Extensive hybridization reveals multiple coloration genes underlying a complex plumage phenotype

Abstract

Coloration is an important target of both natural and sexual selection. Discovering the genetic basis of colour differences can help us to understand how this visually striking phenotype evolves. Hybridizing taxa with both clear colour differences and shallow genomic divergences are unusually tractable for associating coloration phenotypes with their causal genotypes. Here, we leverage the extensive admixture between two common North American woodpeckers—yellow-shafted and red-shafted flickers—to identify the genomic bases of six distinct plumage patches involving both melanin and carotenoid pigments. Comparisons between flickers across approximately 7.25 million genome-wide SNPs show that these two forms differ at only a small proportion of the genome (mean F_ST = 0.008). Within the few highly differentiated genomic regions, we identify 368 SNPs significantly associated with four of the six plumage patches. These SNPs are linked to multiple genes known to be involved in melanin and carotenoid pigmentation. For example, a gene (CYP2J19) known to cause yellow to red colour transitions in other birds is strongly associated with the yellow versus red differences in the wing and tail feathers of these flickers. Additionally, our analyses suggest novel links between known melanin genes and carotenoid coloration. Our finding of patch-specific control of plumage coloration adds to the growing body of literature suggesting colour diversity in animals could be created through selection acting on novel combinations of coloration genes.

1. Introduction

Coloration is a visually striking and extraordinarily variable phenotype in animals that drives both natural and sexual selection, and can ultimately drive the process of speciation [1–3]. In recent decades, biologists have been increasingly interested in connecting variation in coloration to an underlying genotype, with much of the focus placed on genes of large effect that influence whole-body coloration differences [3–10]. However, in recent years the use of anonymous genomic scans and admixture mapping has facilitated the discovery of genomic regions involved in coloration of smaller, discrete patches on the body [11–16]. Increasing empirical evidence of patch-specific control of coloration suggests extensive phenotypic diversity could be created through selection acting on novel combinations of coloration genes [17–22].

Low levels of background genomic divergence—either due to experimental crosses, recent speciation or ongoing introgression—in taxa that differ primarily in colour have allowed for the identification of candidate coloration genes in numerous systems [6,14,21,23]. However, what we know about the genes involved in coloration varies extensively depending on the type of pigment involved. The pathways involved in melanin coloration (greys, blacks, browns and dark reds) are better characterized [5], compared to carotenoid coloration (bright reds, yellows and oranges) for which only a handful of underlying genes have been identified [24,25]. This difference is due to differences in pigment acquisition—melanins are produced endogenously, while carotenoids must be acquired through the diet and are subsequently biochemically processed [24]—and the ability to study melanins in humans and other model systems [5].

Birds with low levels of background divergence have served as particularly powerful non-model systems for discovering the genetic bases of melanin and carotenoid coloration [6,8,11,12,14,18,20,26–28], as they often exhibit discrete feather patches that differ in colour and pigment type across the body [19]. Yet, despite the substantial variation in pigmentation across birds, the genetic bases of melanin and carotenoid coloration have only rarely been studied together in the same system (but see [11,14,16]), though the genes involved are not currently known to overlap in function or co-localize in the genome [3,5]. Here, we leverage the extensive natural phenotypic variation between yellow-shafted (Colaptes auratus auratus) and red-shafted (C. a. cafer¹) flickers, common woodpeckers that hybridize in North America [30], to identify the genomic underpinnings of plumage coloration and explore the connections between melanin and carotenoid pigmentation. The two flickers differ in the coloration of six distinct feather patches: wing and tail (the eponymous ‘shaft’), nuchal patch, ear coverts, throat, crown and male malar stripe (figure 1a; electronic supplementary material, table S1) [31]. The pigments vary depending on the feather patch, with melanins (throat, ear coverts, crown), carotenoids (wing and tail, nuchal patch) and both melanins and carotenoids (male malar stripe) being involved [32,33]. Previous molecular work has highlighted the very low baseline genetic divergence between these two taxa [34–39]. Importantly, there is extensive ongoing hybridization and backcrossing where the flickers meet in a secondary contact zone in the Great Plains of North America (electronic supplementary material, figure S1). Admixed and backcrossed hybrids exhibit the full range of possible trait combinations across the six feather patches [31,40,41] and occasional transgressive phenotypes [42].

Figure 1.

