Research methods and Comparative examination of pinniped craniofacial musculature and its role in aquatic feeding

In situ measurements—maximum length (straight rostral-caudal distance from rostral tip of skull to caudal edge of occipital condyles), width (straight medio-lateral distance from lateral edges of zygomatic arch), and depth (straight dorso-ventral distance from sagittal crest to auditory bullae)—were measured for each muscle when possible. We used scaled photographs to take ex situ measurements of the maximum length and width of each muscle in two-dimensions in ImageJ. Muscle area was calculated from tracing the perimeter of the muscle in scaled photographs. We then calculated the muscle-to-skull area ratio (MSR) for each muscle, which was obtained by dividing each muscle area by the skull area to standardize for head size. We calculated mean MSR for each craniofacial muscle group within each species (e.g., facial expression, mastication, tongue, hyoid, and soft palate) for interspecific comparisons. We ran linear models to compare the relationship between relative muscle size (MSR), species, and feeding strategy and determined significance using ANOVAs (car package; Fox & Weisberg 2011; Bates et al. 2015). We examined the relationship between muscle size and species with species as the predictor variable and between muscle size and feeding strategy with feeding strategy as the predictor variable. We used least-square means to perform Tukey post-hoc pairwise contrasts between each significant predictor variable (lsmeans package; Lenth, 2016). Residual plots for each model were examined for deviations from normality and homoscedasticity. When heteroscedasticity was observed, data were log-transformed. Additionally, we examined variability in MSR for each muscle and species by quantifying the coefficient of variation [CV, standard deviation (s.d.)/mean] for each muscle. The CV measures variation in a trait. A low CV (values close to 0) indicates stereotypy, while a high CV (values close to 1) indicates high variability (Gerhardt, 1991; Wainwright et al., 2008). All statistical analyses were conducted in R v. 3.5.3 (R Core Team, 2019).