Population genomics reveals a representation gap in the conservation of Ardisia gigantifolia (Primulaceae), an endangered species endemic to China

Abstract

Background

Ardisia gigantifolia Stapf is a critically endangered medicinal plant endemic to China that is characterized by a narrow distribution range and small population sizes. Ex situ conservation is an important complementary strategy for in situ, protecting wild plant species from loss of genetic diversity and possible extinction in the wild. However, the extent to which ex situ collections genetically represent their wild populations remains poorly understood.

Results

In this study, we analyzed the genetic diversity and structure of 67 individuals using double digest restriction-site-associated DNA sequencing (ddRAD). The sampled populations included three wild populations from Hekou Town (HK), Gulinqing Town (GLQ), and Nanxi Town (NX), Yunnan Province, as well as one ex situ conservation population originating from Mahuangbao Farm (MHB), Yunnan Province and cultivated at the South China Botanical Garden (SCBG). The results revealed comparatively low genetic diversity across wild populations, with observed heterozygosity of 0.1441–0.1767 and expected heterozygosity of 0.1575–0.2255. Lower genetic diversity was detected in the HK population with small population size and the GLQ population located in severe shading compared to that in the NX population. Notably, the SCBG ex situ population retained a level of genetic diversity comparable to the best-preserved wild population (NX). The genetic structure analysis demonstrated significant differentiation among wild populations. The SCBG ex situ population formed a distinct genetic cluster, showing moderate-to-high differentiation from all extant wild populations.

Conclusion

These findings revealed a critical “representation gap”: while the collection preserves unique genetic variation from an extinct source population, it fails to capture the extant genetic diversity partitioned among the surviving wild populations. Collectively, our findings provide a genomic baseline and strategic insights for the future ex situ conservation and management of A. gigantifolia.

Background

Due to global climate change and frequent human interference, plant diversity is increasingly threatened by the fragmentation of previously continuous habitats and populations [1, 2]. The methods for conservation of plant diversity mainly include conservation in their natural habitats (in situ) and conservation outside their natural range (ex situ). However, not all plant species can be protected in their natural habitats due to loss of habitats and climate change, leading to a growing demand for ex situ conservation [3]. Ex situ collections are key sources for many conservation programs, including reintroduction, restoration, and population enhancement [4, 5]. Retaining high levels of genetic diversity in ex situ collections of endangered plant species is crucial for enhancing the evolutionary potential and fitness of the collection, which maximizes the success of any future reintroduction efforts [6, 7].

Nevertheless, endangered plants face a series of genetic risks during the process of ex situ conservation [8–10]. Ex situ collections usually begin without available genetic information, with characteristics such as incomplete sampling and unclear resources. This leads to uncertainty in terms of whether these ex situ populations formed with limited nursery resources are comparable in genetic diversity to wild populations. In addition, ex situ populations can be considered as small isolated populations due to the limited size of living collections for each endangered plant species in botanical gardens or germplasm nurseries [11]. Therefore, endangered plants in ex situ conservation have the same risk of decline in genetic diversity or extinction as those in small wild populations. These risks reduce the conservation value of ex situ collections. Furthermore, these challenges are compounded by the systemic limitations of the global ex situ conservation network, which is now operating at peak capacity and is constrained by international legislation, collectively hindering the comprehensive conservation of intraspecific genetic diversity [12]. Broad field sampling, sufficient population size and inbreeding management are widely adopted to maintain genetic diversity in ex situ conservation, however, genomic data is essential to guide such practices and optimize the genetic composition of ex situ populations [13–15]. Assessing the genetic diversity of both ex situ and in situ populations is therefore critical to evaluate and improve conservation strategies, including collection strategies and population management [16–19].

