As the most extensively sequenced other bird species, all available zebra finch genes were compared with the chicken genome. This was achieved by clustering 10 validated zebra finch mRNAs and expressed sequence tags, then using chicken protein sequences to search this zebra finch database with Blastx, 11 and subsequently implementing T-Coffee 12 to generate 3,653 pairwise CDS alignments from the Blastx best-hit pairs (for details see supplementary methods).

Pairwise d _N /d _S (ω) was calculated for each CDS alignment using the codeml implementation of the PAML 3.15 package 13 . If synonymous and nonsynonymous mutations are neutral, the relative rates of each are expected to be equal so that ω = 1 14 . Departures from this, where ω > 1 (d _N > d _S ) suggest that nonsynonymous mutations are advantageous, and are maintained under directional selection. If ω < 1 (d _N <d _S ) then the nonsynonymous SNPs may be deleterious since they are not preserved and are likely to be subject to purifying selection. We compared ω by maximum likelihood under two different models: a neutral model where ω was fixed = 1, and a model where ω was free to vary. These models were compared using a likelihood ratio test (LRT) to determine if the variable model was significantly better at explaining the data 13 .

As a consequence of this conservative strategy of calculating ω across the entire gene length, genes may be discounted when the signal of directional selection is focused on specific regions or domains and thus obscured by purifying selection operating on the majority of the gene 15 . Many genes known to be subject to positive selection have 0.5 <ω < 1 16 , so using a lower cut-off point to identify candidate genes that may be subject to selection can be effective. Accordingly, chicken-zebra finch alignments with ω > 0.5 where the variable model was significantly favoured (p < 0.05) were identified. The annotation associated with the best human orthologs from the Panther database 17 was used to identify the function of chicken genes with relevance to the immune system.

The chicken IL-4Rα mRNA sequence (Refseq ID: XM_414885), initially determined by Boardman et al. (GenBank accession: CR407301) and Caldwell et al. 18 , was aligned as a best hit to two clustered zebra finch ESTs, DQ213788 and DQ213787 19 .

A total of 90 chicken samples were acquired: 70 village birds from Asia and Africa (International Livestock Research Institute, Kenya) and 20 commercial broilers (Manor Farms, Co. Monaghan, Ireland). The commercial birds were composed of 10 Ross breed chickens from Ireland and 10 Hubbard Flex from France. The Asian and African samples included 10 chickens from each of 3 Asian (Bangladesh, Pakistan and Sri Lanka) and 4 African (Botswana, Burkina Faso, Kenya and Senegal) populations. One sample for each of six outgroup species were also sequenced. Three were species known to be closely related to the chicken: 20 Chinese bamboo partridge (Bambusicola thoracica, CAS89821), green JF (Gallus varius, CAS85707) and grey francolin (Francolinus pondicerianus interpositus, CAS87894) (Department of Ornithology and Mammalogy, Californian Academy of Sciences). And three other JF were also obtained: Grey (Gallus sonneratii), Ceylon (Gallus lafayetii) and red JF (Gallus gallus) samples (Wallslough Farm, Co. Kilkenny, Ireland). DNA was isolated from the samples using a phenol-chloroform extraction following a proteinase K digestion.

The UCSC http://genome.ucsc.edu, GenBank http://www.ncbi.nlm.nih.gov and Ensembl http://www.ensembl.org browsers were used to investigate the gene structure. At the time of analysis, a portion of the chicken IL-4Rα region was not displayed on these browsers, so the reference assembly (NC_006101) and reference contig (NW_001471454) were aligned with the IL-4Rα mRNA sequence from GenBank (XM_414885) using T-Coffee 12 to determine potential coding regions. A further T-Coffee alignment of the human and chicken IL-4Rα protein sequences identified chicken regions orthologous to variable regions in humans (Additional file 1): according to the Uniprot http://www.uniprot.org entry for human IL-4Rα (Uniprot: P24394), most variation is in the extracellular and cytoplasmic domains. Genscan was used to corroborate the predicted gene structure http://genes.mit.edu/GENSCAN.html.

