Introduction

The speciation of contemporarily diverse taxa like butterflies [1, 2], wood-warblers [3], poplars [4], and other taxa [5] has been associated with changes that occurred during the Pleistocene. This geological epoch was characterized by alternation between periods where ice sheets advanced and covered large land masses (i.e., glacials) and periods where these ice sheets receded (i.e., interglacials). Glacial refugia—contracted habitable regions where organisms retreated to for survival—isolated populations of species and potentially affected their divergence (and possibly subsequent introgression) upon secondary contact during interglacials [6]. The Laurentide ice sheet of North America has been shown to be one of the divides that separated refugia east and west of the Rocky Mountains, where contemporary species now also show repeated phylogenetic divergence in contact zones [6–8]. As a result of these processes, species that recolonized the continent post-refugia may have complex and diverse speciation outcomes. Studying the traits of contemporary species in the context of this glacial legacy is an important pursuit to understand the natural history of species and the divergence of lineages.

In avian species, one behavioral aspect that has received less careful attention in the context of this glacial history is vocalization. Historically, species and subspecies designations in birds have focused on morphological and plumage variants, yet over the past several decades researchers have increasingly recognized the value of considering vocal signals [9]. Studies have linked variation in vocalization and plumage with individual fitness and sexual selection [10]; these vocal traits have therefore been recognized as likely important reproductive isolation barriers in birds, and can therefore possibly provide insights about the extent to which gene flow affects the structure of populations. Moreover, in addition to the vertical inheritance across generations, vocal signals in song-learning birds can change via cultural mutation across a landscape—learning errors during the learning process may result in differentiating localized songs that are eventually passed onto the next generations [11, 12]; thus, it is relevant to assess these traits through a biogeographic lens to understand how songs vary geographically [13, 14]. Different geographic regions may promote divergence simply via isolation or restricted gene flow, or via the effects of adaptation to novel environments or different sexual selection pressures [reviewed in 10]. The result of cultural evolution of birdsong can be represented as geographic variants, which can be identified when the objective acoustic characteristics of songs vary more-or-less discretely across different geographic areas [15, 16].

In continental North America, studies have documented a specific pattern of song variation between eastern and western counterparts of closely related taxa, primarily divided by the Rocky Mountains. For example, Kroodsma [17, 18] described differences between vocalizations of the winter wrens and marsh wrens; both of which sing complex songs and show consistent and statistically detectable differences in song components, song types, and overall repertoires. The winter wrens of North America were eventually split into winter wren (Troglodytes hiemalis) and Pacific wren (T. pacificus) due to both bioacoustic and genetic evidence [19], while the marsh wren status requires additional genetic investigations.

Kroodsma [17] also noted that habitats in western North America are more heterogeneous. This means that there are smaller fragments of each species’ preferred habitat, which might prevent dispersal and result in stronger reproductive barriers in sympatry. Meanwhile in the east, larger fragments of suitable habitat may have allowed for a more gradual divergence and dispersal, as well as physical geographic isolation [17, 18]. This difference in habitat distribution, fragmentation, and history of isolation may affect the transmission of acoustic signals and the formation of dialects in the east and the west differently [9, 17, 20]. Specifically from the example above, western marsh wrens have quantifiably higher song complexity and repertoire size than eastern marsh wrens, despite presumably having the same learning capability [21]. Both species complexes do not hybridize in their sympatric ranges and, in this case, vocalization is likely an important reproductive barrier. Together, these studies show that the east-west distribution of the taxa coinciding with their vocalization differences might be a result of common divergence during geographic isolation and cultural evolution.

There have also been examples of geographic differences between populations within a single species in both eastern and western North America. One of the most prominent and thorough investigations is in white-crowned sparrows, Zonotrichia leucophrys [22–24]. Past studies of this system have focused on comparing differentiation of songs between populations with genetic and geographic structure, although the results have been mixed. Most recently, Lipshutz et al. [25] coupled genomic analysis with territorial playback experiment and found that two subspecies populations whose territories do not overlap also exhibit genetic divergence, as well as ecologically meaningful localized, fine-scale geographic variation in vocalization (i.e., dialects). Though not a direct correlation to genetic divergence, this study affirmed the role of dialects as a possible premating reproductive barrier as they might be driving the separation of populations and decreasing gene flow. In the east, mourning warblers (Geothlypis philadelphia) have been the focus of detailed research and appear to have four regional song types (i.e., regiolect; [26]) across the species’ breeding range based on syllable types and objective frequency and duration of songs, each type encompassing multiple populations [15].

