Geometric morphometric
We analyzed 233 specimens of 32 bat species from Serra da Bodoquena, a karstic region in the southern Brazilian Cerrado (Sallun-Filho et al. 2004; Sallun-Filho et al. 2010). The families (and subfamilies) included were Phyllostomidae (Carolliinae, Desmodontinae, Glossophaginae, Lonchophyllinae, Micronycterinae, Phyllostominae and Stenodermatinae), Emballonuridae (Emballonurinae), Molossidae (Molossinae), Noctilionidae, and Vespertilionidae (Myotinae and Vespertilioninae). All specimens were adult males available from the Zoological Collection of Universidade Federal de Mato Grosso do Sul (ZUFMS), Natural History Museum of Universidade Federal de Lavras (Pigoderma bilabiatum, Micronycteris microtis), and the interactive mammal website of Espírito Santo (Eptesicus brasiliensis) (Table S1-S2). For every specimen, we got 2D images with scale for skull and jaw carried out using Zeiss Discovery V.20 stereoscope microscope.
To get the shape of the species, we selected landmarks and semi-landmarks of skull and jaw following the studies performed with Phyllostomidae (Nogueira et al. 2009), Vespertilionidae (Ospina-Garcés and de Luna 2017), Molossidae (Freeman 1981) and Noctilionidae (Romero 2011) (Figure S1-S2, Table S3-S4). The positioning of landmarks was performed with the software TPSUtil32 and TpsDig 232 (Zelditch et al. 2012; Rohlf 2015). When individuals did not have the structure where the landmarks should be positioned, we estimated their locations from other conspecific specimens containing that structure, through the estimate.missing function in the geomorph package (Adams and Otárola-Castillo 2013). We used Generalized Procrustes Analysis (GPA) to remove effects of rotation, translation, and size. This analysis minimizes the Procrustes distance between corresponding landmarks on each specimen and compute a mean to each sample (Slice 2007). The size of skull and jaw was extracted from the calculation of the centroid size of the landmarks to each species.
Trophic guild and foraging strategies
Species were categorized into five trophic guilds following the classification of Kalko et al. (1996): animalivores, which include sanguivore and carnivores; insectivores; frugivores; nectarivores; and omnivores. For foraging strategy, we classified the species into four groups through combination of the classifications from Kalko et al. (1996) and Denzinger and Schnitzler (2013): active cluttered space foragers (CSA), which are edge insectivores; passive-active cluttered space foragers (CSPA), which are the frugivores, nectarivores, omnivores and hematophagous; passive cluttered space foragers (CSP), which include animalivorous; and active uncluttered space foragers (USA), which are the insectivores of open areas. These classifications are related with habitat use, foraging behavior, and capture mode.
Data analyses
Phylogenetic comparative methods
We used the phylogeny of Shi and Rabosky (2015) for subsequent analyses. This phylogeny was constructed with a supermatrix of mitochondrial and nuclear sequences using Bayesian inference and contains data for 812 bat species (56.78% of the currently known species). So, we pruned this tree to contain the 32 species found in this study. Four species absent in this phylogeny were replaced by phylogenetically closely related species (Pennell et al. 2016): Lonchophylla dekeyseri for L. mordax (Woodman and Timm 2006), Micronycteris sanborni for M. minuta (Morales-Martínez et al. 2021), Artibeus cinereus for Dermanura bogotensis (Agnarsson et al. 2011), and Pteropteryx macrotis for Cormura brevirostris (Lim and Dunlop 2008) (Fig. 1). To visualize the species distribution in the morphospace in relation to the skull and jaw, we first performed a Phylogenetic Principal Component Analysis (pPCA) with the coordinates of the mean shape of each species (Revell 2009) with the gm.prcomp function from geomorph (Adams et al. 2022). We built the phylomorphospace with the first two PCs (Adams and Collyer 2019). To test whether skull and jaw shape are influenced by size (continuous predictor), trophic guild (five-level factor), and foraging strategy (four-level factor) we used a phylogenetic generalized least squares model (PGLS) with lm.rrpp function in geomorph (Collyer and Adams 2018). Thus, we build a model including interaction between the two factors (shape ~ size + guild*strategy). As the factors were significant, we performed a post hoc analysis to test differences between the average distances of the factor level combinations (i.e., Ani:CSA Ani:CSP Fru:CSPA Ins:CSA) (Collyer and Adams 2018).
Phylogenetic signal of shape and size
To estimate the phylogenetic signal of the mean shape and size of the skull and jaw, we used Blomberg's K statistic adapted for multivariate data (Adams 2014b). To calculate this value, we use the physignal function from geomorph, with 10,000 randomizations. A value of K lower than 1 indicates that the species are less similar than expected under Brownian motion (BM), and greater than 1 indicates that the species are more similar than expected by BM model (Adams 2014b; Adams et al. 2022).
Evolutionary allometry
To test for evolutionary allometry of the skull and jaw between guilds and foraging strategies, we built linear models with additive and interaction effects (i.e, shape means ~ size means * guild * strategy, and shape means ~ size means + guild * strategy). In this way, we assessed whether the groups present unique or common trajectories in the variation of shape in relation to size. Afterwards, we evaluated the significance of both models with an ANOVA. Finally, we visually inspected the trajectories with the geomorph's PredLine, which calculates the predicted values from the shape-to-size regression and graphs the component 1 obtained from the shape versus size linear regression (Adams and Nistri 2010; Adams et al. 2022). The significance of the difference between groups in terms of size and shape was calculated with a post hoc test (Collyer and Adams 2018).