

Figure 1. Schematic flowchart showing the type of information obtained in the different modes of the multimodal approach developed.
The cell-line cytosol was separated by ultracentrifugation using
a HimaCs 120GX model (Hitachi, Tokyo, Japan) and fractionated
by size-exclusion chromatography using an Agilent 1100 HPLC
system (Agilent, Wilmington, DE). The metal elution was monitored
by splitting part of the effluent to an Agilent 7500ce ICP
MS (Agilent, Tokyo, Japan).
Procedures. The purpose of the different detection modes
used in this study was schematically illustrated in Figure 1.
µRP HPLC. The column was a C8 Vydac (250 mm ×1 mm i.d.,
5 µm) presented in passivated 316 copper-free stainless steel
housing (Alltech/Grace, Templemars, France). The injection
volume was 5 µL. The elution solvents were eluent A: 5 mM
ammonium acetate pH 6 and eluent B: 5 mM ammonium acetate
(pH 6) in 50% (v/v) acetonitrile. Elution was carried out at 40
µL · min -1 using the following program: 0-50 min: 2-20% B;
50-52 min: 2% B; 52-60 min: 2% B.
µRP HPLC-ICP MS. ICP MS was optimized daily for the
maximum sensitivity by introducing directly a solution containing
1 µg · L -1 89 Y, 7 Li, 205 Tl and 140 Ce. Isotopes monitored in µRP
HPLC-ICP MS were 114 Cd, 112 Cd 63 Cu, 65 Cu, 64 Zn, 66 Zn. The
data were processed using Excel Microsoft software.
µRP HPLC-ESI MS. Chromatographic separation conditions
were identical as given above. For the postcolumn acidification
experiments, the solution added was formic acid:methanol (30/
70%, v/v) delivered at 4 µL · min -1 .
Initial calibration of the mass spectrometer was performed
using a mixture consisting of cafein (195.08765), MRFA peptide
(524.26499) and Ultramark polymer (m/z 1221.99063). The ion
source was operated in the positive ion mode at 3.2 kV. The
vaporizer temperature of the source was set to 120 °C and the
capillary temperature to 280 °C. Nitrogen sheath gas was set to
20, the auxiliary gas to 5 and sweep gas to 0 (arbitrary unit). The
ion lenses were automatically optimized using a MT solution (1
µg · mL -1 in 0.01% formic acid in 50% (v/v) methanol) introduced
by infusion at 4 µL · min -1 and monitored at m/z 1225.843 (z )
5). In all experiments, the most abundant mass of [M + 5H] 5+
ion was systematically monitored for better detection limit.
Throughout this manuscript, the most abundant mass is always
given for multicharged ions or uncharged molecules.
In full scan mode, an m/z 1200-1500 range was scanned for
the detection of apo- and metalated MTs (the resolution was set
at 100 000 (m/∆m, fwhm at m/z 400)). Injection time was 400
ms. In single ion monitoring (SIM) mode, the scan mode was
centered on m/z 1230 ± 50 and m/z 1377.5 ± 65 for the detection
of apo- and metalated rabbit liver MTs, respectively. Pig kidney
apo-MTs were measured at m/z 1250 ± 60 and then at m/z 1170
± 70 whereas the metalated forms at m/z 1367.5 ± 87. The
resolution was set at 100 000 (m/∆m, fwhm at m/z 400) and
injection time was 3 s. The mass accuracy was expressed in ppm
and calculated as the ratio (M theoretical - Mdetermined)/Mtheoretical
multiplied by 10 6 .
µRP HPLC-ESI MS/MS. The [M+5H] 5+ of apo-MT was
selected for top-down sequencing experiments. High energy
collision fragmentation (HCD) was carried out at 50% energy,
using an isolation width of 2 in full scan mode over m/z
100-2000 mass range at a resolution of 100 000. The acquisition
time (60 min) was split into 6 time segments. Segment 1:
0-30.75 min, SIM scan mode (m/z 1100-1300) at 100 000
resolution (m/∆m, fwhm at m/z 400); segment 2: 30.75-36
min, HCD at m/z 1189.4: segment 3: 36-39 min, HCD at m/z
1197.63, and m/z 1194.63; segment 4: 39-44 min, HCD at m/z
1203.24; segment 5: 44-47 min, HCD at m/z 1271.96; segment
6: 47-60 min, SIM scan mode centered at m/z 1200 ± 100 at
100 000 resolution. Data were processed using Xcalibur 2.1
software (Thermo Fisher Scientific) and masses were manually
corrected using the traces of peaks of the Ultramark polymer
calibrant as internal standard. Analyst Q.S. 1.1 software (Applied
Biosystems MDS Sciex, Foster City, CA) was used to obtain
the theoretical y ions lists of the amino acids sequence of MT
subisoforms.
Cell Growth and CdS Nanoparticles Exposure. Kidney pig cell
line (LLC-PK1) were grown in 100 mm cell culture Petri dish
with EMEM (Eagle’s minimal essential medium) media containing
1% antibiotics (penicillin, streptomycin), 2 mM Lglutamin,
1 mM 4-(2-hydroxyethyl)-1-piperazine-ethanesulfonic
acid (HEPES), non-essential amino acids, and 5% of fetal calf
serum. Petri dishes were incubated at 37 °C ina5%CO2
incubator. At subconfluence, cells were exposed during 24 h
with5mLof120µM solution of 10 nm-CdS nanoparticles. Cells
from 11 Petri dishes were pooled. Cells not submitted to CdS
were considered as control.
Recovery of the MT Fraction. After cell growth, the supernatant
was discarded and cells were rinsed twice with 1 mL phosphate
buffered saline (PBS). A 1-mL aliquot of 25 mM Tris/HCl buffer
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6949

(pH 7.2) containing 1 mM dithiothreitol (DTT) and 0.1 mM
phenylmethyl-sulfonylfluoride (PMSF) was added. Cells were
scrapping off and recovered into an eppendorf tube. Another 500
µL of buffer was added to recover the maximum number of cells.
Cellular lysis was carried out by a successive freeze-thaw
procedure by immersing the tube into liquid nitrogen for 3 min
followed by 3 min in a water bath at 37 °C (all steps repeated
three times). An aliquot of 50 µL was kept aside to quantify
proteins amount by the Bradford protein assay. A 1.45 µL PMSF
aliquot was added to the remaining 1.45 mL sample to reach a
final concentration of 0.2 mM. Then, the sample was ultracentrifuged
(4 °C, 120 000 g, 20 min). The recovered supernatant was
heated at 95 °C for 5 min and ultracentrifuged (4 °C, 120 000 g,
20 min). A 100 µL aliquot of the supernatant was fractionated using
a Superdex peptide HR 10/30 (GE Healthcare, Uppsala, Sweden)
size exclusion column (10 × 300 mm × 5 µm) and isocratic elution
at 0.7 mL · min -1 50 mM TRIS HCl pH 7.4. The SEC column
was cleaned beforehand by flushing with mobile phase containing5mM�-mercaptoethanol
and 2 mM EDTA in order to
remove adsorbed metal ions. The main Cd-containing fraction
(8.75 - 10.73 mL) was collected from nine injections, freezedried,
resolubilized in 900 µL of water and desalted using a 5
mL HiTrap column (GE Healthcare, Uppsala, Sweden) by
elution with NH3aq (pH 8.0) at 1.5 mL · min -1 . The Cd-containing
fraction was heartcut, freeze-dried, and resolubilized in 25 µL
prior to µHPLC analyses. No special precautions, for example,
by Ar or N2 saturation during the sample preparation, were
taken to reduce oxidation of MTs.
RESULTS AND DISCUSSION
ESI MS Analysis of Rabbit Liver MT-2 Standard. As poor
detection limits are the principal drawback of the existing methods
for MT detection, effort has been focus on the optimization of
the signal-to-noise ratio. The choice of the scan mode between
full scan and single ion monitoring (SIM) was first investigated.
In the infusion conditions proper for the ionization of the Cd7-
MT form (5 mM AcNH4, pH6in50%(v/v) MeOH) the S/N
ratio of the most intense peak at m/z 1380.691 (δ 0.3 ppm,
[M+H + ] 5+ ) was 10 times higher in the SIM mode (centered
at m/z 1367.5 ± 85) than in the full scan (m/z 1325-1410)
mode. Similar results were observed for the corresponding apo-
MT signal (m/z 1225.848 (δ 0.3 ppm) in 0.01% FA in 50% (v/v)
MeOH, pH 1.6) for which the S/N ratio was 7 times higher in
the SIM mode (centered at m/z 1230 ± 13) than in the full
scan mode (m/z 1215-1250). Note that in full scan the injection
time was decreased from 3000 to 400 ms in order to limit the
number of concomitant ions entering the ion trap but the MT
peak S/N ratio was not affected.
In order to evaluate the detection limits in the chromatographic
mode taking into account the abundance of the individual
isoforms, the purity of the MT standard available was investigated
by µHPLC-ICP MS. The chromatogram (Figure 2a) obtained in
the conditions of full stoichiometric metalation (seven Cd atoms)
shows two major peaks ( 114 Cd). The contribution of Zn and Cu
is insignificant (

difference,
ppm compound
Table 1. Identification of Rabbit Liver MT2 Subisoforms
apo MTs complexed MTs
form time, m/z, z ) 5, most abundant most abundant difference,
amino acid
m/z, z ) 5, most abundant most abundant
N° min measured mass, measured mass, theoretical ppm composition compound compound measured mass, measured mass, theoretical
1 34.3 1230.049 6145.209 6145.212 0.5 Ac-M1D3P2N1C20S9A6T4G4K8E1Q1I1 N Ac-MT-2b 1384.894 6919.435 6919.427 -1.2 Cd7 N Ac-MT2b
1375.300 6871.464 6871.450 -2.0 Cd6Zn N Ac-MT2b
1375.100 6870.464 6870.452 -1.7 Cd6Cu N Ac-MT2b
1365.506 6822.494 6822.476 -2.6 Cd5ZnCu N Ac-MT2b
1365.906 6824.494 6824.474 -2.9 Cd5Zn2 N Ac-MT2b
1365.306 6821.494 6821.478 -2.3 Cd5Cu2 N Ac-MT2b
2 36.45 1217.4455 6082.1911 6082.1909 -0.04 M1D3P3N2C20S8A9T3G4K7Q1I1 MT-2a 1372.290 6856.414 6856.406 -1.1 Cd 7 -MT2a
1362.697 6808.449 6808.429 -2.9 Cd6Zn MT2a
3 36.7 1223.446 6112.194 6112.201 1.2 M1D3P3N2C20S8A8T4G4K7Q1I1 MT-2c / / / / /
4 38.5 1243.862 6214.274 6214.281 1.2 Ac-M1D3P2N1C20S9A7T3G3K8E1Q1I1R1 N Ac-MT-2d 1398.707 6988.499 6988.486 -1.8 Cd7 N Ac-MT2d
1389.113 6940.529 6940.520 -1.2 Cd6Zn N Ac-MT2d
5 38.9 1231.849 6154.209 6154.212 0.5 Ac-M1D3P3N2C20S8A8T4G4K7Q1I1 N Ac-MT-2c 1386.694 6928.434 6928.427 -1.0 Cd7 N Ac-MT2c
1376.900 6879.464 6879.452 -1.7 Cd6Cu N Ac-MT2c
1377.101 6880.469 6880.451 -2.6 Cd6Zn N Ac-MT2c
6 39.3 1225.848 6124.204 6124.201 -0.4 Ac-M1D3P3N2C20S8A9T3G4K7Q1I1 N Ac-MT-2a 1380.693 6898.429 6898.416 -1.8 Cd 7 N Ac-MT2a
1371.299 6851.459 6851.440 -2.7 Cd6Zn N Ac-MT2a
1361.305 6801.489 6801.466 -3.3 Cd5ZnCu N Ac-MT2a
1361.705 6803.489 6803.464 -3.6 Cd5Zn2 N Ac-MT2a
1361.104 6800.484 6800.467 -2.4 Cd5Cu2 N Ac-MT2a
7 42.3 1249.072 6240.324 6240.333 1.5 Ac-M 1D 3P 2N 1C 20S 8A 7T 3G 3K 8Q 1I 1E 1R 1L 1 N Ac-MT-2e 1403.917 7014.549 7014.548 -0.1 Cd 7 N Ac-MT2e
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6951

Figure 3. Comparison of CID and HCD fragmentation modes for the acquisition of sequence information for rabbit liver MT-2a isoform (6.7
µg · mL -1 in 0.01% formic acid, MeOH 50%) directly introduced at 4 µL · min -1 . CID: 40% energy; HCD: 50% energy. Mass range scanned: m/z
335-2000; activation time: 30 ms; injection time: 4000 ms; resolution: 100 000.
smaller than that of 1 order of magnitude observed elsewhere 44
possibly because of the use of a more efficient heated ionization
source.
Mass spectra of the individual MT isoforms [M+H + ] 5+
contributing to the TICs are given in Figure 1-ESI (Supporting
Information). The data are summarized in Table 1. Five MT-2
subisoforms (a-e) could be tentatively identified as N-acetylated
species. Non-acetylated subisoforms MT-2a and MT-2c could also
be detected as minor species. The identification was based on
the accurate mass determination which could be achieved for the
apo-forms with sub-ppm accuracy. The exception was two isoforms
(4 and 7) which were not detected in the ICP MS chromatogram
and for which the measured mass accuracy was 1.5 and 1.2 ppm,
respectively. The achieved mass accuracy allowed the online
determination of the amino acid composition using only few pmole
of MT. The subtraction of the molecular masses of the apo forms
from the corresponding metalated isoforms, the latter determined
with the mass accuracies between 0.1 and 3.7 ppm, allowed the
identification of 20 metal complexes with different rabbit liver
isoforms (Table 1). The overall data confirms the low purity of
the commercial rabbit liver MT-2 preventing its use for most
metalation studies. Note that all the MT-complexes detected could
be efficiently converted to the apo-form by postcolumn acidification
which demonstrates the efficiency of the manifold employed.
(44) Polec, K.; Mounicou, S.; Chassaigne, H.; Lobinski, R. Cell. Mol. Biol. 2000,
46, 221–235.
6952 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
ESI MS Detection Limits. In order to assess the detection
limit of the method an amount of 4.5 ng (ca. 750 fmol, 9-times
less than in Figure 1) of MT-2 standard was analyzed. The five
predominant MT subisoforms could be detected with S/N ratios
comprised between 3 and 66 depending on the MT-isoform
abundance. No significant degradation of the mass accuracy was
observed as metalated isoforms could be measured with mass
accuracies below 3 ppm. Taking the N-Ac-MT2b as example (the
only pure isoform in Figure 1a accounting for ca. 300 fmol of MT)
which was detected with a S/N ratio of 100, the detection limit
(MT concentration producing the signal superior to 3 times
standard deviation of the number of counts on a given mass in
the blank chromatogram) of the apo-isoform could be estimated
as 9 fmol and that of the metalated isoforms for 45 fmol. These
detection limits are by far the lowest ever reported for MT analysis
(500-fold lower than with a former-generation triple quad 44 or with
a recent TOF MS. 28 The HPLC- ICP MS detection limit could be
estimated, for the N-Ac-MT2b subisoform, as 100 fmol of protein
(S/N ) 120 for 4.5 pmol) and was higher than that of ESI MS.
Note that the ESI MS data discussed above were obtained in
the SIM scan mode. When repeated in the full scan mode, the
S/N ratio was found to decrease 2-7 times depending on the
isoform. Mass accuracies obtained for both scan modes were
essentially similar.
Online Sequencing of MT Isoforms. Although the accurate
mass is a valuable parameter for the identification of the MT

isoforms, its successful use, in the absence of other information,
is limited to well-described systems, such as, for example, rabbit
liver. For less characterized systems, especially if genetic information
is not available, the accurate molecular mass information
needs to be completed by sequence information. Consequently,
it was further attempted to investigate the possibility of top-down
sequencing of MT subisoforms eluted in conditions of Figure 2.
Preliminary experiments were carried out in the infusion mode
and consisted in the study of collision induced dissociation (CID)
and higher energy collision dissociation (HCD) modes for the
optimization of the fragmentation of MT-isoforms. The m/z
1225.84 ion corresponding to the +5 charged state of apo-MT2a
was selected as an example.
Using CID, 40% of maximum available energy was sufficient
to fragment the parent ion; an increase in energy did not change
the pattern of the product ion mass spectra. When HCD was
investigated, the energy had to be set at 50% to obtain significant
fragmentation which remained virtually unchanged with a further
increase in energy. In both modes product ions with charge states
from +1 to+5 were generated. Figure 3 shows that a larger
proportion of smaller m/z fragments was obtained in the HCD
mode than in CID mode. More y ions with +3 and +5 charge
states were observed in CID than in HCD; the opposite was true
for y ions with +2 and +1 charge states. y ions with +3 charge
states generated by CID covered the largest part of the sequence
(42%), followed by +4 charge states (34%), +2 charge states (27%),
+1 charge states (21%) and +5 charge states (12.9%). In HCD
fragmentation, the proportion of +3 and +2 charged ions was
inversed. Indeed, +2 charge states covered the largest part of
the sequence (40%), followed by +4 ions (35.5%), +3 ions (30.6%),
+1 ions (27.4%) and +5 charged ions (3.2%). Fragments obtained
using CID and HCD covered 90.3% and 93.5% of the protein
sequence, respectively, with mass accuracies in the 1-2 ppm
range regardless of the charge state. Some y ions (three in each
fragmentation mode) suffered from >3 ppm mass accuracy due
to peak overlap, but in such cases a non-overlapped differently
charged ion could always be found. High energy collision
dissociation (HCD), providing slightly better protein sequence
coverage, was chosen for further studies; in case of a > 90%
sequence coverage was insufficient, it could be complemented
by the CID mode. Using this procedure, the identity assignment
of the rabbit liver MTs in Table 1 could be confirmed. Taking
into account the MT2 concentration (ca. 6.7 µg · mL -1 ) used for
the acquisition of the HCD mass spectrum and the fragments
of lowest abundance, the minimum concentration required to
obtain a sequence coverage superior to 90% could be assessed
to as 2-3 µg · mL -1 .
µHPLC-ICP MS and ESI MS Detection of MT Subisoforms
in Pig Kidney Cell Line Exposed to CdS Nanoparticles. The
developed approach was applied to the characterization of MTs
expressed in a pig kidney cell line exposed to CdS nanoparticles.
A reversed-phase chromatogram with ICP MS 114 Cd detection
of the MT fraction (ca. 10 pM of MT) showed three fairly
intense peaks accompanied by a number of minor ones (Figure
4a). Many of these peaks had their equivalents in the chromatogram
obtained with 63 Cu detection (Figure 4b) which indicates
the existence of mixed metal complexes. The Cu/Cd ratio was
below 10% in all the peaks except peak 8. The presence of Cu-
Figure 4. µHPLC analysis of pig kidney cell line by (a) ICP MS
detection of 114 Cd, (b) ICP MS detection of 66 Zn and 63 Cu, (c) TIC of
ESI MS detection in conditions preserving metalation, (d) TIC of ESI
MS detection with postcolumn acidification. The peak numbers
correspond to data in Table 2.
binding MTs isoforms is not surprising as Cu was present in the
medium used for cell culture and exposure to CdS nanoparticles.
No peaks were observed on the 64 Zn channel. The TIC ESI MS
chromatograms of the metalated MTs (Figure 4c) and apo-MTs
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6953

Table 2. Amino Acid and Elemental Composition of Pig Kidney Metallothioneins
MTs (under demetalation conditions) complexed MTs (under metalation conditions)
mass difference
(demetalatedcomplexed)
time, m/z, z ) 5, most abundant most abundant difference,
amino acid
m/z, z ) 5, most abundant most abundant difference,
min measured mass, measured mass, theoretical ppm
composition
measured mass, measured mass, theoretical ppm
1 33.8 1189.032 5940.124 5940.128 0.7 M1D2P3N1C20S9A7T3G6K6R1Q1I1 1343.877 6714.349 6714.342 -1.0 7Cd - 14H
2 35 1189.837 5944.149 5944.159 1.7 M1D2P2N1C20S10A6T3G6K7R1V2 1344.685 6718.389 6718.374 -2.2 7Cd - 14H
3 37.4 1197.437 5982.149 5982.138 -1.8 Ac-M1D2P3N1C20S9A7T3G6K6R1Q1I1 1352.280 6756.364 6756.353 -1.6 7Cd - 14H
4 38.1 1194.840 5969.164 5969.163a -0.1 M1D2P3N1C20S9A7T3G6K5R2Q1I1 1349.482 6742.374 6742.377 0.5 7Cd - 14H
M1D3P2 N1C20S8A7T2G6K8Q1I1V1 5 38.5 1198.243 5986.179 5986.170 -1.4 Ac-M1D2P2N1C20S10A6T3G6K7R1V2 1353.086 6760.394 6760.384 -1.4 7Cd - 14H
6 40.5 1203.246 6011.194 6011.190 -0.6 Ac-M1D3P2N1C20S8A7T2G6K8Q1I1V1 1358.090 6785.414 6785.405 -1.3 7Cd - 14H
7 41 1203.036 6010.1436 6010.1440 0.1 Ac-M1D2P3N1C20S9A7T3G6K5R2Q1I1 1357.884 6784.384 6784.359 -3.6 7Cd - 14H
8 44.3 1272.167 6355.799 6355.792b -1.0 Ac-M1D3P2N1C20S8A7T2G6K8Q1I1V1 1360.478 6797.354 6797.344 -1.4 4Cd - 8H
form
N°
6954 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
a Theoretical mass for a mixture of MT1a/MT2a (1/1.65). b Mass of N-AcMT2a after thiols oxidation (formation of one disulfide bridge) + 2Cd(II) + 2Cu(I) - 6H.
(Figure 4d) showed a number of fairly intense quite well resolved
peaks in the 33-45 min range which corresponded to the peaks
in the ICP MS chromatograms.
The mass spectrometric data allowed spectra of MT-like ions
([M+5H] 5+ ) to be detected at retention times corresponding
to the ICP MS peaks as demonstrated in Figure 5. In seven
cases a mass spectrum of the apo-isoform and a corresponding
mass spectrum of the Cd7-metalated isoform could be found
resulting in a XIC with a single intense peak. In the case of
peak 8 (Figure 4b), no apo-isoform could be detected under acidic
conditions; the isotopic pattern of the peak revealed the presence
of metals. The elemental composition calculated on the basis of
the accurate molecular mass indicated the presence of two atoms
of Cd and two atoms of Cu. The mass spectrometric data are
shown in Table 2. As it can be seen the mass accuracies fitting a
tentative amino acid composition are usually

Figure 5. Extracted ion current ESI MS chromatograms of MTs subisoforms of pig kidney cell line with mass spectra in insets (upper: metalated
subisoform; lower: subisoform under acidic conditions) at retention time (a) 33.8 min, (b) 35 min, (c) 37.4 min, (d) 38.1 min, (e) 38.5 min, (f) 40.5
min, (g) 41 min, (h) 44.3 min.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6955

Figure 6. List of ions used for the determination of pig kidney MT sequences. (a) MT-1g, (b) MT-1h, (c) N-Ac-MT-1g, (d) MT-1a and MT-2a,
(e) N-Ac-MT-1h, (f) N-Ac-MT-2a, (g) N-Ac-MT-1a, (h) N-Ac-MT-2a. Bold characters highlight residues where modifications are observed.
no mass fragment specific to MT-1e in MS/MS spectra, its
presence or absence cannot be unambiguously confirmed. Even
if present, its contribution to the MTs pool would be very low. By
selecting appropriate and narrow retention time ranges in the MS/
MS chromatogram, y ions detected covered 74% and 87% of the
MT-1a and MT-2a amino acid sequence (Figure 6d) confirming
the identification of Cd7MT-1a and Cd7MT-2a.
An amino acid sequence of MT-2a without the N-terminal was
observed for the peak eluting at 40.5 min (Figure 6f). The mass
difference with regard to MT-2a isoforms indicated N-acetylation
on methionine residue leading to the identification of N-Ac-MT-
2a. This was also confirmed by two peaks of the MT2a specific
fragments (acetylated and non acetylated) in Figure 3-ESI e. The
sequence of the MT eluting at 41 min suggests strongly its
6956 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
identification as N-Ac-MT-1a (δ -0.3 ppm). Indeed, the presence
of the sequence fragment 39 Ala -Gly-Cys-Ala 42 and the overall
sequence coverage of 95% supported this identification.
The molecular masses of the remaining four apo-MTs in Table
2 did not correspond to any of the reported pig MTs subisoforms.
The 42-Da difference between the forms eluting at 33.8 and 37.4
min (peaks 1 and 3) and between the forms eluting at 35 and
38.5 min (peaks 2 and 5) indicated two pairs of acetylated and
non-acetylated isoforms. The amino acid sequence of the MT
subisoform eluting at 33.8 min (5940.124 Da, Figure 6a) was found
to be as a mixture of MT-1a and MT-1e as m/z 1091.924 and m/z
940.374 (charge +2) ions were observed for, respectively, Ala39
and Lys43 residues and matched the theoretical y ions (m/z
1091.929 and m/z 940.376 (charge +2). Furthermore, a y ion at

m/z 1172.935 was observed for Ser14 residue (theoretical m/z
1172.939, z ) 4, δ 3.1 ppm) which differentiated this sequence
from the one of MT-1f which contains a Thr14 residue. As the
sequence matched 98% of pig MT-1 sequence and has never been
reported before, it was denoted here as MT-1g. The sequence
coverage with y ions was 88.5%. Consequently, the subisoform
eluting at 37.4 min (5982.149 Da, Figure 6c) was referred to as
N-Ac-MT-1g.
Regarding the identification of the remaining pair of acetylated
and non-acetylated isoforms eluting at 35.0 min (5944.149
Da) and 38.5 min (5986.179 Da), the HCD spectrum of the
N-acetylated form (m/z 1198.243, charge +5) gives the sequence
information. It was found identical to MT1c until Pro38
and then residues Ala39, Arg43, Gln46, Ile49 of MT1c were,
respectively, replaced by Val39, Lys43, Lys46, Val49 as m/z
1098.4547 (theoretical m/z 1098.4550, δ 0.2 ppm), m/z 933.3862
(theoretical m/z 933.3869, δ 0.7 ppm), m/z 781.813 (theoretical
m/z 781.816, δ 3.2 ppm) and m/z 1274.494 (theoretical m/z
1274.499, δ 4.1 ppm) were detected at +2 charge state and the
last one at +1 charge state. An ExPASy database search
indicated that these modifications on 46 and 49 amino acids
residues were reported for MT of others organisms (pig (in
MT3), sheep, bovine and human). The sequence coverage was
92%. As this MT has not been reported before, this pair of
isoforms was denoted as MT-1h and N-Ac-MT-1h, respectively.
The composition of the complexes formed indicates a release
of Cd ions from the nanoparticles and their complexation by
MTs formed.
The identification of the last peak in the µHPLC-ICP MS
chromatogram (peak 8 at 44.3 min (cf. Figure 4b) was less
straightforward as no apo-isoform could be obtained. The presence
of the fragment sequence 7 Cys-Ala-Ala 9 among +4 charge state
fragments is characteristic for an MT-2 class isoforms. Another
part of an MT sequence could be identified among +2 charge
state fragments as 37 Cys-Pro-Val-Gly-Cys-Ala-Lys-Cys-Ala 45 (δ
3.1 ppm), confirming the identity of the MT subisoform as
MT2a. The end of the MT sequence 46 Gln-Gly-Cys-Ile-Cys-Lys-
Gly-Ala-Ser-Asp 55 is observed as +1 charge state (δ 2.3 ppm).
The metallothionein is obviously metalated and the mass
difference in comparison with the N-Ac-MT2a theoretical mass
(6011.190 Da) corresponds to the addition of Cu2Cd2 (351.665
Da) with a loss of 8H + . This state of metalation is corroborated
by the observation of three y ions corresponding to the 7 Cys-
(47) Meloni, G.; Faller, P.; Vasak, M. J. Biol. Chem. 2007, 282, 16068–16078.
(48) Salgado, M. T.; Stillman, M. J. Biochem. Biophys. Res. Commun. 2004, 318,
73–80.
Ala-Ala 9 fragment (m/z 1417.393, m/z 1391.644 and m/z
1373.883 charge +4) differing by 85.89 mass unit from the
corresponding apo subisoforms. The isotopic pattern was
similar to the theoretical isotopic pattern of fragments with
Cu2Cd2 adduct. Therefore peak 8 (Figure 4b) was identified as
N-terminal acetylated Cu2Cd2-MT2a (6355.799 Da, theoretical
6355.792 Da, δ -1.0 ppm) under acidic conditions and as
N-terminal acetylated Cu2Cd6-MT2a (6797.354 Da, theoretical
6797.344 Da, δ -1.4 ppm) under neutral conditions. The late
elution of a similar complex from a rat kidney extract was
reported elsewhere. 23 The comparison of the molecular mass
of the complex with that of the apoMT indicates the presence
of a disulfide bridge formed by oxidation of Cys by Cu(II)
leading to the loss of additional two protons. 17,47 Indeed, the
product ion mass spectra of Cd2Cu2MT2-a (cf. Figure 6), show
a non-fragmented region (from Gly10 until Cys37) of the sequence
suggesting the binding of four metal ions to Cys. This binding
site is shared between the R and � domains of the MT2-a which
is in agreement with the hypothesis of Vaher et al. 17 We can also
notice that the unfragmented central region binding metals
contains 11 Cys like the R domain of MT2a which would explain
the stabilization of Cd2Cu2-MT2a complex. 48
CONCLUSIONS
µHPLC with the combined elemental detection by ICP MS, subppm
molecular mass determination by electrospray orbital trap MS
in conditions favoring metalation and demetalation, and HCD
fragmentation allowing online protein sequencing was demonstrated
as the, to date, most comprehensive approach to the characterization
of metallothioneins in a biological cytosol. The detection limits down
to the low femtomole levels allowed the characterization of metallothionein
subisoforms induced in cell cultures exposed to CdS
nanoparticles confirming the presence of known and the identification
of new MT subisoforms.
ACKNOWLEDGMENT
The contribution of the Region of Aquitaine and the FEDER
funds via CPER A2E (31486/08011464) project is acknowledged.
We thank ICMCB, Bordeaux, for providing the 10 nm CdS
nanoparticles used in this study.
SUPPORTING INFORMATION AVAILABLE
Figure 1-ESI, Figure 2-ESI, and Figure 3-ESI. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 12, 2010. Accepted July 8, 2010.
AC101245H
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6957

Anal. Chem. 2010, 82, 6958–6968
Cleavable Cross-Linker for Protein Structure
Analysis: Reliable Identification of Cross-Linking
Products by Tandem MS
Mathias Q. Müller, † Frank Dreiocker, ‡ Christian H. Ihling, † Mathias Schäfer,* ,‡ and Andrea Sinz* ,†
Department of Pharmaceutical Chemistry and Bioanalytics, Institute of Pharmacy, Martin-Luther-Universität
Halle-Wittenberg, Wolfgang-Langenbeck-Strasse 4, D-06120 Halle (Saale), and Institute for Organic Chemistry,
Department of Chemistry, Universität zu Köln, Greinstrasse 4, D-50939 Cologne, Germany
Chemical cross-linking combined with a subsequent
enzymatic cleavage of the created cross-linked complex
and a mass spectrometric analysis of the resulting crosslinked
peptide mixture presents an alternative approach
to high-resolution analysis, such as NMR spectroscopy or
X-ray crystallography, to obtain low-resolution protein
structures and to gain insight into protein interfaces.
Here, we describe a novel urea-based cross-linker, which
allows distinguishing different cross-linking products by
collision-induced dissociation (CID) tandem MS experiments
based on characteristic product ions and constant
neutral losses. The novel cross-linker is part of our
ongoing efforts in developing collision-induced dissociative
reagents that allow an efficient analysis of cross-linked
proteins and protein complexes. Our innovative analytical
concept is exemplified for the Munc13-1 peptide and the
recombinantly expressed ligand binding domain of the
peroxisome proliferator-activated receptor r, for which
cross-linking reaction mixtures were analyzed both by
offline nano-HPLC/MALDI-TOF/TOF mass spectrometry
and by online nano-HPLC/nano-ESI-LTQ-orbitrap mass
spectrometry. The characteristic fragment ion patterns of
the novel cross-linker greatly simplify the identification
of different cross-linked species, namely, modified peptides
as well as intrapeptide and interpeptide cross-links,
from complex mixtures and drastically reduce the potential
of identifying false-positive cross-links. Our novel ureabased
CID cleavable cross-linker is expected to be highly
advantageous for analyzing protein 3D structures and
protein-protein complexes in an automated manner.
Chemical cross-linking 1 combined with mass spectrometry
presents an alternative approach to study the tertiary and
quaternary structure of proteins and protein complexes. 2-4
However, an unambiguous, sensitive, reliable, and fast identifica-
* To whom correspondence should be addressed. (A.S.) Phone: +49-345-
5525170. Fax: +49-345-5527026. E-mail: andrea.sinz@pharmazie.uni-halle.de.
(M.S.) Phone: +49-221-4703086. Fax: +49-221-4703064. E-mail: mathias.schaefer@
uni-koeln.de.
† Martin-Luther-Universität Halle-Wittenberg.
‡ Universität zu Köln.
(1) Sinz, A. Angew. Chem., Int. Ed. 2007, 46, 660–662.
(2) Sinz, A. J. Mass Spectrom. 2003, 38, 1225–1237.
(3) Trakselis, M. A.; Alley, S. C.; Ishmael, F. T. Bioconjugate Chem. 2005, 16,
741–750.
6958 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
tion of the amino acids involved in the covalent derivatization by
chemical cross-linking remains challenging. A mass spectrometric
identification of the chemically modified amino acids in the
respective protein is performed after proteolytic digestion of
the cross-linking reaction mixtures. This is often hampered by
the complexity of the created peptide mixtures, wherein only a
relatively small percentage of cross-linked products are present
besides a majority of unmodified peptides. To safeguard for a
selective identification of cross-linked peptides using electrospray
ionization (ESI) 5-7 combined with collision-induced dissociation
(CID) 8,9 tandem MS, 10 several approaches have been suggested.
11-14 They include cross-linking reagents that incorporate
the use of marker ions resulting from low-energy CID 15,16 or
metastable decay, 17 isotope-coding strategies, such as proteolytic
digestion in 18 O water, 18 isotope coding of the cross-linking
reagents 19-22 or of the proteins, 23 and an enrichment of cross-
(4) Sinz, A. Mass Spectrom. Rev. 2006, 25, 663–682.
(5) Fenn, J. B.; Mann, M.; Meng, C. K.; Wong, S. F.; Whitehouse, C. M. Science
1989, 246, 64–71.
(6) Cole, R. Electrospray Ionization Mass Spectrometry: Fundamentals, Instrumentation
and Applications; 1997.
(7) Kebarle, P.; Verkerk, U. H. Mass Spectrom. Rev. 2009, 28, 898–917.
(8) Wells, J. M.; McLuckey, S. A. The Encyclopedia of Mass Spectrometry; 2003;
p1.
(9) Shukla, A. K.; Futrell, J. H. J. Mass Spectrom. 2000, 35, 1069–1090.
(10) Hunt, D. F.; Yates, J. R., III; Shabanowitz, J.; Winston, S.; Hauer, C. R. Proc.
Natl. Acad. Sci. U.S.A. 1986, 83, 6233–6237.
(11) Soderblom, E. J.; Goshe, M. B. Anal. Chem. 2006, 78, 8059–8068.
(12) Soderblom, E. J.; Bobay, B. G.; Cavanagh, J.; Goshe, M. B. Rapid Commun.
Mass Spectrom. 2007, 21, 3395–3408.
(13) Dreiocker, F.; Müller, M. Q.; Sinz, A.; Schäfer, M. J. Mass Spectrom. 2010,
45, 178–189.
(14) Müller, M. Q.; Dreiocker, F.; Ihling, C. H.; Sinz, A.; Schäfer, M. J. Mass
Spectrom., in press.
(15) Back, J. W.; Hartog, A. F.; Dekker, H. L.; Muijsers, A. O.; de Koning, L. J.;
de Jong, L. J. Am. Soc. Mass Spectrom. 2001, 12, 222–227.
(16) Tang, X.; Munske, G. R.; Siems, W. F.; Bruce, J. E. Anal. Chem. 2005, 77,
311–318.
(17) Young, M. M.; Tang, N.; Hempel, J. C.; Oshiro, C. M.; Taylor, E. W.; Kuntz,
I. D.; Gibson, B. W.; Dollinger, G. Proc. Natl. Acad. Sci. U.S.A. 2000, 97,
5802–5806.
(18) Back, J. W.; Notenboom, V.; de Koning, L. J.; Muijsers, A. O.; Sixma, T. K.;
de Koster, C. G.; de Jong, L. Anal. Chem. 2002, 74, 4417–4422.
(19) Petrotchenko, E. V.; Olkhovik, V. K.; Borchers, C. H. Mol. Cell. Proteomics
2005, 4, 1167–1179.
(20) Schulz, D. M.; Kalkhof, S.; Schmidt, A.; Ihling, C.; Stingl, C.; Mechtler, K.;
Zschörnig, O.; Sinz, A. Proteins 2007, 69, 254–269.
(21) Ihling, C.; Schmidt, A.; Kalkhof, S.; Schulz, D. M.; Stingl, C.; Mechtler, K.;
Haack, M.; Beck-Sickinger, A. G.; Cooper, D. M.; Sinz, A. J. Am. Soc. Mass
Spectrom. 2006, 17, 1100–1113.
10.1021/ac101241t © 2010 American Chemical Society
Published on Web 07/22/2010

linked products via affinity tags. 1,24,25 Identification of cross-linked
peptide ions by CID tandem MS or MS n is not trivial as product
ion spectra of these contain a number of product ions, i.e., band
y-type ions, 26,27 originating from both peptides involved in
the cross-linking product. To enable a more effective analysis of
cross-linked peptides by CID tandem MS analysis, we have
developed a novel concept for collision-induced dissociative crosslinking
reagents. As such, the novel cross-linker contains a labile
covalent bond located within the linker region, which is selective
and preferably cleaved by collision activation in the gas phase.
Highly efficient and reproducible cleavage of our novel urea-based
cross-linker is mediated by a nucleophilic attack of a carbonyl
oxygen in the cross-linker’s spacer chain at the urea carbonyl
carbon. The synthesis and application of the predecessor of this
cross-linker, a thiourea derivative, has been studied in detail
previously. 14 However, that thiourea-based cross-linker exhibited
a number of properties which were unfavorable for analyzing
cross-linked products by CID in an automated fashion. Most
importantly, we were unable to differentiate between the different
types of cross-linked species, namely, peptides that are modified
by a partially hydrolyzed cross-linker or intramolecular type 1
cross-linked products.
The urea-based cross-linker presented herein overcomes the
deficiencies of the thiourea-based reagent. It leads to the formation
of indicative fingerprint mass shifted product ions and neutral
losses in the product ion mass spectra, which allow unambiguous
identification of the different types of cross-linking products. The
properties of the novel cross-linker were evaluated for the Munc
13-1 peptide and the 28 kDa ligand binding domain of the
peroxisome proliferator-activated receptor R (PPARR), impressively
indicating its power for efficiently analyzing cross-linked
products in a highly automated fashion.
EXPERIMENTAL SECTION
Materials. All chemicals and solvents used for synthesis of
the cross-linker were used as purchased without further purification
(Acros Organics, Geel, Belgium; ABCR, Karlsruhe, Germany).
Dichloromethane and pyridine were dried by distillation from
calcium hydride. All other solvents were distilled over a column
prior to use.
Buffer reagents and chemicals were obtained from Sigma
(Taufkirchen, Germany). Proteases (trypsin and GluC, sequencing
grade) were obtained from Roche Diagnostics (Mannheim,
Germany). Nano-HPLC solvents were spectroscopic grade (Uvasol,
VWR, Darmstadt, Germany). MALDI matrixes and calibration
standards were purchased from Bruker Daltonik (Bremen,
Germany). Water was purified with a Direct-Q5 water purification
system (Millipore, Eschborn, Germany). The ligand binding
domain of PPARR was expressed in Escherichia coli and purified
(22) Schmidt, A.; Kalkhof, S.; Ihling, C.; Cooper, D. M.; Sinz, A. Eur. J. Mass
Spectrom. 2005, 11, 525–534.
(23) Taverner, T.; Hall, N. E.; O’Hair, R. A.; Simpson, R. J. J. Biol. Chem. 2002,
277, 46487–46492.
(24) Hurst, G. B.; Lankford, T. K.; Kennel, S. J. J. Am. Soc. Mass Spectrom. 2004,
15, 832–839.
(25) Sinz, A.; Kalkhof, S.; Ihling, C. J. Am. Soc. Mass Spectrom. 2005, 16, 1921–
1931.
(26) Roepstorff, P.; Fohlman, J. Biomed. Mass Spectrom. 1984, 11, 601.
(27) Papayannopoulos, I. A.; Biemann, K. Protein Sci. 1992, 1, 278–288.
Scheme 1. Structure of the Symmetric
NHS-BuUrBu-NHS Compound (1) for Chemical
Cross-Linking
according to a previously published protocol. 28 The Munc13-1
peptide was synthesized by Dr. O. Jahn (MPI for Experimental
Medicine, Göttingen, Germany).
Synthesis of the Cross-Linker. Aminobutanoic acid was
dissolved in 4 M NaOH, and 0.175 equiv of triphosgen in dioxane
was added at 0 °C (see Scheme S1 of the Supporting Information).
The reaction mixture was brought to room temperature overnight,
and the solid was removed by filtration. The solvent was removed
under reduced pressure, and the residue was recrystallized from
6 M hydrochloric acid. The product was isolated by filtration,
washed with acetone, and dried in vacuum. HO-BuUrBu-OH was
isolated as a colorless solid in a yield of 26%. 29 An abbreviated
notation, BuUrBu, is used throughout this paper for the backbone
of cross-linker 1, which is derived from urea (Ur) and aminobutyric
acid (Bu) (Scheme 1).
HO-BuUrBu-OH was dissolved in pyridine and treated with
N-(trifluoroacetoxy)succinimidesprepared from N-hydroxysuccinimide
(NHS) and trifluoroacetic anhydridesat 0 °C. 30 The
reaction mixture was brought to room temperature within 2 h.
After addition of ethyl acetate the raw product was isolated by
filtration and dissolved in a mixture of dichloromethane and
methanol. Insoluble components were removed by filtration, and
the filtrate was evaporated. NHS-BuUrBu-NHS (1) was isolated
as a colorless solid in a yield of 82%.
Cross-Linking Reactions. For cross-linking experiments,
aqueous stock solutions of Munc13-1 peptide (100 µg/mL) or
PPARR ligand binding domain (2 mg/mL) were diluted with 20
mM HEPES, 150 mM NaCl, 1 mM TCEP, 10% (v/v) glycerol, pH
8.0, to a volume of 1 mL, giving a final protein/peptide concentration
of 10 µM. The cross-linker (200 mM stock solution in DMSO)
was added in 50, 100, and 200 M excess to the protein/peptide
solution, and the reactions were allowed to proceed for 5, 15, 30,
60, and 120 min. Reactions were quenched with ammonium
bicarbonate (20 mM final concentration). One 200 µL aliquot was
taken from each sample and stored at -20 °C before MS analysis.
Cross-linked Munc13-1 peptide was digested with trypsin, whereas
PPARR was digested either with trypsin or with trypsin/GluC
(each 1:100 (w/w) enzyme to substrate ratio) overnight at 25 °C
according to an existing protocol. 20 The resulting digests were
stored at -20 °C before MS analysis.
Linear Mode MALDI-TOF Mass Spectrometry. MALDI-
TOF mass spectrometry of intact cross-linked PPARR was
(28) Müller, M. Q.; Roth, C.; Sträter, N.; Sinz, A. Protein Expression Purif. 2008,
62, 185–189.
(29) Coe, S.; Kane, J. J.; Nguyen, T. L.; Toledo, L. M.; Wininger, E.; Fowler,
F. W.; Lauher, J. W. J. Am. Chem. Soc. 1997, 119, 86–93.
(30) Rao, T. S.; Nampalli, S.; Sekher, P.; Kumar, S. Tetrahedron Lett. 2002, 43,
7793–7795.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6959

performed in linear and positive ionization mode on an Ultraflex
III instrument (Bruker Daltonik) equipped with a Smartbeam laser
using sinapinic acid (SA) as the matrix. A 5 pmol portion of protein
was applied to a ground steel target (Bruker Daltonik) using the
double layer method as described in the manufacturer’s manual
of the Ultraflex III mass spectrometer.
Nano-HPLC/MALDI-TOF/TOF Mass Spectrometry. Peptide
mixtures derived from enzymatic digestions of Munc13-1 and
PPARR were analyzed by offline coupling of a nano-HPLC system
(Ultimate 3000, Dionex Corp., Idstein, Germany) to a MALDI-
TOF/TOF mass spectrometer (Ultraflex III, Bruker Daltonik).
Samples were injected via the autosampler to a precolumn
(Acclaim PepMap, C18, 300 µm × 5 mm, 5 µm, 100 Å, Dionex)
and desalted by washing the precolumn for 15 min with 0.1% TFA
before the peptides were eluted onto the separation column
(Acclaim PepMap, C18, 75 µm × 250 mm, 5 µm, 100 Å, Dionex),
which had been equilibrated with 95% solvent A (5% ACN, 0.05%
TFA). For PPARR, peptides were separated with a 60 min gradient
(0-60 min, 5-50% B; 60-62 min, 50-100% B; 62-67 min, 100%
B; 67-72 min, 5% B; solvent B ) 80% ACN, 0.04% TFA) at a flow
rate of 300 nL/min with UV detection at 214 and 280 nm. Eluates
were fractionated into 18 s fractions using the fraction collector
Proteineer fc (Bruker Daltonik), mixed with 1.1 µL of matrix
solution (0.8 mg/mL HCCA in 90% ACN, 0.1% TFA, 1 mM
NH4H2PO4) and directly prepared on a 384 MTP 800 µm
AnchorChip target (Bruker Daltonik). For separation of
Munc13-1 peptides, slightly different nano-HPLC conditions
were employed. Tryptic peptide mixtures were separated by
nano-HPLC (30 min, 5-50% B), and fractions were collected
on the AnchorChip 384/800 target in 30 s steps.
MALDI-TOF/TOF-MS/MS analyses were conducted in the
positive ionization and reflectron mode by accumulating 2000 laser
shots in the range m/z 700-5000 to one mass spectrum. Mass
spectra were externally calibrated using Peptide Calibration
Standard II (Bruker Daltonik). Signals with a signal-to-noise ratio
(S/N) > 15 were selected and subjected to laser-induced fragmentation
(LID). In-source decay (ISD) was performed manually
on isolated cross-linking product candidates by increasing the laser
energy by ca. 10% relative to the usual laser energy in LID-MS/
MS. Data acquisition was done automatically by the WarpLC 1.1
software (Bruker Daltonik) coordinating MS data acquisition
(FlexControl 1.3) and data processing (FlexAnalysis 3.0).
Nano-HPLC/Nano-ESI-LTQ-Orbitrap Mass Spectrometry.
Fractionation of tryptic peptide mixtures (Munc13-1 and PPARR)
was carried out on an Ultimate nano-HPLC system (Dionex) using
reversed-phase C18 columns (precolumn, Acclaim PepMap, 300
µm × 5 mm, 5 µm, 100 Å; separation column, Acclaim PepMap,
75 µm × 250 mm, 5 µm, 100 Å; Dionex). After being washed on
the precolumn for 15 min with water containing 0.1% TFA, the
peptides were eluted and separated using gradients from 0% to
50% B (90 min) and from 50% to 100% B (1 min) and 100% B (5
min), with solvent A being 5% ACN containing 0.1% FA and solvent
B being 80% ACN containing 0.1% FA. The nano-HPLC system
was directly coupled to the nano-ESI source (Proxeon, Odense,
Denmark) of an LTQ-orbitrap XL hybrid mass spectrometer
(Thermo Fisher Scientific, Bremen, Germany). 31 MS data were
(31) Hu, Q.; Noll, R. J.; Li, H.; Makarov, A.; Hardman, M.; Graham Cooks, R. J.
Mass Spectrom. 2005, 40, 430–443.
6960 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
acquired over 122 min in data-dependent MS 2 mode: Each highresolution
full scan (m/z 300-2000, R ) 60 000) in the orbitrap
was followed by three product ion scans in the orbitrap (R )
7500) on the three most intense signals in the full-scan mass
spectrum (isolation window 1.5 u). MS 3 was performed on the
three most intense signals in CID tandem mass spectra, and
fragment ions were either analyzed in the linear quadrupole
ion trap (LTQ) or in the orbitrap (R ) 7500). Dynamic
exclusion (exclusion duration 180 s, exclusion window -1 to
+1 Th) was enabled to allow detection of less abundant ions.
Data acquisition was controlled via XCalibur 2.0.7 (Thermo
Fisher) in combination with DCMS link 2.0 (Dionex).
Analysis of Cross-Linked Products. Cross-linked products
were identified by analyzing the MS data using General Protein
Mass Analysis for Windows (GPMAW), 32 version 8.10 (Lighthouse
Data, Odense, Denmark, http://www.gpmaw.com) and the Cool-
ToolBox software program, an upgrade of the VIRTUALMSLAB
software program. 33 The length of the cross-linker was calculated
using the ViewerLight 5.0 software (Accelrys). 34
RESULTS AND DISCUSSION
Cross-Linker Design. The dissociative cross-linker specifically
designed for CID tandem MS is a simple and symmetric
molecule with a central urea moiety (Scheme 1), similar to the
previously introduced thiourea derivative. 14 The practical strategy
of inducing a preferred and defined dissociation of the chemical
cross-linker upon CID was adapted from the Edman degradation
mechanism, which was exemplified for a cross-linker containing
a thiourea moiety that is connected to proline via a glycine
residue. 13 The refined analytical concept of the herein presented
novel cleavable cross-linker relies on an intramolecular nucleophilic
attack at the central urea carbonyl, resulting in an efficient
cleavage of the cross-linker and giving rise to characteristic
fragment ions and constant neutral losses (CNLs). The symmetric
compound is simple as it consists of a central urea moiety that is
flanked by two aminobutyric acids connected to activated NHS
esters, which react with lysines or free N-termini in proteins. 4
An abbreviated notation, BuUrBu, is used throughout this
paper for the novel cross-linker 1, which is derived from urea
(Ur) and aminobutyric acid (Bu) (Scheme 1). The protein
derivatization reactions of 1 are illustrated in Scheme 2S of the
Supporting Information. After one reactive site has reacted with
a primary amine in a protein, the second reactive site has at least
two options to react with proteins: (a) it might form an intermolecular
cross-link (2) with another protein (type 2 cross-link 35 )
or (b) it might react with another amine group of the same protein,
leading to an intramolecular cross-link (type 1, structure 3).
Alternatively, the second NHS functionality is hydrolyzedsin case
no amine group is within the correct distance to be crosslinkedsor
it is aminolyzed during reaction quenching with
ammonium salts (type 0 cross-link linker, 35 structures 4 and 5,
respectively).
(32) Peri, S.; Steen, H.; Pandey, A. Trends Biochem. Sci. 2001, 26, 687–689.
(33) de Koning, L. J.; Kasper, P. T.; Back, J. W.; Nessen, M. A.; Vanrobaeys, F.;
Van Beeumen, J.; Gherardi, E.; de Koster, C. G.; de Jong, L. FEBS J. 2006,
273, 281–291.
(34) ViewerLight 5.0, Accelrys, San Diego, CA.
(35) Schilling, B.; Row, R. H.; Gibson, B. W.; Guo, X.; Young, M. M. J. Am. Soc.
Mass Spectrom. 2003, 14, 834–850.

Scheme 2. Fragmentation Mechanism of Protonated 4 (Scheme S2, Supporting Information) upon CID, Delivering a
Doublet of 26 u Mass Shifted Product Ions [M + H + Bu] + (6) and [M + H + BuUr] + (7) by CNLs of 129 and 103 u
Reactivity and CID Fragmentation. The novel symmetric
urea-based cross-linker is unique in that it allows a discrimination
of different types of cross-linked products, namely, type 0 (a
peptide that is modified by a partially hydrolyzed cross-linker 35 ),
type 1 (intrapeptide cross-link), and type 2 (interpeptide crosslink),
based on characteristic fragmentation reactions. As such,
the cross-linker allows a screening for cross-linked species in a
highly automated fashion. The CID fragmentation mechanism of
peptides modified with hydrolyzed linkers (the type 0 cross-link
linker 4 35 is depicted in Scheme 2). A nucleophilic attack at the
carbonyl carbon of the urea moiety leads to a cleavage of the crosslinker
at either urea amide bond. Depending on which oxygen
attacks the urea carbonyl carbon, either the peptide modified with
4-aminobutyric acid (6) ([M + H + 85 u] + ) corresponding to a
CNL of 129 u (1,3-oxazepan-2-one) or the peptide that is
decorated with 1,3-oxazepan-2-one (7) ([M + H + 111 u] + )
corresponding to a CNL of 103 u (4-aminobutyric acid) is
observed. Intriguingly, the two product ions 6 and 7 formed
are mass shifted by 26 u, which is a unique indicator of a “deadend”
(type 0) cross-link, i.e., a peptide that is modified by a
partially hydrolyzed cross-linker (Scheme 2).
In analogy to the fragmentation behavior of the type 0 crosslink
(Scheme 2), an interpeptide (type 2) cross-link (2) results in
the cleavage of the urea amide bond, yielding a pair of complementary
product ions that originate from the symmetric structure
of the cross-linker (Scheme 3). Either peptide ions are generated
that are modified with 4-aminobutyric acid (ions 6a and 6b) or,
alternatively, peptide product ions are formed carrying a 1,3oxazepan-2-one
(7a and 7b) modification. This specific fragmentation
behavior of an intermolecular cross-linked precursor ion (2)
gives rise to two doublets of peptide ions, which exhibit the
characteristic mass difference of 26 u. This unique pattern of a
“doublet of 26 u doublets” in CID mass spectra is highly indicative
in evidencing the presence of an interpeptide cross-link (type 2),
while a dead-end (type 0) cross-link only generates a single 26 u
doublet as discussed above. Strikingly different from the characteristic
fragment ions created for type 0 and type 2 cross-links is
the fragmentation pattern of a type 1 intramolecular cross-link (3).
As shown in Scheme 4, CID of intramolecular cross-links leads
to the effective formation of pyrolidinone, detectable as a CNL
for 85 u. A summary of the characteristic fragment ions and CNLs
created for the different cross-linked products that allow for a rapid
screening is shown in Table 1. We also observed mixed species,
i.e., peptides containing more than one cross-linker molecule,
which were not considered in this study.
Nano-HPLC/Nano-ESI-LTQ-Orbitrap Mass Spectrometry.
To assess the behavior of the NHS-BuUrBu-NHS cross-linker (1)
in CID experiments, two model substances, the 22-amino acid
Munc13-1 peptide (CRAKANWLRAFNKVRMQLQEAR) and the
28 kDa ligand binding domain of PPARR, were selected. Both were
cross-linked with 1, enzymatically digested and subjected to nano-
HPLC/nano-ESI-LTQ-orbitrap-MS n . The Munc13-1 peptide is
currently under investigation in the Sinz lab 36 and was chosen
as a model peptide for evaluating the properties of 1 as it
contains three primary amines (N-terminus, Lys-4 and Lys-13)
that are separated by four arginines, allowing a tryptic cleavage.
Peptides Modified with a Partially Hydrolyzed Cross-
Linker (Dead-End Cross-Links, Type 0). In Figure 1A, the ESI-
CID-MS/MS product ion spectrum of a modified tryptic peptide
(AKANWLR) of Munc13-1 is presented. The abundant precursor
ion at m/z 1072.60 was selected and completely dissociated by
effective collision activation in the linear quadrupole ion trap. The
ion at m/z 1072.60 was assigned as a type 0 cross-linked ion [M
+ H + BuUrBu-OH] + that is modified with a partially hydrolyzed
cross-linker (4) at the lysine residue. This assignment
is consistent with the fragmentation scheme shown in Scheme
2. The expected CNLs of 129 and 103 u consequently lead to the
observation of specific fragment ions with an indicative mass
difference of 26 u, in this case found at m/z 943.55 and at m/z
969.53 (Figure 1A). Structure identification of these ions was
achieved by interpretation of the LTQ-MS 3 product ion spectra
shown in Figure 1B,C. The former spectrum shows the respective
product ion spectra of [M + H + Bu] + at m/z 943.55 (6), while
(36) Dimova, K.; Kalkhof, S.; Pottratz, I.; Ihling, C.; Rodriguez-Castaneda, F.;
Liepold, T.; Griesinger, C.; Brose, N.; Sinz, A.; Jahn, O. Biochemistry 2009,
48, 5908–5921.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6961

Scheme 3. Fragmentation Mechanism of Protonated 2 upon CID, Delivering Two Complementary Doublets of 26 u
Mass Shifted Product Ions a
a Product ions of peptide 1 are 6a and 7a, and product ions of peptide 2 are 6b and 7b.
Scheme 4. Fragmentation Mechanism of a Protonated Type 1 Modified Peptide (3) upon CID, Delivering a Product
Ion That Is Modified with BuUr [M + H + BuUr] + (7) by a CNL of Pyrolidinone (85 u)
Table 1. Summary of the Characteristic CNLs and Fragment Ion Mass Shifts Used for Discrimination of the
Different Cross-Linking Types (ESI-Ion Trap-MS/MS and MALDI-TOF/TOF-MS/MS)
cross-link type hydrolyzed, type 0 aminolyzed, type 0 intrapeptide, type 1 interpeptide, type 2
CNL 103 and 129 u 102 and 128 u 85 u
no. of 26 u doublets 1 1 2
the latter presents the respective product ion spectrum of [M
+ H + BuUr] + at m/z 969.53 (7). In both experiments the
product ion analysis was conducted in the orbitrap at a mass
accuracy of 1-2 ppm employing a resolving power of 15 000.
In the MS 3 product ion spectra presented in Figure 1B,C, the
precursor ions exhibit frequent losses of water and ammonia, and
more importantly, extensive backbone cleavages allow an efficient
sequencing of the peptide. In the present case, b-type ions
containing the cross-linker dominate the spectra. From the CID
fragmentation pattern (Figure 1A) it is immediately obvious that
fragmentation of the urea cross-linker is strongly preferred, while
the peptide backbone remains virtually intact. The effective
fragmentation of the partially hydrolyzed cross-linker 4 leads to
the formation of prominent “26 u doublet” product ions 6 and 7
(Table 1). As highlighted above, these product ions are indicative
of a selective and sensitive detection of the respective cross-linked
6962 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
species as they cannot be confused with b- and corresponding
a-type ions giving a mass difference of 28 u (CO). Peptides that
are modified by an amidated cross-linker (5), resulting from the
quenching reaction with ammonium bicarbonate, are also observed
in the product ion mass spectra, 37 accordingly exhibiting
CNLs of 128 and 102 u (Scheme 2, Table 1). It should be noted
that peptides modified by hydrolyzed (4) and amidated (5) linkers
were readily separated by our nano-HPLC method with ca. 30 s
elution time difference for a 1000 Da peptide.
An analogous fragmentation behavior was observed for a
modified tryptic peptide of PPARR comprising amino acids
259-272. ESI-CID-MS 2 experiments resulted in characteristic
CNLs of 129 and 103 u with the indicative 26 u pattern due to
(37) Kalkhof, S.; Sinz, A. Anal. Bioanal. Chem. 2008, 392, 305–312.

