We present here a draft genome sequence of the red jungle fowl, Gallus gallus. Because the chicken is a modern descendant of the dinosaurs and the first non-mammalian amniote to have its genome sequenced, the draft sequence of its genome—composed of approximately one billion base pairs of sequence and an estimated 20,000–23,000 genes—provides a new perspective on vertebrate genome evolution, while also improving the annotation of mammalian genomes. For example, the evolutionary distance between chicken and human provides high specificity in detecting functional elements, both non-coding and coding. Notably, many conserved non-coding sequences are far from genes and cannot be assigned to defined functional classes. In coding regions the evolutionary dynamics of protein domains and orthologous groups illustrate processes that distinguish the lineages leading to birds and mammals. The distinctive properties of avian microchromosomes, together with the inferred patterns of conserved synteny, provide additional insights into vertebrate chromosome architecture.
Genome sequence comparison is a modern extension of the long-standing use of other species as models to illuminate aspects of human biology and medicine. Large-scale genome analyses also highlight the evolutionary dynamics of selective and mutational processes at different chronological scales1,2,3,4. We present here results obtained from an extensive analysis of a draft sequence of the chicken genome, which has evolved separately from mammalian genomes for ∼310 million years (Myr)4,5 (Fig. 1). This genome is the first to be sequenced at this particular evolutionary distance from humans, and, as shown previously6,7,8, it provides an excellent signal-to-noise ratio for the detection of functional elements. Our analysis of the 6.6 × coverage draft sequence of the chicken genome resulted in the following main observations.
• The nearly threefold difference in size between the chicken and mammalian genomes reflects a substantial reduction in interspersed repeat content, pseudogenes and segmental duplications within the chicken genome.
• Chicken–human aligned segments tend to occur in long blocks of conserved synteny. We find a relatively low rate of chromosome translocations in both lineages from the last common ancestor, whereas intrachromosomal rearrangements (for example, inversions) are more common.
• Syntenic relationships for certain classes of non-coding RNA (ncRNA) genes differ from those of protein-coding genes. This observation implies a novel mode of evolution for some ncRNA genes.
• Expansion and contraction of multigene families seem to have been major factors in the independent evolution of mammals and birds.
• The sizes of chicken chromosomes, which span a range of nearly two orders of magnitude, correlate negatively with recombination rate, G+C and CpG content, and gene density but positively with repeat density.
• Synonymous substitution rates are elevated for genes in both chicken microchromosomes and in subtelomeric regions of macrochromosomes.
• There is a paucity of retroposed pseudogenes in the chicken genome, in contrast to mammalian genomes, greatly simplifying the classification of chicken gene content. This is explained by the high specificity of the reverse transcriptase from the predominant interspersed repeat element in the chicken genome: the CR1 long interspersed nucleotide element (LINE).
• Unlike all other vertebrate genomes studied so far, no short interspersed nucleotide elements (SINEs) have been active in the chicken genome for the last ∼50 Myr.
• Alignment of the chicken and human genomes identifies at least 70 megabases (Mb) of sequence that is highly likely to be functional in both species.
• Many of the chicken–human aligned, non-coding sequences occur far from genes, frequently in clusters that seem to be under selection for functions that are not yet understood.
Perspectives on the domestic chicken
The chicken (Gallus gallus) is an important model organism that bridges the evolutionary gap between mammals and other vertebrates and serves as the main laboratory model for the ∼9,600 extant avian species. The chicken also represents the first agricultural animal to have its genome sequenced. Modern birds (Ornithurae) evolved from therapod dinosaurs9,10 in the middle of the Mesozoic era (Fig. 1). Chickens were domesticated in Asia at least by 5400 bc, perhaps as early as 8000 bc11,12,13. Darwin14 suggested that the red jungle fowl was the nearest ancestor to the domestic chicken, a view later confirmed by mitochondrial DNA analysis15.
Genetic analysis of the chicken dates back to the start of the twentieth century16,17, and hundreds of well-characterized mutant stocks and inbred lines have been developed18. The chicken embryo has been an especially useful vertebrate system for developmental biologists19 owing to experimental advantages of in ovo embryogenesis. Furthermore, the chicken has been used in seminal studies in virology, oncogenesis and immunology20,21,22. The chicken genetic linkage map, initiated early in the last century23, now includes 2,172 genetic loci with a total length near 4,000 cM24,25. Most avian karyotypes contain chromosomes of markedly different lengths, termed the macro- and microchromosomes, and thus bird karyotypes are quite distinctive as compared with those of mammals26. The chicken karyotype (2n = 78) is made up of 38 autosomes and one pair of sex chromosomes, with the female as the heterogametic sex (ZW female, ZZ male).
Sequencing and assembly
All sequencing libraries were prepared from DNA of a single female of the inbred line of red jungle fowl (UCD 001) to minimize heterozygosity and provide sequence for both the Z and W sex chromosomes, albeit at 50% of the autosomal coverage. The assembly was generated from ∼6.6 × coverage in whole-genome shotgun reads, a combination of plasmid, fosmid and bacterial artificial chromosome (BAC)-end read pairs (Supplementary Table S1). The assembly (Table 1) was generated using PCAP27, a parallel algorithm that exploits both read-pairing constraint information and base quality values (see Supplementary Information for a description of the methods).
A BAC-based physical map for the chicken was developed in parallel with the sequence assembly28. Along with the genetic map25,29,30,31, this provides the main scaffolding for the assembly into larger ordered and oriented groupings (‘ultracontigs’) as well as the mechanism for chromosomal assignment (see Methods). After integrating data from the physical map with the assembly, several additional steps were taken to improve the initial assembly of chicken chromosome sequences. This included using expressed sequence tag (EST) and messenger RNA data to aid the ordering and orientation of sequence, and using map and sequence data to aid in localization of centromeres and telomeres (see Methods). The resulting assembly consists of 574 segments made up of 84 ultracontigs (ordered and oriented by their relationship to the physical map) and 490 ‘supercontigs’ (ordered and oriented by read-pairing data, but not linked to the physical map) anchored to chicken chromosomes. Of the 1.05 gigabases (Gb) of assembled sequence, 933 Mb were localized to specific chromosomes, 907 Mb of which were ordered and oriented along those chromosomes.
Assessment of the coverage and quality of the genome assembly
We estimated the coverage of the assembly using both finished BACs and available mRNA sequences (Methods). In a set of 38 finished autosomal BAC sequences from the same red jungle fowl female (covering 6 Mb of sequence), 98% of finished bases could be aligned with the draft whole-genome shotgun assembly, with an overall substitution rate of 0.02% and no deletions or insertions. Similarly, of a set of 23,212 chicken mRNAs and 485,000 ESTs32, 97% and 96% respectively are at least partially found in the assembly. Of these, 10% are only partially found or are fragmentary. This lack of contiguity contributes in part to the 5–10% of genes estimated to be absent from the Ensembl chicken gene set (see below). Representation of the (G+C)-rich extremes of the genome may be less complete. In one small region of incompletely sequenced BACs (3.6 Mb) orthologous to human chromosome 19 (HSA19) (I. Ovcharenko et al., unpublished data), where the average G+C content was 52% (with some regions exceeding 60%), coverage fell to 82%. Furthermore, we examined a set of 400 genes that were represented in chicken mRNA or ESTs and had single orthologues in five diverse species (human, mouse, rat, Takifugu rubripes (Fugu) and Drosophila) but were predicted to be absent from, or at least partially truncated in, the chicken Ensembl gene set (see below). Over 70% were in fact partially found in the assembly. Of the 400, the largest fraction missing (21%) were HSA19 orthologues from a region known to be unusually rare in chicken clone libraries (L. Gordon et al., unpublished data). The missing genes have a higher G+C content than average and many, including some HSA19 orthologues, are associated with intronic simple sequence repeats (see Methods).
Comparisons to 6 Mb of finished red jungle fowl BAC clone sequence revealed alignment with 311 chicken contigs from 62 supercontigs, which were used to assess possible ordering errors (see Methods). No orientation problems were detected, and only two order discordances (misordered sequence contigs within a supercontig) were discovered. This would extrapolate to a total of ∼400 kilobases (kb) of misordered contigs in the current assembly. In addition, eight cases were found in which a contig was incorrectly inserted into a supercontig, equivalent to ∼1 Mb of incorrect insertions in the full genome.
