Evolutionary history of a relict conifer, Pseudotaxus chienii (Taxaceae), in south-east China during the late Neogene: old lineage, young populations

Abstract

Background and Aims

Many monotypic gymnosperm lineages in south-east China paradoxically remain in relict status despite long evolutionary histories and ample opportunities for allopatric speciation, but this paradox has received little attention and has yet to be resolved. Here, we address this issue by investigating the evolutionary history of a relict conifer, Pseudotaxus chienii (Taxaceae).

Methods

DNA sequences from two chloroplast regions and 14 nuclear loci were obtained for 134 samples. The demographic history was inferred and the contribution of isolation by environment (IBE) in patterning genetic divergence was compared with that of isolation by distance (IBD).

Key Results

Three genetic clusters were identified. Approximate Bayesian computation analyses showed that the three clusters diverged in the late Pliocene (~3.68 Ma) and two admixture events were detected. Asymmetric gene flow and similar population divergence times (~ 3.74 Ma) were characterized using the isolation with migration model. Neither IBD nor IBE contributed significantly to genetic divergence, and the contribution of IBE was much smaller than that of IBD.

Conclusions

These results suggest that several monotypic relict gymnosperm lineages like P. chienii in south-east China did not remain in situ and undiversified for millions of years. On the contrary, they have been evolving and the extant populations have become established more recently, having insufficient time to speciate. Our findings provide a new perspective for understanding the formation and evolution of the relict gymnosperm flora of China as well as of the Sino-Japanese Flora.

Introduction

Geographical isolation between populations within species (allopatry) is viewed as the default means of speciation (Coyne and Orr, 2004; Harrison, 2012) and has been invoked by biogeographers and ecologists to explain the formation of biodiversity hotspots (Halffter, 1987; Qian and Ricklefs, 2000; Whittaker and Fernández-Palacios, 2007; Steinbauer et al., 2016). Given a sufficient period of strict allopatry, the accumulation of endemic variants through mutation, genetic drift and local adaptation will inevitably result in divergence and subsequently speciation (Won and Hey, 2005). However, strict allopatry is not always encountered in nature (Nosil, 2008; Zhou et al., 2010; Anacker and Strauss, 2014; Stankowski et al., 2015; He et al., 2018); species are composed of multiple populations that are dynamic in size, location, selective pressure and amount of gene exchange. Whether isolated populations speciate or not depends on the competition between unifying (gene flow) and diversifying genetic processes (genetic drift and local adaptation) (Hey and Nielsen, 2004).

China is one of the richest countries in plant diversity and one of the world's 17 mega-diversity countries (Mittermeier et al., 1997). One key reason for the high plant diversity in China is that the extreme physiographical heterogeneity, in conjunction with climate and sea-level changes, has provided abundant opportunities for evolutionary radiation through allopatric speciation (Qian and Ricklefs, 2000). However, China (particularly the central and south-east regions) is home to many relict lineages (Sakai, 1971; Ying et al., 1993; Manchester et al., 2009; Chen et al., 2018) that represent the survivors of clades with a large proportion of extinct members or the remains of a larger distributional area (Grandcolas et al., 2014). These relict lineages always contain only one or very few representatives and are narrowly or extremely disjunctly distributed compared with their larger sisters (Grandcolas et al., 2014). The most famous relicts in China include several monotypic gymnosperm genera (e.g. Ginkgo, Metasequoia, Cathaya) and many woody angiosperm genera (e.g. Davidia, Cercidiphyllum, Cyclocarya) that are monotypic or oligotypic (Manchester et al., 2009). Those evolutionary relicts are referred to as ‘living fossils' and their assembly in China over time is key to understanding the origin of the flora of China and even the evolution of Northern Hemisphere floras (Manchester et al., 2009; Chen et al., 2018). Recently, the demographic and divergence history of these relict lineages has gained renewed interest from plant phylogeographers (e.g. Qi et al., 2012; Tian et al., 2015; Kou et al., 2016). However, except for Cercidiphyllum, the range of which extends to Japan (Qi et al., 2012), clear examples of speciation events in relict tree lineages during the late Cenozoic (late Pliocene to Pleistocene) have never been demonstrated in central and south-east China and many relict tree lineages there remain monotypic. This is particularly true for gymnosperms in central and south-east China which each contain only one species (e.g. Cathaya, Ginkgo, Glyptostrobus, Metasequoia, Pseudotaxus, Pseudolarix; Ying et al., 1993). This phenomenon is perplexing given that those relict gymnosperm lineages have experienced long evolutionary histories and recent topographic changes and climatic shifts have provided ample opportunities for allopatric speciation (Qian and Ricklefs, 2000; Harrison et al., 2001; Qiu et al., 2011).

There are three possible explanations for the relict status of China's monotypic gymnosperms. First, several recent studies have demonstrated that relicts may be more dynamic than previously thought. In other words, the stem lineages are ancient, but the extant species originated or became established in China recently (Chou et al., 2011; Hohmann et al., 2018; Chen et al., 2018). The most striking case is Ginkgo biloba. This is an indisputable living fossil, but the genetic footprint of extant Ginkgo biloba populations in China dates back only 390 000 years (Hohmann et al., 2018). Thus, the relict status does not imply that they have not experienced any evolutionary changes. Instead, those lineages might have diversified recently (e.g. Cycas; Nagalingum et al., 2011) whereas other members went extinct. Alternatively, they might have broader ranges, yet China might have been colonized or re-colonized recently, while populations in other places became extinct. Under either situation, those relicts might have had insufficient time to speciate in China. Second, gene flow among populations, which has been assumed to be extremely limited due to long-term glacial and interglacial isolation (Harrison et al., 2001; Qiu et al., 2011), might be otherwise substantial, especially for plants with wind-mediated pollen and/or effective seed dispersal modes (e.g. Bai et al., 2014; Tian et al., 2015). Third, as most of those relict gymnosperms are geographically restricted, their ranges may span fewer environmental gradients and thus experience weak divergent selection (or local adaptation). This scenario might be more plausible in south-east China compared with south-west China where geological complexity could facilitate speciation (e.g. Li et al., 2013; Liu et al., 2013; reviewed by Favre et al., 2014). While one or two of the above hypotheses have been tested explicitly or implicitly (e.g. Chou et al., 2011; Bai et al., 2014), to our knowledge, no study has explicitly considered all the possibilities simultaneously.

