Wings or winds: inferring bat migration in a stepping-stone archipelago

Authors


Manuel Ruedi, Department of Mammalogy and Ornithology, Natural History Museum of Geneva, Route de Malagnou 1, BP 6434, 1211 Geneva (6), Switzerland.
Tel.: 41 22 418 6320; fax: 41 22 418 6301; e-mail: manuel.ruedi@ville-ge.ch

Abstract

Eocene ocean currents and prevailing winds correlate with over-water dispersals of terrestrial mammals from Africa to Madagascar. Since the Early Miocene (about 23 Ma), these currents flowed in the reverse direction, from the Indian Ocean towards Africa. The Comoro Islands are equidistant between Africa and Madagascar and support an endemic land vertebrate fauna that shares recent ancestry predominantly with Madagascar. We examined whether gene flow in two Miniopterus bat species endemic to the Comoros and Madagascar correlates with the direction of current winds, using uni- and bi-parentally inherited markers with different evolutionary rates. Coalescence-based analyses of mitochondrial matrilines support a Pleistocene (approximately 180 000 years ago) colonization event from Madagascar west to the Comoros (distance: 300 km) in the predicted direction. However, nuclear microsatellites show that more recent gene flow is restricted to a few individuals flying against the wind, from Grande Comore to Anjouan (distance: 80 km).

Introduction

The natural colonization of remote islands by terrestrial animals is generally an extremely rare and stochastic event (Simpson, 1940). One example is the land fauna on the western Indian Ocean island of Madagascar. The origin of its extant, nonvolant terrestrial mammals is the result of four independent colonisations (Goodman et al., 2008). Each of these asynchronous, ancient events took place during the Palaeogene, more than 30 million years ago, and led to the diversification of all native carnivorans (Yoder et al., 2003), rodents (Jansa et al., 1999), lemurs (Yoder et al., 1996) and tenrecs (Olson & Goodman, 2003) living on the island today. These mammals, as well as other terrestrial vertebrates (Vences et al., 2003; Raselimanana et al., 2009), originated in Africa and reached Madagascar by over-water dispersal across the currently 415-km-wide Mozambique Channel, reputedly by swimming or rafting on mats of floating vegetation. However, today the prevailing sea currents and winds flow from east to west, contradicting this scenario (Rabinowitz & Woods, 2006). This discrepancy was recently resolved by Ali & Huber (2010), who showed with palaeo-oceanographic modelling that sea currents in the western Indian Ocean were inverted during extensive periods of the Early Tertiary, thus indeed favouring terrestrial animal dispersal from Africa to Madagascar. Westward-blowing currents, which prevail today, were most likely established since the Early Miocene, some 23 Ma (Ali & Huber, 2010).

The Comoro Archipelago lays equidistant (300 km) between mainland Africa and Madagascar and consists of four main islands: Grande Comore, Mohéli, Anjouan and Mayotte (Fig. 1a). These volcanic islands are relatively young (< 10–15 Myr, Nougier et al., 1986) and may have served as stepping-stones, for recent dispersal events (e.g. chameleons, Raxworthy et al., 2002). The current native mammalian land fauna of the Comoros includes only bats (Louette, 2004). Although these flying mammals are less dependent on sea currents to reach oceanic islands, a recent review revealed that nearly all Comorian species have closest ancestry with Malagasy rather than African congeners (Goodman et al., 2010b). This suggests that contemporary winds and cyclonic systems blowing from east to west, like sea currents, would influence regional inter-island dispersal and subsequent colonization in bats.

Figure 1.

 (a) Prevailing wind directions in the western Indian Ocean are marked with black arrows. Miniopterus griveaudi and M. aelleni bats used in this study were sampled at several sites (black squares) on Madagascar and the Comorian islands of Grande Comore and Anjouan, but are unknown from Mohéli and Mayotte (uncoloured). (b) Mitochondrial and (c) nuclear migration rates (m) of M. griveaudi estimated with an isolation-with-migration model and Bayesian inference, respectively. Island surface areas and inter-island distances are not represented to scale.

