Molecular data and ecological niche modelling reveal a highly dynamic evolutionary history of the East Asian Tertiary relict Cercidiphyllum (Cercidiphyllaceae)


  • Xin-Shuai Qi,

    1. Key Laboratory of Conservation Biology for Endangered Wildlife of the Ministry of Education, and Laboratory of Systematic & Evolutionary Botany and Biodiversity, College of Life Sciences, Zhejiang University, Hangzhou 310058, China
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Chen Chen,

    1. Key Laboratory of Conservation Biology for Endangered Wildlife of the Ministry of Education, and Laboratory of Systematic & Evolutionary Botany and Biodiversity, College of Life Sciences, Zhejiang University, Hangzhou 310058, China
    Search for more papers by this author
    • These authors contributed equally to this work.

  • Hans Peter Comes,

    1. Department of Organismic Biology, Salzburg University, A-5020 Salzburg, Austria
    Search for more papers by this author
  • Shota Sakaguchi,

    1. Division of Forest and Biomaterials Science, Graduate School of Agriculture, Kyoto University, Kyoto 6068502, Japan
    Search for more papers by this author
  • Yi-Hui Liu,

    1. Key Laboratory of Conservation Biology for Endangered Wildlife of the Ministry of Education, and Laboratory of Systematic & Evolutionary Botany and Biodiversity, College of Life Sciences, Zhejiang University, Hangzhou 310058, China
    Search for more papers by this author
  • Nobuyuki Tanaka,

    1. Department of Plant Ecology, Forestry and Forest Products Research Institute, Matsunosato, Tsukuba, Ibaraki 305-8687, Japan
    Search for more papers by this author
  • Hitoshi Sakio,

    1. Field Centre for Sustainable Agriculture and Forestry, Faculty of Agriculture, Niigata University, 94-2 Koda, Sado, Niigata 952-2206, Japan
    Search for more papers by this author
  • Ying-Xiong Qiu

    1. Key Laboratory of Conservation Biology for Endangered Wildlife of the Ministry of Education, and Laboratory of Systematic & Evolutionary Botany and Biodiversity, College of Life Sciences, Zhejiang University, Hangzhou 310058, China
    Search for more papers by this author

Author for correspondence:
Ying-Xiong Qiu
Tel: +86 571 88981703


  • East Asia’s temperate deciduous forests served as sanctuary for Tertiary relict trees, but their ages and response to past climate change remain largely unknown. To address this issue, we elucidated the evolutionary and population demographic history of Cercdiphyllum, comprising species in China/Japan (Cercdiphyllum japonicum) and central Japan (Cercdiphyllum magnificum).
  • Fifty-three populations were genotyped using chloroplast and ribosomal DNA sequences and microsatellite loci to assess molecular structure and diversity in relation to past (Last Glacial Maximum) and present distributions based on ecological niche modelling.
  • Late Tertiary climate cooling was reflected in a relatively recent speciation event, dated at the Mio-/Pliocene boundary. During glacials, the warm-temperate C. japonicum experienced massive habitat losses in some areas (north-central China/north Japan) but increases in others (southwest/-east China, East China Sea landbridge, south Japan). In China, the Sichuan Basin and/or the middle-Yangtze were source areas of postglacial northward recolonization; in Japan, this may have been facilitated through introgressive hybridization with the cool-temperate C. magnificum.
  • Our findings challenge the notion of relative evolutionary and demographic stability of Tertiary relict trees, and may serve as a guideline for assessing the impact of Neogene climate change on the evolution and distribution of East Asian temperate plants.


The warm-temperate deciduous forests of East Asia are well known for their richness in woody genera considered to be relicts (or ‘living fossils’) of the palaeotropical northern-hemisphere flora of the Tertiary period (c. 65–2.6 million yr ago, Ma; Takhtajan, 1969; Liu, 1988; Wang, 1988; Axelrod et al., 1998; López-Pujol & Zhao, 2004; Manchester et al., 2009). The greatest concentration of these relicts (e.g. Cercidiphyllum, Davidia, Euptelea, Ginkgo, Metasequoia, Tetracentron) presently extends from Sichuan in southwest China eastward across the Yangtze Valley and then to south Japan (e.g. Ohsawa, 1993; Lu, 1999; Tang & Ohsawa, 2002). The long-term persistence of these monotypic or low-diversity taxa is usually explained by the absence of continental glaciation and, by implication, a smaller magnitude of Quaternary environmental change (≤ 2.6 Ma) compared with other temperate regions, especially Europe and North America (Liu, 1988; Mai, 1995). However, the range dynamics of East Asian warm-temperate deciduous forests under conditions of Late Tertiary/Quaternary climate change, and thus their potential importance for the current genetic structuring in Tertiary relict trees, are still poorly understood (reviewed in Qiu et al., 2011).

Only one phylogeographic study has so far addressed this issue directly (Gong et al., 2008), providing evidence that natural populations of Ginkgo biloba largely persisted throughout glacial/interglacial cycles around the middle-Yangtze (Three Gorges) and in east China (Tianmu Mountains), without exhibiting strong genetic signatures of range expansion. Similar molecular findings in several understorey herbs tend to support this relative stability of warm-temperate deciduous forests, especially in areas south of the Yangtze (i.e. subtropical central/south/east China) and in south Japan (Qiu et al., 2009a,b,c). However, East Asian palaeobiome reconstructions for the Last Glacial Maximum (LGM; c. 21 000 yr before present (BP); Harrison et al., 2001) indicate a more complex picture of the glacial range dynamics of this forest biome, suggesting that massive habitat reductions in some areas (e.g. north China, north Japan) occurred simultaneously with increases in others (e.g. across the exposed East China Sea). The validity of these contentions, and their genetic consequences, could be further addressed by reconciling ecological niche modelling (ENM; Phillips et al., 2006) and phylogeography (e.g. Carstens & Richards, 2007; Alsos et al., 2009; Marske et al., 2011). However, no study to date has used this dual approach to predict past (e.g. LGM) distributions of climatically suitable areas, or to reconstruct locations of glacial refugia and the dynamics of postglacial migrations in a wide-ranging East Asian temperate deciduous forest tree.