Recent duplications are especially difficult to place within whole genome assemblies. Relaxed assembly may collapse duplicated segments into one, and stringent assembly may break sequences into duplicates because of sequencing errors. In the chicken, ‘all-versus-all’ comparison shows that 11% (∼ 123 Mb) of the genome sequence is in pairwise alignments larger than 1 kb with more than 90% sequence identity. The bulk (91% of the 11%), however, are highly similar (> 98%) and might represent false duplication. In a direct test (see Methods), only 22% of duplications with near-perfect sequence identity (> 98%) and 26% (32.3 out of 122.7 Mb) of the full set were confirmed.
Because the assembly process incorporated genetic markers (Supplementary Table S2), the genetic map does not provide independent assessment. However, recent placement of 142 additional genetic markers, mapped after the assembly, suggests that less than 0.75% (∼ 6 Mb) of the sequence has been assigned to a wrong chromosome. Thus, the assembly correctly places the vast majority of the chicken genome in long contiguous stretches. It is well ordered and oriented and faithfully represents older segmental duplications (at the cost of a modest false increase in the most recent duplications). The draft provides an excellent substrate for initial global analysis, recognizing that the elucidation of the full sequence will be critical to allow final, definitive conclusions.
Gene content of the chicken genome
The genome sequence of an organism encodes both ncRNAs and proteins. Extensive analysis of the genome sequences of human1, mouse2 and rat3 has provided our current best assessment of mammalian gene content and has illuminated much about the evolution of genes. The chicken genome provides new perspectives on both the structure and content of mammalian genes, as well as yielding insight into avian gene content and evolution of ncRNA genes.
Non-coding RNA genes
A total of 571 ncRNA genes, from over 20 distinct gene families, were identified within the chicken genome assembly (Table 2) using bioinformatic approaches33,34 (see Methods). Predicted ncRNA pseudogenes are greatly reduced in number relative to their human ncRNA counterparts. The chicken ncRNA predictions therefore represent a set that is mainly functional. If ncRNA genes maintain their placement with respect to neighbouring genes, chicken ncRNA gene locations could be used to identify which mammalian copies are likely to be functional and which are probable pseudogenes. However, few chicken and human ncRNA genes are paired in regions of conserved synteny (Table 2), relative to the high level of shared gene order observed for protein-coding genes (see below). Those classes of ncRNAs that are most often syntenic are microRNAs (miRNAs) and small nucleolar RNAs (snoRNAs), which are often found in the introns of protein-coding genes (or, rarely, of specialized ‘host’ genes35). Most ncRNA genes thus seem to have been translocated to distant genomic sites during vertebrate evolution, without accumulating large numbers of pseudogenes, as would be expected were this process to occur via retrotransposition. This is also in contrast to duplication of genes via unequal crossing over, which results in tandem copies. These insights will require considerably more analysis for a definitive resolution, but it seems that these ncRNAs may not use the same duplication and/or translocation mechanisms as protein-coding genes.
Development of a protein-coding gene set
An evidence-based system (Ensembl36) and two comparative gene prediction methods (Twinscan37 and SGP-2 (ref. 38)) together predicted a common set of 106,749 protein-coding exons, with 85,929 additional exons predicted by one or two methods (Supplementary Table S3). Particular attention was paid to the identification of selenoproteins, which are usually mispredicted in annotated genomes because of their usage of the TGA codon, usually a stop codon, to code for the amino acid selenocysteine (see Methods). Of the human genes predicted using chicken as the “informant”, only 311 genes predicted by SGP-2 are absent from previously identified sets (namely, Vega40, Ensembl41, RefSeq42, MGC43 and H-Invitational44) and have homologous chicken predictions that possess orthologous intron positions. These data, and those of another study (E. Eyras et al., unpublished data), suggest that most of the protein-coding genes conserved among vertebrates are represented in existing complementary DNA sets.
We tested the sensitivity and specificity of the chicken gene predictions. Sensitivity was assessed by comparing predicted exons to those of chicken cDNAs32 representing long open reading frame (ORF)-containing protein-coding genes (Table 3). All three methods correctly predicted about 80% of cDNA-based exons with >80% coverage. An independent SAGE-based analysis (ref. 166, and M. B. Wahl et al., unpublished data) provided a similar, although marginally lower, estimate. Specificity was assessed by testing random exon pairs from the prediction sets using polymerase chain reaction with reverse transcription (RT–PCR) (E. Eyras et al., unpublished data, and ref. 44). Briefly, Ensembl predictions have a false positive rate of ∼4%. When an exon pair is predicted by any two of the three methods (predominantly joint Twinscan plus SGP-2 exons) ∼50% are confirmed, suggesting that some genes are missing from the Ensembl set, but we cannot reliably distinguish these from a similarly large number of Twinscan plus SGP-2 false positives. Using our estimates of specificity and sensitivity, we predict a total of between 20,000 and 23,000 protein-coding genes in chicken, with 80–90% of these found in the present Ensembl set (see Methods). This estimate overlaps the lower bounds in the corresponding ranges for mammalian genomes determined by similar calculations (for example, see refs 2, 3, 45).
Evolutionary conservation of gene components
Alignments of chicken and human orthologous protein-coding genes demonstrate the expected pattern of sequence conservation, with highest identity in protein-coding exons and minimal identity in introns (Fig. 2). These alignments allowed us to examine sequence conservation at different sites within genes.
Alignments of coding regions often did not extend to the previously annotated human protein start codons. Rather, we observed a fourfold increase in the frequency of methionine at the first position of the alignment (Fig. 3), suggesting that these internal ATG codons could be the true start sites for at least some of ∼2,000 human genes. For these proteins, the overall distribution of amino acids upstream of the end of the alignment in human was markedly different from that downstream and was more consistent with a codon distribution derived from non-coding nucleotide sequence. Using this comparative signal and other features, such as the Kozak sequence46, we can potentially improve the annotation of mammalian protein-coding start sites.
Sequence conservation around mammalian splice sites can be predicted by divergence at unselected (non-consensus) base pairs at the neutral rate, coupled with purifying selection on sites matching the splice site consensus47. Given the high level of neutral site divergence that has occurred between mammalian and chicken orthologous sequences (see neutral evolutionary rate, below), one would expect that orthologous mammalian–chicken splice sites should show a level of conservation no different from that of any unrelated pair of splice sites. However, in contrast to analogous comparisons within the mammalian lineage, there is a detectable signal in orthologous splice site comparisons beyond the consensus derived from comparing non-orthologous splice sites (Supplementary Fig. S1). This suggests that either some subtle classes of splice site sequences are conserved beyond the generic consensus that can only be observed at the bird–mammal evolutionary distance, or that there is a significant but weak conservation in mammalian introns that is not detectable in mammalian–bird alignments48.
To explore the role of conserved non-coding sequence segments that are probably regulators of protein-coding genes, we examined the frequency of non-coding alignments of at least 100 base pairs (bp) in, respectively, the 5′ flanking region, 5′ untranslated region (UTR), at least one intron, 3′ UTR, or 3′ flanking region (see Methods) within human–chicken orthologue pairs in relation to gene function (as determined by gene ontology (GO) category, Table 4). Some GO categories (for example, development and transcriptional regulation) showed enrichment for conservation in all five regions, suggesting that conserved regulatory signals exist within all of these locations. However, other categories showed more specific patterns. As one example, introns of ion channel genes are particularly enriched for conserved sequences, in agreement with reports that such introns contain RNA-editing targets49,50.
Pseudogenes and retroposed copies in the chicken genome
Only 51 duplicates of protein-coding genes probably formed by retroposition (that is, exhibiting loss of introns)51 were identified in the chicken genome, in contrast to the more than 15,000 cases observed in mammalian genomes3,52. In mammals, the ancient LINE1 (L1) transposable element is responsible for the origin of most if not all retroposed (pseudo) genes53. Although birds host their own LINE-like elements (chicken repeat 1 (CR1); see below)54, the reverse transcriptase encoded by these elements is unlikely to copy polyadenylated mRNAs55, probably explaining the paucity of processed pseudogenes in chicken. Within the set of 51 (Supplementary Table S4), 36 clearly represent pseudogenes, because their former coding regions are disabled by alterations (including frameshifts and premature stop codons) that preclude protein function. Among the remaining 15 elements, eight show strong evidence for selective constraint (Supplementary Table S4) and therefore may represent functional retroposed genes. We found no clear bias towards either particular gene families or chromosomal locations for the retrocopies (Supplementary Table S4).
