Study Region and Sampling Design

In the Philippines, over 570 species of migrant and resident birds are distributed across a tropical archipelago stretching latitudinally along the Western Pacific Rim [28]. Over 7,000 islands make up the Philippines, and many islands host their own endemic species and subspecies [27]. Avian colonization and vicariance events in the Philippines are likely to have occurred multiple times and from multiple sources [31–33], and heavily restricted gene flow is common between at least some island populations [32, 34]. Unlike many mainland taxa, species ranges in the Philippines are likely to have been quite stable over time, having experienced little Pleistocene fluctuation [32].

During prior research, fieldwork archived museum specimens representing avian assemblages on 9 islands, representing 4 Pleistocene Aggregate Island Complexes, or PAICs [33], to provide a substantial degree of geographic co-distribution to our broad comparisons. This sampling scheme affected sample sizes within populations. Using museum specimens, we obtained between 1 and 11 individuals per population (average n = 4) from 117 populations, including 109 described subspecies of 48 Philippine bird species representing 31 families and 12 orders (Table A in S1 File). Our taxonomic sampling includes lineages with undifferentiated populations, lineages with recognized subspecific differentiation, and lineages with now-recognized full biological species. Thus, among lineages, our samples represent differentiation that spans the divergence and speciation process. Most orders are represented by only one species, whereas the order Passeriformes is represented by 34 species and 79 subspecies. Nearly all (98%) of the subspecies sampled are endemic to the Philippines. Since our study began, several of these species have been split by some authorities [35–37], but we have retained these comparisons because they are complementary and useful in assessing among-lineage divergence patterns in Philippine birds.

Phenotypic Comparisons

We performed 96 pairwise phenotypic comparisons following the Tobias et al. [13] quantitative method for scoring plumage and morphometric characters. We are not using this method to delimit species. Rather, we are using it as a heuristic standard by which to gauge population divergence in phenotypic space; it provides a transparent and broadly applied methodology [36] that has some basic utility when comparing diverse lineages. Using this method, minor, moderate, major, and exceptional differences in plumage received scores of 1, 2, 3, and 4, respectively, and differences among morphometric characters were calculated as Cohen’s d effect sizes (the difference between means divided by standard deviation) for wing chord, tail length, tarsus length, and bill length between populations and scored as follows: |0.2–2| = minor, |2–5| = moderate, |5–10| = major, and |> 10| = exceptional [13]. Morphometric comparisons were performed among samples of n = 2–10 with equal sex ratios for all pairwise comparisons. Whenever possible only males were compared, and immature birds were not included in plumage or morphometric analyses. In total, 728 study skins were examined for phenotypic comparison (Table A in S1 File). We were unable to follow the recommendation from [13] to include vocal scores due to insufficient data being available, and we did not include scores for the presence or absence of hybrid zones, because most of the populations we studied are strictly allopatric. Because selection acts non-independently on morphometric characters more commonly than on plumage characters, each phenotypic pairwise comparison considers the 3 highest scores for plumage and only the single greatest increase and decrease in morphometric effect sizes between populations. Tobias et al. [13] recommended that total scores of 7 or more should be regarded as sufficiently divergent to be considered as full species. Despite the exclusion of vocal scoring and geographic structure, we also consider a phenotypic score of 7, conservatively, to be a conceptual threshold for identifying what we term phenotypically highly divergent taxa.

Genotypic Comparisons

We performed 96 pairwise genetic comparisons using mtDNA sequence data from a single marker (ND2). DNA was extracted from fresh frozen tissues (or temporarily ethanol-preserved tissues for borrowed samples) from 491 individuals from 48 species of Philippine birds using Qiagen DNeasy blood and tissue extraction kits. The mitochondrial gene ND2 was amplified using standard PCR protocols with the forward primer L5215 [38] and reverse primer H6313 [39]. Sanger sequencing was performed by the University of Washington’s High Throughput Genomics lab (www.htseq.org) using the same primers and the following custom internal forward primers designed for this study; Eumy.ND2inter (ACAAAAACCCCAGCACTWAG), Hypsi.ND2inter (TAAACTCAATCAAAACCCTA), OtusND2inter (CCCAACCCTATTGACCMYAA), ParusND2inter (TTCTCCTCCATCTCCCACCT), Phapitre.ND2inter (CTACTAACCTTCTAYCTWTA), SittaND2inter (TATTAACCACCATAGCCATC), Zoster.ND2inter (CTACTCACATGCATAGCCGT), Collo.ND2inter (TCCCATCTCGGATGAATATC), Microhi.ND2inter (ATAATAATTACCTGAACAAA), and Phyllos.ND2inter (ACCGGRCTRCTMCTRTCCACA). Sequence data were visually inspected and cleaned when required then aligned using Sequencher 4.0. Partial ND2 sequences ranged in length from 692 to 1041 bp (average = 955 bp). Genetic divergence was calculated as Jukes-Cantor corrected p-distance in MEGA 5.0 [40], which derives the average distance between all individuals in the two sampled populations. In addition, MtDNA (ND2) haplotype networks were generated for 8 species with paraphyletic population relationships in DnaSP and Network 4.6.1.0 [41, 42]. GenBank accession numbers for all 491 individuals are provided in Table A in S1 File.

