The DNA sequence of the human X chromosome

The X chromosome sequence

We constructed a map of the X chromosome using predominantly P1-artificial chromosome (PAC) and bacterial artificial chromosome (BAC) clones (Supplementary Table 1), which were assembled into contigs using restriction-enzyme fingerprinting and integrated with earlier maps using sequence-tagged site (STS) content analysis⁵. Gaps were closed by targeted screening of clone libraries in bacteria or yeast, and by assessing BAC and fosmid end-sequence data for evidence of spanning clones. Fourteen euchromatic gaps remain intractable, despite using libraries with a combined 80-fold chromosome coverage. Five of these gaps are within the 2.7 megabase (Mb) pseudoautosomal region at the tip of the chromosome short arm (PAR1). This is reminiscent of the situation in other human sub-telomeric regions⁶, and might reflect cloning difficulties in an area with a high content of (G+C) nucleotides and mini-satellite repeats.

We selected 1,832 clones from the map for shotgun sequencing and directed finishing using established procedures⁷. Finished sequences were estimated to be more than 99.99% accurate by independent assessment⁸. The sequence of the X chromosome has been assembled from the individual clone sequences and comprises 16 contigs. These extend into the telomeric (TTAGGG)n repeat arrays at the ends of the chromosome arms, and include both pseudoautosomal regions (PARs). The data were frozen for the analyses described below, at which point we had determined 150,396,262 base pairs (bp) of sequence (Supplementary Table 2). Subsequently, we obtained a further 609,664 bp of sequence. The 14 euchromatic gaps are estimated to have a combined size of less than 1 Mb (see Methods and Supplementary Table 2), and the sequence therefore covers at least 99.3% of the X chromosome euchromatin. There is also a single heterochromatic gap corresponding to the polymorphic 3.0 (±0.4) Mb array⁹ of alpha satellite DNA at the centromere. On this basis, we conclude that the X chromosome is approximately 155 Mb in length.

The coverage and quality of the finished sequence have been assessed using independent data. All markers from the deCODE genetic map¹⁰ are found in the sequence and the concordance of marker orders is excellent with only one discrepancy. DXS6807 is the most distal Xp marker on the deCODE map (4.39 cM), but in the sequence this marker is proximal to three others with genetic locations of 9-11 cM on the deCODE map. Out of 788 X chromosomal RefSeq¹¹ messenger RNAs that were assessed, 783 were found completely in the sequence, and parts of four others are also present (T. Furey, personal communication). The missing segments of GTPBP6, CRLF2, DHRSX and FGF16 lie within gaps 1, 4, 5 and 10, respectively, and the GAGE3 gene is in gap 7 (Supplementary Table 2). The sequence assembly was assessed using fosmid end-sequence pairs that match the X chromosome sequence. The orientation and separation of end-pairs of more than 17,000 fosmids were consistent with the sequence assembly. In two cases, sequences had been misassembled owing to long and highly similar repeats. There were six instances of large deletions in sequenced clones, which were resolved by determining fosmid sequences through the deleted regions. Finally, there were two cases of apparent length variation between the reference sequence and the DNA used for the fosmid library.

Features of the X chromosome sequence

The annotated sequence of the X chromosome is presented in Supplementary Fig. 1, and updates are contained in the Vertebrate Genome Annotation (VEGA) database (http://vega.sanger.ac.uk/Homo_sapiens/). The distribution of a number of sequence features on the chromosome is shown in Fig. 1. Analysis of the sequence reveals a gene-poor chromosome that is highly enriched in interspersed repeats and has a low (G+C) content (39%) compared with the genome average (41%).

Genes

Based on a manual assessment of all publicly available human expressed sequences and genes from other organisms, we have annotated 1,098 genes (7.1 genes per Mb) across four different categories (see Methods): known genes (699), novel coding sequences (132), novel transcripts (166), and putative transcripts (101). We have also identified 700 pseudogenes in the sequence (4.6 pseudogenes per Mb), of which 644 are classified as processed and 56 as non-processed. The gene density (excluding pseudogenes) on the X chromosome is among the lowest for the chromosomes that have been annotated to date. This might simply reflect a low gene density on the ancestral autosomes. Alternatively, selection may have favoured transposition of particular classes of gene from the X chromosome to the autosomes during mammalian evolution. These could include developmental genes for which the protein products are required in double dose in males (or in females after XCI has occurred), or genes for which mutation in male somatic tissues is lethal.

