Taxonomic study on Japanese Salvia (Lamiaceae): Phylogenetic position of S. akiensis, and polyphyletic nature of S. lutescens var. intermedia

Abstract Both Salvia akiensis and S. lutescens (Lamiaceae) are endemic to Japan. Salvia akiensis was recently described in 2014 in the Chugoku (= SW Honshu) region, and each four varieties of S. lutescens distributed allopatrically. Among varieties in S. lutescens, var. intermedia show a disjunctive distribution in the Kanto (=E Honshu) and Kinki (= W Honshu) regions. Recent field studies of S. lutescens var. intermedia revealed several morphological differences between the Kanto and Kinki populations. Here, I evaluated these differences among Salvia lutescens var. intermedia and its allies with morphological analysis and molecular phylogenetic analyses of nuclear ribosomal DNA (internal and external transcribed spacer regions) and plastid DNA (ycf1-rps15 spacer, rbcL, and trnL-F) sequences. Both morphological analysis and molecular phylogenetic analyses showed that S. lutescens var. intermedia from the Kinki region and var. lutescens were closely related to each other. However, var. intermedia from the Kanto region exhibited an association with S. lutescens var. crenata and var. stolonifera, which also grew in eastern Japan, rather than var. intermedia in the Kinki region. These results indicated that S. lutescens var. intermedia is not a taxon with a disjunctive distribution, but a combination of two or more allopatric taxa. Present study also suggested that S. akiensis was most closely related to S. omerocalyx.

There are four varieties known in S. lutescens (Koidz.) Koidz.: var. crenata, var. intermedia, var. lutescens, and var. stolonifera (Murata 1952, Yonekura andKajita 2003 onwards). Fukuoka and Kurosaki (1982) noticed distribution of each taxon does not overlap and clarified that the distribution of var. crenata on the Japan Sea side of Central to Northern Honshu, var. stolonifera on the Pacific side of Central Honshu, var. lutescens around the Suzuka Mountain range (Mie Pref., W Honshu), and the disjunctive distribution of var. intermedia in the Kanto (E Honshu) and Kinki regions (W Honshu) based on herbarium works.
Takano and Okada (2011) conducted molecular phylogenetic analyses of Japanese Salvia and found that the species were distributed among three subclades: (1) S. plebeia (subg. Sclarea), (2) subg. Salvia, and (3) subg. Allagospadonopsis. They also found four varieties of S. lutescens that did not form a monophyletic group; instead, they were dispersed among several clades in phylogenetic trees, based on both plastid DNA (cpDNA) and nuclear ribosomal DNA (nrDNA) data, and their topologies were not concordant with each other. In addition, they became paraphyletic in the phylogenetic trees based on combined cpCNA and nrDNA data (Takano and Okada 2011). Furthermore, during a recent field survey, I noticed that S. lutescens var. intermedia in the Kanto and Kinki regions had different morphological characteristics. The basal part of the anther connective was generally glabrous in the Kanto population, but it was pilose in the Kinki population. Also, in the Kanto population, the stalk of the inflorescence declinated toward the ground after flowering, and it typically became proliferous; in contrast, in the Kinki population, the inflorescence grew erect, and it was never proliferous.
Recently, a new species of Japanese Salvia, S. akiensis A.Takano, T.Sera et Kurosaki has been described from Shimane and Hiroshima Prefectures (Takano et al. 2014). At the moment, this species shows disjunctive distribution, ca. 40 km away from each, and the habitat is also very different between Hiroshima and Shimane: it grows among bamboo by roadsides and on slopes below evergreen mixed forests and plantations in Shimane (Sakoda et al. 2014), but it is found in moist, shallow soil on rock walls by streams in deciduous forests in Hiroshima (Takano et al. 2014). Therefore, it may wonder if the species be monophyletic. Takano et al. (2014) discussed relationships among S. akiensis, S. isensis Nakai ex H.Hara, S. lutescens and S. omerocalyx Hayata based on morphological characters, but molecular phylogenetic position of S. akiensis remains unclear.
As a step toward taxonomic revision of variety of S. lutescens and to confirm monophyly and phylogenetic position of S. akiensis, morphological and molecular phylogenetic analyses were conducted. Takano and Okada (2011) followed the Murata and Yamazaki (1993) system in which treated var. intermedia as a forma f. lobatocrenata and var. lutescens as f. lutescens, however, here I follow the Murata (1952) system (=Ylist, Yonekura and Kajita 2003 onwards), and each infraspecific taxon of S. lutescens is treated as a variety.

