articlesThe pericentromeric regions are structurally very complex, asillustrated for chromosome 21 in Fig. 32a. The pericentromericregions appear to have been bombarded by successive insertions <strong>of</strong>duplications; <strong>the</strong> insertion events must be fairly recent because <strong>the</strong>degree <strong>of</strong> sequence conservation with <strong>the</strong> genomic source loci isfairly high (90±100%, with an apparent peak around 96%). Distinctinsertions are typically separated by AT-rich or GC-rich minisatellite-likerepeats that have been hypo<strong>the</strong>sized to have a functionalrole in targeting duplications to <strong>the</strong>se regions 233,241 .A single genomic source locus <strong>of</strong>ten gives rise to pericentromericcopies on multiple chromosomes, with each having essentially <strong>the</strong>same breakpoints <strong>and</strong> <strong>the</strong> same degree <strong>of</strong> divergence. An example <strong>of</strong>such a source locus on Xq28 is shown in Fig. 32b. Phylogenetic<strong>analysis</strong> has suggested a two-step mechanism for <strong>the</strong> origin <strong>and</strong>dispersal <strong>of</strong> <strong>the</strong>se segments, whereby an initial segmental duplicationin <strong>the</strong> pericentromeric region <strong>of</strong> one chromosome occurs <strong>and</strong> is<strong>the</strong>n redistributed as part <strong>of</strong> a larger cassette to o<strong>the</strong>r such regions 242 .A comprehensive <strong>analysis</strong> for all chromosomes will have to awaitcomplete <strong>sequencing</strong> <strong>of</strong> <strong>the</strong> <strong>genome</strong>, but <strong>the</strong> evidence from <strong>the</strong> draft<strong>genome</strong> sequence indicates that <strong>the</strong> same picture is likely to be seenthroughout <strong>the</strong> <strong>genome</strong>. Several papers have analysed ®nishedsegments within pericentromeric regions <strong>of</strong> chromosomes 2(160 kb), 10 (400 kb) <strong>and</strong> 16 (300 kb), all <strong>of</strong> which show extensiveinterchromosomal segmental duplication 215,219,232,233 . An examplefrom ano<strong>the</strong>r pericentromeric region on chromosome 11 isshown in Fig. 32c. Interchromosomal duplications in subtelomericregions also appear to be a fairly general phenomenon, as illustratedby a large tract (,500 kb) <strong>of</strong> complex duplication on chromosome 7(Fig. 32d).The explanation for <strong>the</strong> clustering <strong>of</strong> segmental duplications maybe that <strong>the</strong> <strong>genome</strong> has a damage-control mechanism wherebychromosomal breakage products are preferentially inserted intopericentromeric <strong>and</strong>, to a lesser extent, subtelomeric regions. Thepossibility <strong>of</strong> a speci®c mechanism for <strong>the</strong> insertion <strong>of</strong> <strong>the</strong>sesequences has been suggested on <strong>the</strong> basis <strong>of</strong> <strong>the</strong> unusual sequencesfound ¯anking <strong>the</strong> insertions. Although it is also possible that <strong>the</strong>seregions simply have greater tolerance for large insertions, manylarge gene-poor `deserts' have been identi®ed 93 <strong>and</strong> <strong>the</strong>re is noaccumulation <strong>of</strong> duplicated segments within <strong>the</strong>se regions. Alongwith <strong>the</strong> fact that transitions between duplicons (from differentregions <strong>of</strong> <strong>the</strong> <strong>genome</strong>) occur at speci®c sequences, this suggests thatactive recruitment <strong>of</strong> duplications to such regions may occur. In anycase, <strong>the</strong> duplicated regions are in general young (with manyduplications showing ,6% nucleotide divergence from <strong>the</strong>irsource loci) <strong>and</strong> in constant ¯ux, both through additional duplications<strong>and</strong> by large-scale exchange among similar chromosomalenvironments. There is evidence <strong>of</strong> structural polymorphism in<strong>the</strong> <strong>human</strong> population, such as <strong>the</strong> presence or absence <strong>of</strong> olfactoryreceptor segments located within <strong>the</strong> telomeric regions <strong>of</strong> several<strong>human</strong> chromosomes 226,227 .Genome-wide <strong>analysis</strong> <strong>of</strong> segmental duplications. We also performeda global <strong>genome</strong>-wide <strong>analysis</strong> to characterize <strong>the</strong> amount<strong>of</strong> segmental duplication in <strong>the</strong> <strong>genome</strong>. We `repeat-masked' <strong>the</strong>known interspersed repeats in <strong>the</strong> draft <strong>genome</strong> sequence <strong>and</strong>compared <strong>the</strong> remaining draft genomic sequence with itself in amassive all-by-all BLASTN similarity search. We excluded matchesin which <strong>the</strong> sequence identity was so high that it might re¯ectartefactual duplications resulting from a failure to overlap sequencecontigs correctly in assembling <strong>the</strong> draft <strong>genome</strong> sequence. Speci-®cally, we considered only matches with less than 99.5% identity for®nished sequence <strong>and</strong> less than 98% identity for un®nishedsequence.We took several approaches to avoid counting artefactual duplicationsin <strong>the</strong> sequence. In <strong>the</strong> ®rst approach, we studied only®nished sequence. We compared <strong>the</strong> ®nished sequence with itself,to identify segments <strong>of</strong> at least 1 kb <strong>and</strong> 90±99.5% sequenceidentity. This <strong>analysis</strong> will underestimate <strong>the</strong> extent <strong>of</strong> segmentalduplication, because it requires that at least two copies <strong>of</strong> <strong>the</strong>segment are present in <strong>the</strong> ®nished sequence <strong>and</strong> because sometrue duplications have over 99.5% identity.The ®nished sequence consists <strong>of</strong> at least 3.3% segmental duplication(Table 16). Interchromosomal duplication accounts forabout 1.5% <strong>and</strong> intrachromosomal duplication for about 2%,with some overlap (0.2%) between <strong>the</strong>se categories. We analysed<strong>the</strong> lengths <strong>and</strong> divergence <strong>of</strong> <strong>the</strong> segmental duplications (Fig. 33).The duplications tend to be large (10±50 kb) <strong>and</strong> highly homologous,especially for <strong>the</strong> interchromosomal segments. Thesequence divergence for <strong>the</strong> interchromosomal duplicationsappears to peak between 96.5% <strong>and</strong> 97.5%. This may indicatethat interchromosomal duplications occurred in a punctuatedmanner. It will be intriguing to investigate whe<strong>the</strong>r such genomicupheaval has a role in speciation events.In a second approach, we compared <strong>the</strong> entire <strong>human</strong> draft<strong>genome</strong> sequence (®nished <strong>and</strong> un®nished) with itself to identifyduplications with 90±98% sequence identity (Table 17). The draft<strong>genome</strong> sequence contains at least 3.6% segmental duplication. Theactual proportion will be signi®cantly higher, because we excludedmany true matches with more than 98% sequence identity (at least1.1% <strong>of</strong> <strong>the</strong> ®nished sequence). Although exact measurement mustawait a ®nished sequence, <strong>the</strong> <strong>human</strong> <strong>genome</strong> seems likely tocontain about 5% segmental duplication, with most <strong>of</strong> thissequence in large blocks (. 10 kb). Such a high proportion <strong>of</strong>large duplications clearly distinguishes <strong>the</strong> <strong>human</strong> <strong>genome</strong> fromo<strong>the</strong>r sequenced <strong>genome</strong>s, such as <strong>the</strong> ¯y <strong>and</strong> worm (Table 18).The structure <strong>of</strong> large highly paralogous regions presents one <strong>of</strong><strong>the</strong> `serious <strong>and</strong> unanticipated challenges' to producing a ®nishedsequence <strong>of</strong> <strong>the</strong> <strong>genome</strong> 46 . The absence <strong>of</strong> unique STS or ®ngerprintsignatures over large genomic distances (,1 Mb) <strong>and</strong> <strong>the</strong> highdegree <strong>of</strong> sequence similarity makes <strong>the</strong> distinction between paralogoussequence variation <strong>and</strong> allelic polymorphism problematic.Fur<strong>the</strong>rmore, <strong>the</strong> fact that such regions frequently harbour intron±exon structures <strong>of</strong> genuine