Our results reveal significant geographic variation in the resting echolocation frequencies (RF) of R. capensis despite substantial historical gene flow across the distribution of this species. Body size, genetic distance and geographic distance play minor roles in the evolution of geographic variation in RF in R. capensis. Further, despite a lack of strong geographic structure in mitochondrial lineages, biome was identified as the best predictor of RF. In support of this we found a significant relationship between RF and increasing vegetation clutter from west to east across the range of this species. Our results suggest that the evolution of geographic variation in RF in the face of homogenizing gene flow in R. capensis was probably influenced by selection for lower echolocation frequencies in less cluttered habitats. However, the relationship we found between RF and habitat type is solely correlative, and therefore a thorough comparison of foraging behaviours of bats between different habitats is required. Thus, alternative evolutionary processes e.g. selection for discrete frequency bands or phenotypic plasticity cannot be excluded at this stage.
Disparities between RF and population genetic structure: trait diversification in the presence of gene flow
Studies investigating the various evolutionary forces shaping phenotypic and genetic divergence between populations often describe trait variation in the context of substantial population structure and limited gene flow . While phylogeographic patterns of maternally (mtDNA) and bi-parentally (e.g. microsatellites) inherited genetic markers usually concur, discordance between phenotypic and genetic structure is also reported. This is usually attributed to demographic processes that characterize species, such as sex-biased dispersal  or secondary contact following historical isolation  (reviewed in ). Similar discordance has been reported in several European and Asian horseshoe bats either as a result of male-biased dispersal and female philopatry (e.g. R. pumilus: ) or historical introgression of mtDNA (R. pearsoni: ; R. sinicus: ; or nuclear genomes (R. yunanensis to R. pearsoni: ) between sister lineages. Despite recent advances in testing the evolutionary processes that shape contemporary population genetic structure in horseshoe bats (e.g. ), few studies have specifically evaluated sensory variation within a phylogeographic or population genetic framework in this genus [66, 69]. In our study, we expected RF divergence in the R. capensis to reflect mtDNA structure given that (i) the fine tuning of the echolocation frequency of young horseshoe bats are partly learned from their mothers, (ii) female philopatry and male dispersal characterize other horseshoe bats studied to date [66, 133], and (iii) the degree of RF divergence observed among populations (range: 1-11 kHz) is similar to that reported for other high duty cycle bats which have corresponding significant genetic structure among maternal lineages . Our results instead reveal minimal genetic structure amongst R. capensis populations. A Bayesian clustering analysis identified four dominant genetic lineages which broadly reflect estimated patterns of regional gene flow that are not limited by biomes; notably only LS and TF (populations at the opposite extremes of the distribution of the species) are identified as unique genetic clusters (Figure 5). A number of mitochondrial haplotypes were also found to be shared between geographically distant populations; these may reflect long-distance dispersal events or the retention of common ancestral polymorphisms. Mitochondrial data clearly reveals a recent evolutionary history of complex reticulations in R. capensis, suggesting that gene flow and not incomplete lineage sorting is responsible for the observed genetic structure or lack thereof.
One caveat of our analytical approach is the statistical non-independence of our populations due to the substantial gene flow we detected between them . Populations that are more closely related to one another or exchange higher numbers of migrants are likely to have similar phenotypic trait values irrespective of the local selection pressures they experience . Ideally, the inclusion of our calculated migration matrices between populations in our ecological models would control for the effect of gene flow, but the development of effective computational methods to include the complex reticulate relationships among population’s remains challenging [134, 135]. Nonetheless, the spatial distribution of the four dominant genetic lineages does not reflect the broad pattern of RF variation among populations of R. capensis. For example, we detected considerable gene flow between populations at SKK, ZPK and DHL, but no direct gene flow between these populations and the population at DHC, even though they were all assigned to the same genetic cluster (Figure 5). Bats from these genetically similar populations use RF’s ranging from 80.6-84.6 kHz. Thus the discordance between patterns of genetic and RF variation among R. capensis lineages suggest that gene flow does not significantly contribute to RF variation across the distribution of R. capensis, although this remains to be explicitly tested. Furthermore, although RF is also characterised by IBD, geographic distance only accounted for a minor proportion of the variation in RF (Figure 6A and B). Interestingly, RF divergence is also positively correlated to some degree with genetic distance, even after controlling for geographic distance (Figure 6C), suggesting that local environmental conditions may influence sensory differentiation in areas with some degree of restricted gene flow .
