Endangered but genetically stable—Erythrophleum fordii within Feng Shui woodlands in suburbanized villages

Abstract Feng Shui woodlands are naturally or artificially formed green areas in southern China. They are precious for maintaining ecosystem balance in modern semiurban environments. However, they are generally small and geographically isolated from each other, and the status of genetic diversity of the plant species within them has been almost neglected. Therefore, we studied the genetic diversity of the endangered Erythrophleum fordii in eight Feng Shui woodlands (a total of 1,061 individuals) in Guangzhou, a large city in southern China, using microsatellites. For comparison, one population with 33 individuals sampled in a nature reserve was also studied. Although our results indicate that significant demographic declines occurred historically in E. fordii, such declines have not resulted in consistent reductions in genetic variation over generations in Feng Shui populations in the recent past, and the levels of genetic variation in these populations were higher than or comparable to the genetic variation of the population in the nature reserve. In addition, our parentage and paternity analyses indicated widespread and potential long‐distance pollen flow within one Feng Shui woodland, indicating the presence of an unbroken pollination network, which would at least partially alleviate the genetic erosion due to habitat fragmentation and the unequal gene contributions of E. fordii parents to their progenies when favorable recruitment habitats are absent under most of the parent trees. Overall, our results suggest that E. fordii in Feng Shui woodlands may not be driven to extinction in the near future. Nevertheless, uncontrolled fast urban development with a lack of awareness of Feng Shui woodlands will cause the local extinction of E. fordii, which has already happened in some Feng Shui woodlands.

Feng Shui woodlands have existed in China for more than 2000 years (Coggins, 2003;Guan, 2002;Hu et al., 2011;. Literally, "Feng" means wind and "Shui" means water in Chinese. Following these definitions, Feng Shui is thought to create harmony, and promote health and wealth in indigenous communities. In addition to maintaining biological diversity, they are valuable for regulating the climate, cleaning the air, and protecting soil and water, and they play important roles in cultural heritage, leisure activities, and the local economy. Feng Shui woodlands can be categorized into three types according to their location: village, cemetery, and temple Feng Shui woodlands, and among them, village Feng Shui woodlands are the most common (Hu et al., 2011;Ye et al., 2013). With the recent development and expansion of cities in southern China, some of these woodlands have become part of cities, such as Guangzhou, which is an ever-growing megacity in southern China. Currently, there are 156 Feng Shui woodlands in and around Guangzhou city with a total of area 521.07 ha . However, because most of these are not included in the local conservation projects of the city government, relentless construction (such as housing developments and road construction) in the city is a constant threat to the woodlands, shrinking their areas and isolating them more from each other. Thus, urbanization contributes greatly to habitat destruction and fragmentation, altering the plant and animal communities of Feng Shui woodlands.
Prolonged clonal growth, long generation times, and long-distance pollen dispersal help species temporarily escape or postpone extinction (Honnay & Bossuyt, 2005;Low, Cavers, Boshier, Breed, & Hollingsworth, 2015), which results in an undetectable erosion of plant species genetic variation, making it less likely for people to start immediate conservation efforts.
So far, only two plant genetic diversity studies in Feng Shui woodlands in China have been carried out (Ge, Liu, Shen, & Lin, 2015;Wang, Ye, Fu, Ren, & Peng, 2008), and their results generally confirmed the results of Honnay and Bossuyt (2005) and Low et al. (2015) and clearly demonstrate the harmful effects of urbanization on species' genetic health. In the work of Wang et al. (2008), they compared the genetic diversity of a common species Cryptocaya chinensis in two Feng Shui woodlands and four natural reserves in the lower subtropical region of southern China. Their results revealed unexpected extensive clonal growth of C. chinensis in two Feng Shui woodlands due to severe fragmentation and the small population sizes in suburban areas. The clonal growth in Feng Shui woodlands maintained a substantial proportion of the genetic variation of the initial populations, and the small sizes of the woodlands did not result in significant genetic differentiation from the larger reserve populations. However, as McDonald, Rice, and Desai (2016) found out that asexual population maintained genetic variation at the cost of fixing substantially deleterious mutations while sexual population allowed natural selection to more efficiently sort beneficial from deleterious mutations and speeded adaptation, extensive clonal growth may threaten the long-term adaptation of C. chinensis in Feng Shui woodlands. In the work of Ge et al. (2015), they compared the genetic diversity of Phoebe bournei in three Feng Shui woodlands and three natural reserves in southern China. Their results showed that the genetic diversity of P. bournei was clearly lower in Feng Shui woodlands than in the reserves, which could be related to its low regeneration rate in Feng Shui woodlands.
Given that some of plant species in Feng Shui woodlands are even endangered, studies on their conservation are thus needed to improve our ability to make relevant recommendations on ways to alleviate the negative impacts of urban development on native species. Therefore, our objectives were to study the genetic diversity and within population gene flow of the endangered species Erythrophleum fordii in Feng Shui woodlands in Guangzhou, China.
Due to severe human disturbance and the small population size, we would expect the genetic diversity of E. fordii in Feng Shui populations to have decreased. Furthermore, because gene flow is crucial to maintain genetic diversity in plant populations, by examining the present gene flow pattern via parentage analysis, we aimed to investigate the relationship between genetic diversity and gene flow patterns. Currently, there are contrasting results regarding such a relationship in the literature. Theoretically, human disturbances change pollen mutualisms in Feng Shui woodlands and, subsequently, alter a population's genetic variation. However, empirical evidence does not always support this in other disturbed habitats (Giombini, Bravo, Sica, & Tosto, 2017;Noreen, Niissalo, Lum, & Webb, 2016;Rosas, Quesada, Lobo, & Sork, 2011). Therefore, the genetic consequences for a small population or due habitat disturbance are both species and location specific (Owusu, Schlarbaum, Carlson, & Cailing, 2016;Schwarcz et al., 2018). In particular, we also examined the demographic history of Feng Shui populations of E. fordii to study how the current level genetic diversity is related to historical events such as bottlenecks. As a valuable timber tree, the endangered status of E. fordii is believed mainly to be the result of large-scale logging in the past. Consequently, it will leave a clear signal of a population decline in genetic diversity.  and Zhu, Wang, Ye, Cao, and Saravanan (2013) have described E. fordii in detail. Briefly, E. fordii is a legume species, belonging to the family Fabaceae. It occurs naturally in China and Vietnam. Due to the hardness of its wood, it is commonly known as the "ironwood" tree in China. In the past, overexploitation has made it endangered in the wild (IUCN 1 ). At present, it is under second-class national protection in China. It is a typical outcrossing species with pollen dispersed by many kinds of insects, such as beetles, butterflies, bees, and wasps . However, self-fertilization could happen in E. fordii , and it was also reported in the congeneric species E. suaveolens . Its seed is flat and ovate-shaped, large and heavy with 1.54-1.67 cm long, 1.28-1.43 cm wide, 0.784-0.917 g in weight (Zhao et al., 2009). It is inedible and has no wing and no particular attributes to attract animals; therefore, it is believed to be dispersed by gravity .

