Introduction

DNA genotyping arrays with hundreds of thousands (or more) single nucleotide polymorphisms (SNPs) are commonly used in genetic studies of model species, but relatively few arrays have been developed in other species because of their high costs of development. In non-model species, alternative techniques such as Restriction Associated DNA sequencing (RAD-seq) are more often used to analyze SNPs, but for a variety of reasons studies using RAD-seq often have an order of magnitude fewer SNPs (tens of thousands of SNPs) than SNP arrays, which can be insufficient for some applications (e.g. genome-wide association studies, GWAS1). Studies using RAD-seq compared to SNP arrays also involve more extensive processing requiring specialized expertise to filter and identify genetic markers. Thus, some researchers have recently tested commercial SNP arrays developed for model species, particularly mammals, on closely related non-model species to determine if they can provide easier access to larger numbers of SNPs2,3,4. In contrast to mammals and plants, relatively few studies have applied SNP arrays to studies of birds. Some studies of passerine birds have used custom Illumina beadchip arrays5,6,7, but they are not commercially available and have relatively fewer SNPs (10–50 K) than arrays for other model organisms. Also, the passerine arrays may not be readily applicable to non-passerine birds, as the two lineages are estimated to have split 60–75 million years ago (mya)8,9. The most recent SNP array manufactured by Affymetrix for a non-passerine poultry species, the domestic chicken Gallus gallus, however, has nearly 600 000 SNPs including a set of SNPs in exonic or coding regions10 and could be potentially useful for applications to non-passerine birds.

One limitation of using SNP arrays is the level of cross-species amplification. In a study of 16 wild mammal species, the percentage of called SNPs declined 1.5% for each million years of evolutionary divergence11. Despite the apparent reduction in call rate, some SNP arrays have been successfully applied to phylogenetically distant lineages. For example, the EquineSNP50, developed for the domestic horse, still yielded a call rate of 25% when applied to species in the suborder Ceratomorpha (tapirs and rhinoceros), which diverged ca. 54 mya from the equine lineage12. On the other hand, cross-species application of the domestic dog CanineHD BeadChip to seals (divergence time ~44 mya) amplified 19% of 173 K loci, but fewer than 200 loci were polymorphic13. Another issue with cross-species use of SNP arrays (and other markers) is ascertainment bias, in which the sample of SNPs does not represent the overall pattern of genetic diversity, in this case because the SNP array was developed in a different species. This difference is likely to lead to an over-representation of polymorphisms shared between species or polymorphisms resulting from artificial selection, when SNPs originally developed for domestic species are cross-amplified in wildlife. Ascertainment bias might also lead to spatial clustering of polymorphisms around selectively non-neutral SNPs, but most commercial SNP arrays developed for domestic animals are based on random selections of evenly spaced SNPs which may help to lessen this bias13. Also, most SNPs on commercial arrays are identified so researchers can screen markers depending on their objectives.

In our study we tested the domestic chicken 600K SNP array on five North American species of Centrocercus (sage-grouse) and Tympanuchus (sharp-tailed grouse and prairie-chickens) grouse to determine if it was suitable for genetic studies in these taxa. The five species, herein collectively referred to as prairie grouse, are estimated to have diverged from chickens about 37 mya14, while the two genera are thought to have split approximately 4.7–9.9 mya15,16. The domestic chicken SNP array is based on the Gallus gallus 4.0 reference genome and contains 580 954 SNPs, including 21 534 SNPs from coding regions. The SNPs on the array are evenly spaced (mean of 1 748 bases between markers) based on genetic map distance to equalize the density (per cM) of SNPs on micro- and macro-chromosomes10. Our cross-species application of chicken SNPs took advantage of the expected high synteny between the genomes of chicken and grouse (see Methods for details) and, to the best of our knowledge, this is the first study to test the Affymetrix 600K SNP array on a species other than Gallus gallus.

