Comparative phylogeography of mainland and insular species of Neotropical molossid bats (Molossus)

Abstract Historical events, habitat preferences, and geographic barriers might result in distinct genetic patterns in insular versus mainland populations. Comparison between these two biogeographic systems provides an opportunity to investigate the relative role of isolation in phylogeographic patterns and to elucidate the importance of evolution and demographic history in population structure. Herein, we use a genotype‐by‐sequencing approach (GBS) to explore population structure within three species of mastiff bats (Molossus molossus, M. coibensis, and M. milleri), which represent different ecological histories and geographical distributions in the genus. We tested the hypotheses that oceanic straits serve as barriers to dispersal in Caribbean bats and that isolated island populations are more likely to experience genetic drift and bottlenecks in comparison with highly connected ones, thus leading to different phylogeographic patterns. We show that population structures vary according to general habitat preferences, levels of population isolation, and historical fluctuations in climate. In our dataset, mainland geographic barriers played only a small role in isolation of lineages. However, oceanic straits posed a partial barrier to the dispersal for some populations within some species (M. milleri), but do not seem to disrupt gene flow in others (M. molossus). Lineages on distant islands undergo genetic bottlenecks more frequently than island lineages closer to the mainland, which have a greater exchange of haplotypes.

main regions. The Lesser Antilles are located on the eastern margin of the Caribbean tectonic plate, and most of these islands were formed more than 20 Ma ago by volcanic activity and have never been connected (Bender et al., 1979;Donnelly, 1988;Pindell, 1994).
In contrast, the Greater Antilles are much older and were formed during the separation of North and South America 170 Ma ago, with many islands remaining united until the Eocene (Iturralde-Vinent & MacPhee, 1999;Kerr, Iturralde-vinent, Saunders, Babbs, & Tarney, 1999;Pindell and Barrett, 1990). Pitman, Cande, Labrecque, and Pindell (1993) hypothesized that during dry periods in the Eocene, a land bridge formed connecting the Greater Antilles with Middle America, potentially resulting in faunal exchange between these landmasses. Iturralde-Vinent and MacPhee (1999) also proposed that a short-lived land bridge connected the Greater Antilles to northwestern South America during the Oligocene, promoting more recent faunal exchange.
Individual species and genera of bats are some of the few extant, native mammals in the Neotropics that occur on both the continental mainland and Caribbean islands. Bats are the second most speciose order of mammals and occupy almost all parts of the globe (Patterson, Willig, & Stevens, 2003), and their broad distribution renders them useful in comparative phylogeographic analyses. They are the only mammals capable of true flight and the dispersal ability of some groups allows these bats to colonize large geographic areas, including oceanic islands (Dávalos, 2007;Russell et al., 2016;Speer et al., 2017). Dispersal is essential to promote gene flow within a species, and the study of barriers that could isolate populations may provide important insights regarding how current and historical evolutionary processes effect speciation (Miller-Butterworth, Jacobs, & Harley, 2003;Tam et al., 2005). Bats with high dispersal abilities usually exhibit little genetic structure among populations due to high rates of gene flow (Carstens, Sullivan, Davalos, Larsen, & Pedersen, 2004;McCracken, McCracken, & Vawter, 1994;Pumo, Goldin, Elliot, Phillips, & Genoways, 1988). These bats are potentially less affected by habitat disturbance and genetic fragmentation than more sedentary groups (Ibáñez, García-Mudarra, Ruedi, Stadelmann, & Juste, 2006;Meyer, Kalko, & Kerth, 2009), although some exceptions have been reported in species of Molossidae inhabiting insular systems (Speer et al., 2017).