Ardisia gigantifolia Stapf (synonym Ardisia dasyrhizomatica) (Primulaceae) is an evergreen subshrub endemic to China, which was only distributed in Yunnan Province, China. This species possesses potential medicinal value and high ornamental value due to its big leaves with purplish-red abaxial surfaces, abundant pink flowers and red fruits [20] (Fig. 1). However, our cultivation experiments in botanical garden and field investigation found that its sexual reproduction is constrained by low natural fruit set, a long fruit ripening period (approximately nine months), and a protracted germination phase about eight months from sowing. Leaves of this species are harvested for medicinal purpose by local people, and mainly from wild populations. Besides, due to the development of artificial economic forests (such as banana and teak) in its native range, the habitat of this species continues to shrink and become fragmented. Because of limited reproductive ability, overharvesting and continuous habitat degradation, only a few wild populations remain in the original distribution areas [21]. According to our recent extensive field investigation in 2024, only three wild populations were found in Hekou Town (HK), Gulinqing Town (GLQ) and Nanxi Town (NX), Yunnan Province, China (Fig. 2a). Because of its rarity and narrow distribution, A. gigantifolia is ranked as one of the critically endangered species on the Red List of Biodiversity in China [22] and as one of the key protected wild plants in Yunnan Province. According to our investigation in 2023, the ex situ conservation programs of A. gigantifolia were only implemented by the South China Botanical Garden (SCBG). The SCBG ex situ population was established in 2016 using seedlings from a wild population at Mahuangbao Farm (MHB), Yunnan Province. Notably, no surviving individuals were found at the MHB site during our field investigation in 2024 (Fig. 2a), indicating the probable extinction of this wild population and thus underscoring the critical importance of the original ex situ conservation effort. To address the critical conservation status of A. gigantifolia, we performed the first genetic assessment of its wild and ex situ populations to evaluate genetic representativeness and inform future conservation strategies.

Fig. 1.

The habitat and morphological characteristics of A. gigantifolia: (a) habitat, (b) plant, (c) flowers and pollinated insect, (d) fruits

Fig. 2.

a Geographic distribution of the three extant wild populations (green spots, GLQ, HK and NX) and the one extinct wild population (red spots, MHB) from which the SCBG ex situ population was collected. b Genetic structure of wild and ex situ populations, with different colors representing putative ancestral components (K = 5). c Principal component analysis of the 67 individuals, with different colors representing different populations. d The maximum likelihood phylogenetic tree for the 67 individuals. e Gene flow analysis of the three wild populations

The rapid development of high-throughput sequencing has led to the application of restriction-site associated DNA sequencing (RAD-seq), which is a rapid, cost-effective approach to discovering hundreds to thousands of single nucleotide polymorphism (SNP) markers in non-model species. Therefore, we employed RAD-seq on the wild and ex situ populations of A. gigantifolia with two main objectives: (1) to determine the levels of genetic diversity and differentiation among the remaining wild populations of A. gigantifolia, and (2) to evaluate the representativeness of the existing ex situ collection in capturing extant genetic variation of the species. The results of this study provide strategies for the ex situ conservation and management of A. gigantifolia.

Results

Population characteristics and reproductive performance of A. gigantifolia

The three wild populations are located on the hillside with slopes larger than 30◦ under the forest. The relative irradiance was in the order of NX > HK > GLQ (Table 1). A total of 137 individuals of A. gigantifolia were recorded in the three extant populations, including 52 individuals in the GLQ population, 25 individuals in the HK population and 60 individuals in the NX population. (Table 1).

Table 1.

Sampling information and habitat characteristics of the four studied populations of A. gigantifolia

Code	GLQ	HK	NX	SCBG
Population type	Wild population	Wild population	Wild population	Ex situ population
Sampling number	21	15	21	10
Population size	52	25	60	10
Latitude (N°)	22.73	22.65	22.6	23.18
Longitude (E°)	103.91	103.86	103.97	113.37
Altitude (m)	786	537	172	34
Slope	38°	47°	42°	-
Relative irradiance	7.40%	12.60%	17.20%	32.00%
Distance to stream (m)	> 30	3–5	3–5	-

Pollination experiments indicated that no fruits were produced via spontaneous self-pollination (348 treated flowers, fruit set = 0), demonstrating the absence of spontaneous self-pollination capacity. The fruit set under artificial self-pollination reached 3.20% (438 treated flowers), whereas artificial cross-pollination yielded a higher fruit set of 7.29% (425 treated flowers). The index of self-incompatibility (ISI) fell within the range of 0.2–0.8 (ISI = 0.56), suggesting that this species is partially self-compatible with a mixed mating system [23].