Pseudotaxus is a genus in Taxaceae established by Cheng (1947). Although two species names have been published in this genus [P. chienii (Cheng) Cheng (Cheng, 1947) and P. liana J. Silba (Silba, 1996)], it was considered to be monotypic (Fu et al., 1999; Wang and Yang, 2007) as leaf morphology variation between P. chienii and P. liana individuals is continuous (Wang and Yang, 2007). Pseudotaxus chienii is a dioecious woody shrub or small tree up to 4 m tall, with seeds partly enclosed within a fleshy white aril at maturity (Fu et al., 1999). The species has a restricted distribution, occurring in isolated montane regions of southern Zhejiang, north-eastern and south-western Jiangxi, north-western and southern Hunan, northern Guangxi, and northern Guangdong, China (Fig. 1). Plants primarily grow on scarp slopes, along the sides of valleys or on cliffs at altitudes of 800–1500 m and are adapted to cool, humid climates (Fu and Jin, 1992; Fu et al., 1999; Xu et al., 2008). Although the fossil record of Pseudotaxus is deficient (Wu et al., 2007), the genus can be considered an evolutionary relict according to two of the criteria proposed by Grandcolas et al. (2014): (1) it contains one living representative and diverged from its sister group Taxus very early (Palaeocene or early Eocene, ~54–65 Ma; Leslie et al., 2012; Lu et al., 2014); and (2) it has a much smaller distribution relative to Taxus (Fu et al., 1999). As an officially protected species in China, its conservation status has led to research efforts. The genetic diversity and genetic structure of P. chienii have been investigated using dominant molecular markers such as random amplified polymorphism DNA (RAPD) and inter-simple sequence repeats (ISSRs) (Wang et al., 2006; Su et al., 2009). Low genetic diversity and significant genetic differentiation were reported, and fragmentation and bottleneck events were postulated based on simple summary statistics. To improve conservation efforts, an in-depth understanding of the evolutionary history that accounts for the present distribution of genetic diversity is needed. In particular, the timing of population divergence and admixture, fluctuation in population size, and natural selection experienced by different populations should be addressed with more powerful molecular tools and more sophisticated statistical algorithms (Qiu et al., 2011; Liu et al., 2012).

Fig. 1.

Distribution of three clusters (W, red; C, yellow; E, blue) revealed by STRUCTURE analysis using K = 3, which was determined to be the optimal K value using the ΔK method.

In this study, we sequenced 14 nuclear loci as well as two chloroplast genomic regions from 12 P. chienii populations. Two coalescent-based analyses [isolation-with-migration (IM) and approximate Bayesian computation (ABC)], which are either likelihood-based (IM) or likelihood-free (ABC), were implemented to explicitly test the first and the second hypotheses, by determining population divergence time and historical gene flow among populations. In addition, the importance of isolation by distance (IBD) and isolation by environment (IBE) were estimated to quantify the relative roles of neutral (genetic drift and gene flow) and adaptive processes (local adaptation) on population divergence, which can provide clues to the third hypothesis. By detailing the evolutionary history of a typical relict genus (Pseudotaxus) in south-east China, we attempted to explain a puzzling paradox: Why do many monotypic gymnosperms in south-east China remain in relict status given their long evolutionary histories and ample opportunities for allopatric speciation.

MATERIALS AND METHODS

Plant material, loci sampling, PCR amplification and sequencing

We investigated 34 wild populations that were recorded in herbarium specimens, reported in the literature, listed on websites of nature reserves, and were identified by experienced botanists. Living trees can be found only in 12 populations because some populations have been removed by deforestation and some remain inaccessible. Leaves of 134 individuals were collected and three to 17 individuals per population were sampled according to population size (Fig. 1 and Supplementary Data S1, Table S1). Leaves were dried in silica gel immediately after collection and subsequently used for DNA extraction. Total genomic DNA was extracted from ~20 mg of dried leaf using a modified CTAB procedure (Doyle and Doyle, 1987).

Two chloroplast (cp) DNA regions (rpl16 and psbA-trnH) were sequenced following our previous studies (Kou et al., 2016, 2017). In addition, 14 low-copy nuclear genes (PT23, PT58, PT107, PT122, PT126, PT140, PT154, PT202, PT203, PT222, PT333, PT353, PT373 and PT399) were developed from transcriptome sequences of P. chienii in this study (Supplementary Data Tables S2 and S3). These loci were classified into various functional categories against four protein databases (Nr, GO, KO and Swiss-Prot), with the exception of PT58 and PT222, which had no known function (Table S4). Transcriptome sequencing was performed on the Illumina HiSeq 2500 platform (Illumina, San Diego, CA, USA) with 101-base paired-end reads at Novogene Biological Information Technology (Beijing, China). The primer pairs were designed using Primer 3 (Untergasser et al., 2012), with PCR product size ranging from 400 to 700 bp, GC content between 40% and 60%, primer length ranging from 18 to 25 bp, and melting temperature between 55 and 65 °C. The raw sequences of the transcriptome were deposited at figshare (https://doi.org/10.6084/m9.figshare.9788927).

PCR amplification was performed in a 20μL volume containing 20–40 ng genomic DNA, 2× Taq PCR MasterMix (Tiangen, Beijing, China), 5 μM of each primer, and double distilled H₂O. The PCR procedure included an initial denaturation for 4 min at 94 °C, followed by 35 cycles of 35 s at 94 °C, 30 s at 55–59 °C and 1 min at 72 °C, ending with a final extension of 7 min at 72 °C. The different annealing temperatures were chosen according to each primer's features (Supplementary Data Table S2). Sequencing reactions were performed with the PCR primers in both directions by the service of Sangon Biotech (Shanghai, China). Sequences of the same locus were aligned using Clustal X 1.81 with default parameters (Thompson et al., 1997) and checked by eye using MEGA 5.05 (Tamura et al., 2011). All obtained DNA sequences have been deposited in the NCBI GenBank database under accession numbers MK331152–MK331582.

Nucleotide diversity, mismatch distribution and neutrality tests

Chloroplast DNA genetic diversity was estimated using the number of haplotypes (N_h), haplotype diversity (H_d) and nucleotide diversity (π; Nei, 1987) in DnaSP 5.10 (Librado and Rozas, 2009). The relationships between haplotypes were constructed using a median-joining network implemented in Network 5.0 (available at http://www.fluxus-engineering.com;Bandelt et al., 1999). The coefficients of differentiation G_ST (coefficient of genetic variation over all populations) and N_ST (coefficient of genetic variation influenced by both haplotype frequencies and genetic distances between haplotypes) were evaluated using Permut 1.0 (Pons and Petit, 1996). A mismatch distribution of the observed number of nucleotide differences between pairs of DNA sequences was computed using DnaSP 5.10, and values of Tajima's D (Tajima, 1989) and Fu's F_S (Fu, 1997) were calculated using Arlequin 3.01 (Excoffier et al., 2005) to infer demographic history.