Divergence Levels and Speciation Processes

We make a key conceptual distinction between divergence and speciation. Populations diverge before and after speciation; it is a continuous process. We therefore focus more on the process, though we consider divergence levels in both categorical and continuous ways (or semi-continuous in the case of phenotypic divergence scoring). We do not propose species limit thresholds in this study. Placing thresholds for species limits on genetic data is an inherently contentious issue, in part because no proposed divergence value clearly indicates the completion of speciation [20, 43, 44]. Here we have chosen a rather high threshold (5% corrected pairwise divergence in ND2) to be conservative in erecting a conceptual category that we term genetically highly divergent lineages, useful to highlight lineages for further studies on species limits. We have chosen this conservative threshold in part to include consideration of the variability of mutation rates among birds [45], especially at shallow levels of divergence [46–49]. Specifically, we set a genetic divergence threshold at 5% Jukes-Cantor corrected ND2 p-distance. The choice of 5% was done blindly with respect to the data, using a value that would be recognized as conservative among those generating the primary literature on species limits and genetic divergence. Using a generic mtDNA divergence rate of ~2% per million years [45, 50], this represents approximately 2.5 Myr of divergence. More than half of the 192 sister-species pairs considered by Weir and Schluter [51] had divergence levels below this threshold, and even lower threshold values have been proposed by others [52, 53]. We emphasize, however, that this is simply an indication of lineage divergence and not an indication of speciation. Genetic divergence on these terms occurs through neutral processes, whereas speciation is thought to be driven largely by selection [12, 54], and legitimate biological species can have lower genetic divergence values [51] while divergences of this threshold level can occur within a single biological species [55]. Indeed, there are numerous ways in which mtDNA can be misleading about species limits and relationships between populations [56–64].

Similarly, while we have used the methods of Tobias et al. [13] to provide a measure of phenotypic divergence, we do not apply their method to delimit species. We do, however, use their phenotypic divergence threshold of a phenotypic score of 7 or greater to denote what we call phenotypically highly divergent taxa. In terms of these genetic and phenotypic thresholds, population pairs were binned into four categories: a) those that are both phenotypically and genetically highly divergent (i.e., above these conceptual thresholds), b) those that are phenotypically highly divergent (i.e., a score of 7 or greater) but not genetically highly divergent, c) those that are highly genetically divergent (i.e., 5% genetic distance or greater) but not phenotypically highly divergent, and d) those populations whose divergence did not cross either conceptual high-divergence threshold. These bins represent different approaches to divergence and speciation (or lack thereof), and categorizing divergence among population pairs in such a way enabled us to test for heterogeneity among taxa. Specifically, we compared frequencies of Passeriformes versus non-Passeriformes species in each of the four bins, estimating chi-squared significance values assuming a null model of homogeneity. Separately from our categorical treatments, divergence data were examined as continuous (or semi-continuous) variables in divergence process space defined by phenotypic and genetic axes following [20].

With appropriate caution, we assume that populational divergences (phenotypic and genotypic) are independent between species, and we have a null expectation that these divergence processes will be independent with respect to higher taxonomy (and we emphasize that this assumption is not being applied to completion of the speciation process). In other words, we expect that divergence between populations within one species is not affected by similar processes occurring within another species, and we use a null model of phylogenetic independence for these processes (e.g., that divergence processes within species are not affected by what order the species is in). Note that this assumption or null model is independent of completed speciation; it considers divergence only. From reviews of genetic data we know this assumption to be violated in a strict sense (mutation rates vary among major avian lineages), but not to such a degree that it has prevented the utility and widespread application of this divergence metric [12]. For example, Weir and Schluter [45], in their review of another commonly used mtDNA gene cytochrome-b, found that “A molecular rate of approximately 2.1% (± 0.1%, 95% confidence interval) was maintained over a 12-million-year interval and across most of 12 taxonomic orders.” Similarly, we have chosen a phenotypic divergence metric that also has wide applicability [13], even though doubts remain about its universal effectiveness within the narrower scope of species delimitation [65, 66].