SNP calling

ddRAD sequencing was performed on 67 individuals of A. gigantifolia, and a total of 597,005,458 raw reads (sample reads range: 6,369,304 − 19,122,386) were generated. After quality filtering, 579,366,728 reads were retained (sample reads range: 6,101,254 − 18,582,040). The average GC content per sample was 39.25% (35.23%-44.83%). We retained 41,168,112 loci using the Stacks module. After retaining one SNP per locus, 24,220 SNPs were finally obtained for further analysis.

Genetic diversity

The genetic diversity of A. gigantifolia was calculated using 24,220 SNPs. Among the four populations, the NX population exhibited the highest H_O, followed by GLQ, and the lowest was observed in HK (Table 2). The highest H_E was detected in the NX population, followed by SCBG, and the lowest was in HK. The magnitude ranking of the π in each population was consistent with the H_E (Table 2). The F_IS value was used to assess the inbreeding level within the populations. The three wild populations and one ex situ population exhibited F_IS values greater than zero, in the order of NX > SCBG > GLQ > HK (Table 2).

Table 2.

Genetic diversity analysis of A. gigantifolia for each population

Population	H O	H E	π	F IS
GLQ	0.1706	0.2082	0.2141	0.1174
HK	0.1441	0.1575	0.1636	0.0558
NX	0.1767	0.2255	0.2319	0.1596
SCBG	0.1666	0.2112	0.2243	0.1529

Genetic structure

Admixture analysis based on the SNP dataset, using cross-validation error, revealed that the CV error reached its minimum value at K = 5 (Figure S1). At K = 5, the four populations of A. gigantifolia were clearly distinguished in accordance with geographical distribution (Fig. 2a, b). Moreover, there was no shared genetic components among the three wild populations (Fig. 2b). In the SCBG ex situ population, two individuals exhibited admixture and at least four genetic components (Fig. 2b). Consistent with Admixture results, both the results of principal components analysis (PCA) and maximum likelihood (ML) trees revealed four major clusters corresponding approximately to the four populations (Fig. 2c, d). However, the result of PCA analysis showed that one individual of the SCBG ex situ population exhibited separation from other individuals of the population (Fig. 2c). The result of ML tree also demonstrated that one individual of the SCBG ex situ population was not grouped on the SCBG branch but on the GLQ branch (Fig. 2d). The result of Admixture showed that more than half of the genetic components of an individual in the SCBG ex situ population come from the GLQ population (green) (Fig. 2b).

Genetic differentiation and gene flow

All the pairwise F_ST values for the four populations were statistically significant (P < 0.01, Table 3). Moderate degrees of differentiation were detected between the SCBG ex situ population and NX population, as well as SCBG and GLQ (0.0953–0.1455) (Table 3). However, strong degrees of differentiation were detected among the three wild populations (0.1502–0.2297) and the highest degree of differentiation was detected between the GLQ and HK population (Table 3). Treemix analysis indicated that no significant gene flow was detected among the three wild populations (Fig. 2e). Analysis of molecular variance (AMOVA) indicated that 54.44% (P < 0.01) of the genetic variation occurred among populations, and 45.96% (P < 0.01) of the variation occurred within populations (Table 4).

Table 3.

Genetic distance (F_ST values) between all pairs of A. gigantifolia populations

	SCBG	HK	GLQ
NX	0.0953**	0.1777**	0.1502**
SCBG	–	0.1831**	0.1455**
HK	–	–	0.2297**

Significance: ^**, P < 0.01

Table 4.

Analysis of molecular variance (AMOVA) performed for the four populations of A. gigantifolia

Source of variation	d.f.	Sum of squares	Variance components	Percentage of variation	Fixation Index
Among populations	3	28,039.30	544.06632	54.04	0.5404**
Within populations	63	29151.51	462.72237	45.96
Total	66	57190.806	1006.78869

Significance: ^**, P < 0.01; d.f., degree of freedom

Discussion

Maximizing genetic diversity is paramount for securing the evolutionary potential and adaptive resilience of conservation collections [24]. However, randomly increasing the number of represented individuals alone is unlikely to achieve such goals [14]. Strategic management must be informed by genomic data to optimize ex situ population composition. Our population genomic analysis of Ardisia gigantifolia establishes an empirical genetic baseline, informing strategic conservation by delivering critical insights for assessing conservation status and guiding future genetic rescue efforts.