Physical characteristics of the genes and pseudogenes are summarized in Supplementary Table 3. Exons of the 1,098 genes account for only 1.7% of the X chromosome sequence. On the basis of the lengths of these gene loci, 33% of the chromosome is transcribed. This is considerably below the recent estimates for chromosomes 6 (ref. ¹²), 9 (ref. ⁶), 10 (ref. ¹³) and 13 (ref. ¹⁴), to which the equivalent gene annotation procedure was applied (Supplementary Table 4), and is a reflection not just of low gene density on chromosome X but also of low gene length. For example, mean gene length is 49 kilobases (kb) on chromosome X compared with 57 kb on chromosome 13. Nevertheless, the X chromosome contains the largest known gene in the human genome, the dystrophin (DMD) locus in Xp21.1, which spans 2,220,223 bp. Consistent with its low gene density, the frequency of predicted CpG islands on the X chromosome is only 5.25 per Mb, which is exactly half of the estimated genome average⁷. There is an association with a CpG island for 49% of the known genes, the category for which the most complete gene structures are expected in the current annotation.

We identified evolutionarily conserved regions (ECRs) by comparing the X chromosome sequence to the genomes of mouse, rat, zebrafish and the pufferfishes Tetraodon nigroviridis and Fugu rubripes (Supplementary Table 5). There are 4,493 ECRs that are conserved between the X chromosome and all of the other species. Of these, 4,393 overlap with 4,373 annotated exons. The remaining 100 ECRs are most likely to be unannotated exons, although some could be highly conserved control or structural elements. From these data we conclude that we have annotated at least 97.8% of the protein-coding exons on the X chromosome ([4,373/(4,373+100)] × 100).

Non-coding RNA genes

The gene set described above includes non-coding RNA (ncRNA) genes only when there is supporting evidence of expression from complementary DNA or expressed-sequence-tag (EST) sources. Using a complementary approach, we analysed the X chromosome sequence using the Rfam¹⁵ database of structural RNA alignments, and predicted 173 ncRNA genes and/or pseudogenes (Supplementary Fig. 1 and Supplementary Table 6). These are physically separate from the genes described in the preceding section and are not included in the total gene count, owing to the difficulty in discriminating between genes and pseudogenes for these ncRNA predictions. Using tRNAscan-SE¹⁶, we predicted only two transfer RNA genes on the X chromosome (Supplementary Table 6), out of the several hundred predicted in the human genome⁷. Thirteen microRNAs from the microRNA registry¹⁷ have also been mapped onto the sequence (Supplementary Table 7).

The most prominent of the ncRNA genes on the X chromosome is XIST (X (inactive)-specific transcript)¹⁸, which is critical for XCI. The XIST locus spans 32,103 bp in Xq13, and its untranslated transcript coats and transcriptionally silences one X chromosome in cis. The RefSeq¹¹ transcript of XIST is an RNA of 19,275 bases, which includes the largest exon on the chromosome (exon 1: 11,372 bp). There is also evidence for shorter XIST transcripts generated by alternative splicing, particularly in the 3′ region of the gene¹⁹. In the mouse, Tsix is antisense to Xist²⁰, and its transcript (or the process of its transcription) is believed to repress the accumulation of Xist RNA. There is evidence for transcription antisense to XIST in human^21,22, but we have been unable to annotate the human TSIX gene as there are no corresponding expressed sequences in the public databases, and because there is a lack of primary sequence conservation between the human and mouse regions. In the human sequence, two other ncRNA genes are annotated in the 400 kb region distal to XIST, which are orthologues of the mouse genes described previously as Jpx and Ftx (ref. ²³). In the mouse, Xist, Jpx and Ftx are located within a smaller area of approximately 200 kb²³.

The cancer-testis antigen genes

On assessing the predicted proteome of the X chromosome for Pfam²⁴ domains, our most prominent finding was the presence of the MAGE domain (IPR002190) in 32 genes (Supplementary Table 8). In comparison, only four other MAGE genes are reported in the rest of the genome: MAGEF1 on chromosome 3, and MAGEL2, NDN and NDNL2 on chromosome 15. The MAGE gene products are members of the cancer-testis (CT) antigen group, which are characterized by their expression in a number of cancer types, while their expression in normal tissues is solely or predominantly in testis. This expression profile has led to the suggestion that the CT antigens are potential targets for tumour immunotherapy. A recent report listed 84 CT antigen genes for the human genome²⁵. The X chromosome gene set we describe above contains 99 CT antigen genes and includes novel members of the MAGE, GAGE, SSX, LAGE, CSAGE and NXF families (Supplementary Table 9). Assessment of the most recent RefSeq¹¹ information shows that this set does not include two known MAGE genes (MAGEA5 and MAGEA7) and seven GAGE genes (GAGE3-7, 7B and 8), which are expected to lie in gaps 14 and 7, respectively (Supplementary Table 2). Furthermore, gaps 6 and 9 are also within regions of CT antigen gene duplication. Therefore, we predict that approximately 10% of the genes on the X chromosome are of the CT antigen type.

