Radiation history of Asian Asarum (sect. Heterotropa, Aristolochiaceae) resolved using a phylogenomic approach based on double-digested RAD-seq data

Abstract

Background and Aims

The genus Asarum sect. Heterotropa (Aristolochiaceae) probably experienced rapid diversification into 62 species centred on the Japanese Archipelago and Taiwan, providing an ideal model for studying island adaptive radiation. However, resolving the phylogeny of this plant group using Sanger sequencing-based approaches has been challenging. To uncover the radiation history of Heterotropa, we employed a phylogenomic approach using double-digested RAD-seq (ddRAD-seq) to yield a sufficient number of phylogenetic signals and compared its utility with that of the Sanger sequencing-based approach.

Methods

We first compared the performance of phylogenetic analysis based on the plastid matK and trnL–F regions and nuclear ribosomal internal transcribed spacer (nrITS), and phylogenomic analysis based on ddRAD-seq using a reduced set of the plant materials (83 plant accessions consisting of 50 species, one subspecies and six varieties). We also conducted more thorough phylogenomic analyses including the reconstruction of biogeographic history using comprehensive samples of 135 plant accessions consisting of 54 species, one subspecies, nine varieties of Heterotropa and six outgroup species.

Key Results

Phylogenomic analyses of Heterotropa based on ddRAD-seq were superior to Sanger sequencing-based approaches and resulted in a fully resolved phylogenetic tree with strong support for 72.0–84.8 % (depending on the tree reconstruction methods) of the branches. We clarified the history of Heterotropa radiation and found that A. forbesii, the only deciduous Heterotropa species native to mainland China, is sister to the evergreen species (core Heterotropa) mostly distributed across the Japanese Archipelago and Taiwan.

Conclusions

The core Heterotropa group was divided into nine subclades, each of which had a narrow geographic distribution. Moreover, most estimated dispersal events (22 out of 24) were between adjacent areas, indicating that the range expansion has been geographically restricted throughout the radiation history. The findings enhance our understanding of the remarkable diversification of plant lineages in the Japanese Archipelago and Taiwan.

INTRODUCTION

Species radiation in a geographically restricted area such as on certain islands serves as an excellent model for studying the history and mode of evolution. However, to disentangle the evolutionary patterns and mechanisms underlying species diversity, it is necessary to obtain an accurate, well-resolved phylogenetic tree. To this end, advances in phylogenetic inference have resulted from widespread use of nucleotide sequencing technologies (Sanger sequencing) during the last 30 years (Delsuc et al., 2005). Nevertheless, one of the most formidable problems with the conventional Sanger sequencing-based approach is resolving radiation that has taken place in a relatively short time. This is because, in such a situation, individual genes are unlikely to contain sufficient information and/or may not necessarily represent the correct species bifurcation history due to the process known as incomplete lineage sorting (Maddison and Knowles, 2006). Phylogenomic approaches, which have only recently become commonly available with the spread of massively parallel sequencing technology, can mitigate these difficulties by integrating information from hundreds to thousands of loci across the genome to identify the signals of speciation histories (Wagner et al., 2013; Giarla and Esselstyn, 2015).

Asarum (Aristolochiaceae) is a genus of perennials comprising 128 species distributed throughout the northern hemisphere, with its centre of diversification in East Asia (Cheng and Yang, 1983; Kelly, 2001; Huang et al., 2003; Sugawara, 2006; Sinn et al., 2015a, b; Takahashi and Setoguchi, 2018; Supplementary data Table S1). The genus is one of the most speciose plant groups in Japan and Taiwan, with 50 % of the diversity (64 spp.) endemic to approx. 10 % of the total distribution range (<60 000 km² of >3600 000 km², based on records from the Global Biodiversity Information Facility). Moreover, the genus is remarkable not only in the number of species found in Japan and Taiwan but also in the marked diversity of floral traits including size, shape, colour and scent (Azuma et al., 2010; Kakishima and Okuyama, 2018). The pollination system of the genus is typical brood-site mimicry without reward for the pollinators (Sugawara 1988; Sinn et al., 2015b), and our preliminary surveys on the pollination biology of some Asarum species in Japan suggest that the floral diversity probably reflects marked diversity in their association with pollinators. Recently, Sinn et al. (2015a) conducted a phylogenetic study based on seven plastid regions and the nuclear ribosomal internal transcribed spacer (nrITS) region, and revealed that the genus can be separated into six sections. Among them, the sect. Heterotropa, which is characterized by base chromosome number x = 12 (Sugawara 1981, 1987; Zhou, 1998), is the most species-rich clade, to which 49 species endemic to Japan belong (62 species in total; Supplementary data Table S1). Indeed, the distribution ranges of many Heterotropa species are narrow and confined, probably due to their extremely low dispersibility (Hiura, 1978; Kelly, 1998). Multiple native species can be found in each area of the Japanese Archipelago, including the isolated small islands. Meanwhile, the characteristic traits shared by Heterotropa species, including reduction in vegetative growth, loss of autonomous self-pollination and the presence of putative fungal-mimicking floral structures, all of which are presumably for an increased investment in sexual reproduction and increased reliance on pollinators, have been reported to have accelerated the diversification rate (Sinn et al., 2015b). These patterns of diversity strongly suggest that an adaptive radiation of Heterotropa has taken place in the Japanese Archipelago. Therefore, addressing its evolutionary history in eastern Asia would enhance our understanding of the dynamic history of the diversity and local endemic flora of the Japanese Archipelago.

It is of particular interest to clarify the diversification history of Asarum to understand how this plant lineage has become species rich, although it has been difficult to achieve good phylogenetic resolution. In this study, we first evaluated the utility of the Sanger sequencing-based phylogenetic approach using plastid genes or nuclear ribosomal DNA, which resulted in a very low resolution for sect. Heterotropa. Secondly, we demonstrated that double-digested RAD-seq (ddRAD-seq) reads have rich phylogenetic information. Using this information, we present the first fully resolved phylogenetic tree of Heterotropa species and illustrate the pattern of diversification centred on the Japanese Archipelago.