Interspersed repeat content of the chicken genome
Interspersed repeats are predominantly copies of transposable elements in various stages of decay. Less than 9% of the chicken genome could be classified as such, and only 11% was annotated when satellites or any lower-copy-number repetitive DNA segments were included (Table 5; see also Methods). This is markedly lower than the 40–50% interspersed repeat density observed in mammalian genomes1,2, and leaves, with coding regions comprising another 4%, over 85% (∼ 900 Mb) of the current assembly unexplained. This large amount of unexplained genetic matter—which could primarily constitute ancient transposable elements that have mutated beyond recognition—and the high divergence (age) of many recognized repeats (Supplementary Fig. S2) suggest that the low interspersed repeat density in chicken is due to low (recent) transposable element activity rather than to a high deletion rate.
Most of the interspersed repeats are CR1 LINE copies
A single type of non-LTR (long terminal repeat) retrotransposon or LINE, CR1 (refs 54, 56, 57), comprises over 80% of all interspersed repeats in the chicken genome (200,000 copies). CR1 resembles the mammalian L1 element in having a (G+C)-rich internal promoter region, followed by two ORFs. A full-length CR1 element is 4.5 kb, but all but 0.6% of the CR1 copies are truncated from their 5′ end. As sequences near the 5′ end are needed for retrotransposition, the success of CR1 in the bird genome seems remarkable. Comparison of the length distribution of CR1 and L1 copies (Supplementary Fig. S3) suggests a higher efficiency of the L1 reverse transcriptase or higher stability of the L1 transcript55. CR1-like elements, unlike L1 elements, do not create target site duplications, probably explaining the absence of copies with 5′ inversions so common for L1 (Supplementary Fig. S3). It is unclear whether CR1 is currently active in the chicken. We found only one full-length element with intact ORFs in the chicken genome: a copy at chromosome 6 (GGA6; 661,970–666,111) that is >2% divergent from the CR1-F consensus or any other element. However, long, minimally divergent CR1 copies are often interrupted by sequence gaps in the present assembly, and a still-active CR1 source gene could have been collapsed and thus missed.
We reconstructed 11 complete CR1 source genes from the copies in the chicken assembly, although the abundance of 3′ end fragments allowed further division into 22 subfamilies. Figure 4 shows the phylogenetic relationship of the 11 source genes based on the alignment of their ORF2 product. The evolution of CR1 in the bird genome seems to differ from that of L1 in mammals, in that several widely divergent elements have been active in parallel, whereas in mammals a single lineage of L1 has been dominant58,59. The most recently active CR1 elements in chicken (CR1-F and CR1-B) are less than 70% identical over their ORF2 coding region, whereas the human and mouse L1 ORF2 products are 78% identical. This high divergence between subfamilies probably led to earlier estimates of only 30,000 (ref. 57) and 100,000 (ref. 55) copies, and the current estimate of 200,000 copies is also probably low. The location of the turtle CR1 element in the ORF2-based phylogenetic tree suggests that the main branches of chicken CR1 elements may predate the turtle–bird speciation. Otherwise the tree follows species phylogeny, suggesting that CR1 elements are ancient, vertically transmitted inhabitants of vertebrate genomes.
The most consequential difference between CR1 and L1 elements is in their 3′ end structures: the L1 3′ UTR has evolved with seemingly little constraint on the primary sequence except the polyadenylated tail59, whereas the CR1 3′ UTR is remarkably conserved between all derived subfamilies55 and ends with a (ATTCTRTG)n microsatellite in all chicken CR1 subfamilies, as well as in the turtle CR1 and the ancient L3 element. The CR1 reverse transcriptase presumably has high substrate specificity, whereas the L1 proteins are known to be highly promiscuous. All polyadenylated transcripts in mammals are potential substrates for the L1 reverse transcriptase, and mammalian genomes are littered with processed pseudogenes and SINEs that have been retroposed by the L1 machinery. As noted above, processed pseudogenes are rare in the chicken genome.
SINEs are small, non-autonomous retroposons derived from structural RNAs; they contain an internal polymerase III promoter and, generally, a 3′ end derived from a LINE-like element in the same genome, which when transcribed is recognized by the LINE machinery for transposition. In all vertebrate and most other animal genomes studied thus far, at least one recently or currently active SINE family has been found, mostly derived from a transfer RNA and often constituting the most numerous interspersed repeat. It is therefore remarkable that the abundant chicken CR1 element does not seem to be associated with a single SINE, especially as the closely related CR1 element in turtles60 and L3 in ancestral mammals are or were paired with SINEs. There are about 10,000 faint matches in the chicken genome to MIR and MIR3 (the SINEs associated with L2 and L3, respectively), which originated before the mammal–bird speciation, but unlike in mammals, no new SINEs seem to have replaced these ancient elements upon their extinction.
As in mammalian genomes, all retrotransposons with LTRs in the chicken genome belong to the vertebrate-specific class of retroviruses, whereas no copies of gypsy or copia retrotransposons were found. We reconstructed 14 internal and 41 LTR sequences for endogenous retroviruses (ERVs) or their non-autonomous companions, with representatives of all three recognized subclasses (class I to III). The assembly contains one of the two copies of the avian sarcoma/leukosis virus provirus known to exist in the sequenced genome (see Methods). The class II chicken GGERVK10 and class III GGERVL endogenous retroviruses may still be active, as their copies in the chicken genome are less than 3% diverged. We reconstructed several subfamilies of the most abundant ERV, GGERVL, which is most closely related to the mammalian ERV-L61. Considering the long descent of ERVL subfamilies in mammals62, it is tempting to propose that GGERVL and its mammalian relatives have been endogenous in both lineages since the mammalian–bird split. However, all GGERVL subfamilies are young (between 0% and 13% divergence), and the nature of retroviruses suggests that independent introduction in the germ line is a valid possibility as well.
Only two ancient DNA transposon families were found in the chicken genome: the unrelated activator-like Charlie12_GG and mariner-like GGMAR elements. Thus, like that of mammals1,63, the bird germ line may be protected from infiltration by DNA transposons. Both Charlie12_GG and GGMAR left copies that are now ∼16% diverged from the consensus, and appear to have been active contemporaneously. This is apparent from a hybrid interspersed repeat consisting of a GGMAR copy within a Charlie12_GG element, from which we infer that the mariner-like element must have transposed into the activator-like element when the latter was still active.
Protein content evolution of chicken and mammalian genomes
Conservation of vertebrate domain and protein content
About 60% of chicken protein-coding genes have a single human orthologue (Fig. 5); for the remainder, orthology relationships are more complex or are not detectable. Chicken and human 1:1 orthologue pairs exhibit lower sequence conservation (median amino acid identity of 75.3%) than rodent and human 1:1 orthologue pairs3 (∼ 88%), as expected (Fig. 6a). Orthologous sequences involved in cytoplasmic and nuclear functions are more conserved than those implicated in reproduction, host defence and adaptation to the environment (Fig. 6b). Sequence conservation of expressed chicken genes, in comparison to their human orthologues, is non-uniformly distributed among chicken tissues. Sequences expressed in the chicken brain, as indicated by EST data, are more conserved than testis-expressed sequences (Fig. 6c). Moreover, genes with ESTs from few (< 4) tissue types are significantly (P < 0.001) less conserved than sequences with ESTs from many (> 6) tissues (median identities of 68.3% and 76.0%, respectively). The latter finding mirrors previous results for mammalian genes64,65,66.
Genes that are conserved between human and chicken are often also conserved in fish: 72% (7,606) of chicken–human 1:1 orthologues also possess a single orthologue in the T. rubripes genome67. These genes represent a conserved core that is likely to be present in most vertebrates (Fig. 5). Their sequences are more conserved than other chicken–human 1:1 orthologues (Fig. 6a), indicating that they have been subjected to a greater degree of purifying selection. Chicken, human and pufferfish genes also encode a similar set of protein domains. Of 1,085 Pfam and SMART domain families in InterPro68,69,70 that are encoded in at least three chicken genes, only two families are unrepresented among human genes, and a further 21 families are absent from Fugu genes.