Conclusive data on the normal functions of the CT antigens, or their involvement in disease conditions, are very limited. However, the remarkable enrichment for CT antigen genes on the X chromosome relative to the rest of the genome might be indicative of a male advantage associated with these genes. Recessive alleles that are beneficial to males are expected to become fixed more rapidly on the X chromosome than on an autosome²⁶. If these alleles are detrimental to females, their expression could become restricted to male tissues as they rise to fixation. Both the concentration of the CT antigen genes on the X chromosome and their expression profiles are consistent with this model of male benefit. The CTantigen genes on the X chromosome are also notable for the expansion of various gene families by duplication. This degree of duplication is perhaps an indication of selection in males for increased copy number. In this context, it is of interest that the MAGE family has independently expanded on the X chromosome in both the human and mouse lineages²⁷.

Repetitive sequences

Interspersed repeats account for 56% of the euchromatic X chromosome sequence, compared with a genome average of 45% (Supplementary Table 10). Within this, the Alu family of short interspersed nuclear elements (SINEs) is below average, in keeping with the gene-poor nature of the chromosome. Conversely, long terminal repeat (LTR) retroposon coverage is above average; but the most remarkable enrichment is for long interspersed nuclear elements (LINEs) of the L1 family, which account for 29% of the X chromosome sequence compared to a genome average of only 17%. The possible significance of this enrichment for XCI is discussed later.

Applying the criterion of at least 90% sequence identity over at least 5 kb (ref. ^​28), we estimate that intrachromosomal segmental duplications account for 2.59% of the X chromosome (Supplementary Table 11 and Supplementary Fig. 2). In contrast, interchromosomal segmental duplications indicated by sequence matches to the autosomes account for a very small fraction (0.24%) of the X chromosome (Supplementary Table 12). Six gaps in the X chromosome map are either flanked by or contained within intrachromosomally-duplicated segments (gaps 2, 3, 6, 7, 9 and 14 in Supplementary Table 2), which might produce instability of clones or otherwise confound mapping progress. The intrachromosomal duplicates are striking in their proximity. Apart from the two segments containing SSX gene copies, which are separated by 4.5 Mb, only six of 229 matches are separated by more than 1 Mb. Among these duplications are well-described cases that are associated with genomic disorders²⁹. In Xp22.32, deletions of the steroid sulphatase (STS) gene, causing X-linked ichthyosis (Online Mendelian Inheritance in Man (OMIM)² entry number 308100), result from recombination between flanking duplications that contain copies of the VCX gene. Also, some instances of Hunter syndrome (OMIM 309900), red-green colour blindness (OMIM 303800), Emery-Dreifuss muscular dystrophy (OMIM 310300), incontinentia pigmenti (OMIM 308300) and haemophilia A (OMIM 306700) result from rearrangements involving duplicated sequences in Xq28. In haemophilia A, mutations are frequently the result of inversions between a sequence in intron 22 of the F8 gene and one of two more distally located copies. A novel finding from our analysis of the X chromosome reference sequence is that the two distal copies are in opposite orientations. Therefore, a large deletion involving F8 and several more distal genes could be an alternative to the inversion rearrangement. A deletion consistent with this prediction has been reported in a family in which carrier females are affected by a high spontaneous-abortion rate in pregnancy³⁰.

The X chromosome centromere

The X chromosome sequence extends from both arms into centromeric, higher-order repeat sequences, which are known to be associated functionally with the X centromere^31-33. The most proximal 494 kb and 360 kb of the Xp and Xq sequences, respectively, consist of extensive regions of satellite DNA, adjacent to euchromatin of the chromosome arms that is exceptionally high in L1 content (Fig. 2). The satellite region on Xp contains small amounts of other satellite families³¹, whereas that on Xq consists entirely of alpha satellite. Similar to all other human chromosome arms that have been examined^33,34, these transition regions consist of monomeric alpha satellite that is not associated with centromere function. Both the Xp and Xq contigs reported here, though, extend more proximally and reach into highly homogeneous, higher-order repeat alpha satellite (DXZ1). Critically, the Xp and Xq contig copies of the DXZ1 repeat are themselves 98-100% identical in sequence, and are oriented in the same direction along the chromosome (Fig. 2). On this basis, the two contigs reach the ‘end’ of each chromosome arm and thus also reach the centromeric locus from either side. This represents a logical endpoint for efforts to complete the sequence of chromosome arms in the human genome, and the first demonstration of this endpoint is provided by the X chromosome sequence.

