Molecular data generation and processing
DNA extraction, mitochondrial COI gene amplification, ddRAD (Peterson et al. 2012) and two-enzyme GBS (Poland et al.2012) library preparation and sequencing, and initial mitochondrial and SNP data processing follows Campbell et al. (2017, 2019) and so are not described in detail here. SNPs were genotyped de novousing the Stacks v. 2.3 pipeline (Catchen et al. 2011; Rochetteet al. 2019) with default parameter settings except for the following: the n parameter, which controls the number of mismatches tolerated per locus during catalog construction incstacks , was set to 2 instead of 1; we only retained loci that were found in 80% of any single population (the “r80” principle of Paris et al. 2017); and a single, random SNP from each locus was output during final processing in the populations program of Stacks to reduce genomic linkage. We conducted additional SNP filtering in vcftools 0.1.14 (Danecek et al. 2011) to retain only loci with a minimum minor allele frequency of 3% and to further reduce the global missing data per locus to a maximum of 20%.