Integrating population genetics to define conservation units from the core to the edge of Rhinolophus ferrumequinum western range

Abstract The greater horseshoe bat (Rhinolophus ferrumequinum) is among the most widespread bat species in Europe but it has experienced severe declines, especially in Northern Europe. This species is listed Near Threatened in the European IUCN Red List of Threatened Animals, and it is considered to be highly sensitive to human activities and particularly to habitat fragmentation. Therefore, understanding the population boundaries and demographic history of populations of this species is of primary importance to assess relevant conservation strategies. In this study, we used 17 microsatellite markers to assess the genetic diversity, the genetic structure, and the demographic history of R. ferrumequinum colonies in the western part of its distribution. We identified one large population showing high levels of genetic diversity and large population size. Lower estimates were found in England and northern France. Analyses of clustering and isolation by distance suggested that the Channel and the Mediterranean seas could impede R. ferrumequinum gene flow. These results provide important information to improve the delineation of R. ferrumequinum management units. We suggest that a large management unit corresponding to the population ranging from Spanish Basque Country to northern France must be considered. Particular attention should be given to mating territories as they seem to play a key role in maintaining high levels of genetic mixing between colonies. Smaller management units corresponding to English and northern France colonies must also be implemented. These insular or peripheral colonies could be at higher risk of extinction in the near future.


