Divergence time dating
Best-fit models of nucleotide substitution of chloroplast haplotypes were chosen by jModelTest2 v2.1.6. The divergence times were estimated by BEAST v1.8.4 (Drummond et al., 2016 ). Lacking fossil records, secondary calibration points according to the diversification of Rosaceae were applied to calibrate node ages. Divergence time dating was calculated in two steps with two trees respectively. Five secondary calibration points were applied to the first tree: Rosaceae Crown, 90.18 Mya (Zhang et al., 2017; Li et al., 2017); Amygdaleae, 84.84Mya (Zhang et al., 2017); Spiraeeae, 84.30 Mya (Zhang et al., 2017); Roseae+Rubeae, 59.51 Mya (Zhang et al., 2017); Maleae Crown, 50.06 Mya (Zhang et al., 2017; Li et al., 2017) to evaluate the divergence time of Prunus(node 1) and the time of Cerasus Crown (node 2). These two nodes were subsequently used to calibrate the node ages in the second tree of haplotypes of Cerasus serrulata . The GTR substitution model and the uncorrelated log-normal relaxed clock and Yule process as for tree prior were selected. MCMC was run for 1 ×107generations with sampling every 5,000 generations. The parameters were checked in Tracer ver. 1.7.1 (Rambaut et al., 2018). Final phylogenetic trees were edited in FigTree v. 1.4.