Intraspecific genetic variation and phylogeographic structure can be influenced by factors such as landscape features, environmental gradients, historical biogeography and organismal traits such as dispersal ability. Since deep genetic structure is often considered a precursor to speciation, identifying the factors that are associated with genetic structure can contribute to a greater understanding about diversification. Here, we use repurposed data to perform a global analysis of volant vertebrates (i.e. bats and birds) to estimate where intraspecific phylogeographic breaks occur and to identify the factors that are important predictors of these breaks. We estimate phylogeographic breaks using Monmonier’s maximum difference barrier algorithm and conduct a random forests analysis using the presence of a phylogeographic break as a response variable. In bats, phylogeographic breaks are concentrated in biodiversity hotspots while breaks estimated in bird species are more widespread across temperate and tropical zones. However, for both clades, geographical features such as maximum latitude, measures of wing morphology and organismal traits associated with feeding ecology were found to be important predictors of phylogeographic breaks. Our analysis identifies geographical areas and suites of organismal traits that could serve as a starting point for more detailed studies of biodiversity processes.

bats, birds, dispersal, macrogenetics, Monmonier’s algorithm, phylogeography, Random Forests

Intraspecific genetic variation is a featured aspect of various models of speciation (Lande 1980; Orr 1996; Nosil 2012). One central tenet of evolutionary theory is that natural selection acts on variation within species to drive diversification (e.g. Grant (1981)). For this reason, entire disciplines have been devoted to characterising intraspecific genetic variation and identifying the landscape features that may influence this variation (e.g. Avise (2000); Manel and Holderegger (2013)). While synthesis has been impeded by confounding factors such as life history traits, environmental variation and biogeographic history, such work has resulted in the accumulation of millions of sequences in databases such as NCBI GenBank (e.g. Garrick et al. (2015)). Given the availability of biodiversity data in natural history collections (e.g. iDigBio, idigbio.org), the aggregation of specimen collection records in data portals (e.g. GBIF, The Global Biodiversity Information Facility (2024)) and collections of phenotypic data (e.g. MorphoBank, O’Leary and Kaufman (2011)), it now appears to be an attainable goal to understand the determinants of intraspecific genetic diversity across broad taxonomic groups (Barrow et al. 2021).

One key aspect of life history that is likely to influence the development of intraspecific genetic variation is dispersal ability, particularly the intrinsic and extrinsic factors that limit the ability of individuals of a species to move throughout the landscape. A lack of dispersal can lead to the accumulation of genetic differentiation due to genetic drift and other forces (Slatkin 1987) and, as such, dispersal limitation has been identified as a key factor in biological processes ranging from speciation rates (e.g. Polato et al. (2018)) to community assembly (e.g. Pigot and Tobias (2015)). Research into empirical systems has indicated that extrinsic landscape features such as mountains (e.g. Kane et al. (2019)), rivers (e.g. Myers et al. (2020)) and environmental gradients (e.g. Sebastián et al. (2021)) can limit dispersal and lead to the formation of intraspecific genetic variation.

Long-term limitation of gene flow between populations, with or without a physical barrier to gene flow (e.g. Irwin (2002)), can result in phylogeographic structure. Phylogeographic structure is often characterised using summary statistics that describe genetic variation in populations (e.g. F_ST), via the identification of population genetic clusters (e.g. PCA) or by testing for spatial autocorrelation (e.g. isolation by distance). Considerably less effort has been devoted to developing theory and methodology for estimating where barriers to gene flow occur. One notable exception is Monmonier’s algorithm (Monmonier 1973), which can be applied to infer spatial genetic structure by converting the genetic data into a network, identifying the edges of the network that are associated with the highest degree of genetic differentiation and projecting the maximum difference boundary on to a spatial connection network (e.g. Manni et al. (2004)). The linear boundary produced by Monmonier’s algorithm is placed on a map without any input from external information such as geospatial or environmental layers, making the algorithm well-suited for macrogenetic investigations which seek to identify the factors that influence genetic structure, but may rely on data that are unevenly sampled across species. Since geographic distance is typically correlated with both environmental distance and landscape features (e.g. Wang (2013); Pelletier and Carstens (2018)), any inference method that incorporates geographic distance amongst samples will likely be influenced by these same factors. By circumventing this problem, Monmonier’s algorithm enables environmental distance and landscape features to be evaluated as potential factors which may contribute to the formation of phylogeographic breaks.

