We visited two known localities of Hosta alata reported by Arakane (2001): Mt. Karasu-dake on September 24, 2020, and Mt. Sobo on September 26, 2020. Mt. Karasu-dake (683 m above sea level) is located at the end of the ridge extending northeast from Mt. Sobo (1756 m above sea level), and there is no rocky area between Mt. Sobo and Mt. Karasu-dake that is suitable for Hosta alata habitats. Below, the population of Mt. Karasu-dake will be referred to as H. alata 1, and the population of Mt. Sobo will be referred to as H. alata 2. We also collected H. kikutii F. Maek. var. kikutii, H. longipes (Franch. & Sav.) Matsum. var. caduca N. Fujita, H. pulchella, and two undescribed species (labeled H. sp. 1 and H. sp. 2) from eight localities on the main island of Kyushu, and an additional sample of H. longipes var. caduca from the Kochi Prefecture, Shikoku Island (Table 1). A total of 29 samples collected from August to September 2020 were included in the following analysis. In addition, we collected samples of Hosta widely from Shikoku and Honshu, examined phylogenetic relationships, and confirmed that other Japanese species are remotely related to Hosta alata (data not shown).

Total DNA was extracted from dried leaves using the CTAB method (Doyle and Doyle 1990). De novo SNP discovery was performed using MIG-seq. Based on methods reported by Suyama and Matsuki (2015), we prepared an MIG-seq library using a two-step PCR amplification process with minor modifications; the annealing temperature of the first PCR was altered from 48 °C to 38 °C. The second-round PCR products were purified in the size range of 300–800 bp and sequenced using the Illumina MiSeq platform (Illumina, San Diego, CA, USA) and the MiSeq Reagent Kit v3 (150 cycles, Illumina). The sequencing of the first 17 bases of reads 1 and 2 (SSR primer regions and anchors) was skipped using ‘DarkCycle’. Low-quality reads and extremely short reads containing adapter sequences were removed using Trimmomatic 0.39 (Bolger et al. 2014). The Stacks 2.41 pipeline software (Catchen et al. 2013; Rochette et al. 2019) was used to obtain individual genotypes with the following parameters: minimum depth of coverage required to create a stack (m) = 3, maximum distance between stacks (M) = 2, maximum mismatches between loci when building the catalog (n) = 2. Three different filtering criteria were considered for quality control of the SNP data. First, any SNP site where one of two alleles had less than three counts was filtered out because it was difficult to distinguish polymorphisms from sequencing errors when the minor allele count of SNPs is too low (Roesti et al. 2012). Second, loci containing SNPs with high heterozygosity (Ho ≥ 0.6) were removed because excess heterozygosity might have resulted from artifactual loci built from several paralogous genomic regions. Third, a SNP was excluded if the number of samples shared by the SNP was below the reference value R; the minimum percentage of samples that retained a SNP. As the resolution of phylogenetic trees depends on R (Wagner et al. 2013), we used four SNP datasets in which the reference value R was changed in the following four steps: R = 0.1, 0.3, 0.5, and 0.8. As is described in Results, the basal topology of phylogenetic trees did vary with R values. The variability of topology with R values is probably derived from the limited sample sizes of H. alata 1, H. alata 2, and H. sp.1, ranging from 3 to 5. Under these low sample sizes, phylogenetic reconstructions using datasets with high R values tend to neglect the presence of SNPs unique to each lineage. Furthermore, phylogenetic reconstruction using datasets with low R values can contain noise with artifacts. The variability of the topology is expected to be reduced by using more samples for each lineage, but we could collect limited number of samples for H. alata 1, H. alata 2, and H. sp.1 because these plants were found on high vertical cliffs.