MATERIALS AND METHODS

Taxon sampling

To select the ingroup species to be analysed, we followed the definitions of the infrageneric sub-divisions of Asarum proposed by Sinn et al. (2015a), although we found that some of the suggested diagnostic characters between sect. Heterotropa and Longistylis were not useful (see the Discussion). Based on this taxonomic treatment, the number of described species of sect. Heterotropa is estimated to be 62 (73 including subspecies and varieties; Supplementary data Table S1), of which we sampled 54 species, one subspecies and nine varieties. Among these, we sampled multiple (2–5) accessions for 39 taxa (including two undescribed species), especially those with a broad distribution (Supplementary data Table S2). We also sampled nine individual plants from Japan, which could not be assigned to any described species, although their affiliation to sect. Heterotropa was obvious. The only taxa of Heterotropa unable to be sampled in the present study were four species and two varieties native to Japan and four species native to Taiwan (Supplementary data Table S1). As the outgroups, we sampled three species from sect. Longistyis and one species each from sections Asarum, Asiasarum and Hexastylis. In total, we sampled 135 plant individuals as operational taxonomic units (OTUs).

We prepared full and reduced sets of OTUs for phylogenetic analyses. The full set included all of the 135 OTUs, whereas the reduced set included only a sub-set of the full set. The reduced set contained 83 OTUs with 55 species, one subspecies and five varieties. The reduced set was used to compare the phylogenetic resolution achieved using Sanger sequencing-based and ddRAD-seq-based approaches, because obtaining the Sanger sequencing-based data sets for all 135 OTUs was very cost-ineffective given the lack of phylogenetic resolution and requirement of extensive sub-cloning for samples. The full set was used for the remaining analyses.

DNA extraction, quantification and Sanger sequencing

Fresh or silica-dried leaves were granulated with liquid nitrogen, washed with 0.1 m HEPES-HCl buffer containing 10 mm EDTA-Na, 0.9 % ascorbic acid, 1 % polyvinylpyrolidone and 2 % mercaptoethanol, and subjected to standard cetyl trimethylammonium bromide extraction. For ddRAD-seq library preparation, the DNA concentration was measured using a Picogreen Double Strand DNA Quantification kit (Thermo Fisher Scientific, Waltham, MA, USA) and a TECAN Infinite F200 spectrophotometer (Tecan Systems, San Jose, CA, USA), and the sample concentrations were adjusted to 10 ng µL^–1.

For Sanger sequencing-based phylogenetic analysis, the plastid matK gene and trnL–F region (CP) and the nrITS were amplified using Ex-Taq (TaKaRa, Kusatsu, Japan) and sequenced using the BigDye Terminator v3.1 Cycle Sequencing kit (Thermo Fisher Scientific). As the amplified fragments of the nrITS sometimes contained heterogeneous sequences, we separated them into individual sequences using the pGEM-T Easy Vector System II (Promega, Madison, WI, USA) when necessary. The primers used for DNA amplification and sequencing are listed in Supplementary data Table S3. New nucleotide sequences obtained by Sanger sequencing in this study were deposited in the DDBJ (accession nos LC530329–LC530601).

Library preparation for ddRAD-seq

The library for ddRAD-seq (Peterson et al., 2012) was created as described by Sakaguchi et al. (2015). In brief, we used EcoRI and BglII (New England Biolabs, Ipswich, MA, USA) for restriction digestion during library preparation. After PCR amplification of DNA, the sequencing libraries were purified with AMPure^®XP (Beckman Coulter, CA, USA) and size-selected at approx. 380 bp by gel electrophoresis using E-Gel^® SizeSelect™ (Life Technologies, Carlsbad, CA, USA). In each library, samples were multiplexed to achieve an average of 1.5 million reads (the first Hiseq2000 run) or 5.0 million reads (the remaining sequencing runs). After quality assessment using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), the library was sequenced with 100 + 100 bp paired-end reads in three lanes of an Illumina HiSeq2000 system and was sequenced with 150 + 150 bp paired-end reads in one lane of an Illumina HiSeqX system (Illumina, San Diego, CA, USA) by Macrogen (Seoul, South Korea). To confirm the stability of the results between the independent library preparations and the different sequencing platforms, six plant materials whose raw read counts were relatively small in the first Hiseq2000 run were used again for the sequencing library for the last HiseqX run. For the sequences from the HiseqX run, only the initial 100 + 100 bp read for each paired-end was used for analyses. Sequences are available at the DDBJ Sequence Read Archive (https://www.ddbj.nig.ac.jp/dra/index-e.html; accession nos DRA008867, DRA008892, DRA008926 and DRA009385).

Data set preparation

The de-multiplexed data for the individual samples were run through PEAR 0.9 software (Zhang et al., 2013) to merge overlapping reads. Next, the merged and unmerged reads (i.e. paired-end reads without sequence overlap) were separated and each data set was subjected to data processing using pyRAD 3.0.4 software (Eaton, 2014). We used the unmerged reads as the principal targets of the analysis and regarded the merged reads as complementary, because the latter are <200 bp fragments that are supposed to be largely removed by the size-selection step. For each individual, we maintained the minimum locus coverage at six reads and allowed for a maximum of four bases in a read with a quality score of <20 (Phred score). We also allowed a maximum of ten shared heterozygous sites for paralogue filtering. We repeated the data processing with three clustering thresholds within and between samples, i.e. 0.84, 0.88 and 0.92, to test the influence of parameters on phylogenetic estimation. In the data sets for phylogenetic analyses, we filtered the loci with missing data in >80 % of the ingroup OTUs.

In the Sanger sequencing-based phylogenetic analyses, we prepared the concatenated CP and nrITS data sets for the reduced set of the OTUs. In the ddRAD-seq-based phylogenomic analysis, we prepared the concatenated data sets of the unmerged reads using each of the three clustering thresholds for both the reduced and full sets of OTUs, resulting in a total of six data matrices. The optimal clustering threshold was determined by the overall superiority of the number of loci shared by 20–60 % of OTUs in the processed data set. After we determined the optimal clustering threshold, the unmerged and merged data matrices generated by the threshold were concatenated and used as the total data set for the Heterotropa phylogenetic analyses. A total of nine nucleotide sequence data matrices (CP dataset, nrITS dataset, reduced ddRAD datasets with three clustering thresholds, full ddRAD datasets with three clustering thresholds, and the total dataset) were used in this study.

Phylogenetic analysis

Prior to the full tree search, we conducted both neighbour–joining (NJ) and maximum parsimony (MP) tree searches using PAUP*4.0a (build 166; Swofford, 2002) to check if all six pairs of the plant materials, whose ddRAD-seq data were obtained in two independent library preparations and sequencing platforms, were placed with the nearest neighbour in the tree with very short branch length in between. After confirming this to be true (data not shown), only the data obtained from HiseqX were retained for these six plant materials, as it contained more data than obtained from the previous Hiseq2000 run.