All nuclear sequences were assigned to coding and non-coding regions by aligning genomic sequences against their corresponding mRNA sequences. For each nuclear locus, basic population genetic parameters were estimated after phasing sequences using the phase algorithm in DnaSP 5.10. We calculated the number of segregating sites (S), the number of haplotypes (N_h), haplotype diversity (H_d), nucleotide diversity (π), Watterson's parameter (θ _w; Watterson, 1975) and the minimum number of recombinant events (R_m). We also tested whether each locus fitted neutral evolution expectations using Tajima's D and Fu and Li's D* and F* statistics (Fu and Li, 1993). These parameters are expected to approach zero under neutrality by comparing the observed value of the summary statistics with their expected distribution.

Population genetic structure

STRUCTURE 2.3.4 (Hubisz et al., 2009) was used to assess population structure with the admixture model and the assumption of correlated allele frequencies using the dataset of 14 nuclear loci. Segregating sites in significant linkage disequilibrium after Bonferroni correction were excluded from this analysis. The number of clusters, K, corresponding to the number of populations, was explored using 20 independent runs per K. Burn-in was set to 20 000 and Markov chain Monte Carlo (MCMC) run length to 200 000. We ran STRUCTURE with K varying from 1 to 12, with 10 runs for each K value. The most likely number of clusters was estimated using LnP(D) (Pritchard et al., 2000) and ΔK statistics (Evanno et al., 2005). The population clusters were visualized using the program DISTRUCT 1.1 (Rosenberg, 2004).

We also used Wright's fixation index (F_ST; Wright, 1949) to assess population genetic differentiation between geographical groups. F_ST values for each nuclear locus and across all nuclear loci were calculated using AMOVA in Arlequin 3.01. The significance of F_ST was tested based on 10 000 permutations as described by Excoffier et al. (1992).

Population divergence and demography

Because cpDNA variation in P. chienii is extremely low (only four substitutions were identified, and nearly half of the sampled populations were invariable; see details in Results), we did not include cpDNA in subsequent analyses. The population divergence models were tested using an ABC approach implemented in DIYABC 2.1.0 (Cornuet et al., 2014). Because two populations in the central cluster contain mixed genetic components in the STRUCTURE analysis (see details in Results), we adopted a three-step ABC analysis. First, we excluded populations with admixed components in the central cluster (CD and YZ, hereafter named C_A), and five possible population divergence scenarios were modelled and tested for the west, central (excluding C_A) and east clusters (W, C_P and E). These scenarios are visualized and described in detail in Supplementary Data Figure S2. Second, after adding genetically mixed populations (C_A) into the most likely model inferred in the first step, five scenarios were formulated (Fig. 2): (1) W, C_P and E diverged from an ancestral population at the same time, t₃, then C_A diverged from C_P; (2) C_A was born from an admixture event between C_P and E, or (3) C_A formed via admixture between W and C_P at time t₁; (4) C_A was then born from an admixture event between W and the descendant of C_P and E; or (5) C_A formed from an admixture event between E and the descendant of W and C_P. Third, six scenarios (Fig. 3) of population demography were tested using ABC for each genetic cluster [W, C_P and E; C_A was excluded in this analysis as well as in IMa2 and LAMARC analyses (see below) due to their admixed origin that probably induces underrated calculation; e.g. Li et al. (2013)]: (1) an expansion at time t₁, (2) a contraction at time t₃ followed by an expansion at time t₁, (3) an expansion at time t₃ followed by a contraction at time t₂, and then followed by an expansion at time t₁, (4) a contraction at time t₁, (5) an expansion at time t₃ followed by a contraction at time t₁, and (6) a contraction at time t₃ followed by an expansion at time t₂, and then followed by a contraction at time t₁. The priors of all parameters were set with a uniform distribution (Supplementary Data Tables S5–S7). All one-sample and two-sample summary statistics were chosen to compare observed and simulated datasets. To ensure statistically robust results, at least 2 000 000 simulated datasets were generated for each scenario. We used 1% of the simulated datasets closest to the observed data to estimate the relative posterior probability [with 95% confidence intervals (CIs)] determined for each scenario via logistic regression and posterior parameter distributions (Cornuet et al., 2014). The generation time of 25 years were assumed for P. chienii as this estimate was applied to Taxus wallichiana in the same family Taxaceae (Liu et al., 2013).

Fig. 2.

Five assumed scenarios for population divergence of clusters W, C_P, C_A and E tested in DIYABC. N_W, N_CP, N_CA and N_E represent the population sizes of clusters W (TZ, JY, LH and DM), C_P (JG and GG), C_A (YZ and CD) and E (SQ, JL, LQ, and FY), respectively. N_C-E and N_C-W represent the population sizes of the admixed populations. N_A represents the ancestral population size. t₁, t₂ and t₃ represent times of population divergence and genetic admixture between clusters W, C_P, C_A and E. The best model is shown in bold type.

Fig. 3.

Six assumed demographic scenarios tested for clusters W, C_P and E in DIYABC. N₁ and N₂ represent current population sizes, and N_A represents the ancestral population size. N_1a, N_1b, N_2a and N_2b represent population sizes between the ancestral population and the current population. t₁, t₂ and t₃ represent times of population changes. The best model is shown in bold type.

The IM model (Hey, 2010a, b) was also used to estimate gene flow and divergence time between genetic clusters. The model assumed neutrality, no selection, no recombination within loci, and random mating in ancestral and descendant populations (Hey and Nielsen, 2004, 2007). After extracting the longest non-recombining region of each locus using the Imgc program (Woerner et al., 2007), the demographic parameters, migration rate (m), effective population size (θ) and divergence time (t), were simulated and estimated using MCMC implemented in the IMa2 software package (Hey, 2010a, b). We ran the simulations using a burn-in of 5 000 000 steps and retaining 1 000 000 steps under the HKY mutation model. The demographic parameters from the IM model are scaled by a mean mutation rate. The mutation rate was estimated according to μ = K_S/2T, where K_S is the average divergence at silent sites between P. chienii and its closest extant relative T. chinensis, and T is the divergence time between the two species (~65 Ma; Leslie et al., 2012). The resulting geometric average mutation rate, 3.44 × 10⁻¹⁰ per site per year, was used to scale the effective population size and divergence time. Although the mutation rate was estimated over seven nuclear loci because PCR amplification of other loci failed for T. chinensis (represented by one individual), it is comparable with estimates in other conifers, such as 2.23–3.42 × 10⁻¹⁰ per site per year for Picea (Bouillé and Bousquet, 2005), 7.0–13.1 × 10⁻¹⁰ for Pinus (Willyard et al., 2007) and 1.94 × 10⁻¹⁰ for Juniperus (Li et al., 2012).