Although the consistent vocal differences between eastern and western groups have been qualitatively described—like in the studies above—more systematic, comparative investigations are important to help reveal whether these differences vary primarily on an individual level, population level, and/or a large-scale, regional level, and whether this is an evolutionary phenomenon related to their historical geographic distributions. To do this also requires combining standardized measures of geographic differences in songs with time estimates of divergence, presumably derived from genetic data [6–8, 27–29]. One prominent example of divergence time estimation in avian species was by Weir & Schluter [30], who hypothesized that the speciation of boreal superspecies complexes were initiated and accelerated during the advancing of ice sheets, which forced their common ancestors into separate eastern versus western ecological ice-free refuges. They employed mtDNA cytochrome b sequences to calculate and compare coalescent timing for several pairs, which they found all fell within the Pleistocene and coincided with the timing of increased glaciation. This result implied that speciation of the examined species was likely initiated with glaciation as a geographic barrier, and potential trait divergence might have happened during this isolation. This divergence time estimation provided the basis for examining the vocal variation of species across the continental divide: if species divergence was a consequence of Pleistocene refugia, subsequent vocal divergence could be one of the key premating reproductive barriers that potentially promotes speciation and prevents hybridization upon post-refugia secondary contact.

The current study aims to describe the macrogeographic variation of songs in eight species pairs whose breeding grounds are eastern and western counterparts of each other, as a direct extension to Weir & Schluter [30] (Table 1). We assess the magnitude of song differences across two scales to explore variation both within and between taxa: (1) measure and compare the acoustic variation between eastern and western North American songbirds, and (2) identify acoustic variation within the range of each taxon and then compare any populations with detectable variation. These different scales characterize the pattern of variation and help determine whether regional differences exist, which will allow us to compare and detect potentially important vocal characteristics among taxa. Specifically, we predict that (a) there are objective, statistical differences in acoustic characteristics that distinguish populations across geographical areas on both scales, and (b) the magnitude of geographic differences between taxa would be larger than that of geographic variation within taxa. To estimate song divergence timing and determine whether that divergence correlates with the timing of Pleistocene events, we analyzed mitochondrial DNA (mtDNA) using phylogenetic analyses and coalescence time analyses. We chose mtDNA because it is maternally inherited and non-recombining and thus retains signals of historical isolation [31]. By contrast, differences in the nuclear genome can be eroded by contemporary gene flow [32]. This is particularly important for the taxa in question that, in many cases, have contact zones or hybrid zones where gene flow is a possibility. We predicted that if mtDNA divergence correlates with song divergence, it implies that song divergence might be consistent with Pleistocene-level isolation timing [30]; if there is no correlation, detectable song divergence might have evolved independently from mtDNA.

Download:

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

Table 1. Sampling details and acoustic measurements.

https://doi.org/10.1371/journal.pone.0312706.t001

Materials and methods

Study species

We selected eight pairs of closely related migratory passerines based on previous molecular phylogeny and coalescent time analysis, whose breeding grounds are in continental North America [30, 33–37]. Each pair consists of an eastern taxon and a corresponding western taxon (i.e., “counterparts”), all listed in Table 1. The chosen boundary between eastern and western taxa is the Rocky Mountains because phylogeographic breaks along this mountain range are found in many species, specifically avian species where they shared recolonization routes post-glaciation [6]. These closely related pairs were chosen so that each taxon is the closest eastern and western representative of their phylogenetic group; three are sister species, two are closely related species, three are subspecies (Fig 1A, Table 1), and many hybridize. Among the eight pairs, Setophaga virens and Setophaga townsendi each have two known song types, which we identified and compared separately [38, 39]. Cardellina pusilla has three subspecies, two of which C. p. pileolata and C. p. chryseola we treated as a single group (western) in comparison to C. p. pusilla (eastern) due to genetic divergence [11, 40].

