Population genetic analyses
We conducted hierarchical clustering analyses of SNPs for the 16 sampled localities in the population genetic dataset using principal components analysis (PCA) and the program Structure 2.3.4 (Pritchard et al.2000). Principal Component Analyses (PCAs) were conducted usingglPca in adegenet (Jombart 2008) implemented in R 3.6.1 (R Core Team 2019), and plotted with ggplot2 (Wickham 2009). Structure was set to use the admixture model and correlated allele frequencies and was run with and without using sampling locations as a prior (locpriorvs. nolocprior ). We tested K = 1-20 with 20 independent replicates per value of K . Each value of K ran for 400,000 MCMC reps with a burn-in period of 200,000 and we averaged runs using CLUMPAK v1.1 (Kopelman et al. 2015). Following the recommendations of Janes et al. (2017), we considered multiple metrics when determining the optimal value of K , including comparison to the PCA, LnPr(X |K ) (Pritchardet al. 2000), ΔK (Evanno et al. 2005), and the statistics proposed by Puechmaille (2016). We calculated the latter with StructureSelector (Li & Liu 2017) using a population map corresponding to collection localities, and a threshold for cluster placement set to 0.5.
SNP pairwise FST was calculated in R using StAMPP (Pembleton et al. 2013) with 1,000 bootstrap permutations and a Benjamini-Hochberg p-value correction. Expected and observed heterozygosity (He and Ho , respectively) were calculated in dartR (Gruber & Georges, 2019). Isolation by distance (IBD) analysis using Euclidean distance and a Mantel test with 10,000 permutations was conducted using the R packages sna (Butts 2019), geosphere (Hijmans 2019), and adegenet (Jombart 2008). Due to potentially different biological scenarios impacting the correlation between genetic and geographic distance (e.g. a single genetic cline versus two or more distinct clines, Meirmans 2012; Teskeet al. 2018; Maitra et al. 2019), the densities between points were visualized with a kernel density estimation function using the package MASS (Venables & Ripley 2002).
PopART (Leigh & Bryant 2015) was used to construct a minimum spanning network of COI haplotypes.