Figure 1. (A) ESI-LTQ-CID-MS 2 product ion spectrum of the type 0 modified tryptic peptide AKANWLR ion [M + H + BuUrBu-OH] + (structure
4 (Scheme S2, Supporting Information) at m/z 1072.60) of Munc13-1. ESI-LTQ-MS 3 product ion spectra of [M + H + Bu] + at m/z 943.55 (B)
and of [M + H + BuUr] + at m/z 969.53 (C).
cleavage of 4 (Figure S1 of the Supporting Information). In
subsequent MS 3 product ion experiments, the peptide backbone
was sequenced, clearly evidencing the modification of Lys-266
in the flexible ω loop of PPARR (Figure S1B,C).
Intrapeptide Cross-Linking Products (Type 1). As the next
step, we aimed to evaluate the behavior of compound 1 in
intrapeptide cross-linked products (3, Scheme 2S, Supporting
Information). In Figure 2, ESI-CID-MS 2 and -MS 3 data of a
derivatized PPARR peptide (aa 196-210) are exemplarily
presented. In contrast to peptides that were modified by a
partially hydrolyzed cross-linker (type 0), intrapeptide crosslinks
(type 1, structure 3, Scheme 2S) exhibit a characteristic
neutral loss of 85 u upon CID (Table 1), resulting from the
elimination of a pyrolidinone as illustrated in Scheme 3. This
property renders our novel cross-linker highly advantageous for
distinguishing different types of cross-links from each other and
thus makes it an extremely valuable tool for a rapid and automated
screening of cross-linking products. ESI-MS 2 and -MS 3 product
ion experiments enabled us to readily identify Lys-204 to be
connected with Lys-208 in the N-terminal helix of PPARR. The
efficient fragmentation of cross-linked species compared to
conventional NHS ester cross-linkers, such as bis(sulfosuccin-
imyl) suberate (BS 3 ), 38 makes it clearly superior to these
noncleavable derivatives.
Interpeptide Cross-linking Product (Type 2). Finally, we
examined the fragmentation behavior of 1 for the analysis of
interpeptide cross-links (type 2, structure 2), which present the
most informative distance constraints with respect to analyzing
3D structures of proteins or to mapping protein-protein interaction
sites. Modified Munc13-1 (Figure 3) and PPARR (273 aa;
Figure S2, Supporting Information) peptides deliver the two
indicative 26 u doublets of fragment ions 6a, 7a and 6b, 7b
(Scheme 3, Table 1), which allowed an unambiguous identification
of cross-linked lysines. For the Munc13-1 peptide, amino acids
3-9 were found to be connected with amino acids 10-15, clearly
pointing to a cross-link between Lys-4 and Lys-13 (Figure 3). For
the PPARR peptide, Lys-208 was demonstrated to be cross-linked
to Lys-216 (Figure S2), a cross-linking product that had not been
identified in previous cross-linking studies of PPARR with noncleavable
cross-linkers. 39 Interestingly, the most abundant signals
in the MS 3 product ion spectra (Figure S2B,C) correspond to
both unmodified peptides comprising the cross-linking product
(38) Dihazi, G. H.; Sinz, A. Rapid Commun. Mass Spectrom. 2003, 17, 2005–
2014.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6963

Figure 2. (A) ESI-LTQ-CID-MS 2 product ion spectrum of the type 1 modified PPARR peptide (aa 195-209) [M + 2H + BuUrBu] 2+ at m/z
945.47. (B) ESI-LTQ-MS 3 product ion spectrum of [M + 2H + BuUr] 2+ at m/z 902.94. $: signal not assigned.
and, as such, originate from cleavage of the amide bond
between the lysine ε-amine group and the cross-linker.
Offline Nano-HPLC/MALDI-TOF/TOF Mass Spectrometry.
To ensure and to probe the general applicability of 1 for
both ESI- and MALDI-MS, we additionally examined the frag-
(39) Müller, M. Q.; de Koning, L. J.; Schmidt, A.; Ihling, C.; Syha, Y.; Rau, O.;
Mechtler, K.; Schubert-Zsilavecz, M.; Sinz, A. J. Med. Chem. 2009, 52,
2875–2879.
6964 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
mentation behavior of the cross-linker in the higher kinetic energy
regime of collision activation (kinetic energy of the precursor ion
in TOF/TOF-CID Ekin ≈ 1 keV) 40-42 in MALDI-TOF/TOF (ISD
and CID) experiments. In the same fashion as for low-energy ESI-
LTQ-orbitrap-MS n product experiments (kinetic energy of the
precursor ion in LTQ-CID Ekin ≈ 1 eV) 43,44 Munc13-1 and PPARR
peptides were created by tryptic digestion and separated by nano-
HPLC before MS analysis.

Figure 3. (A) ESI-LTQ-CID-MS 2 product ion spectrum of the triply protonated type 2 modified Munc13-1 peptide (amino acids 3-9 connected
with amino acids 10-15) at m/z 597.01. (B, C) ESI-LTQ-MS 3 product ion spectra of the modified R peptide [R +2H + Bu] 2+ at m/z 463.76 (B)
and of the � peptide [� + H + Bu] + at m/z 819.48 (C).
Peptides Modified with Hydrolyzed Linkers (Type 0). As
with collision activation in the low-energy regime in an LTQ-CID-
MS 2 product ion experiment of the type 0 modified Munc13-1
peptide (Figure 1A), the MALDI-TOF/TOF-CID spectrum exhibits
the 26 u doublet of product ions 6 and 7 as base peaks
resulting from the highly preferential cleavage of the urea crosslinker
(Figure 4, Table 1). However, due to the overall increased
(40) Pittenauer, E.; Allmaier, G. Comb. Chem. High Throughput Screening 2009,
12, 137–155.
(41) Medzihradszky, K. F.; Campbell, J. M.; Baldwin, M. A.; Falick, A. M.; Juhasz,
P.; Vestal, M. L.; Burlingame, A. L. Anal. Chem. 2000, 72, 552–558.
(42) Shenar, N.; Sommerer, N.; Martinez, J.; Enjalbal, C. J. Mass Spectrom. 2009,
44, 621–632.
(43) Schwartz, J. C.; Senko, M. W.; Syka, J. E. J. Am. Soc. Mass Spectrom. 2002,
13, 659–669.
(44) Douglas, D. J. Mass Spectrom. Rev. 2009, 28, 937–960.
collision activation in the TOF/TOF-CID experiment (Figure 4)
compared to the LTQ-CID experiment (Figure 1A), a significant
series of b and y ion signals resulting from the cleavage of peptide
backbone amide bonds were additionally observed in the former.
ISD-MS/MS data of primary product ions observed in Figure 4 for
signals at m/z 943.6 (corresponding to a CNL of 129 u) and m/z
969.6 (corresponding to a CNL of 103 u) are comparable to those of
the fragmentation product of a Munc13-1 cross-link, in which Lys-4
is connected with Lys-13 (Figure 3). Also, backbone fragmentation
of the peptide was observed in MALDI-ISD experiments.
MALDI-TOF/TOF data of a PPARR peptide comprising amino
acids 259-272 exhibit a fragmentation behavior analogous to that
in an ESI-CID-MS 2 experiment (Supporting Information, Figure
S3).
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6965

Figure 4. MALDI-TOF/TOF product ion spectrum of the type 0 modified Munc13-1 peptide (peptide amino acids 3-9) [M + H + BuUrBu-OH] +
at m/z 1072.6.
Intrapeptide Cross-Linking Products (Type 1). Also for
intrapeptide cross-links, MALDI-TOF/TOF data were comparable
to ESI-CID-MS 2 data for Munc13-1 (data not shown) and a
PPARR peptide (Supporting Information, Figure S4).
Interpeptide Cross-Linking Products (Type 2). For interpeptide
cross-links 2 within the Munc13-1 peptide, MALDI-TOF/
TOF experiments impressively demonstrated the fragility of the
urea cross-linker upon high-energy CID, which is the reason for
the increased abundance of the two 26 u doublets of product ions
dominating the product ion spectrum of the type 2 modified
precursor ion at m/z 1789.2 (Figure 5, Table 1). According to this
spectrum, the fragmentation characteristics of 1 seem to be
perfectly suited for MALDI-TOF/TOF applications, being even
better than for low-energy MS 2 experiments in a quadrupole ion
trap (Figure 3). The two pairs of 26 u mass shifted product ions
reveal that the R-peptide and �-peptide comprising the cross-linked
product are modified either with 4-aminobutyric acid or with 1,3oxazepan-2-one.
Additionally conducted MALDI-ISD-MS/MS experiments
confirmed the amino acid sequences of both peptides
connected via the BuUrBu linker. This fingerprint feature of two
26 u doublets evidencing the interpeptide cross-link was also found
in the MALDI-TOF/TOF fragment ion spectrum of an interpeptide
(type 2) modified PPARR peptide (Lys-208 connected with Lys-
216) as the most abundant signals in the mass spectrum (Figure
S5, Supporting Information; Table 1). A full list of cross-linking
products from PPARR, which were identified by nano-HPLC/
MALDI-TOF/TOF-MS/MS, is presented in the Supporting Information
(Table S1).
CONCLUSIONS
In this paper, we describe the synthesis and application of a
novel dissociative amine-reactive cross-linker that shows enhanced
6966 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
fragmentation capabilities in both ESI-CID-ion trap-MS n and
MALDI-TOF/TOF-CID-MS/MS and ISD-MS/MS and as such
possesses a versatile applicability. The reactivity and the
kinetics of our novel cross-linker were found to be comparable
with that of other NHS esters, such as BS 3 . The cross-linker 1
contains a urea moiety allowing for a facilitated cleavage upon
CID. The key step in the fragmentation mechanism of the
cross-linker is a nucleophilic attack at the urea carbonyl group,
enabling a highly effective cleavage of the urea moiety. The
synthesis of the cross-linker is simple compared to that of the
previously published “Edman” cross-linker 13 and is achieved
in two steps with inexpensive starting materials. The BuUrBu
structure is symmetric, and hence, assignment of cross-linked
product ions is facilitated compared to the rather complicated
product ion spectra obtained with the previously published
unsymmetric tandem MS cleavable linker. 13 The effective
fragmentation properties of the CID cleavable cross-linker
safeguarding for a selective identification of modified peptides
by tandem MS improves the sensitivity of chemical crosslinking
for protein structure analysis. This is highly advantageous
for discriminating cross-linked species compared to
isotope-labeled, i.e., deuterated linkers, which are employed
as 1:1 mixtures of nondeuterated and deuterated derivatives,
and in constrast to those, no decrease in MS signal intensity
is observed for CID labile cross-linkers.
Especially during MALDI-TOF/TOF experiments, higher quality
MS/MS data were obtained compared to those for conventional
noncleavable NHS esters, such as BS 3 . From CID MS 2 product
ion mass spectra, the type of cross-linking present is readily
visible: Peptides that are modified with a hydrolyzed linker
(dead-end or type 0 cross-links) deliver a single pair of product
ions mass shifted by 26 u due to the loss of 103 and 129 u

Figure 5. (A) MALDI-TOF/TOF product ion spectrum of the type 2 modified Munc13-1 peptide at m/z 1789.2. (B) Upper panel: ISD-MS 2
product ion spectrum of the �-peptide [� + H + Bu] + at m/z 819.4. Lower panel: ISD-MS/MS product ion spectrum of the �-peptide [� + H +
BuUr] + at m/z 845.5. (C) Upper panel: ISD-MS 2 product ion spectrum of the R-peptide [R +H + Bu] + at m/z 943.5. (C) Lower panel: ISD-MS/MS
product ion spectrum of the R-peptide [R +H + BuUr] + at m/z 969.5.
(Table 1). The resulting 26 u product ion pair (“doublet”) greatly
simplifies identification of the respective cross-linking type and
therefore allows for an automated analysis of tandem MS data. A
CNL of 85 u (pyrolidinone) indicates an intrapeptide cross-link
(type 1), whereas an interpeptide (type 2) cross-link delivers two
sets of 26 u doublets originating from both R- and �-peptides, each
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6967

eing modified by the cross-linker fragments Bu and BuUr. A
summary of the characteristic fingerprint product ions and neutral
losses that serve for a discrimination of the different cross-linking
types is presented in Table 1. Additionally conducted ISD-MS/
MS and ion trap-MS 3 experiments clearly confirm the amino
acid sequences of cross-linked peptides and allow reliable
structure identification. Our cleavable cross-linker 1 allows an
automated identification of cross-links combined with a database
search based on the characteristic constant neutral losses
as well as the abundant 26 u mass shifted ion pairs. Thus, highabundance
unmodified peptides are excluded from further
analysis and allow a highly efficient analysis of cross-linked
products. Our novel cross-linker is expected to have a bright
future for protein structure analysis.
ACKNOWLEDGMENT
M.Q.M. is supported by the DFG Graduiertenkolleg 1026
“Conformational Transitions in Macromolecular Interactions” at
the Martin-Luther-Universität Halle-Wittenberg. Dr. O. Jahn kindly
provided the Munc13-1 peptide. A.S. gratefully acknowledges
financial support by the DFG and the BMBF (ProNet-T 3 project).
M.S. and F.D. gratefully acknowledge financial support by the
DFG.
APPENDIX
Glossary
aa amino acid
ACN acetonitrile
6968 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
BS 3 bis(sulfosuccinimidyl) suberate
CID collision-induced dissociation
CNL constant neutral loss
Da Dalton
DCC N,N′-dicyclohexylcarbodiimide
DMSO dimethyl sulfoxide
ESI electrospray ionization
FA formic acid
HCCA R-cyano-4-hydroxycinnamic acid
HEPES 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid
HPLC high-performance liquid chromatography
ISD in-source decay
LID laser-induced dissociation
LTQ linear quadrupole ion trap (Thermo Fisher)
MALDI matrix-assisted laser desorption/ionization
MS mass spectrometry
NHS N-hydroxysuccinimide
RT room temperature
TCEP tris(2-carboxyethyl)phosphine
TFA trifluoroacetic acid
Th Thomson (m/z)
TOF time-of-flight
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in the text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 13, 2010. Accepted July 8, 2010.
AC101241T

Anal. Chem. 2010, 82, 6969–6975
Analyte Discrimination from Chemiresistor
Response Kinetics
Douglas H. Read* and James E. Martin
Sandia National Laboratories, P.O. Box 5800, Albuquerque, New Mexico 87185-1415
Chemiresistors are polymer-based sensors that transduce
the sorption of a volatile organic compound into a resistance
change. Like other polymer-based gas sensors that
function through sorption, chemiresistors can be selective
for analytes on the basis of the affinity of the analyte for
the polymer. However, a single sensor cannot, in and of
itself, discriminate between analytes, since a small concentration
of an analyte that has a high affinity for the
polymer might give the same response as a high concentration
of another analyte with a low affinity. In this paper
we use a field-structured chemiresistor to demonstrate
that its response kinetics can be used to discriminate
between analytes, even between those that have identical
chemical affinities for the polymer phase of the sensor.
The response kinetics is shown to be independent of the
analyte concentration, and thus the magnitude of the
sensor response, but is found to vary inversely with the
analyte’s saturation vapor pressure. Saturation vapor
pressures often vary greatly from analyte to analyte, so
analysis of the response kinetics offers a powerful method
for obtaining analyte discrimination from a single sensor.
Polymer sorption is the basis of most simple methods of
sensing vapors of volatile organic compounds. Such devices
include quartz crystal microbalances and surface acoustic wave
sensors that transduce mass sorption into a frequency change 1-4
and methods that transduce mass uptake into a resistance or
capacitance change, such as chemiresistors, chemicapacitors, and
CHEMFETs. 5-13 Regardless of the sensing mechanism, each
* To whom correspondence should be addressed. E-mail: dhread@sandia.gov.
Phone: (505) 844-5338. Fax: (505) 844-4045.
(1) Hierlemann, A.; Ricco, J.; Bodenho, K.; Dominik, A.; Golpel, W. Anal. Chem.
2000, 72, 3696–3708.
(2) Wibawa, G.; Takahashi, M.; Sato, Y.; Takishima, S.; Masuoka, H. J. Chem.
Eng. Data 2002, 47, 518–524.
(3) Ho, C. K.; Lindgren, E. R.; Rawlinson, K. S.; McGrath, L. K.; Wright, J. L.
Sensors 2003, 3, 236–247.
(4) Fang, M.; Vetelino, K.; Rothery, M.; Hines, J.; Frye, G. C. Sens. Actuators,
B 1999, 56, 155–157.
(5) Janata, J. Electroanalysis 2004, 16 (22), 1831–1835.
(6) Wilson, M. W.; Hoyt, S.; Janata, J.; Booksh, K.; Obando, L. IEEE Sens. J.
2001, 1 (4), 256–274.
(7) Patel, S. V.; Mlsna, T. E.; Fruhberger, B.; Klaassen, E.; Cemalovic, S.; Baselt,
D. R. Sens. Actuators, B 2003, 96, 541–553.
(8) Donaghey, L. F. Resistive Hydrocarbon Leak Detector. U.S. Patent
4,631,952, Dec 30, 1986.
(9) Lewis, N. S.; Doleman, B. J.; Briglin, S.; Severin, E. J. Colloidal Particles
Used in Sensing Arrays. U.S. Patent 6,537,498, March 25, 2003.
(10) Lundberg, B.; Sundqvist, B. J. Appl. Phys. 1986, 60 (3), 1074–1079.
(11) Lei, H.; Pitt, W. G.; McGrath, L. K.; Ho, C. K. Sens. Actuators, B 2004,
101, 122–132.
individual sensor, consisting of a single polymer, cannot discriminate
between analytes if only the equilibrium mass uptake is used,
unless somehow the partial pressure of the analyte is either known
or measured. For these polymer-based sensors, analyte discrimination
is currently based on the artificial nose concept, wherein
arrays of sensors having differentiating chemical affinities are
exposed to the vapor. 13-16 Any analyte will then give a more-orless
unique relative equilibrium mass uptake to the array elements,
generating a response fingerprint. This equilibrium approach can
enable the discrimination of analytes having disparate chemical
affinities, but will not be as useful for distinguishing homologous
analytes, such as octane from decane or xylene from mesitylene.
The ability to distinguish between homologous analytes requires
nonequilibrium information.
In this study we use magnetic field-structured chemiresistors
to show that polymer sorption kinetics enables discrimination
between even homologous analytes. The basis for this discrimination
derives in part from Flory-Huggins theory, which shows that,
for analytes having the same chemical affinity for a particular
polymer, the analyte’s equilibrium mass sorption is determined
by the analyte activity alone. 17 This chemical affinity is quantified
by the Flory parameter, �, and activity is defined as the ratio of
the analyte vapor pressure to its saturation vapor pressure, or a
=P/P*. For linear alkanes the saturation vapor pressure decreases
by about a factor of 3 for every additional carbon, so at the same
activity, octane vapor will have roughly 10 times the number
density of molecules as decane, yet will lead to about the same
equilibrium polymer swelling. The flux of the analyte into the
polymer, and therefore the mass transport into the chemiresistor,
is proportional to its diffusivity times the analyte number density,
and the latter can be obtained from equilibrium thermodynamics.
Therefore, we expect swelling will be roughly 10 times faster for
octane than for decane, provided their diffusivities are similar. If
the analyte diffusivities are similar, then to first-order approximation,
the characteristic swelling rate should simply be proportional
to the analyte’s saturation vapor pressure. This swelling time is
expected to be independent of the analyte’s concentration. This
is because when in the linear swelling regime (a valid assumption
(12) Ho, C. K.; Hughes, R. C. Sensors 2002, 2, 23–34.
(13) Doleman, B. J.; Lonergan, M. C.; Severin, E. J.; Vaid, T. P.; Lewis, N. S.
Anal. Chem. 1998, 70 (19), 4177–4190.
(14) Grate, J. W. Chem. Rev. 2008, 108, 726–745.
(15) Kim, Y. S.; Ha, S. C.; Yang, Y.; Kim, Y. J.; Cho, S. M.; Yang, H.; Kim, Y. T.
Sens. Actuators, B 2005, 108, 285–291.
(16) Eastman, M. P.; Hughes, R. C.; Yelton, G.; Ricco, A. J.; Patel, S. V.; Jenkins,
M. W. J. Electrochem Soc. 1999, 146 (10), 3907–3913.
(17) Flory, P. J. Principles of Polymer Chemistry; Cornell University: Ithaca, NY,
and London, 1953; pp 495-514.
10.1021/ac101259w © 2010 American Chemical Society 6969
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/21/2010

Figure 1. Response of an FSCR to a 2000 s exposure of p-xylene
at an activity of 0.019. The conductance drops exponentially and then
recovers on the same time scale during the nitrogen purge.
for most vapor detection applications), twice the analyte vapor
pressure gives twice the swelling, yet also gives twice the number
density of analyte molecules, and thus twice the diffusive flux. A
swelling time that is independent of the analyte concentration is
highly desirable, since it would simplify the interpretation of
kinetics data. In fact, the raw data we collect are sensor response
versus time, so the response must be converted to polymer
swelling before the swelling time can be computed. This is
accomplished using the transduction curve of the particular
chemiresistor, which is the relation between its change in
conductance and equilibrium swelling. In a previous paper we have
shown that this transduction curve is independent of the analyte,
and so can be determined from testing the chemiresistor with
any particular analyte, regardless of its affinity for the polymer. 18,19
The kinetics data we present were taken with the fieldstructured
chemiresistors (FSCRs) we have developed over the
past several years. 18-23 These sensors differ from traditional
carbon black chemiresistors in that the particle phase is composed
of Au-coated Ni particles that are structured into conducting
pathways using magnetic fields. Field-structuring consistently
brings the particle phase to a conducting threshold over a wide
range of particle volume fractions, and the Au-coated particles do
not adsorb typical VOCs. As a result, FSCRs have significantly
increased baseline stability, sensitivity, reversibility, and sensorto-sensor
reproducibility. 18
BACKGROUND
Response Curve. The response of a chemiresistor exposed
to a step increase in analyte concentration is given in Figure 1.
The conductance drops to its equilibrium value as the polymer
swells and then recovers on the same time scale as the polymer
deswells. FSCR response is defined as the conductance ratio G/G0,
where G and G0 are the conductances in the presence and
absence of analyte, respectively. Plotting the equilibrium value
of G/G0 as a function of analyte concentration gives a sigmoidal
(18) Read, D. H.; Martin, J. E. Adv. Funct. Mater. 2010, 20 (11), 1–8.
(19) Read, D. H.; Martin, J. E. Anal. Chem. 2010, 82 (12), 5373–5379.
(20) Martin, J. E.; Anderson, R. A.; Odinek, J.; Adolf, D.; Williamson, J. Phys.
Rev. B 2003, 67, 094207.
(21) Martin, J. E.; Hughes, R. C.; Anderson, R. A. Sensor Devices Comprising
Field Structured Composites. U.S. Patent 6,194,769, Feb 27, 2001.
(22) Martin, J. E.; Hughes, R. C.; Anderson, R. A. Field-Structured Material Media
and Methods for Synthesis Thereof. U.S. Patent 6,290,868, Sept 18, 2001.
(23) Read, D. H.; Martin, J. E. Anal. Chem. 2010, 82, 2150–2154.
6970 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 2. The FSCR equilibrium conductance versus analyte activity
plot forms a sigmoidal curve, which is parametrized by Γ and the
response midpoint, a1/2. For example, the acetone curve has a
response midpoint of ∼0.15, and the detection range is from an
activity of ∼0.01 to an activity of 0.36. Response curves for this single
FSCR exposed to various analytes illustrate selectivity for more
hydrophobic analytes. As evident, these equilibrium data do not
discriminate between chemically similar analytes.
equilibrium response curve, as in Figure 2. Because the relative
resistance change, ∆R/R0, is exponential with the analyte
concentration, the relative conductance can be fit by 18
G
)
G [ 0
1 + eΓa/a1/2 - 1
e Γ -1
- 1 ]
Here a is the analyte vapor activity, Γ is a fit parameter related to
the abruptness of the conductor-insulator transition (typically
∼4), and a1/2 is the response midpoint, defined as the analyte
activity that reduces G0 by half.
Figure 2 shows that a poly(dimethylsiloxane) chemiresistor
is much more sensitive to the hydrophobic analytes toluene,
p-xylene, mesitylene, and undecane than to the hydrophilic analyte
acetone. In fact, the response midpoint for propanol is nearly 10×
that of toluene or p-xylene. Although this polymer is selective for
hydrophobic analytes, the equilibrium response cannot be used
to discriminate between different analytes unless the analyte
activity is known. More typically, these data would be used to
determine the activity of a known analyte.
Transduction Curve. We have previously shown that the
response of an FSCR is a universal function of polymer swelling,
regardless of the chemical nature of the analyte. 18,19 The response
curve in Figure 2 can thus be thought of as a combination of a
solely device-dependent transduction curve (conductance as a
function of swelling) and the solely analyte-dependent mass
sorption isotherm that relates polymer swelling to the vapor
concentration. The transduction curve for a typical sensor is shown
in Figure 3, with the volume fraction of absorbed analyte
determined gravimetrically. This curve is given by
G
)
G [ 0
1 + eΓφ/φ1/2 - 1
e Γ -1
- 1 ]
where φ is the volume fraction of absorbed analyte and φ1/2 is
the response midpoint. The transduction curve is parametrized
by φ1/2 and Γ, and these are strongly dependent on the
fabrication process used to make the sensor. 18,19,23 To a good
(1)
(2)

Figure 3. Equilibrium response as a function of the volume fraction
of absorbed analyte, giving a master transduction curve that is
dependent only on the characteristics of the particular sensor. (The
data for pentane, octane, decane, TMP, TBT, and undecane are
reprinted from Figure 8 of ref 19. Copyright 2010 American Chemical
Society.) Here the response midpoint, φ1/2, is 5.64 × 10 -3 , and Γ is
2.75. a1/2 and � values for each analyte are reported in ref 19.
approximation, the equilibrium sorption is proportional to the
analyte activity, a, and is given by the linearized Flory-Huggins
equation 17,19
φ ) ae -(�+1)
Here � is the Flory interaction parameter, which is a measure of
an analyte’s affinity for the polymer. This transduction curve is of
great importance in this paper, as we will use it to convert
nonequilibrium response curves, such as that in Figure 1, into
the nonequilibrium sorption curves from which we obtain kinetics
data.
EXPERIMENTAL SECTION
Chemiresistor Fabrication. The chemiresistors used in this
research consist of five identically fabricated FSCRs. The FSCR
composite is composed of 15 vol % 3-7 µm gold-plated carbonylnickel
particles (Goodfellow Inc., product no. NI06021) and a
two-part, addition-cure poly(dimethylsiloxane) (PDMS) (Gelest
Inc. optical encapsulant 41, PP2-OE41). The particles are immersion-gold-plated
using Enthone Inc. Lectroless Prep (PCN 210004-
001). The particles are mixed in the PDMS precursors, and a
volume of hexane equal to the volume of PDMS is then added to
thin the viscous composite precursor. A ∼5 µL volume of solventcast
composite precursor is then deposited onto a glass substrate,
forming a ∼3 mm diameter film, which spansa1mmgapbetween
the two gold electrode pads. The sensors are cured at 40 °C for
24hina∼750 G uniaxial magnetic field. The composite is placed
in the magnetic field such that a conductive chainlike particle
network is formed across the two electrodes. 18,22
Analytes. The model analytes used in this research are
acetone, toluene, p-xylene, mesitylene, and undecane. All of the
analytes are from either Fisher Chemicals (ACS certified reagent
grade) or Aldrich Chemicals (ReagentPlus grade). Pertinent
physical data for the analytes (saturation vapor pressures and �
(3)
Table 1. Room Temperature Saturation Vapor
Pressures 24 and � Parameters for the Analytes 19
analyte P*(25 °C) (Torr) �
toluene 28.97 1.268
p-xylene 8.80 1.258
mesitylene 2.55 1.424
acetone 228.19 2.226
undecane 0.39 1.126
parameters) are included in Table 1. 19,24 The aromatic compounds
toluene, xylene, and mesitylene and undecane were chosen for
use as analytes due to their similar � parameters for with PDMS
and dissimilar vapor pressures.
Sensor Testing. The sensors are enclosed in a shielded flow
cell with gas inlet and outlet ports and electrical throughputs.
Known concentrations of analyte vapors are produced by mixing
a controlled flow rate of analyte-saturated nitrogen from a
temperature-controlled bubbler system with a controlled flow of
pure nitrogen. 19 Conductance measurements are made by applying
a constant 10 mV dc voltage across the electrodes with a power
supply (Hewlett-Packard model 6552A) and measuring the current
through the chemiresistors with picoammeters (Keithley Instruments
Inc. model 6485).
Sorption Kinetics. To study diffusion, it is necessary to
determine analyte mass sorption as a function of time. This is
accomplished by using the transduction curve in eq 2 to transform
the FSCR response curves, such as that in Figure 1, into the
volume fraction of absorbed analyte. Solving eq 2 for φ(t) gives
φ(t) ) φ1/2 Γ ln[ 1 + (eΓ - 1) G0 - G(t)
G(t) ] (4)
Using the equilibrium transduction curve to relate the nonequilibrium
response to the nonequilibrium sorption is an approximation,
since it is an unproven assumption that the sensor
conductance depends only on the total mass sorption and is
insensitive to the swelling gradients that accompany diffusion. For
direct mass uptake experiments gradients pose no problems, but
gradients could potentially foil our approach, yet as we show in
the following section, we obtain a response time that is clearly
independent of the analyte activity. Perhaps this assumption is
reasonable because gradients would have no effect except to
second order: If the equilibrium sensor conductance is locally
linear in the analyte activity, then only the curvature of the gradient
would lead to a conductance response that is different from the
zero gradient response at the same analyte uptake. The utility of
this approximation can only be judged by the quality of the final
experimental results shown in the following section.
Figure 4 illustrates a time-dependent mass sorption curve
obtained from the sensor response. In this figure the equilibrium
swelling, calculated from applying eq 4 to the raw conductance
data, differs from the prediction of the linearized Flory-Huggins
equation (eq 3) by ∼10%. This experimental error does not cause
an error in the computed sorption time. Diffusion into a finite slab
(24) Yaws, C. L.; Narasimhan, P. K.; Gabbula, C. Yaws’ Handbook of Antoine
Coefficients for Vapor Pressure [Online], 1st electronic ed.; Knovel: New York,
2005. URL: http://www.knovel.com/web/portal/browse/display?_EXT_
KNOVEL_DISPLAY_bookid)1183, accessed 3/5/2009.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6971

Figure 4. The volume fraction of absorbed analyte, φ, approaches
its asymptotic value exponentially upon exposure to an analyte, in
this case p-xylene vapors, at an activity of 0.054.
Figure 5. The measured characteristic response time, τmeas, isthe
y intercept of the line obtained by plotting A ) ∫0 t ∆φ(s)/φ∞ ds as a
function of B ) ∆φ(t) ) φ∞ - φ(t).
is a mathematically complex problem, but for practical purposes
the kinetic data can be fit by the simple exponential expression
φ(t) ) φ ∞ (1 - e -t/τmeas ) (5)
Here t is time and τmeas is the measured response time of the
chemiresistor to the analyte in question. Fitting to this form
will prove useful in correcting the observed kinetic data for
the time it takes for the analyte concentration to reach the
steady state in the flow cell. Despite this, we can actually obtain
the response time in a model-independent fashion. To do so,
we simply plot the integral A ) ∫0 t ∆φ(s)/φ∞ ds against B )
∆φ(t)/φ∞, where ∆φ(t) ≡ φ∞ - φ(t). We can operationally define
τmeas as the y intercept of this curve in the limit as ∆φ(t) f 0,
but if eq 5 is a reasonable fit, the result will be a straight line
whose y intercept is easily obtained. The data in Figures 5 are
indeed linear, so eq 5 is actually quite a good description of the
raw kinetic data.
RESULTS
In the following we give the theoretical expression for the true
sorption kinetics, which is shown to be independent of the analyte
activity. We then develop an expression for the measured sorption
kinetics, which is a convolution of the true sorption kinetics and
the kinetics of filling the flow cell. We then show how the flow
cell fill kinetics can be determined from the measured sorption
kinetics and how the fill time can be used to extract the true
6972 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
sorption time from the measured time. Both the measured and
true sorption kinetics are shown to be independent of the activity,
but strongly dependent on the analyte saturation vapor pressure.
Predicted Sorption Time. From Raoult’s law, the partial
pressure of a chemical species, P, in a gas is equal to its volume
fraction, φ vap, times the total pressure, or φvap ≡ V/Vtot ) P/Ptot.
Recall that the analyte activity is a ) P/P*; therefore, a ) φvap/
φvap*, where φvap* is the volume fraction of analyte vapor at
saturation. From the linearized form of the Flory-Huggins
equation (eq 3), φ ) a/e 1+� , the partition coefficient, K, is then
K ≡ φ 1
)
φvap φvap *e 1+�
At high inlet stream fluxes the response time of an FSCR is
diffusion limited, and from Fick’s second law of diffusion for the
case of a semi-infinite slab, the sensor’s response time can be
expressed as
τd ∝ K d2 1
)
Dt Dtφvap *e 1+�d2
(6)
(7a)
where d is the thickness of the composite and Dt is the diffusion
coefficient of the analyte into the silicone. Note that even for
analytes of identical chemical affinity there is discrimination
based on sorption kinetics, due to variations in their diffusivity
and saturation volume fraction. From the ideal gas law, the
saturation vapor pressure is given by P* )Fφvap*RT/Mw, so
the sorption time can also be written as
τ d ∝ RT
P*
F
M w D t e 1+�d2
(7b)
The saturation vapor pressure varies over a wide range, so the
sorption kinetics should be a very useful method of discrimination.
Measured Sorption Time. The goal is to determine the true
sorption time, including any dependence this time might have on
the analyte activity. Unfortunately, the measured sorption time is
a convolution of two factors: the time it takes for the analyte
activity in the flow cell to reach its steady-state value (flow cell
fill time) and the true mass sorption time. In a constant-flow-rate
apparatus, and for any particular analyte, neither of these factors
should be dependent on the analyte activity, so the measured
sorption time should be independent of the analyte activity. The
data in Figure 6 show that this is indeed the case and also show
significant differences in the sorption kinetics for different analytes,
as expected. At low concentrations (where the slope of the
response curve is low) and for long response times, sensor drift
becomes an issue and response times are difficult to reliably
measure; such is the case for low concentrations of undecane.
This issue, however, can be addressed by using a sensor with
high sensitivity at low analyte concentrations. 23
The data in Figure 6 are an average for five sensors, each
tested simultaneously in the same flow cell. Because of sensorto-sensor
variations in polymer thickness, each chemiresistor had
a somewhat different response time. For example, the five sensors
exposed to mesitylene at an activity of 0.058 have an average
response time, τmeas,of48± 13 s. To account for these thickness

Figure 6. The average measured response time, τmeas, is independent
of the volume fraction of absorbed analyte, φ, and thus the
analyte activity. This response time, however, is strongly analyte
dependent, enabling discrimination between analytes having nearly
identical chemical affinities such as toluene and undecane. At low
concentrations (where the slope of the response curve is low) and
for long response times, sensor drift becomes an issue and response
times are difficult to reliably measure; such is the case for low
concentrations of undecane.
variations, we normalized the response time of each sensor to
agree with that of an arbitrarily chosen reference sensor, by
exposing the sensors to a mesitylene activity of 0.058, which
resulted in a swelling of 0.005. These renormalized time scales
were then used to determine both the average response time
to an analyte and the associated measurement error in terms
of the standard deviation. The dimensionless correction factors
ranged from 0.6 to 1.3.
To determine if it is a reasonable assumption that the variation
in the measured response time is caused by variations in sensor
thickness, we made 10 chemiresistors on a glass slide using
methods and materials identical to those for the originals. Sensor
thickness measurements, made with a Nikon mm-800 measuring
microscope with a quadra-chek 200 advanced digital readout
system, give a sensor thickness of 216 ± 26 µm. The average
response time, τ meas, of the original five sensors to mesitylene
at an activity of 0.058 was 48 ± 13 s. Using this average response
time and the average composite thickness of the newly made
sensors (i.e., 216 µm), we calculate a measured diffusion
coefficient of 4.5 × 10 -2 cm 2 /s. From this diffusion coefficient
we computed the predicted response times for the 10 new
sensors, yielding 48 ± 8 s. Therefore, the variation in response
times that we measured for the original sensors is similar to
the variation in response time that we predict from the newly
made sensors. It should be noted that this measured diffusion
coefficient is not a true diffusion coefficient because it is
calculated from τmeas, which is a convolution of the actual
diffusion time scale and the time scale from the flow cell fill
time as we will discuss in the following section.
An accurate determination of the sorption kinetics ideally
requires that the flow cell fill time is fast compared to the sorption
kinetics. In the following we will determine this fill time from the
measured sorption kinetics and use this value to extract the true
sorption kinetics from the measured values. The true sorption
time can then be compared to the flow cell fill time to determine
whether this fast fill time condition is met in our experiments.
Flow Cell Fill Time. After the inlet stream starts to deliver
an analyte of fixed activity to the flow cell, the analyte activity
rises continuously, due to the finite volume of pure nitrogen that
must be displaced. The transient analyte activity can be obtained
by modeling the flow cell as a continuously stirred tank and is of
the form
a(t) ) a ∞ (1 - e -t/τf ) (8a)
In terms of the volume fraction of the vapor this is
φ vap (t) ) φ vap (∞)(1 - e -t/τf ) (8b)
The characteristic time for the flow cell to reach a steady-state
concentration is given by τf ) Vf/F, where Vf is the volume of
the flow cell and F is the volumetric flow rate of nitrogen from
the inlet stream.
Convolution of Time Scales. The measured composite
swelling in Figure 4 is a convolution of the increase in the analyte
vapor activity in the flow cell and the diffusive kinetics, which for
simplicity we take to be of the form φ(t) ) φ∞(1 - e -t/τ d) for a
step increase in the analyte activity at t ) 0. Using eq 5, the
expression is
t dφvap (t)
φ(t) ) K∫ 0
t)s dt [1 - e-(t-s)/τd ]ds (9)
Using eq 8a to evaluate the derivative gives
φ(t) ) Kφvap (∞) t
τ ∫ e
0
f
-s/τf -(t-s)/τd [1 - e ]ds (10)
A straightforward integration leads to the final expression for the
sorption kinetics”
φ(t) ) Aφvap (∞)[ 1 - τd e
τd - τf -t/τd +
τ f
e
τd - τf -t/τf]
(11)
The measured lifetime can then be computed from this equation,
and the surprisingly simple result is
τmeas ) 1 ∞
φ ∫ [1 - φ(t)] dt ) τ
0
d + τf ∞
(12)
The true diffusion time can thus be obtained from the measured
time by τd ) τmeas s τf. (The case where τd = τf leads to division
by zero in eq 11. This division looks troublesome, but can be
handled by defining τd ) τf(1 + ε) and carefully taking the limit
as ε f 0.)
Corrected Sorption Times. The correction of the measured
sorption times for the flow cell fill time requires a determination
of the fill time. The fill time can be extracted from the measured
sorption times themselves, through a limiting process, as we will
now describe. To obtain accurate measured sorption times, we
first average these measured times over all analyte activities to
obtain a mean response time we call τjmeas. This averaging is valid,
because the measured sorption time is independent of the
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6973

Figure 7. The measured sorption time, τmeas, is proportional to the
inverse saturation vapor pressure. The nonzero y intercept of 13.8 s
is the flow cell fill time, which when subtracted from the measured
sorption time gives the true sorption time, τd. The sorption time of
undecane is nearly 2 orders of magnitude larger than that of toluene
despite their equilibrium response curves being nearly identical as in
Figure 2.
analyte activity as demonstrated in Figure 6. A plot of this
average time versus the inverse analyte saturation vapor pressure
results in a straight line with a nonzero y intercept, as seen in
Figure 7. Of course, at infinite saturation vapor pressure the
sorption time τd must be zero since the diffusive flux is infinite.
Therefore, this finite intercept of 13.8 s corresponds to the fill
time of the flow cell and can be used to produce the corrected
sorption times, τd, in Figure 7. This measured flow cell fill time
is in good agreement with the theoretical prediction, using Vf/F.
Figure 7 shows a nearly perfect proportionality between the
sorption time, τd, and the analyte saturation vapor pressure, P*.
Equation 7b shows that other analyte parameters, such as the
diffusivity and the � parameter, also determine this sorption time,
but apparently these combined factors are more-or-less constant
for the analytes we tested. Also, the saturation vapor pressure
varies strongly from analyte to analyte. If finer resolution was
desired for differentiation between similarly volatile analytes, these
other analyte parameters could be taken into account. The
homologous series of aromatic analytes, all of which have nearly
identical Flory parameters, and thus identical equilibrium chemiresistor
responses as a function of activity, are now easily distinguished.
For toluene, p-xylene, and mesitylene the sorption times
are 3.7, 10.0, and 37.0 s, respectively, which are in good agreement
with their relative reciprocal saturation vapor pressures, 3.45, 11.4,
and 39.2 (×10 -2 Torr -1 ). These sorption time differences are
large compared to the errors associated with our measurements,
even though the sorption time of toluene is much faster
than the fill time of 13.8 s. The measurement of faster sorption
times is possible, but would require either higher flow rates
and a smaller flow cell volume, so that the fill time is faster, or
thicker sensors, to increase the sorption time. Of course,
increased sorption time comes at the cost of increased sensor
response time, so this would only be desireable if high-volatility
analytes are to be measured.
DISCUSSION
In the pulsed flow experiments the determination of the
sorption time is straightforward, but would require a test unit with
an engineered flow system. This mechanical requirement is
antithetical to the inexpensive and simple chemiresistor concept.
6974 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Other approaches are possible, but are yet untested in the context
of vapor sorption, though analogous approaches have been used
to determine photoluminescence lifetimes. In fact, luminescence
is a useful language in which to describe the other possibilities.
First, a luminescence lifetime can be measured by suddenly
subjecting a material such as a phosphor to steady illumination.
The rise time of the photoemission is the luminescent lifetime.
This approach is analogous to the sorption kinetics reported above.
If instead the steady illumination is suddenly shut off, the emission
decay gives the lifetime. This is analogous to analyzing the
desorption kinetics shown in Figure 1. A third approach is to excite
a phosphor with amplitude-modulated light. The photoemission
will also be amplitude modulated, but shifted in phase, and the
lifetime can be determined from the phase shift and the modulation
frequency. This is analogous to modulating the activity of
the analyte, which is not really practical. The fourth approach is
to excite a phosphor with light that randomly fluctuates in
intensity. The photoemission then also fluctuates in intensity, but
these fluctuations are time correlated. The associated correlation
time can be determined from the photoemission time autocorrelation
function, which is easily computed with a simple shift
register device. This fourth method is applicable to a sensor in
an environment where there is nonstationary air movement that
causes fluctuations in the delivered analyte activity. An engineered
mechanical system needed to deliver a flow pulse can potentially
be supplanted by an electronic chip that computes the correlation
time of the sensor response.
Of course, we cannot expect the activity fluctuations delivered
to the sensor to be purely random. Instead, these activity
fluctuations will themselves have a correlation time (which is an
interesting issue in and of itself). Therefore, the measured
response fluctuations will in general be a convolution of both the
activity fluctuations and the sorption kinetics. Extracting the true
sorption kinetics will require two sensors that have widely
disparate response kinetics (e.g., a thick and a thin polymer film).
The thin sensor will respond rapidly, relative to the correlation
time of the activity fluctuations, to changes in analyte activity. This
sensor will thus measure the correlation time of the activity
fluctuations. The thick sensor will have a sorption time long
compared to this correlation time, and its response will be largely
determined by the sorption kinetics. The thin sensor can be used
to correct this measured sorption time to obtain the true sorption
time. Therefore, by standard time correlation techniques, discrimination
based on sorption kinetics can be developed without
an engineered flow system. This approach will be the subject of
our future research.
CONCLUSIONS
We have used a field-structured chemiresistor to demonstrate
that response kinetics can be used to discriminate between
analytes, even between those that have identical chemical affinities
for the polymer phase of the sensor. To do this, we have used
the analyte-independent transduction curve (conductance versus
polymer swelling) to transform the time-dependent sensor conductance
into a time-dependent polymer swelling. From these
swelling data we can determine the measured sorption time, which
is a combination of the true sorption kinetics and the fill time of
the flow cell. The true sorption kinetics is obtained by correcting

for the fill time, and we find that the sorption kinetics is
independent of the analyte activity, but inversely proportional to
the saturation vapor pressure of the analyte. Saturation vapor
pressures vary greatly from analyte to analyte, making response
kinetics a powerful method of analyte discrimination, even with a
single sensor. Finally, we suggest that when the sensing environment
presents fluctuations in the analyte activity, an analysis of
the fluctuations in the sensor response fluctuations can be used
to extract the sorption kinetics. This approach would obviate the
need for an engineered flow system that can deliver an analyte
pulse and will be the focus of our future work.
ACKNOWLEDGMENT
This work was supported by the Division of Materials Sciences
and Engineering, Office of Basic Energy Sciences, U.S. Department
of Energy. Sandia National Laboratories is a multiprogram
laboratory operated by Sandia Corp., a Lockheed Martin Co., for
the Department of Energy’s National Nuclear Security Administration
under Contract DE-AC04-94AL85000.
Received for review May 13, 2010. Accepted July 1, 2010.
AC101259W
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6975

Anal. Chem. 2010, 82, 6976–6982
Improvement of Sensitivity and Dynamic Range in
Proximity Ligation Assays by Asymmetric
Connector Hybridization
Joonyul Kim, Jiaming Hu, Rebecca S. Sollie, and Christopher J. Easley*
Department of Chemistry and Biochemistry, 179 Chemistry Building, Auburn University, Auburn, Alabama 36849
The proximity ligation assay (PLA) is one of the most
sensitive and simple protein assays developed to date, yet
a major limitation is the relatively narrow dynamic range
compared to other assays such as enzyme-linked immunosorbent
assays. In this work, the dynamic range of PLA
was improved by 2 orders of magnitude and the sensitivity
was improved by a factor of 1.57. To accomplish this,
asymmetric DNA hybridization was used to reduce the
probability of target-independent, background ligation. An
experimental model of the aptamer-target-connector
complex (apt A-T-aptB-C20,PLA) in PLA was developed
to study the effects of asymmetry in aptamer-connector
hybridization. Connector base pairing was varied from
the PLA standard of 20 total bases (C 20) to an asymmetric
combination with 15 total bases (C15). The
results of this model suggested that weakening the
affinity of one side of the connector to one aptamer
would significantly reduce target-independent ligation
(background) without greatly affecting target-dependent
ligation (signal). These predictions were confirmed
using PLA with asymmetric connectors for detection
of human thrombin. This novel, asymmetric PLA
approach should impact any previously developed PLA
method (using aptamers or antibodies) by reducing
target-independent ligation events, thus generally improving
the sensitivity and dynamic range of the assay.
The proximity ligation assay 1 (PLA) is among the most
sensitive protein assays developed to date. PLA can be considered
a descendent of immuno-PCR methods, 2 in which antibodies are
attached to oligonucleotides to allow the coupling of target binding
with exponential amplification by polymerase chain reaction
(PCR). PLA is a homogeneous assay, requiring no washing steps,
with protein detection limits as low as tens of zeptomoles 1 (fM in
1 µL samples). As shown in Figure 1A, the mode of action of PLA
results in a four-part complex composed of the targeted protein
analyte (T), two proximity probes (DNA-conjugated antibodies or
oligonucleotide aptamers; aptA and aptB), and a connector
oligonucleotide (C20,PLA). The three separate DNA molecules
* To whom correspondence should be addressed. Phone: +1 334-844-6967.
Fax: +1 334-844-6959. E-mail: chris.easley@auburn.edu.
(1) Fredriksson, S.; Gullberg, M.; Jarvius, J.; Olsson, C.; Pietras, K.; Gústafsdóttir,
S. M.; Ostman, A.; Landegren, U. Nat. Biotechnol. 2002, 20, 473–
477.
(2) Adler, M.; Wacker, R.; Niemeyer, C. M. Analyst 2008, 133, 702–718.
6976 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 1. Schematics of PLA and model systems described in the
text. (A) Standard PLA, in which a connector length of 20 bases
(C20,PLA, orb ) 20) is used. Ligated signal or background complexes
are not differentiated by qPCR. (B) Asymmetric PLA, in which
connector lengths of 15-19 bases (Cb,PLA, where 15 e b < 20) are
tested. (C) Schematic of the experimental model system used to
predict signal and background in PLA methods. The “signal” complex
is defined as Loop-Cb, whereas the “background” complex is defined
as FreeA-Cb-FreeB. Model connectors were varied from C15 to C20
in this study.
(or moieties) are designed to hybridize into a structure that
promotes a DNA ligase enzyme to covalently couple the two
proximity probes. In the presence of target, the four-part
complex is preferred over target-independent hybridization
(three-part complex) due to a cooperative “proximity effect”
that essentially increases the local concentration of the oligo-
10.1021/ac101762m © 2010 American Chemical Society
Published on Web 07/23/2010

nucleotide tails. 3-5 Thus, the amount of ligation products is
proportional to the amount of protein analyte, and these products
can be made recognizable by PCR for exponential amplification
and highly sensitive quantitation by qPCR. Due to its simplicity
and sensitivity, PLA has found much interest of late and has been
successfully utilized in a variety of applications, such as sensitive
protein detection, 1,6 in situ analysis of protein localization, 7
DNA-protein interactions, 8 clinical diagnostics, 9 and proteinprotein
interaction assays. 10
In the beginning stages of PLA method development, Landegren
and co-workers determined the optimal concentrations of
each proximity probe and the connector oligonucleotide that
maximize the cooperative effect induced by target binding, 1 and
they further proved that this cooperative effect was negatively
related to the Kd of the proximity probes. 6 Major advantages of
PLA are sensitivity, selectivity, and ease of use. Use of at least
two proximity probes, with discrete binding sites on the same
target, allows unprecedented limits of target detection (LOD)
and better specificity compared to other assays with one probe.
Furthermore, the assay requires no washing steps due to its
homogeneous nature. However, PLA has the limitation of a
relatively narrow dynamic range. The reason for this flaw is
the necessity to use very low amounts of proximity probes
(∼10 -16 mol) to minimize a target-independent hybridization
and ligation (background). Such a low amount of proximity
probes allows PLA to assay only up to ∼10 -15 mol (10 times
more analytes than proximity probes) if the Kd of the proximity
probe is less than 0.4 nM. 6
Our goal in this study was to improve the dynamic range of
PLA without sacrificing LOD. To address this issue, we made two
hypotheses: (1) target-independent hybridization (background;
Figure 1A, bottom) is inversely proportional to the Kd of the
connector oligonucleotide at a given concentration of proximity
probes, and (2) the width of the dynamic range is proportional
to the amount of a proximity probe to be used. If confirmed,
these hypotheses suggest that dynamic range and signal to
background could be improved by simply reducing the number
of bases in the connector. To test these hypotheses, first we
developed and tested an experimental model of the proximity
effect. This model suggested that asymmetric connectors with
higher K d values could significantly decrease target-independent
hybridization (background; Figure 1A, bottom). In turn,
this would lead to higher signal-to-background ratios and allow
the use of more proximity probes, which should increase the
dynamic range of PLA. As described below, we successfully
(3) Zhou, H. X. Biochemistry 2001, 40, 15069–15073.
(4) Zhou, H. X. J. Mol. Biol. 2003, 329, 1–8.
(5) Tian, L.; Heyduk, T. Biochemistry 2009, 48, 264–275.
(6) Gullberg, M.; Gústafsdóttir, S. M.; Schallmeiner, E.; Jarvius, J.; Bjarnegård,
M.; Betsholtz, C.; Landegren, U.; Fredriksson, S. Proc. Natl. Acad. Sci.
U.S.A. 2004, 101, 8420–8424.
(7) Söderberg, O.; Leuchowius, K. J.; Gullberg, M.; Jarvius, M.; Weibrecht, I.;
Larsson, L. G.; Landegren, U. Methods 2008, 45, 227–232.
(8) Gustafsdottir, S. M.; Schlingemann, J.; Rada-Iglesias, A.; Schallmeiner, E.;
Kamali-Moghaddam, M.; Wadelius, C.; Landegren, U. Proc. Natl. Acad. Sci.
U.S.A. 2007, 104, 3067–3072.
(9) Nordengrahn, A.; Gustafsdottir, S. M.; Ebert, K.; Reid, S. M.; King, D. P.;
Ferris, N. P.; Brocchi, E.; Grazioli, S.; Landegren, U.; Merza, M. Vet.
Microbiol. 2008, 127, 227–236.
(10) Gustafsdottir, S. M.; Wennström, S.; Fredriksson, S.; Schallmeiner, E.;
Hamilton, A. D.; Sebti, S. M.; Landegren, U. Clin. Chem. 2008, 54, 1218–
1225.
Table 1. Oligonucleotide Sequences Used in the
Experimental Model of Proximity Hybridization a
name sequence b
C20 5′-TAATACTTGCTGAGGAATAA-3′ (similar to that
used in standard PLA)
C19 5′-TAATACTTGCTGAGGAATA-3′
C18 5′-TAATACTTGCTGAGGAAT-3′
C17 5′-TAATACTTGCTGAGGAA-3′
C16 5′-TAATACTTGCTGAGGA-3′
C15 5′-TAATACTTGCTGAGG-3′
FreeA 5′-/IABkFQ/GCA AGT ATT ATT TTT TTT TTT TTT
TTT TTT TTT TTT TTT TTC TCT TTT TC-3′
FreeB 5′-/Phos/TCT TCT CTC TCT CTC TTT TTT TTT TTT
TTT TTT TTT TTT TTT ATT CCT CA/6-FAM/-3′
linker 5′-GAG AGA GAG AGA AGA GAA AAA GAG AAA
AAA-3′
Loop 5′-/IAbFQ/GCA AGT ATT ATT TTT TTT TTT TTT
TTT TTT TTT TTT TTT TTC TCT TTT TC TCT TCT
CTC TCT CTC TTT TTT TTT TTT TTT TTT TTT TTT
TTT ATT CCT CA/6-FAM/-3′
a Bases in bold depict the variable segments of the connectors used
to alter connector affinity (see Table 2 for affinities). b Notes: IABkFQ,
Iowa black FQ; Phos, phosphorylated end; 6-FAM, 5(6)-carboxyfluorescein
(mixed isomer).
improved the sensitivity and substantially widened the dynamic
range for thrombin detection compared to standard PLA. This
novel, asymmetric PLA approach should be generally applicable
to any of the various PLA methods developed to date, using either
aptamer- or antibody-based probes. 1,6-10
EXPERIMENTAL SECTION
Reagents and Materials. All solutions were prepared with
deionized, ultrafiltered water (Fisher Scientific). T4 DNA ligase
was purchased from New England BioLabs. Human thrombin was
obtained from Sigma-Aldrich. All oligonucleotides except the
Taqman probe were obtained from Integrated DNA Technologies
(IDT; Coralville, Iowa), with purity and yield confirmed by mass
spectrometry and HPLC, respectively. Sequences of ssDNA
strands used in the experimental model are given in Table 1.
Sequence Free A was labeled at its 5′-end with Iowa black FQ
quencher (IABkFQ; absorbance maximum at 531 nm). Sequence
FreeB was labeled at its 5′-end with a phosphate group
(Phos) and at its 3′-end with 5(6)-carboxyfluorescein (6-FAM,
mixed isomer; emission maximum at 520 nm). Loop strands
(Table 1) were synthesized by ligation of FreeA and FreeB using
T4 DNA ligase; further details on Loop synthesis are included
in the Supporting Information text and Figure S-2. Sequences
(listed 5′ to 3′) for PLA or asymmetric PLA were as follows.
Thrombin aptamer A, aptA, CAG TCC GTG GTA GGG CAG GTT
GGG GTG ACT TCG TGG AAC TAT CTA GCG GTG TAC
GTG AGT GGG CAT GTA GCA AGA GG; thrombin aptamer
B, aptB, /5Phos/GT CAT CAT TCG AAT CGT ACT GCA ATC
GGG TAT TAG GCT AGT GAC TAC TGG TTG GTG AGG
TTG GGT AGT CAC AAA; symmetric connector, C20,PLA, AAG
AAT GAT GAC CCT CTT GCT AAA A; asymmetric connector,
C18,PLA, TTA TGA TGA CCC TCT TGC TAA AA; asymmetric
connector, C16,PLA, TAG ATG ACC CTC TTG CTA AAA; forward
PCR primer, GTG ACT TCG TGG AAC TAT CTA GCG; reverse
PCR primer, AAT ACC CGA TTG CAG TAC GAT TC; Taqman
probe (Applied Biosystems, ABI), 4-chlorofluorescein-TGT
ACG TGA GTG GGC ATG TAG CAA GAG G-carboxytetramethylrhodamine.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6977