The present study focuses on Cercidiphyllum, the only member of Cercidiphyllaceae, containing two extant species (Supporting Information, Fig. S1a): Cercidiphyllum japonicum is a tall canopy tree occurring mainly in low- to mid-elevation warm-temperate deciduous forests in China and Japan (c. 600–2000 (2700) m above sea level). This species is unusual among East Asian Tertiary relicts in that it widely extends north of the Yangtze and into the far north of Japan (Hokkaido), where it locally descends to near sea level (Isagi et al., 2005; Krassilov, 2010). By contrast, Cercidiphyllum magnificum is a small tree or shrub restricted to cool-temperate/subalpine forests of central Honshu (1500–2800 m above sea level). However, no information is available on the population genetic relationships of the two species and their temporal origin, that is, the genus’s crown group age.

We would expect from palaeodata (e.g. Winkler & Wang, 1993; Yu et al., 2000; Harrison et al., 2001) that during the LGM (and possibly earlier cold periods) the lower-elevation and less cold-tolerant C. japonicum experienced marked habitat reductions at its current northern range limits (i.e. north China, north Japan). On the other hand, when sea levels fell by c. 85–130/140 m during glacial maxima (Millien-Parra & Jaeger, 1999), the East China Sea (ECS) basin was repeatedly exposed and likely covered by warm-temperate deciduous forest (Harrison et al., 2001). Hence, currently isolated populations of Cercidiphyllum in China and Japan may have had ample opportunities for range expansion and genetic admixture via the glacially exposed ECS landbridge, as previously inferred from genetic surveys of several plant species (e.g. Hotta, 1974; Aizawa et al., 2007; Kikuchi et al., 2010; but see Li et al., 2008; Qiu et al., 2009b,c). However, it remains to be tested by molecular markers whether the ECS landbridge served as a migration route for warm-temperate deciduous forest trees such as C. japonicum.

Here, we use a multidisciplinary approach combining molecular phylogeographic, ENM, and phylogenetic approaches to elucidate the evolutionary and population demographic history of Cercidiphyllum. Specifically, we used fossil-calibrated molecular phylogenies to clarify the temporal origin of C. japonicum and C. magnificum and their evolutionary genetic relationships (e.g. are they fully-fledged sister-species, do they form a (paraphyletic) progenitor-derivative pair, or are relationships blurred through ancient or ongoing hybridization?). In addition, we coupled palaeodistribution models with phylogeographic analyses in each species using chloroplast (cp) and nuclear ribosomal (nr) DNA sequence data as well as nuclear microsatellites (nSSRs) to identify the species’ glacial (LGM) distributions and postglacial colonization routes. Together, these analyses should contribute to a better understanding of how East Asia’s Tertiary relict flora responded to previous periods of rapid climate and environmental change in terms of lineage diversification, population demographic processes (e.g. persistence, range contractions/expansion) and adaptive genetic diversity.

Materials and Methods

Study system

Fossil evidence identifies Cercidiphyllum Sieb. et Zucc. as a dominant component of Late Cretaceous/Tertiary forest and woody pioneer communities throughout the Northern Hemisphere (Mai, 1995; Meyer & Manchester, 1997; Manchester et al., 2009). Gradual climate cooling probably exterminated populations in North America and western Asia (during the Miocene) and in Europe (during the Late Pliocene) (Mai, 1995). In East Asia, Cercidiphyllum-like fossils have been reported in northeast China and Japan from the Early Tertiary (Paleocene) to the mid-Pleistocene (Tanai, 1981; Crane & Stockey, 1985; Onoe, 1989; Uemura, 1991; Meyer & Manchester, 1997; Guo et al., 2010; Krassilov, 2010; see Fig. S1b). These data clearly document the long-term presence of the genus as a ‘Tertiary relict’ in East Asia.

Both extant species are diploid (2n = 38), dioecious, wind-pollinated, and wind-dispersed by means of small, flattened, and winged seeds (Fu & Endress, 2001). While originally described as a local variety of C. japonicum Sieb. & Zucc. (Nakai, 1919), C. magnificum Nakai differs, for example, in a lower stature (≤ 10 vs 35–40 m), larger leaves, and seeds winged at both ends (rather than one end; Fu & Endress, 2001). Previous phylogenetic (nrDNA) analyses (Li et al., 2002) supported C. magnificum as a separate species, but in total only seven samples were included. Moreover, fossil-calibrated multilocus phylogenies have recognized Cercidiphyllaceae (represented only by C. japonicum) as a member of the ‘woody clade’ of Saxifragales (along with Altingiaceae, Daphniphyllaceae, Hamamelidaceae), a clade that underwent an ancient and rapid radiation during the mid-Cretaceous (c. 106–90 Ma; Soltis & Soltis, 1997; Fishbein et al., 2001; Jian et al., 2008).

Plant material and sampling design

Silica-dried leaf material was obtained from 48 populations (= 513) of C. japonicum throughout its range in China (populations J1–33) and Japan (J34–48), and five populations (= 64) of C. magnificum (M1–5) (Fig. S1a). All samples (= 577) were surveyed for nSSR variation, whereas subsets were sequenced at both cpDNA and nrDNA regions (= 254 and 25 for C. japonicum and C. magnificum, respectively; Table S1). For the nSSR dataset, 13 C. japonicum populations with small sample sizes (< 8; i.e. 12 from China and one from Japan) were excluded from all population-level analyses of genetic diversity and differentiation, restricting those analyses to 35 populations (marked with an asterisk in Table S1).

DNA extraction, sequencing, and microsatellite genotyping

Total genomic DNA was extracted from the dried leaf tissue using a DNeasy plant tissue kit (Qiagen). For the phylogeographic DNA surveys, we sequenced four intergenic spacer (IGS) regions of cpDNA (petA–psbJ, psbE–petL, rpl32–trnL, F71–R1516), and the entire internal transcribed spacer (ITS) region of nrDNA (i.e. ITS1 + 5.8S + ITS2). A subset of individuals was also sequenced at three additional chloroplast genes (atpB, matK, rbcL) for phylogenetic analysis. The primers and methodology for amplification of these eight DNA regions via PCR were described in Fishbein et al. (2001) and Li et al. (2002, 2008). Sequences were generated with an ABI 377XL DNA sequencer, and edited, assembled, and aligned in geneious (version 4.8.2; Drummond et al., 2009; available at This dataset also included previously published sequences (Fishbein et al., 2001). All haplotype sequences identified in the present study were deposited in GenBank (see Table S2 for accession numbers).

