•Numerous temperate plants now distributed across Eurasia are hypothesized to have originated and migrated from the Qinghai-Tibet Plateau (QTP) and adjacent regions. However, this hypothesis has never been tested through a phylogeographic analysis of a widely distributed species. Here, we use Hippophaë rhamnoides as a model to test this hypothesis.
•We collected 635 individuals from 63 populations of the nine subspecies of H.rhamnoides. We sequenced two maternally inherited chloroplast (cp) DNA fragments and also the bi-paternally inherited nuclear ribosomal ITS.
•We recovered five major clades in phylogenetic trees constructed from cpDNA and internal transcribed spacer (ITS) sequence variation. Most sampled individuals of six subspecies that are distributed in northern China, central Asia and Asia Minor/Europe, respectively, comprised monophyletic clades (or subclades) nested within those found in the QTP. Two subspecies in the QTP were paraphyletic, while the placement of another subspecies from the Mongolian Plateau differed between the ITS and cpDNA phylogenetic trees.
•Our phylogeographic analyses supported an ‘out-of-QTP’ hypothesis for H.rhamnoides followed by allopatric divergence, hybridization and introgression. These findings highlight the complexity of intraspecific evolutions and the importance of the QTP as a center of origin for many temperate plants.
A central objective of biogeography is to identify the underlying causes of how organisms are distributed geographically (Cox & Moore, 2000). The origins and migration patterns of Northern Hemisphere (NH) temperate plants, which exhibit the greatest diversification in the highlands of Asia, have attracted the attention of numerous botanists for many years (e.g. Wulff, 1943; Ozenda, 1988; Donoghue et al., 2001; Donoghue & Smith, 2004). Several have suggested that the majority of these temperate groups originated in the Qinghai-Tibet Plateau (QTP) and adjacent highlands, when global temperature decreased and the QTP uplifted extensively during the Neogene (Wulff, 1943; Ozenda, 1988; Wu, 1988; Axelrod et al., 1996; Wu & Wu, 1996; Zhang et al., 2001, 2004; Kadereit et al., 2008). Evidence in support of this hypothesis has been sought from phylogenetic analyses of genera containing species that are disjunctly and widely distributed in the NH. Three different biogeographic patterns have emerged from such analyses. The first indicates that most of such genera originated in the QTP and adjacent regions, and then migrated to other NH regions where they gave rise to daughter species (e.g. Zhang et al., 2007, 2009; Xu et al., 2010). This is especially true for numerous forest genera containing several lineages that migrated to North America and other temperate regions at different times during the past 30 million yr (Donoghue & Smith, 2004). The second pattern indicates that the disjunct distribution of some genera in the NH originated from local relics of the once continuous Arcto-Tertiary, Tethyan or boreal floras (e.g. Sun et al., 2001; Mao et al., 2010). The third pattern suggests that a limited number of genera originated in other regions of the world and diversified greatly after their ancestors reached the QTP (e.g. Liu et al., 2002; Tu et al., 2010). These different patterns, therefore, suggest that biogeographic connections between the QTP and other NH regions are more complex than previously thought.
Previous phylogenetic studies have focused on species distributions within genera and not below the species level. However, phylogenetic analyses of populations from a single species across its entire range can provide important insights into the origins and migration of organisms, especially over relatively recent historical time (Abbott et al., 2000; Avise, 2000; Qiu et al., 2011). Here we report a phylogenetic analysis of Hippophaë rhamnoides (Sea buckthorn) to test the ‘out-of-QTP’ hypothesis for a temperate plant species. Hippophaë is a small genus of Elaeagnaceae, comprising seven species (Rousi, 1971; Bartish et al., 2002; Swenson & Bartish, 2002). Two of the species are putative homoploid hybrids (H. goniocarpa and H. litangensis), and their taxonomic status is uncertain (Swenson & Bartish, 2002). All species of Hippophaë are restricted to the QTP region and adjacent areas, except for H. rhamnoides, which contains nine subspecies (Swenson & Bartish, 2002; Lian et al., 2003) that are distributed widely but fragmentally in eastern Asia (ssp. yunnanensis, wolongensis, sinensis and mongolica), central Asia (ssp. turkestanica), Asia Minor (ssp. caucasica) and Europe (ssp. carpatica, fluviatilis and rhamnoides; see Fig. 1). Its Eurasian range makes H. rhamnoides a good candidate to examine biogeographic connections between the QTP and Europe through intervening mountain ranges in central Asia and Asia Minor (Kadereit et al., 2008). Although several phylogenetic studies have indicated east-to-west directional dispersal in the genus (Bartish et al., 2000, 2002; Sun et al., 2002), this hypothesis has not been tested at the species level.
