3.1 Variation in Phenotypic Traits
The ANOVA results showed that in each accession, the probability of difference was high for all measured traits (LVN, PL, LA, LLE, LW, FV and FLD) (P≤0.01, Table 2). The leaf traits reported that the LVN value ranged between 4.46±0.65 cm (YM) and 7.83±1.11 cm (YJ). The PL value ranged from 7.22 mm to 17.18 mm, the highest PL values were observed in XY (7.22±1.53mm) and the lowest values, in YJ (17.18±3.29mm). A range of 4.67±3.16cm2 (RH) to 21.53±6.07cm2 (YJ) was found for LA. LLE varied from 2.97±0.34cm (DC) to 8.68±0.34cm (YJ), while LW ranged from 1.78±0.60cm (RH) to 3.66±0.69cm (YJ) (Table 3). The LI value means there was no remarkable differences among all populations. (Table 2). The mean value of LI was 2.15, with a variation ranged from 1.10±0.23 for DC to 2.45±0.32 for ML.
The fruit traits showed that the highest value for FV (43.59±4.67mm) and FLD (46.23±6.42mm) were determined in the LC population, while the ML presented the lowest value for FV (28.39±3.33mm) and the DC for FLD (26.44±3.31mm). The FI value had a average of 1.05, ranging between 0.89±0.09 and 1.39±0.21 for the ML and DC population, respectively. And average FI value did not differ among populations.
The CV for nine traits in the tested Docynia delavayi populations showed (Table 3) that PL and LA exhibited relatively high variation (the CV were 36.85% and 43.71%, respectively). In all phenotypic traits of Docynia delavayi, the LA had the largest CV (43.71%), while FV had the smallest CV (12.17%). The CV value of traits was in the order of LA (43.71%) ＞ PL (36.85%) ＞ LW (25.33%) ＞ LLE (24.47%) ＞LVN (17.50%) ＞ FLD (13.30%) ＞ FV (12.17%). On average, populations showed higher variation in leaf traits than fruits traits (the CV were 28.13% and 12.48%, respectively)
3.2 Correlation Analysis
The correlations between features were found at p ≤ 0.05, and the Pearson correlation coefficients were reported between pairs of traits shown that LVN observed significant and positive correlations with LA, PL, LLE, LW, while PL and LA had the most strongly correlation with LVN (R = 0.816, p＜0.01). Significant positive correlations were also found between PL and LVN (R = 0.816, p＜0.01), LA (R = 0.752, p＜0.01), LLE (R = 0.682, p＜0.01), LW (R = 0.734, p＜0.01), and FV (R = 0.681, p＜0.01); between LA and LLE (R = 0.974, p＜0.01), LW (R = 0.927, p＜0.01), FV (R = 0.735, p＜0.01) and FLD (R = 0.562, p＜0.05); between LLE and LW (R = 0.845, p＜0.01), LI (R = 0.639, p＜0.01), FV (R = 0.679, p＜0.01) and FLD (R = 0.612, p＜0.01); and between FV and FLD (R = 0.767, p＜0.01). FLD was negatively correlated with FI (R = - 0.658, p＜0.01) (Table 4).
Neverless, a significant negative correlation was found between longitude and LVN (R = - 0.543, p＜0.05), LA (R = - 0.789, p＜0.01), LLE (R = - 0.788, p＜0.01), LW (R = - 0.696, p＜0.01), and FV (R = - 0.465, p＜0.01); and between latitude and LVN (R = - 0.454, p＜0.01), LA (R = - 0.660, p＜0.01), LLE (R = - 0.644, p＜0.01), LW (R = - 0.710, p＜0.01), FV (R = - 0.735, p＜0.01), and FLD (R = - 0.488, p＜0.05) (Table 4). In a word, the result showed that traits had no correlated with altitude, but the negative and significant correlated with longitude and latitude. The internal links between individual and geographical conditions was to be further research.
3.3 Principal Component Analysis in Traits
The results of the PCA analysis indicated that over 76% of the observed variance could be explained by the first three components (Table 6). The total variance explained by PC1 axis and PC2 axis was 51.428% and 15.190%, respectively. The Rotated Component Matrix of factorial analysis of Docynia delavayi showed that there is a clear separation between leaf traits, fruit traits, and geographical conditions (Table 6). PC1 represents mainly LW (0.940), LA (0.914), LLE (0.859), PL (0.843) and LVN (0.711). PC2 explains mainly FI (-0.835), FLD (0.721) and FV (0.257). Longitude (0.339) and Latitude (0.216) occupied a certain position in PC3.
The first principal component largely represents these interrelated leaf traits, so it is the most important genetic factor, and it accounts for 51.428% of the total variance. (Table 7). The susceptibility of FV and FLD was positively correlated with the second principal component, and negatively correlated with FI (-0.835). Therefore, PC2 represents the second in importance genetic element, accounting for 15.19% of the overall variance. The third principal component is negatively correlated with longitude and latitude. (r = 0.339 and 0.216, respectively). Therefore, this principal component explains 9.796% of the total variance.
3.4 Clustering Analysis
The Ward clustering of the six most important principal components of nine standardized phenotypic traits resulted in a tree showing that four groups (clusters) were distinguished (Figure 2). The Group I consist of three populations (LC, MS, and YJ). Only one population, DC, belongs to Group II. Group Ⅲ consists of three populations, including CN, LL, and MJ. The ten populations in Group IV represent different geographic sites (YB, RH, NL, LJ, MY, YL, XY, CX, ML, and YM).