In the Caribbean, there are more than 60 species of bats (Dávalos, 2004(Dávalos, , 2006Loureiro, Gregorin, & Perini, 2018;Velazco & Patterson, 2013), but not much is known about the capacity of the different species to disperse among islands. Koopman (1977) suggested that oceanic straits present functional barriers for dispersal in bats in the Caribbean. Similarly, Genoways (1998) proposed that migration between islands was unlikely and predicted that gene flow among island populations was infrequent. Populations of bats are vulnerable to Caribbean hurricanes and volcanic eruptions, which may reduce population sizes and possibly result in accidental dispersal (Pedersen, 1998;Pedersen, Genoways, & Freeman, 1996). Therefore, even in the absence of regular inter-island migration, the genetic diversification among some island populations could be muted by episodic gene flow. Likewise, populations from small and distant islands might be expected to be subjected to genetic drift more frequently than populations from large and less isolated islands, which can potentially decrease the genetic variability on small islands (Nei & Tajima, 1981). Previous studies have shown that although bats have significant capacity for dispersal, ocean straits may act as a barrier for some groups (Carstens et al., 2004;Fleming & Racey, 2013;García-Mudarra, Ibáñez, & Juste., 2009;Larsen et al., 2012;Speer et al., 2017), but may not impose a strong barrier for others (Carstens et al., 2004;García-Mudarra et al., 2009;Larsen et al., 2007;Pumo et al., 1988). This pattern might also have been affected by lower sea levels during the Pleistocene that shortened overwater distance, decreasing the oceanic barrier among some islands (Velazco & Patterson, 2013).
Mastiff bats of the genus Molossus represent an ideal model system for the study of population structuring on a broad geographic scale. Molossus are common aerial insectivores that inhabit a large range of habitats, from dry and humid semideciduous forests and tropical rainforests to pastures and savannas (Eger, 2008;Reid, 2009). Many species in this genus are well-adapted to anthropogenic modifications, and they can be numerous in urban areas and degraded habitats (Taylor et al., 2019). Molossus is nonmigratory, but many species are widely distributed and occur on both sides of prominent geographic barriers (e.g., Andes Mountains, Caribbean Sea) (Dolan, 1989;López-González & Presley, 2001).
Several species of Molossus are environmental generalists and are broadly distributed in the Neotropics, including on islands in the Caribbean. In contrast, other species in the genus are restricted to either mainland or Caribbean islands and prefer specific types of habitats, such as dry grasslands (Taylor et al., 2019). The extensive distribution of Molossus throughout the Neotropics, including the Caribbean, suggests a strong colonizing ability and capacity to fly or to be carried by wind currents and storm systems over water.
Although no studies have measured vagility in Molossus, other molossids have been reported to fly up to 160 km in a single night , to reach speeds over 50 km/hr, to fly for up to 10 hr without resting (Marques, Rainho, Carapuço, Oliveira, & Palmeirim, 2004), and to migrate long distances (Cockrum, 1969;Glass 1958;Russell, Medellín, & Mccracken, 2005). Additionally, bats in the family Molossidae have relatively long, narrow wings with a reduced area, resulting in high wing loadings and high aspect ratios (Norberg & Rayner, 1987). This suite of adaptations is commonly associated with fast, long-distance flight and enhanced dispersal abilities (Peterson, Eger, & Mitchell, 1995;Taylor et al., 2012;Burns & Broders, 2014).
