Despite accumulating evidence that evolution can be predictable, studies quantifying the 2 predictability of evolution remain rare. Here, we measured the predictability of genome-3 wide evolutionary changes associated with a recent host shift in the Melissa blue butterfly 4 (Lycaeides melissa). We asked whether and to what extent genome-wide patterns of evolu-5 tionary change in nature could be predicted (1) by comparisons among instances of repeated 6 evolution, and (2) from SNP × performance associations in a lab experiment. We delineated 7 the genetic loci (SNPs) most strongly associated with host use in two L. melissa lineages 8 that colonized alfalfa. Whereas most SNPs were strongly associated with host use in none 9 or one of these lineages, we detected a ∼two-fold excess of SNPs associated with host use in 10 both lineages. Similarly, we found that host-associated SNPs in nature could also be partially 11 predicted from SNP × performance (survival and weight) associations in a lab rearing exper-12 iment. But the extent of overlap, and thus degree of predictability, was somewhat reduced.
13Although we were able to predict (to a modest extent) the SNPs most strongly associated 14 with host use in nature (in terms of parallelism and from the experiment), we had little to 15 no ability to predict the direction of evolutionary change during the colonization of alfalfa.
16Our results show that different aspects of evolution associated with recent adaptation can be 17 more or less predictable, and highlight how stochastic and deterministic processes interact 18 to drive patterns of genome-wide evolutionary change.