1. Introduction

Ex situ conservation, the conservation of target species outside their natural habitats (Havens et al., 2006), is a widely used approach to protect threatened plant species (Mounce et al., 2017; Bolam et al., 2023). The primary goal of ex situ conservation is to preserve genetic variation (Cohen et al., 1991; Maunder and Byers, 2005), which is crucial for plant individual fitness, population persistence, and ecosystem stability in changing environments (Reed and Frankham, 2003; Jump et al., 2009; Enβlin et al., 2011; Evans et al., 2018; Tito et al., 2020; Backs et al., 2021). Ex situ conservation also provides valuable genetic resources for future wild reintroduction (Schafer et al., 2020).

One critical criterion for assessing the success of ex situ conservation programs is their genetic representativeness, i.e., how well ex situ populations capture the genetic variation of wild populations (Cibrian-Jaramillo et al., 2013; Gargiulo et al., 2019). Genetic variation of wild populations consists of genetic diversity and differentiation (Hoban and Schlarbaum, 2014). In most cases, genetic composition (i.e., alleles and their frequencies) is unevenly distributed across the natural range of wild plant species (Lumibao et al., 2017; Pironon et al., 2017), and this leads to the uneven distribution of genetic diversity and range-wide population genetic structure (Fig. 1). Genetic diversity is closely related to effective population size, which is largely determined by environment suitability (Ellegren and Galtier, 2016). The geographical patterns of genetic diversity are associated with the central-marginal hypothesis, rare-leading edge hypothesis, and post-glacial re-colonization (Petit et al., 2003; Eckert et al., 2008; Duffy et al., 2009; Guo, 2012; Wei et al., 2016; Pironon et al., 2017). For aquatic and riparian plant species, the unidirectional dispersal hypothesis predicts a downstream accumulation of genetic diversity along rivers (Liu et al., 2006; Honnay et al., 2010). Range-wide population genetic structure is primarily determined by landscape connectivity, environmental gradients, and species’ dispersal abilities (Lumibao et al., 2017; Pironon et al., 2017). Owing to the limited gene flow caused by genetic barriers, genetic composition varies considerably among different genetic clusters or conservation units (e.g., evolutionary significant units or management units) (Moritz, 1994; Palsbøll et al., 2007; Funk et al., 2012).

In the last few decades, ex situ conservation genetic studies have increasingly compared genetic variation between ex situ and wild populations. In a global meta-analysis, Wei and Jiang (2021) found that genetic diversity in ex situ plant populations is significantly lower than that in their source wild populations. However, it remains overlooked whether all wild genetic clusters were collected for species with range-wide population genetic structure (Hoban and Schlarbaum, 2014). Different wild genetic clusters are believed to have unique genetic composition that enhance a species’ ability to adapt to changing environments (Mutegi et al., 2014; Volis, 2015; Vlasta and Münzbergová, 2022; Van Natto and Eckert, 2022) and provide unique solutions to evolutionary pressures (Lesica and Allendorf, 1995). The genetic clusters of target species can be determined through the visual and intuitive figures from multiple methods that used to explore genetic structure of both wild and ex situ populations (Backs et al., 2021; Závada et al., 2023). These methods include Bayesian clustering methods (e.g., STRUCTURE and ADMIXTURE; Pritchard et al., 2000; Alexander et al., 2009), principal component analysis (PCA), discriminant analysis of principal components (DAPC; Jombart et al., 2010), agglomerative hierarchical clustering (e.g., unweighted pair-group method with arithmetic, UPGMA; Sokal and Michener, 1958), and iterative clustering methods (e.g., neighbor-joining trees; Saitou and Nei, 1987). The coverage of wild genetic clusters in ex situ populations is commonly included in most empirical studies, but the results are mixed (Li et al., 2005; Lauterbach et al., 2012; Diaz-Martin et al., 2023; Phang et al., 2024; Schumacher et al., 2024).