Fruits of H. rhamnoides are juicy favouring dispersal by birds and the possibility of long-distance dispersal leading potentially to inter-regional hybridization (Rousi, 1971). Such long-distance dispersal and inter- regional hybridization is likely to be further promoted by strong pollen-mediated gene flow because all subspecies of H.rhamnoides are dioecious and wind-pollinated (Rousi, 1971; Hyvönen, 1996; Lian et al., 1998; Bartish et al., 2000). Two previous phylogenetic analyses of the genus Hippophaë based on chloroplast DNA restriction fragment length polymorphisms (Bartish et al., 2002), and nuclear ribosomal internal transcribed spacer (ITS) sequence variation (Sun et al., 2002), included only single representatives of each subspecies of H. rhamnoides and suggested that they clustered as a monophyletic lineage, in spite of being genetically distinct. These subspecies occupy different habitats from high mountains (c. 3800 m) to low-altitude regions (300 m) and also deserts or seashores (0 m) (Lian et al., 1998). Bartish et al. (2006) previously conducted a phylogeographic study of H.rhamnoides throughout its range in Europe and Asia Minor. Although this study was highly informative of the demographic history of the species in the western part of its range, it was not, of course, informative of the species’ biogeographic history across Eurasia, especially in the eastern Asian mountains.
In the present study, we used maternally inherited chloroplast DNA (cpDNA) (Bartish et al., 2002) and bi-parentally inherited ITS data in an attempt to reconstruct historical migration routes and divergence patterns of H. rhamnoides. These two sets of molecular markers have been shown in studies of other Hippophaë species to be highly effective in tracing intraspecific divergence and detecting possible hybridization and introgression events (Wang et al., 2008; Jia et al., 2011). Using phylogenies constructed from sequence variation of both markers we aimed to answer the following specific questions. (1) How many intraspecific clades can be recovered in central and eastern parts of the range of H. rhamnoides, and are these clades consistent with the morphological delimitation and geographical distribution of recognized subspecies? (2) Do intraspecific cpDNA divergence patterns reflect dispersal and migration routes of the species from the QTP to other regions? (3) Are the ITS and cpDNA phylogenies consistent with each other? If not, how much hybridization and introgression might be indicated by noncongruent topology?
Materials and Methods
We collected leaves of 635 individuals of H.rhamnoides L. from 63 localities (Supporting Information Table S1). Our samples include all nine subspecies acknowledged within this species. Because previous studies by Bartish et al. (2002, 2006) indicated that four subspecies (ssp. carpatica, caucasica, rhamnoides and fluviatilis) in Europe and Asia Minor comprise a monophylogenetic lineage, we used 27 individuals collected from nine localities to represent these subspecies (Table S1). We focused our sampling on the remaining five subspecies (54 populations) occurring in the QTP, central Asia and northern China (Fig. 1, Table S1). Based on morphology, each sampled individual from these populations was assigned to one of the subspecies yunnanensis, wolongensis, sinensis, mongolica and turkestanica. We sampled 10–16 trees per population with sampled trees located > 100 m apart in a population. Fresh leaves were dried immediately in the field using silica gel and voucher specimens were deposited in Lanzhou University.
DNA extraction, amplification and sequencing
We used DNeasy™ Tissue Kit (Qiagen) to isolate total genomic DNA from dried leaves and followed Taberlet et al. (1991) and Hamilton (1999) to amplify two cpDNA fragments (trnL-F and trnS-G, respectively), and White et al. (1990) to amplify the nuclear ITS region. Polymerase Chain Reactions (PCR) were performed in 25 μl reaction mixture volumes using reagents and manufacturer’s instructions for Taq polymerase (Takara, Dalian, China) in the GeneAmp® PCR System 9700 (Applied Biosystems, Hayward, CA, USA). PCR cycling programmes followed Jia et al. (2011). PCR products were purified using TIANquick Midi Purification Kit (Tiangen, Beijing, China) and sequencing reactions were conducted with the same PCR primers described above and ABI Prism Bigdye™ Terminator v.3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA, USA). Sequences were obtained using an ABI 3730XL DNA Analyzer.
Elaeagnus umbellata Thunb. was chosen as outgroup in the phylogenetic analyses, based on previous molecular and morphological analyses on Hippophaë of Elaeagnaceae (Bartish et al., 2002). Three congeneric taxa (H. salicifolia, H. neurocarpa ssp. neurocarpa and H. neurocarpa ssp. stellatopilosa) were also included. The sequences of these taxa were taken from GenBank. Sequences from two cpDNA regions (trnL-F and trnS-G) were concatenated into a matrix.
