The PCA showed that E. coccineum was divided in two major genetic groups: North and Center-South (Fig. 1b). The northern more locations were retrieved separated from those corresponding to the Center-South along the first axis of the PCA (PC1, representing 21.54% of the genetic variance), while samples from the Center and the South were separated along the second axis of the PCA (PC2, representing 5.82% of the genetic variance).
Results from the model-based Bayesian clustering approach showed a K = 2 (highest LnP(D) and ΔK) as optimal and K = 4 as a potential secondary level of substructure (Supplementary Fig. 1a-b). The STRUCTURE clustering (K = 2) revealed that all individuals sampled from the North were assigned to the blue genetic cluster while all individuals sampled from Center and South were assigned to the orange genetic cluster (Fig. 2a). In North and South samples, almost no admixture was observed. Individuals from PM, ChlN and ChlS (Center) were majorly assigned to the orange genetic cluster but exhibiting low levels of admixture from the blue genetic cluster. The level of assignation to the blue genetic cluster tends to recede across latitude (from 20% of the genome assigned to the blue genetic cluster in PM to less than 10% in ChlS; Fig. 2a). The STRUCTURE clustering using K = 4 showed two new genetic clusters within the Center-South part of the species range (Fig. 2b). Individuals from the South were assigned to the orange genetic cluster, individuals from Pumalín (PU, Center) were mostly assigned to the new purple genetic cluster and the rest of the individuals from the Center (PM, ChlN and ChlS) were admixed between the orange genetic cluster and the newly detected green genetic cluster (Fig. 2b).
The dendogram reconstructed with the Nei’s distances tree calculated with the PG_dataset of 2.155 SNPs allowed to separate the North and Center-South samples (Fig. 3a), being concurrent with the results obtained by the PCA and the STRUCTURE clustering analyses. When compared with the results obtained with the whole data set (i.e., PG_dataset), the dendogram retrieved with the 59 outlier loci putatively under selection (see below for more details) revealed a much deeper separation between the North and Center-South samples; yet, most sub clustering within major genetic groups was lost (Fig. 3b).
Genome scan for outlier loci detection
Whithin the 2,155 loci included in the PG_dataset, PCAdapt detected 132 outlier loci and BAYESCAN 89 outlier loci. The two methods concurr with the detection of 59 loci (2.73%) that were then identified as potentially adaptive loci in E. coccineum .
The gradient forest revealed four environmental factors with the highest impact on E. coccineum populations (i.e., estimated as the variables presenting the highest values of R2 weighted importance), which correspond to: Precipitation of driest month (bio 14), Mean temperature of wettest quarter (bio 8), Elevation, and the Mean temperature of driest quarter (bio 9) (Fig. 4). The value of the R2 weighted importance for bio 14 (R2 weighted importance > 0.05) was at least two-times higher than those observed for the other variables.
Isolation by distance and isolation by environment
The full RDA model using the PG_datset explained 59.4 % of the total genetic variation and supported an influence of environmental (Env), geographic (Geo) and coancestry (Anc) variables in shaping allelic variation (Env: adjusted R2 = 0.420, p < 0.001; Geo: adjusted R2 = 0.226, p < 0.001; Anc: adjusted R2 = 0.105, p < 0.001; Table 3). With the partial db_RDA, the contribution of the environmental variables to genetic divergence was much higher than the contribution of the geographic distance or the co-ancestry (Env | Geo+Anc: adjusted R2 = 0.288; Geo | Env+Anc: adjusted R2 = 0.118; Anc | Env+Geo: adjusted R2 = 0.011; Table 3). In total, 0.178 of the explained variation was confounded between the effects of IBE, IBD and IBA, and 0.405 of the variation remained unexplained (Table 3). Very similar results were obtained for the subset of 59 loci potentially influenced by selection (i.e., outlier loci). The partial db-RDA highlights the important contribution of the environmental variables to potential local adaptation in E. coccineum (Env | Geo+Anc: adjusted R2 = 0.244; Table 3).
Table 3 : Redundancy analysis (RDA) partitioning among-population genetic variation in Embothrium coccineum into three components: environmental (Env), geographic (Geo) and North / Center-South coancestry (Anc).