Fluorescence Measurements for Model Titrations. Fluorescence
measurements were conducted using a Nanodrop 3300
fluorospectrometer (Thermo Scientific). The intensity of FAM
emission was measured at 520 nm with LED excitation at 485
nm. The buffer used for titrations was 50 mM Tris-HCl (pH 7.5),
100 mM NaCl, 1 mM MgCl2. Titrations sets (14-16 points) were
carried out in duplicate, and each fluorescence measurement was
repeated six times. Fixed amounts of Loop or FreeA and FreeB
strands were incubated with variable amounts of Cb overnight at
room temperature prior to fluorescence measurements.
Proximity Ligation Assays. Temperature control was achieved
using a Mastercycler-EP gradient thermal cycler (Eppendorf).
Thrombin and oligonucleotide solutions were made as described. 1
An amount of 4 µL of a reaction mixture containing human
thrombin, aptA, and aptB was incubated at 37 °C for 15 min
before adding 1 µL of connector (C16,PLA, C18,PLA, orC20,PLA).
An additional 30 min of incubation at 22 °C was performed to
stabilize complexes. These 5 µL samples were then brought
to a total volume of 55 µL for the ligation and amplification
mixture, which contained 50 mM KCl, 10 mM Tris-HCl, pH
8.3, 1.9 mM MgCl2, 0.4 Weiss unit T4 DNA ligase, 73 µM ATP,
0.18 mM dNTPs, 0.45 µM forward and reverse PCR primers,
45 nM Taqman probe for the 5′ nuclease assay, and 1.5 units
of AmpliTaq Gold polymerase (ABI). The reactions were
transferred to a real-time PCR instrument (CFX96, Bio-Rad)
for temperature cycling: 5 min at 22 °C for ligation, 10 min at
95 °C, and then cycling for 15 s at 95 °C and60sat60°C,
repeated 45 times. Signal-to-background (S/BG) values were
calculated from the relative number of ligation products
detected in a sample with thrombin to that in a sample without
thrombin. The relative quantity of the ligated products was
calculated by using the comparative threshold cycle (Ct) method. 11 For measurements of PLA background, 15 pM aptA
and 20 pM aptB were used, with different connector concentrations
as follows: C16,PLA ) 178 nM; C18,PLA ) 112 nM; C20,PLA )
45 nM. For asymmetric PLA, 900 pM aptA and 1200 pM aptB
were used with 480 nM C16,PLA; and for symmetric PLA, 15 pM
aptA and 20 pM aptB were used with 400 nM C16,PLA (same
conditions as previously reported 1 ).
Connector Kd Calculations. Base-pairing dissociation constants
for the invariable segment (Kd,A) and for each variable
segment (Kd,B) of the connectors were calculated using the wellestablished,
thermodynamic nearest-neighbor model. 12 Dissociation
constants were derived from thermodynamic parameters
calculated via the BioPHP melting temperature calculator
(http://insilico.ehu.es/tm.php), a free online service that uses
SantaLucia’s method. 12 Source code is freely downloadable at
http://www.biophp.org/. Kd,B values were also experimentally
confirmed; in Table 2, the column labeled “Kd,B” shows the
BG calculated values, and the column labeled “Kd,eff” shows the
experimentally confirmed values. These values are in good
agreement, as further shown in the Supporting Information
(Figures S-3 and S-4).
(11) Livak, K. J. User Bulletin No. 2: ABI PRISM 7700 Sequence Detection System;
PE Applied Biosystems: Foster City, CA, 1997; pp 11-15.
(12) SantaLucia, J. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 1460–1465.
6978 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Table 2. Analysis of Connector Sequences Used in the
Experimental Model of Proximity Hybridization a
connector,
C b
length
(bases)
Kd,B BG S (nM) Kd,eff (nM) Kd,eff (nM) S/BG [Cn] opt
(nM)
C20 20 16 18.6 ± 2.7 3.25 ± 0.48 5.72 45
C19 19 135 149 ± 11 5.54 ± 0.93 26.9 80
C 18 18 539 446 ± 59 9.52 ± 1.33 46.8 110
C 17 17 3370 2550 ± 220 7.12 ± 1.67 358 170
C16 16 28 600 7100 ± 740 8.47 ± 0.38 838 480
C 15 15 297 000 17.5 ± 6.3 20 000
a
Notes: Kd,B, calculated dissociation constant for variable segment
of connector in 100 mM salt and 1 mM Mg2+ BG S ; Kd,eff and Kd,eff,
measured
effective dissociation constants of background and signal, respectively;
[Cn]opt, optimal connector concentration for PLA; unvaried Kd,A ) 14.8
nM.
RESULTS AND DISCUSSION
Signal and Background in Proximity Ligation. A schematic
of the original, aptamer-based PLA 1,6 is shown in Figure
1A, in the context of protein detection with aptamer probes.
The quantitative capability of PLA is most heavily dependent
upon two steps: the cooperative hybridization of the four-part
complex, aptamer A-target-aptamer B-connector (aptA-TaptB-C20,PLA),
followed by the proximity-dependent ligation of
the two aptamers using a DNA ligase. This process results in
an amount of ligated product that is proportional to the original
amount of protein analyte (“signal” ligations), albeit with some
analyte-independent ligation products present (“background”
ligations). Although qPCR does not differentiate signal from
background ligations, under optimized conditions 1,6 the signal
will be dominant over background to allow highly sensitive,
indirect detection of the protein analyte. Signal enhancement over
background in PLA is based on the proximity effect. The effective
concentrations of two aptamers are significantly increased due to
their proximity, afforded by simultaneous binding to the same
protein. 3-5
The chief interference in PLA is target-independent, background
ligation (Figure 1A, bottom). This background ligation
has been reported to result from two sources. 1,6,13 First, based
simply on binding equilibria, a fraction of aptamers will always
hybridize with the 20-base connector sequences (C20,PLA), even
in the absence of protein analyte, resulting in target-independent
ligations that increase the background levels in the
assay. 1,6 Second, the T4 DNA ligase is capable of ligating a small
fraction of ssDNA sequences, even in the absence of a connector
sequence. 13,14 This will also increase assay background (not
shown in the figure). Importantly, the presence of this background
limits the amount of usable aptamer (or antibody) probes, which
affects the sensitivity and partially defines the upper limit of the
dynamic range. 6 Methods to widen dynamic range would address
a major weakness of PLA and significantly improve the applicability
of the assay.
Design of the Experimental Model of Proximity Hybridization.
We hypothesized that by decreasing connector binding
affinities for aptamers using an asymmetric connector (Figure 1B),
the amount of background ligation products would be greatly
reduced compared to signal ligation products. In turn, this should
(13) Leslie, D. C.; Sohrabi, A.; Ikonomi, P.; McKee, M. L.; Landers, J. P.
Electrophoresis 2010, 31, 1–8.
(14) Kuhn, H.; Frank-Kamenetskii, M. D. FEBS J. 2005, 272, 5991–6000.

allow the use of higher aptamer concentrations, which can
increase the upper limit of dynamic range and improve sensitivity.
This hypothesis was based on the inherent signal enhancement
given by cooperative binding in the proximity effect, 3-5 which does
not apply to background. To understand the impact of connector
binding affinity on S/BG ratio in PLA, we developed a simple
experimental model using DNA hybridization and fluorescence
quenching (Figure 1C). First, we define an “asymmetric connector”
in the model system as a connector with one side shortened,
compared to the original PLA connector, C20,PLA. Herein, model
connector sequences are represented by Cb, where b is equal
to the total number of bases in the connector that are
complementary to the Loop sequence. For example, the
symmetric connector used in the model is represented by C20,
similar to the 20-base connector used in the original PLA
work, 1,6 whereas an asymmetric connector with only six bases
in the variable region is represented by C16. Sequences are noted
in Table 1, with variable regions in bold. (Note that connectors
used later for PLA or asymmetric PLA are differentiated from
model connectors using the subscript “PLA,” such as C20,PLA or
C16,PLA.) In our model system, signal and background were
treated separately. The equilibria shown in Figure 1C represent
signal complex formation with a sequential binding mode and
background complex formation with an independent binding
mode. These are not blind assumptions but are based upon Hill
coefficients which were determined by nonlinear least-squares
fitting of our experimental data (discussed in detail below).
To model target-dependent signal in PLA, a single DNA loop
was hybridized with connector sequences (Figure 1C, top). The
aptA-T-aptB complex in PLA was modeled using a 100nucleotide
DNA loop (Loop) that is capable of hybridizing with
a connector sequence at both ends, representing the ideal case
in which aptamers possess infinite affinities for their protein
analyte (Kd’s ≈ 0). To measure the extent of hybridization, the
Loop was covalently labeled with IABkFQ quencher at the 5′end
(black circle) and with FAM at the 3′-end (gray circle).
Quenching of FAM fluorescence ensued when both the 5′- and
3′-ends of the Loop were hybridized to a connector sequence,
mimicking the proximity hybridization complex in PLA (i.e.,
Loop-Cb represents aptA-T-aptB-C20,PLA). Thus, a decrease
in Loop fluorescence was referred to as “signal” in this case.
The model of background ligations in PLA (Figure 1C, bottom)
follows from the model of signal. The FAM- and IABkFQ-labeled
Loop sequence was essentially split into two free strands (FreeA
and FreeB), representing aptamers unbound by protein analytes.
Upon combination of both free strands and a connector
sequence, the extent of fluorescence quenching represented
target-independent hybridization in PLA and was proportional
to the amount of background ligation products that would be
expected (i.e., FreeA-Cb-FreeB represents aptA-C20,PLA-aptB).
Thus, a decrease in fluorescence was referred to as “background”
in this case.
Complex Formation with Asymmetric Connectors. Moving
toward our goal of reduced background in PLA, the experimental
model (Figure 1C) was used to determine effective dissociation
constants (Kd,eff) of signal and background complexes at
different connector lengths (Cb, where 15 e b e 20). The
assumption was that the Kd,eff values could represent the relative
amounts of signal and background. Separate titrations of the
Loop and Free strands with the Cb strand gave the corresponding
Kd,eff values of the complexes. The S/BG was then
calculated as
S/BG ) K S
a,eff
BG
Ka,eff ) K BG
d,eff
S
Kd,eff This S/BG represents the ideal case in which aptamers have
infinite affinities for the target (Kd’s ≈ 0).
Connectors’ affinities for the FreeB strand (or for one arm of
the Loop) were reduced by decreasing the number of complementary
bases to FreeB or the Loop arm (gray segments in
Figure 1B; bolded text in Table 1). Base-pairing dissociation
constants for the invariable segment (Kd,A) and for each variable
segment (Kd,B) of the connectors were calculated (Table 2)
using the well-established, thermodynamic nearest-neighbor
model. 12 Starting with the 20-base connector (C20) similar to that
used in standard PLA, the variable segment was decreased in
length, resulting in five different asymmetric connectors (C15
through C19). Titrations with these asymmetric connectors and
the symmetric control connector (C20) were carried out on both
the signal and background models (Figure 1C), and the
decrease in fluorescence was used to assay stability.
Background Reduction in the Experimental Model. Modeling
of cooperative complex formation has proven difficult,
particularly in the realm of predicting the affinity of bivalent
ligands based on known affinities of each individual ligand. 3-5
Rather than attempting to develop an extensive model of the
system shown in Figure 1C, titration data was analyzed using the
Hill equation, 15 a simplified model that has found wide use in
biochemistry, physiology, and pharmacology to analyze binding
equilibria and ligand-receptor interactions. 16 According to
Weiss, 16 the Hill coefficient is best described as an “interaction
coefficient” that reflects the extent of cooperativity among multiple
ligand binding sites. Since it is expected that cooperativity is
inherent to the formation of the aptA-T-aptB-C20 complex in
PLA, or the Loop-Cb complexes in our model, we employed
nonlinear least-squares (NLLS) fitting of our data to the Hill
equation (eq 2) to assay this cooperativity and to determine Kd,eff
values.
S ) S i + (S f - S i )
[C b ] n
(K d,eff ) n + [C b ] n
where S is the fluorescence signal measured as a function of [Cb],
the connector concentration, Si and Sf are the initial and final
signals, respectively, and Kd,eff represents the effective dissocia-
S tion constant of the Loop-Cb complex (Kd,eff) or the FreeA-
BG Cb-FreeB complex (Kd,eff). In this way, fluorescence measurements
during titration with increasingly higher [Cb] could be
fit to the Hill equation (eq 2) and the Kd,eff values could be
extracted. Furthermore, each extracted Hill coefficient, n,
represented the extent of cooperativity in that particular system.
Our expectation was that the Loop-Cb complexes (signal)
(15) Hill, A. V. J. Physiol. 1910, 40, iv–vii.
(16) Weiss, J. N. FASEB J. 1997, 11, 835–841.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(1)
(2)
6979

Figure 2. Results of the experimental model. Fluorescence response curves for the signal (Loop, black circles) and background (Free, gray
open squares) complexes when titrated with (A) C20 or (B) C16. NLLS fits to the Hill equation (eq 2) are shown as solid curves. Other curves
(titration with C15,C17,C18, and C19) are given in Supporting Information Figure S-1. Also shown are (C) relative amounts of signal and background
using each connector and (D) signal-to-background ratios as functions of the dissociation constants (Kd,B) of the variable segments of each
connector, Cb.
would have significantly increased cooperativity (higher n)
compared to the FreeA-Cb-FreeB complexes (background).
Example data sets from these titrations are shown in Figure
2, with error bars representing standard deviations of the triplicate
measurements. Figure 2A shows that the Loop-C20 complex
(filled black circles) is more thermodynamically favored
compared to the FreeA-C20-FreeB complex (open gray squares).
This result was expected, owing to the success of various
PLAs 1,6-10 using 20-base connectors such as C20. This data, and
all titrations to follow, were fit to eq 2 via NLLS fitting to
determine Kd,eff values and S/BG (via eq 1) for each asymmetric
connector (Cb); measured and calculated values are shown in
Table 2. With the use of the symmetric connector, C20, the S/BG
value was determined to be 5.72. Figure 2B shows the
fluorescence measurements from a similar experiment using C16.
This figure clearly shows that the relative stability of the signal
complex (Loop-C16, filled black circles) over the background
complex (FreeA-C16-FreeB, open gray squares) is much
greater than that measured with C20, without a large decrease
in stability of the signal complex. In fact, the S/BG value using
C16 was determined to be 838. Thus, a 146-fold increase in
S/BG was achieved by simply decreasing the connector length
in an asymmetric manner. Titration curves for C15, C17, C18,
and C19 (Supporting Information Figure S-1) follow this trend of
increasing S/BG with decreasing connector length (Figure 2, parts
6980 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
C and D). As shown in Figure 2C, the background levels with
C20 were 382-fold larger than the background with C16, whereas
the signal differed by only 3.16-fold. These large decreases in
background and only modest decreases in signal gave the
expected increase in S/BG with decreasing connector affinity
BG (Figure 2D). The agreement between measured Kd,eff and
calculated Kd,B values was also encouraging (Figures S-2 and
S-3 in the Supporting Information). It must be noted that the
background complex titration experiment for C15 was impractical,
due to the expense involved in working with millimolar
concentrations of DNA. Although, using nearest-neighbor
calculations, 12 it is predicted that the S/BG could be as high
as 1.7 × 104 for C15. These results provided a first step toward
confirmation of our hypothesis.
Also shown in Figure 2D is a curve (dashed line) calculated
from data presented by Tian and Heyduk, 5 who developed a
similar experimental model using bivalent ligands of various
affinities. It is clear that the general trend observed in our data
matches with that seen by Tian and Heyduk with an obvious offset
in magnitude. This offset is likely due to the model differences,
in which Tian and Heyduk used a6nmspacer between their
binding sites, whereas our model has essentially no spacer
between the binding sites. Inclusion of an equivalent spacer in
our system would be unreasonable, since PLA requires subse-

quent ligation of the aptamers after hybridization of the signal
complex. Nonetheless, the similarities were encouraging.
Figure 2D presents the relationship between S/BG and the
affinities of the variable segments of the connectors, which are
represented by the calculated Kd,B values 12 (Table 2). This
analysis is useful, since Kd,B will ultimately become the independent
variable in the asymmetric PLA approach. As shown
in the figure, S/BG increases proportionally with Kd,B. In
theory, one could increase Kd,B even further. In the extreme
example, one could use C11 (only one base pair in the variable
segment). However, the S/BG parameter is not useful if there
is insufficient signal present with respect to the noise of
measurement, and it is likely that the amount of either signal
or background (or both) complexes using C11 would be
negligible. The connector length, b, could be considered an
alternative for the independent variable (shown as upper axis
in Figure 2, parts C and D), but care must be taken, since b
remains dependent upon base composition, making it less precise
than Kd,B.
Hill coefficients (n) were also determined by NLLS fitting of
the data to eq 2. The mean n values for formation of all of the
background (FreeA-Cn-FreeB) and signal (Loop-Cn) complexes
were 1.42 ± 0.19 and 1.76 ± 0.11, respectively. These
values were shown to be statistically different (p < 0.01).
Cooperativity enhancement in the signal complex was expected,
since this effect gives PLA the ability to discriminate signal
from background. Furthermore, as reported by Weiss, 16 these
Hill coefficients suggest that the signal complex forms in a
sequential manner, whereas the background complex forms
in an independent manner, as reflected in Figure 1C.
In summary, the improvements in S/BG in the experimental
model is derived from the more than 2 orders-of-magnitude
BG increase in Kd,eff (reduced affinity) that is achieved using
S asymmetric connectors, whereas Kd,eff is increased by less than
1 order of magnitude (Table 2, Figure 2C); thus, by eq 1, S/BG
is increased significantly (Figure 2D). However, application of
asymmetric connectors to PLA requires experimental confirmation,
as given below, since we began our model with the
assumption of infinite aptamer affinities.
Design of Asymmetric PLA. A schematic of asymmetric PLA
is shown in Figure 1B. Since decreased background was achieved
in the experimental model (Figure 2), the next step was to confirm
that this approach was useful for PLA. The real assay is inherently
more complicated than the experimental model, requiring not only
a four-part complex formation but also ligation of aptA to aptB,
followed by qPCR. 1,6 Furthermore, aptamer affinities are finite
in the real system, compared to the assumption of infinite affinities
in the model. On the basis of these factors, we expected that
the observations from the model would be less pronounced in
the real system. Asymmetric PLA (Figure 1B) is based on the
formation of the complex aptA-T-aptB-Cb,PLA, where b < 10 is
achieved by decreasing connector lengths on the side complementary
to aptB. Shorter connectors, therefore, have higher
Kd,B (Table 2), or lower affinity.
For the standard PLA system (C20,PLA, b ) 20), the optimal
connector concentration was determined to be >40 nM. 1 This
correlates well with the model results shown in Figure 2A. At
C20,PLA ) 40 nM, the stability of the signal complex is near
maximal, whereas the stability of the background complex is
minimized as much as possible. Working on this principle, the
results from the titrations in the experimental model (Figure
2, parts A and B and Supporting Information Figure S-1) were
used to estimate the optimal connector concentrations, [Cn]opt,
for each model connector of different length. These values are
reported in the rightmost column of Table 2 and were used for
all asymmetric PLA experiments to follow. This approach allowed
the interrogation of each asymmetric PLA system at its optimal
values of signal and background.
Background Reduction Using Asymmetric PLA. Starting
with the identical reagents used by Landegren and co-workers 1,6
for sensitive detection of human thrombin, connector lengths were
systematically decreased from the initial C20,PLA sequence. Background
ligation levels were measured by carrying out the assay
without the target protein (human thrombin) present. As shown
in Figure 3A, the background levels in asymmetric PLA were
significantly reduced by increasing Kd,B (or decreasing length,
b), matching well with the experimental model (Figure 2C).
Background levels in asymmetric PLA with C16,PLA were 46.7fold
lower than with standard, symmetric PLA (C20,PLA), thus
confirming our hypothesis. It should, therefore, be possible to
use higher concentrations of aptamer probes with asymmetric
PLA to improve the dynamic range and sensitivity.
Dynamic Range and Sensitivity Enhancement with Asymmetric
PLA. A major limitation of standard PLA is the relatively
narrow dynamic range. This disadvantage is directly related to
the generation of background ligations. The original work in PLA
for human thrombin detection determined the optimal concentrations
of aptamer probes to be 15 and 20 pM for aptA and aptB in
the 5 µL incubation mixture, respectively. 1 At higher probe
concentrations, increased background ligations were shown to
be detrimental to the S/BG of the assay. Since the assay relies
on the simultaneous binding of two aptamers to the same
target, simple probability theory shows that the population of
signal complexes (aptA-T-aptB-Cb,PLA) will first increase then
decrease as the concentration of target is increased. The upper
limit of the dynamic range is thus positively related to both
the concentrations and the Kd’s of the aptamer probes. Gullberg
et al. 6 showed that target at 10 times higher concentration than
probes could not be assayed with probe Kd < 0.4 nM but could
be assayed with a probe Kd ) 2.5 nM, albeit with relatively
low S/BG. This implies that in standard PLA, a lower LOD is
achievable with higher affinity probes at the expense of narrow
dynamic range.
Previous publications using standard, aptamer-based PLA for
thrombin detection did not report the LOD or the dynamic
range. 1,6 However, by analyzing the presented data, the reported
LOD was determined to be 16 amol (significantly different from
background, p < 0.05), with a dynamic range from 16 to 400 amol.
In our laboratory, with the standard PLA system using C20,PLA,
we obtained a detection limit about 3-fold higher than 16 amol,
but we were able to achieve a significantly wider dynamic range
compared to the previous reports. As shown in Figure 3B (open
gray squares), our standard PLA detection limit was 50 amol, and
our dynamic range was from 50 to 3000 amol (7.68-fold wider than
previous reports).
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6981

Figure 3. Asymmetric PLA. (A) Amount of background ligations vs
dissociation constant (Kd,B) of the variable segments of each connector,
Cb,PLA, in the absence of target protein. (B) Titration with human
thrombin. Asymmetric connector C16,PLA (filled black circles) and
symmetric connector C20,PLA (open gray squares) were used for
comparison. Error bars represent the standard error of quadruplicate
qPCR measurements. Horizontal bars define the assay dynamic
ranges. (C) Linear scale comparisons of dynamic range, sensitivity,
and limit of detection (LOD) improvements using asymmetric connector
C16,PLA (white bars) compared to the standard connector C20,PLA
(gray bars). Error bars depict standard deviations of linear sensitivities.
The cross-hatched bar represents the LOD obtained in the original
reports of PLA (refs 1 and 6).
With significant reductions in background using asymmetric
PLA (Figure 3A), we hypothesized that the dynamic range could
be improved using higher aptamer concentrations with connector
C16,PLA. This hypothesis was confirmed by titration of the system
with human thrombin (Figure 3B). Asymmetric (closed black
circles) and standard (open gray squares) PLA were carried out
under the following optimal conditions in the 5 µL incubation
mixture: asymmetric, [C16,PLA] ) 480 nM, [aptA] ) 0.9 nM, and
[aptB] ) 1.2 nM; standard, [C20,PLA] ) 45 nM, [aptA] ) 15 pM,
and [aptB] ) 20 pM. Using asymmetric PLA, the dynamic range
6982 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
was increased by 2 orders of magnitude, and the sensitivity
was improved by a factor 1.57 compared to the standard system
with C20,PLA (Figure 3, parts B and C). In addition, the maximum
achievable S/BG was larger by a factor of 3.15. These improvements
were accomplished with an additional 3-fold improvement
in LOD to 5.0 amol of thrombin. When compared to previously
reported data, 1,6 our dynamic range was improved by an even
larger factor of 323. In Figure 3B, horizontal bars mark the
dynamic ranges, with the upper bounds defined as the highest
concentration with a statistically higher (p < 0.05) value of signalto-background
ratio compared to the adjacent lower concentration.
Error bars represent standard errors of quadruplicate runs of
qPCR.
Since logarithmic scales were used in Figure 3B, linear scale
comparisons of dynamic ranges, sensitivities, and limits of detection
are shown in Figure 3C. Sensitivities were determined as the
slopes of the linear regressions of both data sets in the range
from the LOD to 3000 amol. This figure helps to clarify the
significant advantages provided by asymmetric PLA, in comparison
to standard PLA. Most notably, our approach of using
asymmetric connectors has allowed a significant improvement in
one of the major limitations of the assay, dynamic range. This
improvement should be generally applicable to any of the
previously developed PLA methods. 1,6-10
CONCLUSIONS
This report describes a method of utilizing asymmetric connectors
to significantly improve the dynamic range of PLA, thereby
addressing a major limitation of the assay. Sensitivity, LOD, and
maximum S/BG were also improved. By enhancing its already
successful predecessor, this asymmetric PLA method approaches
an ideal assay system, providing a homogeneous and highly
specific protein assay with the capability to detect very small
quantities of protein in free solution, at high throughput, and with
good dynamic range.
Furthermore, an experimental model of proximity hybridization
was successfully developed and tested. This model system
ultimately improved our understanding of the proximity effect and
its inherent cooperative assembly in PLA, highlighting the
importance of the interplay between connector-to-aptamer affinities
and aptamer-to-target affinities in maximizing S/BG and dynamic
range. Generally speaking, low connector affinities can be paired
with high probe affinities (aptamers or antibodies) to achieve this
goal.
ACKNOWLEDGMENT
Support for this work was provided by the Auburn University
Department of Chemistry and Biochemistry. The authors thank
Professor Douglas C. Goodwin for use of his PCR and gel
preparation equipment, as well as Daniel C. Leslie for helpful
discussions about proximity ligation and ligation background.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 14, 2010. Accepted July 8, 2010.
AC101762M

Anal. Chem. 2010, 82, 6983–6990
Improving Pressure Robustness, Reliability, and
Versatility of Solenoid-Pump Flow Systems Using a
Miniature Economic Control Unit Including Two
Simple Pressure Pulse Mathematical Models
Burkhard Horstkotte, † Erich Ledesma, ‡ Carlos M. Duarte, † and Víctor Cerdà* ,‡
Department of Global Change Research, IMEDEA (CSIC-UIB) Institut Mediterráni d’Estudis Avançats, Miquel
Marques 21, 07190 Esporles, Spain, and University of the Balearic Islands, Department of Chemistry, Carreterra de
Valldemossa km 7, 5, 07011 Palma de Mallorca, Spain
In this work we have systematically studied the behavior
of solenoid pumps (SMP) as a function of flow rate and
flow resistance. Using a new, economic, and miniature
control unit, we achieved improvements of the systems
versatility, transportability, and pressure robustness. A
further important improvement with respect to pressure
resistance was achieved when a flexible pumping tube was
inserted between the solenoid pump and the flow resistance
acting as a pressure reservoir and pulsation damper.
The experimental data were compared with two pressure
pulse models for SMP, which were developed during this
work and which were well-suited to describe the SMP
operation.
Solenoid micropumps (SMP) have gained considerable importance
as liquid drivers in analytical flow techniques (FT) since
their first application in flow systems reported by Weeks and
Johnson. 1 The initial motivation for the use of SMP was miniaturization
of field applicable FT systems.
SMP present an economic alterative to (multi)syringe and
peristaltic pumps, typically used for flow injection analysis (FIA) 2
and sequential injection analysis (SIA) 3 systems, respectively.
They provide a semicontinuous flow with highly pronounced
pulsation. It has been considered that this pulsation causes
intermediate turbulent conditions in the flow manifold improving
mixing of the sample and reagents in comparison with other
unsegmented FT, where laminar flow conditions are typical.
While in the first works, this feature was considered as a
disadvantage due to the oscillations of the detector signal, Lapa et
al. proved first that mixing of reagent and sample is enhanced and
sensitivity improvements of 50% are feasible. 4 The authors further
established the denomination “multipumping flow systems” (MPFS).
In contrast to syringe pumps, SMP operate continuously, which
bears the potential to shorten the time of analysis or to perform
* To whom correspondence should be addressed. Phone: +34 971 173 261.
Fax: +34 971 173 426. E-mail: Victor.Cerda@uib.es.
† IMEDEA (CSIC-UIB) Institut Mediterráni d’Estudis Avançats.
‡ University of the Balearic Islands.
(1) Weeks, D. A.; Johnson, K. S. Anal. Chem. 1996, 68, 2717–2719.
(2) Ruzicka, J.; Hansen, E. H. Anal. Chim. Acta 1975, 78, 145–157.
(3) Ruzicka, J.; Marshall, G. Anal. Chim. Acta 1990, 237, 329–343.
(4) Lapa, R. A. S.; Lima, J. L. F. C.; Reis, B. F.; Santos, J. L. M.; Zagatto, E. A. G.
Anal. Chim. Acta 2002, 466, 125–132.
analyte from large sample volumes. 5,6 In contrast to and in
advantage over peristaltic pumps, in MPFS, each flow channel
can be controlled individually. Both FIA and SIA manifold
configurations have been successfully applied so far using SMP,
i.e., confluent, codirectional flows, and contra-directional (aspiration
and dispense), respectively. 4,7 Analytical applications of MPFS
have been reviewed in detail elsewhere. 8-10
However, SMP show two considerable shortcomings: they do
not work reliable in the presence of gas bubbles and particulate
matter affecting both the actuation of the build-in check valves
and they are notably liable to flow backpressure leading to a
decrease of the effective flow with increasing flow resistance.
Although there is an obvious interest in the improvement of the
pressure robustness and flow rate reliability, up to date, there have
been hardly any efforts to complete this objective. In this work,
we studied the pressure robustness of the SMP as a function of
backpressure including two strategies to improve of the pressure
robustness. These have been the adaptation of the activation time,
made possible by the use a versatile and highly economic relay
card and software control, and the use of inflatable pumping tubes
to decrease the peak backpressure. We further present for the
first time simple models of the pressure and flow pulse behavior
as a function of pump dimensions and manifold characteristics,
which were capable of explaining the observations made from the
experiments undertake .
MATERIALS AND METHODS
Distilled water was used throughout. For the evaluation of flow
rates, pumped volumes where quantified by weighing on an
analytical balance, including compensation of the water density
at ambient temperature.
The self-priming SMP from Bio-Chem Fluidics (Boston, NJ)
of nominal 8 µL, 40 µL, and two of 25 µL volumes were used (types
(5) Pons, C.; Santos, J. L. M.; Lima, J. L. F. C.; Forteza, R.; Cerdà, V. Microchim.
Acta 2008, 161, 73–79.
(6) Pons, C.; Forteza, R.; Cerdà, V. Anal. Chim. Acta 2005, 550, 33–39.
(7) Pinto, P. C. A. G.; Saraiva, M.L.M.F.S.; Santos, J. L. M.; Lima, J. L. F. C.
Anal. Chim. Acta 2005, 539, 173–179.
(8) Lima, J. L. F. C.; Santos, J. L. M.; Dias, A. C. B.; Ribeiro, M. F. T.; Zagatto,
E. A. G. Talanta 2004, 64, 1091–1098.
(9) Rocha, F. R. P.; Reis, B. F.; Zagatto, E. A. G.; Lima, J. L. F. C.; Lapa, R. A. S.;
Santos, J. L. M. Anal. Chim. Acta 2002, 468, 119–131.
(10) Santos, J. L. M.; Ribeiro, M. F. T.; Dias, A. C. B.; Lima, J. L. F. C.; Zagatto,
E. E. A. Anal. Chim. Acta 2007, 600, 21–28.
10.1021/ac101250h © 2010 American Chemical Society 6983
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/20/2010

Figure 1. Scheme of the operation of the used solenoid micropumps in (A) activation (aspiration) and (B) deactivation (expulsion). Elements:
1, pretension spring; 2, solenoid; 3, spring; 4, inlet check valve; 5, outlet check valve; 6, metal core of solenoid; 7, membrane; 8, inner volume
of pump; 9, inlet; 10, outlet.
P/N090SP-12-8, P/N110TP-12-40, and P/N120SP-12-25) and are
denoted as pumps 1, 4, 2, and 3, respectively. In the activation
state, the SMP operate in suction whereas at deactivation the pulse
volume is expelled in the forward direction by the pressure
exposed by a metal spring as shown schematically in Figure 1.
The product specifications indicate a pressure height of 1.3 m for
aspiration and 3.5 m for dispensing. Detailed specifications cited
in this article can be downloaded at the producer’s Web site. 11
For control and powering, an eight channel relay card
(SERDIO8R) from EasyDAC (Glasgow, United Kingdom) was
used. It allowed remote and independent control of up to eight
solenoids (SMP or valves) via an RS232 serial interface. Connection
of the solenoid and powering is accomplished via screw
terminals, and no further equipment was required for operation.
For reliable operation, parallel connection of a diode to the
connected solenoid devices in reverse direction is required to
deduct the induced counter-voltage at the solenoid release. The
relay card is highly appropriate for miniaturization of MPFS as
the dimensions are about 11 cm × 10 cm × 3 cm. A thorough
description can be downloaded from the manufacture’s Web site.
Three PTFE tubes of different lengths (L) and inner diameters
(i.d.) were used as model flow resistances (tube 1, 49 cm, 0.2
mm i.d.; tube 2, 30 cm, 0.5 mm i.d.; and tube 3, 175 cm, 0.8 mm
i.d.). For calibration of the effective pressure at a given flow rate,
a Bu4S syringe pump module 12,13 (16.000 steps, 24-1024 s for
total dispense) from Crison Instruments S.A. (Alella, Barcelona,
Spain) was used, equipped with a glass syringe of 5 mL from
Hamilton Bonaduz AG (Bonaduz, Switzerland). Pressure measurements
were performed using a Bourdon pressure gauge of 6 bar
working range connected by a peek tube (20 cm, 0.8 mm i.d.)
and a three-way connector between the solenoid pump and the
pressure resistance. For testing the potential of pressure pulse
dampers, two pieces of purple/black marked peristaltic pumping
tube (2.05 mm i.d.) of 30 and 6.5 cm effective length were used.
(11) http://www.biochemfluidics.com/pdf/micro-pumps_brochure_dn_ibpmp-
01_r1.pdf, accessed February 20, 2010.
(12) Cerda, V.; Estela, J. M.; Forteza, R.; Cladera, A.; Becerra, E.; Altimira, P.;
Sitjar, P. Talanta 1999, 50, 695–705.
(13) Horstkotte, B.; Elsholz, O.; Cerdá, V. J. Flow Injection Anal. 2005, 22,
99–109.
6984 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Remote software control of the relay card as well as of the
syringe pump was done via an RS232C serial interface using the
AutoAnalysis 5.0 platform 14 from Sciware (Palma de Mallorca,
Spain) including specific dynamic link libraries (DLL) for each
communication port and connected instrument.
The DLL made for the used relay card was adapted for both
SMP and three-way solenoid valves. It included a calibration mode
for connected SMP, definition of the mean pulse volume and
minimal activation time for each relay, and the selection and
definition of two of the three operation parameters (flow rate,
dispense volume, and operation time) for each step and for each
individual SMP. The control panels corresponding to configuration
and operation instructions are shown in Figure 2, respectively.
The DLL is commercially available from Sciware SL.
RESULTS AND DISCUSSION
Calibration of the Flow Resistances. For calibration of flow
resistances, a smooth, nonpulsed flow was required. Therefore,
a syringe pump was used. The obtained data are given in Table
1. The relation between flow rate and pressure was linear, and
effective flow resistances of 0.66, 0.062, and 0.017 bar min mL -1
were found corresponding to hydrodynamic diameters of tubes
of 0.27, 0.42, and 0.91 mm i.d.
3.2. Calibration of the Solenoid Micropumps. For calibration,
the volume of water resulting from 240 pump pulses was
weighted in triplicate using an activation frequency of 2 Hz and,
for both the in- and outflow of the SMP, PTFE tubes of 20 cm
length and 1.5 mm i.d., respectively, which corresponded to
neglectible flow resistances.
The activation time was varied between 50 and 400 ms. Longer
activation times were not feasible due to low operation reliability
and excessive heating of the solenoid. The results and mean pulse
volumes are represented in Figure 3.
It was found that the mean pulse volume was approximately
stable for activation times between 150 and 300 ms and exceeded
in every case the nominal volume given by the producer: by 12%
(pump 1), 21% (pump 2), 40% (pump 2), and 120% (pump 4) (P/
(14) Becerra, E.; Cladera, A.; Cerda, V. Lab. Robotics Autom. 1999, 58, 131–
140.

Figure 2. Configuration editor window (above) and instruction command window (below) for the software control of the used relay card using
the program AutoAnalysis.
Table 1. Data from Calibration of Model Flow
Resistances
flow rate
[mL/min]
tube 1
(49 cm,
0.2 mm i.d.)
pressure [bar]
tube 2
(30 cm,
0.5 mm i.d.)
tube 3
(175 cm,
0.8 mm i.d.)
0.50 0.30
0.60 0.35
0.75 0.47
1.00 0.70
1.50 1.00
2.50 0.15
3.75 0.22
5.00 0.32 0.08
6.00 0.40 0.10
7.50 0.50 0.14
pressure/flow rate 0.659 0.0651 0.0175
[bar min mL -1 ]
hydrodynamic diameter
[mm]
0.27 0.42 0.91
N110TP-12-40). For pumps 2 and 3 (type P/N120SP-12-25), the
same behavior of decreasing pulse volume with increasing
activation time was found, while for pumps 1 and 4, the expulsion
volume increased with the activation time. This was reduced to
different responses of the rubber two check valves (see Figure
1). Because of slight differences in fabrication and abrasion, the
characteristics of the pump are assumed to be determined by the
weaker check valve.
For pump 4, activation times below 50 ms or above 400 ms
disabled pumping operation. Activation times above 250 ms lead
to notable pump heating, leading to higher pulse volumes.
3.3. Pressure Pulse Model 1. To simulate the pressure
pulse, a numerical flow model was created. The inner volume of
the SMP (see Figure 1) can be described best by a cone, where
the base is the cross section area of the inner cavity of the pump
and the surface shell is the membrane elevated in its central point
by attraction of the solenoid. The inner volume is therefore
calculated by eq 1, where h i is the lift of the membrane at time
i.
Vin,i ) 1
3 r 2
pump πhi (1)
At the release of the solenoid, the pressure is equal to the spring
force F applied on the inner cross section area A of the pump.
The spring force is a product of the spring constant D and the
sum of the initial lift h0 and the pretension h′ according eq 2.
The initial lift was calculated from the calibrated pulse volume of
the pump by transformation of eq 1.
p 0 ) F spring
A pump
) D(h0 + h′)
)
2
π
r pump
D( 3Vpump + h′)
2
rpump π
2
rpump π
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(2)
6985

Figure 3. Dependency of mean pulse volume on the activation time (240 pulses at 2 Hz, 3-fold repetition).
The initial pressure leads to an outflow Q through the connected
flow resistance characterized by tube length L and inner radius r
and liquid viscosity η at a given temperature given by eq
3(Hagen-Poiseuille equation). 15
Q i ) p i r4 π
8ηL
The outflow leads to a spring and pressure release. The full release
is reached when hi equals 0. The membrane position and
pressure after one time interval ∆t are described by eqs 4 and
2, respectively. Here, we applied a time interval ∆t of 10 ms for
modeling.
hi )hi-1 - 1 Qi∆t 3 2
rpump π
To describe the flow conditions, the Reynold number was
calculated further being the product of flow rate, radius of the
flow resistance tube, liquid density, and dynamic viscosity. 15
Re i ) Q i r2F
η
3.4. Application of Solenoid Valves to Model Flow Resistances.
The model flow resistances (tubes 1-3) were tested
on the SMP for different activation times. The found mean pulse
volume after expulsion of a theoretical volume of 2 mL was set in
relation to the maximal found pulse volume during the former
calibration, denoted further as operation efficiency (OE). The
results are shown in Figure 4A-D.
(15) Glück, B. Hydrodynamische und gasdynamische Rohrströmung; Druckverluste;
Bausteine der Heizungstechnik; VEB Verlag für Bauwesen: Berlin, Germany,
1998 (ISBN: 3-345-00222-1).
6986 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(3)
(4)
(5)
With the use of an activation time of 200 ms, it was found that
even for the lowest flow resistance (tube 3) the OE were
significantly reduced and that efficiency loss increased with the
pump size or pulse volume, respectively. In fact, the found pulse
volumes fitted approximately with the nominal pulse volume of
the SMP given by the producer.
For the lowest resistance, OE was about 90% (same range as
the former reported 1 ) for pumps 1, 2, and 3 and about 40% for
pump 4. For tube 3, OE values further showed to be independent
from the flow rate, while for higher flow resistances, more
pronounced for tube 1 than for tube 2, the OE values decreased
considerably with higher flow rate. The behavior of pumps 2 and
3 were found to be similar when the pulsation frequency was used
as an independent variable of data representation.
Decreasing the activation time from 200 to 50 ms but maintaining
the same operation frequency improved the OE for pumps 2
and 3 for flow resistance tube 1 up to 15%. A similar behavior was
found for pump 1 at operation frequencies >1.5 Hz. This improvement
was reduced to the prolongation of the deactivation time
and in consequence of outflow progress.
In contrast, the OE of pump 4 decreased mainly with the flow
resistance with minimal effect on the flow rate. This was explained
by the high volume of the SMP requiring deactivation times of
several seconds for full expulsion. Leakage of the inlet SMP check
valve has to be considered further. Such leakage would lead to a
continuous loss over the entire pressure exhibition time.
The observations indicate that the mean flow rate is not an
appropriate parameter to describe or calculate the backpressure
exposed by the flow resistances but the pulse volume and
pulsation frequency are the main parameters affecting the flow
rate. Pons et al. 6 recommended “periodic recalibration of the
micropumps”. From the results presented here, it becomes
clear that reliable off-system calibration of the SMP is not
possible but has to be done within the flow system and at the

Figure 4. Dependencies between volumetric efficiency and mean flow
rate for pump 1 (A), 2 (B), 3 (C), and 4 (D) for different activation times
and flow resistances. In part D, the effect of peristaltic pumping tubes
used as a pulsation damper on pressure robustness is shown further.
aimed operation frequency leading otherwise to considerable
volumetric errors. Following this rule, we did not observe any
significant alterations of the pulse volume during this work.
The pressure pulse of pump 3 was simulated. The constant of
the integrated spring was determined by the force exerted on the
plate of an analytical balance on compression. The spring tension
length was measured simultaneously with the depth probe of a
vernier caliber. The spring constant was determined to by about
2500 N/m. Further geometric dimensions measured by the
vernier caliber were a prestressing of 4 mm and an inner diameter
of 14 mm. The initial lift h0 was then calculated from the pulse
volume of 40 µL using eq 1.
Simulations of the pressure pulse release made with model 1
are shown in Figure 5 for pump 3 neglecting any elasticity of the
flow resistance tubes. As expected, the pressure release follows
a first-order lag (PT1) behavior. Since the lion share of the spring
force is the result of the pretension, the pressure decrease over
the pulse duration is low from 0.8 to 0.67 bar. This fact is
fundamental for reliable SMP operation, and a simple way to
improve the SMP pressure stability is to enhance the pretension
and, if required, the operation voltage. However, break-through
of the check valves and excessive solenoid heating probably limits
this modification.
From Figure 5C it becomes clear why the OE of pump 3
decreases at frequencies above 0.5 Hz for tube 1. As a matter of
fact, the deactivation time is not sufficiently long for the flow out
of the entire pulse volume from the SMP. With the use of shorter
activation times, i.e., longer deactivation times while keeping the
operation frequency constant led therefore to improved operation
efficiencies.
As stated, the initial pressure of pump 3 was about 0.8 bar
while the producer guarantees only about 0.33 bar. To evaluate
the maximal pressure applicable by the SMP, they were directly
connected to the aneroid barometer, obtaining stable levels of 1.35
bar for pump 1, 1.55 bar for pumps 2 and 3, and 1.15 bar for pump
4, overcoming the stated pressure stability up to 4-fold but at the
expense of an OE of zero (dispense volume ) 0). The theoretical
pressures at the tested flow rates considering pulse-less flow are
given in Table 2. The OE started to decrease for theoretical
pressures higher than 0.5 bar. It should be pointed out that with
model 1 with pump 3, Reynold numbers higher than 2300 were
not obtained, indicating that the pulsated flow is still laminar,
which contradicts the generally stated turbulent flow conditions
for MPFS.
3.5. Pressure Pulse Model 2. Elasticity of the connected
tubing was neglected in pressure pulse model 1. In reality, tubing
elasticity leads to a faster pulse volume expulsion, faster pressure
decrease, and further delay behavior of both pressure and flow
rate in the downstream tube.
In section 3.6, experiments made with an elastic pumping tube
inserted between the SMP and the flow resistance will be
discussed. To describe the fluid mechanics, we included the
characteristics of an elastic pumping tube in a second pressure
pulse model.
If an elastic cavity is inflated, the wall tension will cause the
increase of the fluid pressure within. The ratio of volume increase
per pressure unit is denoted compliance and is, if simplified as a
linear behavior, given by eq 6.
C ) ∆V
∆p ) V i - V 0
p i - p 0
In consequence, the pressure in the tube ptube could be calculated
according to eq 7.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(6)
6987

ptube,i ) ∆V
C ) Vi - V0 C
Figure 5. Modeling of the pressure and flow pulse of pump 3 for the used
flow resistances (A) tube 1 (49 cm, 0.27 mm i.d.), (B) tube 2 (30 cm, 0.42
mm i.d.), (C) tube 3 (175 cm, 0.91 mm i.d.) with no tube elasticity.
6988 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(7)
The two flow rates out of the pump into the elastic tube QIn and
out of the elastic tube QOut through the flow resistance were
defined. QOut was calculated from the pressure in the elastic
tube and the connected flow resistance analogous to eq 3. QIn
was calculated according eq 8 distinguishing three cases. (1)
The SMP is already empty and consequently both ppump and QIn
are zero.
(2) The pressure in the tubing and the SMP are equal so that
the QIn is equal to QOut. (3) The ppump overcomes ptube so that
QIn is limited by the pressure difference and the flow resistance
of the outlet check valve defined by its passage radius rValve
and its length LValve according to eq 3 and its time response T
and QOut.
Qin,i )
ppump,i ) 0|) 0
{if ppump,i > ptube,i | ) (ppump,i - ptube,i )(1 - e t/T ) r }
4
Valve π
+ Qout,i 8ηLValve ppump,i ) ptube,i | ) Qtube,i (8)
The accumulating volumes of expulsion from the SMP into the
elastic tube VIn and from the compliance elastic tube through
the flow resistance VOut were calculated according to eq 9.
V i )Q i-1 ∆t + V i-1
The inner volume of the elastic tube was calculated by eq 10 being
a simple balance of the initial volume, the flow out QOut, and the
flow in QIn.
(9)
V tube,i )V tube,i-1 + Q In,i-1 ∆t - Q Out,i-1 ∆t (10)
3.6. Improvement of Pressure Resistance. Increasing the
elasticity of the flow resistance connected to the SMP allows a
fast flow out of the pulse volume. This decreases the pulsation
character of the flow and leads to a stabilized but lower pressure
in the system. Since the backpressure is proportional to the 2nd
power of the flow rate, smoothing the flow pulsations by an
integral element such as an elastic pumping tube could decrease
the peak flow rate through the flow resistance and improve the
OE.
Table 2. Theoretical Pressure for the Tested Mean
Flow Rates Using the Model Flow Resistances and
Found Dynamic Inner Diameters
flow rate
[mL/min]
tube 2
(0.27 mm
i.d. 49 cm)
pressure [bar]
tube 2
(0.42 mm
i.d. 30 cm)
tube 3
(0.91 mm
i.d. 175 cm)
0.2 0.13
0.3 0.19 0.02 0.01
0.5 0.31 0.03 0.01
0.8 0.50 0.05 0.01
1.0 0.63 0.07 0.02
1.5 0.94 0.10 0.03
2.0 1.25 0.13 0.03
3.0 1.88 0.20 0.05
5.0 0.33 0.09

Since the time for liquid expulsion increases with the pulse
volume of the used pump, the contribution of any existing leakages
through the inlet check valve of the SMP would increase as well.
Allowing a fast flow-out against an apparently lower flow resistance
can therefore increase the OE because after expulsion the
resistance of two check valves would have to be overcome to
enable a counter-directional flow. By this, the solenoid membrane
can reach its final position of the deactivated status in less time
and can aspirate a full pulse volume again. On the other hand,
another producer of equivalent solenoid micropumps stated that
“it is recommended to use hard tubing for piping” 16 since soft
tubing might absorb the pressure pulse and affect the OE of the
SMP.
For testing the potential of pressure pulse dampers, two pieces
of purple/black marked peristaltic pumping tube (2.05 mm i.d.)
of 30 and 6.5 cm effective length were used and inserted between
solenoid pump 4 and the model flow resistance tube 1. As shown
in Figure 4D, the insertion of the shorter peristaltic pumping tube
doubled the operation efficiency of the pump and a further
improvement of 10% was possible using the longer peristaltic
pumping tube. However, it was observed that the flow was delayed
and lasted several seconds after finalization of the pumping
operation.
The compliance of the elastic tube, i.e., the volumetric
expansions of the elastic tube in function of the inner pressure
was an experimental parameter. It was measured by connecting
the elastic tube and the former flow resistance tube 1 to the
syringe pump and applying different flow rates. When the flow
stops, the lasting flow out was collected and weighed. A linear
relationship between pressure, calculated from the applied flow
rates and resistance, and volume was found in the range of 0.3 to
1.2 bar, leading to a compliance of 1055 µL/bar per meter of the
elastic tube.
It was not possible to measure neither the response time of
the check valve nor the flow path dimensions during the flow
out. Values used for the flow model 2 were estimated to be 1
mm flow path length, 0.4 mm open diameter, and a delay time
of 50 ms. The values have influence on the flow out velocity
from the SMP into the elastic tube but little impact on the flowout
through the flow resistance and would not affect the
simulation of the contribution of the elastic tube of the OE of
the SMP significantly.
The simulation of four consecutive pressure pulses with a
frequency of 0.33 Hz is shown in Figure 6 for neglecting any
elasticity (A) and considering the 30 cm peristaltic pumping tube
(B) for pump 4 and the flow resistance tube 1 applying time steps
of 10 and 2 ms, respectively.
It becomes clear that a single pressure pulse is insufficient
to describe and explain why better OE was achieved using the
elastic tube. However, under repeated operation and allowing
a flow out after stopping the SMP operation renders higher
OE values. Considering further a leakage through the inlet
check valve or higher OE, the difference between both
operation modes (with and without the elastic tube) would even
become more pronounced.
(16) http://www.takasago-elec.co.jp/en/product/pump.html, accessed February
20, 2010.
Figure 6. Modeling of the pressure and flow pulse of pump 4 for
the used flow resistances (A) without elasticity included and (B) with
an elastic pumping tube of 30 cm estimating a delay time of 50 ms.
Further study would be required to optimize the dimensions
of peristaltic pumping tubes as a function of the pulsation volume.
In fact, we did not achieve better but worse operation efficiencies
for the other pumps using the elastic pumping tubes of the given
dimensions.
Model 2 still shows some shortcomings: the aspiration step of
the SMP was not included, and the flow resistances, delay times,
and leakage loss of the check valves were not quantifiable but
were estimated. However, both presented models were suited to
describe the observations of SMP operation and thus present the
first reported intent of SMP simulation.
CONCLUSIONS
Time adaptation and the rational use of a new economic relay
card for the improvement of the operation reliability of SMP was
presented. The OE being the ratio of the calibrated pulse volume
and the experimental one was studied with the dependence of
the connected flow resistance and activation time over a wide
range of operation frequency. For the first time, the operation of
SMP was simulated by pressure pulse models. The following
conclusions and recommendations for future works using SMP
can be made: (1) Activation and deactivation times should both
exceed 150 ms. (2) The pulse volume has to be calibrated in the
flow system at the desired flow rate since flow resistance and
operation frequency both affect the OE considerably. (3) The use
of SMP of small pulse volume at higher frequency promises higher
OE than lower operation frequency using SMP with a larger pulse
volume. (4) The pressure robustness of SMP passed the charac-
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6989

teristics of the producer at least twice. (5) The OE is mainly limited
by the required time of pulse volume expulsion. (6) The use of
elastic pumping tubes can render higher OE when the flow
resistance is high.
ACKNOWLEDGMENT
B.H. was funded by a JAE postdoctoral fellowship from CSIC.
The work was supported from Project CTQ2007-64331 funded by
the MEC (Spanish Ministry of Education and Science).
6990 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
SUPPORTING INFORMATION AVAILABLE
Photo of the used 8-port relay card and scheme of the experiment
for the determination of the back-spring constant. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 17, 2010. Accepted June 28,
2010.
AC101250H