| INTRODUC TI ON
Biodiversity is dramatically declining at an accelerating rate for most animal groups Hoffmann et al., 2010;Sánchez-Bayo & Wyckhuys, 2019). According to the IUCN Red List, more than 26,500 species (27% of all assessed species) are threatened with extinction. This phenomenon results from a combination of ecological factors (e.g., habitat fragmentation and destruction, pollution, introduction of invasive species, and climate change) that affect population sizes and connectivity. As a consequence, these populations become strongly exposed to the negative impacts of inbreeding and genetic drift (Frankham, 2005). Preserving the genetic diversity of such small and isolated populations is therefore essential to avoid inbreeding depression, to maintain genetic variability that may be useful for adaptation, in particular in response to environmental changes, and ultimately, to promote population persistence (Reed & Frankam, 2003). To address this issue and before drawing efficient conservation programs, an important prerequisite is to gather knowledge on population boundaries and demography. In recent decades, population genetics has been combined with more classical ecological studies to infer population demographic features, including the detection of recent demographic declines or the quantification of connectivity between populations (e.g., Vignaud et al., 2014;Vonhof & Russell, 2015). Ultimately, these population genetics studies may help delineating functional and evolutionary conservation units such as "Management Units," which are appropriate for species monitoring and management. There are several ways to define management units, among which are the assessment of panmixia (Moritz, 1994) or the estimation of population genetic divergence (Palsbøll, Bérubé, & Allendorf, 2007). Designing appropriate management units is thus far from being trivial. There is no general framework for determining at which dispersal rate populations become demographically correlated (therefore requiring a single management unit), and there is no easy way to translate gene flow estimates provided by population genetics into dispersal rates. Yet, too large units may increase the risk of extinction of "cryptic" populations that would require specific strategies. On the other hand, splitting a large population into different conservation units with different strategies may lead to excessive management strategies beyond requirements, or to inappropriate strategies limiting connectivity.
The application of population genetics to such issues of conservation biology has been especially important for endangered species that are difficult to monitor with ecological methods, such as bats.
Indeed, bats are very sensitive to climate change and human activities (Jones, Jacobs, Kunz, Willig, & Racey, 2009;Voigt & Kingston, 2016). Almost one-quarter of bat species in the world are considered to be Threatened and another quarter as Near Threatened (Mickleburgh, Hutson, & Racey, 2002). Temperate-zone bats are nocturnal, small, highly mobile, and the location of their roosts are often poorly known, characteristics that make monitoring and assessment of their extinction risk difficult (O'Shea, Bogan, & Ellison, 2003). Conservation programs are often established considering local and national scales. Unfortunately, these scales are usually not defined on the basis of biological knowledge on population delineation and demography, but instead conform to administrative borders that rarely correspond to natural ecological boundaries. This is likely to limit the efficiency and coherence of conservation strategies. Population genetics might therefore improve the definition of appropriate management units of bat populations (e.g., Dool, O'Donnell, Monks, Puechmaille, & Kerth, 2016;Ibouroi et al., 2018).
Besides, demographic inferences based on population genetics may be particularly relevant to highlight the need of conservation management for bat species. As such, Durrant, Beebee, Greenaway, and Hill (2009) have been able to evidence a recent decline and high levels of inbreeding in British populations of Bechstein's bat (Myotis bechsteinii).
Among European bat species, the greater horseshoe bat (Rhinolophus ferrumequinum) is particularly relevant to address conservation issues from population genetics. This insectivorous species-which seasonally uses hibernation and maternity roosts-has experienced dramatic declines, particularly in Northern Europe (e.g., Belgium, Luxembourg, England) where it is now considered rare or extinct (Kervyn, Lamotte, Nyssen, & Verschuren, 2009;Pir, 2009). In some countries, such as the UK, there is evidence of recent population increases . The species is included in Appendix II of Bern Convention, Appendix II of the Bonn Convention, Annex II and Annex IV of the European Directive on the conservation of Natural Habitat and of Wild Fauna and Flora, and is listed in the IUCN Red List of Threatened Animals (International Union for the Conservation of Nature, 2017). The reasons for the disappearance of the populations of R. ferrumequinum are difficult to identify but it is likely that anthropogenic factors (e.g., intensification of agriculture, urbanization, and loss of roosts) are responsible for it (Froidevaux, Boughey, Barlow, & Jones, 2017;Mathews et al., 2018). It is therefore important to implement conservation programs, at adequate geographical scales and based on a solid knowledge of R. ferrumequinum population dynamics.
Previous phylogeographic studies of R. ferrumequinum have revealed a unique genetic cluster in Western Europe mainland, ranging from Portugal to Italy (Switzerland apart), that resulted from the expansion of a single population originating from a Western Asian refugium (Flanders et al., 2009;Rossiter, Benda, Dietz, Zhang, & Jones, 2007). However, these studies are based on sparse sampling, especially in mainland Europe (e.g., one location in France). Yet, where Regional Development Fund; DREAL Nouvelle-Aquitaine K E Y W O R D S Chiroptera, connectivity, conservation, demographic inference, microsatellites, population genetics populations have been intensively sampled, a strong genetic differentiation was observed at smaller spatial scales (over tens to several hundreds of kilometers), for instance, within the United Kingdom (Rossiter, Jones, Ransome, & Barrattt, 2000). These patterns underline the importance of sampling density in population genetics studies to detect finer genetic clustering and particular population functioning (e.g., source-sink dynamics), despite an apparent lack of genetic differentiation detected over thousands of kilometers. More specifically, only one French location of R. ferrumequinum had been included in these previous phylogeographic studies (Flanders et al., 2009;Rossiter et al., 2007Rossiter et al., , 2000. However, R. ferrumequinum distribution in France is very disparate, and its status can be very contrasted between regions (Vincent & Bat Group SFEPM, 2014). These patterns are suggestive of differences in population size, connectivity levels, and therefore extinction risk. Most of the known roosts of R. ferrumequinum in France are located on the Atlantic coast (Vincent & Bat Group SFEPM, 2014 Eckert, Samis, & Lougheed, 2008). In addition, we examined patterns of genetic differentiation to evaluate the connectivity between R. ferrumequinum colonies. We expected a disruption of gene flow between colonies located on either side of the Mediterranean Sea or Channel Sea, as it has already been shown that sea is a barrier to dispersal for several bat species (Castella et al., 2000;García-Mudarra, Ibáñez, & Juste, 2009 Moritz (1994) and Palsbøll et al. (2007).
Both academic and nonacademic partners were involved in this work to guarantee that the results would directly inform conservation and management action (Britt, Haworth, Johnson, Martchenko, & Shafer, 2018 (Ransome & Hutson, 2000). This sampling leads to unbalanced sex and age ratios, with 887 females, 62 males, and one undetermined, and 808 adults, 130 juveniles (less than two years old), and 12 undetermined, respectively. Distances between sampling colonies varied from 2.53 km (closest colonies from western France) to 1,830 km (colonies from England and Tunisia).