We analyzed phylogenetic relationships of the different cpDNA and nuclear ITS sequences through maximum parsimony (MP) and Bayesian analyses using Paup* v.4.0b10 (Swofford, 2002) and MrBayes v.3.0 (Huelsenbeck & Ronquist, 2001; Ronquist & Huelsenbeck, 2003), respectively. GapCoder (Young & Healy, 2003) was used to edit indels as separate characters for inclusion in MP and Bayesian analyses. The MP analyses were performed with characters equally weighted through an heuristic search: 10 replicates of random addition of sequences with ACCTRAN character optimization, MULPARS + TBR branch swapping and STEEPEST DESCENT options on. We calculated bootstrap percentages (BP) using 1000 replicates (Felsenstein, 1985). Bayesian analyses were performed over 5 × 106 generations with one cold and three incrementally heated Monte Carlo Markov chains (MCMCs). Best-fit models of nucleotide substitution for cpDNA and ITS matrices were determined to be TPM1uf + G and SYM + G, respectively, by the AIC using jModelTest v.0.1.1 (Guindon & Gascuel, 2003; Posada, 2008). A standard discrete model (Lewis, 2001) was applied to the indel matrix. Model parameters were unlinked across partitions. We used defaults for the heating scheme as well as for priors on the rate matrix, gamma shape parameter, the proportion of invariable sites and branch lengths. Dirichlet distributions were used to model base frequency parameters and uninformative priors placed on tree topology. We sampled one tree per 500 generations. After discarding the first 2500 trees out of the 10 001 trees as burn-ins, the remaining trees were used to estimate posterior probability (PP).
In order to detect genealogical relationships among sequences with shallow genetic divergences, we also constructed cpDNA and ITS haplotype networks using a statistical parsimony algorithm described by Templeton et al. (1992) as implemented in Tcs v.1.21 (Clement et al., 2000). As the Tcs program collapses sequences that differ only by additive sites, we grouped the 78 different ITS sequences recorded into 28 types and manually selected one representative sequence possessing the lowest number of additive sites for each type (Table S2). These 28 general ribotypes were used in further analyses to reduce bias introduced by sequence ambiguities (Koch & Matschinger, 2007). We ran Tcs with a default parsimony connection limit of 95%. In the case of ITS data, after starting initially with the default limit (11 internal steps allowed), we added ribotypes R68, R76 and R77 with the 90% confidence interval (five steps) and finally R55 and R56 when fixing the connection limit at 33 steps.
The datasets obtained were examined to see if they fitted the hypothesis of a molecular clock using a likelihood-ratio test (LRT; Huelsenbeck & Rannala, 1997) implemented in Paup* v.4.0b10 (Swofford, 2002) and comparing the log likelihood (loge L) of the ML trees with and without assuming a molecular clock. A molecular clock could not be rejected for both cpDNA and ITS data (cpDNA: TPM1uf + G, 2logeLR = 67.889, df = 51, P = 0.057; ITS: SYM + G, 2logeLR = 77.053, df = 80, P = 0.573). We therefore used Beast v.1.5.4 (Drummond et al., 2002; Drummond & Rambaut, 2007) to estimate genetic divergence. We used GTR + G substitution model, a fixed molecular clock, a constant population size coalescent tree prior and a UPGMA starting tree for both datasets. We sampled all parameters once every 2000 steps from 20 × 106 MCMC steps after a burn-in of 5 × 106 steps. The program Tracer (Rambaut & Drummond, 2007) was used to examine convergence of chains to the stationary distribution and the analysis was repeated before combining the two independent runs. The cpDNA substitution rates for most angiosperm species have been estimated to vary between 1 and 3 × 10−9 substitutions per site per year (s s−1 yr−1) (Wolfe et al., 1987), while those for nrITS in shrubs and herbal plants vary between 3.46 and 8.69 × 10−9 s s−1 yr−1 (Richardson et al., 2001). Given the uncertainties in these rate values, we used normal distribution priors with a mean of 2 × 10−9 and a SD of 6.080 × 10−10 for cpDNA, and a mean of 6.075 × 10−9 and a SD of 1.590 × 10−9 for ITS to cover these rate ranges within the 95% range of the distribution for our estimation of divergence times of major clades.
