Phylogenomics confirms most of the existing parulid taxonomy

We sequenced DNA from blood or frozen tissue of 480 individuals (S1 Table) on Illumina HighSeq and NextSeq platforms across 14 total sequencing runs. Average sequencing coverage was 3.79× per individual. To make the parulid phylogeny, we combined reads from separate individuals into species alignments mapped to the Yellow-rumped Warbler (Setophaga coronata) reference (NCBI Accession mywa_2.1), then bioinformatically extracted ~4,000 UCE loci and used them to generate a well-supported phylogenetic hypothesis for Parulidae (Fig 1). The resulting tree contains 100 species and six subspecies of the 115 species in the family, according to the Clements taxonomy [32]. We found strong support for placing Basileuterus lachrymosus in its own genus, which aligns with its previous classification, Euthlypis lachrymosa [33], but otherwise found support for all existing warbler species classifications. The two tree-building methods we used, concatenation and quartet-based gene tree reconciliation, agreed on most of the phylogeny. There were several differences, however, including relationships among some of the youngest species, as well as whether to place Catharopeza bishopi in a monotypic genus sister to the genus Setophaga or in a sub-clade of Setophaga with several other island endemic warblers (S1 Fig). Another recent warbler phylogeny used UCE target-capture data and a mix of frozen tissue and low-quality toepad samples [34]. Our concatenated tree agreed with it on the placement of every genus, including Euthlypis, with the only exception being the placement of the clade containing Protonotaria and Limnothlypis. We used our concatenated tree in downstream analyses as the warbler species tree (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Phylogeny of Parulidae estimated from 4,016 concatenated ultra-conserved element loci.

Nodes receiving less than 100% support by approximate likelihood ratio test (aLRT) are labeled with their aLRT value. Tip labels are current species designations based on the eBird/Clements checklist [32]. Genus labels reflect an updated taxonomy implicated by our results and another recent phylogenetic study of the warblers [34]. The data and code needed to generate this figure can be found at https://doi.org/10.5061/dryad.1zcrjdg3v and from NCBI at BioProject PRJNA630247.

https://doi.org/10.1371/journal.pbio.3003501.g001

Carotenoid metabolism gene BCO2 shows evidence of repeated intergeneric introgression

We previously showed that the gene for carotenoid metabolism enzyme BCO2 has introgressed among species in the genus Setophaga several times and that the introgressed variant may have originated outside the genus [27]. We were therefore motivated to study gene flow at BCO2 across the entire family to understand the full history of BCO2 evolution in warblers. First, we examined gene tree-species tree discordance at the BCO2 locus. The tree derived from a 1.8-kb region within the gene that contained 862 SNPs showed considerable discordance with the species tree, including multiple instances of discordance among genera (Figs 2A and S2). The most striking discordances from this analysis were (1) Geothlypis formosa falling on the Vermivora branch of the gene tree and (2) Cardellina pusilla falling on the Leiothlypis branch of the gene tree, along with six species of Setophaga (Fig 2B and 2C). This topology is consistent with (1) historical gene flow from Vermivora to Geothlypis and (2) from Leiothlypis to Setophaga and Cardellina, or from Leiothlypis to Setophaga and then from Setophaga to Cardellina or vice versa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Topological discordance between the concatenated ultra-conserved element species tree of parulid warblers and a 1.8-kb region of the BCO2 gene.

(A) Full co-phylogeny, with the species tree on the left and gene tree on the right. (B) The Vermivora branch of the tree, showing evidence of gene flow to Geothlypis formosa. (C) The Leiothlypis branch of the tree, showing evidence of gene flow to several Setophaga and one Cardellina species. See S2 Fig for BCO2 gene tree node support values. The data and code needed to generate this figure can be found at https://doi.org/10.5061/dryad.1zcrjdg3v and from NCBI at BioProject PRJNA630247.

https://doi.org/10.1371/journal.pbio.3003501.g002

To support this evidence of introgression with a population allele frequency-based analysis, we calculated the four-taxon introgression statistic f_d, which is an extension of the ABBA-BABA D-statistic for population samples [35], in 10-kb windows across the genome for each pair of species involved in putative BCO2 introgression. The gene trees did not provide evidence of which species within the genera were the likely donor species (Fig 2B and 2C); we therefore chose Vermivora cyanoptera and Leiothlypis ruficapilla ruficapilla as representatives. V. cyanoptera has previously been found to hybridize with G. formosa [36], and L. r. ruficapilla provided the largest population sample for the Leiothlypis introgression analysis. A window inside the BCO2 gene or within 10 kb of the gene was an outlier exceeding the 99.9th percentile in each of the comparisons except Leiothlypis–Setophaga ruticilla, where it exceeded the 99.8th percentile (Fig 3A–3G). The gene tree and f_d analyses strongly support introgression from Vermivora to G. formosa and from Leiothlypis to Setophaga and Cardellina at BCO2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. BCO2 introgression in parulid warblers.