| Sample collection
According to Ye et al. (2013), E. fordii can be found in more than 30 Feng Shui woodlands in Guangzhou city. By comparing their geographic locations and population sizes, we sampled eight of them (Table 1, Figure 1, Figure S1) during 2017 and 2018. During sampling, we carefully examined within and around each population to make sure we would not miss E. fordii populations or individuals nearby.  Figure S3).
We sampled all the individuals that we could find in the populations of TB, WYG, XD, YCG, ZL, and LT villages ( Figure 2, Figure   S1). Unfortunately, due to human interference, the populations of YCG and XD villages, which were previously recorded as having large population sizes, had shrunk to one and two individuals, respectively. Since both ZPT and SKY villages contained hundreds of newly sprouted seedlings, which was more than we could genotype with our budget, we randomly sampled only some of the seedlings but all of the individuals with DBH (diameters at breast height) ≥1 cm or height ≥1 m.
We collected one to three leaf samples per E. fordii individual and put them into sealed plastic bags containing silica gels. We determined the locations of the sampled individuals using GPS, and measured and recorded their DBH if the DBH ≥1 cm, or height if the DBH <1 cm.
In April 2018, we revisited the sampling site in TB village and recorded the flowering status of the individuals. We observed that F I G U R E 1 Map showing the locations of the sampled populations (the background layer was downloaded from Google Maps in this and all the other figures). The full names of the populations are given in Table 1 F I G U R E 2 Spatial distribution of Erythrophleum fordii individuals in TB village in Guangzhou city, China. Red: adults; cyan: juveniles; and yellow: seedlings. Only adults are given proportional circles corresponding to their DBH values. An additional 44 seedling samples were collected from around the adult numbered 303. Since these seedlings grew on the roof of an abandoned hut beside this adult tree, their positions were not recorded by GPS (indicated in Figure S2) and therefore, these 44 seedlings are not shown on the map. According to the plate tagged by the local Forest Management Department in 2013, the individual numbered 1 is more than 200 years old ( Figure S3). The photographs of the two individuals numbered 35 and 276 are shown in Figure  S4 only individuals with DBH >20 cm could flower. By combining flowering status and DBH and/or height information, we classified all of our sampled individuals into three cohorts: adult (DBH ≥20 cm), juvenile (1 cm ≤ DBH < 20 cm), and seedling (DBH <1 cm).
DH Mountain was the first nature reserve in China , and the E. fordii population in it is believed to be well preserved. According to , there were a total of 528 E. fordii individuals including 78 with DBH ≥15 cm. Therefore, we used it as the basis to compare the highly disturbed populations in the Feng Shui woodlands. However, similar to the populations in Feng Shui woodlands, it is in a small area of approximately 2 ha, although there are no obvious environmental limitations that might have prevented it from growing in the surrounding areas (personal observation, WZF). In this study, we resampled 33 individuals with DBH >30 cm and genotyped them using not only the previously isolated loci, but also our newly developed ones (Table S2).