Due to the perception that volant species have the capacity for long distance dispersal, phylogeographic breaks in these species may be more likely to correspond to environmental conditions, differences in food resources and overall habitat suitability than to geophysical barriers. In bats, patterns of phylogeographic structure vary across species on a global scale, but are thought to be influenced by factors including social structure, mating behaviour (e.g. autumnal swarming), migration patterns, habitat connectivity, ecological gradients and geographic barriers (Moussy et al. 2013). In a recent review, Hernández-Canchola et al. (2021) noted that Quaternary climate fluctuations, geographic features such as oceans and rivers and ecological processes such as niche differentiation were the most referenced contributors to phylogeographic structure in bat species. In birds, feeding ecology and annual dispersal were the strongest predictors of high diversification rates (Phillimore et al. 2006), suggesting that species ecology plays a large role in diversification and speciation. Morphological proxies of dispersal show a nuanced response with respect to diversification (Claramunt et al. 2011; Tobias et al. 2020). Social behaviour, particularly song, is also thought to be important at maintaining diversity across the landscape (Uy et al. 2018).

To assess global phylogeographic patterns in volant vertebrates and to identify factors that are important in predicting phylogeographic breaks, we repurposed georeferenced mitochondrial sequence data available from phylogatR (Pelletier et al. 2022). This database that aggregates geographic information from specimen repositories such as GBIF (The Global Biodiversity Information Facility 2024) and sequence data from GenBank (Benson et al. 1993) and BOLD (Ratnasingham and Hebert 2007) stores these data in a convenient framework. While challenges related to data acquisition, quality control and unevenness can be substantial and should be accounted for (Leigh et al. 2021; Pelletier et al. 2022), we apply an automated analysis pipeline modified from previous investigations (Pelletier and Carstens 2018; Barrow et al. 2021; Parsons et al. 2022) to estimate phylogeographic breaks in hundreds of bird and bat species. Bats and birds were chosen for the study as they are both volant vertebrates with high ecological diversity and are generally considered to have high dispersal ability. Additionally, birds tend to be well studied in comparison to bats and, as such, we hoped that the additional avian data would corroborate models of intraspecific genetic structure in bats. We then use machine learning to identify traits that are predictive of phylogeographic breaks in these taxa.

Consistent with previous investigations into population genetic structure (e.g. Martin and McKay (2004); Pelletier and Carstens (2018)), many top predictors of phylogeographic breaks in bats and birds include range characteristics such as occurrence area and measures of latitude. In our study, bird species that contained phylogeographic breaks exhibit maximum latitudes closer to the Equator, consistent with the view that tropical bird species are expected to be more sedentary, have decreased dispersal ability (Sheard et al. 2020) and have more genetic population structure (e.g. Brawn et al. (1996)). Conversely, bat species with a higher maximum latitude more often exhibit phylogeographic breaks, a pattern likely driven by low sampling south of the Equator and the concentration of breaks found north of the Equator in southern Central America and southeast Asia (Suppl. material 1: fig. S5). Additionally, the presence of physical barriers such as a mountain range can still play an important role in structuring genetic diversity even in volant vertebrates as such barriers were a top predictor of phylogeographic breaks in bats. While a recent meta-analysis of phylogeographic breaks in North American mammals found no significant association between the location of phylogeographic breaks in bats and distance to mountains (Jensen et al. 2024), our results indicate that mountains can still be important drivers of phylogeographic structure in highly vagile species. However, many species in both the bird and bat datasets are widely distributed across mountainous regions and lack a phylogeographic break. The organismal traits that facilitate connectivity across mountain ranges are likely related to those required by powered flight, particularly features of the wing that influence flight performance and efficiency.

