We have confirmed 72 natural populations by conducting fieldwork in the Tsushima Islands, Nagasaki [18] . In this study, we investigated the Nagasaki Mainland, Goto Islands, some remote islands, and the western parts of Saga adjacent to the Nagasaki and Koshikijima Islands, based on the report by Itow and Kawasato [16] (Figure 1). When groups of individuals were separated by more than 2 km, we defined each as a different “population”. We confirmed the current status of natural populations reported by Itow and Kawasato [16] and investigated locations of bare rock where C. incana would be expected to grow. We recorded environmental data, such as the latitude/longitude, altitude, and approximate number of individuals in each natural population. Additionally, for the genetic investigation of each population, we collected seeds from mature individuals in all populations.

The management of plant samples from seeds collected from each natural population for the investigations, and the extracting method from the leaves of individuals cultivated followed the previous report [18] .

To investigate the chloroplast DNA sequences, we used three regions that were registered in Genbank under C. incana that were suitable for phylogenetic analysis, and were more subtle than variation within the genus; matK, trnL-trnF, and rpl32-trnL. Therefore, we investigated four intergenic spacer sequences that were used in the previous report and were determined to be suitable for the classification of closely related species by Shaw et al. [20] . Each primer was newly prepared from the original paper or sequence information in Genbank registration regions, and universal primers were used for the remaining regions. The PCR condition, the sequences of primers used, and the equipment used were similar to the former report [18] .

To compare all populations of West Kyushu, we added the base sequences of 72 populations from the Tsushima Islands that were investigated in the former report to the phylogenetic analyses. The base sequences of each population were aligned using the BioEdit software (version 7.2.5) and compared with the registration sequences in the Genbank registration region [21] . Phylogenetic trees were constructed by the neighbor-joining method using the Kimura 2-parameter model in MEGA 6 [22] . The reliability of the topology was assessed with a bootstrap analysis using 10,000 replications. Tripora divaricate (Maxim.) P.D. Cantino, which is a related genus of Caryopteris, was included as outgroup in the phylogenetic analysis. The network figure was drawn using the TCS 1.21 software package, which is able to analyze the insertion-deletion (indel) along with the substitution [23] .

During our fieldwork, we investigated 257 locations throughout West Kyushu, excluding the Tsushima Islands, that were suitable for the growth of C. incana (Nagasaki Mainland = 146; Goto Islands = 80; Koshikijima = 31). As a result, we confirmed 37 natural populations (Table 1): 16 populations in the Nagasaki Mainland, 16 populations in the Goto Islands, and 5 populations in the Koshikijima Islands were confirmed (Figures 2(a)-(c)). The average altitude of these locations was 24.9 ± 3.8 (mean ± S.D.) m for the Tsushima Islands [18] , 175.3 ± 26.9 m in Nagasaki Mainland, 35.75 ± 7.77 m for the Goto Islands, and 123.8 ± 22.0 m for the Koshikijima Islands. On Iki Island, Nagasaki, we were not able to confirm the natural population according to Itow and Kawasato [16] . Among the survey areas, 6 populations were distributed along the shore of the Goto Islands, and several populations were confirmed in the environment near villages, including the open rocky sites along the roadside, like the Tsushima Islands. The populations in the southern part of the Goto Islands tended to include fewer individuals within a population; the populations in the northern part tended to include more individuals within a population. However, the populations in the Nagasaki Mainland, except one population along the shore, were located in the inland and particularly limited to high altitudes. In the Nagasaki Mainland, the populations near Nagasaki City tended to include fewer individuals than populations on Hirado Island of northwestern Nagasaki. In eastern Nagasaki, and on the eastern side of the neighborhood of Saga, no natural populations were confirmed in the environment suitable for the growth of C. incana. On the Koshikijima Islands, there were fewer numbers of populations than other areas; only one population was confirmed on the south island of the Koshikijima Islands. There were many individuals in one population (C107) on the south island of the Koshikijima Islands, but the populations on the north and center island tended to include fewer individuals. In addition, one population (C106), which included many individuals on the north island of the Koshikijima Islands, was located near the green slope faces along the roadside, and it is possible that those were affected by human activities. The populations in West Kyushu tended to have fewer individuals and population densities than the populations on the Tsushima Islands, thus isolated populations which had fewer opportunities to exchange genes with other populations were presumed to have lower fitness due to inbreeding or Allee effect [24] -[26] .

