Ontology engineering

volume 28 number 2 february 2010COMMENTARY131 case study: Never againChristopher Scott133 Going to ridiculous lengths—European coexistence regulations for GM cropsKoreen Ramessar, Teresa Capell, Richard M Twyman & Paul ChristoufeatureRegulations to prevent outcrossing ofGM pollen in the EU, p 133patents137 Changing the rules of the game: addressing the conflict between free access toscientific discovery and intellectual property rightsMiriam Bentwich141 Recent patent applications in antibody fragments© 2010 Nature America, Inc. All rights reserved.NEWS AND VIEWS142 ChIPs and regulatory bitsXin He & Saurabh Sinha144 From genomics to crop breedingRichard Flavell145 Spilling the beans on legume biologyPeter Hare146 Systematic tracking of cell fate changesJonghwan Kim & Stuart H Orkin148 Research highlightscomputational biologyMining the maize genome, p 144Resource149 Rational association of genes with traits using a genome-scale gene network forArabidopsis thalianaInsuk Lee, Bindu Ambaru, Pranjali Thakkar, Edward M Marcotte & Seung Y RheeresearchBRIEF COMMUNICATIONS157 Enhanced antibody half-life improves in vivo activityJ Zalevsky, A K Chamberlain, H M Horton, S Karki, I W L Leung, T J Sproule,G A Lazar, D C Roopenian & J R DesjarlaisComputational discovery of plant genefunction, p 149nature biotechnologyiii

volume 28 number 2 february 2010letters161 Expansion and maintenance of human embryonic stem cell–derived endothelialcells by TGFb inhibition is Id1 dependentD James, H-s Nam, M Seandel, D Nolan, T Janovitz, M Tomishima, L Studer, G Lee,D Lyden, R Benezra, N Zaninovic, Z Rosenwaks, S Y Rabbany & S Rafii167 Real-time imaging of hepatitis C virus infection using a fluorescent cell-basedreporter systemC T Jones, M T Catanese, L M J Law, S R Khetani, A J Syder, A Ploss, T S Oh,J W Schoggins, M R MacDonald, S N Bhatia & C M RiceEndothelial cells from hESCs, p 161172 Rational design of cationic lipids for siRNA deliveryS C Semple, A Akinc, J Chen, A P Sandhu, B L Mui, C K Cho, D W Y Sah,D Stebbing, E J Crosley, E Yaworski, I M Hafez, J R Dorkin, J Qin, K Lam,K G Rajeev, K F Wong, L B Jeffs, L Nechev, M L Eisenhardt, M Jayaraman,M Kazem, M A Maier, M Srinivasulu, M J Weinstein, Q Chen, R Alvarez, S A Barros,S De, S K Klimuk, T Borland, V Kosovrasti, W L Cantley, Y K Tam, M Manoharan,M A Ciufolini, M A Tracy, A de Fougerolles, I MacLachlan, P R Cullis, T D Madden &M J Hope178 CORRIGENDA AND ERRATA© 2010 Nature America, Inc. All rights reserved.Systematicly improving siRNA delivery,p 172careers and recruitment179 Fourth quarter lag in biotech hiringMichael Francisco180 peopleADVERTISEMENTBiotech in ChinaA special report analyzing China’s emerging biotech and pharmaceuticalindustries. The main feature investigates the local strategies of westernand Chinese organisations, assessing massive investment programs and areturning western trained skilled work force. Will China achieve its ambitionto develop and innovative drug discovery industry to compliment its genericsindustry? How long do the experts believe it will be before China has its ownhome grown innovative new drug approved for the global markets? What areactivities of western drug companies in China? The “Biotech in China” specialreport follows Letters on page 176 and is produced with the commercialsupport from the organizations featured in the Advertorial Partnering Profiles.nature biotechnologyv

in this issue© 2010 Nature America, Inc. All rights reserved.Endothelial cell recipeEndothelial cells derivedfrom pluripotent stem cellsmight one day provide theraw material for engineeringor repairing blood vessels. Todevelop an improved methodfor vascular differentiation,Rafii and colleagues generatea human embryonic stemcell line that expresses greenfluorescent protein underthe control of the endothelialcell–specific VE-cadherinpromoter. Using this reportercell line to screen moleculesinvolved in early developmental signaling, they find thatinhibition of transforming growth factor (TGF)β beginning atday 7 of differentiation increases the yield of endothelial cellsand maintains the cells’ vascular phenotype for up to ten celldivisions. Mechanistic investigation identifies the transcriptionfactor Id1 as a key mediator of the effects of TGFβ inhibition.[Letters, p. 161]KAPredicting plant gene functionDespite extensive mutant screening,the functions of many plant genes arestill unknown. Lee et al. predict genefunction in the model plant Arabidopsisthaliana by gauging the likelihood thatpairs of genes are involved in the samebiological processes. Each pair of genesis assigned a score that combines manytypes of experimental and computationaldata gathered in Arabidopsis. Thescore also incorporates data from other organisms, such as yeast,worm, fly and humans, on genes that show substantial sequencesimilarity to Arabidopsis genes. Then, the function of an Arabidopsisgene is predicted based on the scores linking it to other genes withknown function. To demonstrate the utility of the approach, Lee etal. predicted and validated the roles of genes in seed pigmentation,lateral root development and drought sensitivity. By integrating multiplesources of data using methods customized for plants, Lee et al.predict gene function with greater confidence than by using only asingle source of data. This study provides a resource for identifyinggenes that influence agriculturally and economically important planttraits. [Resource, p. 149]CMWritten by Kathy Aschheim, Markus Elsner, Michael Francisco, Peter Hare,Craig Mak & Lisa MeltonPotent siRNA deliveryEmpirical screening has revealed novellipid nanoparticle formulations thathave substantially enhanced in vivodelivery of therapeutic small interfering(si)RNAs. Now Semple et al. haveset a new potency standard for siRNAdelivery to the liver by adopting a morerational approach to the design of cationic lipids. They refined anempirically identified cationic lipid (1,2-dilinoleyloxy-3-dimethylaminopropane),widely regarded as the benchmark for use inlipid nanoparticles, by dividing the structure into three functionalelements and then systematically testing modifications of each elementin isolation. This strategy to reveal structure-activity relationshipswas guided by the putative role of cone-shaped lipids to inducenonbilayer phases, such as the hexagonal H II phase illustrated here.When formulated to silence hepatic gene expression, the best-performinglipid variant conferred in vivo activity at siRNA doses aslow as 0.01 mg/kg in rodents and 0.1 mg/kg in nonhuman primates.[Letters, p. 172]PHLighting up HCV infectionProtocols for detecting hepatitis Cvirus infection without the need forcomplex manipulation of clinicalsamples or the use of genetically engineeredviruses is of prime importancefor many applications in basic researchand drug development. Rice and colleagueshave now developed a fluorescentreporter system that allows thedetection of individual cells infectedby wild-type viruses. The reportermolecule is based on a known target ofa viral protease, interferon-β promoterstimulator protein 1 (IPS-1). TheC-terminal part of IPS-1, includingthe mitochondrial targeting sequenceand the protease cleavage site, is fusedto a fluorescent protein and a nuclearlocalization sequence. Upon expressionof the viral protease the constructis cleaved and the fluorescentprotein relocalizes from the mitochondriato the nucleus. The authors usetheir reporter molecule to study viralpropagation in living cells and thedevelopment of stress responses in cells after infection. They alsodemonstrate that primary hepatocytes can be infected with hepatitisC viruses in vitro. [Letters, p. 167]MEnature biotechnology volume 28 number 2 february 2010v

in this issue© 2010 Nature America, Inc. All rights reserved.Linking antibody half-life to efficacyMutations that improve the affinity of antibodies for the neonatalFc receptor (FcRn) are known to enhance antibody longevityin vivo. Nonetheless, it has never been demonstrated that prolongedPatent roundupBringing a five-year patent dispute to an end, a federal court inBoston has ruled that Roche’s Mircera (methoxy polyethyleneglycolepoetin beta) does, indeed, infringe on Amgen’s patentsand slapped the Swiss drugmaker with a permanent injunctionbanning sales in the US market. Roche has agreed to a limitedlicense agreement with Amgen that will allow it to sell Mircera inthe US until 2014. [News, p. 112]LMTo solve the conflict between free access to scientific discoveryand intellectual property rights, Bentwich proposes a provisionalpatented paper application procedure that could promote earlierdisclosure of novel scientific knowledge and justify the requirementto grant inexpensive licenses for using inventions for theadvancement of other research. [Patent Article, p. 137] MFRecent patent applications in antibody fragments.[New patents, p. 141]MFexposure to Fc-engineered therapeutic antibodies necessarilyenhances in vivo activity. Desjarlais and colleagues use cynomolgusmonkeys and a humanized transgenic mouse model to show that thereduction in antibody clearance caused by increased antibody affinityfor FcRn enhances the antitumor activities of antibodies targetedagainst either an internalizing surface receptor (epidermal growthfactor receptor) or a cytokine (vascular endothelial growth factor).The enhanced pharmacokinetics associated with Fc-engineered variantsmay translate to greater convenience for patients, reduced costsand higher efficacy. [Brief Communications, p. 157] PHNext month in• Screening for drugs that reconfigure metabolism• Leukemia stem cell quiescence and resistance tochemotherapy• Directed evolution of an MRI contrast agent fordopamine• Autophagy harnessed to clear mutant huntingtinproteinvivolume 28 number 2 february 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.www.nature.com/naturebiotechnologyEDITORIAL OFFICEbiotech@us.nature.com75 Varick Street, Fl 9, New York, NY 10013-1917Tel: (212) 726 9200, Fax: (212) 696 9635Chief Editor: Andrew MarshallSenior Editors: Laura DeFrancesco (News & Features), Kathy Aschheim (Research),Peter Hare (Research), Michael Francisco (Resources and Special Projects)Business Editor: Brady HuggettAssociate Business Editor: Victor BethencourtNews Editor: Lisa MeltonAssociate Editors: Markus Elsner (Research), Craig Mak (Research)Editor-at-Large: John HodgsonContributing Editors: Mark Ratner, Chris ScottContributing Writer: Jeffrey L. FoxSenior Copy Editor: Teresa MooganManaging Production Editor: Ingrid McNamaraSenior Production Editor: Brandy CafarellaProduction Editor: Amanda CrawfordSenior Illustrator: Katie VicariIllustrator/Cover Design: Kimberly CaesarSenior Editorial Assistant: Ania LevinsonMANAGEMENT OFFICESNPG New York75 Varick Street, Fl 9, New York, NY 10013-1917Tel: (212) 726 9200, Fax: (212) 696 9006Publisher: Melanie BrazilExecutive Editor: Linda MillerChief Technology Officer: Howard RatnerHead of Nature Research & Reviews Marketing: Sara GirardCirculation Manager: Stacey NelsonProduction Coordinator: Diane TempranoHead of Web Services: Anthony BarreraSenior Web Production Editor: Laura GogginNPG LondonThe Macmillan Building, 4 Crinan Street, London N1 9XWTel: 44 207 833 4000, Fax: 44 207 843 4996Managing Director: Steven InchcoombePublishing Director: Peter CollinsEditor-in-Chief, Nature Publications: Philip CampbellMarketing Director: Della SarDirector of Web Publishing: Timo HannayNPG Nature Asia-PacificChiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843Tel: 81 3 3267 8751, Fax: 81 3 3267 8746Publishing Director — Asia-Pacific: David SwinbanksAssociate Director: Antoine E. BocquetManager: Koichi NakamuraOperations Director: Hiroshi MinemuraMarketing Manager: Masahiro YamashitaAsia-Pacific Sales Director: Kate YoneyamaAsia-Pacific Sales Manager: Ken MikamiDISPLAY ADVERTISINGdisplay@us.nature.com (US/Canada)display@nature.com (Europe)nature@natureasia.com (Asia)Global Head of Advertising and Sponsorship: Dean Sanderson, Tel: (212) 726 9350,Fax: (212) 696 9482Global Head of Display Advertising: Andrew Douglas, Tel: 44 207 843 4975, Fax: 44 207 843 4996Asia-Pacific Sales Director: Kate Yoneyama, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746Display Account Managers:Global Account Development Manager: Graham Combe, Tel: 44 207 843 4914, Fax: 44 207 843 4749New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717New York/Mid-Atlantic/Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481West Coast South: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805West Coast North: Bruce Shaver, Tel: (415) 781 6422, Fax: (415) 781 3805Germany/Switzerland/Austria: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419United Kingdom/Ireland: Jeremy Betts, Tel: 44 207 843 4968, Fax: 44 207 843 4749Scandinavia/Iceland/Spain/Portugal: Evelina Rubio-Hakansson, Tel: 44 207 843 4079,Fax: 44 207 843 4749France/Belgium/The Netherlands/Italy/Israel/Eastern Europe: Nicola Wright, Tel: 44 207 8434959, Fax: 44 207 843 4749Asia-Pacific Sales Manager: Ken Mikami, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746Greater China/Singapore: Gloria To, Tel: 852 2811 7191, Fax: 852 2811 0743NATUREJOBSnaturejobs@us.nature.com (US/Canada)naturejobs@nature.com (Europe)nature@natureasia.com (Asia)US Sales Manager: Ken Finnegan, Tel: (212) 726 9248, Fax: (212) 696 9482European Sales Manager: Dan Churchward, Tel: 44 207 843 4966, Fax: 44 207 843 4596Asia-Pacific Sales Manager: Ayako Watanabe, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746SITE LICENSE BUSINESS UNITAmericas: Tel: (888) 331 6288institutions@us.nature.comAsia/Pacific: Tel: 81 3 3267 8751institutions@natureasia.comAustralia/New Zealand: Tel: 61 3 9825 1160nature@macmillan.com.auIndia: Tel: 91 124 2881054/55npgindia@nature.comROW: Tel: 44 207 843 4759institutions@nature.comCUSTOMER SERVICEwww.nature.com/helpSenior Global Customer Service Manager: Gerald CoppinFor all print and online assistance, please visit www.nature.com/helpPurchase subscriptions:Americas: Nature Biotechnology, Subscription Dept., 342 Broadway, PMB 301, New York, NY10013-3910, USA. Tel: (866) 363 7860, Fax: (212) 334 0879Europe/ROW: Nature Biotechnology, Subscription Dept., Macmillan Magazines Ltd., Brunel Road,Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358Asia-Pacific: Nature Biotechnology, NPG Nature Asia-Pacific, Chiyoda Building,2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746India: Nature Biotechnology, NPG India, 3A, 4th Floor, DLF Corporate Park, Gurgaon 122002,India. Tel: 91 124 2881054/55, Tel/Fax: 91 124 2881052REPRINTSreprint@boston.nature.comNature Biotechnology, Reprint Department, Nature Publishing Group, 75 Varick Street, Fl 9,New York, NY 10013-1917, USA.For commercial reprint orders of 600 or more, please contact:UK Reprints: Tel: 44 1256 302 923, Fax: 44 1256 321 531US Reprints: Tel: (617) 494 4900, Fax: (617) 494 4960

EditorialPhRMA wants you!The biotech brand is in danger of being sullied by the blurring of pharma and biotech boundaries.© 2010 Nature America, Inc. All rights reserved.Early last month, the Pharmaceutical Research and Manufacturers ofAmerica (PhRMA) trumpeted that its drive to diversify membershiphad led to the recruitment of seven biotech firms. The drive began last July,shortly after Roche announced that it was leaving PhRMA and joining theBiotechnology Industry Organization (BIO), characterizing itself as “theworld’s largest biotechnology company.” Tit-for-tat membership battlesof healthcare lobby groups are usually of little importance. However, thestatus that Roche attaches to its biotech identity and the pursuit of biotechfirms by PhRMA suggest that the ‘biotech’ brand has a distinct anddesirable cachet.The pharmaceutical industry does have a public relations problem. AHarris poll from August 2009 found that 84% of US adults blame thepharma industry for the problems with the US healthcare system. Trustratings for pharma are also low, with only 7–14% of US adults willing tobelieve company statements as true (Harris Polls, 2003–2007). In contrast,attitudes toward the biopharmaceutical sector are generally positive.An April 2009 survey from the state industry association, MassBio, forinstance, found 51% of voters had positive feelings towards biotech.So ‘biotech’ is young, thrusting, optimistic and tackling healthcarethrough research while ‘pharma’ is gigantic, profiteering, world-wornand tackling healthcare by increasing market share. In reality, of course,no such clear-cut divisions exist. The biotech and pharma sectors formparts of a continuum, indivisible by the presence or absence of biologics,size, research concentration, clinical competence, greed or vulnerabilityto acquisition.Readers may be surprised to learn that Roche is more of a biologicscompany than a drug company: in 2009, 54% of its revenues came frombiologics and only 24% from small molecules. Other pharmaceutical companieshave been developing and acquiring biologics competence at anincreasing rate over the past decade or so. Pfizer’s acquisition of Wyethlast October was motivated partly by the need to have more biologics.Conversely, a significant portion of the pipeline of Amgen and Genzymeis small molecules.The blurring of the biotech-pharma boundary has led to convergentevolution of lobby groups. PhRMA’s current mission is advocacy on behalfof “pharmaceutical/biotechnology research companies” while BIO’s is tobe “the advocate for its member organizations—both large and small.”Little to choose between the two.There is, however, one very important difference. Whereas PhRMA has34 full members, BIO has 1,200 members, most of which pay only nominalfees. BIO is thus the only advocate for the smaller, younger, nonrevenuedrivencompanies. Its voice on behalf of smaller firms may not always beloud and clear, but it is surely a voice.By donning the ‘biotech mantle’, PhRMA may be hoping to capture forits members some of the public and political sympathies enjoyed by BIO.In the face of this direct competition, BIO needs to make a choice. It canout-PhRMA PhRMA, becoming a better advocate for large pharma, tryingto be all things to all companies. Or it could differentiate itself by becomingan out-and-out advocate for small to medium-sized biotech, deemphasizingthe large company agenda and rededicating itself to the innovative edgeof the industry. We hope they choose the second path.Impossible fixesImpractical solutions to European biotech financing don’thelp anyone.couple of months ago, the European Commission and the industryA trade association European Biopharmaceutical Enterprises held aclosed meeting in Brussels to discuss a survey from ECORYS (Rotterdam),the Danish Technological Institute (DTI: Taastrup) and three other consultantsundertaken as part of a €2.99 ($4.22) million contract from theCommission (see p. 110). The survey pointed to three (obvious) fundinggaps during biotech development. It also estimated that 40% of Europeanbiotechs needed more cash before the end of 2010. Urgent problems,indeed, requiring urgent solutions. So what did the report recommend?First, it proposed that the European Commission should consider sector-specificpolicy measures targeting the biopharmaceutical sector. In arational world, this is plausible and sensible. But the European Union isnot rational. It is an aggregation of formerly warring nation states eachseeking economic advantages while acting out historical resentmentsthrough petty and circular administrative fine-tuning. To counter thenationalistic tendencies, the European Commission’s competitivenesslegislation forbids governments from instituting sector-specific incentivesfor industry. Nations can support companies that are small or young orresearch intensive, but they cannot single out specific sectors, like IT orbiotech, for specific help. Short of overhauling the entire philosophy of theEuropean endeavor, this recommendation, therefore, is a non-starter.A second proposal in the report suggests that national and Europeanpolicy makers should reform the financial markets across Europe to allowventure capitalists to operate across national borders. Again, nice idea, buta pipe dream. In the main, the European Union has 27 distinct financialsystems with 27 national regulatory systems and asynchronicity throughoutits multiple economic cycles. This, too, is a solution for the next decadeor beyond.Thankfully, by the time it went to press, the report had been strippedof suggestions that European biotech should be financed using gold barsfrom the croc at the end of the rainbow or getting Sir Bob Geldof andBono to organize a Bio-Aid telethon. European biotech financing is atough problem in need of imaginative solutions. But solutions must bepractical—not outside the realms of reality.nature biotechnology volume 28 number 2 february 2010 103

NEWS© 2010 Nature America, Inc. All rights reserved.TysabriapprovedTysabri goes back on the marketmid-year, with new safety andreporting measures in placeis now working on PML. Biogen’s scientistshave already developed an assay that identifiescarriers of JC virus. It is estimated thatupwards of 50% of people are infected withthis opportunistic virus, which mostly lieslatent indefinitely. PML is extremely rare inanyone but those with impaired immunesystems, such as people with AIDS, B-cellmalignancies or systemic lupus erythematosus(SLE). Once JC virus is activated itinfiltrates the central nervous system (CNS),causing neurobehavioral symptoms, whichcan become life threatening. Fortunately,Tysabri’s effects are rapidly reversible, soif patients at risk of PML can be identified,their normal immune function can quicklybe restored and most patients recover.The assay Biogen has developed “only tellsyou whether the person has been infected ornot,” says Yednock. “We also need to be ableto tell the level of viral replication or whetherit has transitioned from latent to activatedvirus [capable of infecting brain cells].”“This is such a ubiquitous virus, somethingmust happen to enable it to grow in glialcells,” says Joseph Berger of the Departmentof Neurology at the University of KentuckyCollege of Medicine.One line of research has suggested that thestructure of Tysabri itself might be to blame.Researchers at Genmab in Copenhagen havereported that IgG4 antibodies like Tysabri,which are bi specific and contain a flimsybackbone, can exchange the Fab-arm portionof the antibody with endogenous IgG4s withthe wrong specificity (Nat. Biotechnol. 27,767–771, 2009). If Tysabri molecules recombinedwith endogenously produced anti–JCvirus antibodies, the researchers postulated,the resultant bispecific mAb might capturethe virus and carry it into the CNS.Scientists at PanGenetics in Utrecht, TheNetherlands, who have an IgG4 mAb inRaptiva is voluntarily withdrawnby Genentech in June, after fourconfirmed PML cases andthree deaths2004 2005200620072008 2009Tysabri pulled fromthe market after twoPML cases identifiedFigure 1 Timeline for the reintroduction of Tysabri and PML cases.“Dear Doctor” letter warns thatpatients on Rituxan are at anincreased risk of developing PML.EMA begins a new reviewof TysabriJunePML cases are confirmedin patients takingRituxan and RaptivaOctoberdevelopment, have challenged this theory,pointing out that clinical data suggest there isvery little circulating JC virus to be capturedand no mechanism by which Tysabri couldtransport virus into the CNS (see p. 123;SciBX 2(32); doi:10.1038/scibx.2009.1231).Also, mAbs are not the only drugs that cancause PML. Basel-based Roche’s CellCept(mycophenolate mofetil, which targetsinosine monophosphate dehydrogenase 1),for example, has also been linked to the disease.Most experts appear to be leaning towardother theories. Recent studies suggestthe antibodies can specifically affect themovement of mature and immature lymphocytesin the body. By binding VLA-4,Tysabri prevents lymphocytes from bindingto the vascular cell adhesion molecule-1(VCAM-1), and if hematopoietic stem cellsand pre-B cells are unable to bind VCAM-1, they migrate out from the bone marrow.“We have plenty of data now that showthese progenitor cells are found at higherthan normal rates in the peripheral systemin patients on this drug,” says Gene Majorof the Laboratory of Molecular Medicineand Neuroscience at the National Instituteof Neurological Discoveries and Stroke inBethesda, Maryland. Some of these B cellscontain JC virus that may replicate promptedby DNA-binding proteins (New Engl. J. Med.361, 1041–1043, 2009). This mechanismcould also explain how Rituxan can causePML because this mAb reduces the numberof CD20 + B cells in peripheral blood, whichalso stimulates the release of progenitor cellsfrom the marrow.It is also thought that Tysabri limits thepassage of cytotoxic T cells into the brainbecause they need to bind VCAM-1 tomigrate out of the circulation. Because theseare precisely the cells that would defendagainst the JC virus, the drug could be handicappingpatients’ native immune defense.In the meantime, regulators are keepinga close eye on Tysabri and Rituxan.The European Medicines Agency (EMA)was reviewing the data about Tysabri andPML as Nature Biotechnology went to press.Crystal Rice, an FDA spokesperson, wrotein an e-mail, “We have been receiving andcontinue to receive reports of PML cases inreal time, are monitoring the incidence ofPML both in the United States and worldwideon an ongoing basis, and are continuingto assess the issue to determine the need forfurther regulatory action.”With Tysabri, risk of PML has currentlybeen calculated at one in a thousand, butthat could change as more people take thedrug for extended periods. For Rituxan, thepicture is less clear. PML occurs in rheumatoidarthritis and SLE patients sometimeseven when they are not taking Rituxan. “It’snot clear how much additional risk Rituxanimparts,” says Leonard H. Calabrese ofthe Cleveland Clinic in Cleveland. On thebasis of current data, he believes the risk isso low the number is not even calculable.Nonetheless, rheumatologists are very concernedabout PML. Calabrese says he andcolleagues recently conducted a survey andfound that “the level of concern doesn’tcoincide with the available risk data.” Manyarthritis drugs, he points out, have seriouspotential side effects. But with Rituxan andPML, “there is a disconnect between knowledgeand fear.”It’s also unclear whether other immunemodulatingdrugs that mobilize hematopoieticstem cells and pre-B cells may encounterproblems similar to those Tysabri and Rituxanhave. Copenhagen-based Genmab’s Arzerra(ofatumumab), approved in October forchronic lymphocytic leukemia, Genentech’socrelizumab (in late-stage trials for rheumatoidarthritis and SLE) and Biolex’s BIX-301non-Hodgkin’s B cell lymphoma, all targetCD20. As yet, there are no reports of PMLcases associated with these drugs.“We are perturbing very specific arms ofa highly complex system,” Berger says. Thatpower can clearly backfire if you don’t knowall the drug’s effects. As a result, Biogen’sSandrock expects risk management plansare “here to stay.” Berger says he’s been tryingto convince drug companies to establishregistries for drugs with novel mechanisms.“As a community, we need to be preparedto identify these problems as they arise,” hesays.Malorye Allison Acton, Massachusetts106 volume 28 number 2 february 2010 nature biotechnology

newsPfizer stakes a claim in plant cell–madebiopharmaceuticals© 2010 Nature America, Inc. All rights reserved.On December 1, Pfizer became the first bigpharma to commit to take to market a latestagebiologic drug produced in plant cells.It acquired rights to taligurase alfa, a form ofthe enzyme glucocerebrosidase in developmentfor the treatment of Gaucher’s disease,from Protalix Biotherapeutics in Carmiel,Israel. Protalix has completed phase 3 studiesand has submitted a new drug application forthe drug, also known as prGCD, eyeing USmarketing approval in 2010. At the request ofthe US Food and Drug Administration (FDA)last year, the company has already begun supplyingprGCD to patients in the US underan expanded access program and similarly topatients in the EU under a compassionateuseprotocol. This apparent comfort level ofregulators, along with the interest of a majordrug company, signals a new level of recognitionof plant cell–based manufacturing as aviable and potentially less expensive alternativeto mammalian and bacterial productionof biopharmaceuticals, including biosimilarversions of existing drugs.Protalix has already collected $65 millionfrom the deal, which gives New York–basedPfizer worldwide rights to prGCD, excludingIsrael, and could earn another $50 million inmilestones. Protalix will continue to manufacturethe drug, which it produces in carrotcells, pay 40% of all expenses going forwardand receive the same percentage of revenuesin return. The company’s prGCD will competewith Genzyme’s Ceredase (alglucerase),a form of the enzyme beta-glucocerebrosidasepurified from human placental tissuethat is modified to be terminated with mannose,and Cerezyme (imiglucerase), a recombinanthuman beta-glucocerebrosidase witha His495→Arg substitution and the samesugar modification. Both of Genzyme’sproducts are indicated for the treatment ofGaucher’s disease, a rare lysosomal storagedisorder resulting from a hereditary deficiencyin the glucocerebrosidase enzyme.Gaucher’s disease is the most prevalentamong the group of lysosomal storage disorders,which have been a historic focus forGenzyme in Cambridge, Massachusetts.This is Pfizer’s first move into the area ofrare and neglected diseases, the result of aprocess the company began a year ago toidentify such opportunities. “Protalix’s nameand technology platform and their work inGaucher’s disease came to the top of thatlist. We approached them in the middle oflast year and things moved fairly quickly,”says Andrew Curtis, biosimilar and orphandrugs director for Pfizer’s established productsbusiness.ProtalixProtalix’s bioreactor plant cell system. The GMP-approved system is set up to manufacture a range ofproteins, including antibodies, complex enzymes and plant-derived pharmaceuticals.nature biotechnology volume 28 number 2 february 2010 107

NEWS© 2010 Nature America, Inc. All rights reserved.“This was a unique opportunity to takeadvantage of their plant cell technologyplatform with this lead drug to perhaps providea more cost-effective therapy,” he says,adding that treatments for lysosomal storagediseases have typically been among the mostexpensive, costing several hundreds of thousandsof dollars annually. Protalix’s phase 2data and interim reads on its phase 3 study,including available reports from data safetyand monitoring boards, were reassuring, aswas the fact that regulators had asked Protalixto initiate early access programs. “Typically,companies are not asked to do that unlessFDA or EMEA [the European MedicinesAgency] or both have a good level of comfortaround the safety and efficacy of the drug,”says Curtis. “All those things came to bear onthe comfort level we had with Protalix, eventhough it would be the first plant cell–basedtherapy to be approved.”Curtis is also quick to point out thatalthough there may be future opportunitiesto leverage the Protalix platform, this was aproduct-oriented transaction. “This deal wasborn from the commercial arm of Pfizer. Wesee this obviously as an opportunity to workmore closely with them going forward, tomaximize their technology platform, but thetechnology platform trailed behind. We gotthe drug and along the way discovered howwonderful the platform is.”Beyond prGCD, the Protalix technologyplatform could potentially provide opportunitiesfor Pfizer to work in other raregenetic diseases and even beyond that, inthe production of biosimilars, to lower thecost of goods, says Curtis. Protalix also hasprograms in preclinical development for abiosimilar version of the anti-tumor necrosisfactor alpha fusion receptor protein Enbrel(etanercept) and for enzyme replacement totreat Fabry’s disease, another lysosomal storagedisorder.In plant biotech, as is true for other biotechdeals, the trend “seems to have gone,over the last ten years, away from companiesthat have platform technology much moretowards product-driven investments,” saysCharles Arntzen of the Biodesign Institute atArizona State University in Tempe. “Protalixnever goes around talking that they have acell-culture system, except at plant meetings.They focus on the enzyme they are producingand the product niche it will fit into.”Pittsboro, North Carolina–based BiolexTherapeutics is the only other company witha plant-derived compound in late-stage trials.Their controlled-release form of interferonα-2b made in aquaculture, currentlyBox 1 Green antibody farmingViral vectors have enough going for them that the technique could become a staple forprotein production in whole plants; they can achieve high levels of expression in a veryshort time. “You can use agrobacterium, for example, to essentially bathe the interiorsurface of the leaf,” says Arntzen. “Once you get one copy of the merged virus [it is usuallyintroduced in two parts along with a recombinase delivered as a separate gene] in the cell,boom, off it goes. You adjust conditions so you have a coordinated assembly so every cellin the leaf is simultaneously getting the effect of having a virus starting its replication inthat cell.” Icon was among the first to successfully use these viral vectors, and there are atleast a dozen other companies that have gone that way. “It looks to me like the wave of thefuture,” Arntzen says.The biggest advantage may come in producing monoclonal antibodies, where the bestwork is being done using two separate viruses: one produces the light chain, the other theheavy chain. Essentially, two RNA species are introduced into a single cell at the sametime. “They assemble beautifully and it’s routine now to get about half a gram of antibodyout of a kilogram of plant leaves,” says Arntzen.“There are enough data, not necessarily all published, especially for immunoglobulinsof IgG type made in plants, and enough information and knowledge in several largecompanies to say that all these antibodies are fully active and pharmacologically asgood as those produced in other systems,” says Gleba, adding that Genentech of S.San Francisco, California, and others have shown that their existing molecules are lesseffective because of fucose added to the glycans. “If you remove it, the effector functionimproves dozens of times and that should directly translate into therapeutic effect, so youneed less of the molecule,” he says.MRin phase 2b for the treatment of hepatitis C,received $60 million in venture funding froma group led by Clarus Ventures and OrbimedAdvisors, both known for late-stage, product-orientedinvestments.It’s long been believed that plant cell–based manufacturing has the potential to beless expensive than mammalian or Chinesehamster ovary cell–based methods, in partbecause plants produce protein with a glycosylationpattern closer to human, saysArntzen (Box 1).That’s apparently the case with prGCD. Toproduce Ceredase and Cerezyme, Genzymehas to clip off additional sugars to exposeterminal mannose residues, says DavidAviezer, Protalix’s president and CEO.“With plant technology, by performing ER[endoplasmic reticulum] retention of theprotein during the glycosylation process, wecan obtain the correct mannose glycosylationpattern directly made by the cell.” Thatreduces processing costs—prGCD could bean order of magnitude less expensive to producethan either Cerezyme or another potentialcompetitor such as Shire’s velaglucerase.Chineham, UK–based Shire is expecting anFDA approval decision for their product bythe end of February 2010.Yet some in the field see Protalix’s platformas a somewhat crude first-generationtechnology that only touches on the potentialfor plant-based pharmaceutical produc-tion. “You increase the cost of production bygoing back to fermenters,” says Yuri Glebaof Icon Genetics, now part of the BayerInnovation Group in Halle, Germany. What’smore, expression is not exceptionally highin the Protalix cell line, and the company isusing expression cassettes that have been incirculation for a generation, he says. “Protalixhas shown these fermenters are cheaper, soyou can have 20–30% cheaper product as aresult, which could justify the new businesscase. But with other technologies you can geteven higher expression levels and lower costof goods,” he adds.On the other hand, cell culture brings theadvantages of a plant organism in “the sameregulatory environment FDA has learned andset for the last 20 years” for biopharmaceuticals,says Aviezer—in clean rooms and underthe same standard operating procedures forgrowing mammalian cells and purifying proteins.And Icon’s Gleba acknowledges thatProtalix’s success in product developmentwith prGCD shows they have a keen businesssense. “If you are not strong on one sideyou have to compensate by being excellenton another, and by all accounts, they are,” hesays. The deal with Pfizer and the approvalof prGCD “should open the floodgates, inmy opinion,” he says. “It is by far the mostsignificant development in the plant-madepharmaceuticals arena right now.”Mark Ratner Cambridge, Massachusetts108 volume 28 number 2 february 2010 nature biotechnology

newsAmylin’s $1 billion heavyweight deal© 2010 Nature America, Inc. All rights reserved.In November, Amylin announced a $1 billionpartnership with Osaka, Japan–based Takeda tocodevelop and commercialize obesity treatments.Takeda paid the San Diego–based biotech $75million upfront for Symlin/metreleptin combinationas part of an agreement that could exceed $1billion if certain development and sales-dependentmilestones are hit. The deal also includes Amylin’samylinomimetic compound davalintide, which iscurrently only in phase 2 studies. But, as StephenO’Rahilly, director of the Metabolic ResearchLaboratories at the University of Cambridge in theUK, points out, Symlin (pramlintide) is alreadyapproved and is “used by a lot of type I diabetespatients to smooth out control and prevent theweight gain that happens when on insulin.”The agreement comes amid a surge of dealsin metabolic disease, particularly for diabetestreatments that have potential weight loss benefitsfor the obese. For example, on December 23,Paris-based Sanofi-Aventis paid €100 ($143)million for a 19.9% stake in Zealand Pharma. ThisAmylin has gained a strong position in the already crowded metabolic disease marketplace.Copenhagen-based biotech is developing a peptideanalog of Amylin’s Byetta (glucagon-like peptide1 (GLP-1)/exendin 4) for type 2 diabetes, which has also shown efficacy in promoting weight loss. In what has become a crowdedmarket, products will likely gain a competitive edge if they can fight both metabolic disease and obesity, the latter with a potentialmarket of 300 million people worldwide.Amylin already looks to have consolidated its position. According to Collins Stewart analyst Salveen Kochnover, “They may not befirst to market, but I think Amylin signed a nice partnership, so they’re well-positioned if any of their drugs work.”The nearest competitor, she says, is Vivus, the Mountain View, California–based company, whose once daily capsule Qnexa(phentermine and topiramate) for the diabetes market achieved 15% weight loss in phase 3 trials. Elsewhere, Novo Nordisk haslaunched a long-acting GLP-1 analog Victoza (liraglutide; Nat. Biotechnol. 27, 682–685, 2009) in Europe as a type 2 diabetestreatment. At present, the Copenhagen-based biotech still awaits a decision from the US Food and Drug Administration. Recentstudies show Victoza to be more effective in helping people shed weight than the anti-obesity pill orlistat, marketed as Xenical byRoche and Alli by London-based GlaxoSmithKline.But Amylin’s foothold in the marketplace is further strengthened by its two first-in-class synthetic gut hormone drugs approvedfor type 2 diabetes—the insulin boosting glucagon-like peptide 1 (GLP-1) analog Byetta (exenatide), and Symlin. Symlin is asynthetic version of amylin, a neuroendocrine hormone secreted by pancreatic betaThe path to approval forAmylin and its competitorswon’t be easy.cells, that boosts insulin action and helps regulate appetite, food intakeand glucose control. In diabetes sufferers, sensitivity to Symlin is often reduced overtime, in much the same way that insulin sensitivity is lost. By combining Symlin withmetreleptin, a recombinant version of human leptin (a hormone secreted by fat cellsthat acts on the hypothalamus to regulate food intake), the company hopes to treatobesity. Thus far, the drug combo has shown impressive weight loss in animal andhuman trials, and Amylin is expecting to announce favorable results from phase 2 trials following positive top-line data. The otherdrug included in the Takeda deal is davalintide, a second-generation amylin receptor agonist, which mimics the action of amylin.The path to approval for Amylin and its competitors won’t be easy. Safety issues that led to the withdrawals of the cannabinoidreceptor antagonist Acomplia (rimonabant) and, before that, appetite suppressant Redux (dexfenfluramine), mean that novelproducts are likely to face intense scrutiny. In a recent study, the Symlin/metreleptin combination produced impressive weight lossin a broad population of obese people, but a subanalysis showed less dramatic effects in severely obese individuals. And, accordingto Tom Hughes, CEO of obesity therapeutics company Zafgen, headquartered in Cambridge, Massachusetts, although Amylin’sinjectible is more expensive and inconvenient than an oral drug, the ‘hormone replacement’ approach avoids the idiosyncratictoxicity inherent to small molecules. “I think that may be at the heart of why this deal got as much as it did at the stage that it’sat,” he says. “It is appealing on a number of levels to think that you’re giving back something that is important or missing, whichis restoring the balance to the patient and leading to the effect.” However, John Wilding, a clinical researcher at the Universityof Liverpool, UK, says there’s only a nugget of truth in the claim. Pharmacological doses are orders of magnitude higher thanphysiological concentrations, and in type 2 diabetes, amylin levels actually tend to be high.Hayley BirchAmylinnature biotechnology volume 28 number 2 february 2010 109

NEWS© 2010 Nature America, Inc. All rights reserved.in briefPeruvian GM advocate facescriminal chargesErnesto Bustamante onPeruvian RPP radio.RPPA molecular biologistcould face a prisonsentence for criticizinga report on transgenicgene spread. ErnestoBustamante Donayre,vice president of thePeruvian Collegeof Biologists,a professionalorganization, standsaccused of defamation,a criminal offense,which in Peru can carry a prison term or fine.What triggered the suit was his public criticism ofa report prepared by Antonietta Ornella GutiérrezRosati, a biologist at the La Molina NationalAgricultural University in Lima, identifying a P34Spromoter and NK603 and BT11 transgenes in 14of 42 maize samples from the Barranca region.Gutiérrez sent summaries of her findings to boththe National Agricultural Research Institute andEl Comercio newspaper in 2007 calling for amoratorium on transgenic crops until biosafetyregulations are in place to prevent the spread tohuman food. Bustamante, a frequent contributorto radio and print, with no financial links to cropcompanies, described the alleged detection ofthree simultaneous transgenic events from twofirms as “absurdly improbable” in his newspapercolumn and called for her claims to be peerreviewed. “The main point of my criticism,”Bustamante says, “was her going to the pressinstead of to her peers.” After Bustamanterefused to retract his statements, Gutiérrezfiled a suit for defamation. She later presentedher findings to the Peruvian Genetic Society ofwhich she is president, but would not commenton the case, except to say that “you must userespect” in scientific discussion and that hercritics have “polarized” the debate. AlthoughPeruvian farmers already import transgenicproducts for animal feed, several interest groupsoppose their widespread introduction, which theylabel a foreign intrusion and threat to Peruvianbiodiversity. An ongoing investigation is seeking toreplicate Gutiérrez’s findings, but the governmentlacks the regulations to enforce its biosafety lawseven if it does find transgenic crop outcrossing.The criminal case, however, threatens to stifleall scientific discussion. “Regardless of whetherhe gets sentenced or not I don’t think anyone isgoing to criticize anything,” says plant scientistWayne Parrott, from the University of Georgia, aregular visitor to Peru. Bustamante’s colleagueand supporter Luis Destefano Beltrán of theCayetano Heredia Peruvian University agreesthat “many people have tried to avoid takingsides.” Peru retains criminal defamation laws,which the Inter-American Commission on HumanRights concluded in 1995 are incompatible withthe American Convention on Human Rights.Bustamante, who expects a ruling early thisyear, says, “The point is not whether I’m right orwrong. It’s the fact that for criticizing somebodyon scientific grounds I’m being tried in criminalcourt.”Lucas LaursenEC convenes crisis talks on Europeanbiotech sectorLast December, officials of the EuropeanCommission (EC), together with EmmanuelChantelot, executive director of the industrytrade association European BiopharmaceuticalEnterprises, convened a closed meeting to discussthe plight of the European biotech sector.Held at the EC headquarters in Brussels, themeeting was attended by policymakers, CEOsfrom small-to-medium-sized (SME) companies,national biotech associations, venturecapitalists (VCs) and big pharma representatives.The working group discussed the findingsof an EC-commissioned survey carried out bythe Danish Technological Institute (DTI) inTaastrup on the problems of access to financefaced by the biopharma industry. According tothis study, lack of access to capital is threateninginnovation and competitiveness in the sector,with 40% of companies facing extinction by theend of the year without a further cash injection.On the basis of the group’s discussions, severalpolicy recommendations were put forward withthe potential to increase sustainability of theEuropean sector (Box 1).The DTI’s findings portray European biotechas a rapidly deteriorating sector: 7% of theregion’s biotech SMEs need capital immediately,40% must raise capital within a year and nearly75% over the next 2 years. “Some SMEs are goingto go out of business, and many are stretchingresources and cutting back on programs” saysThomas Saylor, chair of the SME platform ofthe European Association for Bioindustries(EuropaBio). The data were gathered througha survey of 87 biopharma companies in Europethroughout May and June 2009, desk researchof reports and interviews with experts. The surveyis deemed to be representative of the stateof European biopharma, although there is anintentional bias toward smaller and youngercompanies, following the EC’s requested samplingcriterion.“The core of the problem is that there is lessventure capital money for small biotechs,” saysChantelot, and lack of cash creates fundinggaps in the chain from startup to initial publicoffering. The most severe gaps are at the early,high-risk stages, making it hard for fledglingcompanies to get off the ground or even stayafloat. The key reason for this gap in Europe,says Ivica Cerina, a partner at NGN Capital inHeidelberg, Germany, is pressure over the pastfive years to “de-risk,” pushing investors to focuson later stages and avoid risky startups. Privatesources of equity account, on average, for some60% of all SME funding.To address this problem, the DTI report recommendsincreasing public co-investments inmicro-fund and business angels to provide theseed money needed to get an SME rolling, whilesimultaneously creating tax incentives for doingso. As the former type of funding tends to operateon different timescales from venture funding,and with different skill sets and strategic views,several of the VCs at the meeting also pushedfor additional money. As such, Cerina wouldprefer to see more funds go into, or side-bysidewith, existing venture capital funds “eitheras dedicated early stage vehicles or directly intospecialized, early stage–savvy VCs”.Enhancing access to existing pools ofmoney could also support the recovery process,says Saylor, citing the European Union’s7 th Research Framework (FP7). “Funding hasbeen cumbersome to apply for, and for smallcompanies, in particular, it can be a hugeadministrative burden to take on the reportingrequirements under the framework,”Saylor says. He would also like to see followonfunding of the sort established by the UK’sTechnology Strategy Board, which makes aFlying the biotech flag. Policymakers, investorsand companies gathered at the Commission’sheadquarters in Brussels to discuss how toovercome funding shortages faced by Europeanbiotechs.AP Photo/Virginia Mayo110 volume 28 number 2 february 2010 nature biotechnology

news© 2010 Nature America, Inc. All rights reserved.company eligible “for a bigger pot of money”on meeting certain milestones.Another example is the €2 ($2.9) billion fundavailable to SMEs until 2013 through the Risk-Sharing Finance Facility (RSFF) set up by theEC and the European Investment Bank. This is agreat idea in principle, but the eligibility criteriatend to rule out biotech SMEs. To apply, companiesmust be profitable, which many biotechs arenot for several years as they plough their moneyinto R&D. “For biotech SMEs, the RSFF criterianeed to be changed,” says Dirk Carrez, directorof Industrial Biotech at EuropaBio.Others would like to see the LuxembourgbasedEuropean Investment Bank (EIB) andEuropean Investment Fund (EIF) take boldersteps to support fledgling biopharma enterprises.Flexible loans and other financing instrumentsfor companies close to self-sufficiencyare currently made available by EIB and EIF,typically for late-stage co-investments of over€20–30 ($29–43) million a year. “We’d like tosee a bigger share of that funding going to lifescience companies,” says Chantelot, who suggeststhe EC draw up a new mandate to give thesector priority in funding.Chantelot is also keen for big pharma toexpand its corporate venture capital programsfor early stage SMEs—a trend already underway(Nat. Biotechnol. 27, 403– 404, 2009). Publicfunds could help offset the risk, he adds. “Forlate-stage biotech companies in partnershipwith big pharma, EIB could lend or co-investsome money to minimize the risk for the otherparties.”A key issue is determining whether the lackof capital currently stifling many companiesreflects a problem with the financial instrumentscurrently administered by the EIB and EIF ora problem with the companies themselves. Inother words, are deserving companies being letdown by the current system, or should thesecompanies not be receiving funding at all as theyare unlikely to become sustainable enterprises?“The EC is looking carefully at this question,”says Giulia Del Brenna, head of the EC unit oncompetitiveness in the pharmaceuticals industryand biotechnology, speaking in a nonofficialcapacity.European officials are also consideringanother option, and that is to set up a EuropeanBiopharmaceutical Innovation Fund specificallydedicated to biotech startups. VCs arguethat this could indeed be useful, particularly forearly stage companies at risk, but stress the needfor the fund to be administered with the samedue diligence and market considerations thattypically constrain private financing.Some VCs disagreed with the DTI report’sconclusions that capital supply was part of theproblem. “Capital is available in EU, but it’s beenBox 1 Recommendations for European biotechThe report commissioned by the EC (Directorate-General Enterprise and Industry) fromThe Danish Technological Institute is entitled The Financing of BiopharmaceuticalDevelopment in Europe (http://ec.europa.eu/enterprise/sectors/biotechnology/documents/index_en.htm). It provides several suggestions for increasing the access companies haveto capital:• The EC should undertake an in-depth analysis of the effects of different tech transfermodels used within and outside Europe (good practice) to improve the effectivenessof biopharmaceutical R&D and commercialization and ensure that the sector iscompetitive and able to attract private funding.• Early stage investments should be encouraged to ensure that innovative companiescontinue their development activities, perhaps by supporting micro-funds andinvestments by business angels in early stage biopharmaceutical companies throughpublic co-investments and tax incentives.• Establish a European Biopharmaceutical Innovation Fund focused on investingin biotech companies while considering the geographical reach of the existingfunding mechanisms at European and national levels to ensure that global fundingopportunities are exploited.• Improve the framework conditions for both biopharmaceutical companies and theventure capital industry in Europe.kept dry,” says NGN’s Cerina. “While in bettertimes most companies would have found a safeharbor, now there are too many companiesthat are not compelling enough chasing limitedcapital resources, and only the best oneswill find investors.” Michiel de Haan, a VC andgeneral partner of Aescap Venture Management,Amsterdam, agrees. “Out of 100 various proposalswe look at we only invest in one or two. It’sa very strong selection process, and that doesn’tmake us popular. But for high-quality propositionsthere is healthy competition and enoughVCs around to invest.”Despite their reservations, VCs would stillwelcome moves to make investing in biotechSMEs more attractive, such as tax breaks andother forms of risk sharing with public investments.Programs that help companies withsubsidies, guaranteed loans, and technologyand innovation loans on which interest is paidback according to the success of the productwork very well, says de Haan, who cites France’stax-credit system and Holland’s TechnologyStart-Up Programme (a specialized seed fund),as illustrative examples. Indeed, de Haan arguesthat there’s a need to “look at those countriesNew product approvalsActemra (tocilizumab)Genentech(S. San Francisco,California)where these types of financial instruments arevery good for the biotech startup, and learn fromthem.”One other recommendation from the DTIreport is to improve the quality of new venturepropositions through better technology transferfrom universities. For Marja Marakow, currentlythe chief executive of the European ScienceFoundation and former vice president of theUniversity of Helsinki, a key issue is professionalizingtechnology transfer. “Public universitiestypically cannot hire first-class professionalswith the requisite expertise to run tech transferoffices.” Instead, they are frequently staffed bycivil servants who lack the relevant research andprivate-sector experience.The EC recognizes that biotech SMEs areimportant to Europe’s economy, but whetherthe biotech sector should receive targetedhelp remains an open question, and one to beexplored further as a new commission comesinto office this month. What is clear, says theEC’s Del Brenna, is that the EC is listening tothe biotech sector and digesting the meeting’srecommendations.Dan Jones Brighton, UKThe US Food and Drug Administration approvedGenentech’s Actemra to treat rheumatoid arthritis.Actemra is the first US-approved drug to target interleukin-6and is aimed at patients who do not respondto older tumor necrosis factor alpha inhibitors.Actemra is also approved in the EU, India, Brazil,Switzerland and Australia.nature biotechnology volume 28 number 2 february 2010 111

© 2010 Nature America, Inc. All rights reserved.NEWSin briefIrish bioethics council axedStem cell research in Ireland has been throwninto a state of confusion, after a recentgovernment decision to cut all funding for theIrish Council for bioethics at the end of theyear. Paradoxically, the move coincides with arecent Supreme Court decision that removessome of the legal uncertainties surroundinghuman embryonic stem cell research in thecountry. The judges denied a woman the rightto proceed with in vitro fertilization without theconsent of her estranged husband. In doing so,the court ruled that embryos outside the wombare not protected by the country’s constitutionalprotection of the unborn. Although this rulingaffects human embryonic stem cell (hESC)research by providing clarification on the statusof pre-implanted embryos, scientists remain waryof proceeding until a supporting framework isin place. “I’m going to behave responsibly. It’sgoing to be done by the book,” says Barry Mooreat University College Cork (UCC), who has alreadyreceived clearance to carry out hESC researchfrom UCC’s research ethics committee. Irelandhas no laws governing human stem cell researchand scientists have been operating in a legallimbo. “The lack of an independent bioethicsboard will have serious repercussions for howIreland is seen as a hub for medical research,and that will have to be addressed as a matterof urgency,” says scientific director StephenSullivan of the newly formed Irish Stem CellFoundation, which is calling on the governmentto reinstate the council. Cormac SheridanAmgen trumps RocheA 5-year patent dispute between Roche andAmgen over the anti-anemia drug Mircera(methoxy polyethylene glycolepoetin beta) hasended. Roche of Basel acknowledged in courtthat Mircera, its pegylated-erythropoietin,infringed on Amgen’s erythropoietin patent andwould drop its challenges. The ruling ensuresthat Mircera sales are barred and Roche iskept out of the US market until mid-2014,when Amgen’s patents expire. Amgen currentlydominates the US market with erythropoiesisstimulatingagents (ESAs)—Epogen (epoetinalfa) and Aranesp (darbepoetin alfa)—whichtogether generated $5.6 billion in sales lastyear. However, Thousand Oaks, California–basedAmgen may now have to contend with US Foodand Drug Administration (FDA) regulations, asa panel of outside experts expected to meetin 2010 will re-examine safety concerns overESAs (Nat. Biotechnol. 25, 607–608, 2007).Writing in January in the New England Journalof Medicine (doi:10.1056/NEJMp0912328),FDA officials are urging proper dosing of ESAsin individuals with chronic kidney disease, ascertain regimens appear to increase the risk ofcardiovascular events and death. The panel mayimpose regulations on the ESA market or decidethat additional clinical trials are needed. Theoutcome of this meeting, says Eric Schmidt, abiotech analyst at Cowen and Company in NewYork, is that it may bring down sales, as drugcompanies may no longer be allowed to pushhigh-dose regimens.Nazlie LatefiReport blames GM crops for herbicide spike,downplays pesticide reductionsA recent report published by the OrganicCenter, an organic farming advocacy organizationheadquartered in Foster, RhodeIsland, claims that the use of herbicides inweed control has risen sharply since transgeniccrops’ commercial introduction in1996. Increasing cultivation of glyphosate(N-phosphonomethyl glycine)-tolerant transgeniccrops, particularly soybean, has led toan aggregate increase in herbicide use of 383million pounds over the past 13 years, on topof what the Organic Center’s chief scientistCharles Benbrook models suggest would havebeen applied had the technology never beendeployed (http://www.organic-center.org/science.pest.php?action=view&report_id=159).The report also downplayed that transgeniccorn and cotton have delivered reductions ininsecticide use totaling 64.2 million poundsover the same time period.The report’s findings on herbicides are instark contrast to the standard agrochemicalindustry line that transgenic crops have reducedthe chemical load on the environment. Severalcritics have questioned the assumptions underlyingthe analysis and any significance that canbe drawn from it, particularly as the reportcomes from an advocacy group seeking to“communicate the verifiable benefits of organicfarming and products to society.”Rising glyphosate resistance is a plausibleexplanation for the increasing use of herbicides,however. Among plant scientists, thereis little disagreement on the problem of glyphosate-resistantweeds. “It certainly is fairto point out the failure in glyphosate stewardship,that the threat of resistance wasn’tappreciated, that more diverse managementwasn’t used to try to prevent or delay resistanceemerging,” says Chris Boerboom,extension weed scientist at the University ofWisconsin in Madison.The issue of herbicide resistance hasalready become acute in some US states.Report author Benbrook claims that thecotton and soy industries in the Southeastare on “the brink of collapse” because ofthe cost of dealing with glyphosate-resistantweeds. Benbrook goes on to argue thatincreasing reliance on herbicides pairedwith more expensive, engineered tolerancetraits will erode farmers’ profitability, whilecompounding environmental and publichealth risks (through increased chemicalexposure).The report’s other main finding—thatinsect-resistant transgenic crops havehelped cut pesticide use—was downplayedby Benbrook, who claims the increase in thevolume of herbicides applied “swamps” theCrop spraying on the up. Glyphosate-resistant weeds may be driving an increased reliance onherbicide use.Greg Gardnes/istockphoto112 volume 28 number 2 february 2010 nature biotechnology

news© 2010 Nature America, Inc. All rights reserved.benefits of decreased insecticide use attributableto corn and cotton expressing genes thatencode one or more Bacillus thuringiensis (Bt)insect toxins. Bt crops could have a brighterfuture than herbicide-resistant transgenicvarieties, the report states, “but if, and onlyif, [insect] resistance is prevented.”The report is based on extrapolations ofpesticide use survey data compiled by theUS Department of Agriculture’s (USDA)National Agricultural Statistics Service(NASS). Benbrook relies on annual traitacreage data compiled by St. Louis–basedMonsanto to disaggregate transgenic cropsfrom the total crop acreage. However, noNASS data on corn or soy are available for2007 or 2008, years for which Benbrookposits unusually large pesticide increasesof 20% and 27%, respectively. The mainuncertainties stem from gaps in NASS data,which, since 2001, have only been gatheredintermittently, and from that data’s failure todistinguish between pesticide use on transgeniccrop varieties and on their conventionalcounterparts.Benbrook postulates the emergence of glyphosateresistance has fueled a sharp upswingin the use of other herbicides on glyphosatetolerantcrops, whereas levels of herbicide usedon conventional crops have fallen becauseof ongoing improvements in potency. ButJanet Carpenter, formerly of the Washington,DC–based US National Center for Food andAgricultural Policy and now an independentagricultural biotech consultant, disagrees.“That’s all extrapolation,” she says. “The bottomline is we don’t know what has happenedto pesticide use in the last couple of years.”Benbrook says that additional data fromfuture surveys can be factored into his modelwhen it becomes available. “The valid criticism—orvalid question—is these are allaverage numbers,” he says. “I would place afair amount of confidence in these averagesas a reflection of what’s going on out there.”In a published critique of the report,Dorchester, UK–based consultancy PGEconomics argues that Benbrook overestimatesherbicide application rates for biotechcrops and underestimates them for conventionalcrops (http://www.pgeconomics.co.uk/pdf/OCreportcritiqueNov2009.pdf). It citesa new study from the US Geological Survey,which found that concentrations of severalmajor pesticides either declined or remainedconstant in US corn belt rivers and streamsduring 1996–2006 (http://pubs.usgs.gov/sir/2009/5132/). However, the study perioddoes not include the two most recent years,during which Benbrook claims the greatestincrease in herbicide use has occurred. PGEconomics, which also published a lengthystudy on the global socioeconomic andenvironmental impacts of transgenic cropsin May last year, has drawn on two sources:pesticide use data from a commercial source,DMR Kynetec, of St. Louis, which Benbrooksays is in general agreement with his ownfindings; and what he describes as ‘faulty’simulation data generated by the WashingtonDC–based National Center for Food andAgricultural Policy, based on exercises runwith university extension weed scientists.“It’s impossible to reconcile their estimateswith the NASS data,” Benbrook says.In the meantime, several scientists havevoiced support for the general thrust of thestudy. “There’s nothing surprising there,”says Matt Liebman, who holds the H.A.Wallace chair for Sustainable Agriculture atIowa State University in Ames. Dealing withglyphosate-resistant weeds will require alterationsin cropping systems that rely solelyon the marriage of the herbicide-tolerancetrait and the associated herbicide to controlweeds. Widespread convergence on a narrowrange of options, such as the rotationof glyphosate-resistant corn and glyphosateresistantsoybean, has been a significant factor,says Liebman. “You have good conditionsfor rapid selection of herbicide resistance.”Monsanto and its competitors are respondingto the problem by offering farmers subsidiesto include third-party herbicides intheir weed control systems. They are alsostacking additional tolerance traits that canbe paired with other herbicides, such asdicamba (3,6-dichloro-2-methoxybenzoicacid), glufosinate (phosphinothricin) and2,4-d (dichlorophenoxyacetic acid). Externalfactors have hampered progress, however.“The biggest contributor to weed resistancehas been the European Union’s slow approvalprocess for new biotech-enhanced seeds.After many years of delays, the EU finallygranted approval of Liberty Link [phosphinothricin-acetyltransferase]soybeans, whichare resistant to a different active ingredient[l-phosphinothricin],” says Bob Callanan,communications director of the AmericanSoybean Association, located in St. Louis.Critics argue that more diversifiedapproaches will be needed, such as alternativecrop rotations, novel herbicides—it’s20 years since a new mechanism of actionwas commercialized, notes Boerboom—andalternative weed control methods. “If youwant to keep this tool available and effectivethere has to be some way, short of fallowinga field, of delaying the developmentof resistant weeds,” says Robert Kremer, ofthe USDA’s Agricultural Research Service atColumbia, Missouri. The market dominanceof transgenic crop varieties limits some ofthe options, however. “It’s very difficult togo and find nontransgenic soybean,” he says.“Conventional corn rotated with RoundupReady [glyphosate-resistant] soybeans wouldbe very logical,” says Boerboom. “We have anexcellent selection of conventional herbicideswe can use in corn.”That Monsanto’s Roundup Ready croppingsystem has been a major hit with farmersis not in dispute. “The simplicity, the highefficacy and the perceived low cost have beenvery attractive,” says Liebman. In this respect,even Benbrook agrees: “The weed managementsystems that Roundup Ready [crops]replaced were unforgiving and required ahigh level of skill and management to get thebenefits out of them,” he says. What’s more,glyphosate, which inhibits 5-enolpyruvylshikimicacid-3-phosphate synthase, a plantenzyme involved in amino acid biosynthesis(the engineered trait comprises a bacterialform of the enzyme, which is unaffected),has a relatively benign environmental profilein comparison with many other herbicides.Moreover, it has allowed many crop growersto shift to no-till agriculture, which reducesfossil fuel inputs.The problems farmers are encounteringnow are not new. “The selection forglyphosate resistance is not unique. We’veselected for a whole lot of other herbicidefamilies as well,” says Aaron Hager, weedscience extension specialist at the Universityof Illinois at Urbana-Champaign, Illinois.“There’s a plant somewhere in the worldthat’s resistant to an herbicide that hasn’teven been discovered yet. That’s how selectionoccurs.”As glyphosate has played a central role inUS crop production over the past decade, itcan be argued the technology has become avictim of its own success. For many farmers,weed control will, however, soon becomemore complex. Some of the alternativesoffer less favorable environmental profiles.Dicamba, a synthetic auxin or plant hormone,can drift off-site and interfere withflowering plants, for example. “There will beobjections raised to it by the environmentalcommunity because of nontarget effects,”says Liebman.Nevertheless, US agriculture is not facing adoomsday scenario, according to Boerboom.“I don’t think it’s like we’re going into somedark age of chemical use on the landscape,”he says. But a new phase in the moleculararms race between biotech and nature is gettingunderway.Cormac Sheridan Dublinnature biotechnology volume 28 number 2 february 2010 113