Download:

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

Fig 1. Study species description.

(A) Continental North America divided by the Rocky Mountains. All study species are depicted alongside their songs (illustrations courtesy of Cornell Lab or Ornithology’s Birds of the World). (B) An example of an eastern Leiothlypis ruficapilla song, showing the definition of acoustic components measured in the current study: syllable vs. note. This map was made with Natural Earth (free vector and raster map data at naturalearthdata.com) and ArcGIS Online basemap [41].

https://doi.org/10.1371/journal.pone.0312706.g001

Statistical analysis of songs

We implemented three independent statistical analyses to assess the geographic variation in acoustic characteristics of songs. The two clustering analyses are done for each counterpart to assess within-taxa variation, and then for comparison between the counterparts of each pair to assess between-taxa variation, while Δp was used to assess between-taxa variation only. All measurements collected from Raven Pro 1.6 as song features were used for quantifying song divergence via both all three methods. First, we used partitioning around medoids (PAM), a robust k-means clustering method, by running two functions in the R package ‘cluster’: ‘daisy’ to calculate dissimilarity matrices using Gower’s distance, and ‘pam’ to cluster data into ‘k’ using the matrices generated by ‘daisy’ [45]. As opposed to the Euclidean distances that are the root sum-of-squares of differences typically used for numerical-only datasets, Gower’s distance first standardized each variable so that the dissimilarity scores are within the range of [0,1], which makes it possible to measure dissimilarities between categorical values where it is either 0 (no difference) or 1 (different) [46]. This feature allowed for mixed datasets of numerical and categorical data as well as “missing data”, because the variation in number of notes and syllables occasionally caused our datasets to be uneven; the ‘daisy’ algorithm omits the missing data from the dissimilarity calculations so that the scores only contain the individual’s available song features. To determine k for each PAM analysis, we used silhouette width—a metric used in clustering analyses to assess how similar within-cluster data points are to each other comparing to the neighboring cluster. We calculated silhouette width for k = 2–8 using the built-in ‘sil_width’ function from the package and chose the highest value, which signifies the best fitting k. Because PAM generated dissimilarity matrices for each comparison, we also assessed the magnitude of song divergence: (1) within the eastern taxon, (2) within the western taxon, and (3) between the eastern vs. western taxa. We used permutation test for each pair to compare whether the dissimilarity of between-taxa is greater than within-taxa, accounting for the resampling of each individual during the generation of the dissimilarity matrix. We used the R package ‘coin’ to generate permutation tests [47], particularly an approximative reference distribution (Monte Carlo) with 10,000 replicates. We built a model with dissimilarity score as the variable of interest assigned to one of the three categories of comparison: (1) within the eastern taxon, (2) within the western taxon, and (3) between the eastern vs. western taxa. Individuals were then resampled 10,000 times to generate a distribution of the summary statistic under the null hypothesis that there is no difference between populations. Finally, the observed summary statistic from the actual populations was compared to the distribution of summary statistics from the permutations—P-values calculated from these comparisons was adjusted with post-hoc false discovery rate. These analyses allowed us to assess whether further geographic distance (between-taxa) correlates with more diverged songs.

Second, we used the ‘PCAmixdata’ package to perform principal component analysis (PCA), another common clustering method [48]. Similar to PAM, ‘PCAmix’ calculated the principal components and loadings for our mixed and uneven datasets to determine the pattern of song divergence and if so, which variables distinguish them. PCAmix also replaced missing data with means for quantitative variables and with zeros for categorical variables automatically before calculations. Graphical presentation of PCA results and interpretation used chiefly PC1 and PC2 because they explain the majority of variation in each taxon (see Result; Table 2; Tables C & D, & Fig A in S1 File).