We ran Bayes-DIVA analyses using Rasp v.2.0b (Reconstruct Ancestral State in Phylogenies; Yu et al., 2010, 2011) to infer the biogeographic history of H. rhamnoides based on the phylogeny constructed from cpDNA and nrITS. In this analysis we defined five distribution regions for the sampled populations: (Q) the QTP region (including Tibet, Qinghai, northwest Yunnan, southwest Sichuan, south Gansu; populations 1–28); (N) northern China (including east Gansu, Inner Mongolia, Shaanxi, Shanxi, Hebei, Liaoning; populations 29–43); (M) the Mongolian Plateau (including north Xinjiang, Mongolia, Siberia; populations 44–47); (C) central Asia (including west Xinjiang and west Tibet; populations 48–54); (E) Europe and Asia Minor (populations 55–63) (Figs 1, 3). We loaded 10 001 trees previously produced in MrBayes v.3.0 (Huelsenbeck & Ronquist, 2001; Ronquist & Huelsenbeck, 2003) and chose the F81 model for the Bayesian MCMC analyses, allowing for different rates of change among ancestral areas.
Population genetic analyses
Population gene diversity (HS, HT) and between-population divergence (GST, NST) were estimated within each region using the program Permut with 1000 permutations tests (Pons & Petit, 1996; http://www.pierroton.inra.fr/genetics/labo/Software/ PermutCpSSR). A significant difference between NST and GST may suggest a significant phylogeographic structure (Pons & Petit, 1996). We assessed genetic differentiation between all pairs of geographical regions using ΦST as estimator following analyses of molecular variance (AMOVA; Excoffier et al., 1992) in Arlequin v.3.1 (Excoffier et al., 2005). We also conducted two types of hierarchical (3-level) AMOVAs; that is, with the highest level partitioned by either subspecies or regions. Finally, we calculated haplotype diversity (h) and nucleotide diversity (π) for each population and each subspecies using Arlequin.
A mismatch distribution analysis (MDA; Slatkin & Maddison, 1989; Rogers & Harpending, 1992; Schneider & Excoffier, 1999) was conducted to examine the demographic expansions of the five major cpDNA clades identified in the phylogenetic analyses. As population structure has a limited effect on the mismatch distribution (Rogers, 1995; Bernatchez, 2001), we pooled all haplotypes of each clade and did not consider their frequencies. We used 1000 parametric bootstrap replicates to generate an expected distribution using a model of sudden demographic expansion (Excoffier et al., 2005), to calculate the sum of squared deviations (SSD) and raggedness index (HRag) of Harpending (1994) between observed and expected mismatch distributions and to obtain 95% confidence intervals (CIs) around τ. We also calculated Tajima’s D (Tajima, 1989) and Fu’s FS (Fu, 1997) to assess possible expansions. The D and Fs statistic should have large negative values within a clade under the expansion hypothesis due to an excess of rare new mutations. We calculated significance of the tests with 10 000 replicates. All of these demographic tests were performed using Arlequin v.3.1 (Excoffier et al., 2005). When sudden expansions were detected, we used the equation τ = 2ut (Rogers & Harpending, 1992; Rogers, 1995) to estimate expansion times, where t is the expansion time in number of generations, τ is the mode of the mismatch distribution, and u is the mutation rate per generation for the whole analyzed sequence (i.e. chlorotype). We calculated u according to u = μkg, where μ is the substitution rate per nucleotide site per year, k is the average sequence length of the DNA region under study, and g is the generation time in years. We assumed the generation time of H. rhamnoides to be 5 yr (Bartish et al., 2006). We again adopted the substitution rate of 2.0 × 10−9 s s−1 yr−1 for expansion estimations.
From the 635 individuals sampled, we recovered 27 and 29 unique sequences for the trnL-F and trnS-G fragments, respectively. After combining both sequences we identified a total of 49 chlorotypes (C01–C49). Sequence lengths of these haplotypes varied from 1357 to 1514 bp with 59 nucleotide substitutions and 11 indels (1–163 bp) recorded. Thirty-four haplotypes (69.39%) occurred in not more than one population while the remaining 15 (30.61%) were shared between populations (Table S1). However, all chlorotypes were subspecies-specific and therefore none was shared by any two subspecies.
