Multivariate statistics
SNP metrics were recalculated in R for post-filtered data using the ‘DArTR package’ (Gruber, Unmack, Berry, & Georges, 2018). PCoAs were constructed from individual larval SNP genotypes with Gower PCoA ordination and Euclidean distances using the packages ‘Adegenet’ and ‘DArTR’ (Gruber et al., 2018; Jombart, 2008). Clusters of genetically related individuals were inferred using Discriminant Analysis of Principal Components (Jombart, Devillard, & Balloux, 2010) constructed with and without the between- and within- group variance components from PCoA outputs. Multiple loading thresholds were explored, including the default value of the third quartile (75%) of x-values (Grünwald, Kamvar, & Everhart, Jombart & Collins, 2015). Diversity metrics were calculated using ‘DArTR’ following (Sherwin, Chao, Jost, & Smouse, 2017). These analyses are similar to classic population genetic analyses that measure differentiation via Fst, in which high allele frequency differentiation suggests population differentiation although they can be fraught with false positives (Rajora, 2019).
Given the influence of close relationships in STRUCTURE analyses (Anderson & Dunham, 2008), DAPC was used as an alternative method to infer admixture (Grünwald et al., 2010). Hence, “true k ” was calculated based on Bayesian Information Criteria (BIC, Jombart et al., 2010), in which the largest drop in BIC was at k = 5.