Without a reliable phylogeny that encompasses our sampled taxa, we cannot effectively test or correct for phylogenetic effects in our datasets, although we do examine such effects insofar as possible. Additionally, integrating genetic and phenotypic divergence data by making a bivariate plot in our divergence and speciation process space produced multiple pairwise comparisons within some species, introducing non-independence in some cases that confounds analyses. To correct for this when asking questions about possible higher-order effects, we averaged all within-species comparisons for species represented by more than two populations. We then used the corrected datasets to perform analyses of variance (ANOVA and MANOVA) to test for the effect of taxonomy on genetic and phenotypic divergence. Specifically, we tested for the effects of taxonomic order and family on divergence. Because many taxonomic orders were represented only by one species, we performed additional analyses treating all non-Passeriformes species as a single group (n = 14 species). Ordinary least squares regressions were also performed on single-lineage divergence estimates.

We do not have sufficient information to know whether all of our pairwise comparisons are between sister populations. To study the process of divergence and speciation within a lineage, it is common to focus on sister populations or taxa, and given our sampling it is likely that many inter-island comparisons are between sisters. However, when studying the divergence process among many lineages for overarching patterns of divergence, it is not necessary to include only sister lineages. In fact, sister populations are not needed to examine how genetic and phenotypic divergence generally proceed within a lineage. And, because sufficiently detailed intraspecific phylogenies do not yet exist for these 48 lineages (and this is not a phylogeographic study), our approach simply makes contrasts between and among populations. (We know that some comparisons are not between sisters, given that we have multiple populations of some species included, which we correct for when making among-lineage comparisons.) The presence of sister and non-sister lineages in the divergence process space does not materially affect the overarching understanding of how phenotype and genotype diverge—particularly among island populations such as these in which there is so little evidence of gene flow.

Divergence patterns

In considering how populations have diverged in the multidimensional speciation process space, two central results emerged. First, there may be higher-order taxonomic effects in how lineages are undergoing divergence; and, second, phenotypic divergence among lineages appears to have a logarithmic relationship with genetic divergence.

Support for heterogeneity among different lineages in the divergence or speciation processes occurring among them suggest that Passeriformes exhibit less phenotypic divergence relative to genetic divergence than other orders. Or, more precisely, the non-Passeriformes taxa in our study exhibited more plumage and morphometric divergence (traits likely under selection) relative to mitochondrial genetic divergence (likely to be neutral or nearly neutral) than Passeriformes taxa. These findings suggest that higher-order taxonomy probably affects divergence and speciation processes in this assemblage of lineages. For many birds, especially oscine Passeriformes, song can be more important for mate selection than plumage [67]. Because we did not measure vocalizations, phenotypic divergence in Passeriformes is probably underestimated here in the context of Tobias et al.’s [13] divergence quantification method. However, while large phenotypic divergences appeared in both Passeriformes and non-Passeriformes in this study, nearly all large genetic divergences (> 5%) appeared only in Passeriformes (Figs 1 and 2). Rates of mtDNA evolution are known to vary across taxa [45, 50, 68] and may be faster in Passeriformes than in other birds [46–49]. The heterogeneity in divergence patterns we observed between Passeriformes and non-Passeriformes might thus occur because of differential mtDNA mutation rates among birds and/or through underestimates of phenotypic divergence among oscine Passeriformes due to our exclusion of song. We also note that a different subset of Philippine birds (or a complete sampling), or using different high-divergence thresholds, might also affect these findings.

As noted in results, heterogeneity in higher-order taxonomic effects vanished when currently recognized full species were removed from analyses (18 lineages treated here as subspecies from 8 species, four each of Passeriformes and non-Passeriformes; Table B in S1 File). However, when we adopted a phylogenetic species approach higher-order taxonomic effects remained strong. Insofar as these questions involve lineage divergence in genetics and phenotype, examinations need not be restricted to divergences below the species level (particularly in a system where there is so little evidence of gene flow). Determining how and to what extent higher-order taxonomic effects affect the divergence process among Philippine birds will require more comprehensive study. This would ideally include more lineages both above and below the species level and more extensive genotypic and phenotypic data.