Single-nucleotide polymorphisms

A total of 153,146 candidate single-nucleotide polymorphisms (SNPs) have been mapped onto the X chromosome sequence and are displayed in the VEGA database. These include 901 SNPs that result in non-synonymous changes in protein-coding regions, and are therefore candidate functional protein variants. The heterozygosity level on the X chromosome is known to be well below that of the autosomes, and this difference can be explained partly or entirely by population genetic factors³⁵. Included in the mapped SNPs are 62,334 that were identified by alignment of flow-sorted X chromosome shotgun sequence reads to the X chromosome reference sequence. Using comparable sequence data for chromosome 20, we calculated that the heterozygosity level on the X chromosome is approximately 57% of that observed for the autosome.

Evolution of the human X chromosome

Males of the three mammalian groups—Eutheria (‘placental’ mammals), Metatheria (marsupials) and Prototheria (egg-laying mammals)—have X and Y sex chromosomes. Ohno proposed in 1967 that the mammalian sex chromosomes evolved from an autosome pair following their recruitment into a chromosomal system for sex determination¹. A barrier to recombination developed between these ‘proto’ sex chromosomes, isolating the sex-determining regions and eventually spreading throughout the two homologues. In the absence of recombination, the accumulation of mutation events subsequently led to the degeneration of the Y chromosome. The sex chromosomes of birds are not homologous to those of the mammals. The sex chromosome system of birds evolved independently during the last 300 Myr, giving rise to homogametic (ZZ) male birds and heterogametic (ZW) female birds, in contrast to the mammalian system of XY males and XX females.

The autosomal origin of the mammalian sex chromosomes is vividly illustrated by alignment of the human X and chicken whole genome sequences (Fig. 3a). Orthologues of some human X chromosome genes were previously mapped to chicken chromosomes 1q13-q21 and 4p11-p14 (ref. ³⁶). Using genomic sequence alignment, we identified approximately 30 regions of homology that together cover most of human Xq and are confined to a single section of approximately 20 Mb at the end of chicken chromosome 4p (Fig. 3a). In contrast, most of the short arm (Xp11.3–pter), including the pseudoautosomal region PAR1, matches a single block of chicken chromosome 1q. No clear picture emerges regarding the origin of the remainder of the short arm (Xcen–p11.3). We were unable to detect large regions of conserved synteny using sequence alignment, and genes from this region have orthologues on several chicken autosomes, including chromosomes 12, 1 and 4 (ref. ³⁷). This region is also characterized by the expansion of several families of CT antigen genes (Fig. 1), which have no readily detectable orthologues in chicken. The present analysis supports the notion of a mammalian ‘X-conserved region’ (XCR)³⁸, which includes the long arm and is descended from the proto-X chromosome. It also supports a separate, large addition (‘X-added region’ or XAR³⁸) to the established X chromosome by translocation from a second autosome, which occurred in the eutherian mammals before their radiation (~105 Myr ago). In contrast to earlier hypotheses, however, it appears that much of the proximal short arm (Xcen–p11.3) should no longer be considered part of an XCR.

The precise location of genes that demarcate the XAR suggests a possible mechanism for the addition. The annotated genes at the extreme ends of the 47 Mb XAR are PLCXD1 (cU136G2.1 in Supplementary Fig. 1) near Xpter, and RGN in Xp11.3. We also found an unprocessed RGN pseudogene (RGN2P) at Xpter, distal to PLCXD1. The orthologues for these three loci are adjacent on chicken chromosome 1, in the order (tel)–RGN–RGN2–PLCXD1–(cen) (Fig. 3b). The generation of these two different gene orders from a common ancestral sequence would require a minimum of two rearrangements as well as the translocation that added the XAR. A more parsimonious model suggested by these data, however, is that the XAR was acquired by recombination between the X chromosome and a ring chromosome in which the ancestral PLCXD1, RGN and RGN2 sequences were neighbours (Fig. 3b).