Molossus molossus is larger than M. coibensis (Dolan, 1989) and is the most common and broadly distributed species of the genus, occurring from Argentina to the southern United States and the Lesser Antilles (Barquez, Mares, & Braun, 1999;Dolan, 1989;Eger, 2008;Fabian & Gregorin, 2007). This species also is present on both the west and east sides of the Andes. Several geographic populations were originally described as subspecies of M. molossus based on morphological characters but have been relegated as synonyms based on molecular and further morphological analyses (Dolan, 1989;Loureiro et al., 2019;Loureiro, Gregorin, et al., 2018 Several other supposed subspecies of M. molossus have been described that are confined to one or a few Caribbean islands (Dolan, 1989;Loureiro, Gregorin, et al., 2018). However, based on morphological characters, all Caribbean species and subspecies of Molossus were previously synonymized under the name M. molossus (Dolan, 1989;Eger, 2008;Simmons, 2005). Recent studies based on mitochondrial and nuclear genes, however, demonstrated that M. verrilli from the Dominican Republic, and M. milleri from Cuba, the Cayman Islands, and Jamaica are distinct species, restricting the distribution of M. molossus in the Caribbean to the Lesser Antilles and Puerto Rico (Lim et al., 2017;Loureiro et al., 2019;Loureiro, Gregorin, et al., 2018). M. milleri is morphologically similar to M. verrilli and M. molossus and occupies both forests and urban areas in the Greater Antilles (Taylor et al., in press).
Species within Molossus have low genetic variability, and mitochondrial and nuclear markers are often insufficient to resolve the relationship among some morphologically distinct species . However, the use of a next generation sequencing approach (NGS) has shown promise for resolving relationships and population structure in recently diversified groups in which the rate of genetic change is low (Cronin et al., 2015;Enk et al., 2016;Kusza et al., 2014;Lozier, 2014). One of these approaches is the genotyping-by-sequencing (GBS) method, which sequences many small tags in the genome flanking restriction sites. By this method, thousands of single nucleotide polymorphisms (SNPs) are recovered, vastly increasing the size of the overall dataset compared to typical Sanger methods. With this large dataset, consistent variance can be detected among genetically similar groups that are not revealed by standard gene sequencing approaches. In addition, genomic-scale SNP data provide powerful options for testing patterns of genetic structure and demographic trends and to estimate population parameters for identifying changes in population size over time (Excoffier, Dupanloup, Huerta-Sánchez, Sousa, & Foll, 2013;Gutenkunst, Hernandez, Williamson, & Bustamante, 2009;Lozier, 2014). These estimates may be important in a conservation context because they can indicate if a population has the potential to undergo inbreeding depression or has the genetic bandwidth to adapt to future environmental changes (Sovic, Carstens, & Gibbs, 2016).
Herein, we use the GBS approach on three species of Mastiff bats (M. molossus, M. coibensis, and M. milleri) with different ecological histories and geographical distributions to explore population genetic parameters and better understand the role of geographic barriers in dispersal and gene flow in bats. We tested the hypothesis proposed by Koopman (1977) and supported by Genoways (1998) that oceanic straits serve as barriers to dispersal by Caribbean bats. If this hypothesis is correct, we would expect significant genetic structuring and low gene flow among island populations and between islands and mainland populations when compared to levels of variation within island populations or within the mainland.
We also tested the hypothesis proposed by Nei and Tajima (1981) that more isolated populations are likely to experience genetic drift and bottlenecks relative to less isolated ones. To support this hypothesis, we would expect to find a decrease in the effective population size in island populations and a constant or increasing effective population size on the mainland. We would also expect to find that island populations with a mainland source are less affected by bottlenecks than insular populations isolated from other landmasses.
As a result of different isolation patterns, we expect divergent phylogeographic structuring in mainland and insular species based on geographic barriers, habitat selection, and historical fluctuation in climate.