© 2010 Nature America, Inc. All rights reserved.NEWSin briefIndustry gains on moneybackschemesRisk-sharing agreements that assess innovativedrugs based on long-term cost effectiveness maynot be helping governments save money, a newstudy suggests. “In the short term, it’s been to[industry’s] advantage,” says lead investigatorMike Boggild, a neurologist at The Walton Centrein Liverpool. In 2002, the UK governmententered a ‘risk-sharing’ agreement over fivemultiple sclerosis drugs that the UK’s NationalInstitute for Health and Clinical Excellence(NICE) had deemed too expensive. NICE reversedits decision after drug makers dropped theirprices and agreed to reimburse government ifthe drugs did not prove cost effective in the longterm. The study results based on two-years’ datasuggest that the drugs are not cost effective,although Boggild warns it is too early to drawfirm conclusions. “The cost effectiveness of thedrugs can go in either direction, depending onwhich assumptions we use,” he says. This typeof scheme is inherently difficult to run, adds JonNicholl, director of the Medical Care ResearchUnit at Sheffield University, UK, becausestakeholders have conflicting interests: the statewants to reduce costs, whereas industry wants tomaximize profits. A different approach, in whichfirms refund treatment costs for nonresponsivepatients, may be a better way to improve costeffectiveness, he says.Asher Mullard$2 million rice verdictagainst BayerA St. Louis district court has ordered BayerCropScience to pay over $2 million incompensatory damages to two Misssouribasedrice farmers whose crops cross-bredwith the company’s genetically modified (GM)LibertyLink during field testing. When theunwanted presence of transgenic rice wasdiscovered in 2006, several countries haltedUS rice imports, which led to farmers’ economicloss and prompted more than 1,000 similarlawsuits against Bayer CropScience, whose USoperations are based in The Research TrianglePark, North Carolina. This first trial, whoseverdict was issued last December, has beencalled a bellwether case. “We are studying thecourt’s award of monetary damages in detailand are considering our options,” says RichardBreum, corporate spokesperson for BayerCropScience in Monheim, Germany. “Since eachcase is different, we evaluate each separately.Last year the court ruled against the plaintiffsin their efforts to obtain class action statusin the litigation, noting overall differences inindividual plaintiff’s situations and claims.” In2007 the US Department of Agriculture (USDA)decided against pursuing enforcement actionagainst the company. It noted that investigatorswithin the Animal and Plant Health InspectionService (APHIS) at USDA were “unable tomake any definitive determinations” about theinadvertent release, during field trials, of twovarieties of LibertyLink rice that then mixed withcommercial rice crops in Missouri and severalneighboring states.Jeffrey L FoxBiorefineries’ stimulus winNineteen start-ups have landed the bulk of federal stimulus funding earmarked for industrialbiofuel and biomass programs. The US Department of Energy (DOE) in December announced$564 million in funding towards the building and operating of facilities that convert nextgenerationfeedstocks such as switchgrass and wood chips into fuels and products. Grantsrange from $2.5 million to $81.1 million each (Table 1), which dwarf funds allocated torelated areas such as plant genomics research. Small-scale or pilot facilities will receive up to$25 million, demonstration scale $50 million, and one company, Bluefire Ethanol in Irvine,California, more than $81.1 million to build a commercial plant. Amyris Biotechnologies, forexample, will add its $25 million to the $165 million investment money it has accrued overthe last 7 years. The Emeryville, California–based company will use the stimulus grant toexpand its pilot facility, explore feedstocks for making renewable hydrocarbons and scale-upproduction of both fuel and biobased chemicals, says Kinkead Reiling, cofounder. But themoney is not intended to cover all biorefinery building costs—the DOE expects grant winnerscollectively to match prize funds with at least $700 million in nonfederal investment. “[Thegrants] can boost investor confidence in those projects and allow companies to attract the fullamount of the funding needed to get the project done,” says Paul Winters, a spokesperson forthe Biotechnology Industry Organization in Washington, DC. Adds Reiling, “It’s an excellentshot in the arm for the industry, but compared to the size of the problem [energy crisis],it’s small.” The stimulus bill, known as the American Recovery and Reinvestment Act, waspassed in February 2009.Emily WaltzTable 1 Selected biofuel companies receiving US stimulus fundsCompany /locationGrant($ million) Project descriptionBluefire a /California 81.1 To construct a facility that produces ethanol fuel from woodybiomass, mill residue and sorted municipal solid waste. Thefacility will have the capacity to produce 19 million gallons ofethanol per year and will be in Fulton, Mississippi.BioEnergy International b /Lake Providence, LouisianaEnerkem b /Pontotoc, MississippiINEOS New Planet BioEnergy b /Vero Beach, FloridaSapphire Energy b /Columbus, New MexicoAlgenol Biofuels c /Freeport, TexasUOP c /Kapolei, HawaiiZeaChem c /Boardman, OregonHALDOR TOPSOE c/Des Plaines, IllinoisICM c /St. Joseph, Montana50.0 To produce succinic acid from sorghum. The biological processbeing developed displaces petroleum-based feedstocks anduses less energy per ton of succinic acid produced than itspetroleum counterpart.50.0 Located at an existing landfill, this project will use feedstockssuch as woody biomass and biomass removed from municipalsolid waste to produce ethanol and other green chemicalsthrough gasification and catalytic processes.50.0 This project will cultivate algae in ponds that will ultimately beconverted into green fuels, such as jet fuel and diesel, usingthe Dynamic Fuels refining process.50.0 To cultivate algae in ponds that will ultimately be convertedinto green fuels, such as jet fuel and diesel, using the DynamicFuels refining process.25.0 To produce ethanol directly from carbon dioxide and seawaterusing algae. The facility will have the capacity to produce100,000 gallons of fuel-grade ethanol per year.25.0 To integrate existing technology from Wilmington, Delaware–based biofuels firm Ensyn and UOP to produce greengasoline, diesel and jet fuel from agricultural residue, woodybiomass, dedicated energy crops and algae.25.0 To use purpose-grown hybrid poplar trees to produce fuelgradeethanol using hybrid technology. Additional feedstockssuch as agricultural residues and energy crops will also beevaluated in the pilot plant.25.0 To convert wood to green gasoline by fully integrating andoptimizing a multi‐step gasification process. The pilot plantwill have the capacity to process 21 metric tons of feedstockper day.25.0 To modify an existing corn‐ethanol facility to produce cellulosicethanol from switchgrass and energy sorghum usingbiochemical conversion processes.Amyris Biotechnologies c 25.0 To produce a diesel substitute through the fermentation of sweetsorghum. The pilot plant will also have the capacity to coproducelubricants, polymers and other petrochemical substitutes.a Increased funding to existing biorefinery projects. b Demonstration scale. c Pilot scale. Source: US Department of Energy114 volume 28 number 2 february 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.Purpose-built chromosome“It’s functional, and alsoa very good metaphor forwhat the center is tryingto achieve.” Larry Malcic,one of the architects ofLondon’s UK Centre forMedical Research andInnovation (UKCMRI),says scientists exclaimed,“that’s a chromosome,”when he presented thebuilding designs withoutknowing its symbolicsignificance. The new$978 million UKCMRI isbeing built in central London as a partnership between University College London, CancerResearch UK, the Medical Research Council and the Wellcome Trust. It will house fourleading science organizations to conduct biomedical research on genetics, stem cells andcommon diseases, and is expected to open in 2015. (Times, December 8, 2009)in their words“They just wait untilWHO [World HealthOrganization] says‘pandemic’ and activatethe contracts.” WolfgangWodarg, a memberof the German SocialDemocratic Party andchair of PACE healthcommittee, convenientlyshifts blame for Germany’ssurplus H1N1 vaccine stocks on to the companiesthat redirected resources and expertise to make aproduct available in just a few months. (PharmaTimes, January 4, 2010)“These sweetheart deals are being done on thebacks of consumers. From the perspective of theFederal Trade Commission, [they] are one of theworst abuses across the board in healthcare andSelected research collaborationsshould be stopped.” Federal Trade Commission(New York) chairman Jon Leibowitz will press for aprovision in the healthcare reform bill to end dealsin which brand-name drugmakers pay genericproducers to delay copycat versions of best-sellingmeds. (New York Times, January 12, 2010)“The pharmaceutical industry has destroyedso much institutional knowledge over the lastdecade that it makes the Taliban, blowing uptemples, look like high school pranksters.”Anonymous blogger. (In the Pipeline,January 12, 2010)“Cannibalism is rife within the biotech industry!”Barry Canton, a cofounder of Ginkgo Bioworks(Boston), on how his and other companies areacquiring equipment castoffs from universitiesand other companies from online auctioneers.(The Boston Globe, January 4, 2010)newsin briefFDA balks on MedImmune’scell-grown flu vaccineThe shift towards new cell culture–based fluvaccine production has been dealt a blowas MedImmune of Gaithersburg, Maryland,puts its manufacturing efforts on hold. TheAstraZeneca subsidiary took this step afterthe US Food and Drug Administration (FDA)requested follow-on studies that wouldsubstantially increase the cost and time tomarket beyond what the company expected.In its contract with the Department of Healthand Human Services (HHS), MedImmuneproposed an efficacy study comparing immuneresponses in volunteers receiving cellproducedwith those receiving egg-producedvaccines, considering them geneticallyidentical, followed by a large safety trial. Butthe FDA termed cell-grown vaccine a newproduct, requesting Medimmune conduct aclinical trial during an influenza season, aswell as demonstrate efficacy in adults beforevaccinating children. The plan “becamecumbersome and complicated and did notaddress significant scientific and medicalissues we thought we needed to address toadvance this vaccine,” says George Kemble,vice president of vaccine R&D at MedImmune.“I don’t think there is any deliberate delay,”says Anthony Fauci, director of the NationalInstitute of Allergy and Infectious Diseases,noting the move is due to safety and efficacydata gathering. Jose Romero, member of theFDA vaccine advisory committee, comments inan unofficial capacity, “General FDA concernsinclude exposing humans to adventitiousagents that might be lurking in cell linesor the remote possibility of transmitting anoncogene that could create cancer in a humanhost.” Elsewhere, last November, Novartis ofBasel inaugurated a $1 billion cell culture fluvaccine manufacturing facility in partnershipwith the HHS. The plant in Holly Springs,North Carolina, is the first large-scale cellculture flu vaccine and adjuvant productionfacility in the US.Wendy WolfsonPartner 1 Partner 2Alopexx Pharmaceuticals(Cambridge,Massachusetts)Seattle Genetics(Bothell, Washington)Athersys (Cleveland)Sanofi-Aventis(Paris)Millennium/Takeda(Osaka, Japan)Pfizer(New York)Syngenta (Basel) CSR Sugar(Melbourne,Australia)*Financial details not disclosed.$(millions) Details375 Sanofi-Aventis will pay Alopexx for rights to codevelop a monoclonal antibody (mAb) for treating Escherichia coli,Staphylococcus aureus and other infections. Alopexx receives an upfront payment, research funding and is eligiblefor milestone payments that could reach $375 million in total, plus royalties. Sanofi will have the option tolicense the product, which will be in phase 1 trials in 2010.290 Millennium will pay $60 million upfront, plus milestones that could exceed $230 million, to codevelop SeattleGenetics’ brentuximab vedotin (SGN-35). The antibody drug conjugate composed of an anti-CD30 mAb andmonomethyl auristatin E is currently in a pivotal phase 2 trial to treat relapsed and refractory Hodgkin’s lymphoma.Under the agreement, the Takeda Group keeps commercial rights to the drug outside the US and Canadawhere Seattle Genetics retains full rights.111 Pfizer will pay Athersys $6 million initially and up to $105 million in the future for rights to develop Athersys’sstem cells to treat ulcerative colitis and Crohn’s disease. The product, MultiStem, consists of multipotent adultprogenitor cell, and is in early clinical trials for heart attacks and in cancer patients receiving bone marrow transplants.* Syngenta has acquired exclusive global rights, excluding Australia, to CSR Sugar’s SugarBooster, a transgenic technologyto develop cane plants with high sugar content. The license agreement includes milestone payments and royaltieson product sales to CSR Sugar. The terms of the deal were not disclosed.nature biotechnology volume 28 number 2 february 2010 115

© 2010 Nature America, Inc. All rights reserved.data page2009: Turning the cornerWalter YangDespite the shaky start to 2009, the biotech sector regained its financialfooting. Biotech indices were up, as were offerings and partnership monies.Excluding collaborations, the sector raised a total of $24.3 billion.Stock market performanceAlthough biotech indices were up ~16% last year, they underperformedother major indices.Index1,5001,4001,3001,2001,1001,00090080070012/081/09Swiss MarketNASDAQ BiotechNASDAQ2/093/094/095/09S&P 500Dow JonesBioCentury 1006/09Month endingGlobal biotech venture capital (VC) investmentVenture money fell slightly last year, as private companies raised$5.1 billion versus $5.3 billion in 2008.VC amount raised ($ millions)1078,000 1,436Asia-Pacific7,00069 78 58 5,24155Europe1,213 1,451 1,206531,1456,00040 4,045 3,871 4,3151,1304,070 3,947 Americas5,000 7973,1514,0003,0002,0001,00002003AmericasEuropeAsia-Pacific7/098/092004 2005 2006 2007 2008Year2003 2004 2005 2006 2007 2008 2009184775190826182959204818223105921210161968749/092009Table indicates number of VC investments and includes rounds where the amountraised was not disclosed. Source: BCIQ: BioCentury Online IntelligenceNotable 2009 dealsIPOsPercentCompany (lead underwriters)Amountraised($ millions)change instock pricesince offerDatecompletedTalecris (Morgan Stanley, Goldman Sachs, Citigroup, $550.0 17% 30-SepJPMorgan)Movetis (Credit Suisse, KBC) $146.0 3% 3-DecCumberland (UBS, Jefferies, Wells Fargo) $85.0 –5% 10-AugOmeros (Deutsche Bank) $68.2 –30% 7-OctChina Nuokang (Jefferies) $40.7 –13% 9-DecVenture capitalAmountCompany (lead investors)raised($ millions)Roundnumber Date closedClovis Oncology a (Domain, New Enterprise Associates, $145.0 NA 21-MayVersant, Aberdare, Abingworth, Frazier, ProQuest,company management)Zogenix (Clarus, Domain) $71.0 2 07-DecBioVex (Forbion, Morningside, Ventech, MVM) $70.0 6 10-NovPacific Biosciences b (Deerfield, Intel) $68.0 5 12-AugHyperion (Bay City Capital, Panorama Capital) $60.0 3 30-JunNovImmune (BZ Bank) $56.4 NA 12-MaySopherion (Zoticon Bioventures) $55.0 3 18-Feba Lead investor not available. b Series E extension.10/0911/0912/09Although initial public offerings showed signs of resuscitation (at least13 more companies are now in the queue), follow-on financings came inabove $6 billion—the second-best year over the past decade.Global biotech industry financingThe boost in partnership promises to US biotechs and follow-on financingspushed industry funding to $61.3 billion, up 82% from 2008.Year20092008200720062005200420.0 3.2 5.3 3.1 1.9 0.117.3 6.1 5.4 2.7 4.8 1.910.9 8.8 5.3 2.9 3.3 2.620038.9 9.1 4.0 2.2 3.9 0.50 10 20 30 40 50 60 70Amount raised ($ billions)36.9 10.0 5.1 2.2 6.0 0.922.4 11.7 6.8 4.7 4.4 3.019.8 11.9 5.6 4.7 5.6 2.0PartneringDebt and otherVenture capitalPIPEsFollow-onsIPOsPartnership figures are for deals involving a US company. Source: BCIQ: BioCenturyOnline Intelligence, Burrill & Co.PIPEs, private investments in public equity; IPOs, initial public offeringsGlobal biotech initial public offerings (IPOs)The North American IPO market showed signs of life, with four companiesraising $705 million versus only one raising $6 million in 2008.Amount raised ($ millions)5853,5001,055 Asia-Pacific2301,3093,000474Europe1,852982,50015 869930Americas1,0632,0009131,5001,000500043194832003AmericasEuropeAsia-Pacific1211562004 2005 2006 2007 2008Year2003 2004 2005 2006 2007 2008 200981535126182432621323217132433651587052009Table indicates number of IPOs. Source: BCIQ: BioCentury Online IntelligenceMergers and acquisitionsTarget Acquirer Value ($ millions) Date announcedGenentech Roche $46,800 12-MarSepracor Dainippon Sumitomo $2,600 3-SepMedarex Bristol-Myers Squibb $2,400 22-JulCV Therapeutics Gilead Sciences $1,400 12-MarCougar Biotechnology Johnson & Johnson $1,000 21-MayOvation Pharmaceuticals H. Lundbeck $900 09-FebProteolix Onyx Pharmaceuticals $851 12-OctLicensing /collaborationResearcher InvestorValue($ millions) Deal descriptionPTCTherapeuticsRoche $1,924 Develop small molecules against four central nervoussystem disease targetsNektar AstraZeneca $1,505 Worldwide rights to NKTR-118 for opioid-induced constipationand NKTR-119 for pain without constipationIncyte Novartis $1,310 Ex-US rights to INCB18424; in phase 3 for myelofibrosis;worldwide rights to preclinical INCB28060Targacept AstraZeneca $1,240 Worldwide rights to develop and commercialize majordepressive disorder compound TC-5214Exelixis Sanofiaventis>$1,161 Exclusive, worldwide rights to XL147 and XL765 inphase 1b/2 to treat cancerZymoGenetics Bristol-MyersSquibb$1,105 Codevelop and commercialize ZymoGenetics’ phase 1hepatitis C virus compound PEG-interferon lambdaAmylin Takeda >$1,075 Codevelop and commercialize therapeutics for obesityand related indicationsAlder Bristol-MyersSquibb$1,069 Worldwide rights to ALD518 for all indications, exceptcancer116 volume 28 number 2 february 2010 nature biotechnology

news feature© 2010 Nature America, Inc. All rights reserved.The HER2 testing conundrumProblems in interpreting diagnostic tests for HER2 may becompromising patient access to effective treatments. As newversions of therapies targeting HER2 work their way throughclinical trials, will the situation get even murkier? Malorye Allisoninvestigates.A recent study from the University of California,San Francisco, reveals that one in five HER2tests gives the wrong answer 1 . Furthermore, thearticle, which reviews the medical literature,reports that as many as two-thirds of breastcancer patients who should be tested for HER2are not, and consequently a significant fractionof women treated with Genentech’s Herceptin(trastuzumab) have never been tested for HER2overexpression.The health benefit provider Wellpoint,of Indianapolis, might dispute that finding.According to Genentech staff scientist MarkSliwkowski, the insurer has data showing that98% of its breast cancer patients are tested.However, doctors differ in their views on testingbefore prescribing Herceptin. “Some doctorsdon’t know how to interpret test results,Table 1 Selected HER2 testsCompanyLocationBiogenexSan Ramon, CaliforniaDakoGlostrup, DenmarkDakoGenomic HealthInvitrogenCarlsbad, CaliforniaMonogram BiosciencesSiemens Healthcare DiagnosticsErlangen, GermanyVentana-RocheTucsonVentana-RocheVysis (Abbott)Name of testStatusInSite HER2/neu CB11FDA approvedHER2 FISH pharmDx KitFDA approvedHercepTestFDA approvedOncotype DXCLIA validatedSPOT-Light HER2 CISH KitFDA approvedHERmark Breast Cancer AssayCLIA-validatedHER2/neu ELISAFDA approvedthey prefer just to prescribe it and assess thepatient’s progress,” says Michael Liebman ofthe patient stratification company StrategicMedicine of Kennett Square, Pennsylvania.More than a decade after the drug receivedUS Food and Drug Administration (FDA)approval, the personalized medicine paradigmclearly has holes. Many experts are frustratedand troubled by the state of HER2 testing, especiallyas new opportunities for tests are on thehorizon. And as trials testing Herceptin at earlierstages and in combination with other drugscontinue, experts are starting to wonder whatbesides HER2 overexpression might be influencingan individual’s response to the drug.These questions promise to not only spur thedevelopment of a range of new tests to guidebreast cancer therapy but also fundamentallyInform HER2 Silver in situ HybridizationApproved in Europe and elsewhere but notby FDAPathway anti-HER2/neu (Clone CB11)FDA approvedPathVysion HER2 DNA Probe KitFDA approvedCLIA, Clinical Laboratory Improvement Amendment; ELISA, enzyme-linked immunosorbent assay.change understanding of this disease, lead tonew treatments and potentially have an impacton treatment of other cancers.Testing tempestPersonalized medicine proponents point toHerceptin as a paradigm changer: the monoclonalantibody targeting HER2 (also referredto as HER2/neu and ERBB2) evens the playingfield for breast cancer patients overexpressingHER2, whose tumors are typically more aggressive.But testing was problematic from thestart, due to either sloppy execution or complextumor biology. “Giving Herceptin earlyimproves outcome so dramatically that it is anabsolute tragedy to miss patients who shouldbe getting it,” says Jeffrey Ross, from AlbanyMedical College in Albany, New York, whohelped develop a fluorescent in situ hybridization(FISH) HER2 test marketed by DownersGrove, Illinois–based Vysis (Table 1). FISH testsare currently considered the gold standard.As more data become available and theHER2 story evolves, it’s becoming clear thatsome pieces don’t fit together quite as wellas they might. For example, patients whosetumors have progressed on the drug, sometimesrespond to Herceptin when it is givenlater with chemotherapy. Furthermore, fewerthan 50% of HER2-positive metastatic breastTechnologyImmunohistochemistry assay using a monoclonal antibody directed against the internaldomain of HER2/neu available either in automated or manual formatsFISH assay to determine HER2 gene amplification in formalin-fixed, paraffin-embeddedbreast cancer specimens. Gene amplification is determined from the ratio between thenumber of signals from the hybridization of the HER2 gene probe and the number ofsignals from the hybridization of the reference chromosome 17 probe (green signals)Semi-quantitative immunohistochemistry assay for determination of HER2 proteinoverexpression in breast cancer tissues routinely processed for histological evaluationRT PCR–based assay analyzes the expression of a panel of 21 genes, among themHER2. Oncotype DX predicts disease recurrence and assesses benefit from certaintypes of chemotherapyChromogenic in situ hybridization (CISH) using a DNA probe. Quantifiable results arevisualized under a standard brightfield microscope.Proximity-based assay, which provides direct quantitative measurements of HER2total protein and HER2 homodimer levelsSandwich enzyme immunoassay using mouse monoclonal for capture and a differentbiotinylated mouse monoclonal antibody for the detection of human HER2/neu protein.Detection is by direct chemiluminescence. Protein is quantified by spectrophotometryFully automated silver in situ hybridization assay for HER2 and chromosome 17detection. Chromogenic signals are detected through the use of silver depositiontechnology. Results and morphological significance can be interpreted using conventionalbrightfield microscopySemiquantitative immunohistochemistry assay using a monoclonal antibody for thedetection of c-erbB-2 (HER2) antigen using Ventana’s family of automated instrumentplatformsFluorescence in situ hybridization (FISH) assay to determine HER2 amplification,using LSI HER2 probe, which spans HER2, and CEP 17 probe, which hybridizes tothe alpha satellite DNA located at the centromere of chromosomenature biotechnology volume 28 number 2 february 2010 117

NEWS feature© 2010 Nature America, Inc. All rights reserved.tumors respond to Herceptin alone 2 . “It’s notjust a simple translation of gene overamplificationto susceptibility to the drug,” says LarryNorton, of Memorial Sloan-Kettering CancerCenter in New York.Others wonder if the test is even necessary.Studies from two clinical trials (NSABP B31and NCCTG N9831) presented at the AmericanSociety of Clinical Oncologists (ASCO) annualmeeting in 2007 (ref. 3) suggested that someindividuals with HER2-negative tumorscan benefit from Herceptin. In these trials,which comparedchemotherapyalone with chemotherapyplus CentromereaHerceptin, onlywomen who wereHER2 positivecould participate.Upon retesting,however, sometumor samplescame up negative.Nonetheless,some womenwith negativetest results benefitedfrom thedrug, which hasspurred a nowl ong-runningdebate. Althoughmany expertsbelieve this findingto be an artifactof variationin test accuracy,others think thismay be anotherimportant clue.“It’s easy todismiss a findingyou can’texplain, but thisbis forcing us to reexamine our notions of whatbeing HER2 positive or negative means,” saysNorton.The controversy around these particular trialfindings may be resolved soon. Samples fromthe NCCTG N9831 trial are being retested in around-robin fashion by three different groups.Results will then be sent to a central monitoringgroup to identify any discrepancies and totry to pinpoint their cause. Soonmyung Paikof the National Surgical Adjuvant Breast andBowel Project (a National Cancer Institute(NCI)-sponsored cooperative based at theUniversity of Pittsburgh) postulates that individualswith HER2-negative primary tumorsmay have circulating tumor cells that are HER2Her2 (ERBB2)~190 kbLSI HER2positive, but he admits that the discrepancycould just reflect problems with the tests. Hisgroup is doing further microarray analysis ofthe NSABP trial samples. If the earlier findingsare confirmed, a trial could be launched in thesummer of 2010 to test Herceptin in patientswith HER2-negative tumors. “We have an NCIapprovedprotocol,” Paik writes in an e-mail.Abundance of richesFrom the time Herceptin was launched, expertshave warned that existing tests have problems.The most commonlyused test,an immunohis-Telomere tochemistry-17q11.2–q12 regionbased assay,happens to bethe least dependable,especiallywhen performedin laboratoriesthat do onlyoccasional tests.The immunohistochemistrytestmeasures proteinlevels, whereasthe newer FISHbasedtests measuresgene copynumber andare believed tobe more reliable,especiallyin expert hands.Some thinkthe FISH assayshould be thestandard, butclearly, it makesa big differencewho is doing thetesting. In 2006,ASCO and theCollege of American Pathologists releasedstricter guidelines which, according to Ross,forced many laboratories that had low test volumesto send the samples to laboratories withhigher volumes and more experience. But someexperts were dismayed that the new guidelinesdid not recommend FISH over immunohistochemistry.Others, including Norton, are skeptical of allavailable tests. The FISH probe, he points out,is large (190 kilobases in the case of the Vysisprobe), spanning the gene and then some (Fig. 1).“When you see changes in HER2 at the genecopy number level, is that a reflection of HER2itself or of generalized genomic instability?” heasks. Genomic instability, he points out, is notFigure 1 FISHing for HER2. (a) Probe map shows therelative size of the Vysis LSI HER2 probe and the gene.(b) An example of a FISH test for HER2 amplificationshows multiple copies of the HER2 gene (red clusterssignals) compared to chromosome number (green signals).Source: Abbott.a random event and the region where HER2is found is a ‘hot spot’. Norton thinks that theprescribing of HER2-targeting drugs won’t beimproved until we understand how specificmutations influence a tumor’s susceptibilityand sequencing can be routinely done onbiopsies. Comparative genome hybridization(CGH) studies done in his laboratory suggestthat simple ‘amplified’ and ‘nonamplified’ readingsavailable from FISH do not adequatelyreflect the complex changes that can occur inthis region. For example, CGH studies revealedthat in some samples that were HER2 positiveon FISH, the amplified area was actually adjacentto HER2 and not within it 4 . It may endup that higher resolution methods like CGHare needed to get the right information abouta tumor’s status.Meanwhile, companies like Genomic Healthin Redwood City, California, and Labcorp’sMonogram Sciences of S. San Francisco,California, are jumping in with new approachesto testing breast cancer patients. GenomicHealth claims that an advantage of its quantitativePCR-based test, OncotypeDX, is its accuracy.“Our test is more than 95% concordantwith reference labs’ assessment by FISH,” saysSteve Shak, chief medical officer at GenomicHealth. HER2 is one of 21 genes included inOncotypeDX, which is used to quantify therisk of recurrence of early breast cancer and theresponse to particular types of chemotherapy.Starting in 2008, the company began includingestrogen receptor, progesterone receptor andHER2 status in every report it provides.Monogram’s HERMark test measures HER2total protein as well as functional homodimersin a dual-antibody format. The company claimsthe test has advantages over FISH because it is adirect measurement of the protein, and that itis seven to ten times more sensitive than immunohistochemistrytesting. Albany MedicalCollege’s Ross counters, “Monogram has noprospective randomized data to support thattheir test is better.”Neither company’s test is FDA approvedfor use with Herceptin. For now, the tests’use may be confined to confirming or clarifyingresults obtained using other tests. “Weneed not just technical accuracy but to knowif these tests are actually clinically relevant,”says Edith Perez, of the Mayo Clinic Floridain Jacksonville. But Genomic Health seemsoptimistic. “We have extremely positivefeedback on the value of being able to lookat that result, especially in those cases whenthe results for HER2 testing are uncertain,”says Shak. The company is doing additionalstudies of the test’s ability to predict whetherparticular individuals will benefit fromHER2-targeted therapy.118 volume 28 number 2 february 2010 nature biotechnology

news feature© 2010 Nature America, Inc. All rights reserved.Others wonder whether HER2 is eventhe right thing to test. Strategic Medicine’sLiebman, who was at Vysis (now part of AbbottLaboratories of Abbott Park, Illinois) whenthe first FISH test was developed, says that forcertain patients, the immunohistochemistryand FISH results never agree. “Just becauseyou have a change in gene copy number, thatdoesn’t mean it’s expressed,” he says. The factthat a significant fraction of HER2-positivepatients with metastatic disease fail to respondto the drug also suggests that it is not a causalmarker but a surrogate.There may be room for yet more tests.Microarray studies indicate that between 20to 30 distinct classes of breast cancer exist,according to Charles Perou, of the LinebergerComprehensive Cancer Center in Chapel Hill,North Carolina. A few of those make up themost clinically relevant subtypes, but there areenough differences between those types thatmore and better tests are urgently needed.“Our array data show that there are at leasttwo kinds, and maybe many more, of patientswith HER2-positive disease,” he says. The differencecan be seen in how the patients respondto chemotherapy, with 80% responding in onegroup and only 30% in the other.Son of HerceptinOne other HER2-targeting drug, the smallmoleculeTykerb (lapatinib from London-basedGlaxoSmithKline), is now approved for use inbreast cancer, but a bevy of next-generationversions of Herceptin and new combinationswith the drug are nearing the market, potentiallygiving oncologists even more choiceswhen deciding which drug to use and whento use it.Genentech currently has two new HER2-targeting drugs in phase 3 trials. Pertuzumabis a HER2 dimerization inhibitor that binds toa different epitope on HER2 than Herceptin.The drug inhibits HER2 dimer formation withother HER family members, such as HER3and HER1. Genentech is currently studying acombination of both HER2 inhibitors in breastcancer. “With this approach, we are addressingthe question of what happens when youhave more complete HER2/neu blockade,” saysGenentech’s Sliwkowski. The drug has alreadyshown promise in early trials. In one study,about a quarter of women whose disease hadprogressed while they were taking Herceptinhad their tumors shrink by >50% when pertuzumabwas added to their treatment regimen.The company is also optimistic aboutT-DM1 (trastuzumab-DM1), a drug conjugatethat combines Waltham, Massachusetts–basedImmunoGen’s antimitotic maytansinederivativeDM1 cancer-killing agent withHerceptin. In earlier studies, this drug madetumors shrink even in some women withadvanced breast cancer who had been treatedwith a median of seven different drugs. Thedrug is being “moved up the line,” accordingto Sliwkowski, and will be tested in a randomizedphase 2 trial comparing TDM-1 versusHerceptin plus chemotherapy. “We think it isso active that it’s important to try this,” saysSliwkowski. To be able to use a targeted therapywithout chemotherapy is the dream and it maybe very close to realization.Dennis Slamon, a University of California,Los Angeles, oncologist who was part of theteam that developed Herceptin back in the1980s, is enthusiastic about combining it withGenentech’s vascular endothelial growth factor(VEGF) inhibitor, Avastin (bevacizumab).Slamon points out that the two pathways arelinked. When HER2 is amplified, “one of thepathways that consistently goes up is VEGF,”he says. Trials of the combination (Avastin plusHerceptin) have been encouraging. In phase2 trials, “the two antibodies alone, with nochemo, are giving objective response rates in54% of women with metastatic disease.” Usedat earlier stages, Slamon believes the combinationcould be even more powerful and he isoptimistic that this regimen will eventually betested in a phase 3 trial without concomitantchemotherapy.Growing understanding of the pathwaysrelating to HER2 are also leading to new drugtargets. Phosphatidyl inositol 3-phosphate(PI3) kinase and PTEN (phosphatase andtensin homolog) are two other players thatseem connected to HER2. Mutations in PI3kinase occur in 30% of breast cancers andcause the enzyme to be turned on all the time.“It’s a classic oncogenic activating mutation,”says William Sellers, head of oncology researchat Novartis Institutes for Biomedical Researchin Cambridge, Massachusetts. PTEN, meanwhile,is a tumor suppressor gene which canact by blocking the activation of PI3 kinase.Mutations in that gene can again lead to overactivation.“In many instances, the amplificationof HER2 leads to signaling through thispathway,” Sellers says. This has become thenumber one pathway of interest in cancer.The most advanced new compounds targetingthe pathway are mTOR inhibitors, but thesework far downstream, and researchers wouldlike to hit it earlier on. Novartis’ BKM120 is aselective PI3 kinase inhibitor that the companyhopes will do just that. All these targets, includingHER2, are found in other cancers, whichmeans they could have wider use. Herceptin isbeing tested in stomach cancer, for example. Ifthese drugs are used more broadly, new testswill be needed. The question then, Sellers says,will be, “How do we know which therapeuticto use?”Improving outcomesControversy about testing has doggedHerceptin from the beginning. “Slamon battledfor ten years to prove HER2 was important,”recalls Shak, who worked on Herceptin’sdevelopment while at Genentech. “He had tofight that hard because so many groups weredoing their tests without quality control.” Butexperts are adamant that testing for HER2must be improved. “In the adjuvant setting,you are giving a woman absolutely substandardcare if you are wrongly denying her thedrug,” says Ross. Norton concurs: “We mustdo better.”Although standardizing immunohistochemistryseems to be the most obvious next step,it may not necessarily be the best way toimprove outcomes. “Even when countrieswith a national health service do everythingthey can to standardize this test, we still seeunacceptable margins of error,” says Liebman.Strategic Medicine is working with The MayoClinic and Thompson Reuters, headquarteredin New York, to build data models that willreveal which steps are most likely to improvethe quality of HER2 testing. Most laboratoriesare using immunohistochemistry, andthe cost of making improvements across theentire community, he points out, would likelybe prohibitive. “Our goal is to find the weakestpoints in the system, whether it is an issuewith reimbursement, diagnostic developmentor education. What should be the priority fixthat gets us the biggest impact in improvingpatient care?”Others think the answer clearly lies inapplying some newer technologies. “I thinksequencing will give us the final answer, oncewe have inexpensive-enough techniques,” saysNorton.Malorye Allison, Acton, Massachusetts1. Phillips, K.A. et al. Cancer 115, 5166–5174 (2009).2. McArthur, H.L. & Hudis, C. Clin. Cancer Res. 15,6311–6313 (2009).3. Perez, E.A. et al. J. Clin. Oncol. 25, 512, Suppl. 18S(2007).4. McArthur, H.L. et al., abstract 1005, presented atEuropean CanCer Organization 15 and 34th EuropeanSociety for Medical Oncology MultidisciplinaryCongress, Berlin, Sept. 20–24, 2009.nature biotechnology volume 28 number 2 february 2010 119

uilding a businessComing to termsDavid H Oden, Jeffrey A Wolfson & Christina W MarshallBefore taking other people’s money to finance your venture, it pays to fully educate yourself about the strings attached.© 2010 Nature America, Inc. All rights reserved.You’ve found an investor who’s willing tomake a substantial investment in yourbiotech company—that’s great news. Butafter the handshake, the next thing is tonegotiate the term sheet outlining the structureof the transaction to ensure a true meetingof the minds.Term sheets should always be used in complexinvestment transactions—especiallythose involving venture capital investors orother institutional investors. The term sheetsets forth the key terms of the proposedtransaction. A good rule of thumb is that theterm sheet should address any provision thatcould kill the deal.If you skip on drawing up a term sheet,then during the drafting and negotiation ofthe investment documents there may be noclear record of the parties’ understandingson key issues. In the long run, this will causeconfusion and discord, and any subsequentdocuments will probably take more time andcost more to draft and negotiate because theparticipating parties may be unwittinglyusing the definitive documents to negotiate—orrenegotiate—key terms. Worse still,well into the process, it may become apparentthat you are unable to reach agreementon one or more deal-killer terms and thetransaction may collapse (Box 1).In the following article, we guide youthrough the key steps in drawing up a termsheet. Getting this right is important to ensureyou remain in control of your company andreceive your share of returns.David Oden is a partner and Christina Marshallis an associate at Haynes and Boone LLP,Richardson, Texas, USA. Jeff Wolfson is apartner at Haynes and Boone LLP, Washington,DC, USA.e-mail: David.Oden@haynesboone.com,Jeff.Wolfson@haynesboone.com andChristina.Marshall@haynesboone.comBox 1 Potential deal killersDuring negotiations with an investor, you can encounter several hitches. These issues killmore deals than the U.S. Securities and Exchange Commission.• Company technology undervalued by investor(s) or overvalued by founder(s).• Valuation too dependent on issuance of meaningful patent protection.• Partner(s) in joint development arrangements insist on absolute control of patent rights.• Licensing exclusivity in which the partner or licensee in market is not incentivized tocommercialize.• Investor(s) or partner(s) insist on control of bet-the-company litigation.• Founder(s) will lose too much control of the company.• Deal requires clinical milestones that are realistically unreachable.• Future company flexibility is too limited, particularly in partnering and/ordevelopment deals.• Overly cumbersome approval process by investor(s) or partner(s) that could hinder rapidmarket response.• Liability for clinical trials or indemnification in partnering or joint development deals.Before the moneyAlthough some lucky companies areapproached by numerous venture capitalfunds, many have only one investor at atime. The availability and interest of venturecapital often depends on the boom and bustcycles of the biotech industry and the economyas a whole (Box 2). At times, companieshave been lucky to locate a single interestedinvestor, whereas at other times they havehad to fend off multiple investors or limitinvestment. If your transaction is with onlyone investor, it may be a bit simpler, fasterand less expensive, though not by much. Thedownside of having only one investor is thatthere will be fewer pockets to reach into forthe next financing. And, if the sole investordeclines to participate in the next round,you will be in the position of starting fromscratch to attract new ones.If your transaction includes multiple investors,more money and expertise may be availableto you. Additionally, there is a much greaterlikelihood that at least one investor familiar withthe company and its technology will be able toparticipate in subsequent financings. In thiscase, the investors will generally select one tobe the ‘lead’—the party primarily in charge ofdue diligence, negotiations and preparation ofthe definitive investment agreements. Duringdue diligence, the lead investor may examinemultiple aspects of your company, includingthe technical expertise of the founders and keyscientific employees, the market conditions andcompetition, the patent and trademark/brandingpositions of your company and clearanceover any third-party intellectual property (IP)in the space, the R&D pipeline and future patentprotection, the status and estimated cost ofupcoming clinical trials, the status of US Foodand Drug Administration (Rockville, Maryland)interaction and approval, the in-license andout-license agreements the company holds, theagreements with employees and consultantssuch as contract research organizations, andmany other issues.120 volume 28 number 2 FEBRUARY 2010 nature biotechnology

uilding a business© 2010 Nature America, Inc. All rights reserved.That’s a lot to handle, so to ensure a smoothdiligence process with the lead investor, youshould have your legal counsel (preferablyindependent from your regular IP counsel)pre-evaluate the portfolio and IP-related agreementsto help identify and remedy any potentialroadblock issues (like ownership of technology)before seeking investment.The lead investor is usually the one investingthe most money, and that group should bethe main contact for you and your counsel. Inthis situation, a term sheet is absolutely essential,and all investors should participate in thedrafting and negotiation of the term sheet.When all are comfortable with the terms, theother investors should step back and let thelead investor negotiate the rest of the documentsbased on the term sheet.The next step is dealing with the type ofsecurity your investors will be purchasing inreturn for their financing: common stock,preferred stock, a promissory note (normallyconvertible into equity) or some combinationof these (Box 3).But perhaps the most important thing foryou is the valuation the investors assign to yourcompany. Before consummation of the deal, theinvestor and your firm will go through a verydetailed evaluation to determine what portionof the company’s total equity the investor willpurchase. This valuation will involve looking atthe company and its prospects, the values forcomparable companies, the current investmentclimate and the general economic conditions.Valuation is a combination of art and scienceand therefore is open to substantial disagreementand negotiation. It can be particularlydependent on the results of a thorough duediligence investigation.It’s important because the total value willdetermine what percentage of the companythe investor will purchase in exchange for theinvestment. To use a simple example, if theinvestor is investing $1 million in a companywith a pre-money valuation of $1 million, thenthe investor will own 50% of the company afterthe investment (assuming that the company willhave a value of $2 million post-money). If thecompany has a pre-money value of $4 million,then the investor will own 20% of the companypost-money ($1 million being 20% of a $5 millionpost-money value).This determination of value is a key areaof conflict between founders and investors.Not surprisingly, founders usually want ahigher valuation and investors typically seeka lower one.Living with investorsMost founders are familiar with vesting—theconcept that stock options will become exercis-Box 2 Term sheet trendsDuring the past boom for biotech companies (about 9–10 years ago), companies could nothave asked for more advantageous term sheets. At that time, investors were more fearful ofmissing out on a great opportunity than of losing their investment.But the pendulum has inevitably swung back to reflect market conditions, so todayterm sheets tend to be very investor friendly. Biopharma venture capital funding hassubstantially decreased since the boom days, squeezed by conditions in the financialmarkets and, more recently, burned by the global economic downturn. With less biotechventure funding available, companies have had to give up more. Unless your companyis an unusually attractive investment opportunity, do not expect much negotiatingpower at the term sheet stage.able (that is, they will ‘vest’) over time. Vestingis also typical in a venture capital investment,but in a different way: the founder will typicallybe asked to put his or her equity ownership atrisk of being repurchased by the company in theevent that the founder is no longer associatedwith the company for any reason.The rationale behind vesting is that the ventureinvestor is really betting on people (you andyour team) as well as the company and the technology.If you leave, retire, decide to go in a differentdirection or get fired, then you’ll no longerbe in a position to push the company forward.And if you still own a substantial portion of thecompany, this is untenable for your investors.For protection, an investor will typically askyou, the founder, to enter a vesting agreement,whereby all your stock is subject to repurchase bythe company at a nominal price per share (typically,the price originally paid by the founder).The company’s right to repurchase the stock willbe triggered if the founder leaves the companyfor any reason, including the termination ofemployment. This right of repurchase generallydecreases over time, so that at some pointnone of your stock is subject to repurchase. Forexample, in a five-year vesting (which is fairlytypical), the company will have the right (butnot the obligation) to repurchase 100% of thefounder’s stock for the first year after the investment,80% in year two, 60% in year three and soon. After five years, none of the founder’s stockwill be subject to repurchase.As a founder, your risk is the concern overbeing ousted by investors, perhaps to bring ona more business-savvy CEO. This often occurseven if you’re performing well as chief executive.Many founders will seek provisions guaranteeingtheir position for a sufficiently long time,ensuring immediate vesting of rights or otherprotective measures like specifying a reasonablerepurchase price for their stock if involuntarilyor unexpectedly separated from the company.Also up for discussion is the amount of controlinvestors will have over the daily operationsand major decisions of the company. Specifically,particular attention in negotiations should bepaid to whether the investor gains a seat on thecompany’s board, the power the investor has onthe board and the voting rights the investor mayhave as a stockholder.It is fairly normal for an investor to obtain oneor more seats on the company’s board of directorsif the investment is a substantial amount ofmoney and especially if the investor or a designeehas expertise that will be helpful to thefounders. The rationale here is that the investorwants the right to help control the company(and, in turn, try to protect his or her investment)and you want professional assistance inrunning the company.Venture investors specialize in running andgrowing companies—most founders do not. Aventure firm’s presence on the board can reallyhelp those companies that need assistance withbusiness aspects. When properly arranged, thiscan provide founders with a renewed opportunityto focus on what may be their core competency—thetechnology or science.Still, the issue remains of how many boardseats the investor is entitled to and the total sizeof the board. It would be common and expectedthat a large investor would be entitled to at leastone board seat but uncommon to give the investorenough seats to control the board.Investors normally require an agreementwith the company and the other stockholdersregarding the investors’ rights as a stockholder.These voting agreements usually contain provisionspermitting the investor to designate boardmembers and prohibiting the company fromtaking certain actions without the investor’sapproval. Remember to heavily negotiate theseaspects at the term sheet stage of the transactionas they will restrict your ability to run thecompany as you see fit.Exit strategiesBecause an investor’s primary goal is to obtaina substantial return on his or her initial investment,the term sheet will include multipleprovisions focused on how the investor willget the money back—the ‘exit strategy’. Theserights may include a liquidation preference,nature biotechnology volume 28 number 2 FEBRUARY 2010 121

uilding a business© 2010 Nature America, Inc. All rights reserved.redemption of the securities purchased by theinvestor and registration rights.The type of security (Box 3) that the investorwill purchase is directly related to its exit strategy.For example, investors may use a promissorynote to try to protect their investment inthe event that the company is sold or dissolvedby having a ‘liquidation preference’ (liquidationincludes being sold). Essentially, the liquidationpreference says that if the companyis sold or dissolved for whatever reason, theinvestor’s investment (or a multiple thereof)is paid back in full before any funds are paidto other stockholders.This should be of special concern to youbecause it represents an amount of money thatwill be paid out before you, as founder, get onedime of the proceeds. You should try to negotiatethe most narrow liquidation preferencepossible to maximize the amount of money thatwill go to you and other stockholders. Tensionsmay arise only upon liquidation because theliquidation preference can often reveal divergingviews between an investor, who might havelittle incentive to seek additional revenue for thefounders at exit, and the founders, who wouldlike to finally share in a payday after years ofunderappreciated efforts.In the case of a strictly failed biotechcompany (not taken public or acquired, forexample), the investors will typically take anyavailable cash to recover their lost investmentwhen assets are sold off to the highest bidder.The most valuable assets are often the patentrights and in-licensed rights, and they can beaccompanied by trade secret information,such as clinical data from patient trials oreven a Food and Drug Administration drugapproval, as well as real estate, furniture andthe like. Here you could often receive little ornothing due to the liquidation preference, butit may be possible to negotiate around the liquidationpreference and obtain a share of anycash proceeds raised by asset liquidation.Preferred stock that is ‘redeemable’ meansthat the stock must be repurchased by the companyupon the happening of a specified event,such as the passage of time, an insufficient levelof cash, a failed drug trial, poor clinical studyresults, criminal accusations over patient consentor merely at the option of the investor. Thecompany will normally have to purchase thestock back at the investor’s purchase price plusany accrued but unpaid dividends. Redemptionis a feature of preferred stock that is generallydemanded by investors in the current market.Registration rights provide an investor withthe power to register the shares of stock he orBox 3 Defining stockAll types of stock are not equal. The main types of stock that you will encounter fall intothree categories:Common stock. This is the normal type of stock that all companies issue, and the rightsof common stockholders are set forth in the corporation laws of the company’s state offormation. Common stock is usually owned by the founders.Preferred stock. This is usually demanded by most professional investors. Preferred stockis created by amending the company’s certificate of incorporation to include the typeand amount of preferred stock issuable and the rights and privileges of the preferredstockholders. Preferred stock normally has preference over common stock when issuingdividends and distributing assets upon the liquidation or sale of the company. The termsof the preferred stock are typically heavily negotiated and should be discussed in detail inthe term sheet to ensure the parties agree on this fundamental point.Promissory note. This can take the place of stock and is usually convertible to common orpreferred stock upon the occurrence of a certain event (for example, meeting one or morecommercial milestones like successful phase 1, 2 or 3 trials), the passage of time or atthe option of the investor. The terms of the promissory note are also heavily negotiated andshould be addressed in the term sheet. The investor may prefer a promissory note becausein the event of liquidation, noteholders typically recover their investment before anystockholders, even preferred stockholders. Convertible promissory note deals are commonin very early stage investing or in so-called ‘bridge’ financings (short-term loans made inanticipation of subsequent equity financings).she owns during the company’s initial publicoffering (IPO) or after the company hascompleted its IPO. Registered stock is freelytransferable. Even so, it should be noted thatalthough agreements regarding registrationrights are enforceable, the underwriter mayrestrict or eliminate such rights at the timeof an IPO depending on both the respectiveregistration rights of other investors and themarket conditions.ConclusionsRegardless of whether your transaction involvesan investment, an asset purchase, a joint developmentproject or a more complex structure, itis crucial for the parties to enter a term sheet—itwill substantially increase the chances of successfullyclosing a deal. Also, having a written agreementthat outlines the terms of the transactionwill minimize the potential for confusion, costlynegotiation and disagreement between the partiesduring the drafting and negotiation of theinvestment documents.Depending on your need for capital andthe relative attractiveness of your companyto investors, the terms of a financing transactionmay or may not be negotiable. If you donot immediately need funds and the investorfinds your firm attractive, you will have moreleverage negotiating financing terms than ifyou face an immediate cash crisis. Either way,you should pay particular attention to a fewkey terms of the investment. Specifically, tryto negotiate advantageous positions regardingthe percentage of equity the investor will purchasein the transaction, the amount of controlthe investor will have over the company’sdaily operations and major decisions and theamount of money the investor will receiveupon the sale or liquidation of the company.These terms will directly affect the control youand the other founders have over the companypost-investment, as well as your share of theinvestment returns.Money can be hard to find right now,but according to a survey conducted by theUS National Venture Capital Association(Washington, DC) in December 2008,(National Venture Capital Association, 2009Venture Capital Predictions Survey Results,Dec. 17, 2008), the biotech and life sciencesectors are viewed as the second most promisingareas for increasing venture investment.If that’s correct, close scrutiny of term sheetsin biotech ventures is going to become evenmore important than before.To discuss the contents of this article, join the Bioentrepreneur forum on Nature Network:http://network.nature.com/groups/bioentrepreneur/forum/topics122 volume 28 number 2 FEBRUARY 2010 nature biotechnology