A Bayesian method, which accounts for the genealogical relationships among alleles and allows for population size changes, was implemented to estimate the exponential population growth rate (g) in LAMARC 2.1.8 (Kuhner, 2006). This analysis was performed using two runs of LAMARC, which included 10 initial chains and two final chains. The initial chains were run with 5000 samples and a sampling interval of 20 and using a burn-in of 10 000 samples for each chain. The final chains were carried out with the same burn-in and interval sampling as initial chains, but with 50 000 samples.

Isolation by distance and isolation by environment

To understand the effects of geography and environment on spatial patterns of genetic variation, we quantified the contributions of IBD and IBE in determining patterns of population genetic divergence using multiple matrix regression with randomization (MMRR) with all populations included (Wang, 2013). First, 19 bioclimatic variables were extracted from current climatic data with 30 arc-second resolution (http://www.worldclim.org) and used as proxies of environmental gradients. A climatic distance matrix was constructed using population pairwise Euclidean distances across all 19 bioclimatic variables. A geographical distance matrix was also calculated using the GPS coordinates of populations. Genetic distances (F_ST) among each pair of populations were calculated using Arlequin 3.01. Next, after standardizing the three distance matrices by Z scores, we analysed the contributions of geographical distance and environmental distance (explanatory variables) to genetic distance (response variable), using the 'MMRR' function in R (Wang, 2013). Additionally, principal component analysis (PCA) and multivariate statistical analysis were performed on the standardized 19 bioclimatic variables using IBM SPSS Statistics 22 (IBM, New York, USA) to evaluate the climatic divergence of three clusters.

RESULTS

Patterns of variation in cpDNA

One inversion, two indels and four substitutions were detected from concatenated sequences of two cpDNA fragments, rpl16 and psbA-trnH, which had a total length of 1452 bp (Supplementary Data Table S8). Seven haplotypes (H1–H7) were recovered from these sequence variations (Table S1). One haplotype (H1) occurred in all populations except for DM. Three haplotypes (H2–H4) were each shared between a population pair (LQ–JL, TZ–JL and JG–DM) (Fig. S1). The other three haplotypes (H5–H7) occurred in a single population each (YZ, JY and DM, respectively). Network analysis revealed that H2 and H4–H7 derived from H1 with one mutation and H3 was further derived from H6 with two mutations (Fig. S1).

Haplotype diversity (H_d) was 0.5933 at the species level and ranged from 0 to 0.6154 within populations. Species-level nucleotide diversity (π) was 0.0009, ranging from 0 to 0.00138 within populations (Supplementary Data Table S1). Five populations did not have cpDNA variation (SQ, FY, GG, CD and LH). Genetic differentiation was moderate among populations (G_ST = 0.519, N_ST = 0.598), and no significant difference was detected between G_ST and N_ST. The multimodal mismatch distribution for cpDNA (Fig. S3) indicated that populations might have experienced demographic equilibrium or contraction, which was also supported by a non-significant negative value of Tajima's D (−0.74552, P = 0.254) and positive Fu's F_S (0.93509, P = 0.673).

Patterns of variation and neutrality tests for nuclear DNA

Fourteen low-copy nuclear genes were sequenced for all 134 individuals from 12 populations. The total aligned length was 6473 bp, with loci ranging from 334 to 702 bp. Three indels, located in non-coding regions of PT122, PT353 and PT399 with lengths of 18, 1 and 2 bp, respectively, were detected and excluded from the data analyses because they have a different mutation mechanism from substitutions.

A total of 109 segregating sites, included 10 singleton sites, were found from 14 nuclear genes. Haplotype diversity (H_d) varied from 0.2520 (PT58) to 0.9000 (PT353) and nucleotide diversity (π _t) ranged from 0.00067 (PT58) to 0.00974 (PT122) with an average of 0.00265. Silent nucleotide diversity (π _s) varied from 0 (PT23 and PT222) to 0.01192 (PT202) with an average of 0.00358, which was higher than the average non-synonymous diversity (π _a = 0.00212). The minimum number of recombination events (R_m) detected across all nuclear loci ranged between zero and five. Average values of Tajima's D and Fu and Li's D* and F* were all positive, and Tajima's D was close to zero, although these values did not significantly deviate from the expectations of a standard neutral model for all nuclear loci (Supplementary Data Table S9).

Genetic structure and genetic differentiation

The Bayesian clustering algorithm (STRUCTURE) revealed that the most likely number of clusters for the entire dataset was K = 3 based on ΔK statistics (Supplementary Data Fig. S4). The three clusters clearly matched the geographical distribution: the west cluster (W) consisted of populations TZ, JY, LH and DM, the east cluster (E) included SQ, JL, LQ and FY, and the central cluster (C) contained populations JG, GG, YZ and CD (Fig. 1). YZ and CD showed signals of genetic admixture and we therefore defined them as C_A, and the other two populations in the central cluster (JG and GG) were defined as C_P.

We further explored the genetic substructure within each genetic cluster (Supplementary Data Fig. S5). Cluster W was divided into four distinct units when K = 4. While populations in cluster C were grouped into three units when K = 3, populations YZ and CD were not separated when K = 4. For cluster E, STRUCTURE analysis was not able to differentiate the four populations regardless of the K value.

F _ST values were highly significant for each pair of genetic clusters (Table 1). Within cluster C, genetic differentiation between C_P and C_A was non-significant. When mixed populations (C_A) were excluded, the F_ST value between E and C_P increased, whereas it decreased between W and C_P, indicating that cluster E contributed more to the genetic composition of C_A compared with W.

Table 1.

Population differentiation (F_ST) at 14 nuclear loci among clusters W, C (C_P and C_A) and E based on pairwise comparisons.

