**3. Discussion**

Habitat fragmentation changes continuous habitats into small and isolated patches, and populations inhabiting these habitats are often considered to have low genetic diversity [8,34]. In our previous study [12], we found that island populations of *H. chrysotricha* in the TIL region had significantly lower mean genetic diversity than those from the mainland, based on nuclear SSR markers, suggesting that anthropogenic habitat fragmentation can lead to significant loss of genetic diversity over a few decades. In the present study, we further tested whether genetic erosion increases with an increase in spatio-temporal scales of fragmentation. Based on RAD-Seq-derived SNP markers, we observed that the average estimates of genetic diversity for ZA populations were significantly lower than for TIL populations (*π* = 0.247 vs. 0.208, *p* < 0.01; *H*<sup>O</sup> = 0.307 vs. 0.256, *p* < 0.01; *H*<sup>E</sup> = 0.228 vs. 0.190, *p* < 0.01; Tables 1 and 2), and we found no major differences in genetic diversity between island and mainland populations in either the TIL or the ZA region. Although it remains unknown how much of the initial genetic variation has been lost in each region since the formation of islands, these results may indicate that *H. chrysotricha* populations have retained a considerable proportion of their initial genetic diversity over either ~60 years or thousands of years of habitat fragmentation.

Generally, the genetic effects of habitat fragmentation on plants vary, depending on their life-history traits [2]. In this study, the lack of a significant effect of habitat fragmentation on the genetic diversity of *H. chrysotricha* may be explained by demographic factors and/or strong gene dispersal capacity. As a distylous plant species, *H. chrysotricha* has two contrasting flower morphs that reciprocally differ in the spatial separation of stigmas and anthers, which in turn prevents selfing and intramorph mating. Morph bias is expected to reinforce genetic drift effects and inbreeding by reducing mating opportunities [35,36]. We also found that no biased morph ratio existed in either island or mainland populations, and overall levels of inbreeding in both island regions were negative and close to zero, suggesting a sufficient within-population heterozygosity (Tables 1 and 2). In addition, most of the populations showed no genetic signatures of recent bottlenecks (Table 5). Therefore, it is likely that population sizes of *H. chrysotricha* have remained sufficiently large to prevent loss of genetic diversity via inbreeding and genetic drift after fragmentation [37,38].

The gene dispersal ability via pollen or seed is another key characteristic that determines the potential of a plant species to counteract the negative effects of habitat fragmentation [8]. *Hedyotis chrysotricha* is an insect-pollinated plant; its flower visitors include solitary bees (Halictidae and Andrenidae), Diptera (Syrphidae and Bombyliidae), honey bees, small lepidopterans, and/or thrips [13,39]. Due to the limited foraging distances of these pollinators, they seem unlikely to be able to maintain significant levels of gene flow and population connectivity across fragmented habitats. In contrast, the indehiscent fruit of *H. chrysotricha* contains several very small seeds (1.5–2 × 2–2.5 mm), whose dispersal should be easily facilitated by wind [21]. In our previous study, we detected a moderate level of gene flow among 18 *H. chrysotricha* populations in the TIL region [12]. Based on high-quality SNP markers, our genetic analyses here have further verified the undisturbed population connectivity of *H. chrysotricha* populations in the TIL region, with an average of c. 28% individuals in each population being exchanged with other sites (Table S2). Given the relatively large spatial and temporal scales of habitat fragmentation in the ZA region (Figure 5), one might expect high population differentiation and low levels of gene flow between island populations. However, both BAYESASS and DIVMIGRATE analyses suggested that the average level of gene flow among populations in the ZA region was higher than in the TIL region, although the results from these two methods may not be compared directly. Similarly, populations in the ZA region exhibited marginally lower levels of genetic differentiation than those in TIL region (*F*ST = 0.092 vs. 0.103). A previous study showed that water greatly facilitated the seed dispersal of a typical wind disperser across a fragmented landscape [40]. By referencing the population genetic study of the maritime *Hedyotis strigulosa* var. *parviflora* (Hook. et Arn.) Yamazaki [14], we speculated that water flow may also facilitate the fruit dispersal of *H. chrysotricha*. Although the hydration of *H. chrysotricha* could remain speculative due to the lack of detailed morphological and experimental evidence, the genetic evidence presented here further verified that the species has sufficient fruit (seed) dispersal capabilities to maintain moderate-to-high levels of ongoing gene flow and population connectivity across fragmented landscapes at large spatial and temporal scales.

Corresponding to the above results, little evidence of strong genetic structuring was found in *H. chrysotricha* within the two island systems. Based on SNP markers, the STRUC-TURE analysis grouped the TIL populations basically into three clusters (Figure 1) and the ZA populations into two clusters (Figure 2). In the TIL region, most of the island populations were genetically separated from western mainland populations, which is consistent with the results of our previous nSSR analysis, whereas the genetic composition of eastern mainland TIL populations was slightly different from the former study, consisting of a mixture of all three clusters [12]. In the ZA region, the genetic divergence between the mainland and island populations was not as high as expected, and most of the island populations still feature a high percentage of genetic admixtures, even after thousands of years of isolation (Figure 2). The higher levels of population substructure in the TIL region (three clusters) than in the ZA region (two clusters) may be attributed to their contrasting landscape histories. For example, before water flooded the TIL region, mountains (e.g., the Baiji Mountains in the western part of the TIL region) may have acted as a stronger dispersal barrier than water for gene exchange between populations inhabiting the middle and western former hilltops. In the case of the ZA region, which separated from the nearby mainland 7000–9000 years ago due to a rise in sea level, water has been instrumental in facilitating gene exchange between populations for a much longer time than in the TIL region. When combined with the lack of isolation by distance in each island system, these results further suggested that population connectivity in *H. chrysotricha* has not been greatly modified in either island system, regardless of their significant differences in spatial–temporal characteristics.

**Figure 5.** Sample localities of *Hedyotis chrysotricha* in the Thousand-Island Lake (TIL) region and the Zhoushan Archipelago (ZA) region of southeast China. Population codes are identified in Tables 1 and 2.