For each data matrix, a single full maximum-likelihood (ML) tree search combined with a rapid bootstrap algorithm was performed using RaxML 8.2.0 software (Stamatakis, 2014) with the model set to GTR with gamma-distributed rate variation across sites and a proportion of invariable sites. We did not perform model selection prior to the tree search, because it was time-consuming for our ddRAD-seq data sets, and because a recent study has demonstrated that this step is non-essential for phylogenetic tree inference (Abadi et al., 2019). For bootstrapping, we used the majority rule consensus method (MR)-based bootstopping option (Pattengale et al., 2010) with the threshold value set to 0.03. Under this setting, the RAD-seq-based data sets usually converged at 100–150 bootstrap replicates; we conducted 300 replicates for the analyses of the total data set to obtain more accurate branch support values.

Additionally, a coalescent-based tree reconstruction was carried out for the total data set using singular value decomposition scores for species quartets (SVD quartets; Chifman and Kubatko, 2014) as implemented in PAUP*4.0a (build 166; Swofford, 2002). This method was used to infer the topology of the phylogenetic tree in two different manners –either regarding each OTU as an evolutionary unit (full SVD-quartet tree inference) or regarding multiple individuals belonging to a single taxon (either variety or species) as an evolutionary unit (SVD-quartet-based species tree inference). Seven undescribed ‘species’ as well as two distinct clades of A. asperum found in the ddRAD-seq-based ML tree were each regarded as an evolutionary unit. In the SVD quartets, we sampled all quartets and assessed branch support using 300 bootstrap replicates.

Biogeographic analysis

We reconstructed the biogeographic history of the Heterotropa radiation onto the ML tree based on the total ddRAD data sets using the BioGeoBEARS 1.1.1 package (Matzke, 2013) as implemented in RASP 4.2 (Yu et al., 2020). To this end, we first transformed the ML tree into a time-calibrated ultrametric tree (time-tree) using the RelTime method (Tamura et al., 2012) as implemented in MEGA7 (Kumar et al., 2016). In the RelTime analysis, the root of the tree was placed on the branch leading to A. caulescens (sect. Asarum) because the placement of the section Asarum as the sister to the remaining sections sampled here was unequivocally supported in a previous study (Sinn et al., 2015a). Subsequently, three outgroup species, i.e. A. caulescens (sect. Asarum), A. speciosum (sect. Hexastylis) and A. tohokuense (sect. Asiasarum), were excluded from the analyses. Although there is no known fossil record for the genus Asarum, we were able to place two time calibration points. One is the minimum age of 1.3 million years ago (MYA) for the A. lutchuense–A. kumageanum var. kumageannum split. This time calibration point is based on two assumptions; one is that the Asarum species would not be capable of migration between the seas, and the other is the fairly robust geological evidence that the ancient break up of the land connection between the distribution areas of these species (central and northern Ryukyus) dates back at least 1.3 MYA (Kimura, 2000). Our assumption that Asarum species are not capable of migration across seas was based on the seed morphology (i.e. relatively large seed size), seed physiology (susceptible to desiccation) and the presence of ant dispersal syndromes, as well as the fact that there are no instances of disjunct distributions of any of the 62 Heterotropa species (Supplementary data Table S1). We note that although there are some cases of distributions that cross seas within Heterotropa, these are always between adjacent islands that were presumably connected during recent glacial periods. Although these observations do not rule out the possibility of long-distance dispersal over the evolutionary history of Heterotropa, it is reasonable to assume that the probability of long-distance dispersal events within certain lineage splits is very low; thus, this minimum time constraint is justifiable. The other time calibration point is the maximum age of 15.2 million years for the Heterotropa–Longistylis split (= stem age of Heterotropa). This maximum age is the upper limit of the 95 % highest posterior density proposed by Takahashi and Setoguchi (2018), who calculated it from the phylogenetic tree based on ITS sequences and an estimation of the ITS substitution rate in annual/perennial herbaceous plants (Kay et al., 2006).

Next, the time-tree was modified by removing the tips to represent one OTU for one taxon (or evolutionary unit; following the same criterion at the SVD-quartet-based species tree inference). In the time-tree, the taxa with multiple accessions were in most cases monophyletic, hence the decision to remove tips did not affect the resulting tree. The exceptions were the four non-monophyletic taxa, namely A. savatieri var. savatieri, A. takaoi, A. celsum and A. fudsinoi, for which we remove tips arbitrarily. In these instances, the decision to remove tips slightly affected the resulting tree; however, this is unlikely to affect the conclusion of the biogeographic analyses, as the closest taxa have nearly the same distribution states. For the modified (tip-removed) time-tree, a lineage-through-time (LTT) plot was calculated and examined as to whether the speciation pattern fits the pure birth model or the birth–death model using the Akaike information criterion (AIC) as implemented in the phytools package (Revell, 2012) in the R package. Next, the distribution of each taxon was determined, where the geographic regions were categorized into the following 12 areas: A, Mid-East Honshu, Japan Sea side; B, Mid-East Honshu, Pacific Coast side; C, West Honshu; D, Shikoku; E, Kyushu; F, North Ryukyus; G, Amami-oshima (including adjacent Kakeromajima and Ukejima islands); H, Tokunoshima; I, Okinawa; J, South Ryukyus; K, Taiwan; and L, Mainland China (Supplementary data Table S1). The best-fit biogeographic model was selected among the six models (DEC, DEC+J, DIVALIKE, DIVALIKE+J, BAYAREA and BAYAREA+J) with AIC including corrections for small sample size (AICc) using the BioGeoBEARS 1.1.1 package (Matzke, 2013) as implemented in RASP 4.2 (Yu et al., 2020), allowing the maximum number of areas as three without setting any range constraints. However, because the statistical comparisons of the likelihoods between the model with and without ‘+J’ have recently been criticized as inappropriate (Ree and Sanmartín, 2018), we chose the best model among the three without ‘+J’ for the final biogeographic reconstruction and regarded the ‘+J’ model as the sub-optimal one.