According to the report by Itow and Kawasato [16] , there were 26 natural populations of C. incana in West Kyushu, excluding the Tsushima Islands. The populations that we were not able to confirm in this study were at 12 locations reported by Itow and Kawasato [16] . Ten of these were located in the Nagasaki Mainland, suggesting a decreased number of natural populations in this area. Among the survey areas, the distribution of C. incana in Hisakajima in the Goto Islands, Azuchioshima in the northern Nagasaki Mainland, and Naka-Koshikijima of the Koshikijima Islands had not been previously reported and were confirmed as new natural populations.

Sequence variations among populations in six of the chloroplast DNA samlpes were confirmed using primers prepared from the Genbank registration sequences, or universal primers. The intergenic spacer sequence of the trnL-trnF region distinguished three groups by a sequence variation in the number of repetitions (Table 2). Similarly, the sequence of the rpl32-trnL region distinguished five groups by such variation. The sequence of the psbD-trnT region distinguished five groups by sequence variations in the number of repetitions, and three groups by two sites of indels of 9 and 7 bases. The sequence of the ndhF-rpl32 region distinguished two groups

by sequence variations in the number of repetitions, and five groups by four substitution sites. The sequence of the trnQ-rps16 region distinguished five groups by sequence variations in the number of repetitions, three groups by two sites of indels of 7 and 13 bases, and six groups by seven substitution sites. The sequence of the rps16-trnK region distinguished four groups by sequence variations in the number of repetitions, four groups by three sites of indels of 10, 12 and 15 bases, and distinguished eleven groups by nine substitution sites. The sequences variations in total comprising seven sites of repetitions, seven sites of indels, and twenty substitution sites were confirmed between 109 populations of C. incana.

The geographical distribution for each haplotype of chloroplast DNA sequences in the study areas are shown in Figure 3. In each area, five haplotypes were located on the Tsushima Islands [18] (Figure 3(a)). Two haplotypes were mainly separated between the southern part and the northern part of the Tsushima Islands, and several haplotypes were located in the center part and the north island. Six haplotypes were located in the Nagasaki Mainland (Figure 3(b)). Nine haplotypes were located on the Goto Islands (Figure 3(c)). Three haplotypes were located on the Koshikijima Islands (Figure 3(d)).

The family tree consisted of 22 haplotypes from 109 natural populations in West Kyushu, and is shown in Figure 4. These haplotypes were mainly classified into two groups. The lower group on the family tree was designated from H1 to H10. These haplotypes were distributed on the Tsushima Islands and the Nagasaki Mainland. The upper group on the family tree was designated from H11 to H22. These haplotypes were distributed on the Goto Islands, Koshikijima Islands, and the Nagasaki Mainland. The haplotype network that included the indels of base sequences, and the number of repetitions along with the substitutions, is showed in Figure 5. The haplotype network almost reflected the geographical distribution of haplotypes.

Like the sequence variation of the populations on the Tsushima Islands [18] , several substitution sites between Genbank registration sequences and those of the study populations were confirmed in the matK, trnL-trnF and rpl32-trnL regions(Table 2). However, there were no indications of substitution among the study populations in these sites. In Carex conica, haplotypes of chloroplast DNA that included trnL-trnF were common in the Kyushu Mainland, Tsushima Islands, and Jeju Island, Korea, which suggested the result of rapid postglacial expansion from a few refugia in Kyushu [27] . Therefore, the substitutions among the study populations were not confirmed like in the previous study [18] , suggesting that the distribution of C. incana in West Kyushu might have formed relatively recently.

The number of haplotypes distributed in Goto Islands was the largest of all, suggesting that the genetic diversity within this area was rich. Additionally, different haplotypes were confirmed among the nearby populations within an island (Figure 3(c)), suggesting that the fragmentation of each population within this area, and each varied haplotype, was maintained in each population. One population in the center part of Nagasaki Mainland was confirmed as the haplotype that was distributed on the Goto Islands (Figure 3(b), Figure 3(c)). This population was only located in rocky places facing the sea within this area. Moreover, one of haplotypes that was distributed in the northwestern area matched the haplotype distributed on the Koshikijima Islands (Figure 3(b), Figure 3(c)). The population survey showed that this haplotype was located on bare rocky sites, such as the top of a mountain or a natural rock face, and it did not seem to be affected by human activities, such as planting. Thus, we speculated that the populations in the Nagasaki Mainland maintained their origins.