PCR primers were designed using Primer3 software http://frodo.wi.mit.edu and were constructed by VHBio, UK http://www.vhbio.com. The details of the primer sequences and optimal parameters for their usage are available in Additional file 2 (Table S1). Each amplicon was amplified according to the PCR cycle setup (Table S2 in Additional file 2): 8 were successfully amplified for all 96 samples (Figure 1). The use of this large sample size increases probability of identification of a target of selection 21 . The forward and reverse PCR product sequences were determined by Agowa, Germany http://www.agowa.de.

Sequencing reads were assembled into contigs using the Phred-Phrap-Consed-Polyphred pipeline programs http://www.phrap.org/phredphrapconsed.html Phrap v0.990319 and Phred v0.020425.c 22 23 . Bases were called with Consed 24 using P(base is correct) = 1–10^-S/10, where S the base quality score 25 . Any bases with S < 20 were not included in the analysis, so all bases had at least a 99.0% probability of being correct: most had S ≥ 40 (99.99%). Only SNPs with high probability of being accurate (polyphred ranks 1, 2 or 3) were used in further analyses, and only SNPs in polyphred rank 1 were used for the outgroup samples. Polyphred version 5.0 26 27 was used to assemble the data for further processing.

A list of the genotypes for each sample was collated and PHASE 28 was used to infer missing haplotypes. These assigned haplotypes were cross-referenced with haplotypes generated by Arlequin 29 to ensure consistency – both were identical. Any sequence sites with inadequate coverage across populations or continents, which had sub-standard base quality scores, or had insufficient coverage for either forward or reverse sequences, were removed – leaving a total of 5,298 bp for further analysis.

DnaSP 4.0 30 31 was used to analyse the polymorphic characteristics of the data and to perform a series of population genetic analyses. The numbers and types of SNPs were assessed. Nucleotide diversity was measured using π, the average number of nucleotide differences between sequences pairs 32 . The haplotype diversity (Hd, the number and frequency of haplotypes in the sample) 33 , the number of haplotypes, Kelly's Z _nS 34 and θ _W = 4N _e μ 35 were determined. The four gamete test for the minimum number of recombination events (R _M ) 36 and R (the degree of recombination) 37 were calculated, as was the GC content.

A set of summary statistics were used to identify departures from neutrality using coalescent simulations: Fu and Li's D and F 38 , Tajima's D 39 , Fu's F _s 40 , Fay and Wu's H 41 . These tests were performed using DnaSP for 10³ replicates with the following parameters estimated from the resequenced data: numbers of genotypes, segregating sites, total sites, sample sizes and rate of recombination. These simulations generated empirical distributions with which the statistical values were compared to determine the extent of their deviation from neutrality. It is an indication of non-neutral evolution if the observed values lie at the extremes of the distribution.

Median-joining haplotype networks were constructed using Network version 4.2.0.1 http://www.fluxus-technology.com. AMOVA tests 42 were conducted using Arlequin 29 with 10³ permutations. Predictions to estimate the extent of functional impact for each radical substitution were conducted using PMut http://mmb2.pcb.ub.es:8080/PMut/ 43 .

The McDonald-Kreitman test 44 was implemented using DnaSP to examine the relative ratios of fixed and non-fixed nonsynonymous differences to fixed and non-fixed silent changes, which can indicate the presence of non-neutral evolution. Significance was based on a two-tailed Fisher's exact test.

To investigate for evidence of selection in IL-4Rα between chicken and each of the 6 outgroups, CDS alignments were generated and ω was determined under a variety of models using codeml 13 . For this analysis, a chicken sequence from the most numerous haplotype was used (FJ542575). Although the chicken coding haplotypes observed at IL-4Rα were diverse, substituting this for other chicken genotypes yielded no significant changes to results, except at certain sites with model M8 for a divergent sample (FJ542675). The PAML models implemented here are sensitive to low numbers of sequences 45 .