Spatially biased sampling, a common and flawed sampling strategy in previous ex situ conservation practices that did not cover the entire natural distribution range or all of the genetic clusters of the target plant species (Wei and Jiang, 2021), is the primary cause of lower genetic representativeness in ex situ populations (Hoban and Schlarbaum, 2014; Thomas et al., 2023; Schumacher et al., 2024). However, because the uneven distributions of genetic diversity and genetic clusters are determined by different factors, the extent to which they are captured and retained can vary under the same sampling strategy (Fig. 1; Lauterbach et al., 2012; Diaz-Martin et al., 2023; Forgiarini et al., 2023). In other words, spatially biased sampling does not necessarily result in both lower genetic diversity and low coverage of wild genetic clusters in ex situ populations (Fig. 1). Theoretically, if we consciously or unconsciously collect an appropriate number of materials solely from core or marginal populations, ex situ populations may exhibit high or low levels of genetic diversity, respectively. In either case, if the sampled wild populations are genetically distinct from other populations of the target species, the ex situ populations will not cover all the wild genetic clusters (Hoban and Schlarbaum, 2014). To date, empirical studies directly illustrating this phenomenon are lacking (Hoban and Schlarbaum, 2014; Zumwalde et al., 2022).

Myricaria laxiflora, an endangered riparian shrub native to the Yangtze River in central China (Wu et al., 1998; Shen et al., 1999; Liu et al., 2006; Bao et al., 2010), offers an interesting test case for the effects of spatially biased sampling. M. laxiflora primarily inhabits the water-level-fluctuation zone along the main stream of the Three Gorges of the Yangtze River (Shen et al., 1999; Bao et al., 2010). Habitat loss caused by the construction of the Three Gorges project has necessitated ex situ conservation efforts (Zhang and Zhang, 1984; Chen et al., 1994; Shen et al., 1999; Xu et al., 1999; Li et al., 2003; Liu et al., 2006). Early ex situ collections of M. laxiflora (1993–2009) mainly focused on populations projected to be submerged at the Three Gorges Reservoir (Li et al., 2003). Several new populations were subsequently discovered both upstream and downstream of the Three Gorges Reservoir (Bao et al., 2010; Chen, 2018). However, conservation genetic studies have mainly focused on older populations and newly discovered downstream populations (Li et al., 2003; Liu et al., 2006, 2010; Tian et al., 2012). The geographical patterns of genetic variation of all these newfound and remnant wild populations, as well as the genetic representativeness of ex situ populations, remain unclear.

The primary aim of this study is to investigate the extent of spatially biased sampling in living collections and the coverage of wild genetic clusters in plant populations under ex situ conservation worldwide. For this purpose, we first synthesized data from ex situ conservation studies from around the world to investigate the prevalence of spatially biased sampling and its impact on the coverage of wild genetic clusters in ex situ populations. We then assessed how effectively ex situ populations capture the genetic diversity and represent wild genetic clusters of an endangered riparian plant species, Myricaria laxiflora.

2. Material and methods 2.1. Global synthesis

To investigate the extent of spatially biased sampling in living collections and the coverage of wild genetic clusters in plant populations under ex situ conservation worldwide, we searched for peer-reviewed articles, using the same keyword combinations as Wei and Jiang (2021). The search was conducted in April 2024 across multiple databases, including the ISI Web of Science, Scopus, Elsevier, Springer, Wiley, and the China National Knowledge Infrastructure (CNKI; https://ww.cnki.net/). Additionally, we collected 56 articles through cross-referencing. After removing duplicates and screening by title and abstract, we identified 4658 relevant articles. We then filtered these publications based on the following criteria: (1) population genetic structure was analyzed using individuals from both ex situ (i.e., living collection or seed banks) and wild populations; (2) population genetic structure was visually represented (i.e., using STRUCTURE, ADMIXTURE, PCA, DAPC, UPGMA, etc.). These criteria narrowed the collection to 176 relevant articles (including one case from the current study; Table S1).