Results

We successfully genotyped 91% (530 287 of 580 961) of all chicken SNPs on the array (including sex chromosomes), and 53% (308 276 of 530 287) of genotyped SNPs were found in at least one of two sampled individuals of all five species (Fig. 1). Between 73% and 76% of all SNPs on the chip were genotyped per species and 51–52% of SNPs were genotyped for both individuals per species (Table 1). Mean call rate of all genotyped SNPs was 69 ± 0.04%, and 41% of all genotyped SNPs (218 524 of 530 287) were scored in at least nine of ten samples. A total of 488 288 SNPs were genotyped in at least two species (Fig. 1), and of these only 0.2% (1192) showed fixed differences between species (here we only considered a difference as ‘fixed’ if both genotyped individuals of a species were homozygous for an allele that differed from the allele(s) genotyped in the another species). Genotyping success rate of all 21 534 chicken coding SNPs was 63–68% per species.

Figure 1
figure 1

The number of called and polymorphic (given in the parentheses) loci shared among five grouse species. GPCH – Greater Prairie-Chicken, LPCH – Lesser Prairie-Chicken, STRG – Sharp-tailed Grouse, GUSG – Gunnison Sage-Grouse, GRSG – Greater Sage-Grouse.

Full size image
Table 1 Genotyping success of 580 961 SNPs on the 600K Aftymetrix Axiom chicken array and observed heterozygosity of SNPs (HO) in five grouse species (two individuals per species).
Full size table

Among all genotyped SNPs, 77% (407 798 of 530 287) were polymorphic (had two alleles in the sample). Note that this includes SNPs with estimates from one (53% polymorphic) to ten (85%) samples. On average, 7.2 of the ten samples were genotyped across all polymorphic SNPs. Within each species, 48–54% of all SNPs were polymorphic, and 72–75% of them were genotyped for both individuals per species (Table 1). Average observed heterozygosity was 0.66 (Table 1).

For known autosomal SNPs (N = 546 120), the mean genotyping success rate was 74% averaged across species, and it was similar between all chromosome categories, ranging from 73% for intermediate-chromosomes to 75% for micro-chromosomes (all species combined). The proportion of SNPs among all genotyped loci was similar for all chromosome categories, ranging from 69% for macro-chromosomes to 71% for micro-chromosomes. By contrast, the spacing between adjacent markers differed significantly between autosomal chromosome categories (F2,133 = 159.6, P < 0.001). The largest inter-marker distances were found on macro-chromosomes, while the smallest inter-marker distances were found on micro-chromosomes, consistent with the equalization of marker densities across chromosomes (Table 2). In total, 24–26% of markers were located within 1 Kb of adjacent markers and 95–97% of markers were located within 10 Kb of adjacent markers (Figure S1 in the Electronic Supplementary Material, ESM).

Table 2 Number of SNPs and mean (±SD) distance between adjacent marker pairs for three groups of autosomal chromosomes in five grouse species. Chromosome groups are based on the chicken genome.
Full size table

We conducted a principal component analysis (PCA) to examine variation between species using 386 011 autosomal loci that were polymorphic across our samples. The first PC axis accounted for 21.5% of the variability in the SNP dataset and it distinguished between Centrocercus and Tympanuchus genera (Fig. 2). However, species within genera were not distinguished with any of the PC axes. On average, there was little separation between genera and species in terms of shared SNPs. The two genera shared 92.0 ± 0.2% of SNPs, and the species within each genus shared 93.2 ± 0.6% (Centrocercus) and 94.3 ± 0.2% (Tympanuchus) of SNPs (Fig. 3). Similar SNP sharing rates (93.0 ± 0.6% for Centrocercus and 94.8% for Tympanuchus) were recorded between individuals within species (Fig. 3).

Figure 2
figure 2

Clustering of five grouse species (two individuals per species) with principal component analysis using autosomal loci. GPCH – Greater Prairie-Chicken (open squares), LPCH – Lesser Prairie-Chicken (filled squares), STRG – Sharp-tailed Grouse (filled triangles), GUSG – Gunnison Sage-Grouse (open circles), GRSG – Greater Sage-Grouse (filled circles).