| Sample collection and library preparation
We obtained tissue samples from 62 M. molossus from South America, Middle America, and the Lesser Antilles, 20 M. coibensis from South and Middle America, and 19 M. milleri from the Greater Antilles ( Figure 1; Appendix 1). Tissues samples included skeletal muscle, liver, heart, and kidney and were preserved in 95% ethanol or were frozen in liquid nitrogen upon collection of the specimen in the field. DNA extraction was conducted using Qiagen DNeasy extraction kit (Qiagen, Inc.) following standard protocols. Genomic DNA quality was checked by visual inspection on an agarose gel, and the DNA concentration was measured using a Nanodrop spectrophotometer (Nanodrop Technologies). We used 30 μl of DNA samples with concentrations higher than 100 ng/μl for library preparation and for the genotyping-by-sequencing approach following the protocol described by Elshire et al. (2011). All libraries were sequenced on an Illumina HiSeq 2000 in the Cornell Institute of Genomic Diversity (IGD).

| Genotyping
De novo genotyping was performed using the Universal Network-Enable Analysis Kit (UNEAK) pipeline, available on TASSEL 3.0 software (Bradbury et al., 2007). The sequences were trimmed to a 64 bp length, and shorter reads were discarded. In this pipeline, identical reads are clustered into tags and all unique tags are merged.
No reference genome or GBS reference sequences of any species & Rossiter, 2016). All three species were pooled and aligned for the reference genotyping before the dataset for each species was filtered and analyzed separately. Quality control and filtering of the reference genotypes of each species sample were also conducted on TASSEL 3.0 (Bradbury et al., 2007). Tags with depth lower than seven were treated as missing. The minimum allele frequency (MAF) value of 0.02 removed almost half of the SNPs from the original dataset of both the de novo and reference genotyping analyses, while any value above 0.02 had a very small impact on number of SNPs ( Figure S1), which could suggest that these removed SNPs might represent informative rare alleles, rather than sequencing errors. The increase of the MAF value may cause an underestimation of heterozygotes because it may remove rare alleles, instead of the removal of sequencing errors, with the loss of biological information (Ni and Stoneking, 2016). Kim et al. (2011) argued that for rare SNPs (e.g., MAF < 0.01) it is not easy to differentiate between sequencing errors and a true rare allele, and alleles with less the 1% of MAF should be discarded. Linck and Battey (2019) showed that highly accurate population inferences are reached when relatively rare alleles are included (minimum allele count 2% to 8%). Therefore, in this study we set the MAF value at 2%.
We discarded individuals with more than 20% missing data. We also removed invariant SNPs and those with more than 10% missing data. The minimum heterozygosity proportion was set to 0.01.
To remove linked sites in the alignment, SNPs <128 bp apart were removed. We tested for deviations of Hardy-Weinberg equilibrium in each dataset using TASSEL 3.0 (Bradbury et al., 2007). We estimated kinship between individuals within each species by exploring the relationship between identity by state (IBS), when two or more individuals share similar nucleotide sequences, using TASSEL 3.0 to ensure results were not conflicted by kinship (Rodríguez-Ramilo & Wang, 2012).

| Population structure analysis
We assigned individuals to populations under an admixture model with correlated allele frequencies using Structure v.2.3.4 (Pritchard, Stephens, & Donnelly, 2000). This model assumes that the common ancestor of all populations passed part of its genotype to all its descendants (Falush, Stephens, & Pritchard, 2003). The possible number of populations (K) in each species was estimated. To rule out population substructuring within samples of individuals for each species, we allowed K to range from 1 to a number in excess of the geographic locations. For example, we obtained samples of M. milleri from four different Caribbean islands, but we allowed K to vary between 1 and 10 ( Figure S2). Five runs for each K were analyzed under a model of admixture and correlated allele frequency for 10 million generations each. The log likelihoods for each K value were averaged among runs and verified using log (Alpha) plots by interaction and ln L (K) by interaction. We used Structure Harvest v0.694 (Earl & vonHoldt, 2011) to assess the most likely number of populations, using the results of the Structure analyses. Patterns of individual assignment to clusters were also used to make an optimal inference regarding the K value for each species.
We assessed population differentiation of each species by conducting a principal component analysis (PCA) of pairwise individual genetic distances among populations and discriminant analysis of principal components (DAPC). We conducted these analyses converting the observed SNP data into principal components that summarize the variation between samples using the R package poppr (Kamvar, Tabima, & Grünwald, 2014). The relationships among clades within each species were investigated through a coalescence approach, which accounts for differences in genealogical histories of individual loci using the program SVDquartets (Chifman & Kubatkoin 2015) implemented in PAUP 4.0 (Swofford, 2003). Four independent runs were conducted to assess topological convergence, each including 500 bootstrap replicates and exhaustive quartet sampling.
Trees were visualized using FigTree v. 1.4.3. We also generated a genetic distance tree to represent the genetic relatedness of the samples based on the UPGMA algorithm, with 500 bootstrap replicates.
After the populations were defined, we visualized the posterior assignment of each sample using a composite stacked bar plot.
The Fst was performed by calculating genetic distance based on all markers (after quality control and filtering) using a weighted analysis of variance (Cockerham, 1973;Weir & Cockerham, 1984). An analysis of molecular variance (AMOVA) for all pairs of populations was used to compute the divergence among populations using Arlequin 3.5.