unique sequence will complicate effortsto generate a <strong>genome</strong>-wide SNP map. The data indicate that amodest portion <strong>of</strong> <strong>the</strong> <strong>human</strong> <strong>genome</strong> may be relatively recalcitrantto genomic-based methods for SNP detection. Owing to <strong>the</strong>irrepetitive nature <strong>and</strong> <strong>the</strong>ir location in <strong>the</strong> <strong>genome</strong>, segmentalTable 16 Fraction <strong>of</strong> ®nished sequence in inter- <strong>and</strong> intrachromosomalduplicationsChromosome Intrachromosomal (%) Interchromosomal (%) All (%)1 1.4 0.5 1.92 0.1 0.6 0.73 0.3 1.1 1.14 0.0 1.0 1.05 0.6 0.3 0.96 0.8 0.4 1.17 3.4 1.3 4.18 0.3 0.1 0.39 0.8 2.9 3.710 2.1 0.8 2.911 1.2 2.1 2.312 1.5 0.3 1.813 0.0 0.5 0.514 0.6 0.4 1.015 3.0 6.9 6.916 4.5 2.0 5.817 1.6 0.3 1.818 0.0 0.7 0.719 3.6 0.3 3.820 0.2 0.3 0.521 1.4 1.6 3.022 6.1 2.6 7.5X 1.8 3.2 5.0Y 12.1 16.0 27.4Un 0.0 0.5 0.5Total 2.0 1.5 3.3.............................................................................................................................................................................Excludes duplications with identities .99.5% to avoid artefactual duplication due to incompletemerger in <strong>the</strong> assembly process. Calculation was performed on <strong>the</strong> ®nished sequence available inSeptember 2000 <strong>and</strong> re¯ects <strong>the</strong> duplications found within <strong>the</strong> total amount <strong>of</strong> ®nished sequence<strong>the</strong>n. Note that <strong>the</strong>re is some overlap between <strong>the</strong> interchromosomal <strong>and</strong> intrachromosomal sets.NATURE | VOL 409 | 15 FEBRUARY 2001 | www.nature.com © 2001 Macmillan Magazines Ltd891
articlesaSum <strong>of</strong> aligned bases (kb)c1,200,000Sum <strong>of</strong> aligned bases (kb)900,000800,000700,000600,000500,000400,000300,000200,000100,00001,000,000800,000600,000400,000200,0000Intrachromosomal90 90.5 91 91.5 92 92.5 93 93.5 94 94.5 95 95.5 96 96.5 97 97.5 98 98.5 99Similarity (%)bSum <strong>of</strong> aligned bases (kb)dSum <strong>of</strong> aligned bases (kb)1,600,0001,400,0001,200,0001,000,000800,000600,000400,000200,00004,000,0003,500,0003,000,0002,500,0002,000,0001,500,0001,000,000500,000Interchromosomal90 90.5 91 91.5 92 92.5 93 93.5 94 94.5 95 95.5 96 96.5 97 97.5 98 98.5 99Similarity (%)01 2 3 4 5 6 7 8 9 10–19 20–29 30–39 40–49 50+ 1 2 3 4 5 6 7 8 9 10–19 20–29 30–39 40–49 50+Length <strong>of</strong> alignment (kb)Length <strong>of</strong> alignment (kb)Figure 33 a±d, Sequence properties <strong>of</strong> segmental duplications. Distributions <strong>of</strong> length<strong>and</strong> per cent nucleotide identity for segmental duplications are shown as a function <strong>of</strong> <strong>the</strong>number <strong>of</strong> aligned bp, for <strong>the</strong> subset <strong>of</strong> ®nished <strong>genome</strong> sequence. Intrachromosomal,red; interchromosomal, blue.duplications may well be underestimated by <strong>the</strong> current <strong>analysis</strong>. Anunderst<strong>and</strong>ing <strong>of</strong> <strong>the</strong> biology, pathology <strong>and</strong> evolution <strong>of</strong> <strong>the</strong>seduplications will require specialized efforts within <strong>the</strong>se exceptionalregions <strong>of</strong> <strong>the</strong> <strong>human</strong> <strong>genome</strong>. The presence <strong>and</strong> distribution <strong>of</strong>such segments may provide evolutionary fodder for processes <strong>of</strong>exon shuf¯ing <strong>and</strong> a general increase in protein diversity associatedwith domain accretion. It will be important to consider both<strong>genome</strong>-wide duplication events <strong>and</strong> more restricted punctuatedevents <strong>of</strong> <strong>genome</strong> duplication as forces in <strong>the</strong> evolution <strong>of</strong> vertebrate<strong>genome</strong>s.Table 17 Fraction <strong>of</strong> <strong>the</strong> draft <strong>genome</strong> sequence in inter- <strong>and</strong> intrachromosomalduplicationsChromosome Intrachromosomal (%) Interchromosomal (%) All (%)1 2.1 1.7 3.42 1.6 1.6 2.63 1.8 1.4 2.74 1.5 2.2 3.05 1.0 0.9 1.86 1.5 1.4 2.77 3.6 1.8 4.58 1.2 1.5 2.19 2.1 2.3 3.810 3.3 2.0 