Overall, low bootstrap values, indicating the reliability of the family tree (Figure 4), were shown, suggesting a low level of relationship between the sequences. H7 to H10, from the northwestern and southern part of the Nagasaki Mainland, were suggested to be closely related with H3 in the northern part of the Tsushima Islands. However, it seemed that H1 and H2 from the Tsushima Islands had separated earlier which resulted in each haplotype. From this, H13 to H15 on the Koshikijima Islands, and in the northwestern part of the Nagasaki Mainland, were designated as a different sub-group on the Goto Islands, suggesting that these haplotypes had

separated earlier than the diverging haplotypes of the Goto Islands.

The Goto Island group included the richest haplotype network (Figure 5), suggesting that the current distribution had been formed by maintaining the isolated populations, which were separated early within this area. As a whole, it seemed that the haplotypes were mainly separated near the center part of the network by several indels of base sequences and substitutions. Because these classifications indicated a similar result within the family tree, we speculated that the common ancestor of C. incana in Japan diverged into two groups near center part of haplotype network. Several phylogeographical studies in these areas had suggested a southward transfer during the LGM, followed by a northward transfer during the period of de-glaciation [28] [29] . From this, it was suggested that C. incana transferred southward from the Korean Peninsula to the Goto Islands and the Nagasaki Mainland, with the decrease in temperature during the glacial period, followed by an expansion northward or southward to each area. Furthermore, a nested structure did not form among haplotypes between each area, suggesting that the two groups diverged into each area; after that, the differentiation between haplotypes within each area developed. Thus, the genetic composition in these areas would have been formed by the effects of vicariance and migration through historical processes that were linked to the fluctuations of sea levels during the late Pleistocene. The haplotype of the northern part of the Nagasaki Mainland that was distributed on the Koshikijima Islands was different from the other haplotypes distributed in the Nagasaki Mainland; thus, it seemed that the current distribution of the populations in this area was established by a different process. The natural population of C. incana in Kagoshima, within the Kyushu mainland, was reported by Hatsushima [30] , after which those populations vanished [31] . One population indicated a common haplotype with the Goto Islands to that of the center of Nagasaki Mainland, was located in rocky places facing the sea. It seemed this population might be derived from seeds that were washed ashore.

To further define the processes of population distribution in these areas, we would have to analyze other genetic regions that allow the comparison of intraspecific variations and nuclear DNA. East China and the Korean Peninsula were connected by an ECS basin during the last Pleistocene, suggesting the possibility of rapid expansion in species distribution during this period [7] . Accordingly, the C. incana population of Mainland China may be a common haplotype distributed in the Korean Peninsula [32] . Additionally, the degree of separation in the populations between in Japan and in the Korean Peninsula would depend on whether the haplotype in Korean Peninsula is similar to Japanese haplotype or not. Future phylogenetic analysis would be necessary to estimate the periods of distribution in Japan by investigating wild individuals from the Korean Peninsula and Mainland China.

In conclusion, we confirmed the present natural environment of C. incana, which is an endangered species in Japan, and constructed its distribution map in West Kyushu. We confirmed the trend that the number of natural populations in the Nagasaki Mainland has been recently decreasing. Additionally, we investigated the genetic structure among 109 populations in West Kyushu, based on chloroplast DNA sequences. Twenty-two haplotypes were mainly classified into two groups by phylogenetic analysis. We speculated that the common ancestor of C. incana in Japan had diverged into two groups, after that the distribution expanded and has been maintained in each area. Furthermore, a high genetic diversity within the Goto Islands was indicated. We confirmed the populations of the indicated haplotypes that are distributed on the Koshikijima Islands and the Goto Islands in the Nagasaki Mainland, suggesting that a complicated distribution formation had occurred in this area. Additional comparisons of the genetic structure among natural populations in West Kyushu, and phylogenetic analysis by studying wild individuals from the Korean Peninsula and Mainland China, would offer useful information on the evolutionary history of C. incana.