Implications for evolutionary trends from the pairing frequencies among golden‐winged and blue‐winged warblers and their hybrids

Abstract Extensive range loss for the Golden‐winged Warbler (Vermivora chrysoptera) has occurred in areas of intrusion by the Blue‐winged Warbler (V. cyanoptera) potentially related to their close genetic relationship. We compiled data on social pairing from nine studies for 2,679 resident Vermivora to assess evolutionary divergence. Hybridization between pure phenotypes occurred with 1.2% of resident males for sympatric populations. Pairing success rates for Golden‐winged Warblers was 83% and for Blue‐winged Warblers was 77%. Pairing success for the hybrid Brewster's Warbler was significantly lower from both species at 54%, showing sexual selection against hybrids. Backcross frequencies for Golden‐winged Warblers at 4.9% were significantly higher than for Blue‐winged Warblers at 1.7%. More frequent backcrossing by Golden‐winged Warblers, which produces hybrid phenotypes, may contribute to the replacement of Golden‐winged by Blue‐winged Warblers. Reproductive isolation due to behavioral isolation plus sexual selection against hybrids was 0.960. Our analyses suggest that plumage differences are the main driving force for this strong isolation with reduced hybrid fitness contributing to a lesser degree. The major impact of plumage differences to reproductive isolation is compatible with genomic analyses (Current Biology, 2016, 26, 2313), which showed the largest genetic difference between these phenotypes occurred with plumage genes. These phenotypes have maintained morphological, behavioral, and ecological differences during two centuries of hybridization. Our estimate of reproductive isolation supports recognition of these phenotypes as two species. The decline and extirpation of the Golden‐winged Warbler in almost all areas of recent sympatry suggest that continued coexistence of both species will require eco‐geographic isolation.


| INTRODUC TI ON
A central goal of evolutionary biology is to elucidate the processes that generate and maintain the diversity of life, especially those responsible for the origin of species. Evolutionary biologists have long recognized that understanding how reproductive isolating barriers evolve and reduce gene flow between diverging lineages is essential for understanding the origin of species (Sobel, Chen, Watt, & Schemske, 2010). While a number of methodological approaches are used to study speciation, most provide only indirect information on the isolating barriers and evolutionary mechanisms driving speciation (Coyne & Orr, 2004). Consequently, there has been a growing call for more studies that directly estimate the degree to which different isolating barriers reduce gene flow between diverging lineages in nature (Coyne & Orr, 2004;Schemske, 2010;Sobel et al., 2010).
One especially powerful approach for understanding the mechanistic basis of speciation is to estimate the strength of reproductive isolating barriers between sympatric lineages that are incompletely reproductively isolated (Coyne & Orr, 2004;Nosil, 2012;Sobel et al., 2010;Sobel & Streisfeld, 2015). By focusing on lineages that have not yet evolved complete reproductive isolation, one can identify the isolating barriers that reduce gene flow and thus contribute to speciation, as opposed to barriers that evolve after speciation (Coyne & Orr, 2004;Sobel & Streisfeld, 2015). Such an approach can be particularly informative if the lineages in question have been well-characterized genomically. Because historical and ongoing gene flow may homogenize neutral regions of the genome, the genomic regions and traits that contribute to reproductive isolation can be distinguished (Poelstra et al., 2014). Subsequently, the strength of isolating barriers hypothesized to be affected by these traits can be estimated, allowing evaluation of causal links between genomic divergence, trait divergence, reproductive isolating barriers, and ultimately speciation (Seehausen et al., 2014).
Here, we investigate the strength of reproductive isolating barriers in a pair of closely related bird species. Previous work on Golden-winged and Blue-winged Warblers found 3% sequence divergence in the mitochondrial (mtDNA) genomes of these species, with the contemporary distribution of mtDNA lineages corresponding to allopatric populations (Gill, 1997(Gill, , 2004Shapiro, Canterbury, Stover, & Fleischer, 2004). These data suggest that divergence between Golden-winged and Blue-winged warblers was initiated in allopatry roughly 1.5 million years ago (Gill, 2004;Weir & Schluter, 2008). In contrast to these high levels of mtDNA divergence, recent whole-genome sequencing revealed very low divergence in the nuclear genome, with only six small regions showing strong divergence (Toews et al., 2016). Therefore, despite having diverged in allopatry roughly 1.5 million years ago (close to the ~2 million years of geographic isolation required, on average, for bird speciation; Price, 2008) and current high levels of geographic isolation and morphological divergence, introgression between Golden-winged and Blue-winged Warblers in the zone of recent sympatry appears to be high. Toews et al. (2016) noted that of the six genomic regions that are highly divergent between Golden-winged and Blue-winged Warblers, and four were identified as being involved in feather development or pigmentation. Consequently, reproductive isolating barriers affected by plumage divergence (i.e., those related to mate choice) may be quite strong in this system. Indeed, strong reproductive isolation based on plumage differentiation may be a primary mechanism that has maintained the distinctiveness of these lineages, especially since there appear to be no intrinsic or ecologically based reductions in hybrid fitness (Vallender, Friesen, & Robertson, 2007).
Most field studies of interbreeding by Golden-winged and Bluewinged Warblers have omitted quantitative analyses of reproductive isolating barriers, in part due to low sample sizes of social, hybrid pairs. Some studies have found evidence for behavioral isolation (Confer & Larkin, 1998;Ficken & Ficken, 1968b) and sexual selection against hybrids (Confer & Tupper, 2000;Ficken & Ficken, 1968a, 1968bLeichty & Grier, 2006), while others have not (Vallender et al., 2007). We address this uncertainty by compiling data from nine published studies across eight localities on social pairs of Golden-winged Warblers, Blue-winged Warblers, and their hybrids.
This comprehensive dataset allowed us to provide robust estimates of the strength of behavioral isolation and sexual selection against hybrids: the two reproductive isolating barriers that should be directly tied to the plumage and genomic divergence between these species. We further test the effect of plumage divergence on reproductive isolation by quantifying the relationship between plumage divergence and the frequency with which individuals of the two parental species and their hybrids form social pairs.

