Variation between D. magna and D. baiyunshana
The Shapiro-Wilk test revealed that centroid sizes of all populations show a normal distribution (P > 0.05). Leneve’s test showed a significant homogeneity of variances (F(1, 266) = 2.530, P = 0.113 > 0.05) based on mean centroid sizes. A one-way ANOVA of mean centroid sizes of wings showed highly significant differences among all the analyzed populations (F(8, 259) = 29.104, P = 0.000). The result of LSD as a post hoc test after Bonferroni correction on centroid size showed that D. baiyunshana bears the largest centroid size (Fig. 3) and differs significantly from all D. magna populations studied (P = 0.000) (Table 1).
Fig. 3. Boxplot of centroid size of D. baiyunshana (BYS) and D. magna populations (HDT, JLJ, NGS, and TB). White (male), grey (female).
Table 1. Pair-wise ANOVA on wing centroid size of D. baiyunshana (BYS) and D. magna populations (HDT, JLJ, NGS, and TB). * Significant differences, the adjustment of P-value after Bonferroni correction is < 0.006.
In order to discriminate taxa at the interspecific level, we used only the females to avoid the effect of sexual dimorphism. The twenty-three landmarks yield 42 shape variables, and accordingly 42 PCs. Each PC had progressively less variance, with PC1 (18.9%) and PC2 (15.9%) representing more than 34% of the total variance (Fig. 4). Most specimens from BYS are clustered together with D. magna populations on the right of the scatter plots (Fig. 4). Shape variation associated with the first PC reflected a construction of distal and posterior margins of the wing and anterior furcation of veins Sc and R. The second PC represented an expansion of the anterior, distal and posterior margins of the wing, and a construction of the basal part of the wing (Fig. 4).
The shape analysis revealed that the shape variation exists between D. baiyunshana and four different populations of D. magna (MANOVA: F(4, 177) = 3.68; P < 0.0001). CV1 (40.3%, Wilks’ λ = 0.0372; P < 0.001) and CV2 (27.3%, Wilks’ λ = 0.1183; P < 0.001) accounted for 67.6% of total shape variation, clearly separating D. baiyunshana individuals from all D. magna specimens (Fig. 5). Shape changes associated with CV1 explained the most discrimination between species, representing a construction of anterior margin and an expansion of posterior margin of the wing. CV2 also involved in a construction of anterior margin of the wing (Fig. 5).
The assignment by the cross-validation in LDA showed that 87% specimens were correctly assigned to their group (Table 2). Ten individuals of D. baiyunshana showed an HR of 100%. Four of HDT were misclassified, two being assigned to NGS and two to TB (HR 92%). Among 40 JLJ specimens, one was misclassified as HDT, three as NGS, and four as TB (93%). Among 44 NGS specimens, one was misclassified as HDT and four as TB (HR 84%). Twelve of TB were misclassified, two being assigned to HDT, four to NGS, and six to TB (70%) (Table 2).