| Microsatellite isolation and genotyping
Since only nine microsatellites in E. fordii were previously isolated and characterized (Zhu et al., 2009), to increase the discrimination power in parentage analysis, we used the restriction site-associated DNA sequencing (RAD-seq) method to obtain some new microsatellite markers in E. fordii.

By using two E. fordii individuals from the South China Botanical
Garden, we constructed two RAD-seq libraries according to the methods described by Baird et al. (2008). Briefly, whole genome DNA of E. fordii was digested using the restriction enzyme EcoRI (Takara). The digested DNA fragments were then ligated to adaptors and PCR amplified. Approximately 300-500 bp fragments were subsequently selected and sequenced on Illumina HiSeq X Ten genetic analyzer (Illumina) to produce 150 bp paired end sequencing reads.
After sequencing, we obtained a total of 49,971,692 bp of raw reads for one individual and 39,494,352 bp for the other. The raw sequence data are available in the NCBI SRA database with accession numbers SRX5010692 and SRX5010693. Filtering for PCR duplicates and low-quality reads resulted in 16,662,656 and 10,993,602 bp of useful reads for the two individuals, respectively. These reads were assembled using Rainbow 2.0.4 (Chong, Ruan, & Wu, 2012), and the assembled contigs were combined and re-assembled using CAP3 (Huang & Madan, 1999). We used Msatcommander 0.8.2 (Faircloth, 2008) to screen for microsatellites in the re-assembled contigs. In particular, we only chose sequences with at least eight and seven dinucleotide and trinucleotide motifs repeats for the two individuals, respectively. We then randomly chose 35 microsatellite sequences to perform PCRs to test their availability.
We followed the PCR procedures described by Zhu et al. (2009) but with an annealing temperature of 53°C for all the microsatellite loci. We ran the PCR products on 2% agarose gels which revealed that 29 microsatellites could be successfully amplified to produce fragments of the correct size. We then used six individuals from TB village to perform PCRs to study the polymorphism of the 29 microsatellites. After PCR amplification and running the PCR products on an ABI 3730 sequencer, we identified 16 polymorphic microsatellites with clear electrophoretic profiles of alleles in the six individuals.
Using these six individuals, we also tested the seven microsatellites previously isolated by Zhu et al. (2009) and shown to be in Hardy-Weinberg equilibrium (HWE). However, only five of these could be successfully amplified in all of our present samples. Thus, we used these 5 together with the 16 newly identified microsatellites (Table   S2) to genotype 346 individuals from TB village. We tested the HWE of the microsatellites and found that the locus EF-32 not only showed a significant deficit in heterozygosity at the population level, but also in the seedling life stage (Table S3). Therefore, we did not use this locus in our study to avoid null allele errors and instead used a total of 20 loci for all population genotyping and data analyses.