Full size image
Figure 3
figure 3

Heatmap of genetic similarity of autosomal loci between individuals of five grouse species. Yellow/orange squares indicate high allele sharing between individuals, green squares indicate low allele sharing. GPCH – Greater Prairie-Chicken, LPCH – Lesser Prairie-Chicken, STGR – Sharp-tailed Grouse, GUSG – Gunnison Sage-Grouse, GRSG – Greater Sage-Grouse.

Full size image

We also explored potential genetic differences between Centrocercus and Tympanuchus using the two genera as groups in the association test in PLINK (Fig. 4). Here we tested 350 835 polymorphic SNPs with a minor allele frequency >0.05. As a consequence of the large number of tests we performed (350 835), the Bonferroni critical level (1.26 × 10−7) was below the minimum possible P value with our sample size (7.74 × 10−6) and, consequently, none of the tests was significant after Bonferroni correction. Thus, we used the P values as a relative index of importance in separating the two genera. First, we examined the distribution of 5 807 nominally significant (P < 0.05 unadjusted) SNPs that differed between genera. These SNPs were generally distributed evenly across chromosomes relative to the total number of SNPs on each chromosome (median = 0.9% of SNPs were ‘significant’ on each chromosome; range 0.7–1.2%). The only exceptions were small (chr 16; 0.34%) and micro (LGE22C19W2; 1.9%) chromosomes, which is likely an artefact of small numbers of SNPs (<585) on these chromosomes. Next we examined the genes encompassing the most significant SNPs (i.e., P = 7.74 × 10−6) that were found in all 10 samples. In this sample of 53 SNPs, the maximum significance level was achieved because all of the individuals in one genus were homozygous for a different allele than individuals in the other genus (Table S1 in ESM). Several of these SNPs, such as Ephrin type-A receptor 7 (EPHA7) and Keratin, type I cytoskeletal 14 (KRT14) are notable because their associated gene (or a closely related one) has been implicated in feather development. To examine the possible function of these 53 SNPs, we used AgriGo for gene ontology analysis and found significant enrichment (FDR < 0.05) in 15 GO terms (Fig. 5). In terms of biological processes, there were strong differences in anatomical structure development (GO: 0048856) and morphogenesis (GO:0009653), as well as various developmental processes (Fig. 5).

Figure 4
figure 4

Manhattan plot of differences between Centrocercus and Tympanuchus grouse at autosomal loci. P values are unadjusted values from the association test in PLINK 1.9. Chromosome are delineated by altering colours. Chromosome numbers are from the Axiom Genotyping Array10 include unreferenced locations (0). There were 204 SNPs with P < 1 × 10−5 (blue reference line), although the Bonferroni critical level would be 1.26 × 10−7 for 350 837 tests.

Full size image
Figure 5
figure 5

Significant differences between sage and prairie-grouse in terms of enrichment of biological processes. Results are based on 53 SNPs with data from all 10 samples and P = 7.74 × 10−6 from the PLINK analysis of both groups. Adjusted P values (FDR < 0.05) from the Fisher’s exact test in the Singular Enrichment analysis (SEA) in AgriGo are indicated in parentheses in the top row within each box. Also in the box are the GO term identifiers, the GO Description (e.g., developmental process), and the frequencies used in the SEA analysis (bottom row). These are the number of items (UNIPROT ids) in the query list mapping to the given GO term (e.g., 14 of the items mapped to developmental process) out of the total in the query list (29 in this case) and number in the background list (i.e., 3432) relative to the total in the reference (i.e., 24599; Axiom array). Solid and dashed lines represent one and two enriched terms at both ends connected by the arrows, respectively.

Full size image

Discussion

Despite the relatively high divergence time between the domestic chicken and grouse (~37 mya14), the Affymetrix 600K SNP array produced genotypes for 91% of SNPs in at least one of the five species of grouse and 53% of SNPs across all species of grouse. Even with the loss of about half of the SNPs, the 600K SNP array has potential to produce more SNPs than typical RAD-seq methods (that usually generate <50 000 SNPs), particularly at known coding regions. For example, over 10 K coding SNPs were genotyped in each grouse species, which may encourage studies investigating functional divergence among species or populations. In fact, our cross-application of chicken SNP array in grouse allowed us to identify a set of candidate genes that may be responsible for large phenotypic differences between Centrocercus and Tympanuchus. Although the number of SNPs we genotyped here was fewer than might be gained through whole genome resequencing (millions), the loci on the array have already been characterized, and therefore less effort is required to filter and identify loci. Our results do suggest, however, that the loci included in the Affymetrix array may not be particularly suitable to infer phylogenetic relationships in grouse, as only 0.2% of all genotyped SNPs showed fixed differences between the species. While the PCA analysis conducted with autosomal loci could separate Centrocercus from Tympanuchus, it could not distinguish species within genera.