Anal. Chem. 2010, 82, 6991–6999
Integrated Microfluidic System for Rapid Forensic
DNA Analysis: Sample Collection to DNA Profile
Andrew J. Hopwood,* ,† Cedric Hurth, ‡ Jianing Yang, ‡ Zhi Cai, ‡ Nina Moran, † John G. Lee-Edghill, †
Alan Nordquist, ‡ Ralf Lenigk, ‡ Matthew D. Estes, ‡ John P. Haley, † Colin R. McAlister, †
Xiaojia Chen, ‡ Carla Brooks, ‡ Stan Smith, ‡ Keith Elliott, † Pieris Koumi, † Frederic Zenhausern,* ,‡
and Gillian Tully †
Research and Development, Forensic Science Service, Trident Court 2960 Solihull Parkway, Birmingham Business
Park, Birmingham UK B37 7YN, and Center for Applied NanoBioscience and Medicine, The University of Arizona
College of Medicine, 425 N. Fifth Street, Phoenix, Arizona 85004
We demonstrate a conduit for the delivery of a step change
in the DNA analysis process: A fully integrated instrument
for the analysis of multiplex short tandem repeat DNA
profiles from reference buccal samples is described and
is suitable for the processing of such samples within a
forensic environment such as a police custody suite or
booking office. The instrument is loaded with a DNA
processing cartridge which incorporates on-board pumps
and valves which direct the delivery of sample and
reagents to the various reaction chambers to allow DNA
purification, amplification of the DNA by PCR, and collection
of the amplified product for delivery to an integral
CE chip. The fluorescently labeled product is separated
using micro capillary electrophoresis with a resolution of
1.2 base pairs (bp) allowing laser induced fluorescencebased
detection of the amplified short tandem repeat
fragments and subsequent analysis of data to produce a
DNA profile which is compatible with the data format of
the UK DNA database. The entire process from taking the
sample from a suspect, to database compatible DNA
profile production can currently be achieved in less than
4 h. By integrating such an instrument and microfluidic
cartridge with the forensic process, we believe it will be
possible in the near future to process a DNA sample taken
from an individual in police custody and compare the
profile with the DNA profiles held on a DNA Database in
as little as 3 h.
DNA analysis in a forensic context relies on the extraction of
DNA from a sample; quantification and normalization of the DNA;
concurrent PCR-based amplification of a number of specific short
tandem repeat loci and separation/detection of the PCR products,
usually by capillary electrophoresis (CE); thereby allowing precise
sizing of each fragment and accurate typing of the alleles in the
DNA profile. This forensic process allows for the routine analysis
of DNA samples from controls such as buccal swabs in 24-72 h
* To whom correspondence should be addressed. E-mail: andy.hopwood@
fss.pnn.police.uk (A.H.); Frederic.Zenhausern@arizona.edu (F.Z.).
† Forensic Science Service.
‡ The University of Arizona College of Medicine.
from receipt of the sample. 1 Such processes are laboratory-based
and require that the sample, once taken from a suspect in custody
is transported to a laboratory for processing. For convenience,
samples are usually stored and batched by the police force prior
to dispatch to the lab, a process that can take between a few hours
to a few weeks. Even with samples that are dispatched to the
laboratory immediately, the suspect is highly likely to have been
released from custody while the sample is processed. A sample
taken from a suspect in the UK has a 2.3% chance of matching
with a crime sample held on the National DNA Database
(NDNAD) of the UK. 2 Furthermore, evidence suggests that
individuals released on police bail have a high incidence of
offending while on bail. 3 It would therefore be beneficial to the
arresting agency for the individual to remain in custody while the
DNA sample is processed and compared against the NDNAD.
This would require the availability of both rapid DNA processing
and real-time access to the national database.
Within the laboratories of the FSS, current process allows for
reference samples in urgent cases to be prioritized and, once
delivered to the laboratory, these can be processed manually in
as little as 8 h. However, this is a labor intensive and therefore
relatively expensive process. The implementation of a rapid system
whereby a reference sample can be processed within the police
custody area would be of value to the law enforcement community:
a suspect’s DNA sample could be processed and compared to a
database of crime sample DNA profiles while the individual
remains in custody, eliminating the need to locate and rearrest
an individual in response to a match on the database. Rapid
elimination of an individual from an investigation could also be
achieved, freeing up resources for the investigation of alternative
leads in a criminal case.
A number of publications have reported approaches to the
rapid analysis of DNA for forensic science and various methods
have been described including the acceleration of specific parts
(1) Hedman, J.; Albinsson, L.; Ansell, C.; Tapper, H.; Hansson, O.; Holgersson,
S.; Ansell, R. Forensic Sci. Int.: Genet. 2008, 2, 184–189.
(2) Annual Report of The National DNA Database 2007-2009. http://www.
npia.police.uk/en/docs/NDNAD07-09-LR.pdf (Accessed March 2010).
(3) Morgan, P. M.; Henderson, P. F. Home Office Report 184 1998 ISBN 1
84082 074 8 http://www.homeoffice.gov.uk/rds/pdfs/hors184.pdf (Accessed
Feb. 2010).
10.1021/ac101355r © 2010 American Chemical Society 6991
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/15/2010

of the process such as PCR 4-6 or the optimization of the whole
process which requires the intensive use of highly skilled
technicians. 7 The application of Hybeacon technology which allows
for a rapid analysis of STR loci from a crude lysate of buccal cells
without the need for physical separation of the alleles 8-10 has
shown some promise but has the drawback that the resolution of
microvariant alleles is difficult and the analysis of all required
alleles in single tube is not currently achievable. The development
of integrated microfluidic systems, so-called micro total analysis
systems, or microTAS for DNA analysis has been discussed for
many years and a number of groups have demonstrated successful
modules for DNA extraction, 11-14 PCR amplification, 15-18 and
CE. 19-23 Integrated systems for extraction and PCR 24-26 or PCR
and CE 27-29 have also been reported and illustrate that a fully
integrated DNA analysis system should be feasible for forensic
application. Indeed, full integration of all the constituent parts
required for genetic analysis including the application of capillary
(4) Vallone, P. M.; Hill, C. R.; Podini, D.; Butler, J. M. Forensic Sci. Int.: Genet.
Suppl. 2009, 2, 111–112.
(5) Vallone, P. M.; Hill, C. R.; Butler, J. M. Forensic Sci. Int.: Genet. 2008, 3,
42–45.
(6) Wang, D. Y.; Chang, C. W.; Hennessy, L. K. Forensic Sci. Int.: Genet. Suppl.
2009, 2, 115–116.
(7) Hopwood, A.; Fox, R.; Round, C.; Tsang, C.; Watson, S.; Rowlands, E.;
Titmus, A.; Lee-Edghill, J.; Cursiter, L.; Proudlock, J.; McTernan, C.; Grigg,
K.; Thornton, L.; Kimpton, C. Int. Congr. Ser. 2006, 1288, 639–641.
(8) French, D. J.; McDowell, D. G.; Thomson, J. A.; Brown, T.; Debenham,
P. G. Int. Congr. Ser. 2006, 1288, 707–709.
(9) Gale, N.; French, D. J.; Howard, R. L.; McDowell, D. G.; Debenham, P. G.;
Brown, T. Org. Biomol. Chem. 2000, 6, 4553–4559.
(10) French, D. J.; Howard, R. L.; Gale, N.; Brown, T.; McDowell, D. G.;
Debenham, P. G. Forensic Sci. Int.: Genet. 2008, 2, 333–339.
(11) Ji, H. M.; Samper, V.; Chen, Y.; Hui, W. C.; Lye, H. J.; Mustafa, F. B.; Lee,
A. C.; Cong, L.; Heng, C. K.; Lim, T. M. Sens. Actuators, A Phys. 2007,
139, 139–144.
(12) Wolfe, K. A.; Breadmore, M. C.; Ferrance, J. P.; Power, M. E.; Conroy,
J. F.; Norris, P. M.; Landers, J. P. Electrophoresis 2002, 23, 727–733.
(13) Breadmore, M. C.; Wolfe, K. A.; Arcibal, I. G.; Leung, W. K.; Dickson, D.;
Giordano, B. C.; Power, M. E.; Ferrance, J. P.; Feldman, S. H.; Norris, P. M.;
Landers, J. P. Anal. Chem. 2003, 75, 1880–1886.
(14) Oakley, J. A.; Shaw, K. J.; Docker, P. T.; Dyer, C. E.; Greenman, J.;
Greenway, G. M.; Haswell, S. J. Lab Chip 2009, 9, 1596–1600.
(15) Obeid, P. J.; Christopoulos, T. K.; Crabtree, H. J.; Backhouse, C. J. Anal.
Chem. 2003, 75, 288–295.
(16) Yang, J.; Liu, Y.; Rauch, C.; Stevens, R. L.; Liu, R. H.; Lenigk, R.; Grodzinski,
P. Lab Chip 2002, 2, 179–187.
(17) Giordano, B. C.; Ferrance, J.; Swedberg, S.; Hühmer, A. F. R.; Landers,
J. P. Anal. Biochem. 2001, 291, 124–132.
(18) Simpson, P. C.; Roach, D.; Woolley, A. T.; Thorsen, T.; Johnston, R.;
Sensabaugh, G. F.; Mathies, R. A. Proc. Natl. Acad. Sci. U.S.A. 1998, 95,
2256–2261.
(19) Medintz, I. L.; Paegel, B. M.; Mathies, R. A. J. Chromatogr., A 2001, 924,
265–270.
(20) Shi, Y.; Simpson, P. C.; Scherer, J. R.; Wexler, D.; Skibola, C.; Smith, M. T.;
Mathies, R. A. Anal. Chem. 1999, 71, 5354–5361.
(21) Shi, Y. Electrophoresis 2006, 27, 3703–3711.
(22) Zheng, J.; Webster, J. R.; Mastrangelo, C. H.; Ugaz, V.; Burns, M. A.; Burke,
D. T. Sens. Actuators, B 2007, 125, 343–351.
(23) Goedecke, N.; McKenna, B.; El-Difrawy, S.; Carey, L.; Matsudaira, P.; Erlich,
D. Electrophoresis 2004, 25, 1678–1686.
(24) Lee, C. Y.; Lee, G. B.; Lin, J. L.; Huang, F. C.; Liao, C. S. J. Micromech.
Microeng. 2005, 15, 1215–1223.
(25) Legendre, L. A.; Bienvenue, J. M.; Roper, M. G.; Ferrance, J. P.; Landers,
J. P. Anal. Chem. 2006, 78, 1444–1451.
(26) Bienvenue, J. M.; Legendre, L. A.; Ferrance, J. P.; Landers, J. P. Forensic
Sci. Int.: Genet. 2010, 4, 178–186.
(27) Woolley, A. T.; Hadley, D.; Landre, P.; deMello, A. J.; Mathies, R. A.;
Northrup, M. A. Anal. Chem. 1996, 68, 4081–4086.
(28) Lagally, E. T.; Medintz, I.; Mathies, R. A. Anal. Chem. 2001, 73, 565–570.
(29) Liu, P.; Yeung, S. H. I.; Crenshaw, K. A.; Crouse, C. A.; Scherer, J. A.;
Mathies, R. A. Forensic Sci. Int. Gen. 2008, 2, 301–309.
6992 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
electrophoresis has been demonstrated for the detection of
Bacillus anthracis from whole blood 30 but to date we are unaware
of any system which has been used to routinely produce an STR
profile from DNA extraction to DNA profile with no operator
intervention. Integrated microfluidic systems and their application
to high-performance genetic analysis has been recently reviewed 31
and will not be further discussed here.
Within the UK, 75% of individuals arrested are processed and
released from custody within 6 h and about 95% are processed
within 24 h. 32 We defined our target for the process to be capable
of processing the sample to give a DNA profile and returning any
match from a database within 2 h such that the sample could be
processed comfortably, and duplicated if required, within the 6 h
window that a suspect is present in the custody suite.
The fluidic cartridge-based system described here is built using
the principle of closed architecture, which minimizes any opportunity
for contamination of the samplesa critical requirement
for the forensic applicationsand performs DNA extraction, DNA
amplification, resolution of the STR alleles by CE and detection
using laser induced fluorescence (LIF). The polycarbonate cartridge
for DNA extraction, PCR and post PCR manipulation is
prefilled with the reagents required for the entire process and
simply clips into an adapter which also holds the glass micro
capillary electrophoresis (µCE) chip used to resolve and detect
the dye-labeled amplicons. The extraction-PCR cartridge is held
in contact with an electronic circuit board which provides the
functional control, allowing routine use by an unskilled technician
with the minimum amount of training. Fluidic movement is fully
automated and controlled using simple electro-chemical pumps
and single use thermally activated valves as previously described,
33,34 rather than more complex reusable thermally activated
valves 35 avoiding the need for complex fittings and fixings
to integrate the cartridge with the instrumentation, and eliminating
the potential for sample contamination between the different
functions of the cartridge. A servo-controlled magnet is activated
as required for the collection of magnetic particles for DNA
purification. Embedded resistive heaters are used to activate valves
as previously described, 34 and Peltier devices employed for
thermal cycling are programmed to activate in a specific sequence
to facilitate processing. A miniaturized high voltage power supply
is integrated to the hardware and allows for the separation of the
STR amplicons through a matrix of polyvinylpyrolidone (PVP) and
hydroxyethylcellulose (HEC). Data collected from the µCE is
processed manually using commercially available software and
the DNA profile is recorded in a format compatible with the data
requirement for submission to the National DNA Database. The
whole system is designed to allow simple loading of a crude cell
lysate from a buccal scrape to the cartridge and robust walk-away
(30) Easley, C.; Karlinsey, K.; Bienvenue, J.; Legendre, L.; Roper, M.; Feldman,
S.; Hughes, M.; Hewlett, E.; Merkel, T.; Ferrance, J. Proc. Natl. Acad. Sci.
U.S.A. 2006, 103, 19272–19277, Landers.
(31) Liu, P.; Mathies, R. A. Trends Biotechnol. 2009, 27, 572–581.
(32) Philips C. Her Majesty’s Stationary Office, 1981; ISBN 0101809212 ASIN:
B002K6GPAQ.
(33) Liu, R. H.; Bonanno, J.; Yang, J.; Lenigk, R.; Grodzinski, P. Sens. Actuators,
B 2004, 98, 328–336.
(34) Liu, R. H.; Yang, J.; Lenigk, R.; Bonanno, J.; Grodzinski, P. Anal. Chem.
2004, 76, 1824–1831.
(35) Pal, R.; Yang, M.; Johnson, B. N.; Burke, D. T.; Burns, M. A. Anal. Chem.
2004, 76, 3740–3748.

Figure 1. Representation of the Cartridge for DNA extraction,
amplification and post PCR denaturation showing P1-P4 ) electrochemical
pumps; C1 ) lysate input chamber; C2 ) bead chamber;
C3 ) mixing/incubation chamber; C4 ) washing and elution chamber;
R ) PCR chamber; A ) DNA extract archive chamber; D )
denaturation chamber; M ) bead storage chamber; B ) Binding buffer
chamber; W ) wash solution storage chamber; E ) elution buffer
storage chamber; F ) Formamide/ILS storage chamber; X ) output
to CE. A closing valve is represented by b; an opening valve by 0;
and a vent by 1.
processing of the sample culminating in an STR profile which is
compatible with The National DNA Database of the UK.
EXPERIMENTAL SECTION
Samples. Buccal samples were collected from consenting
individuals using an OMNISwab (Whatman, Maidstone, UK) and
processed immediately using the method described below.
Cartridge Design and Fabrication. The cartridge device for
DNA extraction, amplification and post PCR manipulation (Figure
1) was made from 3 mm computed numerically controlled (CNC)
machined polycarbonate (PC) plate stock and comprised a number
of chambers for the storage of reagents and sample manipulation.
The architecture of the fluidic system was designed such that
outlets from one chamber to the next were typically on the bottom
side of the chamber, thus taking advantage of gravity to ensure
that any bubbles generated were removed from the fluid within
the system. The channels linking the chambers were fitted with
opening valves (O) and closing valves (V) formed from paraffin
wax with a melting point of 65 °C (Sigma-Aldrich, Poole, UK) as
previously described 33 which were hot dispensed into their
respective positions. The exception to this was valve V15 which
comprised two chambers connected vertically which were filled
with different paraffin waxes; the lower chamber, closest to the
Table 1. Reagents and Volumes Required for Cartridge
Operation. All Reagents Were Added Prior to Loading
the Cartridge into the Instrument
chamber reagent volume required
pump 1-3 0.25 M NaCl 600 µL
pump 4 0.5 M NaCl 600 µL
M ChargeSwitch (CS) beads 15 µL
water 10 µL
B CS purification buffer 30 µL
W CS wash buffer 200 µL
E CS elution buffer 150 µL
C1 Buccal lysate 150 µL
R PowerPlex ESI 16 multimix
in ReaX beads
two beads
F Hi-Di formamide 45 µL
ILS 500 CC5 5 µL
Peltier was filled with H1 Sasol wax with a nominal melting
temperature of 105 °C (Sasol Wax North America Corp., Hayward,
CA), the second chamber contained the standard paraffin wax.
The H1 wax prevented V15 firing prematurely from exposure to
the heat generated during the thermal cycling of the Peltiers.
Aluminum wire electrodes were installed into the pump chambers,
sealed using Dymax 1180-M UV glue and cured using a Dymax
Bluewave 200 UV Curing System (Dymax, Torrington, CT) in
accordance with the manufacturer’s instructions. The PCR multimix
was formulated into a ReaX bead (Q Chip Ltd., Cardiff, UK)
and added to chamber R prior to binding the 0.5 mm PC cover to
the 3 mm cartridge using a two sided pressure sensitive adhesive
(90106 PSA, Adhesive Research, Glen Rock, PA) and pressing
the assembly at 122 psi for 20 s and then 245 psi for 20 s. The
ReaX bead-packaged multimix reagents are stable for at least 16
weeks when stored at 4 °C (data not shown). The assembled
cartridge was stored at 4 °C until required. Other reagents
required for sample processing were added to the appropriate
reservoirs immediately prior to use as detailed (Table 1). The
single-use cartridge was inserted into a holder in a vertical position
and the 140 mm reusable glass microCE chip was positioned in
a horizontal plane (Figure 2A). The two components were
connected via a 100 mm length of Teflon tubing with an internal
diameter of 200 µm which was manually connected to the output
channel of the plastic cartridge at X in Figure 1, and to the sample
well of the adapter on the electrophoresis chip (Figure 2B) by
sliding over short sections of PEEK tubing (IDEX Corporation,
Northbrook, IL) with an outside diameter of 0.78 mm (Figure 2C)
prior to loading into the instrument. The Teflon tubing was
renewed for each sample.
Electronic Control. The printed circuit board comprises
embedded resistive heaters and control circuitry for managing
the different operations required of the cartridge.
The instrument houses the electronic control systems, the
optical excitation and detection equipment and holds the printed
circuit board to which the cartridge is precisely aligned and
clamped, to allow robust connection to the electrodes for the
electrochemical pumps and surface contact for valve firing by
thermal transfer. The operation of the cartridge is as follows. The
cells on the buccal swab sample were directly lysed by adding 1
mL ChargeSwitch (Life Technologies, Inchinnen, UK) Lysis Buffer
and 10 µL ChargeSwitch Proteinase K solution (20 mg/mL in 50
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6993

Figure 2. Assembly of the polycarbonate sample preparation cartridge and the borosilicate glass microCE chip. (A) The PC cartridge is positioned
in a vertical plane, the glass CE chip in the horizontal plane. (B) Top image shows the adapter with wells for the electrodes. The PEEK tubing
inlet can be clearly seen entering the elongated well of the sample well no. 1. Bottom image shows the adapter lid with gold-plated electrodes.
(C) The adapter lid in position. The attached Teflon tubing can be seen attached to the PEEK inlet tube. (D) Representation of the adapters
fitted to the CE chip and layout of the CE channel in relation to the wells.
mM Tris-HCl, pH 8.5, 5 mM CaCl2, 50% glycerol) in a microcentrifuge
tube and incubating at 60 °C for 15 min.
An aliquot of 150 µL of the sample lysate was introduced into
chamber C1 in the analysis cartridge using a manual pipet. Pump
P1 was fired to pressurize the system and valve O1 was opened
to direct the lysate via the binding buffer chamber B containing
30 µL ChargeSwitch purification buffer, through the paramagnetic
bead storage chamber M containing 25 µL of ChargeSwitch
magnetic beads (1.87 mg/mL in 6 mM MES, pH 5.0, 6 mM NaCl)
and into chamber C2, where the initial sample mixing occurred.
The sample was then passed through two expansion chambers
into chamber C3, where complete sample mixing was ensured
by supplementary activation of the pump to produce a stream of
bubbles which passed through the sample. The sample was
incubated in this chamber for 3 min to facilitate DNA binding.
The sample was then directed into chamber C4 where the
ChargeSwitch magnetic beads were captured by a magnetic field
of 450 ± 100 gauss generated via a servo activated magnet
positioned behind the PCB. The supernatant liquid containing any
unbound DNA and other lysis products was directed to the Waste
chamber. The DNA-bound ChargeSwitch beads were then washed
by flushing the beads with 200 µL of ChargeSwitch wash buffer
from chamber W by activation of pump P2. One hundred and fifty
µL of ChargeSwitch elution buffer (E5) was then directed to
chamber C4 where the purified DNA was released from the beads
by incubating at 60 °C for 3 min with the magnetic field removed.
The beads were then recaptured by activation of the magnetic
field and the eluted DNA solution was directed to the archive
chamber A via the PCR chamber R by activation of pump P3.
6994 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
The PCR chamber was designed to capture a 10 µL total
volume of DNA solution and was preloaded with PowerPlex ESI
16 PCR amplification multimix (Promega, Madison, WI) packaged
in ReaX reagent beads. The PCR chamber was sealed from the
rest of the cartridge prior to amplification by firing valves V13
and V14.
Amplification of the 16 loci multiplex PCR system was
performed as follows: Activation of the PCR mix at 96 °C for 2
min followed by 27 cycles of 94 °C for 30 s, 59 °C for 120 s, 72 °C
for 90 s, and a final incubation at 60 °C for 45 min.
The valves V13 and V14 were then reopened by heating and
the liquid wax pushed through to the paraffin trap chamber O12,
and valve V15 activated to close the PCR chamber bypass channel.
The PCR solution was then collected by flowing 45 µL Hi-Di
formamide (Applied Biosystems, Warrington, UK) containing 5
µL internal size standard (ILS 500 CC5, Promega) from the
formamide reservoir F using pump P4. The formamide/ILS/
sample was collected in the denaturation chamber and incubated
at 95 °C for 3 min prior to pumping to the CE chip.
The CE microchip was custom-made from Schott Borofloat
borosilicate glass (Micronit Microfluidics BV, Enschede, The
Netherlands). The electrophoresis channel had a semielliptic
cross-section and measured 50 µm wide and 20 µm deep. A
machined polycarbonate adapter was interfaced with the powderblasted
access holes on the microchip using UV-curable glue to
provide buffer wells that could contain a maximum volume of 35
µL, and a PEEK attachment point for the Teflon tubing from the
cartridge performing DNA extraction and amplification. A pair of
polycarbonate plates, one with three 1 mm diameter gold-coated

Table 2. Capillary Electrophoresis Injection and
Separation Conditions. The sample is Presented to the
Sample Well No. 1 and Injected into the Separation
Column by a Gated Injection
sample well
no. 1 (V)
sample
waste well
no. 2 (V)
buffer well
no. 3 (V)
buffer
waste well
no. 4 (V) time (s)
loading ground 800 ground 600 40
injection ground 400 350 2000 2
separation ground 400 ground 2700 1000
pins (Figure 2B), one with a single pin were designed as adapter
lids to clip onto the adapters to close the fluidic system (Figure
2C) and connect to the control system. The reusable CE chip was
prepared by flushing with water to remove the polymer from the
previous run, followed by 1 M hydrochloric acid for 10 min and
rinsing with water for 10 min to remove the acid prior to loading
the polymer. The polymer, 3.5% w/v PVP/HEC in a 20/80 ratio
in 1 × ABI 310 running buffer (Applied Biosystems) acted as both
sieving and coating matrix 36 and was loaded for 25 min at 50 °C.
Owing to the design of the cartridge system, the risk of
contamination of the subsequent sample with PCR products from
the CE chip was minimized by loading the reagents for the plastic
cartridge before connecting it to the CE chip. A volume of 25 µL
of 1 × ABI 310 running buffer was added to each of the wells of
the electrophoresis chip prior to clipping the adapter lids in place.
The sample was delivered into the sample well and injected into
the separation channel using a gated injection scheme 37 prior to
separation under the conditions described in Table 2 and the
labeled fragments were detected at a point 110 mm from injection.
The electrophoresis chip was maintained at a temperature of 50
± 0.15 °C through direct contact with a heater plate.
Laser-Induced Fluorescence (LIF) Detection. LIF detection
was achieved with a modified confocal fluorescence setup. The
excitation source, a diode-pumped solid-state (DPSS) laser (Calypso,
Cobolt AB, Solna, Sweden) was cleaned-up using a 480-520
nm band-pass filter (Omega Optical, Brattleboro, VT) and aligned
onto a high-reflectivity elliptic mirror centered on a Raman laser
reflector. The laser output was adjusted to provide 21 mW of
power at 491 nm to the CE microchip and the incident light was
focused onto the CE microchip using a ×40 objective (LucPLFLN,
Olympus, Center Valley, PA), such that a circular 60 µm Gaussian
beam hit the microchannel at a point 110 mm from the injection
point. Alignment and focusing of the chip and detection system
was achieved by a push-button activated servo motor and a
micrometer screw respectively, on a 3 mm translation stage
(Thorlabs, Newton, NJ).
The emission wavelengths of the fluorescently labeled DNA
fragments were collected onto a cooled charge-coupled device
(CCD) camera (Newton 920, Andor, South Windsor, CT) mounted
onto a high-throughput spectrometer (CP140-1605, Horiba Scientific,
Newton, NJ) after a long-pass emission filter (cut-on: 520
nm). The diffraction grating in the spectrometer disperses the
different wavelengths in the emission spectrum over the length
of the CCD, allowing a quantitative measurement of each color
(36) Boulos, S.; Cabrices, O.; Blas, M.; McCord, B. R. Electrophoresis 2008,
29, 4695–4708.
(37) Ermakov, S. V.; Jacobson, S. C.; Ramsey, J. M. Anal. Chem. 2000, 72,
3512–3517.
in the sample over time. The instrument design is further
described in Hurth et al. 38
The raw spectral data from the CCD was deconvoluted to
provide peak information for each of the five colors used in the
multiplex analysis using NanoIdentity v 1.12 (SoftGenetics, State
College, PA) and the data were further analyzed to provide
genotype information using GeneMarker HID V 1.76 (Soft-
Genetics).
During optimization and testing of protocols, DNA quantification
of the DNA extracted from the samples was performed using
the Quantifiler Human DNA Quantification Kit (Applied Biosystems)
in accordance with the manufacturer’s instructions.
Electrophoresis of control samples was executed using the
Applied Biosystems 3130xl in accordance with the PowerPlex ESI
16 kit instructions.
RESULTS AND DISCUSSION
The aim of the process is to generate a DNA profile with all
16 loci present in the electropherogram. To ensure this happens,
a number of variables were identified as critical to the efficiency
of the process: Sample collection; DNA extraction; PCR efficiency;
PCR reaction recovery and robustness of the CE detection
instrumentation. Each of these variables have been investigated
and stabilized as far as possible.
The sample input is the major cause of variation. Different
samples taken from different individuals will have different
quantities of cells, releasing different quantities of DNA upon lysis.
For example, van Wieren-de Wijer et al observed total DNA yield
ranging from 0.08 to 1078.0 µg (median 54.3 µg; mean 82.2 µg). 39
Since there is no simple way to normalize the DNA concentration
on the cartridge by dilution, it was essential to minimize this
variation in DNA concentration extracted from the buccal swab
by some means. The ChargeSwitch gDNA Normalized Buccal Cell
Kit is claimed to produce a normalized yield of genomic DNA at
a concentration of 1-3 ng/µL in150µL under standard conditions.
40 Compared with other available DNA extraction chemistries
the ChargeSwitch protocol is comparatively simple and the
reagents were considered less likely to cause inhibition of the
downstream processes. In our experience, with samples of freshly
collected swabs extracted on cartridge as described above, a DNA
concentration of between 0.47 and 1.92 ng/µL of DNA with a mean
of 0.86 ± 0.41 ng/µL (n ) 27) was routinely attained.
The amount of DNA produced by PCR amplification varied
with a number of factors including the amount of DNA in the
reaction, the total volume of the PCR and the number of cycles
of PCR. The volume of DNA extract added to the reaction was
governed by the size of the PCR chamber and the volume of the
ReaX beads therein. The PCR chamber (Figure 3) was designed
such that it is filled by the DNA solution traveling from the
chamber C4 to the archive chamber A and the volume of the
(38) Hurth, C.; Smith, S.; Nordquist, A.; Lenigk, R.; Surve, A.; Hopwood, A.;
Haley, J.; Chen, X.; Estes, M.; Yang, J.; Cai, Z.; Lee-Edghill, J.; Moran, N.;
Elliott, K.; Tully, G.; Zenhausern, F. Rev. Sci. Instrum. 2010, Submitted.
(39) van Wieren-de Wijer, D. B. M. A.; Maitland-van der Zee, A. H.; de Boer, A.;
Belitser, S. V.; Kroon, A. A.; de Leeuw, P. W.; Schiffers, P.; Janssen,
R. G. J. H.; van Duijn, C. M.; Stricker, B. H. C. H.; Klungel, O. H. Eur. J.
Epidemiol. 2009, 24, 677–682.
(40) ChargeSwitch® gDNA Buccal Cell Kits. Instruction Manual. Invitrogen Life
Technologies. http://tools.invitrogen.com/content/sfs/manuals/chargeswitch_
gdna_buccal_man.pdf (Accessed March 2010).
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6995

Figure 3. Design of the PCR chamber fluidic system. The DNA
solution enters the chamber from the right and fills the chamber. When
the liquid level reaches the constriction channel in the top left-hand
side of the chamber, the pressure required to pass through this
constriction is much greater than that required for the solution to
traverse the bypass. The DNA solution therefore passes over the
bridge, leaving a precise volume of DNA in the PCR chamber
(chamber highlighted by red solution). Following PCR amplification,
the reaction volume is recovered by closing the bridge at Y using
valve V15 and flushing the formamide through the chamber, collecting
the PCR product.
chamber was filled until the solution reached the restriction
channel. The pressure required to breach this channel was greater
than that required for the solution to flow over the bypass and
consequently the flow took that direction, leaving a precisely
measured volume of PCR reaction (9.96 ± 0.21 µL (n ) 20)) in
the PCR chamber. The amount of DNA in the 10 µL PCR chamber
therefore approximated 5.2 ± 2.5 ng, based upon the ReaX bead
volume of 4 µL.
Samples recovered from the PCR chamber and run on the AB
3130xl gave similar peak heights to the control samples amplified
in tube (data not shown).
Following amplification, V15 is closed and the whole of the
PCR sample is recovered from the PCR chamber and delivered
with ILS 500 CC5 in formamide to the loading well of the
electrophoresis chip. We rely on surface effects to smoothly
remove the vast majority of the PCR reaction from the chamber.
As the formamide is driven from its storage chamber, the air in
the channel in front of the formamide is pushed through the PCR
chamber and removes the bulk (80-90%) of the PCR reaction
through the constriction. The formamide then fills the chamber,
taking the remaining PCR solution with it. The surface tension at
this small scale ensures that the bulk of the formamide is also
flushed from the chamber leaving just one or two microlitres in
the 90° corners of PCR chamber. The volume delivered to
chamber D varied between cartridges with a mean volume of 25.1
± 4.8 µL (n ) 25) with a range of 15-35 µL. The variation in
volume was due to three things: (1) The wax in the valves 13 and
14 is molten as the liquid flows by, allowing some liquid to mix
with the wax. This liquid is trapped in the system as the wax
solidifies. (2) Liquid is occasionally lost in the valve structures of
V13 and/or V14. (3) Our channels are machined rather than
injection molded, creating a rough surface that can retain 1-2
µL per inch of channel. The volume of sample released from PCR,
denatured in chamber D and delivered to the µCE is a variable
that could have a significant effect on the final result: If only half
of the PCR volume of a given PCR product was recovered we
would expect to observe lower peak heights in the profile. The
6996 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
system was therefore designed to push the PCR mixture out of
the PCR chamber prior to flushing the chamber with formamide
and we believe that the majority of the PCR product is recovered,
minimizing any variability from this function of the cartridge.
Sample loss in the channels, wax and valve housings in the circuit
before the PCR chamber will reduce the amount of formamide
and size standard reaching the denaturation chamber while the
inefficiencies in the circuitry post PCR chamber would reduce
the volume of PCR product and formamide in proportion to their
volumetric ratios. Current laboratory practice is that the sample
is heat denatured at 95 °C for 3 min and snap-cooled on ice prior
to loading onto the 3130xl platform for injection to the CE. While
the sample was easily heat denatured on the cartridge, no facility
for snap cooling was available and the sample was loaded directly
into the CE. Evaluation of the profiles demonstrated that this
process was acceptable because the DNA profiles produced with
no snap cooling were found to be comparable to those that had
been snap cooled. The electropherogram from the internal lane
standard (GeneScan 500 LIZ) diluted to a volume ratio of 9:1 with
Hi-Di formamide (Applied Biosystems) and heat-denatured at 95
°C for 5 min was used to assess the separation efficiency of the
µCE module. The chromatographic resolution R was calculated
by fitting the peaks with a Gaussian curve to obtain the migration
time, t i, and the full-width at half-maximum, wi for peak i
according to the following equations: 41
R ) (2 ln 2) 1/2 t 2 - t 1
w 1 + w 2
And subsequently: Rbp ) ∆bp/R expresses the resolution in base
pairs, that is, the minimum base pair separation of two peaks
that would still appear separated at the baseline in the
electropherogram. Figure 4A displays an example of an electropherogram
which represents the “binned” data prior to deconvolution
into the target wavelengths, that is, the electropherogram
represents the sum of photon counts over time, and represents
the whole of the signal collected by the CCD by time. Figure 4B
shows the evolution of the resolution Rbp on the µCE system as
a function of the known peak size (red trace) and the
comparison with an electropherogram obtained on a commercial
ABI 310 capillary electrophoresis analyzer (Applied
Biosystems) using the same polymer matrix at 65 °C with a
cathode potential of 10.3 kV. The resolution obtained on the
µCE module is very close to that of a commercial CE apparatus.
Typically, the resolution values obtained for a GeneScan 500
LIZ varied between 1.02 and 1.24 bp for peaks within the 140-160
bp range. Ongoing work aims at further improving the resolution
by (1) using a pinched injection and (2) tailoring the composition
of the HEC/PVP mixture for use in a microchip rather than a
capillary.
A series of allelic ladder runs was recorded on the CE
subsystem using a CC5-labeled ILS 500 size standard (Promega)
to estimate the run-to-run stability of the system. Consecutive runs
were acquired on three different microchips to determine the
variability between microchips. Figure 5 shows a typical sized run
processed with GeneMarker HID 1.76. Inserts B though E are
(41) Manabe, T.; Chen, N.; Terabe, S.; Yohda, M.; Endo, I. Anal. Chem. 1994,
66, 4243–4252.

Figure 4. (A) Typical raw electropherogram obtained for a GeneScan
500 LIZ run on the CE subsystem using a 140 mm microchip at 50
°C. (B) Comparison of the resolution (in base pair) Rbp of the µCE
system (red trace; 2) to that obtained on a 310 apparatus (Applied
Biosystems) at 65 °C in a 36 cm capillary (blue trace; 1).
close-ups of the [7-9] allele region (304-310 bp) on the D22S1045
marker, the [10-11] region (146-150 bp) on the D18S51 marker,
the [9-11] region (97-105 bp) on the TH01 marker, and the
[21.2-26] region (177-195 bp) on the FGA marker respectively.
Table 3 summarizes the observed shifts in base pair sizing for
five different allelic ladder runs using a number of alleles from
Figure 5B-E. The maximum observed variation around the
tabulated value in the positioning of an allele is only about 0.2 bp
on the TH01 marker, 0.7 bp on the D18S51 marker, 0.4 bp on the
FGA marker, and 0.7 bp on the D22S1045 marker. The variability
increases with the fragment size as expected from the absolute
chromatographic resolution of the CE subsystem given in Figure
4B. The allele sizing data gained from running a minimum of five
allelic ladders of known designations was used to define the
expected sizes of the unknown alleles in the amplified DNA
samples.
Integrated Sample Run. We were able to obtain STR
profiles from lysed cells from buccal swabs using a disposable
plastic cartridge attached to a reusable glass microchip device
without any manual interruption of the programmed DNA
extraction, purification, amplification, transfer and CE analysis
process. A typical electropherogram including the ILS is shown
in Figure 6 after applying GeneMarker HID 1.76 to size the
migrated DNA fragments. The DNA profiles obtained as an
integrated automated run were always similar to those obtained
on an Applied Biosystems ABI 3130xl analyzer in terms of the
number and position of the peaks and their relative intensities.
Currently we are unable to routinely label all the peaks with a
definitive allele designation as the migration of the peaks of
some samples has given out of window peaks. We believe this
may be due to a difference in the time base of the instrumentation
and the NanoIdentity software: The software assumes an
equal time base per data point while the instrumentation
collects data on a variable time base which averages 0.21 s with
a maximum of 0.35 s and a minimum of 0.07 s. An updated
Figure 5. (A) Powerplex ESI16 allelic ladder run using CC5 ILS500 size standard (Promega, Madison, WI) sized using GeneMarker HID 1.76
(SoftGenetics LLC, State College, PA) with appropriate panels and bins provided by Promega. (B) Close-up of the [7-9] allele region (304-310
bp) on the D22S1045 marker. (C) Close-up of the [10-11] region (146-150 bp) on the D18S51 marker. (D) Close-up of the [9-11] region
(97-105 bp) on the TH01 marker. (E) Close-up of the [21.2-26] region (177-195 bp) on the FGA marker.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6997

Table 3. Evolution of the Sizing Results for Specific Regions of an Allelic Ladder Run for Three Different Chips and
Five Different Runs a
D22S1045 size of allele no. 7 (bp) size of allele no. 8 (bp)
chip 1 - 1st load 304.2 307.3
chip 2 - 1st load 305.6 308.2
chip 2 - 2nd load 304.1 307.2
chip 3 - 1st load 304.3 307.2
chip 3 - 2nd load 304.0 307.2
D18S51 size of allele no. 10 (bp) size of allele no. 10.2 (bp) size of allele no. 11 (bp)
chip 1 - 1st load 146.1 148.2 150.2
chip 2 - 1st load 147.1 149.2 151.1
chip 2 - 2nd load 146.3 148.3 150.2
chip 3 - 1st load 145.8 147.7 149.7
chip 3 - 2nd load 145.9 147.8 150.0
TH01 size of allele no. 9.3 (bp) size of allele no. 10 (bp)
chip 1 - 1st load 100.2 101.2
chip 2 - 1st load 100.3 101.3
chip 2 - 2nd load 100.2 101.1
chip 3 - 1st load 100.1 101.4
chip 3 - 2nd load 100.2 101.2
FGA size of allele size of allele
no. 21.2 (bp) no. 22 (bp)
size of allele size of allele
no. 22.2 (bp) no. 23 (bp)
size of allele size of allele
no. 23.2 (bp) no. 24 (bp)
size of allele size of allele
no. 24.2 (bp) no. 25 (bp)
size of allele size of allele
no. 25.2 (bp) no. 26 (bp)
chip 1 - 1st load 176.8 178.8 180.8 182.8 184.8 186.8 188.9 190.8 192.8 194.8
chip 2 - 1st load 177.2 179.3 181.0 183.0 184.2 186.0 188.0 189.8 191.6 193.3
chip 2 - 2nd load 177.0 178.9 180.7 182.7 184.5 186.6 188.8 190.7 192.6 194.7
chip 3 - 1st load 176.8 178.6 180.7 182.7 184.5 186.6 188.8 190.7 192.6 194.7
chip 3 - 2nd load 176.6 178.5 180.6 182.5 184.7 186.6 188.6 190.6 192.7 194.5
a The values represent the calculated size of the designated allele in DNA base pairs relative to the ILS in the same run. The markers are
chosen to represent 1 marker per color, cover most of the regions of interest [100-300 bp], and correspond to the most challenging peaks to
separate.
Figure 6. Micro Capillary Electrophoresis profile from a fully integrated run. DNA was purified and amplified on-cartridge from lysed cells from
a buccal swab and transferred automatically via Teflon tubing to the glass microchip. Allele designation was performed using GeneMarker HID
1.76 (SoftGenetics LLC).
version of the software is in development which will use the
time file provided by the instrument and so align the data
correctly on a true time scale.
6998 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
CONCLUSION
We have demonstrated a successful microTAS approach to
the delivery of rapid DNA technology suitable for the forensic

field. DNA profiles were delivered from a crude DNA lysate
with no manual input. The automated process is simple to use
and provided DNA profiles which were concordant with the
expected DNA profile of the donor. Our target is to deliver an
evidential quality DNA profile from sample to profile in less
than 2 h. While we currently have the capability of processing
the sample in under 4 h, we believe that there are a number of
areas where we can speed up the process. A major time saving
(approximately 1.5 h) can be made simply by reducing the PCR
cycling parameters from those published here to thermal
profiles similar to those described previously. 4–6 Optimizing
the cartridge operating process to allow parallel processing of
valve firings and pump control will further reduce the processing
time and we believe we will soon be able to meet the target
time of 2 h for delivery of the genotyped DNA profile. The turn
around between samples is approximately 10 min to load the
solutions into the single-use plastic cartridge, and 75 min for
preparation of the CE chip. This can be carried out while the
previous sample is running, provided that a number of CE chips
are available, reducing the turn around time to little more than
5 min. The resolution obtained on the µCE module is very close
to that of a commercial CE apparatus and the reproducibility
of fragment migration was good: With allelic ladders the
maximum variation in the positioning of an allele was around
0.2 bp on the TH01 locus, 0.7 bp on the D18S51 locus, 0.4 bp
on the FGA locus, and 0.7 bp on the D22S1045 locus. The
migration of DNA profiles from human samples does require
some improvement if the system is to be used for probative
samples as we were unable to successfully type all integrated
runs due to a small variation in the apparent migration of
fragments. This variation is likely to be minimized by implementation
of a new version of the NanoIdentity software which
will take the true time file from the instrument, rather than
assuming all data points are equidistant in time. The largest
variation in the process remains the amount of template DNA
input to the PCR reaction, and while this impacts the peak
heights observed in the electropherogram, this does not affect
the accuracy of the genotyping of a sample.
Simple substitution of the multiplex chemistry in the cartridge
will allow the production of DNA profiles suitable for loading to
any established national DNA database. However, the rapid
delivery of a DNA profile, whether at a crime scene, or in a
custody suite is only a part of the puzzle. If the full impact of these
new technologies is to be gained, the rapid delivery of chemistry
must be supported by a capability to submit and compare the DNA
profile in near real time. Currently neither the CoDIS database,
nor the UK National DNA database have the capability to support
rapid chemistry protocols.
While the operation of the instrument is straightforward, a
good level of training is required before the instrument can be
operated but we believe that any individual with a basic scientific
education could become competent in the operation of the
instrument. Our vision for the future is to have a single fully
integrated injection-molded cartridge with reagents preloaded with
capability for DNA purification, amplification and µCE-based
separation. Such a cartridge, coupled with automatic channel
finding and focusing will minimize set up time and make the
system usable by an individual with very basic training.
ACKNOWLEDGMENT
We gratefully acknowledge the technical assistance of Glen
McCarty, David Nguyen, and Brett Duane for electronics design
and prototyping; Baiju Thomas, Keith Burt, and Matthew Barrett
for plastic device fabrication; Dr. Jian Gu and Peter East for
microfabrication; and Amol Surve for instrumentation design. We
also thank Bruce McCord for the provision of the separation
matrix and helpful discussion around microCE and Promega
Corporation for the generous gifts of PowerPlex ESI 16 STR kits
and ILS 500 CC5.
Received for review May 24, 2010. Accepted July 1, 2010.
AC101355R
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
6999

Anal. Chem. 2010, 82, 7000–7007
Visualization and Recovery of the (Bio)chemical
Interesting Variables in Data Analysis with Support
Vector Machine Classification
Patrick W. T. Krooshof, † Bülent Üstün, ‡ Geert J. Postma, † and Lutgarde M. C. Buydens* ,†
Radboud University Nijmegen, Institute for Molecules and Materials, Analytical Chemistry, P.O. Box 9010,
6500 GL Nijmegen, The Netherlands, and Analytical Sciences Chemicals, Quality Unit API/Biotech, MSD Oss, P.O.
Box 20, 5340 BH Oss, The Netherlands
Support vector machines (SVMs) have become a popular
technique in the chemometrics and bioinformatics field,
and other fields, for the classification of complex data sets.
Especially because SVMs are able to model nonlinear
relationships, the usage of this technique has increased
substantially. This modeling is obtained by mapping the
data in a higher-dimensional feature space. The disadvantage
of such a transformation is, however, that information
about the contribution of the original variables in
the classification is lost. In this paper we introduce an
innovative method which can retrieve the information
about the variables of complex data sets. We apply the
proposed method to several benchmark data sets and a
metabolomics data set to illustrate that we can determine
the contribution of the original variables in SVM classifications.
The corresponding visualization of the contribution
of the variables can assist in a better understanding
of the underlying chemical or biological process.
In the past decade support vector machines (SVMs) have
become a popular technique in pattern recognition and regression
estimation. Applications of SVMs are among the fields of
bioinformatics, 1,2 medicine, 3-6 drug discovery, 7-9 text categorizing,
10 gene expression analysis, 11-13 face recognition, 14 spam
* To whom correspondence should be addressed. Phone: +31 24 3653180.
Fax: +31 24 3652653. E-mail: l.buydens@science.ru.nl.
† Radboud University Nijmegen.
‡ Analytical Sciences Chemicals.
(1) Yang, Z. R. Briefings Bioinf. 2004, 5 (4), 328–338.
(2) Ramo, P.; Sacher, R.; Snijder, B.; Begemann, B.; Pelkmans, L. Bioinformatics.
2009, 25, 3028–3030.
(3) Akay, M. F. Expert Syst. Appl 2009, 36 (2), 3240–3247.
(4) Magnin, B.; et al. Neuroradiology. 2009, 51 (2), 73–83.
(5) Luts, J.; Heerschap, A.; Suykens, J. A. K.; Van Huffel, S. Artif. Intell. Med.
2007, 40 (2), 87–102.
(6) Conforti, D.; Guido, R. Comput. Oper. Res. 2010, 37, 1389–1394.
(7) Burbidge, R.; Trotter, M.; Buxton, B.; Holden, S. Comput. Chem. 2001,
26, 5–14.
(8) Warmuth, M. K.; et al. J. Chem. Inf. Comput. Sci. 2003, 43, 667–673.
(9) Zernov, V. V.; Balakin, K. V.; Ivaschenko, A. A.; Savchuk, N. P.; Pletnev,
I. V. J. Chem. Inf. Comput. Sci. 2003, 43, 2048–2056.
(10) Leopold, E.; Kindermann, J. Mach. Learn. 2002, 46, 423–444.
(11) Furey, T. S.; et al. Bioinformatics. 2000, 16 (10), 906–914.
(12) Clarke, R.; et al. Nat. Rev. Cancer. 2008, 8, 37–49.
(13) Noble, W. S. Nat. Biotechnol. 2006, 24 (12), 1565–1567.
(14) Guo, G.; Li, S. Z.; Chan, K. L. Image Visualization Comput. 2001, 19, 631–
638.
7000 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
categorizing, 15,16 financial forecasting, 17,18 and many others.
Especially because of the possibility to model complex nonlinear
relationships the application of SVMs has grown substantially. 19-22
By transforming the original input space into a high dimensional
feature space, the nonlinear relationships can be presented in a
linear form. This transformation is performed by using a specific
kernel function. 6,23,24 Several kernel functions are proposed in the
literature for this purpose and include variance-covariance based
linear and polynomial kernels, the Euclidean distance based radial
basis function (RBF) and the Pearson VII Universal Kernel (PUK)
functions. 23-25 The transformation by a kernel function has also
been introduced in other algorithms, such as Kernel Principal
Component Analysis, 26 Kernel Partial Least Squares, 27,28 and
Kernel Fisher Discrimination. 29
However, the disadvantage of using such a kernel function is
that the correlation between the obtained SVM model and the
original input space is lost. Therefore it is not possible to
determine which variables (e.g., spectral ranges) contribute to
the final SVM results and a direct interpretation of the SVM model
is not straightforward. 30,31 This seriously hampers the ultimate
(bio)chemical interpretation of the resulting classification model.
In this manuscript we propose a novel method to overcome this
problem and reveal the importance of the original variables.
(15) Drucker, H.; Wu, D.; Vapnik, V. N. IEEE Trans. Neural Networks 1999,
10 (5), 1048–1054.
(16) Guzella, T. S.; Caminhas, W. M. Expert Syst. Appl. 2009, 36 (7), 10206–
10222.
(17) Tay, F. E. H.; Cao, L. Omega. 2001, 29, 309–317.
(18) Kim, K. J. Neurocomputing. 2003, 55, 307–319.
(19) Vapnik, V. Estimation of Dependence Based on Empirical Data; Springer
Verlag: New York, 1982.
(20) Vapnik, V. the Nature of Statistical Learning Theory; Springer Verlag: New
York, 1995.
(21) Cortes, C.; Vapnik, V. Mach. Learn. 1995, 20, 273–297.
(22) Vapnik, V. Statistical Learning Theory; John Wiley and Sons: New York,
1998.
(23) Cristianini, N.; Shawe-Taylor, J. An Introduction to Support Vector Machines
and Other Kernel-Based Learning Methods; Cambridge University Press.:
Cambridge, 2000.
(24) Schölkopf, B.; Smola, A. J. Learning with Kernels; MIT Press.: Cambridge,
2002.
(25) Üstün, B.; Melssen, W. J.; Buydens, L. M. C. Chemom. Intell. Lab. Syst.
2006, 81, 29–40.
(26) Schölkopf, B.; Smola, A. J.; Müller, K. R. Neural Comput. 1998, 10, 1299–
1319.
(27) Walczak, B.; Massart, D. L. Anal. Chim. Acta 1996, 331, 177–185.
(28) Rosipal, R.; Trejo, L. J. J. Mach. Learn. Res. 2001, 2, 97–123.
(29) Mika, S.; Rãtsch, G.; Wetson, J.; Schölkopf, B.; Müller, K. R. Proc. NNSP′99;
1999, 41–48.
10.1021/ac101338y © 2010 American Chemical Society
Published on Web 07/20/2010

Another approach to handle this limitation is automatic relevance
determination, 32,33 which was developed as a feature selection
procedure in SVM models. Even though this method selects the
variables that are important for the model, the relation between
the variables and, for example, the class separation is not
visualized. Generally, researchers report the high performance
obtained by using a SVM model for classification and regression
estimation, but do not comment on the relationship between the
input variables and the modeled output data. SVM is often used
as a black box approach.
In this paper we present an innovative approach to open this
black box for the SVM classifier and give insight in the transformation
by the kernel function to make the SVM model more
transparent. This approach is based on the nonlinear biplot
principles described by Gower and Harding in 1988 34 and is used
to visualize and determine the importance and influence of the
input variables to the final SVM classifier. The resulting information
can then be used to reduce the number of input variables to
improve the performance or to reduce the complexity of the
model. Furthermore, the visualization of the contribution of the
original variables can assist in a better understanding of
the underlying chemical or biological process. We will present
the proposed approach, and illustrate and validate the methodology
by applying it for classification problems: two benchmark data
sets and a relevant metabolomics data set obtained from magnetic
resonance spectroscopic images to diagnose human brain tumors.
The effectiveness of the method is verified by a comparison of
the classification performance obtained by using the entire set of
input variables and a selection of variables, determined by our
approach.
EXPERIMENTAL SECTION
Theory and Computational Strategy. As the theory of SVMs
is described extensively in the literature 13,21-24 and the use of a
kernel transformation results in the loss of information about the
input variables, we will focus in the next section briefly on the
concepts of the kernel function. Subsequently, we will explain
the basic steps of the nonlinear biplot technique, as described by
Gower and Harding, 34 to discuss the proposed method to
determine the contribution of the input variables in SVM classifications.
The full theory and algorithm of the proposed method
can be found in the Supporting Information (SI).
Kernel Transformations. In the various kernel-based methods
a specific mapping function is used to project the original input
data in a higher-dimensional feature space. 24 The data in this new
feature space can subsequently be used as a new input for pattern
recognition. The advantage of such an approach is that the method
can deal with complex nonlinear problems. The typical (twodimensional)
example to illustrate this approach is shown in the
inset of Figure 1. This data set (X) contains two classes that we
would like to separate.
(30) Üstün, B.; Melssen, W. J.; Buydens, L. M. C. Anal. Chim. Acta 2007, 595,
299–309.
(31) Devos, O.; Ruckebusch, C.; Durand, A.; Duponchel, L.; Huvenne, J. P.
Chemom. Intell. Lab. Syst. 2009, 96, 27–33.
(32) Van Gestel, T.; Suykens, J. A. K.; De Moor, B.; Vandewalle, J. Proc. Eur.
Symp. Artif. Neural Networks. 2001, 13–18.
(33) MacKay, D. J. C. In Neural Networks and Machine Learning, NATO Asi
Series. Series F, Computer and Systems Sciences 168; Bishop, C. M., Ed.;
Springer: Berlin, 1998; pp 133-165.
(34) Gower, J. C.; Harding, S. A. Biometrika 1988, 75, 445–455.
Figure 1. The synthetic benchmark data set. Mapping of a twodimensional
data set which contains two nonlinearly separable classes
(outer and inner circles) into a three-dimensional feature space. The
transformation is performed by applying the nonlinear mapping
function φ(x.1, x.2) ) (x.1 2 ,�2x.1x.2, x.2 2 ), which makes the two classes
linearly separable.
The inner circle represents one class and the outer circle
represents the other class. From the figure it is obvious that we
are not able to separate the classes by a linear model. However,
if we project this data in a higher dimensional space by using a
mapping function, we are able to obtain the three-dimensional
feature space which is represented in Figure 1. In this feature
space the two classes can be linearly separated by a plane between
the circles. In this case the transformation is performed by the
mapping function φ. It has been shown that, since the (nonlinear)
mapping function is in general unknown beforehand and is difficult
to determine, the feature space can be constructed implicitly by
invoking a generic kernel function (see, e.g., refs 23-25, 35). Such
a kernel function is a function (K) which operates on two vectors,
such that
K(x i. , x j. ) ) 〈�(x i. ), �(x j. )〉 (1)
where xi. and xj. are two objects in the data set and φ represents
the actual nonlinear mapping function. The use of a kernel
function makes it unnecessary to know the actual underlying
feature map in order to be able to construct a linear model in
the feature space. The application of such a kernel function
will result in a square symmetric matrix: the kernel matrix K.
This matrix is a weighted dissimilarity matrix, of which each
position represents a dissimilarity (distance) measure between
two objects. The specific kernel function and the optimal
parameter settings of the function are determined in combination
with the applied classification algorithm (e.g., SVM) by
optimizing the classification performance. However, because
the kernel function transforms the original input space into a
feature space with a higher dimension, information about the
original variables is not preserved.
Nonlinear Biplots. To explain the concepts of nonlinear biplots,
we will first describe the classical biplots technique which is used
extensively in (chemometric) data analysis.
(35) Gunn, S. R. Support Vector Machines for Classification and Regression.
Technical Report; Image Speech and Intelligent Systems Research Group,
University of Southampton: Southampton, 1997.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7001

Figure 2. Biplots of the Iris benchmark data set. The scores (representing the samples) are visualized as symbols, whereas the loadings are
visualized by the vectors. The loadings are obtained by (a) projection of the first two columns in the loading matrix and (b) projection of the
scores of 10 pseudosamples (for each variable).
A biplot is a visualization in which the samples and the
variables of a data set are represented together. This technique
has been introduced by Gabriel in 1971 36 and is based on singular
value decomposition (SVD) or principal component analysis
(PCA) 37 of the column mean-centered data matrix X, containing
n samples and m variables. By SVD X is decomposed into scores
and loadings according to
T
X (n×m) ) U (n×r) Λ (r×r) V (m×r)
T
) S (n×r) L (m×r)
The positions of the samples (rows of X) in a two-dimensional
biplot are subsequently given by the elements in the columns of
UΛ, often called the scores S. The variables (columns of X) are
usually represented as vectors pointing from the origin to the
coordinates that are given by the elements of the columns of V,
called the loadings L. To construct a biplot for the first two
singular vectors (or principal components, PCs), the elements in
the first two columns of S and in the first two columns of L are
used. This is illustrated for the Iris data set (see the Iris
benchmark Data Set section) in Figure 2a. The coordinates of a
new sample in this biplot can be obtained by premultiplying the
sample as a row vector with V:
x (1×m) V (m×2) ) s (1×2)
The key idea leading to the nonlinear biplot is the interpretation
of the loadings in the classical biplot (the representation of the
variables) as projections of special so-called pseudosamples that
carry all their weight in one variable. This means that these
pseudosamples have a value of 0 for all variables, except for one
variable. Projecting this pseudosample in the two-dimensional PCA
plot yields coordinates that are equal to the loadings of the variable
whose weight it carries. For example, [1, 0, 0, ..., 0] is a (1 × m)
pseudosample with a value 1 for variable 1 and a value 0 for all
other variables. Then
[1, 0, 0, ..., 0] (1×m) V (m×2) ) s (1×2) ) v (1×2)
(36) Gabriel, K. R. Biometrika. 1971, 58, 453–467.
(37) Massart, D. L. Handbook of Chemometrics and Qualimetrics: Part A; Elsevier
Science Publishers: Amsterdam, 1997.
7002 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(2)
(3)
(4)
where v is the first row of V and contains exactly the loading of
variable 1, defining its position in the biplot. If the value of 1 is
replaced by p different values z, a total of p pseudosamples are
obtained. The projection of these pseudosamples in the PCA plot
will result in a trajectory along the direction of the variable vector,
as illustrated in Figure 2b.
Gower et al. 34 extend this idea to the visualization of variable
information in principle coordinate analysis. In principal coordinate
analysis (or classical metrics scaling) an SVD is performed not
on the original rectangular data matrix X, but on the symmetric
matrix of squared Euclidean distances. This approach allows the
visualization of the relative distances of the objects. It is often
remarked that the information of the original variables (such as
in a biplot) is lost. Gower et al., however, showed that this can be
overcome by using the concept of trajectories of pseudosamples,
as explained earlier. To make this approach feasible, one must
be able to calculate the squared Euclidean distances of the
pseudosamples with all other samples (e.g., data X). By projecting
the rows of the resulting distance matrix in the principal
component space, the trajectories that represent the variables are
obtained. Gower et al. showed that this concept can be extended
from Euclidean distances to many nonlinear distance metrics, as
long as the same distance metric can be used to calculate the
distances of the pseudosamples to the original data samples. In
the classical linear biplot, only one pseudosample per variable is
sufficient to represent the variables. For a nonlinear distance
metric, the trajectory will be curved and multiple pseudosamples
per variable are required.
Nonlinear Biplots in SVM Classification. We propose to apply
the above approach to the kernel-based SVM method, since the
kernel is a (nonlinear) distance metric between objects. This
procedure comprises the following steps (shortened):
0. Optimize the kernel function and parameter settings for the
SVM classification used (resulting in the kernel function K(xi.,xj.)).
1. From the data X calculate the kernel matrix K, by using
the optimized kernel function. This matrix is subsequently
centered, resulting in matrix K c of size n × n.
2. Apply SVD on K c and construct different score plots (for
the various combinations of principal components) for the n
samples in K c . Inspect these score plots to find the direction(s)