(a) Coloration differences between red-shafted and yellow-shafted flickers: (1) wing and tail (the eponymous ‘shaft’), (2) nuchal patch, (3) crown, (4) ear coverts, (5) throat and (6) male malar stripe. Pigmentation is based on carotenoids (wing and tail, nuchal patch), melanins (crown, ear coverts, throat) and both carotenoids and melanins (male malar stripe). Illustrations by Megan Bishop. (b) Principal component analysis (PCA) separately clusters yellow-shafted (yellow points), red-shafted (red points) and hybrid (orange points) flickers using the dataset of approximately 7.25 million genome-wide SNPs. (c) PC1 is significantly associated with overall phenotype score (ρ = 0.94, p < 2.2 × 10⁻¹⁶), where variation ranges from 0 for pure yellow-shafted flickers to 1 for pure red-shafted flickers. (d) The distribution of genetic differentiation (F_ST) between allopatric yellow-shafted flickers and allopatric red-shafted flickers across the whole genome. Individual points show the weighted mean F_ST for 25 kb windows. Chromosome positions are based on alignment to the zebra finch genome. (Online version in colour.)

Their combination of low genome-wide divergence, clear phenotypic differences and extensive hybridization makes flickers an exceptional non-model system in which to explore the genomic basis of feather coloration. Further, the variation in both melanin and carotenoid pigmentation provides an opportunity to explore the potential interactions between genes involved in both pigment types. We compare whole genomes of phenotypically admixed individuals from the hybrid zone along with allopatric red-shafted and yellow-shafted individuals. Here, we (i) assess the genomic landscape of divergence between allopatric flickers, and (ii) capitalize on a dataset of phenotypically variable hybrid flickers to perform association tests between the genomic markers and the six plumage traits. We leverage these complementary and independent approaches to identify SNPs that are significantly associated with plumage differences. We then (iii) search for candidate pigmentation genes present near these SNPs and (iv) discuss potential mechanisms connecting melanin and carotenoid genes with individual plumage patches.

2. Results and discussion

(a). The genomic landscape of divergence in flickers

We conducted whole-genome re-sequencing of 10 allopatric red-shafted, 10 allopatric yellow-shafted and 48 hybrid flickers (electronic supplementary material, table S2), resulting in approximately 7.25 million SNPs distributed across the genome. Red-shafted and yellow-shafted flickers clustered separately in a principal component analysis (PCA) with hybrids extending between the two parental taxa on PC1 (figure 1b; 2.07% of the variation) and clustering separately from them on PC3 and PC4 (electronic supplementary material, figure S2; 1.68% and 1.65% of the variation, respectively). We estimated F_ST values between the allopatric red-shafted and allopatric yellow-shafted individuals in non-overlapping 25 kb windows to search for divergent regions of the genome. Differentiation across all windows was low between the allopatric individuals (mean genome-wide F_ST = 0.008, mean autosomal F_ST = 0.007, mean Z-linked F_ST = 0.041), but we identified a number of regions with elevated F_ST estimates relative to the background (figure 1d). Across the entire dataset, we found only a small number of SNPs that were fixed (780 SNPs with F_ST = 1, 0.011% of the total) or nearly fixed (2156 SNPs with F_ST > 0.90%, 0.030% of the total).

We scored the six differing plumage patches (figure 1a) in the flickers to obtain a score ranging from 0 (yellow-shafted) to 1 (red-shafted). See Material and methods for details on the phenotypic scoring. We found that PC1 was strongly correlated with the overall phenotype score (figure 1c; ρ = 0.94, p < 2.2 × 10⁻¹⁶) and with each individual trait separately (electronic supplementary material, figure S3). Further, a PCA based on 780 fixed SNPs between allopatric red-shafted and allopatric yellow-shafted flickers resulted in the first PC axis explaining a majority of the variation (55.56%) and individuals spread along PC1 based on overall phenotype score (electronic supplementary material, figure S4). Taken together, these findings suggest that the few divergent genomic regions between allopatric flickers (F_ST peaks in figure 1d) are associated with the loci responsible for their coloration differences.

