Background

China’s present-day topography is well known for being elevated in the west and low in the east, which could have resulted from the rapid uplift of the Tibetan Plateau (TP) during the Quaternary Period [1]. As a main driving force of eco-environmental evolution, rapid changes have led to corresponding alternations in both geomorphic and climate types and consequently generated various sources and reservoirs of biological diversity in the region surrounding the TP [2]. Further analysis supported that the climate oscillations accompanying this process resulted in alternation between contraction and expansion due to cyclical variations in glacial intervals, which had an important influence on the current distributions of extant species [3]. Recently, many glacial refugia found in mainland China during the species conservation process also well supported this view [4]. For sedentary and/or special species, populations close to glacier refugia generally have greater genetic diversity than those far from them [5]. Therefore, scientists can infer the evolutionary history of a given species according to the distribution of its genetic information.

In the region of the eastern edge of the TP, the alternating distribution of a series of spectacular north‒south trending mountains and rapidly rushing rivers results in various deep valleys, in which numerous species occur due to the large variation in altitudes. Regions harbouring both ancient and newly originated species are the most famous biodiversity cradles in the world [6]. Studies have suggested that many species retreated here for shelter during the ice ages but migrated to other places after the ice ages since the beginning of the Quaternary [7,8,9], which indicated that the area was a major shelter and possibly an important centre of species origin. In addition, the migration of species, especially vascular plants, directly across the TP would have been impeded by various high mountains with perennial frozen earth [10], indicating that subtropical China, both south of the Qinling Mountains and Huaihe River Line and east of the TP [11], is also an ideal geographical area for investigating species’ refuge locations and migration routes.

Species in these areas are useful for biologists studying evolutionary events [12], and there have been various reports on the evolution of species in the region. Based on the evidence from both ancient vegetation and systematic geography, two main models have been proposed to describe the refugia and postglacial population dynamics of plant species in subtropical China [13]. One is the glacial contraction and postglacial expansion model (CE model) [14], which asserts that species experienced sharp contraction during glacial periods and quick expansion after glacial periods. For example, reconstruction of the palaeontological community showed that the subtropical evergreen broad-leaved forest in China retreated c. 1000 km to the south during the last glacial maximum (LGM, c. 0.021 − 0.018 Ma), while large-scale expansion occurred in the northern high-latitude area after that period [14, 15], which provided direct evidence supporting the CE model. In addition, comparative analysis revealed that southern populations usually had greater genetic diversity than northern populations, possibly resulting from the greater accumulation of ancient and private alleles during the longer evolutionary history compared with that of the northern population [16, 17]. The other is the in situ survival model (ISS model), which considers that there are many shelters during glacial periods and relatively few local expansion events after glacial periods [18,19,20]. Most of the evidence supporting this model mainly comes from research on deciduous broad-leaved forest plants [6, 13, 21,22,23]. In fact, the CE model ignores the possibility of many ecological niches with similar climate conditions, while the ISS model poorly accounts for migration effects.

In addition, different plant types can have different evolutionary histories. Thus, further studies of population evolutionary histories across the eastern edge of the TP and in subtropical China with new plant types would be valuable for understanding species history. However, few studies have simultaneously considered populations across the two regions [24, 25]. Broad-leaved forest trees have received little attention because there are very few with wide and continuous distributions in the two regions due to large variations in climate, elevation, and landforms [26], while other plant types, such as vines, have not received much attention in phylogeographic research due to their weakened role in forests compared with that of tall, large woody trees. Furthermore, the two regions together account for one-third of China’s land area [27], so it is difficult to simultaneously sample at this large scale due to a shortage of available ex situ conservation genetic resources. In addition, underdeveloped road networks on the eastern edge of the TP due to natural barriers are also an important reason for the few samples collected in the past. With the establishment of living germplasm banks of various woody perennial plants and the improvement of transportation infrastructure, systemically studying evolutionary history by integrating samples from the two regions is becoming feasible.