(principal components, PCs) in which maximum class separation
is obtained. Note that this is not necessarily along the first
PCs.
3. Construct a matrix Pj for the j th variable which contains p
pseudosamples. The range of these pseudosample values zj
should vary between the minimum and maximum value of the
original variable.
4. Apply the kernel function K(pi.,xj.) to the pseudosample
data Pj to obtain the kernel matrix of the pseudosamples C
(i.e., calculate the kernel distances of the pseudo samples to
the original samples).
5. Project the rows of the centered pseudosample kernel matrix
C c in the score plot that was found in step 2.
6. Repeat steps 3-5 for each variable j.
The resulting plot contains a trajectory of pseudosamples for
each original variable. These trajectories yield information about
the relative contribution of the variables to the SVM classification.
The curvature of a trajectory indicates a nonlinear kernel
transformation.
SVM Classification and Validation Procedure. Each data set was
analyzed by SVM using the RBF kernel. The RBF kernel was
chosen because of its simplicity for optimization. Because the SVM
application that we have used is a binary classifier, each different
class in a data set that contains more than two classes was
analyzed by a one-against-all approach. 38 The kernel parameters
parameter σ and the SVM parameter C (C is a cost parameter 23 )
were optimized by leave-10%-out cross-validation.
All calculations are performed in the software program Matlab
(The MathWorks Inc.) version 6.5 release 13. The nonlinear biplot
approach was implemented in Matlab by using the commercially
available PLS toolbox from eigenvector Research Inc.
Data Sets. To illustrate the applicability of the proposed
method we have used several benchmark data sets (synthetic and
real) and a metabolomics data set based on Magnetic Resonance
Spectroscopic Imaging (MRSI) data.
Synthetic Benchmark Data Set. A synthetic data set was
constructed based on the data presented in Figure 1, which is
generally used to illustrate the power of SVM classification. The
data set contains two variables to represent an inner and outer
circle, both consisting of 50 objects. The two circles can not be
separated in a linear way and therefore a kernel-based classification
method is required to separate the data.
To construct the Synthetic data set we added five variables
containing random noise (normally distributed) to the data. The
variance of these five variables was set to be about 10% of the
variance of the two variables which represent the circles. Because
noise contains no information, the added variables do not
contribute to the classification performance. This data set was
constructed to confirm that the first two variables (representing
the circles) are only (equally) important in the SVM classification.
Iris Data Set. A widely used benchmark data set to exemplify
discriminant and cluster analysis is the data published by Fisher
in 1936. 39 This Iris data set contains fifty specimens of each of
the three species Iris Setosa, Iris Versicolor, and Iris Virginica,
resulting in a total of 150 samples. Four properties of the species
are measured to determine differences between the three classes.
(38) Suykens, J. A. K.; Van Gestel, T.; De Brabanter, J.; De Moor, B.; Vandewalle,
J. Least Squares Support Vector Machines; World Scientific: Singapore, 1999.
(39) Fisher, R. A. Annu. Eugen. 1936, 7, 179–188.
These properties (representing four variables) are Sepal Length,
Sepal Width, Petal Length, and Petal Width, all measured in
millimeters. The Setosa class can easily be separated from the
other two classes using only one variable (either Petal Length or
Petal Width). The two other classes are partly overlapping, as
shown in Figure 2.
We will use the Iris data set to demonstrate the applicability
of our proposed method to determine the most discriminative
variable for the three species.
Metabolomics Data Set. To illustrate the proposed method for
variable selection on a more complex data set, we have used a
metabolomics data set which consists of magnetic resonance
spectroscopic (MRS) spectra obtained from MRSI data. 40 This data
set was constructed during a European project called Interpret,
which was funded by the European Commission to develop new
methodologies to automatically classify tumors in the human brain
(see http://azizu.uab.es/INTERPRET). Data from a total of 24
patients and 4 volunteers were acquired by MRS at different
positions in the brain, according to an acquisition protocol defined
by the Interpret Consortium. The study was approved by the
ethical committee and followed the rules of the World Health
Organization. After reaching consensus about the histopathology,
three tumor types were identified according to the World Health
Organization classification system. These three classes contained
glial tumors with different grades: Grade II (10 cases), Grade III
(4 cases), and Grade IV (7 cases). A fourth class consists of spectra
acquired from patients with Meningioma (3 cases). Additionally,
a class consisting of Healthy tissues was created from patient (4
cases) and volunteer (4 cases) data. For each predefined class a
selection of spectra from the different patients was made. Only
spectra acquired at regions which clearly consisted of tissue
belonging to the particular class were selected. The data for the
Healthy class was selected from the volunteers or from the contralateral
brain region of the patients. 41 The resulting data set
contains 569 spectra, consisting of five different classes. Each
spectrum contains 229 data points, covering the chemical shifts
between 4.0 and 0.5 ppm. Details about the acquisition parameters
and preprocessing of the data are described in Simonetti et al 42
and are beyond the scope of this paper.
RESULTS
Synthetic Benchmark Data Set. Classification of the Synthetic
data set by using SVMs resulted in a leave-10%-out crossvalidated
accuracy of 100%. This accuracy (see also SI Table 1)
illustrates that SVMs are able to separate the two circles. As noise
contains no information, the particular noise-variables should not
have contributed to the class separation. To verify the contribution
of each variable in the final classification first we have to find
and visualize the optimal (linear) separation between the classes
in the resulting kernel matrix. Because the Synthetic data set
contains one hundred objects, the feature space in which the
original data is mapped (i.e., the space of the kernel matrix)
consists as a consequence of one hundred dimensions. Therefore
PCA is used to reduce the dimensionality of the feature space.
Analysis of the score plots of combinations of only the first five
(40) Barker, P. B.; Lin, D. D. M. Prog. Nucl. Magn. Reson. Spectrosc. 2006, 49,
99–128.
(41) Simonetti, A. W.; et al. Anal. Chem. 2003, 75, 5352–5361.
(42) Simonetti, A. W.; et al. NMR Biomed. 2005, 18, 34–43.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7003

Figure 3. Principle component score plots of the synthetic benchmark data set. Pairs-plot of the first five principal components (PCs) are
shown. The separation between the inner circle (blue +-symbol) and the outer circle (red o-symbol) is obtained on PC 5.
Figure 4. Relevant score plot and pseudo sample trajectories for the Synthetic benchmark data set. (a) Projection of the objects of the Synthetic
benchmark data set (after the kernel transformation) in the feature space spanned by PC 1 and PC 5. (b) Pseudosample trajectories projected
in the same feature space. The trajectories of the two variables representing the two circles are indicated by the numbers “1” and “2”. The five
noise-variables (variables “3” to “7”) are distributed around the origin of the plot and are therefore not clearly visible.
PCs show that “PC 5” can be used to visualize the class separation
of the two classes. The pairs-plot for the first five PCs of the
Synthetic data set is given in Figure 3. Apparently, the variance
which accounts for the separation between the classes is captured
by PC 5. The contribution of each original variable can be
determined by projecting the kernel matrix of the corresponding
pseudosamples in the feature space spanned by PC 5 and any of
the other PCs and by subsequently analyzing the obtained
trajectories.
The pseudosample matrices were constructed by varying the
intensities of the individual variables over twenty objects (uni-
7004 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
formly distributed). The application of the kernel function resulted
in seven kernel matrices (one for each variable) of size (20 × 100).
These matrices were then projected in the feature space obtained
by PCA on the original kernel matrix K. The trajectories of the
seven variables obtained by the proposed method are visualized
for the space spanned by “PC 1” and “PC 5” in Figure 4b. As
shown, the trajectories of the five noise-variables (variables 3-7)
are along the direction of the class separation (PC 5), but are
relatively small compared to the two trajectories of the variables
representing the inner and outer circle (variables 1 and 2). This
indicates that the noise-variables have a small contribution to the

Figure 5. Relevant score plots and pseudo sample trajectories for the Iris benchmark data set. Projection of the objects of the Iris data set
after kernel transformation and PCA, to illustrate the optimal separation of the (a) Setosa, (b) Versicolor, and the (c) Virginica class (red o-symbols).
The pseudosample trajectories are projected in the same feature space for the (d) Setosa, (e) Versicolor, and the (f) Virginica class. The
trajectory of each variable is indicated by a different number: “1” for Sepal Length, “2” for Sepal Width, “3” for Petal Length, and “4” for Petal
Width.
class separation, which is in accordance with our hypothesis. As
the trajectories of variable 1 and 2 have a similar length on PC 5,
both variables are important for the class separation. A (linear)
classification based on only one variable is therefore not possible.
A visual inspection of the Synthetic data set (see the Theory and
Computational Strategy section) already confirmed this conclusion.
Inspection of the trajectories in the score plots of combinations
of other PCs showed that the variables representing the two circles
can have a relatively small loading in the particular feature space
(results not shown). For example, in the space spanned by “PC
2” and “PC 4” variable 4 (representing noise) has the largest
loading compared to the other variables. However, no class
separation was found in this particular score-plot (see Figure 3)
and therefore this feature space is not informative.
Iris Benchmark Data Set. The application of the three
possible one-against-all classifications resulted in cross-validated
accuracies of at least 96.7% (see also SI Table 1). After PCA was
applied to the three kernel matrices (for each one-against-all
classifier), we searched for the PCs resulting in the optimal
separation between the classes in the pairs-plots. These optimal
separations are visualized in Figure 5a-c and as shown the Setosa
class is completely separated from the other classes by the
variance captured by PC 1. The Versicolor and Virginica class
requires two principal components to capture the variance which
accounts for the class separation, that is, PC 2 and PC 5 for
Versicolor and PC 1 and PC 2 for the Virginica class.
To determine the contribution of the original variables, pseudosample
trajectories were constructed and projected in the space
spanned by the corresponding PCs. These projections are visual-
ized in Figure 5d-f and as illustrated the variables Petal Length
(variable 3) and Petal Width (variable 4) are most discriminative
for the classifications, confirming our hypothesis. However, the
length of the trajectory of Petal Width is much shorter compared
to Petal Length (for unscaled data) and is therefore less important
for the classifications. To confirm these observations we have
applied SVM to the Iris data set by including only Petal Length.
The resulting cross-validated accuracies are given in SI Table 2.
Although the accuracies for the Versicolor and Virginica class are
somewhat lower compared to the results where all the variables
are used (95.3% versus 96.7%), the accuracies are still comparable.
If one of the other variables was chosen for the classification, that
is, Sepal Width, the accuracies are much lower (

Figure 6. Relevant score plot and pseudo sample trajectories for the metabolomics data set. (a) Space spanned by PC 1 and PC 3, obtained
by PCA applied on the kernel matrix of the MRSI data set. The Healthy class is indicated by red o-symbols and the tumor classes by blue
+-symbols. (b) Pseudosample trajectories, projected in the same feature space as in (a). Each trajectory is indicated by a different number and
represents a different variable. The trajectories are also color-coded using the variable numbers, resulting in similar colors for trajectories of
variables representing specific regions in the NMR spectrum.
One group of trajectories consists of pseudosamples representing
the variable numbers 170-184 (1.40-1.19 ppm; the trajectories
in the direction of PC1, colored orange in Figure 6b), and the
other group the variables 129-132 (2.03-1.98 ppm; the trajectories
more or less in the direction of PC3, colored green in Figure
6b). Because the group of variables within 2.03-1.98 ppm is along
the direction of the class separation we postulate that these
variables have a large contribution to the SVM classification. This
observation corresponds to results published in the literature, in
which the researchers stated that this specific region corresponds
to N-acetyl-aspartate, which is a neuronal marker for viable
neurons, and that the concentration is reduced or absent in most
brain tumors. 43
Note that the other group of trajectories, representing the
variables 170-184 (1.40-1.19 ppm), corresponds to lactate and
lipids regions of the spectra. These trajectories are located in the
direction along PC 1 and direction corresponds to increasing
tumor grade. This observation also corresponds to the results of
Howe et al. 43
With only the variables selected within the region of 2.03-1.98
ppm (four variables), the SVM classification results in a leave-
10%-out cross-validated accuracy of 95.4% (see SI Table 2). This
value is in agreement with the accuracy obtained when all the
variables are included in the SVM model (98.7%), indicating that
these variables have a large contribution to the classification. If a
set of four variables was selected randomly, the cross-validated
accuracy was

CONCLUSIONS
In this paper we have introduced a new method to successfully
visualize and identify the important variables in the classification
by a kernel-based method. By constructing a set of pseudosamples
we are able to determine the effect of the kernel function to the
individual variables. We have applied the proposed method to
several synthetic and real data sets and have shown that our
method is able to find and visualize the most discriminative
variables. To confirm this conclusion, we have selected only the
most discriminative variables and reapplied the proposed method.
The cross-validated classification accuracies of these reduced data
sets are similar to the accuracies obtained by using the data sets
with the full set of variables. This illustrates the validity of our
method to determine the most important variables for classifica-
tion purposes by kernel-based techniques. The interpretation of
SVM models with our proposed visualization method has the
potential to extract relevant chemical and biological knowledge
from complex data, such as omics data. This will greatly enhance
the applicability of the powerful SVM classifiers.
SUPPORTING INFORMATION AVAILABLE
List of symbols, discussion of theory and computational
strategy, two additional figures, and two additional tables. This
material is available free of charge via the Internet at
http://pubs.acs.org.
Received for review May 26, 2010. Accepted July 7, 2010.
AC101338Y
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7007

Anal. Chem. 2010, 82, 7008–7014
Rapid and Sensitive Detection of Protein
Biomarker Using a Portable Fluorescence
Biosensor Based on Quantum Dots and
a Lateral Flow Test Strip
Zhaohui Li, Ying Wang, Jun Wang, Zhiwen Tang, Joel G. Pounds, and Yuehe Lin*
Pacific Northwest National Laboratory, Richland, Washington 99362
A portable fluorescence biosensor with rapid and ultrasensitive
response for protein biomarker has been built
up with quantum dots and a lateral flow test strip. The
superior signal brightness and high photostability of
quantum dots are combined with the promising advantages
of a lateral flow test strip and result in high
sensitivity and selectivity and speed for protein detection.
Nitrated ceruloplasmin, a significant biomarker for cardiovascular
disease, lung cancer, and stress response to
smoking, was used as model protein biomarker to demonstrate
the good performances of this proposed quantum
dot-based lateral flow test strip. Quantitative detection of
nitrated ceruloplasmin was realized by recording the
fluorescence intensity of quantum dots captured on the
test line. Under optimal conditions, this portable fluorescence
biosensor displays rapid responses for nitrated
ceruloplasmin with the concentration as low as 1 ng/mL.
Furthermore, the biosensor was successfully utilized for
spiked human plasma sample detection in a wide dynamic
range with a detection limit of 8 ng/mL (S/N ) 3). The
results demonstrate that the quantum dot-based lateral
flow test strip is capable of rapid, sensitive, and quantitative
detection of nitrated ceruloplasmin and hold a great
promise for point-of-care and in field analysis of other
protein biomarkers.
Rapid and quantitative detection of protein biomarkers is
extremely important for clinical diagnostics, basic discovery, and
a variety of other biomedical applications. As is well-known, a host
of various immunoassays for protein detection have been developed
during the past years. The more established approaches
include microsphere-based arrays, 1 proteome chips, 2 radioimmunoassay,
3,4 surface plasma resonance, 5 microfluidic systems, 6,7
* To whom correspondence should be addressed. Email: yuehe.lin@
pnl.gov. Fax: 1-509-3766242. Tel: 1-509-3716241.
(1) Blicharz, T. M.; Siqueira, W. L.; Helmerhorst, E. J.; Oppenheim, F. G.;
Wexler, P. J.; Little, F. F.; Walt, D. R. Anal. Chem. 2009, 81, 2106–2114.
(2) Zhu, H.; Bilgin, M.; Bangham, R.; Hall, D.; Casamayor, A.; Bertone, P.;
Lan, N.; Jansen, R.; Bidlingmaier, S.; Houfek, T.; Mitchell, T.; Miller, P.;
Dean, R. A.; Gerstein, M.; Snyder, M. Science 2001, 293, 2101–2105.
(3) Wide, L.; Porath, J. Biochim. Biophys. Acta 1966, 130, 257–260.
(4) Harris, E. N.; Boey, M. L.; Mackworthyoung, C. G.; Gharavi, A. E.; Patel,
B. M.; Loizou, S.; Hughes, G. R. V. Lancet 1983, 2, 1211–1214.
(5) Lyon, L. A.; Musick, M. D.; Natan, M. J. Anal. Chem. 1998, 70, 5177–
5183.
7008 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
enzyme-linked immunosorbent assay (ELISA), 8,9 surface-enhanced
Raman scattering, 10 etc. Although these conventional strategies
exhibit promising results for sensitive detection of proteins, there
are still some hindrances including the utilization of radioactive
substances, time-consuming sample purification, incubation and
washing steps before analysis, and specialized equipment. Recently,
a lateral flow test strip (LFTS), also called a dry-reagent
strip biosensor, has been becoming a powerful tool for protein
analysis and has attracted increasing attention. 11-17 In comparison
to the methods mentioned above, LFTS permits a one-step and
rapid low-cost analysis as well as a user-friendly format, very short
assay time, and no requirement for skilled technicians. The most
widely used format of LFTS is the employment of gold nanoparticles
as reporters for colorimetric detection, 18-21 which was either
qualitative or semiquantitative and generally utilized for analyzing
proteins with relatively high concentrations. In order to meet the
requirement of sensitive protein detection, more quantitative LFTS
has been developed recently using various reporters, such as
(6) Zhou, L. J.; Wang, K. M.; Tan, W. H.; Chen, Y. Q.; Zuo, X. B.; Wen, J. H.;
Liu, B.; Tang, H. X.; He, L. F.; Yang, X. H. Anal. Chem. 2006, 78, 6246–
6251.
(7) Duffy, D. C.; McDonald, J. C.; Schueller, O. J. A.; Whitesides, G. M. Anal.
Chem. 1998, 70, 4974–4984.
(8) Engvall, E.; Perlmann, P. Immunochemistry 1971, 8, 871–874.
(9) Kurita, R.; Arai, K.; Nakamoto, K.; Kato, D.; Niwa, O. Anal. Chem. 2010,
82, 1692–1697.
(10) Grubisha, D. S.; Lipert, R. J.; Park, H. Y.; Driskell, J.; Porter, M. D. Anal.
Chem. 2003, 75, 5936–5943.
(11) Millipore, C. A Short Guide: Developing Immunochromatographic Test Strips;
Millipore Corp.: Millipore, MA, 1996.
(12) Hjelle, B.; Jenison, S.; Torrez Martinez, N.; Herring, B.; Quan, S.; Polito,
A.; Pichuantes, S.; Yamada, T.; Morris, C.; Elgh, F.; Lee, H. W.; Artsob, H.;
Dinello, R. J. Clin. Microbiol. 1997, 35, 600–608.
(13) Sorell, L.; Garrote, J. A.; Acevedo, B.; Arranz, E. Lancet 2002, 359, 945–
946.
(14) Liu, G.; Lin, Y. Y.; Wang, J.; Wu, H.; Wai, C. M.; Lin, Y. Anal. Chem. 2007,
79, 7644–7653.
(15) Ho, J. A. A.; Wauchope, R. D. Anal. Chem. 2002, 74, 1493–1496.
(16) Lou, S. C.; Patel, C.; Ching, S. F.; Gordon, J. Clin. Chem. 1993, 39, 619–
624.
(17) Xu, H.; Mao, X.; Zeng, Q. X.; Wang, S. F.; Kawde, A. N.; Liu, G. D. Anal.
Chem. 2009, 81, 669–675.
(18) Shyu, R. H.; Shyu, H. F.; Liu, H. W.; Tang, S. S. Toxicon 2002, 40, 255–
258.
(19) Kim, J. H.; Cho, J. H.; Cha, G. S.; Lee, C. W.; Kim, H. B.; Paek, S. H. Biosens.
Bioelectron. 2000, 14, 907–915.
(20) Choi, D. H.; Lee, S. Ki.; Oh, Y. K.; Bae, B. W.; Lee, S. D.; Kim, S.; Shin,
Y. B.; Kim, M. G. Biosens. Bioelectron. 2010, 25, 1999–2002.
(21) Tanaka, R.; Yuhi, T.; Nagatani, N.; Endo, T.; Kerman, K.; Takamura, Y.;
Tamiya, E. Anal. Bioanal. Chem. 2006, 385, 1414–1420.
10.1021/ac101405a © 2010 American Chemical Society
Published on Web 07/19/2010

electroactive-species loaded liposome, 22 metal ion chelates, 23 and
inorganic nanoparticles 24 as well as fluorescent dyes. 25,26 Because
of its inherently high sensitivity, fluorescence-based LFTS (FLFTS)
might be the most promising for quantifying trace amount of
proteins. To date, most of the FLFTS for protein detection are
mainly based on organic fluorophores. 27,28 However, these fluorophores
have intrinsic limitations such as photobleaching, which
seriously compromises its sensitivity and confines its further
applications. Consequently, a novel fluorescent label with ideal
photostability and immense brightness is highly desirable for
FLFTS applications in protein analysis with ultralow concentrations.
During the past decades, the development of quantum dots
(Qdot) has been of considerable interest in many areas, from
molecular and cellular biology to molecular imaging and medical
diagnostics. 29-35 Compared with organic fluorophores, Qdot has
very high levels of brightness, size-tunable fluorescence emission,
narrow spectral line widths, large absorption coefficients, and
excellent stability against photobleaching. 29,30,34 These unique
characteristics have spurned intense interests in the use of Qdot
for bioassays and biosensors. On the basis of its singular optical
properties such as the superior signal brightness and high
photostability, Qdot is predicted to be a robust reporter for FLFTS.
Therefore, the employment of Qdot for FLFTS is going to be one
of the most important applications in rapid and sensitive protein
analysis. Currently, the investigations of Qdot-based FLFTS are
remaining at a very early stage, and their applications on protein
quantification have not been reported.
In the present work, we introduced Qdot as fluorescence
probes into a portable dry-reagent strip biosensor system successfully.
Due to the advantages derived from Qdot and LFTS, a
rapid, sensitive, selective, and one-step strategy has been developed
for protein analysis. Nitrated ceruloplasmin, a main biomarker
for cardiovascular disease, lung cancer, and stress response
to smoking, 36-41 was used here as target analyte to test the
performances of Qdot-based FLFTS. Experimental results dem-
(22) Yoon, C. H.; Cho, J. H.; Oh, H. I.; Kim, M. J.; Lee, C. W.; Choi, J. W.; Paek,
S. H. Biosens. Bioelectron. 2003, 19, 289–296.
(23) Lu, F.; Wang, K. H.; Lin, Y. H. Analyst 2005, 130, 1513–1517.
(24) Xia, X. H.; Xu, Y.; Zhao, X. L.; Li, Q. G. Clin. Chem. 2009, 55, 179–182.
(25) Oh, S. W.; Kim, Y. M.; Kim, H. J.; Kim, S. J.; Cho, J. S.; Choi, E. Y. Clin.
Chim. Acta 2009, 406, 18–22.
(26) Choi, S.; Choi, E. Y.; Kim, D. J.; Kim, J. H.; Kim, T. S.; Oh, S. W. Clin.
Chim. Acta 2004, 339, 147–156.
(27) Ahn, J. S.; Choi, S.; Jang, S. H.; Chang, H. J.; Kim, J. H.; Nahm, K. B.; Oh,
S. W.; Choi, E. Y. Clin. Chim. Acta 2003, 332, 51–59.
(28) Khreich, N.; Lamourette, P.; Boutal, H.; Devilliers, K.; Creminon, C.; Volland,
H. Anal. Biochem. 2008, 377, 182–188.
(29) Chan, W. C. W.; Nie, S. M. Science 1998, 281, 2016–2018.
(30) Bruchez, M.; Moronne, M.; Gin, P.; Weiss, S.; Alivisatos, A. P. Science 1998,
281, 2013–2016.
(31) Gao, X. H.; Cui, Y. Y.; Levenson, R. M.; Chung, L. W. K.; Nie, S. M. Nat.
Biotechnol. 2004, 22, 969–976.
(32) Alivisatos, A. P. Science 1996, 271, 933–937.
(33) Michalet, X.; Pinaud, F. F.; Bentolila, L. A.; Tsay, J. M.; Doose, S.; Li, J. J.;
Sundaresan, G.; Wu, A. M.; Gambhir, S. S.; Weiss, S. Science 2005, 307,
538–544.
(34) Smith, A. M.; Nie, S. M. Nat. Biotechnol. 2009, 27, 732–733.
(35) Li, Z. H.; Wang, K. M.; Tan, W. H.; Li, J.; Fu, Z. Y.; Ma, C. B.; Li, H. M.;
He, X. X.; Liu, J. B. Anal. Biochem. 2006, 354, 169–174.
(36) Shiva, S.; Wang, X.; Ringwood, L. A.; Xu, X. Y.; Yuditskaya, S.; Annavajjhala,
V.; Miyajima, H.; Hogg, N.; Harris, Z. L.; Gladwin, M. T. Nat. Chem. Biol.
2006, 2, 486–493.
(37) Pignatelli, B.; Li, C. Q.; Boffetta, P.; Chen, Q. P.; Ahrens, W.; Nyberg, F.;
Mukeria, A.; Bruske-Hohlfeld, I.; Fortes, C.; Constantinescu, V.; Ischiropoulos,
H.; Ohshima, H. Cancer Res. 2001, 61, 778–784.
(38) Radi, R. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 4003–4008.
onstrate that Qdot-based FLFTS has good ability for quantitative
analysis of trace amount of proteins and is highly expected for
portable and rapid point-of-care screening in clinical diagnostics
and other biomedical applications.
MATERIALS AND METHODS
Reagents and Materials. Human ceruloplasmin was purchased
from Genway Biotech, Inc. (San Diego, CA). Goat
antinitrotyrosine antibody was obtained from Cayman Chemical
Company (Ann Arbor, MI) while polyclonal human ceruloplasmin
antibody was ordered from Abcam Inc. (Cambridge, MA). A
nitrocellulose membrane, absorbent pad, sample pad, conjugation
pad, and backing cards were purchased from Millipore (Bendford,
MA). Casein (1%) was provided by Bio-Rad (Hercules, CA). A Qdot
585 antibody conjugation kit was obtained from Invitrogen. The
reagent components in the kit include Qdot 585 nanocrystals,
dithiothreitol (DTT) stock solution, succinimidyl trans-4-(Nmaleimidylmethyl)cyclohexane-1-carboxylate
(SMCC) stock solution,
2-mercaptoethanol, dye-labeled marker for antibody elution,
separation media, and exchange buffer. All other chemicals were
ordered from Sigma-Aldrich without further purification. All
buffers and reagent solutions were prepared with purified water,
which was produced from Barnstead Nanopure System (Kirkland,
WA).
Instruments. LM5000 Laminator, XYZ-3050 dispenser consisting
of AirJet Quanti 3000 and BioJet Quanti 3000, and Guillotine
Cutting System CM 4000 were from Biodot Ltd. (Irvine, CA). A
portable fluorescence strip reader ESE-Quant FLUO was purchased
from DCN Inc. (Irvine, CA). A bench-top Eppendorf 5804
centrifuge was obtained from Eppendorf (Hauppauge, NY). An
Ultrospec 2100 Pro UV/Visible spectrophotometer was provided
by Blochrom Ltd. (Cambridge, England).
Preparation of Qdot-Antinitrotyrosine Conjugate.
Qdot-antinitrotyrosine conjugate was prepared according to the
protocol for the Qdot antibody conjugation kit from Invitrogen.
Briefly, 125 µL of Qdot 585 nanocrystals were first activated with
14 µL of 10 mM SMCC at room temperature for 1 h, followed by
being desalted with a NAP-5 desalting column in the presence of
exchange buffer as elution solvent. The colored eluate (∼ 500
µL) was then collected into a centrifuge tube. At room temperature,
300 µL of 1 mg/mL goat antinitrotyrosine antibody was
reduced with 20 mM DTT for 0.5 h. The resulted mixture was
incubated with a dye-labeled marker and purified with a NAP-5
desalting column. The colored fraction (∼500 µL) was collected
and then mixed with the activated Qdot nanocrystals at room
temperature for 1htoform the conjugation complex, followed
by the addition of 10 µL of 10 mM 2-mercaptoethanol to quench
the reaction. The quenched conjugation mixture was then split
into two ultrafiltration devices and concentrated to ∼20 µL for
each half mixture by centrifuging at 7000 rpm on a benchtop
Eppendorf centrifuge. Following the instructions from the conjugation
kit, separation media was gently loaded into the column
and then conditioned with water and PBS, respectively. The
concentrated conjugation mixture was then pipetted into the size-
(39) Lang, J. D.; McArdle, P. J.; O’Reilly, P. J.; Matalon, S. Chest 2002, 122,
314S–320S.
(40) Turko, I. V.; Murad, F. Pharmacol. Rev. 2002, 54, 619–634.
(41) Mitrogianni, Z.; Barbouti, A.; Galaris, D.; Siamopoulos, K. C. Am. J. Kidney
Dis. 2004, 44, 286–292.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7009

Figure 1. (A) Schematic illustration of the test strip and (B1-B4) the detection of nitrated ceruloplasmin using fluorescent Qdot-based FLFTS.
(B1) Aqueous sample containing nitrated ceruloplasmin is applied to sample pad. (B2) Nitrated ceruloplasmin combines with QD-antinitrotyrosine
conjugate and also migrates along the porous membrane by capillary action. (B3) Nitrated ceruloplasmin is captured by anticeruloplasmin
antibodies immobilized on the test line. The excess Qdot conjugates continue to migrate toward the absorption pad. (B4) Fluorescence signal
of Qdot is detected using a test strip reader (solid line). As a control, ceruloplasmin without nitration cannot be recognized by Qdot-antinitrotyrosine
conjugates, so no fluorescence signal can be seen on the test strip (dotted line).
exclusion column followed by the addition of PBS. During the
elution by gravity, the first ten drops only of colored conjugate
(∼200 µL) was collected into a centrifuge tube and stored at 4 °C
until use. Following the instructions from the protocol for the Qdot
antibody conjugation kit from Invitrogen, the conjugate concentration
was determined by measuring the absorbance density of the
conjugate at 585 nm with a Ultrospec 2100 Pro UV/Visible
spectrophotometer and, then, using the formula A ) εcL, where
A is the absorbance, ε is the molar extinction coefficient, c is the
molar concentration of Qdot conjugate, and L is the path length
of the cuvette.
Nitration of Ceruloplasmin. One mg/mL Human ceruloplasmin
in phosphate buffered saline (pH 7.4) was nitrated by
bolus addition of 1 mM authentic peroxynitrite (R&D Systems,
Minneapolis, MN) according to the manufacturer’s recommendations.
The volume of added peroxynitrite was

Figure 2. Qdot-based FLFTS response for (a) 100 ng/mL, (b) 10
ng/mL, and (d) 0 ng/mL nitrated human ceruloplasmin, (c) 10 µg/mL
human ceruloplasmin served as a control.
action. After about 10 min, the cassette containing the test strip
was inserted into an ESE-Quant FLUO reader, followed by the
recording of fluorescence intensity from Qdot on the test line to
quantify the analytes. For the detection of nitrated ceruloplasmin
in human plasma, 20 times diluted plasma spiked with different
quantities of nitrated ceruloplasmin was added onto the sample
pad. The results were obtained by reading the optical response
with the strip reader after 10 min. Meanwhile, these strips after
assay were put under a UV light, and the corresponding fluorescence
images were captured directly by a Sony DSLR-A300 digital
camera.
RESULTS AND DISCUSSION
Assay Principle of Qdot-Based FLFTS. Figure 1 schematically
illustrates the configuration and measuring principle of Qdotbased
FLFTS. This FLFTS is composed of sample application pad,
conjugation pad, absorption pad, and nitrocellulose membrane
(Figure 1A). All the components were assembled onto a plastic
adhesive backing card. During the assay, aqueous sample containing
nitrated ceruloplasmin was applied onto the sample application
pad as shown in Figure 1B1. Subsequently, the analyte migrated
along the porous membrane by capillary action and then bound
with Qdot-antinitrotyrosine on the conjugation zone according
to the specific antibody-antigen interaction (Figure 1B2). The
formed complexes continued to migrate along the membrane and
were captured by anticeruloplasmin antibodies, which resulted
in the accumulation of Qdot on the test line (Figure 1B3). The
excess Qdot conjugates continue to flow into the absorption pad
to the end of the strip. Quantitative analysis was realized by
reading the fluorescence intensities of test line with a portable
strip reader (Figure 1B4). The more analyte in the sample, the
more Qdot conjugates would be captured to the test line, which
leads to the increase of fluorescence intensity. According to the
principle described above, the fluorescence intensity on the test
line would be proportional to the concentration of nitrated
ceruloplasmin in the samples.
Figure 2 shows the typical corresponding responses of Qdotbased
FLFTS to different concentrations of nitrated ceruloplasmin.
Here, human ceruloplasmin without nitration served as a control.
As shown in this figure, well-defined curves were observed in the
Figure 3. Effect of Qdot-antinitrotyrosine conjugates concentration
on the signal-to-noise ratio (S/N) of Qdot-based FLFTS for 1 µg/mL
nitrated human ceruloplasmin.
presence of nitrated ceruloplasmin and the peak area was getting
larger as the target concentration increased from 10 to 100 ng/
mL because more Qdot was captured on the test line based on
the mechanism of Qdot-based FLFTS. In contrast, the presence
of ceruloplasmin without nitration did not contribute to the signal
and exhibited a very low background. The results indicate the
great possibility of Qdot-based FLFTS for sensitive protein
detections.
Determination of Qdot Conjugate Concentration. Following
the instructions for the Qdot antibody conjugation kit from
Invitrogen, the Qdot conjugate eluting from the final column could
be determined according to the Beer-Lambert law 42 by measuring
the absorbance density of the conjugate at 585 nm:
A ) εcL
Where A is the absorbance, ε is the molar extinction coefficient
(as 400 000 M -1 cm -1 provided by Invitrogen for Qdot 585), c
is the molar concentration, and L is the path length of the
cuvette. The result shows that the Qdot conjugate has A ) 0.6
measured in a cuvette with 1 cm path length, so c ) A/ε )
0.6/400 000 ) 1.5 × 10 -6 M, which is the original concentration
of as-prepared Qdot conjugate stock solution.
ParameterOptimization.TheamountofQdot-antinitrotyrosine,
loaded on the glass fiber by physical absorption, directly affects
the fluorescence response of Qdot-based FLFTS since the signal
mainly depends on the amount of Qdot-antinitrotyrosine conjugates
captured on the test line. To probe the optimal amount of
Qdot conjugates for the assay, we diluted them into various
concentrations and investigated the influence on signal-to-noise
(S/N) ratio of the biosensor for 1 µg/mL nitrated human
ceruloplasmin. Ten µg/mL non-nitrated ceruloplasmin served here
as a control. As can be seen from Figure 3, the S/N ratio was
found to be highest for dispensing 150 nM Qdot-antinitrotyrosine.
However, the decrease of S/N at a higher concentration is resulted
from the increase of background signal due to too high of a
concentration of Qdot-antinitrotyrosine conjugate while that at
(42) Ingle, J. D. J.; Crouch, S. R. Spectrochemical Analysis; Prentice Hall: Upper
Saddle Rive, NJ, 1988.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7011

Figure 4. Comparison between blocked (solid line) and unblocked (dash line) strips for the detection of (A) 10 µg/mL nitrated ceruloplasmin
and (B) 10 µg/mL ceruloplasmin without nitration as a control.
lower concentration is ascribed to the decrease of signal due to
too low of an amount of Qdot conjugate availability. Therefore,
150 nM Qdot-antinitrotyrosine conjugate was routinely used as
the optimal concentration throughout the entire study.
Nonspecific binding is one of the likely hindrances in the
development of nanoparticle-based immunoassays. In the current
experiment, there was a fluorescence response obtained from the
control sample (ceruloplasmin without nitration), which resulted
from the nonspecific binding of Qdot conjugates on the test line.
To obtain a maximum response and a minimum nonspecific
absorption, we found that blocking the nitrocellulose membrane
with 1% casein could eliminate the influence of nonspecific binding.
Figure 4 displays the corresponding response of Qdot-based
FLFTS for 10 µg/mL ceruloplasmin with (A) and without (B)
nitration before (dashed line) and after (solid line) blocking. The
results show that the nonspecific binding was effectively reduced
after the blocking while bright fluorescence was well maintained
for the target sample. The significant removal of nonspecific
adsorption maybe attributed to the shield effect of casein, which
was successfully absorbed onto the surface of membrane pad. As
a result, the blocking with 1% casein was employed for the
following experiments.
It is important to evaluate the effect of polycolonal human
ceruloplasmin antibody concentration on the performance of this
biosensor. Consequently, we have investigated different concentrations
of ceruloplasmin antibody (10 ng/mL, 100 ng/mL, 500
ng/mL, 1 mg/mL, 2 mg/mL, and 3 mg/mL) and found that 1
mg/mL has the best response for the detection of nitrated
ceruloplasmin. Furthermore, we have studied the stability of this
biosensor by periodical testing and noticed that this test strip could
be stored at 4 °C (sealed) for at least 3 months and still maintain
good performance.
Analytical Performance of Qdot-Based FLFTS for Nitrated
Ceruloplasmin Detection. To investigate the ability of Qdotbased
FLFTS for protein sensitive quantification, the assay was
examined with different concentrations of nitrated ceruloplasmin.
The fluorescence intensity on test line was recorded and plotted
as a function of various concentrations of nitrated ceruloplasmin.
Figure 5 shows the typical response of this biosensor within 10
min for nitrated ceruloplasmin with different concentrations of 1
ng/mL, 5 ng/mL, 10 ng/mL, 40 ng/mL, 100 ng/mL, 1 µg/mL,
and 10 µg/mL, respectively. Ten µg/mL ceruloplasmin without
7012 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. Qdot-based FLFTS response for 10 µg/mL, 1 µg/mL, 100
ng/mL, 40 ng/mL, 10 ng/mL, 5 ng/mL, and 1 ng/mL nitrated ceruloplasmin
(curves a-g) and 10 µg/mL ceruloplasmin without nitration
(curve h), respectively.
nitration was used as a control. As can be seen from the figure,
well-defined peaks were observed and the peak area increased
along with the increasing of target concentration while no obvious
fluorescence signal for the control sample could be detected.
Meanwhile, trace amount of nitrated ceruloplamin as low as 1 ng/
mL could be responded by this portable biosensor with a 10 min
assay time.
Due to the superior signal brightness and high photostability
of Qdot, fluorescence imaging of this biosensor after assay could
be employed as a conventional approach to qualify or semiquantify
protein analytes visually and rapidly. By observing the fluorescence
image directly, we can easily judge the existence or not of
target proteins. As shown in Figure 6, the fluorescence band
occurred clearly in the presence of nitrated ceruloplasmin. We
observed proportional changes in fluorescence brightness of the
test lines associated with the concentration of nitrated ceruloplasmin.
Such an observation is expected because the test line
captures more Qdot conjugates when the analyte concentration
is higher. Furthermore, a red signal band from the target
concentration as low as 10 ng/mL could be easily seen even with
visual inspection. In the presence of ceruloplasmin without
nitration, no obvious fluorescence band appeared, indicating a very

Figure 6. Fluorescence imaging of Qdot-based FLFTS for (A) 10 µg/mL, (B) 1 µg/mL, (C) 100 ng/mL, and (D) 10 ng/mL nitrated ceruloplasmin
and (E) 10 µg/mL ceruloplasmin without nitration. The bottom curves are the corresponding readout using a strip reader.
low nonspecific absorption of this biosensor. Accordingly, welldefined
peaks were recorded by the strip reader shown on the
bottom of Figure 6. Consequently, the strip reader and fluorescence
imaging can be combined together to show the assay results
of Qdot-based FLFTS with high sensitivity and specificity. Either
of them could also be used alone depending on the conditions
and requirements. By employing the dual reading approaches,
the proposed sensing platform will be a universal and simple
strategy for complicated protein analysis.
Determination of Nitrated Ceruloplasmin in Human
Plasma. To explore the feasibility of Qdot-based FLFTS for
clinical application, the device was then applied to detect nitrated
ceruloplasmin spiked in 20-fold diluted human plasma with
different concentrations such as 10 µg/mL, 1 µg/mL, 100 ng/
mL, 10 ng/mL, and 1 ng/mL, respectively. Ten µg/mL nonnitrated
ceruloplasmin served as control. These samples were
applied to Qdot-base FLFTS, and the fluorescence signals were
recorded by test strip reader and digital camera after 10 min. A
good calibration curve was obtained in a wide range as showed
in Figure 7. The detection limit was 0.4 ng/mL (S/N ) 3), which
is calculated as the concentration corresponding to 3 times the
SD (standard deviation) of the background signal. Error bars are
based on six duplicated measurements of nitrated ceruloplasmin
at different concentrations, and the control was also run six
replicates. Considering 20-fold dilution of plasma sample during
the assay, the detection limit of 0.4 ng/mL is equivalent to 8 ng/
mL for undiluted plasma, which is comparable to the value
indicated by other techniques for nitrated protein detection such
as nitrated fibrinogen and BSA. 43,44 Regarding the detection limit
(43) Franze, T.; Weller, M. G.; Niessner, R.; Poschl, U. Analyst 2003, 128, 824–
831.
Figure 7. Qdot-based FLFTS linear response for 10 µg/mL, 1 µg/
mL, 100 ng/mL, 10 ng/mL, and 1 ng/mL nitrated ceruloplasmin in
human plasma.
of this biosensor and the high concentration (0.5%, i.e., 2.27 µM)
of ceruloplasmin in human plasma, 45 the sensitivity of this assay
is sufficient to detect as low as 0.03% nitration of ceruloplasmin
in human plasma samples even if only one nitration site is present
in the protein. Meanwhile, fluorescence images were also easily
observed as shown in Figure 8, where the fluorescence band
occurred clearly in the presence of nitrated ceruloplasmin and
almost no fluorescence band came up for non-nitrated ceruloplasmin.
Furthermore, the fluorescence band on the test line for 10
ng/mL nitrated ceruloplasmin in human plasma could be directly
(44) Tang, Z.; Wu, H.; Du, D.; Wang, J.; Wang, H.; Qian, W.; Bigelow, D. J.;
Pounds, J. G.; Smith, R. D.; Lin, Y. Talanta 2010, 81, 1662–1669.
(45) Scheinberg, I. H.; Catlin, D. Science 1952, 116, 484–485.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7013

Figure 8. Fluorescence imaging of Qdot-based FLFTS for (A) 100
ng/mL and (B) 10 ng/mL nitrated ceruloplasmin and (C) 10 µg/mL
ceruloplasmin without nitration in human plasma.
viewed by naked eyes, which is equivalent to 200 ng/mL for an
undiluted plasma sample. Successfully detecting the spiked human
plasma samples using a strip reader or fluorescence imaging
displays the promise of Qdot-based FLFTS for various clinical
applications in the near future.
CONCLUSIONS
In summary, Qdot as a promising alternative reporter was
successfully integrated with lateral flow tests strip and first
developed for rapid, sensitive, and one-step quantitative detection
of a trace amount of nitrated ceruloplasmin. This portable
7014 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
biosensor takes advantage of the speed and low cost of conventional
immunochromatographic strip as well as high sensitivity
and photostability of Qdot-based fluorescent immunoassay. Under
optimal conditions, this proposed Qdot-based FLFTS is capable
of detecting a minimum of 1 ng/mL nitrated ceruloplasmina within
10 min. Furthermore, the linear relationship between peak area
and the logarithm of target concentration was observed in the
range of 1 ng/mL to 10 µg/mL with a detection limit of 0.4 ng/
mL in a spiked plasma sample, which is equivalent to 8 ng/mL
for undiluted plasma. Moreover, the presence of non-nitrated
ceruoloplasmin showed no effect on the biosensor response,
illustrating the good selectivity. Overall, the Qdot-based FlFTS,
considered as an advance in alternative immunosensors, has a
great potential for rapid, sensitive, and portable analysis of other
protein biomarkers in clinical diagnostics, basic discovery, and a
variety of other biomedical applications.
ACKNOWLEDGMENT
The work was done at Pacific Northwest National Laboratory
(PNNL) supported by Grant U54 ES16015 from the National
Institute of Environmental Health Sciences (NIEHS), NIH. Its
contents are solely the responsibility of the authors and do not
necessarily represent the official views of the federal government.
PNNL is operated by Battelle for DOE under Contract DE-AC05-
76RL01830.
Received for review May 28, 2010. Accepted July 1, 2010.
AC101405A

Anal. Chem. 2010, 82, 7015–7020
Selection of Column Dimensions and Gradient
Conditions to Maximize the Peak-Production Rate
in Comprehensive Off-Line Two-Dimensional Liquid
Chromatography Using Monolithic Columns
Sebastiaan Eeltink,* ,† Sebastiaan Dolman, ‡,⊥ Gabriel Vivo-Truyols, § Peter Schoenmakers, §
Remco Swart, ‡ Mario Ursem, ‡ and Gert Desmet †
Department of Chemical Engineering, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels, Belgium, Dionex
Corporation, Abberdaan 114, 1046 AA Amsterdam, The Netherlands, and Van ’t Hoff Institute for Molecular Sciences,
University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands
The peak-production rate (peak capacity per unit time)
in comprehensive off-line two-dimensional liquid chromatography
(LC/×/LC) was optimized for the separation
of peptides using poly(styrene-co-divinylbenzene) monolithic
columns in the reversed-phase (RP) mode. A firstdimension
( 1 D) separation was performed on a monolithic
column operating at a pH of 8, followed by
sequential analysis of all the 1 D fractions on a monolithic
column operating at a pH of 2. To obtain the
highest peak-production rate, effects of column length,
gradient duration, and sampling time were examined.
RP/×/RP was performed at undersampling conditions
using a short 10 min 1 D gradient. The peak-production
rate was highest using a 50 mm long 2 D column
applying an 8-10 min 2 D gradient time and was almost
a factor of two higher than when a 250 mm monolithic
column was used. The best way to obtain a higher peakproduction
rate in off-line LC/×/LC proved to be an
increase in the number of 1 D fractions collected. Increasing
the 2 D gradient time was less effective. The potential
of the optimized RP/×/RP method is demonstrated by
analyzing proteomics samples of various complexities.
Finally, the trade-off between peak capacity and analysis
time is discussed in quantitative terms for both onedimensional
RP gradient-elution chromatography and the
off-line two-dimensional (RP/×/RP) approach. At the
conditions applied, the RP/×/RP approach provided a
higher peak-production rate than the 1 D-LC approach
when collecting three 1 D fractions, which corresponds
to a total analysis time of 60 min.
Porous polymer monolithic columns have emerged in the
1990s 1,2 and have become a viable alternative for packed-column
* Corresponding author. Tel.: +32 (0)2 629 3324. Fax: +32 (0)2 629 3248.
E-mail: seeltink@vub.ac.be.
† Vrije Universiteit Brussel.
‡ Dionex Corporation.
§ University of Amsterdam.
⊥ Present address: Bruker Biosciences, 1/28A Albert Street, Preston VIC 3072,
Australia.
(1) Hjertén, S; Liao, J.-L; Zhang, R. J. Chromatogr., A 1989, 473, 273–275.
technology for the gradient-elution separation of peptides and
proteins. 3-5 Macroporous polymer monolithic materials were
initially developed by Svec et al. in large I.D. column formats via
a molding process. 6,7 Macroporous poly(styrene-co-divinylbenzene)
(PS-DVB) monolithic rods were operated at high flow rates
(up to 25 mL/min) for the gradient separations of proteins. 7 This
group also demonstrated that the porous properties could be
controlled by optimizing the polymerization mixture, i.e., type and
composition of the porogenic solvent, and the percentage of
difunctional monomer (“cross-linker”) in the mixture. 8,9 Huber
and co-workers developed styrene-based monoliths in situ in 200
µm I.D. capillary columns and applied these columns for liquidchromatographic
mass-spectrometric (LC-MS) analysis of samples
from proteomics and genomics studies, including peptides,
proteins, single-stranded oligonucleotides, and double-stranded
DNA fragments. 10-12 Excellent separation performance was, for
example, demonstrated by a baseline separation of phosphorylated
oligodeoxyadenylic acids, ranging in size from the 12-mer to the
60-mer, and yielded peak widths of only 5.7 s when applying a
7.5 min gradient, 13 resulting in a peak capacity of approximately
80. Recently, we reported the use of a1mlong poly(styrene-codivinylbenzene)
monolithic column yielding a peak capacity in
excess of 1000 for a peptide separation when applying a 600 min
gradient. 14
(2) Peters, E. C.; Petro, M.; Svec, F; Frechet, J. M. J. Anal. Chem. 1997, 69,
3646–3649.
(3) Ivanov, A. R.; Zang, L.; Karger, B. L. Anal. Chem. 2003, 75, 5306–5316.
(4) Geiser, L.; Eeltink, S.; Svec, F.; Frechet, J. M. J. J. Chromatogr., A 2008,
1188, 88–96.
(5) Levkin, P. A.; Eeltink, S.; Stratton, T. R.; Brennen, R.; Robotti, K.; Killeen,
K.; Svec, F.; Frechet, M. J. M. J. Chromatogr., A 2008, 1200, 55–61.
(6) Svec, F.; Frechet, J. M. J. Anal. Chem. 1992, 54, 820–822.
(7) Wang, Q. C.; Svec, F.; Frechet, J. M. J. Anal. Chem. 1993, 65, 2243–2248.
(8) Viklund, C.; Svec, F.; Frechet, J. M. J. Chem. Mater. 1996, 8, 744–750.
(9) Eeltink, S.; Geiser, L; Svec, F; Frechet, J. M. J. J. Sep. Sci. 2007, 30, 407–413.
(10) Toll, H.; Wintringer, R.; Schweiger-Hufnagel, U.; Huber, C. G. J. Sep. Sci.
2005, 28, 1666–1674.
(11) Holzl, G.; Oberacher, H.; Pitsch, S.; Stutz; Huber, C. G. Anal. Chem. 2005,
77, 673–680.
(12) Walcher, W.; Toll, H.; Ingendoh, A.; Huber, C. G. J. Chromatogr., A 2004,
1053, 107–117.
(13) Oberacher, H.; Mayr, B. M.; Huber, C. G. J. Am. Soc. Mass Spec. 2004,
15, 32–42.
(14) Eeltink, S.; Dolman, S.; Detobel, F.; Swart, R.; Ursem, M.; Schoenmakers,
P. J. J. Chromatogr., A 2010, in press.
10.1021/ac101514d © 2010 American Chemical Society 7015
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/28/2010

For the analysis of complex sample mixtures, as for example
encountered in proteomics research, the peak capacities that can
be realized with one-dimensional ( 1 D) LC often do not suffice to
achieve complete separation of all compounds. Multidimensional
separation approaches have the potential to separate
thousands of components within a reasonable time. In first
approximation, the maximum peak capacity in two-dimensional
LC is the product of the peak capacities of the individual
dimensions, provided that the two retention mechanisms are
orthogonal. 15 Comprehensive two-dimensional liquid chromatography,
where all essential fractions of the sample are being
analyzed in both dimensions, can be subdivided into two main
categories, i.e., online comprehensive two-dimensional LC (or
LC×LC) and off-line comprehensive two-dimensional LC (or
LC/×/LC). 16
In LC×LC, column dimensions must be selected such that the
eluent composition and transfer volume match the LC conditions
used in the second dimension. Schoenmakers showed that the
maximum flow rate in the first-dimension and, consequently, the
diameter of the 1 D column depends on the maximum injection
volume in the second dimension and on the second-dimension
analysis time. 17 The optimization of sampling time in online
two-dimensional LC has recently been reviewed by Guiochon
et al. 18 When the sampling rate is too low, resolution achieved
in the first dimension is partially lost. However, more time is
available to achieve a high peak capacity in the second
dimension. The fraction of potential peak capacity of the twodimensional
combination of columns that is lost due to the
selection of a too small modulation frequency is described by
the “Nobuo factor” (modulation efficiency). 19 Typically, for
LC×LC, two cuts per first-dimension peak provides the best
trade-off between the loss of resolution in the first dimension
and the analysis time in the second dimension. 19,20 However,
from a practical point of view, this is difficult to achieve since
modulation-phase effects and the random sampling process must
be considered.
The off-line approach (LC/×/LC) offers more flexibility for the
selection of column dimensions and elution conditions, since the
second-dimension analysis time can be optimized independently
from the first-dimension sampling time. In addition, the organic
modifier can be evaporated, so that fractions can be concentrated
or dissolved in a different solvent prior to reinjection. 18,21 To
enhance the preconcentration of peptides or proteins on a trap
column, a strong ion-pairing agent can be added prior to the
second-dimension separation. 21 As a result, the flow rates, transfer
volumes, and the compatibility of (first and second dimension)
eluent composition are less critical issues in LC/×/LC. The
LC/×/LC setup can be optimized for proteomics applications. A
large I.D. first-dimension column can be used, which provides
(15) Giddings, J. C. Anal. Chem. 1984, 65, 1258A–1270A.
(16) Schoenmakers, P. J.; Marriott, P.; Beens, J. LC-GC Eur. 2003, 16, 335–
339.
(17) van der Horst, A.; Schoenmakers, P. J. J. Chromatogr., A 2003, 1000, 693–
709.
(18) Guiochon, G.; Marchetti, M.; Mriziq, K.; Shalliker, R. A. J. Chromatogr., A
2008, 1189, 109–168.
(19) Horie, K.; Kimura, K.; Ikegami, T.; Iwatsuka, A.; Saad, N.; Fiehn, O.; Tanaka,
N. Anal. Chem. 2007, 79, 3764–3770.
(20) Davis, J. M.; Stoll, D. R.; Carr, P. W. Anal. Chem. 2008, 80, 461–473.
(21) Eeltink, S.; Dolman, S.; Swart, R.; Ursem, M.; Schoenmakers, P. J.
J. Chromatogr., A 2009, 1216, 7368–7374.
7016 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
sufficient loadability for the analysis of samples with a broad
dynamic range. A small I.D. column can be applied in the second
dimension, minimizing chromatographic dilution and maximizing
ionization efficiency when coupling LC with mass spectrometry
via an electrospray interface. Huber and co-workers compared the
sequence coverage obtained after an LC/×/LC-MS/MS separation
of a tryptic digest of C. glutamicum using either a strong cationexchange
(SCX) column or a reversed-phase (RP) column operated
at high pH in the first dimension and a reversed-phase
monolith operated at low pH in the second dimension. 22,23 They
observed that the SCX/×/RP and RP/×/RP approaches complemented
each other. Recently, our group compared the 1 D-LC
performance of a 50 mm long monolithic column with that of
an LC/×/LC approach employing a weak-anion exchange and
an RP monolithic column in the first and second dimensions,
respectively. 24 At the conditions applied, the WAX/×/RP
approach provided a better peak-capacity-per-analysis-time ratio
for separations requiring a peak capacity of 400 or higher.
The present contribution discusses how to maximize the peakproduction
rate in off-line comprehensive two-dimensional liquid
chromatography for the reversed-phase separations (RP/×/RP)
of peptides performed at high and low pH using monolithic column
technology. The effects of the first-dimension ( 1 D) column
dimensions (length and diameter) and LC conditions (flow rate
and gradient time) on peak width and sampling volume are
discussed. In addition, the effects of 2 D column length and
gradient time on peak-production rate in 2 D-LC are demonstrated
at “undersampling” conditions (i.e., containing fewer
fractions than required to essentially maintain the first-dimension
separation). Finally, the potential of RP/×/RP is demonstrated
with separations of proteomic samples of varying
complexity.
EXPERIMENTAL SECTION
Chemicals and Materials. Acetonitrile (ACN, HPLC supragradient
quality), heptafluorobutyric acid (HFBA, ULC/MS quality),
and trifluoroacetic acid (TFA, ULC/MS quality) were purchasedfromBiosolve(Valkenswaard,TheNetherlands).Ammonium
bicarbonate (min. 99%), dithiothreitol (min. 99%), iodoacetic acid
(approximately 99%), guanidine-HCl, sodium chloride (analytical
reagent grade), cytochrome c (bovine heart), and apo-transferrin
(bovine, g98%) were purchased from Sigma-Aldrich (Steinheim,
Germany). Lysozyme (hen egg white), alcohol dehydrogenase
(yeast), serum albumin (bovine, assay >96%), �-galactosidase, and
sodium phosphate monobasic dihydrate (analytical reagent grade)
were obtained from Fluka (Buchs, Switzerland). Escherichia coli
(E. coli, strain K12) protein sample (lyophilized) was obtained
from Bio-Rad Laboratories (Veenendaal, The Netherlands).
Preconcentration and desalting of peptides prior to the analytical
separation was performed ona5mm× 0.2 mm I.D. monolithic
trap column (Pepswift RP, Dionex Benelux, Amsterdam, The
Netherlands). HPLC separations were performed with 50 mm ×
0.2 mm and 250 mm × 0.2 mm monolithic PepSwift RP columns
(22) Delmotte, N.; Lasaosa, M.; Tholey, A.; Heinzle, E; Huber, C. G. J. Proteome
Res. 2007, 6, 4363–4373.
(23) Toll, H.; Oberacher, H.; Swart, R.; Huber, C. G. J. Chromatogr., A 2005,
1079, 274–286.
(24) Eeltink, S.; Dolman, S.; Detobel, F.; Desmet, G.; Swart, R.; Ursem, M. J.
Sep. Sci. 2009, 32, 2504–2509.