4.711 2.7 1.4 3.712 2.1 1.2 2.813 1.7 1.6 3.014 0.6 0.6 1.215 4.1 4.4 6.716 3.4 3.4 5.517 4.4 1.7 5.718 0.9 1.0 1.919 5.4 1.6 6.320 0.8 1.4 2.021 1.9 4.0 4.822 6.8 7.7 11.9X 1.2 1.1 2.2Y 10.9 13.1 20.8NA 2.3 7.8 8.3UL 11.6 20.8 22.2Total 2.3 2.0 3.6.............................................................................................................................................................................Excludes duplications with identities .98% to avoid artefactual duplication due to incompletemerger in <strong>the</strong> assembly process. Calculation was performed on an earlier version <strong>of</strong> <strong>the</strong> draft<strong>genome</strong> sequence based on data available in July 2000 <strong>and</strong> re¯ects <strong>the</strong> duplications found within<strong>the</strong> total amount <strong>of</strong> ®nished sequence <strong>the</strong>n. Note that <strong>the</strong>re is some overlap between <strong>the</strong>interchromosomal <strong>and</strong> intrachromosomal sets.Gene content <strong>of</strong> <strong>the</strong> <strong>human</strong> <strong>genome</strong>Genes (or at least <strong>the</strong>ir coding regions) comprise only a tiny fraction<strong>of</strong> <strong>human</strong> DNA, but <strong>the</strong>y represent <strong>the</strong> major biological function <strong>of</strong><strong>the</strong> <strong>genome</strong> <strong>and</strong> <strong>the</strong> main focus <strong>of</strong> interest by biologists. They arealso <strong>the</strong> most challenging feature to identify in <strong>the</strong> <strong>human</strong> <strong>genome</strong>sequence.The ultimate goal is to compile a complete list <strong>of</strong> all <strong>human</strong> genes<strong>and</strong> <strong>the</strong>ir encoded proteins, to serve as a `periodic table' forbiomedical research 243 . But this is a dif®cult task. In organismswith small <strong>genome</strong>s, it is straightforward to identify most genes by<strong>the</strong> presence <strong>of</strong> long ORFs. In contrast, <strong>human</strong> genes tend to havesmall exons (encoding an average <strong>of</strong> only 50 codons) separated bylong introns (some exceeding 10 kb). This creates a signal-to-noiseproblem, with <strong>the</strong> result that computer programs for direct geneprediction have only limited accuracy. Instead, computationalprediction <strong>of</strong> <strong>human</strong> genes must rely largely on <strong>the</strong> availability <strong>of</strong>cDNA sequences or on sequence conservation with genes <strong>and</strong>proteins from o<strong>the</strong>r organisms. This approach is adequate forstrongly conserved genes (such as histones or ubiquitin), but maybe less sensitive to rapidly evolving genes (including many crucial tospeciation, sex determination <strong>and</strong> fertilization).Here we describe our efforts to recognize both <strong>the</strong> RNA genes <strong>and</strong>protein-coding genes in <strong>the</strong> <strong>human</strong> <strong>genome</strong>. We also study <strong>the</strong>properties <strong>of</strong> <strong>the</strong> predicted <strong>human</strong> protein set, attempting to discernhow <strong>the</strong> <strong>human</strong> proteome differs from those <strong>of</strong> invertebrates such asworm <strong>and</strong> ¯y.Noncoding RNAsAlthough biologists <strong>of</strong>ten speak <strong>of</strong> a tight coupling between `genesTable 18 Cross-species comparison for large, highly homologous segmentalduplicationsPercentage <strong>of</strong> <strong>genome</strong> (%)Fly Worm Human (®nished)*. 1 kb 1.2 4.25 3.25. 5 kb 0.37 1.50 2.86. 10 kb 0.08 0.66 2.52.............................................................................................................................................................................* This is an underestimate <strong>of</strong> <strong>the</strong> total amount <strong>of</strong> segmental duplication in <strong>the</strong> <strong>human</strong> <strong>genome</strong>because it only re¯ects duplication detectable with available ®nished sequence. The proportion <strong>of</strong>segmental duplications <strong>of</strong> . 1 kb is probably about 5% (see text).892 © 2001 Macmillan Magazines Ltd NATURE | VOL 409 | 15 FEBRUARY 2001 | www.nature.com