Low genetic diversity

Genetic diversity is closely related to the adaptive evolution and evolutionary potential of species [25]. Describing the distribution of the remaining diversity is critical to the long-term in situ survival of threatened species and the establishment of representative ex situ populations [26, 27]. Compared with other endangered shrub species in which RAD-seq was also used for population genomics analyses, including Camellia tunghinensis (H_E = 0.247–0.262, π = 0.252–0.271) [28] and Lonicera oblata (H_E = 0.2863, π = 0.2874) [29], the three wild populations of A. gigantifolia exhibited low levels of H_E and π (Table 2). These results suggest that the genetic diversity of A. gigantifolia is comparatively low. Low genetic diversity is commonly the result of a combination of several factors, including restricted gene flow, historical events and recent bottlenecks [30]. Previous studies have shown that the presence of frequent gene flow between fragmented and small populations is effective in maintaining high genetic diversity, while the lack of gene flow is not conducive to maintain high genetic diversity [31, 32]. The results of Admixture indicated that there was no shared genetic component among the three wild populations, suggesting a lack of gene flow among them (Fig. 2b). The results of Treemix supported the absence of gene flow among the wild populations (Fig. 2e). Besides, disturbances such as road constructions, selective farming and excavation for medical usage have changed locally ecological and genetic processes, perhaps resulting in population declines, further aggravating the loss of genetic diversity [30].

Different A. gigantifolia populations possessed different levels of genetic diversity. The NX population (H_E = 0.2255, π = 0.2319) exhibited higher genetic diversity than the GLQ (H_E = 0.2082, π = 0.2141) and HK population (H_E = 0.1575, π = 0.1636) (Table 2). This divergent genetic diversity pattern is potentially associated with inter-population differences in population size and microhabitat conditions. Our field surveys confirmed that the HK population had the smallest wild population size (only 25 individuals recorded) and displayed the lowest level of genetic diversity, which is consistent with the general expectation that small populations tend to harbor reduced genetic variation due to genetic drift and inbreeding. Notably, the NX (n = 60) and GLQ (n = 52) populations had comparable population sizes, while the NX population exhibited relatively higher level of genetic diversity. Although we did not directly quantify the causal effects of microhabitat on population genetic variation, the observed habitat differences may serve as a plausible explanatory factor. The NX site features moderate light intensity (relative irradiance = 17.2%) and proximity to water sources (around 3–5 m away from the closest stream), whereas the GLQ site is characterized by heavy shading (relative irradiance = 7.4%) and a longer distance to water (> 30 m) (Table 1). Our previous study demonstrated that seedlings at NX exhibited significantly higher light energy utilization efficiency but lower malondialdehyde (MDA) content than those at GLQ [21]. Combined with these physiological differences, the relatively favorable microhabitat at NX may potentially facilitate seedling regeneration and population recruitment, which could indirectly maintain higher genetic diversity. A similar positive association between suitable habitat conditions and elevated genetic diversity has been reported in Paeonia ludlowii [30]. Furthermore, microhabitat variation may indirectly influence population genetic patterns via pollination dynamics, though such pollinator-mediated effects were not directly tested in the present study. Our artificial pollination experiments showed that the breeding system of A. gigantifolia is a mixed mating system with characteristic of partial self-compatibility, similar to Ardisia humilis [33]. Positive inbreeding coefficients (F_IS) across all populations also indicated ubiquitous inbreeding within the studied populations (Table 1). Species in the genus Ardisia generally produce pollen-only reward flowers and are mainly pollinated by small generalist bees [34, 35]. Previous studies have suggested that heavily shaded understory habitats may reduce the activity and abundance of generalist pollinators, thereby restricting pollen dispersal [35, 36]. Accordingly, the severe shading at the GLQ site potentially limits pollen flow and aggravates inbreeding, which may partly explain its relatively lower genetic diversity compared with the NX population.