The two small species of Miniopterus bats found in the Comoros, M. griveaudi (Harrison, 1959) and M. aelleni Goodman, Maminirina, Weyeneth, Bradman, Christidis, Ruedi, Appleton, 2009, occur also on Madagascar and share no common recent ancestor with African taxa (Weyeneth et al., 2008). Several matrilines are co-distributed on Madagascar and the Comoros, but their precise geographical origin has not been determined. We used these bat species as models to examine whether colonization events and gene flow between islands in the western Indian Ocean correlate with the direction of prevailing winds (Fig. 1a). Both female-transmitted (mitochondrial DNA) and bi-parentally inherited (nuclear microsatellites) markers are used to determine indirectly if dispersal is sex-dependent (see e.g. Castella et al., 2001; Biollaz et al., 2010) and provide insight into migration patterns across different evolutionary timescales.

Materials and methods

Sampling

Microchiropteran bats belonging to two species, Miniopterus griveaudi and M. aelleni, were sampled during field surveys conducted in 2006–2008 from three populations on Grande Comore (mean number of individuals per population 17.3), from two populations on Anjouan (mean 26.5), and from three populations on Madagascar (mean 17.3) (Fig. 1a). The mitochondrial DNA (mtDNA) variation of 96 of those 157 samples has already been reported in Weyeneth et al. (2008). As not one of these populations differed significantly within islands (no significant Fst; see below), populations were grouped by species and island to obtain three population samples for M. griveaudi (Grande Comore: = 58; Anjouan: = 48; Madagascar: = 42), and two population samples for M. aelleni (Anjouan: = 5; Madagascar: = 10) (see Table S1, for details on sampling locations). Pectoral muscle samples were collected from analysed specimens and stored in a tissue buffer containing NaCl-saturated 25% DMSO and 250 mm EDTA.

DNA amplification and sequencing

Total genomic DNA was isolated from approximately 25-mg tissue sample using a DNeasy Blood and Tissue kit (Qiagen GmbH, Hilden, Germany), eluted and stored in a final volume of 200-μL AE Buffer (provided elution buffer). The second hyper-variable domain (HVII) of the mitochondrial D-loop region was newly sequenced for 55 M. griveaudi and six M. aelleni, as described in Weyeneth et al. (2008). These new sequences were aligned together with those obtained from GenBank (accession numbers: FJ232807FJ232873), which resulted in 157 sequences of 358 base pairs (bp) in length.

These 157 specimens were also genotyped at 10 autosomal microsatellite loci. Loci MS1, MS4 and MS5 were developed by Miller-Butterworth et al. (2002) for M. natalensis, and loci MM02, MM22, MM23, MM28, MM29, MM30 and MM31 were developed by Wood et al. (2011) for Australian M. cf. schreibersii. Three primer-pair combinations containing 2 μm of each primer were used for multiplex PCR amplification (see Table S2 for PCR and thermal profile). Fluorescent PCR products were analyzed on an ABI377 automated sequencer, using the Genescan software version 3.1 (Applied Biosystems, Rotkreuz, Switzerland) to score alleles.

Genetic diversity

For the D-loop data set, the number of polymorphic sites and haplotypes (nh), as well as the percentage of private haplotypes (nhP), haplotype diversity (h) and nucleotide diversity (π), was calculated for each species and island, using Arlequin version 3.5 (Excoffier et al., 2005). The parameters of microsatellite diversity included total number of alleles and observed heterozygosity per locus, as well as the number of alleles (A) and allelic richness (AR standardized with the rarefaction method; Petit et al., 1998) across all loci, the percentage of private alleles (AP) and the observed heterozygosity (HO) averaged between loci for each species and island using Arlequin and Fstat version 2.9.3 (Goudet, 2001). Null alleles, highly related individuals within a sampled population or genotypic disequilibrium, violate the model assumptions of many programs which implement population structure and/or migration rate analyses. Hence, to test whether each locus represents independent information, tests of genotypic disequilibrium were computed with Fstat. To test for null alleles and examine the relatedness of individuals, inbreeding coefficients (FIS) were computed within each population for each locus, and over all loci using Fstat. Corresponding probabilities for genotypic disequilibrium and FIS were inferred with a permutation procedure, and probability values were adjusted for multiple simultaneous tests using the sequential Bonferroni adjustment (Rice, 1989).