All DNA samples were genotyped at seven nSSR loci using primers and amplification protocols developed for Cercidiphyllum (Chen et al., 2010; see Table S5). PCR products were separated on a MegaBACE 1000 (GE Healthcare Biosciences, Sunnyvale, California, USA). Alleles were scored manually using genetic profiler (version 2.2; GE Healthcare Biosciences).

Phylogeographic analyses of cpDNA, ITS, and nSSRs

For both cpDNA and ITS, indices of haplotype (h) and nucleotide diversity (π) were estimated according to Nei (1987) at the levels of populations (hS, πS) and species (hT, πT) using dnasp (version 4.1; Rozas et al., 2003). Haplotype networks were constructed in tcs (version 1.21; Clement et al., 2000) under the 95% statistical parsimony criterion. For ITS, we had to increase the tcs connection limit to 30 steps to link the divergent networks of the two species. Gaps (indels) detected in the cpDNA and ITS dataset were treated as single mutation events, and coded as substitutions (A or T).

For nSSRs, measures of genetic diversity were assessed for each population with  8, and across all loci, by calculating the total number of detected alleles (NA), allelic richness (RS), and gene diversity (HS). Differentiation between populations (with  8) was computed using FST (Weir & Cockerham, 1984), and its significance (at each locus, and overall) was tested using 1000 bootstrap permutations. All these analyses were performed in fstat (version 2.9.3; Goudet, 2001) and genepop (version 4.0; Raymond & Rousset, 1995). To characterize population structure, structure (version 2.3; Pritchard et al., 2009) was run on the entire nSSR dataset (i.e. 53 populations) for 100 000 Markov chain Monte Carlo (MCMC) cycles following 10 000 burn-in cycles, using the admixture model with independent allele frequencies. Ten replications were performed for each K, in the range = 1–53, and the optimal K was estimated according to Evanno et al. (2005). For each dataset (cpDNA, ITS, nSSRs), analyses of molecular variance (AMOVAs) were carried out in arlequin (version 3.1; Excoffier et al., 2006).

Phylogenetic divergence time estimation

To estimate the crown group age of Cercidiphyllum, sequences of three cpDNA regions (atpB, matK, rbcL) were obtained from GenBank, representing all other genera of the ‘woody clade’ of Saxifragales (i.e. Altingia, Corylopsis, Daphniphyllum, Exbucklandia, Liquidambar, Hamamelis, Rhodoleia), plus a functional outgroup, Paeonia brownii (Fishbein et al., 2001; Jian et al., 2008; see Table S2 for accessions numbers). Sequences were assembled with those of four atpB-matK-rbcL haplotypes (HA–D) identified among 24 C. japonicum and four C. magnificum individuals (see Table S2), chosen to represent all 12 cpDNA-IGS haplotypes identified in the phylogeographic survey (see the Results section).

Bayesian searches for tree topologies and node ages of this ‘woody clade’ cpDNA (atpB, matK, rbcL) dataset were conducted in beast (version 1.61; Drummond & Rambaut, 2007) using a GTR+G substitution model, selected by modeltest (version 3.6; Posada & Crandall, 1998), and an uncorrelated lognormal relaxed clock (Drummond et al., 2002). A Yule process was specified as tree prior. Following Jian et al. (2008), two fossil dates were used to assign minimum age constraints on two internal stem nodes: Cercidiphyllum (Cercidiphyllaceae: 65–71 Ma; Magallón et al., 1999) and Liquidambar (Altingiaceae: 11.6–15.9 Ma; Pigg et al., 2004). Assuming a normal distribution, these calibration points were modelled with a mean (± SD) of 68 ± 1.53 and 13.7 ± 1.07 Ma, respectively (see nodes 1 and 2 in Fig. 4a).

In a second step, we applied a similar beast analysis to all cpDNA-IGS haplotype sequences of Cercidiphyllum to determine intrageneric node ages. We used the same settings as the first step, except for an HKY substitution model (selected by modeltest), and a constant-size coalescent tree prior. To calibrate the root node, we used the median crown group age of Cercidiphyllum approximated from the ‘woody clade’ (atpB, matK, rbcL) phylogeny, that is, 6.52 Ma (lognormal mean, 1.875; SD, 0.368; zero offset, 0; range, 3.17–13.41 Ma).

For each beast analysis, MCMC runs were performed, each of 1 × 107 generations, with sampling every 1000th generation, following a burn-in of the initial 10% cycles. MCMC samples were inspected in tracer (version 1.5; Rambaut & Drummond, 2009) to confirm sampling adequacy and convergence of the chains to a stationary distribution. Resulting chronograms were visualized in figtree (version 1.3.1;

Demographic analyses

For particular cpDNA-IGS (sub)lineages of Cercidiphyllum, we tested the null hypotheses of a spatial expansion and a pure demographic expansion using mismatch distribution analysis (MDA) in arlequin. For each model, goodness-of-fit based on the sum of squared deviations (SSD) and Harpending’s (1994) raggedness index (HRag) was tested using parametric bootstrapping (Schneider & Excoffier, 1999) with 1000 pseudoreplicates. In addition, we used tests of selective neutrality (Tajima’s (1989)D; Fu’s (1997)Fs) to infer potential population growth and expansion.