| Biological material
For each bat, a tissue sample was collected from the wing membrane (patagium) using a 3-mm diameter biopsy punch. Samples were either preserved in 95° ethanol and quickly stored at 4°C until DNA extraction or preserved in silica-gel (Puechmaille et al., 2011).
Biopsy punches were cleaned with bleach, water, and then ethanol between each sample collection.

| DNA extraction and microsatellite genotyping
DNA was extracted from each wing sample using the EZ-10 Spin Column Genomic DNA Minipreps Kit for Animal (BioBasic) following the manufacturer's instruction with a final elution of twice 50 µl in elution buffer. We amplified 17 microsatellite loci using primers modified from those previously designed for R. ferrumequinum (Dawson, Rossiter, Jones, & Faulkes, 2004;Rossiter, Burland, Jones, & Barratt, 1999) and the lesser horseshoe bat (Rhinolophus

| Genetic diversity and relatedness within colonies
We assessed genetic diversity within colony by estimating the allelic richness and private allele richness corrected for minimal sample size (A r and pA, N = 13), the expected (H e ) heterozygosity, using FSTAT v.2.9.3.2 (Goudet, 1995), HP-RARE (Kalinowski, 2005) We estimated the fixation index F IS (Weir & Cockerham, 1984) using GENEPOP v4.6. We next performed relatedness analyses to verify whether population genetic structure could not be due to the comparison of different family units rather than populations (Schweizer, Excoffier, & Heckel, 2007). We estimated the maximum likelihood pairwise coefficient of relatedness r between pairs of individuals on an absolute scale (0, unrelated to; 1, identical individuals), using the ML-RELATE software (Kalinowski, Wagner, & Taper, 2006). Indeed maximum likelihood estimate is less biased than commonly used estimators (Milligan, 2003). The coefficient r corresponds to the probability for each locus that individuals share zero, one, or two alleles that are identical by descent. We used the genetic clusters identified from further clustering analyses (see below) as population references for estimating r within each colony.