Akebia trifoliata (three-leaf Akebia) is a woody perennial climbing vine in the family Lardizabalaceae [28]. Various studies have suggested that A. trifoliata has many advantages for elucidating the phylogeography of mainland China. First, according to the records from the Plants for a Future database (www.pfaf.org), it originated in China (https://pfaf.org/user/Plant.aspx? LatinName = Akebia + trifoliata), although South Korea and Japan also have populations of A. trifoliata [29], and it is continuously distributed from the eastern edge of the TP to the eastern coast of mainland China; therefore, A. trifoliata is a typical species of the region spanning the eastern edge of the TP and subtropical China [28]. In addition, characteristics such as a small genome [30], a relatively short juvenile stage, a large breeding coefficient, cross-pollination and numerous discernible phenotypic traits make A. trifoliata an ideal model plant, especially for perennial woody species [31, 32]. Third, the availability of sufficient wild germplasm from the whole mainland of China [33] is highly advantageous for systemically studying the phylogeography of A. trifoliata. There have been few reports on the genetic diversity of A. trifoliata, but the employed samples were narrow in scope and usually small in size [34]. Therefore, further study of phylogenetic relationships will be valuable for understanding the evolutionary history of A. trifoliata. In the present study, conserved DNA sequences, including those of the maternally inherited chloroplast gene rps16 and the biparentally inherited nuclear ribosomal internal transcribed spacer (ITS), were used to explore the genetic diversity and population structure of A. trifoliata in the region across the eastern edge of the TP and subtropical China. Our objectives were to determine the locations of the main shelters of A. trifoliata, to infer the migration route, and to identify the evolutionary model responsible for the distribution of extant A. trifoliata populations.

Methods

Plant genotypes

A total of 391 genotypes of 61 populations from the A. trifoliata ex situ conservation germplasm bank located at the Chongzhou Research Station of Sichuan Agricultural University [31] were chosen for the study. These genotypes originated from a large geographical area spanning from 99°57′E to 120°11′E and from 24°49′N to 34°42′N in mainland China.

DNA extraction, amplification and sequencing

Total genomic DNA was extracted according to the previously published CTAB method. The rps16 chloroplast gene and ITS were selected as maternally and biparentally inherited molecular markers, respectively [35, 36]. The corresponding primers were synthesized and then subjected to polymerase chain reaction (PCR) according to a previously described reaction system and program [31]. Sequencing reactions were conducted with corresponding forward and reverse primers commercially provided by Tsingke Biotechnology Co., Ltd. (Chengdu, China).

Population genetic diversity and genetic differentiation

For the rps16 and ITS data, the sequence was first manually edited using BioEdit 7.0.1 [37] and then aligned with MAFFT 7.2.2 software [38]. Finally, the conserved region of the aligned sequence was extracted using Gblock v 0.91b [39]. We used DnaSP 5.10 software to calculate genetic information statistics, including haplotype number, Hd and π, of the conserved sequence [40]. In addition, we used PERMUT 2.0 [41] to calculate the hT and hS between populations with 1,000 repeats and to assess the difference between the GST and NST of distinct populations, excluding populations with sample sizes less than 3. Analysis of molecular variance (AMOVA) and correlation analysis between genetic distance and geographical distance were performed by Arlequin 3.0 [42] and by GenAlEx 6.5 [43], respectively.

Phylogenetic analyses and molecular dating

The haplotype (ITS haplotypes and rps16 haplotypes) network diagram was constructed using PopART 1.7 [44], and a sampling distribution map was constructed using ArcGIS 10.2 [45]. ML trees were reconstructed according to conserved haplotype sequences using PhyloSuite 1.2.2 [46] with the best-fit models (TPM3 + F + I and TIM3 + F + R2 for rps16 and ITS, respectively) produced from the AIC program of ModelFinder software [47], in which Archakebia apetala was used as the outgroup.