Locus	W vs. E	E vs. C	W vs. C	E vs. CP	W vs. CP	CP vs. CA
PT23	−0.08964	0.29237	−0.00048	0.40364	0.05304	−0.27831
PT58	0.02059	0.02934	−0.05446	−0.02606	−0.17540	0
PT107	0.07616	0.17826**	−0.07264	0.20517**	−0.11832	0.01705
PT122	0.21718***	−0.02168	0.17286***	−0.01229	0.10909*	−0.37503
PT126	0.12006*	0.21293***	0.10458*	0.32433*	0.11656	−0.01298
PT140	0.54957***	0.18721*	0.13424	0.21929*	0.07369	−0.08641
PT154	0.79650*	0.10853	0.42214	0.16007	0.32832	−0.12108
PT202	−0.08902	−0.00700	−0.08521	0.04835	−0.07607	0.03947
PT203	0.23256***	0.11600*	−0.04306	0.24987**	−0.07565	0.48263
PT222	−0.00540	0.12427**	0.08535	0.09315*	0.04920	0.00004
PT333	0.24574***	0.11564*	0.43376***	0.14059	0.40948**	0.22827
PT353	0.31210***	0.12404***	0.33513***	0.17002***	0.35686***	0.06306
PT373	0.00454	0.10006*	0.11531	0.09470	0.11458	−0.30421
PT399	0.10195***	0.10905*	0.00759	0.12134*	−0.05415	0.00942
Total	0.22659***	0.11358***	0.17654***	0.15774***	0.13901***	−0.08293

W, western populations (TZ, JY, LH and DM); E, eastern populations (SQ, JL, LQ and FY); C, central populations (JG, GG, YZ and CD); C_P, central populations dominated by one genetic component (JG and GG); C_A, central populations with admixed genetic components (YZ and CD). *P < 0.05; **P < 0.01; ***P < 0.001.

Population divergence and demographic history

To yield more robust results, ABC modelling of the population divergence history of P. chienii was performed in two steps. First, two mixed populations (YZ and CD), detected from genetic assignment analysis, were excluded from the modelling. Scenario 1, which hypothesized that three clusters (W, C_P, and E) diverged from an ancestral population at the same time, received the highest support (59.56 % posterior probability; Supplementary Data Fig. S6, Table S10) among five potential population divergence models (Fig. S2). Second, based on the best model (scenario 1) in the first step, we simulated five scenarios that included two mixed populations (YZ and CD) (Fig. 2). The results suggested that scenario 4 had a high posterior probability (49.05 %) (Fig. S7, Table S11). Posterior parameter estimates for scenario 4 suggested that three clusters (W, C_P and E) diverged from a common ancestor around 3.68 Ma (95 % CI: 2.60–4.63 Ma). The origin of C_A was quite complex; it was derived from two admixture events: one between C_P and E at around 0.68 Ma, and the other between the derivative of C_P-E and W at ~0.40 Ma (Table 2; Fig. S8).

Table 2.

Posterior estimates of demographic parameters for the best model (scenario 4) of population divergence in Fig. 2 revealed by Approximate Bayesian Computation.

Parameter	Mean	Median	Mode	95 % CI
N W	2.80 × 105	2.89 × 105	3.03 × 105	1.98–3.35 × 105
N CP	1.80 × 105	1.83 × 105	1.89 × 105	1.11–2.37 × 105
N CA	4.86 × 104	4.69 × 104	4.72 × 104	1.16–9.19 × 104
N E	3.62 × 105	3.68 × 105	3.81 × 105	2.31–4.79 × 105
N C-E	5.20 × 104	5.20 × 104	3.77 × 104	0.81–9.54 × 104
N A	3.85 × 105	3.87 × 105	3.48 × 105	0.19–5.69 × 105
T 1 (years)	4.00 × 105	3.73 × 105	2.19 × 105	0.56–8.58 × 105
T 1 (generations)	1.60 × 104	1.49 × 104	8.76 × 103	0.22–3.43 × 104
T 2 (years)	6.75 × 105	5.65 × 105	3.63 × 105	0.16–1.58 × 106
T 2 (generations)	2.70 × 104	2.26 × 104	1.45 × 104	0.64–6.32 × 104
T 3 (years)	3.68 × 106	3.73 × 106	3.85 × 106	2.60–4.63 × 106
T 3 (generations)	1.47 × 105	1.49 × 105	1.54 × 105	1.04–1.85 × 105
r 1	0.665	0.716	0.917	0.185–0.966
r 2	0.274	0.233	0.136	0.051–0.638

N _W, population size of cluster W (TZ, JY, LH and DM); N_CP, population size of cluster C_P (JG and GG); N_CA, population size of cluster C_A (YZ and CD); N_E, population size of cluster E (SQ, JL, LQ and FY); N_C-E, population size of admixed population; N_A, ancestral population size; T, divergence time of clusters W, C and E from the ancestral population; r₁ and r₂, admixture rates.

Among six potential demographic scenarios in ABC modelling (Fig. 3), cluster E conformed to scenario 1, which indicated a population expansion at 0.62 Ma, while clusters W and C_P conformed to scenario 4, which indicated population contractions at 0.48 and 0.49 Ma, respectively (Table 3 and Supplementary Data Table S12, Figs S9 and S10). The LAMARC analysis also suggested that the three clusters have undergone changes in population size. For clusters W and C_P, the estimated g values were −151.75 and −59.27, respectively, suggesting population declines. In contrast, the g value of cluster E was 219.77, indicating population growth.

Table 3.

Posterior estimates of demographic parameters for the best models of population demography in Fig. 3 revealed by Approximate Bayesian Computation.

Cluster (best model)	Parameter	N A	N 1/N2	T (years)	T (generations)
W (scenario 4)	Mean	1.79 × 105	9.13 × 104	4.78 × 105	1.91 × 104
	Median	1.77 × 105	8.75 × 104	4.63 × 105	1.85 × 104
	Mode	1.52 × 105	8.45 × 104	2.23 × 105	8.92 × 103
	95 % CI	0.76–2.87 × 105	0.38–1.59 × 105	0.68–9.33 × 105	0.27–3.73 × 104
CP (scenario 4)	Mean	1.79 × 105	5.70 × 104	4.90 × 105	1.96 × 104
	Median	1.79 × 105	5.19 × 104	4.70 × 105	1.88 × 104
	Mode	1.62 × 105	3.86 × 104	3.18 × 105	1.27 × 104
	95 % CI	0.70–2.87 × 105	0.17–1.14 × 105	0.93–9.38 × 105	0.37–3.75 × 104
E (scenario 1)	Mean	1.25 × 105	2.21 × 105	6.20 × 105	2.48 × 104
	Median	1.11 × 105	2.23 × 105	6.93 × 105	2.77 × 104
	Mode	4.39 × 104	2.49 × 105	9.93 × 105	3.97 × 104
	95 % CI	0.33–2.59 × 105	0.60–3.79 × 105	0.79–9.83 × 105	0.32–3.93 × 104

N _A, ancestral population size; N₁/N₂, current population sizes; T, population expansion/contraction time.