For a single expanding subclade identified (see the Results section), the MDA-derived expansion parameter (τ), and its 95% confidence interval (CI), were converted to absolute estimates of time (T, in number of generations) since expansion began using τ/2u (Rogers & Harpending, 1992; Rogers, 1995), where u is the neutral mutation rate for the entire sequence per generation. The value for u was calculated as =  μkg, where μ is the substitution rate in substitutions per site yr–1 (s/s/y), k is the average sequence length of the cpDNA region under study (here, 3278 bp; see the Results section), and g is the generation time in yr (i.e. age of first reproduction; approximated as 10 yr, as observed for C. japonicum in cultivation; Wan & Zhang, 2002). The mean substitution rate of the four combined cpDNA-IGS regions was obtained from the corresponding clock-calibrated beast tree of Cercidiphyllum.

Present and past distribution modelling

Ecological niche modelling was carried out in maxent (version 3.2.1; Phillips et al., 2006) to predict suitable past (LGM) and present climate envelopes for each species. Information on the geographic distribution of C. japonicum was based on the 48 populations included in this study (see Table S1, Fig. S1a), combined with 170 presence records in Japan obtained from the Phytosociological Releve Database (PRDB), Japan (see Fig. 5a). However, because of its rarity, only14 records were available for C. magnificum, that is, five localities of this study plus nine obtained from major herbaria in Japan (TNS, TI, KYO, OSA; see Fig. 5c). In this case, background points were extracted only from Japan (Kyushu to Hokkaido), as no extant or fossil records are known from elsewhere.

The climatic niche of each species was modelled as a function of four (out of 19) BIOCLIM variables available from the WorldClim dataset (Hijmans et al., 2005), that is, warmest and coldest quarter mean values for temperature and precipitation, respectively. This restricted dataset was used to avoid having to include highly correlated variables (data not shown), and thus prevent potential overfitting (Peterson & Nakazawa, 2008). To construct ENMs, we used the default parameters of maxent and the following user-selected features: regularization multiplier of 3.0, application of a random seed, duplicate presence records removal and logistic probabilities used for the output (Phillips & Dudík, 2008). Model performance was evaluated using the area under the (receiver operating characteristic) curve (AUC) calculated by maxent. Values between 0.7 and 0.9 indicate good discrimination (Swets, 1988). We modelled the modern distribution 10 times, using a different 70% of localities to train the model and 30% to test the model, and visually compared AUC scores and jackknife tests of variable importance to assess consistency between runs. The potential distribution was then obtained by projecting the estimated ecological niche on to the reconstructed LGM climatic conditions simulated by community climate system model (version 3.0; Collins et al., 2006; LGM palaeoclimate layers in 2.5 arcmin resolution were then prepared from these data following Sakaguchi et al. (2010); however, to estimate the palaeocoast lines (−130 m than at present) and the palaeoclimate surfaces of the exposed seafloor area during the LGM, seafloor topography data (ETOPO1) from the National Geophysical Data Center of National Oceanic and Atmospheric Administration (NOAA, USA) were used.


CpDNA phylogeography and diversity of Cercidiphyllum

The four cpDNA-IGS regions surveyed across the 278 individuals (53 populations) of Cercidiphyllum were aligned, with a total length of 3278 bp. Nucleotide substitutions occurred at 17 sites, and one indel (7 bp) was present in the F71–R1516 region (Table S3). In combination, these polymorphisms identified a total of 12 haplotypes (‘chlorotypes’; H1–12). Of those, 10 were specific to C. japonicum (H1, H2, H5–12), whereas H3 and H4 were shared between C. japonicum and C. magnificum (Figs 1a,b). In the unrooted chlorotype network (Fig. 1c), these two sets of chlorotypes formed distinct clades (cp-group I and II, respectively), separated by eight mutational steps.

Figure 1.

Distribution of chlorotypes in (a) Cercdiphyllum japonicum and (b) Cercidiphyllum magnificum. Population codes are identified in Table S1. (c) TCS-derived network of genealogical relationships between the 12 chlorotypes. The small open circles represent missing chlorotypes. The sizes of circles are approximately proportional to sample size (n), with the smallest circles representing = 1 and the largest representing = 84.

In C. japonicum, two subclades comprising cp-group I, but differing by only a single mutational step, had nearly nonoverlapping distributions (Fig. 1a,c): chlorotypes of the ‘northern’ subclade mainly occurred in localities north of the middle reaches of the Yangtze, where H2, in particular, was predominant; chlorotypes of the ‘southern’ subclade were largely confined to areas south of the Yangtze, with H1 predominating in southwest and east China but being fixed in south Japan. By contrast, populations in north Japan consisted almost entirely of group II chlorotypes (H3/H4) as found in C. magnificum (Fig. 1b). However, a single population of C. japonicum in central Honshu (J41) harboured chlorotypes of both groups (H1, H3).

Most populations of each species were fixed for a single chlorotype (Table S1). There was much stronger population subdivision in C. japonicum than in C. magnificum (93.6 vs 56.6% of the total variation; Table 1). By contrast, only 22.2% of the total variation was partitioned between species, reflecting both the lack of unique chlorotypes in C. magnificum and the sharing of chlorotypes H3/H4 between the two species.

Table 1.   Analyses of molecular variance (AMOVAs) based on cpDNA chlorotype frequencies, nrDNA (internal transcribed spacer (ITS)) ribotype frequencies, and nuclear microsatellite (nSSR) allele frequencies for populations of Cercidiphyllum japonicum and Cercidiphyllum magnificum
Source of variationcpDNAnrDNA ITSnSSR
dfPercentage of total variance (%)dfPercentage of total variance (%)dfPercentage of total variance (%)
  1. All levels of variation were significant.