RESULTS

Phylogenetic trees derived from the Sanger-based approach were poorly resolved and topologically unstable

Both the CP and nrITS data sets resulted in poorly resolved ML trees, in which only 14 % (11/80) and 13 % (14/104) of the overall branches were supported with >80 % bootstrap values (Figs 1 and 2). Also, the topologies of the trees based on the CP and nrITS data sets were incompatible (Figs 1 and 2). In the CP-based tree, the monophyly of sect. Heterotropa was not supported, and the species belonging to the sections Asiasarum (A. tohokuense) and Longistylis (A. petelotii) were nested among the Heterotropa species, whereas, in the nrITS-based tree, the monophyly of Heterotropa was strongly supported.

Fig. 1.

Maximum-likelihood (ML) tree of Asarum sect. Heterotropa based on the combined DNA sequences of the plastid matK gene and the trnL–F region with the reduced set of operational taxonomic units (OTUs). Bootstrap values >50 % are indicated above branches, while the branches without >50 % bootstrap support are indicated with dashed lines. The five taxa with multiple accessions that formed exclusively monophyletic groups are highlighted. Note that sections Heterotropa and Longistylis were not supported as monophyletic.

Fig. 2.

ML tree of Asarum sect. Heterotropa based on the DNA sequences of the nuclear ribosomal ITS region with the reduced set of OTUs. Bootstrap values >50 % are indicated above branches, while the branches without >50 % bootstrap support are indicated with dashed lines. The four taxa with multiple accessions that formed exclusively monophyletic groups are highlighted. Note that sect. Heterotropa and the Sakawanum clade therein were supported as monophyletic.

Also, the phylogenies based on the Sanger sequencing approach were only partly compatible with the morphology-based taxonomic system of Heterotropa. Only five and four of the 18 taxa with multiple accessions formed exclusively monophyletic groups in the CP- and nrITS-based trees, respectively (Figs 1 and 2). Moreover, several species such as A. hexalobum perfectum and A. takaoi had highly heterogeneous sequences of their nrITS, and these sequences appeared in divergent clades within the tree (Fig. 2). Meanwhile, the series Sakawanum, which consists of three Asarum species with a common morphological feature (i.e. only longitudinal ridges are present on the inner surface of the calyx tube), was well supported only in the nrITS-based tree (Figs 1 and 2).

The ddRAD-seq data set and the optimal clustering threshold

After de-multiplexing and quality filtering, the number of reads per sample varied between 466 142 and 19 269 552. Using only unmerged reads, the three clustering thresholds, 0.84, 0.88 and 0.92, resulted in a total of 3954, 4535 and 4718 loci (819 803, 931 744 and 962 255 bp, respectively) in the final data matrices, respectively, although the numbers of sequenced loci per individual were 115–2134, 118–2436 and 97–2514, respectively. For individual samples, there was a clear correlation between the number of reads and the number of loci in the final data matrices, and the number of loci was saturated to approx. 2100 (approx. 45 % of the total loci in the data matrix) in the samples with >6 000 000 reads, regardless of the sequencing platforms (Fig. 3). The total number of loci in the final data matrices increased as the clustering threshold increased. However, the data processing statistics revealed that the number of loci shared by 30–60 % of OTUs was slightly lower at a clustering threshold of 0.92 than at 0.88, suggesting that the former caused oversplitting of loci (Fig. 4). Therefore, 0.88 was the optimal clustering threshold.

Fig. 3.

Scatterplot of the relationship between the number of loci (unmerged only) in the final data set and the total raw reads for an individual OTU in the ddRAD-seq analysis. The data points in black and white correspond to Heterotropa materials sequenced by Hiseq 2000 and Hiseq X, respectively, while the data points in grey refer to the outgroup materials.

Fig. 4.

Cumulative curves of the number of loci (unmerged only) shared among OTUs in the final data set for three clustering thresholds in the ddRAD-seq analysis. Note that the number of loci shared among 40–80 OTUs was greatest at a clustering threshold of 0.88.

The final total data set using 0.88 as the clustering threshold with unmerged and merged reads combined had a total of 11 000 loci (1 744 676 bp).

Fully resolved phylogenetic tree of Heterotropa based on ddRAD-seq-based phylogenomics

The resolution of the ML tree based on the unmerged reads of ddRAD-seq was remarkably high, irrespective of the clustering threshold. For the reduced data set generated by the three clustering thresholds (0.84, 0.88 and 0.92), 81.3, 85.0 and 82.5 % of the branches, respectively, were supported with >80 % bootstrap values (Supplementary data Figs S1–S3).

The ML tree based on the full data set resulted in a phylogenetic tree concordant with that based on the reduced data set. Again, for the full data set generated by the three clustering thresholds (0.84, 0.88 and 0.92), 78.8, 84.8 and 82.6 % of the branches, respectively, were supported with >80 % bootstrap values (Supplementary data Figs S4–S6). In all three clustering thresholds, the monophyly of sect. Heterotropa was strongly supported, and A. forbesii was placed as sister to all other Heterotropa species (core Heterotropa). The core Heterotropa was further divided into nine clades (Fig. 5), each of which consisted of species with distributions restricted to a particular geographic range. The nine clades were categorized as follows: Nipponicum, Asperum, Sakawanum (series Sakawanum), Asarum hexalobum, Asaroides, Lutchuense, Amami, Okinawa–Taiwan and South-east Mainland China (A. ichangense). The only major topological difference observed among the three clustering thresholds was the placement of the A. hexalobum + Asaroides + Luchuense clade, which was sister to the Nipponicum + Asperum + Sakawanum clade at the thresholds 0.84 and 0.92, but was sister to the Amami + Okinawa–Taiwan + Mainland China clade at the threshold 0.88 (Supplementary data Figs S4–S6).

Fig. 5.

ML tree of Asarum sect. Heterotropa based on the total ddRAD-seq data set. Bootstrap values of 80–99 % are indicated above branches, while the branches with <80 % bootstrap support are indicated with dashed lines. The solid branches without bootstrap values were those with 100 % bootstrap support.

The resolution and topology of the phylogenetic tree based on the total data set (Fig. 5) did not differ from the phylogenetic tree based on the full data set of unmerged reads generated with the optimal clustering threshold of 0.88 (Supplementary data Fig. S5). Overall, 111 of 132 branches (84.1 %) were supported with >80 % bootstrap values in the ML tree based on the total data set. Moreover, 34 of the 39 taxa with multiple accessions were recovered as exclusively monophyletic groups. The six taxa not recovered as monophyletic were A. savatieri var. savatieri, A. fauriei var. takaoi, A. asperum var. asperum, A. celsum and A. fudsinoi.

