3.5 Population genetic structure
Analysis of molecular variance (AMOVA) based on the AFLP dataset from 43 populations of P. villosa showed that the proportion of genetic variation among populations was 35.84%, while that within populations was 64.16%, and the value of average pairwiseF ST was 0.35841 (P < 0.001) (Table 4), which showed that the genetic variation of P. villosamainly occurred within populations. More interestingly, when aggregating populations into Groups 1 and 2, we found that 22.38% of the genetic variation occurred among populations within groups (F CT = 0.28501, P < 0.001), while most of genetic variation (56.14%) existed within populations (F ST = 0.43865, P < 0.001) (Table 4). Overall, the genetic variation at the population- and local geographic-scale was much higher than regionally in P. villosa . Additionally, the result of neutrality test suggested the value of Tajima’s D and Fu’s Fs was positive, but non-significant for all populations of P. villosa (Table 3).
The Mantel test revealed that there was a significant positive correlation between geographic distance and F STfor 43 populations (r = 0.282, P < 0.05) (Figure 1). Similarly, we detected a strong, significant, positive correlation between geographic distance and F ST for Group 1 (r = 0.622, P < 0.05) and a weak but significant positive correlation for Group 2 (r = 0.372, P < 0.05). Simultaneously, results from UPGMA tree, SplitsTree network, PCoA, and STRUCTURE suggested that 43 populations of P. villosawere divided into two groups, which were largely consistent with our assessment using SAMOVA (Figure 2 - 5).
3.6Distributional change of P. villosa
ENMs for P. villosa yielded relatively high AUC, demonstrating reliable model performance (AUC = 0.969, Figure S3). For the eleven non-biological variables used for modeling, the most significant factor for the spatial distribution pattern of P. villosa was altitude (Alt), followed by temperature annual range (bio 7) and precipitation of warmest quarter (bio 18), whose contribution rates were 40.0%, 17.2% and 16.7%, respectively (Table 5). In comparison with the LIG, we observed a contraction in highly suitable habitat during the LGM based on the MaxEnt models (Table 6 & Figure 6). Similarly, the spatial distribution of the present was continuously shrinking compared to the potential range during the LGM (Figure 7). The simulated distribution based on present climate data was mostly congruent with the actual distribution range of P. villosa , which was mainly distributed in the Inner Mongolia Plateau with an area of approximately 111.2450 × 104 km2 (Table 6 & Figure7). Simultaneously, we estimated the future changes in the potential spatial distribution under the RCP 2.6 and RCP 8.5 scenarios for the 2050 s and 2070 s. According to the future model predictions, the areas of suitable habitat is likely to remain stable under the climatic scenario of RCP 2.6 for the 2050s and 2070s, whereas there was an increase of highly suitable areas based on RCP 8.5 (Table 7 & Figure 8).
When we compared the niches of hypothesis of niche identity was rejected when the empirically observed value for D and/or I was significantly lower than the values expected from the pseudo-replicated data sets. Therefore, identity tests between two groups indicated that there was distinct niche differentiation (P < 0.01) (Figure 9). The niche of two groups differs mainly in that it was characterized by high elevation and temperature.