(b). Multiple, discrete genomic regions shape the complex plumage phenotype

We took advantage of the plumage trait variation among hybrid flickers to conduct genome-wide associations (GWAs) for each of the six plumage patches to test whether particular F_ST divergence peaks were associated with plumage coloration (see electronic supplementary material, figures S5 and S6 for illustrations and trait variation of hybrids). By focusing only on hybrid individuals, the results of our GWA analyses are independent from our assessment of genomic divergence between allopatric individuals (shown in figure 1d). Because red-shafted and yellow-shafted flickers do not differ ecologically [30] and hybrid flickers with variable trait combinations were sampled from the same geographical transect, we expect any associations identified in the GWAs to be related to differences in plumage coloration. 368 SNPs (0.005% of the total) were significantly associated with plumage patches using a significance threshold of α = 0.0000001 (−log₁₀(α) = 7), with 19 SNPs identified in more than one analysis (figure 2; electronic supplementary material, table S3 and figure S7). We found significant associations between multiple SNPs and plumage for four of the six focal traits, excluding throat colour (only 1 SNP identified) and crown colour (no SNPs identified). We validated our associations to ensure the identified regions represent real associations between plumage patches and genotype using randomized GWA analyses (electronic supplementary material, figure S8; see Material and methods for details).

Figure 2.

Results from the genome-wide associations (GWAs) of hybrid flickers comparing individual SNPs with coloration differences in the six plumage patches. Pigment type is indicated by the square next to the trait name (red = carotenoid, black = melanin, red and black = carotenoid and melanin). Some peaks of significant SNPs are present in GWAs of multiple phenotypic traits, while other peaks are unique to a single GWA, suggesting multiple mechanisms influence coloration in flickers. The red line represents the significance threshold of −log₁₀(p) = 7. Chromosome positions are based on alignment to the zebra finch genome. For visualization purposes we show only points with −log₁₀(p) > 2.5. (Online version in colour.)

The GWA analyses revealed several genomic regions that were significantly associated with the coloration of the wing and tail, nuchal patch, ear coverts and male malar stripe (figure 2; electronic supplementary material, table S3). In several cases, we identified regions of the genome that were significantly associated with multiple plumage traits (e.g. at the end of chromosome 1 and the beginning of chromosome 3). However, we also identified regions of the genome that were unique to a single GWA analysis (e.g. associations between wing and tail colour and regions on chromosomes 5, 8 and 12). These findings suggest multiple mechanisms influencing coloration in flickers: some genomic regions exert pleiotropic control over the coloration of multiple plumage patches, while other genomic regions control the coloration of a single plumage patch (perhaps as loci of large effect). The presence of potential genetic incompatibilities influencing wing and tail colour in hybrid flickers makes understanding the interactions between these genomic regions of particular importance [42]. By taking complementary, yet independent, approaches in the GWAs and F_ST analyses, we find that genomic regions identified in the GWAs of hybrid flickers largely lie within regions of the genome with elevated F_ST between allopatric flickers (peaks in figure 1d; electronic supplementary material, figure S9). However, not all genomic regions with elevated F_ST were associated with variation in coloration (e.g. the first peak on chromosome 4A, the peak on chromosome 10 and multiple peaks on the Z chromosome).

(c). Melanin and carotenoid genes both associate with carotenoid plumage in flickers

To identify candidate genes associated with plumage variation, we searched for all genes within 20 kb of SNPs that were significantly associated with plumage patches. Using this approach, we identified a total of 112 genes (electronic supplementary material, table S4). Here, we highlight 12 genes (table 1 and figure 3) that are known or suspected to be involved in melanin or carotenoid pigmentation in other systems: seven of these candidates are known to be directly involved with pigmentation [6,8,43], three are suspected to be involved in pigmentation based on the function of related genes [8] and two were identified in previous associations with feather coloration in birds [11,26].

Table 1.

Candidate coloration genes located within 20 kb of SNPs that were significantly associated with plumage patches in the GWAs. Pigment types of the trait(s) significantly associated with the SNPs are indicated by the coloured squares (red, carotenoid; black, melanin; red and black, carotenoid and melanin). The full list of identified genes is presented in electronic supplementary material, table S4.