| Population structure and genetic differentiation between colonies
The genetic differentiation between colonies was quantified using estimates of global and pairwise F ST (Weir & Cockerham, 1984).
Significance was assessed using exact G-test of differentiation implemented in GENEPOP v4.6 (Rousset, 2008). Pairwise F ST and exact G-tests were computed for each pair of colonies and each pair of genetic clusters identified from further clustering analyses.
Analyses were performed with and without juveniles. We accounted for multiple testing using false discovery rate (FDR). In order to control for potential effects of null alleles on genetic differentiation, we also estimated pairwise F ST corrected for null alleles using the "Excluding Null Alleles" (ENA) correction implemented in FreeNA (Chapuis & Estoup, 2007).
Genetic structure was also investigated using several complementary approaches to give a robust cross-validation of our results and prove that the observed genetic signature is robust despite the potential violation of the underlying hypothesis. First, we used the clustering approach implemented in the STRUCTURE program v2.3.4 (Pritchard, Stephens, & Donnelly, 2000) to determine the presence of genetic discontinuities without any a priori knowledge.
We determined the most likely number of genetic clusters using the log-likelihood of K and ∆K statistic (Evanno, Regnaut, & Goudet, 2005) implemented in the website STRUCTURE HARVESTER (Earl & vonHoldt, 2012). We used the admixture model with uncorrelated frequencies and an alpha-value of 1/K, as recommended by Wang (2017) in the case of unbalanced sampling. The same results were obtained with the default alpha-value and when applying or not the LOCPRIOR model to the population model (Hubisz, Falush, Stephens, & Pritchard, 2009) and were therefore not presented here.
Puechmaille (2016) demonstrated that the ∆K statistic was biased in the case of uneven sampling (as in our current study). However, in the present study, given the fact that three recovered clusters (see Section 3) were geographically coherent, were composed of colonies with nonsignificant or very low F ST values (see Section 3), and were consistent with previous genetic findings, we did not perform further subsampling or used alternative estimators. We performed 20 independent runs with a burn-in period of 1,000,000 iterations and 50,000 MCMC repetitions after burn-in, testing K = 1 to K = 28 (N colonies + 1). We used the R package pophelper (Francis, 2017) to compute the plots. In addition, we also performed a principal component analysis (PCA) implemented in the R packages ade4 (Dray & Dufour, 2007) and factoextra (Kassambara & Mundt, 2017). Contrary to STRUCTURE algorithm, this approach does not rely on any specific population genetic assumption including Hardy-Weinberg equilibrium and linkage equilibrium (Pritchard et al., 2000).
We next used MAPI program (Mapping Averaged Pairwise Information, Piry et al., 2016), implemented in the R package mapi, to detect spatial genetic discontinuity. This approach has low sensitivity to potential confounding effects resulting from isolation by distance (IBD) and does not require predefined population genetic model. It is based on a spatial network in which pairwise genetic distance between georeferenced samples is attributed to ellipses. A grid of hexagonal cells covers the study area and each cell receives the weighted arithmetic mean of the pairwise genetic distance associated to the ellipses intersecting the cell (Piry et al., 2016). We used the Rousset's coefficient â (Rousset, 2000) computed with SPAGeDi 1.4 (Hardy & Vekemans, 2002) as an index of pairwise genetic differentiation between individuals and default parameter value for the eccentricity of the ellipses (0.975). We used the permutation procedure (1,000 permutations) to identify areas exhibiting significantly higher or lower levels of genetic differentiation than expected by chance.
Lastly, isolation by distance (IBD) was analyzed with the regression of the genetic distances between colonies (F ST /1−F ST ; Rousset, 1997)

| Inference of demographic parameters
We first inferred the demographic history of R. ferrumequinum colonies. We used the software MIGRAINE v.0.5.1   We then inferred contemporary levels and directions of migration between the main genetic clusters using the program BAYESASS v3.0.4 (Wilson & Rannala, 2003). We performed five independent runs of 10,000,000 iterations sampled every 2000 iterations, with a burn-in of 1,000,000. For each run, we calculated the Bayesian deviance using the R script provided by Meirmans (2014). We used this deviance as a criterion to find the run that provided the best fit and to identify runs with convergence problems (Faubet, Waples, & Gaggiotti, 2007;Meirmans, 2014

| RE SULTS
Three of the 17 microsatellites genotyped were excluded from further genetic analyses: one of them was monomorphic and the other two had poor quality profiles. Results gathered using MICROCHECKER showed no large allele dropout or scoring inconsistencies due to stuttering. Null alleles were suspected at loci Rferr06 in three colonies from France ("AIR," freq = 0.074; "FEN," freq = 0.046; "XAI," freq = 0.041), and at loci RHD103 (France, "ARL," freq = 0.078), Rferr27 (France, "BED," freq = 0.152), and Rferr01 (England, "BUC," freq = 0.149). Nevertheless, we did not detect any F I G U R E 2 Expected heterozygosity (H e ) estimated for each colony. Blue to red colors indicate low to high levels of H e . The geographic range of R. ferrumequinum is shaded in gray F I G U R E 3 Distribution of the pairwise relatedness coefficient r estimated with the ML-relate software (Kalinowski et al., 2006) Figure 2).