Network (mtDNA) and clustering (ncDNA) analyses

To compare mitochondrial and nuclear relationships of Miniopterus bats, we assigned D-loop haplotypes and multilocus genotypes to clades and clusters, respectively. We used TCS version 1.21 (Clement et al., 2000) to construct statistical parsimony D-loop networks with an 85% connection limit. Bayesian clustering analyses of multilocus genotypes were conducted with Structure version 2.3 (Pritchard et al., 2000). This program assigns individuals to K clusters based on their allelic frequencies and estimates the posterior probability of the data given each particular K. The admixture model of Structure was run for = 2 to = 7 for 15 repetitions of 106 iterations following a burn-in period of 105 iterations for each K. We determined the number of biologically meaningful clusters based on the ad hoc statistic delta K (Evanno et al., 2005) computed by averaging the posterior probabilities across the five best log likelihood score runs for each K. The outcome of the Structure analysis was graphically represented with the program Distruct version 1.1 (Rosenberg, 2004). The smaller data set for M. aelleni was not used for further population-level analyses.

Population structure

To assess the degree of genetic variation among island populations within M. griveaudi, Ф (D-loop) and F (microsatellites) statistics were calculated using a hierarchical analysis of molecular variance (amova) and pairwise population differentiation, as implemented in Arlequin. The Tamura-Nei model and gamma rate evolution (α = 0.8269) was adopted for the D-loop data set to correct for multiple hits. Corresponding probabilities for Ф and F-statistics were inferred with 10 000 permutations. Sequential Bonferroni correction for multiple tests (Rice, 1989) was applied to assess the probabilities of pairwise differentiation.

Ancestry and migration among populations

The first aim of this project was to address the directionality of the initial colonization of Miniopterus griveaudi between the three islands (Grande Comore, Anjouan and Madagascar) (Fig. 1a). To infer which population is ancestral, and hence the origin of M. griveaudi, we adopted the split parameter s (0 < < 1), which is the proportion of the ancestral population that founded descendent populations 1 (s) and 2 (1 – s), as implemented with the isolation-with-migration coalescence model (IM) (Nielsen & Wakeley, 2001) using the mtDNA data set. As IM only handles data from two populations, the parameter s was estimated in a pairwise manner between the three island populations. This coalescent approach to estimate directionality of mitochondrial gene flow was successfully used in a previous study on colonization patterns in Malagasy bats (Russell et al., 2008).

This investigation also addresses whether current gene flow, if any, is correlated with prevailing winds. The more recent, isolation-with-migration coalescence model implemented in IMa2 (Hey & Nielsen, 2007) estimates mutation scalar (u, neutral mutations per generation), unidirectional migration rates (M/u; M is the migration rate per generation per gene copy) and population splitting times (Tu; T is the time since common ancestry in generations) between several populations. We estimated the posterior probability of gene flow since populations of Grande Comore (G), Anjouan (A) and Madagascar (M) diverged using the mtDNA data set and the following population string: [((G, A): 1, M): 2] where numbers represent ancestral populations. To obtain demographic estimates of mutation-scaled parameters, a divergence rate of 20% per million years was used, as estimated for a homologous fragment of the D-loop in bats by Petit et al. (1999), resulting in = 7.16 × 10−5. For IM and IMa2 analyses, the generation time was set to 1 year, the inheritance scalar to 0.25, and the HKY model of evolution was selected. We used preliminary runs to determine the starting values of prior distributions for the subsequent runs. For each analysis, two final runs were conducted with different random seeds, setting prior values to = 2, = 100–200, Θ = 200–400, including the parameter s in IM. We adopted the geometric heating scheme using 20–40 simultaneous chains run for 105 steps after a ‘burn-in’ of 105 steps. Independent runs produced similar posterior distributions with effective sampling sizes of > 100 for all parameters, suggesting convergence to the stationary distributions. Parameter values were inferred from the maximum marginal posterior estimates (Nielsen & Wakeley, 2001; Hey & Nielsen, 2007).