gene	figure	chromosome	associated trait(s)	rationale
EED	figure 3a	1	nuchal	known melanin gene [8]
PLCB1	figure 3b	3	malar	known melanin gene [8]
PLCB4	figure 3b	3	malar, wing and tail	known melanin gene [8]
CYP2J19	figure 3c	8	wing and tail	known carotenoid gene [6,43]
SEMA3B	figure 3d	12	wing and tail	related to known melanin gene family (SEMAs) [8]
MFSD12	figure 3e	28	ear coverts, wing and tail	candidate melanin gene [23,26]
FKBP8	figure 3f	28	malar, wing and tail	known melanin gene [8]
RAB8A	figure 3f	28	malar, wing and tail	related to known melanin gene family (RABs) [8]
MYO9B	figure 3f	28	malar, wing and tail	related to known melanin genes (MYO5A, MYO7A) [8]
PAM	figure 3g	Z	malar	known melanin gene [8]
APC	figure 3h	Z	malar	known melanin gene, candidate carotenoid gene [8,12]
RGP1	figure 3i	Z	malar	candidate melanin gene [11]

Figure 3.

Patterns of genetic differentiation and GWA significance around nine genomic regions of interest containing 12 candidate coloration genes (table 1). Significance values from the GWA analyses are shown as points coloured by the analyses they were identified in (legend in (a)). The horizontal red lines indicate the GWA significance threshold of −log₁₀(p) = 7. Weighted mean F_ST between allopatric red-shafted and allopatric yellow-shafted flickers estimated in 25 kb (b,d,f,h) or 5 kb (a,c,e,g,i) windows are shown as black lines in each panel. Genes contained within the plotted area are shown as bars at the bottom of each panel, with the red bars indicating the locations of focal genes. When multiple focal genes are located within a single panel, they are listed at the top of the panel in the order of their physical location (from left to right). Pigment types of the trait(s) significantly associated with the focal genes are indicated by the squares under the gene names (red = carotenoid, black = melanin, red and black = carotenoid and melanin). Chromosome positions are based on alignment to the zebra finch genome. (Online version in colour.)

We find a strong association between wing and tail colour and the genomic region on chromosome 8 containing the gene CYP2J19 (figure 3c, table 1), which codes for a cytochrome P450 enzyme. CYP2J19 upregulation via an introgressed variant is causal in changing the typical yellow-feathered canary (Serinus canaria) into the ‘red factor’ canary [6] and the lack of a functional copy in zebra finch (Taeniopygia guttata) is implicated in the ‘yellowbeak’ phenotype in which the normally red beak and legs are instead yellow [43]. It is currently one of only two genes known to be involved in red coloration in birds, and evidence of its functioning in natural systems is increasing [44–47]. Our identification of CYP2J19 in the GWA for wing and tail coloration suggests that it mediates this yellow versus red trait difference in flickers and provides further support for its importance in red coloration across diverse avian lineages.

The majority of our identified candidate genes for carotenoid plumage patches in the flickers (table 1) are known or suspected to affect melanin pigmentation in other organisms (henceforth ‘melanin genes’). In some cases, we find melanin genes are associated with both melanin and carotenoid traits in a single region of the genome (PLCB1 and PLCB4 (figure 3b) on chromosome 3, and MFSD12 (figure 3e) and FKBP8, RAM8A, and MYO9B (figure 3f) on chromosome 28), while in other cases, we find melanin genes associated with a single trait (malar stripe) that uses both pigment types (PAM (figure 3g), APC (figure 3h) and RGP1 (figure 3i) on the Z chromosome). Most unusually, we identify two regions containing known or suspected melanin genes—EED (figure 3a) on chromosome 1 and SEMA3B (figure 3d) on chromosome 12—that are associated only with carotenoid-based traits. To our knowledge, of these 12 melanin genes only APC has previously been linked to carotenoid pigmentation (in an associational study [12]), in addition to its known link to melanin pigmentation [8].

(d). Potential mechanisms linking melanin genes with carotenoid traits

Melanin and carotenoid pigmentation derive from different biochemical pathways [48] and the genes involved in the different processes are not currently known to co-localize in the genome or exert influence over each other [3,5]. Thus, our finding in flickers of repeated associations between different carotenoid traits and melanin genes was unexpected. Although we lack a complete annotation of the flicker reference genome and therefore may have missed some causal genes, we repeatedly found associations between known carotenoid traits and melanin genes from different regions of the genome. Here, we outline three non-mutually exclusive explanations for these associations that link melanin genes with carotenoid plumage coloration.