Materials and methods
Morphological analyses on S.lutescens in herbaria Murata (1952) studied morphological variations in the plants of subgen. Allagospadonopsis in Japan and found hairiness, number or shape of leaflets, presence /absence of glandular hairs were so variable and could not be used as diagnostic characters. Diagnostic characters separated each variety of S. lutescens are indumentums of the basal part of the anther connective and floral color (Nakai 1950, Murata 1952. Among varieties, var. lutescens shows pale yellow flowers and pilose at the base of anther connective, var. intermedia shows deep violet corolla and pilose at the base of anther connective, var. crenata does purple corolla and glabrous base of anther connective. Floral color and indumentums of var. stolonifera is same as var. crenata, however, var. stolonifera extends its stolon after anthesis (Nakai 1950). Since it is difficult to know exact floral color by examining dry specimens, the indumentums at the base of the anther connective were observed for glabrousity in selected specimens, which bore at least several flowers. A total of 89 specimens of S. lutescens var. intermedia, including its syntypes, of the 34 specimens are from Kanto region and 55 from Kinki, were examined in the following herbaria: the Museum of Nature and Human Activities, Hyogo (HYO); the Kanagawa Prefectural Museum (KPM); Kyoto University (KYO); Tokyo Metropolitan University (MAK), the Osaka Museum of Natural History (OSA), and The University of Tokyo (TI) (Appendix 1). Additionally, all the specimens of S. lutescens var. lutescens including its holotype at KYO were examined on the same characters, since no information on that character is available.

DNA extraction, PCR, and DNA sequencing
The protocols for DNA extraction, polymerase chain reaction (PCR), purification, and DNA sequencing were described previously by Takano and Okada (2011). The PCR conditions and the PCR and sequencing primers for rbcL, the trnL-F intergenic spacer region of cpDNA (trnL-F), and the internal transcribed spacer (ITS) region of nuclear ribosomal DNA (nrDNA) were also described previously by Takano and Okada (2011). To amplify in the ycf1-rps15 spacer region found in cpDNA (ycf1-rps15), 5711f and rps15r (both from Drew and Sytsma 2011) were used as PCR primers in PCR assays, and ETS-bdf1 Sytsma 2011) and18S-E (Baldwin andMarkos 1998) were used to amplify the external transcribed spacer (ETS) sequence from 18S-26S ribosomal DNA. The four PCR primers were also used for sequencing. The PCR conditions for amplifying the two loci were: denaturation at 95 °C for 3 min, followed by 40 cycles at 95 °C for 30 s, 54 °C for 30 s, and 72 °C for 30 s; and a final extension at 72 °C for 5 min.