Migration rates (m) estimated under the IM model using IMa2 were asymmetric, with much higher gene flow from E to W (6.69), E to C_P (10.73), and C_P to W (10.06) than in the reverse directions (approaching zero) (Fig. 4 and Supplementary Data Fig. S11, Table S13). The divergence time estimates using IMa2 between clusters C_P and W and between E and C_P were unreliable due to a lack of convergence in their marginal posterior densities. However, we obtained a robust time estimate for the divergence between E and W [3.74 Ma; 95% highest posterior density (HPD) interval: 1.31–11.80 Ma) (Fig. 4; Fig. S11, Table S13), which is almost identical to the divergence time of three clusters in ABC modelling.

Fig. 4.

Posterior probability distributions of the migration rate (m) between each pair of clusters (W, E and C_P) from both directions and the divergence time (t) between clusters W and E estimated by the IM model.

Contributions of IBD and IBE to population divergence

MMRR analysis revealed non-significant effects of both IBD (β _IBD = 0.121, P = 0.425) and IBE on population genetic divergence (F_ST) (β _IBE = 0.062, P = 0.758). The impact of IBD on spatial patterns of genetic variation was about twice as much as IBE. Almost none of the 19 bioclimatic variables had a significant contribution to genetic divergence, except for BIO 8 (mean temperature of wettest quarter), which was marginally significant (P = 0.047; Fig. 5A). PCA did not find clear climatic differentiation among three genetic clusters (W, C and E) in either of the first two principal components (PC1: 52.87%, PC2: 23.23%) (Fig. 5B); multivariate statistical analysis also showed no significant differentiation among them (P = 0.207).

Fig. 5.

(A) The effects of 19 bioclimatic variables on population genetic divergence analysed using the MMRR function in R. The red dotted line represents a significance level of 0.05. (B) Principal component analysis of 19 bioclimatic variables among three genetic clusters using IBM SPSS Statistics 22. The red, green and blue circles represent population clusters W, C and E, respectively.

Discussion

Recent establishment of Pseudotaxus chienii populations in south-east China

The divergence time of three genetic clusters (W, C_P and E) of P. chienii estimated by ABC was about 3.68 Ma (95 % CI: 2.60–4.63 Ma), indicating that the first hypothesis is most plausible. Given the stem lineage of Pseudotaxus branched very early (~54–65 Ma; Leslie et al., 2012; Lu et al., 2014), this result is quite striking. However, this timing is almost identical to the divergence time of the E and W clusters in IM (3.74 Ma; HPD: 1.31–11.80 Ma), an approach that has a different philosophy from ABC. In ABC analysis, a calibrated mutation rate for sequences is not needed for parameter estimation (such as migration rate, divergence time) (Beaumont, 2010); in contrast, the mutation rate is required to convert the rate-scaled parameter estimates to more interpretable demographic units (years or generations) in the IM model (Hey, 2010a, b). Although there are wide confidence intervals around the time estimates, the recent time estimates by two different approaches suggests that: (1) Pseudotaxus might have existed in south-east China and other locations initially – however, the lineage of P. chienii branched recently in south-east China and other members and populations of Pseudotaxus elsewhere went extinct; and (2) P. chienii might have broader ranges (perhaps including south-east China) but south-east China could have been colonized (never appeared in south-east China) or re-colonized (appeared in south-east China initially but disappeared later) recently, while populations in other locations went extinct. Whichever scenario is more plausible, the results of this study suggest that the extant P. chienii lineage is relatively young and the populations of P. chienii started to differentiate in south-east China recently (possibly started in the late Pliocene). Such a short duration of population isolation (~3.68–3.74 Ma) may be insufficient to result in species-level divergence within a slow-growing tree species with long generation time, a conclusion consistent with the idea of macroevolutionary stasis in tree lineages since the Pliocene (5.30–2.60 Ma; Huntley, 1991).

With the recent interest in phylogeographical patterns and evolutionary history of Chinese plants, Pliocene to Pleistocene origins or establishment of relict gymnosperms in central to south-east/south China have been increasingly documented. Taiwania cryptomerioides, for example, is also a relictual monotypic conifer in south China (occasionally extending to northern Vietnam). Molecular clock estimation demonstrated that the mean divergence time between two major cpDNA lineages (insular and mainland) of Taiwania occurred at ~3.23–3.41 Ma (Chou et al., 2011), a value very close to the population divergence time of P. chienii. Recently, an analysis of whole chloroplast genomic sequences of wild extant Ginkgo populations revealed that the deepest temporal footprint unexpectedly dated back to ~0.39 Ma (Hohmann et al., 2018). Although a single locus (cpDNA) may suffer from genealogical stochasticity and overestimate population divergence time (Maddison, 1997; Carstens and Knowles, 2007), previous timings are consistent with the population divergence time of P. chienii inferred from multiple nuclear sequence data, enhancing the possibility that the duration of population/lineage differentiation of relict gymnosperm taxa in central to south-east/south China may be much shorter than the ages of lineages seemingly indicate (Mao and Liu, 2012).

Fossils provide the best evidence for the presence of a species within a past window of space and time (Gavin et al., 2014). Unfortunately, the fossil record of P. chienii is very limited (Wu et al., 2007), providing little information about its evolutionary history. The situation for P. chienii is very similar to Amborella trichopoda that is sister to all other extant flowering plants (Soltis et al., 2008), on the island of New Caledonia, but has no fossil record. Amborella trichopoda has long been viewed as an autochthonous remnant of the Gondwanan biota (Heads, 2009), but geological evidence precludes the long-term existence of A. trichopoda in New Caledonia (Grandcolas et al., 2014). A palaeogeographical study of the flora of New Caledonia showed that 'Amborella' was probably represented in many places by many lineages initially, but only the sole representative (i.e. A. trichopoda) remained in New Caledonia. However, any initial New Caledonia lineage could have been lost during the Eocene, regained during the Oligocene and subsequently extinguished in other places (Lee et al., 2001). Similarly, the monsoon climate to which P. chienii is adapted did not exist before the early Miocene (~23 Ma) because south-east China was controlled by a broad, roughly W–E dry belt caused by the planetary circulation system during the early Cenozoic (Guo et al., 2008), precluding the possibility of long-term existence of Pseudotaxus in south-east China after the stem lineage initially branched out (54–65 Ma). Pseudotaxus was able to colonize south-east China only after the early Miocene (Kou et al., 2016) and the most likely colonization of P. chienii into southeast China could have occurred during the late Miocene to Pliocene when the climate deteriorated globally (Zachos et al., 2001) and the Asian monsoon was intensified (An et al., 2001), a period that is consistent with the crown ages of several relict lineages (Wang et al., 2009; Qi et al., 2012; Tian et al., 2015; Zhang et al., 2015). This conclusion implies that P. chienii could have had a broader distribution in history, the lack of fossil records might be due to the morphological similarity to Taxus (personal communication with Dr Y. Yang, Institute of Botany, The Chinese Academy of Sciences) or incomplete investigations.