(A–G) Plots of introgression metric f_d in 10-kb windows with Vermivora cyanoptera (A) or Leiothlypis ruficapilla (B–G) as the presumed “donor” species and the recipient labeled on each plot. Black arrows indicate where a window within or adjacent to BCO2 is a 99.9% outlier. (H and I) Plots showing the genotype at biallelic SNPs in the BCO2 region. (H) Sites with ≥90% frequency of one allele in Vermivora, ≥90% frequency of a different allele in Geothlypis (omitting G. formosa), and <10% missing data. (I) Sites with ≥90% frequency of one allele in Leiothlypis and ≥90% frequency of a different allele in Setophaga and Myioborus (omitting S. ruticilla, S. magnolia, S. fusca, S. petechia, S. discolor, and S. vitellina). Warbler illustrations courtesy of Cornell Lab of Ornithology, illustrated by David Quinn, from Birds of the World (https://doi.org/10.2173/bow). The data and code needed to generate this figure can be found at https://doi.org/10.5061/dryad.1zcrjdg3v and from NCBI at BioProject PRJNA630247.

https://doi.org/10.1371/journal.pbio.3003501.g003

Next, to obtain the precise chromosomal locations of the introgressed regions, we examined genotypes in the groups of interest and filtered to retain those fixed or nearly fixed for different alleles between the genera, excluding recipients of gene flow in the filter (Fig 3H and 3I). Evidence of introgression was apparent in a region both upstream and within the BCO2 gene from Vermivora to G. formosa and from Leiothlypis to S. ruticilla, S. magnolia, S. fusca, S. petechia, S. discolor, S. vitellina, and C. pusilla (hereafter, we refer to this group of species as “BISC,” for “BCO2 introgression in Setophaga and Cardellina”). C. rubrifrons, though it did not fall on the Leiothlypis branch of the gene tree (Fig 2A), also showed evidence of introgression in the same region but with fewer sites fixed for Leiothlypis alleles (Fig 3I).

We calculated a range of dates for the most recent common ancestor between genera that appeared to have been involved in BCO2 introgression using the branch lengths from our concatenated UCE tree and three previous estimates of the Parulidae crown age (S3 Fig) [9,37,38]. We estimated that the split between Leiothlypis and Setophaga and between Vermivora and Geothlypis occurred between 5.3 and 8.1 Mya (mean 6.7 Mya), and the split between Setophaga and Cardellina occurred between 2.6 and 4.0 Mya (mean 3.3 Mya). These divergence times are not unprecedented for hybridizing species, including in birds [39,40]. However, adaptive introgression between genera is a rare phenomenon—or at least rarely observed thus far—and further study is needed to evaluate whether these instances in warblers cover unusually great evolutionary distances.

A unique amino acid substitution is retained in BCO2 introgression recipient species

Because BCO2 variants commonly co-occur with carotenoid coloration differences in animals [22–27,41–43] and because the introgression signal at BCO2 was strong in each comparison we examined (Fig 3A–3G), we hypothesized that the introgressed regions in Setophaga, Cardellina, and G. formosa contained one or several mutations with important implications for the gene’s expression or the protein’s function.

We aligned the BCO2 coding sequences and amino acid sequences from all species to explore whether possible functional mutations affected the protein itself (S4 Fig). We found no amino acid substitutions common only to Vermivora and G. formosa. We also found none that were common to only Leiothlypis and the BISC species. However, at the third position in exon three, these species had amino acids that were distinct from the rest of the warblers: Leiothlypis had an asparagine, and the BISC species had a serine (S4A Fig). Our previous work identified this amino acid as distinct in the Setophaga BISC species compared to 68 other bird genomes [25,27]. The nucleotide sequence revealed that a G > T mutation in the third codon position, conferring an asparagine in Leiothlypis, was inherited by all of the BISC species, and they all shared a subsequent A > G mutation in the second codon position, conferring a serine (S4B Fig).