C. japonicum C. magnificum
 Among species122.22149.62115.18
 Among populations within species5169.145116.68386.71
 Within populations2258.6445933.71101078.10
C. japonicum
 Among populations4793.664729.96347.89
 Within populations2058.9418670.0488792.11
C. magnificum
 Among populations456.62460.5547.73
 Within populations2043.381839.4512392.37

ITS phylogeography and diversity of Cercidiphyllum

The ITS sequences of the 260 individuals (53 populations) of Cercidiphyllum were aligned with a total length of 682 bp, revealing 43 nucleotide substitutions (ITS-1: 21; ITS-2: 22) and one indel (1 bp) in each spacer region (Table S4). Together, these 45 polymorphic sites yielded 25 ITS haplotypes (‘ribotypes’, R1–25; Table S4). Of those, 20 were specific to C. japonicum (R1–5, R11–25; including 12 population-specific ribotypes, ‘RN’ in Fig. 2a) and five to C. magnificum (R6–10; Fig. 2b). These species-specific ribotypes formed separate tcs clades (ITS-group I and II, respectively), differing by 22 mutational steps (Fig. 2c). In C. japonicum, all 15 Japanese populations were fixed for the most abundant and interior (ancestral) ribotype R1, whereas 25 (out of 33) populations in China were polymorphic, consisting primarily of R1 and few mutational derivatives, of which only R3 showed a distinct geographical distribution (middle-Yangtze and Daba/Qinling Mountains; Fig. 2a).

Figure 2.

Distribution of internal transcribed spacer (ITS) ribotypes in (a) Cercdiphyllum japonicum and (b) Cercidiphyllum magnificum. Population codes are identified in Table S1. An uppercase N represents population-specific ribotypes. (c) tcs-derived network of genealogical relationships between the 25 ribotypes. The small open circles represent missing ribotypes. The sizes of circles are approximately proportional to sample size (n), with the smallest circles representing = 1 and the largest representing = 169.

There was markedly higher population differentiation in C. magnificum (60.5%) than in C. japonicum (29.6%; Table 1), reflecting its lack of ribotype variation in Japan (Fig. 2a). In fact, Chinese C. japonicum had, on average, higher degrees of within-population diversity (hS = 0.429; πS = 0.917 × 10−3) than C. magnificum (hS = 0.280; πS = 0.644 × 10−3; Table S1). Overall, differentiation among the two species was much larger for ITS (49.6%) than for cpDNA (22.2%; Table 1).

Nuclear microsatellite genotyping

The 40 populations of Cercidiphyllum (= 525) revealed a total of 254 alleles across the seven nSSR loci surveyed, along with high mean per-locus estimates of allele and gene diversity (NA = 36.3; HS = 0.804; HT = 0.901; see Table S5). Population differentiation (FST) was significant at each locus (< 0.05), and averaged 0.109 across all loci (Table S5).

For the entire nSSR dataset (53 populations, = 577), structure yielded the highest likelihood when samples were clustered into three groups (= 3, data not shown). All individuals of C. magnificum formed a separate cluster (‘blue’ in Fig. 3a), whereas those of C. japonicum were variously assigned to clusters I (‘green’) vs II (‘red’). Both clusters (or ‘gene pools’) were present at high frequencies in China (44 vs 56% of all local samples) but without showing a distinct geographical pattern. In Japan, where both clusters were present, the great majority of individuals were assigned to cluster I (88%), while those of cluster II (12%) apparently decreased in frequency from the south to the north (Fig. 3b). A separate structure analysis of C. japonicum resulted in similar patterns (data not shown).

Figure 3.

(a) Histogram of the structure assignment test for 53 populations (577 individuals) of Cercidiphyllum based on variation at seven nuclear microsatellite (nSSR) loci. Each vertical bar represents an individual and its assignment proportion into one of three population clusters corresponding to Cercidiphyllum magnificum and two clusters (I and II) of Cercidiphyllum japonicum. The more uniform the shades of the bar, the higher the probability that the individual belongs to a given cluster (or ‘gene pool’). Each cluster is represented by a different colour. (b) Geographic origin of the 48 C. japonicum populations and their colour-coded grouping according to the structure analysis. Population codes are identified in Table S1.

There were similar population subdivisions in C. japonicum (7.89%) and C. magnificum (7.73%), whereas their separation explained c. 15.2% of the total variation (Table 1). Average within-population diversity was significantly higher in C. japonicum than in C. magnificum (NA = 66 vs 44, RS = 7.665 vs 5.104, HS = 0.838 vs 0.626; all = 0.001; Table S1). Notably, several populations of C. japonicum from southwest China (e.g. southwestern Sichuan Basin) and south Japan (Shikoku, central Honshu) had HS values above average, identifying these areas as main centres of gene diversity (Fig. S2); however, relatively high HS values were also observed in single populations from north-central China (J14/Qinling Mountains) and north Japan (J46/Hokkaido). Allelic richness (RS) largely followed the same pattern (data not shown).

Phylogeny-based estimations of divergence times

The beast-derived cpDNA (atpB, matK, rbcL) chronogram of the ‘woody clade’ of Saxifragales (Fig. 4a) recovered Cercidiphyllum as sister to Liquidamber/Altingia, albeit with relatively low support (posterior probability, PP = 0.82; node 1). Based on this chronogram, we estimated the crown group age of Cercidiphyllum (PP = 1.00; node 3), comprising chlorotypes either unique to C. japonicum (HB, HC) or shared with C. magnificum (HA, HD), as c. 6.52 Ma (95% highest posterior density, HPD: 1.34–13.42 Ma) (Table S6). Using the latter estimate as root prior for the cpDNA-IGS chronogram of Cercidiphyllum (Fig. 4b), we obtained a similar point estimate for the coalescent time (node 4) of all 12 chlorotypes (c. 5.32 Ma; 95% HPD: 1.93−9.25 Ma), suggesting a Late Mio-/Early Pliocene split between cp-groups I (C. japonicum; PP = 0.939) and II (C. magnificum/C. japonicum; PP = 1.00; Table S6). By contrast, ‘northern’ vs ‘southern’ subclades of C. japonicum (PP = 0.997 vs 0.374) probably diverged at the Plio-/Pleistocene boundary (c. 1.89 Ma; 95% HPD: 0.36−4.15 Ma; node 5 in Fig. 4b, and Table S6). Based on this cpDNA-IGS chronogram, the beast analysis provided an average substitution rate of 3.18 × 10−10 s/s/y. This is much lower than the average values generally reported for noncoding regions of the chloroplast genome (e.g. 1.2–1.7 × 10−9 s/s/y; Graur & Li, 2000) but consistent with the notion that woody taxa and/or phylogenetic relicts (viz. ‘living fossils’) should have slower rates of molecular evolution (e.g. Avise et al., 1994; Kay et al., 2006).