The divergence time of the haplotype lineages of A. trifoliata was determined with a secondary calibration method by BEAST 2 [48]. The first step was to determine the crown age of A. trifoliata. Four matK gene sequences of A. trifoliata and 15 matK gene sequences of 15 different species were employed by searching GenBank (Table S2), consisting of six from Lardizabalaceae, five from Menispermaceae, two from Berberidaceae, one from Eupteleaceae and one from Ranunculaceae. In addition, four calibrated time points, Ranzania-Mahonia-Berberis (45.0 Ma) [49], Chasmanthera-Odontocarya (33.9 Ma) [50], Tinospora-Chasmanthera-Odontocarya (55.2 Ma) [51, 52], and Stephania-Menispermum-Chasmanthera-Tinospora-Odontocarya (89.3 Ma) [53], were used. The chronogram of Ranunculales was produced using BEAST2 with the Yule model, an uncorrelated lognormal relaxed clock [54] and the TVM + F + G4 best-fit model selected by the BIC approach in ModelFinder software [47]. For each BEAST2 analysis, Markov chain Monte Carlo (MCMC) was run for 1 × 108 generations, sampling once every 1000 generations, and the first 10% of the samples were removed as burn-in. The sufficiency and convergence of the sampling results were checked by Tracer 1.7.2.

In the second step, chronograms of A. trifoliata haplotypes were further derived using a method similar to that used to date intraspecific nodes, in which the root nodes were corrected by using the above estimated crown group age (median value) of A. trifoliata in the first step, and A. apetala was still an outgroup. GTR + F + G4 and F81 + F + G4 were the best-fit models for the ITS and rps16 haplotypes, respectively. A. apetala was still an outgroup.

Demographic history

The values of Tajima’s [55], Fu and Li’s [56] and Fu and Li’s [57] of all sequences were calculated by DnaSP 5.10 software [40], and the observed paired difference pattern (mismatch distribution) was also analysed using the same software with a constant population size model. Both the sum of squared deviations (SSD) and Harpending’s raggedness index (HRag) were analysed by Arlequin 3.0, and we further analysed population expansion only when the corresponding p value was greater than the 0.05 level. The expansion time was calculated using the formula T = τ/2µkg, where µ, k and g refer to the replacement rate (s s− 1 yr− 1) of each site in the ITS every year, the length of the sequence and the age of the first reproduction, respectively. In this study, a mean value of µ for shrubs and herbal plants ranging from 3.46 to 8.69 × 10− 9 (s s− 1 yr− 1) [58] was used as the µ value of the A. trifoliata ITS sequence. The g value of A. trifoliata is usually three years.

Ancestral area reconstructions

Based on the chronograms of the ITS and rps16 haplotypes, the geographical diversification of A. trifoliata was further reconstructed via Bayesian binary MCMC (BBM) analysis in RASP 4.x [59]. During this process, the distribution of A. trifoliata was divided into four regions according to natural geographical boundaries: A (the eastern Tibetan Plateau), B (central northern China), C (central China) and D (eastern China). MCMC trees were constructed with 5000 randomly chosen trees from all post-burn-in trees from BEAST2 analysis. The parameters of the BBM analyses were set as follows: null for the root distribution, 10 chains for optimization with the fixed JC model for 5 × 105 cycles, the posterior distribution sampled every 100 generations, and four regions for the permitted maximum value.

Results

Sequence variation, genetic diversity, and genetic structure

The lengths of the rps16 and ITS sequences obtained from 391 samples of A. trifoliata were 744 bp with a 33.9% G + C content and 695 bp with a 63% G + C content, respectively. We first identified a total of 22 rps16 haplotypes (C1 ∼ C22) and 75 ITS haplotypes (R1 ∼ R75) according to 59 and 88 polymorphic sites in the rps16 and ITS sequences, respectively. ITS sequences had greater haplotype diversity (Hd) (0.86), nucleotide diversity (π) (3.07 × 10− 3), total genetic diversity (hT) (0.88), average genetic diversity (hS) (0.31) and fixation index (FST) (0.64) values among the 61 small populations than did rps16 sequences (Table S1). In addition, we also found that for the ITS dataset, the GST (0.44) was lower than the NST (0.68), while for the rps16 dataset, the GST (0.44) was larger than the NST (0.31), which indicated that the ITS dataset had an obvious phylogeographic structure, while the rps16 dataset did not.