| ME THODS
We compiled data on social pairs from studies published by five of the authors. In addition, we included data from Ficken (1968a, 1968b), and from Will (1986) with supplemental data from Will (personal communication). This provided a total of nine, chronologically distinct studies in eight study areas. For each study area and for pooled data, we compiled the frequency of social pairing for Golden-winged and Blue-winged Warblers and hybrid phenotypes.
Not all studies could be used for all calculations because of limitations in the recorded data. Ficken (1968a, 1968b) compiled phenotypic pairing frequencies and pairing success rates for Vermivora spp. during four seasons spanning 7 years (1961)(1962)(1963)(1964)(1965)(1966). The habitat was a single successional site with an elevation range of 284-315 m. Will (1986) monitored pairing by Vermivora spp. for 3 years  within old field habitat. The study area consisted of one site with an elevation range of 205-209 m. We compiled pairing success frequencies for his study using data from Will (1986) and supplemental information (Will, personal communication).

| Old field succession in Oswego County, New York
Confer and Larkin (1998) described pairing frequencies by Vermivora spp. over seven consecutive years (1988)(1989)(1990)(1991)(1992)(1993)(1994) across 21 sites where elevation ranged from 80 to 130 m. The sites provided dry successional habitat although some predominately dry sites included adjacent ephemeral wetlands. Unpaired birds were not determined for this study and these results could not be used to calculate pairing success rates.

| Diverse habitats in Orange County, New
York (1998-1999) Confer and Tupper (2000) observed pair formation for resident, male Golden-winged and Brewster's Warblers in Sterling Forest State Park. Study sites (n = 6) ranged in elevation from 200 to 350 m and included utility rights-of-way and other successional habitats. Data from this study were insufficient to calculate pairing success rates or hybridization for male Blue-winged Warblers, but were used to calculate the frequency of primary hybridization and the frequency of backcrossing by Golden-winged Warblers.

| Lightly grazed pastures in Randolph and Pocahontas Counties, West Virginia
Phenotypic pairing frequencies and pairing success rates were monitored at 14 sites during 2008-2014 in grazed pastures described by Aldinger et al., 2014, Aldinger, 2018. Sites were at 800-1,000 m elevation in Randolph County and at 700-1,250 m in Pocahontas County.

| Managed forest in Pike and Monroe Counties, Pennsylvania
In Pennsylvania's Delaware State Forest, Vermivora spp. pairing was monitored across seven managed forest sites ranging from 400 to