Figure 4.

beast-derived chronograms of: (a) the ‘woody clade’ of Saxifragales based on cpDNA (atpB, matK, rbcL) sequences with calibration points denoted by numbers 1 and 2 (see the Materials and Methods section for further explanation); and (b) the crown group of Cercidiphyllum based on cpDNA-IGS (petA–psbJ, psbE–petL, rpl32–trnL, F71–R1516) sequences. Grey bars on each node indicate 95% highest posterior densities (HPDs) of time estimates (in million yr ago, Ma). Mean divergence dates and 95% HPDs for major nodes (1–7) are summarized in Table S6. Posterior probabilities (PP > 0.6) are labelled above nodes. Chlorotypes are indicated by letter codes (HA–D; H1–12).

Demographic analyses

Demographic analyses (MDAs) were conducted for all cpDNA-IGS (sub)lineages of Cercidiphyllum, except for cp-group II, as it comprised chlorotypes of both species. Mismatch distributions were unimodal for cp-group I (C. japonicum) and its ‘northern’ and ‘southern’ subclades (figures not shown); however, only for the latter subclade were both spatial and demographic expansions supported by uniformly nonsignificant SSD and HRag values, as well as significantly negative Tajima’s D and Fu’s FS values (Table 2). Based on the corresponding τ values, and assuming a substitution rate of 3.18 × 10−10 s/s/y (see previous section), we dated the spatial expansion of the ‘southern’ subclade to the last glacial cycle(s) (c. 0.24 Ma; 95% CI: 0.07–0.59 Ma).

Table 2.   Summary of mismatch distribution parameters and neutrality tests for the pooled samples of cp-group I, and individual subclades (northern, southern) of Cercidiphyllum japonicum. Estimates were obtained under models of spatial or pure demographic expansion using arlequin
ModelGroupParameter (τ)Expansion time (t) in yr bpSSD P H Rag P Fu’s FS P Tajima’s D P
  1. Note that for the FS test, = 0.02 is considered to be significant at the α = 0.05 level (Fu, 1997). NC, not calculated (see the Materials and Methods section).

Spatial expansioncp-group I0.894 (0.520–2.241)NC0.04500.0890.0221.4030.7130.1570.658
Northern subclade0.402 (0.197–0.943)NC0.0030.0320.2130.446−0.0400.395−0.1700.434
Southern subclade0.503 (0.136–1.226)241,499 (65 234–588 064)0.0030.0560.1550.414−5.0950.007−1.6090.021
Demographic expansioncp-group I0.922 (0.678–1.320)NC0.05500.0890.0111.4030.7510.1570.658
Northern subclade2.981 (0–8.307)NC0.2290.1390.2130.126−0.0400.407−0.1700.413
Southern subclade0.520 (0.107–1.184)249,198 (51 324–567 918)0.0030.3000.1550.344−5.0950.007−1.6090.024

Present and past (LGM) distribution of C. japonicum and C. magnificum

For C. japonicum, the test AUC (mean ± SD) for the ENM, averaged across all 10 runs, was very high (0.97 ± 0.006). Although the current distributional predictions (Fig. 5a) were quite good representations of the species’ extant distribution (Fig. S1), there were also some predicted areas where the species does not occur at present (e.g. southeastern Qinghai–Tibetan Plateau, Taiwan, South Korea). By contrast, the distribution model for the LGM (Fig. 5b) differed substantially from the present, indicating a general southward range shift. Most evident is a loss of suitable habitat in north Japan and north-central China (e.g. northern Sichuan Basin, Daba/Qinling Mountains). By contrast, isolated patches of suitable habitat probably persisted around the middle-Yangtze and in central Taiwan. However, the most extensive stretches of hospitable habitat were predicted for southwest China (e.g. southwestern Sichuan Basin); and disjunct areas further east, extending from southeast China across the exposed ECS basin to south Japan (Kyushu, Shikoku, south/central Honshu) (Fig. 5b). For C. magnificum, the test AUC for the ENM was moderate (0.86 ± 0.15) as a result of limited collection records. Predicted suitable habitats for this species at the LGM were restricted to the Pacific Ocean side of central Honshu (Fig. 5d), indicating an even smaller than present geographic range (Fig. 5c).

Figure 5.

Potential distributions as probability of occurrence for Cercidiphyllum in East Asia at the present (0 kaBP) (a) and at the Last Glacial Maximum (LGM, 21 kaBP) for C. japonicum (b), and at the present (c) and at the LGM for C. magnificum (d). Ecological niche models were established with current bioclimatic variables on the basis of extant occurrence points (black dots) of the species using maxent (Phillips et al., 2006). Predicted distribution probabilities (in logistic values) are shown in each 2.5 arcmin pixel. The distribution of river systems on the exposed East China Sea basin during the LGM was drawn from Kimura (1996).


Species boundaries and cytoplasmic-nuclear discordance in Cercidiphyllum

Some authors have questioned the validity of C. magnificum, arguing that it may instead be a local variety of C. japonicum (Bean, 1970; Griffiths, 1994; Flint, 1997; Krassilov, 2010; see also Nakai, 1919). The two taxa often inhabit the same mountain ranges in Japan (central Honshu), but they are segregated altitudinally (Kubo et al., 2010), and intermediate forms between them have never been recorded (Lindquist, 1954; Li et al., 2002; S. Sakaguchi pers. obs.). Our results based on nuclear DNA (ITS, nSSRs) clearly support the distinctiveness of these two named species of Cercidiphyllum (Figs 2c, 3a), consistent with earlier ITS data (Li et al., 2002). However, the cpDNA analyses provided an unexpected result in that C. japonicum in central-north Japan shares the same two chlorotypes (H3, H4) with C. magnificum (Fig. 1a,b). This cytoplasmic-nuclear discordance might be explained by incomplete lineage sorting of chloroplast genomes from a polymorphic ancestor, or unidirectional (asymmetric) ‘chloroplast capture’ through introgressive hybridization (e.g. Comes & Abbott, 2001; Wu & Campbell, 2005; Minder et al., 2007; Tagane et al., 2008).

