Intraspecific adaptive variation
To test for adaptive differences among populations of NIDGS and SIDGS, we performed partial redundancy analyses (pRDA) on each dataset separately. The environmental variables that we used differed among datasets, resulting in 19 variables for NIDGS, while only four variables were kept for SIDGS due to a large number of correlations (Figures S3B and S3C, respectively, Supporting information).
For NIDGS, PCA of the genomic data showed that none of the PCs have eigenvalues greater than random, suggesting K = 1 (Figure S7A, Supporting information). However, this result does not mean there is no genetic structure in the data, but rather that it might not be particularly strong. Thus, because we have evidence from previous studies that NIDGS are divided in at least two groups (Garner et al. 2005; Hoisington-Lopez et al. 2012; Zero et al.2017), we decided to condition the pRDA using PC1 (i.e. assumingK = 2). From the 19 environmental variables used, we excluded four (BIO_1, BIO_12, BIO_15 and LF_VA) due to high VIF. We then obtained an adjusted r2 = 0.06 across 15 axes. The ANOVA showed that the pRDA was significant at p < 0.001, and that the first three axes were significant at p< 0.001. The first pRDA axis explained 3.3% of the variance, while pRDA2 and pRDA3 explained 3.0% and 2.7%, respectively. The remaining axes combined (pRDA4-15) explained 14.0% of the remaining variance. From this analysis, three populations were the most distinct in association to particular environmental variables: Lower Butter (LB) mostly associated with Local Ridge in Plain (LF_LRP) and slope, Rocky Top (RT) mostly associated with Ridges and Peaks (LF_RP), and Tamarack (TA) mostly associated with soil particle size (soilPartSize) (Figures 3A and 3B).
For SIDGS, the PCA on the genomic data showed that PC1 had eigenvalues greater than random, suggesting K = 2 (Figure S7B, Supporting information). From the four environmental variables used, all were kept for the final pRDA, which had an adjusted r2=0.01. The ANOVA showed that the pRDA was significant at p< 0.001, and that only the first axis was significant atp < 0.001, explaining 3.7% of the variance. The remaining three axes combined (pRDA2-4) explained 6.7% of the remaining variance. From this analysis, Paddock (PA) was the most distinct population, correlating mostly with grassland areas (LC_GH), as opposed to the other populations which are more associated with shrub/scrubland, and Annual Mean Temperature and Isothermality (Figure 3C).