| Data analysis
Since only one and two E. fordii individuals were found in YCG and XD villages, respectively, these three individuals were only used to estimate the overall genetic diversity in E. fordii but excluded from the other data analyses.
We first estimated null allele frequencies at 20 loci in each population with INEST v2.2 under the individual inbreeding model (IIM) with default parameters (Chybicki & Burczyk, 2009). IIM was implemented by a Bayesian approach which showed better statistical properties than maximum likelihood and the other approaches (Chybicki & Burczyk, 2009). We then calculated genetic diversity parameters, observed and unbiased expected heterozygosity (H O , H E ) using GenAlEx 6.501 (Peakall & Smouse, 2012), and the inbreed- ing coefficient (f) using GENEPOP 4.3 (Rousset, 2008 wilcox.test function in R software and one-sided p-values ("less" or "greater") were reported.
We also used GENEPOP 4.3 to assess the deviation from HWE and genotypic linkage disequilibrium (LD) among all pairs of loci. The levels of significance for HWE and LD were adjusted by using the sequential Bonferroni correction (Holm, 1979). At the population level, since the association analysis between loci via the LD tests may be strongly influenced by any family structure (Flint-Garcia, Thornsberry, & Buckler, 2003) present in our data, we only used adults or adults and juveniles (in WYG and LT villages) in the LD tests for our studied loci. At locus Gm2024, some individuals produced abnormal alleles whose sizes did not follow the rule for the gain or loss of repeated unit; therefore, we treated these alleles as missing values for those individuals in subsequent bottleneck and demographic history inference analyses, but not for other analyses.
Since only TB village was sampled thoroughly, especially for seedlings, we only performed a parentage analysis in this population using the Cervus 3.07 program (Kalinowski, Taper, & Marshall, 2007). Before performance parentage analysis, the power of exclusion for the microsatellite loci was estimated by Cervus. The cryptic gene flow was then assessed by 1−(1−P parent-pair ) Na following Dow and Ashley (1996), where P parent-pair was combined nonexclusion probability of parent pair, and Na was the number of adults used for parentage analysis (18 in this study). After determining the power of exclusion for the microsatellite loci, using the allele frequency data calculated from all the samples in this population, we ran a simulation to estimate the critical Delta scores necessary for parentage assignments at a 95% confidence level. The simulation parameter values were as follows: 100,000 tests, 18 for the candidate parents (the adults in our samples of TB village), 0.9 for the proportion of candidate parents sampled, 1 for the proportion of loci genotyped, 18 for the minimum genotyped loci, self-fertilization was allowed, and the default setting were used for the other parameters. Although we could identify the parents at a 95% confidence level for most of the seedlings, there were some seedlings that we could not identify the parents of. Considering the restricted seed dispersal ability of E. fordii for the rest of the seedlings, we assumed the mother tree to be the nearest tree to the seedlings geographically, and then used the same program to perform paternity analysis. In this analysis, we used the same simulation parameters as in the above parentage analysis.
After parentage and paternity analyses, the actual pollen immigration rate was estimated as number of seedlings with undetermined paternity/total number of seedlings. The effective pollination neighboring area (A ep ) was calculated by A ep = 2πσ 2 , where σ 2 was the variance of the pollen dispersal distance (Levin, 1988).
We also estimated the pollen immigration using the spatially explicit neighborhood model (Burczyk, Adams, Birkes, & Chybicki, 2006) in the NM+ 1.1 (Chybicki & Burczyk, 2010) which simultaneously estimated seed immigration. Using maximum likelihood, we estimated the self-fertilization rate (s), the pollen immigration rate (mp), pollen dispersal distance (dp), the seed immigration rate (ms), and seed dispersal distance (ds). If initial parameter values were far from the true values, the maximum likelihood algorithm could fail to reach convergence. After trying different initial parameters, the final parameter settings used for estimation were as follows: exponentialpower dispersal kernel for both seed and pollen, genotyping error rates 0.01 for all loci, seed immigration rate 0.01, average seed dispersal distance 4.57, shape parameter of seed dispersal kernel 0.42, pollen immigration rate 0.134, average pollen dispersal distance 160, shape parameter of pollen dispersal kernel 1.6, selfing rate 0.12, and default for the other parameters.
We finally examined genetic structure among populations by STRCTURE 2.3.4 (Pritchard, Stephens, & Donnelly, 2000). Assuming admixture model with correlated allele frequency, twenty independent runs were performed for each possible cluster (K, from 1 to 8) using a 5 × 10 5 Markov Chain Monte Carlo (MCMC) iterations after a burn-in period of 5 × 10 5 on total multiloci genotypes of adults without prior concerning of their origin populations. The choice of the probable K value was made both as recommended in STRUCTURE user's manual and by ΔK method (Evanno, Regnaut, & Goudet, 2005).
For the inferred K, we used CLUMPP 1.1.2 (Jakobsson & Rosenberg, 2007) to calculate the average membership coefficient for each individual by combing the results of 20 runs. All these analyses were performed by a combination of functions from the StrataG 2.1 (Archer, Adams, & Schneiders, 2017) in R package.