correspondenceFab-arm exchange© 2010 Nature America, Inc. All rights reserved.To the Editor:In a recent Letter, Labrijn et al. 1 reportedthat therapeutic wild-type IgG4s engagein Fab-arm exchange with endogenoushuman IgG4 in vivo. The work presentedembellishes a theme that was revived by vander Neut Kolfschoten et al. 2 , who confirmedin an excellent paper previous hypothesesand findings 3,4 , that is, that IgG4s aredynamic molecules that exchange Fab armsby swapping a heavy chain and attachedlight chain (half molecule) with a heavylightchain pair from another molecule,resulting in bispecific antibodies. Whereasvan der Neut Kolfschotenet al. 2 suggest that futurestudies should addressthe contribution of IgG4Fab-arm exchange to invivo activity of therapeuticmonoclonal IgG4 antibodies,Labrijn et al. 1 demonstratethat Fab-arm exchangebetween natalizumab(Tysabri) and endogenoushuman IgG4 can indeed beobserved in blood samplesfrom natalizumab-treatedindividuals. It is unfortunatethat Labrijn et al. 1 do not address theintriguing suggestion raised by van der NeutKolfschoten et al. 2 , and no data are presentedon how the phenomenon of Fab-armexchange may affect the therapeutic activityof therapeutic wild-type IgG4. Instead,Labrijn et al. 1 repeat what was alreadypostulated by van der Neut Kolfschoten etal. 2 ; that is, that Fab-arm exchange could havebiological consequences in that the bindingto the cognate antigen could change in timefrom an avidity to an affinity interaction,thereby possibly decreasing binding strengthand changing homologous cross-linkingto non-cross-linking behavior. Indeed, thisis a significant take-home message and itcontributes to the overall knowledge of animportant subclass of therapeutic antibodies.The awareness of Fab-arm exchange andthe subsequent undesired introduction ofunpredictability for human immunotherapy,as created by van der Neut Kolfschoten etal. 2 and confirmed by Labrijn et al. 1 , is ofsubstantial value for antibody developmentcompanies who choose IgG4 as the preferredsubclass for their products. We thereforeagree with the conclusion that mutations thatprevent Fab-arm exchange in vivo should beconsidered when designing therapeutic IgG4.In a Perspectives article that accompaniedthe original Fab-arm exchange paper by vander Neut Kolfschoten et al. 2 in Science, Burtonand Wilson 5 rightly drew attention to thesignificance of the work. They concluded bystating, “[In instances where wildtype IgG4molecules have been used inclinical trials] the possibilitythat Fab arm exchange couldcontribute to adverse effectsin IgG4 therapy [6,7] shouldbe explored immediately,”thereby referring to paperspresenting the cytokinestorm precipitatedby TGN1412 (ref. 6)or the occurrence ofprogressive multifocalleukoencephalopathy (PML)in natalizumab-treatedpatients 7 . This concludingstatement introduced an alleged link betweenwild-type IgG4 and adverse events.Labrijn et al. 1 refer to exactly this statementin the last part of their letter where they recitethat the potential exchange with preexistingIgG4 with undesired specificity raises thepossibility that Fab-arm exchange couldhave contributed to some of the adverseevents reported for wild-type IgG4. It isregrettable that, in contrast to Burton andWilson’s 5 suggestion, Labrijn et al. 1 do notfurther explore the possibility that Fabarmexchange could have contributed toeither the occurrences of PML or cytokinestorm experimentally. Instead, Labrijn etal. 1 briefly summarize the current theoryof reduced immune surveillance, whichplausibly explains the occurrence of PMLin natalizumab-treated patients, but goon to counter that JC virus (JCV)–verylate activation antigen 4 (VLA4; or alpha4beta1 chain integrin) bispecific antibodiesmediate transport of JCV into the centralnervous system (CNS) to cause PML. Notonly have they failed to provide supportingexperimental data but they offer nosubstantiating evidence from the literature.The mechanism by which JCV-VLA4bispecifics might mediate transfer of virusto the CNS is presumably based on thefollowing assumptions. First, sufficientamounts of anti-JCV IgG4 should bepresent in the circulation to form JCV-VLA4bispecifics. Labrijn et al. 1 do not reportdata on potential detection of JCV-VLA4bispecific antibodies in natalizumab-treatedpatients, nor do they predict what the chancesof formation of these would be. In a recentstudy, Egli et al. 8 measured the prevalence ofJCV infection and replication in 400 healthydonors. They report the IgG seroprevalencefor JCV to be 58%. However, it is not furtherspecified what proportion of the JCV IgG wasIgG4. But assuming that IgG4 anti-JCV willbe present in a proportion of healthy donors,JCV-VLA4 bispecifics are only expected toexist transiently, given the dynamic natureof Fab-arm exchange 2 . Second, if JCV-VLA4bispecifics were “mediating the capture ofJCV,” as postulated by Labrijn et al. 1 , free viralparticles would need to be in the circulationor in tissues. Egli et al. 8 report that JCV DNAcould not be detected in any of the 400 bloodsamples from healthy donors. In addition,Iacobaeus et al. 9 analyzed the cerebrospinalfluid, cerebrospinal fluid cells and bloodfrom 217 patients with multiple sclerosis(MS) and 212 controls for detection of JCVDNA. They reported a low copy number ofJCV DNA in only four samples (two MS andtwo controls), none in the other 425. Thesefour individuals had no sign or symptomof PML nor did they develop the diseaseduring follow-up. The combined publicationsby Egli et al. 8 and Iacobaeus et al. 9 thusdemonstrate that free virus particles couldnot be detected in plasma or cerebrospinalfluid samples of the majority of healthydonors and MS patients. This is in linewith previous studies that report residenceof JCV in the kidney in an asymptomaticnature biotechnology volume 28 number 2 february 2010 123

correspondence© 2010 Nature America, Inc. All rights reserved.Table 1 IgG4 monoclonals that have been in trials or on the marketCompany (location)Antibody (brand name)Stabilizedhinge a Type Target StageWyeth (Madison, NJ, USA) UCB (Brussels) Gemtuzumab (Mylotarg) YES Humanized CD33 MarketBiogen-Idec (Cambridge, MA, USA), Elan (Dublin) Natalizumab (Tysabri) NO Humanized VLA-4 (CD49d) MarketBiogen-Idec, UCB CDP 571 (Humicade) No Humanized Tumor necrosis factor alpha(TNFalpha)AstraZeneca (London) with Medimmune-CambridgeAntibody Technology (CAT) (Gaithersburg, MD, USA)PDL Biopharma (Redwood City, CA, USA)with Biogen-IdecCAT-152/lerdelimumab N/A Human Transforming growth factor(TGF)-beta 2Volociximab N/A Chimeric α5β1 integrin Phase 2Discontinued afterphase 3Discontinued afterphase 3Bristol Myers Squibb (Princeton, NJ, USA) BMS-663513 N/A Human CD137 (4-1BB) Phase 2Genzyme (Cambridge, MA, USA) with AstraZeneca GC-1008 N/A Human TGF-beta (1,2,3) Phase 2(Medimmune-CAT)Tanox (Houston) with Biogen-Idec TNX-355/ibalizumab No Humanized CD4 Phase 2AstraZeneca (Medimmune-CAT) CAT-354 No Human Interleukin 13 Phase 2iCo Therapeutics (Vancouver, BC, Canada) with iCo-008/CAT 213)/bertilimumab No Human Eotaxin Phase 2Medimmune-CATAltor Bioscience (Palm Beach, FL, USA) with Tanox ALT-836/TNX-832 N/A Chimeric Tissue factor Phase 2bBiogen-Idec IDEC-151/clenoliximab YES Chimeric CD4 Discontinued afterphase 2Genzyme, AstraZeneca (Medimmune-CAT) CAT-192/metelimumab No Human TGF-beta 1 Discontinued afterphase 2Biotest (Dreieich, Germany) BT-062 N/A Humanized Syndecan-1 (CD138) Phase 1Innate Pharma (Marseille, France) IPH 2101 Yes Human Natural killer inhibitory Phase 1receptorHuman Genome Science (Rockville, MD, USA)/ HGS-TR2J No Human TNF-related apoptosisinducingPhase 1Kirin (Tokyo)ligand receptor 2(TRAIL-2R)Human Genome Science HGS004 No Human CC-motif chemokine Phase 1receptor 5 (CCR5)PanGenetics PG102 NO Humanized CD40 Phase 1PanGenetics ch5D12 NO Chimeric CD40 Phase 1Johnson & Johnson (New Brunswick, NJ, USA) hOKT3γ4 N/A Humanized CD3 Discontinued afterphase 1GPC Biotech (Martinsreid, Germany) 1D09C3 No Human Human leukocyte antigen(HLA)-DRTeGenero Immuno Therapeutics(now closed; Wurzburg, Germany)a Uppercase data (YES/NO) confirmed; sentence case data (Yes/No) based on patent/literature information; N/A, not available.Discontinued afterphase 1TGN1412 NO Humanized CD28 Discontinued afterphase 1state and the tropism of the virus for(pre-)B cells and CD34 + hematopoieticprogenitor cells (reviewed in ref. 10). So inthe event that JCV-VLA4 bispecifics wereto be circulating in natalizumab-treatedpatients, the chance that these bispecificswould indeed mediate capture of free virusparticles would be vanishingly low. Lastly,Labrijn et al. 1 postulate that JCV, capturedby JCV-VLA4 bispecifics, is transported intothe CNS by infiltrating activated (VLA4 + )leukocytes. This last step would requirean active infiltration of leukocytes intothe CNS of natalizumab-treated patients.Although the influx of leukocytes intoareas of disease activity is a pathologicalhallmark of MS, it is exactly this featurethat is inhibited by natalizumab. Numerousstudies, including both animal and humandata, have demonstrated that antibodiesagainst α4-integrins effectively prevent theaccumulation of leukocytes in the CNS. Morespecifically, it has been demonstrated thatcompared with patients with MS not treatedwith natalizumab, cerebrospinal fluid fromnatalizumab-treated patients has significantlyfewer white blood cells, CD4 + T cells, CD8 +T cells, CD19 + B cells and CD138 + plasmacells. These levels remain low, even 6 monthsafter cessation of natalizumab (reviewed inref. 11).On the basis of the above evidence, weconclude the following: first, JCV-VLA4bispecifics in natalizumab-treated patientshave not been demonstrated and, if theywere to exist, would be transient; second, thechance for JCV-VLA4 bispecifics to capturefree virus particles is infinitesimal; andthird, any facilitation of transport of JCVby activated leukocytes is prevented by anoverall natalizumab-mediated inhibition ofleukocyte entry into the CNS.We wish to emphasize that it is wellknownthat PML tends to arise inchronically immunosuppressed patients.CD4 + and CD8 + T lymphopenia resultingfrom HIV infection, chemotherapy orimmunosuppressive therapy are the primaryrisk factors. In addition to the cases reportedfor natalizumab treatment, PML has beenreported to occur in patients treated withIgG1-based biologicals, including rituximab(Rituxan), efalizumab (Raptiva) andalemtuzumab (Campath) 10,12,13 .Furthermore, we would like to point outthat Labrijn et al. 1 fail to explain how Fabarmexchange might have caused cytokinestorm in TGN1412-treated patients. However,the authors do prominently mention thatthe kinetics of these adverse events are124 volume 28 number 2 february 2010 nature biotechnology

correspondence© 2010 Nature America, Inc. All rights reserved.compatible with the first detection of Fabarmexchange in their study. We wish to stressthat none of the many papers publishedthat studied the mechanism for TGN1412-associated cytokine storm consider Fab-armexchange. In contrast, almost all point tospecific CD28 target biology (e.g., see refs.14,15).Although we acknowledge that a Lettercan have room for some speculation, therepeated insinuated link between Fab-armexchange and adverse events to natalizumaband TGN1412 is misleading. We believe thatFab-arm exchange poses no generic safetyissue but recognize that a major disadvantagewill be the unpredictability for humanimmunotherapy due to the dynamics ofthe transient existence of specific Fab-armcombinations and reduced ability to crosslinkthe originally targeted antigen.The absence of any generic safety issue issupported by the number of independentclinical studies performed with monoclonalIgG4 to date (Table 1). We fear that,through repetition, the notion that Fab-armswitching causes adverse events will becomeaccepted, which will undermine the effortsof biopharmaceutical companies developingwild-type IgG4 candidate drugs. This concernis substantiated by the observation that, sincethe publication of the previously mentionedPerspectives article by Burton and Wilson 5 ,the alleged link between Fab-arm exchangeand adverse events to wild-type IgG4 didreach biotechnology stakeholders. Onebiopharmaceutical company announcedthe discontinuation of its antibody programmentioning the “General concerns ofFab-arm exchange,” although the drugcandidate had not raised any unexpectedor unacceptable safety concerns in initialclinical testing (Supplementary Note).On top of that, the Distillery section ofthe newsletter SciBX 16 , which each weeksummarizes the most essential scientificfindings in techniques of commercial interest,summarizes Labrijn et al. 1 by reciting “Fabarmexchange occurs when the arm fragmentof a therapeutic antibody exchanges withthe arm fragment of an endogenous plasmaantibody. The result is a bispecific antibodythat reduces the binding affinity of thetherapeutic antibody and potentially leadsto side effects such as progressive multifocalleukoencephalopathy (PML).” A later issueof SciBX 17 amplified this summary. Thisdemonstrates that reinforcement of theunfounded assertion has already begun.In conclusion, we are not aware of anyevidence or theory to support a generic safetyissue for Fab-arm switching and would like toalert the antibody community on misleading‘arm-waving’.Note: Supplementary information is available on theNature Biotechnology website.COMPETING INTERESTS STATEMENTThe authors declare competing financialinterests: details accompany the full-text HTMLversion of the paper at http://www.nature.com/naturebiotechnology/.Ellen Broug 1 , Philip A Bland-Ward 2 ,John Powell 2 & Kevin S Johnson 21 PanGenetics BV, Utrecht, The Netherlands.2 PanGenetics BV, Royston, United Kingdom.email: ellen.broug@pangenetics.com1. Labrijn, A.F. et al. Nat. Biotechnol. 27, 767–771(2009).2. van der Neut Kolfschoten, M. et al. Science 317,1554–1557 (2007).3. Schuurman, J. et al. Immunology 97, 693–698(1999).4. Aalberse, R.C. & Schuurman, J. Immunology 105,9–19 (2002).5. Burton, D.R. & Wilson, I.A. Science 317, 1507–1508(2007).6. Suntharalingam, G. et al. N. Engl. J. Med. 355, 1018–1028 (2006).7. Kleinschmidt-DeMasters, B.K. & Tyler, K.L. N. Engl. J.Med. 353, 369–374 (2005).8. Egli, A. et al. J. Infect. Dis. 199, 837–846 (2009).9. Iacobaeus, E. et al. Mult. Scler. 15, 28–35 (2009).10. Carson, K.R. et al. Lancet Oncol. 10, 816–824(2009).11. Stüve, O. et al. J. Neurol. 255 suppl.6, 58–65(2008).12. Carson, K.R. et al. Blood 113, 4834–4840 (2009).13. Waggoner, J., Martinu, T. & Palmer, S.M. et al. J. HeartLung Transplant. 28, 395–398 (2009).14. Schraven, B. & Kalinke, U. Immunity 28, 591–595(2008).15. Gogishvili, T. et al. PLoS ONE 4, e4643 (2009).16. Anonymous. SciBX 2, doi:10.1038/scibx.2009.1195.17. Edelson, S. SciBX 2, doi:10.1038/scibx.2009.1231.Labrijn et al. reply:Our papers test the hypothesis of IgG4 Fabarmexchange 1,2 . In a series of in vitro andin vivo experiments, we provide convincingexperimental evidence for the occurrenceof Fab-arm exchange and show that thismechanism represents an intrinsic activityof IgG4 antibodies that affects treatmentwith IgG4-based therapeutics in patients.Broug et al. criticize our approach andsuggest that we are reiterating previousinsights. Surely, however, Broug et al.understand that we are employing the timehonoredapproach of hypothesis-drivenresearch. Thus, a hypothesis should notbe accepted until rigorous testing showsit to be true, which then leads to novelhypotheses. IgG4 Fab-arm exchange therebyhas only recently become scientific fact 1,2 .The hypothesis that states that Fab-armexchange of therapeutically administeredand endogenous IgG4 molecules maygenerate bispecific IgG4 with undesiredspecificity that induce adverse events 3still requires (clinical) testing. Brouget al. specifically object to a discussionin which we visit the idea that Fab-armexchange of natalizumab with patientIgG4 might play a role in the pathogenesisof PML 2 . Examining this notion, wedraw the conclusion (albeit not veryplausible under normal conditions)that our analysis cannot fully excludethe possibility that cross-linking undercertain (pathological) conditions may besufficient for undesired biological effects.This limitation, however, also applies tothe discussion brought forward by Brouget al., despite their comprehensive citing ofliterature. We fully acknowledge the absenceof reported evidence for generic safetyissues with IgG4 molecules and do notclaim that natalizumab is ‘unsafe’. As ourpharmacokinetic modeling already suggests,the frequency of IgG4 with undesiredcross-linking ability is probably very low 1 .Studies with large patient populations willbe required to investigate their presenceand effect. In the case of natalizumab,the potential safety-risk, furthermore,has already been mitigated by extensivescreening for JCV.Does our research undermine efforts ofbiopharmaceutical companies developingwild-type IgG4 candidate drugs, asargued by Broug et al.? The mechanismof Fab-arm exchange, in our view, makesthe development of nonstabilized IgG4antibody therapeutics highly problematicas it complicates manufacturing,affects pharmacokinetics and makespharmacodynamics unpredictable. Thelast of these problems is brought homeby observations with the therapeuticIgG4 anti-HLA DR antibody 1D09C3,which was not discontinued because ofunsubstantiated concerns, as suggestedby Broug et al., but instead was foundto become inactive following Fab-armexchange 4 . With the demise of 1D09C3, itcan be learned from Table 1 in Broug et al.that Pangenetics’ CD40 antibodies ch5D12and PG102 remain the only two confirmednonstabilized IgG4 antibodies in clinicaldevelopment. Biotechnology companies,of course, are responsible for plotting theirown strategy. Nevertheless, as discussedby De Rubertis et al., successful companiesshould combine strong science with acapability to absorb failure, adapt and moveon 5 .It can certainly be argued that theimpact of the unpredictable behavior ofnonstabilized, wild-type, IgG4 antibodiesin terms of manufacturing, consistency,nature biotechnology volume 28 number 2 february 2010 125

correspondence© 2010 Nature America, Inc. All rights reserved.pharmacology and clinical safety canbe minimized by installing appropriatecounter measures. The costs and effortsinvolved as well as the disadvantages inpositioning relative to stabilized competitordrugs, however, strongly suggest to us thatsuch development routes should no longerbe considered.COMPETING INTERESTS STATEMENTThe authors declare competing financialinterests: details accompany the full-text HTMLversion of the paper at http://www.nature.com/naturebiotechnology/.Aran F Labrijn, Janine Schuurman,Jan G J van de Winkel & Paul W H I ParrenGenmab, Utrecht, The Netherlands.e-mail: p.parren@genmab.com1. Labrijn, A.F. et al. Nat. Biotechnol. 27, 767–771(2009).2. van der Neut Kolfschoten, M. et al. Science 317, 1554–1557 (2007).3. Burton, D.R. & Wilson, I.A. Science 317, 1507–1508(2007).4. Hansen, K. et al. Mol. Immunol. 46, 3269–3277(2009).5. De Rubertis, F., Fleck, R. & Lanthaler, W. Nat.Biotechnol. 27, 595–597 (2009).An economic and technicalevaluation of microalgal biofuelsTo the Editor:In her News Feature “Biotech’s green gold” 1 ,Emily Waltz details the ‘hype’ being propagatedaround emerging microalgal biofueltechnologies, which often exceeds the physicaland thermodynamic constraints that ultimatelydefine their economic viability. Ourcalculations (Supplementary Box 1) countersuch excessive claims 1,2 and demonstrate that22 MJ m –2 d –1 solar radiationsupports practical yieldmaxima of ~60 to 100 kl oilha –1 y –1 (~6,600 to 10,800gal ac –1 y –1 ) and an absolutetheoretical ceiling of ~94 to155 kl oil ha –1 y –1 , assuminga maximum photosyntheticconversion efficiency of 10%(ref.3) (results summarizedin Table 1). To evaluateclaims and provide an accurateanalysis of the potentialof microalgal biofuel systems,we have conductedindustrial feasibility studies and sensitivityanalyses based on peer-reviewed data andindustrial expertise. Given that microalgalbiofuel research is still young and its developmentstill in flux, we anticipate that thestringent assessment of the technology’s economicpotential presented below will assistR&D investment and policy development inthe area going forward.If sustainable and profitable processescan be developed, the potential benefitsof these technologies for the commongood appear compelling and include theproduction on nonarable land of biodiesel,methane, butanol, ethanol, aviationfuel and hydrogen, using waste or salinewater, as well as CO 2 from industrial oratmospheric sources. We have examinedindustrial feasibility models of microalgalsystems to identify the key economicdrivers and provide an industrially relevantupdate on previous economic analyses 4,5 .Two of our models are described here as‘base case’ (that is, integrating currenttechnology) and ‘projectedcase’, which is consideredachievable but has notyet been demonstratedat commercial scale. The30-year internal rate ofreturn (IRR; Fig. 1 andSupplementary Fig. 1) andnet present value (NPV;Supplementary Fig. 2) areused as a measure of theprofitability of differentproduction scenarios.IRR values of 15% andabove are considered toindicate the potential for economic viability.Importantly, all subsidies including carboncredits have been deliberately excluded, ashave financial optimization techniques, toprovide a substantial financial contingency.The base case is intended to represent anemerging scenario from the industry andinvolves the following assumptions:(i) production of microalgal biomass using500 ha of microalgal production systems; (ii)the extraction of oil; (iii) the co-productionand extraction of a high value product(HVP; e.g., β-carotene at 0.1% of biomass,$600/kg); and (iv) the sale of the remainingbiomass as feedstock (e.g., soymeal orfishmeal substitute). In contrast, theprojected case is intended to represent themicroalgal biofuel industry at maturity andno longer incorporates the co-productionof HVPs. The base case is essentially a selfsubsidizing,co-production model. Althoughit produces ~100 times more oil than HVPon a per tonnage basis, the revenue fromHVPs is ~10 times that of oil due to theirdifference in value. In reality, deployment ofthis co-production approach will require theservicing of a diversity of HVP markets, asHVP markets are small and easily saturable.Consequently, a major consideration is thatthe technical developments required forthe commercialization of individual HVPscan be as challenging as those required forbiofuel production. Therefore, the existenceof one or more suitable market-readyHVPs represents a central decision pointfor would-be biofuel producers. Improvedmicroalgal productivity approaching thetargets identified in the projected case willreduce the reliance on co-production (Table1) as the industry matures.All assumptions (variable settings) in thismodel are detailed in the SupplementaryData and are based on what are consideredto be realistically achievable and publishedpeer-reviewed values. The key findingsof this model are summarized in Figure1 (also see Supplementary Figs. 1 and2) as sensitivity analysis plots in whichindividual or multiple settings (e.g.,biomass productivity and constructioncosts) are varied to evaluate their effect onthe IRR. For example, as construction costsare reduced (Fig. 1a), the IRR increases.The appropriateness of using the IRR asa measure for profitability in this study isdemonstrated in the Supplementary Dataand Supplementary Figure 2. Detailedfigures for NPV are also provided. Thismodel deliberately does not discriminatebetween open pond and closed bioreactorsystems, the pros and cons of whichare hotly debated; instead, it comparesconstruction costs versus yield (Fig. 1e,f) asthis is the critical factor (that is, low cost/lowyield and high cost/high yield reactors cantheoretically be equally profitable).We illustrate how key factors affect theIRR of both case studies (Fig. 1). Thesewere (i) capital costs for construction ofthe ponds/reactors (Fig. 1a) and (ii) thebiomass productivity (g m –2 day –1 ) (Fig.1b). In the base case, the third key factorwas the role of HVPs (Fig. 1c), whereas inthe projected case, the corresponding effectwas oil price (Fig. 1d), but biomass oilcontent (see Supplementary Fig. 1d) was126 volume 28 number 2 february 2010 nature biotechnology

correspondence© 2010 Nature America, Inc. All rights reserved.Table 1 Practical and theoretical yield maxima for microalgal biomass and oil production aPhotosyntheticconversion efficiencyBiomassenergyBiomassenergyonly a minor factor. In each case, HVPs oroil represent the dominant revenue streamsat ~60% and ~50%, respectively. Thediversification of products (HVPs, biomassand oil) in the base case has the potential toprotect against oil price fluctuations and tosubsidize oil prices.Production area scales nonlinearly inour analysis (Supplementary Fig. 1a) withsizes

correspondence© 2010 Nature America, Inc. All rights reserved.systems. Examples include the possibilityof a resurgence in oil price from $30/bblrecently (as of December 2008; http://tonto.eia.doe.gov/dnav/pet/hist/rwtcd.htm) back to and beyond the previoushigh of $147/bbl (July 2008) in the nearfuture (modeled here as $100 per barrelin both cases), the introduction of morestringent CO 2 emissions targets andcarbon trading schemes (potentiallyrising to ~$200 per ton CO 2 by 2050; refs.7,8) and the increased demand for foodand fuel by a population rising from 6.8billion in 2009 to 9.4 billion in 2050 (refs.9,10). All of these factors appear to bestrong drivers for the economic viabilityof this technology. This analysis suggeststhat although microalgal biofuel systemsremain in an early stage of development,they are now approaching profitability ifthe co-production systems in the base case,and/or the increased productivities in theprojected case can be attained. A recentreport by Huntley and Redalje 11 estimatesthat current technology could produceoil for $84/bbl (with no value attributedto the non-oil fraction), with reasonableadvancements in technology reducing thiscost to $50/bbl or less. This supports ourconclusion that co-production is requiredin the short term and that at increased oilprices (that is, $100 in this model) an IRRof 15% could be obtained.Considerable synergies also exist betweenmicroalgae biofuel production and a widerange of other industries, including humanand animal food production, veterinaryapplications, agrochemicals, seed suppliers,biotech, water treatment, coal seam gas,material supplies and engineering, fuelrefiners and distributors, bio-polymers,pharmaceutical and cosmetic industries,as well as coal-fired power stations (CO 2capture) and transport industries, such asaviation. Sound opportunities therefore existfor the development of a rapidly expandingsustainable industry base whose productivityis independent of soil fertility and lessdependent on water purity. Thus, thesetechnologies can conceivably be scaled tosupply a substantial fraction of oil demandwithout increasing pressure on waterresources while potentially contributingto food production. Furthermore, as thisstudy was conservatively modeled onpublished data, excluding subsidies (whichare actually commonly used to developother renewable energy sectors, for example,photovoltaics) and proprietary technologies,it follows that strategic partnerships andgovernment policy decisions will play a largepart in facilitating a coordinated scale-upand deployment of these technologies tocontribute to future energy security.Note: Supplementary information is available on theNature Biotechnology website.ACKNOWLEDGMENTSThe authors gratefully acknowledge the support ofthe Australian Research Council, IMBcom, and theeconomic advice of Liam Wagner.COMPETING INTERESTS STATEMENTThe authors declare competing financial interests:details accompany the full-text HTML version of thepaper at http://www.nature.com/naturebiotechnology/.Evan Stephens 1 , Ian L Ross 1 , Zachary King 2 ,Jan H Mussgnug 3 , Olaf Kruse 3 , Clemens Posten 4 ,Michael A Borowitzka 5 & Ben Hankamer 11 The University of Queensland, Institute forMolecular Bioscience, St. Lucia, Queensland,Australia. 2 IMBcom, St. Lucia, Queensland,Australia. 3 University of Bielefeld, Department ofBiology, AlgaeBioTech Group, Bielefeld, Germany.4 University of Karlsruhe, Institute of Life ScienceEngineering, Bioprocess Engineering, Karlsruhe,Ontology engineeringTo the Editor:Gene Ontology (GO) 1 and similarbiomedical ontologies are critical tools oftoday’s genetic research. These ontologiesare crafted through a painstaking process ofmanual editing, and their organization relieson the intuition of human curators. Herewe describe a method that uses informationtheory to automatically organize thestructure of GO and optimize thedistribution of the information within it. Weused this approach to analyze the evolutionof GO, and we identified several areas wherethe information was suboptimally organized.We optimized the structure of GO andused it to analyze 10,117 gene expressionsignatures. The use of this new versionchanged the functional interpretations of97.5% (P < 10 –3 ) of the signatures by, onaverage, 14.6%. As a result of this analysis,several changes will be introduced in thenext releases of GO. We expect that theseformal methods will become the standard toengineer biomedical ontologies.Every year, over 400,000 new articlesenter the biomedical literature 2 , creating anunprecedented corpus of knowledge that isimpossible to explore with traditional meansof literature consultation. This situationmotivated the development of biomedicalontologies, structured informationGermany. 5 Murdoch University, School ofBiological Sciences and Biotechnology, Algae R&DCenter, Murdoch, Western Australia, Australia.e-mail: b.hankamer@imb.uq.edu.au1. Waltz, E. Nat. Biotechnol. 27, 15–18 (2009).2. Mascarelli, A. Nature 461, 460–461 (2009).3. Melis, A. Plant Sci. 177, 272–280 (2009).4. Weissman, J.C. & Goebel, R.P. Design and Analysis ofPond Systems for the Purpose of Producing Fuels (SolarEnergy Research Institute, Golden, Colorado, SERI/STR-231–2840, 1987).5. Benemann, J.R. & Oswald, W.J. Systems and EconomicAnalysis of Microalgae Ponds for Conversion of CO 2to Biomass. Final Report to the Pittsburgh EnergyTechnology Center. Grant no. DE-FG22–93PC93204(1996).6. Huggett, B. Nat. Biotechnol. 26, 1208–1209 (2008).7. McFarland, J.R., Reilly, J.M. & Herzog, H.J. EnergyEcon. 26, 685–707 (2004).8. Zhu, Z., Graham, P., Reedman, L. & Lo, T.A. Decis. Econ.Finance 32, 35–48 (2009).9. Anonymous. World Population Data Sheet (PopulationReference Bureau, Washington, DC, 2009). 10. Anonymous. Soziale und Demographische Daten zurWeltbevölkerung (Deutsche Stiftung Weltbevölkerung,Hannover, Germany, 2009). 11. Huntley, M.E. & Redalje, D.G. Mitig. Adapt. StrategiesGlob. Change 12, 573–608 (2007).repositories that organize biomedicalfindings into hierarchical structures andcontrolled vocabularies. GO is arguably themost successful example of a biomedicalontology. GO is a controlled vocabulary todescribe gene and gene product attributesin any organism and includes 26,514 termsorganized along three dimensions: molecularfunction, biological process and cellularcomponent. GO has become even moreintensively used with the introduction ofhigh-throughput genomic platforms becauseof its ability to categorize large amounts ofinformation using a controlled vocabulary togroup objects and their relationships 1,3,4 .Today, GO and other biomedicalontologies are the result of a painstaking,costly and slow process of manual curationthat requires reaching a consensus amongmany experts to implement a change.Furthermore, the topology of GO hasbecome critically important because ofthe introduction of gene set enrichmentmethods. These methods have allowedinvestigators to characterize the results ofa high-throughput experiment in termsof coherent, knowledge-defined sets ofgenes (e.g., pathways, functional classesor chromosomal locations) rather thanin terms of anecdotal evidence about128 volume 28 number 2 february 2010 nature biotechnology

correspondence© 2010 Nature America, Inc. All rights reserved.systems. Examples include the possibilityof a resurgence in oil price from $30/bblrecently (as of December 2008; http://tonto.eia.doe.gov/dnav/pet/hist/rwtcd.htm) back to and beyond the previoushigh of $147/bbl (July 2008) in the nearfuture (modeled here as $100 per barrelin both cases), the introduction of morestringent CO 2 emissions targets andcarbon trading schemes (potentiallyrising to ~$200 per ton CO 2 by 2050; refs.7,8) and the increased demand for foodand fuel by a population rising from 6.8billion in 2009 to 9.4 billion in 2050 (refs.9,10). All of these factors appear to bestrong drivers for the economic viabilityof this technology. This analysis suggeststhat although microalgal biofuel systemsremain in an early stage of development,they are now approaching profitability ifthe co-production systems in the base case,and/or the increased productivities in theprojected case can be attained. A recentreport by Huntley and Redalje 11 estimatesthat current technology could produceoil for $84/bbl (with no value attributedto the non-oil fraction), with reasonableadvancements in technology reducing thiscost to $50/bbl or less. This supports ourconclusion that co-production is requiredin the short term and that at increased oilprices (that is, $100 in this model) an IRRof 15% could be obtained.Considerable synergies also exist betweenmicroalgae biofuel production and a widerange of other industries, including humanand animal food production, veterinaryapplications, agrochemicals, seed suppliers,biotech, water treatment, coal seam gas,material supplies and engineering, fuelrefiners and distributors, bio-polymers,pharmaceutical and cosmetic industries,as well as coal-fired power stations (CO 2capture) and transport industries, such asaviation. Sound opportunities therefore existfor the development of a rapidly expandingsustainable industry base whose productivityis independent of soil fertility and lessdependent on water purity. Thus, thesetechnologies can conceivably be scaled tosupply a substantial fraction of oil demandwithout increasing pressure on waterresources while potentially contributingto food production. Furthermore, as thisstudy was conservatively modeled onpublished data, excluding subsidies (whichare actually commonly used to developother renewable energy sectors, for example,photovoltaics) and proprietary technologies,it follows that strategic partnerships andgovernment policy decisions will play a largepart in facilitating a coordinated scale-upand deployment of these technologies tocontribute to future energy security.Note: Supplementary information is available on theNature Biotechnology website.ACKNOWLEDGMENTSThe authors gratefully acknowledge the support ofthe Australian Research Council, IMBcom, and theeconomic advice of Liam Wagner.COMPETING INTERESTS STATEMENTThe authors declare competing financial interests:details accompany the full-text HTML version of thepaper at http://www.nature.com/naturebiotechnology/.Evan Stephens 1 , Ian L Ross 1 , Zachary King 2 ,Jan H Mussgnug 3 , Olaf Kruse 3 , Clemens Posten 4 ,Michael A Borowitzka 5 & Ben Hankamer 11 The University of Queensland, Institute forMolecular Bioscience, St. Lucia, Queensland,Australia. 2 IMBcom, St. Lucia, Queensland,Australia. 3 University of Bielefeld, Department ofBiology, AlgaeBioTech Group, Bielefeld, Germany.4 University of Karlsruhe, Institute of Life ScienceEngineering, Bioprocess Engineering, Karlsruhe,Ontology engineeringTo the Editor:Gene Ontology (GO) 1 and similarbiomedical ontologies are critical tools oftoday’s genetic research. These ontologiesare crafted through a painstaking process ofmanual editing, and their organization relieson the intuition of human curators. Herewe describe a method that uses informationtheory to automatically organize thestructure of GO and optimize thedistribution of the information within it. Weused this approach to analyze the evolutionof GO, and we identified several areas wherethe information was suboptimally organized.We optimized the structure of GO andused it to analyze 10,117 gene expressionsignatures. The use of this new versionchanged the functional interpretations of97.5% (P < 10 –3 ) of the signatures by, onaverage, 14.6%. As a result of this analysis,several changes will be introduced in thenext releases of GO. We expect that theseformal methods will become the standard toengineer biomedical ontologies.Every year, over 400,000 new articlesenter the biomedical literature 2 , creating anunprecedented corpus of knowledge that isimpossible to explore with traditional meansof literature consultation. This situationmotivated the development of biomedicalontologies, structured informationGermany. 5 Murdoch University, School ofBiological Sciences and Biotechnology, Algae R&DCenter, Murdoch, Western Australia, Australia.e-mail: b.hankamer@imb.uq.edu.au1. Waltz, E. Nat. Biotechnol. 27, 15–18 (2009).2. Mascarelli, A. Nature 461, 460–461 (2009).3. Melis, A. Plant Sci. 177, 272–280 (2009).4. Weissman, J.C. & Goebel, R.P. Design and Analysis ofPond Systems for the Purpose of Producing Fuels (SolarEnergy Research Institute, Golden, Colorado, SERI/STR-231–2840, 1987).5. Benemann, J.R. & Oswald, W.J. Systems and EconomicAnalysis of Microalgae Ponds for Conversion of CO 2to Biomass. Final Report to the Pittsburgh EnergyTechnology Center. Grant no. DE-FG22–93PC93204(1996).6. Huggett, B. Nat. Biotechnol. 26, 1208–1209 (2008).7. McFarland, J.R., Reilly, J.M. & Herzog, H.J. EnergyEcon. 26, 685–707 (2004).8. Zhu, Z., Graham, P., Reedman, L. & Lo, T.A. Decis. Econ.Finance 32, 35–48 (2009).9. Anonymous. World Population Data Sheet (PopulationReference Bureau, Washington, DC, 2009). 10. Anonymous. Soziale und Demographische Daten zurWeltbevölkerung (Deutsche Stiftung Weltbevölkerung,Hannover, Germany, 2009). 11. Huntley, M.E. & Redalje, D.G. Mitig. Adapt. StrategiesGlob. Change 12, 573–608 (2007).repositories that organize biomedicalfindings into hierarchical structures andcontrolled vocabularies. GO is arguably themost successful example of a biomedicalontology. GO is a controlled vocabulary todescribe gene and gene product attributesin any organism and includes 26,514 termsorganized along three dimensions: molecularfunction, biological process and cellularcomponent. GO has become even moreintensively used with the introduction ofhigh-throughput genomic platforms becauseof its ability to categorize large amounts ofinformation using a controlled vocabulary togroup objects and their relationships 1,3,4 .Today, GO and other biomedicalontologies are the result of a painstaking,costly and slow process of manual curationthat requires reaching a consensus amongmany experts to implement a change.Furthermore, the topology of GO hasbecome critically important because ofthe introduction of gene set enrichmentmethods. These methods have allowedinvestigators to characterize the results ofa high-throughput experiment in termsof coherent, knowledge-defined sets ofgenes (e.g., pathways, functional classesor chromosomal locations) rather thanin terms of anecdotal evidence about128 volume 28 number 2 february 2010 nature biotechnology

correspondence© 2010 Nature America, Inc. All rights reserved.0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 BitsBiological processMacromoleculemetabolismDefenseresponseCell proliferationsingle genes 5,6 . GO has become a primaryprovider of these gene sets and researchersuse its graphical structure to identify thespecificity of a gene class so that they willcompare classes of the same specificty 7 .Previous studies have found that thestructure of GO does not conform toexpected intuitions regarding the structureand distributions of ontology terms 8,9 .Gene enrichment methods typically usethe structure of ontologies as a proxy forthe specificity of a term 10,11 or, in somecases, use automated procedures to identifystructural biases and to compensate forthem in the analysis 7,8,12 . Unfortunately,in some cases, even these compensativemethods are unable to reach the sameconclusions of a well-calibrated ontology(Supplementary Notes 1).The approach we advocate here aims tosolve the problem at its root by optimizingthe structure of the ontology so that it willindeed be an accurate representation ofthe informational specificity of any termin the ontology. This approach would notonly avoid the necessity to compensatefor biases but also improve the semantictransparency of the ontology structure.To do so, we introduce an automatedmethod for engineering the structure ofGO based on the information content ofeach single term. The intuition behind thismethod is that ontologies are informationsystems and, as such, they can be optimizedusing the well-established mathematics ofinformation theory. Given its mathematicalnature, this optimization process can beautomated, thus producing a principled andscalable architecture to engineer GO and,analogously, other biomedical ontologies.Our approach starts from thequantification of information contained inthe terms of the ontology. The informationcontent of a term is computed fromthe amount of annotation available forit relative to all other terms, and it is aHeterocyclemetabolismFructosemetabolismFigure 1 Spectrum of GO terms: examples ranging from 1 to 14 bits.TerpenemetabolismGanglion mother cellfate determinationmeasure of the surprise caused by labelinga gene with this term rather than withany other term (Supplementary Notes 2).For instance, if a term contains all genes,then it is not surprising for a given geneto be labeled with it, so this term does notcontain much information. Thus, the moregenes or gene products associated with aterm, the less specific the term is and theless information is conveyed by it. This‘surprise factor’ is called ‘self-information’,and information theory provides a formaldefinition for it 13 (Fig. 1).Using information theory, we analyzed theevolution of the information content of GOacross time, examining 2 million genes acrossall the organisms encoded in the ontologyannotations. This process highlightedinformation biases and inefficiencies thatmay affect the usage of GO and identifiedthose areas of the ontology that were suboptimallyorganized. The analysis identifiedthree types of information inefficiencies inthe structure of GO.Topological metric0.610.60.590.580.570.560.552.5 × 10 1 , 7.6 × 10 −2 , 5.9 × 10 −10.550.0780.0740.072Inter-level metric 0.0720The first type of inefficiency arises fromthe variability of the information contentamong the terms within a given ontologylevel. By the principle of maximum entropy,an even a priori distribution of information(where all terms in a level are equally specificand hence equally informative) is mostefficient because a random experimentis most informative if the probabilitydistribution over outcomes is uniform 13 .Furthermore, gene set enrichment methodsoften use GO level (that is, distance from thetop of the graph) as a proxy for degree ofspecificity 7,10,11 ; this strategy implicitly relieson within-level uniformity of informationcontent. Optimally, then, all the terms in agiven level would have equal specificity and,therefore, the same information content. Ouranalysis revealed that the original version ofGO contained a large degree of such intralevelvariability of information content.For example, the term ‘pilus retraction’ wasoriginally at level 2, at the same level of termslike ‘cell cycle’ and ‘cell development’ that areactually much more general.The second type of structural inefficiency,inter-level variability, arises from deviationsin information content between levels. Ingeneral, terms become more specific as theinformation content of a level increases withdepth in the graph. In some areas of GO,however, the mean information contentdecreases from one level to the next, creatingan information bottleneck. In this case,most of the annotation information of theprevious level is transmitted to the nextthrough only a few terms. The larger thedecrease in information content, the moresevere the bottleneck. The presence of these2.9 × 10 1 , 7.7 × 10 −2 , 6.1 × 10 −13.0 × 10 1 , 7.9 × 10 −2 , 5.9 × 10 −13.0 × 10 1 , 7.6 × 10 −2 , 6.0 × 10 −13.2 × 10 1 , 7.7 × 10 −2 , 5.9 × 10 −12.9 × 10 −1 ,7.6 × 10 2 ,5.9 × 10 −13.0 × 10 1 , 7.6 × 10 −2 , 5.8 × 10 −12.5 × 10 1 , 7.5 × 10 −2 , 5.7 × 10 −12.0 × 10 1 , 7.6 × 10 −2 , 5.6 × 10 −1 2.5 × 10 1 , 7.4 × 10 −2 , 5.6 × 10 −1262824Inter-level metricFigure 2 Three-dimensional evolution of GO over ten releases from 2005 to 2007 along the threedimensions of structural inefficiency. An ontology with no inefficiency across these metrics wouldbe at the origin (0,0,0).223032nature biotechnology volume 28 number 2 february 2010 129

correspondence© 2010 Nature America, Inc. All rights reserved.areas of suboptimal information distributionviolate the assumption of gene set enrichmentanalysis methods 7,12 that the specificity in GOterms effectively increases from one level tothe next (Supplementary Notes 3).The third type of structural inefficiency,topological variability, arises from thesuboptimal organization of the branches.The principle of maximum entropy dictatesthat the closer a topological structure is touniform, the greater is the information thatexperiments can derive from it 8 . We usedentropy rate to quantify the uniformity ofthe GO branch structure (SupplementaryNotes 4) so that a higher entropy rateindicates that the ontology structure iscloser to uniform.We analyzed the evolution of GO alongthese three dimensions of structuralinefficiency using ten releases of GOcontaining over 2 million unique genes 14 .Figure 2 plots their structural inefficienciesfor each release of GO and illustrateshow they have been decreasing over time(Supplementary Notes 5). For instance,with time point 8 (February 1, 2007), interlevelvariability and topological variabilitysaw substantive improvements, coincidingwith introduction of the [‘is_a complete’]property in GO 15 . In contrast, intra-levelvariability saw comparatively modestimprovements over the evolution of GO.One of the greatest dangers of structuralinefficiencies in GO is the impact they canhave on the functional interpretation of theresults of high-throughput experiments. Wethus optimized the information distributionof GO by introducing single-level changesand modifying 1,001 relationships and 11%of GO terms, thus significantly improvingthe overall intra-variability (P < 10 −3 )(Supplementary Notes 6).We used this optimization methodto create a modified, improved GO andwe compared it to the current GO in theinterpretation of 10,117 gene expressionsignatures from DNA microarrayexperiments 16 . Each signature containsgenes differentially expressed between twobiological conditions, and we comparedthe results of gene enrichment analysis ofthese signatures obtained by the originaland the modified GO. We found that thesechanges significantly affected the functionalinterpretations of 97.5% (P < 10 −3 ) of theexperimental gene signatures and alteredthe resulting set of GO categories by 14.6%on average (Supplementary Notes 7). Onthe basis of this analysis, we presented 14recommendations to the GO Consortiumand most of these new annotations (12)will be introduced in the next release of GO(Supplementary Notes 8).Finally, as a result of our analysis, weapplied this approach to more complicatedmulti-level structural changes. We suggestedthe GO Consortium move 12 terms. Theterms all underwent the standard curatorialvalidation of the GO consortium, and 11 ofthem are now included in the current releaseof GO. The twelfth term, pigmentation(GO:0043473) had few annotations at thetime but was not moved as it was expectedthat many more genes would be annotatedwith that term in the future.The most striking result of ourexperiment was to show the convergenceof mathematical optimality and biologicalvalidity and that a formal, automatedanalysis is able to uncover sound biologicalinformation hidden in the structure of theontology. By altering the ontology itself, ourapproach improves gene enrichment resultsin ways that cannot be obtained by simplychanging the underlying gene enrichmentmethod (Supplementary Notes 1).Our analysis also reveals that GOcontains more information than iscurrently used. By optimizing thedistribution of information within GO,our method can be used to aid the designof more efficiently organized knowledgerepositories—leading to a more effectiveuse of biological information. This methodis already being used to achieve this aim bythe GO Consortium and other ontologies,such as the Phenotypic Quality Ontology(PATO) 17 in the OBO Foundry 18 . We expectthat formal and automated methods willbecome the standard for the engineering ofbiomedical ontologies.Note: Supplementary information is available on theNature Biotechnology website.ACKNOWLEDGMENTSThis work was supported in part by the NationalLibrary of Medicine (NLM/NIH) under grants1K99LM009826 and 5T15LM007092 and by theNational Human Genome Research Institute(NHGRI/NIH) under grants 2P41HG02273,1R01HG003354, and 1R01HG004836. The authorsare grateful to the anonymous reviewers for theirhelpful suggestions.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.Gil Alterovitz 1–3 , Michael Xiang 1,2 ,David P Hill 4 , Jane Lomax 5 , Jonathan Liu 6 ,Michael Cherkassky 2 , Jonathan Dreyfuss 1,2 ,Chris Mungall 7 , Midori A Harris 5 , Mary E Dolan 4 ,Judith A Blake 4 & Marco F Ramoni 1,21 Children’s Hospital Informatics Program,Harvard-MIT Division of Health Sciencesand Technology, Harvard Medical School,Boston, Massachusetts, USA. 2 PartnersHealthcare Center for Personalized GeneticMedicine, Boston, Massachusetts, USA.3 Department of Electrical Engineering andComputer Science, Massachusetts Institute ofTechnology, Cambridge, Massachusetts, USA.4 Jackson Laboratory, Bar Harbor, Maine,USA. 5 EMBL-EBI, Wellcome Trust GenomeCampus, Hinxton, UK. 6 Department ofBiology, Massachusetts Institute of Technology,Cambridge, Massachusetts, USA. 7 LawrenceBerkeley National Laboratory, Berkeley,California, USA.e-mail: gil_alterovitz@hms.harvard.edu ormarco_ramoni@harvard.edu.1. Ashburner, M. et al. Nat. Genet. 25, 25–29 (2000).2. Davis, D.A., Ciurea, I., Flanagan, T.M. & Perrier, L.Med. J. Aust. 180, S68–S71 (2004).3. Camon, E. et al. Nucleic Acids Res. 32, D262–D266(2004).4. Harris, M. et al. Nucleic Acids Res. 32, D258–D261(2004).5. Subramanian, A. et al. Proc. Natl. Acad. Sci. USA 102,15545–15550 (2005).6. Doniger, S.W. et al. Genome Biol. 4, R7 (2003).7. Al-Shahrour, F., Diaz-Uriarte, R. & Dopazo, J.Bioinformatics 20, 578–580 (2004).8. Alterovitz, G., Xiang, M., Mohan, M. & Ramoni, M.F.G.O.Nucleic Acids Res. 35, D322–D327 (2007).9. Ogren, P.V., Cohen, K.B. & Hunter, L. Pac. Symp.Biocomput. 174–185 (2005).10. Dennis, G. Jr. et al. Genome Biol. 4, 3 (2003).11. Zhou, M. & Cui, Y. In Silico Biol. 4, 323–333(2004).12. Raychaudhuri, S., Chang, J.T., Sutphin, P.D. & Altman,R.B. Genome Res. 12, 203–214 (2002).13. MacKay, D.J.C. Information Theory, Inference, AndLearning Algorithms, xii (Cambridge University Press,Cambridge, U.K.; New York, 2003).14. Wu, C.H. et al. Nucleic Acids Res. 34, D187–D191(2006).15. The Gene Ontology Consortium. Nucleic Acids Res. 36,D440–D444 (2008).16. Yi, Y., Li, C., Miller, C. & George, A.L. Jr. Genome Biol.8, R133 (2007).17. Gkoutos, G.V. et al. Comp. Funct. Genomics 5, 545–551 (2004).18. Smith, B. et al. Nat. Biotechnol. 25, 1251–1255(2007).130 volume 28 number 2 february 2010 nature biotechnology

case studyNever againcommentaryChristopher ScottIn 2008, Roche acquired Genentech, ending the most successful symbiotic business relationship the biotech/pharma sector has ever seen. Morphing biotech business models, pharma management’s short-termism anddwindling investor patience means we’ll never see the like of it again.© 2010 Nature America, Inc. All rights reserved.When Roche’s management acquired a $2.1 billion stake inGenentech in 1990, it eyed not only a deep pipeline of drugs,but also a team of dedicated, independent scientists. At the sametime, Roche executives also saw a financial market buzzed aboutGenentech stock. So the Swiss company bought a majority stake,with an option to buy up the rest later, and left Genentech as a standaloneunit.In 1999, the Basel-based pharma exercised that option, takingGenentech off the market for a few brief months before placing itback on the New York Stock Exchange in another public offering,taking advantage of the bullish environment. Roche then reducedits ownership by selling off shares twice more, bringing in nearly$8 billion for the three offerings (each public sale at a higher pershareprice) and Roche got to keep the majority of product revenueevery year, as a primary stakeholder.By all accounts, this relationship was mutually beneficial. Thepharma’s stake in Genentech was essentially a gift that kept on giving.For its part, Genentech retained product marketing rights inside theUnited States and was able to foster its science in a ‘small company’culture free from interference by its Swiss partner.But Roche couldn’t resist taking the final bite: in March last yearthe pharma acquired the remaining 44% stake it didn’t already ownin a $46.8 billion deal. Analysts fretted that the entrepreneurial culturewould yield to a staid Swiss lockstep. Even Roche chairmanFranz Humer, who drove the deal, had previously said that a buyoutwould “never work, because if we owned all of Genentech we wouldkill it.”In market share alone, Roche-Genentech is now the seventh largestUS pharmaceutical company, projected to generate $17 billion inrevenue yearly. In the months since the merger, Roche pulled out ofthe Washington, DC, lobbying group Pharmaceutical Research andManufacturers of America, saying it would work instead through thebiotech equivalent, the Washington, DC–based Biotechnology IndustryOrganization. And the June American Society of Clinical Oncologymeeting was chock-full of rosy reports from Roche-Genentech’s growingoncology portfolio, with ten new molecular entities in clinical development.In August, the US Food and Drug Administration approvedAvastin (bevacizumab) for kidney cancer. New uses for establisheddrugs, such as Tarceva (erlotinib), Lucentis (ranibizumab) and Avastin,are also moving through confirmatory trials.The union produced its share of changes in senior management.Founder Arthur Levinson moved up to the board of directors andthe president of product development, Susan Desmond-Hellmann,departed. Sandra Horning, a longtime Stanford academic with nopharma experience, took over as senior vice president of global clinicaldevelopment. The chief of operations, Richard Scheller, and his toplieutenants stayed. Time will tell whether the new structure can sustainthe track record of the independent biotech, but even before thebuyout, Genentech’s business had become more reliant on extensionsChristopher Scott is Contributing Editor at Nature Biotechnology.of its existing product franchise (similar to any other big company)rather than production of new therapeutic entities.Few pharma-biotech relationships look anything like the Roche-Genentech courtship. Consider Johnson & Johnson (J&J) of NewBrunswick, New Jersey, locking up Centocor for $4.9 billion in 1999.Centocor was another biotech with an early blockbuster and goodscience to back it up. But J&J treated it like any other merger; thoughit left the biotech as a stand-alone unit, the latter was not publicand had no scientific autonomy. In 2008, J&J merged Centocor withOrtho Biotech, slashed 400 jobs at Ortho and deployed the remainingresources to support Centocor’s high-selling anti-tumor necrosisfactor alpha compound Remicade (infliximab). Today, J&J partiallyowes its standing as a leading drug maker to Remicade: in 2008, ithad worldwide revenue of $5.3 billion. That’s a big hitter in J&J’slineup, but it has made Centocor look an awful lot like a one-trickpony. Genentech, it was not.More diversifying was Basel-based Novartis’ purchase of Emeryville,California–based Chiron for $5.1 billion in 2006. Here again, unlikeRoche’s hands-off approach with Genentech, the acquisition immediatelysent executives from the biotech packing and put scientistsand their projects on the bubble. That year, Novartis launched athree-year reorganization plan, and Chiron went inside Novartis’vaccine and diagnostics division, which last year accounted for just4% of its $42.6 billion in sales. Though Chiron’s vaccine expertise hasbeen important in developing chicken egg–free flu vaccine productionmethods, Novartis did not buy Chiron to play parent; it boughtit to hoard the valuables.Is there another young Genentech and nurturing pharma parentstill out there? Staring across today’s biotech landscape (economicdownturn or not), the answer seems to be no. Another Genentechwould need not only to have a promising and established pipeline,but also to be surrounded by a free-wheeling scientific culture. Thatis awfully hard to find these days, in part because biotech companiesare no longer built with flexible business plans designed to chase bigquestions down dark alleys, as Genentech was.Firms are designed for a quick and total buyout and a return onventure money. Any burgeoning Genentech wannabe would becherry-picked long before its first product launch. What’s more,once consummated, big mergers today tend to isolate resources forthe probable winning compounds, leaving the rest of the science towither on the vine. The priorities for today’s top management aremore myopic: how does an acquisition and its compounds affect thebalance sheet; science is secondary.One final reason that Genentech-Roche is a once-in-a-lifetimerelationship: luck. Though Genentech was a clear biotech leader in1990, and its vaunted pipeline has long been discussed as one of theindustry’s best, the business of making drugs is scored by failure andserendipity alike. It took the financial stability of Roche, the geniusof Genentech’s scientists, an inspiring intellectual culture and providenceto spin out product after product. But lightning does notstrike twice.nature biotechnology volume 28 number 2 february 2010 131

commentaryGoing to ridiculous lengths—Europeancoexistence regulations for GM cropsKoreen Ramessar, Teresa Capell, Richard M Twyman & Paul ChristouEven if a GM crop can surmount Europe’s excessive product registration process, any farmer hoping to plant it mustthen navigate tortuous, arbitrary and scientifically unjustifiable coexistence regulations.© 2010 Nature America, Inc. All rights reserved.Genetically modified (GM) crops nowcover over 100 million hectares of arableland in >20 countries, and this trend towardincreased uptake and deployment is growingat a steady rate 1 . Inevitably, GM and non-GMcrops of the same species will be grown neareach other, a concept defined by the term‘coexistence’ 2 . There has been an extraordinaryand sustained campaign mainly inthe European Union (EU; Brussels) thathas united certain stakeholders, includingorganic producers, certification bodies andenvironmental groups, against GM/non-GMcoexistence. The escalating battle has drawnin producers, retailers, governments, regulatorybodies, scientists and, of course, thegeneral public. The outcome in the EU is amess: a haphazard and inconsistent set ofrules that has no rational scientific underpinning,which obstructs GM producers,misleads the public and adds unnecessarylayers of complexity to international trade.GM/non-GM coexistence is now a loadedterm, used by opponents as a de facto criticismof GM agriculture and a self-fulfillingreason to impose restrictions. Is there anyway to encourage a rational approach to thecoexistence debate?Koreen Ramessar and Teresa Capell are at theDepartament de Producció Vegetal I CiènciaForestal, University of Lleida, Lleida, Spain;Richard M. Twyman is at the Department ofBiological Sciences, University of Warwick,Coventry, UK; and Paul Christou is at theDepartament de Producció Vegetal I CiènciaForestal, University of Lleida and the InstitucióCatalana de Recerca i Estudis Avançats, PasseigLluís Companys, Barcelona, Spain.e-mail: christou@pvcf.udl.esSpecial treatment required? Keeping GMcorn pollen grains (like this one pictured ata magnification of 795×) segregated fromconventional corn is one of the purposes ofEurope’s coexistence regulations.Adventitious presenceThe basis of the campaign against GM/non-GM coexistence is “adventitious presence,”which is defined (in the context of GMagriculture) as the presence of unwantedGM material in non-GM commodities. Theadventitious presence of GM material canoccur in many ways (Fig. 1), but most oftenthrough outcrossing, the growth of volunteerplants from stray seeds and admixtureafter harvest 3 . The adventitious presence ofGM material in non-GM commodities isoften presented as disastrous by opponentsof GM technology and described usingterms such as ‘contamination’ and ‘adulteration’.However, it is important to recognizethat the reasons it is thus regardedANDREW SYRED/SCIENCE PHOTO LIBRARYdiffer according to different stakeholders.Environmental pressure groups are keen topromote uncertainties about the impact ofGM crops on human health and the environmentand oppose coexistence on thebasis that the adventitious presence of GMmaterial is a safety issue, even though thesafety of GM crops must be demonstratedto regulators before licensing for commercialproduction. Organic producers, on theother hand, oppose coexistence becausethey fear their organic status and associatedorganic price premium may dependon the absence of GM material, promptinglegal challenges and lobbying againstGM agriculture both within the EU andelsewhere 2,4 . The European Commission(EC; Brussels) has confirmed that coexistenceis purely an economic issue by definingit as “…issues relating to the economicconsequences of adventitious presence ofmaterial from one crop in another and theprinciple that farmers should be able tocultivate freely the agricultural crops theychoose, be it GM crops, conventional ororganic crops...” 5 .Intimately intertwined with the politicalcoexistence debate is the European public’santipathy to GM products and preferencefor non-GM products. Public uncertaintyabout the safety of GM products is exaggeratedby environmental pressure groupsand some parts of the media, thus helpingto create the preference. Many suppliers andretailers, responding to consumer pressure,have therefore imposed restrictions on theuse of GM material and its presence in foodproducts, encouraging producers to segregateGM and non-GM crops. A viciouscircle has been created.nature biotechnology volume 28 number 2 february 2010 133