Phylogenetic trees based on SNPs was inferred using the maximum likelihood method implemented in RAxML 8.2.10 (Stamatakis 2014). We used a GTRCAT model and performed 1,000 replicates of parallelized tree search bootstrapping. Considering a possibility of reticulate relationship due to past hybridization, we also reconstructed a split network using SplitsTree4 4.14 (Huson and Bryant 2006) by implementing neighbor-net algorithm with the uncorrelated P distance matrix calculated from the SNPs matrix. We also performed an analysis of population genetic structure using Structure 2.3.4 (Pritchard et al. 2000). This software assumes a population genetic model in which there are K populations, each of which has a unique set of allele frequencies at each locus, and the individuals in the sample are probabilistically assigned to populations. If a multi-locus genotype of an individual indicates a sign of mixture, it is assigned to more than one population. To estimate the log-likelihood for each model with a different number of populations (K = 1–10), we performed 20 independent runs with a burn-in of 100,000 steps and additional 100,000 steps with the admixture model. Optimal K values were determined by using the Delta K method of Evanno et al. (2005) in Structure Harvester (Earl and Vonholdt 2012).

A total of 17,666,364 raw reads (609,185 ± 16,586 reads per sample) were obtained by performing MIG-seq. After quality control, 16,924,537 reads (583,605 ± 16,419 reads per sample) were used for further analyses. After de novo SNP detection and filtering, the following four datasets with different R values were used: R = 0.1, 16,510 loci, 31,769 SNPs; R = 0.3, 8,163 loci, 18,296 SNPs; R = 0.5, 3,345 loci, 8,037 SNPs; and R = 0.8, 460 loci, 1,069 SNPs.

A phylogenetic tree reconstructed using MIG-seq by setting R = 0.5 (Fig. 1; SNPs shared among 15 samples of the 29 total were used) recovered Hosta alata as sister to H. pulchella (98% bs). The two known populations of H. alata, Karasu-dake (H. alata 1: JPN2281–2283) and Sobo (H. alata 2: JPN2350, JPN2351, JPN2372) were each recovered monophyletic (100 bs), and these form a well-supported clade (100 bs). Three samples from Mt. Mukabaki (JPN2208–2210) and two samples from Mt. Oninome (JPN2012, 2013) formed a clade supported by a 100% bootstrap value, designated as H. sp. 1 (Fig. 1). The clade that included H. alata, H. pulchella, and H. sp. 1 (100 bs) was sister to the remaining sampled species including H. kikutii, H. longipes var. caduca, and a sample from Mt. Sobo designated as H. sp. 2 (100 bs).

Figure 1. 

A molecular phylogeny of Hosta alata and its relatives reconstructed on the basis of MIG-seq data. Bootstrap values are shown on the internodes. The scale bar represents the average number of substitutions per SNP site.

The topology of the phylogenetic reconstructions using MIG-seq data varied with the setting of R, the minimum percentage of samples that shared a SNP (Fig. 2). When R = 0.1 (SNPs shared by three or more samples are used; Fig. 2A), or R = 0.3 (SNPs shared by nine or more samples are used; Fig. 2A), Hosta alata 1, H. alata 2, and H. pulchella were trichotomous, and the monophyly of the clade including these three was supported by a 100% bootstrap value; H. sp. 1 was sister to this clade. When R = 0.5 (SNPs shared by 15 or more samples were used; Fig. 2B), the two H. alata populations were sister to H. pulchella. When R = 0.8 (SNPs shared by 24 or more samples are used; Fig. 2C), H. alata 1 and H. alata 2 were not monophyletic; H. alata 1 was sister to H. sp. 1 and H. alata 2 was sister to H. pulchella (Fig. 2C).

Figure 2. 

Three different topologies of Hosta phylogenetic trees reconstructed on the basis of MIG-seq conducted using different settings for the proportion of missing loci, R A the topology when R = 10 or 30 B the topology when R = 50 C the topology when R = 80. Bootstrap values are shown on the internodes.

A split network reconstructed using R = 0.5 (Fig. 3) showed that Hosta pulchella, H. alata 1, H. alata 2, and H. sp. 1 formed four distinct clusters. Hosta alata 2 was placed between H. pulchella and H. alata 1. Two samples of H. longipes var. caduca were clustered together, but nested within H. kikutii var. kikutii. Hosta sp. 2 was placed outside the cluster including H. longipes var. caduca and H. kikutii var. kikutii.