| RE SULTS
The number of alleles per locus varied from 2 to 11 (Table S3)  in older generations (adult and juvenile) than in seedlings, but the populations of WYG, ZL, and SKY villages had the highest in seedlings, the youngest generation (Table 1). However, for each Feng Shui population, the differences in genetic variation among generations were not significant (Table S6). All f values in the life stages of the populations were smaller than zero, but not all of them significant deviation from zero ( Table 1) The mean Log-likelihood values in STRUCTURE analysis indicated K = 6 was the "optimal" genetic clusters ( Figure S5) because at K = 6 the Log-likelihood values began to reach "more-or-less plateaus" according to the STRUCTURE manual. The ΔK showed two obvious peaks with the highest at K = 2 and the second highest at K = 6. However, because ΔK value was smaller at K = 6 than at K = 3, we then illustrated these three K results (

| Genetic diversity
Among approximately 15 species in the Erythrophleum genus, which is mainly found in Africa, only E. fordii is found in China. Since it is an endangered species, it is valuable to compare its level of genetic diversity to that of its congeners (Cole, 2003;Gitzendanner & Soltis, 2000 We previously investigated the genetic diversity of E. fordii in one well-preserved population using nine microsatellites in the DH F I G U R E 5 Membership probability results of the STRUCTURE analyses based on adults for seven Erythrophleum fordii populations. STRUCTURE shows genetic group K = 2, 3, and 6. In the plot each vertical bar represents one individual and the colors in the bar show the assignment probability to the genetic clusters Mountain, a nature reserve. Although our previous study reported higher genetic diversity in the DH Mountain (a H O of 0.606 and H E of 0.586,  than all the Feng Shui populations we investigated here, our present study using a new set of microsatellites indicated this was not the case (Table 1). In fact, the genetic diversity of the population of DH Mountain was only significantly higher than that of the population of WYG village in A P value (Table S5). This implies, on one hand, that E. fordii in Feng Shui woodlands are not deprived of genetic diversity; on the other hand, that at population level, the genetic diversities within populations measured using different sets of markers should be compared with caution.
We observed that among all the populations, SKY harbored the highest genetic diversity but not the highest number of private alleles (Table 1) (Table 1), which indicates random mating.
Such random mating also existed in the well-preserved population of DH Mountain whose f value was 0.0082 (Table 1). This is con-

| Pollination and pollen flow
Our parentage analysis for the TB village population (Figure 3) further supports the conclusion of random mating in E. fordii. Although this figure shows that mating events seem to be biased to a few individuals, our field investigation indicated this may not be true in nature. TB village is an environment with lots of human interference, and for most of the mother trees, their existing environment is clearly unsuitable for seedling survival ( Figure S4). Below these trees, few or no recruits were found (Figure 2). Since our parentage analysis was based on seedlings collected in the field, these few mother trees would display higher mating events than most of the others, resulting in skewed mating patterns in the population (Figure 3). Given this, we believe that the pollination network required random mating in E. fordii populations, at least in the TB village population, was not destroyed by urbanization.
According to the review by Senapathi, Goddard, Kunin, and Baldock (2017), pollinator abundance and composition in urbanized areas are not always inferior to less disturbed areas. The flowers of E. fordii can attract a diverse range of pollinators . In this study, we found pollen immigration rate (mp) of 0.144 and 0.122, and mean pollen dispersal distance (dp) of 160.085 and 220.910 m based on parentage assignments and neighborhood model, respectively, in the population of TB village. Effective pollination neighborhood (A ep ) for E. fordii was 11.24 ha between seed trees, equivalent to a circle with a radius of 189.21 m around a seed tree. Hence, considering the very limited seed immigration rate of 0.010 and dispersal distance of 4.666 m we observed, long-distance gene flow in E. fordii is primarily by pollen dispersal. As both mp and dp are directly related to the sizes and degrees of isolation of areas, they should vary among species and populations within species with similar pollination insects (Braga & Collevatti, 2011;Manoel et al., 2012;Monthe, Hardy, Doucet, Loo, & Duminil, 2016;Noreen et al., 2016;Sebbenn et al., 2011;Tambarussi, Boshier, Vencovsky, Freitas, & Sebbenn, 2015). For example, for a Neotropical tree Copaifera langsdorffii in a highly isolated and fragmented forest fragment (4.8 ha), Manoel et al. (2012) found its mp and dp values were 0.08 and 66 m, respectively; while for a tropical tree Entandrophragma cylindricum in a relatively large and continuous forest, Monthe et al. (2016) found its mp and dp values were 0.32-0.40 and 506-540 m, respectively. Therefore, we consider the pollen flow in the population of TB village moderate. Because pollination insects, such as bees, could carry pollen to very long distances (Braga & Collevatti, 2011;Dick, Etchelecu, & Austerlitz, 2003;Manoel et al., 2012;Noreen et al., 2016;Sebbenn et al., 2011;Tambarussi et al., 2015), it is possible to find larger mp and dp in larger and more continuous populations of E. fordii than those in our study, based on pollen-mediated gene dispersal capacities of E. fordii. Because our study here is mainly focused on local pollination dynamics, we provide no indication of the potential pollen sources for immigrated pollens. Continued research including more surrounding populations on characterizing within and among population pollen flow patterns is a priority in this system in the future.