| Genetic diversity and relatedness
The northern French colony "MON" showed lower allelic richness estimate than the other French colonies (adjusted p = .030 < p critical = .037) but similar H e estimate (adjusted p = .059 > p critical = .037).
The Tunisian colony "GHA" showed similar levels of allelic richness and expected heterozygosity than the French colonies (adjusted p > p critical ). Private allelic richness was low in all colonies except the Tunisian one ("GHA"; Table 1).
The distributions of the pairwise coefficient of relatedness r within each colony showed a common pattern in all colonies except in the northern French colony "MON" (Figure 3). For all colonies, the distribution of the r coefficient was L-shaped with a peak of unrelated individuals (r = 0) and a decreasing proportion of related individuals. In the northern French colony "MON," we observed a more uniform proportion of unrelated and relatively closely related individuals (r ranging between .0 and .3). When considering pairwise relatedness between individuals from different colonies, we still observed high levels of relatedness (r > .5; Figure 4). The high levels of relatedness involved female-female pairs and female-male pairs but never male-male pairs. It concerned 537 females, 33 males, and one undetermined individual. The majority of the females were adults (474 adults, 59 juveniles, and four undetermined), but this was not the case when considering males (12 adults, 19 juveniles, and two undetermined).
F I G U R E 4 Geographic distribution of the higher values of pairwise relatedness coefficient (r > .5) between R. ferrumequinum colonies from the continental genetic cluster (France and the Spanish Basque Country)

| Population structure and genetic differentiation between colonies
F ST estimates calculated with and without excluding null alleles (ENA) were very similar, and analyses without juveniles did not substantially change F ST estimates. Therefore, we only reported the uncorrected estimates for the dataset including juveniles. The pairwise F ST estimates between colonies and associated G-tests are presented in Table S2. Low but significant genetic differentiation was observed between colonies within western France (F ST < 3%, 65.24% of the G-tests with p < .05; Table S2). The northern French colony "MON" exhibited higher estimates of pairwise F ST than the other colonies Using STRUCTURE, we found that the ∆K statistics (Evanno et al., 2005) were highest for K = 2, but the likelihood of the number of genetic clusters Ln (L(K)) showed similar values for K = 2 and K = 3 ( Figure S1). For K = 2, the first genetic cluster included all colonies from England, France, and Spanish Basque Country and the sec-

| Inference of demographic parameters
Using MIGRAINE, we detected a significant signature of expansion  (Table S3). The marginally significant signature of contraction observed for the western French colony "CHE" in the Poitou-Charentes region was also considered to be nonsignificant because the higher value of the confidence interval was very close to 1 (0.99) and because we performed a high number of tests.
We estimated the scaled current population size θ (4N e µ) for all stable colonies, and θ estimates ranged between 1.854 and 6.465 (Table S4, Figure 8). The lowest estimates were found for the English colonies ("BRY" θ = 1.854; "BUC" θ = 2.107) and for the northern French colony "MON" (θ = 3.030). Other colonies and pool of colonies exhibited similar levels of θ estimates (from 4.066 to 6.465). Kruskal-Wallis tests and post-hoc pairwise comparisons of θ estimates using Wilcoxon test (with FDR correction for multiple testing) revealed a significant difference of θ estimates between the western French-Spanish Basque and the English genetic clusters (p = .002) and nonsignificant differences between the other clusters (p > .115). We observed a significant negative relationship between the estimated θ of each colony and the distance of the colony to the centroid of our sampling (p < .05; Figure   S3).
The five runs implemented in BAYESASS provided similar results and converged well, with values ranging from D run2 = 67,128.6 to D run4 = 67,134.7. None of the migration rates estimated between the three main genetic clusters was significant (