Table 1. Effect of 1 D-LC Conditions on Sampling Time and Number of Fractions Collected When Applying a
Modulation Ration (MR) ∼2 per 1 D Peak a
I.D. (mm) flow rate (µL/min) gradient time (min) 4σ peak width (s) sampling time (s) sampling volume (µL) no. fractions
0.2 2 10 5.5 2.8 0.09 214
1 60 10 5.8 2.6 2.4 231
a
Column length is 50 mm; 1 µL injection of the six-protein digest; aqueous acetonitrile gradient from 1-26% (ACN) at pH ) 8; detection at 214
nm.
and with a 50 mm × 1 mm monolithic ProSwift RP-10R column
(Dionex).
Preparation of Tryptic Digests. The six-protein mixture
prepared from transferrin, bovine serum albumin, �-galactosidase,
alcohol dehydrogenase, lysozyme, cytochrome, and E. coli proteins
were digested according the following procedure. Proteins
were reduced for1hat60°C in the presence of 7 mol/L guanidine
and 1 mol/L dithiothreitol, followed by alkylation for 30 min at
room temperature by adding 1 mol/L iodoacetic acid. To consume
any unreacted iodoacetic acid, 1 mol/L dithiothreitol was added.
The reduced and alkylated proteins were then dialyzed against
50 mM ammonium bicarbonate (pH 8) for 24 h in dialysis sacks
(Sigma) with a cutoff 0.997).
Therefore, the relationship 2 wb ) a + b( 2 tG) can be established,
which holds at a constant length. After fitting, the a and b
parameters were used to predict the peak capacity as a function
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7017

Figure 1. Effect of gradient time on peak width (A) and peak capacity
(B) using a 250 mm × 0.2 mm (closed symbols) and a 50 mm × 0.2
mm (open symbols) monolithic column. Sample: six-protein digest
(transferrin, bovine serum albumin, �-galactosidase, alcohol dehydrogenase,
lysozyme, and cytochrome c), 1 µL injection (0.5 pmol/
µL); flow rate: 2 µL/min; aqueous acetonitrile gradient from 1% to
35% with 0.05% TFA ion-pairing agent; column temperature: 60 °C.
Detection at 214 nm using a3nLflowcell.
of tG (Figure 1B). For gradient times below 60 min, the peak
capacity of the 50 mm long monolith was higher than that of the
250 mm monolith. This can be explained by the difference in
morphology of the two monoliths. 14
At undersampling conditions, the total peak capacity ( 2D nc) and
total analysis time (ttot) in off-line two-dimensional LC are given
by
1
tG
2D
nc )
st t tot ) 1 t +
×
1 tG
s t
2 tG
2 W
(1)
· 2 t (2)
where 2 tG is the second-dimension gradient time, 2 W is the
average second-dimension peak width (equivalent to 4σ), 1 t is
the first-dimension analysis time, 2 t is the second-dimension
analysis time, and st is the sampling time. The first-dimensional
analysis time is the sum of the column holdup time ( 1 t0), the
dwell time ( 1 tdwell), and the gradient time ( 1 tG). The seconddimension
analysis time is the sum of 2 tdesalt, 2 t0, 2 tdwell, a wash
step ( 2 twash), and the second-dimension column equilibration
time ( 2 teq), which corresponds to the time needed to flush the
column with three column volumes to obtain good retention
time stability. The peak production rate (�, min -1 ) is defined
as
7018 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
� )
2D nc
Figure 2 illustrates the effect of the 2 tG on the � using a 50 mm
and 250 mm long monolithic column. The 1 D RP separation
was performed using a 50 mm × 1 mm monolithic column,
applying a 10 min gradient at pH ) 8. The 2 D analysis included
a preconcentration and desalting step on a monolithic trap
column ( 2 tdesalt) and a RP gradient separation at pH ) 2onthe
capillary column. Initially, a steep increase in � can be observed.
In this region, � is dominated by the contributions of 2 t0 and
2 teq time to the total analysis time in the second dimension.
With increasing gradient time, the peak capacity initially
strongly increases (Figure 1B) and the peak-production rate
reaches a maximum. When applying even longer gradients, the
peak-production rate decreases. This is because the peak capacity
marginally increases with the gradient time at longer gradient
duration (see Figure 1B). The existence of a maximum can also
be demonstrated by taking the derivative in eq 3 with respect to
2 tG. Forcing this derivative to be 0 yields
2 tG,max ) � a
t tot
b� s t(
1 td
1 tG
+ 1) + 2 t d
where 2 tG,max refers to the gradient time in the second dimension
that yields the highest value for �. 1 td ) 1 t0 + 1 tdwell, and 2 td )
2 tdesalt + 2 t0 + 2 tdwell + 2 twash + 2 teq. The maximum peakproduction
rate using the 50 mm long monolithic column is
obtained at a 2 tG ∼ 9 min, and the maximum shifts to a higher
Figure 2. Effect of 2 D gradient time on peak-production rate using
250 mm (a) and a 50 mm (b) long second-dimension monolithic
column with a sampling time of 30 s (solid line), 50 mm column with
sampling time of 60 s (- --; c), and 50 mm column with sampling
time of 120 s (---; d). First-dimension separation on a 50 mm × 1
mm long column, 1 D delay time is 4 min, 10 min 1 D gradient time. 1 D
gradient from 1% to 26% acetonitrile at pH ) 8 (10 mM aqueous
ammonium carbonate buffer) at a flow rate of 60 µL/min. Seconddimension
separation on a 250 mm column operating at a flow rate
of 2 µL/min included a 1 min preconcentration, delay time of 0.4 min,
t0 time of 3.53 min, 0.5 min wash step, and 10.6 min equilibration
time; 2 D gradient from 1% to 35% with 0.05% TFA ion-pairing agent.
Second-dimension separation on a 50 mm column operating at a
flow rate of 2 µL/min included a 1 min preconcentration, delay time
of 0.4 min, t0 time of 1.36 min, 0.5 min wash step, and 4.1 min
equilibration time; 2 D gradient from 1% to 35% with 0.05% TFA ionpairing
agent.
(3)
(4)

Figure 3. Peak-production rate versus total analysis time when either
a 50 mm (closed symbols) or a 250 mm (open symbols) long 2 D
monolithic column is used at optimal 2 tG and two or more fractions
are collected. LC conditions described in Figure 2.
value (20 min) when the 250 mm long monolith is used. This
is because the a/b ratio increases with the column length (see
Figure 1A); the slope of 2 tG vs 2 wb is lower for the 250 mm
column, which implies that the b is lower. Also, the a decreases
with column length.
In addition, the � of the 50 mm long monolith is more than a
factor of 2 higher than that of the 250 mm monolith. This is due
to two effects. First, the region of 2 tG considered (from 1 to 60
min) contemplates situations in which the 250 mm column
yields broader peaks than the 50 mm column (see Figure 1A),
reducing 2 nc for the 250 mm column compared to the 50 mm
monolith for a short gradient duration (Figure 1B). Second, with
increasing column length, t0 and teq increase, affecting 2 td and,
hence, the � (see eq 3). Figure 2 also shows that the optimal 2 tG
depends on the sampling rate. This is clearly perceived from
eq 4, where 2 tG,max depends on st. The higher st, the higher the
value of 2 tG,max.
Effect of Sampling Time on Peak-Production Rate. Figure
3 illustrates the � and total analysis time when either a 50 or a
250 mm long 2 D monolithic column is used at optimal 2 tG and
two or more fractions are collected.
When a 50 mm 2 D column is used, first, a strong increase
in � is observed. At higher sampling rates, the increase levels
off and � tends to reach a maximum around 10 peaks/min.
When a 250 mm 2 D column is used, � increases slightly from
4.5 peaks/min (2 fractions) to a maximum of 5 peaks/min after
sampling only three fractions. Apparently, when a 50 mm long
2 D column is used and a slow sampling rate is applied, � is
significantly affected by the contribution of the 1 D analysis time
to the total analysis time. With decreasing sampling time or
when longer 2 D columns (longer t0 and teq) are used, � is
dominated by the 2 D analysis time.
Figure 4 shows the RP(pH)8)/×/RP(pH)2) separation of a sixprotein
digest when applying a sampling time of 60 and 30 s,
respectively. The 1 D separation was performed using a 50 mm
× 1 mm monolithic column applying a gradient time of 10 min.
The 2 D separation was executed on a 50 mm × 0.2 mm long
monolithic column, applying a 2 tG of 7.5 min. The maximum
theoretical peak capacity of the separation shown in Figure 4A
is 1200, and the separation was completed in 98 min. The
Figure 4. RP (pH ) 8)/×/RP (pH ) 2) separation of a digest of six
proteins showing the effect of sampling time (st ) 60 s (A); st ) 30 s
(B)) on peak capacity and total analysis time. 50 mm × 1mm 1 D
column and 50 mm × 0.2 mm 2 D column. Further conditions as
described in Figure 2.
Figure 5. Off-line 2 D-LC separation of an E. coli digest using a
sampling time of 15 s. 50 mm × 1mm 1 D column and 50 mm × 0.2
mm 2 D column. Further conditions as described in Figure 2.
separation in Figure 4B yielded a maximum 2 D-LC peak capacity
of 2400 in 164 min. Whereas the theoretical peak capacity
doubles, the total analysis time increased only by 60%. It should
be noted that the separation space was not completely filled.
This is because the retention mechanisms of the reversedphase
separations performed at pH ) 8 and 2, respectively,
are not completely independent. As a consequence, the fraction
of the total peak capacity that is actually used (“sample peak
capacity”) is lower than the theoretical peak capacity.
For the analysis of more complex samples, such as an E. coli
digest, the most productive way to obtain a higher peak capacity
in LC/×/LC, while still working at undersampling conditions, is
to increase the number of 1 D fractions collected; see Figure 2.
Increasing 2 tG is less effective. Figure 5 shows the LC/×/LC
separation of E. coli digest. The sampling rate (15 s) was selected
such that a maximum theoretical peak capacity of 6700 could be
achieved within a total analysis time of 740 min.
Trade-Off between 1 D-LC and the Optimized LC/×/LC
Approach. Figure 6 illustrates the trade-off between the 1 D-LC
system using a 50 mm and 250 mm long monolithic column
and the RP(pH)8)/×/RP(pH)2) using a 50 mm long monolithic
column in each dimension. The total 1 D-LC analysis time is
the sum of the desalting time (0.5 min), the column holdup
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7019

Figure 6. Trade-off between 1 D-LC performance using 50 mm (solid
line) and 250 mm (dotted line) long monolithic columns and the
optimized off-line 2 D-LC approach collecting g2 fractions (solid
symbols). 1 D-LC conditions described in Figure 1; 2 D-LC conditions
described in Figure 2.
time, the dwell time, and the gradient time. The lines depict
the 1 D-LC performance. The closed circles represent the
RP(pH)8)/×/RP(pH)2) performance collecting discrete numbers
of fractions (two and more). It is evident that below 45 min
the highest � is obtained using 50 mm long monolithic columns
in the 1 D-LC mode. This yields a maximum peak capacity of
320. At a total analysis time of 70 min, 50 mm and 250 mm
long monolithic columns show comparable performance. For
1 D-LC analysis longer than 70 min, the 250 mm monolithic
column provides the �. The maximum 1 D-LC peak capacity of
475 can be achieved within 3 h. However, the � is much smaller
than what can be realized by LC/×/LC. A peak capacity of 480
7020 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
can already be achieved in 60 min using LC/×/LC when
analyzing only three 1 D fractions. The peak capacity linearly
increases with the number of fractions collected and approximately
doubles every 60 min.
CONCLUDING REMARKS
An approach to select column dimensions and LC conditions
in LC/×/LC to maximize � have been demonstrated. Although
the LC/×/LC optimization strategy has been performed with
monolithic columns, this approach can be used when applying
other types of LC columns. At undersampling conditions, the 1 D
separation performance is subsidiary of sampling rate. As a
result, a short first-dimension analysis time is optimal. This
can be achieved using short 1 D column length and applying a
short 1 tG. The optimal 2 tG depends on 2 D column length and
on sampling rate. The optimal 2 tG shift to lower values when
shorter 2 D columns are used yields higher values for �.
For separations requiring a maximum peak capacity up to 340,
1 D-LC using a 50 mm long monolithic column was found to be
superior to LC/×/LC in terms of analysis time. For more
demanding separations, 250 mm long monoliths can be used,
applying longer gradient times (nc, max = 475). However, the �
in LC/×/LC is superior for an analysis taking longer than 60
min (collecting three fractions and more), and the achievable
peak capacity increases linearly with analysis time.
ACKNOWLEDGMENT
Support of this work by a grant of the Research Foundation
Flanders (G.0919.09) is gratefully acknowledged.
Received for review June 8, 2010. Accepted July 17, 2010.
AC101514D

Anal. Chem. 2010, 82, 7021–7026
Cell-Free Expression of Soluble and Membrane
Proteins in an Array Device for Drug Screening
Ruba Khnouf, † Daniel Olivero, ‡ Shouguang Jin,* ,§ Matthew A. Coleman,* ,| and Z. Hugh Fan* ,†,‡
Department of Biomedical Engineering, University of Florida, P.O. Box 116131, Gainesville, Florida 32611, Department of
Mechanical and Aerospace Engineering, University of Florida, P.O. Box 116250, Gainesville, Florida 32611, Department of
Molecular Genetics and Microbiology, University of Florida, P.O. Box 100266, Gainesville, Florida 32610, and Lawrence
Livermore National Laboratory, 7000 East Avenue, Livermore, California 94551
Enzymes and membrane protein receptors represent
almost three-quarters of all current drug targets. As a
result, it would be beneficial to have a platform to produce
them in a high-throughput format for drug screening. We
have developed a miniaturized fluid array device for cellfree
protein synthesis, and the device was exploited to
produce both soluble and membrane proteins. Two membrane-associated
proteins, bacteriorhodopsin and ApoA
lipoprotein, were coexpressed in an expression medium
in the presence of lipids. Simultaneous expression of
ApoA lipoprotein enhanced the solubility of bacteriorhodopsin
and would facilitate functional studies. In addition,
the device was employed to produce two enzymes, luciferase
and �-lactamase, both of which were demonstrated
to be compatible with enzyme inhibition assays.
�-lactamase, a drug target associated with antibiotic
resistance, was further used to show the capability of the
device for drug screening. �-Lactamase was synthesized
in the 96 units of the device and then assayed by a range
of concentrations of four mock drug compounds without
harvesting and purification. The inhibitory effects of these
compounds on �-lactamase were measured in a parallel
format, and the degree in their drug effectiveness agreed
well with the data in the literature. This work demonstrated
the feasibility of the use of the fluid array device
and cell-free protein expression for drug screening, with
advantages in less reagent consumption, shorter analysis
time, and higher throughput.
There is an escalating need to discover new drug targets for
various diseases that impact the quality of life of countless
individuals. Once a target is validated, it is necessary to find drug
candidates by screening a large library of compounds. For both
target discovery and drug screening, novel platforms such as highthroughput
screening (HTS) are required in the modern pharmaceutical
era. 1-3
* To whom correspondence should be addressed. E-mail: hfan@ufl.edu
(Z.H.F.). Fax: 1-352-392-7303 (Z.H.F.).
† Department of Biomedical Engineering, University of Florida.
‡ Department of Mechanical and Aerospace Engineering, University of Florida.
§ Department of Molecular Genetics and Microbiology, University of Florida.
| Lawrence Livermore National Laboratory.
(1) Drews, J. Science 2000, 287, 1960–1964.
(2) Carnero, A. Clin. Transl. Oncol. 2006, 8, 482–490.
Although cell-based HTS has become increasingly popular due
to its ability to deliver phenotype information at the cellular level,
there is still a considerable need to have innovative platforms that
provide biological and chemical information at the molecular
level. 4 Cell-based screening can result in “off target hits” because
different portions of a cellular process other than the target could
be affected by library compounds. 5 In addition, those compounds
that cause cytotoxicity would result in reduced signal even though
they do not inhibit the desired target. 5 Galarneau et al. showed a
great example that both cell-free and cell-based assays were
required to provide complementary and indispensible information
for their study in protein-protein interactions. 4
Cell-free protein synthesis (CFPS) is a platform technology that
addresses the shortcomings mentioned above for cell-based
assays. CFPS utilizes an efficient, often-coupled transcription and
translation reaction to produce recombinant proteins. 6-12 It
eliminates the time-consuming steps of conventional cell-based
protein production, including transformation, cell culture maintenance,
and expression optimization. The proteins synthesized
could be used for functional and structural studies, drug screening,
and other applications. Since there is no cellular control mechanism,
CFPS will not suffer from cytotoxicity or other challenges
associated with cell-based protein expression (e.g., inclusion
body). 6 A large number of soluble proteins have been synthesized
using CFPS, and many of them are enzymes and functionally
active. 13
More importantly, CFPS is an open system and, thus, can be
modified by simply adding various components. This open nature
has been used for producing membrane proteins. Lipids, liposome,
and lipoprotein particles have been introduced into the cell-free
(3) Pereira, D. A.; Williams, J. A. Br. J. Pharmacol. 2007, 152, 53–61.
(4) Galarneau, A.; Primeau, M.; Trudeau, L. E.; Michnick, S. W. Nat. Biotechnol.
2002, 20, 619–622.
(5) von Ahsen, O.; Bomer, U. ChemBioChem 2005, 6, 481–490.
(6) Spirin, A. S.; Baranov, V. I.; Ryabova, L. A.; Ovodov, S. Y.; Alakhov, Y. B.
Science 1988, 242, 1162–1164.
(7) Kigawa, T.; Yabuki, T.; Yoshida, Y.; Tsutsui, M.; Ito, Y.; Shibata, T.;
Yokoyama, S. FEBS Lett. 1999, 442, 15–19.
(8) Madin, K.; Sawasaki, T.; Ogasawara, T.; Endo, Y. Proc. Natl. Acad. Sci.
U.S.A. 2000, 97, 559–564.
(9) Shimizu, Y.; Inoue, A.; Tomari, Y.; Suzuki, T.; Yokogawa, T.; Nishikawa,
K.; Ueda, T. Nat. Biotechnol. 2001, 19, 751–755.
(10) Angenendt, P.; Nyarsik, L.; Szaflarski, W.; Glokler, J.; Nierhaus, K. H.;
Lehrach, H.; Cahill, D. J.; Lueking, A. Anal. Chem. 2004, 76, 1844–1849.
(11) Jewett, M. C.; Swartz, J. R. Biotechnol. Prog. 2004, 20, 102–109.
(12) Katzen, F.; Chang, G.; Kudlicki, W. Trends Biotechnol. 2005, 23, 150–156.
(13) Spirin, A. S. Trends Biotechnol. 2004, 22, 538–545.
10.1021/ac1015479 © 2010 American Chemical Society 7021
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/28/2010

expression media to enhance solubility of membrane proteins. 14-19
Several membrane proteins have been synthesized using either
commercially available CFPS systems or ones created by individual
laboratories. 14-19 CFPS simplified the process and reduced
the time and labor required to produce membrane proteins. Many
toxic effects attributed to overproduction of recombinant proteins
in cell-based systems are eliminated by cell-free expression since
viable host cells are no longer required. 14,18 As a result, CFPS
has become an emerging alternative tool for the high level
production of membrane proteins. 14
Because CFPS is able to produce both soluble and membrane
proteins, it could play a significant role in supporting the drug
discovery pipeline. Membrane proteins, mainly G protein-coupled
receptors, represent 45% of all drug targets, while enzymes account
for 28% of them. 1 Thus, these two categories make up a total of
73% of all current drug targets. 1
In this report, we describe the use of a miniaturized fluid array
device for expression of both soluble and membrane proteins. The
device was designed and optimized for high-throughput cell-free
protein synthesis. The fabrication and characterization of the
device have been previously reported. 20-22 We used the device
for producing two membrane-associated proteins, bacteriorhodopsin
and ApoA lipoprotein, in the presence of lipids. Coexpression
of ApoA lipoprotein in the same CFPS addressed the solubility
concern of bacteriorhodopsin.
In addition, the array device was also employed to produce
two enzymes, luciferase and �-lactamase, both of which were
demonstrated to be compatible with enzyme inhibition assays.
Further, we used �-lactamase to show the adaptability of the device
for drug screening. �-Lactamase is an enzyme produced by some
bacteria to generate resistance to a �-lactam class of antibiotics
such as penicillin and cephalosporins. 23 All of these antibiotics
have a four-atom ring structure known as �-lactam. �-lactamase
breaks the amide bond in �-lactam, deactivating its antibacterial
properties and causing antibiotic resistance. One common practice
is to combine antibiotics with a �-lactamase inhibitor that
inactivates �-lactamase. 24 Since �-lactam-associated antibiotics
constitute 50% of antibiotic consumption around the world, 25 it is
important to have an efficient method to identify additional
�-lactamase inhibitors. Therefore, we investigated the exploitation
(14) Klammt, C.; Schwarz, D.; Lohr, F.; Schneider, B.; Dotsch, V.; Bernhard, F.
FEBS J. 2006, 273, 4141–4153.
(15) Kalmbach, R.; Chizhov, I.; Schumacher, M. C.; Friedrich, T.; Bamberg, E.;
Engelhard, M. J. Mol. Biol. 2007, 371, 639–648.
(16) Cappuccio, J. A.; Blanchette, C. D.; Sulchek, T. A.; Arroyo, E. S.; Kralj,
J. M.; Hinz, A. K.; Kuhn, E. A.; Chromy, B. A.; Segelke, B. W.; Rothschild,
K. J.; Fletcher, J. E.; Katzen, F.; Peterson, T. C.; Kudlicki, W. A.; Bench,
G.; Hoeprich, P. D.; Coleman, M. A. Mol. Cell. Proteomics 2008, 7, 2246–
2253.
(17) Schwarz, D.; Dotsch, V.; Bernhard, F. Proteomics 2008, 8, 3933–3946.
(18) Katzen, F.; Fletcher, J. E.; Yang, J. P.; Kang, D.; Peterson, T. C.; Cappuccio,
J. A.; Blanchette, C. D.; Sulchek, T.; Chromy, B. A.; Hoeprich, P. D.;
Coleman, M. A.; Kudlicki, W. J. Proteome Res. 2008, 7, 3535–3542.
(19) Kubick, S.; Gerrits, M.; Merk, H.; Stiege, W.; Erdmann, V. A. Membr. Protein
Cryst. 2009, 63, 25–49.
(20) Mei, Q.; Fredrickson, C. K.; Lian, W.; Jin, S.; Fan, Z. H. Anal. Chem. 2006,
78, 7659–7664.
(21) Mei, Q.; Fredrickson, C. K.; Simon, A.; Khnouf, R.; Fan, Z. H. Biotechnol.
Prog. 2007, 23, 1305–1311.
(22) Khnouf, R.; Olivero, D.; Jin, S.; Fan, Z. H. Biotechnol. Prog. 2010, DOI:
10.1002/btpr.474.
(23) Bush, K. Curr. Opin. Invest. Drugs 2002, 3, 1284–1290.
(24) Poole, K. Cell. Mol. Life Sci. 2004, 61, 2200–2223.
(25) Livermore, D. M. J. Antimicrob. Chemother. 1998, 41 (Suppl D), 25–41.
7022 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 1. (a) Picture of a device in the format of a 96-well plate for
protein expression. The middle layer (membrane) can be observed
in the wells at the bottom left due to light reflection. (b) The crosssectional
view of one well unit, consisting of three access holes and
one reaction chamber in the top layer, a dialysis membrane in the
middle layer, and a feeding chamber in the bottom layer. (c) The
expression product of bacteriorhodopsin and apolipoprotein coexpressed
in the miniaturized fluid array device. (d) The expression
product of bacteriorhodopsin and apolipoprotein coexpressed in a
conventional microplate.
of CFPS to generate �-lactamase that functioned as a drug target,
followed by employing four mock drug compounds to test their
inhibitory effects on �-lactamase. The results demonstrated the
feasibility of the use of the fluid array device for drug screening
with advantages in reagent consumption, analysis time, and
throughput.
MATERIALS AND METHODS
Device Fabrication. The details of the device fabrication have
been described elsewhere. 20-22 Briefly, two polypropylene sheets
were machined to have desired wells with an appropriate size. A
dialysis membrane with molecular cutoff of 8 KD was sandwiched
between these two sheets by an adhesive. Figure 1a shows a
picture of the assembled device in the format of a 96-well plate.
The cross-sectional view of one unit is shown in Figure 1b. The
top layer consists of a reaction chamber and three access holes
for loading the feeding solution. The bottom layer contains a
feeding chamber that encompasses both the access holes and
reaction chamber. The dialysis membrane was used to retain
newly synthesized proteins while allowing the nutrients (e.g.,
amino acids and ATP) in the feeding chamber to continuously
replenish the reaction chamber. The membrane is sturdy and does
not break if touched by a pipet tip by accident. The device was
designed to be disposable; thus, each well was used once.
Protein Expression. Luciferase and �-lactamase were expressed
using an RTS 100 wheat germ kit (Roche). The vector of

luciferase (T7 control vector) was obtained from Promega while
the �-lactamase vector was cloned in house using pIVEX-1.4
plasmid and verified by restriction enzyme digestion and gel
electrophoresis. 22 The reaction and feeding solutions were prepared
according to the manufacturer’s instructions. The reaction
solution was composed of 15 µL of wheat germ lysate, 15 µL of
reaction mix, 4 µL of amino acids, 1 µL of methionine, and 15 µL
of an individual DNA vector (2 µg). For negative controls, the
DNA vector was replaced with the same volume of nuclease free
water. The feeding solution was prepared by combining 900 µL
of feeding mix, 80 µL of amino acids, and 20 µL of methionine.
(All of these were provided in the kit and the concentration of
each component was fixed by the manufacturer.)
Membrane proteins were expressed in the RTS 500 E. coli kit
(Roche). The reaction solution was made by mixing 525 µL ofE.
coli lysate, 225 µL of reaction mix, 270 µL of amino acids without
methionine, and 30 µL of methionine, 2 mg/mL 1,2-dimyristoylsn-glycero-3-phosphocholine
(DMPC, Avanti Polar Lipids), 50 µM
retinal (Sigma), and 5 µg/mL DNA vectors for bacteriorhodopsin
and apolipoprotein. A stock solution of DMPC lipid (68 mg/mL)
was prepared by adding DMPC in nuclease-free water, sonicating
with a Vibra-Cell probe sonicator at a power of 6 W until the
solution became clear, and then centrifuging it to retain the
supernatant. The retinal solution (10 mM) was prepared in pure
ethanol. DNA vectors for bacteriorhodopsin and apolipoprotein
were prepared as reported previously. 16 The feeding solution was
made by mixing 8.1 mL of feeding mix, 2.65 mL of amino acids
without methionine, and 0.3 mL of methionine.
To carry out reactions in the device for either soluble or
membrane proteins, 200 µL of the feeding solution was pipeted
to the feeding chamber and 10 µL of the reaction solution was
added to the reaction chamber. After sealing the device using a
PCR tape (to prevent evaporation), the device was placed on an
orbital shaker for four hours, rotating at a speed of 30 rpm. The
synthesized proteins were then analyzed as discussed below.
Protein Assays. Luciferase was detected by injecting 30 µL
of luciferase assay reagent (Promega) into the reaction chamber
in the device, shaking the device for 2 s, and measuring
luminescence over 10 s. All of these steps were carried out in a
Mithras microplate reader (Berthold Technologies, Germany).
�-Lactamase was measured using m-{[(phenylacetyl)glycyl]oxy}benzoic
acid (PBA, Calbiochem), a chromogenic substrate that
changes color upon being enzymatically cleaved. 26 PBA (90 µL,
2 mM) was added into the expression product, followed by a 5
min incubation and absorbance measurement at 314 nm using a
BioRad spectrophotometer.
Expression of the membrane proteins was indicated by color
observation as discussed in the Results and Discussion. They were
also verified by gel electrophoresis, followed by either Coomassie
blue staining or Western blotting. For Coomassie blue staining,
the sample aliquot was diluted at a ratio of 1:20 using 2× sodium
dodecyl sulfate (SDS) sample loading buffer. A five µL aliquot of
the resulting solution was heated at 99 °C for 10 min (to denature
proteins) and then loaded onto a 15% SDS polyacrylamide gel.
Electrophoresis was carried out with a Precision Plus protein
standard ladder (BioRad) in a Mini-Protean III Cell system
(26) Govardhan, C. P.; Pratt, R. F. Biochemistry 1987, 26, 3385–3395.
(BioRad). After electrophoresis at 150 V for 3 h, the gel was
transferred to a container for Coomassie blue staining.
For the Western blotting, the samples were mixed in a ratio
of 1:100 with the sample loading buffer. After the same electrophoresis
procedure, the gel was transferred onto a nitrocellulose
membrane under an electric current of 220 mA for 30 min. The
membrane was then blocked with 5% fat free milk in PBS/Tween-
20 buffer for 1 h, followed by overnight incubation in anti-Hisperoxidase
antibody (Roche) that was prepared in the blocking
solution in a ratio of 1:50 000. After washing three times, the
membrane was placed in a solution of 1:40 ECL Plus Western
blotting reagents (GE Healthcare) for 5 min. The image of the
membrane was obtained by exposing it onto a 9 × 10 in. Kodak
film.
Enzyme Inhibition Assays. Proteins synthesized in the CFPS
device were used directly for enzyme inhibition assays without
harvesting and purification. Luciferase was chosen to be tested
first using a known inhibitor D-luciferin 6′-methyl ether (LME,
also known as 4,5-dihydro-2[6-methoxy-2-benzthiazolyl]-4-thiazole
carboxylic acid). 27,28 LME solutions with a range of concentrations
were prepared in nuclease-free water. To carry out the inhibition
assay, 1 µL of a LME solution was added to the CFPS reaction
product, followed by the luciferase assay as described above.
For the �-lactamase inhibition assay, three clinically used
drugs, tazobactam (Sigma), potassium clavulanate (Sigma), and
sulbactam (Astatech), as well as an additional compound with
similar chemical structure, cefotaxime (Sigma), were used to study
inhibition. To carry out the inhibition assay, a series of concentrations
of each compound was prepared. Five microliters of each
solution was added to the synthesized �-lactamase in the reaction
chamber, followed by a 15 min incubation. The resulting mixtures
were analyzed using the same protein assay procedure described
above for �-lactamase. The degree of inhibition was calculated
relative to the positive control (no inhibitors) and the negative
control (no DNA vector for protein expression).
RESULTS AND DISCUSSION
Miniaturized Fluid Array Device. A picture of the miniaturized
fluid array device is shown in Figure 1a. The device consists
of 96 units, which are in agreement with a conventional 96-well
microplate. The device was found to be compatible with commercially
available microplate readers and reagent dispensing
apparatuses; thus, it can be used for high-throughput applications
such as drug screening. For each unit (Figure 1b), the top layer
contains a reaction chamber for gene transcription and protein
translation, the bottom layer contains a feeding chamber for
replenishing nutrients (e.g., amino acids and ATP), and the middle
is a dialysis membrane that connects these two chambers. As
discussed in the literature, 6,12,29 the functions of the membrane
are to (1) achieve continuous supply of additional nutrients; (2)
retain proteins produced and large-molecule synthesis machinery;
and (3) dilute the reaction byproducts (e.g., pyrophosphates) and
reduce their effects on the reaction equilibrium. Compared to the
commercially available CFPS systems (e.g., RTS 500 kit), 29 one
major advantage of the fluid array device is lower reagent
(27) Wang, J. Q.; Pollok, K. E.; Cai, S.; Stantz, K. M.; Hutchins, G. D.; Zheng,
Q. H. Bioorg. Med. Chem. Lett. 2006, 16, 331–337.
(28) Barros, M. P.; Bechara, E. J. Free Radical Biol. Med. 1998, 24, 767–777.
(29) Betton, J. M. Curr. Protein Pept. Sci. 2003, 4, 73–80.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7023

Figure 2. (a) Gel electrophoresis of membrane proteins. Lane 1 is
a protein ladder. Lanes 2-4 are for the negative control. Lanes 5-7
are for expression of bacteriorhodopsin. Lanes 8-10 are for expression
of apolipoprotein. Lanes 11-13 are for coexpression of bacteriorhodopsin
and apolipoprotein. The first lane in each group is for
the reaction mixture; the second lane is for the pellet, and the third
lane is for the supernatant. The arrow in lane 5 indicates the position
of bacteriorhodopsin while the arrow in lane 8 indicates the position
of apolipoprotein. (b) Western blotting of membrane proteins with the
same lane designation as in (a). (c) Pictures of tubes containing
expressed protein products (after centrifugation). They are in the same
order as in (a) from the left to the right.
consumption (10 µL versus 1 mL). The cost-savings would be
substantial when a large number of arrays are required for highthroughput
applications. The additional advantage of miniaturization
is to enable a high-density array, which makes it possible to
have high-throughput CFPS for applications such as drug screening.
An alternative way to run high-throughput CFPS is in a
conventional 96-well or 384-well microplate. Without a membrane
in the conventional microplate, we found the synthesis yield was
much lower. 20-22 For the membrane protein to be discussed
below, coexpression of bacteriorhodopsin and apoA lipoprotein
in the fluid array resulted in a mixture with the characteristic
purple color of bacteriorhodopsin as shown in Figure 1c, demonstrating
the functional production of bacteriorhodopsin. Note
that the protein expression product in the device was transferred
to a transparent tube for the purpose of taking a picture. In
contrast, when the same expression was carried out in a
conventional microplate, the amount of proteins expressed was
too low to have an effect on the color of the mixture as shown in
Figure 1d. The yellow color of the mixture is very similar to the
negative control that contains no DNA vectors.
Membrane Proteins. The fluid array device was used to
produce two membrane-associated proteins, apoA lipoprotein and
bacteriorhodopsin. ApoA lipoprotein can form nanoparticles due
to the presence of lipid in the expression medium. 16 Its expression
was confirmed by polyacrylamide gel electrophoresis (PAGE) as
shown in Figure 2a (arrow in lane 8). Since the lipoprotein particle
is soluble, it was in the supernatant after centrifugation (lane 10).
Expression of bacteriorhodopsin in the fluid array was also
confirmed by PAGE (arrow in lane 5). After centrifugation of the
7024 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
mixture, most bacteriorhodopsin molecules precipitated to form
pellets since they were insoluble (lane 6). 15,16,30
Bacteriorhodopsin and ApoA lipoprotein could be coexpressed
in a single reaction; the lipoprotein’s ability to form lipid bilayer
patches from lipid micelles can be used to increase the solubility
of other membrane-associated proteins such as bacteriorhodopsin.
15,16,30 The parallel expression of both proteins was
confirmed by PAGE (lane 11). After centrifugation of the mixture,
most bacteriorhodopsin molecules stayed in the supernatant due
to the increased solubility in the presence of lipoprotein particles
(lane 13).
Since there are many protein bands belonging to the components
of the protein expression system, we carried out Western
blotting to verify the presence of bacteriorhodopsin and nanolipoprotein
in the samples. The blot in Figure 2b shows much
cleaner protein bands for the proteins of interest. Note that
Western blotting in Figure 2b is more sensitive than Coomassie
blue staining in Figure 2a; hence, a minute amount of bacteriorhodopsin
undetectable in lane 7 in Figure 2a was shown as a
band in Figure 2b.
The effect of coexpressed lipoprotein on the solubility of
bacteriorhodopsin is evident from the physical appearance of the
protein product solutions at the end of expression, followed by
centrifugation, as shown in Figure 2c. These proteins were
expressed in the device, and the product mixtures were transferred
to a tube, followed by centrifugation to separate soluble
and insoluble materials for visualization. The negative control with
no DNA vectors shows the color of the cell-free reaction mixture.
The product mixture for expression of apoA lipoprotein is
homogeneous since apoA lipoprotein is soluble. However, the
product mixture for expression of bacteriorhodopsin contains a
purple pellet and clear supernatant after centrifugation. This is in
agreement of the result of gel electrophoresis in Figure 2a that
bacteriorhodopsin is in the pellet. When two proteins were
simultaneously coexpressed, the solubility of bacteriorhodopsin
was increased in the presence of apoA lipoproteins. Importantly,
the bacteriorhodopsin sample was purple, which indicates the
formation of a functional protein in our device.
Soluble Proteins. As we have reported previously, several
soluble proteins can be expressed in the miniaturized fluid array
device, including luciferase, green fluorescent protein, �-glucoronidase,
alkaline phosphatase, �-lactamase, and �-galactosidase. 20-22
In this work, we chose two soluble proteins, luciferase and
�-lactamase, to demonstrate the enzyme inhibition assay. Luciferase
gene has been extensively used as a reporter gene for
visualizing biomolecular processes in living animals, 27,31 and
luciferase-based luminescence assays have been increasingly used
for high-throughput drug screening. 32 As a result, we selected
luciferase to verify if the protein expressed in the CFPS device
can be used for enzyme inhibition assays without harvesting and
purification from the protein expression mixture that contains a
number of cofactors, enzymes, and others. We first studied the
kinetics of luciferase assay reactions in the presence of an
inhibitor, luciferin 6′-methyl ether (LME). LME is a known
luciferase assay inhibitor and has been used by others in the
(30) Bayburt, T. H.; Grinkova, Y. V.; Sligar, S. G. Arch. Biochem. Biophys. 2006,
450, 215–222.
(31) Prescher, J. A.; Contag, C. H. Curr. Opin. Chem. Biol. 2010, 14, 80–89.
(32) Fan, F.; Wood, K. V. Assay Drug Dev. Technol. 2007, 5, 127–136.

Figure 3. Inhibitory effects of luciferin 6′-methyl ether on the
luciferase assay. (a) The luciferase assay reaction kinetics in the
presence of various concentrations of inhibitors. (b) The inhibitory
effects as a function of the inhibitor concentration. The decrease in
percentage of the luminescence signal in the y-axis is relative to the
positive control, in which no inhibitor was added. The lines are the
best fit in the power relationship among the experimental data points.
The error bars are invisible, as they are smaller than the data markers.
literature. 27,28 The inhibitory assay was performed using luciferase
synthesized in the miniaturized fluid array device; the expression
product was directly used for the assay without any treatment. Figure
3a shows the reaction kinetic curves we obtained. The resemblance
of these curves with those in the literature using purified luciferase 33
indicates that the enzyme synthesized in the cell-free medium can
be used for the enzyme inhibition assay without harvesting and
purification. As a result, we can use CFPS products directly for the
enzyme inhibition assay. Figure 3b shows the inhibitory effects on
the luciferase assay as a function of the LME concentration.
In addition, we evaluated the specificity of each enzyme assay.
For example, the UV absorbance-based enzyme assay of �-lactamase
using m-{[(phenylacetyl)glycyl]oxy}benzoic acid (PBA)
should be specific to �-lactamase. Figure 4 shows the assay
kinetics, in which signal increased over a period of time when
�-lactamase was cleaving the chromogenic substrate. It leveled
off to form a plateau when the reactions reached equilibrium.
However, when the same assay was applied to luciferase, there
was negligible signal increase, except for the initial point that was
due to the addition of a solution in a way similar to a negative
control. These results suggest that the �-lactamase assay is
specific. The data also indicate that reliable assay results for
�-lactamase should be recorded at 5 min after the assay reagents
were added into the expression products.
Drug Screening. To demonstrate the utility of the miniaturized
fluid array for drug screening, �-lactamase was chosen as a
mock drug target. �-Lactamase is an enzyme that hydrolyzes the
amide bond of �-lactam antibiotics, causing the antimicrobial
agents to lose their effectiveness. Three known �-lactamase
(33) DeLuca, M.; Wannlund, J.; McElroy, W. D. Anal. Biochem. 1979, 95, 194–
198.
Figure 4. Enzyme assay kinetics of �-lactamase using a chromogenic
agent, m-{[(phenylacetyl)glycyl]oxy}benzoic acid. The same assay
was applied to luciferase, showing negligible increase over the
background signal.
Figure 5. Inhibition of clavulanate acid (open cirles), tazobactam
(diamonds), and sulbactam (triangles) on the enzymatic activities of
�-lactamase. The concentrations of these inhibitors in the x-axis are
in the log scale. Also shown is negligible and concentrationindependent
inhibition of cefotaxime (solid circles) on �-lactamase.
Each data point represents an average obtained from three repeat
experiments, and the error bars indicate one standard deviation. The
chemical structures of all compounds tested are shown at the top;
all of them have a �-lactam ring that consists of three carbon atoms
and one nitrogen atom with a cyclic amide functional group.
inhibitors (clavulanate acid, tazobactam, and sulbactam) and one
noninhibitory chemical (cefotaxime) were used as mock compounds.
All of these compounds contain a �-lactam ring that can
be hydrolyzed by �-lactamase, as shown in Figure 5. However,
cefotaxime is a very large molecule, difficult to be hydrolyzed by
�-lactamase. �-Lactamase was produced in each unit of the array
device, and a series of varying concentrations of each compound
was added into several wells to obtain a response curve. The
relationship between the concentrations of each compound and
the level of �-lactamase inhibition is also shown in Figure 5. The
percentage of inhibition is calculated against the one without any
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7025

inhibitor (positive control). The results show that clavulanate acid,
tazobactam, and sulbactam had inhibitory effects on the enzymatic
activities of �-lactamase, while cetotaxime did not show any
inhibitory effect. The I50 values (the concentration required to
produce 50% inhibition 34 ) of clavulanate acid, tazobactam, and
sulbactam are 0.11, 0.87, and 24.8 µg/µL, respectively. As a
result, clavulanate acid had the highest inhibitory effect among
them, which agrees well with the previously published
results. 35,36
The result in Figure 5 suggests that protein expressed in a
cell-free system can be used for drug screening just as those
obtained from cell-based systems. The power of the fluid array
device is evident from the fact that the inhibition-concentration
curves of multiple compounds can be simultaneously obtained in
one device through judicial experimental designs. Importantly,
the device allowed us to determine not only if the compound is a
drug candidate but also the effectiveness of the drug by comparing
the various degrees of inhibition. An additional benefit of the use
of the fluid array is that �-lactamase is in solution phase, so that
its enzymatic activities are retained due to its appropriate threedimensional
configuration in an aqueous solution. As a result, it
is superior over ELISA or protein chips in which proteins are
attached to a solid surface and efforts must be made to minimize
the loss of biological activities of the proteins.
CONCLUSION
A miniaturized fluid array device has been demonstrated
capable of carrying out high-throughput cell-free protein synthesis.
Compared to the conventional cell-free protein synthesis using
commercially available RTS 500 kits, 29 the reagent consumption
is 2 orders of magnitude less (10 µL of the reaction solution in
our device versus 1 mL in the commercial kits). As a result, the
cost-savings would be substantial when a large array is required
for high-throughput applications (e.g., drug screening).
Cell-free expression of both soluble and membrane proteins
has been achieved in the array device. A membrane protein
(34) Duggleby, R. G. Biochem. Med. Metab. Biol. 1988, 40, 204–212.
(35) Payne, D. J.; Cramp, R.; Winstanley, D. J.; Knowles, D. J. Antimicrob. Agents
Chemother. 1994, 38, 767–772.
(36) Bou, G.; Martinez-Beltran, J. Antimicrob. Agents Chemother. 2000, 44, 428–
432.
7026 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
synthesized in the device can be solubilized by coexpressing it
with a lipoprotein in the presence of reagents necessary for correct
folding, which would allow functional and structural studies of
the membrane protein. Since membrane protein receptors represent
the largest portion of drug targets, this capability could
have great impact in searching for therapeutics.
Proteins synthesized in the device have been exploited for
enzyme inhibition assays without protein harvesting and purification.
Therefore, laborious and time-consuming procedures such
as magnetic bead-based separation and chromatography-based
purification can be eliminated. Potential losses of proteins and
possible structure changes in the separation procedures have been
reduced. In addition, a solution array (proteins in solution) does
not compromise the kinetics of binding 37 whereas proteins
immobilized on a solid surface in conventional protein chips are
likely compromised by the fact that (1) only a portion of protein
structures are exposed to ligands, (2) spatial hindrance exists
during interactions between immobilized proteins and ligands in
a sample, and (3) protein tertiary structures may change during
immobilization.
The array format of the device enabled simultaneous expression
of proteins, screening for a variety of inhibitors, and
comparison of the inhibition levels among the inhibitors studied.
We showed the capability of the device for drug screening by the
use of �-lactamase, a drug target associated with antibiotic
resistance, and four mock drug compounds. The inhibitory effects
of these compounds on �-lactamase were measured in parallel,
and their degrees in drug effectiveness were compared.
ACKNOWLEDGMENT
This work was supported in part by Defense Advanced
Research Projects Agency (DARPA) via Micro/Nano Fluidics
Fundamentals Focus Center at the University of California at
Irvine, the University of Florida via UF Opportunity Fund, and
the University of California Discovery Grant Program.
Received for review June 10, 2010. Accepted July 18,
2010.
AC1015479
(37) Zhou, H.; Roy, S.; Schulman, H.; Natan, M. J. Trends Biotechnol. 2001,
19, S34–39.

Anal. Chem. 2010, 82, 7027–7034
Patterning of Metal, Carbon, and Semiconductor
Substrates with Thin Organic Films by
Microcontact Printing with Aryldiazonium Salt Inks
Joshua Lehr, †,‡ David J. Garrett, †,‡ Matthew G. Paulik, ‡,§ Benjamin S. Flavel, ‡,
Paula A. Brooksby, ‡ Bryce E. Williamson, ‡ and Alison J. Downard* ,†,‡
MacDiarmid Institute for Advanced Materials and Nanotechnology, Private Bag 4800, Christchurch, 8140, New Zealand,
and Department of Chemistry, University of Canterbury, Private Bag 4800, Christchurch, 8140, New Zealand
Surface modification through reduction of aryldiazonium
salts to give covalently attached layers is a widely investigated
procedure. However, realization of potential applications
of the layers requires development of patterning
methods. Here, we demonstrate that microcontact printing
with poly(dimethylsiloxane) stamps inked with aqueous
acid solutions of aryldiazonium salts gives stable
organic layers on gold, copper, silicon, and graphitic
carbon surfaces. Depending on the substrate-diazonium
salt combination, the layers range from relatively irregular
multilayers to smooth films with close to monolayer
thickness. After printing, surface attached aminophenyl
and carboxyphenyl groups retain their usual reactivity
toward amide bond formation with solution species, and
hence, the method is a simple route to patterned, covalently
attached, reactive tether layers. Multicomponent
patterned films can be prepared by printing a second
modifier onto a film-coated surface. Microcontact printing
using aryldiazonium salt inks is experimentally very
simple and is applicable to the broad range of substrates
capable of spontaneously reducing aryldiazonium salts.
Localized immobilization of molecular species is an important
step in the fabrication of surfaces for applications that include
chemical and biological sensors, biochips, molecular electronics,
and tissue engineering. A large research effort over many years
has established a suite of patterning methods, compatible with
the best-known surface modification strategies, particularly those
based on assembly of alkanethiols at noble metal surfaces, and
reactions of silanes at oxide surfaces and alkenes at silicon. A
relatively recently developed surface modification method is
grafting from aryldiazonium salts solutions. 1,2 The mechanism,
scope, and applications of this method have been widely
* To whom correspondence should be addressed. E-mail: alison.downard@
canterbury.ac.nz. Fax: +64-3-3642110.
† MacDiarmid Institute for Advanced Materials and Nanotechnology.
‡ University of Canterbury.
§ Present address: AgResearch Ltd, Private Bag 4749, Christchurch, 8140,
New Zealand.
Present address: School of Chemistry, Physics and Earth Sciences, Flinders
University, Bedford Park, SA 5042, Australia.
(1) Delamar, M.; Hitmi, R.; Pinson, J.; Saveant, J. M. J. Am. Chem. Soc. 1992,
114, 5883–5884.
(2) Allongue, P.; Delamar, M.; Desbat, B.; Fagebaume, O.; Hitmi, R.; Pinson,
J.; Saveant, J.-M. J. Am. Chem. Soc. 1997, 119, 201–207.
investigated; 3-9 however, patterning techniques appropriate for
use with this approach have received little attention.
Surface grafting from aryldiazonium salt solutions proceeds
via reduction of the aryldiazonium cation and elimination of
dinitrogen to yield an aryl radical capable of reaction with the
surface. 2 For some substrates, a covalent linkage between the
modifier and the surface has been directly demonstrated; 10-12 for
others, covalent attachment has been inferred on the basis of the
orientation of surface groups 13,14 or the stability of the layer. 2,15,16
Under commonly used grafting conditions, attack by aryl radicals
on pregrafted groups gives a multilayered, covalently coupled film
structure. 17,18 The stability of the grafted layers is a key advantage
of this modification method; other attractive features are its wide
substrate compatibility (from noble and industrial metals to
semiconductors 9-11,19-22 ) and the opportunity for further chemistry
involving the grafted groups. Electrografting at an externally
(3) Adenier, A.; Bernard, M.-C.; Chehimi, M. M.; Cabet-Deliry, E.; Desbat, B.;
Fagebaume, O.; Pinson, J.; Podvorica, F. J. Am. Chem. Soc. 2001, 123,
4541–4549.
(4) Barriere, F.; Downard, A. J. J. Solid State Electrochem. 2008, 12, 1231–
1244.
(5) Downard, A. J. Electroanalysis 2000, 12, 1085–1096.
(6) Fave, C.; Leroux, Y.; Trippe, G.; Randriamahazaka, H.; Noel, V.; Lacroix,
J. C. J. Am. Chem. Soc. 2007, 129, 1890–1891.
(7) Pinson, J.; Podvorica, F. Chem. Soc. Rev. 2005, 34, 429–439.
(8) Ranganathan, S.; Steidel, I.; Anariba, F.; McCreery, R. L. Nano Lett. 2001,
1, 491–494.
(9) Stewart, M. P.; Maya, F.; Kosynkin, D. V.; Dirk, S. M.; Stapleton, J. J.;
McGuiness, C. L.; Allara, D. L.; Tour, J. M. J. Am. Chem. Soc. 2004, 126,
370–378.
(10) Bernard, M. C.; Chausse, A.; Cabet-Deliry, E.; Chehimi, M. M.; Pinson, J.;
Podvorica, F.; Vautrin-Ul, C. Chem. Mater. 2003, 15, 3450–3462.
(11) Boukerma, K.; Chehimi, M. M.; Pinson, J.; Blomfield, C. Langmuir 2003,
19, 6333–6335.
(12) Combellas, C.; Kanoufi, F.; Pinson, J.; Podvorica, F. I. Langmuir 2005,
21, 280–286.
(13) Anariba, F.; Viswanathan, U.; Bocian, D. F.; McCreery, R. L. Anal. Chem.
2006, 78, 3104–3112.
(14) Ricci, A.; Bonazzola, C.; Calvo, E. J. Phys. Chem. Chem. Phys. 2006, 8, 4297–
4299.
(15) D’Amours, M.; Belanger, D. J. Phys. Chem. B 2003, 107, 4811–4817.
(16) Shewchuk, D. M.; McDermott, M. T. Langmuir 2009, 25, 4556–4563.
(17) Doppelt, P.; Hallais, G.; Pinson, J.; Podvorica, F.; Verneyre, S. Chem. Mater.
2007, 19, 4570–4575.
(18) Kariuki, J. K.; McDermott, M. T. Langmuir 2001, 17, 5947–5951.
(19) Delamar, M.; Desarmot, G.; Fagebaume, O.; Hitmi, R.; Pinson, J.; Saveant,
J. M. Carbon 1997, 35, 801–807.
(20) Liang, H. H.; Tian, H.; McCreery, R. L. Appl. Spectrosc. 2007, 61, 613–
620.
(21) Mahmoud, A. M.; Bergren, A. J.; McCreery, R. L. Anal. Chem. 2009, 81,
6972–6980.
10.1021/ac101785c © 2010 American Chemical Society 7027
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/28/2010

applied potential is the most widely used strategy for the reduction
step, but simple immersion of the substrate in a solution of an
aryldiazonium salt can also lead to formation of surface layers.
This spontaneous, or open-circuit potential (OCP), reaction has
been reported for a range of metals, semiconductors, and carbon
materials 4 and appears to involve electron transfer from the
substrate to the aryldiazonium cation in solution.
Currently, there are few examples of patterned organic layers
prepared by reduction of aryldiazonium salts. In the earliest report,
we used mechanical scribing with an atomic force microscope
(AFM) tip to remove regions of electrografted film from a carbon
substrate. 23 A second aryldiazonium salt was then electrografted
to the bare regions, creating a surface with dual chemical
functionality. In a soft lithographic approach, we patterned a
carbon substrate by adhering a poly(dimethylsiloxane) (PDMS)
mold to the surface (either bare or film-coated) to form microchannels.
24 The channels were subsequently filled, either with
aryldiazonium salt solution for site-specific electrografting or with
reagents used for electrochemical or chemical conversion of the
pre-existing surface film. Most recently, we demonstrated that
conventional photolithography can be coupled with electrografting
to give large areas of micrometer-sized patterns of modifiers on
highly doped silicon. 22 Charlier, Palacin, and co-workers, 25 and
Cougnon, Bélanger, and co-workers 26 have established that the
scanning electrochemical microscope is a useful tool for localized
surface grafting from aryldiazonium salts, while the former
research group has also reported elegant patterning methods
specific to silicon substrates. In one example, they used ionic
implantation to create locally doped areas of silicon and, thus,
achieved site-specific electrografting of an aryldiazonium salt. 27
In another example, they illuminated p-type silicon through a mask
to locally increase the substrate conductivity and allow electrografting
to proceed. 28 Palacin and co-workers have also explored
the use of a patterned agarose hydrogel containing an aryldiazonium
salt solution sandwiched between two electrodes as an
electrochemical “printing” method. 29 Finally, Corgier and Bélanger
have adapted the methods of colloidal nanolithography to electrograft
organic groups to the nanoscale spaces between polystyrene
beads assembled on carbon and gold surfaces. 30
All of the patterning methods described above involve electrochemical
generation of aryl radicals at an externally applied
potential. In an earlier communication, we established that
spontaneous, OCP reduction of aryldiazonium salts by carbon
substrates can also be used. 31 We showed that microcontact
(22) Flavel, B. S.; Garrett, D. J.; Lehr, J.; Shapter, J. G.; Downard, A. J.
Electrochim. Acta 2010, 55, 3995–4001.
(23) Brooksby, P. A.; Downard, A. J. Langmuir 2005, 21, 1672–1675.
(24) Downard, A. J.; Garrett, D. J.; Tan, E. S. Q. Langmuir 2006, 22, 10739–
10746.
(25) Ghorbal, A.; Grisotto, F.; Charlier, J.; Palacin, S.; Goyer, C.; Demaille, C.
ChemPhysChem 2009, 10, 1053–1057.
(26) Cougnon, C.; Gohier, F.; Belanger, D.; Mauzeroll, J. Angew. Chem., Int.
Ed. 2009, 48, 4006–4008.
(27) Charlier, J.; Palacin, S.; Leroy, J.; Del Frari, D.; Zagonel, L.; Barrett, N.;
Renault, O.; Bailly, A.; Mariolle, D. J. Mater. Chem. 2008, 18, 3136–3142.
(28) Charlier, J.; Clolus, E.; Bureau, C.; Palacin, S. J. Electroanal. Chem. 2008,
622, 238–241.
(29) Mouanda, B.; Eyeffa, V.; Palacin, S. J. Appl. Electrochem. 2009, 39, 313–
320.
(30) Corgier, B. P.; Belanger, D. Langmuir 2010, 26, 5991–5997.
(31) Garrett, D. J.; Lehr, J.; Miskelly, G. M.; Downard, A. J. J. Am. Chem. Soc.
2007, 129, 15456–15457.
7028 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
printing (MCP) using PDMS stamps and aryldiazonium salt inks
gave micrometer-scale patterns of modifiers on glassy-carbon-like
thin films (pyrolyzed photoresist film; PPF). MCP is a very simple
and relatively fast patterning method and, in these respects, has
obvious advantages over the electrochemical methods outlined
above. 32
In this paper, we investigate the characteristics (surface
concentration, thickness, homogeneity, and chemical reactivity)
of layers prepared using MCP and OCP reduction of aryldiazonium
salts on PPF. We demonstrate that the method can be extended
to metal (Au and Cu) and semiconductor (Si) substrates and
provide guidelines concerning its general applicability in terms
of substrate/diazonium cation combinations and the characteristics
of the resultant films. We also demonstrate a unique feature
of MCP of aryldiazonium salts: covalently coupled two-component
surfaces can be prepared simply by printing a second layer onto
a previously modified surface.
EXPERIMENTAL SECTION
Materials. Aqueous solutions were prepared using Millipore
Milli-Q water (>18 MΩ cm). Tetrafluoroborate salts of 4-nitrobenzenediazonium
(NBD) and 4-carboxybenzenediazonium (CBD)
were synthesized using standard procedures. 33 4-Aminobenzenediazonium
salt (ABD) was synthesized as a 20 mM solution in
0.5 M HCl. 34 Procedures for preparing citrate-capped Au nanoparticles
(∼13 nm diameter), 35 PPF, 36 and planar Au films (Au/
NiCr/Si), 37 and drying acetonitrile (ACN) have been described
previously.
Uncut single walled carbon nanotubes (SWCNTs) (Carbon
Nanotechnologies Incorporated) were acid-treated by adding 25
mg to 27 mL of 3:1 concentrated H2SO4 and HNO3 and sonicating
for 10 h while adding ice to the ultrasonicator bath to maintain
a temperature close to 20 °C. Following sonication, the solution
was poured into 500 mL of distilled water. After standing
overnight, the solution was filtered under suction through
Millipore 0.22 µm hydrophilic polyvinylidene fluoride filter
membranes and then washed with copious amounts of water.
The dried SWCNT cakes were peeled from the filters and
resuspended in DMSO to give a1mgmL -1 stock solution.
Si(100) wafers (1-20 Ω cm, Silicon Quest and Micro Materials)
were cut into ∼15 × 15 mm 2 tiles, immersed in 40% HF (Sigma-
Aldrich) for 3 min (Caution: HF is hazardous; handle with care
and appropriate personal protective clothing), washed with
methanol, dried in a stream of N2 gas, and used within 10 min
of HF treatment. Small pieces of Cu plate were immersed in
16 M HNO3 for 10 s, washed with water, immersed in 17 M
acetic acid for 30 s, and dried in a stream of N2 gas. 38
Fabrication and solvent extraction of PDMS stamps followed
previously described procedures. 24 The stamps were either
nonpatterned or had a test pattern with micrometer-sized
features.
(32) Xia, Y. N.; Whitesides, G. M. Angew. Chem., Int. Ed. 1998, 37, 551–575.
(33) Saunders, K. H.; Allen, R. L. M. Aromatic Diazo Compounds, 3rd ed.; Edward
Arnold: London, 1985.
(34) Lyskawa, J.; Belanger, D. Chem. Mater. 2006, 18, 4755–4763.
(35) Grabar, K. C.; Freeman, R. G.; Hommer, M. B.; Natan, M. J. Anal. Chem.
1995, 67, 735–743.
(36) Brooksby, P. A.; Downard, A. J. Langmuir 2004, 20, 5038–5045.
(37) Lehr, J.; Williamson, B. E.; Flavel, B. S.; Downard, A. J. Langmuir 2009,
25, 13503–13509.
(38) Chamoulaud, G.; Belanger, D. J. Phys. Chem. C 2007, 111, 7501–7507.