| Demographic inference
We generated joined population folded site frequency spectra (SFS) for each population within a species using Arlequin 3.5 (Excoffier & Lischer, 2010) with 100 bootstraps. To infer historical demography patterns in all populations, each SFS was imported into FastSimCoal26 (Excoffier et al., 2013) and the likelihood of three demographic models was compared using the likelihood ratio test (LRT) and Akaike's information criterion (AIC). The maximum log likelihood given the observed SFS was calculated to determine how well each model fit the data. The first model represents a constant population size, the second model represents a population expansion, and the third model imposes a bottleneck in each population. The last two models assume changes in population size and also calculate the present population size, the ancestral population size, and the time in number of generations since the growth or decline began.
All parameters used in the demographic models were selected from a uniform distribution, and we considered the mutation rate as 2.5 × 10 −8 per nucleotide per generation. This value was calculated from human genomic data (Keightley, 2012;Nachman & Crowell, 2000) and is thought to provide a conservative estimate of population genetic parameters in mammals, including bats (Sovic et al., 2016). We calculated the confidence interval for each variable by performing a parametric bootstrap using point estimates from each parameter. For each model, we performed 100 runs (500,000 simulations per run) on FastSimCoal26 (Excoffier et al., 2013) and used the highest likelihood value in the LRT and AIC comparison. To ensure simulations were not stuck in a local maxima, individual runs of FastSimCoal26 under each model were tested for consistency and similar likelihood. Linkage disequilibrium is not common within subpopulations, except between very close or adjacent sites. In addition, isolated populations are less likely to exhibit strong correlations among loci (Pritchard et al., 2000). The initial removal of SNPs <128 bp apart likely removed almost all linked loci from our dataset, since only a few linked loci were found after the application of this filter. In addition, the low admixture among populations within all three study species might have also reduced linkage disequilibrium in our data.

| GBS data
The IBS analysis did not reveal any closely related individuals in any of the three species, and no individuals were discarded ( Figure   S2).

| Population structure analyses
The Structure analyses indicated a best fit of K = 3 for M. coibensis, K = 3 for M. molossus, and K = 2 for M. milleri (Figures 1 and   2a; Figure S3). Structure plots based on higher K values did not show any substructuring within populations ( Figure S4 In M. milleri, samples were divided into two distinct populations, one from Jamaica and the other from the Cayman Islands and Cuba (Figure 2b; Figure S5).

| Phylogenetic relationships
The phylogenetic and distance trees generated for each species also support the genetic structure found in the other analyses, although there were minor differences in relationships within each population between the analyses (Figure 3; Figure S7). In M. coibensis, individuals were primarily clustered by geographic location, and the three main groups had 100% bootstrap support (Figure 3a; Figure S7a).

| Population diversity
The AMOVA results indicated significant genetic differentiation between each pair of populations within each analyzed species (p < .01; Table S1). Observed heterozygosity was lower than expected heterozygosity for all populations, except for M. molossus in the Middle America-northern South America population (Table 1). Designating a threshold of Fst > 0.2 as high population genetic differentiation (Frankham, Bradshaw, & Brook, 2014) (Table 1). The estimated number of migrants was above one for all population pairs within species, except between M. molossus from southern South America and the Caribbean (0.699, Table 1).  (Table 2). In contrast, the population  Cayman and Cuba in comparison with the group from Jamaica and had reduced population sizes of 40% and 56%, respectively (Table 3).

| D ISCUSS I ON
We examined three species of the mastiff bat (Molossus) with distinct geographic and ecological distribution patterns and found that their population structures have a low admixture of haplotypes and vary with habitat preferences, level of population isolation, and historical fluctuations in climate, resulting in mainland and insular species having different phylogeographic patterns. Mainland geographic barriers such as the Andes did not correspond to lineage breaks, except for the Amazon River, which acts as a filter barrier for M. molossus and also correlates with differences between rainforest and savanna habitats.
Oceanic straits pose a partial barrier for some bats but not others, isolating populations of M. milleri between islands in the Greater Antilles, but with gene flow occurring among island populations of M. molossus within the Lesser Antilles. M. molossus populations from the mainland are distinct from those in the Lesser Antilles, indicating a degree of isolation. However, migrants from the continent found in the archipelago (such as Grenada) demonstrate ongoing gene flow between these two regions. For this group of bats, oceanic straits appear to act as a partial, filter barrier to dispersal. Our dataset also is consistent with the expectation that more isolated lineages on islands undergo genetic bottlenecks more frequently than lineages closer to the mainland. The riverine barrier hypothesis proposes that rivers act as a barrier to gene flow, promoting divergence between populations on opposite banks (Wallace, 1852). The Amazon River in particular, which originated during the Miocene and attained its present course during the Pliocene, is thought to have contributed to allopatric speciation and population differentiation in many taxa (Baker et al., 2014;Nazareno, Dick, & Lohmann, 2017). Indeed, Amazonian rivers seem to be acting as dispersal barriers to several taxa of volant (Hayes & Sewlal, 2004) and nonvolant animals (Ayres & Clutton-Brock, 1992;Bonvicino, Lindbergh, Faria, & Bezerra, 2012;Valdez & D'Elía, 2013).

