Drivers of adaptive genetic variation
To evaluate the relative contributions of demographic history, environment, and geography to shaping the genetic variation ofScutellaria in Taiwan, partial redundancy analysis (pRDA) was performed. First, the geographic coordinates were converted into dbMEMs, and the first three axes were retained as proxies of geographic structure. Second, PCA was performed using the allele frequencies among populations, and PC1 and PC2 were retained as proxies of demographic history. Variation partitioning was carried out using the allele frequencies of each population as responses and either of the groups of predictors, including (1) geographic structure, (2) demographic history, and (3) the eight environmental factors (Fig. 3C) we selected by VIF as explanatory variables, while conditioning the remaining predictors. The variation explained by the full model and each testable portion was assessed and tested for significance. Significant predictors were selected by forward selection. The pRDA analyses, dbMEM, and forward selection were conducted using the R package vegan (Oksanen et al., 2007).