Download:

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

Table 2. The statistical result of PCA, PAM, and all-trait Δp for each between-taxa comparison.

https://doi.org/10.1371/journal.pone.0312706.t002

For consistency and comparative purposes, the tools were used on all counterparts regardless of whether their dataset contains categorical variables. Both tools allowed us to assess acoustic differences without a priori eastern versus western grouping and assess whether these differences match the known geographic ranges, as well as detecting any potential within-taxa variation. For both PAM and PCA, we expected to observe clear clustering of groups whose acoustic characteristics are distinguishable from each other, as well as some high-loading variables that drive the differences. Using the clusters blindly determined by PAM, we mapped the locality of each singer onto their respective range maps produced by BirdLife International [49] to visually identify geographic structure. Geographic variants would be identified if we observed (1) clear clustering of groups as shown in graphical representation of PAM and PCA and (2) these groups can be distinguished by their geographic locality.

Finally, we quantified the magnitude of divergence of each pair using Δp–a metric of phenotypic distance that allowed for multi-trait comparison between two populations [50]. Due to the nature of song diversity among the study species, it was not possible to compare each measurement directly (e.g., the difference in note A between Vireo solitarius and V. cassinii would be different from the difference in note A between S. c. coronata and S. c. auduboni); therefore, we calculated Δp as a standardized “divergence score” to compare the pairs to each other. Δp is a non-parametric distance measure calculated based on a joint cumulative distribution function (CDF), where a joint CDF was first calculated for each raw song feature across both populations. Then, for each population, a median percentile coordinate in the dimension of each feature was determined relatively to this CDF. Repeating this calculation for the remaining song features, the consequent Δp is the Euclidean distance between two populations calculated using the median percentile coordinates of all song features. Δp are relative values to the data set, and it is robust against unequal variance and sample sizes and considerable missing data, particularly compared to Hedge’s g [50]. We executed the MATLAB script developed and provided in the original paper (http://sourceforge.net/projects/deltap/files/) to import the data and calculate Δp, Hedge’s g, and descriptive statistics for each song feature and the pairwise whole-population Euclidean distance. The reported Δp in Table 2 that we used as “divergence score” is the whole-population distance.

Results

Song divergence between taxa (Eastern vs. western)

The divergence between counterparts of each pair is reported in Table 2. Overall, each pair’s songs fall into two PAM clusters and variation is primarily explained by the first two dimensions of PCA, except for the Icterus pair which have songs that best fit best into 3 clusters (Fig 2, Table 1, Figs A, C in S1 File). The dissimilarity statistics PAM generated for within- and between-taxa is shown in Fig 4; for the majority of comparisons, the dissimilarities between taxa are significantly larger than the dissimilarity within each taxon, tested with permutation analyses with Bonferroni correction for significance criteria, P ≤ 0.00185 (Fig 4, Table B in S1 File). The groupings determined by both PAM and PCA correlated with the biogeography of each pair, though to varying degrees. The east-west divide is very distinct in S. magna vs. neglecta (Δp = 187.19), G. philadelphia vs. tolmiei (Δp = 166.31), and S. virens vs. townsendi (Δp = 136.42 for type A, 117.70 for type B), within all of which the groupings of most individuals match their locality (Fig 3). This divide is less prominent for the rest of the pairs, where L. r. ruficapilla vs. ridgwayi (Δp = 128.18) and S. c. coronata vs. auduboni (Δp = 83.66) showed moderate geographic divergence (Fig 3), and I. galbula vs. bullockii (Δp = 118.51), C. p. pusilla vs. pileolata and chryseola (Δp = 115.60) and V. solitarius vs. cassinii (Δp = 71.93) showed low geographic divergence (Fig 3). Interestingly, the stratification of Δp did not always match the categorization by PCA and PAM. Although the highly diverged pairs have the highest Δp values and the least diverged pair has the lowest Δp value, the magnitude of divergence of the “moderate” pairs are incongruent. For example, I. galbula vs. bullockii pair have low PC1 and 2 values and moderate PAM percentage while having high Δp values; the C. p. pusilla vs. pileolata & chryseola pair have very low PC1 and 2 and quite high PAM percentage and have high Δp. We also found that the differences between the sets of measurements for each pair did not affect the divergence score measured as Δp values; for example, the Icterus pair has 42 measurements, yet its Δp value is similar to that of the Leiothlypis pair which only has 29 measurements.

