Initial sequencing and analysis of the human genome - Vitagenes

articles30% to 50%. The correlation appears to be due primarily to intronsize, which drops markedly with increasing GC content (Fig. 36c).In contrast, coding properties such as exon length (Fig. 36c) or exonnumber (data not shown) vary little. Intergenic distance is alsoprobably lower in high-GC areas, although this is hard to provedirectly until all genes have been identi®ed.The large number of con®rmed human introns allows us toanalyse variant splice sites, con®rming and extending recentreports 281 . Intron positions were con®rmed by applying a stringentcriterion that EST or mRNA sequence show an exact match of 8 bpin the ¯anking exonic sequence on each side. Of 53,295 con®rmedintrons, 98.12% use the canonical dinucleotides GT at the 59 splicesite and AG at the 39 site (GT±AG pattern). Another 0.76% use therelated GC±AG. About 0.10% use AT±AC, which is a rare alternativepattern primarily recognized by the variant U12 splicingmachinery 282 . The remaining 1% belong to 177 types, some of whichundoubtedly re¯ect sequencing or alignment errors.Finally, we looked at alternative splicing of human genes. Alternativesplicing can allow many proteins to be produced from asingle gene and can be used for complex gene regulation. It appearsto be prevalent in humans, with lower estimates of about 35% ofhuman genes being subject to alternative splicing 283±285 . Thesestudies may have underestimated the prevalence of alternativesplicing, because they examined only EST alignments coveringonly a portion of a gene.To investigate the prevalence of alternative splicing, we analysedreconstructed mRNA transcripts covering the entire coding regionsof genes on chromosome 22 (omitting small genes with codingregions of less than 240 bp). Potential transcripts identi®ed byalignments of ESTs and cDNAs to genomic sequence were veri®edby human inspection. We found 642 transcripts, covering 245 genes(average of 2.6 distinct transcripts per gene). Two or more alternativelyspliced transcripts were found for 145 (59%) of these genes.A similar analysis for the gene-rich chromosome 19 gave 1,859transcripts, corresponding to 544 genes (average 3.2 distinct transcriptsper gene). Because we are sampling only a subset of alltranscripts, the true extent of alternative splicing is likely to begreater. These ®gures are considerably higher than those for worm,in which analysis reveals alternative splicing for 22% of genes forwhich ESTs have been found, with an average of 1.34 (12,816/9,516)splice variants per gene. (The apparently higher extent of alternativesplicing seen in human than in worm was not an artefact resultingfrom much deeper coverage of human genes by ESTs and mRNAs.Although there are many times more ESTs available for human thanworm, these ESTs tend to have shorter average length (because manywere the product of early sequencing efforts) and many match nohuman genes. We calculated the actual coverage per bp used in theanalysis of the human and worm genes; the coverage is onlymodestly higher (about 50%) for the human, with a strong biastowards 39 UTRs which tend to show much less alternative splicing.We also repeated the analysis using equal coverage for the twoorganisms and con®rmed that higher levels of alternative splicingwere still seen in human.)Seventy per cent of alternative splice forms found in the genes onchromosomes 19 and 22 affect the coding sequence, rather thanmerely changing the 39 or 59 UTR. (This estimate may be affected bythe incomplete representation of UTRs in the RefSeq database andin the transcripts studied.) Alternative splicing of the terminal exonwas seen for 20% of 6,105 mRNAs that were aligned to the draftgenome sequence and correspond to con®rmed 39 EST clusters. Inaddition to alternative splicing, we found evidence of the terminalexon employing alternative polyadenylation sites (separated by. 100 bp) in 24% of cases.Towards a complete index of human genes. We next focused oncreating an initial index of human genes and proteins. This index isquite incomplete, owing to the dif®culty of gene identi®cation inhuman DNA and the imperfect state of the draft genome sequence.Nonetheless, it is valuable for experimental studies and providesimportant insights into the nature of human genes and proteins.The challenge of identifying genes from genomic sequence variesgreatly among organisms. Gene identi®cation is almost trivial inbacteria and yeast, because the absence of introns in bacteria andtheir paucity in yeast means that most genes can be readilyrecognized by ab initio analysis as unusually long ORFs. It is notas simple, but still relatively straightforward, to identify genes inanimals with small genomes and small introns, such as worm and¯y. A major factor is the high signal-to-noise ratioÐcodingsequences comprise a large proportion of the genome and a largeproportion of each gene (about 50% for worm and ¯y), and exonsare relatively large.Gene identi®cation is more dif®cult in human DNA. The signalto-noiseratio is lower: coding sequences comprise only a few percent of the genome and an average of about 5% of each gene;internal exons are smaller than in worms; and genes appear to havemore alternative splicing. The challenge is underscored by the workon human chromosomes 21 and 22. Even with the availability of®nished sequence and intensive experimental work, the gene contentremains uncertain, with upper and lower estimates differing byas much as 30%. The initial report of the ®nished sequence ofchromosome 22 (ref. 94) identi®ed 247 previously known genes,298 predicted genes con®rmed by sequence homology or ESTs and325 ab initio predictions without additional support. Many of thecon®rmed predictions represented partial genes. In the past year,440 additional exons (10%) have been added to existing geneannotations by the chromosome 22 annotation group, althoughthe number of con®rmed genes has increased by only 17 and somepreviously identi®ed gene predictions have been merged 286 .Before discussing the gene predictions for the human genome, itis useful to consider background issues, including previous estimatesof the number of human genes, lessons learned from wormsand ¯ies and the representativeness of currently `known' humangenes.Previous estimates of human gene number. Although direct enumerationof human genes is only now becoming possible with the adventof the draft genome sequence, there have been many attempts in thepast quarter of a century to estimate the number of genes indirectly.Early estimates based on reassociation kinetics estimated the mRNAcomplexity of typical vertebrate tissues to be 10,000±20,000, andwere extrapolated to suggest around 40,000 for the entire genome 287 .In the mid-1980s, Gilbert suggested that there might be about100,000 genes, based on the approximate ratio of the size of a typicalgene (,3 ´ 10 4 bp) to the size of the genome (3 ´ 10 9 bp). Althoughthis was intended only as a back-of-the-envelope estimate, thepleasing roundness of the ®gure seems to have led to it beingwidely quoted and adopted in many textbooks. (W. Gilbert,personal communication; ref. 288). An estimate of 70,000±80,000genes was made by extrapolating from the number of CpG islandsand the frequency of their association with known genes 129 .As human sequence information has accumulated, it has beenpossible to derive estimates on the basis of sampling techniques 289 .Such studies have sought to extrapolate from various types of data,including ESTs, mRNAs from known genes, cross-species genomecomparisons and analysis of ®nished chromosomes. Estimatesbased on ESTs 290 have varied widely, from 35,000 (ref. 130) to120,000 genes 291 . Some of the discrepancy lies in differing estimatesof the amount of contaminating genomic sequence in the ESTcollection and the extent to which multiple distinct ESTs correspondto a single gene. The most rigorous analyses 130 exclude asspurious any ESTs that appear only once in the data set and carefullycalibrate sensitivity and speci®city. Such calculations consistentlyproduce low estimates, in the region of 35,000.Comparison of whole-genome shotgun sequence from the puffer®shT. nigroviridis with the human genome 292 can be used toestimate the density of exons (detected as conserved sequences898 NATURE | VOL 409 | 15 FEBRUARY 2001 | www.nature.com© 2001 Macmillan Magazines Ltd