The Genom of Homo sapiens.pdf

286 OVCHARENKO AND LOOTSmodulation of gene expression is achieved through thecomplex interaction of regulatory proteins (trans-factors)with specific DNA regions (cis-acting regulatory sequences)(Krivan and Wasserman 2001). Intensive experimentalefforts over several decades have identifiednumerous regulatory proteins, transcription factors (TF),and their DNA-binding specificities. The sequence-specificDNA-binding activity of TFs is central to transcriptionalregulation and regulatory networks. For most of theknown TFs, the DNA-binding sequence motifs werefound to be short (6–12 bp) and highly degenerate, andsuch specificities are cataloged in the TRANSFACdatabases (http://www.biobase.de/) (Wingender et al.2001; Matys et al. 2003). Pattern-recognition programs(MATCH or MatInspector) (Quandt et al. 1995) use thisdatabase to carry out the reverse-engineering task of predictingsignificant motif matches in DNA sequences,which could serve as an in silico strategy for detectingtranscription factor binding sites (TFBSs). Due to theirhighly degenerate nature, TF motifs occur very frequentlyin short genomic intervals, and only a very smallfraction of the predicted sites are biologically significant.This fact limits the use of TFBS databases for sequencebaseddiscovery of transcriptional regulatory elements(Fickett and Wasserman 2000).Multispecies comparative sequence analysis or phylogeneticfootprinting has been suggested as a strategy tocounter the large numbers of TFBS false positives derivedfrom the analysis of a single sequence (Gumucio et al.1996; Duret and Bucher 1997; Levy et al. 2001). In general,it is believed that TFBSs are the building blocks ofgene regulation. Several studies have carefully shown thattranscriptional regulatory elements are evolutionarily conserved,supporting the use of genomic comparisons for thede novo discovery of gene regulatory elements (Hardisonet al. 1997; Oeltjen et al. 1997; Loots et al. 2000). In addition,in complex organisms, gene expression results fromthe cooperative action of many different proteins simultaneouslyrequired to cooperatively activate and modulategene expression (Berman et al. 2002). Therefore, a potentialavenue for improving the discovery of functional regulatoryelements is to identify multiple TFBSs that arespecifically clustered together (Kel et al. 1999; Zhu et al.2002). This strategy has been implemented successfully inthe analysis of regulatory regions involved in muscle(Wasserman and Fickett 1998) and liver-specific gene expression(Krivan and Wasserman 2001).To facilitate the efficient and accurate prediction of biologicallyfunctional regulatory sequences present in largegenomic intervals, we have developed a computationaltool, Regulatory VISTA or rVISTA (http://rvista.dcode.org/) that enriches for functional TFBSs using evolutionaryconservation (Loots et al. 2002). The rVISTAtool combines TFBS motif recognition, orthologous sequencealignments, and TFBS cluster analysis to overcomesome of the limitations associated with TFBS predictionsof sequences derived from a single organism. Theanalysis proceeds in four steps: (1) identification of TFBSmatches in the individual sequences, (2) identification oflocally aligned noncoding TFBSs, (3) calculation of localconservation extending upstream and downstream fromeach orthologous TFBS, and (4) visualization of individualor clustered noncoding TFBSs. Alignments generatedby the zPicture (http:zpicture.dcode. org/) program can beprocessed by rVISTA by following the rVISTA link providedin the results page. The user can also submit alignmentfiles generated by PipMaker along with the correspondingsequence annotations (optional) to identifyconserved TFBS matches present only in noncoding genomicintervals. Pre-computed matrices imported fromthe TRANSFAC database or user-defined consensus sequencescan be used to identify TFBS motifs in the inputsequences. The alignment and annotation files are usednext to identify all the aligned TFBSs present in noncodingDNA and to calculate the degree of DNA conservationencompassing each TFBS. rVISTA calculates the maximumDNA conservation surrounding each aligned bindingsite in a dynamically shifting window, 20 bp in length,and filters out the sites present in regions that are less than80% conserved (Fig. 2) (Loots et al. 2002).The data generated by rVISTA are compiled into twotypes of outputs: (1) static data files and (2) a dynamicWeb-interactive graphical user interface that maps TFBSon top of the conservation plot. The static text files includedata tables with detailed statistics for all alignedand conserved TFBSs, alignment files depicting the textfor each TFBS match in a different color, and files withthe numerical position of each TFBS within the referencesequence. The visualization module allows the user tocustomize the data and graphically visualize TFBS togetherwith the alignment conservation plot. Since regulatoryregions in higher eukaryotes are represented byconglomerates of multiple TFBSs that act in concordanceto directly modulate the expression patterns of the linkedgenes (Pilpel et al. 2001), rVISTA calculates the distancebetween all neighboring TFBSs and allows the user toperform customized clustering of individual or multipleunique transcription factors. One clustering module allowsthe user to selectively cluster two or more sites ofthe same TF present in regions of user-defined lengths,and a second clustering module allows the user to identifygroups of multiple sites specific for different TFs, to predictDNA regions with unique regulatory signatures (Fig.2) (Loots et al. 2002).Recently, the ECR Browser has included an rVISTAportal that takes advantage of the available precomputedwhole-genome pair-wise alignments, eliminating theneed to create an alignment file prior to TFBS analysiswhile using the rVISTA tool. Users now have the optionto browse the human genome and to perform rVISTAanalysis on individual highly conserved noncoding elementsby using the button, or on long genomicintervals containing several blocks of conservationby using the link. rVISTA analysis canthen be conducted on any available pair-wise alignmentsby pushing the button. From this point on,TFBS analysis proceeds as described in Figure 2B–E, andis processed by the rVISTA software.Annotating the noncoding portion of the humangenome still remains one of the greatest challenges post-