Significant genetic differentiation

An understanding of genetic structure is vital for identifying the correct conservation and management strategies [19, 37]. The Admixture analysis identified K = 5 as the optimal clustering solution for the four studied populations (Figure S1, Fig. 2b). This result reflects fine-scale genetic structure within the wild populations and the ex situ population. The GLQ population exhibited admixture of two distinct genetic components, likely reflecting historical secondary contact and gene flow between two previously isolated lineages [38]. The HK and NX populations were each dominated by a single genetic cluster, demonstrating their genetic distinctiveness. The SCBG ex situ population was dominated by a unique genetic component with minor admixture from other populations (the NX and GLQ populations), consistent with its wild origin. Thus, the five clusters correspond to both major geographical populations and the cryptic substructure within the GLQ population.

The analyses of Admixture, PCA and ML tree demonstrated that the three wild populations of A. gigantifolia are genetically separated in accordance with their geographical distribution (Fig. 2). The genetic differentiation among all the four populations was further evaluated. Typically, F_ST values in the range of 0.05–0.15 and 0.15–0.25 are indicative of moderate and large differentiation, respectively [39]. Using this criterion, A. gigantifolia showed a large differentiation among the three wild populations (F_ST range from 0.1502 to 0.2297). For rare and endangered plant species, geographic isolation is generally considered as a key factor driving inter-population genetic differentiation [40]. The sampled wild populations are geographically isolated by distances of more than 30 km and high mountain terrain, which may potentially hinder inter-population gene flow. Although we did not directly monitor pollen dispersal and seed propagation dynamics in the field, the limited gene flow observed in this study could be plausibly associated with the species’ reproductive and dispersal traits documented in previous studies and our field investigation. Specifically, A. gigantifolia is mainly pollinated by small generalist bees that typically exhibit short foraging distances [41], which may restrict long-distance pollen exchange among populations. Seed dispersal of this species primarily depends on rolling along mountain slopes, a trait that likely results in a narrow seed dispersal range. Furthermore, our field surveys showed that all wild populations inhabit steep slopes with gradients over 30°, and seedlings are sporadically distributed around maternal individuals. These habitat and biological characteristics may collectively contribute to the restricted gene flow and pronounced genetic differentiation among A. gigantifolia populations. This differentiation pattern is consistent with the findings reported for the endangered species Rhododendron meddianum [40].

Salvaged yet incomplete ex situ collection

Our results revealed that the small, rescue-derived ex situ collection of A. gigantifolia represents a notable salvage achievement yet a strategic paradox. The pivotal finding is that this collection retains a level of genetic diversity comparable to the best-preserved wild population (NX). This success is a direct result of its timely founding from seedlings of the now-extinct MHB population, a method that effectively captured and preserved a broad sample of maternal genetic lineages [28, 42]. This underscores the irreplaceable value of proactive, provenance-based rescue translocations. However, the three extant wild populations are strongly differentiated from each other, and the SCBG ex situ population forms a distinct fourth genetic cluster, showing moderate-to-high differentiation from all of them (F_ST = 0.0953–0.1831). This exposes a critical “representation gap” in the current conservation strategy, given that over half (54.04%) of the species’ total genetic variation is partitioned across the four distinct populations (Table 4).

This “representation gap” revealed a critical methodological flaw, whereby ex situ collections capture only a biased subset of a species’ genetic diversity due to non-representative sampling. The prevalence and consequences of this sampling bias are corroborated by genomic studies across multiple threatened plant species. For instance, a genomic assessment of the critically endangered oak Quercus hinckleyi revealed that its ex situ collection, although containing novel alleles, was sourced from only one of the three major genetic clusters identified in the wild, resulting in a significant conservation gap [43]. Similarly, for the critically endangered Tanzanian tree Karomia gigas, the ex situ collection represented only 32.6% of the wild genetic variation, leading to severe under-representation and genetic bias [44]. The analyses of ex situ conservation worldwide revealed that the majority of ex situ populations fail to cover all genetic clusters, largely due to spatially biased sampling with low geographic coverage [45]. The contrasting success case of Cupressus chengiana, where a scientifically managed ex situ population harbored higher genetic diversity, demonstrates that this bias is not inevitable and can be overcome with deliberate, genetically informed sampling [46].

