Outlier analysis

Two independent methods were applied to identify putative loci under selection. The first program used was LOSITAN (Antao et al., 2008) which utilizes the method of Beaumont & Nichols (1996a) to identify loci under selection based on the joint distributions of expected heterozygosity and F_ST under an island model of migration. The following settings were used for the SNP and the microsatellite datasets. The neutral mean setting was selected in which during an initial run (100,000 simulations), a candidate subset of selected loci (outside the 95% confidence interval) were identified and removed. Then the distribution of neutral F_ST was computed using 100,000 simulations and a bisection approximation algorithm (Antao et al., 2008), with the following options, force mean F_ST, infinite alleles mutation model, and a confidence interval 0.99. A FDR <0.1 correction for multiple testing was applied. Loci outside the upper and lower confidence areas were identified as candidates affected by positive and balancing selection, respectively (Table S4). All geographic regions were analyzed together. Outliers identified as being under balancing selection were not considered as these are more likely to be false positives (Lotterhos & Whitlock, 2014). The positive outlier loci (p < 0.01) were blasted against the NCBI nr, UniProt, and Trembl databases with parameters of expected value = 0.00001, gap opening penalty = 11, gap extension penalty = 1, length of initial exact match (word size) = 6 and scoring matrix = BLOSUM62 using BLASTX 2.2.32 + (Altschul et al., 1997).

The Stacks exported Genepop dataset was also reformatted with PGDSpider version 2.0.5.2 (Lischer & Excoffier, 2012) to a GESTE file. The method of Foll & Gaggiotti (2008) was performed using Bayescan 2.0 (http://www-leca.ujf-grenoble.fr/logiciels.html). For each locus, the probability of it being under selection was inferred using the Bayes factor (BF). Based on Jeffreys’ (1961) scale of evidence, a log10 BF of 1.5–2.0 is interpreted as “strong evidence” for departure from neutrality at that locus and corresponds to a posterior probability between 0.97–0.99. For our analysis, the estimation of model parameters was set as 20 pilot runs of 5,000 iterations each, followed by 50,000 iterations.