Download:

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

Fig 2. Principal component analysis (PCA) for each species pair.

This figure includes analyses for both song types of S. virens and S. townsendi. Individuals of Eastern counterpart are shown in blue, individuals of Western counterpart are shown in yellow.

https://doi.org/10.1371/journal.pone.0312706.g002

Download:

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

Fig 3. Map of geographical variation in songs of eight closely related avian pairs based on partitioning around medoid (PAM) analysis.

This figure includes analyses for both song types of S. virens and S. townsendi. The polygons underneath the data points are range maps of each counterpart: eastern ranges are in blue, western ranges are in yellow. Taxa of the same pair share the same shape on each map; eastern individuals are in blue and western individuals are in yellow. This map was made with free basemap available via geoBoundaries [53], an open database of political administrative boundaries. CC BY 4.0.

https://doi.org/10.1371/journal.pone.0312706.g003

Discussion

Here we present a systematic analysis of song divergence in various oscines of North America using three independent metrics, which build upon and complement previous studies on each species pair. We found evidence of differences between east-west distributions and vocal differences on a continental scale—considered between-taxa geographic differences—where closely related species differ in their vocalizations. The magnitude of these differences ranges from high, moderate, to low, depending on the species pair. Although there were many shared variables in each pairwise comparison, we did not detect any acoustic features that consistently contribute to the overall acoustic differences across the pairs (i.e., no variable had consistently high PC loadings in multiple pairs; Table D in S1 File). We also tested the relationship between the magnitude of these song differences with divergence time at cytb and found no correlation between the two. This lack of relationship implies that the differences between pairs in their song is not entirely predictable based on their divergence time, at least based on mitochondrial dating. This means that song differences may change more rapidly or idiosyncratically than is captured by isolation time alone, but also suggests that markers from detailed, whole genome data might be more useful to better characterize historical differences that might relate to song variation. We also found that for each pair, the magnitude of song differences between eastern-western counterparts are consistently larger than the magnitude of song differences within each counterpart—this affirms that the comparisons we made captured the song difference relative to each pair, and that our method of determining geographic variation based on range maps and individual locality is sufficient in revealing the geographic pattern (or the lack thereof). On a smaller geographic scale, we did not find many potential within-taxa acoustic variants that separate adjacent populations of the same taxa, except for a song type of one S. townsendi population. Although no common pattern of variation was observed for all taxa, our findings further suggest that song as a trait varies across geographical space, which could be a precursor for the formation of well-defined regiolects and dialects.

The lack of clear within-taxa geographic differences was an unexpected pattern, particularly due to the previously reported complex geographic variation in some prominent examples (e.g., G. philadelphia; 15). We did not observe strong geographic bioacoustic differences within the taxa presented in the current study; however, it is possible that this is because there were slight differences in the set of measurements; for instance, Pitocchelli [15] stereotyped the primary syllable and its variation while we did not. Further playback studies will be needed to assess song discrimination to determine the importance of subtle differences between within-taxa populations. The sole potential local geographic variant detected in one of the two song types in S. townsendi could serve as a case study for this work—our results suggest a separation between the highly populated area around the city of Vancouver versus the natural areas throughout the rest of its range. Specifically, this difference was mostly driven by the maximum frequency of the third syllable (also the maximum frequency of the whole song), of which the putative Vancouver local geographic variant is significantly higher (Fig D in S1 File). Different habitats associated with vocalization divergence has been documented in previous studies of oscines, mostly related to vegetation density, which affects the transmission of vocalization [9, 20, 54]. Several species have been shown to have gradually changed their vocalizations to sing louder in the urban landscapes compared to forests [55]. Moreover, urban areas have been shown to have higher attenuation and reverberation compared to rural areas and urban songs tend to have short whistles, faster trills, and narrower bandwidth, which are evidence that the landscapes might affect both sound transmission and the song characteristics themselves [56]. Counter to our findings in S. townsendi, Phillips and colleagues [56] found that the city landscape correlated with lower maximum frequency. However, the difference in S. townsendi is likely between urban/suburban vs. mountain habitat, which might differ from the “rural” landscape outlined in [56]. Recent work by Ore and colleagues [57] identified a similar geographic pattern where Vancouver S. townsendi individuals sang a geographically localized type B song (Type I in that study), which confirms and extends our own findings. We suggest future work could benefit from additional sampling as well as playback experiments in S. townsendi to fully assess the possible sound transmission difference in these different landscapes and understand the development of this local geographic variant in S. townsendi.