In addition, the abundant fossil records of other gymnosperms, such as Metasequoia glyptostroboides, may provide clues for the evolution of P. chienii. Metasequoia glyptostroboides was once widely distributed through high latitudes of the northern hemisphere. With the climate deteriorating, its range shrank sharply and the latest fossils were found in Japan in the Pliocene or early Pleistocene. However, living M. glyptostroboides was surprisingly rediscovered in south-eastern China (LePage et al., 2005), suggesting that it might have become established in China during the early Pleistocene, possibly immigrating from Japan (LePage et al., 2005). The bulk of Metasequoia fossils, coupled with molecular dating by Chou et al. (2011), Hohmann et al. (2018) and this study, clearly demonstrates that the relict gymnosperm flora in south-east China and even the Sino-Japanese Flora are much younger than previously thought (Chen et al., 2018). Formation of the relict gymnosperm flora may have resulted from colonizations or recolonizations from other parts of the world (Thorne, 1999; Manchester et al., 2009) driven by global aridity and cooling since the Miocene (Zachos et al., 2001) and the formation of the Asian monsoon since the Miocene (Guo et al., 2008).

Strong asymmetrical historical gene flow

We hypothesized that gene flow among populations of P. chienii might be substantial enough to prevent allopatric speciation. In this study, ABC analysis found that two historical genetic admixture events occurred at 0.68 Ma (between E and C_P; 95% CI: 0.16–1.58 Ma) and 0.40 Ma (between derived populations of E-C_P and W; 95% CI: 0.06–0.86 Ma). IM analysis further indicated that genetic exchange between clusters was asymmetrical, with moderate to strong gene flow from east to west, whereas gene flow approached zero in the reverse direction (Fig. 4), supporting that gene flow has indeed counteracted population divergence in P. chienii.

The timing of the first admixture event is largely consistent with the population expansion time of cluster E (0.62 Ma; 95% CI: 0.08–0.98 Ma), a period corresponding to the Cromerian Glaciation immediately following the middle Pleistocene transition (0.80–1.20 Ma) (Ehlers and Gibbard, 2007). This is considered to be the largest glaciation during the Pleistocene; for example, the ice sheet during the Wangkun glaciation (700–500 ka) was 18 times larger than the present ice sheet on the Qinghai–Tibetan Plateau (Shi, 1998). This glaciation may be associated with several major intraspecific evolutionary events in other plant species of south-east China (e.g. Tian et al., 2015). It is likely that cold-tolerant P. chienii expanded its range (especially in the east populations) during favourable glacial conditions and this expansion resulted in genetic admixture. This possibility is further supported by the LAMARC analysis, which also detected an expansion event in cluster E.

The second admixture event occurred at ~0.40 Ma, a period corresponding to the Elsterian Glaciation (Marine Isotope Stage 12; Ehlers and Gibbard, 2007). However, ABC analyses showed that clusters W and C_P experienced population declines starting at about 0.48 and 0.47 Ma, respectively. The population declines are also supported by the g values of LAMARC analyses and the mismatch analysis of cpDNA variation (Supplementary Data Fig. S3). These events would have reduced the possibility of genetic exchange between west and central populations. However, the central and west populations are geographically contiguous, and pollen-mediated gene flow was highly likely (Bai et al., 2014) given that the pollen of P. chienii is transmitted by wind (Su et al., 2009). This interpretation is further supported by the widely distributed chloroplast haplotype (H1) because chloroplast genomes in Taxaceae are paternally inherited (Collins et al., 2003) and transmitted by pollen, although the retention of ancestral polymorphism and insufficient marker resolution may also contribute to the sharing of haplotypes. In addition, DIYABC is usually sensitive to the strongest population change during a given time period (Cornuet et al., 2014), so we cannot exclude the possibility that clusters W and C_P would have experienced slight range expansion during the Elsterian Glaciation (~0.40 Ma) after the population decline events (~0.48 and 0.47 Ma).

The finding of strong historical gene flow within P. chienii has profound implications for understanding the evolutionary history of temperate tree species in China during the late Neogene. In contrast to the expansion–contraction model for temperate tree species in response to the Quaternary climate oscillations in Europe and North America, early plant phylogeographical studies that were preferentially based on maternally inherited cpDNA sequences identified a multi-refugium model in China (particularly in central to southeast China) (Qiu et al., 2011; Liu et al., 2012). This model describes a general pattern of multiple localized refugia and little admixture among refugial populations throughout the Quaternary glacial–interglacial cycles. This model is consistent with the allopatric speciation hypothesis of Harrison et al. (2001) that population fragmentation of temperate forest species occurred during both the Quaternary glacial and interglacial periods, which may have resulted in allopatric radiations. However, to our knowledge, no clear examples of a late Pliocene–Pleistocene tree speciation event have been documented in south-east China thus far, possibly due to range shift-induced genetic exchange and pollen-mediated gene flow as demonstrated in P. chienii. This finding has two implications. First, temperate tree species in south-east China could have lower speciation rates than previously thought, and the high speciation rate in woody taxa might be ascribed to some speciose taxa (e.g. Castnopsis, Ilex, Eurya or Camellia) in evergreen broadleaved forests (Lu et al., 2018). Second, while the role of pollen-mediated gene flow has been recognized (e.g. Bai et al., 2014), the contribution of range shift to species cohesion has not been fully evaluated and needs to be further investigated in south-east China using coalescent-based methods relying on multiple nuclear loci or genomic data.