Ideally, estimates of split parameter, migration rates and divergence times should have been obtained from a single analysis. However, the currently available versions of IM and IMa2 do not allow the consideration of more than two populations (IM), or the incorporation of s in the model (IMa2). In addition, the full isolation-with-migration model in IM and IMA2 used to infer the directionality of dispersal is computationally demanding, with run times exceeding several weeks to approximate the true values and obtain repeatable parameter estimates (e.g. McCormack et al., 2008). These coalescence-based analyses were only conducted with the mtDNA data set, as the high polymorphism of our microsatellite data would entail excessively intensive computational analyses (Hey & Nielsen, 2004).

To understand historical and recent inter-island movements and whether dispersal is sex-dependent, it is crucial to obtain independent migration rates for mtDNA and ncDNA. Hence, nuclear migration rates were assessed from the multilocus genotypes with the Bayesian prior assignment model in Structure (Pritchard et al., 2000), using = 4 groups, as described above. Following the approach suggested by Vonholdt et al. (2010), individuals with assignment probabilities higher than 80% (averaged over the five best runs) for discordant sampled locality were considered as recent migrants. These animals most probably represent first- or second-generation migrants, whereas animals with discordant assignment probabilities lower than 50% were considered as nonmigrant. Intermediate discordant assignment probabilities (50–80%) would be inconclusive (see Vonholdt et al., 2010). The rate of migration m was then calculated as the proportion of recent migrant versus nonmigrant individuals present in each population.

Results

mtDNA and ncDNA diversities

The D-loop data set for Miniopterus griveaudi consists of 142 sequences: 52 from Grande Comore, 48 from Anjouan and 42 for Madagascar, whereas the data set for M. aelleni comprises 15 sequences: five from Anjouan and 10 from Madagascar. The D-loop alignment (358 bp) for all 157 Miniopterus individuals identified 110 haplotypes based on 149 variable nucleotide positions (see Table S1 for GenBank accession numbers). No haplotype was shared between islands or between species (see Table S3 for details on geographic partition of haplotype diversity).

Microsatellite genotypes were obtained for these 157 specimens at all 10 loci (microsatellite genotypes are available on request), with the exception of one individual, for which alleles at loci MS1 and MS5 could not be interpreted. The total number of alleles per locus ranged from 5 to 20 (mean = 14.5), and observed heterozygosity from 0.108 to 0.834 (mean = 0.557). Private alleles were found at the species level (M. griveaudi: AP = 49%; M. aelleni: AP = 17%) and within most islands for M. griveaudi (Grande Comore: AP = 3%; Anjouan: AP = 0%; Madagascar: AP = 28%) and M. aelleni (Anjouan: AP = 10%; Madagascar: AP = 13%; Table S3). The observed heterozygosity per island and species averaged over all loci ranged from 0.516 to 0.730. The allelic richness per population (AR, standardized for five individuals per population) ranged from 3.3 to 5.4 (Table S3).

After Bonferroni correction (Rice, 1989), all loci were in linkage equilibrium. Inbreeding coefficients (FIS) for individual loci were not significantly different from zero, which suggests that null alleles are absent or rare. FIS were also low and nonsignificant at the population level, suggesting that bats sampled from the same population are not inbred.

Individual assignment and population structure

The relationships of Miniopterus griveaudi and M. aelleni were inferred with parsimony D-loop networks and Bayesian clustering analyses of microsatellite genotypes (Fig. 2). Consistent with previous analyses (Weyeneth et al., 2008), Clade 1 (subclades 1A–1C) haplotypes are specific to M. griveaudi, and Clade 2 haplotypes are specific to M. aelleni (Fig. 2). Within M. griveaudi, subclades 1A and 1B occur on all three islands, whereas 1C is restricted to Madagascar (Fig. 2). Bayesian clustering analyses based on the microsatellite data, conducted with the program Structure (Fig. S1), resulted in = 4 groups across the three islands, including one group of M. aelleni and three groups of M. griveaudi (Fig. 2). These groups do not correspond to any of the mtDNA clades inferred within Mgriveaudi (Fig. 2).

Figure 2.

