RESULTS
In the mark-recapture experiment conducted on the mainland, we marked and released 1714 snails in total. We re-encountered 337 of the snails released in the field (257 living and 80 dead snails). We recaptured some snails more than once, and the maximum number was four times. Most of the snails on the mainland had bright shells, and the distribution of the shell colour was unimodal (Figure 2A), while the shell colour distribution of the island was bimodal (Figure 2B).
Some recaptured dead snails had evidence of shell crushing by predators (Figure 3). There were 37 and 8 such snails on the mainland and island, respectively. More snails were predated on the mainland than the island (Figure S1; Fisher-exact test, odds = 9.46, P < 0.001). In the GLMM analysis, the number of predated snails was not significantly correlated with the shell colour in both study sites (Table 1; z-value = 2.86, P > 0.05). However, the number of predated snails was correlated with growth stage on the mainland (Table 1). This result implies that predators attacked more adult snails than juvenile ones on the mainland. We conducted trail-camera experiments to identify the predators and filmed eight mammals, four birds, and three reptiles within the monitoring snails’ camera range (Table S1). Of these, the only large Japanese field mouseApodemus speciosus was attacking the snails (Figure 3C; Movies S1-S3 in Dryad).
The P-spline regression showed that the survival rates excluding predation effects of mainland adult snails were not bimodal but monomodal in terms of shell colour (Figure 4A). The adult survival rate was estimated using a second-order polynomial function of shell colour in the MCMC simulations based on the Bayesian framework, and the median of the nonlinear selection gradient \(\gamma\) was -1.05 (Table 2). The negative value of \(\gamma\) supports that the P-spline regression curve was concave upward. Although the 95% Bayesian confidence interval (BCI) of \(\gamma\) was [-2.47, 0.33], 94.2% of \(\gamma\) values estimated in the simulation runs were negative. In contrast, the survival rates of mainland juvenile snails were not monomodal (Figure 4A), and the 95% BCI of \(\gamma\) included 0 (Table 2). The other parameters estimated from the Bayesian method are shown in Tables S2-S4.