In C. rubrifrons, we generated higher coverage sequencing data for all five individuals to confirm the allelic makeup of the introgression haplotype. We examined aligned reads and found that in the immediate vicinity of this pair of nucleotides, one individual had two copies of the ancestral Cardellina haplotype, one had two copies of the introgression haplotype, and three were heterozygous (S5 Fig). This pattern is consistent with introgression that has not progressed to fixation, but deeper insights into the evolutionary forces driving the pattern in C. rubrifrons require additional population sampling.

We cannot say conclusively whether the Leiothlypis G > T mutation and/or the subsequent BISC A > G mutation are the functional variant(s) likely to have been under selection following introgression in these species, but we note that they occur in the very small 68-nucleotide stretch at the center of the introgression block where all BISC species have sites fixed for Leiothlypis alleles (Fig 3I). The ancestral lysine is large and basic, while the Leiothlypis asparagine and the BISC serine are both polar neutral; asparagine is smaller, and serine is much smaller than lysine. We also note that all BISC species show considerable carotenoid-based feather coloration, a phenotype linked in several other birds to reduced BCO2 expression [25]. Another bright yellow species, Protonotaria citrea, had a separate mutation in the first position of the same codon, conferring a glutamic acid (S4 Fig). Glutamic acid is acidic, while the ancestral lysine is basic.

Together, these facts all point to a possible role for the amino acid substitution in disrupting or altering BCO2 enzyme function or expression in the BISC warblers, leading to increased deposition of feather carotenoids. Alternatively, a nearby mutation in the noncoding sequence could alter BCO2 expression. In the 68-nt sequence described above that was common to all BISC species, there were three SNPs in the intron immediately before exon 3 that fit the same inheritance pattern as the amino acid substitution. One of these SNPs could affect BCO2 expression. In a study identifying BCO2 introgression underlying yellow plumage in manakins, Lim and colleagues [44] found that the highest concentration of introgressed alleles fell in the same intron, which could suggest a conserved role for this region in harboring transcription factor binding sites or other regulatory sequences that control when or where the gene is expressed.

Whether or not the serine substitution is the functional change under selection, the lysine-to-arginine-to-serine sequence of mutations can help inform a preliminary hypothesis of how BCO2 gene flow occurred in Parulidae. The inheritance pattern strongly suggests that there was a single introgression event from Leiothlypis into one of the BISC species; the asparagine-to-serine mutation occurred in that species; and the new BCO2 sequence was subsequently spread to the rest of the species in the group. Because Setophaga discolor and S. vitellina share very similar introgression haplotypes (Fig 3I), and because S. vitellina most likely originated as an isolated island population of S. discolor [45], the initial introgression event from Leiothlypis to the BISC species probably predates the isolation of S. vitellina on the Cayman and Swann Islands. However, it most likely occurred, at a minimum, after the split between S. ruticilla and S. citrina. From previous estimates of the Parulidae crown age [9,37,38] and branch lengths in our concatenated phylogeny, we calculated that the initial introgression from Leiothlypis occurred between 532 thousand and 1.9 million years ago (mean 679 kya–1.6 Mya). The donor species may have been the ancestor of several extant Leiothlypis species, including L. celata, L. crissalis, L. ruficapilla, L. virginiae, and L. luciae, which we estimated to have diverged between 615 and 984 kya (mean 785 kya). These dates are summarized in S3 Fig.

The precise order and timing of subsequent introgression events among the BISC species are less clear. If one BISC species received initial BCO2 gene flow from Leiothlypis and subsequently spread the haplotype to each other species in the group, the first species’ haplotype would result from one instance of introgression, while the rest would result from two. Alternatively, the haplotype may have spread several times sequentially from species to species, resulting in third-order or higher introgression haplotypes (S6A Fig). Extrapolating from the model of Veller and colleagues [46] in which introgressed DNA is rapidly purged from the recipient genome and then levels off, we would predict second- or higher-order introgression events would result in progressively smaller introgression haplotypes in the recipient species (S6B and S6C Fig). With this simple model as a guide, S. magnolia would be the likeliest recipient of the original Leiothlypis introgression, since it has the largest remaining introgression haplotype (Fig 3I). However, each species may have different demographic histories, which would render the effects of drift more or less important. Additional population sampling among these species is needed to illuminate the order and direction of historical BCO2 introgression.