Phylogenetic trees constructed using MP and Bayesian methods were largely consistent in topology (Fig. 2a). All haplotypes from H. rhamnoides comprised a monophyletic lineage with five clades (A–E) showing no exact ‘one to one’ match with subspecies. All haplotypes of ssp. yunnanensis clustered into three major clades (A, B and C) apart from C32 (found in population 11) which was placed in clade E. Haplotypes placed in clades A, B or C were exclusively distributed in the QTP region (Fig. 1). Clade E comprised most haplotypes of ssp. sinensis, plus one from ssp. mongolica (C26) in addition to the one from ssp. yunnanensis (C32) mentioned above. However, C08 of ssp. sinensis nested within clade B. A monophyletic subclade comprising four haplotypes of ssp. wolongensis also nested within clade B. Clade D comprised two subclades: one consisted of haplotypes of ssp. turkestanica distributed in central Asia, and the other included ssp. caucasica, carpatica, fluviatilis and rhamnoides distributed in Asia Minor and Europe (Figs 1, 2a). The haplotype network (Fig. S1a) was largely congruent with the phylogenetic trees (Fig. 2a), but depicted relationships between haplotypes in more detail. The dating analysis estimated that the five major clades diversified from the late Miocene to the late Pliocene (Fig. 2a), with the first (node a1) and last major events (node a4) occurring 5.46 (95% HPD: 2.23–10.10) million yr ago (Ma) and 2.05 (95% HPD: 0.72–3.86) Ma (Table S3), respectively.
We recovered 78 different ITS sequences (ribotypes, R01–R78), of which 50 (64.10%) were unique to single populations. A total of 108 base substitutions and five indels (1–2 bp) were found and the two phylogenetic analyses generated topologically similar trees (Fig. 2b). Two ribotypes (R55 and R56) of ssp. yunnanensis clustered with H. neurocarpa ribotypes into a clade also containing H. salicifolia ribotype with high support, indicating possible past hybridization and introgression between these species. The other 76 ribotypes comprised a monophyletic H. rhamnoides lineage with high support. However, phylogenetic relationships between subspecies differed somewhat from those inferred from cpDNA sequences.
Ribotypes from ssp. yunnanensis clustered into three basal clades (I, II and III) with those of ssp. wolongensis also placed within clade III. Clade V consisted of only ribotypes from ssp. sinensis, while clade IV included all the ribotypes from the other six subspecies. Three ribotypes were shared between subspecies: R13 between ssp. yunnanensis (occurred in populations 6, 10, 11, 13, 14 and 15) and ssp. wolongensis (population 17), R16 between ssp. wolongensis (population 16) and ssp. sinensis (population 21), R08 between ssp. turkestanica (populations 48, 49, 50, 51 and 52) and ssp. mongolica (populations 45, 46 and 47) and R77 among ssp. caucasica (populations 57, 58 and 59), ssp. carpatica (population 61) and ssp. rhamnoides (population 63). R18, the only ribotype of ssp. sinensis placed in clade III, occurred in population 21. The first divergence within H.rhamnoides (node b1; Fig. 2b) resulting in the origin of clade II and the ancestor of clades III, IV and V in the ITS tree, was estimated to have occurred 3.02 (95% HPD: 1.42–5.18) Ma. The second divergence (node b2) occurred 2.56 (95% HPD: 1.22–4.33) Ma, while the third divergence (node b3) occurred 2.44 (95% HPD: 1.07–4.19) Ma (Table S3).
Biogeographic analyses and intersubspecies differentiation
Bayes-DIVA analyses of cpDNA data support the QTP region (area Q) as the ancestral area for H. rhamnoides (Fig. 3b). This species most likely diversified early in the QTP and subsequently multiple dispersals (for each node only the most likely reconstruction was considered) (Fig. 3a,b) occurred from this area to central Asia (C), Asia Minor/Europe (E), northern China (N) and the Mongolian Plateau (M), respectively. Bayes-DIVA analysis based on nrITS data resulted in similar reconstructions, but required additional dispersals events from central Asia to the Mongolian Plateau (Fig. 3a,c). However, neither Bayes-DIVA analysis based on cpDNA and nrITS detected any distinct vicariance signal, although vicariance between C and E or between N and M seem to be likely.
For both cpDNA and ITS datasets, a significantly larger NST than GST value estimated across all populations indicated that genetic variation of the species was geographically structured across its distribution (Table 1). The highest genetic diversity occurred in the QTP (HT = 0.955 and 0.811 for cpDNA and ITS data, respectively), corresponding to the occurrence of the most divergent haplotypes (Fig. 1). Pairwise ΦST showed high levels of divergence among regions for both cpDNA and ITS sequences (Table 2) and further analysis by hierarchical AMOVA showed that for cpDNA 54% of the total genetic variation was partitioned by subspecies, and 33% partitioned by regions (Table 3), while for ITS data, 77% of the total genetic variation was attributed to the differences among subspecies and 59% to the differences among regions.
