b2. Frequency-dependent selection: are allele effects inconsistent across lakes?
We found some cases where the allele effects are inconsistent across lakes (significant MHC allele × site interaction effects), suggesting there were potentially Red Queen style co-evolutionary arms race between stickleback population and parasites. We obtained 788 generalized linear models for each moderately prevalent MHC-parasite combinations which were present in more than one site. 53 models had a significant interaction term but an insignificant MHC term (11 of them involved parasites with birds as final hosts), suggesting inconsistent allelic effects across sites. For example, the allele prot_110 confers susceptibility (higher infection) in McCreight Stream, but has no effect in McCreight Lake (Fig. 4C). Although none of the models were significant after correction for multiple-comparison with BH method, the number of significant models was greater than null expectation with a significance level of 0.05 (792 * 0.05 = 39.6), so some of the model results were truly significant despite the issue of multiple comparison. 66 models had a significant MHC term but an insignificant interaction term (13 of them involved parasites with birds as final hosts) (Fig. 4B), suggesting that those alleles had consistent effects on the focal parasite across sites, and thus did not fit localized Red Queen race dynamics. Parasites with birds as final hosts were not more likely to be found in the models with significant MHC term than in the models with significant interaction term (Chi-squared test, X2=0.004, p=0.95).