Electrochemistry. Electrochemical measurements were made
at room temperature in an N2 atmosphere. The electrochemical
cell exposed a circular area (0.18 cm 2 for PPF or 0.79 cm 2 for
Au) of substrate to the cell solution, as described previously. 36
The auxiliary and reference electrodes were Pt and SCE,
respectively. Cyclic voltammograms were recorded with a scan
rate of 100 mV s -1 .
Surface concentrations of grafted nitrophenyl (NP) groups
were estimated from cyclic voltammograms of modified surfaces
in 0.10 M H2SO4. The area under the irreversible reduction peak
at Ep,c ≈ -0.6 V and the area under the oxidation peak at Ep,a
≈ 0.3 V was determined by fitting the data with polynomial
baselines and mixed Lorentzian-Gaussian curves, using the
Levenberg-Marquardt algorithm implemented via Linkfit software.
36 The surface concentration was then calculated using
Faraday’s law, as previously described, 36 with an estimated
±20% uncertainty in the absolute surface concentration.
Microcontact Printing. Within2hofuse, PDMS stamps were
treated with O2 plasma (100 W at 0.1 Torr for 5 min). Stamps
were immersed in 20 mM aryldiazonium cation ink for 2 min,
dried to tackiness in a stream of N2 gas, and placed on the
substrate for 30 min. Unless stated otherwise, all printed
samples were ultrasonicated for 5 min in Milli-Q water and
dried in a stream of N2 prior to analysis or further treatment.
Each stamp was used with only one type of ink. The compositions
of printing inks are denoted “diazonium cation/solvent”.
Control samples were prepared in the same way but using a
solution from which the aryldiazonium salt had been omitted
(“blank” ink).
Characterization of Films and Patterns. AFM (Digital
Instruments Dimension 3100) depth-profiling measurements were
performed on modified PPF samples by scratching with the AFM
tip to remove a section of film, as described previously. 36 Three
average line profiles were obtained from each of two 10 × 1.25
µm 2 scratches per sample. Each line profile gave two thickness
values: one from the step down into the scratch and the other
from the step out of the scratch. Thus, the reported thickness
for each sample is a mean of 12 values, and the uncertainties
are two standard deviations of the mean. Condensation figures
were obtained by allowing water vapor to condense on surfaces
and imaging using an Olympus BX60 inverted light microscope
equipped with a polarizer and an Olympus DP10 camera.
Scanning electron microscopy (SEM) images were obtained
using a JEOL 7000 high-resolution instrument with an accelerating
voltage of 15 kV.
The procedure for measuring water contact angles has been
described previously. 31 Measurements were made using two 1
µL drops of Milli Q water placed on each duplicate sample or
control. The stated values are the means of the four measurements,
and the uncertainties are two standard deviations of the
mean.
RESULTS AND DISCUSSION
This section is divided into four parts. The first describes
printing with aryldiazonium salt inks on PPF, Au, and Si substrates
using nonpatterned PDMS stamps. The second part demonstrates
that printing with patterned PDMS stamps gives micrometer-scale
patterned layers on Au, PPF, Si, and Cu. The reactivity of the
printed layers is described in the third part, and finally, we show
Figure 1. First (s) and second scan (---) cyclic voltammograms in
0.1MH2SO4 of a PPF surface printed with NBD/1 M H2SO4 ink for
30 min.
that a second modifier can be printed onto an already modified
surface to create patterned, covalently coupled, two-component
surfaces.
Nonpatterned Printing on PPF, Gold, and Silicon. Preliminary
results establishing successful MCP of PPF using DMFand
1 M H 2SO4-based aryldiazonium salt inks have been
reported earlier. 31 In both that work and the present study,
PPF was used as the carbon substrate because aryldiazonium
salts spontaneously graft to its surface at OCP from acidic
aqueous solution 31 and its low surface roughness facilitates
measurement of film thickness. 39 Here, two aryldiazonium
cations were chosen for detailed examination of the printing
process. The NBD derivative was selected as a relatively easily
reduced species (in cyclic voltammograms, Ep,c ≈ 0VvsSCE
in 0.1 M H2SO4 at a scan rate of 100 mV s -1 ) that is additionally
convenient because the resultant grafted NP moieties can be
readily detected and quantified by electroreduction. The CBD
cation was selected as a more difficult to reduce species (Ep,c
≈ -0.3 V vs SCE in 0.1 M H2SO4 at a scan rate of 100 mV s -1 ),
but which gives grafted carboxyphenyl (CP) films whose
chemical reactivity makes them useful as tether layers for
subsequent coupling reagents. CP groups are not electroactive
in the potential range accessible in 0.1 M H2SO4, but their
presence can be detected by water contact angle measurements.
Figure 1 shows consecutive cyclic voltammograms of a PPF
surface printed with NBD/1 M H2SO4 ink using a nonpatterned
stamp. The first scan, from 0.8 to -0.9 V, shows the characteristic,
irreversible reduction (Ep,c ≈ -0.60 V) of surfaceattached
NP groups to aminophenyl (eq 1) and hydroxyaminophenyl
groups (eq 2) but no evidence for physisorbed
aryldiazonium cations, which, if present, would be reduced at Ep,c
≈ 0 V. The return scan shows the reversible oxidation (Ep,a )
0.35 V) of hydroxyaminophenyl to nitrosophenyl groups (eq
3), 40 with the nitrosophenyl/hydroxyaminophenyl redox couple
appearing as the only significant feature in subsequent cycles.
These results are consistent with immobilized NP groups,
spontaneously grafted to the surface during printing. Table 1 lists
electroactive-NP surface concentrations, ΓNP, determined for
three nonpatterned NBD/1 M H2SO4 samples from the
(39) Ranganathan, S.; McCreery, R. L. Anal. Chem. 2001, 73, 893–900.
(40) Ortiz, B.; Saby, C.; Champagne, G. Y.; Belanger, D. J. Electroanal. Chem.
1998, 455, 75–81.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7029

Table 1. Surface Concentration and Film Thickness
Data for NP Films Printed on PPF
1 (5 ± 1) × 10
film
thickness/nm
-10 1.5 ± 0.4
2 (6 ± 1) × 10-10 2.4 ± 0.6
3 (16 ± 3) × 10-10 1.7 ± 0.4
sample ΓNP /mol cm -2
first-scan, along with the corresponding film thicknesses
obtained by AFM depth profiling.
Surface---C6H4-NO2 + 6H + + 6e - f
Surface---C6H4-NH2 + 2H2O (1)
Surface---C6H4-NO2 + 4H + + 4e - f
Surface---C6H4-NHOH + H2O (2)
Surface---C 6 H 4 -NHOH h Surface---C 6 H 4 -NO + 2H + + 2e -
Previously, we have found that single layers of NP groups
electrografted to PPF have ΓNP ) (2.5 ± 0.5) × 10 -10 mol cm -2 , 36
and similar values were obtained for methylphenyl and CP
groups electrografted to flat Au substrates. 41 A monolayer of
vertically oriented NP groups has a calculated thickness of 0.8
nm; 36 hence, both the concentration and thickness data in Table
1 indicate the formation of multilayer domains with an average
thickness of 2 to 3 layers. 18 Table 1 also shows significant
variations between films prepared under the same conditions and
no systematic relationship between ΓNP and average AFMdetermined
film thickness. The initial OCPs differ significantly
between PPF samples, particularly from different preparation
batches, and we attribute the sample-to-sample variations in
Table 1 to this variability of “activity”. We, 37 and others, 37,42,43
have observed that the substrate potential increases as the
spontaneous reduction of aryldiazonium cations proceeds and that
film growth stops when the potential becomes too positive to
sustain reduction. The initial OCP, therefore, influences the
amount of charge that can be transferred before the “cutoff”
potential is reached, and consequently, it helps to determine the
amount of material that can be attached to the surface in the
absence of an externally applied potential.
The lack of a clear relationship between Γ NP and the “average”
film thickness is attributed to differences in the measurement
scales of the associated methods. The surface concentration
data are averages over a large area (0.18 cm 2 ), whereas the
AFM film thicknesses are derived from three depth profiles
across two 10 × 1.25 µm 2 scratches per sample. For films of
highly variable thickness, the average results of a few measurements
performed over a small area may not yield results that
are representative of the whole-sample average. The possibility
of multilayer grafting, coupled with an inhomogeneous distri-
(41) Paulik, M. G.; Brooksby, P. A.; Abell, A. D.; Downard, A. J. J. Phys. Chem.
C 2007, 111, 7808–7815.
(42) Le Floch, F.; Simonato, J.-P.; Bidan, G. Electrochim. Acta 2009, 54, 3078–
3085.
(43) Smith, R. D. L.; Pickup, P. G. Electrochim. Acta 2009, 54, 2305–2311.
7030 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(3)
bution of aryldiazonium cations on the inked stamp, is likely
to lead to films of variable thickness, and the large uncertainties
for the AFM thickness values are consistent with this. Clearly,
such variability will be a limitation when highly reproducible
and homogeneous films are required.
Successful modification of PPF after printing with CBD/1 M
H2SO4 ink was confirmed by a decrease of the water contact
angles from 68 ± 2° to 31 ± 2°, consistent with the attachment
of hydrophilic CP groups. In contrast, control surfaces gave
an unchanged postprinting contact angle of 68 ± 11°. Cyclic
voltammograms (not shown) revealed no signals between 0.8
and -0.5 V, confirming the absence of physisorbed CBD
(expected reduction peak at -0.3 V). AFM depth-profiling
measurements on two films gave an average film thickness of
1.0 ± 0.2 nm, close to that expected for a monolayer. Printed
CP films are, thus, on average, thinner and significantly more
uniform than printed NP films. The different morphology can
be attributed to the lower reduction potential for CBD. Because
CBD is more difficult to reduce than NBD, its reduction ceases
at a lower substrate OCP and a correspondingly smaller amount
of CP is grafted to the surface. The more limited degree of
spontaneous reduction diminishes the prevalence of significant
multilayer “outgrowths”, and a more uniform film thickness
results. Vautrin-Ul and co-workers have reported results
consistent with this interpretation, 44 finding that spontaneous
reduction of NBD rapidly gave thick, nonuniform films on zinc
but formed thin homogeneous layers more slowly on a lessreducing
nickel surface. Hence, for substrates that act as the
reducing agent for aryldiazonium cation-based grafting, when
the potential driving force for reduction is large (the aryldiazonium
derivatives are easily reduced in comparison with the
reducing power of the substrate), thicker and more irregular
films will be formed than when the driving force is lower.
The feasibility of printing using other aryldiazonium salt/
substrate combinations was also examined in experiments described
in the Supporting Information. Electrochemical (Figures
S-1, S-2) or AFM depth-profiling measurements confirmed that
printing of ABD on PPF, NBD on Au, and NBD on Si gave the
expected modified surfaces. For Si samples, the oxide layer was
removed or significantly thinned by HF treatment prior to printing.
Samples with an intact native oxide layer could not be modified.
To summarize, we expect printing to be successful for all
aryldiazonium salt-substrate combinations for which the grafting
reaction proceeds spontaneously at OCP in solution. As is found
for layers grafted from solution, the stability of attachment will
depend mainly on the substrate; for example, very stable layers
are formed by grafting onto graphitic carbon, 2,15 but the stability
of the layers on Au is significantly less. 16
Patterned Microcontact Printing of Gold, PPF, Silicon,
and Cu. Having successfully demonstrated that printing with
aqueous aryldiazonium salt inks leads to surface modification, we
next investigated patterning of surfaces. Figures 2-4 show images
of Au, PPF, Si, and Cu substrates printed with aryldiazonium salt
inks and with blank inks.
In Figure 2, SEM images of Au substrates patterned with NP,
aminophenyl (AP), and CP groups are compared with images from
(44) Adenier, A.; Cabet-Deliry, E.; Chausse, A.; Griveau, S.; Mercier, F.; Pinson,
J.; Vautrin-Ul, C. Chem. Mater. 2005, 17, 491–501.

Figure 2. SEM images of Au substrates patterned by printing with
(a) NBD/1 M H2SO4, (b)1MH2SO4, (c) ABD/0.5 M HCl, (d) 0.5 M
HCl, and (e) CBD/1 M H2SO4 inks.
the corresponding controls. The aryldiazonium salt inks give
clearly defined patterns with feature sizes down to 20 µm, whereas
the blank inks give faint patterns, which are attributed to PDMS
residues.
SEM is not a good technique for imaging organic films on
carbon substrates. Consequently, the image contrast for PPF
patterned with NP (Figure 3a) is just marginally better than that
for the control sample (Figure 3b). To overcome this inherent
limitation, alternative methods were used to image surfaces
patterned with AP and CP groups. For the former, after printing
with ABD/0.5 M HCl (or blank 0.5 M HCl), the surfaces were
immersed for 40 min, at pH ∼ 5, in a solution of citrate-capped
Au nanoparticles. Preferential assembly of nanoparticles (via
electrostatic interactions) on the AP-modified areas clearly revealed
the patterns (compare Figure 3c,d). For surfaces patterned
with CP groups (by printing with CBD/1 M H2SO4 ink) and the
corresponding blanks, condensation figures were imaged by
optical microscopy (Figures 3e,f). Although the pattern is welldefined
in both the CP-printed sample and the blank, the relative
sizes of water droplets in the stamped and “bare” areas are the
opposite in the sample and blank. Water droplets are larger in
the CP areas than on bare PPF, consistent with addition of
hydrophilic CP groups to the surface; in contrast, areas contacted
by the stamp inked with 1MH2SO4 only are smaller than on
bare PPF, suggesting hydrophobic contaminants have been
transferred to the surface by the stamp. (Note that the sizes
of water droplets on bare PPF in Figures 3e,f are not the same
because the size depends on the extent of evaporation prior to
image capture.)
Figure 3. SEM images (a-d) and optical micrographs (e, f) of PPF
surfaces printed with (a) NBD/1 M H2SO4, (b, f) 1 M H2SO4, (c) 20
mM ABD/0.5 M HCl, (d) 0.5 M HCl, and (e) CBD/1 M H2SO4 inks.
The surfaces shown in (c) and (d) were immersed in Au nanoparticle
solution for 40 min before imaging; surfaces shown in (e) and (f) were
treated with water vapor before imaging.
Patterning of NP and AP groups on Si was also successful
as revealed by the SEM images of Figure 4a-d. There is strong
contrast between the grafted and bare areas in Figure 4a,c, in
comparison with the faint patterns of the controls (Figure
4b,d).
As a final example to demonstrate the wider applicability of
MCP with aqueous aryldiazonium salt inks, a Cu surface was
patterned with NP groups. Copper is known to react spontaneously
with NBD at OCP in both aqueous and nonaqueous conditions. 38,45
The SEM image in Figure 4e shows strong contrast between NPprinted
areas and bare Cu, whereas only a very faint pattern is
seen on the control (Figure 4f). The roughness of the Cu surface,
evident in both images, leads to incomplete contact with the stamp
and accounts for the “patchy” appearance of the pattern in Figure
4e.
These examples confirm that MCP is a very simple route to
patterning conducting surfaces. The scope of the method and the
characteristics of the patterned layers will be determined by the
substrate-diazonium salt combination as described in the previous
section.
Printed Tether Layers for Further Immobilization Chemistry.
The utility of MCP can be enhanced by printing layers that
act as tethers for further immobilization reactions. Two examples
are demonstrated here, on the basis of coupling of secondary
reagents (4-nitroaniline (NA) and SWCNTs) to primary layers of
CP or AP printed on PPF. The selection of these reagents was
based on their ease of detection: NA by its redox chemistry and
SWCNTs by AFM imaging.
(45) Hurley, B. L.; McCreery, R. L. J. Electrochem. Soc. 2004, 151, B252–B259.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7031

Figure 4. SEM images of (a-d) Si samples patterned by printing
with (a) NBD/1 M H2SO4, (b)1MH2SO4, (c) ABD/0.5 M HCl, and (d)
0.5 M HCl inks; (e, f) Cu samples patterned by printing with (e) NBD/1
MH2SO4 and(f)1MH2SO4 inks.
Table 2. Film Thicknesses for NA Layer Coupled to CP
Film and Associated Controls
surface film thickness/nm a
PPF-CP 1.0 ± 0.3
PPF-CP/SOCl2/NA 2.0 ± 0.5
PPF-CP/NA 1.0 ± 0.4
PPF-H2SO4/SOCl2/NA 0.4 ± 0.2
a
AFM line profiles are shown in Figure S-4 (Supporting Information).
NA + CP on PPF. CP groups were printed on PPF using
nonpatterned stamps and CBD/1 M H2SO4 ink. The films were
activated by immersion in SOCl2 for 30 min and then transferred
to a 20 mM NA/ACN solution at room temperature for
24 h to promote coupling of NA groups to the CP layer via the
formation of amide bonds. These samples are denoted PPF-CP/
SOCl2/NA. Controls were also prepared: PPF-CP blanks were
obtained by printing CP films onto PPF without subsequent
activation or immersion in NA/ACN; for PPF-CP/NA blanks,
only the activation step was omitted; and for PPF-H2SO4/
SOCl2/NA, blanks were prepared by printing PPF with blank
1MH2SO4, followed by “activation” and immersion in NA/
ACN. Prior to their analysis, all samples and controls were
sonicated for 5 min in ACN.
AFM depth profiling results are shown in Table 2, and cyclic
voltammograms of the modified surfaces and typical AFM line
profiles are shown in Figures S-3 and S-4 (Supporting Information).
Activation of the CP film and reaction with NA increased the film
thickness from 1.0 ± 0.3 to 2.0 ± 0.5 nm, consistent with the
coupling of NA groups to the CP layer. When the activation step
7032 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 5. AFM image of VACNTs tethered to a patterned AP layer
on PPF.
was omitted, there was no change in film thickness, confirming
that NA does not physisorb to the CP film. Interestingly, a thin
surface layer of NA was detected electrochemically (Figure S-3,
Supporting Information) and by AFM measurements on the
PPF-H2SO4/SOCl2/NA controls. The measured thickness (0.4
± 0.2 nm) of this film is less than expected for a monolayer of
NA groups (0.8 nm), indicating a submonolayer coverage. We
assume that NA couples directly to a low concentration of
carboxylate functionalities on the (otherwise) bare PPF surface.
SWCNT + AP on PPF. This second example of the utility of
printed films as tethers is based on recent work in which we
assembled and characterized vertically aligned carbon nanotube
(VACNT) forests on AP films electrografted to PPF. 46 To test
whether printed AP films could be used similarly, PPF surfaces
were patterned using ABD/0.5 M HCl ink and then immersed in
a DMSO solution (2 mL) of cut SWCNTs (0.2 mg mL -1 ) and N,N′dicyclohexylcarbodiimide
(1 mg mL -1 )for24hat65°C. These
conditions promote formation of amide bonds between surfaceimmobilized
AP groups and carboxylate groups at the cut ends
of the SWCNTs. The resultant surfaces were sonicated in
acetone for 10 s and then in isopropyl alcohol for 10 s prior to
imaging by AFM. The image shown in Figure 5 (and the SEM
image in Figure S-5, Supporting Information) is similar to those
previously obtained for VACNTs on electrografted AP tether
layers. 46
These examples demonstrate that MCP yields tether layers
with their usual reactivity, and hence, the method can be used to
prepare patterned substrates which form the basis of more
complex structures.
Buildup MCP Patterning of Two-Component Surfaces.
Two- or multicomponent films in which secondary modifiers are
patterned on top of a continuous base film have potential
applications in sensing, where (for example) the base film is
tailored to reduce nonspecific interactions with the analyte while
the patterned secondary modifiers act either as tethers for
attachment of recognition species or as the recognition elements
themselves. This “buildup” method relies on the ability of the
substrate to reduce the secondary modifiers by electron transfer
across the base film. Reduction of the printed secondary modifier
generates radicals which couple to the base film. Hence, a
covalently coupled structure spontaneously forms in a single,
simple step requiring no additional reagents. The buildup method
(46) Garrett, D. J.; Flavel, B. S.; Shapter, J. G.; Baronian, K. H. R.; Downard,
A. J. Langmuir 2010, 26, 1848–1854.

Table 3. Film Thickness Measurements on PPF
Substrates Modified with CP Films with and without
Overprinting of a NP Layer a
film thickness/nm b
sonication solvent and time CP region CP/NP region
H2O (5 min) 1.6 ± 0.3 2.2 ± 0.3
H2O (5 min) 1.4 ± 0.3 2.1 ± 0.2
H2O (5 min) + ACN (30 min) 1.7 ± 0.6 2.1 ± 0.3
H2O (5 min) + ACN (30 min) 1.5 ± 0.4 2.0 ± 0.2
a Samples were prepared in duplicate. b AFM line profiles are shown
in Figure S-7 (Supporting Information).
Figure 6. SEM images of surfaces that were immersed in 10 mM
CBD/0.1 H2SO4 for 30 min and subsequently printed with (a) ABD/
0.5 M HCl and (b) 0.5 M HCl inks. After printing, the surfaces were
immersed in Au nanoparticle solution for 40 min.
was demonstrated by printing the secondary modifiers NP and
AP.
CP + NP on PPF. NP groups were printed onto a base CP
film grafted spontaneously to PPF at OCP. Two ∼3.0 × 1.5 mm 2
PPF samples were immersed in 10 mM CBD/0.1 M H2SO4
solution for 30 min to form the base films. One half of each
sample was then printed with a nonpatterned stamp using
NBD/1 M H2SO4 ink. Cyclic voltammetry (Figure S-6, Supporting
Information) confirmed that NBD had reacted in the
expected manner giving an NP layer.
AFM depth profiling measurements of printed and unprinted
sections (Table 3) show that printing increases the film thickness,
consistent with attachment of NP to the CP layer. The magnitude
of the increase (∼0.4-0.7 nm) corresponds to the addition of a
submonolayer of NP groups on top of the CP film. However, NP
is also expected to couple within the CP film, and hence, the
concentration of printed NP groups may be higher than indicated
by film thickness data.
To test the stability of the printed layers, the samples were
sonicated in ACN for 30 min and the AFM measurements were
repeated. The data in the lower part of Table 3 indicate that the
film thicknesses did not change significantly, consistent with
covalent attachment both to the substrate surface and between
the base and printed layers.
CP + AP on PPF. In the second example of the buildup
printing approach, AP groups were patterned onto a spontaneously
grafted layer of CP groups using a patterned stamp with ABD/
0.5 M HCl ink. A blank was also prepared on a CP layer by printing
with blank 0.5 M HCl ink. The printed surfaces were immersed
for 40 min in a solution of Au nanoparticles and then sonicated in
H 2O for 30 s prior to analysis by SEM. Figure 6 shows the SEM
images where the assembled Au nanoparticles clearly reveal the
patterned immobilization of AP (Figure 6a) in comparison with
the control (Figure 6b).
The buildup approach for preparing patterned two- or multicomponent
surfaces should be applicable to all substrates at which
grafting from aryldiazonium salt solutions proceeds spontaneously
at OCP. However, the requirement that the second aryldiazonium
cation be reduced by the substrate places some limitations on
the nature and thickness of the base film and also on the
aryldiazonium cation derivative. For strongly reducing substrates
(such as zinc), it should be possible to print even relatively difficultto-reduce
aryldiazonium cations onto thick base films. On the
other hand, printing on less-reducing substrates (such as Au) is
likely to be successful only with easily reduced aryldiazonium salt
derivatives on thin base films. In the examples above, the base
films were formed by spontaneous grafting from a aryldiazonium
salt solution. There is no reason why MCP, electrografting, and/
or other classes of modifiers cannot be used. Methods such as
electro-oxidation of primary amines 47 or arylhydrazines; 48 electroreduction
of iodonium, 49,50 sulfonium 51 salts, or vinylic compounds;
52 photolytic or thermal grafting of alkenes and alkynes; 53-56
or photografting of arylazides 57 would greatly widen the range of
functionalities that could be added to the surface.
CONCLUSION
Microcontact printing using aryldiazonium salt inks has been
applied to carbon, metal, and semiconductor substrates to give
stable, covalently attached, thin films. The thickness and the
morphology of the printed film appear to depend on the potential
driving force for reduction of the aryldiazonium cation by the
substrate. Aminophenyl and carboxyphenyl groups in printed
layers retain the ability to form amide bonds with solution species
and, consequently, provide useful tethers for more complex
surface structures.
Microcontact printing using aryldiazonium salts is applicable
to all substrate-diazonium salt combinations for which surface
modification proceeds spontaneously at open circuit potential in
solution. The variable film thickness and roughness, which is
substrate- and film-dependent, may be a limitation for some
potential applications; however, compared with other methods for
patterning layers using aryldiazonium salts, microcontact is low
cost and simple to implement with no requirement for electrochemical
capability. Tightly defined patterns with feature sizes
tens of micrometers upward can be routinely prepared, and
(47) Barbier, B.; Pinson, J.; Desarmot, G.; Sanchez, M. J. Electrochem. Soc. 1990,
137, 1757–1764.
(48) Malmos, K.; Iruthayaraj, J.; Pedersen, S. U.; Daasbjerg, K. J. Am. Chem.
Soc. 2009, 131, 13926–13927.
(49) Vase, K. H.; Holm, A. H.; Norrman, K.; Pedersen, S. U.; Daasbjerg, K.
Langmuir 2007, 23, 3786–3793.
(50) Vase, K. H. j.; Holm, A. H. k.; Pedersen, S. U.; Daasbjerg, K. Langmuir
2005, 21, 8085–8089.
(51) Vase, K. H.; Holm, A. H.; Norrman, K.; Pedersen, S. U.; Daasbjerg, K.
Langmuir 2008, 24, 182–188.
(52) Palacin, S.; Bureau, C.; Charlier, J.; Deniau, G.; Mouanda, B.; Viel, P.
ChemPhysChem 2004, 5, 1469–1481.
(53) Lasseter, T. L.; Cai, W.; Hamers, R. J. Analyst 2004, 129, 3–8.
(54) Ssenyange, S.; Anariba, F.; Bocian, D. F.; McCreery, R. L. Langmuir 2005,
21, 11105–11112.
(55) Sun, B.; Colavita, P. E.; Kim, H.; Lockett, M.; Marcus, M. S.; Smith, L. M.;
Hamers, R. J. Langmuir 2006, 22, 9598–9605.
(56) Yu, S. S. C.; Downard, A. J. Langmuir 2007, 23, 4662–4668.
(57) Gross, A. J.; Yu, S. S. C.; Downard, A. J. Langmuir 2010, 26, 7285–7292.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7033

multicomponent patterned surfaces can be fabricated simply by
printing on top of a base film. The two-component structures were
shown to be stable to prolonged sonication, indicating covalent
coupling of the second modifier to the base layer. The availability
of such a method will facilitate application of the aryldiazonium
salt surface modification approach in areas such as the fabrication
of bio- and chemical sensors.
ACKNOWLEDGMENT
This work was supported by the MacDiarmid Institute for
Advanced Materials and Nanotechnology. J.L. and D.J.G thank
the New Zealand Tertiary Education Commission for doctoral
scholarships and B.S.F. thanks the Australian Government’s
Endeavour Research Fellowship program. We thank Dr. John
Loring for use of the Linkfit curve fitting software.
7034 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
SUPPORTING INFORMATION AVAILABLE
Characterizations of the following modified surfaces prepared
by printing using nonpatterned stamps: AP on PPF, NP on Au,
and NP on HF-treated Si; cyclic voltammograms of CP layers and
bare PPF after activation with SOCl2 and reaction with 4-nitroaniline;
repeat cyclic voltammograms of a PPF surface bearing
a CP film, overprinted with an NP layer; figures of AFM line
profiles of data presented in Tables 2 and 3; SEM image of the
surface shown in Figure 5. This material is available free of charge
via the Internet at http://pubs.acs.org.
Received for review July 6, 2010. Accepted July 10, 2010.
AC101785C

Anal. Chem. 2010, 82, 7035–7043
Quantitation, Visualization, and Monitoring of
Conformational Transitions of Human Serum
Albumin by a Tetraphenylethene Derivative with
Aggregation-Induced Emission Characteristics
Yuning Hong, † Chao Feng, ‡ Yong Yu, †,‡ Jianzhao Liu, † Jacky Wing Yip Lam, † Kathy Qian Luo, ‡,§
and Ben Zhong Tang* ,†,#
Nano Science and Technology Program, Department of Chemistry, Institute of Molecular Functional Materials,
Bioengineering Program, and Department of Chemical and Biomolecular Engineering, The Hong Kong University of
Science and Technology (HKUST), Clear Water Bay, Kowloon, Hong Kong, China, School of Chemical and
Biomedical Engineering, Nanyang Technological University, Singapore 637457, and Department of Polymer Science
and Engineering, Institute of Biomedical Macromolecules, Key Laboratory of Macromolecular Synthesis and
Functionalization of the Ministry of Education, Zhejiang University, Hangzhou 310027, China
Human serum albumin (HSA) is a major protein component
of blood plasma, and its assay is of obvious value to
biological research. We, herein, present a readily accessible
fluorescent bioprobe for HSA detection and quantitation.
A nonemissive tetraphenylethene derivative named
sodium 1,2-bis[4-(3-sulfonatopropoxyl)phenyl]-1,2-diphenylethene
(BSPOTPE) is induced to emit by HSA, showing
a novel phenomenon of aggregation-induced emission
(AIE). The AIE bioprobe enjoys a broad working range
(0-100 nM), a low detection limit (down to 1 nM), and a
superior selectivity to albumins. The fluorescent bioassay
is unperturbed by the miscellaneous bioelectrolytes in the
artificial urine. The AIE luminogen can also be used as a
rapid and sensitive protein stain in gel electrophoresis for
HSA visualization. Utilizing the AIE feature of BSPOTPE
and the Förster resonance energy transfer from HSA to
BSPOTPE, the unfolding process of HSA induced by
guanidine hydrochloride is monitored, which reveals a
multistep transition with the involvement of molten globule
intermediates. Computational modeling suggests that
the AIE luminogens dock in the hydrophobic cleft between
subdomains IIA and IIIA of HSA with the aid of hydrophobic
effect, charge neutralization, and hydrogen bonding
interactions, offering mechanistic insight into the
microenvironment inside the hydrophobic cavity.
Human serum albumin (HSA) is the most abundant protein
in the circulatory system and plays multiple biological functions
in the human body. 1 For example, it regulates water balance
between blood and tissues and serves as a physiological carrier
for various endogenous and exogenous substances that are
* To whom correspondence should be addressed. E-mail: tangbenz@ust.hk.
† Nano Science and Technology Program, Department of Chemistry, Institute
of Molecular Functional Materials, and Bioengineering Program, HKUST.
‡ Department of Chemical and Biomolecular Engineering, HKUST.
§ Nanyang Technological University.
# Zhejiang University.
(1) Peters, T., Jr. Adv. Protein Chem. 1985, 37, 161.
partially soluble in the bloodstream, such as bilirubin, fatty acids,
and drugs. 2-5 The albumin is synthesized in the liver. A low level
of albumin in the blood serum, known as hypoproteinemia, may
indicate liver failure, cirrhosis, and chronic hepatitis. 6 When blood
passes through healthy kidneys, the kidneys filter out waste
products but keep the substances the body needs, such as albumin
and other proteins. Appearance of an excess amount of proteins
in urine is a sign of chronic kidney disease, which may result in
diabetes, high blood pressure, and problems associated with
inflammation in the kidneys. 7 Many studies have been done on
diabetic nephrosis, with microalbuminuria being identified as an
early sign. 8 Microalbuminuria is commonly diagnosed by elevated
protein concentration (30-300 mg/L) in the urine on at least two
occasions. A higher albumin value is regarded as albuminuria.
The urine-testing dipsticks impregnated with pH sensitive dyes
are normally used for daily health monitoring. 9 The level of
albumin caused by microalbuminuria, however, cannot be accurately
assayed by the dipsticks due to their low sensitivity. 10 It
is, thus, of clinical value to develop effective methods for urinary
protein detection and quantitation.
Colorimetric methods, such as Brandford and Lowry assays,
have traditionally been used for protein quantitation in solutions. 11,12
These methods, however, generally lack sensitivity and accuracy,
(2) Carter, D. C.; Ho, J. X. Adv. Protein Chem. 1994, 45, 153.
(3) (a) Neuzil, J.; Stocker, R. J. Biol. Chem. 1994, 269, 16712. (b) Hu, Y. J.;
Liu, Y.; Xiao, X. H. Biomacromolecules 2009, 10, 517.
(4) Berde, C. B.; Hudson, B. S.; Simoni, R. D.; Sklar, L. A. J. Biol. Chem. 1979,
254, 391.
(5) Sjoholm, I.; Ekman, B.; Kober, A. Mol. Pharmacol. 1979, 16, 767.
(6) Murch, S. H.; Winyard, P. J. D.; Koletzko, S.; Wehner, B.; Cheema, H. A.;
Risdon, R. A.; Phillips, A. D.; Meadows, N.; Klein, N. J.; Walker-Smith, J. A.
Lancet 1996, 347, 1299.
(7) Hoogenberg, K.; Sluiter, W. J.; Dullaart, R. P. F. Acta Endrocrinol. 1993,
129, 151.
(8) (a) Mogensen, C. E.; Keane, W. F.; Bennett, P. H.; Jerums, G.; Parving,
H. H.; Passa, P.; Steffes, M. W.; Striker, G. E.; Viberti, G. C. Lancet 1995,
346, 1080. (b) Nomata, S.; Haneda, K.; Moriya, T.; Katayama, S.; Iwamoto,
Y.; Sakai, H.; Tomino, Y.; Matsuo, S.; Asano, Y.; Makino, H. Jpn. J. Nephrol.
2005, 47, 768.
(9) Rose, B. D. Pathophysiology of Renal Disease; McGraw Hill: New York, 1987.
(10) Martinez, A. W.; Phillips, S. T.; Butte, M. J.; Whitesides, G. M. Angew. Chem.,
Int. Ed. 2007, 46, 1318.
10.1021/ac1018028 © 2010 American Chemical Society 7035
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/20/2010

in addition to their lengthy procedures. 13 Biosensors based on
fluorescent (FL) materials have attracted much attention due to
their sensitivity, selectivity, and rapidity. Organic molecules and
organometallic complexes have been developed for protein
assays. 14,15 Upon complexation or conjugation with proteins, the
luminophores show emission enhancements and/or spectral
shifts, which enable the biopolymers to be detected and quantified.
Some of the FL probes, however, are insoluble in aqueous media,
while others are unstable under ambient conditions. 16 Some
luminophores show spectacular performances in protein assays
but are extremely expensive because of their painstaking
syntheses. 14,15 These drawbacks greatly limit the scope of their
real-life high-tech applications. This calls for the development of
synthetically readily accessible and environmentally stable bioprobes
with high sensitivity and selectivity.
A thorny problem associated with the emissions of conventional
luminophores in aqueous medium or physiological buffer
is aggregation-caused quenching (ACQ). 17 The luminophores are
in close vicinity in the aggregates suspended in the aqueous buffer,
which favors the formation of such detrimental species as
excimers and exciplexes and causes nonradiative relaxations of
the excited states. We have recently discovered that tetraphenylethene
(TPE) behaves in a way exactly opposite to the ACQ
dyes: it is nonemissive in the solution state but becomes highly
luminescent in the aggregate state. 18 We coined “aggregationinduced
emission” (AIE) for the phenomenon because the
nonemissive TPE is induced to emit by aggregate formation.
Decorating TPE with ionic or polar functional groups yields watersoluble
derivatives, which can be utilized as FL probes for
bioanalyses. 19,20 Our and other research groups have successfully
used the TPE-based AIE luminogens for nucleic acid detection,
enzymatic activity assay, and metallic ion tracing. 21,22 The successes
prompted us to further explore their potentials in bioanaly-
(11) Bradford, M. M. Anal. Biochem. 1976, 72, 248.
(12) Lowry, O. H.; Rosebrough, N. J.; Farr, A. L.; Randall, R. J. J. Biol. Chem.
1951, 193, 265.
(13) Haugland, R. P. Handbook of Fluorescent Probes and Research Chemicals;
Molecular Probe: Leiden, 2002, p. 420.
(14) (a) Giepmans, B. N. G.; Adams, S. R.; Ellisman, M. H.; Tsien, R. Y. Science
2006, 312, 217. (b) Royer, C. A. Chem. Rev. 2006, 106, 1769. (c) Suzuki,
Y.; Yokoyama, K. J. Am. Chem. Soc. 2005, 127, 17799. (d) Matulis, D.;
Baumann, C. G.; Bloomfield, V. A.; Lovrien, R. E. Biopolymers 1999, 49,
451. (e) Yarmoluk, S. M.; Kryvorotenko, D. V.; Balanda, A. O.; Losytskyy,
M. Y.; Kovalska, V. B. Dyes Pigm. 2001, 51, 41. (f) Hawe, A.; Sutter, M.;
Jiskoot, W. Pharm. Res. 2008, 25, 1487.
(15) (a) Eryazici, I.; Moorefield, C. N.; Newkome, G. R. Chem. Rev. 2008, 108,
1834. (b) Keefe, M. H.; Benkstein, K. D.; Hupp, J. T. Coord. Chem. Rev.
2000, 205, 201.
(16) (a) Matulis, D.; Lovrien, R. Biophys. J. 1998, 74, 422. (b) Davis, D. M.;
Birch, D. J. S. J. Fluoresc. 1996, 6, 23.
(17) Birks, J. B. Photophysics of Aromatic Molecules; Wiley: London, 1970.
(18) Hong, Y.; Lam, J. W. Y.; Tang, B. Z. Chem. Commun. 2009, 4332.
(19) Li, Z.; Dong, Y. Q.; Lam, J. W. Y.; Sun, J.; Qin, A.; Haussler, M.; Dong,
Y. P.; Sung, H. H. Y.; Williams, I. D.; Kwok, H. S.; Tang, B. Z. Adv. Funct.
Mater. 2009, 19, 905.
(20) (a) Yuan, C. X.; Tao, X. T.; Wang, L.; Yang, J. X.; Jiang, M. H. J. Phys. Chem.
C 2009, 113, 6809. (b) Zhao, M.; Wang, M.; Liu, H.; Liu, D.; Zhang, G.;
Zhang, D.; Zhu, D. Langmuir 2009, 25, 676. (c) Wang, M.; Zhang, D.;
Zhang, G.; Zhu, D. Chem. Commun. 2008, 4469. (d) Suzuki, Y.; Yokoyama,
K. J. Am. Chem. Soc. 2005, 127, 17799.
(21) (a) Tong, H.; Hong, Y.; Dong, Y.; Haussler, M.; Lam, J. W. Y.; Li, Z.; Guo,
Z.; Guo, Z.; Tang, B. Z. Chem. Commun. 2006, 3705. (b) Tong, H.; Hong,
Y.; Dong, Y.; Haussler, M.; Li, Z.; Lam, J. W. Y.; Dong, Y.; Sung, H. H. Y.;
Williams, I. D.; Tang, B. Z. J. Phys. Chem. B 2007, 111, 11817. (c) Hong,
Y.; Haussler, M.; Lam, J. W. Y.; Li, Z.; Sin, K. K.; Dong, Y.; Tong, H.; Liu,
J.; Qin, A.; Renneberg, R.; Tang, B. Z. Chem.sEur. J. 2008, 14, 6428.
7036 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Chart 1. Chemical Structure of Water-Soluble AIE
Luminogen of BSPOTPE
ses. In this work, we examined the utility of a TPE salt, sodium
1,2-bis[4-(3-sulfonatopropoxyl)phenyl]-1,2-diphenylethene(BSPOTPE;
Chart 1), as a bioprobe for HSA detection and quantitation. The
FL “turn-on” attribute of BSPOTPE by its complexation with
albumin facilitated the quantitative assay and visual observation
of HSA in the aqueous buffer and gel electrophoresis, respectively.
It is well-known that proper biological functions of proteins
are associated with their specific strand conformations and folding
structures. 23 Understanding of protein folding is of fundamental
importance for proteomic and pharmaceutical research. 24 HSA is
a polypeptide chain with three R-helical domains (I-III), which
are further divided into two subdomains (A and B). 1,25 Its crystal
structure shows that the main ligand-binding sites in the albumin
are located in the hydrophobic cavities of subdomains IIA and
IIIA, which are sometimes referred to as Sudlow sites I and II,
respectively. 26 Conformation analyses of the hydrophobic cavities
play an important role in drug development, especially pharmacokinetic
and pharmacodynamic investigations. 27
Study of conformation transitions of proteins in the presence
of denaturants is a topic of great interest because it can offer
mechanistic insights into folding and unfolding processes of the
biopolymers. 28,29 Though intermediate states have been suspected
to be involved in the unfolding and refolding processes of many
proteins, they are often not detected due to the lack of appropriate
probes. 29 Characterization of the intermediate states becomes even
more complex in the multidomain proteins, such as HSA, in which
each domain is capable of unfolding and refolding independently. 30
Whether intermediate states are involved in the unfolding pathway
of HSA has been an issue of debate. In this study, we made use
of the AIE feature of BSPOTPE and investigated the unfolding
process of HSA. A stable molten-globule intermediate was
observed in its unfolding process induced by guanidine hydrochloride
(GndHCl), a well-known denaturant. Förster resonance
energy transfer (FRET) study proved the accessibility of HSA by
BSPOTPE and suggested probable location of the FL bioprobe
in the hydrophobic cavity in the protein folding structure.
Combination of the FL technique with circular dichroism (CD),
(22) (a) Wang, M.; Gu, X.; Zhang, G.; Zhang, D.; Zhu, D. Anal. Chem. 2009,
81, 4444. (b) Wang, M.; Zhang, D.; Zhang, G.; Tang, Y.; Wang, S.; Zhu, D.
Anal. Chem. 2008, 80, 6443.
(23) Flora, K.; Brennan, J. D.; Baker, G. A.; Doody, M. A.; Bright, F. V. Biophys.
J. 1998, 75, 1084.
(24) Jusko, W. J.; Gretch, M. Drug Metab. Rev. 1976, 5, 43.
(25) He, X. M.; Carter, D. C. Nature 1992, 358, 209.
(26) (a) Sudlow, G.; Birkett, D. J.; Wade, D. N. Mol. Pharmacol. 1975, 11, 824.
(b) Sudlow, G.; Birkett, D. J.; Wade, D. N. Mol. Pharmacol. 1976, 12, 1052.
(27) Lavinder, J. J.; Hari, S. B.; Sullivan, B. J.; Magliery, T. J. J. Am. Chem. Soc.
2009, 131, 3794.
(28) Abou-Zied, O. K.; Al-Shihi, O. I. K. J. Am. Chem. Soc. 2008, 130, 10793.
(29) (a) Nolting, B.; Andert, K. Protein Struct. Funct. Genet. 2000, 41, 288. (b)
Santra, M. K.; Banerjee, A.; Krishnakumar, S. S.; Rahaman, O.; Panda, D.
Eur. J. Biochem. 2004, 271, 1789.
(30) Ahmad, B.; Ahmed, M. Z.; Haq, S. K.; Khan, R. H. Biochim. Biophys. Acta
Protein Proteomics 2005, 1750, 93.

Figure 1. (A) FL spectra of BSPOTPE in the PBS buffer containing different amounts of HSA. (B) Change in the FL intensity at 475 nm with
HSA concentration; I0 ) FL intensity in the absence of HSA. Inset: linear region of the binding isotherm of BSPOTPE to HSA. [BSPOTPE] )
5 µM; λex ) 350 nm.
differential scanning calorimetry (DSC), and molecular modeling
helped draw a mechanistic picture on the protein unfolding
process.
MATERIALS AND METHODS
General Information. BSPOTPE was prepared according to
our previously published procedures. 21b HSA, GndHCl, and other
biomolecules were all purchased from Sigma and used as received.
Phosphate buffered saline (PBS) with pH of 7.0 was purchased
from Merck. Water was purified by a Millipore filtration system.
All the experiments were performed at room temperature unless
otherwise specified.
UV spectra were measured on a Milton Roy Spectronic 3000
Array spectrophotometer, and FL spectra were recorded on a
Perkin-Elmer LS 55 spectrofluorometer with a Xenon discharge
lamp excitation. CD spectra were recorded on a Jasco J-810
spectropolarimeter in a 1 mm quartz cuvette using a step
resolution of 0.2 nm, a scan speed of 100 nm/min, a sensitivity of
0.1°, and a response time of 0.5 s. Each spectrum is the average
of three scans. Details about the artificial urine preparation,
cytotoxicity assay, FRET study, DSC and pH-dependent FL
measurements, and computation modeling are given in the
Supporting Information.
Sample Preparation. Stock solutions of BSPOTPE and HSA
with a concentration of 1.0 mM were prepared by dissolving
appropriate amounts of the luminogen and protein in the PBS
buffer. The final concentration of HSA in PBS was double checked
by measuring its absorbance at 279 nm. In the HSA unfolding
study, HSA (0.4 µM) in PBS was incubated in the presence of
different amounts of GndHCl (0.2-7.0 M) at 25 °C for 30 min.
Afterward, BSPOTPE was added to the mixtures and incubated
for another 30 min before spectral measurements. FL spectra were
recorded in the wavelength range of 370-670 nm using 350 nm
as the excitation wavelength. For the intrinsic tryptophan fluorescence
study, FL spectra were collected from 310 to 570 nm
using 290 nm as the excitation source. In the refolding experiment,
0.2 mM HSA was incubated in 8 M GndHCl for 24 h at 25 °C to
ensure that all the proteins were fully unfolded. The denatured
sample was then diluted with PBS until the final concentrations
of HSA and GndHCl were equal to 0.4 µM and less than 0.1 M,
respectively. Fraction of refolded proteins (F r) was calculated
from the following equation:
F r ) 1 - I N - I R
I N - I D
where IN and ID are the FL intensities of native and denatured
HSA-BSPOTPE complexes, respectively, and IR is the FL
intensity of BSPOTPE recovered from the refolded HSA.
Electrophoresis Assay. A poly(acrylamide) gel electrophoresis
(PAGE) experiment was performed on a Hoefer miniVE system
under nondenaturing conditions using 5% stacking and 12%
resolving native poly(acrylamide) gel at 100 V for 3hat4°C.
After electrophoresis, the gel was soaked in an aqueous solution
of BSPOTPE (1 mg in 100 mL ddH2O) at room temperature for
5 min. Alternatively, the gel was stained with a Coomassie
Brilliant Blue (CBB) solution (0.1% w/v Coomassie Blue R250
in an aqueous mixture containing 10% methanol and 7% acetic
acid) for6htoovernight on a rotary shaker with gentle mixing,
followed by destaining in an aqueous solution containing 10%
methanol and 7% acetic acid for 1 to 2 h until the background
of the gel became transparent. An AlphadigiDoc system with
a DE-500 MultiImage II light cabinet and an ML-26 UV
transilluminator (Alpha Innotech) was used for data collection
and analysis.
RESULTS AND DISCUSSION
Protein Quantitation in Solution. BSPOTPE is soluble in
water but insoluble in common organic solvents, such as acetonitrile,
THF, and chloroform. Its FL quantum yield (ΦF) is
increased from 0.37% in water to 17.5% in acetonitrile, where
the luminogen molecules aggregate. The BSPOTPE solution
in PBS is feebly luminescent at 390 nm in the absence of HSA
(Figure 1A). When a small amount of HSA is added, the
BSPOTPE solution becomes luminescent. Its FL intensity at 475
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(1)
7037

Figure 2. (A) Binding isotherm of HSA to BSPOTPE in the artificial urine and PBS buffer. (B) Dependence of the FL intensity of BSPOTPE at
470 nm on different proteins in artificial urine and PBS buffer. [BSPOTPE] ) 5 µM; [protein] ) 100 µg/mL; λex ) 350 nm.
nm keeps rising with an increase in the HSA concentration. The
rate of the FL enhancement is fast at lower HSA concentrations
and becomes almost constant at [HSA] > 1 µM (Figure 1B). At
an HSA concentration of 10 µM, the FL intensity is increased by
as high as ∼300-fold. The detection limit can be squeezed to as
low as 1 nM. In the protein concentration range of 0-100 nM,
the plot of FL enhancement (I/I0 - 1) as a function of HSA
concentration is a linear line with a high correlation coefficient
(0.995).
To examine the feasibility of protein assay in body fluids, we
performed HSA quantitation in artificial urine. 31 As shown in
Figure 2A, the binding isotherm is practically identical to that in
PBS. The assay sensitivity is not affected by the presence of
miscellaneous bioelectrolytes and physiological level of urea.
Thanks to the intriguing AIE characteristic of BSPOTPE, the
detection limit and linear range can be tuned by adjusting the
luminogen concentration. 21b Besides the superior sensitivity,
BSPOTPE shows excellent selectivity to albumin proteins, such
as HSA and bovine serum albumin (BSA; Figure 2B). A variety
of human proteins [pepsin, human immunoglobulin G (IgG),
trypsin, etc.] with isoelectric points ranging from 1 to 10 and DNA
[e.g., calf thymus DNA] were chosen for the binding tests. None
of these biomolecules, however, can turn on the BSPOTPE
emission. The selectivity of BSPOTPE toward the albumin proteins
remains unperturbed in the artificial urine. These exciting results
encourage us to further explore its clinical utility.
Protein Visualization in PAGE. The “lighting-up” of BSPOTPE
luminogen by albumin in the PBS buffer inspired us to check the
possibility of the use of the AIE luminogen as a protein staining
reagent in the PAGE assay. Figure 3A shows the gel images of
electrophoresed HSA after staining with a BSPOTPE solution for
∼5 min. The protein bands become visible under the illumination
of a hand-held UV lamp. The detection limit can be down to 50
(31) Brooks, T.; Keevil, C. W. Lett. Appl. Microbiol. 1997, 24, 203.
7038 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Figure 3. PAGE analyses of HSA using (A) BSPOTPE and (B)
Coomassie Blue R-250 as the staining reagents. Lanes correspond
to the protein bands with amounts of HSA of (1) 1000, (2) 500, (3)
100, (4) 50, and (5) 2.5 ng. [BSPOTPE] ) 1 mg/100 mL; staining
time ≈ 5 min.
ng per lane (Lane 4). CBB is one of the most commonly used
protein stains in the bioassay. 32 The gel stained by CBB is shown
in Figure 3B for comparison. Although we have made efforts to
enhance the contrast of the image, only a few bands are
discernible in the gel. Those fine bands containing trace amounts
of HSA are ignored by CBB. It must be pointed out that a high
CBB concentration is generally needed for staining. Evidently,
BSPOTPE shows a much higher sensitivity than CBB.
Another advantage of the use of BSPOTPE as gel stain is its
simplicity and promptness, in contrast to the conventional methods.
Commercially available protein staining reagents such as
colorimetric CBB and fluorimetric SYPRO Ruby require a long
staining time (normally overnight). A destaining step is often
required to remove the excess dye in the gel. 33 Silver staining is
expensive and entails several labor-intensive and time-sensitive
steps. 13 Unlike the above probes, neither careful timing nor a
destaining step is necessary for BSPOTPE staining. The protein
bands can be visualized after 5 min of staining. Soaking the gel
(32) (a) Blakesley, R. W.; Boezi, J. A. Anal. Biochem. 1977, 82, 580. (b)
Candiano, G.; Bruschi, M.; Musante, L.; Santucci, L.; Ghiggeri, G. M.;
Carnemolla, B.; Orecchia, P.; Zardi, L.; Righetti, P. G. Electrophoresis 2004,
25, 1327.
(33) Lopez, M. F.; Berggren, K.; Chernokalskaya, E.; Lazarev, A.; Robinson, M.;
Patton, W. F. Electrophoresis 2000, 21, 3673.

Figure 4. (A) Variation in the FL intensity of BSPOTPE at 470 nm with concentration of GndHCl in the presence or absence of HSA. (B)
Dependence of the FRET ratio (I470/I350) on the concentration of GdnHCl. [BSPOTPE] ) 5 µM; [HSA] ) 2 µM; λex ) 350 nm (for A), 295 nm (for
B).
in the luminogen solution for longer time does not cause
overstaining. As the resolution of the protein band is appreciably
high, no washing step is required, which vastly shortens the time
and lowers the workload.
Irritant acetic acid and methanol are used in the gel electrophoresis
assays using CBB and SYPRO Ruby dyes as staining
reagents. BSPOTPE is soluble and stable in aqueous media, and
the use of toxic chemicals can, thus, be avoided. To learn whether
BSPOTPE is a “safe” stain, cytotoxicity assays were performed
with HeLa cells on the basis of reduced activity of methyl thiazolyl
tetrazolium (MTT). As can be seen from Figure S1 (Supporting
Information), the living cells grow as normally as they do in the
control experiment in the absence of BSPOTPE. The luminogen
neither inhibits nor promotes the growth of the cells. In other
words, it is cytocompatible without interfering with the metabolisms
of the living cells.
Conformation Studies of HSA by BSPOTPE. A few FL dyes
have been used to investigate the HSA unfolding process induced
by GndHCl, but each of them gives a different result. 14c,29b,34
Whether intermediate states are formed and how many steps are
involved in the unfolding process remain to be a controversial
issue. In this study, we tried to find out answers by utilizing the
AIE characteristic of the BSPOTPE luminogen.
Free BSPOTPE is practically nonluminescent in the aqueous
buffer, which is virtually unaffected by the addition of different
amounts of GndHCl (open circles in Figure 4A). Addition of HSA,
however, turns on the BSPOTPE emission. The FL intensity at
475 nm changes when the HSA is denatured by GndHCl. Analysis
of the isotherm shown by the solid circles in Figure 4A reveals
that three transition steps are involved in the GndHCl-induced
denaturation process. The first transition step occurs at low
concentrations of GndHCl (

Figure 5. (A) CD spectra of HSA in the PBS buffer in the presence of different amounts of GndHCl. (B) Plot of ellipticity of HSA at 220 nm
versus GndHCl concentration.
and HSA in the different steps are estimated. (See Supporting
Information for details.)
The change in the secondary structure of HSA during the
GndHCl-induced denaturation process was monitored by far-UV
CD. The CD spectrum of HSA in the PBS buffer shows an R-helix
profile with two minima at 208 and 220 nm (Figure 5A). 35 When
the GndHCl concentration is lower than 1.5 M, the CD signals
remain almost unchanged. Afterward, the ellipticity starts to
decrease progressively and continuously, indicating that the
protein chains are gradually transformed from helical strands to
random coils.
The change in the ellipticity at 220 nm against the GndHCl
concentration is plotted in Figure 5B. No intermediate states are
detected by the CD technique. It is known that the Cotton effects
in the far-UV region provide valuable information on the secondary
structures of proteins. 30 Details on the tertiary structure, however,
are not available in this spectral region. The plateau of the
transition curve suggests that the secondary structure of HSA
remains intact at the low GndHCl concentrations. The attenuation
of the FL signals by GndHCl in this region, however, is inconsistent
with the CD data. This suggests that the tertiary structure
of HSA, rather than its secondary structure, is perturbed at the
low GndHCl concentrations. The partial recovery of the fluorescence
of BSPOTPE at ∼2.0 M GndHCl hints the formation of a
stable intermediate. At the high GndHCl concentrations, the HSA
strand is unfolded and the CD signal is accordingly weakened.
Accompanying the unfolding process, the luminogen molecules
are released back to the buffer solution, leading to the observed
decrease in the FL intensity.
DSC has been commonly used for the studies of stability of
biological macromolecules and folding of proteins. 36 The thermal
transition midpoint (Tm), where 50% of biomolecules are unfolded,
can directly be obtained from the DSC measurement.
The higher the Tm value, the more stable is the structure. The
Tm values for HSA in the presence of 0, 0.8, 1.5, and 2.0 M
(35) Sreerama, N.; Venyaminov, S. Y.; Woody, R. W. Protein Sci. 1999, 8, 370.
(36) Bruylants, G.; Wouters, J.; Michaux, C. Curr. Med. Chem. 2005, 12, 2011.
7040 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
GndHCl are 63.5, 46.9, 34.9, and 46.5 °C, respectively (Supporting
Information, Figure S4). Note that there is no transition
midpoint for HSA in the presence of 6.0 M GndHCl, where the
protein folding structure is completely collapsed. From this set
of data, we can elicit that the stability of the molten-globule
intermediate at 2.0 M GndHCl is lower than that of the native
folding structure but much higher than that of the unfolded
random coil.
Protein denaturation by GndHCl is reversible when GndHCl
is removed. 37 Refolding of the unfolded protein was initiated by
dilution in PBS. A control experiment was carried out by dilution
using the 6 M GndHCl solution. Whereas BSPOTPE is almost
nonluminescent at6MofGndHCl, its luminescence is revitalized
with ∼98% recovery after GndHCl has been removed (Figure 6A).
The refolding fraction is calculated according to eq 1 given in the
Materials and Methods section. Similarly, the CD spectral data
show that the secondary structure of HSA is ∼95% recovered after
refolding (Figure 6B).
Mechanistic Understanding. The unfolding process of HSA
induced by the GndHCl denaturant has previously been investigated
using intrinsic tryptophan fluorescence, bilirubin and
bromophenol blue absorption, and far-UV CD analyses. The
unfolding transition was proposed to follow a single-step, two-state
pathway. 38 On the other hand, the studies using extrinsic FL dyes,
such as Nile red and ANS, suggest that the protein denaturation
occurs through a two-step, three-state transition. 29b Our study in
this work, however, indicates a three-step course for the folding
structure of the protein to transform to random coil.
Mechanistic understanding of the emission behaviors of
BSPOTPE is a prerequisite to interpret the unfolding pathway of
HSA. In our previous work, we have proved that the restriction
of intramolecular rotation is the main cause for the AIE effect. 18
This mechanism is applicable to the case of BSPOTPE. In the
(37) Fanali, G.; De Sanctis, G.; Gioia, M.; Coletta, M.; Ascenzi, P.; Fasano, M.
J. Biol. Inorg. Chem. 2009, 14, 209.
(38) (a) Halim, A. A. A.; Kadir, H. A.; Tayyab, S. J. Biochem. 2008, 144, 33. (b)
Muzammil, S.; Kumar, Y.; Tayyab, S. Protein Struct. Funct. Genet. 2000,
40, 29.