Table 1. Estimates of average gene diversity within populations (HS) of Hippophae rhamnoides, total gene diversity (HT), interpopulation differentiation (GST), and number of substitution types (NST) for chlorotypes and ribotypes across regions
ITS, internal transcribed spacer; #Pop, no. of populations; N, no. of plants; n, no. of chlorotypes/ribotypes.
*, P < 0.005.
Qinghai-Tibetan Plateau (QTP)
Table 2. Pairwise genetic differentiation (ΦST) among regions estimated from internal transcribed spacer (ITS) sequences (upper part) and cpDNA sequences (lower part)
All values are significant at the 0.01 level in a permutation tests (1000 permutations).
Qinghai-Tibetan Plateau (QTP)
Table 3. Analysis of molecular variance (amova) of chlorotypes and ITS ribotypes for Hippophae rhamnoides populations, partitioned by subspecies and regions, respectively
Source of variation
df, degrees of freedom; SS, sum of squares; VC, variance components; PV, percentage of variation.
Under a model of population expansion, only clade E identified in the cpDNA phylogeny, and containing haplotypes mainly of ssp. sinensis, showed a strongly unimodal mismatch distribution (Fig. S2), indicating that it underwent an expansion in the recent past. Nonsignificant SSD and raggedness index values (PSSD = 0.12 and PHRag = 0.10), as well as a significantly large negative FS value (−15.557, P < 0.001; Table 4), also suggested a historical demographic expansion within clade E. Assuming cpDNA mutation rates of 2 × 10−9 s s−1 yr−1, the expansion of clade E was estimated to have occurred 90.4 (95% CI: 40.1–139.5) thousand yr ago.
Table 4. The results of Tajima’s D and Fu’s FS tests, and mismatch distribution analyses (MDA) for the five lineages (A–E) of Hippophae rhamnoides chlorotypes
Tajima’s D test
Fu’s FS test
Previous studies of the historical biogeography of northern hemisphere, widespread, temperate plant groups having greatest diversification in the highlands of Asia have tended to focus on the phylogenetic analysis of genera showing such distribution (e.g. Zhang et al., 2007, 2009; Mao et al., 2010; Tu et al., 2010; Xu et al., 2010). Here we have focused on the historical biogeography of a widespread, temperate species in Eurasia, H. rhamnoides, rather than a genus, and shown that a phylogeographic analysis of cpDNA and ITS sequence variation in this species strongly supports an ‘out-of-QTP’ origin and migration pattern, as suggested by phylogenetic analyses at the genus level (e.g. Zhang et al., 2007, 2009; Xu et al., 2010). Our results further suggest that following dispersal from the QTP region, allopatric divergence took place, sometimes followed by secondary contact, hybridization and introgression between divergent intraspecific taxa.
Out of the Qinghai-Tibet Plateau
‘Dispersal is often inferred if the taxa in one area are phylogenetically derived from lineages that have more ‘primitive’ (phylogenetically basal) members in another area, which is inferred to be the source area’ (Futuyma, 1998; p. 208). Our phylogenetic trees of both ITS ribotype and cpDNA haplotype variation showed that the phylogenetically basal clades of H. rhamnoides (II and III in the ITS tree, and A, B and C in the cpDNA tree) were exclusively distributed in the QTP, thus supporting the ‘out-of-QTP’ hypothesis for the origin and dispersal of the species (Figs 1, 2). The phylogeographic analysis of cpDNA variation showed that the two other clades identified (D and E) contained haplotypes found only in individuals distributed in northern China, the Mongolian Plateau, central Asia, Asia Minor and Europe. Because clades D and E were nested within two QTP clades (A and B) and sister to QTP clade C, our results further indicated an origin of H. rhamnoides in the QTP region and subsequent colonization by dispersal of other parts of Eurasia. Further support for this hypothesis came from Bayes-DIVA biogeographic analyses of the cpDNA and nrITS trees, which suggested that multiple dispersals from the QTP accounted for most of the species’ range expansion in Asia (Fig. 3). The inferred migration route of H. rhamnoides to Europe through intervening mountain ranges in central Asia and Asia Minor is one suggested previously for many plants of the European Alps and neighbouring mountain ranges that originated in the QTP region (Kadereit et al., 2008). It was estimated that dispersal to central Asia (represented by clade D of cpDNA and clade IV of nrITS) preceded that to northern China (represented by clade E of cpDNA and clade V of nrITS) with both dispersal events having their source in the QTP. Because cpDNA is maternally inherited in Hippophaë (Bartish et al., 2002), the phylogeographic patterns observed are likely to reflect past migration of the species through long-distance seed dispersal mediated by birds. The fleshy fruits of the species are preferred by migrating birds that stay at high-altitude in the central QTP in summer (to avoid high temperature) and return to northern China or central Asia in autumn (e.g. Blanford’s snow finch; Cramp & Perrins, 1994; Qu et al., 2010).