In fact, the results of the Mantel test suggested that for the rps16 dataset, there was no obvious correlation between genetic distance and geographical distance among populations (R2 = 0.0005, p = 0.17) (Fig. 1a), while for the ITS dataset, the relationship was significant at the p = 0.05 level, although R2 was only 0.21 (Fig. 1b). Furthermore, the AMOVA results showed that the variation in the ITS region mainly occurred among populations, while that in the rps16 region mainly occurred within populations (Table 1).

Fig. 1
figure 1

Mantel test results for the geographic distance matrix (GGD) and genetic distance matrix (GD) based on (a) the rps16 dataset and (b) ITS dataset

Full size image
Table 1 Molecular variance (AMOVA) of rps16 haplotypes and ITS haplotypes for Akebia trifoliata populations
Full size table

Network and distribution of haplotypes

The TCS network diagrams of the ITS haplotypes exhibited a multiple star-shaped radial pattern, and the haplotypes clearly corresponded to distinct geographical regions (Fig. 2a). In contrast, the distribution of rps16 haplotypes showed a single star-shaped radial pattern (Fig. 2b), which indicated that there was no geographical structure. According to the geographical distribution of the ITS haplotypes (Fig. 2a), no haplotype was common to all regions, while 63 (84%) of the haplotypes were region specific. Among the 63 region-specific haplotypes, 12 (R3 ∼ R10, R33, R73 ∼ R75), 13 (R27 ∼ R32, R36, R38, R39, R41, R67 ∼ R69), 28 (R14 ∼ R16, R21 ∼ R26, R42 ∼ R56, R63 ∼ R66) and 10 (R18 ∼ R20, R57 ∼ R60, R70 ∼ R72) haplotypes were specific to the A, B, C and D regions, respectively (Fig. 2c and Table S1). We further found that 26 haplotypes, including 12 of the A region-specific haplotypes, 13 of the B region-specific haplotypes and one (R11) haplotype common to the A, B and C regions, were derived from R2 by one to ten steps of mutation; 28 of the C region-specific haplotypes were derived from R12 by one to four steps of mutation; and 10 of the D region-specific haplotypes were derived from R61 by one to three steps of mutation. Obviously, R2 is the ancestral haplotype because of its large proportion and wide distribution.

Fig. 2
figure 2

(a) TCS-derived network of genealogical relationships between the 75 ITS haplotypes. (b) TCS-derived network of genealogical relationships between the 22 haplotypes (rps16). The circle size of the two network diagrams is proportional to the sample size of each haplotype. Black filled dots indicate missing haplotypes. The haplotypes from the four regions are represented by four different colours. The dashes on the straight line represent the number of abrupt steps. (c) Sampling distribution map of 61 populations of Akebia trifoliata. Red, yellow, green and blue represent the different geographical locations of A (the eastern Tibetan Plateau), B (central northern China), C (central China) and D (eastern China), respectively

Full size image

Moreover, we detected 22 rps16 haplotypes, 16 (72.7%) of which were region specific, while three (13.6%) rps16 haplotypes (C1, C4 and C6) were common to all four lineages. We also found that C1 produced four B region-specific rps16 haplotypes (C10-C12 and C21), two C region-specific rps16 haplotypes (C8 and C9) and one D region-specific rps16 haplotype (C19) via one mutation step and directly produced two A region-specific rps16 haplotypes (C2 and C22) via multiple steps of mutation. Moreover, all rps16 haplotypes of the D region were simply derived from C1 by only one mutation step. The results suggested that C1 was the ancestral rps16 haplotype. Overall, the distributions of both the ITS and rps16 haplotypes revealed obvious ITS sequence divergence, while the rps16 sequence lacked such divergence across geographical regions.