Conclusion

Our study on the genetic diversity and structure of A. gigantifolia provides important conservation implications for this critically endangered species. Specifically, the genomic evidence presented here necessitates a strategic shift from general ex situ collection to precise, genetics-informed management for A. gigantifolia. The immediate priority is to augment the ex situ collection by sampling across the three remnant wild populations, thereby closing the identified “representation gap” and capturing the species’ full genetic variance. In addition, the small and isolated ex situ collection is inherently vulnerable to genetic drift and inbreeding [47], necessitating active genetic management such as pedigree control to preserve its genetic diversity. Furthermore, establishing a long-term genetic monitoring regime will allow managers to quantify genetic erosion, evaluate the efficacy of augmentation efforts, and make data-driven decisions to ensure the viability and conservation value of the ex situ population. Finally, reintroduction of the distinct genetic variation back into natural habitat could enhance the species’ genetic resilience and adaptive capacity, provided it is strategically integrated with in situ conservation.

Materials and methods

Plant materials and sampling

In this study, three wild populations (HK, GLQ and NX) and one ex situ population (SCBG) of A. gigantifolia were selected for this study (Table 1). Measurements of habitat characteristics and population size were referenced to our previous study [21]. Sample collection was conducted in the summer of 2024 and was permitted by Forestry and Grassland Bureau of Hekou County and Protection Bureau of Gulinqing Nature Reserve, complying with national guidelines. The plant materials utilized in our study were meticulously identified by our team and the corresponding voucher was deposited at the Herbarium of the South China Botanical Garden (voucher specimen number: IBSC1104115). Fresh leaves were collected from the A. gigantifolia populations at four sites: 21 plants from Gulinqing Town (GLQ), 15 plants from Hekou Town (HK), 21 plants from Nanxi Town (NX), and 10 plants from South China Botanical Garden (SCBG) which were transplanted in 2016 at seedling stage from their native habitat at Mahuangbao Farm (Fig. 2a; Table 1). The distance between sampling plants within each wild population was greater than 5 m. After collection, leaf samples were quickly dried and preserved with silica gel.

Pollination experiments

To evaluate the reproductive system of A. gigantifolia, five healthy individuals were chosen from the SCBG ex situ population, with three randomly selected inflorescences per individual assigned to different pollination experiments: (a) no emasculation, no artificial pollination, and bagging to assess spontaneous self-pollination capacity; (b) emasculation followed by artificial outcross pollination and bagging to determine outcrossed fruit set; and (c) emasculation followed by artificial self-pollination and bagging to evaluate selfed fruit set. These pollination experiments were conducted in greenhouse at the South China Botanical Garden during the species’ flowering period (April and May) in 2026. The total number of treated flowers and fruits (one seed per drupe) for each treatment were recorded to calculate the fruit set. To assess the strength of self-incompatibility, the index of self-incompatibility (ISI) was calculated with the following formula: ISI = 1 − (selfed fruit set rate / outcrossed fruit set rate) [23, 48].

Library preparation and sequencing

The total genomic DNA was extracted from silica-dried leaflets with modified CTAB (Cetyltrimethyl ammonium Bromide) protocol [49]. The DNA quality was assessed using agarose gel electrophoresis and subsequently quantified using a Qubit 3.0 Fluorometer, ensuring that each sample met the following criteria: total mass > 0.5 µg, concentration > 12 ng/µl, and OD260 / OD280 = 1.80-2.00. The quality-checked DNA was digested with the DpnII and MspII restriction enzymes, followed by ligation of barcode adapters using T4 ligase. Fragments ranging from 220 to 450 bp were isolated and purified, followed by PCR amplification, ligation of sequencing adapters and purification. The constructed libraries were sent for 150-bp paired-end sequencing on the Illumina Novaseq 6000 platform to yield raw data. Sequencing was completed by Shanghai Personalbio Biotechnology Co., Ltd.

