The vertebrate cranium is a complex anatomical structure with diverse mechanical and sensory functions. Shifts between modularity and integration in both sets of functions, especially mechanical function, have been implicated in adaptive diversification. However, how mechanical and sensory systems and functions have coevolved and how their interrelationship contributes to phenotypic disparity remains largely unexplored. To examine the modularity, integration, and evolutionary rates of sensory and mechanical structures within the head, we analyzed hard and soft tissue scans from ecologically diverse bats from the superfamily Noctilionoidea, which range from generalized insectivores to derived frugivores and nectarivores. We identified eight cranial regions as distinct modules — five associated with bite force and three linked to the olfactory, visual, and auditory systems, respectively — whose interrelationships differ between Neotropical leaf-nosed bats (Family Phyllostomidae) and other noctilionoids. Our analyses suggest that the peak rates of sensory module evolution predate those of mechanical modules. This finding is consistent with transitions to new diets first involving changes in the detection of novel food items, followed by adaptations to process them. We propose the coevolution of structures influencing bite force, olfaction, vision, and hearing constituted a structural opportunity that allowed the phyllostomid ancestor to take advantage of existing ecological opportunity and the group to become a classic example of adaptive radiation.

To quantify the sizes and shapes of mechanical structures, the olfactory bulb, and cochlea, we visualized the heads of bats using a combination of standard computed tomography (CT) for hard tissues. Bat specimens were scanned using a Nikon Metrology (X-Tek) HMXST225 microCT system at the Center for Nanoscale Systems at Harvard University. Three-dimensional (3D) images were processed following (Hedrick et al. 2020). We generated image stacks using proprietary software associated with the X-Tek scanner (CTPro, Nikon Metrology Inc., Japan), segmented image stacks using Mimics v. 16.0 (Materialise, Leuven, Belgium), created meshes using VGStudio Max 3.0 (Volume Graphics Inc., Germany) and exported them as PLY files.

We used reconstructions of eyes previously published by (Hall et al. 2021), which include diffusible iodine-based contrast-enhanced computed tomography (DiceCT; (Gignac et al. 2016; Hedrick and Dumont 2018)) from the same specimens (S1 Table).  Briefly, we defined the orbital space surrounding the left eye by its muscular boundaries and eyelid. We used the volume and location of the orbital space as a proxy for eye location and volume because it is less subject to distortion than the globe itself in fluid-preserved museum specimens (Hedrick and Dumont 2018).

Placing Landmarks

Our data set included a total of 322 landmarks: 43 fixed landmarks, 160 sliding semi-landmarks on curves, 55 surface landmarks on the eye, and 64 surface landmarks on a patch placed on the palate (S2 Table, S1 & 2 Video) that were placed on 3D meshes using IDAV Landmark Version 3.6 (Wiley et al. 2005). Sliding semi- and surface landmarks are adjusted to reduce their weight in the analysis, thus reducing the potential effect of their representation by large numbers of semi- and surface landmarks. These methods involve sliding neighboring landmarks along curves and surfaces to minimize bending energy and ensure that the arbitrary spacing of semi-landmarks does not influence shape variation (Bookstein 1997; Gunz and Mitteroecker 2013). We placed external landmarks on the external surface of the cranium as per Hedrick et al. (2020). To landmark structures on internal surfaces of the bony skull, 3D models were digitally dissected in Geomagic Studio 2014 (3DSystems, SC, USA) into three parts to reveal the cochlea, the impression left by the olfactory bulb on the internal surface of the skull (anterior cranial fossa), and the internal surface of the cranial base.

We identified landmarks to represent the eye in three steps. First, we used Mimics to calculate the volume and centroid of the orbital space reconstructed by Hall et al (2021). We then generated a sphere of equal volume around the centroid and placed 55 landmarks on its surface. Reducing the eye landmarks to seven (one in the center and six across perpendicular poles) did not change the results of tests for modularity (S9 Table)), and so we proceeded with the data set of 55 eye landmarks. Because our eye data are based on idealized and identical shapes, the results of our analyses reflect variation in the orientation and overall size of the eye. Throughout the landmarking process, we kept all parts of each specimen in the same coordinate system so that landmarks could be concatenated into a single file using custom R code.

