Sample design and population clustering
For landscape-scale analyses it is important to not confound diversity
within meta-populations with divergence among historically isolated
lineages, as have previously been identified in each of the species
considered here (Suppl. Mat. S1). To visualize the population structure
and divergence among samples using the SNP data, we first performed a
principal coordinate analysis (PCoA) based on the genetic distance
matrix between individuals using Euclidean distance to identify discrete
clusters. We then ran two different analyses to verify clusters and test
for admixture among them — FastStructure (Raj, Stephens, &
Pritchard, 2014) and conStruct (Bradburd, Coop, & Ralph, 2018).FastStructure is a model-based clustering method that looks for
the probability of an individual belonging to a cluster, considering the
population structure, and was calculated with model complexity (K)
between 1-7. The admixture proportions were visualized with aDistruct plot. ConStruct infers continuous and discrete
patterns of population structure by estimating ancestry proportions for
each sampled individual from two-dimensional population layers, where
within each layer a rate at which relatedness decays with distance is
estimated (Bradburd et al., 2018). In essence, this approach has IBD as
a null-model, against which isolated populations are inferred, whereasFastStructure (and related methods) assumes panmixia within
genetic clusters. For ConStruct we used spatial data (considering
Isolation by distance), with a K between 1 and 7, and compared the
predictive performance of the models by running cross-validation with 50
replicates. From here on, we will refer to these populations within
described species with strong structuring as ‘lineages’, even if low
admixture is observed between them in one or more of the above analyses
(Table 1). These groups are well supported in previous studies using
mitochondrial DNA and, in most cases, multilocus sequencing (more
details in Suppl. Mat. S1).