| A large and stable population of R. ferrumequinum ranging from Spanish Basque Country to northern France
Our study revealed high and homogeneous levels of genetic diversity within the western French and Spanish Basque colonies examined.
These levels were similar to those previously detected in colonies from the eastern part of R. ferrumequinum distribution (Rossiter et al., 2007(Rossiter et al., , 2000, or in other bat species such as Rhinolophus euryale and Myotis myotis (Budinski et al., 2019;Castella, Ruedi, & Excoffier, 2001 Flanders et al., 2009;Rossiter et al., 2007). In addition to these findings, we could infer the scaled effective size θ (4N e µ) of this population. We found homogeneous θ among the western French and Spanish Basque colonies and for the pool of these colonies. This result was congruent with our finding of one large population with each colony representing a replicate of the whole population. We could not transform this estimate θ into an effective number of individuals (N e ) because it requires fixing the mutation rate of microsatellite markers in this species. This is a crucial issue because estimates of mutation rates can greatly vary between and within species and any inaccuracy in the mutation rate estimate is propagated in the estimation of N e (Waples, 2010).
We did not reveal any signature of demographic decline, neither for the colonies of the French Poitou-Charentes region nor TA B L E 2 Isolation by distance characteristics using the genetic differentiation parameter (F ST /1−F ST ) between colonies against the logarithm of the Euclidian geographical distance  for the delineated western French and Spanish Basque population of R. ferrumequinum. Several biases could have lead to the nondetection of a demographic event. First, Leblois et al. (2014) showed that the capacity to detect demographic events from genetic data depends on the number of genetic markers used, the strength of the event, and the time when it happened. Therefore, because the decline was recent and we used 14 microsatellites, our study may have suffered from a lack of power that could explain the absence of signature of demographic decline in our data. Second, the maximum lifespan of these bats is 30 years (Caubère, Gaucher, & Julien, 1984), and simulations of microsatellite data analyzed under similar conditions using MsVar (Beaumont, 1999;Storz & Beaumont, 2002) on the eastern red bat Lasiurus borealis have shown a significant delay in the response of coalescent-based N e estimates to recent population declines (Munster, 2015). However, our re-

| Dispersal and reproduction of R. ferrumequinum
Our results revealed that the Mediterranean Sea and the English LGM; Flanders et al., 2009;Rossiter et al., 2007). Our demographic inferences also revealed the absence of current gene flow between the two English and western French-Spanish Basque clusters. The high genetic differentiation observed between Tunisian and western French-Spanish Basque colonies may also rely on historical colonization history. R. ferrumequinum seems to have been present in North Africa before the Late Glacial Maximum (Flanders et al., 2009;Rossiter et al., 2007), and there is no evidence that North Africa was recolonized from Europe post-LGM. These historical patterns of  Similarly, as we only had a few samples from both sides of the Pyrenees, we could not assess the potential effect of this mountain as a barrier to gene flow. The absence of (or weak) gene flow disruption associated with western Pyrenees in this study may therefore suggest that this mountain may not impede dispersal or that R. ferrumequinum uses the shoreline as corridor along both the Atlantic and Mediterranean edges of the Pyrenees, where altitude is lower than 500 m, as has been seen for migratory birds (Galarza & Tellería, 2003). The situation is different from another Rhinolophus species, R. hipposideros, where the Pyrenees most likely act as a strong barrier (Dool et al., 2013). Future genetic and ecological studies including dense sampling from both sides and all along the Pyrenees are now required to assess whether R. ferrumequinum movements and genetic mixing are restricted by mountains.
In the absence of important landscape barriers such as seas or, potentially, mountains, we showed that R. ferrumequinum is able to move over hundreds of kilometers, as exemplified by the low levels of genetic differentiation observed at large geographical scales, the inference of significant migration rates between the western French-Spanish Basque maternity colonies, and the high levels of relatedness observed between individuals sampled in distant colonies.
These results also revealed the high genetic mixing that occurs at large scale between R. ferrumequinum western French and Spanish Basque colonies. Several demographic processes may underlie this genetic mixing. First, mating dispersal at large distance would lead to extracolony copulations and to the relaxation of colonies' genetic borders (Veith, Beer, Kiefer, Johannesen, & Seitz, 2004). Second, because we also found high levels of relatedness between juveniles (under two years old) sampled at considerable distances (up to 861 km), natal dispersal (one-way movement of juveniles during their first year, from their colony of birth to another) and/or movements of adults from one maternity colony to another between two consecutive reproduction events (years) are also potential mechanisms shaping genetic mixing. These alternatives are still difficult to evaluate due to a lack of knowledge with regard to R. ferrumequinum mating behavior (where and when) and dispersal, in particular the one of males. Ringing data from R. ferrumequinum from across Europe suggest the species is mostly sedentary but with occasional movements over 100 km. Indeed, although rare, there are documented movements of 180 km in Spain, 320 in Hungary (reviewed in Hutterer, Ivanova, Meyers-Cords, & Rodrigues, 2005), and 500 km in France (Saint Girons, 1973), clearly demonstrating the species is occasionally able to move over large distances. In the future, long-term capture-mark-recapture surveys of adults and juveniles could provide invaluable information to assess the relative importance of mating, natal, and breeding dispersal in the genetic mixing of colonies within management units. Given the scale at which the species is suspected to move given the available ringing data and our current genetic results (weak population structure), it would be important to monitor sites (for recaptures) not only close to the ringing sites but also several hundreds of kilometers away.
This will make these studies, which would ideally be long-term studies, logistically challenging.