We argue for the latter explanation for the following reasons. First, the shared chlorotypes are not distributed throughout the range of C. japonicum, as expected for incomplete lineage sorting, but are restricted to central Honshu (H3, H4), where the species are broadly sympatric, or to allopatric areas further north, where C. japonicum is fixed for H4 (Fig. 1a,b). Secondly, according to neutral theory, lineage sorting through drift will be four times slower in nuclear than in chloroplast genes (4Ne vs Ne generations) for a dioecious species (Schaal et al., 1998; Palumbi et al., 2001). In consequence, given reciprocal monophyly between the two species for nuclear genes, the same should hold true, even more so, for cpDNA, but this is not observed. We therefore conclude that the two species are not undergoing primary divergence in central Honshu, but instead are fully-fledged sister-species that underwent an allopatric phase of divergence before coming into secondary contact in this region, resulting in the capture of both chlorotypes native to C. magnificum (H3, H4) by C. japonicum.

A Late Tertiary speciation event in Cercidiphyllum

If interpreted as a signature of cpDNA capture, the nonmonophyly of Cercidiphyllum for cpDNA should present no problem to date the divergence between C. japonicum (cp-group I) and C. magnificum (here equated with cp-group II). Based on our fossil-calibrated cpDNA (atpB-matK-rbcL, IGS) phylogenies, this speciation event occurred at the Mio-/Pliocene boundary (see nodes 3 and 4 in Fig. 4a,b; Table S6). This timing is reasonable as it postdates the earliest fossil records of several now extinct Cercidiphyllum species in central/north Japan from the Eocene to Early Miocene (e.g. Celongatum, C. eojaponicum, C. crenatum), but predates the first fossil appearance of the two modern species in the Mid-Pleistocene of central Honshu (Tanai, 1981; Crane & Stockey, 1985; Onoe, 1989; see Fig. S1b). Unlike its closest extant, and likely fossil relatives, C. magnificum inhabits cool-temperate/subalpine rather than warm-temperate deciduous forests; it is feasible, therefore, that its divergence, possibly from a C. japonicum-like ancestor, was initiated by the transition from a globally warm to a cooler climate towards the end of the Miocene (Cerling et al., 1997; Zachos et al., 2001, 2008; see also Nagalingum et al., 2011). The speciation time reported here thus qualifies the two extant species of Cercidiphyllum as Late Tertiary relicts, even though they can scarcely be regarded as ‘living fossils’. In fact, our estimated coalescent times of all chlorotypes native to C. japonicum (cp-group I) and C. magnificum (cp-group II) are relatively recent, that is, 1.89 Ma (0.36–4.15 Ma) and 0.88 Ma (0.02–2.11 Ma), respectively (see nodes 5 vs 6 in Fig. 4b; Table S6). This timing places current patterns of genetic diversity within each species, and historical processes associated with them, mostly likely within the (Late) Quaternary.

Late Quaternary population history of C. japonicum

The wide-ranging C. japonicum revealed a remarkable differentiation in cpDNA between the populations from north China (i.e. north of the Yangtze; ‘northern’ subclade) and those from subtropical China/south Japan (‘southern subclade’). This split long predates the LGM (c. 1.89 Ma, see node 5 in Fig. 4b; Table S6), and might have been fostered through recurrent range expansions/contractions and isolation through time, with the Yangtze further acting as geographic barrier to seed dispersal (see also Gao et al., 2007; Zhang et al., 2007). Although the MDA failed to detect an expansion event in this northern subclade (Table 2), the ENM data in conjunction with the northern predominance of a relatively derived chlorotype (H2) strongly suggest that C. japonicum recolonized areas north of the Yangtze after the last glacial(s), that is, following the retreat of arid steppe and desert vegetation there (Winkler & Wang, 1993; Yu et al., 2000; Harrison et al., 2001). Based on the distribution of H2, the species most probably spread from the western Sichuan Basin and/or the middle-Yangtze area (Figs 1a, 5a,b). Exceptionally high degrees of nSSR gene diversity further support the former area as a major glacial refuge (Fig. S2). Notably, postglacial northward expansion from around the Sichuan Basin has also been indicated in a conifer characteristic of north China’s cool-temperate deciduous forests, Pinus tabulaeformis (Chen et al., 2008).

All three markers (cpDNA, ITS, nSSR) failed to detect any deep subdivision within the ‘southern’ subclade. Moreover, the MDA indicated a relatively recent spatial (and demographic) expansion within this subclade at c. 0.24 Ma (95% CI: 0.07–0.59 Ma; Table 2), possibly coinciding with the penultimate (Riss) glacial (c. 0.12–0.35 Ma). On the other hand, several populations from around the middle-Yangtze were characterized by derived (‘tip’) chlorotypes (e.g. H6, H8, H9, H11) that probably evolved in situ from the widespread and ancestral (‘interior’) chlorotype (H1) (Fig. 1a,c). The ENM revealed these populations largely remained isolated from the major refugia in southwest and southeast China during the LGM (Fig. 5b). Overall, these data are strongly suggestive of a once wider southern distribution of C. japonicum during earlier (possibly colder) periods, spanning areas from southwest through southeast China, and across the ECS landbridge to south Japan. Accordingly, molecular traits presently observed on either side of the ECS (i.e. ancestral haplotypes H1 and R1, and nSSR pools I/II) are best interpreted as shared ancestral polymorphisms persisting since at least the latest submergence of this landbridge (< 16 000 yr bp; Chung, 2007). At the same time, these data do not allow us to infer any directional spread of C. japonicum from China to Japan (or vice versa).