| Mainland populations
Conversely, rivers might not always act as barriers to gene flow in bats or small mammals. Many groups of bats have been reported to use smaller rivers as landmarks for orientation and migration pathways (Furmankiewicz & Kucharska, 2009;Serra-Cobo, Lopez-Roig, Marqués-Bonet, & Lahuerta, 2000), and as preferred foraging habitats (Smith & Racey, 2008), suggesting that rivers might also act as dispersal corridors. In addition, the genetic patterns of some terrestrial small mammals do not corroborate the riverine hypothesis, with population substructure more pronounced along the length of the margins of the rivers, rather than between opposite sides (Patton & Da Silva, 1998;Patton, Silva, & Malcolm, 1994). In M. molossus, the Amazon River appears to act as a partial barrier to gene flow, despite the vagility of this bat, in accord with the riverine barrier hypothesis.
An additional explanation is that ecological factors also likely  (Almeida, Bonvicino, & Cordeiro-Estrela, 2007;Voss, 1991). Within each M. molossus lineage from the mainland, there is low intra-population genetic divergence, a lack of obvious phylogeographic structure, and high levels of gene flow. This pattern is perhaps not surprising in a species with high dispersal abilities (Burns & Broders, 2014;Norberg & Rayner, 1987;Speer et al., 2017;Taylor et al., 2012).

Molossus coibensis was structured as three populations from: (a) tropical forests in Panama; (b) tropical forests of northern South
America; and (c) savannas of Venezuela and Guyana. The Amazonian region of South America is composed primarily of rainforest, but patches of savanna are distributed from Venezuela to Suriname and southern Brazil (Haffer, 2002;Vanzolini & Williams, 1970).

Repeated cycles of expansion and fragmentation of savannas in the
Neotropics occurred most recently in the last 2 My, due to changes in temperature and sea level (Bennett, 1990), promoting opportunities for periodic connection and isolation between populations.
Previous studies based on mitochondrial and nuclear genes have reported that the population of M coibensis from the savannas formed a distinct clade relative to M. coibensis from forests of South and Middle America, suggesting that this population might belong to a putative new species (Lim & Engstrom, 2001;Lim & Lee, 2018;Loureiro et al., 2019). Although this population is genetically distinctive, it has a low degree of isolation with a high number of immigrants per generation (Table 1). Therefore, our data suggest that there is gene flow among these structured populations of M. coibensis, and the savanna population likely does not represent a distinct species. Koopman (1977) and Genoways (1998)  isolation is incomplete. Similarly, Muscarella, Murray, Ortt, Russell, and Fleming (2011), Carstens et al. (2004), and Larsen et al. (2012 found varying population structuring in different species of phyllostomid bats in the Caribbean, ranging from some species having monophyletic populations confined to individual islands, and other species lacking any genetic structuring among islands. Similar patterns have also been observed in other groups of volant animals, such as butterflies (Davies & Bermiham., 2001) and birds (Khimoun et al., 2016).

| Caribbean populations
Two individuals from off-shore islands (Bonaire and Grenada) had higher genetic similarity with the Middle America-northern South America group than to the Caribbean population. Bonaire is located approximately 80 km off the coast of Venezuela, and geologically considered a part of the continent, and Grenada is located about 160 km north of Trinidad, but geologically considered part of the Lesser Antilles. Due to the proximity of both islands to South America, they share similar fauna and flora with the mainland (Baker & Genoways, 1978), and these results are not unexpected, especially for Bonaire. However, some specimens from Grenada shared more haplotype similarity with other Lesser Antilles populations and clustered within the Caribbean group. These results indicate that there are two different haplotypes in Grenada, but because only one of our specimens from this island grouped within the Middle Americanorthern South America population, it suggests at least infrequent dispersal from the mainland (Pedersen et al., 1996). Speer et al. where we could examine potential population differentiation, no population structuring was found within the same island, except for Grenada, which suggests lower genetic variation and higher gene flow within as compared to among different islands or between island and mainland populations.