Sequence alignment and phylogenetic analysis
Raw sequence data were assembled and edited manually, with BioEdit software (ver. 7.2. 5 Hall 1999) DNA sequences were aligned with the CLUSTALW 1.83 software package, with default settings and multiple alignments (Thompson et al. 1994). Alignments of the rbcL, trnL-F, and ycf1-rps15 sequences of cpDNA, and the ITS and ETS sequences of nrDNA were combined. Gaps were deleted.
Compared to Takano and Okada (2011), the ETS (Baldwin and Markos 1998) and ycf1-rps15 of cpDNA (Dong et al. 2015) were newly sequenced for all samples. Further, two individuals of S. akiensis and three of S. lutescens var. intermedia, three of S. lutescens var. crenata, and one each of S. isensis, S. japonica var. japonica, S. lutescens var. lutescens, and S. plebeia were newly added for the analysis. The sampling sites of S. lutescens group were shown in Fig. 1. A total of 36 individuals of Salvia were used, including all the Salvia taxa from Japan and one Taiwanese Salvia (S. arisanensis Hayata). Salvia polystachya M. Martens et Galeotti and S. plebeia were selected as outgroup; the former species belonged to clade II sensu Maria and Classen-Bockhoff (2014), which became a sister to group IV and contained the East Asian Salvia; the latter species became a sister to a species of the subgenus Allagospadonopsis and Salvia (Hu 2015). Materials, accession numbers for the sequences, vouchers, and references to the literature are presented in Table 1. The sampling sites for the varieties of S. lutescens are shown in Fig. 1.
The incongruence length difference (ILD) test (Farris et al. 1994) was used to evaluate congruence between the chloroplast and the nuclear data sets. 100 replications were performed using PAUP*4.010b (Swofford 2002). As the ILD test (P < 0.01) suggested incongruence between the two datasets, and the topologies also exhibited discordance, I performed separate analyses for the cpDNA and the nrDNA data. Maximum Likelihood (ML) and Bayesian inference (BI) were used. Nucleotide substitution model parameters were determined for each partition by gene was evaluated with  analyses were completed using TREEFINDER version March 2011 (Jobb et al. 2004). A replicated (500 iterations) partitioned analysis was performed with bootstrap (1000 rounds) using AICc separated model for nrDNA dataset and AICc proportional model for cpDNA dataset. Bayesian evolutionary analysis using partitioned datasets were run in BEAST v.1.8.3 (Drummond et al. 2012, Heled andDrummond 2010) with 20 million Markov Chain Monte Carlo (MCMC) iterations, under an uncorrelated relaxed clock (Drummond et al. 2006), Yule process of speciation with a random starting tree for each partition. Convergence of the chains was checked using the program Tracer 1.6 (Rambaut et al. 2014). High effective sample sizes were observed for all parameters (posterior ESS values > 200 for the combined analyses). Maximum clade credibility trees with divergence times means and 95% highest probability densities (HPDs) were produced using Tree Annotator (Drummond et al.2012).

Morphological characteristics
Among the 89 specimens of S. lutescens var. intermedia examined, 52 specimens from the Kinki region were pilose at the base of the anther connective (Fig. 2), and no speci-  mens from the Kanto region shared this characteristic (Appendix 1). Ten specimens collected from the Kanto region had at least one, but less than 10 hairs. Twenty-four specimens from the Kanto region (Fig. 2)  Totally, 18 specimens of S. lutescens var. lutescens were deposited at KYO and examined, 13 of these had pilose at the base of the anther connective (Appendix 1). Four of these had no flowers, and only one specimen, M.Hara s.n., collected from Mt. Takami, Maze-Mura, Iinan-gun, Mie Pref. showed glabrousity.

Phylogenetic positions of Japanese taxa in the genus Salvia
A likelihood analysis using the concatenate cpDNA datasets (rbcL+trnL-F+ycf1-rps15 spacer) for 36 individuals in 23 taxa resulted in a ML tree with -lnL = 5295.264. The ML and Bayesian trees had similar topology; the Bayesian maximum clade credibility tree is shown with ML bootstrap (ML-BS) and Bayesian posterior probability (BI-PP) in Figure 4. The Japanese and Taiwanese species of subg. Allagospadonopsis formed a well supported clade (ML-BS/BI-PP, 100/0.97). Two subclades were found in the subg. Allagospadonopsis clade: (1) S. japonica + S. pygmaea + one S. akiensis + S. arisanensis + five individuals of S. lutescens in E Japan subclade, and (2) one S. akiensis (S1), two S. isensis, S. lutescens in Kinki + S. ranzaniana + two S. lutescens in the Kanto region

Discussion
This study suggests that S. lutescens var. intermedia is polyphyletic. Four individuals of var. intermedia, two from the Kanto and two from the Kinki region fell into different subclades in both molecular phylogenetic trees using cpDNA and nrDNA datasets, although the two from the Kinki region were always in the same subclade (Figs 4, 5). The plants of var. intermedia from the Kanto region (Tanzawa and Mt.Mikuni) fell into the same subclade in the nrDNA tree together with var. crenata, var. stolonifera, Figure 5. The Bayesian maximum clade credibility tree derived from nuclear ribosomal DNA (concatenate dataset of ETS and ITS). ML-bootstrap/Bayesian-PP numbers are shown above or below the corresponding branch. Thick lines denote a clade that was strongly supported with ML-bootstrap and/or Bayesian-PP values greater than 95 %. ML: maximum likelihood; PP: posterior probability.

