Bayesian sample assignment
To assign individuals to genetic clusters, a two-step approach was
taken. First, genotypes for all North American samples were added to the
dataset presented in Andersen et al. (2019a), and the probability of
assignment (Q ) of sampled individuals to one of two distinct
genetic clusters (K ) representing either pure winter moth or pure
Bruce spanworm was calculated using Structure v.2.3.2 (Falush,
Stephens, & Pritchard 2003; Pritchard, Stephens, & Donnelly, 2000).
These analyses were based on the analysis of 12 polymorphic
microsatellite loci that co-amplify between the two species, and ten
independent analyses were run using the admixture model, correlated
allele frequencies, and default settings, with random starting values,
runtimes of 1,000,000 generations, and burn-in periods of 100,000
generations. Results were then summarized across runs using
Clumpak (Kopelman, Mayzel, Jakobsson, Rosenberg, & Mayrose,
2015), and hybrid individuals were identified as those receiving scores
of Q < 0.75 to both the Bruce spanworm and the winter
moth genetic clusters. Hybrid individuals were then removed from the
dataset, and the filtered dataset (≥ 20 polymorphic loci for each
individual) was used to estimate values of Q for all individuals
(both native and introduced regions) for values of K= 2 throughK = 14 in Structure using the run parameters described
above. To determine the optimal number of clusters present in the
dataset, the ∆K statistic (Evanno, Regnaut, & Goudet, 2005) was
calculated in StructureHarvester (Earl & vonHoldt, 2012), and
independent runs were again summarized for major and minor partition
schemes using Clumpak.
For each value for ∆K with a distinct peak representing a
positive rate of change identified using StructureHarvester,
the summarized ‘popfile’ of cluster membership coefficients for the
major mode calculated in Clumpak was used to create a distance
matrix using the ‘dist’ function in R v. 4.0.0 (R Core Team, 2020). The
resulting matrices were then used to calculate ‘NeighborNet’ networks
using SPLITSTREE v.4.14.2 (Huson & Bryant, 2006), and the outputs were
examined to identify geographic patterns.