| Hybrid phenotypes and genotypes
Lawrence's Warbler (Vermivora lawrencii, Herrick (1874)). Parkes described the color patterns as if they were due to two genes each having a dominant and a recessive allele. This two gene model provides a fairly accurate predictor of the pattern of phenotype inheritance (Toews et al., 2016), although it is insufficient to explain occasional intermediate phenotypes. According to Parkes' model, Brewster's Warblers are the F1 product of primary hybridization between genetically pure Golden-winged and Blue-winged Warblers, but can also result from matings of other genotypes within the Golden- While there is a strong correlation between phenotype and genotype of individuals in this system (Toews et al., 2016), it is important to note that some phenotypically "pure" individuals show signs of introgression in their genetic background (Dabrowski, Fraser, Confer, & Lovette, 2005;Vallender et al., 2009;Wood et al., 2016).
The presence of these "cryptic hybrids" will inflate our estimates of reproductive isolation (see below) and overestimate the reduction in gene flow due to a given barrier. Nonetheless, assortative mating by plumage phenotype and/or sexual selection against males with intermediate phenotypes would still act to reduce gene flow between lineages, thus promoting speciation. The main goal of this study was to determine whether there is nonrandom mating based on these phenotypic differences, and thus whether patterns of mating in the field are consistent with patterns of genomic divergence primarily in regions related to plumage development (Toews et al., 2016). We note that an imperfect relationship between phenotype and genotype is precisely what is expected in systems that are in the early stages of speciation (Dobzhansky, 1958;Roux et al., 2016) and thus not unique to Golden-winged and Blue-winged Warblers.

| Residency, pairing, primary hybridization, and backcrosses
Males were considered as a resident at each study area if they were heard or seen on at least 3 days over a week's span of time within an area approximately the size of Vermivora spp. territories (e.g., Confer, Allen, & Larkin, 2003). Almost all males were seen over a much longer period. Following the methods of Will (1986) and others (Canterbury, 2012;Confer et al., 2003;Vallender et al., 2007), we considered males to have formed a pair with a female if they were observed feeding nestlings or fledglings or if they were seen on a perch close to the nest on several occasions. We considered pairing attributes for a banded male that returned to breed in another year as an additional, independent event.
Conspicuous singing with type 1 calls from one or a few song posts (Gill & Murray, 1972a) by paired or unpaired males provides a strong clue about the location of an established or desired breeding territory. After searching on three mornings for a total of at least six hours and spanning at least a week, a male was thought to be unpaired if no evidence of nesting was found near such song posts.
Females are very cryptic, and almost all observed females were engaging in reproductive activities (e.g., nest building, carrying food, and alarm behavior). This provides a very biased sample of the proportion of females that are paired. Consequently, we estimated pairing success rates only for males. We quantified the pairing success rate at each study area as the fraction of the resident males that formed a social pair averaged for all years of each study. We equate primary hybridization to the formation of a social pair between phenotypes of Golden-winged and Blue-winged Warblers.

| Estimating the strength of reproductive isolating barriers
We estimated the strength of one prezygotic reproductive isolating barrier (BI or behavioral isolation) and one postzygotic reproductive Sexual selection against hybrids, which we refer to as hybrid fitness, was estimated as.
where Hyb denotes the proportion of phenotypically hybrid males that formed a social pair with a female and Pur denotes the proportion of phenotypically pure males that formed a social pair.
These equations produce symmetrical values that represent the proportional reduction in gene flow relative to expectations under random mating (Sobel & Chen, 2014). A slope of two ensures that values of RI range from −1 to 1, with 1 denoting complete reproductive isolation. The strength of both reproductive isolating barriers was estimated for each population, and 95% confidence intervals for individual RI indices were estimated using bootstrap resampling with 1,000 replicates using the boot package (Canty & Ripley, 2015) We used the methods outlined in Coyne and Orr (1989) and Ramsey, Bradshaw, and Schemske (2003) to estimate the absolute contribution of each sequentially and independently acting reproductive isolating barrier (AC) to total reproductive isolation resulting from BI and SH. Because behavioral isolation acts first, AC BI = BI.
The absolute contribution of sexual selection against hybrids (AC SH ) equals SH(1-AC BI ). Total reproductive isolation is the sum of the absolute contributions of behavioral isolation and sexual selection against hybrids (AC BI + AC SH ).

| RE SULTS
We obtained data on breeding Golden-winged and Blue-winged Warblers and their hybrids from nine studies at eight study areas over 47 years of field work. The total sample provided information on 2,679 resident males and females for Golden-winged (n = 1,852) and Blue-winged Warblers (n = 667) and their hybrids (n = 160; Table 1).

| Behavioral isolation
Across all studies, primary hybridization occurred with 0.9% (n = 14) of the 1,680 paired Golden-winged Warblers. The tabulation and calculations of paired individuals ( Blue-winged Warblers across sites (r 2 = 0.03, df = 7, p = 0.67), suggesting that the strength of behavioral isolation (the observed rate of interbreeding, corrected for random expectations) is unaffected by variation in local relative abundance of the two lineages.