 Mitochondrial (mtDNA) and nuclear (ncDNA) assignment of Miniopterus aelleni (Clade 2) and M. griveaudi (subclades 1A-1C). Samples were ordered by species, and within M. griveaudi by island and by mtDNA-subclade affiliation. Subclades 1A and 1B co-occur on all three islands (light-grey and white stripes), subclade 1C is Madagascar specific (white), and Clade 2 is M. aelleni specific (dark grey). Using ncDNA multilocus genotypes, bats assigned to four clusters, M. aelleni (dark grey) and the three island populations of M. griveaudi (light grey, white and black). The solid horizontal line marks the 50% threshold of individual assignment based on multilocus genotypes and highlights bats with higher probabilities to belong to a different cluster than assigned to.

The island structure within M. griveaudi was also tested with hierarchical amovas and pairwise population differentiation. For D-loop, 29.4% (< 0.001) of the total variance was explained among islands, whereas all pairwise ФST’s were highly significant with the highest differentiation found between Madagascar and Grande Comore (Madagascar – Grande Comore/Anjouan: ФST = 0.36–0.27, < 0.001; Grande Comore – Anjouan: ФST = 0.24, < 0.001). The same pattern was observed for the microsatellites with 9.4% (< 0.001) found among islands and maximum differentiation found between Madagascar and Anjouan or Grande Comore (Madagascar – Anjouan/Grande Comore: FST = 0.11–0.10, < 0.001; Grande Comore – Anjouan: FST = 0.07, < 0.001).

Origin and migration of Miniopterus griveaudi

IM analyses under the isolation-with-migration coalescence model support a Malagasy origin of extant M. griveaudi populations, as the highest proportion of the ancestral population is assigned to Madagascar (0.9955–0.9785) versus Grande Comore (0.0045) or Anjouan (0.0215) [range of 90% highest probability density (HPD): 0.0005–0.9995; Fig. S2]. Marginal posterior probability distributions of all parameters inferred with IM and IMa2 were well resolved showing one single peak, except for the parameter s (= probability of ancestry) between Grande Comore and Anjouan (Fig. S2). Assuming the [((G, A): 1, M): 2] population string in IMa2, the Malagasy and Comorian populations diverged approximately 183 000 years ago (95% HPD: 127 000–2 792 000 years ago), whereas populations on Grande Comore and Anjouan diverged about 149 000 years ago (95% HPD: 54 000–214 000 years ago).

IMa2 (mtDNA) and Structure (ncDNA) revealed no gene flow across the 300-km sea strait between Madagascar and the Comoros, but significant migration between Grande Comore and Anjouan (Fig. 1b,c). Female migration rates (as evidenced by mtDNA only) between Madagascar and the Comoros and from Anjouan to Grande Comore were all close to zero (mean: = 0.001; range of 95% HPD: 0.000–0.097), with a higher rate from Grande Comore to Anjouan (= 0.033, 95% HPD: 0.000–0.161). These very low rates of female migration are consistent with the absence of shared haplotypes between the two islands in the Comoros (Table S3). Nuclear genotypes, that reflect both male and female contributions to gene flow, analysed with Structure resulted in five M. griveaudi that had a higher posterior probability (> 80%) of belonging to an island other than that where they were sampled (Fig. 2). One bat was an adult male obtained on Grande Comore that was assigned (mean 82% ± 0.003 SD) to Anjouan. Four other adult bats from Anjouan show a high probability (88–97%) of belonging to the Grande Comore population: three were males and one a female, all carrying unique mtDNA haplotypes. They can therefore all be considered as first- or second-generation migrants. A further male from Anjouan had an intermediate posterior probability (68% ± 0.002 SD) of belonging to Grande Comore and was considered as an ambiguously assigned, nonmigrant bat. The fraction of recent migrants estimated from these multilocus genotype assignments suggest a minimum rate of 0.019 (one in 52 sampled individuals) from Anjouan to Grande Comore, and 0.083 (four in 48 sampled individuals) in the reverse direction (Fig. 1c).

Discussion

Mitochondrial DNA has been widely used in phylogeographic studies to address initial colonization histories (e.g. Russell et al., 2008) and divergence of populations or species (e.g. Castella et al., 2001), whereas the faster evolutionary rate of microsatellites make them excellent markers to identify recent gene flow among populations (Zhang & Hewitt, 2003; Biollaz et al., 2010). It is therefore crucial to infer dispersal patterns separately for mitochondrial and nuclear markers to understand how historical and current factors may have influenced population dynamics.

