Phenotypic divergence
Total length – Model comparisons showed that TL varied across sites rather than habitats (Table 2). This pattern was predominantly driven by stickleback from HS2 site being shorter than stickleback from other sites (Figure 1). TL was negatively correlated with both temperature (Table 3) and bird density (Table 3), but the model with temperature as a predictor fit the data better (i.e., had a lower DIC; Table 3). All traits, except plate number and GRN, were positively correlated with TL (Table 2), and all results presented hereafter refer to effects on size-corrected traits.
Defence traits - There were no sex differences in the relative length of either dorsal or pelvic spines, but males had more armour plates than females (Table 2). Model comparisons suggested that defence traits tended to vary according to habitat rather than site, although there was only very weak statistical support for this effect (Table 2, Figure 1). Whilst none of the ecological variables predicted relative length of DS1 and armour plate number (Table 3), relatively length of DS2 increased as the density of piscivorous birds increased (Table 3) and relative length of PS increased in deeper water (Table 3, Figure 2). However, statistical support for these environmental associations was weak (the ΔDIC to the null model was within 2, and the lower credible interval of the posterior distribution of the linear coefficient was only just above zero; Table 3).
Individuals with relatively longer DS1 had correspondingly longer DS2 and PS (pairwise phenotypic covariances (CoV), posterior mode and 95%CI: DS1:DS2 = 0.224(0.165, 0.289); DS1:PS = 0.133(0.077, 0.190); DS2:PS = 0.122(0.070, 0.190)), indicating that spine traits covaried at the individual level. However, relative length of spines did not seem to covary with armour plate number (see Table S2). There was no evidence for phenotypic covariance across the habitats between any of the defence traits, suggesting that spatial divergence in defence traits was not correlated at the habitat level (see Table S2). However, to increase statistical power for detecting divergence in phenotypic covariances, greater level of replication is likely needed.
Trophic traits - Males had relatively longer GRL2 and GRL3, and relatively more gill rakers (GRN), than females (Table 2), but there were no sex differences in GRW (Table 2). Model comparisons showed that gill raker traits varied according to site rather than habitat (Table 2), but this variation was not associated statistically with any of the ecological variables (Table 3). Specifically, stickleback from the CS site had relatively longer and fewer gill rakers than stickleback from other sites, and stickleback from site CS and DN had fewer gill rakers than stickleback from all other sites (Figure 1). There was evidence for phenotypic covariances at the residual (i.e., individual) level for some gill raker traits. For instance, individuals with relatively longer GRL2 had correspondingly longer GRL3 (CoV, posterior mode and 95%CI: GRL2:GRL3 = 0.288(0.188, 0.403)). Furthermore, individuals with more gill rakers also had narrower GRW (CoV, posterior mode and 95%CI: GRW:GRN = -0.184(-0.311, -0.052))). There was no evidence for phenotypic covariances at the site level between any of the gill raker traits, suggesting that spatial divergence in gill raker characteristics was not correlated (see Table S2).
Gut length - Males had relatively shorter guts than females (Table 2). Model comparisons suggested that relative gut length varied among sites (Figure 1, Table 2), but this variation was not associated to any of the ecological variables (Table 3).