The application of SNP arrays in research on non-model vertebrates has recently become an alternative tool in addition to whole-genome sequencing and RAD-sequencing methods. Although most SNP arrays were originally developed for model or domestic organisms, those same arrays can potentially be applied to wild relatives to obtain large panels of SNP markers12. Cross-species application of SNP arrays has been particularly common in mammals3,11,13,17, but this approach has not received as much attention in avian research. As far as we are aware, our study provides the first attempt to apply a commercially available SNP array to non-passerine bird species. So far, non-commercial Illumina SNP arrays have been developed for several passerines, such as great tit Parus major5, house sparrow Passer domesticus6, and Ficedula flycatchers7. These SNP arrays allow researchers to genotype tens of thousands (10–50 K) of SNPs, but they may not be effectively applicable in non-passerine birds due to the long evolutionary times since the divergence of these lineages. Thus, genome-wide SNP genotyping in non-passerines has often been based on RAD-seq methods18,19,20,21. However, RAD-seq applications in non-model bird species do not typically exceed 100 K SNPs22,23,24, and it has been shown that RAD-seq-based genome scans are likely to miss a considerable proportion of loci under selection, as the median density of markers is ca. 4 SNPs/Mb25. Thus, the usefulness of RAD-seq data in studies of local adaptation has been recently debated25. This problem can potentially be avoided with the Affymetrix SNP array, which has a density of 400–1300 SNPs/Mb at each autosome10. In our cross-species study of prairie grouse the Affymetrix array still yielded a density of 250–500 SNPs/Mb, depending on the size of the autosome. Over 21 K annotated SNPs from the Affymetrix array have been identified in exonic regions, while 190 K SNPs have been identified in intronic regions10. Thus, considering an even distribution of SNPs across the genome, it can be expected that the majority of the estimated 20 000–23 000 chicken genes26 will be covered with the array. The array also includes an additional 266 K SNPs that have been annotated from intergenic regions of the chicken genome10. Thus, cross-application of the Affymetrix array in wild avian species may provide researchers with easier means of studying population genetics, local adaptation and genome-wide selection. It is important to note, however, that our small-scale study was designed primarily to test the feasibility of the array, and these other applications may require using greater number of samples.

Our application of the Affymetrix 600K SNP array in prairie grouse yielded surprisingly high levels of polymorphism (48–54% within species, 77% across all five species). This is in stark contrast to previous research on mammals, which showed an exponential decline of shared polymorphism with the time to the last common ancestor11. In general, mammalian taxa, which diverged >5 mya shared less than 10% of polymorphic SNPs11. Unfortunately, we are aware of no previous attempts to apply SNP arrays across phylogenetically distant avian species, but our analyses suggest that some bird lineages may be much more conservative in the retention of shared polymorphism than mammals. Although this hypothesis certainly needs more empirical testing, it appears supported by the high similarity of draft genomic sequences of several grouse species to the reference chicken genome27,28,29. This high similarity between chicken and grouse also suggests that mispriming was not a likely bias in using the SNP array. At the same time, we acknowledge that genome synteny is unlikely to provide an exclusive explanation for high levels of shared polymorphism between grouse and chicken. In mammals, very high genome synteny between closely related taxa may not necessarily be associated with high levels of shared polymorphism. For example, cross-species application of SNPs developed for the domestic sheep Ovis aries in a recently diverged (ca. 2 mya) wild species, the bighorn sheep Ovis canadensis, showed very low level of retained polymorphism (less than 2%), despite high genome synteny between the two species17.