First, some of the association patterns we identify in the GWAs suggest pleiotropic effects of melanin genes: we find multiple plumage patches (carotenoid and melanin) associated with the same region of the genome (figure 3b,e,f). This could occur through regulatory genes typically involved in melanin pigmentation evolving to control the expression of both melanins and carotenoids. Similar pleiotropy has been found in two different warbler species (Setophaga), where associations between a single genomic region and multiple aspects of carotenoid and melanin plumage differences have been identified [11,16]. The finding of possible pleiotropic effects on melanin and carotenoid plumage in woodpeckers and warblers, distantly related bird taxa, suggests pleiotropic effects of melanin genes may be widespread.

Second, melanin genes could be associated with carotenoid traits because the trait differences we observe are actually due to a combination of both pigments. Melanin genes associated with the male malar stripe (figure 3b,f–i) exemplify this mechanism: red pigments are present in the malar stripes of both red-shafted and yellow-shafted flickers, and yellow-shafted flickers subsequently overlay melanin to produce a black malar stripe that masks the red pigment [32,49]. Beyond this one situation where melanic pigments completely overlay a carotenoid trait, it is also possible that the two pigments are used in concert within the feathers to produce the observed colour (as in [50,51]). Additionally, melanins serve a number of other functions in feathers apart from coloration (e.g. feather structure and stability [52,53], UV protection [54], resistance to bacterial degradation [55]), so differences in patterns of plumage pigmentation relating to these other factors could also exist.

Finally, rather than controlling the upregulation and co-deposition of melanins with carotenoids, our results suggest the intriguing possibility that these genes may instead control the absence of melanin within the feathers. A reduction of melanin is necessary for the bright red and yellow coloration to be visible, and it is possible that the two taxa have differential levels of melanin reduction in their feathers or that they use different molecular pathways to reduce melanin deposition. In particular, the associations between the nuchal patch with EED (figure 3a) and the nuchal patch and wing and tail colour with SEMA3B (figure 3d), suggests the potential for melanin genes to play a direct role in carotenoid traits. This finding opens up a novel area of inquiry aimed at understanding the interactions between melanin genes and the production and display of carotenoid traits. Exploring differences in gene expression in these coloured feather patches could help to better understand the mechanisms underlying these associations.

3. Conclusion

The extensive hybridization between red-shafted and yellow-shafted flickers, in combination with their clear phenotypic differences, has allowed us to separately connect phenotypic differences with individual genomic regions. Here, we identify a complex relationship linking pigmentation genes with modular plumage patches: we find that some genomic regions associate with multiple plumage patches, while others associate with a single plumage patch. We provide evidence for a novel link between known melanin genes and carotenoid traits, and additionally identify CYP2J19 as a strong candidate related to red versus yellow coloration differences. The patch-specific control of plumage coloration that we identify here, and increasingly found in other systems (reviewed in [19]), suggests the possibility that colour diversity across birds could be created through the selection to produce novel combinations of coloration genes each exerting control on a separate body patch.

4. Material and methods

(a). Sample collection and plumage scoring

We obtained tissue samples from allopatric yellow-shafted flickers from New York (n = 5) and Florida (n = 5), and allopatric red-shafted flickers from Oregon (n = 5) and California (n = 5). These allopatric samples allowed us to characterize genomic differentiation between the flickers far from the region of current hybridization. Additionally, we sampled flickers with variable phenotypes (n = 48) from a sampling transect across the hybrid zone in Nebraska and Colorado following the Platte River. See electronic supplementary material, table S2 for details on included samples.

Red-shafted and yellow-shafted flickers differ in colour across six distinct plumage characters: wing and tail, nuchal patch, crown, ear coverts, throat and male malar stripe (figure 1a) [30]. Hybrids exhibit various combinations of parental traits and traits intermediate to the parental traits (electronic supplementary material, figure S5). We scored plumage characters on a scale from 0 (pure yellow-shafted) to 4 (pure red-shafted) following a protocol slightly modified from [31] (electronic supplementary material, table S1), an approach that has been taken in many previous studies of the northern flicker hybrid zone (e.g. [40,41,56,57]). We additionally calculated an overall phenotype score by adding the scores for the six individual traits and standardizing to range from 0 to 1 to include both sexes (as all females lack a malar stripe). To ensure consistency, all scoring was conducted by a single individual (SMA). Hybrid flickers were chosen for genotyping in this study to maximize power in the GWA analyses: we selected a panel of hybrids that exhibited high variation in their combination of plumage traits (electronic supplementary material, figure S6).

