Discussion

The relationship between the within and between group variance did not comply with the predictions of the model for drift. As predicted, geographic variation in the skulls and mandibles of both lineages was thus likely the result of selection, in accordance with our first prediction. Modularity was only supported in R. simulator skulls, the cranium and muzzle evolving as separate and integrated units. Contrary to our second prediction, the mandible of R. simulatorand both the skull and mandible of R. cf. simulator , did not show modularity. Thus the two closely related lineages (Dool et al. , 2016) showed contrasting results with respect to modularity. We also found evidence consistent with our hypothesis that more prominent variation in shape were seen in the nasal dome in R. cf. simulator than in R. simulator . Thus Lande’s model indicates that selection rather than drift has shaped the skulls of these two lineages. The results on modularity (see reference 43) suggest that the selection responsible for the diversification of R. simulator is predominantly directional (in the skull) and stabilising in the mandibles, whereas in R. cf. simulator , is mainly stabilising for both the skull and the mandible.
To some extent, these results contrast with previous findings on the relative contributions of drift and adaptation based on Lande’s model which identified signals of drift in some instances (Marroig & Cheverud, 2004; Weaver et al. , 2007). Mutumi et al . (Mutumi et al. , 2017) report signals for drift but their study was based on a broader range of phenotypic features including flight, size and echolocation parameters. Perhaps the fact that the skull incorporates several functions (e.g., feeding and echolocation) crucial to fitness causes it to be under severe selection pressure that could eliminate or obscure any drift that might have occurred. The head is under the influence of multiple selective pressures because it houses the structures used for a variety of crucial survival and reproduction functions, particularly echolocation. Both lineages appear to have experienced selection pressure associated with echolocation, a key survival trait. Echolocation is a sophisticated sense that varies strongly with the task at hand and environmental conditions (Schnitzleret al. , 2003; Jakobsen et al. , 2013; Luo et al. , 2014).
It surprising that modularity was present only R. simulator and not in R. cf. simulator because modularity has been reported across 22 African and Asian species of rhinolophids (Santana & Lofgren, 2013). However, stabilizing selection is thought to mitigate against the evolution of modularity (Melo & Marroig, 2015). The absence of modularity in R. cf. simulator may be a consequence of stabilizing selection to retain the adaptive complex between flight, body size and echolocation. In this respect the evolution of R. cf. simulator is similar to Phyllostomidae which is tightly integrated and probably evolved under the constraint of preserving adaptive complexes (Hedrick et al. , 2019). Body size, wing loading and echolocation frequency in bats are associated allometrically and are indicative of an adaptive complex (Jones, 1999; Jacobs et al. , 2007; Jacobs & Bastian, 2018). With respect to these allometric relationships R. cf. simulator is an average rhinolophid. Its echolocation frequency and wing loading fall within the allometric relationships of the genus (Jacobs et al. , 2007; Jacobs & Bastian, 2018).
In contrast to R. cf. simulator there was evidence of modularity in R. simulator suggesting that its skull was under directional selection (48). Unlike R. cf. simulator , the adaptive complex between echolocation frequency and body size is absent. Although its wing loading scaled allometrically with body size, R. simulatorecholocated at a lower frequency for its body size (Jacobs et al. , 2007; Jacobs & Bastian, 2018). Furthermore, it also had lower echolocation frequencies than would be predicted by the volume of its nasal capsules (Jacobs et al. , 2014). This suggests directional selection for lower frequency echolocation, possibly to increase the operational range of its echolocation, reflected in the phenotype of the skull associated with echolocation. Lower frequency sound undergoes less atmospheric attenuation than high frequency sound (Lawrence & Simmons, 1982) and, all else being, the echolocation of R. simulatorshould therefore have longer operational ranges than R. cf. simulator , unless it emits echolocation pulse at lower intensities. Currently the intensities at which these two lineages emit their echolocation pulses are unknown. If the same, the fact that R. simulator and R. cf. simulator were sometimes caught at the same locality and from the same cave, suggests that their use of different echolocation frequency with consequent differences in the operational range of their echolocation pulses, may be a means of partitioning their foraging habitat, if not their diet. In both lineages, the mandible evolved as one complete module (ascending ramus and alveolar bone) contrary to the mandibular modularity found in R. ferrumequinum(Jojić et al. , 2015). The mandible has therefore possibly evolved under constraint and might be following a line of least evolutionary resistance as in the phyllostomids (Hedrick et al. , 2019). The mandible variations across localities did not show any difference between the two species except the variations on the position of the incisors that were seen in R. cf. simulator but not in R. simulator . The similarities between the mandibles signifying close similarities in diet between the two species.
The marked influence of echolocation on the skull of both R. simulator and R. cf. simulator is also reflected in variations in the shapes of cochlea in both species. This suggests that selection has acted strongly on both sound production and perception functions in the two lineages. Variations in the morphology of the cochlea are related to variations in perceptions of sound, particularly in rhinolophids (Davies et al. , 2013). For example, in rhinolophids the cochlear basal turn is expanded, more so than in other bats (Davieset al. , 2013), probably because of the well-developed auditory fovea in this taxon allowing the Doppler shift compensation upon which high duty cycle echolocation is based (Neuweiler, 2003). The frequency of echolocation pulses in rhinolophids are also negatively associated with the length of the basilar membrane length and positively associated with the number of cochlear turns (Davies et al. , 2013). These relationships suggest that the cochlea of these bats probably track the acoustic properties of the habitats they occupy, hence the geographic variation reported here in both echolocation frequency and cochlea morphology. The finer details of the mechanistic association between cochlea morphology and echolocation parameters still need to be elucidated (Davies et al. , 2013).
The differences in the selection pressures experienced by the two lineages are remarkable given the genetic similarity of the two lineages at least in the genetic markers considered by Dool et al. (Dool et al. , 2016). The two lineages were indistinguishable across 6 nuclear introns and one mitochondrial fragment. It has been suggested that R. cf. simulator is possibly a cryptic lineage, sister to R. simulator (Dool et al. , 2016), a view supported by the differences in the evolution of skulls reported here. However, the two lineages occur at the same sites and sometimes in the same caves and there is some evidence of hybridization between the two lineages. This raises the question of how they can maintain such divergent and non-overlapping echolocation frequencies. The answer to this question requires evolutionary development studies to identify the loci which code for echolocation frequency (e.g., Sun et al (Sun et al. , 2020)) and how these loci are assorted during gamete formation.