Using a manual approach combining synteny information and highly sensitive search parameters, we were able to identify chicken orthologues of several immune-related genes previously thought to be specific to mammals. Owing to high sequence divergence (Fig. 6b), these were not predicted during the automated gene build process. These include the antimicrobial protein cathelicidin, colony-stimulating factor, leukaemia inhibitory factor, interleukin-3 (IL-3), -4, -7, -9, -13 and -26, and three secretoglobins. IL-26 was previously known only in humans (it is a pseudogene in mouse and rat71), thus chicken represents the only available model organism with which to investigate IL-26 function.
Gene and family differences
We next sought to identify those protein and domain families that are over- or under-represented in chicken, compared with mammals (Table 6 and Fig. 7). However, this type of analysis can be complicated by artefacts inevitably present in a draft genome project such as those introduced by gaps in sequencing coverage, assembly, or gene prediction. In chicken, although many are present in partial, fragmented form within the genome assembly, we estimate that roughly 5–10% of genes are substantially truncated in or missing from the Ensembl gene set (see Methods). Nonetheless, selecting our approaches with this in mind there are many insights to be gained, particularly in those areas not affected by these issues such as gene losses in human and gene innovations and expansions in chicken. Here, we document gene innovations, losses and expansions in chickens and mammals, and the evolution of function in orthologues.
Gene innovations in chicken. Expansion of and innovation within gene families in different lineages are often correlated with divergent phenotypes. For example, scales, claws and feathers of birds are formed using an avian-specific family of keratins, whereas hair fibre formation in mammals involves a distinct keratin family, which has greatly expanded within this lineage (Fig. 7). Of the ∼150 avian keratins identified in the chicken genome, 30 are found in tandem arrays on GGA27. We also considered proteins that are specific to the eggshell, such as ovocleidin 116, for which homologues were not previously known outside of birds. However, with further analysis, we predict that ovocleidin 116 has mammalian orthologues, namely primate matrix extracellular phosphoglycoprotein (Blastp, E = 2 × 10-4), indicating that, despite low sequence identity (35% over 80 amino acids), both avian and mammalian genes perform similar roles in calcification. Gene innovation may also occur through domain accretion1. These events are extremely rare among mammalian genes72, and we succeeded in detecting only a single instance within birds for a gene (ENSGALG00000000805) that encodes both scavenger receptor cysteine-rich (SRCR) and fibrinogen-related domains. A single exon, encoding the SRCR domain, appears to have been inserted into this gene (Supplementary Fig. S4) relatively recently, because comparison to its paralogue (ENSGALG00000000732) shows considerably fewer synonymous substitutions (KS value of 0.034) than do the majority of chicken–human orthologue pairs (median KS value of 1.66).
Genes absent from chicken. Genes encoding vomeronasal receptors, casein milk proteins, salivary-associated proteins (statherin and histatins) and enamel proteins seem to be absent from the chicken: from within both the EST sets and the genome. This is unlikely to result from imperfections in the chicken genome assembly because it preserves orthologues of closely linked (syntenic) mammalian genes. These absences might therefore mirror the evolution of the vomeronasal organ and mammary glands in mammals, and the loss of teeth in birds. The presence in fish of enamel-associated genes and their absence in chicken, together with absence of chicken casein and salivary-associated genes, which all cluster together in mammalian genomes, is consistent with a previous suggestion73 that these have all descended, by gene duplication and rapid sequence diversification, from a common ancestor, possibly an enamel-associated gene.
Loss of an entire domain family in the protein repertoire is unusual. Nonetheless, the SCAN domain family seems to be unrepresented in the chicken gene set and genome. SCAN domains are dimerization motifs that are found, often with zinc finger domains, in more than 60 human proteins74,75. In Fugu and zebrafish, a total of three SCAN domains are found, not associated with zinc fingers but instead within large retrovirus-like proteins. This phyletic distribution indicates a possible acquisition of new function in the synapsid or mammalian lineages, whereas in the avian lineage the domain might have remained associated with retroviral sequences that died out subsequently.
Overall, we find that among the predicted chicken genes there is a notable under-representation of genes that are widely conserved and were presumably present in the mammal–bird common ancestor. We estimated gene innovation and loss by considering patterns of the presence or absence of orthologous genes for nine metazoan genomes, and by reconstructing the most parsimonious gene contents of ancestral root species, using the plant Arabidopsis thaliana as an outgroup (Fig. 8; see also Methods). Despite uncertainties stemming from the incompleteness of current genome sequences, all metazoan lineages seem to be gaining orthologous core genes, with the notable exception of the ancestor of the Diptera76,77. Among the vertebrates, the chicken seems to be the only species thus far sequenced to have lost more of those genes than it has gained over an extended period of time. At present, it is difficult to estimate to what extent this observation is influenced by gaps in the assembly. Indeed, 57% of the apparent losses that are otherwise conserved as single copy genes in mammals and insects are represented in chicken EST libraries (reciprocal best match, bitscore ≥ 200). However, this coverage is lower than the average for this gene class (88%), suggesting that not all of the absences can be attributed to assembly gaps. There are a number of potential explanations for these data. The most straightforward is that some of these core genes were deleted within the avian lineage. An alternative is that they have experienced accelerated evolution. Targeted investigations will be required in order to resolve whether birds have lost unusually large numbers of genes that are otherwise conserved among metazoa.
Expansion of domains and gene families in chicken. Three domain families are significantly over-represented in chicken with respect to human (Fig. 7). The most notable case is an apparent 40% expansion of SRCR domains. Many of these domain sequences are similar to those of human DMBT1, a proposed regulator of mucosal homeostasis78. This suggests a link between the abundance of SRCR domains in chicken and its adaptive requirements in mucosal defence.
To gain a more in-depth view of family expansions in chicken we identified, from the set of automatically derived orthologues, nine chicken gene families containing at least fourfold more representatives than in human (Fig. 7). Most of these families seem to have roles in immunity and host defence. Family sizes were further refined by re-predicting genes and pseudogenes from chicken and human assemblies using Exonerate (G. Slater and E. Birney, unpublished software) and Genewise79 (Table 7). The most marked expansion involves at least 218 non-identical chicken genes that are predicted to be orthologous to one of two olfactory receptor genes (OR5U1 and OR5BF1). In humans, OR5U1 and OR5BF1 genes lie within olfactory receptor gene clusters that each are positioned next to paralogous major histocompatibility complex (MHC) class I gene clusters80. These olfactory receptors have been tentatively proposed to be involved in odorant-mediated detection of MHC diversity81. Duplication of and/or gene conversion within OR5U1-like and OR5BF1-like genes in chicken appear to have been both extensive and relatively recent in this lineage (Supplementary Fig. S5).
The OR5U1/OR5BF1-like genes contribute the majority of at least 283 olfactory receptor genes in the current chicken assembly, a similar number to that found for humans82. The large size of the olfactory receptor family seems to run counter to the textbook view (for example, see ref. 83) that birds have a poor sense of smell84. Individual chicken researchers, however, had already suspected that chickens are not particularly anosmic (for example, see ref. 84). On five chicken chromosomes, these olfactory receptor genes are present in subtelomeric clusters, which, by analogy with human olfactory receptor clusters85, may be associated with their rapid evolution86. As with these genes in mammals, chicken olfactory receptor genes are interspersed with many (∼ 100) homologous pseudogenes.
Gene innovations in mammals. The largest mammal-specific gene family is the high/ultrahigh sulphur hair keratins (also named keratin-associated proteins). Thirty-seven copies have been reported on HSA17q12-21 (ref. 87) embedded in the larger type I keratin cluster, which is present also in other mammals88. Another large orthologous family with multiple members in human, but none in chicken, is a subgroup of the α-interferons (Fig. 7). The expansion and diversification of interferons into various subfamilies is thought to be a mammalian innovation in response to different pathogen challenges89,90. Some highly diverged interferon homologues have been found outside mammals (for example, see ref. 89), and the distant relatives in chicken and Fugu appear to have duplicated independently from the mammalian radiation.