In comparing geographic differences between eastern-western taxa, the divergence score Δp presented a spectrum of acoustic divergence: the three pairs with the highest Δp (S. magna vs. neglecta, G. philadelphia vs. tolmiei, and S. virens vs. townsendi) also have the most variance explained by the first two dimensions in both PCA and PAM (Table 2). Similarly, all three metrics agreed on the low divergence across their geographic range for the least diverged pair V. solitarius vs. V. cassinii. We consider the other four pairs (L. ruficapilla subspecies, C. pusilla subspecies, S. coronata subspecies, and Icterus species) to have moderate to low song divergence, as the methods gave discordant results on where they would fall on the divergence spectrum (Table 2, Figs 2 and 3). Our findings are consistent with the current knowledge that S. magna vs. S. negleta, G. philadelphia vs. tolmiei, and S. virens vs. townsendi are highly diverged genetically and show strong geographic structure across their breeding ranges [33, 35, 36, 58]. Although we did not examine reproductive isolation in this study beyond reasonable speculation, the systematic and comparative quantification of song divergence presented in this study contributes to the current knowledge of each taxon and may serve as the basis for subsequent studies such as playback studies to further examine the reproductive isolation via songs in these species.

Hybridization and the maintenance of hybrid zones have provided some interesting insights into the dynamics of song variation between eastern and western taxa. In the case of S. virens vs. S. townsendi, previous studies found that song types match species membership in allopatry, and individuals in the hybrid zones would sing one song or the other regardless of whether they are genetically “more S. virens” or “more S. townsendi” [42]. Reciprocal playback experiments in the same study showed that allopatric birds have a stronger territorial response to conspecific songs and not heterospecific songs, while in the hybrid zone there were no observed differences in responses. This implies that song may be a strong reproductive barrier in allopatry, but this breaks down in the hybrid zone, where it is most relevant [42]. In contrast, while having a similarly narrow zone of extensive hybridization, S. philadelphia and S. tolmiei hybrid song is a blend of the two allopatric songs that differed from both, and the song types in the hybrid zone are not predictive of species membership determined by genetic markers [59, 60]. The implication from this study was also that song was not a reliable reproductive barrier for S. philadelphia and S. tolmiei hybrids, due to the lack of song learning discrimination. In both cases, singing and learning songs that are not conspecific does not appear to contribute to selection against hybrids, but the influence of these song patterns within the hybrid zones also does not seem to go beyond the zones and affect allopatric birds. Since our analysis excluded potential hybrid songs, it is consistent with the literature that we found high song divergence between these taxa in allopatry, implying that these taxa have a strong barrier via songs despite maintaining hybrid zones.

The S. coronata subspecies pair and Icterus species pair are both previously shown to hybridize, despite evidence showing a strong reproductive barrier in plumage and a weak barrier in songs, and that genetic divergence is significant in both pairs [35, 61–63]. The species status of both systems are highly debated because both actively hybridize, but only in their narrow hybrid zone—the rest of the continental range remains differentiable in both plumage and genetics. The songs between counterparts in each pair also diverged per our findings here, though the magnitude of the differences is smaller compared to other hybridizing Setophaga, as discussed above. Previous playback experiments for S. coronata subspecies also suggest that song is not a reliable reproductive barrier both within and outside of the hybrid zone [61], whereas this is unknown for Icterus species due to the lack of playback experiments. Additionally, recent genomic and plumage analyses of Icterus species suggest a gradient of genomic admixture within the hybrid zone which correlates with the phenotypes of hybrid plumages. This implies moderate reproductive isolation within the hybrid zone, possibly due to an unknown form of selection against hybrids [64, 65]. For both pairs, there is substantial genomic differentiation between them, yet the precise phenotypes relating to those genetic differences are still unknown. Thus, while speculative, it is possible that one or some of these genomic regions might affect their vocalizations indirectly, such as via beak morphology or song-learning ability, and subsequently alter hybrid fitness [66, 67]. It is also notable that although the Icterus taxa are currently considered two species, S. coronata coronata and S. c. auduboni are currently not.