| Differences in the functioning of centralperipheral and island-continental colonies
Our results revealed contrasting patterns of genetic structure within and between populations when comparing R. ferrumequinum western French-Spanish Basque population with colonies from Tunisia, England, and northern France ("MON"). In these latter colonies, we detected lower levels of genetic diversity (H e up to 20% lower) and smaller estimates of ϴ (4N e µ; up to half the size) than in the western French-Spanish Basque population. They were also more genetically differentiated than the western French-Spanish Basque ones, as revealed by F ST estimates and isolation by distance and clustering analyses. Some of these particular colonies might be insular (England), but are commonly located near the edge of R. ferrumequinum distribution range. Contrasting levels of genetic diversity between insular and continental populations are common in animals (Frankham, , 1996 and have already been observed between England and the continent in several bat species such as R. ferrumequinum (Rossiter et al., 2000), Myotis bechsteinii (Wright et al., 2018), R. hipposideros (Dool et al., 2013), Plecotus austriacus (Razgour et al., 2013), and Eptesicus serotinus (Moussy et al., 2015).
The contemporary isolation of colonies from UK with the continent might have maintained higher levels of genetic drift, as shown by the low N e estimates, which reinforces their vulnerability (Newman & Pilson, 1997). These colonies may face stochastic reduction of genetic diversity that could limit their evolutionary potential, in particular in the face of environmental changes.
More surprisingly, the northern French colony at Montreuil-sur-

| Implications for conservation
In this study, we have shown that connectivity, genetic diversity levels, and effective population size are high and homogeneous in the western French-Spanish Basque population, when excluding the northern French colony "MON." Therefore, the French Poitou-Charentes region does not need to be considered as a management unit by itself. We rather recommend considering the large population as a unique management unit. The development of new partnerships or the reinforcement of existing ones between NGOs from different neighboring countries (Spain, France) and French administrative regions are needed to improve the knowledge and conservation of this (and other) bat species in France. This large population could be resilient to local disturbance because of its strong interconnection between colonies. However, we cannot exclude that some particular colonies within this population might be vulnerable. It is therefore still important to pursue colony surveys at local scales and also to standardize monitoring procedures at the national and international scale (Battersby, 2010). Our sampling scheme did not enable us to identify the eastern boundaries of this population or to test for large geographical barriers to gene flow. It would require more sampling in Eastern France and in neighboring countries (e.g., Belgium, Germany, Luxembourg, and Switzerland). In the future, this delineation and inference of bat population demography might be of particular importance as R.
ferrumequinum experienced severe declines there, so that some populations might be at higher risk of extinction and deserve special management attention (see Ransome & Hutson, 2000).
We have also shown in this study that peripheral colonies are genetically poorer than those at the core range, because of genetic drift, low gene flow, and small effective population size. Thus, these colonies are more vulnerable to extinction and deserve particular management efforts. Interestingly, these colonies located at the edge of the species range are genetically divergent and may harbor some genetic and phenotypic variability that could be important for adaptation to global changes (Lesica & Allendorf, 1995). For example, these colonies may play a key role in the face of climate change by facilitating species range shift northward (Rebelo, Tarroso, & Jones, 2010).
Lastly, our results advocate for paying particular attention to mating territories and to movement pathways that enable extracol- have drafted much of the manuscript. All authors read, criticized, and approved the final manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Microsatellite genotypes for this study are available at: https ://doi. org/10.5061/dryad.r44t5dk.