Weak climate-induced divergent selection in south-east China

Understanding the mechanisms generating differences among populations and species is a major objective in evolutionary biology. Geography and environment represent two key landscape components that can potentially influence gene flow and population connectivity (Wang and Bradburd, 2014). In this study, MMRR analysis showed that the contribution of IBD to genetic divergence of P. chienii is much higher than that of IBE. These results suggest that neutral processes associated with geography, such as mutation and genetic drift, are stronger driving factors for population divergence of P. chienii than processes associated with environments, echoing the hypothesis of limited environmental gradients within narrow geographical ranges in south-east China. In addition, MMRR analysis found that none of 19 bioclimatic variables except mean temperature of the wettest quarter (BIO 8) contributed to genetic divergence significantly, indicating that environmental gradients (using 19 climatic variables as proxies) across the range of P. chienii may not be strong enough to facilitate genetic differentiation, possibly due to the prevalence of the East Asian monsoon. PCA and multivariate statistical analysis of climatic variables did not find significant differentiation of climatic envelopes among the three genetic clusters (P = 0.207; Fig. 5B), further supporting that IBE has limited contribution to genetic divergence in P. chienii.

Divergent selection and adaptation associated with different habitats have been viewed as a major cause of allopatric speciation in plants (Rieseberg and Burke, 2001; Funk et al., 2006; Harmon et al., 2008). Because many mechanisms can generate IBE, the detection of IBE alone is not evidence of local adaptation (Wang and Bradburd, 2014). However, the effects of climatic variables such as precipitation and temperature have been implicated in cases of incipient or completed speciation events (Nosil et al., 2005; Lowry et al., 2008). It is reasonable to assume that IBE can be simplified and considered as a proxy of divergent selection. Therefore, in this case, divergent selection exerted by environmental gradients (i.e. climatic gradients) in different populations of P. chienii could be too weak to generate enough of a driving force (i.e. divergent selection or local adaptation) for speciation or incipient speciation to occur. This weak divergent selection may be partially responsible for the lack of recent tree speciation in south-east China as well.

By contrast, in south-west China and particularly the Himalaya–Hengduan Mountains (HHM) region, where the East Asian Monsoon and Indian Monsoon converge and extreme physiographical changes occur within relatively short distances, geologically recent speciation or incipient speciation events have been frequently reported (e.g. Xu et al., 2010; Jia et al., 2011; Liu et al., 2013; Li et al., 2013; Luo et al., 2016). For example, Liu et al. (2013) found that Taxus wallichiana, a species within the sister genus to Pseudotaxus, initiated incipient speciation at ~4.20 Ma, an estimate slight older than the divergence time of the three Pseudotaxus clusters. The underlying factor may be that population isolation was initiated by the rise of the Qinghai–Tibetan Plateau and then reinforced by subsequent differential ecological (climatic) adaptations (Liu et al., 2013). However, such conditions (strong physical barriers and environmental gradient) do not exist in south-east China and that is why south-west China became a cradle of plant diversity, but south-east China a preservation centre for relict lineages (Thorne, 1999).

Conclusions

The results of our study on P. chienii, taken together with those recently reported for other monotypic gymnosperms that occur in central and south-east China, clearly demonstrate that the relict gymnosperms may have become established very recently despite the fact that the stem lineages of those taxa may have had a much longer evolutionary history (Mao and Liu, 2012). The current relicts are dynamic and do not imply conservation of ancestral characters in terms of morphology, ecology and spatial distribution (Grandcolas et al., 2014). The reason that these monotypic gymnosperms have a relict status in China is that the prerequisites (enough time for strict isolation and divergent selection) for allopatric speciation in south-east China have not been fulfilled even though the opportunities for allopatric speciation prevail. This study may provide a new perspective for understanding the evolution of relict gymnosperm flora in south-east China as well as the evolution of the Sino-Japanese Flora.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: Sampling information, the distribution of chloroplast haplotypes and chloroplast genetic variation of Pseudotaxus chienii. Table S2: Primer sequences, annealing temperatures, PCR product sizes and gene ID of 14 nuclear loci used in this study. Table S3: Transcriptome sequences corresponding to the loci in Supplementary Data Table S2. Table S4: Functional categories of 14 nulcear loci against four protein databases. Table S5: Description of the prior distribution of parameters from five scenarios in Supplementary Data Fig. S2 used in ABC. Table S6: Description of the prior distribution of parameters from five scenarios in Fig. 2 used in ABC. Table S7: Description of the previous distribution of parameters from six scenarios in Fig. 3 used in ABC. Table S8: Variable sites in seven haplotypes from two cpDNA fragments. Table S9: Nucleotide variation, haplotype diversity and neutrality tests across all 14 nuclear loci. Table S10: Posterior probabilities and credibility intervals of five scenarios in Supplementary Data Fig. S2 used in ABC modelling. Table S11: Posterior probabilities and credibility intervals of five scenarios in Fig. 2 used in ABC modelling. Table S12: Posterior probabilities and credibility intervals of six scenarios in Fig. 3 used in ABC modelling. Table S13: Maximum likelihood estimates and the 95 % highest posterior density intervals of demographic parameters between each two clusters estimated in IMa2. Figure S1: The geographical distribution and haplotype network of chloroplast haplotypes in Pseudotaxus chienii. Figure S2: Five assumed scenarios of population divergence for clusters W, C_P and E tested in DIYABC. Figure S3: Mismatch distribution for cpDNA sequences. Figure S4: The most likely number of clusters inferred by STRUCTURE using LnP(D) and ΔK statistics. Figure S5: STRUCTURE analysis applied to clusters W, C and E when K = 2–4 was assumed. Figure S6: Posterior probabilities for five scenarios used in Supplementary Data Fig. S2 estimated using logistic regression of 1 % of the closest datasets in DIYABC. Figure S7: Posterior probabilities for five scenarios used in Fig. 2 estimated using logistic regression of 1 % of the closest datasets in DIYABC. Figure S8: Prior and posterior distributions of demographic parameters under scenario 4 in Fig. 2 estimated using DIYABC. Figure S9: Posterior probabilities of six scenarios used in Fig. 3 for W, C_P and E estimated using logistic regression of 1 % of the closest datasets in DIYABC. Figure S10: Prior and posterior distributions of demographic parameters for W (scenario 4), C_P (scenario 1) and E (scenario 1) in Fig. 3 estimated using DIYABC. Figure S11: Posterior probability distributions of effective population size (θ) for each cluster and divergence time between each pair of clusters estimated by the IM model.

Funding

This work was supported by the National Natural Science Foundation of China (41461008, 31560064), a grant from Yunnan Provincial Government (2017HA014), and a grant for key discipline (ecology) of Jiangxi Agricultural University.