C. pusilla subspecies is the only pair in this study that has a completely contiguous breeding range; interestingly, there has been substantial evidence of genetic divergence between eastern and western subspecies despite not exhibiting stark differences in plumage and morphology [40, 68–70]. The eastern subspecies C. p. pusilla—distributed throughout the large fragments of suitable habitat of the continental east—forms a genetically cohesive group with minimal geographic variation within and is distinct from the western subspecies [70]. Although we used songs from both western subspecies to represent a sole “western” group, we still detected some geographic structure of song variation that might be concordant with the distributions of C. p. pileolata and C. p. chryseola, but a larger sample size is needed to confirm this pattern (Table A in S1 File). Consistent with genetic studies, the difference between these western subspecies is seemingly smaller than the difference between the eastern and the western groups according to dissimilarity score generated by PAM analysis (Fig 4, Table B in S1 File). Moreover, individuals of the western group whose distribution matches C. p. pileolata always grouped with the eastern populations in the eastern vs. western song comparison (Fig 3). We suspect due to the continuous distribution in the north, but not throughout the central US, the eastern song might have been transmitted and blended into songs along the coast, while the western inland song stays moderately different. Moreover, since the population boundaries between the C. pusilla subspecies are not known, we could not confirm the identity of the individuals we chose to analyze along the proposed Rocky Mountains boundary, which might have made our analysis less accurate. Further investigation into song recognition is needed between the subspecies to make conclusions about song variation.

Download:

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

Fig 4. Magnitude of differences in within- and between-taxa comparisons.

Comparison of the magnitude of differences (i.e. dissimilarity) among within eastern counterpart (blue), within western counterpart (yellow), and between eastern vs. western counterpart (green) based on dissimilarity scores for each species pair. Each point is an overall dissimilarity score based on all song features, which represents a comparison between two individuals. Thus, blue points are comparisons among individuals of eastern counterparts, yellow points are comparisons among individuals of western counterparts, and green points are comparisons among individuals of both groups. Significance values shown are from permutation: *: P ≤ 0.05, **: P ≤ 0.01, ***: P ≤ 0.001, ****: P ≤ 0.0001. Bolded values are those remain significant following Bonferroni correction, P ≤ 0.00185.

https://doi.org/10.1371/journal.pone.0312706.g004

For the non-hybridizing pairs, we draw a contrast between the most and the least diverged songs: the Sturnella and the Vireo pairs, respectively. Beam and colleagues [58] suggested that eastern Sturnella magna and western S. neglecta are highly diverged both acoustically and genetically from each other, as well as from the southwestern Lilian’s subspecies (now split and renamed as S. lilianae). Given the virtually identical plumages and the additional evidence of substantially diverged songs we provided here, yet a lack of hybridization, future playback experiments will be crucial to assess whether song would play a role in species recognition between S. magna and S. neglecta. In the least diverged pair, V. solitarius and V. cassinii, all three metrics PCA, PAM, and Δp agreed on their low song divergence across their geographic range. Coupled with low plumage difference between them, this outcome raises questions about which reproductive barrier is most important for the species boundary, since the two common mechanisms seem to be weak. mtDNA evidence justified their split into separate species [34, 71, 72], but since then there has been much back and forth debate on the phylogenetic relationship within the vireo complex that also includes V. plumbeus of the western interior.