In order to examine more recent patterns of evolution, we compared the human X chromosome with other mammalian sequences. We saw nine major blocks of sequence homology between human and mouse X chromosomes, and eleven between human and rat (Fig. 4). The homology blocks occupy almost the entirety of each X chromosome, confirming the remarkable degree of conserved synteny of this chromosome within the eutherian mammalian lineage. This is consistent with Ohno’s law, which predicts that the establishment of a dosage compensation mechanism had a stabilizing effect on the gene content of the mammalian X chromosome¹. On the long arm, just two blocks of homology account for the entire alignment of the human and corresponding mouse sequences, but the mouse homologous regions are punctuated with three additional segments, each containing long and very similar repeats (arrowed in Fig. 4). Alignment of human Xq with the rat sequence reveals four discrete homology blocks; the greater fragmentation compared with the mouse alignment would be explained by a minimum of two rearrangements, one in each of the two mouse-human homology blocks, specifically on the rat lineage. The mouse-specific repeat segments are not detected in the current version of the rat genome sequence. On the short arm of the human X chromosome, seven major blocks of homology with each rodent account for most of the human sequence (Fig. 4). Using the dog as an outgroup, we established that the human and dog X chromosome sequences are essentially collinear (K. Lindblad-Toh, personal communication). Therefore, all of the rearrangements indicated in Fig. 4 occurred in the rodent lineage, and the human X chromosome appears to have been remarkably stable in its organization since the radiation of eutherian mammals. This is consistent with the recent prediction, derived from a comparison of human, rodent and chicken chromosomes, that the human X chromosome is identical to the putative ancestral (eutherian) mammalian X chromosome³⁹.

The most notable difference we found between the human and rodent X chromosomes is the existence of 9 Mb of sequence at the tip of the human short arm (including human PAR1) that is apparently missing from the rodent X chromosomes (Fig. 4). There are 34 known and novel protein-coding genes in this segment of the human X chromosome (Supplementary Fig. 1), enabling us to investigate how this difference arose. A comprehensive database search of the rodent genome sequences revealed convincing orthologues for only thirteen of these genes in rat and five in mouse. Most of the rat orthologues are located in two groups on chromosome 12, and the only genes for which X-linked orthologues could be found in both rodents were PRKX and STS. In contrast, we found 24 of these 34 genes on chicken chromosome 1, and the order of these genes is perfectly conserved between the two genomes. Therefore, we conclude that this large terminal segment was present in the XAR and was subsequently removed from the X chromosome in a common murid ancestor of mouse and rat. The relative paucity of rodent ECRs in this segment of the X chromosome sequence (Supplementary Fig. 1) suggests that much of the region may be absent altogether from the genomes of Mus musculus and Rattus norvegicus.

Comparison of the human X and Y chromosomes

The evolutionary process has eradicated most traces of the ancestral relationship between the human X and Y chromosomes. At the cytogenetic level, the Y chromosome has a large and variably sized heterochromatic block and is considerably smaller than the X chromosome, and the euchromatic part of the X chromosome is six times longer than that of Y. Few genes on human chromosome X have an active counterpart on the Y chromosome, and the majority of these are contained in regions where XY homology is of relatively recent origin.

A detailed comparison of the human X and Y chromosome sequences reveals the extent of Y chromosome decay in non-recombining regions. All of the large homologous blocks visible in Fig. 5 (and represented schematically in Fig. 6) are descended from material that was added to the established sex chromosomes. The tip of the short arm of the X and Y chromosomes comprises the 2.7 Mb pseudoautosomal region PAR1. Homology between the X and Y chromosomes in PAR1 is maintained by an obligatory recombination in male meiosis; gene loci in this region are present in two copies in both males and females and are not subject to dosage compensation by XCI. At the tip of the long arm of X and Y is a second pseudoautosomal region, the 330 kb PAR2, which was created by duplication of material from X to Y since the divergence of human and chimpanzee lineages⁴⁰. Some genes in PAR2 are subject to XCI, presumably reflecting their status on the X chromosome before the duplication event. Outside the PARs, homologies between the X and Y chromosomes are in non-recombining regions, predominantly in other parts of the XAR, together with a large ‘X-transposed region’ (XTR)⁴¹ in Xq21.3 and Yp11.2–p11.3 (see below). It is thought that the XAR originally formed a large pseudoautosomal region with an equivalent YAR, which is now largely eroded. At a gross level, the homology between the XAR and YAR is continuous for 6 Mb proximal to the pseudoautosomal boundary on the X (PABX), but is considerably more fragmented on the Y chromosome (Figs 5b and 6). Beyond this, the remaining 38.5 Mb of the XAR detects few other remnants of the YAR. Homologies are mostly in small islands around genes with functional orthologues on both sex chromosomes (for example, AMELX/AMELY, ZFX/ZFY, see Table 1).