We also acknowledge that the application of the chicken SNP array in wild grouse could carry some risk of ascertainment bias, where the cross-amplified panel of SNPs is not representative of the overall pattern of genetic diversity in the study species. This could be caused by over-representation of evolutionarily old shared polymorphisms and under-representation of rare and novel variants11. Also, application of arrays originally developed for domestic animals in wildlife could introduce an over-representation of SNPs resulting from artificial selection. While these limitations have to be carefully considered while using SNP arrays, it should also be acknowledged that the problem of ascertainment bias is not restricted to SNP arrays, but is inherent in any marker development and can be compensated for during data analysis11,30. Cross-amplified loci should always be subject to the process of initial filtering, which may include removal of loci that show low allele frequencies (usually with MAF <5%), deviate from neutrality, or show strong linkage disequilibrium (LD) within a given basepair window. The latter approach (LD based SNP pruning) is particularly useful for reducing the effects of ascertainment bias, especially when estimating differentiation measures between populations (e.g. genetic distances), inbreeding coefficients or kinships31. The low sample size in this study prevented us from exploring these possibilities in any detail, but we recommend using these filtering steps in future studies that apply SNP arrays across species based on larger sample sizes.

Successful application of the Affymetrix 600K SNP array in North American prairie grouse suggests that the same method could be used for genotyping other galliform species. Chickens (Tribe Gallini) and grouse (Tribe Tetraonini) are thought to have diverged relatively early in the radiation of the family Phasianidae. Thus, it seems likely that the Affymetrix array might be successfully applied across the entire Phasianidae clade. However, galliform species that are more distantly related to chickens, such as New World quail (Odontophoridae) or guineafowl (Numididae) would probably yield lower genotyping success than in the case of grouse. Cross-application of the array in phasianids may provide an effective tool for molecular studies of this group, including those focused on conservation related issues. In general, phasianids are one of the most threatened groups of birds, since they are often affected by direct exploitation, as well as challenged with severe habitat loss and degradation32. Nearly 40% of all 187 extant phasianid species face some level of conservation risk and 12 species have been classified by the International Union for the Conservation of Nature (IUCN) as endangered (EN) or critically endangered (CR), and thus facing a very high risk of extinction in the near or immediate future33. All five taxa studied here are of conservation concern; Gunnison sage-grouse is classified as endangered and both species of prairie-chicken are classified as vulnerable by IUCN. As a consequence of habitat loss, these species have undergone long-term and rapid population declines, and have increasingly small and fragmented ranges34. All five of the study species are under active conservation programs, including translocations of birds to restore genetic variation35, or captive-breeding and reintroduction programs19,36. Thus, we suggest that the Affymetrix 600K SNP array may become a new and efficient tool to obtain genome-wide data that are increasingly important for the management of threatened populations of grouse and other phasianids35,37,38.

Our study provided support for high levels of shared polymorphisms between grouse species, especially within genera. This pattern seems to be consistent with previous phylogenetic studies on prairie grouse that suggest a recent divergence among species within the genera Tympanuchus and Centrocercus. For example, the fossil-calibrated species tree using both autosomal and Z-linked nuclear loci estimated the time of divergence, within Tympanuchus, between the greater prairie-chicken and lesser prairie-chicken at less than 300 K years39. Also, autosomal and mtDNA loci showed extensive allele sharing among Tympanuchus species with minimal haplotype clustering corresponding with taxonomy31. A low degree of reciprocal monophyly between species within Centrocercus and Tympanuchus genera was also found based on the analysis of several innate immune genes40. This pattern was attributed to incomplete lineage sorting and female-biased introgression, as hybridization between Tympanuchus species has been widely reported in the areas of geographic overlap41,42,43 and female hybrids have been documented breeding in the wild39. Although our results are in general agreement with previous research, we explicitly acknowledge that the statistical power of our analyses was low due to small sample sizes (two individuals per species, ten individuals in total). Consequently, analyses using larger sample sizes are needed to investigate whether the Affymetrix 600k SNP array could be used to investigate phylogenetic relationships within Centrocercus and Tympanuchus. Despite these limitations, our study suggests that even with hundreds of thousands of autosomal SNPs, there may be relatively few SNPs contributing to fixed differences between prairie grouse species, especially within Tympanuchus. This view is reinforced by several recent studies that find relatively small regions of the genome are responsible for the large phenotypic differences between some closely related species of birds44,45,46. At the same time, we acknowledge that having de novo SNPs discovered directly in prairie grouse could change this pattern and reveal more fixed difference between our study species.