Introgression of carotenoid dehydrogenase BDH1L and carotenoid ketolase CYP2J19 point to a possible single origin of red plumage in warblers

Research into the genetic basis of red coloration in birds has implicated the combined action of two genes, BDH1L and CYP2J19, in the conversion of yellow dietary carotenoids to red ketocarotenoids [28,29]. Though many warblers show carotenoid pigmentation, only seven species are believed or known to deposit ketocarotenoids in feathers [47,48], including two species of Setophaga, three species of Cardellina, and two species of Myioborus. The former two have orange plumage, and the latter five are red. The five red species predominantly occur in Central America—or, in the case of Myioborus miniatus, the red subspecies are in Central America—which raises the question of whether their similar distributions reflect a history of red plumage introgression between two or more of the species.

To test for evidence of gene flow at BDH1L and CYP2J19 in warblers, we compared the gene trees to the warbler species tree. On the gene tree derived from an 11-kb region within and upstream of BDH1L that contained 977 SNPs, M. pictus and M. miniatus miniatus, both of which have bright red underparts, fell on the branch of the tree with the red Cardellina species, C. rubrifrons, C. rubra, and C. versicolor (S7A Fig), consistent with gene flow from Cardellina to M. pictus and M. m. miniatus. In the genomes of both M. pictus and M. m. miniatus, a 10-kb window overlapping BDH1L was among the highest outliers in f_d, indicating a strong signal of introgression from Cardellina (Fig 4B and 4C). Finally, genotypes across the gene region, filtered for sites with fixed or nearly fixed differences between red Cardellina and yellow Myioborus species, showed clear regions of Cardellina ancestry in both M. pictus and M. m. miniatus within and upstream of BDH1L (Fig 4E). The major fixed region in the M. pictus introgressed haplotype encompassed the entire major fixed introgressed region in M. m. miniatus and was roughly 2 kb larger. We conclude that the most likely scenario is one in which a portion of the BDH1L gene introgressed from a red Cardellina species into M. pictus and subsequently introgressed from M. pictus into M. m. miniatus. We estimated that Cardellina and Myioborus diverged between 2.6 and 4.0 million years ago (mean 3.3 Mya) (S3 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. BDH1L and CYP2J19 introgression in red warblers.

(A) Images of the species involved, including red Cardellina species and Myioborus pictus, and M. miniatus subspecies. M. m. miniatus occurs in Mexico, and M. m. verticalis and subsimilis occur across much of South America. Other subspecies exist but were not sequenced. (B–D) Introgression metric f_d calculated in 10-kb windows. (B and C) Cardellina rubra as the presumed donor and (B) M. pictus and (C) M. m. miniatus as the presumed recipients. (D) M. pictus as the presumed donor species and M. m. miniatus as the recipient. Black arrows indicate where windows within or adjacent to BDH1L or CYP2J19 are among 99.9% outliers. (E and F) Genotypes at biallelic SNPs in the (E) BDH1L region, filtered to retain sites fixed for one allele in red Cardellina species and with ≥ 90% frequency of a different allele in yellow Myioborus species and (F) CYP2J19 region, filtered to retain sites fixed for one allele in M. pictus and both fixed for a different allele in yellow M. miniatus and with ≥90% frequency of the yellow allele in the other yellow Myioborus species. Warbler illustrations courtesy of Cornell Lab of Ornithology, illustrated by David Quinn, from Birds of the World (https://doi.org/10.2173/bow). The data and code needed to generate this figure can be found at https://doi.org/10.5061/dryad.1zcrjdg3v and from NCBI at BioProject PRJNA630247.

https://doi.org/10.1371/journal.pbio.3003501.g004

The gene tree derived from a 37-kb region within and upstream of CYP2J19 that contained 2657 SNPs did not show evidence of gene flow between Cardellina and Myioborus, but M. m. miniatus, which has red underparts, was sister to M. pictus, which also has red underparts, rather than M. m. subsimilis and M. m. verticalis, which both have yellow underparts (S7B Fig). This pattern is consistent with gene flow between M. pictus and M. m. miniatus at CYP2J19. Three windows surrounding CYP2J19 were f_d outliers exceeding the 99.9th percentile in M. m. miniatus (Fig 4D). We also examined genotypes across the gene region, filtered for sites where M. pictus had a different allele from yellow M. miniatus and 90% of other Myioborus species’ alleles. M. m. miniatus had more M. pictus alleles within and just outside the CYP2J19 gene than in flanking regions (Fig 4F). These results are all consistent with gene flow from M. pictus to M. m. miniatus at CYP2J19.