Allopatric divergence, glacial refugia and local expansion
Clade divergence within the QTP (i.e. clades A–C for cpDNA, and clades II and III for ITS) and also outside the QTP (clades D and E for cpDNA, and IV and V for ITS) indicates that H. rhamnoides underwent allopatric divergence in different parts of its distribution during its evolutionary history. Clades A, B and C comprise cpDNA haplotypes found mainly in the western, eastern and central parts of the species’ distribution range in the QTP, respectively, while clades D and E comprise haplotypes that occur mainly in Europe/central Asia and northern China/Mongolia, respectively (Figs 1, 2a). Similarly, ITS clades (I, II and III) are distributed mainly in western-central, southeastern, and northeastern parts of its range in the QTP, while ribotypes in clades IV and V occur in Europe/central Asia/Mongolia and northern China, respectively (Figs 1, 2b). Although there is no ‘one to one’ match between the five cpDNA clades and the nine subspecies sampled within H. rhamnoides, haplotypes contained within ssp. wolongensis form a subclade of clade B, while those present in ssp. turkestanica, caucasica, carpatica and fluviatilis comprise clade D, and those in ssp. mongolica and sinensis mainly comprise clade E. Thus, there is evidence of an association between phylogenetic and taxonomic divergence within the species. The same is also apparent from an examination of the ITS phylogeny (Fig. 2b).
The major cpDNA and ITS clades were estimated to have diverged from the earliest Pliocene to the middle Quaternary (Table S3). The available fossil record of Hippophaë in Europe is too recent (the middle Pleistocene; Krupinski, 1992; Lang, 1994) for dating analyses based on ITS or cpDNA, while the fossil record of this genus in eastern Asia is very poor. Therefore, divergence dates were estimated from conventional substitution rates of 1–3 × 10−9 s s−1 yr−1 for cpDNA (Wolfe et al., 1987) and 3.46–8.69 × 10−9 s s−1 yr−1 for ITS sequences (Richardson et al., 2001). There was a reasonable match in divergence dates for the main cpDNA and ITS clades, that is, spanning the Pliocene and middle Quaternary periods. It is thought that the QTP underwent an extensive uplift during the Pliocene and Quaternary periods (Li & Li, 1991; Shi et al., 1998) with climates oscillating greatly during this period (Shi, 2002). It is feasible, therefore, that these marked geological and climatic changes could have fragmented the distribution of H.rhamnoides within the QTP and triggered the phylogenetic divergence recorded. Clearly, acceptance of this possibility needs to be treated with caution given the uncertainty of the clade divergence dates; however, multiple glacial refugia in the QTP during the Quaternary (Wang et al., 2009; Tang et al., 2010; Wu et al., 2010; Li et al., 2011) are thought to have caused similar deep phylogenetic divergence within several other plant species found in this region (e.g. Wang et al., 2009; Tang et al., 2010), including a congener H. tibetana (Wang et al., 2010; Jia et al., 2011).
The latest divergence within H. rhamnoides clades (i.e. between central Asia and Asia Minor/Europe; node a5 in Fig. 2a and node b4 in Fig. 2b) was estimated to have occurred between 1.45 (95% HPD: 0.53–2.82) and 1.87 (95% HPD: 0.82–3.25) Ma (Table S3), indicating that populations of the species have survived in the QTP, northern China, the Mongolian Plateau, central Asia and Asia Minor/Europe since at least this time. The divergences within the Asia Minor/Europe lineage should be later, around the middle Pleistocene, which is congruent with fossil records in central Europe (Lang, 1994). In our analysis, we were only able to detect a significant range expansion in clade E, which is distributed in northern China/Mongolian Plateau. This expansion was dated to 90.4 (95% CI: 40.1–139.5) thousand yr ago, which mostly corresponds to the Last Interglacial Period (Zheng et al., 2002) in the late Pleistocene and earlier than the Last Glacial Maximum (LGM; 18–21 thousand yr ago). Interestingly, this age is highly consistent with the estimate for range expansion of the species in Asia Minor/Europe (Bartish et al., 2006). Failure to detect demographic expansions in the other clades might stem from the limited number of samples and haplotypes recorded or for other reasons currently unknown.