To ensure that the number of landmarks selected did not affect the number of modules recovered, we conducted a sensitivity analysis by downsampling the number of landmarks by 25%, 50%, and 75% and testing whether this changed the number of modules detected. Custom code was written in R to subsample within each module systematically; every second landmark (50%), one in every four landmarks (25%), and the first three of every four landmarks (75%). The same number of modules were recovered using EMMLi from the complete set of noctilionoid landmarks and all down-sampled datasets, and vector congruence correlation matrices of the data sets were strongly correlated (R2 = 0.73 – 0.95; S3 Table). We further tested the effect of landmark number on modularity by regressing the number of landmarks against the ρ coefficient of each module (a measure of integration), first for the whole noctilionoid group and then for the sub-groups of phyllostomids and other noctilionoids. We found no evidence that the number of landmarks in each module influenced modularity (S8 Table). Based on these analyses, we used the full set of 322 landmarks for further analyses. To employ phylogenetic comparative methods and adjust for unequal sample sizes in our analyses, we used species means for taxa represented by multiple individuals (7 species, S1 Table).

Defining Modules

To test hypotheses about modularity of sensory and mechanical systems we grouped landmarks into eight hypothetical modules that encompass anatomical structures associated with specific functions (Fig 1; S1 & 2 Video. Among mammals, the size of sensory structures is often directly related to function. To conserve the size component in our analyses of shape variation, we did not adjust data for allometry. Therefore, our landmark data still carry shape variation that is associated with size. The shapes of some mechanical regions are clearly linked to functional variation (e.g., Dumont 2004; Dumont et al. 2009, 2012; Santana et al. 2012; Neaux et al. 2021). Here we make inferences based on those linkages but do not directly measure mechanical variables. The shapes of sensory structures (e.g., olfactory bulb, cochlear region, and eye) are unlikely to perfectly reflect sensory ability. Likewise, it is likely that some shape changes don’t reflect functional change at all (Gould and Lewontin 1979). Nevertheless, we propose that there is at least circumstantial evidence to suggest that each of our eight modules carries some functional signal. 

We defined three sensory modules that encompass the olfactory bulb (smell), the cochlea (sound), and the eye (vision) (Figure 1; S1 & 2 Videos). Enlarged eyes are associated with increased visual acuity  (Müller and Peichl 2005; Müller et al. 2007; Land and Nilsson 2012; Eklöf et al. 2014; Veilleux and Kirk 2014; Sadier et al. 2018) and eye orientation is related to activity pattern in primates (Heesy 2008). Larger olfactory bulbs support more expansive epithelia and therefore larger surface areas for odor detection (Barton et al. 1995; Buschhüter et al. 2008; Corfield et al. 2015). Enlarged olfactory epithelia are associated with frugivory among noctilionoid bats (Yohe et al. 2021). Finally, cochlear volume and shape are correlated with aspects of cochlear morphology that influence hearing performance (Kössl and Vater 1995; Kirk and Gosselin-Ildari 2009; Vater and Kössl 2011; Davies et al. 2013a, 2013b). Among bats, variation in the relative volumes of sensory structures often tracks variation in foraging strategy and diet (Barton et al. 1995). For example, the ancestral phyllostomid diverged from its sister taxa in having relatively larger olfactory bulbs and eyes, which is characteristic of frugivory (Hall et al. 2021). Similarly, the shift to plant-based diet in palaeotropical fruit bats (Pteropodidae) allowed larger skulls and brain regions associated with vision, olfaction, and spatial memory (Thiagavel et al. 2018). We defined olfactory, cochlear, and eye modules with landmarks that encompass the three-dimensional volume as well as shape of those structures (Figure 1, S1 & 2 Videos). Olfactory module landmarks were placed on the impression of the olfactory bulb on the internal surface of the skull, cochlea module landmarks were placed on the region of the cranial base that encompasses the cochlea, and the eye module was defined by landmarks placed on the surface of the reconstructed orbital sphere. 

We used predominantly R statistics https://www.r-project.org/. For rates through time we used BayesTraits http://www.evolution.reading.ac.uk/BayesTraitsV4.0.0/BayesTraitsV4.0.0.html