Together, these results suggest that the red underparts of several M. miniatus subspecies, including M. m. miniatus, were likely acquired through gene flow from M. pictus, and that the red underparts of M. pictus were either acquired or enhanced through gene flow from Cardellina. The geographic pattern of plumage color in M. miniatus supports this idea, as red-bellied subspecies are found in northern Central America—close to and overlapping with the M. pictus range—while yellow-bellied subspecies are found in South America, with orange-bellied subspecies in the middle [49]. In at least one instance, hybrid offspring of M. pictus and M. miniatus have been produced in Arizona, USA (https://macaulaylibrary.org/asset/31405941). In addition, the distribution of M. pictus overlaps with all three red Cardellina species in the USA and Central America. While red plumage was already known to be rare among parulid warblers, these findings raise the possibility that its occurrence in the family has a single evolutionary origin in Cardellina.

Genome-wide signatures of gene flow

The above analyses pointed to particular instances of gene introgression between warbler species. We next asked whether these introgression events were indicative of greater genome-wide gene flow between these species, or whether there were distinct broad gene flow patterns among the warbler genera. To address this, we calculated the D-statistic and f4-ratio, which estimates the admixture proportion in the genome between two taxa [50], for pairwise comparisons of the species involved in the introgression events we identified above and several other species in the same genera (S8 Fig). In some cases, species involved in color gene introgression did show a greater signal of hybridization, including red Cardellina with M. pictus, S. fusca with S. magnolia, and S. petechia with S. discolor. But in most cases, including Leiothlypis–Setophaga comparisons, Setophaga–Cardellina comparisons, and Geothlypis–Vermivora comparisons, the species sharing color genes as uncovered in this study did not show greater hybridization signals genome-wide than those not sharing color genes.

Because many of these pairings were between distantly related species (S3 Fig), hybridization between them was probably quite rare even when the color gene introgression occurred, and donor DNA was efficiently purged from the recipient genome. The fact that a signal of Cardellina–M. pictus hybridization is clearly visible in this analysis may reflect the more recent Cardellina–Myioborus divergence, which we estimate to be 2.61–4.02 Mya (mean 3.33 Mya) (S3 Fig), or it may suggest that BDH1L introgression is recent, and Cardellina DNA is still being purged from the M. pictus genome.

Introgression, sexual selection, and diversification

In many instances of adaptive introgression, the selective mechanism is clear. Introgression of wing patterns in Heliconius butterflies facilitates their mimicry [51]; introgression of resistance genes in mice protects them from the rodent poison warfarin [52]. In the cases we have identified in warblers, the most likely traits spreading via introgression are yellow and red plumage coloration, and the most likely driver of their introgression is sexual selection. These cases add substantially to known examples of apparent adaptive introgression likely driven by sexual selection. Most previous studies that discovered similar situations involved traits that had only partially introgressed into recipient populations [31,53–56], rather than those that introgressed completely and became fixed in recipients (but see Lim and colleagues [44]). The latter case may be as common or more so than the former, but only discoverable by looking for evidence of past hybridization rather than active hybrid zones. The fact that colorful plumage appears to have spread many times among warbler species raises the question of whether some color phenotypes are or were universally attractive across the family. Under this idea, once one species acquired the trait as an evolutionary novelty, it would eventually spread to relatives, as long as there was at least some interbreeding among them. Such a preference could arise through several evolutionary processes, including perceptual bias [57] or “honest signaling” [58]. As the genetics underlying sexual traits continue to be uncovered and hybridization research continues to expand beyond contemporary hybrid zones, similar situations to that in the warblers will likely be discovered, especially among animals where color signaling plays a prominent role in sexual selection.

In contrast to our finding that carotenoid-related genes have frequently introgressed between warbler species, previous studies have found that despite a small suite of genes often underlying traits related to melanism, these genes seem to repeatedly evolve in parallel to produce melanistic phenotypes rather than spread via introgression. For example, Baiz and colleagues [27] found no evidence of gene flow on the ASIP gene tree, even though ASIP divergence was associated with plumage differences among several pairs of warbler species. Selection on separate melanin genes or separate regulatory regions controlling the same gene have also been implicated in color divergence in Solomon Islands flycatchers and a radiation of South American seedeaters, respectively [59,60]. This trend requires more research to confirm, but it suggests that the evolutionary “hurdles” presented by altering carotenoid pathways are more difficult to solve than those related to melanin pathways and, therefore, more likely to be “borrowed” from a related species.