Hybridization and/or introgression
Several examples of phylogenetic incongruence between the cpDNA and ITS trees indicated that historical hybridization and introgression had occurred between H.rhamnoides and a close relative, and also between some subspecies of H. rhamnoides. Interspecific hybridization between H. rhamnoides and H. neurocarpa was suggested by the finding that two ribotypes (R55 and R56) present in a few individuals of ssp. yunnanensis clustered with H. neurocarpa in the ITS tree, whereas their cpDNA haplotypes were placed firmly within H. rhamnoides in the cpDNA tree. Bartish et al. (2002) previously proposed that hybridization between these two species resulted in the origin of the two homoploid hybrid species, H. goniocarpa and H. litangensis (Lian et al., 1998; Sun et al., 2003). Hybridization between ssp. yunnanensis and/or ssp. wolongensis and ssp. sinensis was also indicated by the following: several individuals of ssp. sinensis in population 20 (Table S1; Fig. 1) possessed cpDNA haplotypes characteristic of ssp. yunnanensis and wolongensis (i.e. placed in clade B), but ribotypes characteristic of ssp. sinensis (i.e. placed in clade V of the ITS tree); and certain individuals in population 21 of ssp. sinensis possessed a cpDNA haplotype characteristic of this subspecies (i.e. placed in clade E), but a ribotype that clustered with those of ssp. yunnanensis and wolongensis in clade III of the ITS tree. Finally, there was an indication that ssp. mongolica originated from hybridization between ssp. sinensis and ssp. turkestanica in that its cpDNA haplotype (C26) was placed within those of ssp. sinensis in the cpDNA tree (clade E; Fig. 2a), while its ribotypes were placed in a subclade with those of ssp. turkestanica (clade IV; Fig. 2b).
Hippophaë rhamnoides is one of seven species recognized in the genus (Swenson & Bartish, 2002). Whereas the six other species, H. goniocarpa, H. litangensis, H.gyantsensis, H. neurocarpa, H. salicifolia and H. tibetana, are restricted in their distribution to the QTP and the adjacent Himalaya region, H. rhamnoides is disjunctly distributed across Eurasia occurring in the QTP, northern China, Mongolia, Siberia, central Asia, Asia Minor and Europe. Our phylogeographic analysis of H. rhamnoides indicates that the species originated in the QTP and then dispersed out of the QTP to other parts of Eurasia. Thus, our results support the hypothesis that many Northern Hemisphere, widespread, temperate plant groups having the greatest diversification in the highlands of Asia originated in the QTP and adjacent highlands, before migrating to other northern hemisphere regions and undergoing divergence. Both within and outside the QTP H. rhamnoides is divergent at the molecular level with several distinct clades resolved in both cpDNA and ITS phylogenetic trees. It was not possible to date the origin of these clades precisely; however, it is feasible that they originated during the Pliocene epoch and Quaternary period when the QTP and other parts of Eurasia underwent considerable geological and/or climatic oscillations. Such oscillations are likely to have fragmented the distribution of H. rhamnoides and triggered allopatric divergence and the formation of deep clades. We were only able to detect a significant expansion for one of the cpDNA clades, which using a mean substitution rate was estimated to have occurred before the Last Glacial Maximum. Although there was no ‘one to one’ match between the molecular clades and the nine subspecies we examined in H. rhamnoides, there was an association between phylogenetic and taxonomic divergence. A comparison of the topologies of the cpDNA and ITS trees revealed several examples of phylogenetic incongruence indicating that historical hybridization and introgression had occurred between H. rhamnoides and a closely related species, and also between some subspecies of H. rhamnoides.
We thank Dr M. Rabadanov for assistance in collecting material from eastern Caucasus, Dr K. Rumpunen for allowing us to collect wild materials of Hippophaë cultivated at Balsgård, Sweden, and Dr Y-M. Yuan for collecting material from Switzerland on our behalf. We are grateful to two anonymous reviewers for their constructive criticism of an earlier version of the manuscript. This research was supported by grants from the Natural Science Foundation of China (30725004 and 40972018), the Key Project of International Collaboration Program, the Ministry of Science and Technology of China (grant number 2010DFB63500) to J.Q.L., a Royal Society-NSF China International Joint Project award 2010/R4 to R.J.A. and J.Q.L., and an International Collaboration ‘111’ Project to J.Q.L and Purkine fellowship from the Academy of Sciences of the Czech Republic to I.V.B.