However, estimates of maternal gene flow support significant regional connectivity in the recent past and is at odds with the pattern of structuring in RF we observe across populations. This is in contrast to previous studies of high duty cycle bats where population genetic structure generally reflects the variation of sonar frequencies across often widespread species (e.g. Pteronotus parnellii: , R. clivosus: , R. hildebrandtii: , R. rouxii: ). As we discuss below, this anomaly may be a consequence of the complex interactions between gene flow, diversifying selection and environmentally mediated selection for phenotypic plasticity. Phenotypic plasticity can be advantageous if it results in the expression of different phenotypes that increase an individual’s fitness in diverse environments [19, 138]. In this way plasticity can minimise the costs incurred from dispersal into environments with different selection regimes [18, 29]. If plasticity influences RF diversification, it may lead to a situation in which selection does not significantly constrain gene flow among populations, leading to adaptive phenotypic variation in the presence of gene flow .
Body size and habitat structure predict RF: influence of diversifying selection and adaptive phenotypic plasticity
The frequency of acoustic signals scales negatively with body size in a range of organisms [139–141] including bats ; larger bat species produce echolocation calls of lower frequencies than smaller bat species. This relationship also exists within R. capensis with larger bats generally using lower RF’s. However, body size only explained a minor proportion of the variation in RF because concomitant changes in body size and RF were not evident across all populations. For example, although BAV bats are not the smallest, they use the highest echolocation frequencies. In contrast, individuals in populations at both ends of the distribution of this species (LS in the west and TF in the east) were similar in size but echolocated at the lowest and second highest frequencies respectively (Table 1, Figure 3). This may explain why previous research on echolocation variation in the species did not support a relationship between body size and RF, and instead suggested the decoupling of echolocation divergence from the evolution of body size . This result was likely an artefact of under sampling trait variation in R. capensis and highlights the difficulties of inferring evolutionary processes from data sets which inadequately sample the true distribution of a trait. Our results here suggest that the allometric relationship between body size and RF collapses in populations situated towards the edge of the range of this species. This is perhaps not unexpected given that range edges and ecotones provide novel environments to which species can become locally adapted, leading to significant phenotypic divergence of edge populations from those at the centre of the distribution of a species . This may explain why populations at either end of the species’ range (LS in the west, and TF in the east) were identified as unique genetic clusters, while the nine sampled populations between them were assigned to only two genetic clusters (Figure 5).
Paleoenvironmental change from the Miocene to Pleistocene profoundly impacted the evolution of population divergence and speciation in a wide range of southern African taxa; several studies report a strong link between divergent genetic lineages and the biomes or ecogeographical regions of southern Africa for various organisms including reptiles , invertebrates  and small mammals [145, 146] including bats [82, 87]. In our study we did not find any clear population genetic divergence with topographical features. Neither did we find any support for the humidity hypothesis which proposes that divergence in RF is the result of selection against higher frequencies in humid environments . Instead we found that bats in the Desert Biome used significantly lower frequencies than those occupying Forest and areas of transition between multiple biomes (Figure 4, Table 3). This clinal increase in RF across populations of R. capensis from west to east may be the result of the gradient of increasing vegetation cover and density from west to east (Figure 4). At LS in the west the vegetation was very sparse and consisted of low shrubs (< 1m in height). In the east the vegetation ranged from dense Fynbos to Forest (Figure 4). Thus, there is a steep habitat gradient between LS (genetic cluster 1) and its nearest neighbours (SKK and ZPK: genetic cluster 2), and selection may therefore dominate in this region. In contrast, plasticity may be favoured as an explanation for RF variation amongst the other populations because the habitat gradients are not as steep. A notable exception is that of the population at TF. At TF (genetic cluster 4) bats use similar RF’s to other populations situated in ecotones (genetic cluster 3), and yet appears to be relatively isolated genetically (Figure 5). TF is situated at the interface between aseasonal and summer rainfall zones  which, together with the intrinsic habitat heterogeneity of ecotones, may serve as a significant barrier to gene flow. This pattern has been observed for various other species in this region (e.g. four-striped mouse, Rhabdomys pumilio; forest shrew, Myosorex varius).
