Morphometric data.
We examined 335 adult specimens (fully erupted M3; following age classes from Arellano et al. 2012) deposited in mammal collections and classified at the time of collection as R. mexicanus . However, due to the cryptic diversity reported for this species, only individuals that, 1) had a known genetic identity or 2) were collected within less than 60 km of an individual with a known genetic identity were used in the morphometric analyses. In total, 69 specimens (31 males, 38 females) were selected for morphometric analysis (Fig. 1; Appendix 1) and grouped according to the species proposed by the molecular delimitation methods. Individuals from El Salvador and Colombia (R. mexicanus clade IIB and clade IIIA in Fig. 2) were excluded due to the low sample size (n ≤ 5). Digital images of the dorsal and ventral view of the skull of each specimen were taken using an Olympus DP73 Digital Camera and a millimeter rule as a scale bar.
For both views, we digitized landmark and semilandmark configurations as in Martínez-Borrego et al. (2022b) using TPSdig 2.31 (Rohlf 2015). Configurations were alignment, rotate, and scaled under a Generalized Procrustes Analysis (Rohlf and Slice 1990) in the R package geomorph 4.04.0 (Adams et al. 2022). In the case of semilandmarks, they were aligned by sliding points along their tangent vectors until reaching the point of minimum bending energy (Bookstein 1997; Zelditch et al. 2004). Configurations superimposition outputs were the shape variables (Procrustes Distances and Procrustes Coordinates) and the centroid size (CS).
Morphometric comparison between delimited species.
We first explored the shape variation in both views of the skull using a Principal Component Analysis (PCA) of the Procrustes Coordinates to visualize clustering and reduce the dimensionality of the data. We tested skull shape differences related to sex and delimitated species using a Procrustes ANOVA model (Klingenberg and McIntyre 1998), where we used the Procrustes distance variance to estimate shape variance in each factor. The factorial design included shape as the dependent variable, sex and species as the main factor, and CS as a covariate to also evaluate the effect of the skull size on the shape variation (allometry). Additionally, we used a residual resampling procedure, on the reduced model (Shape~CS), based on 1000 iterations, to assign the significances of the F statics of the model. These analyses were performed in the package geomorph. We also performed pairwise comparison tests to quantify differences in skull shape between delimited species in the R package MORPHO 2.4 (Schlager 2016) employing the Procrustes Distances, 1000 permutations, and a significance level ofp ≤ 0.05. Differences in the CS between sexes and delimitated species were assessed in the R package RRPP (Collyer and Adams 2018) using an OLS estimation method and the same resampling procedure described above.
To evaluate cranial shape differences between putative species included in the morphometric analyses we performed a canonical variates analysis (CVA). Because CVA cannot be performed when the number of variables is greater than the sample size per group (Kovarovic et al. 2011), we only used the shape information from the first 10 principal components (90% and 92% of the variance explained for dorsal and ventral views, respectively). Finally, we performed Linear discriminant analysis (LDA) to assess the discrimination of individuals against the different delimited species based on skull shape. A cross-validation procedure from the CVA scores was employed to obtain the correct discrimination rate, which allowed us to verify the effectiveness of the LDA to discriminate taxa into their correct groups. For these analyses, we used the R packages geomorph and MASS 7.3-51.4 (Ripley et al. 2020).