(b). Reference genome assembly and annotation

We sequenced and assembled the genome of a male yellow-shafted flicker (CUMV 57446). DNA was extracted using the Gentra Puregene Tissue Kit following the manufacturer's protocol (Qiagen, California, USA) to isolate high molecular weight DNA. Three libraries were prepared and sequenced by the Cornell Weill Medical College genomics core facility—one 180 bp fragment library and two mate-pair libraries (3 kb and 8 kb insert size)—across three lanes of an Illumina HiSeq2500 using the rapid run mode. The two mate-pair libraries were multiplexed on a single lane, while the fragment library was run across two lanes. The three lanes of sequencing generated approximately 481 million raw paired-end reads.

We assembled the reference genome using ALLPATHS-LG v. 52488 [58] and assessed the quality of the assembly using QUAST v. 4.0 [59] and BUSCO v. 3.1.0 [60]. The reference assembly had a total length of 1.10 Gb distributed across 22 654 scaffolds with an N50 of 1.57 Mb. Using BUSCO, we searched for a set of 2586 conserved, single-copy orthologues found across vertebrates. Our flicker reference genome contained a single, complete copy of 87.2% of these genes and a fragment of an additional 6.7%. Of the remaining genes, 0.4% were identified multiple times and 5.7% were completely missing. To obtain more precise information on chromosome position, we additionally assigned individual scaffolds to chromosomes based on assignments in the Ensembl zebra finch (T. guttata) reference genome v. 3.2.4 release 91 [61] using the ‘Chromosemble’ function in Satsuma [62].

We annotated the reference genome using the MAKER v. 2.31 pipeline [63]. We first used RepeatModeler v. 1.0 [64] to generate a library of repetitive sequences present in the assembly and RepeatMasker v. 4.0 [65] to soft mask these repeats. We then produced gene models by running two iterations of MAKER: the first iteration produces ab initio gene predictions, while the second iteration uses the gene models predicted from the first to improve performance. We used the Ensembl expressed sequence tags (ESTs) and protein database from the zebra finch (v. 3.2.4 downloaded July 2017) [61] to train MAKER. This pipeline annotated a total of 12 141 genes (62.4% of the total). 97.3% of the proteins predicted by MAKER matched zebra finch proteins using BLAST [66].

(c). Low coverage re-sequencing and variant discovery

We performed low coverage re-sequencing of the genomes of 68 flickers. Genomic DNA was extracted from each sample using DNeasy blood and tissue extraction kits (Qiagen, California, USA) and DNA concentrations were determined using a Qubit fluorometer (Life Technologies, California, USA). We used 200 ng of DNA from each sample to prepare individually barcoded libraries with a 550 bp insert size following the protocol for the TruSeq Nano DNA Library Prep kit (Illumina, California, USA). The libraries were pooled into three groups and sequenced separately on an Illumina NextSeq500 lane (2 × 150 bp) at the Cornell University Biotechnology Resource Center.

We assessed the quality of individual libraries using FastQC v. 0.11.5 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and subsequently performed trimming, adapter removal and initial quality filtering using AdapterRemoval v. 2.1.1 [67]. We required a minimum Phred quality score of 10 and merged overlapping paired-end reads. Filtered sequences were aligned to the northern flicker reference genome using Bowtie2 v. 2.2.8 [68] with the very sensitive, local option. Alignment statistics were obtained from Qualimap v. 2.2.1 [69]; the average alignment rate across all samples was 92.3%. Alignment rates were comparable across the different taxa: red-shafted flickers (91.9%), yellow-shafted flickers (91.8%) and hybrid flickers (92.6%). After filtering and aligning to the reference, the mean depth of coverage was 4.1X (range: 1.6X–11.4X).

All resulting SAM files were converted to BAM files, sorted and indexed using SAMtools v. 1.3 [70]. We used Picard Tools v. 2.8.2 (https://broadinstitute.github.io/picard/) to mark PCR duplicates and subsequently realigned around indels and fixed mate-pairs using GATK v. 3.8.1 [71]. Variant discovery and genotyping for the 68 flickers was performed using the unified genotyper module in GATK. We used the following hard filtering parameters to remove variants from the output file: QD < 2.0, FS > 40.0, MQ < 20.0 and HaplotypeScore > 12.0. Subsequently, we filtered out variants that were not biallelic, had a minor allele frequency less than 5%, had a mean depth of coverage less than 3X or greater than 50X, or had more than 20% missing data across all individuals in the dataset. This pipeline produced 7 233 334 SNPs genotyped across all 68 flickers. We repeated the analyses with a variety of other SNP calling tools, including ANGSD [72] and the haplotype caller module in GATK [71]. We obtained qualitatively similar results across all analyses, and so here choose to present results from SNP calling with unified genotyper in GATK.