COMMENTARY© 2010 Nature America, Inc. All rights reserved.Coexistence practices in non-GMagricultureWhat often gets forgotten in the heat of theGM/non-GM coexistence debate is that differentvarieties of the same crop species havecoexisted for generations and that adventitiouspresence is recognized as an inevitableconsequence of coexistence that can be minimizedbut not entirely eliminated. Therefore,almost all traded agricultural commoditiesanticipate some degree of inadvertent mixing,and thresholds exist that are recognizedin laws, regulations and/or voluntarystandards.Such thresholds have resulted in thedevelopment of a series of measures that areapplied during cultivation, harvest, transportand storage to minimize outcrossing, thegrowth of volunteer plants and inadvertentmixing 3 . These best practices were establisheddecades ago and have evolved to deliver highpurity seed and grain to support the production,distribution and trade of products fromdifferent agricultural systems. The principlesof these coexistence practices are dependenton context (which crops and where they aregrown), consistent, proportionate to need,fair and practical. Examples of successfulcoexistence practices in non-GM agricultureinclude production systems for certified seeds(e.g., hybrid seed), organic crops coexistingwith conventional crops and commoditycrops coexisting with specialty crops (e.g.,field corn with sweet corn and/or popcorn,and specialty corns such as high-amylose,high-oil, white, waxy, hard endosperm andnutritionally dense varieties) 6 .Perhaps one of the best-studied examplesof coexistence in conventional agricultureis standard rapeseed varieties and specialtyhigh erucic acid rapeseed (HEAR) varietiesfor industrial use, particularly because HEARis regarded as antinutritional and undesirablein food (and therefore constitutes an actualrisk rather than a consumer preference, as isthe case for GM crops). Contracts for growingHEAR crops require that only certifiedHEAR seed is used, equipment should becleaned and segregated and that there shouldbe an isolation distance of between 50 m (e.g.,in the UK) and 100 m (e.g., in Germany)from other rapeseed crops. The admixturethreshold for HEAR in food rapeseed is 2%although recorded levels are usually muchlower. For example, the 100-m separation distancein Germany generally delivers seed lotswith HEAR levels 0.5%. In the UK, coexistenceresearch shows that separation distances aslow as 9 m still provide bulk rapeseed harvestscontaining

COMMENTARY© 2010 Nature America, Inc. All rights reserved.Spain is arguably the most enthusiasticadopter of GM agriculture in the EU, allowingthe cultivation of GM crops without acomplete regulation regime. The establishmentof coexistence rules has been preventedby disputes between the Spanish Ministry ofAgriculture (influenced by farmers’ lobbies)and the Ministry of Environment (influencedby ecological lobbies). Coexistenceis currently determined by seed companyguidelines together with some specificregulations 12 , but there are no compulsorytraining courses, no specific liability rulesand 50-m isolation distances are standard 13 .Despite successful coexistence in Spain,market forces have created region-by-regionsegregation. In the productive agriculturalregions of Catalonia and Aragon, 55% and42% of corn, respectively, is GM 14 . In contrast,Asturias and the Basque Country havedeclared themselves GM free with the supportof regional governments and somefarmers’ associations.Several EU member states require farmersto gain official approval before they areallowed to plant GM crops. In Austria, farmersneed approval from local authorities foreach field and crop (similar procedures arebeing considered in Hungary, Ireland andthe Slovak Republic). Austria has the strictestregime, and even though there are somecoexistence measures (zero-risk seed purityregulation, compulsory training coursesand strict liability policies), the Austrianauthorities are against GM crops and striveto avoid coexistence instead of promotingit 15 . Austrian provinces have approachedthe EU to establish GM-free regions, butin September 2007 the European Court ofJustice finally rejected general statutoryregional bans on GM crops, arguing that astatutory ban is a denial of the freedom ofchoice for farmers and consumers 16 . Polandand Belgium are also seeking to avoid thedeployment of GM crops (120 communitiesin Belgium have already declared themselvesGM free). Portugal has a complete system ofregulation (established before commercialplanting) with compulsory training courses,strict anti-cross-pollination measures and apublic compensation fund. Even so, this stillallows some flexibility in isolation measuresdepending on voluntary agreements amongneighbors. This kind of collective initiativeavoids complicated anti-cross-pollinationmeasures and expensive double farmfacilities.How far is far enough?EU coexistence guidelines (Recommendation2003/556/EC) state that “…Managementmeasures for coexistence should reflect thebest available scientific evidence on the probabilityand sources of admixture between GMand non-GM crops…” 5 , but it is quite clearthat this recommendation is being ignoredin many EU member states. Some countriesrequire vast isolation distances that bear norelationship to the underpinning scientificevidence. For example, Luxemburg requires800 m between GM and non-GM corn and3 km between GM and non-GM rapeseed.Latvia requires 4 km between GM and conventionalnon-GM rapeseed and 6 km if thenon-GM rapeseed is organic (SupplementaryTable 1). Such isolation distances imposeimmense costs on GM farmers because theyhave to negotiate with a much larger numberof neighboring farms and, in practicalterms, simply remove their choice in relationto adopting GM crops 17 .The minimum isolation distances imposedon GM producers in the EU should be thosethat are sufficient to maintain the adventitiouspresence of GM material below 0.9% inneighboring organic and conventional plots.The current isolation distances were basedon the assessment of biological and physicalprocesses that affect outcrossing 3,4,18,19 , andthese tend to differ between studies if factorssuch as pollen viability; male sterility; floweringsynchrony; wind speed and direction;weather conditions; field size and shape; anddistance, topography and vegetation betweenthe pollen donor and recipient fields are notstandardized. Maize pollen is released in verylarge quantities, between 4.5 and 25 millionpollen grains per plant over a typical 5- to8-day period 20 , but is larger (90–125 µm)and heavier than the pollen of most otherwind-pollinated plants, and therefore dispersalis limited to about 10% of the range coveredby other species, often settling within afew hundred meters of its source.A research study conducted by the SpanishInstitute for Agriculture & Food Research andTechnology (Madrid) demonstrated that infield trials, the average presence of the Btgene in conventional maize separated from Btmaize by just 2–10 m is

COMMENTARY© 2010 Nature America, Inc. All rights reserved.Note: Supplementary information is available on theNature Biotechnology website.ACKNOWLEDGMENTSResearch in our laboratory is funded by MICINN(Ministry of Science and Innovation), Spain(BFU2007-61413); European Union Framework 6Program – The Pharma-Planta Integrated Project LSH-2002-1.2.5-2; European Union Framework 7 Program-SmartCell Integrated Project 222716; European UnionFramework 7 European Research Council IDEASAdvanced Grant (to P.C.) Program-BIOFORCE;Acciones Complementarias (MICINN) BIO2005-24826-E. Centre CONSOLIDER on Agrigenomicsfunded by MICINN, Spain.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.1. James, C. International Service for the Acquisition ofAgri-biotech Applications Brief No 37 (ISAAA, Ithaca,NY, USA, 2008).2. Brookes, G. & Barfoot, P. Co-existence in North AmericanAgriculture: Can GM Crops Be Grown with Conventionaland Organic Crops? (PG Economics Ltd., Dorchester,UK; 2004).3. Devos, Y. et al. Agron. Sustain. Dev. 29, 11–30 (2009).4. Sanvido, O. et al. Transgenic Res. 17, 317–335(2008).5. European Commission. Off. J. Eur. Commun. L189,36–47 (2003).6. Brookes, G. Coexistence of GM and Non-GM Crops.Current Experience and Key Principles (PG EconomicsLtd., Dorchester, UK; 2004).7. Kalaitzandonakes, N. in Proceedings to the SecondInternational Conference on Co-existence BetweenGM and non-GM-based Agricultural Supply Chains,Montpellier, France, November 14–15, 2005 (ed.Messéan, A.) 29–30 (Agropolis Productions, 2005).8. Ramessar, K., Capell, T., Twyman, R.M., Quemada, H.& Christou, P. Nat. Biotechnol. 26, 975–978 (2008).9. Ramessar, K., Capell, T., Twyman, R.M., Quemada, H.& Christou, P. Mol. Breed. 23, 99–112 (2009).10. European Commission. Off. J. Evr. Commun. L268,24–28 (2003).11. Abbott, A. & Schiermeier, Q. Nature 450, 928–929(2007).12. Asociación Profesional de Empresas Productoras deSemillas Selectas (APROSE). Guía 2007 de BuenasPrácticas para el Cultivo de maíz Bt. (AsociaciónProfesional de Empresas Productoras de SemillasSelectas, Madrid, Spain; 2007).13. Corti-Varela, J. in Proceedings to the 2 nd AnnualCambridge Conference on Regulation, Inspection andImprovement, Cambridge, UK, September 11–12,2007. 14. Binimelis, R. J. Agric. Environ. Ethics 21, 437–457(2008).15. Beckmann, V., Soregaroli, C. & Wesseler, J. Am. J. Agric.Econ. 88, 1193–1199 (2006).16. EuropaBio. EU Court rejects Austrian biotech ban—supportsright to choose biotech crops (EuropaBio, Brussels) (14 September 2007).17. Messean, A., Angevin, F., Gomez-Barbero, M., Menard,K. & Rodriguez Cerezo, E. in Technical Report Series ofthe Joint Research Center of the European Commission(Institute for Prospective Technological Studies, Sevilla,Spain; 2006). 18. Aylor, D.E. Agric. For. Meteorol. 123, 125–133(2004).19. Devos, Y., Reheul, D. & de Schrijver, A. Environ.Biosafety Res. 4, 71–87 (2005).20. Paterniani, E. & Stort, A.C. Euphytica 23, 129–134(1974).21. Brookes, G. et al. GM Maize: Pollen Movement andCrop Co-existence. (PG Economics Ltd., Dorchester,UK; 2004).22. Luna, V.S. et al. Crop Sci. 41, 1551–1557 (2001).23. Cruz de Carvalho, P. (coordinator). Coexistência EntreCulturas Geneticamente Modificadas e Outros Modosde Produção Agrícola. Relatórios de Acompanhamento.(Ministério da Agricultura, do DesenvolvimentoRural e das Pescas/Direcção-Geral de Agricultura eDesenvolvimento Rural, Lisbon, 2009). 24. Scientific Committee on Plants (SCP). Opinion ofthe Scientific Committee on Plants Concerning theAdventitious Presence of GM Seeds in ConventionalSeeds (European Commission, Brussels, Belgium;2001).136 volume 28 number 2 february 2010 nature biotechnology

patentsChanging the rules of the game: addressing theconflict between free access to scientific discoveryand intellectual property rightsMiriam BentwichA provisional patented paper application procedure could promote earlier disclosure of novel scientific knowledge.© 2010 Nature America, Inc. All rights reserved.major constraint on the advancementA of scientific research in biotech is thegrowing barriers to free exchange of scientificknowledge associated with the increasednumber of patents linked to intellectualproperty rights (IPR). This matter, whichcan be tagged as the ‘secrecy threat’, has beenaddressed in numerous publications 1–7 , andcan be described as essentially revolvingaround two issues. One issue concerns thepossible reluctance of patent applicants andowners to fully disclose their acquired novelscientific knowledge through early and easilyaccessible scientific publications. The otherissue was broadly termed the “tragedy of theanticommons” by Eisenberg and Heller andhas been further developed by other scholars8–10 . It involves the possible constrainingof other scientists from using patented novelknowledge for further scientific research foreconomic reasons.Indeed, numerous empirical studies haveshown that scientists delay scientific publicationof research they expect to patent and arereluctant to freely share research results andmaterials with other scientists 11–15 . Otherstudies and judicial decisions have shownthat, contrary to its intended purpose 6,16,17 ,the patent system does not necessarily offera practical incentive for patent applicantsto disclose much information regardingtheir inventions in a comprehendible manner18–21 . In fact, the notion of prior art, socrucial in determining the eligibility of a patentapplication with respect to its novelty, atbest, significantly limits the grace period inMiriam Bentwich is in the Department ofPolitical Science, The Hebrew University,Mt. Scopus, Jerusalem, Israel.e-mail: mbentwich@gmail.comwhich potential patentees may publicly discloseinformation regarding their inventionbefore a corresponding patent application issubmitted 22–24 . Finally, additional case-inpointresearch illuminates instances such asthat of Myriad Genetics’ policy regarding itspatented breast cancer genes and their mutations,where patentees exploited their IPR toincrease their profits at the direct expense ofsharing novel scientific knowledge 10,25–27 .Admittedly, there have already been extensiveand contrasting attempts to address thechallenge at hand 3,28–30 , the bulk of whichrevolves around practical steps that may betaken to mitigate the secrecy threat and itsnegative consequences for scientific researchwithout deserting the patent system 20,31–33 .However, these practical steps appear to beflawed with respect to the two main concernsassociated with the secrecy threat.Thus, suggested practical measures againstthe tragedy of the anticommons, such aspatent pools and governmental compulsorylicensing for use of patents, lack a theoreticaljustification from an IPR advocate’s viewpoint.Meanwhile, other suggested practicalsteps, like extending the grace period for ‘selfprior art’ or using the US provisional patentapplication (PPA) option more extensively,could allow potential patentees to publishtheir scientific knowledge at an earlier stageand in a fuller manner. Yet these practicalsteps have so far neither obligated norstimulated potential patentees to do so inpractice.We present here an alternative and, wehope, more successful approach to tacklethe secrecy threat without undermining thepatent system. This is achieved by a revisedmandatory version of PPAs, designed for patentsin biotech and administered by leadingscientific journals, which would be officiallyrecognized by the US Patent and TrademarkOffice (USPTO) and subsequently by otherpatent offices as well.From optional PPAs to mandatory PPPAsSince 1995, the USPTO has offered theoption to submit provisional patent applications34 that are easier and substantiallycheaper to file than a full-fledged patentapplication 35–37 . Submitting a PPA is easierbecause, unlike regular patent applications,PPAs do not require the inclusion of the patent’sclaims, which define the extent of theprotection sought in a patent application indetailed technical terms. Additionally, PPAsdo not necessitate discussion of relevant priorart or use of applicable legal terms. Instead,PPAs need only adequately describe theinvention and its scope 38,39 . As a result, thistype of patent application lacks fundamentalingredients of any full-fledged patent application,and therefore, patents cannot evolvedirectly from PPAs.However, PPAs allow potential patenteesto secure an earlier valid ‘priority date’(that is, official filing date) for their futurepatented invention before the latter is fullyprepared, thereby decreasing the chances ofpossible competition over their inventions 36 .Such an advantage, though, is granted only iftwo conditions are met. First, the nonprovisionalpatent application must be submittedno later than 12 months past the filing dateof the PPA. Second, the provisional applicationmust adequately provide a writtendescription of the full scope of the inventionexplicated in the later full application 36,39 .Additionally, when these two conditions aremet, and a corresponding non-US patentapplication is submitted within 12 monthsnature biotechnology volume 28 number 2 february 2010 137

patents© 2010 Nature America, Inc. All rights reserved.How the PPPA procedure worksStep 1: Creating required components for PPPA through online paper submission systemOptionaldata filesSubmittedmanuscriptGenerate DTScertification(s)Step 2: Full patent application submissionFull patentapplicationProof of articlepublicationStep 3: PPPA validity determination by USPTOCondition 1: FPA due no laterthan 12–18 months pastDTS certification dateafter the filing date of the US PPA, in principle,the filing date of the PPA may also serveas the priority date for the non-US patentapplication. Consequently, US PPAs can beused in conjunction with other patent offices(e.g., the European Patent Office and JapanPatent Office), thereby potentially expandingthe grant of IPR beyond US soil.As already mentioned, although PPAstheoretically enable their submitters to sharetheir novel knowledge with other researchers,they do not obligate patent applicants to doso. In fact, because they are not full patentapplications, PPAs constitute an anti-incentivefor sharing novel scientific knowledgeoutside the patent system. For instance, if apatent applicant is unable to submit the fullpatent application within the allowed oneyearperiod, the applicant will lose the abilityto patent the invention, at least insofar as itis explicated in the early publication as thelatter now constitutes patent-blocking priorart 22 .The importance of the PPA procedureis that it opens the door to a contingentacknowledgment of underdeveloped andinformally submitted patent applications.Charge applicablefundsSubmitted manuscript& optional data filesSend author DTScertification filesDTS certificationfilesCondition 2: Submitted manuscript andoptional data files sufficiently disclose theinvention described in FPAFigure 1 A graphical outline demonstrating how the PPPA procedure can be practically employed.PPPA, provisional patented paper application; DTS, digital time-stamping service; FPA, full patentapplication; USPTO, US Patent and Trademark Office.Hence, PPAs are informally submitted in thesense that, for themselves, they are neitherpublished nor examined by the USPTO; andthey are underdeveloped (by design) becausethey purposely accept patent applicationslacking fundamental ingredients like claimsand the discussion of prior art, as long as theseapplications contain a sufficient descriptionof the invention and its scope. Consequently,in principle, a scientific paper that includesa description of an invention’s purpose andthe method(s) used to attain it may qualifyas a PPA. What’s more, insofar as the contentand submission date of such a paper canbe verified, the paper need not be physicallysubmitted to USPTO before the filing of thefull patent application, because other thanrecording their existence, USPTO does notexecute any action with regard to PPAs.From this perspective, therefore, scientificpublications would not constitute an externalprior art, undermining the eligibility of patentapplications. Instead, such publications,if recognized as provisional patented paperapplications (PPPAs), would provide the verysame protection that traditional PPAs granta future intended patent, including theirsupposed compatibility with the regulationsof non-US patent offices. At the same time,contrary to the existing model of PPAs, thesuggested PPPA alternative version will obligatepatent applicants to share their novelknowledge with colleagues through scientificpublications (Fig. 1).Thus, given the proper tools for date andcontent verification, PPPAs may be entirelyadministered by peer-reviewed academicjournals instead of the USPTO, therebyfurther binding such provisional patentapplications to scientific publications, whilesignificantly lowering the fee charged for theseapplications. Moreover, because the scientificpublication constitutes either the very PPPAor at least amounts to a significant integralportion of it (as will be explained later), thereis a direct incentive for potential patenteesto fully disclose the information concerningtheir inventions in the corresponding scientificpublications. Otherwise, they risk theireligibility to use the PPPA as the basis for thefull patent application, thereby underminingtheir ability to be granted with the desiredpatent and its conferred IPR. Similarly,potential patentees would be interested indisclosing that information as early as possiblebecause any delay in the submission ofPPPAs naturally increases the risk of potentialcompetition over their inventions.Meanwhile, as the suggested PPPA procedureis inseparable from scientific publications,the potential patentee might very wellchoose not to use this procedure. Therefore,we suggest that the PPPA be defined as anobligatory step on the way to filing a full patentapplication in biotech. This being thecase, it might also be advisable to extend thedeadline for filing the full patent applicationfor, say, another six months, so that potentialpatentees would have a larger safety net, coveringinstances where the paper is not immediatelyaccepted.Moreover, paradoxically, the supposedreluctance of potential patentees to usePPPAs, should this be left to their choice,appears to be grounded on a sort of prisoner’sdilemma, where the assumed leastrisky choice is conflated with and mistakenlypreferred over the best outcome choice, dueto lack of cooperation 40,41 . Hence, lackingknowledge regarding the decisions of otherpotential patentees concerning the use ofPPPAs and their entailed extra risks, a potentialpatentee would not want to take theseextra risks. For, by taking such extra risks, shemight put her invention in a more vulnerableposition than competitors who choose not touse PPPAs. Yet, theoretically, if all potentialpatentees agreed (that is, cooperated among138 volume 28 number 2 february 2010 nature biotechnology

patents© 2010 Nature America, Inc. All rights reserved.themselves) to use PPPAs before their fullpatent applications were submitted, such anagreement would result in a much better outcomefor all involved parties.By reaching this agreement, all PPPAsubmitters would be granted the followingbenefits:1. Elimination of the extra risks associatedwith PPPAs, as the latter will entail nowmerely standard risks shared by all potentialpatentees.2. Preliminary professional assessments of theinvention’s validity through the journal’speer-review process, thereby indicatingwhether the invention is worth investing in.3. An earlier ‘certificate of quality’, should thescientific paper describing the invention beaccepted, thereby potentially increasing theability to raise more funds.4. Possible advice for prepatenting improvementsto the invention, leading to a potentialincrease in the patent’s value.Mandatory PPPAs are, therefore, justifiable,even from the viewpoint of potentialpatentees, while being an acceptable form ofprovisional patent application (PPA). At thesame time, PPPAs address the first portionof the secrecy threat, regarding the possiblereluctance of potential patentees to publiclyshare new scientific knowledge in a full andeasily accessible manner.From theory to practiceAssuming the employment of mandatoryPPPAs is logically plausible, even from apotential patentee’s standpoint, it is still necessaryto explain whether and how they canbe applied in practice. Essentially, enablingvalid and effective PPPAs necessitates thatscientific publications provide papers’ submitterswith trustworthy certifications of thedate and content of the relevant informationthat is needed to constitute such PPPAs. Thatway, though PPPAs are not administered bya patent office, but rather by academic journals,PPPAs would be amply reliable for useby both potential patentees and the USPTO.Accordingly, to supply such trustworthy certifications,academic journals must complywith the following three requirements:1. Provide authors, upon their request, withthe necessary certification of the papersubmitted for publication that describesthe invention for which a patent may besought.2. Grant authors, upon their request, a separatecertification for additional optionaland unpublished data that they wish toinclude as part of the PPPA. Such certifiedbut unpublished data contain materialsthat are not suitable for a scientific paper,but might be technically important forsecuring the full patent application (e.g.,certain claims or additional figures).3. Ensure that the granted certifications canbe independently verified by the USPTOin a simple manner, once full patent applicationsare submitted. That is, allow theUSPTO to (i) check the authenticity ofthese certifications and (ii) ensure that,insofar as the content of the certifiedmanuscript is concerned, it is adequatelycovered by the finalized published paper.All of the aforementioned requirementscan be realized by a combination of two existingand widely accessible technologies. Onetechnology is the digital time-stamping service,issuing a trusted third-party time-stampthat associates a date and time with a digitaldocument in a cryptographically strong way.The digital time-stamp can be used at a laterdate (e.g., by USPTO) to independently verifythat an electronic document with a particularcontent existed at the time stated on its timestamp,thereby certifying the document’s contentand creation date 42,43 . In this respect, it isimportant to note two points. First, a digitaltime-stamp, understood as a specific type ofdigital signature 44 , is legally supported andrecognized by applicable legislations in manycountries worldwide, and specifically by theUnited States 45 . Second, commercial digitaltime-stamping services 46 are already availableon the Internet with costs as low as 40 centsper use, making the financial barriers negligibleto end-users.The other technology is the widelyemployed online paper submission processingsoftware (e.g., Editorial Manager, ManuscriptCentral) already used, according to a recentstudy, by 76% of the peer-reviewed academicjournals, and still growing 47 . The significanceof these web-based computer programs istwofold. First, they can provide the optionof date and content certification within thepaper submission process, through seamlesslyembedding the already available digitaltime-stamping toolkits that were designed tobe set-in within a third-party software environment.Second, these programs include abundle of features like financial transactionand automatic e-mail communication withthe submitters of papers 48 . As such, thesefeatures provide further necessary supportfor employing PPPAs, like the ability to sendauthors their needed certified documents andtheir finalized articles’ proofs, as well as collectingthe applicable fees for USPTO and thedigital time-stamping service.PPPAs limit the effects of the tragedy ofthe anticommonsOnce used, PPPAs can also provide an adequatetheoretical justification, from a patents’advocate viewpoint, for significantly limitingthe negative effects of the tragedy of theanticommons. This theoretical rationalizationis based on patentees’ interest in securingthe main benefit that is gained by theirpatents, namely IPR, as patents effectivelygrant patentees exclusive property rights overtheir inventions, even if for a limited periodof time. The original notion of exclusiveproperty rights and its justification is usuallyattributed to John Locke, the forefatherof classical liberalism, though he simply usedthe term ‘property rights’ 49 . In fact, Locke’sframing of property rights is still influentialamong contemporary libertarians and neoliberals,who usually perceive themselves ashis successors, and therefore Locke’s thoughtis still relevant for contemporary discussionsof exclusive property rights 50–52 .According to Locke, the initial justificationfor private exclusive property rights stemsfrom an individual’s “mixing of his labor” ina previously unowned object. Thus, an individual,naturally, has exclusive ownershiprights over his own labor, namely, it is hisproperty. Consequently, by mixing his labor,it becomes inextricably part of the object,thereby rendering anyone else who wouldthen use that object as effectively infringingthe property that the first person has‘invested’ in the object through his labor. Byannexing her labor to the object, therefore,the individual restricts others’ right to use it,making it exclusively her own 49,53 .Acquiring a previously unowned object,then, necessitates that the new owner bethe sole investor of labor in this object (beit directly from his own work, or indirectlythrough transfer of other people’s labor tohis disposal). Inversely put, so long as otherpeople have invested their direct or indirectlabor in the object at hand as well, withoutbeing compensated for their loss of proportionateinvestment, no one can claim exclusiveproperty rights over the object.From this perspective, therefore, a patent,pertaining per se to a new invention, involvesan ownership of a previously unowned object,and may be granted to the patent applicantonly if at the time the patent was granted,the applicant (and his party) could benature biotechnology volume 28 number 2 february 2010 139

patents© 2010 Nature America, Inc. All rights reserved.considered as the direct or indirect solelaborer(s) over the invention. However,by using PPPAs, potential patentees enjoythe labor that the anonymous referees haveinvested or mixed in the knowledge underlyingthe inventions intended to be patented.Similar to Locke’s example of a person’smixing his labor in a natural and unownedland, thereby cultivating it and rendering itas more profitable, the anonymous referees’work, as already indicated above, also assistsin increasing the value of the knowledge atthe basis of the intended patent.Consequently, in principle, a patent applicanthas to ‘purchase’ the referees’ labor inenhancing the knowledge on which theinvention is based, so that the applicantwould be the only party to have invested herlabor in the invention she is seeking to patent.Yet because the referees are anonymous,the patent applicant cannot transfer thepayment to a particular person or group ofpeople. Moreover, the referees themselves areable to perform their job (that is, labor) basedon the current relevant scientific knowledge,acquired through the previous labor of otherscientists, who have gained their knowledgeon the grounds of their predecessors, and soon.In other words, only through compensatingthe scientific community as a whole canthe potential patentee buy the labor, investedby other parties that are affiliated with thiscommunity in the theoretical knowledge,underlying the invention for which a patentis sought. This compensation would be madeby unequivocally guaranteeing the community’smembers an inexpensive license to usethe theoretical knowledge behind the inventionto which the community has contributed.Consequently, such a guarantee wouldbe applicable, for example, to theoreticalknowledge concerning particular mappedgenes and their mutations (e.g., BRCA1 andBRCA2), but not to specific devices and servicesthat further use this knowledge, as in thecase of a breast cancer diagnostic kit. Still, itwas precisely with regard to the type of patentassociated with BRCA1 and BRCA2 and itspotential damage to the free flow of novel scientificknowledge that the concern about thetragedy of the anticommons was expressedin the first place.Meanwhile, notice that the argumentadvanced here is not only theoretically plausible,but also practically allowable. Thus,the requirement to furnish other scientistswith cheap licenses to use the theoreticalknowledge underlying a patented inventiondecreases the patentee’s income fromsuch licensing. However, it also decreasesthe same patentee’s expenditures on gettingsimilar licenses to other patentees’ theoreticalknowledge. Moreover, the justified compensationscope is merely limited to theoreticalknowledge used only within the conduct ofresearch. Therefore, the pledge to inexpensivelicensing for research endeavors neithercovers using services and devices that usethis theoretical knowledge, nor permits sellingthem without the explicit agreement ofthe patentee, thereby preserving the overallintegrity of patentees’ IPR and their conferredprivileges.ConclusionsWe have presented a novel and relativelyeasy solution to the secrecy threat entailedin patents and their undesired effects on thefree access to scientific knowledge and to theadvancement of biotechnological research.By suggesting a mandatory PPPA procedure,largely based on the already available andinternationally recognized PPA procedure,we have demonstrated how the two mainconcerns associated with the secrecy threatcan be addressed. We have shown how PPPAsmay promote earlier and fuller disclosuresof novel scientific knowledge and providean ample justification for requiring patenteesto grant inexpensive licenses for use oftheir inventions by other parties conductingresearch, while protecting the integrity ofpatents and IPR.COMPETING INTERESTS STATEMENTThe author declares no competing financial interests.ACKNOWLEDGMENTSI am grateful to Zvi Bentwich and Isaac (Zahon)Bentwich for their invaluable comments on a previousversion of this article. I also thank Moshe Maor andAvner de-Shalit for preliminary discussions regardingthe article’s theme. Finally, I would like to expressmy gratitude to the Lady Davis Trust at the HebrewUniversity for its generous financial support thatenabled the pursuit of this study.1. Andrews, L.B. Nat. Rev. Genet. 3, 803–808 (2002).2. Caulfield, T., Cook-Deegan, R.M., Kieff, F.S. & Walsh,J.P. Nat. Biotechnol. 24, 1091–1094 (2006).3. Cook-Deegan, R., Chandrasekharan, S. & Angrist, M.Nature 458, 405–406 (2009).4. Cukier, K.N. Nat. Biotechnol. 24, 249–251 (2006).5. David, P.A. J. Inst. Theor. Econ. 160, 9–34 (2004).6. Eisenberg, R.S. Yale LJ 97, 177–231 [225] (1987).7. Holman, C.M. Trends Biotechnol. 25, 539–543(2007).8. Heller, M.A. & Eisenberg, R.S. Science 280, 698–701(1998).9. Shapiro, C. in Innovation Policy and the Economy, vol. 1(eds. Jaffe, A.B., Lerner, J. & Stern, C.) 577–579 (MITPress, Cambridge, Massachusetts, 2001).10. Merz, J.F., Kriss, A.G., Leonard, D.G.B. & Cho, M.K.Nature 415, 577–579 (2002).11. Blumenthal, D., Campbell, E.G., Anderson, M.S.,Causino, N. & Louis, K.S. J. Am. Med. Assoc. 277,1224–1228 (1997).12. Schissel, A., Merz, J.F. & Cho, M.K. Nature 402, 118(1999).13. Campbell, E.G. et al. J. Am. Med. Assoc. 287, 473–480 (2002).14. Grushcow, J.M. J. Leg. Stud. 33, 59–84 (2004).15. Murray, F. & Stern, S. J. Econ. Behav. Organ. 63,648–687 (2007).16. 35 US Code § 112, 113.17. Walsh, J.P., Cho, C. & Cohen, W.M. Science 309,2002–2003 (2005).18. Brenner v. Manson 383 US (1996), 519, 534.19. Vitronics Corp v. Conceptronic Inc. 90 F3d (Fed. Cir.1996), 1576, 1583.20. Fromer, J.C. Iowa Law Rev. 94, 539–606 (2009).21. Long, P. Univ. Chic. Law Rev. 69, 625–680 (2002).22. 35 US Code § 102(b).23. Tokkyo Ho [Patent Law] § 30(31) & 30(33).24. Anonymous. European Patent Convention. Article 54(European Patent Office, Munich, 1973).25. Benowitz, S. J. Natl. Cancer Inst. 94, 80–81 (2002).26. Borger, J. Rush to patent genes stalls cures for disease.Guardian (London) 15 December 1999, p.1.27. Williams-Jones, B. Health Law J. 10, 123–146(2002).28. Caulfield, T., Gold, E.R. & Cho, M.K. Nat. Rev. Genet.1, 227–231 (2000).29. Fabrizio, K.R. & Di Minin, A. Res. Policy 37, 914–931(2008).30. Petherbridge, L. Maine Law Rev. 59, 339–384(2007).31. Bagley, M.A. Boston Coll. Law Rev. 47, 217–274(2006).32. Ebersole, T.J., Guthrie, M.N. & Goldstein, J.A. Intellect.Prop. Technol. Law J. 17, 6–13 (2005).33. Franzoni, C. & Scellato, G. Proc. Acad. Innov.Entrepreneurship 2008, 388–401 (2008).34. Anonymous. Federal Register. 73, 47535, 47540(2008).35. http://www.uspto.gov/web/offices/pac/provapp.htm36. 35 US Code § 119(e), 120.37. 35 US Code § 102(e).38. 35 US Code § 111(b).39. 35 US Code § 112 (first paragraph).40. Axelrod, R. Am. Polit. Sci. Rev. 75, 306–318(1981).41. Rapoport, A. in Game Theory as a Theory of ConflictResolution (ed. Rapoport, A.) 17–34 (Springer-Verlag,New York, 1974).42. Cipra, B. Science 261, 162–163 (1993).43. Skevington, P.J. & Hart, T.P. BT Technol. J. 15, 39–44(1997).44. http://www.digistamp.com/FAQts.htm#legality45. Feng, H. & Wah, C.C. Inf. Manage. Comput. Secur. 10,159–164 (2002).46. http://www.guardtime.com; http://www.digistamp.com47. Tananbaum, G. & Holmes, L. Learn. Publ. 21, 300–306 (2008).48. Wood, D. Learn. Publ. 14, 151–158 (2001).49. Locke, J. in Two Treatises of Government (ed. Lasslet, P.)Second Treatise §5 (Cambridge University Press,Cambridge, 1967).50. Machan, T.R. in Liberty for the Twenty-First Century:Contemporary Libertarian Thought (ed. Machan, T.R.)209–226 (Rowman and Littlefield, Lanham, Maryland,1995).51. Narveson, J. in Liberty for the Twenty-First Century:Contemporary Libertarian Thought (ed. Machan, T.R.)19–40 (Rowman and Littlefield, Lanham, Maryland,1995).52. Nozick, R. Anarchy, State and Utopia (Basic Books,New York, 1974).53. Hailwood, S.A. Exploring Nozick: Beyond Anarchy,State and Utopia (Avebury, Aldershot, UK, 1996).140 volume 28 number 2 february 2010 nature biotechnology

patents© 2010 Nature America, Inc. All rights reserved.Recent patent applications in antibody fragmentsPatent number Description Assignee InventorWO 2009126730,WO 2009126730US 20090297439FR 2931481,WO 2009141458WO 2009138714FR 2929519,WO 2009136031US 20090269277WO 2009129521WO 2009127046WO 2009099545,US 20090202557WO 2009092014KR 2009011215A method of preparing nucleotides of single-chain variable fragmentsencoding an antigen-specific binding domain by amplifying the variableregions of the antibody’s heavy chain and the lambda and kappa lightchains using PCR with a set of primers.An immuno-imaging agent for the detection of a tumor in a subject,comprising an anti-Met monoclonal antibody, its fragment and a geneticallyengineered/humanized antibody containing the epitope bindingregion or complementarity-determining regions of the antibody.New isolated antibodies or at least one of their functional fragments,which is specific to an epitope comprising at least one lysyl compounduseful, e.g., for detecting epitope in a sample in vitro or in vivo.A method for the separation of a fragment antibody, e.g., single-chainvariable fragment, involving contacting a medium containing the fragmentantibody with a synthetic affinity ligand attached to a support matrix underconditions where the fragment antibody binds to the ligand.Use of a monoclonal antibody secreted by a hybridoma or its functionalfragments to prepare a medicament to inhibit the growth of a primarytumor for early cancer treatment where the cancer is, e.g., colon, lungprostate cancer.A method for delivering an agent useful for diagnosing or treating, e.g.,cancer or cardiovascular disease, by administering a hexameric stablytethered structure comprising the agent, an IgG antibody and antibodyfragments or cytokines to the subject.An antigen composition for early detection of Mycobacterium tuberculosisor immunizing against infection, comprising, e.g., M. tuberculosisproline threonine repetitive protein fragments having specific sequencesthat bind antibody specific for protein.A new antibody or an antigen-binding fragment comprising a complementarity-determiningregion Gly-X1e-X2e-X3e-X4e-X5e-X6e-X7e-X8e-His(SEQ ID NO. 65); useful for reducing the growth of prostate tumor cells.A method of preparing crystals of an antigen-binding fragment (Fab) ofan antibody, comprising mixing the Fab with a reservoir solution comprisingpolyethylene glycol (PEG) and a buffer.Purifying a nonaggregated antibody or an immunoreactive antibody fragmentfrom an impure preparation containing the antibody or antibodyfragment, comprising contacting the impure preparation with an apatitechromatography support.A surface expression vector expressing short-chain variable fragmentsof porcine epidemic diarrhea virus (PEDV)-neutralizing antibodyencoded by a gene having a defined sequence of 777 amino acids(Seq. id. no. 1) on membrane surface of bacteria; useful for expressingshort-chain variable fragments of PEDV-neutralizing antibody on thesurface of Escherichia coli, where the bacteria is useful in compositionfor preventing or treating diarrhea.University ofPennsylvania(Philadelphia)MetheresisTranslationalRsesearch(Lugano,Switzerland)Covalab(Villeurbanne,France)Avecia Biologics(Manchester, UK)PriorityapplicationdatePublicationdateMason N 4/9/2008 10/15/2009,12/30/2009Carminati P,Comoglio PM,van Dongen GCeylan I,El Alaoui EBS,Thomas V6/2/2008 12/3/20095/23/2008 11/27/2009,11/26/2009Liddell JM 5/16/2008 11/19/2009Pierre Fabre Haeuw JF 4/4/2008 10/9/2009,Medicament11/12/2009(Boulogne, France)IBCPharmaceuticals(Morris Plains, NJ,USA)New YorkUniversity(New York)ProScan RxPharma (Montreal)Abbott Bioresearch(Worcester, MA,USA) Argiriadi MA,Borhani DW,Ghayur T, Wu C,Xiang TChang C,Goldenberg DM,Rossi EALaal S,Zolla-Pazner SCuello AC,Gold P,Melancon D,Moffett S,Saragovi HUArgiriadi MA,Borhani DW,Ghayur T, Wu C,Xiang T10/19/2005 10/29/20094/19/2008 10/22/20094/14/2008 10/22/20091/30/2008 8/13/2009,8/13/2009Gagnon PS Gagnon PS 1/18/2008 7/23/2009Republic ofKorea Ministry ofAgriculture andForestry (Seoul)Cho S, Hyun B,Kim I, Kim S,Pyo H, Song J7/25/2007 2/2/2009Source: The status of each application is slightly different from country to country. For further details, contact Thomson Scientific, 1800 Diagonal Road, Suite 250, Alexandria, Virginia22314, USA. Tel: 1 (800) 337-9368 (http://www.thomson.com/scientific).nature biotechnology volume 28 number 2 february 2010 141

news and viewsChIPs and regulatory bitsXin He & Saurabh SinhaMachine learning reveals combinatorial patterns of transcription factor binding that drive gene expression.© 2010 Nature America, Inc. All rights reserved.Gene expression patterns are orchestrated largelyby promoters and enhancers, which integratea multitude of signaling and transcriptionalinputs 1 . But exactly how these cis-regulatorymodules control transcription on a global scalehas been devilishly hard to decipher. A recentstudy by Zinzen et al. 2 in Nature marks excitingprogress on this problem. The authors’ key innovationis a technique to predict the expressionactivity driven by any genomic segment usingtranscription factor occupancy data generatedfrom high-throughput chromatin immunoprecipitation(ChIP) assays. The approach led tothe discovery of dozens of novel cis-regulatorymodules involved in mesoderm and muscledevelopment in the Drosophila melanogasterembryo, while providing new insights into theunderlying regulatory code.In parallel with experimental studies, therehave been two main computational directionsfor explaining the mode of action of cis-regulatorymodules—that is, how several transcriptionfactor binding sites in a module integrate theintracellular concentrations of transcription factorsto determine gene expression. Approachesbased on fundamental physical-chemical principleshave made significant headway in predictingfunction directly from sequence 3,4 buthave been limited by gaps in our understandingof the underlying molecular mechanisms,such as the combined effects of simultaneousprotein-DNA and protein-protein interactions.In addition, statistical approaches have beenused to associate specific expression states withthe patterns of binding site occurrence sharedby the corresponding regulatory sequences 5 . Acommon hurdle in both paradigms is the nontrivialnature of predicting transcription factoroccupancy from sequence alone.Xin He and Saurabh Sinha are at theUniversity of Illinois at Urbana-Champaign,Urbana, Illinois.e-mail: xinhe2@illinois.edu orsinhas@illinois.eduZinzen et al. 2 adopted a pragmatic solutionthat circumvents this challenge by usingChIP-chip technology to directly measure theoccupancy levels of five transcription factorsknown to be involved in mesoderm specification(Twi, Bin, Tin, Mef2 and Bap). Theyidentified candidate cis-regulatory modulesas clusters of ChIP-binding peaks and tackledthe problem of mapping quantitative measurementsof transcription factor occupancywithin these modules to one of five predeterminedclasses of expression patterns, eachof which corresponded to a specific tissue ordevelopmental stage (Fig. 1).A machine learning technique called supportvector machine was used to achieve highaccuracy in this classification task, sidesteppingthe need for mechanistic details of theregulatory process. Loosely speaking, thistechnique treats each module’s occupancyprofile as a point in space and learns how tobest draw boundaries that separate pointscorresponding to known modules with differentexpression patterns. Then, any candidatemodule’s expression pattern can be predictedbased on where it lies in space relative tothese boundaries. The authors experimentallyvalidated 35 of 36 predicted cis-regulatorymodules for mesoderm specification, demonstratinga remarkable success rate.The method of Zinzen et al. 2 uses transcriptionfactor occupancy information onan entire genomic segment rather than atindividual binding sites. Its success thereforeseems to support the ‘information display’model of cis-regulatory module function 6 ,which contends that regulatory functiondepends on the number and types of bindingsites in a genomic segment and not on theirprecise arrangement. However, arrangementsof sites may have an important role in determiningnet transcription factor occupancyor in fine-tuning the expression activity; thepaper’s findings do not rule out either possibility.It is also worth noting that the transcriptionfactor occupancy information used by Zinzenet al. 2 was obtained from whole-embryo measurements.That the method does not requirespatial information on the concentrations ofregulatory proteins is a practical advantage butalso raises questions about its generalizability.For example, would it be possible to modelthe expression patterns of genes involved inanterior-posterior axis specification withoutreading the spatial patterns of the transcriptionfactors themselves? Or is there a fundamentaldifference between models appropriatefor regulatory networks that respond to a morphogengradient versus more downstream networksthat impart tissue-specific expression?Notably, Zinzen et al. 2 found that differenttranscription factor occupancy profiles mayyield the same regulatory activity. Althoughthis has been implicit in existing models of cisregulatorylogic 4 , definitive examples such asthose highlighted here are rare. Moreover, theauthors observed that diverse cis-regulatorymodules with similar activity were bound bya key common regulator (e.g., Twi for mesodermand Bin for visceral muscle), with othertranscription factors acting on specific cis-regulatorymodules to modulate the gene expressionpattern. This may prove to be a generaldesign principle for achieving expression patternsthat exhibit the same tissue specificitywhile allowing for minor differences.The novelty of the authors’ approach liesin predicting expression solely from quantitativetranscription factor occupancy values.This raises the possibility of incorporatingdirect occupancy measurements (from ChIPchipor ChIP-Seq experiments) into previous,sequence-based models of expression 4 , whichattempt to predict transcription factor occupanciesand use these to explain the expressionpattern. Such a combination might leadto greater predictive power compared withapproaches based on sequence or occupancyalone. Future modeling efforts are also likely142 volume 28 number 2 february 2010 nature biotechnology

news and views© 2010 Nature America, Inc. All rights reserved.TwiTinMef2BinBapChIPTranscription factorTwiTinMef2BinBapExtract occupancy profileof candidate module5−78−9Stage10−1112−1314−15Predict expressionof candidate modulesProfilesSupport vector machineM SM VMIn vivo validationCandidate cis-regulatory modulesExtract occupancy profilesof known modulesExpression patternsMesoderm (M)Somatic muscle (SM)Visceral muscle (VM)Mesoderm and somaticmuscle (MSM)Mesoderm and visceralmuscle (MVM)OthersMSM MVM OthersTemporal stageTrain classifierFigure 1 Pipeline for discovery of cis-regulatorymodules involved in mesoderm specification.ChIP-chip assays provide genome-wide occupancyinformation for each of five relevant transcriptionfactors at five different temporal stages ofembryonic development. Clusters of ChIP peaksare designated as candidate cis-regulatorymodules. Transcription factor occupancy profilesare generated for each candidate module (left).The same ChIP-chip data are used to generateoccupancy profiles of previously identified cisregulatorymodules (right). These profiles, togetherwith experimentally determined expressionpatterns driven by each module, which arecurated from the literature, are used to train asupport vector machine classifier. The classifieris used to predict the expression pattern (visceralmuscle in this example) driven by the candidatecis-regulatory module. The prediction is verifiedin vivo by a transgenic reporter assay. Reporterresults reprinted from ref.2, with permission of theauthors.expression patterns are easier to come by 8 ;thus, adapting the authors’ approach to workwith gene, rather than module, expressionpatterns as training data would go a long waytoward ensuring broader application.The new method may also be useful in syntheticbiology. Whether for ab initio design ofa sequence that drives a desired tissue-specificpattern 9 or for the refinement of an existingsequence to be used in a synthetic circuit 10 , theutility of quantitative models of expression iswell recognized. The working model proposedhere could help to identify several endogenoussequences with the same regulatory functionand could even suggest the variants (by specifyingtargets of mutation) that are best suitedfor the specific engineering goal.As genome-wide assays of transcriptionfactor–DNA binding become more common,tools that interpret the resulting data toelucidate combinatorial gene regulation willbe needed. The study by Zinzen et al. 2 offersan innovative approach to building such toolsand sets the stage for more in-depth explorationsof regulatory networks.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.to involve whole-genome assays of chromatinstate, such as nucleosome occupancy or varioushistone modifications 4 .There are some practical considerations inapplying the proposed strategy more broadly.First, the method relies on prior knowledgeof all relevant transcription factors, which inthe case of mesoderm specification was availablefrom extensive prior work. For studies ofother regulatory networks, this requirementmight be mitigated using existing statisticaltechniques 7 that identify binding sites overrepresentedin known cis-regulatory modulesof the network, thus inferring the relevanttranscription factors. Second, the model has a‘training phase’ that requires expression measurementson a large number of cis-regulatorymodules—the authors used 139 moduleswith previously characterized expression inmesoderm and/or muscle. Such data are notavailable for most regulatory systems and aredifficult to generate. On the other hand, gene1. Davidson, E.H. The Regulatory Genome: Gene RegulatoryNetworks in Development and Evolution (Academic Press,2006).2. Zinzen, R.P., Girardot, C., Gagneur, J., Braun, M. &Furlong, E.E. Nature 462, 65–70 (2009).3. Janssens, H. et al. Nat. Genet. 38, 1159–1165 (2006).4. Segal, E. & Widom, J. Nat. Rev. Genet. 10, 443–456(2009).5. Beer, M.A. & Tavazoie, S. Cell 117, 185–198 (2004).6. Arnosti, D.N. & Kulkarni, M.M. J. Cell. Biochem. 94,890–898 (2005).7. Warner, J.B. et al. Nat. Methods 5, 347–353 (2008).8. Tomancak, P. et al. Genome Biol. 8, R145 (2007).9. Venter, M. Trends Plant Sci. 12, 118–124 (2007).10. Haseltine, E.L. & Arnold, F.H. Annu. Rev. Biophys. Biomol.Struct. 36, 1–19 (2007).nature biotechnology volume 28 number 2 february 2010 143

news and viewsFrom genomics to crop breedingRichard FlavellNew insights into the maize genome must be incorporated into breeding programs to realize the potential of cropgenomics.© 2010 Nature America, Inc. All rights reserved.A collection of recent papers in Science andPLoS Genetics 1–7 represents a landmark inplant biology and an indispensable resourcefor efforts to improve maize by breeding. Thepapers include the draft genome sequence ofan inbred maize line 1 ; analyses of the chromosomalorganization of genes, microRNAsand transposable elements 1,3 ; comparisonsbetween the genomes of some agriculturallyrelevant inbred lines 4 ; and transcriptionalprofiles of some of the same inbred lines andof hybrids generated from crosses betweenthem 2,5 . The wealth of information in thesereports should accelerate breeding projectsaimed at generating superior varieties ofcorn and other crops.Over 1 billion people eat maize or meatfrom maize-fed livestock. Maize also providesthe raw materials for manufactured productsranging from coatings for paper and cloth tobiodegradable plastics and biofuel. Averagemaize yields over the past 40 years havedoubled in the United States, although notelsewhere. This success was due in part tobreeding of better-performing hybrids, whichare generated by combining the genomes ofinbred plants from different genetic groups.There remains much scope for continuingto improve maize yields by exploiting theyield gain in hybrids relative to their inbredparents, a phenomenon known as heterosis.But to support the world’s growing population,it will be necessary to enhance the rateof increase in the productivity of maize andother crops, especially in inhospitable climates.This challenge will likely be addressedthrough better farming, more reliable seedsupplies and more stable markets, as well asby the application of genomics technologiesto breeding of superior varieties.As most commercially relevant plant phenotypesdepend on the interactions of largenumbers of genes and also on the positionsof genes with respect to one another and tosites of recombination, plant breeding is anactivity that involves whole genomes. Anyplant breeder would like to know, first, howevery chromosomal segment, gene or allele—Richard Flavell is at Ceres, Inc., Thousand Oaks,California, USA.e-mail: rflavell@ceres.netalone and in combination with others—contributesto specific traits, and, second, howto alter the genome to manipulate traits atwill. Historically, these goals remained out ofreach because knowledge of gene–trait associationswas limited by the need to measuretraits in the field and to define their geneticbasis by the inadequate trait mapping proceduresof classical and statistical genetics(right side of Fig. 1). With the advent ofgenomics technologies, breeders can characterizethe allelic content of their particulargermplasm in exquisite detail throughoutthe breeding program and so preserve themost valuable allele combinations (left sideof Fig. 1).Progress in identifying gene–trait associationsfor maize is being achieved by constructingand analyzing “nested associationmapping” lines. In this approach, chromo-Genome-widegenotypingDiverse population ofplants/chromosomesGene-trait associationsSelection of potentialinbred parents of higheryieldinghybridsCrossing of inbredsto make hybridsSelection andevaluation of heterotichybrids for commercesome segments from diverse inbreds arerecombined into sets of inbreds by a verylarge number of recombination events, andthe plants are compared in ways that allowthe roles of individual chromosome segmentsto be inferred 6,7 . The chromosome segmentsare also being defined by their molecularpolymorphisms and by the genes they carry,allowing them to be tracked throughoutthe breeding process. However, a very largenumber of defined polymorphisms will berequired to uniquely mark all alleles. Thishas recently become possible with wholegenomesequencing technologies 1,3 .Integration of the new genomics technologieswith traditional breeding strategies willalso empower breeders in their efforts todesign and select the best combinations ofchromosome segments, genes and allelesavailable in the species to meet commercialTrait analysis infields at multiplelocationsFigure 1 A highly simplified scheme illustrating how genomics can contribute to steps in a maize-hybridbreeding pipeline. Plants selected from parent inbred improvement programs are used in crosses tomake F 1 hybrids that are then evaluated. F 1 hybrids with improved performance are then adopted forcommercialization. Historically, traits could not be managed efficiently as the causal DNA sequencesremained largely unknown and their existence was recognized only through the use of extensive fieldtrials (right side). New genomics technologies to determine complete genomes, DNA polymorphisms,whole genome expression patterns and chromosomal haplotype blocks (reflecting recombinationpatterns) now make it possible to build detailed gene-trait associations and manage such associationsthroughout the breeding program, including the selection of combinations of alleles (left side). Thecombined use of trait assessments in the field and genomics technologies increases the efficiency ofcrop breeding.144 volume 28 number 2 february 2010 nature biotechnology