Quality filtering and SNP calling

Quality assessment of the raw data was performed using Fastp v0.20.0 [50]. Adapter sequences, low quality reads, cut sites (the first six base pairs of reads) and ambiguous bases in the raw sequences were filtered with Fastp v0.20.0 [50]. Specifically, 3’-end adapter sequences were removed. A 5 bp sliding window was used for quality filtering from the 3’ to 5’ end: bases with an average Phred quality score < 20 were removed. Any paired-end reads with a length ≤ 50 bp or more than five N bases were filtered. Subsequently, SNPs with a missing data rate higher than 20% were excluded. Loci showing significant deviation from Hardy-Weinberg equilibrium (P < 0.01) and abnormal heterozygosity were also removed. The clean reads were then analyzed using Stacks v2.59 [51]. The ustacks module was used to cluster single sample sequences to form loci, with the parameters most influencing the number and error rate of SNPs were set as follows: minimum depth of coverage required to create a stack (-m), 3; maximum distance (in nucleotides) allowed between stacks (-M), 3. Loci data of all samples were merged into a catalog using cstacks module with default parameters. The sstacks module was run to compare the single sample loci to the catalog so as to obtain the SNPs. Finally, the populations module was used to screen SNPs, with parameters set as follows: minimum number of populations a locus must be present in to process a locus (-p), 4; minimum percentage of individuals in a population required to process a locus for that population (-r), 0.7; minimum minor allele frequency required to process a nucleotide site at a locus (--min_maf), 0.01; maximum observed heterozygosity required to process a nucleotide site at a locus (--max-obs-het), 0.5. The output result was exported in VCF format using the populations module.

Analysis of genetic diversity and structure

Based on filtered high-quality SNP data, genetic diversity parameters including heterozygosity (H_O), expected heterozygosity (H_E), nucleotide diversity (π), and inbreeding coefficient (F_IS) were calculated by the population module of Stacks v2.59 [51].

Arlequin v3.5.2 [52] was used to estimate pairwise genetic differentiation coefficient (F_ST) between populations (1000 permutations) and to calculate the genetic variation between and within populations with analysis of molecular variance (AMOVA). Bayesian clustering was performed on all individuals using Admixture v1.3.0 [53] with clustering values (K) set from 2 to 10. The optimal K value was determined based on the minimum cross-validation (CV) error [54]. Complementary to the Bayesian clustering, principal components analysis (PCA) was performed via Plink v1.07 [55]. A phylogenetic tree was constructed for the 67 individuals using the maximum likelihood (ML) method in FastTree v2.1.11 [56] software, with bootstrap replications conducted 1000 times. Treemix v1.12 [57] was used to detect gene flow between the wild populations. The migration events (m) were assumed to be between 0 and 5.

Abbreviations

RAD

Restriction-site associated DNA

HK

Wild population from Hekou Town, Hekou County, Yunnan Province

GLQ

Wild population from Gulinqing Town, Maguan County, Yunnan Province

NX

Wild population from Nanxi Town, Hekou County, Yunnan Province

SCBG

Ex situ conservation population conserved at the South China Botanical Garden, Guangzhou City, Guangdong Province (origin: Mahuangbao Farm)

MHB

Mahuangbao Farm, Hekou County, Yunnan Province

SNP

Single nucleotide polymorphism

AMOVA

Analysis of molecular variance

PCA

Principal components analysis

Authors' contributions

RL and ZLN designed the experiments. RL, DX, MLL, YJY, JGZ, YMS and ZLN carried out the experiments. RL, DX, JW and ZLN analyzed the data and wrote the main manuscript. All authors have read and approved the final manuscript.

Funding

The research was supported by the Guangzhou Municipal Science and Technology Project (2025A04J4134), the Guangdong Science and Technology Program (2022B1111230001), the Guangdong Flagship Project of Basic and Applied Basic Research (2023B0303050001), and the Yunnan Ecological Monitoring Network and Project (2026-YN-06).

Data availability

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA041821) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.