The XTR arose by duplication of material from X to Y since the divergence of the human and chimpanzee lineages⁴². The duplicated region spans 3.91 Mb on X, but the corresponding region is only 3.38 Mb on the Y chromosome (Fig. 5c). We have aligned the entire X and Y copies of this region. Excluding insertions and deletions, sequence identity between the copies is 98.78%. We estimate that the transposition event occurred approximately 4.7 Myr ago (Supplementary Discussion 1), which is close to the suggested date of the speciation event that led to humans and chimpanzees, assumed here to be 6 Myr ago. The sequence alignment demonstrates the substantial changes to the XTR on the Y chromosome since the transposition. An inversion is known to have separated a 200-kb section from the rest of the XTR⁴³ (Fig. 5c). Also, the main block of homology is 540 kb shorter on Y than X, owing in particular to the absence of four large regions from the Y chromosome (Fig. 5c). The detection of these sequences at the expected positions on the chimpanzee X chromosome confirms that they were deleted from the Y chromosome after the transposition.

We found that only 54 of the 1,098 genes annotated on the X chromosome have functional homologues on the Y chromosome (Table 1). We obtained direct evidence for 24 genes in PAR1. Twenty-three of them are annotated (Supplementary Fig. 1), and the location of the 5′ end of CRLF2 indicates that the rest of this gene is in gap 4 of the human X sequence (see the VEGA database). On the basis of the excellent conservation of synteny between human PAR1 and the chicken sequence, we infer that a stromal antigen gene (orthologue of chicken Ensembl gene ENSGALG00000016716) lies in gap 1 (see Fig. 3b). As the annotated putative transcript cM56G10.2 might represent the 3′ end of this gene, we conclude that PAR1 contains at least 24 genes. Together with the five annotated genes in PAR2, 29 genes lie entirely within the recombining regions of the sex chromosomes. Additionally, the XG locus spans the boundary between PAR1 and X-specific DNA, but has been disrupted by rearrangement on the Y chromosome.

Outside the XY-recombining regions of the X chromosome, we observed 25 genes that have functional homologues on the Y chromosome (Table 1). Fifteen of these are within the XAR, and a further three genes are shared by the X and Y copies of the XTR. The seven other XY gene pairs are believed to have descended from the proto-sex chromosomes. Only five cases have been described previously^44,45: the X chromosome genes are SOX3, SMCX, RPS4X, RBMX and TSPYL2, which are located on the long arm and proximal short arm (Table 1). The two additional cases we report here involve heat-shock transcription factor genes, designated HSFX1 and HSFX2. They are assigned to the category of XCR genes on the basis of a high degree of divergence from their Y chromosome homologues and their location distal to SOX3 within the XCR. HSFX1 and HSFX2 lie within the separate copies of a palindromic repeat in Xq28 and are identical to each other. By analogy, their Y chromosome homologues (HSFY1 and HSFY2) lie within the arms of a Y chromosome palindrome, the similarity of which is thought to be maintained by gene conversion⁴¹.

On the basis of this and previously published information⁴¹, we can conclude that approximately 15 protein-coding genes on the Y chromosome have no detectable X chromosome homologue.

The progressive loss of XY recombination

The barrier to recombination between the proto-X and Y chromosomes initially encompassed the sex-determining locus on the Y (SRY) and possibly other loci affecting male fitness. It is proposed that rearrangement of the Y chromosome led to the development of this barrier. Thereafter, successive rearrangements that encompassed parts of the pseudoautosomal region resulted in segments of Y-linked DNA that could no longer recombine and consequently degenerated over time. Evidence for the role of Y-specific (as opposed to X-specific) rearrangement in this phenomenon is most clearly illustrated by our analysis of the XAR, which shows very little rearrangement between human and avian lineages (Fig. 3a).