news and views© 2010 Nature America, Inc. All rights reserved.criteria for crop improvement (Fig. 1). Therecent papers 1–7 will guide selection of parentsby clarifying the substantial differencesin active gene contents between differentmaize inbreds. Assuming that a major basisof heterosis is the complementation of thesedifferences in inbreds 1,2 , the new informationprovides pointers as to which parents shouldcomplement well in heterotic hybrids.Interestingly, the genome-wide expressionstudies of Swanson-Wagner et al. 5 suggestthat there is much differential gene expressionbetween alleles in hybrids relative to theinbred parents and that this is driven primarilyby the paternal allele of trans-actingexpression quantitative trait loci—genes thatexert their effect by regulating the expressionof other genes, often elsewhere in thegenome. This provocative finding raises thequestion of the extent to which heterosisdepends also on epigenetic and parent-oforiginimprinting effects, and of whetherbreeders should therefore focus on these loci,in particular when making crosses in specificdirections.The positions and frequencies of recombinationevents relative to the positions ofdiverse alleles influence tremendously theefficiencies of improved plant productionbecause they determine how readily newcombinations of alleles are created. Goreet al. 4 revealed that some 21% of genes liein low-recombination regions around thecentromeres and, thus, that variation inthese genes is difficult to exploit in breeding,except by means of new chromosomecombinations and hybrids. Vielle-Calzadaet al. 3 defined >100 regions characterizedby low genetic diversity among maize lines.These regions may therefore be associatedwith traits that contributed to domestication(such as a larger number of kernels thatremain attached to the cob and lack a stonycasing) and consequently have been selectedThe wealth of information inthese reports should acceleratebreeding projects aimed atgenerating superior varieties ofcorn and other crops.in all modern maize breeding programs.Other regions are highly polymorphic andmay therefore be associated with adaptationto different geographic regions and sourcesof variation that breeders select and elaborate.The largest maize-breeding companieshave probably already sequenced many corngenomes, or parts of them, and they havemuch more accurate information on phenotypesof lines in a wide array of environmentsthan the public sector does. Nevertheless,the studies 1–7 will provide these companieswith much new data and analyses that willbe used to drive their breeding programs.Unfortunately, most smaller companies lackthe resources (e.g., databases and informationtechnology systems) to fully exploit the newgenomic information, suggesting that the gapbetween those who have this capability andthose who do not will continue to widen.Those who stand to benefit especially fromthese reports 1–7 are the breeders in the publicsector and small companies that are seekingto provide improved lines for poorer societies.Foremost among them is the ConsultativeGroup on International AgriculturalResearch and its tropical maize breedingefforts spearheaded by the InternationalMaize and Wheat Improvement Center(Centro Internacional de Mejoramientode Maíz y Trigo; CIMMYT). CIMMYT hasstruggled to fully embrace genomics andhas lagged behind the leading private sectorcompanies in exploiting genomics inits breeding program, on which so manydepend. Let us hope that these publicationswill prove sufficiently compelling to inspiretheir government funders and other publicsectorbreeders to expedite the application ofgenomics in crop breeding.COMPETING INTERESTS STATEMENTThe author declares no competing financial interests.1. Schnable, P.S. et al. Science 326, 1112–1115(2009).2. Springer, N.M. et al. PLoS Genet. 5, e1000734(2009).3. Vielle-Calzada, J.-P. et al. Science 326, 1078 (2009).4. Gore, M.A. et al. Science 326, 1115–1117 (2009).5. Swanson-Wagner, R.A. et al. Science 326, 1118–1120 (2009).6. McMullen, M.D. et al. Science 325, 737–740(2009).7. Buckler, E.S. et al. Science 325, 714–718(2009).Spilling the beans on legume biologySoybean is the most recentaddition to the rapidly growinglist of crops for which a highqualitydraft genome is nowavailable. Writing in Nature,Schmutz et al. 1 report that the1.1-gigabasesoybeangenome—thelargest shotgunsequencedplantgenome—ispredictedto encode46,000 genes.Two genomeduplicationevents are likelyRoy Kaltschmidtto account for the observationthat ~75% of these genesare found in multiple copies.Although the importanceof soybean as a source ofprotein and oil alone testifiesto the potential implicationsof understanding its geneticmakeup, this genome willalso serve as the reference for~20,000 leguminous speciesthat play a critical ecologicalrole through their unique abilityto fix nitrogen with the help ofrhizobial bacteria. Availability ofthe genome should acceleratethe association of quantitativetrait loci of nutritional,economic and ecologicallyimportant traits with the causalDNA sequences from soybeanin the near future. In the longerterm, the genome will likelyalso be leveraged to improvethe way in which a range ofleguminous subsistence cropsare used to both replenish soilnitrogen through crop rotationand meet the expanding needsof developing nations forprotein and energy.Peter Hare1. Schmutz, J. et al. Nature 463, 178–183 (2010).nature biotechnology volume 28 number 2 february 2010 145

news and views© 2010 Nature America, Inc. All rights reserved.criteria for crop improvement (Fig. 1). Therecent papers 1–7 will guide selection of parentsby clarifying the substantial differencesin active gene contents between differentmaize inbreds. Assuming that a major basisof heterosis is the complementation of thesedifferences in inbreds 1,2 , the new informationprovides pointers as to which parents shouldcomplement well in heterotic hybrids.Interestingly, the genome-wide expressionstudies of Swanson-Wagner et al. 5 suggestthat there is much differential gene expressionbetween alleles in hybrids relative to theinbred parents and that this is driven primarilyby the paternal allele of trans-actingexpression quantitative trait loci—genes thatexert their effect by regulating the expressionof other genes, often elsewhere in thegenome. This provocative finding raises thequestion of the extent to which heterosisdepends also on epigenetic and parent-oforiginimprinting effects, and of whetherbreeders should therefore focus on these loci,in particular when making crosses in specificdirections.The positions and frequencies of recombinationevents relative to the positions ofdiverse alleles influence tremendously theefficiencies of improved plant productionbecause they determine how readily newcombinations of alleles are created. Goreet al. 4 revealed that some 21% of genes liein low-recombination regions around thecentromeres and, thus, that variation inthese genes is difficult to exploit in breeding,except by means of new chromosomecombinations and hybrids. Vielle-Calzadaet al. 3 defined >100 regions characterizedby low genetic diversity among maize lines.These regions may therefore be associatedwith traits that contributed to domestication(such as a larger number of kernels thatremain attached to the cob and lack a stonycasing) and consequently have been selectedThe wealth of information inthese reports should acceleratebreeding projects aimed atgenerating superior varieties ofcorn and other crops.in all modern maize breeding programs.Other regions are highly polymorphic andmay therefore be associated with adaptationto different geographic regions and sourcesof variation that breeders select and elaborate.The largest maize-breeding companieshave probably already sequenced many corngenomes, or parts of them, and they havemuch more accurate information on phenotypesof lines in a wide array of environmentsthan the public sector does. Nevertheless,the studies 1–7 will provide these companieswith much new data and analyses that willbe used to drive their breeding programs.Unfortunately, most smaller companies lackthe resources (e.g., databases and informationtechnology systems) to fully exploit the newgenomic information, suggesting that the gapbetween those who have this capability andthose who do not will continue to widen.Those who stand to benefit especially fromthese reports 1–7 are the breeders in the publicsector and small companies that are seekingto provide improved lines for poorer societies.Foremost among them is the ConsultativeGroup on International AgriculturalResearch and its tropical maize breedingefforts spearheaded by the InternationalMaize and Wheat Improvement Center(Centro Internacional de Mejoramientode Maíz y Trigo; CIMMYT). CIMMYT hasstruggled to fully embrace genomics andhas lagged behind the leading private sectorcompanies in exploiting genomics inits breeding program, on which so manydepend. Let us hope that these publicationswill prove sufficiently compelling to inspiretheir government funders and other publicsectorbreeders to expedite the application ofgenomics in crop breeding.COMPETING INTERESTS STATEMENTThe author declares no competing financial interests.1. Schnable, P.S. et al. Science 326, 1112–1115(2009).2. Springer, N.M. et al. PLoS Genet. 5, e1000734(2009).3. Vielle-Calzada, J.-P. et al. Science 326, 1078 (2009).4. Gore, M.A. et al. Science 326, 1115–1117 (2009).5. Swanson-Wagner, R.A. et al. Science 326, 1118–1120 (2009).6. McMullen, M.D. et al. Science 325, 737–740(2009).7. Buckler, E.S. et al. Science 325, 714–718(2009).Spilling the beans on legume biologySoybean is the most recentaddition to the rapidly growinglist of crops for which a highqualitydraft genome is nowavailable. Writing in Nature,Schmutz et al. 1 report that the1.1-gigabasesoybeangenome—thelargest shotgunsequencedplantgenome—ispredictedto encode46,000 genes.Two genomeduplicationevents are likelyRoy Kaltschmidtto account for the observationthat ~75% of these genesare found in multiple copies.Although the importanceof soybean as a source ofprotein and oil alone testifiesto the potential implicationsof understanding its geneticmakeup, this genome willalso serve as the reference for~20,000 leguminous speciesthat play a critical ecologicalrole through their unique abilityto fix nitrogen with the help ofrhizobial bacteria. Availability ofthe genome should acceleratethe association of quantitativetrait loci of nutritional,economic and ecologicallyimportant traits with the causalDNA sequences from soybeanin the near future. In the longerterm, the genome will likelyalso be leveraged to improvethe way in which a range ofleguminous subsistence cropsare used to both replenish soilnitrogen through crop rotationand meet the expanding needsof developing nations forprotein and energy.Peter Hare1. Schmutz, J. et al. Nature 463, 178–183 (2010).nature biotechnology volume 28 number 2 february 2010 145

news and viewsSystematic tracking of cell fate changesJonghwan Kim & Stuart H OrkinHigh-throughput measurements across several regulatory levels provide a comprehensive view of ES-celldifferentiation.© 2010 Nature America, Inc. All rights reserved.Deciphering the regulation of eukaryoticgene expression is a formidable challengebecause of the multilayered nature of regulatorymechanisms. In an effort to decodethe complexity of molecular processes governingcell fate changes in mouse embryonicstem (ES) cells, Lu et al. 1 , in a recent issueof Nature, integrated multiple ‘omics’ datasets and systematically monitored temporalchanges in some of the key regulatory events(Fig. 1). In offering this broad view, their systemsapproach to understanding dynamicfate changes provides insight into how todeconstruct complex networks in other cellularcontexts, such as lineage specification,differentiation and somatic cell reprogramming.Cells modulate gene expression in responseto external and/or internal stimuli. Owingto the complexity of regulatory mechanisms,efforts to date have focused largely on oneaspect at a time. For example, investigatorshave studied epigenetic modification, transcription,post-transcriptional modification,translation or post-translational modificationoccurring within a sequence of events,rather than all processes in aggregate.The recent development of high-throughputtechnologies—including gene expressionprofiling; global mapping of protein-DNAinteractions; mapping of histone modificationby microarray or sequencing; proteinproteininteraction mapping and proteinabundance measurement by mass spectrometry;and gene knockdown by RNAinterference—offers the potential to observebiological phenomenon at a global level 2 . Aseach of these methods generates a wealth ofdata, data handling and analysis tools becomerate limiting in the interpretation. This is thechallenge addressed by Lu et al. 1 .ES cells are distinguished by their capacityfor perpetual self-renewal and pluripotency(the ability to differentiate into all tissues).Nanog is one of the key transcription factorsin ES cells and is known to be required formaintenance of pluripotency in mouse ESJonghwan Kim and Stuart H. Orkin are at theChildren’s Hospital Boston and Dana FarberCancer Institute, Boston, Massachusetts, USA.e-mail: stuart_orkin@dfci.harvard.eduES cellsDifferent regulatory levelsNanog downregulationDay 0 Day 1Day 3Temporal regulationHistone modificationRNA polymerase II occupancymRNA abundanceDifferentiated cellsDay 5Nuclear protein abundanceFigure 1 Systematic monitoring of differentiating ES cells. Upon knockdown of the keytranscription factor Nanog, changes in histone acetylation, RNA polymerase II occupancy, mRNAabundance and nuclear protein abundance were measured over five days of differentiation bysystems biology tools. Acquired data sets were integrated, analyzed and visualized by GATEsoftware.cells 3,4 . In their elegant experiments, Lu etal. 1 sought to introduce a single genetic perturbationby knocking down Nanog with aninducible small hairpin RNA and to examinethe consequences of this perturbation in aglobal fashion. Upon knockdown of Nanog,ES cells exit the stem cell state and differentiate.During the accompanying dynamicchanges, Lu et al. 1 measured the outcomes atfour different regulatory layers: (i) histoneH3 lysine 9 and 14 acetylation by ChIP-chip,indicating an active epigenetic signature,(ii) RNA polymerase II occupancy by ChIPchip,indicating active gene transcription,(iii) mRNA transcript abundance by expressionmicroarray, indicating an outcome oftranscription, and (iv) nuclear protein abundanceby mass spectrometry. They collecteddata for each regulatory layer at four timepoints (days 0, 1, 3 and 5 after Nanog knockdown)to monitor the temporal sequence ofevents.By comparing changes in each regulatorylayer, Lu et al. 1 observed that, in general,changes between different gene expressionsteps are moderately well correlated.However, they also found discordant regu-DataacquisitionandintegrationInteractive analysis andvisualization of datausing GATE softwareHeat map movieand/orDynamic scatter plotlation for a large number of genes. Notably,~42–53% of all proteins whose levels changesubstantially did not correlate with changesin their respective mRNA abundances. Thisdiscrepancy has been observed in lowerorganisms, such as yeast, but has been ratherunclear in the mammalian context owingto either technical limitations or improperexperimental settings 5 . Although Lu et al. 1did not address this issue further, theirresults clearly imply that additional layersof regulation beyond transcriptional control,such as translational and post-translationalmodifications, are of critical importance incell fate decisions.Using gene-ontology analysis, Lu et al. 1also observed that chromatin-modifyingenzymes are regulated by RNA polymerase IIoccupancy but not by the other three regulatorylayers. This suggests that although chromatinremodeling is known to be importantin ES cell fate decision, primary control isexerted through transcriptional regulationvia transcription-factor occupancy.A challenge in handling such large, diversedata sets is the difficulty of visualizing thedata without adequate graphical software.146 volume 28 number 2 february 2010 nature biotechnology

news and views© 2010 Nature America, Inc. All rights reserved.The authors provide a tool, Grid analysis oftime-series expression (GATE), which usesa correlation-based clustering algorithm forthe comparison and visualization of multilayeredtime-course data sets as interactiveimages or movies 6 . Using GATE, Lu et al. 1conducted temporal and dynamic analysesamong different regulatory layers with clusteringof genes that are regulated in a similarpattern, revealing connections betweenthe regulatory mechanisms underlying EScell fate changes. Their analysis reveals thetemporal order of regulatory network configurationsafter a single perturbation, thedownregulation of Nanog.The study by Lu et al. 1 does not plumbthe full complexity of cell fate determination.Rather, it should be viewed as a first step inwhat will have to be a series of attempts toincorporate multiple data sets into a comprehensiveunderstanding of gene regulation. Luet al. 1 introduced an artificial perturbationin ES cells by knocking down a single criticalfactor. Surely, during normal development,changes in multiple regulators may occursimultaneously. The necessity to oversimplifythe regulatory problem is evident inother ways. For example, different kinds ofhistone modifications involved in positiveor negative gene regulation that are likely tobe functionally relevant were not examined.Moreover, several forms of regulation werenot studied, including post-transcriptional/translational regulation by microRNAs 7 ,post-translational modification of proteins,and modulation of protein localization.A truly comprehensive accounting of thedynamics of fate changes will require considerationof these additional regulatorylayers.Tracking cell-state transitions by multiplehigh-throughput assays, as well as the integrationof such observations in a systematicfashion, is a monumental task. Ultimately,one would like to develop ways to use thekinds of large data sets analyzed by Lu etal. 1 to predict the specific outcomes in EScells of other regulatory perturbations.Given the heterogeneity in gene expressionamong cells 8,9 , it is likely that additionaltechnologies will be needed, such as quantitativemonitoring of gene expression andof epigenetic modifications at the single-celllevel. The development of readily accessibledatabases for storing large data sets fromvarious platforms, as well as user-friendlyanalysis and visualization tools, will also benecessary to facilitate the comprehensiveunderstanding of multilayered gene regulatorynetworks during dynamic cellularprocesses.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.1. Lu, R. et al. Nature 462, 358–362 (2009).2. MacArthur, B.D., Ma’ayan, A. & Lemischka, I.R.Cold Spring Harb. Symp. Quant. Biol. 73, 211–215(2008).3. Mitsui, K. et al. Cell 113, 631–642 (2003).4. Chambers, I. et al. Cell 113, 643–655 (2003).5. de Sousa Abreu, R., Penalva, L.O., Marcotte, E.M. &Vogel, C. Mol. Biosyst. 5, 1512–1526 (2009).6. MacArthur, B.D., Lachmann, A., Lemischka, I.R. &Ma’ayan, A. Bioinformatics 26, 143–144 (2010).7. Marson, A. et al. Cell 134, 521–533 (2008).8. Singh, A.M., Hamazaki, T., Hankowski, K.E. & Terada,N. Stem Cells 25, 2534–2542 (2007).9. Chambers, I. et al. Nature 450, 1230–1234 (2007).nature biotechnology volume 28 number 2 february 2010 147

esearch highlights© 2010 Nature America, Inc. All rights reserved.Adaptive optics in microscopyLight microscopy of wholetissues is often complicated bydistortions caused by the opticalinhomogeneities in the biologicalspecimen. By borrowing andadapting approaches from astronomy,Ji et al. develop adaptive optics tocorrect aberrations. The correctionis achieved using a ‘spatial light modulator’, an active opticalelement that permits adjustment of the tilt and phase of thelight passing through more than 100 individual segments at therear pupil of the objective lens. The extent of the modulation foreach of the segments is determined by an algorithm that firstmeasures and corrects the spatial deflections caused by thesample inhomogeneities and then corrects errors in the phase foreach of the segments. The improvements in signal strength, imagefidelity and resolution that can be achieved using adaptive opticsare demonstrated by imaging fluorescent beads and neurons in300- to 500-µm thick brain slices. The technique can be used toimprove the performance of many wide-field and point scanningmicroscopy technologies, including the latest super resolutiontechniques that are especially sensitive to optical aberrations.(Nat. Methods advance online publication, December 27, 2009,doi:10.1038/nmeth.1411)MEDigestible plant wallsRigid cell walls give plants strength, but they also confound attemptsby plant genetic engineers to convert woody plants to biofuels.Creating rigid cell walls requires intermolecular cross-linking ofpectins, facilitated by de-methyl-esterified homogalacturonans(HGA). Releasing fermentable sugars from plant cell walls, on theother hand, requires environmentally unfriendly chemicals or hightemperatures. Now Lionetti et al. show that plants transformedwith enzymes that inhibit the de-esterification of HGA polymersare more accessible to enzymatic degradation. Twice as much sugarwas released from the leaves of transgenic Arabidopsis thaliana plantsoverexpressing fungal polygalacturonase, 60% more when plantswere transformed with a pectin methylesterase inhibitor (PMEI).To show that pectin architecture was being modified, the researchersreacted the transgenic leaves with an antibody that binds to blocksof de-esterified HGA and found reduced binding to the transgenicplants. They were able to replicate these findings in wheat (Triticumdurum), an industrially important plant. Finally, they found thatArabidopsis expressing PMEI had more biomass than the controlplants (due to cell expansion). This contrasts with polygalacturonase-expressingtransgenic plants that generally have less biomass.The group suggests that regulating polygalacturonase expression intime and space might prevent the loss of biomass. (Proc. Natl. Acad.Sci. USA 107, 616–621, 2010)LDMicroarray SNP detection heats upGresham et al. have discovered new rules to enhance the accuracyof DNA microarrays. These rules substantially improve the abilityWritten by Kathy Aschheim, Laura DeFrancesco, Markus Elsner &Craig Makof probes on the array to bind the correct sequence when similarsequences are present in a sample, which is particularly useful whenidentifying variation at the level of single nucleotide polymorphisms(SNPs). The authors varied different experimental parameters anddiscovered that a probe is best able to discriminate SNPs when itsmelting temperature (T m ) is ~2–5 °C below the temperature used tohybridize samples to the array. This knowledge was used to designmicroarrays with probes between 16 and 35 nt in length—in contrastto the common practice of making uniform-length probes—such that all probes have the same melting temperature, and thus allachieve optimal performance at the same hybridization temperature,~2–5 °C above the T m . These so called isothermal microarrays outperformedconventional arrays at identifying heterozygous SNPs.The use of an optimum hybridization temperature in tandem withuniform T m probes (varying in length) may be useful for designingarrays for other applications. (Proc. Nat. Acad. Sci. USA, publishedonline January 8, 2010, doi:10.1073/pnas.0913883107) CMCo-workers in transcription factoriesChromosome conformation capture on a chip (4C) is capable ofdetecting remote chromatin interactions on a genome-wide scale.Schoenfelder et al. use a modification of this method to analyze thegenome-wide repertoire of transcriptional interactions associatedwith globin genes in erythroid tissues. Their technique, dubbedenhanced ChIP-4C, first cross-links proteins and DNA to generate asnapshot of the spatial organization of the nucleus. However, the 4Cassay is then modified to incorporate an RNA polymerase–recognizingantibody that identifies DNA that is near to, but not necessarily onthe same chromosome as, actively transcribed copies of the ‘bait’ gene(that is, globin). Subsequently, a biotinylated bait-specific DNA probeis used to enrich for ‘prey’ sequences, which are cross-linked to thebait. The DNA in the enriched sample is then identified by microarrayanalysis. Analysis of the promoters of actively transcribed globingenes in mouse cells reveals a transcription factor, Klf1, requiredfor regulating genes in globin-containing ‘transcription factories’ inthe nucleus. The approach should facilitate understanding of howgenes are brought together in the nucleus to regulate their expression.(Nat. Genet. 42, 53–61, 2010)CMPlatelet allyBiotech has no shortage of new ideas on how to staunch bleeding.Where traditional therapy amounts to little more than the applicationof pressure or absorbent material, research in the past decade hassought to enhance the body’s intrinsic mechanisms of coagulation.Allogeneic platelets, recombinant clotting factors, red blood cellsdisplaying the cell-adhesive RGD sequence, self-assembling peptides,liposomes and a block copolymer of hemoglobin and fibrinogenare some of the strategies that have been tried, but none hasdemonstrated adequate safety and efficacy. Now Bertram et al. haveproposed to control bleeding with “synthetic platelets,” or nanoparticlesconsisting of poly(lactic-co-glycolic acid)-poly-l-lysine blockcopolymer cores carrying polyethylene glycol chains that are cappedwith RGD sequences. Working with a rat model of major injury tothe femoral artery, the authors found that the nanoparticles bindto platelets and boost clot formation more effectively than existingtherapies. Moreover, the nanoparticles were rapidly cleared fromthe circulation, and no adverse effects were observed. (Sci. Transl.Med., published online December 16, 2009, doi:10.1126/scitranslmed.3000397)KA148 volume 28 number 2 february 2010 nature biotechnology

e s o u r c eRational association of genes with traits using agenome-scale gene network for Arabidopsis thalianaInsuk Lee 1,2,5 , Bindu Ambaru 4,5 , Pranjali Thakkar 4 , Edward M Marcotte 2,3 & Seung Y Rhee 4© 2010 Nature America, Inc. All rights reserved.We introduce a rational approach for associating genes with plant traits by combined use of a genome-scale functional networkand targeted reverse genetic screening. We present a probabilistic network (AraNet) of functional associations among 19,647(73%) genes of the reference flowering plant Arabidopsis thaliana. AraNet associations are predictive for diverse biologicalpathways, and outperform predictions derived only from literature-based protein interactions, achieving 21% precision for 55%of genes. AraNet prioritizes genes for limited-scale functional screening, resulting in a hit-rate tenfold greater than screens ofrandom insertional mutants, when applied to early seedling development as a test case. By interrogating network neighborhoods,we identify AT1G80710 (now DROUGHT SENSITIVE 1; DRS1) and AT3G05090 (now LATERAL ROOT STIMULATOR 1; LRS1) asregulators of drought sensitivity and lateral root development, respectively. AraNet (http://www.functionalnet.org/aranet/) providesa resource for plant gene function identification and genetic dissection of plant traits.Manipulating plant traits that affect the production of food, fiberand renewable energy has important agricultural and economic consequences.What is the best approach for identifying genes for importantplant traits? Forward genetics is limited as mutations in manygenes may generate only moderate or weak phenotypes. Similarly,although reverse genetics allows for directed assay of gene perturbations1 , saturated phenotyping for many plant traits is impractical. Apragmatic near-term solution is the computational identification oflikely candidate genes for desired traits, allowing for focused, efficientuse of reverse genetics. This solution is not unlike rational drug designin which computer-assisted and expert knowledge are combined withtargeted screening for the desired drug, or in this case, trait.One emerging approach for prioritizing candidate genes is networkguidedguilt by association. In this approach, functional associationsare first determined between genes in a genome on the basis of extensiveexperimental data sets, encompassing millions of individual observations.Such a map of functional associations is often represented as agraph model and referred to as a functional gene network 2 .Probabilistic functional gene networks integrate heterogeneousbiological data into a single model, enhancing both model accuracyand coverage. Once a suitable network is generated, new candidategenes are proposed for traits based upon network associations withgenes previously linked to these traits. Such network-guided screeninghas been successfully applied to unicellular organisms 3,4 andCaenorhabditis elegans 5,6 , and is a proposed strategy for identifyinghuman disease genes 4,5,7–10 .We demonstrate here that this approach successfully identifies genesaffecting specified traits for a reference flowering plant, A. thaliana,and we introduce a genome-wide, functional gene network forArabidopsis suitable for prioritizing candidate genes for a wide varietyof traits of economic and agricultural importance.RESULTSReconstruction of an Arabidopsis gene networkWe integrated diverse functional genomics, proteomics and comparativegenomics data sets into a genome-wide functional gene network,using data integration and benchmarking methods customized forArabidopsis genes (Supplementary Methods). The data sets includedmRNA co-expression patterns measured from DNA microarray datasets (Supplementary Table 1), known Arabidopsis protein-proteininteractions 11–14 , protein sequence features including sharing ofprotein domains, similarity of phylogenetic profiles 15–17 or genomiccontext of bacterial or archaebacterial homologs 18–20 , and diversegene-gene associations (mRNA co-expression, physical proteininteractions, multiprotein complexes, genetic interactions, literaturemining) transferred from yeast 21 , fly 11,22–24 , worm 5 and humangenes based on orthology 25 (Supplementary Table 2). In total,24 distinct types of gene-gene associations, encompassing >50 millionindividual experimental or computational observations, were scoredfor their ability to correctly reconstruct shared membership inArabidopsis biological processes. Then, these were incorporated intoa single integrated network model, dubbed AraNet. AraNet contains1,062,222 functional linkages among 19,647 genes (~73% of the totalArabidopsis genes), with each linkage weighted by the log likelihood ofthe linked genes to participate in the same biological processes.Integrating data improves network coverage and accuracy,as tested by recovery of known functional associations (Fig. 1aand Supplementary Fig. 1). AraNet extends substantially beyond1 Department of Biotechnology, College of Life Science and Biotechnology, Yonsei University, Seodaemun-gu, Seoul, Korea. 2 Center for Systems and Synthetic Biology,and 3 Department of Chemistry and Biochemistry, Institute for Cellular and Molecular Biology, University of Texas at Austin, Austin, Texas, USA. 4 Department of PlantBiology, Carnegie Institution for Science, Stanford, California, USA. 5 These authors contributed equally to this work. Correspondence should be addressed to I.L.(insuklee@yonsei.ac.kr) or E.M.M. (marcotte@icmb.utexas.edu) or S.Y.R. (rhee@acoma.stanford.edu).Received 10 June 2009; accepted 23 December 2009; published online 31 January 2010; doi:10.1038/nbt.1603nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 149

e s o u r c eFigure 1 Construction, accuracy and coverage of AraNet, a functionalgene network for Arabidopsis. (a) Pairwise gene linkages derived from 24diverse functional genomics and proteomics data sets, representing >50million experimental or computational observations, were integrated intoa composite network with higher accuracy and genome coverage than anyindividual data set. The integrated network (AraNet) contains 1,062,222functional linkages among 19,647 (73%) of the 27,029 protein-codingArabidopsis genes. The plot x axis indicates the log-scale percentage ofthe 27,029 protein-coding genes 13 covered by functional linkages derivedfrom the indicated data sets (curves); the y axis indicates predictivequality of the data sets, measured as the cumulative likelihood ratio oflinked genes to share GO biological process term annotations, tested using0.632 bootstrapping and plotted for successive bins of 1,000 linkageseach (symbols). Data sets are named as XX-YY, where XX indicates speciesof data origin (AT, A. thaliana; CE, C. elegans; DM, D. melanogaster;HS, H. sapiens; SC, S. cerevisiae) and YY indicates data type (CC,aLikelihood ratio for linked genes toparticipate in same biological process(cumulative likelihood ratio, log scale)AraNet148AT-CXAT-DCAT-GNAT-LC90AT-PGCE-CCCE-CX55CE-GTCE-LCCE-YH33DM-PIHS-CXHS-DC20HS-LCHS-MSHS-YH12SC-CCSC-CXSC-DCSC-GT7SC-LCSC-MSSC-TS4SC-YH1 10100Coverage of 27,029 protein-coding genes (%)bCoverage of 27,029 protein-codinggenes (%)806040200AT-LCaccuracylevelReliableannotation onlyAll annotationAraNetco-citation; CX, mRNA co-expression; DC, domain co-occurrence; GN, gene neighbor; GT, genetic interaction; LC, literature-curated protein interactions;MS, affinity purification/mass spectrometry; PG, phylogenetic profiles; PI, fly protein interactions; TS, tertiary structure; YH, yeast two-hybrid. Relativecontribution of each data type and combining different evidences for inferring function is discussed in Supplementary Discussion (SupplementaryFigs. 13–15). (b) AraNet spans ~73% of the protein-coding genes, far in excess of current GO biological process annotations for Arabidopsis, for which~12.2% of genes are annotated by reliable experimental evidence (GO evidence codes IDA, IMP, IGI, IPI, IEP) or traceable author statements (GOevidence code TAS), or ~45% annotated by any evidence including computational inferences or sequence homology. The subset of AraNet linkagesstronger than the likelihood ratio for literature-curated protein interactions (AT-LC, corresponding to a likelihood ratio of 35:1) covers 55% of the genes.© 2010 Nature America, Inc. All rights reserved.well-characterized Arabidopsis genes (Fig. 1b): 23,720 Arabidopsis genesare unannotated with Gene Ontology (GO) biological process annotationsby reliable experimental evidence 13 . AraNet includes more thanhalf (7,465) of genes lacking even sequence homology–based annotations(14,847 genes). These genes’ functions can now be hypothesizedbased upon their network neighborhoods. AraNet implicates specificprocesses for 4,479 uncharacterized genes using guilt by association.There are 2,986 uncharacterized genes associated only with otheruncharacterized genes in AraNet, suggesting many still-uncharacterizedcellular processes in plants.Evaluating the accuracy of AraNetTo verify the reliability of functional associations in AraNet, wetested their consistency with known Arabidopsis gene annotationsby applying guilt by association in AraNet to identify genes associatedwith specific biological processes. Each gene in the genome wasscored for association with a particular process by summing networkedge weights connecting that gene to known genes in that process.A gene’s resulting score corresponds to the naive Bayes estimate forthe gene to belong to that process given network evidence (Fig. 2a).Performing cross-validation of this test allows us to assess predictivepower with a receiver-operator characteristic (ROC) curve, measuringthe true-positive prediction rate versus false-positive predictionrate as a function of prediction score. We use the area under the ROCcurve (AUC) to summarize performance. AUC values of ~0.5 and 1indicate random and perfect performance, respectively.Using cross-validation, we tested AraNet’s ability to correctlyassociate genes with each GO biological process, observing significantlybetter than random predictability for the majority of biologicalprocesses (P < 10 −53 ; Wilcoxon signed rank test unless notedotherwise) (Fig. 2b). AraNet incorporates data from other organisms;we correspondingly observed higher predictability for evolutionarilyconserved processes than for GO processes annotated onlywith plant genes (P < 10 −24 , Wilcoxon rank sum test) (Figs. 2c,d).Nonetheless, genes were correctly associated with plant-specificprocesses at significantly higher rates than expected by chance (P

e s o u r c e© 2010 Nature America, Inc. All rights reserved.Figure 2 Predictive power of AraNet for conservedand plant-specific biological processes. AraNet’spredictive capacity was measured using crossvalidatedreceiver operator characteristic (ROC)curve analysis, as illustrated in (a). For a givenprocess, each gene in the Arabidopsis genomeis rank-ordered by the sum of its network linkagescores to the set of ‘bait’ genes already associatedwith that process (omitting each bait gene from thebait set for purposes of evaluation). High-scoringgenes are most tightly connected to the bait set andare the most likely new candidates to participatein that process. This trend is evident in a ROC plotmeasuring recovery of bait genes as a function ofrank, calculating the true-positive prediction rate(sensitivity; TP/(TP+FN)) versus the false-positiveprediction rate (1−specificity; FP/(FP+TN)). If baitgenes are highly interconnected (red circles), unlikerandom genes (blue circles), additional genesconnected to the bait genes (green circles) are morelikely to be involved in the same process. The areaunder the cross-validated ROC curve (AUC) providesa measure of predictability, ranging from ~0.5 forrandom expectation (blue curve) to 1 for perfectpredictions (red curve). (b) Distributions of AUCvalues are plotted for network-based identificationof genes for each of the 318 GO biological processterms with annotations, (c) for each of the 151biological process terms with annotations sharedbetween plant and animal or between plant andyeast and (d) for each of the 167 biological processterms with annotations found in plants but absentfrom animals and fungi. In bar-and-whiskers plots,abArea under ROC curveeTrue-positive rate001False-positive rateGO BP (all) c GO BP (conserved)Response to0 oxidative stress (0.82) 00 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.01.01.00.90.90.90.80.80.80.70.70.70.60.60.60.50.50.50.40.40.4Abiotic responseRandom (0.50)1.0f 1.0Organ development0.80.60.40.2Response to waterdeprivation (0.73)Response to hydrogenperoxide (0.73)Cold acclimation (0.72)Response to heat (0.80)Response to highlight intensity (0.79)0.80.60.40.2False-positive rateFalse-positive rateRandomAraNetAraNet plantdata onlyAraNet noplant dataArea under ROC curveTrue-positive rate1RandomAraNetAraNet plantdata onlyAraNet noplant datadArea under ROC curveGO BP (plant-specific)1.0RandomAraNetAraNet plantdata onlyAraNet noplant dataRandom (0.50)Root development (0.62)Cuticledevelopment (0.61)Stamendevelopment (0.64)Stomatal complexmorphogenesis (0.71)Trichomemorphogenesis (0.73)Carpeldevelopment (0.75)Ovuledevelopment (0.81)the central horizontal line in the box indicates the median AUC and the boundaries of the box indicate the first and third quartiles of the AUC distribution. Whiskersindicate the 10th and 90th percentiles, and circles indicate individual outliers. AraNet specifically identified genes associated with (e) plant abiotic stress responsegenes and (f) organ developmental processes, as annotated by GO. AUC values are indicated in parentheses.True-positive rateLinked genes share cell type–specific expression patternsMany traits in multicellular organisms pertain to specific tissues orcell types. The predictive strength shown by AraNet for such processesraises the question of how a global gene network, incorporatingdiverse samples and data from orthologs, can correctly identifygenes for cell type– and tissue-specific processes. Using measurementsof transcript observations in 20 root cell types 31 that werenot used in building AraNet, we measured the extent to which geneslinked in AraNet were spatiotemporally co-expressed in these cells.We find that linked genes show strong cell-specific co-expression inArabidopsis (Fig. 3c)—indeed, far stronger than in previous networksof Arabidopsis genes (Supplementary Table 3) 27–30 —with linkedgenes four times more likely to be expressed in the same cell typesthan expected by chance. Thus, although different individual networkswere not constructed for each cell type, such cell and tissue specificityis nonetheless at least in part implicitly encoded in AraNet linkages.This correlation between functional association and spatiotemporalco-expression of genes likely enhances prediction strength for manytraits, and is evident even for linkages between characterized anduncharacterized genes (Fig. 3c), supporting applicability of AraNetto uncharacterized genes.Associating genes with specific mutant phenotypesBecause linked genes in AraNet tend to operate in the same processes(Figs. 1–4), we might expect that they often affect the same phenotypictraits 3,5 . This allows association of new candidate genes with traits ofinterest based on network connections. To test this, we used resultsfrom large-scale mutant seed phenotyping 32 and analyzed geneswhose disruption induced embryonic lethality or changes in seed(embryo) pigmentation. Genes involved in each trait were interlinkedsignificantly more often compared to chance (p < 10 −31 for embryoniclethality and P < 10 −10 for seed pigmentation, normal distribution)(Fig. 3d). Unlike AraNet, previous Arabidopsis gene networks 27–30 donot significantly predict either phenotype (Supplementary Fig. 4).Thus, AraNet offers a feasible approach for selecting genes likely tobe associated with specific plant traits.Tenfold enrichment for seed pigmentation genesTo experimentally test the association of new genes with a trait, weused 23 known seed pigmentation genes (Supplementary Table 4) tosearch AraNet for new pigmentation genes. Genes in this phenotypicclass generally affect chloroplast development or photomorphogenesis,and mutant seedlings show early developmental defects, with albino,pale green, purple or variegated leaves 33 .From AraNet’s top 200 candidate genes, we screened all geneswith available homozygous T-DNA insertional mutant lines(Supplementary Table 5). We screened 90 candidate genes (representedby 118 mutant lines), of which 14 genes (represented by 17lines) exhibited color and morphology defects in young seedlings,reminiscent of seed pigmentation mutants (Supplementary Tables 6and 7). This represents a tenfold enrichment in the discovery rate ofthe mutant phenotype (P ≤ 10 −12 , binomial distribution) over thatobserved during screens of T-DNA insertional lines 33 (see OnlineMethods). This discovery rate compares well to animal networks, forexample, in C. elegans 16 tumor suppressor effectors were identifiedfrom 170 candidates 5 .nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 151

e s o u r c e© 2010 Nature America, Inc. All rights reserved.a bc dArea under ROC curve1.00.90.80.70.60.50.4GO CCRandomAraNetAraNet plant data onlyAraNet no plant dataArea under ROC curve1.00.90.80.70.60.50.4Isozyme-free KEGGRandomAraNetAraNet plant data onlyAraNet no plant dataLikelihood ratio of linked genesco-expressing across 20 root cell typesOf the 14 genes with mutant phenotypes, 3 genes (AT5G45620,AT4G26430 (also known as CSN6B) and AT5G50110) exhibited thephenotypes in two alleles, 6 genes in only one of the two alleles, and5 genes were tested in only one allele (Fig. 5a and SupplementaryTable 7). The 6 genes in which only one of the two alleles showedphenotype are likely to be untagged and were not characterizedfurther. Expressivity of the phenotypes of the 11 lines representing8 genes (6 lines for 3 genes and 5 lines for 5 genes) varied among individualplants within the homozygous population, ranging from delayedor failed germination, arrested or delayed development, anthocyaninaccumulation, clear or white patches on the shoot to pale green shoot.As expected from known seed pigmentation mutants, survival rate insoil was

e s o u r c e© 2010 Nature America, Inc. All rights reserved.Table 5). The CSN and the lid subcomplex of 26S proteasome complexesshare structural and functional similarities 34 , suggestinginvolvement of other protein degradation machineries in photomorphogenesisand early seedling development. The self-pollinatedprogeny of mutants that survived to make seeds did not show theseedling defects under standard growth conditions as their progenitors(data not shown). However, when grown in dark or under bluelight, the mutants showed slight (5–25%) but significant (P < 0.01,paired t-test, see Online Methods) differences in hypocotyl length(Fig. 5c,d). CSN6B mutants showed reduced hypocotyl length in darkbut slightly increased hypocotyl length when grown under blue lightcompared to wild type. The other two mutants had longer hypocotylsthan wild type under both dark and blue light conditions. All threegenes have paralogs in the genome. As double mutants of CSN6Aand CSN6B show a constitutive photomorphogenic phenotype, severeseedling dwarfism and high levels of anthocyanin accumulation 36 , itis possible that AT4G26430 and AT5G45620 may also exhibit geneticredundancy with their respective paralogs.The remaining hits from our screen link to each other and knownseed pigmentation genes in three components relevant to thylakoidbiogenesis and chlorophyll biosynthesis, processes affecting chloroplastdevelopment and function (Fig. 5b). Supplementary Table 9details possible roles for the newly discovered genes. These results confirmthat AraNet can efficiently associate new genes with a specificphenotypic trait.Discovering functions for uncharacterized Arabidopsis genesGiven that AraNet can successfully associate genes with traits ofinterest, we wished to test hypothesized roles for uncharacterizedArabidopsis genes in planta. AraNet predicts biological roles for 4,479aWild typeAT5G45620-1 AT5G45620-2 AT4G26430-1 AT4G26430-2 AT5G50110-1 AT5G50110-2previously uncharacterized genes. We selected three uncharacterizedgenes (AT1G80710, AT2G17900 and AT3G05090) based on severalcriteria: (i) no known biological process assigned; (ii) predicted byAraNet to be involved in developmental regulatory processes; and(iii) exist as single-copy genes. These represent extremely stringenttests of the network-based association method, and are all cases inwhich prediction based on sequence homology has failed.AraNet predicts GO biological process annotations, ordering predictionsby the sum of the log likelihood scores linking a gene to genesalready annotated by each term (Supplementary Table 10). For thethree genes selected, we tested for morphological and physiologicalphenotypes in the top ten predicted processes. Two control genes,AT1G15772 and AT2G34170, were chosen randomly from genes lackingAraNet functional predictions. Mutant plants were confirmedfor homozygosity (Supplementary Table 11) and lack of detectabletranscripts (data not shown). Self-pollinated progeny of homozygousplants were subjected to a bank of phenotypic assays based on the topten predictions (see Online and Supplementary Methods). Of thethree mutants, two exhibited phenotypes in the predicted processes.AT1G80710 is a regulator of drought sensitivityAraNet implicated the gene AT1G80710 in the response to water deprivation,among other processes, drawing support from affinity purificationsof yeast orthologs (SC-MS) 37,38 (Supplementary Table 10).This gene is expressed in all tissues examined, with highest abundancein flowers (Supplementary Fig. 5). We asked whether the ability toretain water differed in the mutants. Under drought, mutant plantsretained ~80% of the water of wild type (P ≤ 0.001, unpaired t-test,Fig. 6a). Reduced water retention was not observed in control mutants(Supplementary Fig. 6).cWild typeAT4G26430-1AT4G26430-2AT5G45620-1AT5G45620-2AT5G15610-1DarkbCSN8AT4G26430CSN1CSN4CSN7PhotomorphogenesisAT5G45620AT5G15610COP1AT5G04110AT5G50110DET1ALB3AT5G11480CHLDCHLI1CHLGCHLHChloroplast developmentAT3G20160DXSKnown seed pigmentation genesNew seed pigmentation genes(two alleles tested)New seed pigmentation genes(one allele tested)Linkage by plant-derived dataLinkage by animal-derived dataLinkage by plant + animal data30d2520151050–5–10Figure 5 Discovery of seed (embryo) pigmentation defective genes predicted by AraNet guilt byassociation. (a) Seedling pigmentation defects are apparent in each of two independent alleles forthe genes AT5G45620, AT4G26430 (CSN6B) and AT5G50110, all predicted based on networkconnections to known pigmentation defect genes. (b) Eight new seed pigmentation defective genesare organized into five network components by connections to known seed pigmentation genes, withevidence for the connections coming both from plant- and animal-derived data sets. (c) Mutantslinked to known CSN genes show longer hypocotyl length than wild type in dark and under blue light,except CSN6B mutants, which show slightly shorter hypocotyls in dark. Scale bar, 1.3 cm. (d) Mostof the differences in hypocotyl length of these mutants are slight (5–25%) but significant. SignificantDifference in hypocotyllength (%)Blue lightAT4G26430-1DarkBlue light* *differences from wild type are indicated by asterisks (P < 0.01, paired t-test, n = 26 (dark), 32 (blue light)). n indicates the number of plates, each platecontaining 7–8 plants of wild type and mutant genotype. Results are from seven (dark) or eight (blue light) independent experiments, plotting mean ± s.e.m.AT4G26430-2**AT5G45620-1**AT5G45620-2***AT5G15610-1nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 153

e s o u r c e© 2010 Nature America, Inc. All rights reserved.aRelative water content(Fw – Dw)/(Tw – Dw)bWild type0.70.60.50.4drs1-1121080 µM ABA2.5 µM ABA5 µM ABA10 µM ABA0.360.20.14200Wild typeWatered Drought 0 5 10 15 20Time (h)Water loss (% fresh weight)Figure 6 Discovery of regulators of drought sensitivity andlateral root development from previously uncharacterizedgenes using AraNet. (a) Plants carrying a T-DNA insertion(drs1-1) in a previously uncharacterized gene, AT1G80710,retained significantly less water than wild type under drought.Relative water loss was calculated as (Fw − Dw)/(Tw − Dw)(Fw, fresh weight; Dw, dry weight; Tw, turgor weight).Significant differences were observed between the relativewater loss of wild type and mutant plants (P ≤ 0.001,unpaired t-test, n = 15) and between watered and droughtconditions of the same genotype (P ≤ 0.0001, unpairedt-test, n = 15). (b,c) Transpiration was reduced in wild typeplants in the presence of abscisic acid (ABA) in a dosagedependent manner (b), whereas mutant plants were insensitivecWater loss (% fresh weight)121086420Wild typeDrought response is mediated by several signaling pathways inArabidopsis, including the hormone abscisic acid (ABA), transcriptionfactor DREB2A, ERD1 and E3 SUMO ligase SIZ1 39,40 . To determinewhether the reduced water retention upon drought stress is mediatedby ABA, we examined the effect of ABA on transpiration of detachedleaves. The mutant was insensitive to ABA on water loss, whereas thewild type lost significantly less water in the presence of 10 µM ABA(P ≤ 0.0004, unpaired t-test, Fig. 6b,c). At this ABA concentration,mutant leaves lost 30% more water than wild type (P ≤ 0.04, unpairedt-test, Fig. 6b,c). ABA showed no effect on germination rate in themutant (data not shown), indicating that not all ABA-mediated processesare affected in the mutant.Both the water retention and ABA-insensitive water transpirationresponse segregated as a single recessive Mendelian locus and linked tothe T-DNA insertion (Supplementary Table 12 and SupplementaryFig. 7). We designate AT1G80710 as DRS1 (DROUGHT SENSITIVE 1)and the T-DNA insertion allele (Salk_001238C) as drs1-1. An independentT-DNA allele (Salk_149366C) that we designate as drs1-2 exhibitedthe same phenotypes in relative water content after drought and ABAinsensitivewater transpiration (Supplementary Fig. 8), confirmingthat the phenotypes are linked to mutations in DRS1. DRS1 is a WD-40repeat family protein containing a DWD (DDB1 binding WD-40)motif 41 . Some DWD-containing proteins are substrate receptorsfor DDB1-Cul4 ubiquitin ligase machinery in humans, yeast andArabidopsis 41,42 . Combination of AraNet prediction and experimentaltesting thus demonstrates that DRS1 promotes tolerance to droughtstress, possibly mediated by ABA, and suggests involvement of DDB1-Cul4–mediated protein degradation in drought response. Given thatthe a priori odds of selecting a gene affecting the response to waterdeprivation are ~1 in 318 (currently only 85 of 27,029 Arabidopsisd0 µM ABA2.5 µM ABA5 µM ABA10 µM ABAdrs1-10 5 10 15 20Time (h)Irs1-1eNumber of LR14121086420Total LRLRprimordiabeforeemergence(stages IV-VII)EmergedLRmeristems(stage VIII)Wild typeIrs1-1ElongatedLRIrs1-1 Wild type(pGWB2-LRS1) (pGWB2-LRS1) 1 nM lAA 10 nM lAAWild type Irs1-1 Wild type Irs1-1to ABA (c). (d) The number of lateral roots is strongly reduced in lines carrying a T-DNA insertion (lrs1-1) in another previously uncharacterized geneAT3G05090. This phenotype can be complemented by reintroduction of the functional gene. When additional copies of the gene are expressed in a wildtype strain, lateral roots increase, whereas the primary root decreases in length. Six other independent transformants in each background gave similarphenotypes (data not shown). 1 nM Auxin (IAA) increases the number and length of lateral roots in both the wild-type and mutant seedlings. Contrarily,10 nM IAA severely reduces the primary root length in both genotypes. Scale bar, 1.4 cm. (e) Different stages of the lateral root formation are affected inthe lrs1-1 mutant. Wild-type lateral roots are distributed fairly evenly among lateral root primordia, emerged lateral root and elongated lateral root.The mutant has reduced numbers of the lateral roots in all of these stages, though the reduction is more severe in the emerged and elongated lateralroot than that in the lateral root primordia. LR, lateral root. Error bars indicate s.e.m.genes are annotated for response to water deprivation), these testsstrongly support the network-based approach to rationally associateeven entirely uncharacterized genes with plant traits.AT3G05090 is a regulator of lateral root developmentThe second candidate gene, AT3G05090, was implicated in cell proliferationand meristem organization, drawing support from phylogeneticprofiling of bacterial homologs of Arabidopsis proteins anddomain co-occurrence patterns of yeast orthologs (SupplementaryTable 10). We examined both shoot and root development inat3g05090-1 seedlings. We did not observe shoot phenotypes, butthe number of lateral roots was significantly reduced (P ≤ 10 −37 ,unpaired t-test, Fig. 6d,e and Supplementary Fig. 9). This phenotypesegregated as a single recessive Mendelian locus linked to theT-DNA insertion (Supplementary Table 12 and Supplementary Fig.10a). The length of the primary root was shorter than in wild type(Fig. 6d) but this phenotype was unlinked to the T-DNA insertion(Supplementary Fig. 10b), showing that the lateral root phenotypeis separable and independent from the primary root phenotype. Wedesignate AT3G05090 as LRS1 (LATERAL ROOT STIMULATOR 1)and the at3g05090-1 allele as lrs1-1. Homozygous lines transformedwith a wild-type coding sequence driven under a 35S CaM viruspromoter complemented the lateral root phenotype (Fig. 6d andSupplementary Fig. 9). To determine if the lateral root formationis blocked before the lateral root meristem emergence, we examinedthe number of lateral root primordia and meristems (lateral rootstages IV–VIII 43 ). Wild-type lateral roots are distributed fairly evenlyamong lateral root primordia, emerged lateral root and elongatedlateral root (Fig. 6e). The mutant has reduced numbers of lateralroots at all of these stages, though the reduction is more severe in the154 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

e s o u r c e© 2010 Nature America, Inc. All rights reserved.emerged and elongated lateral root than in the lateral root primordia(Fig. 6e). Transforming wild-type lines with the 35S::LRS1 constructdid not increase the number of lateral roots, but we observed a dramaticincrease in the length of the lateral roots and a decrease in theprimary root length (Fig. 6d and Supplementary Fig. 9).Regulation of root architecture and function, modulated by bothintrinsic and extrinsic signals, is critical for efficient nutrient andwater use for plants. Auxin, a plant hormone, is a key regulator forlateral root development, including lateral root initiation, primordiumdevelopment and emergence 44 . The reduction in number oflateral roots in the mutant and the increase in lateral root lengthconcomitant with the decrease in the primary root length in the overexpressedlines evoke defects in auxin accumulation or perception 44 .We thus asked whether exogenous auxin could alleviate the phenotypeby growing plants in the presence of native auxin, indole acetic acid(IAA). IAA increased the lateral root number in the mutant (Fig. 6dand Supplementary Fig. 11a,b), demonstrating that auxin perceptionwas not altered in the mutant and suggesting that auxin accumulationis compromised in the mutant. Auxin accumulation can be altered bychanging synthesis, degradation, sequestration or transport 45 . To testfor auxin transport defects, we examined effects of an auxin transportinhibitor, N-(1-naphthyl)phthalamic acid (NPA) on root growth.NPA decreases both the number and length of lateral roots in bothgenotypes (Supplementary Fig. 11a,b). LRS1 encodes another DCAFprotein 41 , suggesting involvement of DDB1-Cul4–mediated proteindegradation in lateral root development. These results demonstratethat the lrs1-1 mutant is defective in lateral root development and suggestroles for DDB1-Cul4–mediated protein degradation in regulatingauxin accumulation during lateral root primordium development andlateral root meristem emergence, consistent with its hypothesizedroles in cell proliferation and meristem organization.DISCUSSIONWe demonstrate here that genes can be rationally associated with planttraits through guilt by association in a gene network. For this purpose,we created AraNet, a genome-wide gene network for A. thaliana,a reference organism for flowering plants, including many crops.AraNet is the most extensive gene network for any plant thus far; geneannotations derived by network guilt by association extend substantiallybeyond current gene annotations. We validated the network’spredictive power by cross-validation tests, independent pathway andphenotype data sets, cell-specific expression data sets, and by experimentson computationally selected candidate genes.AraNet generates at least two main types of testable hypotheses. Thefirst type uses a set of genes known to be involved in a specific processas bait to find new genes involved in that process. This test is useful ifthe bait genes are well connected (that is, high AUC). We used the setof genes conferring seed pigmentation defects (AUC = 0.68) as bait andfound a tenfold enrichment in identifying mutants with comparablephenotypes. Of the 318 GO biological processes with more than fivegenes, ~43% have AUCs of at least 0.68 (Supplementary Table 13), suggestingthat AraNet will be useful in identifying new genes in nearly halfof these biological processes. Similar distributions of AUCs are foundfor GO cellular component terms and KEGG pathways (SupplementaryTables 14 and 15). In practice, this translates into identifying a smallset of new genes from a relatively limited-scale screen of the topnetwork-predicted candidates (e.g., computer simulations suggestfinding an average of 4–7 novel genes from tests of the top 200 candidatesfor biological processes with AUC >0.6; Supplementary Fig. 12).The second type of hypothesis involves predicting functions foruncharacterized genes. We assayed predicted phenotypes for threeuncharacterized genes, two of which showed phenotypes in the predictedprocesses, response to drought and meristem development. Thereare 4,479 uncharacterized genes in AraNet (16% of protein-codinggenes) with links to characterized genes, suggesting broad utility forAraNet in identifying candidate functions. Both of these modes ofoperation can be easily performed on the AraNet website.Although AraNet currently shows high accuracy for many processes(Figs. 2–4), there are nonetheless specific processes that are poorlyrepresented, with this trend stronger among plant-specific processes(Fig. 2d). This trend manifested in our experimental validation of onlytwo of three tested candidate genes, although these intentionally representedchallenging cases lacking any current functional annotationand for which sequence homology approaches had failed. Althoughwe observed that non-plant–derived data sets helped identify genesfor plant-specific processes, it is clear that more plant data sets willenhance the utility of gene networks for finding trait-relevant genes.Three major causes underlie such cases of poor predictive performance.First, our current knowledge of genetic factors for a process may be sosparse that AraNet cannot link them efficiently. Second, AraNet maylack linkages or data relevant to the poorly predicted processes. Thesetwo trends likely explain the lower performance among plant-specificprocesses relative to more broadly studied, evolutionarily conservedprocesses. Additional plant-specific data sets, such as protein interactions,should help here, as should considering both indirect and direct networklinkages for ranking candidates. Third, strongly implicated candidategenes that nonetheless test negative for a trait, resulting in apparent falsepositives, might be masked by epistatic effects, thus actually representingtrue predictions and false-negative assay results. This trend may bereasonably common and has been previously observed in yeast 46 .AraNet represents a step toward the goal of computationally identifyinggene-trait associations in plants. This work suggests that genenetworks for food and energy crops will facilitate manipulation oftraits of economic importance and crop genetic engineering.MethodsMethods and any associated references are available in the onlineversion of the paper at http://www.nature.com/naturebiotechnology/.Note: Supplementary information is available on the Nature Biotechnology website.AcknowledgmentsWe are grateful to M. Ahn, A. Noorani and V. Bakshi for technical assistance, J. Shinfor assistance on AraNet web design, T. Nakagawa (Shimane University, Japan) forproviding pGWB2, K. Barton for providing lab space and D. Meinke, M. Running,W. Briggs, Z. Wang and K. Dreher for helpful discussions. This work was supportedby Carnegie Institution for Science (B.A., S.Y.R.), a grant from the National ScienceFoundation (MCB-0520140) to S.Y.R. and by the National Research Foundation ofKorea (NRF) grant funded by the Korean government (no. 2009-0063342, 2009-0070968) and Yonsei University (no. 2008-7-0284, 2008-1-0018) to I.L. and fromthe National Science Foundation, National Institutes of Health, and Welch (F1515)and Packard Foundations to E.M.M.AUTHOR CONRIBUTIONSI.L. and S.Y.R. conceived the project, I.L. created AraNet using approachesdeveloped with E.M.M., B.A. performed the experimental tests, P.T. assisted inseed pigmentation mutant analysis, S.Y.R. supervised the experimental tests, I.L.,E.M.M. and S.Y.R. analyzed AraNet and wrote the manuscript.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.Published online at http://www.nature.com/naturebiotechnology/.Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/.1. Alonso, J.M. et al. Genome-wide insertional mutagenesis of Arabidopsis thaliana.Science 301, 653–657 (2003).nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 155