Gene losses in mammals. The chicken genome enables increased precision in dating gene losses in the lineage leading to humans: we can identify losses that happened within the synapsid lineage, but before the mammalian radiation (an interval of ∼110–310 Myr ago). Only one loss affects an entire domain family, the avidins, which are represented in the chicken but not in sequenced mammalian genomes (Fig. 7). Avidins are present in oviparous vertebrates as minor egg-white proteins, and closely similar homologues in zebrafish, sea urchin and bacterial genomes indicate a loss of this domain family in mammals. Mammals also seem to have lost genes encoding vitellogenin I and II, which are yolk storage proteins providing nutrients to the early embryo in the egg. These losses were probably concordant with reduction in egg size and with internalization of the embryo during mammalian development. Other mammalian losses appear to reflect a presumed episode of nocturnal lifestyle in early mammalian history91. Apart from the known loss of CPD-photolyase, which is a DNA-repair enzyme dependent on light energy92, we find that mammals appear to have lost at least one pigmentation gene, indigoidine synthase A, whose bacterial homologue is an enzyme involved in generating blue colour pigments93.
Gene expansions in mammals. The largest mammalian expansion observed at the domain level involves the Kruppel-associated box (KRAB) domain, whose presence coincides with C2H2 zinc fingers in transcription factors. Whereas the KRAB domain has been found in more than 400 human genes94, the chicken seems to have fewer than 140. The complete absence of this domain in Fugu underlines the marked expansions of this family in mammals and to a lesser degree in birds. Orthology analysis identifies an apparent gene expansion of G-protein-coupled taste receptors in mammals (Fig. 7a, b). These fast-evolving receptors were previously studied only in mammals95,96. In the chicken genome, we find similar numbers of type I receptors (those responsible for sweet and umami taste; that is, certain amino acids such as Glu and Asp, and related compounds) but only three type II receptors, compared with around 30 in mammals. Type II taste receptors are thought to be responsible for the sensing of bitter tastants97,98, an adaptive avoidance system that may be more variable than the other taste systems. This might indicate that birds have a limited capacity for bitter taste, or that they have recruited other G-protein-coupled receptor subtypes for sensing bitter compounds.
Gene duplication events in chicken and human. To understand gene repertoire differences between avian and mammalian lineages, we quantified the number of gene duplication events since the divergence of chicken and human lineages by examining species-specific duplicate pairs (paralogues that reciprocally have their best hit in the same genome). Paralogues exceeding 95% pairwise identity were excluded from both data sets to avoid chicken genome assembly artefacts. In chicken, 13% (221 out of 1,712) of duplicate pairs have their closest hit in the chicken genome. In contrast, 27% (647 out of 2,426) of paralogues with <95% identity from the human genome are human specific, a pattern that is significantly different compared with that seen in the chicken (Fisher's exact test, P < 10-15). Thus, despite similar numbers of older duplications in both genomes (1,779 in humans versus 1,491 in chicken), there have been significantly fewer paralogues specific to the chicken lineage.
Evolution of function in orthologous gene clusters
Five alcohol dehydrogenase (ADH) genes plus one pseudogene fragment are present on GGA4, whereas seven genes (plus one pseudogene) are found in the orthologous segment on HSA4p23 (ref. 99) (Fig. 9; see also Supplementary Fig. S6). The ADH2 gene appears only recently to have been silenced in chicken because it is still functional in the ostrich100. Two recent duplication events have given rise to three chicken ADH genes that might compensate for some of the lost ADH2 functionality. These duplications, and others in human and other vertebrate lineages99 (Supplementary Fig. S6), might indicate changes in expression patterns or even reveal episodes of adaptation to nutritional and detoxification requirements involving changes to limited gene repertoires.
Change of function in orthologues
For the past 40 years, avian species have been described as lacking a functional urea cycle because of the absence in the chicken liver of the enzymatic activity that initiates the cycle: carbamoyl phosphate synthetase 1 (CPS1)101. This observation has been linked to a difference in excretion of nitrogenous waste: whereas mammals excrete urea, birds excrete uric acid as a possible stratagem in reducing the build-up of soluble urea during development in the egg. Contrary to these expectations we have identified a full-length, apparently functional, chicken CPS1 gene on GGA7 in the orthologous location to the human CPS1 gene (Supplementary Fig. S7a). Characterization of the expression profile of this candidate gene revealed expression in brain and muscle, and in immune tissues (Supplementary Fig. S7b). These data, and the presence of all the other enzymes of the mammalian urea cycle in the chicken genome (data not shown), indicate that this cycle is intact in birds but might perform a function distinct from the generation of urea for excretion.
Exploring chicken genome architecture
The assembly of the chicken genome, with its distinctive karyotype and smaller size, provides an opportunity to explore important issues in genome structure and evolution. Cytogenetic, molecular and mapping data have suggested previously that microchromosomes are (G+C)-rich, CpG-rich and gene-rich102,103,104, exhibit features that are correlated with transcriptionally active DNA105,106,107,108,109,110, and have structural counterparts in the (G+C)-rich mammalian chromosomal R-bands (Table 8). Our current analysis shows that the different size classes of chicken chromosomes exhibit a number of correlated attributes. Although these trends are well documented in mammals (for example, see refs 1, 2, 111, 112–113), the sequence of the chicken genome assembly demonstrates levels of complexity, breadth and resolution that previously could not be achieved.
Comparison of physical and genetic distances
A comparison of the physical distance along each chicken chromosome derived from the sequence assembly with the genetic distance between markers (based on the sex-averaged map25,29; see Methods) reveals wide variation in recombination rates that has a strong negative association with chromosome length (Fig. 10a). The recombination rate varies over an eightfold range among chromosomes (2.5 to 21 cM Mb-1); rates are much higher on microchromosomes (median value of 6.4 cM Mb-1) than on macrochromosomes (median value of 2.8 cM Mb-1). These results contrast with the narrow range and overall lower rates found in mammalian chromosomes of only 1–2 cM Mb-1 for human and 0.5–1.0 cM Mb-1 for the mouse1,2. The increased recombination rate of microchromosomes is such that all are likely to have total genetic lengths between 50 and 100 cM, as was expected, given the obligatory chiasma per bivalent to ensure normal segregation during meiosis114.
Chromosomal distributions of sequence features
Our analysis of the distributions of G+C content, CpG islands, and gene density and size was restricted to chromosomes with nearly complete sequence coverage; this excluded chromosomes GGA16, GGA22, sex chromosomes and microchromosomes smaller than GGA28 (Supplementary Table S2 and Fig. S8). We arbitrarily designated three chromosome size groups: large macrochromosomes (GGA1–5), intermediate chromosomes (GGA6–10) and 28 microchromosomes (GGA11–38). The overall G+C content declines markedly with increasing chicken chromosome length (Fig. 10b). The G+C content in macrochromosomes shows a narrow distribution, whereas for microchromosomes the distribution is broad and shifted to higher values (Supplementary Fig. S9a). Indeed, some individual chromosomes have almost no overlap in their distributions of G+C content (Supplementary Fig. S9b). These patterns differ from the considerable overlap observed in G+C content among individual mammalian chromosomes (Supplementary Fig. S9c; see also ref. 3).
Thus both G+C content and recombination increase with decreasing chromosome size. The association between G+C content and recombination may be viewed as a dynamic cycle. A possible explanation is that chromosomal regions with a high recombination rate might elevate the G+C content by “biased gene conversion”115. In biased gene conversion, mismatch repair within the heteroduplexes formed during recombination results in G+C-biased gene inclusion116. Biased gene conversion may also increase the neutral mutation rate of microchromosomes. Furthermore, studies on a wide range of species (summarized in ref. 111) have shown that (G+C)-rich DNA has an increased recombination rate. Thus, a cycle of high recombination may lead to increased G+C content, which then leads to a further increase in recombination rate until equilibrium is reached when selective forces prevent further change. This model suggests that changes in recombination brought about by a chromosome rearrangement may lead to changes in G+C content, as recently shown in mouse117.
In the chicken genome 48% of CpG islands (see Methods and Supplementary Table S5), often associated with promoters and other sites of regulation118,119, overlap a gene. We found that ∼38% of chicken CpG islands are conserved in the human genome. However, taking into account proximity to protein-coding genes and overlaps with ESTs (see Methods and Supplementary Table S5), we find that 10% of CpG islands conserved with the human genome are not near a gene. These may represent distant regulatory regions, flank undetected genes, or serve another function.