The free-ratio (M1) model was used to calculate tree branch lengths and ω for each species lineage in the sample. To identify specific codon sites with evidence of selection, site-specific models estimated ω for each site across the whole sequence by using a random sites model under Bayes empirical Bayes (BEB) 13 46 47 . For each model both the ω values and the fractions of sites affected are informative. For M1a, a neutral model, only two (K = 2) fixed ω values are permitted: ω ₀ = 0 (conserved) and ω ₁ = 1 (neutral). For M2a, a variable model, these two classes are used with an additional class (K = 3) where ω is freely estimated to allow for deviations from neutrality. Similarly, M7 is a neutral model that models K = 4 site classes sampled from a β-distribution, all of which have 0 ≤ ω ≤ 1. Variable model M8 has the same four β-distributed classes as M7 with an additional class where ω > 1 (K = 5). A LRT was conducted between the paired neutral and variable models (M1a vs M2a, M7 vs M8). BEB was used to determine the posterior Bayesian probability of ω for each amino acid site: a significantly high posterior probability for this variable ω class suggests that a particular site is under selection, if ω > 1 and M8 (or M2a) is significantly favoured by the LRT 48 . Candidate positively selected sites from M8 were examined using PMut to assess the functional impact for each nonsynonymous substitution.

Tests on 3,653 chicken and zebra finch CDS pairwise alignments identified genes with ω > 0.5 where the variable model was significantly favoured from (Table S3 in Additional file 2; Additional file 3). From these, IL-4Rα was selected for further analysis because of its critical function in the immune response, including an implicated role in the anti-viral response 1 . Interestingly, another chicken immune gene identified by the pairwise comparison method (Progesterone-induced blocking factor) had a human ortholog that binds IL-4Rα.

IL-4Rα was resequenced in 7 closely related bird species: chicken, red JF, grey JF, Ceylon JF, green JF, grey francolin and bamboo partridge. An excess of nonsynonymous compared to synonymous substitutions was observed in all birds except red JF (Table S4 in Additional file 2). Branch-specific models of evolution, implemented in PAML 13 were used to investigate evidence of selection among the sequenced lineages. Using the free-ratio model, the branch leading to the Gallus genus was determined to have a high ω value (0.92) (Additional file 4), though this cannot be taken as strict evidence of positive selection. Consequently, site-specific models were implemented to investigate whether particular codon sites contributed to the evidence of selection. Model M8, one the most conservative models of site-specific evolution was determined to be significantly more favoured in comparison to the neutral M7 model (p = 5 × 10^-23; Table 1). Bayes Empirical Bayes (BEB) was used to estimate the proportion of sites under positive selection: 48 (9.8%) of the sites had ω > 9.5, values much greater than that expected under neutrality 47 . Under M8, 28 sites were identified with a BEB posterior probability of at least 95% for ω > 1 (Table 2). There were substitutions between the chicken and red JF sample or genome sequence at 6 of these sites (5, 517, 547, 590, 628 and 665). PMut found 4 substitutions at these sites would have a neutral effect on protein structure (Table S5 in Additional file 2).

Of the 100 SNPs observed among the chicken populations, 7 were singletons. In protein-coding regions 17 SNPs were observed: 10 were nonsynonymous and 7 were synonymous. Assuming red JF was the primary ancestral origin of diversity at this gene, some replacement mutations between red JF and chicken are potentially associated with the domestication process. In the chicken 7 nonsynonymous substitutions were identified as segregating at high frequencies (55% or more): F5L, L520P, S590G, L594R, M665R, S670Y and T692S (Table 3, Additional file 5).

The generation of median-joining networks (Figure 2) illustrated a high degree of allele diversity among samples and little geographical structuring among populations. The number of genetically divergent high-frequency haplotypes showed a trend of balanced diversity (Figure 2). When only the nonsynonymous SNPs were examined, an interesting pattern of dominant haplotypes emerged (Figure 3); when silent SNPs were included, recombination obfuscated these groups (Additional files 6 &7). Four haplotypes containing 81% of the 180 genotypes were characterised by substitutions at two sites: F5L and L520P. The 4 alleles possible at these 2 sites (F-L, F-P, L-L and L-P) were present in all 8 populations. No single variant was dominant among the samples: 32 were F-L, 38 were F-P, 46 were L-L and 64 were L-P. Both sites 5 and 520 showed evidence for positive selection in the site-specific test in codeml (Table 2, Additional file 8). Here, red JF and chicken both shared L520 and P520 alleles as well as F5, but L5 was unique to chicken.