Colonization history: dispersal in the direction of the wind

Coalescence-based analyses, using a mitochondrial marker, inferred a Late Pleistocene (about 180 000 years ago) divergence of Malagasy and Comorian M. griveaudi populations, with the latter separating in Anjouan and Grande Comore soon thereafter (about 150 000 years ago). The most common recent ancestor of this taxon has a clear Malagasy origin (Fig. S2). Hence, the flight direction of M. griveaudi that dispersed over the 300-km open water expanse from Madagascar to the Comoros was in the same direction as the prevailing winds blowing westwards since the Miocene (Fig. 1a). Several mitochondrial matrilines co-occur on these islands, which suggest either incomplete lineage sorting since population divergence, or two separate colonisations from Madagascar to the Comoros represented by matrilines 1A and 1B (Fig. 2). In the latter case, we would expect two peaks in the marginal posterior probability distributions of divergence time if the Comoro Archipelago has been colonized twice. However, the distribution of the divergence parameter between Comorian and Malagasy populations was well resolved and showed one peak, suggesting a single successful colonization that was dated some 180 000 years ago. As this initial event post-dates the in situ emergence of Grande Comore, the youngest island in the archipelago (maximum age of 0.5 Myr, Nougier et al., 1986), the two Comorian islands were potentially colonized simultaneously by bats carrying either matriline 1A or 1B (Fig. 2), also found in Madagascar. Stepping-stone colonization from one Comorian island to the other cannot be excluded, as neither Grande Comore nor Anjouan was inferred as the ancestral population within the Comoros (Fig. S2).

Current inter-island gene flow

Over-water dispersal is not uncommon in bats (Russ et al., 2001; Russell et al., 2008; Garcia-Mudarra et al., 2009), and the colonization of remote islands can be favoured by prevailing winds, storms or cyclones (Cowie & Holland, 2006). For example, tropical storms in the Lesser Antilles move primarily from southeast to northwest and correlate with the direction of gene flow (Carstens et al., 2004) or to the South American origin of Lesser Antillean bat species (Genoways et al., 2007). However, shorter distances to mainland source areas also influenced movements in the Caribbean bat fauna (Hedges, 1996), suggesting that colonization events do not strictly depend on the direction of prevailing winds.

No mitochondrial or nuclear gene flow was evident across the 300-km-wide sea gap between Madagascar and the Comoros after the initial colonization by M. griveaudi (Fig. 1b,c). Further, mitochondrial and nuclear gene flow is only significant from Grande Comore to Anjouan, which is against prevailing winds (Fig. 1). Dispersal between Madagascar and the Comoros might still have occurred, but without successful colonization (i.e. no reproduction). Matrilines 1A, 1B and, for Madagascar, 1C co-occur on the three islands, but no haplotype is shared between them, suggesting that current female gene flow is very low or absent, and significant estimates of migration from Grande Comore to Anjouan must be because of more ancient movements between these islands. Relationships based on multilocus genotypes suggest that enough time has passed for these nuclear markers to segregate according to island, with only five individuals showing clear discordant ancestry (Fig. 2). The majority of these bats were sampled in the Comoros and showed mixed ancestry between Grande Comore and Anjouan, supporting occasional movements across the 80 km of open sea separating these two islands. In contrast to M. griveaudi, with an average body mass of 5.4 g, Comorian populations of larger bodied bats, such as Mops leucostigma G.M. Allen, 1918 (21 g) or Rousettus obliviosus Kock, 1978 (56 g), are weakly structured within the Comoros (Ratrimomanarivo et al., 2008; Goodman et al., 2010a), suggesting regular inter-island gene flow across sea distances of 40–80 km, which in turn may be related to their increased flight capacity associated with body size.

