2.5. Discriminant analysis of principal components method

We also assessed sub-structure using the multivariate Discriminant Analysis of Principal Components (DAPC) method [27], which does not rely on maximizing linkage disequilibrium between loci and Hardy–Weinberg equilibrium. We performed DAPC in the adegenet package [28] in the R environment v. 3.2.2 [26]. The find.clusters function was used first to assign individual samples to proposed sub-populations. We retained all 80 principal components for this initial step. We then applied the DAPC analysis function to the number of clusters exhibiting the lowest Bayesian Information Criterion (BIC) to produce a scatterplot, retaining 40 principal components which accounted for 90% of the observed variance, and all linear discriminants.