| Demographic history
Our data indicate that all of our populations clearly suffered from bottlenecks. It is possible that such bottlenecks might have been caused by recent (approximately 100 years ago) wood demands due to long periods of war and poverty. However, the results from skyline plots (Figure 4)

| Conservation implication
Our results show that despite moderate gene flow via pollen, the limited seed dispersal distance may result in significant relatedness among E. fordii individuals at short distances. Furthermore, they reveal clearly disproportionate contributions of adults to the recruit landscape due to absence of suitable recruitment environment under some adults, which could remain across reproductive cycles if present unfavorable conditions for seedling establishment continue. Together with substantial self-fertilization which is often associated with isolated and fragmented small populations (Cheptou, Hargreaves, Bonte, & Jacquemyn, 2017), all these factors may increase the rates of mating among relatives, producing negative fitness effects in future generations, and slowing down adaptation in the face of climate change.
Seed collection for ex-situ conservation of E. fordii should include Feng Shui woodlands as high genetic diversity harbored in most them and contain many trees as many trees as possible. The mean pollen dispersal distance suggests that seed trees must be separated by at least 220 m. In addition, since the gene pool of the population of SKY village is different from the others, seeds collected from it should be separated and only mixed with those from other populations after being sure no outbreeding depression effects happening.

| CON CLUS ION
Feng Shui woodlands, as part of CPFs, are valuable supplements for urban forests in southern China, especially with the present policy aiming to build national forest cities, because they are more natural and have higher species diversities than modern artificial green lands . Furthermore, they can provide a source for rural afforestation and play a key role in ecological networks. Globally, CPFs are distinctive elements of worldwide vegetations, and they naturally and seminaturally distribute in and around urban areas (Avtzis et al., 2018;Bossart & Antwi, 2016;Hu et al., 2011;Lee et al., 2019). Similar to Feng Shui woodlands in China, most of the plant species in CPFs are not considered being threatened in the near future. However, monitoring their gene pools is essential to prevent genetic erosion caused by anthropogenic effects. Therefore, an extension of this study to other species with different life stages and different landscape configures in CPFs would be recommended to know their gene flow and how such flow determines the microevolutionary changes in them.
Overall, our results suggest that E. fordii may have suffered serious demographic declines before large-scale human settlements in southern China and has not recovered at the present time due to consistently high demands for its high quality wood. However, parentage analysis indicated that its pollen-mediated gene flows were not severely affected within the disturbed suburban areas, and genetic diversity was stably maintained across different generations. A previous simulation study of E. fordii also indicated that its longevity with iteroparity provided the potential to maintain genetic diversity in small isolated populations , and our present study supports this conclusion. However, for most of the Feng Shui woodlands, the major threat to the long-term adaptation and evolution of E. fordii is the lack of suitable regeneration habitats. The maintenance of large and continuous populations to guarantee high gene flow is also required for long-term species sustainability.

ACK N OWLED G M ENTS
We thank the anonymous reviewers and editor for their extremely constructive comments on the manuscript. We thank Yasi Liu for her helping in field sampling. This work was supported by the Reserves (1210-1741YDZB0401-1).

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.

AUTH O R CO NTR I B UTI O N S
WZF conceived and designed the project and carried out the laboratory procedures and data analyses. WZF and LHL carried out the field collections. All authors contributed to writing the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The RAD sequences data are available in the NCBI SRA database with accession numbers SRX5010692 and SRX5010693. The microsatellites are available in the NCBI database with accession numbers shown in