ML phylogenetic tree of A. trifoliata haplotypes

The maximum likelihood (ML) phylogenetic tree of all 75 A. trifoliata ITS haplotypes with A. apetala as an outgroup had three major branches: I, II and III (Fig. 3a). Branch I mainly consisted of all (12) specific haplotypes from the A region, most (nine out of 13) specific haplotypes from the B region, four specific haplotypes (R21, R22, R46 and R56) from the C region and five other common haplotypes (R1, R2, R11, R35 and R40) in two or three regions; branch II mainly consisted of all (10) specific haplotypes from the D region, two specific haplotypes (R32 and R67) from the B region, four specific haplotypes (R45, R65, R50 and R51) from the C region and one common haplotype (R61) in the B and D regions; and branch III mainly consisted of almost all (24 out of 28) specific haplotypes from the C region, two B region-specific haplotypes (R27 and R31) and six other common haplotypes (R12, R13, R17, R34, R37 and R62) in two or three regions. In contrast, there was no obvious branch on the ML phylogenetic tree of all 22 A. trifoliata rps16 haplotypes (Fig. 3b), in which the haplotypes were evenly distributed among the four regions.

Fig. 3
figure 3

Maximum likelihood tree for A. trifoliata based on (a) ITS haplotypes and (b)rps16 haplotypes; Archakebia apetala was the outgroup. Bootstrap probabilities are indicated on the branches. Red, yellow, green and blue in colour represent A (the eastern Tibetan Plateau), B (central northern China), C (central China) and D (eastern China) region-specific haplotypes, while others in grey represent haplotypes from two to four different regions. I, II and III are the three major branches of the ITS haplotype ML tree (a)

Full size image

Divergence time of A. trifoliata haplotypes

The BEAST2-derived matK chronogram showed that the crown group age of A. trifoliata was 2.06 Ma (Fig. 4), and when it was used as a root prior, the coalescence times of 75 haplotypes (node f in Fig. 5a) and 22 haplotypes (node g in Fig. 5b) were 2.82 Ma and 2.12 Ma, respectively.

Fig. 4
figure 4

BEAST2-derived chronogram of Ranunculales based on matK sequences with calibration points denoted by nodes a-d. The blue bars indicate the 95% HPD credibility intervals for node ages (in Myr ago, Ma). The age estimate (Ma) (mean 95% highest posterior density: HPD) for each node is shown beside the nodes. The posterior probability values are labelled on each branch in turn

Full size image
Fig. 5
figure 5

BEAST2-derived chronograms of ITS haplotypes (a) and rps16 haplotypes (b) calibrated by the crown time of Akebia trifoliata. The age estimate (Ma) (HPD) for each node is shown beside the node. The regions with a greater than 95% posterior probability are indicated by thick branches. The red, yellow, green and blue branches represent haplotypes from the A (the eastern Tibetan Plateau), B (central Northern China), C (central China) and D (eastern China) regions, respectively, and the other black branches except those of the outgroup represent haplotypes from two to four different regions

Full size image

In the ITS chronogram, four nodes (f1-f4) covered 37 haplotypes: 12 specific to the A region, 10 specific to the B region, seven specific to the C region and eight non-region specific; node f5 covered 38 haplotypes: three specific to the B region, 21 specific to the C region, 10 specific to the D region, and four that were non-region specific.

The coalescence time of the five nodes varied from 2.27 Ma to 1.25 Ma, which falls within the early Pleistocene (2.58–0.78 Ma) (Fig. 5a), during which the divergence of all 22 ITS haplotypes and five rps16 haplotypes occurred. However, the remaining 53 (70.67%) ITS haplotypes and 17 (77.27%) rps16 haplotypes originated during the middle Pleistocene (0.78 − 0.12 Ma).

Demographic history

All neutrality test values, including Tajima’s D, Fu and Li’s D and Fu and Li’s F, of all ITS and rps16 datasets in the whole population and all four regions were negative, although those of the rps16 haplotypes were not significant in the D region (Table 2). The mismatch distribution of the ITS haplotypes was unimodal in the A and D regions (Fig. 6), while that of the rps16 dataset was peak free in the whole population and each region (Fig. S1). The results of the fit test using the sudden expansion model showed that the SSD (0.0267) and HRag (0.322) values of the ITS haplotypes in the whole population were not significant at the p = 0.05 level, which suggested the occurrence of one expansion event (possibly 150 Ka). However, the rps16 haplotypes of A. trifoliata mainly experienced balancing selection throughout their evolutionary history because there were no detectable SSD or HRag values in the whole population or any of the four geographical regions.