e s o u r c e© 2010 Nature America, Inc. All rights reserved.2. Marcotte, E.M., Pellegrini, M., Thompson, M.J., Yeates, T.O. & Eisenberg, D.A combined algorithm for genome-wide prediction of protein function. Nature 402,83–86 (1999).3. McGary, K.L., Lee, I. & Marcotte, E.M. Broad network-based predictability ofSaccharomyces cerevisiae gene loss-of-function phenotypes. Genome Biol. 8, R258(2007).4. Fraser, H.B. & Plotkin, J.B. Using protein complexes to predict phenotypic effectsof gene mutation. Genome Biol. 8, R252 (2007).5. Lee, I. et al. A single gene network accurately predicts phenotypic effects of geneperturbation in Caenorhabditis elegans. Nat. Genet. 40, 181–188 (2008).6. Zhong, W. & Sternberg, P.W. Genome-wide prediction of C. elegans geneticinteractions. Science 311, 1481–1484 (2006).7. Lage, K. et al. A human phenome-interactome network of protein complexesimplicated in genetic disorders. Nat. Biotechnol. 25, 309–316 (2007).8. Franke, L. et al. Reconstruction of a functional human gene network, with anapplication for prioritizing positional candidate genes. Am. J. Hum. Genet. 78,1011–1025 (2006).9. Linghu, B., Snitkin, E.S., Hu, Z., Xia, Y. & Delisi, C. Genome-wide prioritization ofdisease genes and identification of disease-disease associations from an integratedhuman functional linkage network. Genome Biol. 10, R91 (2009).10. Huttenhower, C. et al. Exploring the human genome with functional maps. GenomeRes. 19, 1093–1106 (2009).11. Hermjakob, H. et al. IntAct: an open source molecular interaction database. NucleicAcids Res. 32, D452–D455 (2004).12. Alfarano, C. et al. The Biomolecular Interaction Network Database and related tools2005 update. Nucleic Acids Res. 33, D418–D424 (2005).13. Swarbreck, D. et al. The Arabidopsis Information Resource (TAIR): genestructure and function annotation. Nucleic Acids Res. 36, D1009–D1014(2008).14. de Folter, S. et al. Comprehensive interaction map of the Arabidopsis MADS Boxtranscription factors. Plant Cell 17, 1424–1433 (2005).15. Huynen, M., Snel, B., Lathe, W. III & Bork, P. Predicting protein function by genomiccontext: quantitative evaluation and qualitative inferences. Genome Res. 10,1204–1210 (2000).16. Pellegrini, M., Marcotte, E.M., Thompson, M.J., Eisenberg, D. & Yeates, T.O.Assigning protein functions by comparative genome analysis: protein phylogeneticprofiles. Proc. Natl. Acad. Sci. USA 96, 4285–4288 (1999).17. Wolf, Y.I., Rogozin, I.B., Kondrashov, A.S. & Koonin, E.V. Genome alignment,evolution of prokaryotic genome organization, and prediction of gene function usinggenomic context. Genome Res. 11, 356–372 (2001).18. Bowers, P.M. et al. Prolinks: a database of protein functional linkages derived fromcoevolution. Genome Biol. 5, R35 (2004).19. Dandekar, T., Snel, B., Huynen, M. & Bork, P. Conservation of gene order: a fingerprintof proteins that physically interact. Trends Biochem. Sci. 23, 324–328 (1998).20. Overbeek, R., Fonstein, M., D’Souza, M., Pusch, G.D. & Maltsev, N. The use ofgene clusters to infer functional coupling. Proc. Natl. Acad. Sci. USA 96,2896–2901 (1999).21. Lee, I., Li, Z. & Marcotte, E.M. An improved, bias-reduced probabilistic functional genenetwork of baker’s yeast, Saccharomyces cerevisiae. PLoS ONE 2, e988 (2007).22. Breitkreutz, B.J. et al. The BioGRID Interaction Database: 2008 update. NucleicAcids Res. 36, D637–D640 (2007).23. Chatr-aryamontri, A. et al. MINT: the Molecular INTeraction database. Nucleic AcidsRes. 35, D572–D574 (2007).24. Giot, L. et al. A protein interaction map of Drosophila melanogaster. Science 302,1727–1736 (2003).25. Remm, M., Storm, C.E. & Sonnhammer, E.L. Automatic clustering of orthologs andin-paralogs from pairwise species comparisons. J. Mol. Biol. 314, 1041–1052(2001).26. Ogata, H. et al. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic AcidsRes. 27, 29–34 (1999).27. Cui, J. et al. AtPID: Arabidopsis thaliana protein interactome database anintegrative platform for plant systems biology. Nucleic Acids Res 36, D999–D1008(2007).28. Geisler-Lee, J. et al. A predicted interactome for Arabidopsis. Plant Physiol. 145,317–329 (2007).29. Gutierrez, R.A. et al. Qualitative network models and genome-wide expression datadefine carbon/nitrogen-responsive molecular machines in Arabidopsis. Genome Biol.8, R7 (2007).30. Ma, S., Gong, Q. & Bohnert, H.J. An Arabidopsis gene network based on thegraphical Gaussian model. Genome Res. 17, 1614–1625 (2007).31. Brady, S.M. et al. A high-resolution root spatiotemporal map reveals dominantexpression patterns. Science 318, 801–806 (2007).32. Meinke, D., Muralla, R., Sweeney, C. & Dickerman, A. Identifying essential genesin Arabidopsis thaliana. Trends Plant Sci. 13, 483–491 (2008).33. McElver, J. et al. Insertional mutagenesis of genes required for seed developmentin Arabidopsis thaliana. Genetics 159, 1751–1763 (2001).34. Wei, N., Serino, G. & Deng, X.W. The COP9 signalosome: more than a protease.Trends Biochem. Sci. 33, 592–600 (2008).35. Peng, Z., Serino, G. & Deng, X.W. Molecular characterization of subunit 6 of theCOP9 signalosome and its role in multifaceted developmental processes inArabidopsis. Plant Cell 13, 2393–2407 (2001).36. Gusmaroli, G., Figueroa, P., Serino, G. & Deng, X.W. Role of the MPN subunits inCOP9 signalosome assembly and activity, and their regulatory interaction withArabidopsis Cullin3-based E3 ligases. Plant Cell 19, 564–581 (2007).37. Gavin, A.C. et al. Proteome survey reveals modularity of the yeast cell machinery.Nature 440, 631–636 (2006).38. Krogan, N.J. et al. Global landscape of protein complexes in the yeast Saccharomycescerevisiae. Nature 440, 637–643 (2006).39. Catala, R. et al. The Arabidopsis E3 SUMO ligase SIZ1 regulates plant growth anddrought responses. Plant Cell 19, 2952–2966 (2007).40. Shinozaki, K. & Yamaguchi-Shinozaki, K. Gene networks involved in drought stressresponse and tolerance. J. Exp. Bot. 58, 221–227 (2007).41. Lee, J.H. et al. Characterization of Arabidopsis and rice DWD proteins and theirroles as substrate receptors for CUL4-RING E3 ubiquitin ligases. Plant Cell 20,152–167 (2008).42. Jin, J., Arias, E.E., Chen, J., Harper, J.W. & Walter, J.C. A family of diverse Cul4-Ddb1-interacting proteins includes Cdt2, which is required for S phase destructionof the replication factor Cdt1. Mol. Cell 23, 709–721 (2006).43. Casimiro, I. et al. Dissecting Arabidopsis lateral root development. Trends PlantSci. 8, 165–171 (2003).44. Fukaki, H., Okushima, Y. & Tasaka, M. Auxin-mediated lateral root formation inhigher plants. Int. Rev. Cytol. 256, 111–137 (2007).45. Vanneste, S. & Friml, J. Auxin: a trigger for change in plant development. Cell 136,1005–1016 (2009).46. Li, Z. et al. Rational extension of the ribosome biogenesis pathway using networkguidedgenetics. PLoS Biol. 7, e1000213 (2009).156 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.ONLINE METHODSAll analyses are based on the set of 27,029 predicted protein coding loci ofArabidopsis (genome release version TAIR7) 13 . Reference and benchmark sets,raw data sets, and the construction and computational validation of AraNet aredescribed in Supplementary Methods (Supplementary Figs. 16–18).Targeted reverse genetics screening for seed pigmentation mutants.We searched AraNet with 23 confirmed seed pigmentation mutants(Supplementary Table 4) from the SeedGenes database 32 as bait. We retrievedthe homozygous T-DNA insertional lines for the top 200 candidates from theSIGnAL database 1 and obtained the stocks from the Arabidopsis BiologicalResource Center (Supplementary Tables 5 and 6). Seven to 9 seeds for eachline were sterilized as described below (Plant material). Seeds were stratifiedat 4 °C for 2 d in the dark and grown in Murashige and Skoog (MS) mediumwith 1% sucrose under continuous illumination of 50–80 µmol/m 2 s at 22 °C.Seedlings were observed under a dissecting microscope (Leica MZ125) 6 d aftergermination and followed up 10–12 d after germination. For each of the lineswhere sufficient seeds were available, the assay was conducted at least twice.T-DNA insertions and genotypes of the seeds for the 11 lines with the mutantphenotypes described in this study (6 lines representing two alleles of threegenes and 5 lines representing single alleles of five genes) were confirmed usingPCR (Supplementary Methods and Supplementary Table 16).To determine the significance of the discovery rate, we used the results of alarge-scale screening of T-DNA insertional mutants for embryo defective orseed pigmentation mutants 32 as the background rate. This study found ~1,260seed pigmentation mutants from screening 120,000 T-DNA lines. Because thiswas a forward-genetics screening whereas our screen was reverse-genetic (thatis, preselected for intragenic insertions), we adjusted the total number of linesin the background to 84,000 based on the genome-wide distribution of T-DNAinsertion sites of 70% insertion events in intragenic regions 1 .Candidate selection for uncharacterized genes and experimental validationof mutant phenotypes. To test the predictive power of AraNet in planta, weanalyzed mutant phenotypes of genes of unknown function, whose biologicalroles were inferred by the annotations of the neighbors of these genes inAraNet. Of the 27,029 protein-encoding genes in Arabidopsis, 14,847 have noinformation about the biological processes in which they are involved. Morethan half of these uncharacterized genes (7,465 genes) are included in AraNet.Of these, 4,479 genes are inferred to be associated with specific biologicalprocesses based upon annotated AraNet neighbors (using only IDA, IMP, IGI,IPI, IEP and TAS evidence). To test the accuracy of such inferences made byAraNet, we chose three genes to characterize experimentally. These were chosenon the basis of available homozygous knockout lines, absence of paralogsand AraNet inferences of involvement in specific biological processes. Thegenes chosen were AT3G05090, AT1G80710 and AT2G17900, whose top tenAraNet predictions are shown in Supplementary Table 10. From the predictions,we assayed all of the phenotypes that could be measured with availableresources at Carnegie. For AT1G80710, the following were tested: response towater deprivation (rank 3); trichome differentiation (rank 8); leaf development(rank 9). For AT3G05090, the following phenotypes were tested: trichomedifferentiation (rank 1); leaf development (rank 3); cell proliferation (rank 5);meristem organization (rank 8); and regulation of flower development(rank 10). For AT2G17900, the following were tested: meristem organization(rank 2); leaf morphogenesis (rank 3); hyperosmotic salinity response (rank 4);brassinosteroid mediated signaling (rank 5); multidimensional cell growth(rank 6); response to auxin stimulus (rank 7); detection of brassinosteroidstimulus (rank 8). In addition, we selected two genes randomly, AT1G15772and AT2G34170, which were included in AraNet but were neighbors of otheruncharacterized genes, to test specificity of all observed phenotypes.Plant material. Seeds of homozygous T-DNA knockout mutants were obtainedfrom the Arabidopsis Biological Resource Center. The stock numbers for the118 seed pigmentation candidate genes are listed in Supplementary Table 6.Seeds for the five uncharacterized genes were SALK_059570C (AT3G05090),SALK_001238C (AT1G80710), SALK_127952C (AT2G17900), Salk_118634C(AT1G15772) and Salk_099804C (AT2G34170). For experiments conducted insoil, seeds were sown in soil (Premier Pro-mix) supplemented with fertilizer(Osmocote Classic, Hummert International). Seeds were stratified at 4 °C inthe dark for 2 d and grown under 16/8 h of light/dark (90–100 µmol/m 2 s) and30% humidity at 22 °C. For experiments conducted in agar plates, seeds weresurface sterilized with 15% commercial bleach (6.25% sodium hypochlorite),containing a few drops of Tween-20 detergent and rinsed with sterile waterfive times. Seeds were sown on agar plates containing 0.43% MS salts, 0.5%2-(N-Morpholino)ethanesulfonic acid (MES), 0.5% sucrose, 0.8% agar,pH 5.7. Plants on agar plates were grown under constant illumination of50–80 µmol/m 2 s at 22 °C. For root assays, 50 ml of the MS medium wasprepared poured into 100 × 100 × 15 mm square plates (Fisher Scientific) 1 dbefore planting to minimize plate-to-plate variability.T-DNA insertions were confirmed, and genetic linkage, complementation,and overexpression tests were performed as described in SupplementaryMethods (Supplementary Tables 16 and 17).Visible phenotype assays. The following traits were observed by naked eyeand using dissecting (Leica MZ125) and compound (Nikon Eclipse E600 withNomarski optics) microscopes throughout the life cycle of the mutant plants:trichome differentiation (observations made on rosette and cauline leaves andsepals), leaf development and morphogenesis, cell proliferation, meristemorganization and multidimensional cell growth (observations made on leaf,floral, inflorescence and root organs). To detect phenotypes in the regulationof flower development, we observed floral organs and flowering time underlong days (16/8 of light/dark) and short days (8/16 of light/dark).Hypocotyl length measurements. Seeds were germinated and grown verticallyfor 4 d in dark or under 4 µM/m 2 s continuous blue light. Seven to eightseeds of mutant and wild-type Col-0 (a reference strain that is the geneticbackground for the T-DNA mutants) were planted per plate and two platesper genotype were tested in each experiment. Each condition was tested in 7–8independent experiments. Hypocotyl length was measured using ImageJ onphotographs of the plates after 4 d of growth. The average hypocotyl length ofeach genotype was determined from each plate and the difference in hypocotyllength between wild type and mutant was determined using one-tailed,paired t-test.Root length and number measurements. Seeds were germinated and grownvertically. Root measurements were taken 10–11 d after germination. Lateralroots were counted using a dissecting microscope or from digital images ofplants using ImageJ. Different stages of the lateral roots were determined usinga compound microscope. The root length was measured by tracing the lengthof the root using ImageJ on digital images of the seedlings.Auxin response. Auxin and auxin transport inhibitor treatments were carriedout as described 47 . Seeds were sown on MS agar medium and grownunder continuous light. After 4 d, seedlings were transferred to either MS agarmedium (control) or MS agar medium containing 1 nM, 5 nM, 10 nM or 30 nMindole acetic acid (IAA, Sigma) or 1 nM, 10 nM, 100 nM or 1 µM of naphthylphthalamicacid (NPA, Chem Service). Both wild type (Col-0) and mutantseedlings were transferred to the same plates. On the tenth or eleventh day ofgrowth, the primary root length, number of lateral roots and length of lateralroots were measured as described above. Significant differences were determinedby unpaired t-test. Each experiment was conducted with 2–3 plates of7–10 plants each of wild type and mutant per plate. At least three independentexperiments were carried out for each hormone assay.Abscisic acid (ABA) response. The effect of ABA on detached leaf transpirationwas determined as described 48 with some modifications: plants weregrown in soil under long-day conditions (16/8 h light/dark) under whitelight of 90–100 µM m −2 s −1 at 22 °C. The largest, fully-open rosette leaves of4–week-old plants were excised at the bottom of the petioles and were placedinto a Parafilm-sealed 1.5 ml centrifuge tubes containing 1.4 ml of 0, 2.5 µM,5 µM and 10 µM of ABA in an artificial xylem sap solution (15 mM KNO 3 ,1 mM CaCI 2 , 0.7 mM MgSO 4 , and 1 mM (NH 4 ) 2 HPO 4 , with pH adjustedto 5.0 with 1 M phosphoric acid 49 ). Transpiration was measured by weighingtotal weight of the tubes at times 0, 2, 4, 6 and 22 h. All of the excisionstook place between 10 am and noon (4–6 h after the onset of illumination).doi:10.1038/nbt.1603nature biotechnology

For the F2 linkage test of drs1-1, four leaves were excised from each plant andtwo were treated with the sap solution only and the other two were treatedwith 10 µM of ABA in the sap solution. Details on the linkage test are foundin Supplementary Methods. For wild-type and mutant comparisons, eachexperiment used two to four leaves from three to four plants per genotypeat each time point and was conducted in triplicate. Four independentexperiments were conducted.Drought response assay. Response to water deprivation was determinedby measuring relative water content as described 50 . Plants were grownin soil under long-day conditions (16/8 h light/dark) under white light of90–100 µM m −2 s −1 at 22 °C for 4–5 weeks. Watering was stopped for thedrought treatment and relative water content was measured on day 0, 4, 7and 10 of droughting. Control plants were watered every 2–3 d. To measurerelative water content, we excised plants at the shoot/root junction, removedany bolts and weighed rosettes to determine the fresh weight (Fw). Therosettes were then completely submerged in water for 4 h and weighed todetermine the turgid weight (Tw). Rosettes were then dried overnight at80 °C and weighed to obtain the dry weight (Dw). Three plants from eachgenotype for each condition were measured. Relative water content wascalculated as (Fw − Dw)/(Tw − Dw) and the significance of differenceswas determined by unpaired t-test. Three plants of each genotype wereused for each time point per experiment. Four independent experimentswere conducted.An interactive web tool for AraNet-based candidate gene selection is availableat http://www.functionalnet.org/aranet/.47. Cho, H.T. & Cosgrove, D.J. Regulation of root hair initiation and expansin geneexpression in Arabidopsis. Plant Cell 14, 3237–3253 (2002).48. Munns, R. & King, R.W. Abscisic acid is not the only stomatal inhibitor in thetranspiration stream of wheat plants. Plant Physiol. 88, 703–708 (1988).49. Goodger, J.Q., Sharp, R.E., Marsh, E.L. & Schachtman, D.P. Relationships betweenxylem sap constituents and leaf conductance of well-watered and water-stressedmaize across three xylem sap sampling techniques. J. Exp. Bot. 56, 2389–2400(2005).50. Giraud, E. et al. The absence of ALTERNATIVE OXIDASE1a in Arabidopsis resultsin acute sensitivity to combined light and drought stress. Plant Physiol. 147,595–610 (2008).© 2010 Nature America, Inc. All rights reserved.nature biotechnologydoi:10.1038/nbt.1603

B r i e f c o m m u n i c at i o n s© 2010 Nature America, Inc. All rights reserved.Enhanced antibody half-lifeimproves in vivo activityJonathan Zalevsky 1,3 , Aaron K Chamberlain 1 ,Holly M Horton 1 , Sher Karki 1 , Irene W L Leung 1 ,Thomas J Sproule 2 , Greg A Lazar 1 , Derry C Roopenian 2 &John R Desjarlais 1Improved affinity for the neonatal Fc receptor (FcRn) is knownto extend antibody half-life in vivo. However, this has neverbeen linked with enhanced therapeutic efficacy. We testedwhether antibodies with half-lives extended up to fivefold inhuman (h)FcRn transgenic mice and threefold in cynomolgusmonkeys retain efficacy at longer dosing intervals. We observedthat prolonged exposure due to FcRn-mediated enhancementof half-life improved antitumor activity of Fc-engineeredantibodies in an hFcRn/Rag1 –/– mouse model. This bridges thedemand for dosing convenience with the clinical necessity ofmaintaining efficacy.The well-established role of FcRn in IgG turnover has been the foundationfor Fc engineering efforts aimed at improving the pharmacokineticproperties of therapeutic antibodies 1,2 . Despite contraryresults about the relationship between FcRn affinity and half-life 3,4 ,several such efforts at pharmacokinetic engineering in nonhumanprimates, whose FcRn is similar to that of humans, have demonstratedthat engineered antibody variants have a prolonged half-life 5–8 . Yet,although the successful extension of half-life in pharmacokineticexperiments bodes well for the prospect of improving clinical dosing,a critical gap remains. For half-life extension technologies to beof practical use, efficacy of a biotherapeutic with longer half-life mustbe preserved at longer dosing intervals. Although the relationshipbetween drug exposure and efficacy is well-established, this correlationhas not thus far been established for antibodies engineered forlonger half-life.We coupled rational design methods with high-throughput proteinscreening to engineer a series of Fc variants with greater affinity forhuman FcRn. Variants were constructed in the context of the humanizedanti-vascular endothelial growth factor (VEGF) IgG1 antibodybevacizumab 9 (Avastin), which is currently approved for the treatmentof colorectal, lung, breast and renal cancers. A description of theconstruction, production and binding studies of the antibodies is providedin Supplementary Methods. As FcRn binds IgG at the lower pHof the early endosome (pH 6.0–6.5) but not at the higher pH of blood(pH 7.4), we used Biacore to screen antibodies for binding to humanFcRn at pH 6.0. Our engineered variants demonstrated between3- and 20-fold greater binding to FcRn at pH 6.0, with improvementsdue almost exclusively to slower off-rate (k off ) (Supplementary Fig. 1and Supplementary Table 1). A lead variant, M428L/N434S, subsequentlyselected principally based on its pharmacokinetic performance(see below), provided an 11-fold improvement in FcRnaffinity at pH 6.0. We refer to this double substitution in the context ofbevacizumab as Xtend-VEGF.Details of a pharmacokinetic study in cynomolgus monkeys(Macaca fascicularis) to evaluate the capacity of the variants toimprove serum half-life are provided in the Supplementary Methods.Binding improvements of the variants to monkey FcRn at pH 6.0were comparable to improvements for human FcRn, and the rankorder of the variants in FcRn affinity was the same (data not shown).When three monkeys per group were injected intravenously with4 mg/kg variant or native IgG1 anti-VEGF antibody, we observed alarge improvement in half-life for the variants relative to native IgG1(Supplementary Fig. 2a). Fitted parameters for the full set of variants(Supplementary Table 2) indicated increases in β-phase half-life, areaunder the curve (AUC) measurements and the rate of antibody clearancefrom serum. The observed 9.7-d half-life for native IgG1 bevacizumabagrees with the published value (9.3 d) for a slightly lower(2 mg/kg) dose 10 . Among the engineered antibodies that were tested,the Xtend double variant performed best (Fig. 1a). It prolonged halflifefrom 9.7 to 31.1 d, a 3.2-fold improvement in serum half-liferelative to native IgG1 (Supplementary Table 2). Simple allometricscaling extrapolations suggest that such improvement can potentiallytranslate into human half-lives >50 d.We then sought to further challenge the applicability of pharmacokineticengineering by targeting an internalizing cell-surface antigenthat potentially provides a competing sink for antibody clearance.aConcentration (µg/ml)1001010.10 1020 30 40 50DaysXtend-VEGFIgG1 Bevacizumab60 70 80 90bConcentration (µg/ml)10010Xtend-EGFRIgG1 Cetuximab10 5 10 15 20DaysFigure 1 Increasing antibody affinity to FcRn promotes half-life extensionin cynomolgus monkeys. (a) Log-linear changes in serum concentrations foranti-VEGF (bevacizumab) antibodies in cynomolgus monkeys. All antibodieswere administered by single 60-min intravenous infusion at 4 mg/kg andserum antibody concentrations were determined using a VEGF antigendownimmunoassay. Results are means ± s.e.m. (n = 2 for bevacizumaband n = 3 for variants). (b) Log-linear changes in serum concentrations foranti-EGFR antibodies in cynomolgus monkeys. Monoclonal antibodies wereadministered by single 30-min intravenous infusion at 7.5 mg/kg and serumantibody concentrations were determined using an EGFR antigen-downimmunoassay. Results are means (n = 2 animals per test article).1 Xencor, Inc., Monrovia, California, USA. 2 The Jackson Laboratory, Bar Harbor, Maine, USA. 3 Present address: Takeda San Diego, Inc., San Diego, California, USA.Correspondence should be addressed to J.R.D. (jrd@xencor.com).Received 20 October 2009; accepted 14 December 2009; published online 17 January 2010; doi:10.1038/nbt.1601nature biotechnology volume 28 number 2 February 2010 157

i e f c o m m u n i c at i o n s© 2010 Nature America, Inc. All rights reserved.a100Concentration (µg/ml)cMean tumor volume (mm 3 )1010.1 Xtend-VEGFIgG1 Bevacizumab0 5 10 15 20DaysDay2565 751,200 Vehicle1,000 IgG1 BevacizumabXtend-VEGF800600400200035 45 55*85Concentration (µg/ml)Several studies have demonstrated that antibodies to epidermalgrowth factor receptor (EGFR) are internalized. Moreover, nonlineardose-dependent clearance has been observed in monkeys andhumans, leading to the hypothesis that receptor-dependent internalizationmakes a major contribution to clearance of anti-EGFRantibodies 11,12 . The M428L/N434S Xtend variant was constructed ina humanized version (huC225) of the anti-EGFR antibody cetuximab(C225) 13 (Erbitux), which is approved for the treatment of colorectaland head and neck cancers. We refer to this pharmacokineticallyenhanced anti-EGFR antibody as Xtend-EGFR. The improvementin affinity for human FcRn resembled that observed for anti-VEGF;binding to human EGFR antigen was unperturbed, and both cetuximaband humanized cetuximab cross-react with cynomolgus EGFR 14(data not shown). The 7.5 mg/kg dose chosen for this study is in arange where the dose-clearance relationship is nonlinear 14 . In ourhands cetuximab had a half-life of 1.5 d (Supplementary Table 2),similar to previously published data at the same dose (2.7–3.1 d) 14 .Consistent with the bevacizumab results, the Xtend variant anti-EGFR increased half-life to 4.7 d, reflecting a 3.1-fold improvement(Fig. 1b and Supplementary Table 2). We have thus demonstratedpharmacokinetic improvements conferred by Fc engineering of aninternalizing antibody, even when it is dosed within the nonlinearclearance regime.We performed pharmacokinetic experiments in C57BL/6J (B6)-background mice that are homozygous for a knockout allele of murineFcRn and heterozygous for a human FcRn transgene (mFcRn –/– ,bdMean tumor volume (mm 3 )1001010.10.01Xtend-EGFRIgG1 Cetuximab0 5 10 15 20DaysDay251,000Vehicle800 IgG1 CetuximabXtend-EGFR600400*200010 20 30 40Figure 2 Improved antibody half-life translates to greater in vivo efficacy.(a) Log-linear changes in serum concentrations of anti-VEGF antibodies inhFcRn mice. All antibodies were administered via single intravenous bolusat 2 mg/kg, and serum antibody concentrations were determined using ahuman immunoglobulin recognition immunoassay. Results are means± standard errors (n = 6). For some data points, errors are smaller than canbe indicated. (b) Log-linear changes in serum concentrations of anti-EGFRantibodies in hFcRn mice. The study design was identical to that describedin a, except that serum concentrations were measured with an EGFRantigen-down immunoassay. (c) Xenograft study in hFcRn/Rag1 –/– micecomparing activity of native IgG1 and Xtend variant versions of bevacizumabagainst established SKOV-3 tumors. Tumor volume is plotted against dayafter tumor cell injection. Antibodies were dosed at 5 mg/kg every 10 dstarting on day 35 (indicated by the arrows). n = 8 mice/group. *, P = 0.028at 84 d. (d) Xenograft study in hFcRn/Rag1 –/– mice comparing activity ofanti-EGFR antibodies against established A431 tumors. Tumor volume isplotted against day after tumor cell injection. Antibodies were dosed 5 mg/kgevery 10 d starting on day 10 (indicated by the arrows). n = 9 mice/group.*, P = 0.005 at 35 d.hFcRn + ) 15 , referred to here as hFcRn mice. A description of theseexperiments is provided in the Supplementary Methods. Serumconcentration data for native IgG1 and Xtend anti-VEGF antibodiesshowed a dramatic enhancement in half-life for the variant relativeto native IgG1 (Fig. 2a), improving half-life fourfold from ~3–12 d(Supplementary Table 2). In the anti-EGFR context, the Xtendvariant improved half-life to 13.9 d relative to 2.9 d for cetuximab,resulting in an enhancement of about fivefold (Fig. 2b andSupplementary Table 2). The IgG1 version of huC225 also had a relativelyshort half-life of 2 d (data not shown). We observed a generalcorrelation between antibody half-life and FcRn affinity at pH 6.0across two anti-VEGF studies and one anti-EGFR hFcRn pharmacokineticstudy. The pharmacokinetic results for individual variantsand native IgG1 were consistent and reproducible between the threestudies (Supplementary Fig. 2b–c and Supplementary Table 2).To test whether the slower clearance of our pharmacokineticengineeredantibodies results in improved exposure-related pharmacology,we developed an hFcRn transgenic, Rag1 –/– immunodeficientmouse strain (Supplementary Methods and Supplementary Fig. 3).For VEGF, SKOV-3 tumors were established to 25–60 mm 3 and thentreated with either vehicle or 5 mg/kg native IgG1 or Xtend variantbevacizumab every 10 d. This dosing schedule approximated the halflifeof the Xtend variant, but was three to four half-lives longer thanthe half-life of the native IgG1 version (Supplementary Table 2). Astatistically greater level of tumor reduction (P = 0.028 at study termination)was observed for the Xtend variant relative to the native IgG1version (Fig. 2c). A similar study in hFcRn/Rag1 –/– mice comparingXtend-EGFR to a native IgG1 version showed similar improvementsin tumor reduction (P = 0.005) against established A431 epidermoidcarcinoma tumors (Fig. 2d). Consistent with the pharmacokineticresults in hFcRn mice (Fig. 2a–b), the variants reduced clearance inthe hFcRn/Rag1 –/– mice (Supplementary Fig. 3a–b), demonstratingan inverse correlation between tumor volume and serum concentrationof antibody at study termination. These results indicate that theslower clearance of the variant antibodies leads to higher drugexposure and consequently superior tumor-suppressing pharmacology.Additional studies comparing various dosing intervals of theXtend variants and parent antibodies will be necessary to preciselydefine dosing regimens for optimal clinical benefit. However, theresults described here firmly establish a positive correlation betweenpharmacokinetic enhancement and in vivo efficacy.Despite the reasonably long half-lives of monoclonal antibodies,market pressures for higher patient convenience and compliance continueto drive antibody drug programs toward less frequent dosingschedules. Yet, because of the potential loss in efficacy when the dosingfrequency is not justified by the pharmacokinetics of the drug, thecritical issue of whether slower antibody clearance through Fc engineeringleads to superior exposure-dependent efficacy has remainedunresolved. Our results indicate that, for at least some therapies,efficacy can be preserved with extended dosing intervals enabled bypharmacokinetic engineering. This work thus paves the way for a newgeneration of antibody therapies and biologically superior versions ofapproved antibody drugs that deliver finer control over dosing whileproviding greater convenience to patients.Note: Supplementary information is available on the Nature Biotechnology website.AcknowledgmentsWe thank The Jackson Laboratory JAX West and SNBL USA for carrying outpharmacokinetic experiments, B. Dahiyat for helpful discussions, andA. Eivazi, D.-H.T. Nguyen, H. Herman, J.M. Jacinto and U.S. Muchhal fortechnical contributions.158 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

i e f c o m m u n i c at i o n sAUTHOR CONTRIBUTIONSJ.Z., A.K.C., H.M.H., G.A.L., D.C.R. and J.R.D. designed the research, J.Z., A.K.C.,H.M.H., S.K., I.W.L.L. and T.J.S. carried out experiments, and J.Z., G.A.L. andJ.R.D. wrote the manuscript.COMPETING INTERESTS STATEMENTThe authors declare competing financial interests: details accompany the full-textHTML version of the paper at http://www.nature.com/naturebiotechnology/.Published online at http://www.nature.com/naturebiotechnology/.Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/.1. Roopenian, D.C. & Akilesh, S. Nat. Rev. Immunol. 7, 715–725 (2007).2. Presta, L.G.. Curr. Opin. Immunol. 20, 460–470 (2008).3. Datta-Mannan, A., Witcher, D.R., Tang, Y., Watkins, J. & Wroblewski, V.J. J. Biol.Chem. 282, 1709–1717 (2007).4. Gurbaxani, B., Dela Cruz, L.L., Chintalacharuvu, K. & Morrison, S.L. Mol. Immunol.43, 1462–1473 (2006).5. Dall′Acqua, W.F., Kiener, P.A. & Wu, H.. J. Biol. Chem. 281, 23514–23524(2006).6. Hinton, P.R. et al. J. Biol. Chem. 279, 6213–6216 (2004).7. Hinton, P.R. et al. J. Immunol. 176, 346–356 (2006).8. Yeung, Y.A. et al. J. Immunol. 182, 7663–7671 (2009).9. Presta, L.G. et al. Cancer Res. 57, 4593–4599 (1997).10. Lin, Y.S. et al. J. Pharmacol. Exp. Ther. 288, 371–378 (1999).11. Fan, Z., Lu, Y., Wu, X. & Mendelsohn, J. J. Biol. Chem. 269, 27595–27602(1994).12. Lammerts van Bueren, J.J. et al. Cancer Res. 66, 7630–7638 (2006).13. Naramura, M., Gillies, S.D., Mendelsohn, J., Reisfeld, R.A. & Mueller, B.M. CancerImmunol. Immunother. 37, 343–349 (1993).14. Imclone Systems, Inc Biologic License Application 125084, Erbitux (Cetuximab)(US Food and Drug Administration, Feb. 12, 2004). 〈http://www.accessdata.fda.gov/drugsatfda_docs/bla/2004/125084_ERBITUX_PHARMR_P2.PDF〉.15. Petkova, S.B. et al. Int. Immunol. 18, 1759–1769 (2006).© 2010 Nature America, Inc. All rights reserved.nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 159

lettersExpansion and maintenance of human embryonicstem cell–derived endothelial cells by TGFb inhibitionis Id1 dependentDaylon James 1 , Hyung-song Nam 2,7,8 , Marco Seandel 1,3,8 , Daniel Nolan 1 , Tyler Janovitz 1 , Mark Tomishima 4 ,Lorenz Studer 4 , Gabsang Lee 4 , David Lyden 1 , Robert Benezra 2 , Nikica Zaninovic 5 , Zev Rosenwaks 5 ,Sina Y Rabbany 1,6 & Shahin Rafii 1© 2010 Nature America, Inc. All rights reserved.Previous efforts to differentiate human embryonic stem cells(hESCs) into endothelial cells have not achieved sustainedexpansion and stability of vascular cells. To define vasculogenicdevelopmental pathways and enhance differentiation, we usedan endothelial cell–specific VE-cadherin promoter driving greenfluorescent protein (GFP) (hVPr-GFP) to screen for factors thatpromote vascular commitment. In phase 1 of our method,inhibition of transforming growth factor (TGF)β at day 7 ofdifferentiation increases hVPr-GFP + cells by tenfold. In phase 2,TGFβ inhibition maintains the proliferation and vascular identityof purified endothelial cells, resulting in a net 36-fold expansionof endothelial cells in homogenous monolayers, which exhibited atranscriptional profile of Id1 high VEGFR2 high VE-cadherin +ephrinB2 + . Using an Id1-YFP hESC reporter line, we showedthat TGFβ inhibition sustains Id1 expression in hESC-derivedendothelial cells and that Id1 is required for increasedproliferation and preservation of endothelial cell commitment.Our approach provides a serum-free method for differentiationand long-term maintenance of hESC-derived endothelial cells ata scale relevant to clinical application.Human embryonic stem cells (hESCs), which self-renew indefinitely 1 ,offer a plentiful source of endothelial cells for therapeutic revascularization.However, few studies have identified specific developmental stimulisufficient to support the specification and maintenance of large numbersof functional and vascular-committed endothelial cells from hESCs 2–7 .Although small numbers of hESC-derived endothelial cells have beengenerated in short-term cultures, these cells have not been subjected tosustained expansion, angiogenic profiling or interrogated as to the stabilityof vascular fate. As a result, molecular pathways that maintain vascularidentity and long-term expansion of hESC-derived endothelial cellsremain unknown.To detect the emergence of endothelial cells from differentiating hESCsin real time, we generated a cell line for endothelial cell–specific lineagetracing. We cloned a 1.5-kilobase fragment from a bacterial artificial chromosome(BAC) containing the genomic locus of the human endothelialcell–specific gene VE-cadherin (CDH5). The promoter sequence of thisgene, encompassing a region upstream of exon 1, was inserted into a lentiviralvector upstream of GFP (hVPr-GFP; Fig. 1a). Human endothelial cellstransduced with this vector showed robust expression of GFP, in contrastto transduced human mesenchymal and fibroblastic cells, which did notexpress GFP (Supplementary Fig. 1a–c). Endothelial-specific expressionof the reporter was also evident in transduced, spontaneously differentiatinghESCs (RUES1 line; Fig. 1b and Supplementary Fig. 1d–j): hVPr-GFP +cells were organized into capillary-like structures expressing endothelialcell markers, including VE-cadherin, CD31 and CD34 (SupplementaryFigs. 1d–g and 2a,b), and were negative for alpha smooth muscle actin(α-SMA) and CD45, a marker of hematopoietic cells (SupplementaryFigs. 1h–j and 2c).Using the hVPr-GFP hESC reporter line, we tracked the chronology andgeometry of vasculogenic differentiation in differentiating embyroid bodiesby time-lapse confocal microscopy. Beginning at day 5, we observed thespecification and emergence of hVPr-GFP + cells (Supplementary Video 1and Supplementary Fig. 3), and by day 8, hVPr-GFP + cells co-expressingvascular endothelial growth factor receptor (VEGFR)2 and CD31 (Fig. 1c)formed motile vessel-like structures (Supplementary Video 2). These datavalidated the ability of the hVPr-GFP reporter construct to specificallyidentify and track hESC-derived nascent endothelial cells.We used the reporter line to develop a chemically defined, serum-freemethod for enhancing vascular differentiation. In phase 1, heterogenousembryoid body cultures of hVPr-GFP hESCs were sequentially stimulatedwith bone morphogenetic protein (BMP)4, activinA, fibroblast growthfactor (FGF)-2 and VEGF-A 8–10 (Fig. 1d). Although these growth conditionspromoted formation of hVPr-GFP + structures (Supplementary Fig. 4and Supplementary Videos 3 and 4), the yield of dissociated hVPr-GFP +endothelial cells obtained by fluorescence-activated cell sorting (FACS)was low, and these few isolated endothelial cells could not be expandedwithout the majority of cells assuming a non-endothelial cell phenotype1 Howard Hughes Medical Institute, Ansary Stem Cell Institute, Department of Genetic Medicine, Weill Cornell Medical College, New York, New York, USA. 2 Program inCancer Biology and Genetics, Memorial Sloan-Kettering Cancer Center, New York, New York, USA. 3 Division of Medical Oncology, Department of Medicine, MemorialSloan-Kettering Cancer Center, New York, New York, USA. 4 Developmental Biology Program, Memorial Sloan-Kettering Cancer Center, New York, New York, USA.5 Ronald O. Perelman and Claudia Cohen Center for Reproductive Medicine, New York, New York, USA. 6 Bioengineering Program, Hofstra University, Hempstead, NewYork, USA. 7 Present address: Weill Cornell Medical College, New York, New York, USA. 8 These authors contributed equally to this work. Correspondence should beaddressed to S.R. (srafii@med.cornell.edu).Received 1 December 2009; accepted 8 January 2010; published online 17 January 2010; doi:10.1038/nbt1605nature biotechnology volume 28 number 2 february 2010 161

lettersa1.5 kb hVPrIntron 112.5 kbATGbhVPr-GFPc10 510 4LTRExon 164 bpExon 2226 bpα-CD3110 3dU3 U5 GA RRE eGFP WPRE U3U5SuspensionAdherent10 20−10 2−100−10 2 010 2−203Day (−1) 0 1 247 1410 3 10 4α-VEGFR2Isolation10 5BMP-4Activin AFGF-2Phase 1differentiationMesodermal and vascular specificationVEGF-ASB431542EC amplificationPhase 2isolation© 2010 Nature America, Inc. All rights reserved.ehVPr-GFPfNuc VECad hVPr-GFPPropidium iodide10 310 210 1g−SBh+SB10 0 0.15% 1.8%10 0 10 1 10 2 10 3 10 0 10 1 10 2 10 3hVPr-GFPi% hVPr-GFP + cells2.01.61.20.80.4Day 14Grouped IsolatedP < .001 P < .01d0 + N d7 + N d7 +Figure 1 Sequential TGFβ activation followed by inhibition during phase 1 differentiation promotes a tenfold expansion of hVPr-GFP + hESC-derived cells.(a) A 1.5-kb fragment of the putative human VE-cadherin promoter (hVPr) region was isolated from a BAC clone and placed upstream of GFP in a lentiviralexpression vector (hVPr-GFP). (b) Spontaneously differentiating embryoid bodies exhibited expression of hVPr-GFP in tubular structures. Inset, merge of GFPand brightfield views. (c) Flow cytometric analysis showed hVPr-GFP + cells were positive for the vascular markers CD31 and VEGFR2. (d) Schematic diagramshowing the sequence in which BMP4, activinA, FGF-2, VEGF-A and SB431542 were added and removed from differentiation cultures. EC, endothelial cell.(e) Adherent hVPr-GFP cultures stimulated with SB431542 (10 µM) resulted in differentiation of hESCs into monolayers of hVPr-GFP + adherent cells.Inset, hVPr-GFP + cells alone. (f) Human VPr-GFP + cells were immuno-positive for VE-cadherin. Blue, nuclear counterstain. Inset, hVPr-GFP + cells alone.(g,h) The proportion of hVPr-GFP + cells was measured by flow cytometry at day 14 after culture in the absence (–SB; g) and presence (+SB; h) of SB431542.(i) Measurement of hVPr-GFP + cells at day 14 when embryoid bodies were cultured either in groups or as isolated embryoid bodies and SB431542 was addedat day 0, day 7 or not at all (N). Error bars represent s.d. of experimental values performed in triplicate. Scale bars, 100 µm.(data not shown). We therefore screened for bioactive small moleculesthat would improve vascular differentiation. Screening of >20 moleculesassociated with early developmental signaling pathways (SupplementaryTable 1) showed that the TGFβ-inhibitory molecule SB431542 (ref. 11)reproducibly increased the yield of hVPr-GFP + cells. Adding SB431542to differentiation cultures at day 7 resulted in the formation of hVPr-GFP + VE-cadherin + monolayers (Fig. 1e,f), which, upon dissociation,yielded tenfold more hVPr-GFP + endothelial cells than cultures stimulatedby cytokines alone (Fig. 1g–i). No hVPr-GFP + cells were generatedif SB431542 was added at the onset of differentiation (day 0), suggestingthat vascular commitment depends on active TGFβ/activin/nodal signalingbefore day 7.Kinetic analysis of differentiation suggested a shift from a pluripotentphenotype (Oct3/4 + ; Fig. 2a) to a vascular phenotype (CD31 + ;Fig. 2b,c) through a mesodermal intermediate (brachyury + ; Fig. 2a).Addition of SB431542 to differentiating hESC cultures at day 7 acceleratedthe reduction of Oct3/4 and brachyury and increased the numberof hVPr-GFP + CD31 + cells beginning at about day 9, while reducingexpression of α-SMA (Fig. 2b,c). After isolation from heterogenouscultures by FACS, endothelial cells grown in the absence of TGFβ inhibitionretained high expression of CD31 but also expressed α-SMA(Supplementary Video 5), indicating that these endothelial cell–likecells had not assumed a terminally committed vascular fate.Expression of α-SMA in hESC-derived endothelial cells suggesteda degree of plasticity that is not present in terminally differentiatedendothelial cells (human umbilical vein endothelial cells; HUVEC,Fig. 2b). Indeed, extended culture (>10 d after FACS isolation) of hESCderivedendothelial cells in the absence of TGFβ inhibition yielded asubstantial number of cells co-expressing VE-cadherin and α-SMA(Fig. 2d). One explanation for the increased percentage of endothelialcells in SB431542-stimulated cultures is maintenance of the vascularcommittedstate after specification. To test the capacity of TGFβ inhibitionto promote expansion of pure populations of hESC-derivedendothelial cells, we dissociated day 14 differentiation cultures, isolated162 volume 28 number 2 february 2010 nature biotechnology

lettershVPr-GFP + cells by FACS and expanded them for an additional 5 d withor without SB431542 (Phase 2, Fig. 2e–i). SB431542-treated culturesyielded more cells in the 5-d period, and a higher percentage of thepopulation retained an α-SMA − CD31 + VE-cadherin + phenotype (Fig.2e–h). In addition to preserving the vascular phenotype, SB431542also increased cell proliferation, as indicated by a higher percentageof phospho-histoneH3 + (PHH3) mitotic endothelial cells (Fig. 2i andSupplementary Videos 6 and 7).In aggregate, TGFβ inhibition in phase 1 and 2 resulted in a 36-foldexpansion in the total number of vascular-committed hESC-derivedendothelial cells, with 7.4 CD31 + VE-cadherin + endothelial cells generatedfrom every one hESC input over 20 d, compared with 0.2 endothelial cellsper input hESC derived from control culture conditions (Fig. 2j). Similarlevels of expansion of hESC-derived endothelial cells were achieved withfour additional hESC lines and one induced pluripotent stem cell lineusing the same protocol except that either SB431542 or soluble TGFβRII© 2010 Nature America, Inc. All rights reserved.Relative expression (a.u.)a10,0001,000dj100101Brachyury − SBBrachyury + SBNuc SMA VECadeOct3/4 − SBOct3/4 + SBNucUndifferentiated hESCsNon-vascular cellsSmooth muscle cellsEndothelial cellsbRelative expression (a.u.)600500400300200100VECad −SBfαSMA − SBαSMA + SBCD31 − SBCD31 + SBNuc−SBVECad+SBCell number (k)

letters© 2010 Nature America, Inc. All rights reserved.abα-CD31 antibodyhVPr-GFP EBsd14 (−SB)10 510 410 3hESC derivedSorted hVPr-GFP+d14 (+SB)Sorted hVPr-GFP+d24 (+SB)CD31 +id1 low CD31 +10 2 10 2 10 3 10 4 10 5ld1 promoter: YFPUmbilical cord derivedEndothelium(HUVEC)Smooth musclecellsCord bloodCD34 + cells0.0 1.0GenesCXCR4EphrinB2EphB4Lyve-1Podoplaninc ld1:YFP dExpression0.5 2 3 4 5: Belowdetection thresholdProx-1VEGFR1 (Flt1)VEGFR2VEGFR3 (Flt4)CD34CD133 (Prominin1)E-SelectinP-SelectinPECAM (CD31)ThrombomodulinVE-cadherinvonWillebrand factorHOXA9ld1ld2ld3α-Smooth muscle actinld1:YFPefld1:YFP MFI (k)ld1 low (−SB)302418126IIIICD31 + ld1 low−SBI+SBd14 id1:YFP FACS3 Days (+SB)ld1 low (+SB)IVld1:YFP MFIII4 Daysld1 high (−SB)CD31 + ld1 high−SBVCD31 MFIII+SB5k 5k 11.2k 21.7k 18.3k 53.8kIII IV V VITotal cellsld1 high (+SB)VI352821147CD31 MFI (k)Figure 3 Molecular profiling of hESC-derived endothelial cells reveals a signature defined by high Id1 expression. Human VPr-GFP embryoid bodies andhighly purified hVPr-GFP + cells were compared to mature vascular cells by microarray analysis. (a) RNA was extracted for microarray analysis from humanVPr-GFP embryoid bodies cultured in the presence of recombinant cytokines alone until day 14; isolated endothelial cells (99.8% pure) from hVPr-GFPembryoid bodies cultured in the presence of recombinant cytokines and the TGFβ inhibitor SB431542 until day 14; isolated endothelial cells (>95%pure) from hVPr-GFP embryoid bodies cultured in the presence of recombinant cytokines and the TGFβ inhibitor SB431542 until day 14, followed by 10 dadditional culture in the presence of cytokines and SB431542; HUVECs; human umbilical vein smooth muscle cells; and CD34 + umbilical cord blood cells.Expressed factors are displayed in an ordered array, as labeled, with highly expressed factors shown in red, minimally expressed factors shown in blue andfactors for which transcripts are below a significant expression level shown in gray. (b–f) Following the endothelial cell differentiation protocol (Fig. 1d), Id1-YFP hESC-derived cells were sorted by FACS, separating the CD31 + population into Id1-YFP high -expressing cells (b (green) and c) and Id1-YFP low -expressingcells (b (red) and d). Insets, brightfield views on the day after isolation of Id1-YFP high (c) and Id1-YFP low (d) cells. (e) After 3 d culture in the presence ofSB431542, both populations were transferred to conditions with and without SB431542 for an additional 4 d (+SB and –SB, respectively). (f) Total cellsand mean fluorescence intensity (MFI) measurements of Id1:YFP (black) and CD31 + (white) were measured for: CD31 + Id1 low (I) and CD31 + Id1 high (II)populations upon isolation; and for four populations following culture conditions (as shown in e, III–VI). Scale bars, 100 µM.receptor decoys was used interchangeably to inhibit activation of theactivin/nodal branch of TGFβ superfamily signaling (SupplementaryFig. 5a–e). These results demonstrate that the effect of TGFβ inhibitionshown for the RUES1 line is applicable to other pluripotent cell lines.To define the vasculogenic transcriptional signature of hESC-derivedendothelial cells at different time points during phases 1 and 2, we carriedout Affymetrix microarray analyses of several hESC-derived populationsand mature cell types (Fig. 3a). The yield of freshly isolated phase 1endothelial cells in the absence of TGFβ inhibition was insufficient formicroarray analyses, underscoring the value of our approach for generatingsufficient expanding (phase 1) and vascular-committed (phase 2)endothelial cells for molecular profiling.Phase 1 hESC-derived endothelial cells showed increased levels of factorstypical of arterial-like endothelial cells (VEGFR2, VEGFR1, Id1, CD31,CD34, VE-cadherin, vWF, thrombomodulin, ephrinB2 and E-selectin) butnot of lymphatic endothelial cells (Prox1 and podoplanin). Markers ofvascular progenitor cells, including CD133 and Id1 (refs. 12–17), were alsohighly expressed in phase 1 endothelial cells and downregulated upon in vitroculture. Transcription factors expressed primarily in committed endothelialcells, including HoxA9 (ref. 18), were not expressed in phase 1 endothelialcells. Accordingly, we defined a comprehensive vasculogenic expressionprofile of the hESC-derived endothelial cell population as VE-cadherin+ VEGFR2 high Id1 high thrombomodulin high ephrinB2 + CD133 + HoxA9 − ,whereas mature endothelial cells were identified by a VE-cadherin + VEGFR2 low Id1 low ephrinB2 + CD133 − HoxA9 + phenotype.Id1 was one of numerous transcription factors upregulated in phase 1endothelial cells. Because it has been shown to modulate differentiationand maintenance of vascular cell fate 19 , we focused on Id1 as a potentialmediator of the pro-angiogenic effect of TGFβ-inhibition observed in ourstudy. To track Id1 expression in live hESC differentiation cultures, we used164 volume 28 number 2 february 2010 nature biotechnology

lettersα-CD31e10 510 410 3ahVPrGFPhVPrGFP1.31%(+/− 0.3)VEGFR214.6%(+/− 2.6)fbα-VEGFR2hVPrGFPhVPrGFP0.56%(+/− 0.2)VEGFR26.2%(+/− 1.4)10 2 −140 0 10 2 10 3 10 4 10 5 −170 0 10 2 10 3 10 4 10 5Relative Id1 expressionc20015010050gScrId1ScrId1d75k50k25khScrTotal cellsCD31 + cellsHUVEC hVprGFP −SB +SBd14+8 (+SB) VE-cadherin Nuc hVPr-GFP hVPr-GFP GIB4−Cell numberId1ScrId1i© 2010 Nature America, Inc. All rights reserved.Figure 4 TGFβ inhibition upregulates Id1 expression and is necessary for the increased yield of functional endothelial cells capable of in vivo neoangiogenesis.(a,b) Human VPr-GFP hESCs that were stably transduced with control (a) or Id1-specific (b) shRNAs were differentiated according to theprotocol shown in Figure 1d and assessed at day 14 for the prevalence of VEGFR2 + (blue) and hVPr-GFP + (green) cells. The insets show plots of sidescatter on the y axis and hVPr-GFP on the x axis. (c) Control and Id1-specific shRNAs were added to HUVEC or freshly isolated (at day 14) hVPr-GFP +cells, and the relative Id1 transcript levels were measured after 3 d. *, P < 0.05. Error bars, s.d. of experimental values performed in triplicate.(d) Control and Id1-specific shRNAs were added to freshly isolated hVPr-GFP + cells, which were cultured in the absence or presence of SB431542.After 5 d, the total cell number and proportion of CD31 + cells was measured by flow cytometry. Error bars, s.d. of experimental values performed intriplicate. Scr, scrambled control shRNA. (e–g) Human VPr-GFP + cells were isolated by FACS at day 14 and expanded in monolayer culture (e) for8 d while retaining expression of both the endogenous VE-cadherin (f) and the hVPr-GFP transgene (g). Panel e shows a mosaic view of one well ofa 24-well dish. A magnified view of the boxes in e and f are shown in f and g, respectively. (h,i) Expanded cells were injected in Matrigel plugs intoimmunodeficient mice and excised after 10 d following intravital labeling of functional vasculature with lectin (GIB4, blue). h, View of hVPr-GFP + cellsalone; i, view of hVPr-GFP + cells merged with GIB4 + cells. Scale bars, 100 µM.a stable BAC transgenic hESC line 20 containing yellow fluorescent proteindriven by the Id1 promoter (Id1-YFP) (Fig. 3b–f) (Nam, H.S. and Benezra,R., unpublished data). Differentiated endothelial cells were isolated at day14 from Id1-YFP cultures (Fig. 1d), sub-fractionating the CD31 + populationinto Id1-YFP high-expressing (Fig. 3c) and low-expressing (Fig. 3d)cells, and these populations were serially expanded for 7 d with or withoutthe TGFβ inhibitor (Fig. 3e,f). Flow cytometric analysis of these cellsrevealed a direct relationship between upregulation of Id1 expression andTGFβ inhibition. Notably, although SB431542 increased the percentage ofthe CD31 + population, the mean fluorescence intensity of CD31 on thesecells was lower than that of unstimulated cells. These data suggested thatTGFβ inhibition increased expansion of hESC-derived endothelial cellsby maintaining high levels of Id1 expression and preserving an immatureproliferative phenotype.To determine the requirement for Id1 in mediating endothelial cellcommitment, we transduced hVPr-GFP + cells with lentiviral short hairpin(sh)RNA targeted against the Id1 transcript (Fig. 4a,b). In the presence ofSB431542, knockdown of Id1 reduced the numbers of both VEGFR2 + vascularprogenitors and hVPr-GFP + cells at day 14. When the Id1 shRNA constructwas introduced after isolation of the hVPr-GFP + fraction (Fig. 4c),it elicited a marked decrease in CD31 + endothelial cells after 5 d ofSB431542 treatment (Fig. 4d). These results identified TGFβ inhibition–mediated Id1 upregulation as a primary effector in promoting endothelialcell expansion and maintaining long-term vascular identity.To demonstrate that our cultured endothelial cells could form functionalvessels, we grew purified hVPr-GFP + cells from day 14 differentiationcultures for an additional 8 d in the presence of SB431542. Theseendothelial cells showed high proliferative potential (up to ten cell divisions)and generated homogenous hVPr-GFP + VE-cadherin + monolayers(Fig. 4e–g) with retention of hVPr-GFP fluorescence at the single-celllevel (arrowheads in Fig. 4g). These cells were subcutaneously injected inMatrigel plugs into nonobese (NOD)/severe combined immunodeficient(SCID) mice and 10 d later extracted after intravenous injection of lectininto live animals. In Matrigel plugs, hVPr-GFP + cells co-localized withlectin + cells, forming chimeric vessels along with host cells (Fig. 4h–i andSupplementary Videos 8 and 9). These data indicated that the endothelialcells generated by our methods could function in vivo.A prerequisite to therapeutic vascularization using hESC-derived cellsis generation of abundant durable endothelial cells that upon expansionmaintain their angiogenic profile without differentiating into nonendothelialcell types. Here, we show that differentiation of hESCs into alarge number of stable and proliferative endothelial cells can be achievedby early-stage TGFβ-mediated mesoderm induction followed by TGFβinhibition beginning at day 7 (phase 1) and after isolation at day 14 (phase2). Using this approach, we achieved a 36-fold net expansion of committedendothelial cells. The increased yield allowed transcriptional analysis,which revealed a molecular signature that sheds light on the regulatoryinfluences that govern embryonic vasculogenesis. Indeed, genes encodingnature biotechnology volume 28 number 2 february 2010 165