| Sexual selection against hybrid males
For pooled values, the pairing success rate for male Golden-winged Warblers was 83% and for male Brewster's Warblers was 54% ( Some of the variance would be due to random variation especially with the very small sample of hybrids in each study. To account for potential real differences in pairing success from one study area to another, we used paired t tests. With the differences paired by study area, male Golden-winged Warblers had higher pairing success than TA B L E 2 Numbers of paired Golden-winged (GWWA) and Blue-winged Warblers (BWWA) individuals for each study area, the proportion of paired individuals that were GWWA, and the number and frequency of primary hybrid pairs.   (Table 3; two-tailed, paired t test: df = 8, t = 3.25, p = .012) as did male Blue-winged Warblers (

| Total reproductive isolation
The combined action of behavioral isolation plus sexual selection against hybrids results in strong reproductive isolation between Golden-winged and Blue-winged Warblers at all study sites, ranging from 0.882 to 1 (mean: 0.960; lower 95% CI: 0.928; upper 95% CI: 0.983). The individual contribution of behavioral isolation to total reproductive isolation was much greater than that of sexual selection against hybrids (two-tailed, paired t test: df = 5, t = 9.89, p < .001).

| Relationship between plumage divergence and pairing frequency
Of 849 paired, male Golden-winged Warblers, 4.9% (n = 42) paired with a female Brewster's Warbler (Table 4). This backcross frequency is 5.4 times greater than the rate of primary hybridization (χ 2 = 24.29, p < .0001). For the 288 paired, male Blue-winged Warblers, 1.7% (n = 5) paired with a Brewster's Warbler. Backcross frequency by male Blue-winged Warblers was less than but not significantly different from their frequency of primary hybridization (χ 2 = 3.48, p = .062). The frequency of backcrossing by male Golden-winged Warblers with Brewster's Warblers was 2.9 times greater than the rate for male Blue-winged Warblers (χ 2 = 4.81, p = .028). Sample sizes for Lawrence's Warblers were too small for statistical analyses.
For the 309 paired, female Blue-winged Warblers, 3.6% (n = 11) formed a social pair with a Brewster's Warbler. Backcross frequency by female Blue-winged Warblers was not significantly different than their frequency of primary hybridization (χ 2 = 0.888, p > .10). The

frequency of backcrossing by female Golden-winged Warblers with
Brewster's Warblers was not different than the rate for female Bluewinged Warblers (χ 2 = 0.0001, p > .10). Sample sizes for Lawrence's Warblers were too small for statistical analyses.

| D ISCUSS I ON
The degree of difference between Golden-winged and Bluewinged Warblers is difficult to quantify. In regions of sympatry, the two species often nest in old field successional habitat   (Bennett et al., 2017;Kramer et al., 2017). Further, Blue-winged Warblers arrive earlier on their sympatric breeding grounds (Canterbury & Stover, 1999;Ficken & Ficken, 1968a, 1968b. Golden-winged Warblers weigh more and have larger wing chords, but smaller tarsi (Confer, 1992;Gill, Canterbury, & Confer, 2001). The primary song of Goldenwinged and Blue-winged Warblers, which is used to attract mates, is readily distinguished (Ficken & Ficken, 1966, 1968a, 1968b, 1969Gill & Murray, 1972a;Highsmith, 1989) with small variation among males of the same phenotype (Gill & Murray, 1972b Notably, each song type seemed quite normal. Kramer et al. (2019) provide audio/visual documentation and analyses of this bivalent singing. These differences in habitat preference, range, behavior, song, and morphology surely have a genetic foundation, but their contribution of the differentiation between these two species is not readily quantified.
Our study provides a measure of the degree of speciation by compiling the pairing frequencies for sympatric populations of  Vallender et al. (2007). This suggests that behavioral isolation from social pairing data would be minimally confounded by the presence of extra-pair or extra-species copulations in this system.
Hybrid fitness significantly influences our understanding of differentiation between Golden-winged and Blue-winged Warblers, and of the factors that may drive speciation. To assess hybrid fitness, we used data for only males because they are usually caught near singing posts, which are used by both mated and unmated males, and which seems to provide an unbiased sample of pairing frequency. We exclude females who are most often caught in nets placed near a known nest, which would provide a biased sample of pairing frequency. Ficken (1968a, 1968b) compiled data from several sources that showed a significant difference in the ratio of paired to unpaired males for "pures" versus hybrid: 93% (n = 32:3) vs. 46% (n = 6:7) (chi-square = 11.781, p = .018). Confer and Tupper (2000) found that only 1 of 13 resident male Brewster's Warblers formed a social pair. Experimental manipulation of plumage pattern (Leichty & Grier, 2006) showed reduced pairing success for hybrid-looking males. For our pooled results for males, hybrid fitness was significantly lower with a 35% reduction in pairing success rate for hybrids compared to Goldenwinged Warblers. Vallender et al. (2007) analyzed male and female pairing success for a study in southeastern Ontario. Based on these data, Kramer et al. (2018) suggest that "there is little evidence of costs to producing hybrid young." However, considering just males, the data showed a pairing success rate of 42% (55 of 132) for Golden-winged Warblers and 18% (2 of 11) for Brewster's Warblers (Vallender et al., 2007), a 57% reduction in pairing success for hybrids compared to Golden-winged Warblers. The trend for this data for males agrees with the significant results reported by Ficken and Ficken (1968a) and the extremely low pairing success for hybrids observed by Confer and Tupper (2000), and to the significant reduction in hybrid fitness shown by our pooled results and by the paired t tests for our individual studies. Collectively, the published data suggest that male hybrids have a significant loss in reproductive fitness compared to both Golden-winged and Blue-winged Warblers. Although our estimate of behavioral isolation is much higher than our estimate of sexual selection against hybrids (Figure 2), recent work suggests that even weak postzygotic reproductive isolating barriers can potentially play a larger role in reducing gene flow than strong prezygotic barriers (Irwin, 2020 Despite the near-complete levels of reproductive isolation between Golden-winged and Blue-winged Warblers that we document, other studies have documented high levels of introgression (Dabrowski et al., 2005;Shapiro et al., 2004;Vallender et al., 2007) and weak genome-wide differentiation (Toews et al., 2016) in this system. Our estimates of reproductive isolation might underestimate the actual level of gene flow between Golden-winged and Blue-winged Warblers. Nevertheless, our primary data analysis assessed whether divergent plumage phenotypes contribute to nonrandom mating in this system, not to an assessment of the actual levels of gene flow.
The percentage of the total population composed of individuals with hybrid phenotypes averaged across sites (5.2%; lower 95% CI: 4.1; upper 95% CI: 6.7) is reasonably consistent with the probability of gene flow (estimated as 1 − total RI) based on the joint effects of behavioral isolation and sexual selection against hybrids averaged across sites (3.4%; lower 95% CI: 1.1%; upper 95% CI: 5.8%).