In a previous study⁴⁶, four broad physical and evolutionary regions were defined on the X chromosome. The X chromosome genes within a given region all showed a similar level of divergence from their Y chromosome counterparts. However, between regions, levels of divergence were very different, presumably reflecting the stepwise loss of recombination between the X and Y chromosomes. The physical order of the four regions on the X chromosome was seen to parallel their evolutionary ages, and therefore the chromosome was described as having four “evolutionary strata”⁴⁶. In general, gene pairs were found to be less divergent moving through the strata from Xqter to Xpter. The first two strata (S1 and S2) encompass the long arm and proximal short arm, respectively, and were defined by the genes that survive from the proto-sex chromosomes. Gene pairs were found to be increasingly similar moving through strata 3 and 4, which occupy the proximal and distal sections of the XAR, respectively.

We re-evaluated XY homology in S4 and S3 using finished, genomic sequences from the two chromosomes. For S4 in particular, substantial blocks of homology exist between the chromosomes (blocks 1-10 in Fig. 5b and Fig. 6). Aligning the X and Y chromosome sequences across this region, we observed a bipartite organization, with markedly greater XY identity in the distal 1.0 Mb compared with the proximal 4.5 Mb (Fig. 7a). On this basis, the distal portion containing the GYG2, ARSD, ARSE, ARSF, ADLICAN and PRKX genes can be redefined as a new, fifth stratum, S5 (Figs 1 and 7a). A most parsimonious series of inversions, from the current arrangement of homologous blocks on X to that on Y, is consistent with the proposed strata (Fig. 7b). These data refine the picture of loss of XY recombination during evolution, which occurred by migration of the PABX in a stepwise manner distally through the XAR. The available evidence now suggests that there have been at least four PABX positions within the XAR, which are at the S2/S3, S3/S4 and S4/S5 boundaries (~47 Mb, ~8.5 Mb and ~4Mb from Xpter, respectively), and at the current position (2.7 Mb from Xpter). We estimate that the two most recent PABX movements, which created first S4 and then S5, occurred 38-44 Myr ago and 29-32 Myr ago, respectively (Supplementary Discussion 2).

In addition to the varied degree of XY sequence identity within S3, S4, S5 and PAR1, we found marked differences in their sequence composition, which were presumably also caused by the loss of recombination in each region during evolution. Specifically, we observed that L1, L2 and mammalian interspersed repeat (MIR) coverage decrease with each more distal stratum and PAR1 (Table 2 and Fig. 1), but (G+C) levels and Alu repeat content increase abruptly at the boundary between S4 and S5 (Table 2 and Fig. 8); variations in the incidence of different Alu subfamilies (Y, S and J) also contribute to the distinct character of each stratum and PAR1 (Supplementary Table 13). The compositional differences between S4 and S5 provide additional support for the subdivision of the original stratum 4 (Fig. 8).

X-chromosome inactivation

XCI in mammals achieves dosage compensation between males and females for X-linked gene products. Inactivation of one X chromosome occurs early in female development and is initiated from the X-inactivation centre (XIC). The XIST transcript is expressed initially on both X chromosomes, but later the transcript from the chromosome that is destined for inactivation becomes more stable than the other. Finally, the transcript is expressed only from the inactive X chromosome (X_i). Coating with the XIST transcript is the earliest of many chromatin modifications on X_i.

XCI was first proposed based partly on the study of X:autosome translocations in female mice⁴⁷. Studies of derivative chromosomes containing inactivated X chromosome segments later concluded that the inactivation could spread across the translocation boundary to the autosomal segment, but that inactivation of this segment was incomplete. More recently, it has become clear that more than 15% of the genes on the human X chromosome, including many without functional equivalents on the Y, escape from XCI, as presented in detail elsewhere⁴⁸. The majority of the genes that escape XCI lie within the distal regions of the XAR (Fig. 1): all genes studied in PAR1, S5 and S4 were found to escape from XCI, but there is a lower proportion of escapees in S3, and very few examples in the XCR⁴⁸. This observation correlates with our picture of X chromosome evolution: XCI follows Y chromosome attrition⁴⁹, which is less advanced in the distal strata of the XAR.