Regardless of these uncertainties, fossil data do indicate the presence of both C. japonicum and C. magnificum in central Honshu during the Mid-Pleistocene (e.g. Crane & Stockey, 1985; Onoe, 1989). Moreover, our ENM data provide firm evidence for the glacial survival of C. japonicum in south Japan (Kyushu, Shikoku, south/central Honshu; Fig. 5b). In support of this, many C. japonicum populations show above-average degrees of nSSR diversity in these areas (Fig. S2), and all of those have previously been suggested as refugia for other warm-temperate deciduous forest taxa more widely distributed in Japan today (e.g. Tsukada, 1988; Takahashi et al., 2005; Okaura et al., 2007; Sugahara et al., 2011). The ENM further indicates that C. japonicum was extirpated from north Honshu and Hokkaido at the LGM, and recolonized these areas from southern refugia upon recent climate amelioration (Fig. 5b). Possibly as a consequence, most C. japonicum populations in the recolonized areas (except J46) contain low degrees of nSSR diversity (Fig. S2), as would be expected under a ‘leading edge’ model of range expansion (Hewitt, 2000). Interestingly, for C. japonicum, the ENM also predicted suitable habitat along the Pacific Ocean side of central Honshu during the last glacial(s) (Fig. 5c); this may have fostered secondary contact with C. magnificum, which probably occurred contemporaneously in the same area, though in isolated patches (Fig. 5d).

Asymmetric cpDNA introgression between C. japonicum and C. magnificum

Our data strongly support the hypothesis that C. japonicum invaded the area already occupied by C. magnificum in central Honshu, possibly during a glacial period, resulting in the capture of C. magnificum cpDNA by C. japonicum. It is likely that this event occurred relatively recently because all introgressed individuals of C. japonicum analysed contained the same IGS-chlorotypes (H3, H4) as found in C. magnificum. Following postglacial climate warming, one of the introgressed chlorotypes (H4) would have become fixed in the leading-edge populations of C. japonicum while expanding to the very far north of Japan.

Considering the genetic-demographic consequences of introgression, this scenario is remarkably consistent with two predictions raised by a recent model of Currat et al. (2008). These include, first, that the direction of introgression of neutral genes should proceed from the locally established species towards the expanding relative, because introgressed genes at the front of a range expansion are likely to ‘surf’ on the wave of advance as a result of high amounts of genetic drift (Excoffier et al., 2009); and secondly, that introgression is expected to be stronger for neutral genes experiencing little gene flow with conspecific populations, which may partly explain the more frequent introgression of organellar compared with nuclear genes (Currat et al., 2008; Petit & Excoffier, 2009). In agreement with the model’s first prediction, C. japonicum was shown to be the expanding species that was fixed for the captured chlorotype H4 in Japanese populations further north along the expansion wave. Likewise, the species’ generally stronger population substructure and thus lower amounts of gene flow for cpDNA (through seed only) than for nDNA (ITS, nSSRs; through seed and pollen) agrees well with the model’s second prediction (see Table 1), and recent empirical evidence of seed-transmitted markers being more readily introgressed than markers dispersed by both pollen and seed (Du et al., 2011). Although stochastic surfing of introgressed organellar genomes during postglacial range expansions has recently been reported in perennial herbs from Japan (Veratrum: Kikuchi et al., 2010) and the seaweed Fucus ceranoides from Europe (Neiva et al., 2010), the present study is the first that obtained independent evidence on the status of plant species (resident vs expanding) and their past ranges by using ENM approaches.

Concluding remarks

East Asian warm-temperate deciduous forests have long been viewed as sanctuary for Tertiary relict trees during the climatic fluctuations of the Quaternary and earlier (e.g. Liu, 1988; Mai, 1995). Recent phylogeographic studies of Ginkgo biloba and associated understorey herbs sampled in subtropical China and/or south Japan have tended to support the implicit stability of this forest biome (Gong et al., 2008; Qiu et al., 2011). By contrast, the present results demonstrate that Late Tertiary/Quaternary climate change profoundly influenced the evolutionary and population demographic history of Cercidiphyllum, the most wide-ranging representative of this relict flora. Thus, Late Tertiary climate cooling was reflected in a relatively recent speciation event, dated at the Mio-/Pliocene boundary. Moreover, our ENM and phylogeographic analyses indicate that, during glacials, one of the resultant species, C. japonicum, experienced massive habitat losses in north-central China and north Japan, but simultaneously expanded its range within three major refugia (southwest China, southeast China/ECS landbridge, south Japan) or persisted in isolated patches around the middle-Yang-tze. Postglacial northward range expansion most likely occurred from all these refugia, except for southeast China.

In addition, our data raise the intriguing possibility that postglacial range expansion of the warm-temperate C. japonicum to the very far north of Japan may have been facilitated through introgressive hybridization with its cool-temperate congener, C. magnificum, in central Honshu. The role of introgressive hybridization as an evolutionary mechanism leading to increased geographic range and adaptive genetic diversity is widely recognized (Abbott, 1992; Milne & Abbott, 2000; Ellstrand & Schierenbeck, 2000; Comes & Abbott, 2001; Petit et al., 2004), even though the specific mechanisms are still poorly understood (Rieseberg et al., 2007). Clearly, efforts need to be undertaken now to obtain more precise estimates of the reproductive and ecogeographical isolation between these two Cercidiphyllum species in the mountain ranges of central Honshu, for example, by examining intrinsic aspects of their biology along altitudinal gradients. In addition, it would be particularly interesting to determine from molecular mapping studies whether quantitative trait loci (QTLs) increasing cold-tolerance have been added from the locally adapted, cold-hardy C. magnificum to the gene pool of the northern Japanese C. japonicum. If so, genetic adaptation to newly encountered environments, rather than climatic or other factors (e.g. biotic interactions, migration ability), would probably be the most influential for the species’ northward expansion in Japan (see also Sakaguchi et al., 2010). In sum, the above inferences may provide a useful guideline for future studies assessing the impact of past climate change on the current genetic structuring of East Asian temperate plants on a broad geographical scale.


The authors thank C.X. Fu and P. Li in China, and Y. Isagi and T. Oda in Japan for their great help in collecting plant materials, and R. J. Abbott and three anonymous reviewers for very helpful comments on earlier versions of this paper. This research was supported by the National Basic Research Program of China (973 Program, grant no. 2007CB411600), the National Science Foundation of China (grant no. 30970197), and the Open Research Foundation of the LSEB (State Key Laboratory of Systematic and Evolutionary Botany, Institute of Botany, Chinese Academy of Sciences).