Most of the organismal traits that were found to be important predictors of phylogeographic breaks in bats and birds are associated with wing morphology (Fig. 3) and feeding ecology. Wing loading (i.e. the ratio of mass to wing surface area) and wingspan are likely to be important because these factors influence the efficiency of flight in flying objects that range in size from Drosophila to the Airbus A380. In bats, higher wing loading is positively correlated with flight speed and distances (Norberg and Rayner 1987) and genetic connectivity (Taylor et al. 2012), whereas high aspect ratios are found in species with long, thin wings that exhibit lower drag and increased aerodynamic efficiency (Norberg 1995). Both characteristics are likely correlated with higher dispersal ability, larger geographic ranges (e.g. Luo et al. (2019)) and greater potential to traverse physical barriers in the landscape (Burns and Broders 2014). We found that body-size corrected wingspan was a top predictor of phylogeographic breaks in bats and is significantly smaller in species which exhibit a phylogeographic break (Fig. 2B). Wing loading was amongst the top ten important predictors in bats and was higher in bats that were estimated to not contain a phylogeographic break, though this relationship was not statistically significant (p = 0.201; Suppl. material 1: table S1).

Figure 3. 

A) Bat illustration with wingspan measurement indicated and equation for wing loading; B) Bird with important morphological measurements indicated and the equation for Kipp’s distance. Illustrations by Sydney K. Decker.

Similarly, in birds, increased dispersal ability with more efficient long-distance flight is associated with elongated wings with higher hand-wing index, an approximation of aspect ratio (Claramunt et al. 2011). Our results are consistent with findings that bats and birds with wing morphologies adapted for more efficient flight, such as higher hand-wing index, exhibit fewer phylogeographic breaks, possibly due to increased dispersal ability and genetic connectivity across their geographic ranges (Fig. 2D).

Wing shape is also implicated in bat and bird feeding ecology as determinants of how species can navigate their environment to forage (e.g. Norberg and Rayner (1987); Sheard et al. (2020)). In birds, habitat specialisation and diet specialisation are interrelated (Reif et al. 2016). We suspect that this is also true of bats, although we cannot identify a comparable study. However, it is well established that habitat variation is associated with dietary differences within species (e.g. Clare et al. (2011)), that habitat influences the guild structure of bat communities (e.g. Rocha et al. (2018)) and that wing shape influences both the extent of and types of specialisation in feeding ecology (e.g. Fenton (1982)). In birds, invertivory and granivory trophic niche categories were included in top predictors of phylogeographic breaks. Wing morphology in passerines is associated with diet, particularly as it relates to foraging behaviour and food availability (Sheard et al. 2020). Variation in genetic differentiation across biogeographic barriers in South American bird species is also explained by foraging stratum (Burney and Brumfield 2009), suggesting a considerable interaction between feeding ecology and genetic diversity.

Phylogeographic breaks are not necessarily indicative of species-level diversity, as phylogeographic patterns can be caused by social structure or sex biased dispersal (e.g. Dávalos and Russell (2014)), historically allopatric populations (e.g. Fleming et al. (2010)), limited dispersal or small population sizes (Irwin 2002). Some phylogeographic breaks estimated here correspond to boundaries of named sub­species, for example, the eastern and western subspecies of the North American Wilson’s Warbler (Cardellina pusilla; species specific results available on https://github.com/skdecker/PhylogeographicBreaks). Nevertheless, phylogeographic structure and intraspecific genetic diversity is positively associated with speciation rates in some taxa, for example, in New World birds (Harvey et al. 2017), while no association is apparent in other groups such as the diverse radiation of Australian lizards (Singhal et al. 2018). Furthermore, a large portion of phylogeographic breaks estimated here occur in the Tropics, indicating that, within species, phylogeographic structure follows similar latitudinal gradients as intraspecific genetic diversity (e.g. Smith et al. (2017); Fonseca et al. (2023)), predictions of cryptic species (Parsons et al. 2022) and species richness (e.g. Gaston (2000); Mittelbach et al. (2007)).

