Population structure: genetic differentiation and disparity
To visualize population clustering, we used principal component analyses (PCA), starting from genotype-frequency-based covariance matrices (pcangsd v.0.99, Meisner & Albrechtsen 2018). We further visualized coancestry between individuals for each species using heatmaps (GPLOTS v.3.0.1.1; Warnes et al. 2020).
We used two measures of genetic distance to test whether populations were significantly different from each other. First, we converted pcangsd-derived covariance matrices to distance matrices (dist.from.cov , rwc v.1.11; Hanks 2018). Second, we calculated pairwise genetic distances between all individuals of each species (ngsDist). We then tested for significant differences in genetic diversity between populations of the same species (adonis , VEGAN; Oksanen et al. 2019). We used pairwise.adonis (with corrected p-value estimation) as post-hoc test to assess which populations were significantly different. In order to test whether certain populations were more disparate (genetically variable) than others, we usedbetadisper and TukeyHSD as post-hoc test (results presented in SI).
In order to assess population genetic differentiation, we calculatedFST values on genotype likelihoods starting from unfolded, pairwise SFS (core Merianieae) or folded SFS (Ad. adscendens ) using realSFS in ANGSD.
We further calculated the admixture coefficient for each individual by estimating the likelihood of genetic clustering in the data (NGSadmix, Skotte et al. 2013). We randomly selected only one SNP per locus and used ten random initializations to estimate admixture from K = 1 to K = n+1 ancestral populations, n being the total number of sampled localities in each species. We compared the rate of change in the log-likelihood of different successive K-values (Evanno 2005) to select the K-value best describing clustering in the data and used bar plots for visualization.