Table 2 The results of neutrality tests for various populations
Full size table
Fig. 6
figure 6

Mismatch distribution analyses of the five ITS clades consisting of both four regional populations and the whole population, in which the observed mismatch frequencies and best-fit curves of the sudden expansion model are shown

Full size image

Ancestral area reconstructions

According to the topology of the reconstructed ancestral area of the ITS region, the marginal probabilities of the A region at h1, h2, h4 and h8 were 0.64, 0.45, 0.52 and 0.77, respectively (Fig. 7a). The populations of other regions, especially the C region, colonized from the A region. The marginal probability of B at h3 was the largest (0.25), which suggested that region C was also partly colonized by the population from region B. The marginal probabilities of C at h5, h6 and h7 were 0.97, 0.97 and 0.95, respectively, indicating that the population of the D region was largely derived from the C region.

Fig. 7
figure 7

Ancestral area reconstructions based on the Bayesian binary Markov chain Monte Carlo (BBM) method performed in the RASP using the BEAST2-derived chronogram of Akebia trifoliata based on (a) ITS and (b)rps16 haplotypes (see Fig. 5a and b). The circle at each node represents the marginal probability of each alternative ancestral region derived from the BBM analysis. The results are based on a maximum area of four. Inferred dispersal events are indicated by blue arrows. (c) Distribution area map of Akebia trifoliata: A (the eastern Tibetan Plateau), B (central northern China), C (central China) and D (eastern China). (d) Speculative diagram of expansion or propagation events in Akebia trifoliata, with bold solid lines representing the main expansion or propagation events

Full size image

Similarly, the marginal probability of the A region at i1 was 0.97 in the rps16 topology (Fig. 7b), also suggesting that the A region was the ancient distribution area of A. trifoliata. The value (0.40) at i2 indicated that the populations of both the B and C regions could have colonized from the A region. The marginal probability of the C region at i4 was 0.39, suggesting that the population of the D region mainly colonized the C region. We also noticed that the information at both i3 and i5 was confusing, which could result from the unclear pedigree of rps16.

BBM analyses showed the potential for many seed dispersal events (see the blue arrow in Fig. 7), with only the absence of directional dispersal events from the population in the D region to the A region. Generally, the species migration route was from the ancestral distribution A region simultaneously to the B and C regions, then from the B region to the C region, and finally from the C region to the D region (Fig. 7c and d).

Discussion

The characteristics of the population and conserved DNA sequences help elucidate the phylogeographic structure of A. trifoliata

Genetic diversity parameters such as the hT, Hd and π of conserved DNA sequences provide the best evidence for elucidating the phylogeographic structure of a given species or a larger classification unit, such as genus, family and even order [40]; however, population characteristics, including geographical distribution, breeding system and effective size, also have an important influence on genetic diversity [60]. Whether we can accurately detect phylogeographic structure is largely dependent on two key factors: population characteristics and DNA sequence conservation.

Previous studies demonstrated that A. trifoliata originated in China and exhibited a constrained continuous distribution from the eastern edge of the TP to the eastern coast of mainland China and even to South Korea and Japan [28, 29]. In addition, the biological characteristics [30] and availability of sufficient germplasms via the ex situ living conservation method [31] have been reported. The 391 genotypes employed in this study originated from various places in mainland China, and there was a small geographic interval between adjacent samples (Fig. 2c), providing an ideal population for deducing the evolutionary history of species, especially climbing vine plants, in mainland China.