The SVD-quartet method, which regarded each OTU as an evolutionary unit, also resulted in a highly resolved tree (henceforth full SVD-quartet tree), in which 95 of 132 (72.0 %) branches were supported with >80 % bootstrap values, and its topology was mostly concordant with that of the ML tree (Supplementary data Fig. S7). The major difference between the full SVD-quartet tree and the ML tree was the placement of A. hexalobum. In the ML tree, A. hexalobum was placed as sister to the Asaroides clade, while, in the full SVD-quartets tree, their placement among other species was unclear (Supplementary data Fig. S7). Moreover, the monophyly of the Okinawa–Taiwan clade was not supported.

The SVD-quartet-based species tree inference resulted in a topology almost identical to that of the full SVD-quartet tree, and 57 of 75 (76.0 %) branches were supported with >80 % bootstrap values (Fig. 6). The only difference was the monophyly of the Okinawa–Taiwan clade, which was weakly supported (67 % bootstrap support) in the SVD-quartet-based species tree.

Fig. 6.

Multispecies coalescent species tree of Asarum sect. Heterotropa based on the singular value decomposition scores for the species quartet (SVD-quartet) method using the total ddRAD-seq data set. The geographic origins of the individual OTUs are also illustrated. Bootstrap values are indicated as in Fig. 5. The number in parentheses after the taxon name indicates those of the individuals analysed, with only taxa with multiple accessions indicated. Photographs are of flowers of the representative species of individual clades in sect. Heterotropa. From top to bottom: A. savatieri var. iseanum, A. savatieri, A. nipponicum, A. nipponicum var. nankaiense, A. kurosawae, A. asperum, A. blumei, A. tamaense, A. minamitanianum, A. hexalobum, A. subglobossum, A. asaroides, A. kumageanum var. satakeanum, A. yakushimense, A. pellucidum, A. leucosepalum, A. hatsushimae, A. fudsinoi, A. gelasinum, A. dissitum, A. macranthum, A. yaeyamense, A. ichangense and A. forbesii.

Biogeographic reconstruction of the radiation of Heterotropa using the time-calibrated ML tree based on the ddRAD-seq data set

The RelTime method, which uses the ML tree of the total ddRAD-seq data, resulted in a time-tree with the stem age of Heterotropa calculated as 10.3 MYA (Fig. 7). An LTT plot based on the time-tree was fit to a pure birth model (Yule, 1925) with λ = 0.2457 (Supplementary data Fig. S8).

Fig. 7.

A diversification–extinction–cladogenesis (DEC) model-based biogeographic history reconstruction onto the time-calibrated, modified ML tree of Heterotropa. The most likely states of ancestral distribution are shown on the major internal nodes, with alternative estimates also shown with pie charts. The number of estimated dispersal events are also shown on the map, illustrating the distribution areas of Heterotropa. See the Materials and Methods for the description of the distribution areas A–L.

Although BioGeoBEARS selected DEC+J as the best model for biogeographic history, we regarded DEC as the optimal model (Table 1; see the Materials and Methods). The DEC model estimated 27 events for dispersal between areas, 16 events for vicariance, 60 events for speciation within areas and no extinction. It also estimated that the range expansion from mainland China toward Japan had occurred at the split of the core Heterotropa clade approx. 7.2–9.2 MYA; the major vicariance between the Pacific Coast side of Mid-East Honshu and mainland China + Kyushu subsequently occurred (Fig. 7). Within Japan and Taiwan, most dispersal events (22 out of 24) were between adjacent areas (Fig. 7), indicating that the mode of range expansion of Heterotropa plants was geographically restricted throughout the radiation history. Meanwhile, the DEC+J model estimated 31 events for dispersal between areas, 20 events for vicariance, 54 events for speciation within areas and no extinction, which was not greatly different from the estimate based on the DEC model (Supplementary data Fig. S9). However, under the DEC+J model, all internal nodes were estimated to have distribution at a single area, suggesting that the biogeographic reconstruction under this model is somewhat unrealistic.

Table 1.

Results of model selection for historical biogeographic reconstruction using the total ddRAD-seq-based data set of Asarum spp.

model	LnL	No. of free parameters	d	e	j	Corrected AIC (AICc)	AICc weight
DEC	–177.6	2	0.0068	0.0076	0	359.4	4.60E-10
DEC+J	–155.6	3	0.0033	1.00E-12	0.016	317.6	0.57
DIVALIKE	–178.1	2	0.0081	1.00E-12	0	360.3	3.00E-10
DIVALIKE+J	–156	3	0.0034	1.00E-12	0.015	318.4	0.37
BAYAREALIKE	–200.6	2	0.0063	0.13	0	405.4	4.80E-20
BAYAREALIKE+J	–157.9	3	0.0029	1.00E-07	0.017	322.1	0.06

DISCUSSION

Phylogenomic reconstruction of Heterotropa radiation using ddRAD-seq

We fully resolved the phylogenetic relationships of genus Asarum sect. Heterotropa (sensuSinn et al., 2015a). This was achieved only by a phylogenomic approach using an ddRAD-seq data set obtained via massively parallel sequencing. A clear advantage of ddRAD-seq over Sanger sequencing-based approaches was that a greater percentage of taxa were recovered as monophyletic. In the phylogenetic trees reconstructed from the Sanger sequencing-based CP or nrITS data sets, <28 % of the taxa (species or varieties) with multiple accessions formed exclusively monophyletic groups (Figs 1 and 2), whereas >77 % of the same taxa formed exclusively monophyletic groups in the tree constructed from the ddRAD-seq data set (Supplementary data Figs S1–S3). The results demonstrated that most of the taxa circumscribed in the present taxonomic system have phylogenetic support and highlighted some taxa with potential taxonomic issues.