The sex chromosomes were excluded from our clustering analysis because we had an unbalanced mix of both sexes among species in our small sample set that impacted our ability to investigate phylogenetic relationships (samples clustered largely based on sex; Figure S2 in ESM). As sex chromosomes often show a higher divergence than autosomes between species47, they may contribute disproportionately to maintaining species boundaries, as previously reported for prairie grouse39,48. Since the chicken SNP array includes over 25 K Z-linked loci, future studies with larger sample sizes and a more balanced representation of sexes among taxa could use the array to investigate phylogenetic relationships among grouse (and other galliform taxa).

We identified differences between sage-grouse (Centrocercus) and prairie-chickens (Tympanuchus; Fig. 2), which differ most notably in body mass and plumage, even though our study had small sample sizes (four and six individuals in each genus, respectively). Sage-grouse are over two times heavier than prairie-chickens (2.4–3.9 kg versus 0.8–1 kg, respectively). In addition, sage-grouse have much longer tail feathers and exaggerated feathers around the neck (a white ruff) and head (filoplumes), which are used in mating displays49 and are not present in prairie-chickens. Recent genome scans of pigeons and doves with feather crests (or exaggerated neck feathers similar to sage-grouse) have found associations between those traits and SNPs at the Ephrin receptor B2 (EphB2) gene, which is a receptor for tyrosine kinase. The Eph family of molecules has 14 members, which play a role in tissue patterning and morphogenesis50 and, in birds, influence the development of the feather cytoskeleton51,52. In our study Ephrin type-A receptor 7 (EphA7) showed strong differences between the two genera of grouse. The A and B type receptors are generally thought to be activated by different types of ligands, but there appears to be high levels of interaction between them (e.g., between EphB2 and EphA553), suggesting that other Eph receptors could also be involved in feather development. We also found differences between the two genera of grouse at several other genes that have been associated with feather morphology (KRT14, a member of the keratin gene family which is associated with frizzled feathers; KRT7554), melanin pigmentation (FZD4 and LIMK155) and body mass (miR-1656) in birds. GO analysis revealed that these genes were primarily enriched in processes that influenced morphology and development, consistent with the large phenotypic differences between these two genera of grouse.

Although the results of our clustering and gene ontology analyses must be treated with caution because of low sample sizes, the main conclusion that emerges from our study is that the Affymetrix 600K array is a promising tool for future research on molecular ecology and conservation genetics of grouse. SNP arrays have the potential to advance studies of closely related non-model species, but currently they may be too expensive for the budgets of many researchers (e.g., $205 per sample at GeenSeek, Lincoln, NE, USA). Over time, however, these costs may decrease. One of the advantages of the array is that it enables much easier access to an annotated genome and the traits associated with variable SNPs. Such information is becoming critically important for population genetic studies of animals in the wild and endangered species, in particular. Our results should provide some encouragement for other researchers willing to investigate the utility of using commercial SNP arrays in other non-model species.

Methods

We collected samples from all five species of North American prairie grouse in the genera Centrocercus (greater sage-grouse, C. urophasianus, and Gunnison sage-grouse, C. minimus) and Tympanuchus (greater prairie-chicken, T. cupido; lesser prairie-chicken, T. pallidicinctus; and sharp-tailed grouse, T. phasianellus). Blood samples were taken for DNA extraction from two individuals per species. Samples were collected from wild populations between 1996 and 2006 in Colorado (greater sage-grouse and Gunnison sage-grouse), Kansas (lesser prairie-chicken), Minnesota (greater-prairie chicken), and North Dakota (sharp-tailed grouse). The sampling of sexes was unbalanced among species, resulting in a total of six females (two greater sage-grouse, two greater prairie-chicken, one lesser prairie-chicken, and one sharp-tailed grouse) and four males (two Gunnison sage-grouse, one lesser prairie-chicken, and one sharp-tailed grouse). Genomic DNA was extracted using protocols described elsewhere39,57,58. Sample collection complied with the current laws of the USA and with the approval of IACUC committees at Colorado Department of Parks and Wildlife, Kansas State University, and the Univ. of Wisconsin-Milwaukee.