S. lut. v. lutescens (Mie)
Subg. Allagospadonopsis anther connective pilose KINKI Reg. and S. isensis whereas they fell into different subclades in the cpDNA tree. Such a contradiction might indicate that var. intermedia from the Kanto region have multiple origins, or might have originated via hybridization or introgressive gene flow between nighbouring taxa (e.g., Sudarmono and Okada 2007). The discordance between nr DNA and cpDNA data is common in the mint family (Trusty et al. 2004, Moon et al. 2010, Drew and Sytsma 2013, Deng et al. 2015, and chloroplast-based phylogeny often does not reflect their morphological relationships, which can be explained by chloroplast capture (Rieseberg and Soltis 1991). Morphological analysis also supports the contention that var. intermedia is polyphyletic, as the specimens of var. intermedia studied showed in the indumentums at the base of the anther connective: pilose in the plants from the Kinki region, and glabrous in the plants from the Kanto region (Fig. 3).Therefore it is clear that var. intermedia from the Kinki region and the taxon from the Kanto region are different entities, suggesting that var. intermedia is not a taxon that shows disjunctive distribution, but is instead admixture of two or more biological entities. Additionally, as mentioned in introduction, after flowering the stalk of the inflorescence becomes declinate to ground and usually proliferous in case of the plants from the Kanto region, but never become declinate in the plants from the Kinki region. The indumentums at the base of anther connective is effective to select pollinators to avoid intrusion of insects which could not be effective pollinators (R.Classen-Bockhoff pers. Comm..) However, pollinators of var. intermedia in the Kinki and the Kanto region are not different (=Bombus (Diversobombus) diversus diversus, some Halictidae, and Syrphidae. Takano 2017). Habitat is also similar: halfshaded, on mesic soils along streamlet on the forest floor of deciduous forests. They might have begun to be diverged from each other after long geographical isolation.
On the contrary, present morphological and molecular phylogenetic analyses indicated that S. lutescens var. lutescens and var. intermedia from the Kinki region are closely related to each other. In molecular phylogenetic analysis, they formed a cluster in both cpDNA-and nrDNA trees, though ML-BP/BI-PP support was not strong in cpDNA tree. The morphological study revealed var. lutescens is pilose at the base of the anther connective: therefore, S. lutescens var. intermedia in the Kinki region share the same morphological status with var. lutescens. The distribution of var. lutescens is very near to that of var. intermedia in the Kinki region (Mie, Shiga, Nara Prefs.), although var. lutescens and populations of the Kinki regions of var. intermedia have never been found to grow together.
Salvia lutescens var. intermedia in the Kanto region may be more closely related to var. crenata and var. stolonifera. Murata (1952) mentioned that the base of anther connective is glabrous in var. stolonifera and var. crenata. The present study revealed that var. intermedia in the Kanto region shares this character with those two taxa. Salvia lutescens var. intermedia in the Kanto region formed a strongly supported sucblade with var. crenata, var. stolonifera and S. isensis in nrDNA phylogenetic tree. In the cpDNA phylogenetic tree, S.lutescens var.intermedia from the Kanto region (Mt.Mikuni) was included in the subclade containg S. akiensis, S. japonica, S. lutescens var. crenata, and S. pygmaea whereas S. lutescens var. intermedia (Tanzawa) formed a subclade with var. stolonifera and was included in the subclade containing S. akiensis, S. omerocalyx, S. ranzaniana, and S.lutescens var. intermedia from the Kinki + S. isensis. These findings suggest a close relationship among var. crenata, var. stolonifera, and var. intermedia from the Kanto region. Var. intermedia from the Kanto region may belong to var. stolonifera and var. crenata. The identity of var. intermedia and other varieties of S. lutescens are needed to be re-evaluated, and further study is necessary towards revision of varieties of S. lutescens.
The phylogenetic analyses also suggest that S. akiensis comprises a monophyletic group, as indicated by nrDNA tree, and that most of the species allied to S. akiensis was the S. omerocalyx group. Salvia akiensis and S. omerocalyx group comprised a subclade in nrDNA (ML-BS/BI-PP: 89/1.00). These two taxa did not form a subclade in cpDNA, but it may be of introgression/chloroplast capture /hybridization as mentioned above. In contrast, S. akiensis and S. omerocalyx share following characters: bearing the largest flowers among species in the subg. Allagospadonopsis, flower from May to June, and exhibit gynodioecy (Takano 2013;Takano et al. 2014). These characters are assumed to be symapomorph.