Several phylogenetic studies of the genus Asarum using Sanger sequencing-based approaches have been performed (Kelly, 1998; Sinn et al., 2015a; Takahashi and Setoguchi, 2018), but the results were inconsistent. For example, Sinn et al. (2015a) recognized six sections in the genus, and included A. forbesii, a species native to south-east China, in sect. Heterotropa together with those native to Japan, while assigning other species native to China, such as A. maximum, to the newly established sect. Longistylis, which is sister to sect. Hexastylis, a monophyletic group of species native to north-eastern America. In contrast, Takahashi and Setoguchi (2018) concluded that the species native to Japan and those in China (including both A. forbesii and A. maximum) constitute distinct sister clades, whereas the monophyly of sect. Hexastylis was not supported. In any case, because the main focus of previous studies has been to circumscribe the major infrageneric groups in Asarum and to clarify the relationships among them, they lack sufficient information to discuss the relationships within sect. Heterotropa. This is because taxon sampling for sect. Heterotropa has not been comprehensive and/or only one individual per taxon was included in previous data sets.

As the present ddRAD-seq-based phylogenetic tree was mostly consistent and complementary with the report by Sinn et al. (2015a) in terms of the circumscription of sect. Heterotropa, we followed their circumscription of sect. Heterotropa and Longistylis. However, as Takahashi and Setoguchi (2018) pointed out, the morphological characteristic defining the section Longistylis proposed by Sinn et al. (2015a), i.e. yellow pollen and overhanging ovoid stigmas placed sub-laterally below highly developed style extensions that are bifid to varying degrees, would not be appropriate because the suggested diagnostic characters are in fact shared among the species in both sect. Heterotropa and Longistylis. Rather, we suggest that the chromosome number can be used to differentiate these sections. In the genus Asarum, 67 % (87/128) of the species had information on their chromosome number (Sugawara et al., 1990, 2005; Sugawara and Ogisu, 1992; Whittemore and Gaddy, 1997; Whittemore et al., 1997; Zhou, 1998; Wang et al., 2004; Sugawara, 2006; Shi et al., 2008; Lu and Wang, 2009, 2914; Lu et al., 2009, and references therein), and the chromosome number of species in sect. Heterotropa is x = 12, whereas that of species in sect. Longistylis is x = 13 (Supplementary data Table S1). For example, Sugawara and Ogisu (1992) and Zhou (1998) independently reported a chromosome count of 2n = 24 for A. ichangense and A. forbesii, which are species from mainland China found to be nested in Heterotropa in the present study, but a chromosome count of 2n = 26 or 39 for the other 20 species, including A. maximum and A. petelotii (but see Shi et al., 2008, which reports a chromosome count of 2n = 26 for A. ichangense). The observation that the species of sect. Hexastylis have the chromosome number x = 13 (Sugawara, 1982; Soltis, 1984), similar to that of sect. Longistylis, is also consistent with the phylogenetic tree of Sinn et al., (2015a), which placed Hexastylis as sister to Longistylis.

We report the highest resolution phylogenetic tree of Heterotropa and provide insight into Heterotropa radiation in East Asia. Asarum forbesii, which is native to the south-east area of mainland China and is the only deciduous species in the section, was found to be sister to all other evergreen Heterotropa species (henceforth we term them as ‘core Heterotropa’ as in Sinn et al., 2015a). The core Heterotropa was further sub-divided into nine subclades, each of which had a limited geographic distribution, and among which the major split was between the Honshu + Shikoku and south-western Japan clades, which showed only slight distributional overlap (Fig. 6).

Resolution of the ddRAD-seq-based phylogenetic tree within individual clades in the core Heterotropa and its implications for taxonomy

Below we discuss the taxonomic implications for individual clades within the core Heterotropa of the ML tree based on the total ddRAD-seq data sets.

Nipponicum clade (Supplementary data Fig. S10).

The Nipponicum clade consisted of ten of the 54 Heterotropa species sampled, although no morphological traits characterizing the clade were found. The Nipponicum clade has a relatively wide distribution in mainland Japan – ranging from Honshu to Shikoku; the present biogeographic history reconstruction estimated that their ancestor had arisen in the Pacific Coast side of Honshu with the split from the common ancestor with Asperum and Sakawanum clades (Fig. 7). The circumscription of the species within this clade is the most problematic among the Heterotropa. For example, A. takaoi, A. fauriei, A. nipponicum, A. rigescens and A. savatieri were not recovered as monophyletic. Also, some undescribed taxa such as A. sp. ‘Natadera’ and A. sp. ‘Echizen’ constitute distinct clades. Thus, additional phylogenomic analyses with more thorough sampling of populations is needed for the taxonomic revision of the Nipponicum clade in a phylogenetically relevant system.

Asperum clade (Supplementary data Fig. S11).

The Asperum clade consisted of five of the 54 Heterotropa species sampled, although, again, no morphological traits characterizing the clade were found. The Asperum clade had a wide distribution from mid-Honshu west to Shikoku and the northern half of Kyushu, and their common ancestor was estimated to have distribution across the Pacific Coast side of Mid-East Honshu and Shikoku (Fig. 7). Each species formed a distinct clade. The fact that A. tamaense, A. muramatsui and A. curvistigma were included in this clade was somewhat unexpected as these species had been considered to be closely related to A. asaroides and were traditionally placed in the (sub-)section Asaroides (Maekawa, 1932). This result indicates that the traditional infrageneric taxa (especially sub-sections) have little phylogenetic support and a major reclassification is required.

Sakawanum clade (Supplementary data Fig. S11)

The circumscription of the Sakawanum clade, in which three species and one variety were included, was the only consistent result between the ddRAD-seq-based and the Sanger sequencing-based phylogenetic analysis. The distribution of the species in the Sakawanum clade is mostly confined to Shikoku, with one species (A. minamitanianum) distributed in the adjacent area of Kyushu (coastal side of Miyazaki prefecture), which was surmised to be involved in the dispersal and vicariance events from the ancestral distribution in Shikoku (Fig. 7). These species are characterized by the unique trait that only longitudinal ridges are present on the inner surface of the calyx tube (Sugawara 2006; Takahashi et al., 2018). A detailed analysis of the morphological and genetic differentiation within this clade was performed by Takahashi et al. (2018), who discovered low genetic differentiation among the taxa despite the morphological divergence of calyx lobes, suggesting that divergent selection acted on the calyx lobe length. Taxonomic analysis of this discordance between the morphological and genetic divergence has not yet resulted in consensus.

Asaroides clade and Asarum hexalobum (Supplementary data Fig. S12).