The positive correlation we found between RF and NDVI suggests that variation in the degree of habitat clutter might explain variation in RF. Across our sampling sites, the mean detection distance for large prey and background vegetation edge was significantly different between populations; bats occupying more open habitats have lower RF’s and thus longer detection distances than those in more cluttered habitats, allowing them to detect larger prey or background targets at greater distances. While this result was statistically significant, LS bats only had a 10 cm and 50 cm greater detection distance than their nearest neighbours for large prey and vegetation edge, respectively (Additional file 2: Table S1). Although all rhinolophids supposedly fly close to vegetation and may experience even relatively open habitats as cluttered, the 50 cm greater detection distances of the vegetation edge may be advantageous during orientation and commuting flight in the sparse vegetation of LS where the distance between clumps of vegetation are greater than in the Fynbos or the Forest (Figure 4). Further, prey density is also likely to be lower at LS and selection is likely to favour the evolution of lower RF’s that would allow the detection of larger prey at greater distances. There is sufficient empirical evidence to suggest that even rhinolophids, constrained by their echolocation to hunt in narrow space , display some degree of flexibility in the foraging habitat they exploit [148–151] or the foraging style they adopt (aerial hawking vs. perch hunting) [59, 152] as a result of resource partitioning [150, 153], habitat structure  or seasonal changes in prey resources . At least one species of rhinolophid also appears to vary its echolocation frequency in response to different degrees of clutter . Within the same nature reserve, greater horseshoe bats (R. ferrumequinum) use a variety of habitats with differing degrees of clutter and use significantly lower echolocation frequencies in relatively open habitats than in cluttered habitats . It is likely therefore that R. capensis uses both aerial hawking and perch hunting styles to different degrees in the different habitats perhaps also altering its call frequency to deal with different degrees of clutter. The distinctly lower call frequency at LS could possibly be explained by the pronounced clutter gradient between LS and the other habitats occupied by R. capensis. However, our detection distance calculations must be interpreted with caution because different populations may have different echolocation call intensities, which could greatly influence the calculation of maximum detection distance .
Although differences in habitat structure provide a compelling explanation for different RF’s in R. capensis, an interesting anomaly needs to be accommodated; R. damarensis (forearm length 49.5 ± 1.7 mm; n = 20) is sympatric with R. capensis in the extremely arid area of LS but uses frequencies (mean ± SD = 85.4 ± 1.4 kHz; range = 82–89 kHz; ) as high as those used by R. capensis in highly cluttered habitats e.g. TF. It seems reasonable to assume that selection would have favoured similar call frequencies, and therefore detection distances, in R. damarensis and R. capensis where they are sympatric in the arid region of LS. However, it is possible that R. damarensis exploits a different foraging niche to R. capensis in this area of sympatry. Alternatively, this species may compensate by using higher intensity calls to achieve more or less the same detection distances as R. capensis. Of course it is possible that the presence of R. damarensis at LS has selected for the low frequencies we observe in R. capensis. Observed frequencies are much lower in the LS population than in the other arid populations (Figure 4) despite similarities in body size to the other arid populations (Table 1). On the basis of its body size, R. capensis at LS should echolocate at approximately 82 kHz (Figure 3). Instead we observe a mean RF of 75.7 kHz. Bats at LS may have shifted their frequency below 82 kHz to avoid acoustic overlap with R. damarensis (82-89 kHz), and maintain effective intraspecific communication. However, the closest known roosts of these two species are 40 km apart and it is not known if the foraging areas of these two species overlap, i.e. if they are syntopic, as required by the Acoustic Communication Hypothesis.