Our finding that carotenoid genes seem to readily spread between warbler species has potential implications for the group’s diversification. If the phenomenon increases the likelihood of interbreeding between species, it has the potential to counteract forces leading to speciation, at least if the species are in the early stages of divergence. Our analyses here had no ability to detect introgression between sister species, so we cannot say definitively whether this process occurred in warblers. On the other hand, if one species or population changed color due to introgression from a distant relative, this might introduce a new premating barrier with any close relatives who used color to select mates, thereby accelerating speciation and contributing to diversification. Past studies have found that hybridization can, in fact, contribute to diversification, e.g., by providing new genomic substrates for adaptation in African cichlids [6] and Hawaiian silverswords [61]. The parulid warblers may be a fruitful system to continue this area of research or even extend it: much of their radiation appears to have been characterized by divergence in sexual traits at least as much as divergence of ecological niches [62], and recent evidence from birds of paradise suggests intergeneric hybridization may not be a rare phenomenon in such a group [40,63].

Parulid warblers have been noted to produce intergeneric hybrids and more generally hybridize outside of hybrid zones fairly frequently [16,17]. The present study suggests that such crosses could have played a major role in producing the pattern of plumage color variation that makes the family so distinctive. We present evidence from three genes that can have large effects on carotenoid-based phenotypes. Similar processes may have played out across the genome, but because genome-wide signals of hybridization may not point to these events, a targeted approach appears to be necessary to detect them.

Data collection and sequencing

Samples for DNA extraction and sequencing were obtained from a combination of frozen museum tissues, blood in lysis buffer collected from mist-netted birds in the field, and, in the case of extinct Vermivora bachmanii, toepads cut from museum specimens (S1 Table). All samples collected for this study were obtained under a protocol approved by IACUC at Cornell University (protocol no. 2015-0065) or Penn State University (protocol no. 201900879) and under a USGS federal master bird banding permit (24043 to D.P.L.T.), as well as the New York State Department of Environmental Conservation (banding permit no. 156), the Pennsylvania Game Commission (banding permit 46121), and the Pennsylvania State Forest (SFRA permit 2220). DNA was extracted using Qiagen DNeasy Blood and Tissue kits. We used 2 mm³ of tissue and 75 μL of blood as inputs for digestion. The sequencing library preparation procedure for Setophaga, Catharopeza, V. cyanoptera, and V. chrysoptera DNA is outlined in [27]. The procedure for V. bachmanii is described in [64]. For all other samples, paired-end libraries were prepared using Illumina DNA Prep kits and dual-indexed, with a 350-bp insert size targeted. Libraries were sequenced across seven lanes of an Illumina NextSeq 2000. Reads were trimmed using AdapterRemoval [65] and aligned to the S. coronata reference genome using Bowtie 2 [66], and PCR duplicates were marked with PicardTools. The final average coverage across all samples was 3.79×. We evaluated whether mapping rates were uneven across the warbler tree, which could bias results that rely on phylogenetic relationships [67,68]. Mapping rates were obtained from bowtie log files. We found very consistent mapping rates across the tree (S9 Fig), with only a few species showing reduced mapping, and these species were not clustered in the tree.

Phylogenetic analysis

We only required a single sequence for each species for phylogeny-building, so to take advantage of replicate individuals per species and increase confidence in base calling via higher coverage, we combined alignment BAM files for all individuals of each species to make species-BAM files. Next, we made a master variant call format (VCF) file by running bcftools mpileup and call commands [69] with a ploidy of 2, then filtered to remove low-quality SNPs (QUAL < 10) and set low-quality genotypes (GQ < 10) to uncalled using bcftools. Finally, we filtered genotypes by read depth using bcftools. Since some species’ data varied in the number of individuals they contained, we customized read depth filtering for each species after examining depth distributions across a reduced dataset. Minimum read depth permitted ranged from 2 to 5; maximum depth permitted ranged from 20 to 125. The code with precise cutoffs for all species is available from Dryad [70].

We next extracted UCE loci and flanking sequences for each species with a custom script that used bcftools view to subset the VCF file to the region of interest, then samtools faidx [69] and bcftools consensus to extract each species’ sequence, incorporating variants from the VCF. Detailed methods for identifying UCE loci and flanking sequences in the S. coronata reference genome are described in [27]. Briefly, the Tetrapods-UCE-5kv1 probe set [71] was aligned to the S. coronata reference; only uniquely aligning sequences were retained; loci were thinned to 3 kb separation distances; and 1 kb of flanking sequence on each side was included in final fasta files for 2-kb total lengths. We aligned and trimmed the species UCE sequences using mafft [72] and GBlocks [73] implemented in PHYLUCE [74]. No UCEs were missing for any species, so we retained all 4016 loci.