The Asaroides + Asarum hexalobum clade consisted of eight of the 54 Heterotropa species sampled. For species for which we sampled multiple accessions (A. asaroides, A. kiusianum, A subglobossum and A. hexalobum), we confirmed that their circumscription in the current taxonomic system is consistent with the phylogenetic tree. Overall, the distribution of this clade is centred on Kyushu Island and adjacent small islands, with the exceptions of A. asaroides and A. hexalobum var. hexalobum, whose distribution extends to western Honshu Island. Another possible exception is A. satsumense, a species considered endemic to southern Kyushu but recently discovered in Taiwan (Lu et al., 2010). However, we doubt the species identity in this record because the study was based on one specimen from Taiwan, and A. satsumense is morphologically very similar to A. macranthum of the Okinawa–Taiwan clade, which is endemic to Taiwan. As we did not find any morphological characters that could differentiate between the Asaroides and Okinawa–Taiwan clades, this enigmatic ‘A. satsumense’ from Taiwan must be included in a phylogenomic analysis of Heterotropa to clarify its identity.

Notably, there was major topological uncertainty regarding the placement of A. hexalobum. In contrast to the ML tree, which placed A. hexalobum as sister to the Asaroides clade (Fig. 5), the SVD-quartet tree placed this species as sister to the Asaroides + Lutchuense + Amami + Okinawa–Taiwan + Mainland China clade although the branch support for this relationship was weak (<50 %; Fig. 6). Although the cause of this incongruence is beyond the scope of this study, it is possible that the genetic introgression between A. asaroides and A. hexalobum var. hexalobum might be responsible for this pattern because these two species occasionally grow in sympatry in locations where natural hybrids are reported (e.g. Asarum × miyakeanum F. Maekawa nom.nud.; Oka et al., 1972). Careful inspection of any contradicting phylogenetic signals within the ddRAD-seq data sets will be helpful in addressing this problem.

Lutchuense clade (Supplementary data Fig. S13).

The Lutchuense clade consisted of five of the 54 Heterotropa species sampled. We confirmed that all species except one were recovered as distinct monophyletic groups. The only exception was A. kumageanum, two varieties of which formed distinct, non-sister clades, indicating that these two varieties should be treated as different species. The distribution of the described species of the Lutchuense clade is confined to northern and central Ryukyus. However, we revealed that the two undescribed species from southern Kyushu, A. sp. ‘Hiuganum’ and A. sp. ‘Toimisakiensis’, also belong to this clade. A notable pattern of the distribution of the species in Lutchuense is that of A. kumageanum and A. lutchuense, a pair of sister species that are distributed across north and central Ryukyus. Given that the land connection between north and central Ryukyus dates back to at least 1.3 MYA (Kimura, 2000), and there are no instances of disjunct distributions among the 62 Heterotropa species, over-sea migration is unlikely to have occurred, and thus the split of this species pair most probably took place prior to 1.3 MYA.

Amami clade (Supplementary data Fig. S14).

The circumscription of the Amami clade, which consisted of ten species endemic to the Amami Islands, is a major finding of this study. We found that all species native to the Amami Islands except one (A. lutchuense in the Lutchuense clade) belong to this clade. This is somewhat surprising because these species are diverse in both their habitat and their floral morphology, indicating that local radiation has taken place on a small landmass (approx. 960 km² at present). Although each species is endemic to one of the two major Amami Islands (Amami-oshima Island including the adjacent Kakeroma Island, and Tokunoshima Island), it is noteworthy that the two subclades of this clade do not correspond to the species of the two islands, i.e. one subclade was composed of A. pellucidum, A. nazeanum, A. celsum, A. gusk, A. tabatanum and A. fudsinoi (endemic to Amami-oshima Island), and the other subclade was composed of A. trinacriforme (endemic to Amami-oshima Island), A. hatsushimense, A. simile and A. leucosepalum (endemic to Tokunoshima Island). We estimated that the ancestors of the species endemic to Tokunoshima Island migrated from Amami-oshima Island (Fig. 7), most probably via a land bridge that may have existed between the two islands in the past. Despite the presumed rapid radiation of the clade, our analysis in part supported the current circumscription of the species (A. pellucidum, A. trinacriforme, A. simile and A. leucosepalum) in this clade. It also suggested the non-monophyly of A. fudsinoi, which probably consists of at least two distinct clades with different floral morphologies (Y. Okuyama, pers. obs.). In any case, the sampling of more populations and a morphological re-evaluation are necessary to precisely delimit the species of this clade.

The species of the Amami clade grow either sympatrically or allopatrically in a variety of sloped forest floors along streams, on ridge-tops and in limestone areas. Matsuda et al. (2017) discovered that very little gene flow occurred among the four sympatric species on Amami-oshima Island (A. pellucidum, A. celsum, A. gusk and A. fudsinoi) despite the potential for cross-fertilization and very low genetic differentiation, indicating that mechanisms to reproductively isolate these species exist. Although the ecological basis of the origin and maintenance of species diversity in the Amami clade is unknown, it is likely that both pollination system and flowering phenology have played important roles, because each species has highly differentiated floral traits, including flowering seasons and floral colour, shape and scent (Matsuda et al., 2017; Y. Okuyama, pers. obs.). Overall, further efforts are needed to elucidate the putative rapid speciation in this clade.

Okinawa–Taiwan clade (Supplementary data Fig. S15).

The Okinawa–Taiwan clade comprised 12 of 54 Heterotropa species sampled in the present study. Members of this clade are distributed along the Ryukyu arc between Okinawa island and Taiwan. Three sub-groups are recognizable within this clade, two of which are found on Iriomote Island (southern Ryukyus) and Taiwan. The other sub-group, consisting of A. okinawensis, A. senkakuinsulare and A. dissitum, is native to the north-eastern part of the distribution range of the Okinawa–Taiwan clade, and was sister to the other two subclades in the phylogenetic tree. This pattern suggests that a single colonization northward from the southern Ryukyus (or the ancient Taiwan–southern Ryukyus area) occurred during the speciation of this clade (Fig. 7).

For the circumscription of the species, the monophyly of all six species with multiple accessions, including A. yaeyamense, which is native to Iriomote Island and Taiwan, was supported. Nevertheless, a more thorough sampling of the species and populations native to Taiwan, particularly the four species endemic to Taiwan that were not included in our study, is needed to understand the pattern of diversity and species circumscription in this clade.

South-east Mainland China clade.