CpG island density presents a strong negative association with chromosome length, being highest on the intermediate chromosomes and microchromosomes (Fig. 10c). Similarly, gene density shows a strong negative association with chromosome length (Fig. 10c), confirming, at higher resolution, earlier studies102,104. Furthermore, we find a strong correlation between the length of a gene and the size of the chromosome in which it is found, an effect that is determined largely by variation in intron size (Fig. 10d). Although exon lengths (Fig. 10d) and numbers (data not shown) do not vary significantly among chromosome types, intergenic distances do increase with chromosome size (Fig. 10e). Intron length in the chicken correlates negatively with recombination (rs = -0.774, P < 0.001), G+C content (rs = -0.934, P < 0.001) and gene density (rs = -0.961, P < 0.001), as has been reported previously for other genomes120,121,122,123. The proportion of a chromosome covered by interspersed repeats increases with chromosome length (Fig. 10f; rs = 0.973, P < 0.001, for the 14 largest chromosomes excluding GGA8; rs = 0.940, P < 0.001, for all chromosomes when excluding the outliers GGA8, GGA24, GGA27 and GGA28). Not surprisingly, repeat density correlates positively with intron and intergenic gap length, as repeats primarily exist in these locations. Repeat density correlates negatively with recombination rate, G+C content and gene density (Fig. 10). The correlation with recombination rate is also expressed in the twofold higher than expected repeat density on sex chromosome GGAZ (Table 5) and an increased density with distance from the centromere (Supplementary Fig. S10). Recombination rate will influence repeat density if transposable element insertions on average are slightly deleterious and selection is more effective in regions of high recombination124. Indeed, recently a strong negative correlation between recombination rate and repeat density was reported in primates and rodents, where recombination rate was found to be proportional to distance from the centromere111. The correlation may be more obvious in chicken because of the larger differences in chromosome size and recombination rates. The same mechanism can be proposed to explain the correlation between the size of introns and intergenic gaps and recombination rate, assuming that longer introns and intergenic gaps are slightly deleterious.
The distinctive characteristics of microchromosomes are retained even after fusion with a macrochromosome, as illustrated by an ancestral chromosome fusion leading to GGA4. The p arm of GGA4 is orthologous to an ancestral microchromosome, as shown by comparative mapping using chromosome painting and by sequence matches with microchromosome 9 in turkey (see Methods). Interestingly, the telomeric DNA signals from fluorescence in situ hybridization analyses show that GGA4p possesses an interstitial telomere adjacent to the centromere (Supplementary Fig. S11). Analysis of the sequence assembly shows that GGA4p has not taken on characteristics of the macrochromosome, but rather it still has the properties of its ancestral microchromosome, including high recombination rate and high gene density (Table 9).
Synonymous substitution rates
Neutral substitution rate variation was analysed by examining synonymous substitution rates (KS), estimated using Codeml125, in a set of 7,529 human–mouse–chicken 1:1:1 orthologues (Fig. 11). As expected from the greater phylogenetic distance, the level of nucleotide substitution deduced from human–chicken alignments is much higher than that observed in human–mouse comparisons. The median value of KS for 1:1 orthologues between human and chicken is 1.66, indicating that synonymous sites on average have changed one to two times in the combined lineages to humans and chickens. For the analyses presented below, we acknowledge that with such high substitution rates, the conclusions should be corroborated with data from comparisons of much more closely related taxa, in particular from two or more bird lineages.
Synonymous substitution rates are elevated in two types of chromosomal region. First, chicken–human KS values are significantly higher among genes on microchromosomes (median KS = 2.01) than they are on intermediate chromosomes (1.60) and on macrochromosomes (1.54) (Table 10; P < 2.2 × 10-16, Kolmogorov–Smirnov test). Despite significant differences in G+C content for the chicken chromosomes (Fig. 10b), the elevation in KS is not entirely due to non-equilibrium base compositions of chicken or human sequences in the time since their ancestral genes diverged. Specifically, we found that KS distribution differences between genes on chicken micro- and macrochromosomes also remain significant when only considering genes exhibiting similar G+C contents, or G+C percentage at fourfold degenerate sites126 (Table 9). Higher neutral divergence rates are also seen on microchromosomes when genes are compared between chicken and turkey genomes127.
Second, subtelomeric sequences, which are often duplicated and polymorphic86,128, also show elevated neutral substitution rates. The synonymous substitution rates of chicken genes in regions up to 10 Mb from the ends of assembled macrochromosomes are elevated to levels that are indistinguishable (P > 0.05 in a Kolmogorov–Smirnov test) from those of microchromosomal genes when chicken–human alignments are considered (Table 10). It thus seems possible that the elevated KS values associated with chicken microchromosomal genes result from the more general phenomenon of rate elevation in subtelomeres: the KS value for a microchromosomal gene is, on average, higher than a gene on a macrochromosome because all microchromosomal genes lie within 10 Mb of one or both of their telomeres. However, it is important to note (Fig. 12) that genes on chicken microchromosomes are not, in general, subtelomeric in mammalian genomes. Nevertheless, the mammalian orthologues of chicken microchromosomal genes show significantly (P = 0.003) elevated KS values even when those values are derived from mouse–human comparisons. Thus, this gene set seems to have retained some characteristic other than chromosome location that has elevated neutral divergence during mammalian, as well as avian, evolution.
The overall increase in the rate of synonymous substitution for microchromosomal genes was not accompanied by significant changes in the chicken–human non-synonymous substitution rate, KA (median values of 0.121, 0.118 and 0.116 for macro-, intermediate and microchromosomes, respectively). Microchromosomal genes on average have smaller KA/KS ratios (median value of 0.052, compared with 0.073 and 0.068 for macrochromosomes and intermediate chromosomes, respectively), indicating that they have been subjected to greater degrees of purifying selection. This highlights a hitherto unforeseen contribution of genomic location to coding sequence evolutionary constraint.
One unusual aspect of the human genome with respect to other sequenced genomes is the abundance (4%) of large (> 20 kb), nearly identical duplications129. In marked contrast to observations for mammalian genome assemblies129,130,131,132, of the confirmed segmental duplications in chicken (see above), few exceeded 10 kb in length (Fig. 13), and none greater than 50 kb were detected. Almost all mapped duplications (93%) within the chicken genome are intrachromosomal (excluding unplaced sequences) (Supplementary Fig. S12). Analysis of exon content of the duplicated regions revealed that only 3.7% of Ensembl predicted genes showed evidence of being recently duplicated—a slight enrichment as compared to the genome average of confirmed duplicated bases (3.0%). The proportion of uncharacterized genes within these recent segmental duplications is nearly fivefold greater than that observed for other unique sequences. Characterization of these may uncover lineage-specific genes important for the adaptation of the chicken.
Evolution of vertebrate genomes
Amniote chromosomal rearrangements
The evolutionary history of genome organization can be inferred from the comparative analysis of gene orders on chromosomes. Recent studies of mammalian genomes suggest a larger number of rearrangements than previously estimated1,2,133,134,135,136, largely due to the underestimation of intrachromosomal rearrangements. Using the chicken genome sequence as an outgroup provides an opportunity to understand better the rate of rearrangements in mammalian lineages and the architecture of the ancestral mammalian genome.
We generated two maps of conserved synteny—defined as orthologous chromosomal segments with a conserved gene neighbourhood133,134,137—among chicken, human and mouse, using DNA- and protein-level methods (see Methods). These two sets of anchors generated highly similar maps, with comparable sizes of orthologous segments and amounts of coverage of the human and mouse genomes (Table 11). We then used two approaches to quantify chromosomal dynamics: (1) by counting the number of synteny breaks where the ancestral state is confirmed by conserved synteny to an outgroup species; and (2) by estimating the minimum number of rearrangements that could explain the current genome organization. Both approaches are consistent, revealing a slow rate of rearrangement in the human lineage, about one-third that of the rodent lineage (Fig. 14). The synteny analysis using orthologous gene pairs is more sensitive at long evolutionary distances than DNA-based methods, and it allowed us to use recently sequenced fish genomes of Fugu67 and Tetraodon138 as outgroups. This analysis revealed an even slower rate of rearrangements in the chicken lineage (Fig. 14). The synteny maps confirm an earlier observation139 that the human genome is closer to the chicken than to rodents in terms of chromosomal organization of genes (Supplementary Figs S13 and S14). As indicated from the analysis of the Tetraodon genome, this surprising level of similarity arises from an unusually low rate of interchromosomal shuffling in the lineage leading to the earliest mammal.