| Demographic histories
All mainland populations showed a constant or expanding effective population size through time, whereas the isolated Caribbean populations showed a higher probability of experiencing bottlenecks. These results support the hypothesis that island populations are more susceptible to bottlenecks than their continental relatives (Luikart, Sherwin, Steele, & Allendorf, 1998). Bottlenecks might cause heterozygosity deficiency in natural populations (Nei and Graur, 1984), decrease the genetic diversity of a population through random genetic drift (Groombridge, Jones, Bruford, & Nichols, 2000), reduce reproductive function (Madsen, Shine, Olsson, & Wittzell, 1999), and be involved in the speciation process (Mayr, 1963). Population persistence is highly connected to its evolutionary potential, which is enhanced by genetic variation (Frankel & Soulém, 1981;Newman & Pilson, 1997). Thus, level of genetic variance has direct implications for conservation management (Bouzat, 2010), especially in determining the minimum viable sizes of wild populations (Lande, 1988). These estimates may be important in a conservation context and can indicate if a population has the potential to undergo inbreeding depression or has the genetic breadth to adapt to future environmental changes (Sovic et al., 2016).
The recent divergence times associated with population expan-  (Pacifici et al., 2013), the effective population size decline in M. milleri started around 1.1 mya in the group from Cuba and the Cayman Islands, and about 719 ka in the group from Jamaica. These dates correspond to the early and beginning of the middle Pleistocene, respectively, which was characterized by several cycles of glacial and interglacial climates (Raymo, Ganley, Carter, Oppo, & McManus, 1998). These cycles likely caused fluctuations in temperature and sea level, resulting in the extinction of some Caribbean bat species during the late Pleistocene (Dávalos & Russell, 2012;Morgan, 2001). Therefore, we hypothesize that climate change in the early Pleistocene likely generated ecological changes and repetitive fluctuations in size of island landmasses, affecting population sizes of these bats. However, the confidence in- Populations that are isolated on islands might decrease drastically in size due to environmental changes and catastrophic events (Pedersen, 1998;Pedersen et al., 1996), and genetic drift in these small populations may result in the loss or decrease of some allele frequencies (Keller et al., 2001). Increased levels of immigration can lead to very different genetic outcomes from those expected in isolated populations. Island lineages with immigration sources can recover much faster from a bottleneck than isolated populations, despite increases in average inbreeding on islands (Keller et al., 2001). Buerlke and Gompert (2012) found that population allele frequencies were more variable for smaller samples of individuals (2-10), which could potentially affect results in both populations structuring and demographic analyses. However, Pluzhnikov and Donnelly (1996) and Beerli (2004) argue that the effect of small sample size on the estimates of population structuring and gene flow is minimum, except that the confidence intervals are somewhat larger with fewer individuals. Nadeau et al. (2011) proposed that a deeper genetic coverage could ameliorate the effect of a small number of individuals. In addition, other studies with next generation sequencing have conducted similar population genetic and demographic analyses using different population sizes (from 5 to 27 individuals) and found consistent results (Sovic, Fries, Martin, & Gibbs, 2018). Robison, Coffman, Hickerson, and Gutenkunst (2014) also found accurate parameter and demographic estimates for populations with more ancient demographic events (in the order of 0.5Ne generations ago) in small numbers of sampled individuals. In our data, when the models that included changes in population size had a higher likelihood, all the events (bottlenecks or expansions) occurred more than 0.5Ne generations ago, corroborating with Robison et al. (2014).
Our study of phylogeographic patterns in mainland and island populations in a group of highly mobile bats (Molossus) found that the Amazon River and ecological habitats (rainforest and savanna), but not the Andes Mountains, have an effect on the genetic structuring of M. molossus in South America. By contrast, oceanic barriers in the Greater Antilles play a role in isolation of some species (M. milleri), and that these isolated populations are more subject to bottlenecks and therefore vulnerable to environmental change. We expect these patterns to be even more pronounced in populations of bats with lower dispersal abilities, such as fruit and nectar feeding phyllostomids. Demographic research on the bat fauna as a whole would provide important information relevant to biological conservation in the Caribbean as climate change and environment vulnerability accelerate.

ACK N OWLED G M ENTS
This work was supported by Coordenação de Aperfeiçoamento de (UFRRJ), and J. A. Oliveira (MNRJ). We also thank Oliver Haddrath for providing constructive feedback on this manuscript.

CO N FLI C T O F I NTE R E S T
None declared.