Among the 54 Heterotropa species sampled, only A. ichangense was classified into this unique clade. Although there was slight topological inconsistency, the South-east Mainland China clade was closely related to the Okinawa–Taiwan and Amami clades (Figs 5 and 6), suggesting that their common ancestor originated at the time when mainland China, mainland Japan and Ryukyus were connected (Fig. 7). At present, we do not know if any more species, presumably those native to mainland China, belong to this clade. However, judging from the karyological, morphological and phylogenetic data available for the species native to China (most of them probably belonging to sect. Longistylis; Supplementary data Table S1), it is likely that few, if any, other species belong to this clade.

Future perspectives on the evolutionary ecology of the Heterotropa radiation

The geographic pattern of a plant phylogeny may have implications for the distribution of other organisms associated with the plants, provided that the host use of such organisms is evolutionarily conservative (Winkler and Mitter, 2008). For example, the butterfly species Luehdorfia japonica (Papilionidae), which is a specialist herbivore of the Heterotropa (Nishida, 1994), is endemic to Honshu Island, where the Nipponicum and Asperum clades dominate. Luehdorfia japonica exhibits a strong preference for species of Heterotropa such as A. takaoi, A. blumei, A. megacalyx (the only tetraploid species of Heterotropa not included in the present analysis) and A. asperum. Because the genus Asarum has poor long-distance dispersal capability with ant-dispersed seeds (Kelly, 1998), it would be reasonable to assume that highly mobile butterfly species have much greater dispersal ability than the host plants. If this is true, the phylogeny of Heterotropa species may be the major limiting factor of the distribution of this butterfly. A fine-scale phylogeographic study of L. japonica with a close look at population-level host use would enable the intimate relationships, such as the chemical coevolutionary arms race, between these herbivores and plants to be elucidated.

One of the most intriguing topics in species radiation is the ecological factors that facilitate species diversity. Heterotropa species are remarkably diverse in floral traits including size, colour, shape and scent, and this is true even at smaller scales, such as among the closely related species of the Sakawanum clade (Kakishima and Okuyama, 2018; Takahashi et al., 2018; Supplementary data Figs S10–S15, which show the diversity of floral morphology among closely related species within each subclade of Heterotropa). Because of the pattern of floral diversity, one can reasonably assume that the differentiation of pollination systems might have been responsible for the radiation of Heterotropa, and this view has been partially supported by a previous study (Sinn et al., 2015b). Although there have been few observations of their pollination (e.g. Sugawara, 1988), a common feature of Heterotropa species is the rewardless, chamber-like flower with a characteristic scent profile, suggesting that deceptive pollination (most probably brood-site mimicry systems using small flies) prevails (see also Mesler and Lu, 1993, for evidence of brood-site mimicry in other sections of the genus Asarum). This pattern somewhat parallels the radiation of other plant lineages with deceptive flowers, such as stapeliads (Apocynaceae; Bruyns et al., 2014, 2015), among which variable floral traits including morphology and scent have been observed in association with the diversity of the fly pollinators (Jürgens et al., 2006; Ollerton et al., 2009, 2017). Our findings will enable further studies of plant radiation associated with deceptive flowers and comparative studies of independent plant groups to identify the ecological and evolutionary mechanisms underlying such radiation processes.

Conclusions

Using a ddRAD-seq-based phylogenomic approach, we resolved the overall history of the radiation of Heterotropa centred on Japan and Taiwan. The resultant phylogenetic tree identified A. forbesii, the only deciduous Heterotropa species native to mainland China, as sister to all other species (core Heterotropa) and recognized nine subclades in the core Heterotropa. It also identified described species whose current circumscription does and does not have a phylogenetic basis, which is necessary for the improvement of their taxonomy and systematics. Moreover, our findings will facilitate further study of the ecological diversity of Heterotropa in an evolutionary context. In particular, using phylogenetic comparative methods, the highly resolved phylogenetic tree will allow the remarkable diversity of the floral traits of Heterotropa to be linked to pollinator diversity, which is probably a key factor in the ecological speciation of this fascinating group of plants.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Figure S1: ML tree of Asarum sect. Heterotropa based on the reduced, unmerged-only ddRAD-seq data set using the clustering threshold of 0.84. Figure S2: ML tree of Asarum sect. Heterotropa based on the reduced, unmerged-only ddRAD-seq data set using the clustering threshold of 0.88. Figure S3: ML tree of Asarum sect. Heterotropa based on the reduced, unmerged-only ddRAD-seq data set using the clustering threshold of 0.92. Figure S4: ML tree of Asarum sect. Heterotropa based on the full, unmerged-only ddRAD-seq data set using the clustering threshold of 0.84. Figure S5: ML tree of Asarum sect. Heterotropa based on the full, unmerged-only ddRAD-seq data set using the clustering threshold of 0.88. Figure S6: ML tree of Asarum sect. Heterotropa based on the full, unmerged-only ddRAD-seq data set using the clustering threshold of 0.92. Figure S7: a coalescent-based phylogenetic tree of Heterotropa samples reconstructed based on full SVD-quartet tree inference. Figure S8: an LTT plot for Heterotropa phylogeny fitting to the pure birth model with λ = 0.2457. Figure S9: a biogeographic history reconstruction onto the time-calibrated, modified ML tree of Heterotropa based on the DEC+J model. Figure S10: a partial ML tree that illustrates the phylogenetic relationships of species within the Nipponicum clade. Figure S11: a partial ML tree that illustrates the phylogenetic relationships of species within the Asperum and Sakawanum clades. Figure S12: a partial ML tree that illustrates the phylogenetic relationships of species within the Asaroides clade and A. hexalobum. Figure S13: a partial ML tree that illustrates the phylogenetic relationships of species within the Lutchuense clade. Figure S14: a partial ML tree that illustrates the phylogenetic relationships of species within the Amami clade. Figure S15: a partial ML tree that illustrates the phylogenetic relationships of species within the Okinawa–Taiwan clade. Table S1: the tentative classification of 128 spp. into the six sections of the genus Asarum proposed in the present study. Table S2: the plant specimens and nucleotide sequence data analysed in the present study. Table S3: amplification and sequencing primers used in the present study.

FUNDING

This study was supported by the research programme of the National Museum of Nature and Science (Biological Properties of Biodiversity Hotspots in Japan), by the Japan Society for the Promotion of Science KAKENHI grants [grant nos 24657065, 15H05604, and 19H03292 (to Y.O.) and 25290085 (to G.K.]) and by the 28th Botanical Research Grant of the Ichimura Foundation For New Technology (to Y.O.).