We reconstructed the putative mammalian ancestor genome (more precisely, the common ancestor of human–mouse–rat and all the Euarchontoglires140) by looking for a scenario minimizing the number of rearrangement events on the evolutionary tree141 (see Methods). Because the pairwise distances between the initial genome are substantial, it is possible to find alternative ancestors that also minimize the total number of rearrangement events. By exploring some of these alternative ancestors, we partitioned all the pairs of adjacent synteny blocks of the recovered ancestor into ‘strong’ and ‘weak’ pairs depending on whether they are present or not in all of the observed alternative ancestors (Fig. 12a). Many of the previously postulated chromosome associations of the placental ancestor correspond to strong pairs of adjacent blocks in the mammalian ancestor136,142. The synteny of six human chromosomes and of ten chicken chromosomes is preserved in the mammalian ancestor, whereas synteny only of the rodent X chromosome is preserved in the mammalian ancestor141. This is consistent with the suggestion that interchromosomal rearrangements have been more frequent in rodents139. The scenario recovered for the X chromosome also reveals variable rates of intrachromosomal rearrangements (Fig. 12b). There appear to have been no rearrangements between HSAX and the X chromosome of the mammalian ancestor (that is, human order is ancestral), yet there are three inversions between the mouse X chromosome (MMUX) and the mammalian ancestor X chromosome, and there are 14 rearrangements (13 inversions and 1 fusion) between the two chromosome segments in chicken (from GGA1 and GGA4) and the mammalian ancestor X chromosome.
Variation in human–chicken genome expansion ratio
The large variation in genome size between organisms143 has been addressed previously by comparing entire genomes, thereby deriving an aggregate estimate for expansion or contraction in one lineage. The change in overall genome size between chicken and mammals coupled with the conserved synteny allows us to isolate different factors that contribute to this variation (see Methods). The chicken genome is about 40% the length of the human genome, but this ratio is not constant across all orthologous regions (Fig. 15a). The variation in length ratios for human–chicken alignments is much greater than that seen for human–rodent alignments (Fig. 15b). The higher density of interspersed repeats in the human genome accounts for ∼10% of this variation (correlation coefficient 0.326, P < 10-4), with the strongest positive association at intermediate ranges (Fig. 15c). G+C content also accounts for a portion of the residual variability: after subtracting repeats from human and chicken in each window, the resulting length and G+C content ratios are strongly and negatively related (correlation coefficient -0.256, P < 10-4) (Fig. 15d).
More extensive multivariate regression analyses using these and other genomic parameters derived from human–chicken alignments explain only 20–25% of the variability in length ratios (see Methods and Supplementary Table S6). Features not measured in the current analysis are thus the main contributors to changes in genome size. One potential hidden variable is the density of ancient repeat sequences that have diverged beyond the ability to be found confidently by RepeatMasker144. For example, repeat expansions in G+C-poor regions during early mammalian evolution might contribute to length ratio variation.
Illuminating the human genome: conserved non-coding segments
One important use of the chicken sequence is to identify functional non-coding elements in the human genome. The high synonymous substitution rates described above would preclude neutral regions from aligning between human and chicken. Thus the regions that do align are likely to be subject to purifying selection and hence are predicted to be functional. Compared with approximately 5% of the human sequence estimated to be under purifying selection, based on human–mouse comparisons2,145, only 2.5% of the human sequence aligned with chicken. Of aligned positions, 44% are in protein-coding regions, 25% are intronic and 31% are intergenic. Aligned sequences within genes were discussed above in relation to chicken gene content. Here we investigate the global properties of the complete set of non-coding aligned segments.
Conservation patterns in proximal cis-regulatory regions
We examined functional non-coding elements in the human genome to analyse how frequently they are conserved with chicken. For sets of known regulatory elements, functional promoters, predicted promoters, CpG islands, and predicted transcription-factor-binding sites that are conserved between human and rodents, we find that only 30–40% are conserved between human and chicken (Fig. 16). Even for coding exons, which are among the most highly constrained elements in the genome, only 75% aligned between human and chicken. Thus, conservation between mammalian and chicken genomes greatly increases specificity in the search for functional elements, but at a price in sensitivity that varies depending on the functional category. The conserved regulatory modules (CRMs; for example, promoters and enhancers) that still align between human and chicken are not a coincidental result of a lower local rate of neutral divergence because the distribution of neutral substitution values2,126 for the regions containing conserved CRMs was indistinguishable from that for the diverged CRMs. We assessed whether the biological functions of genes regulated by the conserved CRMs might differ from those with divergent CRMs. We found that genes from three GO categories (development, metabolism and structural component of muscle) are over-represented in the set regulated by the conserved CRMs, whereas genes within the signal transducer and hydrolase activity GO categories are over-represented in the set with non-conserved regulatory elements (Fig. 17).
Conserved non-coding regions are clustered and far from genes
Several recent human–rodent comparisons146,147,148 within limited regions of the genome have observed that conserved non-coding segments tend to lie relatively far from genes. To investigate whether this holds true across the genome in human–chicken comparisons, we partitioned the human genome into non-overlapping intervals, and in each interval (after excluding repetitive nucleotides) computed the fraction that falls in a coding exon annotated by RefSeq or Ensembl (the ‘coding fraction’) and the fraction that aligns with chicken but does not intersect coding exons (the ‘conserved non-coding fraction’ or CNF), taking care to keep the two quantities logically independent (see Methods). The two quantities are negatively associated (correlation coefficient -0.197, P = 0.000; Supplementary Fig. S15). This inverse correlation is robust to variation in the interval size (between 100 and 1,000 kb), the set of gene annotations, and the set of non-coding alignments, and it is not explained by G+C content alone (see Methods). Proximal cis-regulatory elements (that is, those within a few kilobases of an exon) thus seem not to be the primary contributors to overall non-coding conservation.
Non-coding conservation is not uniformly distributed across the human genome. Fifty-seven segments of high non-coding conservation (average length 1.176 Mb and average CNF 13.1%, compared with a genome average of 1.7%; see Methods and Supplementary Table S7) were found to be gene poor (they cover 2.3% of the human genome but contain only 0.3% of the exons in RefSeq genes). They were also G+C poor (38.8% overall) and depleted for all classes of interspersed repeats, as are their chicken orthologues. They contain none of the 731 break points identified through an analysis of regions of conserved gene order between human and chicken with length exceeding 200 kb, and—from human–mouse neutral substitution rates estimated using interspersed repeats126—do not seem to have experienced particularly low mutation rates. The genes within or overlapping these 57 high-CNF segments, however, are significantly enriched for the GO categories associated with gene regulation (Fig. 17). Thus, the degree of non-coding sequence conservation is related to the biological function of the genes in the general genome neighbourhood. This enrichment is not merely a by-product of the gene-poor nature of these segments, because human gene-poor regions with low CNF are not enriched for the same GO categories149.
The 57 high-CNF segments were found to contain 60.7% of 417 human–chicken ultraconserved elements (UCEs): sequences defined as being 200 bp or longer in orthologous chromosomal locations and 100% identical between the two species. Only 27.3% of these human–chicken UCEs overlap the 481 previously studied human–rodent UCEs150, although all of them are found in the draft genomes of either mouse or rat, at 87–100% identity in the more conserved rodent. Human–chicken UCEs differ from human–rodent UCEs in containing far fewer exon-intersecting elements (7.9% versus 23%) and in showing only weak enrichment for proximity to genes whose products have a GO and InterPro classification associated with RNA splicing. The two UCE sets share a strong bias against harbouring human-verified single-nucleotide polymorphisms (P < 10-40 compared to the genome-wide average in both cases). In both, the set of elements with no expression evidence (non-exonics in ref. 150) is found in or next to genes whose products are highly enriched for transcriptional regulation, DNA binding, homeodomains and developmental functions; many of these UCEs are found more than 100 kb away from the corresponding genes. Finally, in both sets, no UCEs except a handful of coding regions could be traced back through sequence similarity to Ciona intestinalis, Caenorhabditis elegans or Drosophila melanogaster. At the moment, little is understood about the functional significance of either the UCEs or the high-CNF segments that are far from genes.