We used two different methods for tree building: concatenation and species tree estimation from gene trees. For the former, we concatenated all 4016 loci and used the resulting alignment as the input to IQ-TREE [75]. We first repeated all of the above steps with just 25 species (for computational tractability) and ran the model finder function in IQ-TREE. The best supported DNA evolution model was TVM + F + I + R3, so we used that model when running the program with the full dataset. The final run also included 1,000 replicates of an approximate likelihood ratio test (aLRT) to compute branch support.

Separately, we used the individual aligned, trimmed UCE loci as inputs to runs of IQ-TREE using the same DNA evolution model as the full dataset and 1,000 aLRT replicates per tree. The resulting 4,016 trees were concatenated and used as inputs for weighted TREE-QMC, a quartet-based method for species tree estimation from gene trees [76]. Because this method does not estimate terminal branch lengths, we finally re-ran the concatenated tree procedure in IQ-TREE described above, but constrained to the TREE-QMC topology. A comparison between the two trees is displayed in S1 Fig.

In downstream analyses that required a species tree, we used the concatenated tree with estimated branch lengths from IQ-TREE. To display the phylogeny as an ultrametric tree in Fig 1, we first rooted the tree on Icteria virens, dropped the I. virens tip, then ran the chronos function from the package ape [77] in R [78] with the “clock” model of substitution rate variation among branches and a lambda of one.

To examine the effect of combining individuals of the same species together into a single BAM file before variant calling, we made a tree following the same procedure but with three individuals per species and one species per genus (with the exception of Catharopeza, for which we only had two samples). We subset the resulting tree to one branch per genus and compared relative node ages to our final concatenated UCE tree subset to the same taxa. Older nodes were at similar positions between the two trees, but the separate-individual tree’s younger nodes were comparatively older than the combined-individual tree (S10 Fig). Because we did not do further analyses to evaluate our tree-building method, we cannot say which method is superior. However, our interpretations of gene tree topologies did not rely on branch lengths, and windowed f_d analyses used separate individual datasets. The study’s conclusions were therefore unaffected by this methodological choice.

Color gene introgression analyses

To test for evidence of gene flow among species at key genes related to coloration, we compared the UCE species tree to gene trees for BCO2, CYP2J19, and BDH1L. We first generated gene annotations for the S. coronata reference genome from Zebra Finch assembly (accession GCF_003957565.2) using LiftOff with default options [79]. Gene sequences were then extracted using bcftools consensus as described above for UCE loci, then aligned with mafft. Trees were made using IQ-TREE, and in each case, we used the best supported DNA substitution model as determined by the model finder. We reduced the regions of interest to roughly 1.8-kb, 10.9-kb, and 36.9-kb stretches of DNA based on the largest mostly contiguous sequences that made up apparent introgression haplotypes for BCO2, BDH1L, and CYP2J19, respectively (S11 Fig). For CYP2J19 and BDH1L, after viewing the gene trees with all species included, we created a new version of the master VCF file described above for just the genomic regions of interest in Cardellina, Myioborus, and B. lachrymosus as an outgroup, separating the red-breasted M. miniatus samples (subspecies miniatus) from the yellow-breasted samples (subspecies subsimilis and verticalis). This was done to avoid redoing the computationally intensive step of calling variants across the entire genome for all species. Filtering for the two subspecies followed the filtering scheme described above, with read depth filtering cutoffs chosen based on the range of cutoffs for similar numbers of input individuals in the full dataset.

In several instances where species relationships on a gene tree did not match the species tree and were thus suggestive of gene flow, we ran windowed analyses calculating the introgression statistic f_d [35], an extension of the ABBA-BABA D-statistic [80] which uses population samples to estimate the proportion of a genomic window shared between species due to introgression. To run these analyses, we first called variants from individual BAM files of species of interest (in contrast to the species-combined BAM files described above). Although we were most interested in introgression at the genes themselves, we ran these analyses across the entire genome in 10-kb windows to put introgression around the gene of interest in the context of genome-wide introgression patterns. In the case of BCO2, we ran a version with V. cyanoptera as the “donor” (i.e., P3 from Martin and colleagues [35]) and G. formosa as the “recipient” (i.e., P2 from Martin and colleagues [35]) and a series with L. ruficapilla as the donor and S. ruticilla, S. magnolia, S. fusca, S. petechia, S. discolor, and C. pusilla as the recipients. For CYP2J19, we used M. pictus as the donor and M. m. miniatus as the recipient. For BDH1L, we used C. rubra as the donor and M. pictus and M. m. miniatus separately as the recipients. The presumed donor versus recipient taxa can be determined from the topology of the gene tree: the recipient taxon will fall on the branch of the tree normally occupied by the donor taxon. For example, in Fig 2C, G. formosa is the recipient, and a Vermivora species is the donor. This is because the sequence of the gene will have followed the evolutionary trajectory of the donor species before subsequently moving into the recipient species via gene flow.

