Automatic Procedures for Compilation of Promoter sequences and ...

Automatic Procedures for Compilation of Promoter sequences and their
Evaluation based on Signal Content and Positional Distributions
Christoph D. Schmid 1 , Viviane Praz 1,2 , Mauro Delorenzi 1,2 , Rouaida Perier 1 , Philipp Bucher 1,2
1 Swiss Institute of Bioinformatics (SIB), 2 Swiss Insitute of Experimental Cancer Research (ISREC), Ch. des boveresses, CH-
1066 Epalinges s/Lausanne (Switzerland)
Christoph.Schmid, Viviane.Praz, Mauro.Delorenzi, Rouaida.Perier, Philipp.Bucher@isb-sib.ch
Keywords: Bioinformatics, Computational tools, Functional motifs, Genomic organization, Regulation of transcription
Introduction:
Collections of precisely mapped eukaryotic transcription start sites (TSS) are important resources for studying gene
control elements and for developing promoter prediction algorithms. The present study consists of two parts. First,
we present a new clustering method to infer TSS positions from EST 5' ends. Second, we present a quality
evaluation of the promoter sequences defined by TSS positions from the first part of the study.
Methods:
Briefly, TSS were determined using an in silico version of standard primer extension experiments. As input, cDNA
sequences from carefully selected full-length sequencing projects were aligned to genomic sequences. For each gene
the genomic positions of the 5'end of the cDNAs were recorded in a data structure called a "cDNA 5' end profile". A
new program named "MADAP" was applied to define zero to several significant TSS clusters representative of a
promoter. This program attempts to model the cDNA 5'end profiles by zero to several gaussian probability
distributions.
Results:
Promoter compilation:
Based on the approach described above, two novel promoter sequence sets were generated.
(i) TSS clusters based on 5' ESTs from the MGC project [1], and
(ii) TSS clusters based on the 5' ends of oligo-capped cDNAs from DBTSS [2].
As comparison we used two existing promoter sets available from public websites.
(iii) promoters as described in the literature and manually compiled in EPD [3],
(iv) automatically compiled promoters from the PRESTA database [4].
DBTSS MGC PRESTA EPD
Genes with a minimum of 10 cDNA transcripts 1522 8806 N.A. N.A.
Genes with at least one promoter found and mapped to a
1034 1038 424 255
genomic EMBL sequence
Genes with 1 promoter 971 965 292 239
Genes with 2 alternative promoters 57 69 63 10
Genes with 3 alternative promoters 5 4 2 4
Genes with 4 alternative promoters 1 0 0 2
Total number of promoters 1104 1115 490 279

Promoter evaluation:
To evaluate the quality of the four promoter sets, we searched the sequences around the TSS positions for common
promoter elements using the Signal Search Analysis server [5]. The positional occurrence frequencies of elements
such as the Initiator, TATA-box, CCAAT-box, GC-box, as well as the presence of CpG islands, were used as
indicator of the percentage of correctly mapped promoters in each of the different sets. Note that for a single
promoter sequence, the presence or absence of these elements is not significant, as these elements occur only in
(variable) subsets of promoters [6]. However, for sets of promoters of equal quality, we expect a similar average of
the signal content.
We find that the frequencies of the CCAAT-box and the GC-box are similar for all four data sets. Based on this
criterion, all sets appear to be of comparable quality and reasonably pure if we assume that the manually compiled
EPD collection contains only few false entries. We also tried to estimate the precision of the initiation site mapping
on the basis of the width of the TATA-box and Initiator element peaks. Both elements are believed to occur at nearly
fixed distances from the TSS [6]. The narrowest TATA-box peak appears in the signal occurrence profile for the
DBTSS set at the expected position -27. The same promoter set also contains the highest proportion of TSS that
exactly coincide with an Initiator motif. These two tests distinguish the promoter set derived from DBTSS, which is
based on oligo-capped cDNAs, as the most precisely, mapped one, providing strong evidence for the effectiveness
of this cloning technique in generating full-length cDNAs.
Remarkably, EPD appears to be enriched in TATA-box containing promoters with a single initiation site. This may
result from a publication bias against TATA-less promoters with diffuse TSS patterns in the time period before 1990
when most EPD entries were collected.
In summary, the goal of this study is to evaluate the quality of automatically compiled promoter sets as compared to
the manually compiled promoters from EPD. Another contribution of this work is a new method to infer
transcription start sites (TSSs) from collections of cDNA 5’ends mapped to the same genome region.
Our results demonstrate that 5'end sequencing of oligo-capped cDNA libraries in conjunction with our newly
developed data processing tools constitutes the most efficient, accurate, and reliable method for large-scale
eukaryotic promoter mapping. It is hoped that these results will encourage public support for promoter mapping
projects based on this technology for various organisms.
References
[1] Strausberg, R.L. et al. (2002) Generation and initial analysis of more than 15,000 full-length human and mouse cDNA
sequences. Proc. Natl. Acad. Sci. U.S.A. 99, 16899-16903.
[2] Suzuki, Y., Yamashita, R., Nakai, K.and Sugano, S. (2002) DBTSS: DataBase of human Transcriptional Start Sites and fulllength
cDNAs. Nucleic Acids Res. 30, 328-331.
[3] Praz, V., Perier, R., Bonnard, C. and Bucher, P. (2002) The Eukaryotic Promoter Database, EPD: new entry types and links to
gene expression data. Nucleic Acids Res. 30, 322-324.
[4] Mach, V. (2002) PRESTA: associating promoter sequences with information on gene expression. Genome Biol. Research
3(9):0050.1.
[5] Ambrosini et al., The Signal Search Analysis server Nucleic Acids Res. in press, 2003.
[6] Bucher, P. (1990) Weight matrix descriptions of four eukaryotic RNA polymerase II promoter elements derived from 502
unrelated promoter sequences. J. Mol. Biol. 212, 563-578.