Molecular markers based on conserved DNA elements, including maternal chloroplast and biparental nuclear genomic markers, have been widely employed to investigate population genetic diversity. In this study, a previously reported chloroplast gene (rps16) and a nuclear DNA marker (ITS) were used to determine the genetic diversity of the A. trifoliata population [35, 36]. We first checked 22 and 75 haplotypes from 59 to 88 polymorphic sites in the rps16 and ITS sequences, respectively (Table S1). Based on the haplotypes, the genetic diversity (hT=0.881, π × 10− 3=3.07) of A. trifoliata at the nuclear DNA level (Table S1) was greater than that of Zingiber officinale [61], Pinus attenuata [62] and Antirrhinum charidemi [63]. We found that the genetic diversity (hT=0.313, π × 10− 3=0.90) of the A. trifoliata population at the chloroplast sequence level was lower than that at the nuclear DNA level (Table S1), which could result from the low variation rate of the chloroplast compared with the nuclear DNA sequence [64]. This also provides a reasonable explanation for why the ITS and rps16 sequences are used to determine the genetic structure of intraspecific and interspecific populations, respectively [63, 65]. To obtain more information on different DNA sequences, we simultaneously used rps16 and ITS markers to determine genetic diversity in the present study, which would be beneficial for elucidating phylogeographic structure.

A. trifoliata populations exhibited clear phylogeographic structure

Usually, molecular variation [66], the relationship between GGD and GD [67], the number of regional private haplotypes [19] and the relative sizes of GST and NST [20] are important information for describing population structure. First, we found that there was large molecular variation in the ITS sequence among different regional populations, while the variation in the rps16 sequence was small (Table 1). Second, we also found a significant positive relationship between GGD and GD based on the ITS sequence, while there was no obvious relationship between them based on the rps16 sequence (Fig. 1); in this study, the weak relationship could partly result from uneven samples from various regions (Table S1). Third, some regional private haplotypes of both ITS and rps16 existed in each region (Fig. 2a and b; Table S1) and even on each branch of the ML tree (Fig. 3), possibly related to the high differentiation of different regions (Table 1). Finally, for the ITS sequence, the GST (0.437) was obviously lower than the NST (0.677). Taken together, these findings suggest that there was no obvious geographical structure at the chloroplast DNA level, possibly resulting from different inheritance modes [64], dispersal strategies [68] and slow variation rates because of their high conservation [65]; however, the clear phylogeographic structure of the A. trifoliata population at the nuclear DNA level is beyond all reasonable doubt.

Multiple historical ice shelters could have affected the current distribution status of A. trifoliata

Studies have shown that plant populations around glacial habitats usually have high genetic diversity and many private alleles or haplotypes [17, 69]. Here, we found that many haplotypes of both ITS and rps16 were region-specific (Figs. 2a and b and 3), which indicated that there were multiple refuge regions. According to the number of haplotypes, Hd and π, three putative main refuge regions (the Qinba Mountains, Nanling Mountains and Yunnan-Guizhou Plateau) can be suggested.

First, the Qinba Mountains could be the main refugia region of A. trifoliata because the populations, including Bazhong from the A region and both Ankang and Xian from the B region, had many ITS haplotypes, high Hd and large π, and they commonly showed the ancestral haplotype R2 (Fig. 2a), which was also well supported by the high polymorphism of A. trifoliata resources previously reported in the region [34]. Second, the Nanling Mountains, as representative mountains in the subtropical region and one of the major biodiversity hotspots in China [70], could also be a refuge region because the Hezhou, Yongzhou, Guilin, and Ganzhou populations from the C or D region had similarly many haplotypes and high Hd and π values (Fig. 2b), and R12 was their common haplotype. Various studies have suggested that the Nanling Mountains provide shelter for many plants [71], such as Castanopsis eyrie [72] and Loropetalum chinense [73], which supports the Nanling Mountains as a historically important refuge region for A. trifoliata. Third, the Yunnan-Guizhou Plateau is still a refuge region because of the very high π values for the rps16 sequence in addition to the ITS sequence in both the Bijie and Zunyi populations from the C region [74, 75].