The chicken genome represents an intermediate test case as a target for genome sequence assembly and analysis. Although less than half the size of mammalian genomes, it is still much larger than those of D. melanogaster, C. elegans and even T. rubripes, and it lacks the dense linkage map platform that helped to assemble the first two. Unlike the rat and chimpanzee genomes, there was no closely related, high-quality genome sequence already available to provide a framework for assembly. Nevertheless, a relatively high-quality draft of the chicken genome has been achieved on the basis of only 6.6 × whole-genome shotgun coverage, owing in part to the remarkably low level of recent transposon activity it has endured (Table 5).
The quality of this draft genome sequence makes it a key resource for comparative genomics. Natural selection and evolution provide us with many perspectives from which to view our own genome. Genomes of distant species resolve key processes that have been conserved over millennia, whereas those of our close relatives allow an analysis of rapidly changing sequence. In many respects, the chicken genome provides insights that were unavailable from previous sequences. For nearly every aspect of biology, it allows us to distinguish features of mammalian biology that are derived or ancient, and it reveals examples of mammalian innovation and adaptation.
The chicken is sufficiently distant that little unselected sequence has survived unchanged along the separate evolutionary paths to birds and mammals from their last common ancestor. Against this background, conserved non-coding sequences stand out clearly. Some of these represent known regulators (Fig. 16) and others use novel mechanisms yet to be identified. On the other hand, the counterparts of many functional mammalian sequence elements could not be identified in the chicken sequence. Either these represent mammalian innovations or else any commonality has been lost over the course of >310 Myr of separate mutation and fixation.
Chicken breeding, based on quantitative genetic methods, represents one of the most remarkable examples of directed evolution. Even after 50 yr of intensive selection, annual genetic progress in production traits remains undiminished151. An impressive list of chicken quantitative trait loci has already been identified152, many with combined agricultural and medical relevance. The chicken genome sequence promotes both the development of more refined polymorphic maps (see the accompanying paper153) and the framework for discovering the functional polymorphisms underlying interesting quantitative traits, thus fully exploiting the genetic potential of the chicken.
The chicken genome is invaluable for shedding light on functional elements of the human genome and our unique evolutionary history. It also points the way forward to the great utility we can expect from the genome sequences of other carefully chosen species. The data presented here demonstrate both the unique value of the chicken as a model species and emphasize the incomplete nature of our collective understanding of complex organisms. This chicken genome sequence will both integrate and stimulate the expanding array of contributions from this versatile species.
Domain matching and ranking
To identify known families of genes and domains we scanned respective proteomes for characteristic HMM profile signatures from Pfam69 and SMART databases using HMMER (http://hmmer.wustl.edu/) software and applying corresponding family-specific cutoffs. The identified families were ranked by the number of matching genes requiring at least one matching transcript, and only counting repetitive matches once.
Orthologous relationships between genes of chicken, human, Fugu and others were inferred through systematic similarity searches at the level of the predicted proteins. We retained only the largest predicted ORF per locus, and compared those in an all-against-all fashion using the Smith–Waterman algorithm. We then formed orthologous groups using a variant of a strategy used earlier137,154,155. First, we grouped recently duplicated sequences within genomes into ‘paralogous groups’, to be treated as single sequences subsequently. For this, there was no fixed cutoff in similarity, but instead we started with a stringent similarity cutoff and relaxed it on each successive step, until all paralogous proteins were joined, thereby satisfying the following criteria: all members of a group had to be more similar to each other than to any other protein in any other genome; and all members of the group had to have hits that overlapped by at least 20 residues, to avoid ‘domain walking’. After grouping paralogous proteins, we started to assign orthology between proteins by joining triangles of reciprocal best hits involving three different species (here, paralogous groups were represented by their best-matching member). Again, a stringent similarity cutoff was used first and relaxed on each successive step, and all proteins in a group were required to have hits overlapping by at least 20 residues. Finally, we joined any remaining nodes by allowing not only reciprocal triangles, but also reciprocal tuples.
Detection of gene loss in mammals
The orthologous relations defined above were used to infer losses when a gene was found in chicken and in at least one earlier-branching animal, but not in any mammal. Of 122 candidate losses obtained in this manner, many were manually discounted after Blastp searches in mammalian genomes (thus hinting that several as-yet-unannotated genes in mammals remain to be predicted).
Detection of orthologous introns
For each orthologous group we created a multiple alignment and mapped intron positions and protein features onto it. This procedure is incorporated into the SMART web server. To minimize errors due to erroneous alignments, introns flanking alignment gaps were discarded (less than 1% of all introns). To compensate for effects of intron sliding and to reduce further the impact of possible alignment errors, we allowed a window of 12 nucleotides in which we considered a position as conserved. Previous estimates indicated that the chance of independent intron insertion in such a window is <1%. To avoid biases due to incomplete gene predictions, we omitted 18,910 introns in regions that were missing from some of the predicted genes.
Deriving tissue expression data
Chicken ESTs were mapped to the assembly, and to Ensembl genes (± 1 kb), using BLAT and a 95% identity threshold, and were partitioned into ten (brain; fat and skin; bone and connective tissues; heart; kidney and adrenal tissues; immune; liver; female reproduction; alimentary; testis) distinct tissue types. Percentage amino acid sequence identities of 1:1 chicken–human orthologues were calculated as described previously (Fig. 6). Note that single genes may be assigned to multiple tissues.
Human–chicken whole-genome alignments were obtained by using the program BLASTZ156 to produce short (typically 100–1,000 bp) local alignments, and then assembling gap-free segments of those alignments into ‘chains’ in which aligned segments occur in the same order and orientation in both species134. These alignments—which were used to generate data for Figs 2, 15, 16, 17 and Table 4, and Supplementary Figs S1, S15 and Supplementary Tables S6 and S7—can be obtained from the U.C. Santa Cruz Browser (http://genome.ucsc.edu/). To compute the CNF of a human genomic interval, we limited consideration to non-repetitive bases that are not in a local alignment that intersects a protein-coding region, and determined the fraction of those bases that are within an alignment.
Evolution of vertebrate genomes
The maps of conserved synteny (orthologous chromosomal segments with a conserved gene neighbourhood133,134,137) between chicken, human and mouse were produced using whole-genome DNA alignments post-processed into chains and nets134 as well as looking for a conserved neighbourhood of orthologous genes as described previously133,137. We used gene-based synteny as input to MGR157 and GRIMM133,158 to look for parsimonious scenarios of rearrangements, starting from a set of 6,447 four-way orthologous genes pre-filtered for evidence of conserved pairwise synteny using SyntQL (E. Zdobnov, unpublished program) and applying GRIMM-Synteny133 for more stringently defined 586 four-way human–mouse–rat–chicken synteny blocks.
Detailed descriptions of all methods are provided in the Supplementary Information.
The Gallus gallus sequence and map generation at Washington University School of Medicine's Genome Sequencing Center was supported by grants from the National Human Genome Research Institute (NHGRI). For work from other groups, we acknowledge the support of the Biotechnology and Biological Sciences Research Council, Center for Integrative Genomics Funds, Childcare and Lejeune Foundations, Chinese Academy of Sciences and Ministry of Science and Technology, Department of Energy, Desiree and Niels Yde Foundation, European Union, European Molecular Biology Laboratory, Fonds Quebeçois de la Recherche sur la Nature et les Technologies, Howard Hughes Medical Institute, National Institute for Diabetes and Digestive and Kidney Diseases, NHGRI, National Institutes of Health, National Natural Science Foundation of China, National Science Foundation, Novo Nordisk Foundation, Stowers Institute for Medical Research, Swedish Research Council, Swiss NCCR Frontiers in Genetics, Swiss National Science Foundation, USDA/CSREES National Research Initiative, USDA/CSREES National Animal Genome Research Program, Wallenberg Consortium North and the AgriFunGen program at the Swedish University of Agricultural Sciences, UK Medical Research Council, University of California Presidential Chair Fund, University of Pennsylvania Genomics Institute Award, University of Texas at Arlington, and the Wellcome Trust. Resources for exploring the sequence and annotation data are available on browser displays available at Ensembl (http://www.ensembl.org), UCSC (http://genome.ucsc.edu) and the NCBI (http://www.ncbi.nlm.nih.gov). We thank R. Waterston for advice regarding the manuscript.
Correspondence and requests for materials should be addressed to R.K.W. (email@example.com). This G. gallus whole-genome shotgun project has been deposited at DDBJ/EMBL/GenBank under the project accession number AADN00000000. The version described in this paper is the first version, AADN01000000.
About this article
BMC Genomics (2019)