While repurposed, single locus, mitochondrial data may not be representative of overall patterns of population genetic structure (e.g. Bazin et al. (2006); Galtier et al. (2009)), extensive data quality control steps and established automated analysis pipelines allow initial investigations of global biodiversity patterns (e.g. Barrow et al. (2021)). We used rigorous filtering to exclude samples that were highly differentiated from other sequences in species alignments, which could cause an underestimation of deep phylogeographic breaks. However, it was our observation that most sequences excluded at this step were from species with known cryptic lineages currently undergoing taxonomic revision and likely reflect species-level boundaries rather than intraspecific phylogeographic breaks (e.g. Kerivoula hardwickii; Tu et al. (2018)).

The use of Monmonier’s algorithm to identify phylogeographic breaks has been fairly limited despite its long history as an estimate of population structure. For example, Dupanloup et al. (2002) compared Monmonier’s algorithm with spatial analysis of molecular variance (samova) and found that the algorithm performed better than samova in finding genetic breaks of simulated single locus data. However, across different methods of phylogeographic break estimation, isolation by distance (IBD) can be misidentified as a break if sampling is sparse across the range of a species (e.g. Templeton et al. (1995); Irwin (2002)). As we used repurposed data with imperfect sampling, it is likely that some estimated breaks represent sampling gaps; however, breaks were often estimated between samples that were not the most geographically distant. For example, if sampling gaps were the largest contributor to estimation of phylogeographic breaks, we would expect a break to be estimated that separate Desmodus rotundus samples in southern Brazil from all other samples, rather than the estimation of a break between eastern and western populations (Fig. 1B).

Once phylogeographic breaks were estimated, 50 independent RF classifiers were trained due to low accuracy and instability of variable importance across preliminary models, likely caused by the use of small datasets. It was our intention that analysis of phylogeographic breaks in birds would yield a more accurate model due to increased availability of data compared to bats; however, over 89% of bird species in the original download from phylogatR were removed by our data quality control filters. To verify that low model accuracy was due to small sample size and intrinsic variation, we trained a second RF classifier using IBD, a more widely used metric of how intraspecific genetic diversity is structured across the landscape, as the response variable against the trait datasets (Suppl. material 1). The models trained with IBD also exhibited low accuracy (66.67% for the bat classifier, 60.88% for birds; Suppl. material 1: table S3), this secondary analysis suggesting that low accuracy in preliminary models is not due to the phylogeographic break response being uninformative. Though overall accuracy of both bat and bird models was still relatively low, averaging across multiple independent classifiers allowed us to have more confidence in the measures of predictor variable importance. As RF models predict a complex of traits that interact to affect the response variable, measures such as MDA can help tease apart these relationships to infer biological meaning of influential traits even when there is no statistical difference in trait variables between classes.

Moreover, traits do not evolve independently of one another and it is important to consider the effects of phylogeny in a comparative framework. In the RF analysis, we used taxonomic classification as a proxy for phylogeny and found that only the avian orders Procellariiformes and Caprimulgiformes were amongst the top ten important predictors of phylogeographic breaks in birds (Fig. 2C), while no taxonomic family was a top ten predictor of breaks in bats (Fig. 2A). Notably, the Order Procellariiformes exhibit extreme wing morphologies because they are adapted to foraging over the open ocean and diving for prey (Warham 1996). This specialised ecological niche may facilitate population connectivity, as none of the Procellariiformes species in our analysis was estimated to exhibit a phylogeographic break. Of the Caprimulgiformes species in our analysis, all are predicted to contain a phylogeographic break possibly corresponding to the presence of 6–13 named subspecies in each of these species. This strongly implies that phylogenetic effects are not the primary explanation for the results reported here. Our analysis focuses on traits that influence breaks comprehensively across species, while there likely exist more context-dependent phylogeographic responses to environmental conditions and dispersal ability within individual species. To further explore the effect of phylogeny on the variables in our dataset, we conducted a mean decrease in accuracy (PFA; Suppl. material 1). PFA is a multivariate approach to understanding trait evolution under a Bayesian phylogenetic framework. While the PFA did find that some morphological, geographic and ecological trait variables likely did combine to influence diversification, the phylogeographic break variable did not contribute significantly to any of the identified latent evolutionary factors (see Suppl. material 1). This supports the assertion that phylogeographic breaks, as we have estimated here, are not strongly influenced by phylogeny.

