2.6 Inference of demographic history
For ancestral area reconstruction, we used seven groups of C. chuniana for S-DIVA (statistical dispersal-vicariance analysis) analysis implemented in RASP v3.2 (Ronquist 1997; Yu, Harris, Blair, & He, 2015). The analysis was based on the BEAST MCMC trees and the maximum clade credibility tree derived from the Bayesian analysis with BEAST and TreeAnnotator (Matuszak, Muellner-Riehl, Sun, & Favre, 2016). With this method, the frequencies of an ancestral area at a node in the ancestral reconstructions are averaged over all trees. Dispersal or vicariance events were also detected with S-DIVA.
We applied coalescent simulations with the program fastsimocoal2 (FSC2; Excoffier, Dupanloup, Huerta-Sanchez, Sousa, & Foll, 2013) to provide model evidence of divergence, secondary contact, bottleneck effects, and demographic expansion. The populations in the Nanling Mts., which formed a monophyletic group and showed a distinct geographic location, were considered as one group (NL), and the remaining populations as another (ES). We used easySFS (https://github.com/isaacovercast/easySFS) to transform SNPs into a folded site frequency spectrum (SFS), based on the construction of 10 demographic models with the two groups (Figure S2). The models are: without isolation (NIS), isolation only (IS), isolation followed by migration (MIG), bottleneck effect (BOT), or secondary contact (SEC). Models including ancient (AMIG) or recent migration (RMIG), bidirectional or one-way migration, and demographic expansion (EXP) were also applied. In each model, NL or ES were alternatively used as the split source that was subjected to each scenario. We estimated effective population size (Ne ), time (T ) and migration rates in individual migrants per generation (MNL-ES and MES-NL ) for the two groups in each model from posterior distributions. To scale parameter estimates into real values, we used the substitution/site/generation mutation rate of 1.16×10-7 based on our research (Liu et al., unpublished data, 2020), because no genomic mutation rate has been calculated for Cercis . We estimated a generation time of five years for C. chuniana based on estimates for its congeners (Aldworth, 1998; Chen & Mao, 1999). We ran 100 replicate FSC2 analyses under each model with 10,000 simulations for optimal parameters and composite likelihood estimation. All 10 demographic models were compared (Figure S2, Tables S2‒S4). The composite likelihood of arbitrarily complex demographic models under the given SFS was calculated by using best-fit models based on the Akaike information criterion (AIC). The models with the lowest AIC were chosen as the best fit of the data.