This study allows us to assess the unique situation of the L. ruficapilla subspecies, which have allopatric ranges with a presumably strong dispersal boundary separating them (i.e., the Great Plains). L. r. ruficapilla and L. r. ridgwayi have diagnosable differences in plumage, behavior, and calls [73]; here, we found that songs also differentiate them, though it is unclear whether these acoustic differences are the result of the range separation, or whether stochastic cultural mutations had occurred and spread before range separation. Allopatric species pairs have overall greater plumage and song divergence compared to sympatric pairs [74]; this is potentially because allopatric species tend to inhabit different habitats and sympatric species tend to inhabit similar habitats, which can be important in the context of plumage perception and sound transmission [9, 75]. Together with previous knowledge, the song divergence we found here is a starting point for further investigation regarding the species status of L. ruficapilla subspecies, as the magnitude of divergence between their songs are higher than pairs currently considered different species (e.g., Icterus and Vireo species). L. ruficapilla is also the least studied of all taxa in the current study: the most recent genetic work examined only the mtDNA, which showed two distinct haplotypes corresponding with the east-west grouping [76]. It is crucial for future studies to assess the genetic difference and song discrimination between the two subspecies to determine whether the acoustic differences we found here are important, and whether these differences constitute a reproductive barrier.

Genetic divergence and song divergence

We used the divergence of cytochrome b sequences as a proxy for divergence between taxa [30] and assessed whether the time of the split between the taxa correlated with the magnitude of song divergence. This approach is similar to the one taken by [77] in different Amazonian forest systems, which found a correlation between overall song structure and genetic distance. Like our study, this approach assumes that genetic divergence in mtDNA is an accurate measure of divergence times, which has been quantified in other avian systems [78]. In our case, we found that the song differences did not correlate with mtDNA divergence, implying that these differences are either not related to geographic isolation during the Pleistocene or could not be captured via the coalesced timing of mtDNA (Fig 5 using Δp as song divergence, Fig F in S1 File using PCA and PAM output as song divergence). Such timing could be confounded by processes such as mitochondrial introgression, which has occurred in one of the species pairs (S. c. coronata and S. c. auduboni; 54,56), though we attempted to control for this issue in this pair. It is also possible that divergence resulted in genome-wide effects, which might not have been adequately quantified by a single genetic marker like mtDNA. Future work reassessing the relationship between song divergence and whole genome data will be important to better identify the divergence patterns. We also cannot rule out the possibility that the timing of the isolation was not important, and these taxa would have diverged as long as they are isolated regardless of when. We also acknowledge that vocal traits could potentially evolve rather quickly; some species have been shown to change song characteristics over a short time frame: chestnut-sided warbler (Setophaga pensylvanica) showed new song elements within 19-year intervals [79]; song recognition in white-crowned sparrows (Z. leucophrys) for historical local songs is less in magnitude compared to current local songs and is proportional to how acoustically different the songs are, within about 20 years [80]. That said, we believe that among our study species, qualitative song characteristics have not changed since at least the 1950s, when the earliest recordings are available (e.g., a 1956 recording of S. virens is identical to a typical contemporary S. virens song; Macaulay Library Asset #ML67887).

Download:

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

Fig 5. Correlation between coalescent estimated mtDNA divergence and song divergence score Δp of all species.

https://doi.org/10.1371/journal.pone.0312706.g005

Conclusion

This work complements and extends previous comparative studies on the evolution of avian song evolution, song discrimination, and genetic divergence [92, 93]. Vocalization divergence in songbirds in North America has likely undergone both divergent and parallel evolution, which was greatly affected by isolation during glacial periods [84]. While our analysis of pairs shows this may not be directly related to time since their isolation, the consistent divergence between—though not necessarily within—each pair demonstrates these differences persist today. Future work incorporating genomic data will hopefully provide a more nuanced characterization of the size and diversity of refugial populations and, subsequently, how this may have influenced their cultural trajectory.

Supporting information

Acknowledgments

The authors thank Matthew Medler and the Macaulay Library at the Cornell Lab of Ornithology for providing us with all audio recordings for this work. We thank members of the Toews lab for their feedback and support throughout the process.