Figure 6. (A) FL spectra of BSPOTPE bound with refolded HSA in the PBS buffer and unfolded HSA in the 6 M GndHCl solution. (B) CD
spectra of refolded HSA in the PBS buffer and unfolded HSA in the 6 M GndHCl solution. The spectra for the native HSA are given for comparison.
aqueous solution, the luminogen molecules are well dissolved and
dispersed as isolated molecules. The active torsional/rotational
motions annihilate their excited states, making them nonluminescent
in the solution state. Their conformations are rigidified
when they are aggregated, which effectively blocks the nonradiative
channels and activates their radiative transitions. We have
previously observed that cationic TPE derivatives bind to DNA
mainly through electrostatic interaction. 21c However, the noncanonical
change in the emission of the HSA-bound BSPOTPE with
pH suggests that the electrostatic attraction is not the main driving
force for BSPOTPE to bind with HSA (Supporting Information,
Figure S5).
To gain insights into the binding mode and to better understand
the favorable interactions between BSPOTPE and HSA, we
conducted computation modeling according to the procedures
described in the Supporting Information. The 50 conformations
of the docked BSPOTPE molecules are summarized in Figure 7.
BSPOTPE shows a high preference to the hydrophobic cavity
surrounded by subdomains IIIA, IB, and IIA. This cavity is
adjacent to W214, with 22 out of the 50 docked conformations
located in it, indicative of a high probability of occurrence of the
luminogen molecules in this structural region.
We then compared the root-mean-square-deviation (rmsd)
values between any two of the 50 docked conformations and
grouped the similar conformations into clusters with a rmsd
tolerance of 2.0 Å (Supporting Information, Figure S6). We chose
two of the biggest clusters, namely clusters 1 and 2, in the
hydrophobic cavity for the analysis of the binding sites for
BSPOTPE in HSA. Since the detailed binding modes of conformations
in each cluster are almost identical, the conformation with
the lowest estimated binding energy in each cluster is shown as
a representative to illustrate the interaction details (Figure 8).
Generally speaking, the luminogen molecule docks in the same
location and adopts a similar pose in the two clusters. The
BSPOTPE luminogen inserts into the interdomain cleft formed
by the R-helices from subdomains IIA (green helix) and IIIA (blue
Figure 7. Summary of the 50 conformations of BSPOTPE docked
on HSA, whose domains are denoted by different colors, with red,
green, and blue for domains I, II, and III, respectively. Each of the
docked conformation of BSPOTPE is represented by a sphere placed
at the average position of the coordinates of all the atoms in that
conformation. The tryptophan residue in domain II (W214) is marked
by a purple circle. BSPOTPE shows a high preference (22 out of 50)
to the hydrophobic cavity (circled by red color) adjacent to the W214
residue.
helix). The aromatic core of the luminogen stacks on the R-helix
via hydrophobic interaction, while its sulfonate peripheral arms
interact with the polar patches inside the pocket of the protein
folding structure.
In the binding conformations of cluster 1 (Figure 8A), Lys195,
Lys199, and Arg218 in HSA serve as proton donors to form
hydrogen bonds with oxygen atoms of the sulfonate group of
BSPOTPE. Lys274 on the adjacent helix forms another hydrogen
bond with the other sulfonate group of BSPOTPE. The nitrogen
atom in the side chain of Trp214 orients toward the electrondeficient
sulfonate group. On the other hand, Arg218 and Arg222
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7041

Figure 8. Details of the best binding conformations with the lowest binding free energies in clusters (A) 1 and (B) 2 (cf., Supporting Information,
Figure S6). In each case, the BSPOTPE molecule is shown in a stick representation with a semitransparent van der Waals surface and with a
color scheme of gray for carbon, blue for nitrogen, red for oxygen, and yellow for sulfur. The residues interacting with BSPOTPE are shown as
sticks (with nonbinding apolar hydrogen atoms omitted for clarity) and labeled on R carbon. Yellow dashed lines denote hydrogen bonds.
accommodate the aromatic core of BSPOTPE via cation-π
interaction. A similar binding mode is found in clusters 2 (Figure
8B). Additionally, Pro447 contacts the hydrophobic part of the
luminogen, while His440 interacts with its sulfonate group. All
the aforesaid interactions rigidify the molecular conformation of
the BSPOTPE luminogen and hamper its intramolecular rotation,
hence making it highly emissive in the binding state. The lowest
binding energies of the conformations in clusters 1 and 2 are -8.60
and -7.54 kcal/mol, corresponding to estimated inhibition constants
(K i) of 0.49 and 2.98 µM, respectively. These small Gibbs
free energies indicate that the interaction between BSPOTPE
and HSA is highly spontaneous and energetically favorable.
The detailed binding modes of the conformations from the
other clusters in the hydrophobic cavity closely resemble those
of clusters 1 and 2, although they are grouped into different
clusters. This is not unexpected because the ligand is treated as
fully flexible in the computational modeling process, and a small
rmsd tolerance is used for clustering. Thus, any difference in the
coordinates of an atom between two conformations will increase
the rmsd value, even though the atom is not important for binding
interactions.
HSA is a physiological carrier in the human body, and its
hydrophobic regions are capable of binding insoluble endogenous
compounds like fatty acids and drugs. With close scrutinization
of other proteins, such as trypsin, pepsin, and papain, one can
hardly find a hydrophobic pocket as in HSA. This may explain
why the BSPOTPE luminogen shows excellent selectivity toward
the albumin proteins over the other types of proteins.
GndHCl is known to work as a bulky ionic cosolvent. It not
only weakens hydrophobic interaction but also interferes with
association between charged solutes. 39 When HSA unfolds, a large
number of nonpolar side chains are exposed to the aqueous
medium and stabilized by GndHCl. GndHCl may also intrude into
the hydrophobic pocket of HSA and break down the spatial
architecture of the protein folds. CD spectral data indicate that
no significant alternation in the secondary structure occurs during
the first unfolding transition in the low concentration range of
GndHCl (6.0 M). No secondary
structure or strand helicity is retained in this region, as confirmed
by the CD spectral data. The luminogen molecules are fully
released from the protein, and no emission signals can be
collected. The unfolding process discussed above is summarized
in Scheme 1. Upon removal of the GndHCl denaturant by dilution,
the protein returns to its native form and the emission of
BSPOTPE is recovered, indicating that the interaction between
HSA and BSPOTPE is fully reversible.
CONCLUDING REMARKS
In summary, in this work, we developed an environmentally
stable and synthetically readily accessible FL probe for HSA
detection and quantitation. The nonluminescent BSPOTPE becomes
emissive in the presence of HSA. The AIE bioprobe shows
a linear calibration curve at [HSA] ) 0-100 nM, enabling the
protein quantitation over a wide concentration range. It enjoys a
low detection limit (down to 1 nM) and a high selectively toward
albumins. The FL bioassay is tolerant of the species in the artificial
urine and is promising for applications in real-life urinary protein
detection. Design and fabrication of simple bioassay kits for clinical
diagnostics are ongoing in our laboratories.
BSPOTPE is successfully employed as a protein staining
reagent in the PAGE analysis. It offers a simple, rapid, effective,
and economic way for visualization of protein bands in the gel
assay. Studies on the interaction between BSPOTPE and HSA

Scheme 1. Proposed Mechanism for Fluorescent Probing of GndHCl-Induced HSA Unfolding Process a
a BSPOTPE and HSA are denoted by stars and springs, respectively.
reveal the essential role of the hydrophobic cavities of the protein
folding structure. The change in the FL intensity reflects the global
stability of the protein. The unfolding process of HSA induced by
GndHCl monitored by the AIE luminogen reveals a three-step
transition with the molten-globule intermediate involved. Studies
of other biologically important processes by AIE luminogens are
in progress in our laboratories.
ACKNOWLEDGMENT
This work reported in this paper was partially supported by
the Nano Science and Technology Program at HKUST, the
Research Grants Council of Hong Kong (603008 and 602706), the
University Grants committee of Hong Kong (AoE/P-03/08), and
the National Science Foundation of China (20974028). B.Z.T.
thanks the support from Cao Guangbiao Foundation of Zhejiang
University.
SUPPORTING INFORMATION AVAILABLE
Text describing the procedures for artificial urine preparation,
cytotoxicity assay, FRET measurement, DSC analysis, and computational
modeling. Figures showing viability of the HeLa cells
incubated in the presence of BSPOTPE, spectral overlap between
BSPOTPE absorption and HSA emission, FL spectra of BSPOTPE
and DSC thermograms of HSA in the presence of GndHCl, FL
intensities of BSPOTPE at 470 nm in the presence and absence
of HSA at different pH, and summary of the docked conformations
of BSPOTPE clustered with a rmsd tolerance of 2.0 Å. This material
is available free of charge via the Internet at http://pubs.acs.org.
Received for review July 7, 2010. Accepted July 7, 2010.
AC1018028
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7043

Anal. Chem. 2010, 82, 7044–7048
Direct Fluorescence Polarization Assay for the
Detection of Glycopeptide Antibiotics
Linliang Yu, Meng Zhong, and Yinan Wei*
Department of Chemistry, University of Kentucky, Lexington, Kentucky 40506
Glycopeptide antibiotics are widely used in the treatment
of infections caused by Gram-positive bacteria. They
inhibit the biosynthesis of the bacterial cell wall through
binding to the D-alanyl-D-alanine (D-Ala-D-Ala) terminal
peptide of the peptidoglycan precursor. Taking advantage
of this highly specific interaction, we developed a direct
fluorescence polarization based method for the detection
of glycopeptide antibiotics. Briefly, we labeled the acetylated
tripeptide Ac-L-Lys-D-Ala-D-Ala-OH with a fluorophore
to create a peptide probe. Using three glycopeptide
antibiotics, vancomycin, teicoplanin, and telavancin, as
model compounds, we demonstrated that the fluorescence
polarization of the peptide probe increased upon
binding to antibiotics in a concentration dependent manner.
The dissociation constants (Kd) between the peptide
probes and the antibiotics were consistent with those
reported between free D-Ala-D-Ala and the antibiotics
in the literature. The assay is highly reproducible and
selective toward glycopeptide antibiotics. Its detection
limit and work concentration range are 0.5 µM and
0.5-4 µM for vancomycin, 0.25 µM and 0.25-2 µM
for teicoplanin, and 1 µM and 1-8 µM for telavancin.
Furthermore, we compared our assay in parallel with
a commercial fluorescence polarization immunoassay
(FPIA) kit in detecting teicoplanin spiked in human
blood samples. The accuracy and precision of the two
methods are comparable. We expect our assay to be
useful in both research and clinical laboratories.
Glycopeptide antibiotics are cyclic peptides that bind to the
peptidoglycan precursor D-alanyl-D-alanine to inhibit the biosynthesis
of bacterial cell walls. They are widely used to treat
infections caused by Gram-positive bacteria, including Methicillin
Resistant Staphylococcus aureus (MRSA), and are regarded as the
drug of “last resort”. 1 Vancomycin, teicoplanin, and telavancin are
three representative glycopeptide antibiotics that are currently in
clinical use. 2-4 Monitoring of drug levels in patients’ serum during
* To whom correspondence should be addressed. Address: 305 Chemistry-
Physics Building, University of Kentucky, Lexington, KY 40506-0055. Telephone:
(859) 257-7085. Fax: (859) 323-1069. E-mail: yinan.wei@uky.edu.
(1) Perkins, H. R. Pharmacol. Ther. 1982, 16, 181–197.
(2) Levine, D. P. Clin. Infect. Dis. 2006, 42, S5–S12.
(3) Delalla, F.; Nicolin, R.; Rinaldi, E.; Scarpellini, P.; Rigoli, R.; Manfrin, V.;
Tramarin, A. Antimicrob. Agents Chemother. 1992, 36, 2192–2196.
(4) Higgins, D. L.; Chang, R.; Debabov, D. V.; Leung, J.; Wu, T.; Krause, K. M.;
Sandvik, E.; Hubbard, J. M.; Kaniga, K.; Schmidt, D. E.; Gao, Q.; Cass,
R. T.; Karr, D. E.; Benton, B. M.; Humphrey, P. P. Antimicrob. Agents
Chemother. 2005, 49, 1127–1134.
7044 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
the treatment is critical in maximizing the effectiveness of the
treatment while minimizing drug toxicities and side effects. 5-7
Such data also provide important pharmacokinetic and pharmacodynamic
parameters for clinical studies 8-12 and, thus, guide the
optimization of therapies. 7,13,14 In addition, vancomycin has been
extensively used as a model antibiotic in tissue engineering and
controlled drug release studies. 15-20 Various assays have been
developed to measure the concentration of glycopeptide antibiotics,includingmethodsbasedonUVabsorbance,
17 immunoassay, 21-24
high performance liquid chromatography (HPLC), 25-28 and agar
diffusion bioassay. 29,30 Each of these existing methods suffer from
(5) James, C. W.; Gurk-Turner, C. Proc. (Bayl. Univ. Med. Cent.) 2001, 14,
189–190.
(6) Williams, A. H.; Gruneberg, R. N. J. Antimicrob. Chemother. 1984, 14, 441–
445.
(7) Brink, A. J.; Richards, G. A.; Cummins, R. R.; Lambson, J.; Teicoplanin,
G. U. Int. J. Antimicrob. Agents 2008, 32, 455–458.
(8) Svetitsky, S.; Leibovici, L.; Paul, M. Antimicrob. Agents Chemother. 2009,
53, 4069–4079.
(9) Wood, M. J. J. Antimicrob. Chemother. 1997, 40, 147–147.
(10) Cobo, J.; Fortun, J. J. Antimicrob. Chemother. 1996, 38, 1113–1114.
(11) Wood, M. J. J. Antimicrob. Chemother. 1996, 38, 919–919.
(12) Wood, M. J. J. Antimicrob. Chemother. 1996, 37, 209–222.
(13) Pea, F.; Brollo, L.; Viale, P.; Pavan, F.; Furlanut, M. J. Antimicrob. Chemother.
2003, 51, 971–975.
(14) Mohammedi, I.; Descloux, E.; Argaud, L.; Le Scanff, J.; Robert, D. Int. J.
Antimicrob. Agents 2006, 27, 259–262.
(15) Adams, C. S.; Antoci, V., Jr.; Harrison, G.; Patal, P.; Freeman, T. A.; Shapiro,
I. M.; Parvizi, J.; Hickok, N. J.; Radin, S.; Ducheyne, P. J. Orthop. Res. 2009,
27, 701–709.
(16) Antoci, V., Jr.; King, S. B.; Jose, B.; Parvizi, J.; Zeiger, A. R.; Wickstrom, E.;
Freeman, T. A.; Composto, R. J.; Ducheyne, P.; Shapiro, I. M.; Hickok, N. J.;
Adams, C. S. J. Orthop. Res. 2007, 25, 858–866.
(17) Radin, S.; Chen, T.; Ducheyne, P. Biomaterials 2009, 30, 850–858.
(18) Radin, S.; Ducheyne, P.; Kamplain, T.; Tan, B. H. J. Biomed. Mater. Res.
2001, 57, 313–320.
(19) Perelman, L. A.; Pacholski, C.; Li, Y. Y.; VanNieuwenhz, M. S.; Sailor, M. J.
Nanomedicine 2008, 3, 31–43.
(20) Lai, C. Y.; Trewyn, B. G.; Jeftinija, D. M.; Jeftinija, K.; Xu, S.; Jeftinija, S.;
Lin, V. S. Y. J. Am. Chem. Soc. 2003, 125, 4451–4459.
(21) Kitzis, M. D.; Goldstein, F. W. Clin. Microbiol. Infect. 2006, 12, 92–95.
(22) Wilson, J. F.; Davis, A. C.; Tobin, C. M. J. Antimicrob. Chemother. 2003,
52, 78–82.
(23) Lee, H. B.; Kwak, B. Y.; Lee, J. C.; Kim, C. J.; Shon, D. H. J. Microbiol.
Biotechnol. 2004, 14, 612–619.
(24) Lam, M. T.; Le, X. C. Analyst 2002, 127, 1633–1637.
(25) Valle, M. J. D.; Lopez, F. G.; Navarro, A. S. J. Pharm. Biomed. Anal. 2008,
48, 835–839.
(26) Abu-Shandi, K. H. Anal. Bioanal. Chem. 2009, 395, 527–532.
(27) Saito, M.; Santa, T.; Tsunoda, M.; Hamamoto, H.; Usui, N. Biomed.
Chromatogr. 2004, 18, 735–738.
(28) Shen, J.; Jiao, Z.; Zhou, Y. N.; Zhu, H. L.; Song, Z. J. Chromatographia 2007,
65, 9–12.
(29) Kureishi, A.; Jewesson, P. J.; Bartlett, K. H.; Cole, C. D.; Chow, A. W.
Antimicrob. Agents Chemother. 1990, 34, 1642–1647.
(30) Bantar, C.; Durlach, R.; Nicola, F.; Freuler, C.; Bonvehi, P.; Vazquez, R.;
Smayevsky, J. J. Antimicrob. Chemother. 1999, 43, 737–740.
10.1021/ac100543e © 2010 American Chemical Society
Published on Web 07/20/2010

one or more drawbacks including poor selectivity, requirement
of large amount of samples, and involvement of cumbersome
pretreatment processes. Fluorescence polarization immunoassay
(FPIA) is the most popular method currently used in the
quantification of glycopeptide antibiotics in clinical samples. 23,31
For example, in the FPIA assay for vancomycin, fluorescentlabeled
vancomycin forms a complex with antivancomycin antibody.
Free vancomycin in test samples competitively binds to the
antibody to displace the labeled one, decreases the fluorescence
polarization signal, and thus allows the quantification of the
analyte. 32 In this study, we developed a direct fluorescence
polarization based assay, taking advantage of the specific interaction
between glycopeptide antibiotics and their therapeutic target,
the dipeptide D-alanyl-D-alanine (D-Ala-D-Ala). Briefly, we labeled
acetylated peptide Ac-L-Lys-D-Ala-D-Ala-OH with a fluorophore and
monitored the increase of fluorescence polarization of the labeled
peptide in the presence of drugs. Our method is a class-specific
assay, which can be used to detect glycopeptide antibiotics in
general. In addition, since no protein (such as the antivancomycin
antibody) is used in the assay, we expect the shelf life and storage
stability of our assay to greatly exceed those of the immunoassay.
In this study, we have characterized the interaction between
the fluorescent peptide probes and three representative glycopeptide
antibiotics, vancomycin, teicoplanin, and telavancin. We
were able to measure the concentration of these drugs in serum
and blood samples without complicated pretreatment. In addition,
we compared the performance of our assay with a commercial
FPIA kit in determining the concentration of teicoplanin in human
blood samples. We expect our assay to be useful in both research
and clinical applications.
EXPERIMENTAL SECTION
Chemicals. The acetylated Ac-L-Lys-D-Ala-D-Ala-OH peptide,
teicoplanin, fluorescein isothiocyanate (FITC), HPLC grade acetonitrile,
and heat inactivated, sterile-filtered fetal bovine serum
(FBS) were from Sigma-Aldrich (St. Louis, MO). Vancomycin
hydrochloride was from Duchefa (Haarlem, The Netherlands).
Telavancin was from Astellas (Deerfield, IL). Alexa Fluor 680maleimide
(AF680) was from Invitrogen (Eugene, OR). Teicoplanin
FPIA kit was from LABfx (Portland, OR). Human whole blood
was from Bioreclamation (Westbury, NY). All other reagents were
of analytical grade and purchased from Sigma-Aldrich.
Fluorescent Labeling of the Peptide. The acetylated peptide
Ac-L-Lys-D-Ala-D-Ala-OH (3 mM) was incubated with fluorescein
isothiocyanate (FITC, 4.5 mM) in a phosphate buffer (PBS) (10
mM Na2HPO4,2mMKH2PO4, 2.7 mM KCl, 137 mM NaCl, pH
7.4) at room temperature overnight. The reaction was terminated
through the addition of a Tris-Cl buffer (5 mM, pH 7.4)
followed by an incubated of2hatroom temperature. The FITClabeled
peptide was purified using a reverse-phase HPLC
(Waters, Milford, MA) on a C18 column (Waters, Milford, MA).
The purified FITC-labeled peptide was dried under vacuum and
then resuspended in water. The concentration of the peptide
was determined based on the absorption at 515 nm. Labeling
with AF680 was performed similarly. The peptide probe
concentration was determined by absorption at 680 nm. The
molecule weight of the labeled peptides was confirmed by mass
spectroscopy analysis.
Fluorescence Polarization Measurements. Fluorescence
polarization measurements were performed using a Perkin-Elmer
LS-55 fluorescence spectrometer (Perkin-Elmer, Waltham, MA).
The temperature was kept at 20 °C. Fluorophore-conjugated
peptide (FITC-peptide or AF680-peptide) was added into 400 µL
of PBS or FBS to a final concentration of 1 µM. The excitation
and emission wavelengths were 479 and 515 nm for FITC-peptide
and 679 and 702 nm for AF680-peptide.
Titration and Data Fitting. For titration experiments, FITCpeptide
or AF680-peptide was added into 400 µL PBS or FBS to
a final concentration of 1 µM, and then, small aliquots of
glycopeptide antibiotics were titrated into the samples. The
increases of fluorescence polarization upon the formation of
drug-peptide complexes were recorded.
The titration curve was fitted using a one-to-one binding model
according to the following equation. 33
∆P ) ∆Pmax( ([F] + x + Kd ) - √([F] + x + Kd ) 2 - 4[F]x)
2[F]
(1)
Where ∆P is the monitored change of fluorescence polarization
at each drug concentration, ∆Pmax is the maximum change of
the fluorescence polarization upon saturation, [F] is the peptide
probe concentration, Kd is the dissociation constant between
the peptide probe and the drug, and x is the drug concentration.
The titration data were fitted using Origin (Northampton, MA).
Binding Selectivity Assay. For each antibiotic, AF680-peptide
was added into 400 µL of FBS to a final concentration of 1 µM.
Next, antibiotics were added to a final concentration of 8 µM. The
change of fluorescent polarization upon the addition of antibiotics
was recorded.
Determination of Drug Concentration in Test Samples.
For tests in serum, AF680-peptide was added to 400 µL of FBS to
a final concentration of 1 µM. Then, known quantities of antibiotics
were added into the sample. The concentrations of the drugs were
determined based on the calibration curves derived in FBS. For
tests in blood, three human blood samples were spiked with
teicoplanin in the therapeutic range by a person not involved in
this project. The exact values were kept blind to the authors during
the analysis. The blood samples were centrifuged at 2000g for 10
min. The supernatant (plasma) was collected and analyzed
together with a series of teicoplanin standards in FBS as described
above. For each plasma sample, we prepared two sets of dilutions
so the final concentration of teicoplanin would be in the working
concentration range of the assay. The first set of three test samples
contain 10 µL of plasma and 390 µL of FBS with 1 µM of AF680peptide.
The second set of three test samples contain 20 µL of
plasma and 380 µL of FBS with 1 µM of AF680-peptide. The
teicoplanin concentration in each set of samples was determined.
The original teicoplanin concentration calculated from the set of
samples falls in the working concentration range. The same
(31) Urakami, T.; Maiguma, T.; Kaji, H.; Kondo, S.; Teshima, D. J. Clin. Pharm.
Ther. 2008, 33, 357–363.
(32) Adamczyk, M.; Grote, J.; Moore, J. A.; Rege, S. D.; Yu, Z. G. Bioconjugate
Chem. 1999, 10, 176–185. (33) Yu, L. L.; Fang, J.; Wei, Y. N. Biochemistry 2009, 48, 2099–2108.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7045

plasma samples were analyzed using the teicoplanin FPIA kit
following the instruction of the manufacturer.
RESULT AND DISCUSSION
Design and Synthesis of Peptide Probes. We used aminereactive
fluorophores to label the acetylated peptide Ac-L-Lys-D-
Ala-D-Ala-OH. We chose this peptide for two reasons. First, L-lys
is naturally occurring in the peptidoglycan precursor peptide L-Ala-
D-Glu-L-Lys-D-Ala-D-Ala. We expected the incorporation of the L-lys
would not affect the interaction between the peptide and the
antibiotics. Second, the side chain of Lys contains a primary amine
group, which is distant from the binding site judged by the crystal
structure of the vancomycin and diacetyl-L-Lys-D-Ala-D-Ala complex.
34 As expected, we found that the binding affinities between
the labeled peptide and antibiotics were similar to the values
reported for the binding between the free peptide and antibiotics,
indicating that the modification did not affect their interaction (see
below).
Effect of Fluorophores. The fluorescence polarization signal
is determined by the Perrin Equation. 35
( 1 1
-
P 3) ) ( 1
-
P0 1 kT
1 + 3)( ηV τ)
Where τ is the fluorescence lifetime of the fluorophore, P0 is the
limiting polarization, k is the Boltzmann constant, T is the
absolute temperature, η is the viscosity, and V is the molecular
volume (molecular weight). The fluorescence polarization is
negatively correlated with the lifetime of the fluorophores and
positively correlated with the molecular volume (molecular
weight). In order to maximize the observed change of polarization
upon antibiotic binding, we tested two fluorophores with
different lifetimes, FITC (4 ns lifetime) and AF680 (1 ns
lifetime) (Figure 1). We found that AF680-peptide yielded a larger
change of signal upon drug binding and, thus, exhibited a better
signal-to-noise ratio.
Titration Curves of Glycopeptide Antibiotics in Phosphate
Buffer. The binding between the antibiotics and peptide probes
was monitored following the increase of fluorescence polarization
of the samples upon addition of increasing concentrations of the
antibiotics (Figure 1). The dissociation constants (Kd) between
the drugs and peptide probes were estimated to be around 1.1,
0.2, and 1.8 µM, respectively, for vancomycin, teicoplanin, and
telavancin (Table 1). The identity of the fluorophores did not have
a drastic effect on the binding affinity, suggesting they were not
directly involved in the interaction. The determined affinities are
similar to the values reported for the bindings between these
antibiotics and diacetyl-L-Lys-D-Ala-D-Ala (∼1 µM for vancomycin,
and ∼0.6 µM for teicoplanin). 36-39 Compared with the FITC-
(34) Nitanai, Y.; Kikuchi, T.; Kakoi, K.; Hanmaki, S.; Fujisawa, I.; Aoki, K. J.
Mol. Biol. 2009, 385, 1422–1432.
(35) Ye, B. C.; Ikebukuro, K.; Karube, I. Nucleic Acids Res. 1998, 26, 3614–
3615.
(36) Popieniek, P. H.; Pratt, R. F. Anal. Biochem. 1987, 165, 108–113.
(37) Azad, M.; Hernandez, L.; Plazas, A.; Rudolph, M.; Gomez, F. A. Chromatographia
2003, 57, 339–343.
(38) Rao, J. H.; Yan, L.; Xu, B.; Whitesides, G. M. J. Am. Chem. Soc. 1999, 121,
2629–2630.
(39) van Wageningen, A. M. A.; Staroske, T.; Williams, D. H. Chem. Commun.
1998, 1171–1172.
7046 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
(2)
Figure 1. Calibration curves of FITC-peptide (diamonds) and AF680peptide
(squares) with vancomycin (A), teicoplanin (B), and telavancin
(C) in PBS. The concentration of the peptide probes were 1 µM. The
concentrations of the glycopeptide antibiotics increased from 0 to 7.7
µM during the titration. The data were fitted using a one-to-one binding
model to derive the dissociation constants listed in Table 1.
Table 1. Fitting Parameters of Peptide-Drug
Interaction
glycopeptide
antibiotics probe matrix
dissociation
constant Kd (µM)
vancomycin FITC-peptide PBS 0.91 ± 0.06
AF680-peptide PBS 1.07 ± 0.19
AF680-peptide FBS 1.81 ± 0.13
teicoplanin FITC-peptide PBS 0.14 ± 0.07
AF680-peptide PBS 0.22 ± 0.02
AF680-peptide FBS 0.87 ± 0.08
telavancin FITC-peptide PBS 3.22 ± 0.38
AF680-peptide PBS 1.81 ± 0.36
AF680-peptide FBS 5.36 ± 0.35
peptide probe, the AF680-peptide yielded a better signal-to-noise
ratio. Therefore, it was used in fluorescence polarization assays
for the rest of the study.
To directly examine if the free fluorophores (FITC or AF680)
interacted with the glycopeptide antibiotics, we compared the
change of fluorescence polarization of 1 µM FITC, AF680, FITCpeptide,
or AF680-peptide upon addition of 8 µM vancomycin,
teicoplanin, or telavancin (Figure 2). No significant change of
fluorescence polarization was observed for samples containing free
fluorophores, indicating the lack of interaction between the drugs
and the free fluorophores.
Drug Detection in Serum. To further examine the usefulness
of the current method in the detection of glycopeptide antibiotics
in clinical samples, we measured the titration curves using AF680peptide
for the detection of vancomycin, teicoplanin, and telavancin
in FBS. The potential matrix effects from serum could affect the
assay in two ways. First, the higher viscosity in serum slows down
the tumbling of the fluorophores and, thus, increases fluorescence
polarization. Second, certain species in the serum may interact
with the peptide probe or the drugs to prevent accurate detection.

Figure 2. Free fluorophores do not interact with the antibiotics. The
fluorescence polarization did not change when vancomycin (black),
teicoplanin (white), or telavancin (gray) was added into PBS containing
the free fluorophores (FITC or AF680). FITC and AF680 were
first incubated with a Tris-Cl buffer (5 mM, pH 7.4) to block the amine
reactive groups. The concentration of fluorophores or peptide probes
was 1 µM. The concentration of the drugs was 8 µM.
Figure 3. (A) Titration curves for vancomycin (squares), teicoplanin
(triangles), and telavancin (diamonds) in FBS. AF680-peptide (1 µM)
was used in the assays. (B) Intraday and interday variance of the
assay. Three calibration curves of teicoplanin were collected at three
different times during a day: 9:00 a.m. (filled squares), 1:00 pm (open
diamonds), and 5:00 pm (open circles). A fourth curve was obtained
at 9:00 a.m. of the following day (filled triangle). The same teicoplanin
calibration curve as shown in Figure 3A was also plotted (gray) to
illustrate the long-term reproducibility of the assay.
The fluorescence polarization signals measured in serum increased
both for the free peptide probes and the drug-peptide
complex (Supporting Information). The increase was more dramatic
for the complex, which resulted in an increase in the ∆P
values compared to the ∆P values measured in PBS (Figure 3A).
The binding affinities decreased slightly compared to the affinities
measured in PBS (Table 1). We speculated this change of affinity
was due to the interaction between certain components in the
serum with the peptide probes and/or glycopeptides antibiotics,
which weakened their affinities for each other through a competition
mechanism.
Furthermore, we examined the interday and intraday variances
of the method by collecting the calibration curves of teicoplanin
in serum at three different times of one day and again in the
following day (Figure 3B). The curves superimposed well onto
each other, indicating the assay was reproducible. At the same
teicoplanin concentration, the relative variance among all four
curves was less than 5%. In addition, we plotted the teicoplanin
calibration curve from Figure 3A in Figure 3B again to illustrate
the good reproducibility of the method over a period of 6 months,
Figure 4. Selectivity of the assay. Different antibiotics (8 µM) were
added into solutions containing AF680-peptide (1 µM). Antibiotics
tested were ampicillin (1), gentamicin (2), tetracycline (3), nalidixic
acid (4), chloramphenicol(5), vancomycin (6), telavancin (7), and
teicoplanin (8). Error bars indicate the standard deviations from three
independent measurements.
as the trace in Figure 3A was collected approximately 6 months
ago. As all binding based assays, the detection limit and working
concentration range are determined by the binding affinity
between the probe and the analyte, as well as the signal-to-noise
ratios of the measurements. The detection limit and working
concentration range of the current method were determined to
be 0.5 µM and 0.5-4 µM for vancomycin, 0.25 µM, 0.25-2 µM
for teicoplanin, and 1 µM and 1-8 µM for telavancin.
Selectivity Study. In many situations, several antibiotics may
be used together as a cocktail in research or clinical applications.
It is desirable to have a method that can selectively detect a
specific one or specific class of antibiotics. The interaction between
the peptide probe and glycopeptide antibiotics is highly specific.
We examined the binding selectivity of the AF680-peptide probe
toward different antibiotics in FBS (Figure 4). The assay was
highly selective toward the three glycopeptides. No significant
change of fluorescence polarization was observed for other
antibiotics tested.
Recovery Test in Human Blood. Finally, we compared the
performance of the current assay with a commercial FPIA kit in
the quantification of teicoplanin in human blood. Human blood
samples were spiked with different concentrations of teicoplanin
based on the drug’s therapeutic concentration window. Measurement
was conducted blindly using the current assay side by side
with a commercial FPIA teicoplanin detection kit. For the current
assay, each spiked blood sample was centrifuged to remove the
blood cells. The supernatant (plasma) was collected and diluted
to the working concentration range. Since the therapeutic window
of teicoplanin is approximately 10-30 µg/mL, 40-42 while our
working concentration range is between 0.25 and 2 µM (corresponding
to 0.47-3.8 µg/mL), we diluted each plasma sample by
FBS to two scales: 20-fold and 40-fold. Thus, at least one of the
diluted final concentrations would fall in the working concentration
range. On the basis of the calibration curve established in FBS,
the antibiotic concentrations in the diluted serum solution were
determined, which were then multiplied by the fold of dilution to
(40) Soy, D.; Lopez, E.; Ribas, J. Ther. Drug Monit. 2006, 28, 737–743.
(41) Lortholary, O.; Tod, M.; Rizzo, N. Antimicrob. Agents Chemother. 1996,
40, 1242–1247.
(42) Lamont, E.; Seaton, R. A.; Macpherson, M.; Semple, L.; Bell, E.; Thomson,
A. H. J. Antimicrob. Chemother. 2009, 64, 181–187.
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7047

Table 2. Detection of Teicoplanin in Human Blood
Using the Current Assay or a Commercial FPIA Kit
spiked
(µg/mL) this method recovery (%) FPIA recovery (%)
9.5 9.2 ± 0.2 97 9.8 ± 0.6 103
19 19.6 ± 0.4 103 18.5 ± 0.6 97
38 39.0 ± 1.0 103 37.8 ± 4.4 99
obtain the original drug concentration. Meanwhile, each plasma
sample was also analyzed using a commercial FPIA kit. The spiked
and determined concentrations were summarized in Table 2.
When human blood samples from different donors were used as
the background matrix for spiking, the variation between the
determined drug concentrations was similar to the deviation
among multiple tests performed using blood samples from the
same donor. The accuracy and precision of the current assay is
comparable to the commercial kit.
In conclusion, we developed a simple analytical assay for the
detection and quantification of selected glycopeptides antibiotics,
including vancomycin, teicoplanin, and telavancin, in various
samples. The method is highly selective toward glycopeptide
7048 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
antibiotics and does not need complicated pretreatment for serum
and blood samples. In addition, no antibody is used in our method;
therefore, we expect the shelf life and storage stability of our assay
to greatly exceed those of the commercial FPIA kits. We have
stored the AF680-peptide probe at room temperature in the dark
for 9 months. No significant loss of signal intensity or binding
affinity to antibiotics could be detected.
ACKNOWLEDGMENT
This work was supported by a RCTF fellowship and the faculty
startup fund from University of Kentucky. The authors thank Mr.
Raymond Miracle from LABfx LLC for suggestions on the
teicoplanin analysis using the FPIA kit.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review February 28, 2010. Accepted July 7,
2010.
AC100543E

Anal. Chem. 2010, 82, 7049–7052
Development of a High Sensitivity Rapid Sandwich
ELISA Procedure and Its Comparison with the
Conventional Approach
Chandra Kumar Dixit, †,‡ Sandeep Kumar Vashist, †,|,# Feidhlim T. O’Neill, † Brian O’Reilly, †
Brian D. MacCraith, †,§ and Richard O’Kennedy* ,†,‡,§
Centre for Bioanalytical Sciences (CBAS), National Centre for Sensor Research, Applied Biochemistry Group, School
of Biotechnology, and Biomedical Diagnostics Institute (BDI), Dublin City University, Dublin 9, Ireland, and
Bristol-Myers Squibb (BMS), Swords Laboratories, Watery Lane, Swords, Co. Dublin, Ireland
A highly sensitive and rapid sandwich enzyme-linked
immunosorbent assay (ELISA) procedure was developed
for the detection of human fetuin A/AHSG (r2-HSglycoprotein),
a specific biomarker for hepatocellular
carcinoma and atherosclerosis. Anti-human fetuin A
antibody was immobilized on aminopropyltriethoxysilanemediated
amine-functionalized microtiter plates using
1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride
and N-hydroxysulfosuccinimide-based heterobifunctional
cross-linking. The analytical sensitivity of the
developed assay was 39 pg/mL, compared to 625 pg/
mL for the conventional assay. The generic nature of the
developed procedure was demonstrated by performing
human fetuin A assays on different polymeric matrixes,
i.e., polystyrene, poly(methyl methacrylate), and polycyclo-olefin
(Zeonex), in a modified microtiter plate format.
Thus, the newly developed procedure has considerable
advantages over the existing method.
Conventional enzyme-linked immunosorbent assay (ELISA)
procedures have been followed for decades for the detection of
analytes of importance in industrial, healthcare, and academic
research. However, improvement on existing ELISA technologies
are continuously attempted by many groups. 1,2 We report the
development of a high-sensitivity ELISA-based assay for human
fetuin A (HFA), with lower detection limits and a higher sensitivity
than commercially available assays. The biological importance of
HFA, a member of the cystatin superfamily, which is commonly
present in the cortical plate of the immature cerebral cortex and
hemopoietic matrix of bone marrow, is discussed elsewhere. 3-10
There are many commercially available ELISA kits for fetuin A,
* To whom correspondence should be addressed. Email:
richard.okennedy@dcu.ie. Tel.: +353 1 700 7810. Fax: +353 1 700 5412.
† Centre for Bioanalytical Sciences (CBAS), National Centre for Sensor
Research, Dublin City University.
‡ Applied Biochemistry Group, School of Biotechnology, Dublin City University.
§ Biomedical Diagnostics Institute (BDI), Dublin City University.
| Bristol-Myers Squibb (BMS).
# Current Address: Nanoscience and Nanotechnology Initiative (NUSNNI),
National University of Singapore, Engineering Drive 1, Singapore 117576.
(1) Kaur, J.; Boro, R. C.; Wangoo, N.; Singh, R. K.; Suri, C. R. Anal. Chim.
Acta 2008, 607 (1), 92–99.
(2) Jia, C. P.; Zhong, H. Q.; Liu, M. Y.; Jing, F. X.; Yao, S. H.; Xiang, J. Q.; Jin,
Q. H.; Zhao, J. L. Biosens. Bioelectron. 2009, 24 (9), 2836–2841.
which are listed in Supplementary Table 1, Supporting Information.
The assay was demonstrated on different commercially
relevant solid supports. The developed ELISA is better than the
conventional procedure in terms of greatly reduced overall assay
duration, higher sensitivity, and greater reproducibility.
HFA was taken as the model assay system to demonstrate the
utility of our developed ELISA procedure since all the assay
components were commercially available in kit form. This enabled
us to do robust and highly precise comparison of the developed
ELISA with the commercially existing conventional ELISA procedures,
as the same assay components were used under the same
conditions. The developed procedure can be employed on any
commercially relevant substrate. Therefore, this approach of
immobilizing antibody on chemically modified solid supports
(Figure 1) has potential applications in many other assays and
formats.
EXPERIMENTAL SECTION
Plate Preparation, Amine-Functionalization with Silane,
and Cross-Linking Carboxyl Groups of Anti-HFA Antibody
to the Amino Groups of the Surface of Microtiter Plate Wells.
Each well of the 96-well plate was treated with 100 µL of absolute
ethanol for 5 min at 37 °C and washed five times with 300 µL of
deionized water (DIW). Subsequently, each well was treated with
(3) Kalabay, L.; Gráf, L.; Vörös, K.; Jakab, L.; Benk|Ado˜, Z.; Telegdy, L.; Fekete,
B.; Rohászka, Z.; Füst, G. BMC Gastroenterol. 2007, 7, 15.
(4) Srinivas, P. R.; Wagner, A. S.; Reddy, L. V.; Deutsch, D. D.; Leon, M. A.;
Goustin, A. S.; Grunberger, G. Mol. Endocrinol. 1993, 7, 1445–1455.
(5) Mathews, S. T.; Srinivas, P. R.; Leon, M. A.; Grunberger, G. Life Sci. 1997,
61 (16), l383–1392.
(6) Kalabay, L.; Prohászka, Z.; Füst, G.; Benkõ, Z.; Telegdy, L.; Szalay, F.; Tóth,
K.; Gráf, L.; Jakab, L.; Pozsonyi, T.; Arnaud, P.; Fekete, B.; Karádi, I. In
Liver Cirrhosis: New Research; Chen, T. M., Ed.; Nova Science: New York,
2005; pp 63-75.
(7) Reynolds, J. L.; Skepper, J. N.; McNair, R.; Kasama, T.; Gupta, K.;
Weissberg, P. L.; Dechent, W. J.; Shanahan, C. M. J. Am. Soc. Nephrol.
2005, 16, 2920–2930.
(8) Jethwaney, D.; Lepore, T.; Hassan, S.; Mello, K.; Rangarajan, R.; Dechent,
W. J.; Wirth, D.; Sultan, A. A. Infec. Immun. 2005, 73 (9), 5883–5891.
(9) Ix, J. H.; Shlipak, M. J.; Brandenburg, V. M.; Ali, S.; Ketteler, M.; Whooley,
M. A. Circulation 2006, 113, 1760–1767.
(10) Hermans, M. M. H.; Brandenburg, V.; Ketteler, M.; Kooman, J. P.; Sande,
F. M. V. D.; Boeschoten, E. W.; Leunissen, K. M. L.; Krediet, R. T.; Dekker,
R. T. Kidney Int. 2007, 72, 202–207.
(11) Park, S. J.; Jung, W. Y. J. Colloid Interface Sci. 2002, 250, 93–98.
(12) Cass, T.; Ligler, F. S. Immobilized biomolecules in analysis: a practical
approach; Oxford University Press Inc.: New York, 1998.
10.1021/ac101339q © 2010 American Chemical Society 7049
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
Published on Web 07/20/2010

Figure 1. Schematic of the developed sandwich ELISA procedure for the HFA/AHSG assay. Plate functionalization was performed by treatment
with 1% (w/v) KOH and then silanization with 2% (v/v) APTES. KOH oxidizes the organic moiety of the polymer generating a carbonyl (-CO)
or a hydroxyl (-OH) group according to the nature of the polymer. The alkoxy groups of APTES subsequently react with the carbonyl or hydroxyl
groups of the surface in a hydrolysis-dependent reaction. 11,12 Anti-HFA/AHSG antibody was then captured using EDC and sulfo-NHS chemistry
on the aminated 96-well plate, where the EDC-SNHS was used in a ratio of 1:100 with the antibody. The EDC-SNHS-activated antibody molecules
were then immobilized on the aminated ELISA plate, where a bond was formed between amine groups on the plate and the activated carboxyl
groups of the antibody.
100 µL of 1.0% (w/v) KOH at 37 °C for 10 min followed by five
washings with 300 µL of DIW. The KOH-treated wells were then
functionalized with amino groups by incubation with 100 µL of
2% (v/v) aminopropyltriethoxysilane (APTES) per well at 80 °C
for 1 h inside a vacuum-desiccator, in order to achieve maximum
silanization. The reaction mechanism following the KOH activation
and APTES-based surface functionalization involves mild oxidation
followed by a hydrolytic APTES polymerization on the oxidized
surface through its alkoxy groups. 11,12 The desiccator was
equilibrated to room temperature for 20 min. The amine-functionalized
plate was subsequently washed five times with 300 µL
of DIW in order to remove excess unbound 3-APTES from the
surface. Afterward, the anti-HFA/AHSG (R2-HS-glycoprotein; 990
µL of4µg/mL) was incubated with 10 µL of a premixed solution
of 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride
(EDC) (4 mg/mL) and N-hydroxysulfosuccinimide (SNHS) (11
mg/mL) for 15 min at 37 °C. The resulting EDC cross-linked anti-
HFA/AHSG antibody solution was added to each of the functionalized
wells (100 µL) and incubated for 1hat37°C. The anti-
HFA/AHSG-coated wells were then washed five times with 300
µL of PBS. The methodology pertaining to the conventional assay
is described in detail in the Supporting Information methods
section.
ELISA on Anti-HFA/AHSG Antibody-Immobilized Microtiter
Plates. Plates with covalent and passively immobilized anti-
HFA antibody were blocked with 1% (v/v) BSA diluted in 0.1 M
7050 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
phosphate buffered saline (PBS), pH 7.4, for 30 min at 37 °C and
subsequently washed five times with 300 µL of PBS. Varying
concentrations of HFA/AHSG (from 4.8 pg/mL to 20 ng/mL)
were prepared in 0.1 M PBS, pH 7.4, and 100 µL of each of these
concentrations were incubated in the antibody-coated plates for
1hat37°C and, subsequently washed five times with PBS. One
hundred microliters of biotinylated anti-HFA/AHSG detection
antibody (200 ng/mL) was added and incubated for 1hat37°C
followed by five PBS washes. HRP-conjugated streptavidin (100
µL per well), at a dilution of 1:200, was added and then incubated
for 20 min at 37 °C followed by five washes with PBS. The 3,3′,5,5′tetramethylbenzidine
(TMB) substrate was subsequently added
(as per the manufacturer’s recommendations), and the reaction
was stopped after 20 min by addition of 50 µL of1NH 2SO4. The
absorbance was recorded at a primary wavelength of 450 nm
with a reference wavelength of 540 nm. The dual wavelength
system of absorbance measurement in 96-well plates is designed
to eliminate or greatly reduce the optical imperfections
at the well-to-well level.
All the experiments were carried out in triplicate. The control
for this study was 0 ng/mL HFA in 0.1 M PBS, pH 7.4, and the
absorbance of the control was subtracted from all the assay values.
The respective analytical sensitivity (detection limit) of the assay
was determined where analytical sensitivity was calculated using
the formula [mean absorbance of blank +3σ (standard deviation
of the blank)]. Additionally, data sets obtained from the modified

Figure 2. Comparative analysis of developed sandwich ELISA (black
circles) with the conventional format. Anti-human fetuin A (anti-HFA)
antibody was covalently bound to the amine-modified surface in the
developed ELISA, whereas antibody adsorbed on the unmodified
surface was used in the conventional assay format. The conventional
assay was performed with the manufacturer-recommended protocol
(white circles) and the protocol developed in this study (inverted
triangle). The error bars represents standard deviations.
Table 1. Comparison of Assay Performance
Parameters (Precision and Sensitivity) for
Conventional and Developed ELISAs a
intraday precision
range (CV %),
(n ) 5)
interday precision
range (CV %),
(n ) 3)
sensitivity
(pg/mL)
developed ELISA 2.4-10.4 1.7-17.6 39
conventional ELISA 4.7-17.4 3.6-20.0 624
a Intraday precision was calculated from the five repeats (n ) 5) of
the same assay on a single day but at different times, while interday
precision was calculated from assay repeats on three different days (n
) 3). All the assays were carried out in triplicate. Analytical sensitivity
was calculated using the formula [average absorbance of the blank +
3(SDblank)].
and conventional ELISA were analyzed using standard curve
analysis of Sigma Plot software, version 11.0.
RESULTS AND DISCUSSION
The range of detection of HFA/AHSG for the developed ELISA
(Figure 2) was 9-20 pg/mL (r 2 ) 0.99). An analytical sensitivity
of 39 pg/mL was recorded for developed ELISA, where
analytical sensitivity was calculated using the formula [average
absorbance of the blank + 3(SDblank)]. Assay variability
parameters for this high-sensitivity assay for HFA are described
in Table 1, where intraday assay variability was calculated from
5 repeats performed on a single day, and interday assay variability
was calculated from three repeats performed on three different
days in triplicate. The percentage recovery for lower concentrations
(4.88-625 pg/mL) was 50-80% while for higher concentrations
(1.2-20 ng/mL) it was 75-100%. The percentage recovery
was calculated as the obtained value/expected value × 100.
The half-effective concentration (EC50) obtained from the
dose-response curve, 13 which is a measure of analyte-ligand
(13) Armbruster, D. A.; Schwarzhoff, R. H.; Hubster, E. C.; Liserio, M. K. Clin.
Chem. 1993, 39, 2137–2146.
interaction, was found to be on an average of 3.3 ng/mL. The
EC50 for all the assay repeats was found to be in the range of
3-3.5 ng/mL. A lower EC50 (3.3 ng/mL), which was obtained
for the developed ELISA, suggests a strong interaction between
HFA and anti-HFA antibody. This interaction behavior is
attributed to the significant increase in the total amount of
immobilized antibody achieved using the covalent immobilization
strategy, which increased the availability of anti-HFA
antibody per HFA molecule, thus increasing the resultant HFA
capture on the well surface.
Conversely, EC50 for the conventional ELISA format was 23
ng/mL, (r 2 ) 0.99), which was significantly higher than the
developed ELISA. This suggests a less sensitive conventional
assay (Supplementary Table 2, Supporting Information). The
most important factor that may be attributed to the enhanced
sensitivity of the modified assay is the covalent immobilization of
the anti-HFA/AHSG capture antibody since the covalently crosslinked
antibodies should not leach out during the assay procedure
in comparison to the use of passively adsorbed antibodies, where
leaching may occur more easily.
Performance of the developed sandwich ELISA procedure was
compared with the conventional protocols (Figure 2), both
involving passively adsorbed antibody on normal unmodified
microtiter plates (one performed using the modified protocol and
the other as per the manufacturer’s recommendations). All the
assays were performed simultaneously on the same day under
same set of conditions in order to minimize variability, where
variability was reported as percentage coefficient of variation (%
CV) (Table 1).
The technique employed here for chemically modifying the
polymeric surface was developed and standardized by our group,
and a standard EDC-based cross-linking strategy was deployed
to immobilize anti-HFA/AHSG antibody. The developed sandwich
ELISA procedure decreased the overall assay duration from 20
to 6 h, which is more than a 3-fold decrease. Thus, the reported
procedure is a rapid ELISA format having the additional benefit
of being generic in nature, i.e., it can be used with sandwich
ELISAs on different polymeric matrixes for a range of different
analytes. Hence, using the surface modification technique reported
here, it is possible to develop rapid and high sensitivity assays
on various categories of solid supports including those that are
chemically inert.
To the best of our knowledge, this developed ELISA procedure
is the most sensitive assay reported for the detection of HFA/
AHSG (Supplementary Table 1, Supporting Information), which
is based on the comparison between different commercially
available ELISA kits for HFA. The sensitivity of the assay
developed by our group may be attributed to the use of monoclonal
antibody at the capture stage, which was covalently
immobilized on the surface, and a polyclonal antibody to detect
the antigen. The covalently cross-linked monoclonal antibodies
that are used as capture antibodies in this study were found to
capture significantly higher amounts of HFA, which has improved
the detection limit and the overall assay sensitivity of the
developed ELISA procedure in comparison to the conventional
format. However, the commercially available ELISA kits, as
described in Supplementary Table 1 (Supporting Information),
have different capture and detection antibody combinations, which
Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
7051

Figure 3. Performance of the developed ELISA on different solid
support matrixes. Three categories of the solid support were chosen
on the basis of their chemical properties, namely polystyrene (PS)
which is hydrophobic (white circles); polymethyl methacrylate (PMMA),
a hydrophilic polymer (black circles); and polycyclo-olefin polymer,
trade name Zeonex, which is a neutral matrix (inverted black
triangles). The error bars represent standard deviations.
may be a factor in their reduced sensitivities. 14 The sensitivity
and specificity of any immunoassay is dependent on the antigen
capture efficiency of an antibody which is subsequently governed
by the nature of antibody 15,16 and accounts for the different assay
sensitivities of the commercially available kits employing different
type of antibodies that are monoclonal or polyclonal at the capture
stage in ELISA.
Three different types of commercially relevant substrates,
polystyrene (PS) (hydrophobic), polymethyl methacrylate (PMMA)
(hydrophilic), and cyclo-olefin polymer (Zeonex) (inert), were
used in the modified microtiter plate format. The HFA assay was
performed on these matrixes in the modified microtiter plate
format. The HFA assays performed on different solid substrates
could not be directly compared because of their different thick-
(14) Wild, D. The Immunoassay Handbook, 3rd ed.; Elsevier: Oxford, 2005.
(15) Lacy, A.; Dunne, L.; Fitzpatrick, B.; Daly, S.; Keating, G.; Baxter, A.; Hearty,
S.; O’Kennedy, R. J. AOAC Int. 2006, 89 (3), 884–892.
(16) Byrne, B.; Stack, E.; Gilmartin, N.; O’Kennedy, R. Sensors 2009, 9, 4407–
4445.
7052 Analytical Chemistry, Vol. 82, No. 16, August 15, 2010
nesses. Theoretically, the optical path length contributes toward
the absorbance (A). Therefore, a thick solid support increases
the path length and, hence, significantly changes the absorbance.
However, the generic nature of the developed ELISA procedure
was successfully demonstrated (Figure 3). The HFA assay
performed on the Zeonex support, which is an inert cyclic poly
olefin derivative, had an analytical sensitivity of 19.5 pg/mL, while
for the assay performed on PS and PMMA, it was 78 pg/mL. The
analytical sensitivities for the HFA assay and the dose-response,
which determines the slope-dependent linearity range of the assay,
of HFA-anti-HFA for all the three solid supports used in this study
suggests that this modified protocol increases the overall assay
sensitivity and is not confined to the nature of the solid support
used for capturing antibody.
CONCLUSIONS
A rapid sandwich ELISA procedure was developed for the
highly sensitive detection of HFA/AHSG. It was based on the
covalent immobilization of anti-HFA capture antibody on a
3-APTES-functionalized microtiter plate. The developed ELISA has
comparatively better analytical sensitivity (39 pg/mL) and less
variability in interday and intraday assay repeats than the
conventional format, which has an analytical sensitivity of 624 pg/
mL. The surface chemistry developed for this study is generic
and can be employed to activate polymer matrixes irrespective of
their chemical nature as demonstrated on PS (hydrophobic),
PMMA (hydrophilic), and polycyclo-olefin (inert). Therefore, this
strategy can also be used for rapid assay development on various
commercially relevant substrates and the screening of substrates
for particular biosensor/diagnostic applications.
ACKNOWLEDGMENT
C.K.D. and S.K.V. contributed equally to this work. We
acknowledge Bristol Myers Squibb (BMS), Syracuse, USA, and
Industrial Development Agency, Ireland, for the financial support
under the CBAS Project Code 116294.
SUPPORTING INFORMATION AVAILABLE
Additional information as noted in text. This material is
available free of charge via the Internet at http://pubs.acs.org.
Received for review May 21, 2010. Accepted July 7, 2010.
AC101339Q