Figure 3. 

A split network of Hosta alata and its relatives reconstructed on the basis of MIG-seq data by setting R = 0.5.

In the Structure analysis, using R = 0.5, K = 4 or 5 was optimal: delta K increased from K = 3 to K = 4, took almost the same value at K = 4 and K = 5, and decreased from K = 5 to K = 6. When K = 4 (Fig. 4), H. pulchella, H. alata 1, H. sp. 1, and a group including H. kikutii var. kikutii, H. longipes var. caduca represented genetically unique populations, and H. alata 2 showed a mixture of genetic elements derived from H. pulchella and H. alata 1. Genetic elements from H. sp. 1 were found in three other genetically unique populations, albeit infrequently. When K = 5, H. alata 2 and some individuals of H. pulchella represented the fifth genetically unique populations, and thus H. pulchella was shown to be genetically heterogeneous. Hosta alata 2 had some genetic elements derived from H. alata 1.

Figure 4. 

Population genetic structure of Hosta alata and its relatives at K = 4 and 5 inferred from MIG-seq data by setting R = 0.5.

Figure 5. 

Holotype of Hosta alata Hatusima ex Yahara. (Nogami s.n. KAG151816, available from https://dbs.kaum.kagoshima-u.ac.jp/musedb/s_plant/picture/KAG151816/KAG151816.jpg).

On Mt. Karasu-dake, a small population of Hosta alata 1 is found on cliffs along the ridge line at 676 m elevation. On Mt. Sobo, H. alata 2 grows along the ridge line at 1500 m elevation, where we also found a population of H. pulchella. While H. alata 2 grows on steep cliffs, H. pulchella grows in crevices of rocks along the mountain path. In the vicinity of the peak of Mt. Sobo at 1756 m elevation, we found only H. pulchella growing in rock crevices. We found several (fewer than 10) flowering H. alata 1 in Mt. Karasu-dake on September 24, 2020, but all H. alata 2 plants observed in Mt. Sobo on September 26, 2020, were fruiting (fewer than 10) or sterile (ca. 20). In contrast, we collected a few flowering H. pulchella specimens at the peak of Mt. Shojidake, located 2 km south of Mt. Sobo, on September 25, 2020; however, other plants observed in Mt. Shojidake were fruiting or sterile.

According to the photographs, sketched illustrations, and description of H. alata 1 (Arakane 2001) and our own observations of H. alata 1 and H. alata 2, Hosta alata plants in Mt. Karasu-dake and Mt. Sobo were indistinguishable in both floral and vegetative traits. Hosta alata was similar to H. pulchella, as evidenced by the presence of dark purple veins inside perianth lobes (Fig. 6C); bracts in anthesis are vivid (not withering) and erect or diagonally spreading, and leaf blades are ovate or oblong-ovate with smooth veins on the lower surface. However, H. alata was distinguished from H. pulchella by the presence of more leaves (5–9 vs. 3–4), larger leaf blades (8.5–24.5 cm long vs. 2.7–8.0(–8.9) cm long), more lateral veins (5–9 pairs vs. 3–4 pairs), wider-winged petioles (0.4–1.4 cm wide vs. 0.2–0.4 cm wide), more flowers with 9–40 flowers per scape vs. 3–4 flowers, longer pedicels (1.1–2.3 cm long vs. 0.5–0.8 cm long), and fertile bracts which are purplish green in color (vs. pale green).

Figure 6. 

Depiction of Hosta alata Hatusima ex Yahara in its natural habitat in Mt. Karasu-dake, Oita Pref., Japan. A leaves with winged petioles B flowering plant C front view of a flower. Scale bars: 2 cm (A); 10 cm (B); 1 cm (C). Photographs taken by K. Fuse on September 24, 2020.

On Mt. Mukabaki and Mt. Oninome, plants of Hosta sp. 1 were found on high vertical cliffs. When we visited these localities late September, all plants we observed were sterile. These plants were similar to H. alata in the number of leaves, leaf size, the number of lateral veins, and width of petiole wings.