RESULTS
Morphological differences between populations
The Principal Component Analysis reveals no visually obvious shape differentiation among populations along the two main axes of variation. The first two principal components together account for 35% of the total shape variation (PC1 = 24.58%, PC2 = 11.58%) (Supplementary Figure 3). An example of shape changes along the first principal component axis with allometric shape changes put together with other unrelated shape changes is illustrated in Figure 1.
Figure 1 : Shape changes associated with the First Principal Component (above) and with allometry (below). Spheres are the landmarks used in this study. Warmer colours represent higher landmark variation between mean shapes. Vectors show direction and magnitude of shape variation.
Despite the low differentiation of populations within a PCA, the Procrustes ANOVA demonstrates that at least one of the five populations differ in shape (F(4,91) = 3.125, R-squared = 0.121, p < 0.001) (Table 1). The post-hoc pairwise comparisons between the shape means of each population reveal significant differences in shape among all populations. Intriguingly, the only sex-biased population (Kimberley, which consisted mostly of males) shows no clear difference with the remaining four populations (Figure 2). Groote Eylandt specimens show a generally narrower skull as revealed by the greater interlandmark distances in the zygomatic arches. The four mainland populations have shorter muzzles than the Groote specimens, as revealed by the shortening of the nasal and frontal areas. Northern Territory specimens display elongated frontal bones. Pilbara specimens exhibit an expansion of the braincase size and shortest muzzles relative to the rest of the skull. Shape disparity (shape variance) differences between island and mainland specimens were not significant (p = 0.604).
Figure 2 : Pairwise comparisons between means of each population and visualization of interlandmark variation between populations mean shapes. Warmer colours represent higher landmark variation between mean shapes. Top left, comparisons of absolute mean shape of each population; bottom right, comparisons of size-corrected mean shapes of each population. Map on bottom left shows all specimen locations used in this study.
Sexual dimorphism and allometry
We first confirmed that known sexual dimorphism in animal weight and skeletal measurements (Oakwood, 1997; Schmitt et al., 1989) are reflected in cranial size (Supplementary Figure 4) and shape (Table 1 and Supplementary Figure 5). Size differences are significant between males and females (F(1,90) = 23.9, R-squared = 0.21, p < 0.001), island and mainland populations (F(1,99) = 14.15, R-squared = 0.125, p < 0.001) and populations (F(4,91) = 8.361, R-squared = 0.269, p < 0.001).
Females and males show the same homogenous allometric relationship with no significant difference between slopes (p = 0.087), such that small males and large females overlap on the allometric slope (see Supplementary Figure 6). Just over half of sexual shape differences are due to size (R-squared for shape between sexes = 0.102), with a component of shape changes not due to size (R-squared of allometry-adjusted shape differences between sexes = 0.043) (Table 1). In other words, although there is some non-size-related variation between sexes, small males and large females are similarly shaped according to their common size. We therefore included individuals of both sexes in our analyses of population differences and other pertinent analyses.
In the full dataset, and among all variables tested, size manifests as the strongest determinant of shape variation in northern quolls (p < 0.001), accounting for 15.1 % of the total shape variation. A Homogeneity of Slopes Test suggests no significant differences among allometric slopes of each population (p = 0.203) (Table 1 and Figure 3), meaning that the hypothesis of populations following the same allometric slope is not rejected. Allometry-corrected shape analysis also reveals the shape differences between populations; Procrustes ANOVA performed on the residuals of allometry revealed significant differences between populations (F(4,91) = 3.419, R-squared = 0.131, p < 0.001). Pairwise comparisons between these size-corrected shapes show similar significant differences between populations. Thus allometry does not play a role in differentiating the shape of populations.
Figure 3 : Allometry plot, centroid sizes (proxy for body size) versus shape scores obtained from the regression of shape on size (Drake & Klingenberg, 2008). Results of Homogeneity of Slopes Test for allometric slopes of populations are shown on Table 1.
Association of shape variation with geography, climate and size
Shape differences show significant differences along both latitudinal (F(1,87) = 4.051, R-squared = 0.044, p = 0.002) and longitudinal (F(1,87) = 3.023, R-squared = 0.034, p = 0.003) gradients on mainland specimens. Size is significantly different along the latitudinal gradient (F(1,87) = 8.833, R-squared = 0.092, p = 0.004; but not along the longitudinal gradient, p = 0.117). Temperature and precipitation contribute small, but significant, effects to shape (Temperature: F(1,87) = 2.006, R-squared = 0.023, p = 0.029; Precipitation: F(1,87) = 3.411, R-squared = 0.038, p = 0.002). Size differences are only significantly predicted by the effect of precipitation (F(1,87) = 8.236, R-squared = 0.086, p = 0.005), but not by the effect of temperature (p = 0.794).
We dissected the influence of size, geography and climate (precipitation + temperature) with a variation partitioning analysis (Figure 4). The full model [a + b + c + d + e + f + g] shows a significant effect of these three factors on cranial shape variation (F(13,75)= 3.136, adjusted R-squared = 0.24, p < 0.001). Climatic variables alone [c] do not explain any of the variation (p = 0.224), however, they contribute to the model when geography is considered jointly [f] (adjusted R-squared = 0.05). Pure geographical distances [a] explain 3 % (adjusted R-squared = 0.033) of the shape variation (F(10,75) = 1.371, p = 0.004). Finally, in accordance with our predictions, size alone [b] contributes mostly to the model by accounting for 17 % of the total cranial shape variation (F(1,75) = 17.539, adjusted R-squared = 0.165, p < 0.001).
Figure 4 : Schematic representation of the variation partitioning analysis (VARPART), which included effect of geography, size and combined climatic variables (precipitation and temperature) on cranial shape. The values shown in the diagram represent the individual fractions for each set. The outer numbers are the adjusted R-squared values of pure geography [a], pure size [b] and pure climate [c] and the inner values are the adjusted R-squared values of the interaction of the corresponding explanatory variables. The individual fraction for the interaction of all three variables [g] is negligible and not shown. The amount of unexplained shape by this model is depicted by the residuals (76%). Circle sizes are schematic and do not represent the amount of shape explained by the model.