Nevertheless, character displacement, mediated by some form of resource partitioning or local adaptation, in sonar parameters can contribute to the initial stages of lineage divergence by causing populations in sympatry with heterospecifics to diverge from their respective conspecific populations in allopatry. The result would be reduced gene flow between populations found in different selective regimes or heterospecific assemblages . For example, Lemmon  showed that acoustic traits important for female preference and mate choice in populations of the chorus frog, Pseudacris feriarum, diverged to maximize differences from the heterospecific assemblage present, ultimately promoting reproductive isolation between conspecific populations via sexual selection. Ecologically adaptive traits can also promote divergence if divergence has a pleiotropic effect on reproductive isolation via assortative mating; so called ‘magic traits’ . In R. philippinensis assortative mating has evolved between size morphs as a by-product of selection for different frequencies used to exploit different prey sizes . The significant and consistent sexual dimorphism we observe in the RFs of R. capensis (female’s echolocate at higher frequencies than males: Table 1), may serve a role in sex-specific communication in the species. Recent experimental evidence reveals that horseshoe bats (R. euryale and R. mehelyi) and emballonurid bats (Saccopteryx bilineata) are indeed able to recognise the sex of conspecifics based on their echolocation calls [44, 156]. Thus, the limited gene flow between LS and other populations may be a consequence of LS bats not effectively recognizing other R. capensis as potential mates. Divergence in RF may have under-appreciated consequences for the evolution of reproductive isolation via female preference for male RFs in different populations of horseshoe bats. Evaluating female preference in LS bats for local versus allopatric RFs may provide intriguing insights into the causes and consequences of sexual selection in horseshoe bats. Finally, because R. damarensis only occurs in sympatry with R. capensis at a single site (LS), we were not able to test whether habitat or the presence of R. damarensis offered a better explanation for RF differentiation in R. capensis. An experimental approach is required to test whether R. capensis has shifted its RF to avoid acoustic overlap with R. damarensis. Such an approach would evaluate whether R. capensis from LS responds differentially to the echolocation calls of heterospecifics and acoustically divergent conspecifics using playback experiments.
At a ‘local to regional’ scale geographic distance is not a significant barrier to gene flow in R. capensis, and the evolution of sensory divergence in the presence of this gene flow may also reflect a degree of adaptive phenotypic plasticity in RF. Despite the tight coupling between RF and the acoustic fovea in high duty cycle bats , empirical studies have shown that species are able to shift their RF’s in response to both neighbouring conspecifics (maximum shift 3.9 kHz: ) and different ambient noise conditions (maximum shift <0.5 kHz: ). Such small shifts in frequency may explain the range of RF variation in the southern and eastern populations of our study (approximately 3 kHz) where plasticity in response to slightly varying degrees of vegetation clutter towards the east might occur. However, it is unlikely to explain the 9 kHz shift in the Lekkersing bats. Nonetheless, it appears that southern and eastern populations of R. capensis use RF’s within the best hearing range of the acoustic fovea of their nearest neighbours, possibly facilitating gene flow and promoting relatively flexible RF’s in these populations. While small shifts in the acoustic fovea and its corresponding reference frequency are possible in high duty cycle bats [46, 81], we do not know the precise limits of the flexibility of the acoustic fovea. Long-term experimental studies evaluating the change in RFs in response to the RFs of bats from acoustically divergent populations may shed light on the degree of plasticity in this system. Alternatively, the relative influence of plasticity versus selection can be evaluated indirectly. For example, selection may better explain our observations if variation in functional genes involved in hearing co-varies with RF variation across populations. Recent studies reveal a wide range of candidate hearing genes which show strong signals of ancestral positive selection in the evolution of echolocation in bats and cetaceans [157, 158]. Selection may also be favoured as an explanation for variation in RF if there is a strong correlation, independent of variation in body size, between RF and morphological features directly involved in echolocation production and emission (such as dorsal nasal chambers) . Previous research investigating morphological correlates of RF in R. capensis revealed that RF is best predicted by nasal chamber length , but a thorough evaluation of skull morphology variation across the distribution of the species is required. The social life of bats may also influence the relative roles of diversifying selection versus plasticity across the distribution of a species. If bats are able to recognise conspecific calls from a range of acoustically divergent populations this might suggest that selection for some degree of plasticity in the trait is also favoured. Classic playback experiments can be used to assess the sensitivity of individuals to the range of frequencies exhibited by a species. These two hypotheses are clearly not mutually exclusive; and their relative influences across the highly heterogeneous environments of R. capensis certainly merit further attention.