3.2. Genetic Diversity Analysis
At the species level, the values of PPL, He, and I were 100%, 0.333, and 0.504, respectively. At the population level, the results are listed as follows: PPL ranged from 53.19% (Xuanen, Hubei (HX)) to 78.72% (Shimen, Hupingshan, Hunan (HH)), with an average of 66.31%; He ranged from 0.177 (Shucheng, Wanfoshan, Anhui (AW)) to 0.288 (Huangshan, Anhui (AH)), with an average of 0.226; and I varied from 0.274 (AW) to 0.423 (AH), with an average of 0.338 (
Table 3). According to the values of He and I, the results showed that the genetic diversity of Chinese dogwood at the population level was slightly lower than that at the species level. The order of I of the 12 populations from high to low was: AH, Jixi, Qingliangfeng, Anhui (AQ), Hefeng, Hubei (HHF), Leshan, Emei, Sichuan (SE), HH, Shennongjia, Hubei (HS), Jinzhai, Tian Tang Zhai, Auhui (AT), Nanjing, Baohua, Jiangsu (NB), Jiujiang, Lushan, Jiangxi (JL), Ji’an, Jinggangshan, Jiangxi (JJ), HX, and AW. Furthermore, the highest genetic diversity among the populations, as revealed by the values of He and I, was in AH (0.288 and 0.423, respectively), while the lowest was in AW (0.177 and 0.274, respectively).
3.4. Genetic Structure and Genetic Relationship Analysis
The genetic distance and genetic similarity in 12 natural populations of Chinese dogwood are listed in
Table 5. The genetic distance ranged from 0.037 (between AQ and AH) to 0.297 (between JJ and AT), with an average of 0.177. Nei’s coefficient, varying from 0.743 (between JJ and AT) to 0.964 (between AQ and AH), with an average of 0.840, indicated small differentiation among populations. According to the Mantel test, there was no significant correlation between genetic distance and geographical distance based on the ISSR data (
R2 = 0.061,
p = 0.08), indicating that geographical isolation was not the main factor influencing the genetic differentiation of Chinese dogwood and that its genetic variation mainly exists within each population.
A dendrogram was constructed according to the genetic similarity between the populations. Twelve populations were grouped into three clusters at a coefficient of 0.85 (
Figure 2). The correlation coefficient (r) was 0.85, indicating that the clustering result was well. Cluster I, the largest, included seven populations (HS, HH, HX, AH, AQ, JL, and NB) mainly from Hubei, Anhui, Jiangsu, and Jiangxi provinces. The populations AW and AT from Anhui were placed in Cluster II. Cluster III, containing three populations (HHF, JJ, and SE), was mainly distributed in Hubei, Jiangxi, and Sichuan provinces. In addition, Cluster I and Cluster II can be combined into one larger group with a coefficient of 0.80. Additionally, the values of Nei’s coefficient varying from 0.80 to 0.96 also indicate little differentiation among the groups (
Figure 2). Among all individuals, three clusters were formed at a coefficient of 0.40 while the correlation coefficient (r) was 0.70. Individual samples in HS, HHF, and NB were grouped into cluster I, some samples of HH were placed to cluster II, and remaining samples were grouped into cluster III (SM 1). On the contrary, greater differentiation among the individual was verified by the values of Nei’s coefficient varying from 0.34 to 0.96.
Principal coordinate analysis (PCoA) can be used to analyze the phylogenetic relationships between species at different levels more intuitively than with systematic cluster analysis. The total percentage of variation explained by the first three axes was 62.78%, and it also showed that 12 populations could be divided into three groups at the 3D level (
Figure 3). The three groups were as follows: the populations of HHF, SE, and JJ belonged to group I; the populations of HX, HH, HS, AH, AQ, JL, and NB belonged to group II; and the populations of AT and AW belonged to group III. In addition, this result was consistent with the result of the system clustering with a coefficient of 0.85.
The structure of a Chinese dogwood population containing 223 individuals was calculated based on 94 combined ISSR loci using a model-based approach in STRUCTURE. One hundred data sets were obtained by setting the number of possible clusters (
K) from 1 to 12 with ten replications each. The LnP (D) value for each given
k increased with increasing
k, but no significant change was observed when
k was increased from 1 to 12 (
Figure 4A). The structure analysis showed that Δ
K had a maximum value of 494.19 when
K was 2, and Δ
K for
K = 3 (Δ
K = 488.50) did not differ significantly from that for
K = 2 (
Figure 4B). Therefore, the 12 populations can be divided into two or three groups (
Figure 4C). When
K = 2, the model-based groups were divided into two groups (
Figure 4C). Group I included the populations of HX, AH, AQ, HHF, SE, and JJ, and group II included the populations of HS, HH, NB, JL, AT, and AW (
Table 6). When
K = 3, the populations NB and JL were clustered into a group that was distributed on the edge of the Yangtze River (
Figure 5). However, the model-based groups were not consistent with the isolated topographical distribution regions of the individuals. It also suggests that geographical isolation would not be the main factor responsible for genetic differentiation in Chinese dogwood. The three groups correspond to the three major germplasm sources in China, which was seven populations from three provinces in eastern China (Anhui, Jiangxi and Jiangsu), four populations from two provinces in central China (Hunan and Hubei), and one population from Sichuan province in southwestern China. Among them, group I was distributed in three major germplasm resource areas, while group II (except for HH) was concentrated north of the Yangtze River; the AT and AW populations were distributed in the Dabie Mountains (
Figure 4C and
Figure 5).
The result of the STRUCTURE analysis was not consistent with that of the UPGMA and PCoA analysis based on populations when
K = 3, but was similar to that of the UPGMA analysis using samples at a coefficient of 0.52. The HHF, JJ, and SE populations were clustered in a group by UPGMA and PCoA analysis, while the proportion of membership for these populations was indeed larger than that of the others populations in group I according to the STRUCTURE analysis (
Table 6). The STRUCTURE analysis showed that populations of JL and NB were grouped together and they deviated from their group in PCoA analysis. According to the results of AMOVA analysis based on three groups, the genetic variation among population was more significant (22.27%). Further analysis shows that the genetic backgrounds of three clusters have high similarity, and even some populations contain genetic characteristics of multiple groups.