letters© 2010 Nature America, Inc. All rights reserved.factors associated with vascular progenitor identity (Id1 high , VEGFR2 high ,CD133) 12–17,19 as well as vascular markers (PECAM, VE-cadherin, ephrinB2)were highly expressed in hESC-derived endothelial cells and, amongthese factors, Id1 was found to act downstream of TGFβ inhibition toincrease endothelial cell yield by promoting proliferation and preservingvascular commitment. These studies establish TGFβ modulation of Id1expression as a determinant of hESC-derived endothelial cell identity andset the stage for large-scale generation of authentic long-lasting humanendothelial cells for therapeutic vascularization.Our use of vascular-specific hVPr-GFP and Id1-YFP hESC reporterlines in small-molecule screens allowed the discovery of the TGFβ inhibitorSB431542 as a key stimulus for human endothelial cell differentiationand proliferation in serum-free conditions. In murine ESCs, TGFβ andserum factors promote smooth muscle cell differentiation, whereas inhibitionof this pathway promotes formation of CD31 + cells 21 . Our datashow that stage-specific TGFβ inhibition, beginning on day 7 at a pointfollowing TGFβ-mediated mesoderm induction, increases the mitoticindex and maintenance of hESC-derived endothelial cells by upregulationof Id1 expression. Differentiation of hVPr-GFP hESCs with TGFβinhibition generated endothelial cells at yields tenfold greater than thoseof cells differentiated with angiogenic factors alone, and after purification,TGFβ inhibition supported endothelial cell expansion for up to ten celldivisions while retaining the angiogenic surface phenotype. The ability ofTGFβ inhibition to increase endothelial cell yield in both differentiating(phase 1) and purified (phase 2) cultures resulted in a 36-fold increasein the absolute number of hESC-derived endothelial cells, with 95% ofthe population maintaining endothelial cell identity. As such, we haveestablished a means of generating a homogenous population of stableendothelial cells in ratios that greatly exceed hESC input and are relevantto therapeutic vasculoplasty.Expression of Id1 has been shown to inhibit cell differentiation andgrowth arrest in multiple cell types 22 . The TGFβ signaling pathway,through the effectors Smad3 and ATF3, has been shown to repress Id1 promoteractivity 23 . The link between TGFβ signaling, Id1 and preservationof proliferation and phenotypic identity of hESC-derived endothelial cellsprovides insight into the molecular mechanisms that regulate vascularontogeny during human development. Indeed, these results point towarda biphasic role for TGFβ signaling during vasculogenesis, whereby earlyactivation of this pathway is required for specification of mesodermal progenitors,and inhibition after vascular commitment functions to increasemitotic index and prevent the loss of endothelial identity. Our approachfor vascular monitoring and differentiation may enable the identificationof as-yet unrecognized vasculogenic and angiogenic modulators forpreclinical studies aimed at the cell-based therapeutic revascularizationof ischemic tissues.METHODSMethods and any associated references are available in the online versionof the paper at http://www.nature.com/naturebiotechnology/.Accession codes. GEO: GSE19735.ACKNOWLEDGMENTSWe thank A. Brivanlou for providing the RUES1 hESC line. D.J., M.S. and G.L. areFiona and Stanley Druckenmiller Fellows of the New York Stem Cell Foundation. S.R. issupported by Howard Hughes Medical Institute; Ansary Stem Cell Institute; Anbinderand Newmans Own Foundation; National Heart, Lung, and Blood Institute R01 grantsHL075234 and HL097797; Qatar National Priorities Research Program; and EmpireState Stem Cell Board and New York State Department of Health, NYS C024180.Author ContributionsD.J. designed and performed the experiments and wrote the manuscript. H.-s.N.and R.B. designed and created the Id1-YFP BAC transgenic vector. M.S. performedexperiments and contributed to the manuscript. D.N. performed flow cytometricexperiments. T.J. performed molecular cloning. M.T. and L.S. generated the Id1-YFP BAC transgenic hESC line. L.S. and G.L. generated the FD iPSC line. N.Z.and Z.R. generated the hESC lines WMC2, WMC8 and WMC9. D.L. and S.Y.R.designed experiments and performed data analysis. S.R. designed experiments andwrote the manuscript.Note: Supplementary information is available on the Nature Biotechnology website.COMPETING INTERESTS STATEMENTThe authors declare no competing financial interests.Published online at http://www.nature.com/naturebiotechnology/.Reprints and permissions information is available online athttp://npg.nature.com/reprintsandpermissions/.1. Thomson, J.A. et al. Embryonic stem cell lines derived from human blastocysts. Science282, 1145–1147 (1998).2. Yamahara, K. et al. Augmentation of neovascularization in hindlimb ischemia by combinedtransplantation of human embryonic stem cells-derived endothelial and muralcells. PLoS ONE 3, e1666 (2008).3. Sone, M. et al. Pathway for differentiation of human embryonic stem cells to vascularcell components and their potential for vascular regeneration. Arterioscler. Thromb.Vasc. Biol. 27, 2127–2134 (2007).4. Lu, S.J. et al. Generation of functional hemangioblasts from human embryonic stemcells. Nat. Methods 4, 501–509 (2007).5. Goldman, O. et al. A boost of BMP4 accelerates the commitment of human embryonicstem cells to the endothelial lineage. Stem Cells 27, 1750–1759 (2009).6. Nourse, M.B. et al. VEGF induces differentiation of functional endothelium from humanembryonic stem cells: implications for tissue engineering. Arterioscler. Thromb. Vasc.Biol. 30, 80–89(2009).7. Bai, H. et al. BMP4 regulates vascular progenitor development in human embryonicstem cells through a smad-dependent pathway. J. Cell Biochem. published online,doi:10.1002/jcb.22410 (30 November 2009).8. Huber, T.L., Kouskoff, V., Fehling, H.J., Palis, J. & Keller, G. Haemangioblast commitmentis initiated in the primitive streak of the mouse embryo. Nature 432, 625–630(2004).9. Levenberg, S., Zoldan, J., Basevitch, Y. & Langer, R. Endothelial potential of humanembryonic stem cells. Blood 110, 806–814 (2007).10. Yang, L. et al. Human cardiovascular progenitor cells develop from a KDR+ embryonicstem-cell-derivedpopulation. Nature 453, 524–528 (2008).11. Inman, G.J. et al. SB-431542 is a potent and specific inhibitor of transforming growthfactor-beta superfamily type I activin receptor-like kinase (ALK) receptors ALK4, ALK5,and ALK7. Mol. Pharmacol. 62, 65–74 (2002).12. Gehling, U.M. et al. In vitro differentiation of endothelial cells from AC133-positiveprogenitor cells. Blood 95, 3106–3112 (2000).13. Kelly, M.A. & Hirschi, K.K. Signaling hierarchy regulating human endothelial cell development.Arterioscler. Thromb. Vasc. Biol. 29, 718–724 (2009).14. Peichev, M. et al. Expression of VEGFR-2 and AC133 by circulating human CD34(+)cells identifies a population of functional endothelial precursors. Blood 95, 952–958(2000).15. Rafii, S. & Lyden, D. Cancer. A few to flip the angiogenic switch. Science 319,163–164 (2008).16. Gao, D. et al. Endothelial progenitor cells control the angiogenic switch in mouse lungmetastasis. Science 319, 195–198 (2008).17. Lyden, D. et al. Impaired recruitment of bone-marrow-derived endothelial andhematopoietic precursor cells blocks tumor angiogenesis and growth. Nat. Med. 7,1194–1201 (2001).18. Rossig, L. et al. Histone deacetylase activity is essential for the expression of HoxA9and for endothelial commitment of progenitor cells. J. Exp. Med. 201, 1825–1835(2005).19. Ruzinova, M.B. & Benezra, R. Id proteins in development, cell cycle and cancer. TrendsCell Biol. 13, 410–418 (2003).20. Placantonakis, D.G. et al. BAC transgenesis in human embryonic stem cells as a noveltool to define the human neural lineage. Stem Cells 27, 521–532 (2009).21. Watabe, T. et al. TGF-beta receptor kinase inhibitor enhances growth and integrity ofembryonic stem cell-derived endothelial cells. J. Cell Biol. 163, 1303–1311 (2003).22. Jankovic, V. et al. Id1 restrains myeloid commitment, maintaining the self-renewalcapacity of hematopoietic stem cells. Proc. Natl. Acad. Sci. USA 104, 1260–1265(2007).23. Kang, Y., Chen, C.R. & Massague, J. A self-enabling TGFbeta response coupled to stresssignaling: Smad engages stress response factor ATF3 for Id1 repression in epithelialcells. Mol. Cell 11, 915–926 (2003).166 volume 28 number 2 february 2010 nature biotechnology

letters© 2010 Nature America, Inc. All rights reserved.ONLINE METHODSHuman ESC culture. The experiments delineated in this report were performedprimarily with the recently approved RUES1 hESC (generous gift fromA. Brivanlou 24 ), and corroborated using WMC2, WMC7, WMC8, generated atWeill Cornell Medical College (courtesy of Z. R./N.Z.) and H9 (Id1-YFP, courtesyof R.B./H.-s.N. and L.S./M.T.) and IPSc (courtesy L.S./G.L.). The permissions foruse of these cell lines were obtained after comprehensive review by the Cornell-Rockefeller-Sloan Kettering Institute ESC research oversight committee. The fundingfor execution of these studies was secured from approved non-federal fundingresources. Human ESC culture medium consisted of Advanced DMEM/F12(Gibco) supplemented with 20% knockout serum replacement (Invitrogen), 1×non-essential amino acids (Gibco), 1× l-glutamine (Invitrogen), 1× penicillin/streptomycin (Invitrogen), 1× β-mercaptoethanol (Gibco) and 4 ng/ml FGF-2(Invitrogen). Human ESCs were maintained on Matrigel using hESC mediumconditioned by mouse embryonic fibroblasts (Chemicon).Lentiviral vectors and transduction. Supernatants containing infectious particleswere collected 40 and 68 h after transfection of HEK 293T with hVPr-GFP alongwith accessory vectors as previously described 25 . Viral supernatants were concentratedby ultracentrifugation and used to transduce undifferentiated RUES1hESCs. After two passages, hESCs were disaggregated by accutase to form singlecells, which were isolated and expanded to form multiple parallel cultures, eachcontaining a relatively consistent level of viral incorporation. After expansion,these cultures were differentiated as adherent embryoid bodies and screened forthe presence of GFP + cells.Id1-YFP hESC reporter line and lentiviral Id1 shRNA knockdown. A bacterialartificial chromosome (BAC) was modified to place YFP under control ofthe endogenous Id1 promoter locus. This construct was electroporated into theH9 hESC line, selected for BAC integration using antibiotic resistance and subcloned.Clones were assessed and selected based on expression of YFP in Id1hESC derivatives after spontaneous differentiation. The Id1 and control shRNAlentiviral constructs were obtained from Open Biosystems and viral particles wereassembled according to the manufacturer’s recommendations (pLKO LentiviralPackaging System).Embryoid bodies. Human VPr-GFP hESCs were grown to confluence on Matrigel(BD Biosciences) and then incubated in 5 units/ml dispase (Gibco) until colonieswere completely detached from the substrate. Human VPr-GFP embryoidbodies were washed and cultured in hESC medium on ultra-low attachmentplates (Corning) and cultured in the conditions described, with replacement ofcytokine-supplemented medium every 48 h. Embryoid bodies were fixed in 4%paraformaldehyde and frozen for cryosectioning and staining.Endothelial differentiation protocols. Embryoid bodies were generated andcultured in base hESC medium, supplemented with the cytokines as shown.Sequential administration of cytokines was implemented (Fig. 1d). Briefly,embryoid bodies were generated in hESC base medium without FGF-2. On themorning after generation of embryoid bodies (day 0), medium was supplementedwith 20 ng/ml BMP4 (R&D Systems) (removed at day 7); on day 1, medium wassupplemented with 10 ng/ml activinA (R&D Systems) (removed at day 4); on day2, medium was supplemented with 8 ng/ml FGF-2 (Peprotech) (remained forthe duration of culture); on day 4, embryoid bodies were transferred to adherentconditions on Matrigel-coated plates and medium was supplemented with25 ng/ml VEGF-A (Peprotech) (remained for the duration of culture); on day7, SB431542 (Tocris) was added at 10 µM concentration and remained for indicatedduration. Cultures were dissociated using 0.5% Trypsin/EDTA (Gibco) orAccutase (eBioscience). Absolute yield as well as ratio of input hESCs to differentiatedendothelial cells was calculated from the number of live cells recoveredfrom differentiation cultures at days 0, 14 and 20. Purified endothelial cells couldbe frozen and thawed in 10% DMSO with >90% recovery.Quantitative PCR. Total RNA was prepared from cultured cells using theRNeasy extraction kit (Qiagen) and reverse transcribed using Superscript IIreverse transcriptase (Invitrogen) according to the manufacturer’s instructions.Relative quantitative PCR was performed on a 7500 Fast Real Time PCR System(Applied Biosystems) using either TaqMan PCR mix along with Id1 and β-actinprimer pairs, or SYBR Green PCR mix (Applied Biosystems). Human-specificSYBR green primer pairs used were: PECAM – f, 5′-tctatgacctcgccctccacaaa–3′,r, 5′ gaacggtgtcttcaggttggtatttca-3′; Oct3/4 - f, 5′-aacctggagtttgtgccagggttt-3′, r,5′-tgaacttcaccttccctccaacca-3′; Brachyury – f, 5′-cagtggcagtctcaggttaagaagga-3′,r, 5′-cgctactgcaggtgtgagcaa-3′; and a-SMA, f, 5′-aatactctgtctggatcggtggct-3′, r,5′-acgagtcagagctttggctaggaa-3′. Cycle conditions were: one cycle at 50 °C for 2min followed by 1 cycle at 95 °C for 10 min followed by 40 cycles at 95 °C for15s and 60 °C for 1 min. Primers were checked for amplification in the linearrange and primer dissociation and verified. Threshold cycles of primer probeswere normalized to the housekeeping gene β-actin (ACTB) and translated torelative values.Endothelial cell isolation and flow cytometry. Endothelial cells were isolatedfrom differentiation cultures using Magnetic Activated Cell Sorting (MACS;Miltenyi Biotech) with an antibody against CD31 conjugated to magnetic microbeads.Alternatively, cells were isolated by virtue of the expression of GFP/YFPor a fluorophore conjugated antibody to human CD31 or VEGFR2 (BD) usinga FACSAriaII (BD).Microarray analysis. The Affymetrix Human Genome U133 2.0 array was usedto analyze gene expression. In brief, using Qiagen RNeasy kits, total RNA wasextracted from: Human VPr-GFP embryoid bodies that were cultured in thepresence of recombinant cytokines alone until day 14; MACS-sorted endothelialcells isolated from hVPr-GFP embryoid bodies cultured in the presence of recombinantcytokines alone until day 14; MACS-sorted endothelial cells isolated fromhVPr-GFP transduced embryoid bodies cultured in the presence of recombinantcytokines and the TGFβ inhibitor SB431542 until day 14; MACS-sorted endothelialcells isolated from hVPr-GFP embryoid bodies cultured in the presence ofrecombinant cytokines and the TGFβ inhibitor SB431542 until day 14, followedby 10 d additional culture in the presence of cytokines and SB431542; humanumbilical vein endothelial cells; human umbilical vein smooth muscle cells; andCD34 + umbilical cord blood cells. The Superscript choice kit (Invitrogen) wasused to make cDNA with a T7-(dT)24 primer incorporating a T7 RNA polymerasepromoter. The biotin-labeled cRNA was made by in vitro transcription(Enzo Diagnostics). Fragmented cRNA was hybridized to the gene chips, washed,and stained with streptavidin phycoerythrin. The probe arrays were scanned withthe Genechip System confocal scanner and Affymetrix Microarray suite 4.0 asused to analyze the data.Matrigel plug. Human VPr-GFP embryoid bodies were differentiated for 14 d byour differentiation protocol followed by expansion in the presence of SB431542for 10 d and injected subcutaneously into NOD/SCID mice in a suspension ofMatrigel. After 2 weeks, Griffonia simplificolia IB4 lectin and/or Ulex europusagglutinin lectin were administered intra-vitally to Matrigel plug–bearing miceand plugs were harvested, fixed overnight in 4% paraformaldehyde and equilibratedin 30% sucrose before freezing and cryosectioning.Immunofluorescence. Cryosections were immunocytochemically stained asprevious described 24 . Briefly, samples were permeabilized in PBST and blockedin 5% donkey serum. Samples were incubated for 2 h in primary antibodiesblocking solution, washed 3 times in PBS and incubated in CY3-conjugated secondaryantibodies (Jackson Laboratories) for 1 h. After washing, some sectionswere counterstained for nucleic acids by TO-PRO3 (Invitrogen) before mountingand imaging by confocal microscopy. Primary antibodies included CD31(DAKO), CD34 (DAKO), Phospho-HistoneH3, Smooth Muscle Actin (DAKO)and VE-cadherin (R&D). All imaging was performed using a Zeiss 510 METAconfocal microscope.Live imaging and 3D rendering. Human VPr-GFP embryoid bodies were culturedin a TOKAI-HIT live cell-imaging chamber on a Zeiss 510 META confocal microscope.Laser intensity and interval were optimized to ensure viability of cells forthe duration of the experiments. Three-dimensional reconstruction and renderingof optical z-stacks was performed using Improvision Volocity software.24. James, D., Noggle, S.A., Swigut, T. & Brivanlou, A.H. Contribution of human embryonicstem cells to mouse blastocysts. Dev. Biol. 295, 90–102 (2006).25. Naldini, L. et al. In vivo gene delivery and stable transduction of nondividing cells bya lentiviral vector. Science 272, 263–267 (1996nature biotechnology volume 28 number 2 february 2010 1

l e t t e r sReal-time imaging of hepatitis C virus infection using afluorescent cell-based reporter systemChristopher T Jones 1 , Maria Teresa Catanese 1 , Lok Man J Law 1 , Salman R Khetani 2,5 , Andrew J Syder 1,5 ,Alexander Ploss 1 , Thomas S Oh 1 , John W Schoggins 1 , Margaret R MacDonald 1 , Sangeeta N Bhatia 2–4 &Charles M Rice 1© 2010 Nature America, Inc. All rights reserved.Hepatitis C virus (HCV), which infects 2–3% of the worldpopulation, is a causative agent of chronic hepatitis and theleading indication for liver transplantation 1 . The ability topropagate HCV in cell culture (HCVcc) is a relatively recentbreakthrough and a key tool in the quest for specific antiviraltherapeutics. Monitoring HCV infection in culture generallyinvolves bulk population assays, use of genetically modifiedviruses and/or terminal processing of potentially precioussamples. Here we develop a cell-based fluorescent reportersystem that allows sensitive distinction of individual HCVinfectedcells in live or fixed samples. We demonstrate use ofthis technology for several previously intractable applications,including live-cell imaging of viral propagation and hostresponse, as well as visualizing infection of primary hepatocytecultures. Integration of this reporter with modern image-basedanalysis methods could open new doors for HCV research.For over two decades, advances in HCV assay systems have been hardwon.Methodologies have ranged from adapted selectable genomesand detection methods that require fixation or cell lysis, such asimmunostaining and quantitative RT PCR, to the use of infectiousreporter viruses 2,3 . Broadening the scope of HCV research, however,will require versatile new assays that allow sensitive single-cell analysisof infection events using unmodified viral genomes.To construct a cellular marker of HCV infection, we adapted aknown substrate of the HCV NS3-4A protease 4–6 , the mitochondriallytethered interferon (IFN)-β promoter stimulator protein 1 (IPS-1;ref. 7), also termed MAVS 8 , VISA 9 or Cardif 4 . The C-terminal region ofIPS-1, encompassing the NS3-4A recognition site and a mitochondrialtargeting sequence, was fused to green fluorescent protein (EGFP-IPS,Fig. 1a) or to the red fluorescent proteins (RFPs) mCherry or TagRFP.We also introduced an SV40 nuclear localization sequence (NLS)between the RFP variant and IPS-1 segment (RFP-NLS-IPS, Fig. 1a).Human hepatoma (Huh-7.5) cells stably transduced with lentivirusesencoding EGFP-IPS or RFP-NLS-IPS exhibited punctate fluorescenceconsistent with mitochondrial localization of the reporter, which wasconfirmed by colocalization with native IPS-1 (Fig. 1b).We determined the reporter phenotype of the EGFP-IPS or RFP-NLS-IPS constructs in the presence of NS3-4A by transduction intoHuh-7.5 cells stably expressing an autonomously replicating HCVsubgenome 10 (JFH-1 strain, SG-JFH). Replicon-harboring cellsexpressing EGFP-IPS showed diffuse fluorescence, whereas an NS3-4Acleavage–resistant form 11 of the reporter (EGFP-IPS(C508Y); Fig. 1c)exhibited a punctate pattern. Similarly, replicon-containing Huh-7.5 cells expressing RFP-NLS-IPS, but not RFP-NLS-IPS(C508Y),showed nuclear translocation of fluorescence (Fig. 1c). Both reportersdisplayed a punctate pattern in the absence of the HCV replicon(Fig. 1c). These results indicate that cleavage of EGFP-IPS and RFP-NLS-IPS are dependent on an intact NS3-4A recognition site and thatHCV-dependent fluorescence relocalization (HDFR) can be used asa marker of viral replication.HCV exists as multiple genotypes, which exhibit extensive sequencedivergence as well as differences in pathogenesis and treatment susceptibility12 . Evasion of the innate immune response by cleavage ofnative IPS-1, however, is likely to be a conserved feature of HCV infection.In addition to JFH-1 (genotype 2a), Huh-7.5 cells harboring H77(genotype 1a) or Con1 (genotype 1b) subgenomes 10 were transducedwith EGFP-IPS or EGFP-IPS(C508Y). Regardless of the HCV strain,EGFP-IPS transduction resulted in diffuse fluorescence, and EGFP-IPS(C508Y) expression led to punctate EGFP (Fig. 1c). Whereas thelack of replicon systems for other genotypes precludes comprehensiveanalysis, these results indicate that cleavage of EGFP-IPS can be usedas a marker of several diverse HCV strains. In contrast, replicationof other positive-strand RNA viruses, such as yellow fever virus orVenezuelan equine encephalitis virus, did not lead to fluorescencerelocalization (Supplementary Fig. 1a). These results suggest thatthe HDFR reporter system achieves a high level of HCV specificitycombined with genotype independence.Although replicon-containing cells constitutively express theviral proteins, monitoring authentic virus infection is important foranalyses of HCV biology and therapeutic inhibition. To determinethe ability of HDFR to detect infection, we inoculated Huh-7.5 cellsexpressing RFP-NLS-IPS with an HCVcc reporter virus expressingsecreted Gaussia luciferase, Jc1FLAG2(p7-nsGluc2A) 13 , followed by1 Center for the Study of Hepatitis C, Laboratory of Virology and Infectious Disease, The Rockefeller University, New York, New York, USA. 2 Division of Health Sciencesand Technology, Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 3 HowardHughes Medical Institute, 4 Division of Medicine, Brigham & Women’s Hospital, Boston, Massachusetts, USA. 5 Present addresses: Hepregen Corporation, Medford,Massachusetts, USA (S.R.K.) and iTherX Pharmaceuticals, San Diego, California, USA (A.J.S.). Correspondence should be addressed to C.M.R. (ricec@rockefeller.edu).Received 17 August 2009; accepted 4 January 2010; published online 31 January 2010; doi:10.1038/nbt.1604nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 167

l e t t e r sabIPS-1 Reporter Merge WT C508YcIPS-11CARDPRO508462 TM540Huh-7.5Huh-7.5EGFP-IPSRFP-NLS-IPSNLSEGFP-IPSSG-JFH© 2010 Nature America, Inc. All rights reserved.RFP-NLS-IPSEGFP-IPSSG-H77dLog 10 RLU87654PBSIFN-βIgGα-CD81RFPHoechstUninfected+HCVccPBS IFN-β IgG α-CD81Figure 1 An IPS-1-based reporter system for detection of HCV infection. (a) Schematic of IPS-1 and derivativereporter constructs. The caspase recruitment domain (CARD) and proline-rich (PRO) domains of IPS-1 areindicated. The HCV NS3-4A protease cleaves IPS-1 at C508 (arrow). The C-terminal transmembrane domain (TM)directs IPS-1 to the outer membrane of mitochondria. EGFP-IPS encodes EGFP fused to residues 462–540 of IPS-1. RFP-NLS-IPS encodes ared fluorescent protein (mCherry or TagRFP) and an SV40 nuclear localization signal (NLS, PKKKRKVG) fused to residues 462–540 of IPS-1.(b) EGFP-IPS and RFP-NLS-IPS localize to mitochondria in Huh-7.5 cells. Native IPS-1, detected by immunofluorescent staining (IPS-1), and EGFPor RFP autofluorescence (Reporter) were visualized in untransduced (Huh-7.5) or transduced (EGFP-IPS or RFP-NLS-IPS) cells by confocal microscopy.Merge images also depict Hoechst nuclear dye (blue). (c) EGFP-IPS and RFP-NLS-IPS relocalize in response to HCV replication. Huh-7.5 cell linesharboring subgenomic (SG) neomycin-selectable replicons were transduced with lentiviruses expressing wild type (WT) or mutant (C508Y) EGFP-IPSor RFP-NLS-IPS. H77, genotype 1a; Con1, genotype 1b; JFH-1, genotype 2a. Wide-field fluorescence images of unfixed cells are shown. (d) RFP-NLS-IPS relocalizes in HCV-infected cells. Huh-7.5 cells expressing RFP-NLS-IPS were infected with secreted Gaussia luciferase HCVcc reportervirus, Jc1FLAG2(p7-nsGluc2A), in the presence of PBS, IFN-β, blocking antibody (α-CD81) or isotype control (IgG). Luciferase activity in the culturesupernatants (left) and reporter (RFP) or nuclear dye (Hoechst) fluorescence (right) were monitored at 48 h post-infection. Wide-field fluorescenceimages of fixed cells are shown. Scale bars, 20 µm. RLU, relative light units.RFP-NLS-IPSSG-Con1Huh-7.5SG-JFHincubation for 48 h. Uninfected cells showed punctate fluorescence,whereas HCV-infected cultures displayed a distinct nuclear signal(Fig. 1d). Inoculation in the presence of IFN-β largely abolished thefluorescence translocation phenotype. Similarly, cells infected inconjunction with a monoclonal antibody targeting a known HCVentry factor (α-CD81) did not show nuclear fluorescence. Detectionof Gaussia luciferase in the culture supernatants yielded correspondingresults (Fig. 1d). Staining for viral replicase protein NS5Ain infected EGFP-IPS–expressing cells supported the correlationbetween fluorescence relocalization and HCV replication at thesingle-cell level (Supplementary Fig. 1b).Monitoring infection by fluorescence relocalization does notrequire cells to be fixed, lysed or processed. These advantages suggestedthe possibility of real-time visualization of HCV infectionin live cells. Huh-7.5 cells expressing RFP-NLS-IPS and a constitutivemitochondrial marker (EGFP-cytochrome c oxidase subunitVIII fusion protein; mito-EGFP) were inoculated with Jc1FLAG2(p7-nsGluc2A) and monitored by live-cell microscopy beginning at6 h post-infection (Fig. 2a,b; DMSO). Translocation of RFP-NLS tothe nucleus could be detected as early as 10–12 h post-inoculation,with complete cleavage by 16–18 h (Fig. 1b and SupplementaryVideo 1a). In contrast, cells infected in the presence of a viral RNAdependentRNA polymerase inhibitor (2′CMA) 14 showed verylimited nuclear fluorescence (Fig. 2b and Supplementary Video 1b).We then investigated whether drug treatment of cells with establishedHCV infection could lead to observable reconstitution ofmitochondrially localized fluorescence. RFP-NLS-IPS reporter cellswere infected with Jc1FLAG2(p7-nsGluc2A) for 24 h before treatmentwith VX-950, an inhibitor of the NS3-4A 15 protease, or DMSOas a vehicle control and imaged for an additional 24 h (Fig. 2a,c).Over the time course of the experiment, RFP-NLS-IPS localizationin DMSO cells remained unchanged, whereas steady reconstitutionof punctate fluorescence was seen in the majority of infected cellstreated with the protease inhibitor. These results indicate that thereporter system can be used to visually monitor NS3-4A inhibitionin real time (Supplementary Video 1c,d).168 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

l e t t e r sFigure 2 Time-lapse live-cell imaging of HCVccinfection. (a) Schematic of live-cell imagingtime course. Huh-7.5 cells stably expressingRFP-NLS-IPS and a mitochondrially targetedEGFP-cytochrome c oxidase subunit VIII fusionprotein (mito-EGFP) were infected with HCVccreporter virus, Jc1FLAG2(p7-nsGluc2A)(time = 0 h). (b) Cells were infected in thepresence of DMSO or HCV RNA-dependentRNA polymerase inhibitor 2′CMA. (c) Cells wereinfected for 24 h before removal of the inoculumand addition of imaging medium containingDMSO or the NS3-4A protease inhibitor VX-950.Images were captured every 30 min startingat 6 h (b) or 24.5 h (c) post-infection. RFPfluorescence is shown in grayscale. Time (h) fromthe start of infection (b) or drug addition (c) areindicated. Scale bar, 20 µm. See SupplementaryVideo 1a–d for the full time course.ab6cTime (h)DMSO2′CMAbc0249 12 15 18 21 24 272 4 8 12 16 20 24 2848© 2010 Nature America, Inc. All rights reserved.DMSOThe availability of spectrally distinct HDFRreporters (EGFP-IPS and RFP-NLS-IPS) suggestedthe possibility of discerning infectionVX950in two separate cell populations simultaneously.We applied this advantage to visualizethe recently described phenomenon of CD81-independent HCV infection.Circulating HCV enters hepatocytes through a complex pathwayinvolving multiple co-receptors. CD81, SCARB1 and two tight-junctionproteins, CLDN1 and OCLN, have been shown to be essential forthis process 16–19 . Recent reports, however, suggest a second, CD81-independent route of virus entry, which may entail particle transferthrough close cell-cell contacts 20,21 . This transmission mode may behighly biologically relevant in the context of chronic infection, and thedevelopment of inhibitors targeting this entry pathway necessitatesa reliable method of detection. To monitor routes of HCV spread,we employed cells expressing RFP-NLS-IPS and EGFP-IPS as producerand target populations, respectively. EGFP-IPS target cells wereengineered to stably express a short hairpin (sh)RNA targeting CD81(EGFP-IPS/CD81 − ) or an irrelevant sequence (EGFP-IPS/IRR), andtested for permissiveness to cell-free virus using an adapted HCVcc(J6/JFH clone 2), which exhibits superior titers to J6/JFH 22 . Cellsexpressing CD81 shRNA had undetectable levels of CD81 protein(Supplementary Fig. 2a). At 48 h post-infection, the majority ofEGFP-IPS/IRR cells exhibited diffuse EGFP, whereas EGFP-IPS/CD81 −cells were largely nonpermissive (Fig. 3a). Fluorescence-activated cellsorting (FACS) analysis of fixed samples stained with an NS5A antibodysupported these observations, indicating that

l e t t e r sRFP-NLS-IPSEGFP-IPSIRR CD81 –RFP-NLS-IPS+IRRRFP-NLS-IPS+CD81 –cWT+DMSOWT+2′CMAC508Y+DMSOab+HCVcc–HCVccCo-culture© 2010 Nature America, Inc. All rights reserved.MergeEGFP-G3BPRFP-NLS-IPS32 32.5 33.5 34.5 36.5 37 39.5 48Figure 3 Use of the IPS-1-based reporter to expand HCVcc culture systems. (a) Co-culture of spectrally distinct HCV reporter cell lines for visualizingCD81-independent infection. Wide-field fluorescence images of unfixed mono- and co-cultures of Huh-7.5 cell lines expressing RFP-NLS-IPS orEGFP-IPS in the presence (+HCVcc) or absence (−HCVcc) of J6/JFH clone 2. EGFP-IPS cells stably express shRNA targeting CD81 (CD81 − ) or anirrelevant sequence (IRR). Monocultures were infected for 72 h before imaging. In co-culture experiments, RFP-NLS-IPS cells were infected withHCVcc for 36 h before mixing with uninfected EGFP-IPS/IRR or EGFP-IPS/CD81 − cells. Co-cultures were incubated for an additional 48 h beforeimaging. (b) Multiplexing the HCV reporter with a marker of the stress response. An Huh-7 cell line expressing RFP-NLS-IPS and an EGFP-taggedstress granule marker (EGFP-G3BP) was infected with Jc1FLAG2(p7-nsGluc2A). Live-cell imaging was initiated at 6 h post-infection with imagescaptured every 30 min. Montage shows selected time points beginning at 32 h post-infection; times (h) from the start of infection are indicated.See Supplementary Videos 2a–c. (c) Visualization of HCVcc infection in primary hepatocytes. Primary human hepatocytes maintained as MPCCswere transduced with lentiviruses expressing wild type (WT) or mutant (C508Y) RFP-NLS-IPS. At 24 h post-transduction, MPCCs were infectedwith Jc1FLAG2(p7-nsGluc2A). After 12 h, virus was removed and MPCC medium containing DMSO or 2′CMA was added. Unfixed MPCCs wereimaged by wide-field fluorescence microscopy at 48 h post-infection. Representative phase contrast (top row) and corresponding RFP fluorescenceimages (middle row) are shown. Enlarged fluorescence images (bottom row) correspond to area denoted by white dotted box (middle row). The numberof cells per MPCC island exhibiting nuclear RFP at 48 h post-infection is plotted for each condition. For WT RFP-NLS-IPS+DMSO, n = 40;WT RFP-NLS-IPS+2′CMA, n = 30; C508Y RFP-NLS-IPS+DMSO, n = 35. Bar, mean number of positive cells/island. Scale bars, 20 µm (a and b),200 µm (c, top row), 10 µm (c, lower row).Infected cells/island (number)1614121086420WT+DMSOWT+2′CMAC508Y+DMSOdifficulties, and explored its use for detecting HCV infection inthe recently developed micropattern co-culture (MPCC) system 26 .MPCCs consist of primary adult human hepatocytes seeded onislands of collagen and surrounded by mouse fibroblast ‘feeder’ cells.These conditions allow hepatocytes to be maintained for extendedperiods without the rapid decline in cellular functions seen in conventionalmonocultures or random co-cultures 27 . To visualize HCVinfection in MPCC hepatocytes, we transduced cultures with RFP-NLS-IPS or RFP-NLS-IPS(C508Y) and infected them 48 h later withJc1FLAG2(p7-nsGluc2A), allowing MPCC infection to be monitoredin parallel by Gaussia luciferase secretion 28 (SupplementaryFig. 3). MPCC islands were examined by live cell microscopy and thecells per island exhibiting nuclear RFP were enumerated (Fig. 3c).In cultures transduced with RFP-NLS-IPS followed by DMSO treatment,~98% of the islands observed contained cells with nuclearRFP, averaging four cells/island. In the RFP-NLS-IPS populationtreated with 2′CMA, only 6% of islands exhibited infected cells,corresponding to an average of 0.06 cells/island. Cells transducedwith RFP-NLS-IPS(C508Y) did not show nuclear RFP in any of theislands examined (Fig. 3c). These results indicate that infection ofprimary hepatocytes can be readily detected using the HDFR system.To our knowledge, visualization of HCV infection in live primaryhepatocytes has not been demonstrated previously.Although systems for tracking HCV replication in culture haveexpanded rapidly in recent years, robust detection methods applicableto imaging of individual live cells have not been available. Wedescribe a sensitive HCV reporter that allows easy distinction ofinfected and uninfected cells in live or fixed cultures by standardfluorescence microscopy. The robust signal of the reporter systemderives from the efficiency of NS3-4A cleavage and the constitutivehigh-level expression of the substrate; nuclear translocation increasesvisualization, as the reporter becomes concentrated in a region withlow autofluorescence, a particular advantage when working withhepatocytes. Although reporter cleavage does not occur in the absenceof an active protease, transient signal may be expected in the presenceof a polymerase inhibitor—this high sensitivity may have to befactored in as ‘background’ for some applications. The HDFRsystem does not require genetic modification of the viral genome andshowed efficient detection of all HCV genotypes tested. This raisesthe possibility of using the reporter to identify novel infectious isolatesdirectly from patient samples, potentially expanding the HCVccsystem beyond the currently available genotype 2a strain. Coaxing170 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

l e t t e r s© 2010 Nature America, Inc. All rights reserved.HCVcc to infect biologically relevant primary cell types may also be keyto understanding authentic viral processes and patient-specific responses.The low level of replication observed in these cultures may reflectheterogeneity between individual cells or viral genomes, and underscoresthe value of single-cell analysis in dissecting the often subtle or variablephenotypes associated with chronic infection. Combining HDFR-basedvisualization with laser capture microscopy and analysis of neighboringinfected and uninfected cells could begin to unravel the determinants ofpathogenesis or virus control. The value of single-cell analysis was furtherillustrated by multiplexing HCV detection with a fluorescent marker ofcellular stress, allowing direct visual correlation of viral and host events.Recent advances in automated microscopy and ‘high-content’ screeninghave made a large number of cellular phenotypes, including drugtoxicity profiles, accessible to interrogation in a multiparametric format(reviewed in ref. 29). Addition of a robust fluorescent translocation assayrequiring minimal sample processing has the potential to integrate HCVresearch into this burgeoning field. We anticipate that the ability of HDFRto increase the flexibility and diversity of HCV culture systems will beimportant for basic virology and antiviral drug development.MethodsMethods and any associated references are available in the online versionof the paper at http://www.nature.com/naturebiotechnology/.Note: Supplementary information is available on the Nature Biotechnology website.AcknowledgmentsWe acknowledge the expert support of The Rockefeller University BioimagingCore Facility, with special thanks to A. North, S. Galdeen and S. Bhuvanendran.We thank The Rockefeller University Flow Cytometry Resource Center, supportedby the Empire State Stem Cell Fund through NY State Department of Health(NYSDOH) contract no. C023046; opinions expressed here are solely those of theauthors and do not necessarily reflect those of the Empire State Stem Cell Fund,the NYSDOH, or the State of NY. We are grateful to C. Stoyanov (The RockefellerUniversity) for YF17D(5′C25Venus2AUbi), J. Tazi for G3BP (Institut de GénétiqueMoléculaire de Montpellier) and I. Frolov (UTMB) for Venezuelan equineencephalitis virus-EGFP. We thank M. Holz, A. Forest, M. Panis and A. Websonfor laboratory support and C. Murray for critical reading of the manuscript. Thiswork was supported by Public Health Service grants R01 AI075099 (C.M.R.)and R01 DK56966 (S.N.B.). This work was also funded by the Office of theDirector/National Institutes of Health (NIH) through the NIH Roadmap forMedical Research, Grant 1 R01 DK085713-01 (C.M.R. and S.N.B.). Informationon this Roadmap Transformative R01 Program can be found at http://grants.nih.gov/grants/guide/rfa-files/RFA-RM-08-029.html. Additional funding was providedby the Greenberg Medical Research Institute and the Starr Foundation (C.M.R.).S.N.B. is an Howard Hughes Medical Investigator investigator. C.T.J. was supportedby National Research Service Award DK081193; M.T.C. was supported by a Women& Science Fellowship; L.M.J.L. is supported by a Natural Sciences and EngineeringResearch Council of Canada fellowship. A.P. is a recipient of a Kimberly Lawrence-Netter cancer research discovery fund award.AUTHOR CONTRIBUTIONSC.T.J. and C.M.R. designed the project, analyzed results and wrote the manuscript.C.T.J., M.T.C., L.M.J.L., A.J.S., S.R.K., T.S.O., A.P. and J.W.S. performed theexperimental work. S.R.K., J.W.S., T.S.O., M.R.M. and S.N.B. contributed reagentsand technical expertise.COMPETING INTERESTS STATEMENTThe authors declare competing financial interests: details accompany the full-textHTML version of the paper at http://www.nature.com/naturebiotechnology/.Published online at http://www.nature.com/naturebiotechnology/.Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/.1. Shepard, C.W., Finelli, L. & Alter, M.J. Global epidemiology of hepatitis C virusinfection. Lancet Infect. Dis. 5, 558–567 (2005).2. Bartenschlager, R. & Sparacio, S. Hepatitis C virus molecular clones and theirreplication capacity in vivo and in cell culture. Virus Res. 127, 195–207(2007).3. Bartenschlager, R. Hepatitis C virus molecular clones: from cDNA to infectious virusparticles in cell culture. Curr. Opin. Microbiol. 9, 416–422 (2006).4. Meylan, E. et al. Cardif is an adaptor protein in the RIG-I antiviral pathway and istargeted by hepatitis C virus. Nature 437, 1167–1172 (2005).5. Foy, E. et al. Control of antiviral defenses through hepatitis C virus disruption ofretinoic acid-inducible gene-I signaling. Proc. Natl. Acad. Sci. USA 102,2986–2991 (2005).6. Li, X.D., Sun, L., Seth, R.B., Pineda, G. & Chen, Z.J. Hepatitis C virus proteaseNS3/4A cleaves mitochondrial antiviral signaling protein off the mitochondria toevade innate immunity. Proc. Natl. Acad. Sci. USA 102, 17717–17722(2005).7. Kawai, T. et al. IPS-1, an adaptor triggering RIG-I- and Mda5-mediated type Iinterferon induction. Nat. Immunol. 6, 981–988 (2005).8. Seth, R.B., Sun, L., Ea, C.K. & Chen, Z.J. Identification and characterization ofMAVS, a mitochondrial antiviral signaling protein that activates NF-kappaB and IRF3. Cell 122, 669–682 (2005).9. Xu, L.G. et al. VISA is an adapter protein required for virus-triggered IFN-betasignaling. Mol. Cell 19, 727–740 (2005).10. Tscherne, D.M. et al. Superinfection exclusion in cells infected with hepatitis Cvirus. J. Virol. 81, 3693–3703 (2007).11. Loo, Y.M. et al. Viral and therapeutic control of IFN-beta promoter stimulator 1during hepatitis C virus infection. Proc. Natl. Acad. Sci. USA 103, 6001–6006(2006).12. Simmonds, P. Genetic diversity and evolution of hepatitis C virus-15 years on.J. Gen. Virol. 85, 3173–3188 (2004).13. Marukian, S. et al. Cell culture-produced hepatitis C virus does not infect peripheralblood mononuclear cells. Hepatology 48, 1843–1850 (2008).14. Carroll, S.S. et al. Inhibition of hepatitis C virus RNA replication by 2′-modifiednucleoside analogs. J. Biol. Chem. 278, 11979–11984 (2003).15. Lin, K., Perni, R.B., Kwong, A.D. & Lin, C. VX-950, a novel hepatitis C virus (HCV)NS3–4A protease inhibitor, exhibits potent antiviral activities in HCV replicon cells.Antimicrob. Agents Chemother. 50, 1813–1822 (2006).16. Pileri, P. et al. Binding of hepatitis C virus to CD81. Science 282, 938–941(1998).17. Scarselli, E. et al. The human scavenger receptor class B type I is a novel candidatereceptor for the hepatitis C virus. EMBO J. 21, 5017–5025 (2002).18. Evans, M.J. et al. Claudin-1 is a hepatitis C virus co-receptor required for a latestep in entry. Nature 446, 801–805 (2007).19. Ploss, A. et al. Human occludin is a hepatitis C virus entry factor required forinfection of mouse cells. Nature 457, 882–886 (2009).20. Timpe, J.M. et al. Hepatitis C virus cell-cell transmission in hepatoma cells in thepresence of neutralizing antibodies. Hepatology 47, 17–24 (2008).21. Witteveldt, J. et al. CD81 is dispensable for hepatitis C virus cell-to-cell transmissionin hepatoma cells. J. Gen. Virol. 90, 48–58 (2009).22. Diamond, D.L. et al. Temporal proteome and lipidome profiles reveal HCV-associatedreprogramming of hepatocellular metabolism and bioenergetics. PLoS Pathog. 6,e1000719 (2010).23. Anderson, P. & Kedersha, N. RNA granules. J. Cell Biol. 172, 803–808(2006).24. Schutz, S. & Sarnow, P. How viruses avoid stress. Cell Host Microbe 2, 284–285(2007).25. Tourriere, H. et al. The RasGAP-associated endoribonuclease G3BP assembles stressgranules. J. Cell Biol. 160, 823–831 (2003).26. Khetani, S.R. & Bhatia, S.N. Microscale culture of human liver cells for drugdevelopment. Nat. Biotechnol. 26, 120–126 (2008).27. Bhatia, S.N., Balis, U.J., Yarmush, M.L. & Toner, M. Effect of cell-cell interactionsin preservation of cellular phenotype: cocultivation of hepatocytes and nonparenchymalcells. FASEB J. 13, 1883–1900 (1999).28. Ploss, A. et al. Persistent hepatitis C virus infection in microscale primary humanhepatocyte cultures. Proc. Natl. Acad. Sci. USA (in the press).29. Wollman, R. & Stuurman, N. High throughput microscopy: from raw images todiscoveries. J. Cell Sci. 120, 3715–3722 (2007).nature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 171

© 2010 Nature America, Inc. All rights reserved.ONLINE METHODSCell culture. Huh-7 and Huh-7.5 cells were cultured at 37 °C, 5% CO 2 inDulbecco’s Modified Eagle Medium (DMEM, Invitrogen) containing 10% FBSand 0.1 mM nonessential amino acids (NEAA) (complete media), unless otherwisenoted. For time-lapse imaging, cells were maintained in CO 2 -independentmedia (Invitrogen) containing 10% FBS, 0.1 mM NEAA, 1 mM sodium pyruvateand 2 mM l-glutamine (imaging media). Huh-7.5 cell lines harboringselectable subgenomic replicons 10 were grown in complete media containing0.5 mg/ml G418. Huh-7.5 cells stably expressing the pLenti-3′-U6-EC-EP7vector encoding an shRNA against CD81 (nt 268-288, cDNA numbering) ora predicted nontargeting sequence (IRR) have been previously described 21and were grown in complete media containing 6 µg/ml blasticidin. MPCCcultures were generated as previously described 26 and maintained in highglucose DMEM, 10% FBS, 0.5 U/ml insulin, 7 ng/ml glucagon, 7.5 µg/mlhydrocortisone and 1% penicillin-streptomycin. For HCV inhibition, culturemedia was supplemented with 0.2% DMSO, 14 mM 2′CMA, 24 mM VX-950, or1,000 U/ml IFN-β (Peprotech). For neutralization experiments, HCVcc infectionswere performed in the presence of 10 µg/ml antibody directed againstCD81 (BD Biosciences) or an isotype control (IgG1κ, BD Biosciences).Virus stocks. Jc1 30 and Jc1FLAG2(p7-nsGluc2A) 13 are fully infectious HCVccviruses that have been previously described. J6/JFH clone 2 is a passaged derivativeof J6/JFH 31 that contains a number of adaptive mutations that increaseinfectious titers 22 . Bi-Ypet-Jc1FLAG2 is a bicistronic reporter virus in whichthe HCV IRES drives expression of Ypet, an EGFP variant with enhancedbrightness, in the first cistron; the EMCV IRES drives expression of thesecond cistron, which encodes the Jc1 polyprotein with a FLAG epitope at theN terminus of E2. YF17D(5′C25Venus2AUbi) is a monocistronic yellow feverreporter virus (kindly provided by C. Stoyanov, The Rockefeller University)encoding the Venus fluorescent protein, a yellow-shifted variant of EGFP.Venezuelan equine encephalitis virus-EGFP (kindly provided by I. Frolov,UTMB) is a double subgenomic EGFP reporter virus derived from the TC83vaccine strain of Venezuelan equine encephalitis. Virus stocks were generatedby electroporation of in vitro transcribed RNAs into the appropriate cell lines,as described previously 31–33 .Plasmid constructs. Constructs were created by standard methods; plasmidand primer sequences are available upon request. IPS-1–based reporters andsubcellular localization markers were constructed in a lentivirus backbonederived from TRIP-EGFP 34 . Residues 462–540 of IPS-1 (IPS) were obtainedby PCR from a human hepatocyte cDNA library (Ambion) and inserted intothe BsrGI/XhoI sites of TRIP-EGFP to generate TRIP-EGFP-IPS. IPS-1 mutationC508Y was generated by overlap PCR mutagenesis. TRIP-mCherry andTRIP-TagRFP were used to construct TRIP-RFP-NLS-IPS plasmids. TRIPmCherrywas derived from the TRIP-mCherry-CLDN1 plasmid 18 . TagRFPsequence was obtained from pTagRFP-C (Evrogen). TRIP-RFP-NLS-IPSplasmids encode the SV40 nuclear localization signal (NLS, PKKKRKVG) andIPS fused to the C terminus of RFP. TRIP-mito-EGFP, encodes the mitochondrialtargeting sequence of human cytochrome c oxidase subunit VIII fused to theN terminus of EGFP. TRIP-EGFP-G3BP encodes the Ras-Gap-SH3 domainbinding protein (G3BP, kindly provided by J. Tazi 25 ) fused to the C terminusof EGFP.Generation of lentivirus pseudoparticles and transductions. Pseudoparticles(pp) were generated by co-transfection of 293T cells with TRIP provirus, HIVgag-pol, and vesicular stomatitis virus envelope protein G (VSV-G) plasmidsusing a weight ratio of 1:0.8:0.2, as described previously 18 . Huh-7 and Huh-7.5 cells were transduced by incubation for 6 h at 37 °C with TRIPpp diluted1:3 in complete media supplemented with 4 µg/ml polybrene and 20 mMHEPES. In some cases, transduced hepatoma cell populations were enrichedusing a FACSAria II high-speed flow cytometry cell sorter (BD Biosciences).For transduction of MPCC, cultures were first treated for ~20 s with 0.025%EDTA/Trypsin, before washing and overnight incubation with 1:3 dilutedTRIPpp stocks.Immunofluorescence staining and FACS analysis. For NS5A immunostaining,cells grown on glass coverslips were washed with PBS before fixation in formaldehyde(3.7% wt/vol in PBS) and incubation in blocking buffer (3% BSA,0.2% saponin in PBS). After overnight incubation at 4 °C with monoclonalantibody 9E10 (ref. 31) (1:2,000 in blocking buffer) and 1 h incubation at 25 °Cwith AlexFluor-594-conjugated secondary antibody to mouse (Invitrogen,1:1,000 in blocking buffer), cell nuclei were stained with Hoechst dye (ThermoScientific). Coverslips were mounted using ProLong Gold Antifade reagent(Invitrogen). For IPS-1 immunostaining, cells grown on glass bottom multiwallplates (MatriCal) were washed with PBS and fixed with 2% paraformaldehyde.After incubation in blocking buffer for 30 min, and polyclonal anti-IPS-1 antibody(Cell Signaling Technology, 1:50 in blocking buffer) for 2 h, AlexaFluor-555or AlexaFluor-488 conjugated anti-rabbit IgG secondary antibodies (CellSignaling Technology, 1:500 in blocking buffer) were added for 1 h at 25 °C.Cells were then stained with Hoechst nuclear dye followed by applicationof ProLong Gold Antifade reagent. For FACS analysis, cells were harvestedusing AccuMax (eBioscience) and fixed using Fixation/Permeabilization buffer(BD Biosciences) for 10 min at 4 °C. Fixed cells were washed with BD Perm/Wash buffer (BD Biosciences), incubated 30 min at 25 °C with AlexaFluor-647-conjugated 9E10 antibody (1:4000 in BD Perm/Wash buffer), washed twicewith BD Perm/Wash buffer and once with FACS buffer (PBS/3% FBS) beforeanalysis using a BD LSR II flow cytometer and BD FACSDiva software. Analysisof FACS data was performed using FlowJo software.Microscopy. Wide-field fluorescent images were captured using an EclipseTE300 (Nikon) inverted microscope and SPOT imaging software or theDiscovery-1 system and MetaXpress software (Molecular Devices). Confocalimaging of fixed samples was performed using an inverted Axiovert 200 laserscanning microscope (Zeiss). For long-term live cell imaging, cells were grownon rat-tail collagen–coated (BD Biosciences) no. 1.5 Lab-Tek II 4-chambered(Thermo Fisher Scientific) coverslips. Live cells maintained at 37 °C in imagingmedia were imaged using a Zeiss Axiovert 200 inverted microscope equippedwith an UltraView spinning disk confocal head (Perkin-Elmer), an Orca ERcooledCCD camera (Hamamatsu), a 20×/0.75 N.A. Plan-Apochromat objective,and an environmental chamber (Solent Scientific). Solid-state 491 and561 nm lasers (Spectral Applied) and ET 530/50 and ET 605/70 emissionfilters (Chroma) were used for excitation and emission of EGFP and RFPfluorescence, respectively. Alternatively, time-lapse images were captured usingan Olympus IX71 inverted microscope equipped with an Orca ER cooledCCD camera, a 20×/0.75 N.A. UPlan SApo objective and an environmentalchamber. Image acquisition was performed using Metamorph (MolecularDevices) and processing was performed using ImageJ64.30. Pietschmann, T. et al. Construction and characterization of infectious intragenotypicand intergenotypic hepatitis C virus chimeras. Proc. Natl. Acad. Sci. USA 103,7408–7413 (2006).31. Lindenbach, B.D. et al. Complete replication of hepatitis C virus in cell culture.Science 309, 623–626 (2005).32. Lindenbach, B.D. & Rice, C.M. Trans-complementation of yellow fever virus NS1reveals a role in early RNA replication. J. Virol. 71, 9608–9617 (1997).33. Petrakova, O. et al. Noncytopathic replication of Venezuelan equine encephalitisvirus and eastern equine encephalitis virus replicons in mammalian cells. J. Virol.79, 7597–7608 (2005).34. Zennou, V. et al. HIV-1 genome nuclear import is mediated by a central DNA flap.Cell 101, 173–185 (2000).nature biotechnologydoi:10.1038/nbt.1604

