We investigated all of Tsushima based on the report by [20] to confirm six locations of exposed rock where C. incana would be expected to grow. In this article, when groups of individuals were separated by more than 2 km, we defined each as a different “population”. We recorded environmental data, such as the latitude/longitude, altitude, population area, and numbers of individual in each natural population. In addition, for the genetic investigation of each population, we collected seeds from mature individuals in all population. We performed the statistical analysis of environmental data by comparing with shore and inland populations using the Student’s t-test in SPSS.

We planted the seeds from the natural populations and grew them under the same conditions in a greenhouse. We transplanted seeds into seven bowls and managed the plants in a non-temperate/climate controlled environment from June to July after planting in April 2013. The growth medium consisted of red soil: peat moss: perlite = 7:2:1 without basal fertilizer. Fresh young leaves were gathered from each individual and were kept at −80˚C until DNA extraction. Genomic DNA was extracted by the modified cetyltrimethylammonium bromide (CTAB) method [23] [24] . We adjusted extraction DNA to a concentration of 100 ng/µl by a spectrum altimeter and used polymerase chain reaction (PCR). We used a sample of one or two individuals for the chloroplast DNA sequence analysis, and a bulk sample, which mixed DNA of the same concentration that we extracted and adjusted from 10 individuals for the nuclear DNA sequence analysis.

To investigate the chloroplast DNA sequence, We used three regions that were registered in Genbank in C. incana that were suitable for the phylogenetic analysis and were more subtle than variation within the genus; matK [25] [26] , trnL-trnF [27] [28] , and rpl32-trnL [29] [30] . The Genbank accession numbers of matK, trnL- trnF, and rpl32-trnL are AF315295, JF301359, and JQ669280 respectively. Therefore, we chose eight intergenic spacer sequences that were determined to be suitable for the classification of closely related species by [29] : trnQ-rps16, atpI-atpH, ndhF-rpl32, petL-psbE, psbD-trnT, psbJ-petA, rps16-trnK, and trnV-ndhC. On the other hand, we used the ITS [31] -[33] of the Genbank registration sequence for the nuclear DNA sequence. The Genbank accession number of the ITS is EF508064. Each primer was prepared from the original paper or sequence information newly in Genbank registration regions, and universal primers were used in the remaining regions (Table 1). The PCR solution was adjusted according to the instruction of the KAPA extra Taq kit (NIPPON Genetics). All reactions were performed with the following program using a Veriti Thermal Cycler (Life technologies): initial denaturation for 2 min at 95˚C; 35 cycles of 20 s at 95˚C, 15 s of annealing at 50˚C for all the chloroplast regions and 56˚C for the ITS region, and 1 or 2 min at 68˚C; a final extension for 2 min at 68 ˚C; and kept at 4˚C until further processing. The amplification was confirmed by agarose gel electrophoresis. PCR products were purified using ExoSAP-IT (GE), which reacted for 30 min at 37˚C, 15 min at 80˚C. Sequencing reactions were performed using a Bigdye^TM Terminator v3.1 Cycle Sequencing Kit (Life technologies), and gel filtration was performed for each sample. Sequences were analyzed in a 3500 Genetic Analyzer (Life technologies).

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 [34] . Phylogenetic trees were constructed by the neighbor-joining method using the Kimura 2-parameter model in MEGA 6 [35] . The reliability of the topology was assessed with the bootstrap analysis by 1000 replications. Tripora divaricate (Maxim.) P.D. Cantino, which is a related genus of Caryopteris, was included as an outgroup in the phylogenetic analysis. The network figure was drawn using the SplitsTree 4.0 software package [36] .

^aK = G or T, R = A or G, Y = C or T.

From our field work, we confirmed 72 natural populations in 111 locations throughout Tsushima that were suitable for the growth of C. incana (Figure 2). In agreement with the report of [20] , it seems to be widely distributed in each place over the islands. We confirmed that C. incana grew locally on open rocky sites, such as roadsides, mountains, and shorelines. In such places, it grew wild while avoiding competition with plants at some environments, such as gaps in a natural forest, artificial rocky places by infrastructure maintenance, gaps in concrete surfaces, and bare rock places facing the sea. There were few individuals of C. incana on rocky places covered by creeping vines. Natural populations were not confirmed on rock surfaces along the shore exposed to strong winds, and there were many individuals in inlets. Thus, natural populations of C. incana tended to be locally discontinuous. In addition, populations were confirmed near the slope faces for greening along roadside, and it is possible that their presence at such locations was affected by human activities, such as planting.

There were 39 populations near shores at around 20 m from the sea, and the average altitude in these locations was 5.6 ± 1.4 m above the high tide line (Table 2). On the other hand, there were 33 populations on mountains or small inlands, and the average altitude of these locations was 47.7 ± 6.3 m, covering a wider range than shore populations. There were 22 populations on the south island, at an average altitude of 38.2 ± 8.8 m, and 50 populations on the north island, at an average altitude of 19.0 ± 3.6 m. It was thought that these values were due to higher altitudes and a higher ratio of inland populations on the south island than on the north island, but there were few populations in the center of either island. In addition, the number of individuals in populations along

^aover 3000 m². ^bover 500 individuals.

the shore tended to be lower than in inland populations (120.5 ± 39.3 < 172.9 ± 36.5, F (2.591) = 0.112, P = 0.268; no significant difference). Because the places where it could grow were more limited along the shore than inland and were surrounded by trees and the sea, it seemed that the shore populations were less likely to expand their distribution in the future. Particularly, isolated populations with few individuals were presumed to have lower fitness from inbreeding depression [37] [38] or Allee effect [39] .

Sequence variations among populations in six of the 11 regions of amplified chloroplast DNA were confirmed using primers prepared from the Genbank registration sequence or universal primers. The intergenic spacer sequence of trnL-trnF region distinguished H2 from other haplotypes by sequence variation in the number of repetitions (Table 3). Similarly, the sequence of the rpl32-trnL region distinguished H1 from other haplotypes by such variation. The sequence of the psbD-trnT region distinguished H1, H2, and H3 from other haplotypes by sequence variations in the number of repetitions and an insertion-deletion (indel) of 9 bases. The sequence of the trnQ-rps16 region distinguished H2 from other haplotypes by sequence variations in the number of repetitions and an indel of 13 bases, and distinguished H1, H2, and H4 from other haplotypes by three substitution sites. The sequence of the rps16-trnK region distinguished H1 and H2 from other haplotypes by sequence variations