Our investigation is a contribution to the growing body of research described as “macrogenetics” (Blanchet et al. 2017) or “automated comparative phylogeography” (Gratton et al. 2017). This research relies on the collection and re-analysis of genetic data that were originally collected for investigation into a single species in order to make inferences about large-scale evolutionary processes. Facilitated by the accession of these data in databases such as NCBI GenBank (Benson et al. 1993) and the development of data aggregators such as phylogatR (Pelletier et al. 2022), researchers now have the capacity to ask in-depth questions about genetic diversity in a particular clade (e.g. Manel et al. (2020); Theodoridis et al. (2021); French et al. (2023)) or to investigate patterns across clades on a global scale (Miraldo et al. 2016; Fonseca et al. 2023). Macrogenetic investigations are subject to important questions about scale and over-interpretation (Paz-Vinas et al. 2021, but see Millette et al. (2021)) and it is important to recognise the limitations of single locus data (Edwards and Bensch 2009; Knowles 2009). Given these difficulties and limitations, we recognise that macrogenetic investigations enable novel synthesis of biodiversity data. Our data were drawn from databases containing genetic, environmental, geographic and organismal trait data. These data can be analysed synthetically using machine learning or other AI approaches (e.g. Carstens et al. (2018); Parsons et al. (2022); Yang et al. (2023)). Our study, conducted using data from two clades of volant vertebrate animals, indicates that feeding ecology is likely a more important influence on intraspecific genetic structure than has been previously appreciated. It could be that this result is due to the unique dispersal abilities of the focal clades or it may be that this also holds true for non-volant vertebrates.

Volant vertebrates are not equal in their dispersal capabilities. Our macrogenetic investigation and random forests analyses suggest that species which are capable of flying long distances are less likely to contain intraspecific genetic structure. Phylogeographic breaks tend to be associated with larger ranges at higher latitudes in bats and ranges closer to the Equator in birds. Additionally, ecological traits including habitat breadth in bats and migration, granivory and invertivory in birds were important predictors in our analysis, suggesting an intricate interplay between organismal traits and features of their geographic range contributing to phylogeographic structure. Measures of wing morphology, such as shorter wingspans in bats and lower hand-wing index in birds, were also important predictors of phylogeographic breaks. Since species-rich clades in bats (i.e. Vespertilionidae) and birds (i.e. Passerines) generally contain species that exhibit these wing morphologies, this suggesting that wing shape may play a role in diversification rate shifts within the Order Chiroptera and Class Aves. While diversification rates appear to be influenced by wing morphology in moths (Aiello et al. 2021) and there has been some exploration of these factors within Class Aves (e.g. Kennedy et al. (2016)), a comprehensive phylogenetic investigation of diversification rates and wing morphology in Chiroptera and Aves is needed.

We thank members of the Carstens lab for their helpful comments on this work. Support for this work was provided by the National Science Foundation (NSF) (DBI-1661029 and DBI-1910623 to B.C.C.) and the Ohio Supercomputing Center (PAA1174).

Sydney K. Decker: Conceptualisation, Methodology, Data curation, Analysis, Writing – Original draft, Writing – Review and editing. Kaiya L. Provost: Methodology, Data curation, Analysis, Writing – Review and editing. Bryan C. Carstens: Conceptualisation, Methodology, Writing – Original draft, Writing – Review and editing, Supervision, Funding acquisition.

Code and source data, including DNA sequence alignments, trait data and analysis files, used in the manuscript are available at https://github.com/skdecker/PhylogeographicBreaks. All other data are provided in Suppl. materials 1, 2.