At present, the CE and ISS models represent two prevalent views about the contraction and expansion of plant species in the subtropical region of China during the ice age cycle [14, 20]. Unfortunately, the CE model largely ignores the existence of many similar ecological niches, while the ISS model overlooks the effects of migration on species distributions. In this study, among the 61 populations, 55 had a sample size larger than two, of which 26 populations had an Hd value of ITS greater than 0.6 (Table S1). Of these 26 populations, 12 were geographically close to the three main refuge regions, while 14 were far from the main refuge regions, which clearly showed that the three main refuge regions (the Qinba Mountains, Nanling Mountains and Yunnan-Guizhou Plateau) harboured many independent, small ice shelters. In addition, many haplotypes, including 53 ITS haplotypes and 17 rps16 haplotypes, originated during the middle Pleistocene (0.78 − 0.12 Ma), while the other haplotypes were produced before this stage, which indicated that whole or partial population expansion occurred in the past. The results of the neutrality test and ITS mismatch distribution analysis clearly demonstrated that population expansion events occurred in at least the A and D regions. The view of three main refuge regions with some independent, small ice shelters provides a reasonable explanation for the continuous and wide distribution of A. trifoliata in many regions of mainland China.

Putative migration route and differentiation events of A. trifoliata

Molecular dating can provide information about the differentiation time of conserved DNA sequences, which is helpful for inferring population differentiation processes [76]. In this study, molecular dating revealed that the differentiation times of the common ITS and rps16 haplotypes in A. trifoliata were 2.82 Ma and 2.12 Ma, respectively (Fig. 5a and b), and both occurred during the period spanning the late Pliocene to early Pleistocene. During this period, the third uplift of the Qinghai‒Tibet Plateau [77], the Asian monsoon [78] and the Asian drought [79] gradually intensified, causing habitat fragmentation and species differentiation [80]. For example, the Hainan Island population of Tetrastigma hemsleyanum differentiated at 2.78 Ma [22]. Second, most (72.2%) of the haplotypes of A. trifoliata differentiated in the middle Pleistocene (0.78 − 0.12 Ma), which is similar to the haplotype differentiation time of Sargentodoxa cuneata in this region [76], but the remaining 27.8% of the A. trifoliata haplotypes differentiated before this period (Fig. 5a and b). Hence, the putative divergence times of the common ITS and rps16 haplotypes in A. trifoliata could be reasonable. In addition, both the frequency and distribution of haplotypes (Fig. 2a and b) suggested that region A was the ancestral region because of the high frequency and wide distribution of the ancestral haplotypes R2 and C1 (Table S1).

The results of further BBM analysis revealed the putative migration route (Fig. 7c and d), and in this model, the route from the ancestral A region simultaneously from both the B and C regions, from the B region to the C region, and then from the C region to the D region was the main route of A. trifoliata migration, and the corresponding reverse route could be a minor migration route. In fact, this hypothesis of species migration off the Qinghai‒Tibet Plateau [81] also agreed well with the putative migration model. In addition, we also found minor migration from the ancestral A region to the D region, while there was no reverse route. Finally, migration from west to east could be a major direction, while reciprocal migration between the south and the north could be minor, which could have resulted from climate oscillation in the past [17]. The main route and direction of A. trifoliata migration could be closely related to its natural spread through west-east-flowing rivers, especially the Yangtze River [31], animals such as birds [82] and even commercial action [83]. Moreover, incomplete coverage of glaciers during this process [84] may also explain the widespread differentiation of A. trifoliata.

The last concern was the origin of the A. trifoliata distributed in Korea and Japan, and in our opinion, these A. trifoliata possibly migrated from the D region of mainland China, which could be resolved by a future study with samples from these two countries.

Conclusions

Both conserved DNA sequences and continuously and widely distributed large populations provide a valuable opportunity to elucidate the evolutionary history of a given species. In this study, a maternally inherited chloroplast gene (rps16) and a biparentally inherited nuclear ITS sequence were simultaneously employed to assess the phylogeographic structure of A. trifoliata populations in the region across the QTP eastern edge and subtropical China, where more than 90% of publications about the species originated. Our results suggested that A. trifoliata originated on the eastern edge of the QTP and then migrated eastwards to the east coast of mainland China, even to Korea and Japan, and that multiple ice shelters in the Qinba Mountains, Nanling Mountains and Yunnan-Guizhou Plateau region supported its proliferation throughout subtropical China. Overall, the unique origin, long-distance migration, high intraspecific divergence, multiple ice shelters and regional expansion of A. trifoliata resulted in its continuous and wide distribution.