(d). Population genomic analyses

We visualized genetic clustering in the SNP dataset by performing a PCA using the ‘snpgdsPCA’ function in the SNPRelate package [73] in R v. 3.5.2 [74]. We characterized genome-wide patterns of divergence between allopatric red-shafted and allopatric yellow-shafted flickers by calculating F_ST using VCFtools v. 0.1.16 [75] across 5 and 25 kb windows and individual SNPs. We visualized the results using the ‘manhattan’ function in the qqman package [76] in R.

(e). Genotype–phenotype associations

We used GEMMA (Genome-wide Efficient Mixed Model Association) v. 0.98 [77] to associate genotypic variation at SNPs with variation in the six plumage traits for the 48 hybrid flickers while controlling for levels of relatedness. The GEMMA analysis requires a complete SNP dataset, so we first used BEAGLE v. 4.1 [78] to impute missing data in the dataset. We transformed the imputed dataset into binary PLINK BED format using VCFtools v. 0.1.16 [75] and PLINK v. 1.09 [79]. We calculated a relatedness matrix in GEMMA using the centred relatedness matrix option (-gk 1). We conducted separate univariate linear mixed models for each phenotypic trait and used the Wald test (p_wald) with a significance threshold of α = 0.0000001 (−log₁₀(α ) = 7) to identify significant associations between SNPs and phenotypes. To visualize the results, we used the ‘manhattan’ function in the qqman package [76] in R.

To validate the resulting associations, we repeated the GEMMA analysis using a dataset with randomized phenotypes. Instead of generating artificial phenotypic scores, we retained the true phenotypic scoring across all plumage traits, but randomized the individual assignment. If the GEMMA analysis was identifying real associations between genotype and phenotype, we expected few SNPs to exceed our significance threshold in this randomized analysis. In strong contrast to the true results, we found only five significant SNPs across the six randomized analyses and no clustering of significant SNPs in any genomic region (electronic supplementary material, figure S8).

(f). Functional characterization of candidate genes

We compiled a list of genes within a 20 kb buffer of SNPs significantly associated with phenotype using Geneious v. 11.1.5 [80]. To characterize putative candidate genes, we used ontology information from the zebra finch Ensembl database [61] and functional information from the Uniprot database [81]. We additionally compared the identified list of genes to known genes involved in pigmentation. We were able to compare our gene list to 428 genes known to be involved in melanin pigmentation [8], and searched for the three gene families known to be involved in carotenoid pigmentation (β-carotene oxygenases, scavenger receptors and cytochrome P450s) [24] and genes identified in recent analyses of pigmentation in other bird species [11,12,14,16,26].

Ethics

This research was approved by the Cornell University Institutional Animal Care and Use Committee (protocol no. 2015-0065). Field permits were provided by the US Geological Survey (permit no. 24043), Colorado Parks and Wildlife (licence no. 16TRb2098) and Nebraska Game and Parks Commission.

Data accessibility

Reference genome and re-sequencing data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.7wm37pvrj [82]. Scripts for all analyses are available at https://github.com/stepfanie-aguillon/.

Authors' contributions

S.M.A. and I.J.L. conceived the study. S.M.A. collected the data. S.M.A. analysed the data with input from J.W. S.M.A., J.W. and I.J.L. wrote and reviewed the manuscript.

Funding

This work was supported by the Cornell Lab of Ornithology Athena Fund, the Garden Club of America Frances M. Peacock Scholarship, the Cornell University EEB Richard G. Harrison Fund, the Cornell University EEB Betty Miller Francis Fund, the Cornell University Andrew W. Mellon Student Research Grant and the Cornell Sigma Xi chapter (all to S.M.A.). S.M.A. was supported by the US National Science Foundation Graduate Research Fellowship Program (DGE-1144153) and the AAUW American Dissertation Fellowship.

Competing interests

We declare we have no competing interests.