Contrasting patterns can be expected when female and bi-parentally inherited markers are used in population genetic analyses. Male-biased dispersal and female philopatry are common in mammals (Greenwood, 1980), resulting in stronger population structure being inferred with mitochondrial than nuclear DNA. Significant structure for both marker types is rare. Accounting for the four-fold difference in effective population size for uni-parental haploid versus bi-parental diploid markers, M. griveaudi populations are similarly and strongly structured for both mitochondrial D-loop and microsatellite data, suggesting that dispersal is not particularly sex-biased. Additional analytical methods and larger sample sizes are needed to address this question more precisely (see e.g. Whitlock & McCauley, 1999; Goudet et al., 2002).

Population histories of continental South African Miniopterus natalensis (A. Smith, 1834) and continental Portuguese Miniopterus schreibersii (Kuhl, 1817) are comparable to the pattern observed in insular M. griveaudi, with significant population structures inferred with mitochondrial and nuclear markers (Miller-Butterworth et al., 2003; Pereira et al., 2009). Ecological factors and inter-colony distances lead to population divergence in these continental species. In contrast, the mechanism of diversification that shaped the current inter-island population structure in M. griveaudi is open water acting as a permeable barrier to gene flow.

Conclusion

The biogeographic pattern inferred for M. griveaudi based on gene flow data suggests that its colonization of the volcanic Comoros was a unique event and perhaps favoured by the prevailing westward winds or storms (Fig. 1a). Wind direction has less impact on post-colonization gene flow, but flight distance and availability of day roost sites, in this case lava tubes, might have influenced movements in these bats. Indeed, Miniopterus bats are unknown on the Comorian islands of Mohéli and Mayotte, where lava tubes have filled in or collapsed (Goodman et al., 2010b) and patterns of historical geology may act as the limiting factor to differential extinction and contemporary distribution patterns. Miniopterus griveaudi and M. aelleni do not occur on mainland Africa, which has a suite of similar sized members of this genus (Monadjem et al., 2010), and in this case, prevailing winds would have favoured colonization of Africa from the Comoros. The Comorian land vertebrate fauna has numerous endemics, a large percentage of which share recent ancestry with Malagasy rather than African taxa (e.g. chameleons Raxworthy et al., 2002; birds Louette, 2008; bats Goodman et al., 2010b). This general pattern suggests that the direction of initial colonization in the western Indian Ocean indeed correlates with ocean currents and prevailing winds. Contrary to our findings, Russell et al. (2008) showed that bats belonging to the genus Triaenops dispersed on at least two occasions across the Mozambique Channel, from Africa to Madagascar, during the Late Pleistocene, i.e. against current prevailing winds. However, these bats are known to colonize remote islands (Goodman & Ranivo, 2008) and are therefore markedly less philopatric than Miniopterus. In addition to species-specific dispersal capacities, other variables such as wing shape and size or foraging strategies may also account for the propensity of bats and other volant animals to cross water channels, whether or not against prevailing winds.

The mammal fauna of the Comoros is depauperate in comparison with that of Madagascar (Goodman et al., 2008), with bats being the only group that has reached the archipelago via natural means of dispersal (Louette, 2004; Goodman et al., 2010b). This suggests that mammalian colonization of the Comoros is a rare event. Once bats colonized these islands successfully, populations remained isolated, leading to population divergence as observed for M. griveaudi or even speciation (e.g. R. obliviosus, Goodman et al., 2010a).

Acknowledgments

We thank the Direction des Eaux et Forêts and Association National pour la Gestion des Aires Protégées (Madagascar), and the Centre National de Documentation et de Recherche Scientifique (Union des Comores) for providing authorizations for the capture, collection and exportation of specimens. We acknowledge Yahaya Ibrahim and Ishaka Saïd for their aid with fieldwork and Lawrence R. Heaney, John D. Phelps and William T. Stanley of the Field Museum of Natural History for facilitating access to tissue samples. We thank the Unité de Phylogénie et Génétique moléculaires des Conservatoire et Jardin botaniques of the City of Geneva for access to laboratory facilities to analyse microsatellite markers. This project was supported by financial aid from the Basler Stiftung für biologische Forschung, an Augustin Lombard grant from the Société de Physique et d’Histoire Naturelle de Genève and from the city of Geneva to NW, and funding from CABS of Conservation International, John D. and Catherine T. MacArthur Foundation and Volkswagen Foundation to SMG.

Data deposited at Dryad: doi: 10.5061/dryad.8797

Ancillary