We sought to estimate the divergence times between genera that showed evidence of color gene exchange. Because we did not time-calibrate the phylogeny we generated, we used a combination of our phylogeny and previous time-calibrated trees to estimate divergence times of interest. We used the warbler crown age from Barker and colleagues [37], Oliveros and colleagues [9], and Claramunt and colleagues [38], and branch lengths from our phylogeny to calculate a range of divergence times, displayed in S3 Fig.

To more precisely determine the locations of introgressed DNA in the genome, we filtered the master VCF file based on several subsets of species genotype data and visualized the resulting variants using IGV [81]. We filtered each data subset to retain sites where signatures of introgression would be most apparent: sites that were fixed or nearly so for different alleles between two lineages of interest. In each case, we used trial-and-error to adjust allele frequency and missingness thresholds to omit sites not fitting this pattern but retain as many others as possible. In each case below, we retained only biallelic SNPs. For Vermivora–Geothlypis BCO2 introgression, we subset the data to sites where Vermivora species were fixed for one allele and Geothlypis, excluding G. formosa, had ≥90% frequency of the other allele. For Leiothlypis–BISC species BCO2 introgression, we included only sites where Leiothlypis had ≥90% frequency of one allele and Setophaga (excluding the BISC species) and Myioborus both had ≥90% frequency of the other allele. We included Myioborus in the filter rather than Cardellina due to uncertainty over which Cardellina species received BCO2 introgression. In both BCO2 genotype subsets, we allowed <10% missing genotypes. For BDH1L introgression, we included sites where the three red Cardellina species were fixed for one allele and yellow Myioborus species, including yellow M. miniatus subspecies, had ≥90% frequency of a different allele. We allowed for 40% missing genotypes in red Cardellina and 30% missing genotypes in yellow Myioborus. For CYP2J19 introgression, we included only sites where M. pictus was fixed for one allele, yellow M. miniatus was fixed for the other allele, and the rest of Myioborus, excluding red M. miniatus, had ≥90% frequency of the yellow M. miniatus allele. No missing data was allowed for M. pictus or yellow M. miniatus. We allowed <30% missing genotypes in the rest of Myioborus, excluding red M. miniatus.

In order to better understand potential functional changes to BCO2 that may be related its introgression between species, we generated a coding sequence alignment for the gene using the gene alignment described above, manually extracting the exons based on the gene annotation and translating to an amino acid alignment in Geneious Prime 2021.2.2. To display the alignment in S4 Fig, we chose species that comprised most of the variation shared between any two species from our genera of interest in the gene’s first two exons. We noted that the nucleotide of greatest interest, which conferred a serine in the BISC species, was uncalled in S. vitellina, rather than called a G as in the rest of the BISC species. Viewing the individual BAM files revealed that, though there was low coverage in the region, raw reads only contained a G at the nucleotide in question, and none contained the presumed ancestral A (S12 Fig). As a result, in S4 Fig we display the S. vitellina nucleotide sequence with a G and the amino acid sequence with a serine. Similarly, at the first position of the same codon, the P. citrea genotype was uncalled, but 18 out of 19 reads at the position show a G rather than the presumed ancestral A (S13 Fig). Therefore, we display the P. citrea nucleotide sequence with a G and the amino acid sequence with a glutamic acid in S4 Fig.

Genome-wide signatures of gene flow

To evaluate genome-wide signals of gene flow, we calculated the D-statistic and f4-ratio for species pairs discovered to exchange color genes in the previous analyses and several relatives from the same genera as comparisons. We first called variants from individual BAM files using bcftools, setting a minimum minor allele count of 2 and setting to uncalled any genotype with a genotype quality of less than 10. Three variant files were created: one for Leiothlypis–Setophaga–Cardellina comparisons, one for Geothlypis–Vermivora comparisons, and one for Cardellina–Myioborus comparisons. We then ran the command dtrios from DSuite [82], providing the species topology and using 200 jackknife blocks across the genome. To display the results, we used a custom R script to select the P1 species resulting in the maximum D and f4 values for each comparison and transform the results into a matrix.