Overall, our data show an apparent logarithmic increase of phenotypic divergence with respect to genetic divergence (Fig 3, Table D in S1 File). This is of interest for two reasons. First, it suggests that speciation in this system might be dominated by selection promoting divergence rather than by a gradual accumulation of differences in allopatry (neutral processes would more likely be linear). Secondly, it suggests limits to phenotypic divergence over time in this assemblage of lineages. This is not unexpected, as there are likely to be evolutionary constraints on phenotypic divergence that exceed any constraints on the continued accumulation over time of putatively neutral genetic divergence. In other words, on the genetic axis there is relentless ongoing mutation promoting divergence, while on the phenotypic axis such predictably directional forces are lacking. Interestingly, Delmore et al. [69] found a similar relationship among an assemblage of North American migratory bird species. Such findings require further research into how phenotypic divergence accrues within and among lineages.

In considering the generalized routes of divergence toward speciation in Figs 1 and 2, we found substantial numbers of lineages that appear to have followed the general routes a and b. Population pairs that were both genetically and phenotypically highly divergent (Fig 1 bin A, Table 1 A1-12) are the most likely candidates for elevation to full biological species. Large genetic distances suggest long periods of evolutionary isolation, and large phenotypic differences imply divergent selection, although selection is difficult to disambiguate from phenotypic plasticity [70]. This group includes taxa with large divergences in the Dicrurus hottentottus and Otus megalotis complexes, species complexes that include multiple contentious subspecies and putative species [21, 35, 36, 71, 72]. Population pairs among 14 species that were highly divergent phenotypically but not genetically (Fig 1 bin B, Table 1 B1-19) also seem likely to have undergone divergent selection (with the caveat that phenotypic plasticity may influence these results). Although the species in this group did not exceed our genetic threshold of 5%, relatively large distances (e.g., 3–4%) separated many populations (e.g., Table 1 B16-23), and many may well be full biological species [13]. Some of the highly divergent populations in this group (e.g., within the Phapitreron leucotis and Chrysocolaptes lucidus species complexes) have recently been elevated to full species [36].

The third general route to speciation (c in Fig 2) was less well represented among our 48 lineages. Population pairs that were not phenotypically divergent but were separated by greater than 5% genetic distances (Fig 1, bin C) included 6 species of Passeriformes (Table 1, C1-8). The divergences observed here likely result from extended time in evolutionary isolation, as changes in mtDNA sequence data are usually interpreted to be neutral or nearly neutral [73, 74, 75]. Canalization of phenotypic characters, notably plumage color and pattern, and body size and shape, may contribute to the lack of phenotypic divergence observed between these oscine Passeriformes populations [20], as occurs in some other noteworthy avian genera (e.g., Scytalopus, Empidonax). The populations in this group (Table 1, C1-8) represent deep divergence between cryptic populations (e.g., Ixos philippinus guimarasensis separated from other subspecies by more than 12% genetic distance despite very similar phenotype) and may merit elevation to full biological species as recently done by Dickinson and Christidis [37].

Initial taxonomic implications

While we are cautious about applying simplistic threshold values to determine species limits, our data suggest that cryptic species remain within many currently recognized species of Philippine birds. Using conservative conceptual speciation thresholds we found at least 29 populations, currently recognized as Philippine endemic subspecies, that may with further study warrant consideration as full biological species. Average genetic distance from 96 within-species comparisons was 3.15%, and distances less than this separate many sympatric species in well-studied mainland systems [20, 52]. Our results support the suggestion of Lohman et al. (2010) that the lower rates of endemism in birds (compared to other vertebrates) in the Philippines may be an artifact of misclassifying distinct island populations of birds as subspecies, rather than species. Lohman et al. [10] also predicted that, upon further investigation, a more accurate measure of endemism among birds in the Philippines may exceed 50%, as occurs in other terrestrial vertebrates [18]. With highly divergent populations occurring in more than half of the species we studied, our data provide empirical support for this prediction. Taxonomic revision, taking into consideration multiple types of comparison (e.g., genomic comparisons, phenotypic scoring, behavioral and ecological traits, etc.) is needed in this system. Uniting phenotypic and genetic datasets, as we have done here, will prove essential to such revisions.

Implications for conservation

Taxonomic designation can have real-world consequences on the conservation of populations [19], and in the Philippines birds and other wildlife are severely threatened by anthropogenic forces [17, 18]. Our results reemphasize the urgent need for a reappraisal of Philippine avian diversity expressed by Peterson [29] and Lohman et al. [10]. Further research is also warranted on the 29 endemic populations in 25 species that our data suggest to be largely through the speciation process (e.g., estimates of gene flow and evaluation of traits promoting reproductive isolation). In addition, our results show surprisingly little haplotype sharing among populations (i.e., high rates of genetic endemism), revealing a level of diversity with important implications for management and conservation below the species level for populations that have likely not yet achieved speciation.