The feature of high population diversity and low geographic partitioning in the networks was apparent in the analysis of variation using AMOVA with the Arlequin package 29 . This assessed the extent of partitioning of diversity at different levels of population structure. Most variation lay within the populations (94.1%, p < 1 × 10^-5), a trend seen in other studies of chicken populations 48 49 ; the remainder partitioned between the populations (1.8%, p = 0.060) and the continents (4.1%, p = 0.033).

There was further evidence for the trend of elevated allelic diversity: 115 haplotypes were observed in just 180 genotypes. This was reflected in the high Hd value, a statistical measure of haplotype diversity (Table 4). Fu's F _S was highly negative, signifying an excess of rare alleles. Nucleotide, haplotype and SNP diversity were all higher in Asia than in Africa as expected, despite sampling fewer birds in Asia (30) than in Africa (40).

The significantly positive Tajima's D in Asia and Africa (Table 4) and in each of their populations (Table S6 in Additional file 2) was paralleled by a highly negative Fay and Wu's H, an indicator of an excess of derived alleles. Together, these metrics suggest a clear tendency for alleles to rise to mid- or high- frequency levels. Tests on the protein-coding portion of the gene alone indicated a significantly negative Fay and Wu's H (-3.02, p = 0.04) and a less positive Tajima's D (0.61); the latter may be a consequence of stronger conservation in coding regions, which appears to limit diversity, except at sites 5 and 520.

Moderate recombination was detected at IL-4Rα: for the given value of the recombination rate R, coalescent simulations showed the minimum number of recombination events (R _M ) was significantly high among all groups (Table 5).

Evidence of non-neutral evolution was evident from the McDonald-Kreitman 44 test results. The McDonald-Kreitman test examines the relative ratios of fixed and non-fixed nonsynonymous differences to fixed and non-fixed silent changes between species. Purifying selection may explain a rate of fixation of nonsynonymous differences much lower than that for silent substitutions. Alternatively, if there is a significant excess of fixation of nonsynonymous changes compared to silent ones, then directional selection may be present. The chicken genotypes were tested against the red JF genome sequence and also against each of the outgroup samples. Both tests showed an excess of fixed nonsynonymous substitutions (p = 0.002 with the genome sequence, p = 0.040 with all the outgroup samples; Table 6), indicating that selection may have affected the evolution of this gene.

This study shows evidence for high and balanced diversity at the chicken IL-4Rα gene, which was initially identified through the evaluation of the rate of nonsynonymous to synonymous substitutions in pairwise comparisons of chicken and zebra finch orthologs. This strategy incorporated functional and literature information to detect a suitable gene for resequencing in African, Asian and commercial chicken samples, as well as in related JF and bird species. Haplotype networks, tests of neutrality and summary statistics indicated a signal of balanced nonsynonymous polymorphisms at two sites in the IL-4Rα gene. Networks showed that red JF is the primary source of diversity at this gene. The elevated and balanced diversity present in all the populations might be a result of the chicken's history of multiple domestications 72 73 74 , introgressions 76 77 78 and subsequent admixture of different types 79 80 81 . However, the identification of two potentially functionally significant SNPs as fulcrums of the balancing signal suggest that the functions of IL-4Rα in the immune system may affected by selective processes for specific allelic variants in response to new pathogenic challenges during domestication.

CDS: coding sequence; IL-4Rα: interleukin-4 receptor alpha-chain gene; JF: jungle fowl; LRT: likelihood ratio test; π: nucleotide diversity; ω: the rate of nonsynonymous mutations per nonsynonymous site (d _N ) divided by the rate of synonymous mutations per synonymous site (d _S ); UCSC: University of California Santa Cruz.

DB, COF, AL and DL designed the study. TD and DL completed the bioinformatic gene identification. OH, TD, AB, PS, AN, RS, RSS and BP carried out sample collection. TD and SC prepared the samples. TD did the resequencing, data assembly and conducted population genetic tests. TD, DL, DB, COF and AL wrote the manuscript. All authors read and approved the final manuscript.