| CON CLUS IONS
The taxonomy of Golden-winged and Blue-winged Warblers has been debated for well over a century, since the initial description of Brewster's and Lawrence's Warblers (Brewster, 1874;Herrick, 1874). The intensity of this debate has heightened recently, in light of genomic analyses (Toews et al., 2016) whereby genetic differences between Golden-winged and Blue-winged Warblers are associated primarily with genes that control feather attributes.
Indeed, the low level of genome-wide divergence has been cited as evidence that Golden-winged and Blue-winged Warblers may be plumage morphs of a single-species complex (Kramer et al., 2018).
Alternatively, our work suggests these genomic differences relate to plumage divergence that in turn leads to high levels of reproductive isolation. We document strong behavioral isolation and significant sexual selection against hybrids, which together provide a value of 0.960 for reproductive isolation. These attributes support the view that Golden-winged and Blue-winged Warblers are distinct species under the biological species concept.
The taxonomic treatment of these lineages is especially important given the conservation challenges that Golden-winged and Blue-winged Warblers face (Sauer et al., 2017). Extensive population decline in the Golden-winged Warbler has resulted in a petition for listing on the Endangered Species Act (Sewell, 2010), and the decision regarding listing will be highly influenced by one's interpretation of speciation in the Vermivora spp. complex.
Ecological and genetic interactions between Blue-winged and Golden-winged Warblers as the former moves into sympatry with the latter appear to be a major cause of the decline of Goldenwinged Warblers (Gill, 1980;Confer et al., 2011;Rohrbaugh et al., 2016;Rosenberg et al., 2016). The final genetic outcome of this expansion is unclear. On the one hand, introgression (Vallender et al., 2009;Wood et al., 2016) may prevent further divergence between lineages (Karrenberg et al., 2019;Nosil, Harmon, & Seehausen, 2009;Sambatti, Strasburg, Ortiz-Barrientos, Baack, & Rieseberg, 2012;Strasburg & Rieseberg, 2008), while sexual selection against hybrids may enhance isolation. Despite the high levels of reproductive isolation between Golden-winged and Bluewinged Warblers we document, we suggest that their continued existence may require conservation efforts that maintain or repair eco-geographic isolation (Roth, Rohrbaugh, Will, & Buehler, 2019).

ACK N OWLED G M ENTS
The data for the analyses of this manuscript came from studies that were published previously. This compilation was not directly

CO N FLI C T O F I NTE R E S T
There is no conflict of interest with this manuscript for any of the authors.