All ten individuals from the five species were genotyped using the Axiom Genome-Wide Chicken Genotyping array (Affymetrix, Santa Clara, CA, USA, now owned by ThermoFisher Scientific) that was originally developed for the domestic chicken10. The Affymetrix 600K array was originally developed to contain a large catalogue of SNPs that are segregating within diverse populations of the domestic chicken and to address a variety of purposes, including the detection of genetic associations with complex traits, fine mapping of quantitative trait loci, detecting signatures of selection, exploitation of the linkage disequilibrium structure of the genome, and implementation of genomic selection10. The array contains exactly 580 961 biallelic SNPs across 28 autosomes, two sex chromosomes, and two linkage groups (LGE64 and LGE22C19W28_E50C23) that have not yet been assigned to a chromosome, but are expected to reside on micro-chromosomes59. Genotyping was performed by GeneSeek (Lincoln, NE, USA) using Genotyping Console Software v. 4.2 (Affymetrix). A minimum quality control score >0.82 (default), which is based on the signal to noise ratio, was used to filter each sample. The ten grouse samples were run on a single Axiom array (96 samples total) that also included domestic chicken samples. The grouse samples were analysed separately from the chicken samples following Affymetrix guidelines, because otherwise the statistical thresholds used to call SNPs would be based on the much larger chicken sample and this would lead to the exclusion of many SNPs among the grouse samples.

Although we have no direct information on the genome organization in our study grouse species, birds in general are notable for possessing high level of karyotypic conservation and synteny across chromosomes60,61. Traditionally, genomic analyses with phasianids often take advantage of their close relationship with the domestic chicken because it is well characterized in terms of genomic organization. To date, draft genomes of various grouse species28,29, including sage-grouse27, have been assembled and mapped to the reference chicken genome. For example, the draft genome of black grouse Tetrao tetrix had high synteny with the main chicken autosomal chromosomes 1–28, as well as sex chromosomes28. No major genomic rearrangements have been reported for grouse, when compared to the chicken genome27,28. Thus, based on the previous research, we assumed that genomes of our study grouse species shared high synteny with the chicken genome.

We performed a principal component analysis (PCA) of the SNP genotypes to examine patterns of genetic diversity in the 10 samples. This analysis used SNPs that were successfully genotyped in at least one individual (see Results). PCA was conducted using the snpgdsPCA function in the SNPRelate package62 developed for the R statistical environment63. The gdsfmt package62 in R was used for managing the genomic data structure (GDS) files. To avoid confounding sex and species differences in our small sample, we removed the sex chromosomes (Z and W) from the analyses comparing different taxa (n = 21 787 SNPs removed), but we included a separate analysis of the sex chromosomes in the Electronic Supplementary Material. Previous studies have indicated that the sex chromosomes are likely important in speciation and tend to differentiate faster than autosomes between species47,64. Indeed, a previous study of Centrocercus showed that the Gunnison and greater sage-grouse showed much stronger differentiation at SNPs on the Z chromosome than on the autosomes48 (see also Galla & Johnson39 for comparisons between Tympanuchus spp.). Allele sharing rates were calculated as average identity-by-state (IBS) pairwise identities using the snpgdsIBS function in SNPRelate.

Following the PCA, we used PLINK v. 1.965 to test individual SNPs for an association with taxonomy, here specifically between genera because of the PCA results (see Results). To understand the biological function of SNPs that differed between genera, we performed gene ontology (GO) analysis as implemented in Agrigo (http://bioinfo.cau.edu.cn/agriGO/index.php66) using the chicken Affymetrix array data for the genetic background. Fisher’s exact test with a false discovery rate (FDR) <0.05 was used to test for significant enrichment in each GO term. All values are presented as means ± SE.