l e t t e r sRational design of cationic lipids for siRNA deliverySean C Semple 1,6 , Akin Akinc 2,6 , Jianxin Chen 1,5 , Ammen P Sandhu 1 , Barbara L Mui 1,5 , Connie K Cho 1 ,Dinah W Y Sah 2 , Derrick Stebbing 1 , Erin J Crosley 1 , Ed Yaworski 1 , Ismail M Hafez 3 , J Robert Dorkin 2 , June Qin 2 ,Kieu Lam 1 , Kallanthottathil G Rajeev 2 , Kim F Wong 3 , Lloyd B Jeffs 1 , Lubomir Nechev 2 , Merete L Eisenhardt 1 ,Muthusamy Jayaraman 2 , Mikameh Kazem 3 , Martin A Maier 2 , Masuna Srinivasulu 4 , Michael J Weinstein 2 ,Qingmin Chen 2 , Rene Alvarez 2 , Scott A Barros 2 , Soma De 2 , Sandra K Klimuk 1 , Todd Borland 2 ,Verbena Kosovrasti 2 , William L Cantley 2 , Ying K Tam 1,5 , Muthiah Manoharan 2 , Marco A Ciufolini 4 ,Mark A Tracy 2 , Antonin de Fougerolles 2 , Ian MacLachlan 1 , Pieter R Cullis 3 , Thomas D Madden 1,5 & Michael J Hope 1,5© 2010 Nature America, Inc. All rights reserved.We adopted a rational approach to design cationic lipids foruse in formulations to deliver small interfering RNA (siRNA).Starting with the ionizable cationic lipid 1,2-dilinoleyloxy-3-dimethylaminopropane (DLinDMA), a key lipid component ofstable nucleic acid lipid particles (SNALP) as a benchmark,we used the proposed in vivo mechanism of action of ionizablecationic lipids to guide the design of DLinDMA-based lipidswith superior delivery capacity. The best-performing lipidrecovered after screening (DLin-KC2-DMA) was formulatedand characterized in SNALP and demonstrated to havein vivo activity at siRNA doses as low as 0.01 mg/kg in rodentsand 0.1 mg/kg in nonhuman primates. To our knowledge, thisrepresents a substantial improvement over previous reports ofin vivo endogenous hepatic gene silencing.A key challenge in realizing the full potential of RNA interference(RNAi) therapeutics is the efficient delivery of siRNA, the moleculesthat mediate RNAi. The physicochemical characteristics of siRNA—high molecular weight, anionic charge and hydrophilicity—preventpassive diffusion across the plasma membrane of most cell types.Therefore, delivery mechanisms are required that allow siRNA toenter cells, avoid endolysosomal compartmentalization and localizein the cytoplasm where it can be loaded into the RNA-inducedsilencing complex. To date, formulation in lipid nanoparticles (LNPs)represents one of the most widely used strategies for in vivo deliveryof siRNA 1,2 . LNPs represent a class of particles comprised of differentlipid compositions and ratios as well as different sizes and structuresformed by different methods. A family of LNPs, SNALP 3–6 , is characterizedby very high siRNA-encapsulation efficiency and small,uniformly sized particles, enabled by a controlled step-wise dilutionmethodology. LNPs, including SNALP, have been successfully used tosilence therapeutically relevant genes in nonhuman primates 6–8 andare currently being evaluated in several clinical trials.An empirical, combinatorial chemistry–based approach recentlyidentified novel materials for use in LNP systems 7 . A key feature ofthis approach was the development of a one-step synthetic strategythat allowed the rapid generation of a diverse library of ~1,200 compounds.This library was then screened for novel materials capableof mediating efficient delivery of siRNA in vitro and in vivo. Here, weinstead used a medicinal chemistry (that is, structure-activity relationship)approach, guided by the putative in vivo mechanism of actionof ionizable cationic lipids, for rational lipid design. Specifically,we hypothesized that after endocytosis, the cationic lipid interactswith naturally occurring anionic phospholipids in the endosomalmembrane, forming ion pairs that adopt nonbilayer structures anddisrupt membranes (Fig. 1) 9–12 . We previously advanced the conceptFigure 1 Proposed mechanism of action for membrane disruptive effects ofcationic lipids and structural diagram of DLinDMA divided into headgroup,linker and hydrocarbon chain domains. In isolation, cationic lipids andendosomal membrane anionic lipids such as phosphatidylserine adopt acylindrical molecular shape, which is compatible with packing in a bilayerconfiguration. However, when cationic and anionic lipids are mixed together,they combine to form ion pairs where the cross-sectional area of the combinedheadgroup is less than that of the sum of individual headgroup areas inisolation. The ion pair therefore adopts a molecular ‘cone’ shape, whichpromotes the formation of inverted, nonbilayer phases such as the hexagonalH II phase illustrated. Inverted phases do not support bilayer structure and areassociated with membrane fusion and membrane disruption 9,21 .+ – + –Cylindrical shape supportsbilayer structureBilayerCone shape disruptsbilayer structureHexagonal H IIDLinDMAHeadgroupLinkerHydrocarbonchains1 Tekmira Pharmaceuticals, Burnaby, British Columbia, Canada. 2 Alnylam Pharmaceuticals, Cambridge, Massachusetts, USA. 3 Department of Biochemistry andMolecular Biology and 4 Department of Chemistry, University of British Columbia, Vancouver, British Columbia, Canada. 5 Present address: Alcana Technologies,Vancouver, British Columbia, Canada. 6 These authors contributed equally to this work. Correspondence should be addressed to S.C.S. (ssemple@tekmirapharm.com)or A.A. (aakinc@alnylam.com).Received 16 September 2009; accepted 17 December 2009; published online 17 January 2010; doi:10.1038/nbt.1602172 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

l e t t e r saRelative serum factor VIIprotein (%)1201008060402000.01 0.1 1 10 100Factor VII siRNA dose (mg/kg)cMe 2 NOOMe 2 NRROORDLinDMAMe 2 NROORRMe 2 NMe 2 NODLinDAPO RRO Me2 NDLin-K-DMA DLin-KC2-DMA DLin-KC3-DMA DLin-KC4-DMAR =bRelative serum factor VIIprotein (%)1201008060402000.01 0.1 1 10Factor VII siRNA dose (mg/kg)OOROROORRFigure 2 In vivo evaluation of novel cationic lipids. (a) Silencingactivity of DLinDAP (), DLinDMA (), DLin-K-DMA () and DLin-KC2-DMA (•) screening formulations in the mouse Factor VII model.All LNP-siRNA systems were prepared using the preformed vesicle(PFV) method and were composed of ionizable cationic lipid, DSPC,cholesterol and PEG-lipid (40:10:40:10 mol/mol) with a FactorVII siRNA/total lipid ratio of ~0.05 (wt/wt). Data points are expressedas a percentage of PBS control animals and represent group mean(n = 5) ± s.d., and all formulations were compared within the samestudy. (b) Influence of headgroup extensions on the activity ofDLin-K-DMA. DLin-K-DMA () had additional methylene groups addedbetween the DMA headgroup and the ketal ring linker to generateDLin-KC2-DMA (•), DLin-KC3-DMA () and DLin-KC4-DMA (). Theactivity of PFV formulations of each lipid was assessed in the mouseFactor VII model. Data points are expressed as a percentage of PBScontrol animals and represent group mean (n = 4) ± s.d. (c) Chemicalstructures of novel cationic lipids.© 2010 Nature America, Inc. All rights reserved.of using ionizable cationic lipids with pK a s < pH 7.0 to efficientlyformulate nucleic acids at low pH and maintaining a neutral or lowcationic surface charge density at pH 7.4 (ref. 13). This strategy shouldprovide better control of the circulation properties of these systemsand reduce nonspecific disruption of plasma membranes. As positivecharge density is minimal in the blood but increases substantiallyin the acidic environment of the endosome, this should activatethe membrane-destabilizing property of the LNP. Although theseattributes may account for the activity of these systems upon internalizationby hepatocytes, they do not necessarily explain the highlevels of hepatic biodistribution observed for many LNPs, includingSNALP. Although these LNPs do not specifically include a targetingligand to direct them to hepatocytes after systemic administration,it is possible that these LNPs associate with one or more proteins inplasma that may promote hepatocyte endocytosis.The ionizable cationic lipid DLinDMA has proven to be highlyeffective in SNALP, has been extensively tested in rodents and nonhumanprimates, and is now being evaluated in human clinicaltrials. Therefore, we selected it as the starting point for the designand synthesis of novel lipids. We chose the mouse Factor VII model 7 ,as the primary in vivo screening system to assess functional LNPmediateddelivery to hepatocytes. Briefly, C57BL/6 mice receiveda single dose of LNP-formulated Factor VII siRNA through bolustail vein injection and serum was collected from animals 24 h afteradministration to analyze Factor VII protein level. The initial screeningof LNP-siRNA systems was conducted using LNPs prepared bya preformed vesicle method 14 and composed of ionizable cationiclipid, distearoylphosphatidylcholine (DSPC), cholesterol and PEGlipid(40:10:40:10 mol/mol), with a Factor VII siRNA/total lipid ratioof ~0.05 (wt/wt). Although not a bilayer-destabilizing lipid, a smallamount of phosphatidylcholine was incorporated into the LNP tohelp stabilize the LNP both during formulation and while it was incirculation. A short acyl chain PEG-lipid was incorporated into theLNP to control particle size during formulation, but is designed toleave the LNP rapidly upon intravenous injection. As our goal wasto identify novel ionizable cationic lipids for use in LNPs, we aimedto minimize other effects by using a single robust composition andset of formulation conditions suitable for all novel lipids tested. Thepreformed vesicle method employing the composition listed aboveprovides a convenient platform for such testing, but uses a differentformulation process, a different lipid composition and a differentsiRNA/lipid ratio than SNALP. The structure of DLinDMA can bedivided into three main regions: the hydrocarbon chains, the linkerand the headgroup (Fig. 1). A detailed structure-function study toinvestigate the impact of increasing the number of cis double bondsin the hydrocarbon chains found the linoleyl lipid containing twodouble bonds per hydrocarbon chain (DLinDMA) to be optimal 15 .We therefore maintained the linoleyl hydrocarbon chains present inDLinDMA as an element in our lipid design, and focused on optimizingthe linker and headgroup moieties.The linker region in a bilayer structure resides at the membraneinterface, an area of transition between the hydrophobic membranecore and hydrophilic headgroup surface. Our approach to linkermodification of DLinDMA involved introducing groups expected toexhibit different rates of chemical or enzymatic stability and to spana range of hydrophilicity. A variety of these rationally designed lipidswere made, characterized and tested (Supplementary Syntheses 1 andSupplementary Table 1). LNPs based on the ester-containing lipidDLinDAP showed substantially reduced in vivo activity comparedto LNPs based on the alkoxy-containing lipid DLinDMA (Fig. 2).Further, LNPs based on DLin-2-DMAP, a lipid with one alkoxy linkageand one ester linkage, yielded activity intermediate betweenDLinDAP- and DLinDMA-based LNPs (Supplementary Table 1).Although it is uncertain why the ester-containing lipids are considerablyless active in vivo, we speculate that the diester lipid (DLinDAP)is relatively inactive because it is more readily hydrolyzed in vivothan the alkoxy analog (DLinDMA), and therefore, unable to eitherprotect the siRNA adequately before release from the endosomeand/or survive long enough in the endosome to disrupt the membrane.These hypotheses are being investigated. LNPs based on lipidscontaining carbamate or thioether linkages also resulted in dramaticallyreduced in vivo activity. Interestingly, the introduction of a ketalring linker into DLinDMA resulted in LNPs that were ~2.5-fold morepotent in reducing serum Factor VII protein levels relative to theDLinDMA benchmark, with an ED 50 (that is, dose to achieve 50%gene silencing) of ~0.4 mg/kg versus 1 mg/kg, respectively (Fig. 2).Given the importance of positive charge in the mechanismof-actionhypothesis guiding the lipid design, the effects of structuralchanges in the amine-based headgroup were investigated in the contextof DLin-K-DMA as the new benchmark lipid. A series of headgroupmodifications were made, characterized and tested to explorethe effects of size, acid-dissociation constant and number of ionizablegroups (Supplementary Syntheses 2 and Supplementary Table 2).Piperazino, morpholino, trimethylamino or bis-dimethylamino modificationstested were not better than the benchmark dimethylaminoheadgroup of DLin-K-DMA. As an additional parameter, the distancebetween the dimethylamino group and the dioxolane linker was variedby introducing additional methylene groups. This parameter cannature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 173

l e t t e r s© 2010 Nature America, Inc. All rights reserved.Table 1 Biophysical parameters and in vivo activities ofLNPs containing novel lipidsCationic lipidApparentlipid pK aaL II to H II phase transition In vivo ED 50temperature (°C) b (mg/kg)DLinDMA 6.8 ± 0.10 27 ~1DLinDAP 6.2 ± 0.05 26 40–50DLin-K-DMA 5.9 ± 0.03 19 ~0.4DLin-KC2-DMA 6.7 ± 0.08 20 ~0.1DLin-KC3-DMA 7.2 ± 0.05 18 ~0.6DLin-KC4-DMA 7.3 ± 0.07 18 >3a pK a values ± s.d. (n = 3 to 9). b L II to H II phase transition was measured at pH 4.8 in equimolarmixtures with DSPS, using differential scanning calorimetric, repeat scans reproducibleto within 0.1 °C.affect both the pK a of the amine headgroup as well as the distanceand flexibility of the charge presentation relative to the lipid bilayerinterface. Inserting a single additional methylene group into the headgroup(DLin-KC2-DMA) produced a dramatic increase in potencyrelative to DLin-K-DMA. The ED 50 for this lipid was ~0.1 mg/kg,making it fourfold more potent than DLin-K-DMA and tenfold morepotent than the DLinDMA benchmark when compared head-to-headin the Factor VII model (Fig. 2a). Further extension of the tether withadditional methylene groups, however, substantially decreased activity,with an ED 50 of ~0.6 mg/kg for DLin-KC3-DMA and >3 mg/kgfor DLin-KC4-DMA (Fig. 2b).As changes in lipid design and chemistry may affect the pharmacokinetics,target tissue accumulation and intracellular delivery of LNPformulations, we investigated the relative importance of these parameterson LNP activity at an early stage in this research program.Several of the novel lipids were incorporated into LNP-siRNA formulationscontaining cyanine dye (Cy3)-labeled siRNA. Plasma,liver and spleen levels of siRNA were determined at 0.5 and 3 h afterinjection at siRNA doses of 5 mg/kg, and the results are presented inSupplementary Table 3. In general, formulations that were the mostactive in the mouse Factor VII screens achieved the highest liver levelsof siRNA at 0.5 h; however, delivery of siRNA to the target tissue wasnot the primary factor responsible for activity. This is supported bythe observations that most formulations accumulated in the liver andspleen quite quickly and that some formulations with similar liver levelsof siRNA had large differences in activity. Moreover, plasma pharmacokineticsalone did not predict activity. For example, althoughDLin-KC2-DMA and DLinDMA had virtually indistinguishableblood pharmacokinetic profiles in mice (data not shown), the activityof DLin-KC2-DMA in LNPs is approximately tenfold greater than thesame formulation with DLinDMA. Taken together, these results ledus to conclude that rapid target tissue accumulation was important,but not sufficient, for activity. Moreover, other parameters were morecritical for maximizing the activity of LNP-siRNA formulations.Two important parameters underlying lipid design for SNALPmediateddelivery are the pK a of the ionizable cationic lipid and theabilities of these lipids, when protonated, to induce a nonbilayer(hexagonal H II ) phase structure when mixed with anionic lipids.The pK a of the ionizable cationic lipid determines the surfacecharge on the LNP under different pH conditions. The charge stateat physiologic pH (e.g., in circulation) can influence plasma proteinadsorption, blood clearance and tissue distribution behavior 16 ,whereas the charge state at acidic pH (e.g., in endosomes) can influencethe ability of the LNP to combine with endogenous anioniclipids to form endosomolytic nonbilayer structures 9 . Consequently,the ability of these lipids to induce H II phase structure in mixtureswith anionic lipids is a measure of their bilayer-destabilizing capacityand relative endosomolytic potential.The fluorescent probe 2-(p-toluidino)-6-napthalene sulfonicacid (TNS), which exhibits increased fluorescence in a hydrophobicenvironment, can be used to assess surface charge on lipid bilayers.Titrations of surface charge as a function of pH can then be usedto determine the apparent pK a of the lipid in the bilayer (hereafterreferred to as pK a ) of constituent lipids 17 . Using this approach, thepK a values for LNPs containing DLinDAP, DLinDMA, DLin-K-DMA,DLin-KC2-DMA, DLin-KC3-DMA and DLin-KC4-DMA were determined(Table 1). The relative ability of the protonated form of theionizable cationic lipids to induce H II phase structure in anionic lipidswas ascertained by measuring the bilayer-to-hexagonal H II transitiontemperature (T BH ) in equimolar mixtures with distearoylphosphatidylserine(DSPS) at pH 4.8, using 31 P NMR 18 and differential scanningcalorimetric analyses 19 . Both techniques gave similar results.The data presented in Table 1 indicate that the highly active lipidDLin-KC2-DMA has pK a and T BH values that are theoretically favorablefor use in siRNA delivery systems. The pK a of 6.4 indicatesthat LNPs based on DLin-KC2-DMA have limited surface charge incirculation, but will become positively charged in endosomes. Further,the T BH for DLin-KC2-DMA is 7 °C lower than that for DLinDMA,suggesting that this lipid has improved capacity for destabilizingbilayers. However, the data also demonstrate that pK a and T BH donot fully account for the in vivo activity of lipids used in LNPs. Forexample, although DLin-KC3-DMA and DLin-KC4-DMA haveidentical pK a and T BH values, DLin-KC4-DMA requires a more thanfivefold higher dose to achieve the same activity in vivo. Moreover,Figure 3 Efficacy of KC2-SNALP in rodentsand nonhuman primates. (a) Improvedefficacy of KC2-SNALP relative to theinitial screening formulation tested inmice. The in vivo efficacy of KC2-SNALP() was compared to that of the unoptimizedDLin-KC2-DMA screening (that is, PFV)formulation ( • ) in the mouse Factor VII model.Data points are expressed as a percentageof PBS control animals and represent groupmean (n = 5) ± s.d. (b) Efficacy of KC2-SNALPin nonhuman primates. Cynomolgus monkeys(n = 3 per group) received a total dose of either0.03, 0.1, 0.3 or 1 mg/kg siTTR, or 1 mg/kgsiApoB formulated in KC2-SNALP or PBSas 15-min intravenous infusions (5 ml/kg)aRelative serum factor VII protein (%)1201008060402000.001 0.01 0.11 10Factor VII siRNA dose (mg/kg)bRelative liver TTR/GAPDH mRNA levels1.41.21.00.80.60.40.20PBS1mg/kgsiApoB0.03mg/kg**0.1 0.3mg/kg mg/kgsiTTRthrough the cephalic vein. Animals were euthanized 48 h after administration. TTR mRNA levels relative to GAPDH mRNA levels were determined inliver samples. Data points represent group mean ± s.d. *, P < 0.05; **, P < 0.005.**1mg/kg174 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

l e t t e r sTable 2 Clinical chemistry and hematology parameters for KC2-SNALP–treated ratsVehiclesiRNA dose(mg/kg) a ALT (U/L) AST (U/L)Total Bilirubin(mg/dl)BUN(mg/dl)RBC(× 10 6 /µl)Hemoglobin(g/dl)WBC(× 10 3 /µl) PLT (× 10 3 /µl)PBS 56 ± 16 109 ± 31 2 ± 0 4.8 ± 0.8 5.5 ± 0.3 11.3 ± 0.4 11 ± 3 1,166 ± 177KC2-SNALP 1 58 ± 22 100 ± 14 2 ± 0 4.4 ± 0.6 5.6 ± 0.2 11.6 ± 0.6 13 ± 2 1,000 ± 272KC2-SNALP 2 73 ± 9 81 ± 10 2.2 ± 0.4 4.3 ± 0.6 5.9 ± 0.3 11.6 ± 0.3 13 ± 4 1,271 ± 269KC2-SNALP 3 87 ± 19 100 ± 30 2 ± 0 5.0 ± 0.8 6.0 ± 0.2 11.9 ± 0.4 15 ± 2 958 ± 241a Nontargeting, luciferase siRNA. Sprague-Dawley rats (n = 5) received 15-min intravenous infusions of KC2-SNALP formulated siRNA at different dose levels. Blood samples were taken 24 hafter administration. ALT, alanine aminotransferase; AST, aspartate aminotransferase; BUN, blood urea nitrogen; RBC, red blood cells; WBC, white blood cells; PLT, platelets.© 2010 Nature America, Inc. All rights reserved.DLin-KC2-DMA and DLin-KC4-DMA, which have very similar pK aand T BH values, exhibit a >30-fold difference in in vivo activity. Thisresult suggests that other parameters, such as the distance and flexibilityof the charged group relative to the lipid bilayer interface, mayalso be important. Thus, although the biophysical parameters of pK aand T BH are useful for guiding lipid design, the results presented inTable 1 support the strategy of testing variants of lead lipids, evenones with very similar pK a and T BH values.The lipid composition chosen for the initial formulation andscreening of novel ionizable cationic lipids (cationic lipid/DSPC/cholesterol/PEG-lipid = 40:10:40:10 mol/mol, siRNA/total lipid~ 0.05 wt/wt) was useful for determining the rank-order potencyof novel lipids, but is not necessarily optimal for in vivo delivery. Inaddition, the in vivo activity of resultant LNP-siRNA formulationsis affected by the formulation process employed and the resultingparticle structure. Improvements in activity were possible with thepreformed vesicle process by modifying and optimizing lipid ratiosand formulation conditions (results not shown). However, we choseto further validate DLin-KC2-DMA activity specifically in the contextof the SNALP platform, currently the most advanced LNP formulationfor delivery of siRNA in vivo. We therefore tested in vivo a versionof SNALP (termed KC2-SNALP), which uses less PEG lipid thanreported previously 6 and in which DLinDMA was replaced with DLin-KC2-DMA. The incorporation of DLin-KC2-DMA into SNALP ledto a marked improvement in potency in the mouse Factor VII model;the measured ED 50 decreased from ~0.1 mg/kg for the unoptimizedscreening formulation to ~0.02 mg/kg for the KC2-SNALP formulation(Fig. 3a). KC2-SNALP also exhibited similar potency in rats(data not shown). Furthermore, after a single administration in rats,KC2-SNALP–mediated gene silencing was found to persist for over10 d (Supplementary Fig. 1).In addition to efficacy, tolerability is another critical attribute ofa suitable LNP-siRNA delivery system for human use. We thereforestudied the single-dose tolerability of KC2-SNALP in rats—a popularrodent model for assessing the toxicology of siRNA and nucleic acid–based therapeutics. As doses near the efficacious dose level were foundto be very well tolerated (data not shown), single-dose escalationstudies were conducted starting at doses ~50-fold higher (1 mg/kg)than the observed ED 50 of the formulation. To understand formulationtoxicity in the absence of any toxicity or pharmacologic effectsresulting from target silencing, we conducted the experiments usinga nontargeting control siRNA sequence directed against luciferase.KC2-SNALP containing luciferase siRNA was prepared in the exactsame manner as that containing Factor VII siRNA, and the resultantsize, lipid composition and entrapped siRNA/lipid ratio were similar.Clinical signs were observed daily and body weights, serum chemistryand hematology parameters were measured 72 h after dosing.KC2-SNALP was very well tolerated at the high dose levels examined(relative to the observed ED 50 dose) with no dose-dependent,clinically significant changes in key serum chemistry or hematologyparameters (Table 2).Given the promising activity and safety profile observed in rodents,studies were initiated in nonhuman primates to investigate the translationof DLin-KC2-DMA activity in higher species. For these studies,we chose to target transthyretin (TTR), a hepatic gene of hightherapeutic interest 20 . TTR is a serum protein synthesized primarilyin the liver, and although amyloidogenic TTR mutations are rare,they are endemic to certain populations and can affect the peripheralnerves, leading to familial amyloidotic polyneuropathy, and theheart, leading to familial amyloid cardiomyopathy. Currently, theonly disease-modifying therapy is liver transplantation. We treatedcynomolgus monkeys with a single 15-min intravenous infusion ofKC2-SNALP–formulated siTTR at siRNA doses of 0.03, 0.1, 0.3 and1 mg/kg. Control animals received a single 15-min intravenous infusionof PBS or KC2-SNALP–formulated ApoB siRNA at a dose of1 mg/kg. Tissues were harvested 48 h after administration andliver mRNA levels of TTR were determined. A clear dose responsewas obtained with an apparent ED 50 of ~0.3 mg/kg (Fig. 3b).A toxicological analysis indicated that the treatment was well toleratedat the dose levels tested, with no treatment-related changes in animalappearance or behavior. No dose-dependent, clinically significantalterations in key clinical chemistry or hematological parameterswere observed (Supplementary Table 4).In summary, we applied a rational approach to the design of novelcationic lipids, which were screened for use in LNP-based siRNAdelivery systems. Lipid structure was divided into three main functionalelements: alkyl chain, linker and headgroup. With DLinDMAas a starting point, the effect of each of these elements was investigatedin a systematic fashion, by holding the other two constant.First, the alkyl chains were established, then linker was varied and,finally, different headgroup structures were explored. Using thisapproach, important structure-activity considerations for ionizablecationic lipids were described and lipids with improved activityrelative to the DLinDMA benchmark were identified. A SNALPformulation of the best-performing lipid (DLin-KC2-DMA) waswell-tolerated in both rodent and nonhuman primates and exhibitedin vivo activity at siRNA doses as low as 0.01 mg/kg in rodents,as well as silencing of a therapeutically significant gene (TTR) innonhuman primates. Although the scope of the current work hasbeen limited to hepatic delivery in vivo, the TTR silencing achievedin this work (ED 50 ~ 0.3 mg/kg) represents a substantial improvementin activity relative to previous reports of LNP-siRNA mediatedsilencing in nonhuman primates.MethodsMethods and any associated references are available in the onlineversion of the paper at http://www.nature.com/naturebiotechnology/.Note: Supplementary information is available on the Nature Biotechnology website.AcknowledgmentsThe authors thank K. McClintock for assistance with animal studies. The authorsalso thank the Centre for Drug Research and Development at the Universitynature biotechnology VOLUME 28 NUMBER 2 FEBRUARY 2010 175

l e t t e r s© 2010 Nature America, Inc. All rights reserved.of British Columbia for use of the NMR facilities and M. Heller for his expertassistance in setting up the 31 P-NMR experiments.AUTHOR CONTRIBUTIONSJ.C., M.A.C., P.R.C., T.D.M., M.J.H. and K.F.W. designed and advised on novellipids. J.C., K.F.W. and M.S. synthesized novel lipids. M.J.H., T.D.M., J.C., K.F.W.,M.M., K.G.R., M.A.M., M.T. and M.J. analyzed and interpreted lipid data. T.D.M.,M.J.H. and M.A.T. co-directed novel lipid synthesis and screening program. S.C.S.designed and directed rodent in vivo studies. S.C.S., S.K.K., B.L.M., K.L., M.L.E.,M.K., A.P.S., Y.K.T., S.A.B., W.L.C., M.J.W. and E.J.C. generated rodent in vivodata, including Factor VII and tolerability analyses. L.N., V.K., T.B., R.A., Q.C.and D.W.Y.S. developed novel siRNAs targeting TTR. R.A. and A.A. designed anddirected NHP in vivo studies. S.C.S., S.K.K., A.A., B.L.M., I.M., A.P.S., Y.K.T., R.A.,T.B., D.W. Y. S., S.A.B., J.Q., J.R.D. and A.d.F. analyzed and interpreted in vivo data.B.L.M., K.L., A.P.S., S.K.K., S.C.S. and E.J.C. generated and characterized preformedvesicle formulations with novel lipids. D.S. and C.K.C. developed methods anddesigned and conducted HPLC lipid analyses of preformed vesicle formulations.E.Y. and L.B.J. prepared SNALP formulations. P.R.C. directed biophysical studiesand advised on methods. A.P.S., I.M.H., S.D. and K.W. performed biophysicalcharacterization studies (pK a , NMR, differential scanning calorimetric) of novellipids and formulations. M.J.H., P.R.C., T.D.M., A.P.S., I.M.H. and K.F.W. analyzedbiophysical data. S.C.S., M.J.H., A.A. and P.R.C. co-wrote the manuscript. T.D.M.,M.M., M.A.M., M.A.T. and A.D.F. reviewed and edited the manuscript. S.C.S.,M.J.H., A.A., P.R.C., I.M. and A.D.F. were responsible for approval of the final draft.COMPETING INTERESTS STATEMENTThe authors declare competing financial interests: details accompany the full-textHTML version of the paper at http://www.nature.com/naturebiotechnology/.Published online at http://www.nature.com/naturebiotechnology/.Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/.1. de Fougerolles, A.R. Delivery vehicles for small interfering RNA in vivo. Hum. GeneTher. 19, 125–132 (2008).2. Whitehead, K.A., Langer, R. & Anderson, D.G. Knocking down barriers: advancesin siRNA delivery. Nat. Rev. Drug Discov. 8, 129–138 (2009).3. Judge, A.D. et al. Confirming the RNAi-mediated mechanism of action ofsiRNA-based cancer therapeutics in mice. J. Clin. Invest. 119, 661–673 (2009).4. Judge, A.D. et al. Sequence-dependent stimulation of the mammalian innateimmune response by synthetic siRNA. Nat. Biotechnol. 23, 457–462 (2005).5. Morrissey, D.V. et al. Potent and persistent in vivo anti-HBV activity of chemicallymodified siRNAs. Nat. Biotechnol. 23, 1002–1007 (2005).6. Zimmermann, T.S. et al. RNAi-mediated gene silencing in non-human primates.Nature 441, 111–114 (2006).7. Akinc, A. et al. A combinatorial library of lipid-like materials for delivery of RNAitherapeutics. Nat. Biotechnol. 26, 561–569 (2008).8. Frank-Kamenetsky, M. et al. Therapeutic RNAi targeting PCSK9 acutely lowersplasma cholesterol in rodents and LDL cholesterol in nonhuman primates. Proc.Natl. Acad. Sci. USA 105, 11915–11920 (2008).9. Hafez, I.M., Maurer, N. & Cullis, P.R. On the mechanism whereby cationic lipidspromote intracellular delivery of polynucleic acids. Gene Ther. 8, 1188–1196(2001).10. Xu, Y. & Szoka, F.C. Jr. Mechanism of DNA release from cationic liposome/DNAcomplexes used in cell transfection. Biochemistry 35, 5616–5623 (1996).11. Zelphati, O. & Szoka, F.C. Jr. Mechanism of oligonucleotide release from cationicliposomes. Proc. Natl. Acad. Sci. USA 93, 11493–11498 (1996).12. Torchilin, V.P. Recent approaches to intracellular delivery of drugs and DNA andorganelle targeting. Annu. Rev. Biomed. Eng. 8, 343–375 (2006).13. Semple, S.C. et al. Efficient encapsulation of antisense oligonucleotides in lipidvesicles using ionizable aminolipids: formation of novel small multilamellar vesiclestructures. Biochim. Biophys. Acta 1510, 152–166 (2001).14. Maurer, N. et al. Spontaneous entrapment of polynucleotides upon electrostaticinteraction with ethanol-destabilized cationic liposomes. Biophys. J. 80,2310–2326 (2001).15. Heyes, J., Palmer, L., Bremner, K. & Maclachlan, I. Cationic lipid saturationinfluences intracellular delivery of encapsulated nucleic acids. J. Control. Release107, 276–287 (2005).16. Semple, S.C., Chonn, A. & Cullis, P.R. Interactions of liposomes and lipid-basedcarrier systems with blood proteins: Relation to clearance behaviour in vivo. Adv.Drug Deliv. Rev. 32, 3–17 (1998).17. Bailey, A.L. & Cullis, P.R. Modulation of membrane fusion by asymmetric transbilayerdistributions of amino lipids. Biochemistry 33, 12573–12580 (1994).18. Cullis, P.R. & de Kruijff, B. The polymorphic phase behaviour of phosphatidylethanolaminesof natural and synthetic origin. A 31P NMR study. Biochim. Biophys.Acta 513, 31–42 (1978).19. Epand, R.M., Robinson, K.S., Andrews, M.E. & Epand, R.F. Dependence of thebilayer to hexagonal phase transition on amphiphile chain length. Biochemistry 28,9398–9402 (1989).20. Sekijima, Y., Kelly, J.W. & Ikeda, S. Pathogenesis of and therapeutic strategies toameliorate the transthyretin amyloidoses. Curr. Pharm. Des. 14, 3219–3230(2008).21. Cullis, P.R., Hope, M.J. & Tilcock, C.P. Lipid polymorphism and the roles of lipidsin membranes. Chem. Phys. Lipids 40, 127–144 (1986).176 VOLUME 28 NUMBER 2 FEBRUARY 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.ONLINE METHODSSynthesis of cationic and PEG-lipids. A detailed description of the cationiclipid syntheses is available in the Supplementary Syntheses 1 and 2. Thesynthesis of N-[(methoxy poly(ethylene glycol) 2000 )carbamoyl]-1,2-dimyristyloxlpropyl-3-amine(PEG-C-DMA) was as described 22 . The synthesis ofR-3-[(ω-methoxy poly(ethylene glycol) 2000 )carbamoyl)]-1,2-dimyristyloxlpropyl-3-amine(PEG-C-DOMG) was as described 7 . These lipids were interchangeablein the formulation without substantially affecting activity (datanot shown), and are collectively referred to as PEG-lipid.siRNA synthesis. All siRNAs were synthesized by Alnylam and were characterizedby electrospray mass spectrometry and anion exchange highperformanceliquid chromatography (HPLC). The sequences for the sense andantisense strands of Factor VII, ApoB and control siRNAs have been reported 7 .The sequences for the sense and antisense strands of the TTR siRNA isas follows:siTTR sense: 5′-GuAAccAAGAGuAuuccAudTdT-3′; antisense: 5′-AUGGAAuACUCUUGGUuACdTdT-3′.2′-O-Me–modified nucleotides are in lowercase. siRNAs were generatedby annealing equimolar amounts of complementary sense andantisense strands.Preformed vesicle method to formulate LNP-siRNA systems. LNP-siRNAsystems were made using the preformed vesicle method 14 . Cationic lipid,DSPC, cholesterol and PEG-lipid were solubilized in ethanol at a molar ratioof 40:10:40:10, respectively. The lipid mixture was added to an aqueous buffer(50 mM citrate, pH 4) with mixing to a final ethanol and lipid concentrationof 30% (vol/vol) and 6.1 mg/ml, respectively, and allowed to equilibrate at 22 °Cfor 2 min before extrusion. The hydrated lipids were extruded through twostacked 80 nm pore-sized filters (Nuclepore) at 22 °C using a Lipex Extruder(Northern Lipids) until a vesicle diameter of 70–90 nm, as determined bydynamic light scattering analysis, was obtained. This generally required1–3 passes. The siRNA (solubilized in a 50 mM citrate, pH 4 aqueous solutioncontaining 30% ethanol) was added to the pre-equilibrated (35 °C)vesicles at a rate of ~5 ml/min with mixing. After a final target siRNA/lipidratio of 0.06 (wt/wt) was reached, the mixture was incubated for a further30 min at 35 °C to allow vesicle reorganization and encapsulation of the siRNA.The ethanol was then removed and the external buffer replaced with PBS(155 mM NaCl, 3 mM Na 2 HPO 4 , 1 mM KH 2 PO 4 , pH 7.5) by either dialysisor tangential flow diafiltration.Preparation of KC2-SNALP. siRNA were encapsulated in SNALP using a controlledstep-wise dilution method process as described 23 . The lipid constituentsof KC2-SNALP were DLin-KC2-DMA (cationic lipid), dipalmitoylphosphatidylcholine(DPPC; Avanti Polar Lipids), synthetic cholesterol (Sigma) andPEG-C-DMA used at a molar ratio of 57.1:7.1:34.3:1.4. Upon formation ofthe loaded particles, SNALP were dialyzed against PBS and filter sterilizedthrough a 0.2 µm filter before use. Mean particle sizes were 75–85 nm and90–95% of the siRNA was encapsulated within the lipid particles. The finalsiRNA/lipid ratio in formulations used for in vivo testing was ~0.15 (wt/wt).In vivo screening of cationic lipids for Factor VII activity. Eight- to 10-week-old,female C57BL/6 mice were obtained from Harlan. Mice were held in a pathogen-freeenvironment and all procedures involving animals were performed inaccordance with local, state and federal regulations, as applicable, and approvedby the Institutional Animal Care and Use Committee (IACUC). LNP-siRNAsystems containing Factor VII siRNA were diluted to the appropriateconcentrations in sterile PBS immediately before use and the formulations wereadministered intravenously through the lateral tail vein in a total volume of10 ml/kg. After 24 h, animals were anesthetized with ketamine/xylazine and bloodwas collected by cardiac puncture and processed to serum (microtainer serumseparator tubes; Becton Dickinson). Serum was tested immediately or stored at−70 °C for later analysis for Factor VII levels.Measurement of Factor VII protein in serum. Serum Factor VII levels weredetermined using the colorimetric Biophen VII assay kit (Aniara) 7 . Briefly,serially diluted pooled control serum (200–3.125%) and appropriatelydiluted serum samples from treated animals (n = 4–5 animals per doselevel) were analyzed in 96-well, flat bottom, nonbinding polystyrene assayplates (Corning) using the Biophen VII kit according to manufacturer’sinstructions. Absorbance was measured at 405 nm and a calibration curvewas generated using the serially diluted control serum to determine levelsof Factor VII in serum from treated animals, relative to the saline-treatedcontrol animals. ED 50 values for each formulation were derived from linearinterpolation of the Factor VII activity profile, and included data pointswithin 10–90% residual Factor VII activity (typically three to six points).Formulations containing novel lipids were always screened with one or morebenchmark formulations to control and assess assay variability over time,and formulations with promising activity were repeated, with an expandednumber of dose levels.In situ determination of pK a using TNS. The pK a of each cationic lipid wasdetermined in LNPs using TNS 17 and preformed vesicles composed of cationiclipid/DSPC/cholesterol/PEG-lipid (40:10:40:10 mol%) in PBS at a concentrationof ~6 mM total lipid. TNS was prepared as a 100 µM stock solution in distilledwater. Vesicles were diluted to 100 µM lipid in 2 ml of buffered solutionscontaining 1 µM TNS, 10 mM HEPES, 10 mM 4-morpholineethanesulfonicacid , 10 mM ammonium acetate, 130 mM NaCl, where the pH ranged from2.5 to 11. Fluorescence intensity was monitored in a stirred, thermostatedcuvette (25 °C) in a PerkinElmer LS-50 Spectrophotometer using excitationand emission wavelengths of 321 nm and 445 nm. Fluorescence measurementswere made 30 s after the addition of the lipid to the cuvette. A sigmoidal bestfit analysis was applied to the fluorescence data and the pK a was measured asthe pH giving rise to half-maximal fluorescence intensity.Differential scanning calorimetry. Analyses were performed using the samesamples used for 31 P NMR, on a TA Instruments Q2000 calorimeter using aheat/cool/heat cycle and a scan rate of 1 °C/minute between 10 °C and 70 °C.Repeat scans were reproducible to within 0.1 °C. The temperature at the peakamplitude of the endo- and exotherms was measured for both the heating andcooling scans, and the T BH values observed correlated closely with the phasetransition temperatures measured using 31 P NMR.Determination of siRNA plasma levels. Plasma levels of fluorescently labeledCy3 siRNA were evaluated at 0.5 and 3 h after intravenous injection of selectedLNP (preformed vesicle) formulations, administered at an siRNA dose of5 mg/kg, in C57BL/6 mice. Blood was collected in EDTA-containing Vacutainertubes, processed to plasma at 2–8 °C, and either assayed immediately or storedat –30 °C. An aliquot of the plasma (100 µl maximum) was diluted to 500 µlwith PBS (145 mM NaCl, 10 mM phosphate, pH 7.5); methanol (1.05 ml)and chloroform (0.5 ml) were added; and the sample was vortexed to obtain aclear, single-phase solution. Additional water (0.5 ml) and chloroform (0.5 ml)was added and the resulting emulsion was sustained by periodic mixing. Themixture was centrifuged at 500g for 20 min and the aqueous phase containingthe Cy3-labeled siRNA was collected and the fluorescence measured using anSLM Fluorimeter at an excitation wavelength of 550 nm (2 nm bandwidth)and emission wavelength of 600 nm (16 nm bandwidth). A standard curvewas generated by spiking aliquots of plasma from untreated animals withthe formulation containing Cy-3-siRNA (0 to 15 µg/ml), and the resultingstandards were processed as indicated above.Determination of siRNA biodistribution. Tissue (liver and spleen) levels ofsiRNA were evaluated at 0.5 and 3 h after intravenous injection in C57BL/6mice after administration of LNP (preformed vesicle) formulations containingselected novel lipids. After blood collection, animals were perfused withsaline to remove residual blood from the tissues; liver and spleen were thencollected, weighed and divided into two pieces. Portions (400–500 mg) ofliver or whole spleens were weighed into Fastprep tubes and homogenizedin 1 ml of Trizol using a Fastprep FP120 instrument. An aliquot of the homogenate(typically equivalent to 50 mg of tissue) was transferred to an Eppendorftube and additional Trizol was added to achieve a final volume of 1 ml.Chloroform (0.2 ml) was added and the solution was mixed and incubatedfor 2–3 min, before being centrifuged for 15 min at 12,000g. An aliquot(0.5 ml) of the aqueous phase was diluted with 0.5 ml of PBS and thedoi:10.1038/nbt.1602nature biotechnology

sample fluorescence was measured as described above. The data wereexpressed as the percent of the injected dose (in each tissue).In vivo nonhuman primate experiments. All procedures using cynomolgusmonkeys were conducted by a certified contract research organization usingprotocols consistent with local, state and federal regulations, as applicable, andapproved by the IACUC. Cynomolgus monkeys (n = 3 per group) receivedeither 0.03, 0.1, 0.3 or 1 mg/kg siTTR, or 1 mg/kg siApoB (used as control)formulated in KC2-SNALP as 15-min intravenous infusions (5 ml/kg) throughthe cephalic vein. Animals were euthanized 48 h after administration, anda 0.15–0.20 g sample of the left lateral lobe of the liver was collected andsnap-frozen in liquid nitrogen. Prior studies have established uniformity ofsilencing activity throughout the liver 6 . TTR mRNA levels, relative to GAPDHmRNA levels, were determined in liver samples using a branched DNA assay(QuantiGene Assay) 6 . Clinical chemistry and hematology parameters wereanalyzed before and 48 h after administration.Statistical analysis. P-values were calculated for comparison of K2C-SNALP–treated animals with PBS-treated animals using analysis of variance (ANOVA,single-factor) with an alpha value of 0.05. P < 0.05 was considered significant.22. Heyes, J., Hall, K., Tailor, V., Lenz, R. & MacLachlan, I. Synthesis and characterizationof novel poly(ethylene glycol)-lipid conjugates suitable for use in drug delivery.J. Control. Release 112, 280–290 (2006).23. Jeffs, L.B. et al. A scalable, extrusion-free method for efficient liposomalencapsulation of plasmid DNA. Pharm. Res. 22, 362–372 (2005).© 2010 Nature America, Inc. All rights reserved.nature biotechnologydoi:10.1038/nbt.1602

corrigenda & errataCorrigendum: Microdroplet-based PCR enrichment for large-scale targetedsequencingRyan Tewhey, Jason B Warner, Masakazu Nakano, Brian Libby, Martina Medkova, Patricia H David, Steve K Kotsopoulos, Michael L Samuels,J Brian Hutchison, Jonathan W Larson, Eric J Topol, Michael P Weiner, Olivier Harismendy, Jeff Olson, Darren R Link & Kelly A FrazerNat. Biotechnol. 27, 1025–1031 (2009); published online 1 November 2009; corrected after print 11 November 2009In the version of this article initially published, the email address for K.A.F. should have been kafrazer@ucsd.edu. The error has been corrected inthe HTML and PDF versions of the article.Corrigendum: The valuation high groundJeffrey J Stewart & Ben BonifantNat. Biotechnol. 27, 980–983 (2009); published online 24 September 2009; corrected online 6 November 2009; pdf corrected 5 February 2010In the version of this article initially published, the email address for Ben Bonifant was incorrect. The email address is bbonifant@campbellalliance.com. The error has been corrected in the HTML and PDF versions of the article.© 2010 Nature America, Inc. All rights reserved.Corrigendum: Receptor-binding specificity of pandemic influenza A (H1N1)2009 virus determined by carbohydrate microarrayRobert A Childs, Angelina S Palma, Steve Wharton, Tatyana Matrosovich, Yan Liu, Wengang Chai, Maria A Campanero-Rhodes,Yibing Zhang, Markus Eickmann, Makoto Kiso, Alan Hay, Mikhail Matrosovich & Ten FeiziNat. Biotechnol. 27, 797–799 (2009); published online 9 September 2009; corrected after print 5 February 2010In the version of this article initially published, two acknowledgments were inadvertently omitted: NCI Alliance of Glycobiologists for Detectionof Cancer and Cancer Risk; and Biotechnology and Biological Sciences Research Council. The error has been corrected in the HTML and PDFversions of the article.Corrigendum: Small but not simpleMarkus ElsnerNat. Biotechnol. 28, 42 (2010); published online 8 January 2010; corrected after print 5 February 2010In the version of this article initially published, the organisms in question were incorrectly identified as Mycobacterium pneumoniae andMycobacterium genitalium. The correct names are Mycoplasma pneumoniae and Mycoplasma genitalium, respectively. The error has been correctedin the HTML and PDF versions of the article.Erratum: A nuclear magnetic resonance technique for determininghybridoma cell concentration in hollow fiber bioreactorsAnthony Mancuso, Erik J. Fernandez, Harvey W. Blanch & Douglas S. ClarkBiotechnology 8, 1282–1285 (1990); corrected online 5 February 2010In the version of this article initially published online, a graph published in print as Figure 2 was erroneously duplicated and appeared as bothFigure 1 and Figure 2. The original Figure 1 has been restored in the online PDF version of the article.Erratum: Can web 2.0 reboot clinical trials?Malorye AllisonNat. Biotechnol. 27, 895–902 (2009); published online 8 October 2009; corrected after print 5 February 2010In the version of this article initially published, Sharib Khan was incorrectly identified as the CEO of TrialX. He is cofounder. The error has beencorrected in the HTML and PDF versions of the article.178 volume 28 number 2 february 2010 nature biotechnology

careers and recruitmentFourth quarter lag in biotech hiring© 2010 Nature America, Inc. All rights reserved.Michael FranciscoTougher times returned for the fourth quarter of 2009, as the slighthiring increase seen in the previous quarter at the 25 largest biotechsand 10 largest pharma companies (Nat. Biotechnol. 27, 1056, 2009) didnot hold, according to three representative job databases (Tables 1 and2). Big biotech and pharma companies across the board posted fewerjob listings with the exception of Bayer (Leverkusen, Germany), whoseUS job listings jumped from 4 to 78 on Monster.com.The landscape for existing jobs is mixed as well, with Seattlebasedbiotech VLST reportedly reducing its workforce by 30% aspart of a restructuring. Contract research organization Charles RiverLaboratories (Wilmington, MA, USA) is cutting 300 jobs and plans tosuspend operations at a preclinical services facility by the middle of2010, after the completion of ongoing projects.On the pharma side, Pfizer has informed New Jersey officials of itsplans to lay off 400 employees at Wyeth’s old research center, as partof its acquisition. In addition, Merck will eliminate 500 jobs—mostlysales and administrative positions—in conjunction with its takeoverof Schering-Plough. The company, however, plans to safeguard 500jobs as part of its acquisition of Avecia (Billingham, UK). Merckwill acquire all of Avecia’s assets, including all process developmentand scale-up, manufacturing, quality and business support operations.And Lonza Group of Basel, Switzerland, a large supplier ofactive pharmaceuticals ingredients, is closing three of its manufacturingsites in 2010. One hundred and seventy-five jobs are expectedTable 1 Who’s hiring? Advertised openings at the 25 largest biotech companiesCompany aNumber of employeesNumber of advertised openings bMonster Biospace NaturejobsMonsanto 21,700 0 0 112Amgen 16,800 35 45 2Genentech 11,186 11 22 73Genzyme 11,000 54 0 105Life Technologies 9,700 33 23 1PerkinElmer 7,900 16 0 0Bio-Rad Laboratories 6,600 5 10 0Biomerieux 6,140 2 0 0Millipore 5,900 4 22 0IDEXX Laboratories 4,700 11 1 0Biogen Idec 4,700 35 28 0Gilead Sciences 3,441 2 12 0WuXi PharmaTech 3,172 0 0 0Qiagen 3,041 0 0 0Cephalon 2,780 0 0 0Biocon 2,772 0 0 0Celgene 2,441 11 2 0Biotest 2,108 7 5 0Actelion 2,054 1 2 0Amylin Pharmaceuticals 1,800 21 8 0Elan 1,687 5 1 0Illumina 1,536 26 0 10Albany Molecular Research 1,357 0 0 0Vertex Pharmaceuticals 1,322 19 33 4CK Life Sciences 1,315 0 0 0a As defined in Nature Biotechnology’s survey of public companies (27, 710–721, 2009). b Assearched on Monster.com, Biospace.com and Naturejobs.com, January 10, 2010. Jobs may overlap.to be lost. The sites are located in Conshohocken, Pennsylvania,Shawinigan, Canada and Wokingham, UK. “The economic pressuresof the past 18 months have clearly accelerated the cost reductionefforts of the pharmaceutical industry,” according Lonza’s pressrelease. Adds Lonza CEO Stefan Borgas, “The closure of the threesites will help to optimize our global operational network and furtherincrease the competitiveness for our customers.” The company saysit will increase its platform in Asia.In a sign of confidence of biotech’s role in growing the economy,the governor of the state of Missouri embraced a proposal to “directtens of millions of tax dollars to Missouri’s biotechnology industry,”according to the Associated Press. The plan would divert an annualportion of the new tax revenues generated by biotech companies toa special state fund, from which incentives could be given to new orexpanding entrepreneurs in the same field, perhaps in the form of aidto startups, providing infrastructure to lure existing out-of-state firmsand subsidizing college-based training for the employees of biotechcompanies. The move is expected to result in a sizable number ofbiotech jobs. Missouri already is home to some top university andprivate-sector researchers in the life sciences. But economic developmentofficials say Missouri is lagging when it comes to converting thatresearch into commercial ventures.Nature Biotechnology will continue follow hiring and firing trendsthroughout 2010.Table 2 Advertised job openings at the ten largest pharma companiesCompany aNumber of employeesNumber of advertised openings bMonster Biospace NaturejobsJohnson & Johnson 119,200 203 1 0Bayer 106,200 78 30 1GlaxoSmithKline 103,483 1 3 4Sanofi-Aventis 99,495 0 0 0Novartis 98,200 3 30 60Pfizer 86,600 39 33 8Roche 78,604 8 17 4Abbott Laboratories 68,697 30 0 0AstraZeneca 67,400 18 7 5Merck & Co. 59,800 0 0 0a Data obtained from MedAdNews. b As searched on Monster.com, Biospace.com and Naturejobs.com,January 10, 2009. Jobs may overlap.Michael Francisco is Senior Editor, Nature Biotechnologynature biotechnology volume 28 number 2 february 2010 179

people© 2010 Nature America, Inc. All rights reserved.Inovio Biomedical (San Diego) has namedMark L. Bagarazzi as chief medical officer.Bagarazzi joins Inovio from Merck & Co.,where he was director of worldwide regulatoryaffairs for vaccines and biologics. Before joiningMerck in 2001, he was director of the HIV/AIDS program for St. Christopher’s Hospitalfor Children in Philadelphia.IRX Therapeutics (New York) has namedNeil L. Berinstein CSO, succeeding companyfounder John W. Hadden, who remains amember of the board of directors. Berinsteinpreviously served as assistant vice presidentand global program leader at Sanofi Pasteur,where he was in charge of leading the developmentof the company’s cancer vaccines bothstrategically and operationally.Genzyme (Cambridge, MA, USA) has namedRon Branning as its new senior vice president ofglobal product quality. Branning is tasked withensuring the quality of all Genzyme productsmanufactured at 17 sites around the world. Hebrings 30 years of experience in product qualityand regulatory affairs at biotech and pharmacompanies including Johnson & Johnson, GileadSciences and Genentech as well as companiesacquired by Baxter, Wyeth and Pfizer.Makefield Therapeutics (Newtown, PA, USA) has announcedthe appointment of Jim Ballance (left) as CSO. Most recently acorporate strategic consultant to the biotech industry, Ballancebrings more than 20 years of experience in the development andmanufacturing of novel therapeutics. His experience includes suchpositions as vice president, technology development at BioRexisPharmaceutical, director of biotech evaluation at Aventis Behring,chief technology officer at Genesis Therapeutics and head of R&Dat Delta Biotechnology.team following its recently completed mergerwith Neuromed. Mark Corrigan, former executivevice president of R&D at Sepracor and amember of the CombinatoRx board of directors,will assume the role of president and CEOof CombinatoRx and Christopher Gallen, formerCEO of Neuromed, will serve as executivevice president of R&D. CombinatoRx’s interimpresident and CEO Robert Forrester has decidedto leave his position at the company to pursueother opportunities.Roger Hickling has been appointed R&Ddirector and a board director at Phytopharm(Godmanchester, UK). He previously workedat SmithKline Beecham where he oversawboth in-house and partnered early stage neurosciencedevelopment projects, and was mostrecently R&D director and a board member atAlizyme Therapeutics.Prana Biotechnology (Melbourne, Australia)has named Paul Marks as a director of thecompany. Marks was previously vice presidentof foreign exchange with Prudential-Bache Securities and senior FX strategistwith National Australia Bank. He also servesas director of Conquest Mining and severalprivate companies.annual shareholder meeting. The company hasstarted a search for Mullen’s successor.CrystalGenomics (Seoul, S. Korea) has chosenEric M. Nelson to serve as vice presidentof business development for its USsubsidiary, Emeryville, California-based CGPharmaceuticals. Nelson brings 22 years ofexperience in business development andlicensing for pharma and biotech companies.Most recently, he was global head of businessdevelopment at Advinus Therapeutics.Proteonomix (Mountainside, NJ, USA)has announced the following managementchanges: Joel Pensley has resigned as secretary,director and general counsel and RogerFidler has joined the company as director andgeneral counsel; and Steven Byle, presentlya director, has assumed the role of secretary.Pensley resigned as an officer and director dueto health concerns. Fidler has been the soledirector, president, CEO and CFO of GlobalAgri-Med Technologies.Eric Ruby has joined Presidio Pharmaceuticals(San Francisco) as vice president of regulatoryaffairs. He served previously as seniordirector, regulatory affairs at AlnylamPharmaceuticals, where he was responsiblefor regulatory filings to support clinical trialsin the US and Europe.Protagen (Dortmund, Germany) hasappointed two new members to its executiveboard: Peter Schulz-Knappe, former CSO atProteome Sciences, will head the company’sdiagnostics business unit, and co-founder andCOO Martin Blüggel will manage the proteinservices unit.Ian Brown has been named CEO of BioCeramicTherapeutics (London) as part of a successionplan developed by the board of directors to succeedthe company’s outgoing founding CEO,Daniel Green. Brown has held senior executivepositions with Chromogenix (previouslyKabi Pharmacia), Instrumentation Laboratory,Cordlife, SDP Technology and Avanti Capital.Green will remain as a company director.CombinatoRx (Cambridge, MA, USA) hasannounced changes to its senior managementBIND Biosciences (Cambridge, MA, USA) hasannounced the appointment of Scott Minickas president and CEO. Minick was formerly amanaging director at ARCH Venture Partnersand previously president and COO of SequusPharmaceuticals/Liposome Technology.Biogen Idec (Cambridge, MA, USA) hasannounced that James Mullen will retire aspresident and CEO as of June 8. Mullen willalso retire from Biogen’s board at the end of hiscurrent term as a director at the company’s 2010Pieris (Freising-Weihenstephan,Germany) hasannounced theappointment ofStephen S. Yoder(left) as CEO. Hesucceeds interimCEO and co-founderClaus Schalper, who will remain CFO. Yoderjoins Pieris from MorphoSys, where he servedas general counsel and head of licensing & IP.180 volume 28 number 2 february 2010 nature biotechnology