Inefficient inactivation of the autosomal segment in X_i:autosome translocations led to the proposal that ‘way stations’ on the X chromosome boost the spread of XCI. According to this model, way stations are present throughout the genome but are enriched on the X chromosome, particularly in the region of the XIC⁵⁰. Lyon suggested that L1 elements are good candidates for acting as way stations on account of their enrichment on the mammalian X chromosome⁵¹. We observe a distribution of L1 elements on the chromosome that is consistent with both the way station and the Lyon hypotheses (Fig. 1 and Table 2). The coverage of L1 repeats is very high in the XCR, especially around the XIC. As noted previously⁵², this enrichment in L1 levels is accounted for particularly by elements that were active more recently in mammalian evolution⁵³ (L1P in Fig. 1). In the XAR, L1 coverage is close to autosome levels, whereas L1 levels are particularly low in the distal evolutionary strata of the XAR, where genes consistently escape inactivation. The XIST locus itself lies in a 60 kb region that is virtually devoid of L1 elements, whereas L1 levels are extremely high in the adjacent regions. Based on their distributions, other interspersed repeats are not strong candidates for way stations. For example, although L2 and MIR elements are reduced in S4, S5 and particularly PAR1 relative to the rest of the chromosome, their overall levels on the X chromosome are not enriched relative to the autosomes but are slightly reduced. Furthermore, L2 and MIR levels are low in the region distal to the XIC. These characteristics do not preclude an involvement in XCI, but are not consistent with a role as way stations.

The possible causal relationship of L1 elements to the spread of XCI remains a subject of debate. Some studies have reported significant associations between L1 coverage and inactivation⁵², and others have refuted this⁵⁴. Our observations on regional differences in composition emphasize that such studies should compare active and inactivated genes (or domains) from the same evolutionary stratum, in order to avoid correlations that are unrelated to XCI.

Medical genetics and the X chromosome sequence

The X chromosome holds a unique place in the history of medical genetics. Ascertainment of X-linked diseases is enhanced by the relative ease of recognizing this mode of inheritance. More important, however, is the fact that a disproportionately large number of disease conditions have been associated with the X chromosome because the phenotypic consequence of a recessive mutation is revealed directly in males for any gene that has no active counterpart on the Y chromosome. Thus, although the X chromosome contains only 4% of all human genes, almost 10% of diseases with a mendelian pattern of inheritance have been assigned to the X chromosome (307 out of 3,199; information obtained from OMIM²). These two aspects of the medical genetics of the X chromosome have greatly stimulated progress in the positional cloning of many genes associated with human disease. To date, the molecular basis for 168 X-linked phenotypes has been determined, and the X chromosome sequence has aided this process for 43 of them, by providing positional candidate genes or a reference sequence for comparison to patient samples (Supplementary Table 14).

Identifying genes involved in rare conditions yields important biological insights. For example, discovery of mutations in the SH2D1A gene⁵⁵ (involved in X-linked lymphoproliferative disease (XLP, OMIM 308240)) led to identification of a new mediator of signal transduction between T and NK cells, and a novel family of proteins involved in the regulation of the immune response. Mental retardation is one of the most common problems in clinical genetics, and affects significantly more males than females. To date, 16 genes from the X chromosome have been associated with cases of non-syndromic X-linked mental retardation (NS-XLMR), in which mental retardation is the only phenotypic feature. These genes encode a range of protein types, and some are also involved in syndromic forms of mental retardation. For example, the ARX gene encodes an aristaless-related homeobox transcription factor and is linked to NS-XLMR cases, as well as to syndromic mental retardation associated with epilepsy (infantile spasm syndrome, ISSX, OMIM 308350) or with dystonic hand movements (Partington syndrome, PRTS, OMIM 309510)⁵⁶. The MECP2 gene, which encodes a methyl-CpG-binding protein, was initially linked to cases of Rett syndrome in girls⁵⁷ (RTT, OMIM 312750) but was later also seen to be mutated in males or females with NS-XLMR⁵⁸. The molecular defect has been determined in only a minority of families affected by NS-XLMR, which has led to speculation that there could be as many as 100 genes on the X chromosome that are associated with NS-XLMR⁵⁹. Discovering the genes for these and other rare, monogenic disorders is of critical value in extending our understanding of fundamental new processes in human biology, and the annotated X chromosome will further facilitate this process.

Concluding remarks

The completion of the X chromosome sequencing project is an essential component of the goal of obtaining a high-quality, annotated human genome sequence for use in studies of gene function, sequence variation, disease and evolution. It also means that for the first time, we now have the finished sex chromosome sequences of an organism. The study of these sequences gives a greater insight into mammalian sex chromosome evolution and its consequences. As these analyses are extended to other genomes, we will gain a greater appreciation of the different evolutionary forces that shape sex chromosome and autosome evolution. It will be important to study differences in the rates of mutational processes, and to consider the influence of the unusual pattern of male recombination on these processes. Clearly, this analysis should not be restricted to a consideration of mammalian sex chromosomes, and it will be of great interest to make comparisons with non-mammalian systems that arose independently in evolution.

Supplementary Material

Supplementary Information accompanies the paper on www.nature.com/nature.