Here, we consider spatially biased sampling as a collection range of living materials (e.g., seeds, seedlings, or cuttings) for ex situ conservation that does not cover most (≥ ca. 75%) of the natural distribution range of target species. To identify spatially biased sampling in previous studies, we extracted and compiled data on target species, taxonomic family, life form, continent (Fig. 2), country/region, numbers of wild and ex situ populations, sample size, collection range for ex situ conservation, and coverage of wild genetic clusters in ex situ populations, as well as information on publication year, molecular marker used. If available, information of collection range was obtained from the original article; if not available, we obtained the natural distribution range of target species from the GBIF website or other available literature, and then determined whether the collection range covered most of the natural distribution range. As the sample size varies among different case studies and there is no standard on how many individuals is sufficient to represent each wild genetic cluster, here we only tested whether ex situ populations covered all wild genetic clusters and did not consider how many individuals of each genetic cluster were collected in ex situ populations. Specifically, we screened the visual figures presenting population genetic structure of wild and ex situ populations. If all wild genetic clusters of a target species were represented in ex situ populations, we considered all wild genetic clusters covered in the case study (Fig. 1). If one or more wild genetic cluster was not represented in ex situ populations, we considered a subset of the wild genetic clusters uncovered (Fig. 1).

We divided the cases into two groups based on whether sample collection for ex situ conservation covered or did not cover most of the natural geographic range of target species. We then further divided each group into two subgroups, one in which the ex situ populations covered wild genetic clusters and one in which it did not cover wild genetic clusters. We then used a chi-squared test to assess the effect of spatially biased sampling on the coverage of wild genetic clusters in ex situ populations.

2.2. Case study 2.2.1. Study species

Myricaria laxiflora (Tamaricaceae) is a diploid (2n = 24; Li et al., 2003) evergreen shrub and a habitat-specialist species, with its range confined to riverbanks and eyots along the Yangtze River (Liu et al., 2006). This riparian plant produces insect-pollinated flowers and very small seeds that are dispersed by wind and/or water. Seed-germinated seedling is the primary means of recruitment. Due to ongoing anthropogenic disturbances and its restricted distribution area, M. laxiflora has been designated as one of China's National Key Protected Wild Plants.

2.2.2. Sample collection and DNA extraction

We collected leaf samples from 222 individuals of Myricaria laxiflora across all known populations, including eight wild populations (WU1–WU5, n = 114; WD1–WD3; n = 72) and three ex situ populations (ES1–ES3; n = 36) in central-west China between 2019 and 2021 (Fig. 3A and Table 1). The leaf samples were preserved in liquid nitrogen or silica gel in the field and stored at −80 ℃ in the laboratory until DNA extraction. The number of samples per population ranged from 12 to 24 (Table 1). The wild populations were distributed with five located upstream (WU1–WU5) and three downstream (WD1–WD3) of the TGD (Fig. 3A and Table 1).

Genomic DNA was extracted using a CTAB protocol (Doyle and Doyle, 1987). DNA concentration was tested using an Invitrogen^TM Qubit^TM 3 Fluorometer (Fisher Scientific, Loughborough, UK) after verifying DNA integrity and overall quality. All 222 samples met the sequencing requirements.

2.2.3. Genotyping-by-sequencing

The genotyping-by-sequencing library preparation was conducted as follows. First, 200 ng of genomic DNA was fully digested with the restriction enzyme Mse I Sacl (New England Biolabs). The enzyme digestion products were ligated to a custom-designed specific adapter, and a specific amount of the connection products was pooled and purified using AMPure XP paramagnetic beads (Beckman Coulter, Brea, CA, USA). PCR enrichment was then performed using high-fidelity polymerase KOD-Plus-Neo (TOYOBO). Finally, all products were pooled and subjected to low-pressure overnight electrophoresis with Bio-Rad Certified Megabase Agarose. Fragments between 380 and 480 bp was selected and purified using agarose (QIAGEN).

The quality of the genomic libraries was rigorously tested before proceeding to sequencing. Initial quantification was performed using Qubit 3.0 (Thermo Fisher Scientific). The insert size of the library was tested using Agilent 2100 Bioanalyzer. Further analysis was only performed after the insert size met the expectations and no adapter contamination was detected. The effective concentration of the library was accurately quantified using the QTOWER real-time fluorescence quantitative PCR instrument (ANALYTIKJENA, Germany), with > 2 nM considered a qualified library. The genotyping-by-sequencing libraries were sequenced on the Illumina HiSeq platform using paired-end 150 bp (PE150) reads.

Single-nucleotide polymorphisms (SNPs) were extracted using stacks 2.5.9 (Rochette et al., 2019). The process_radtags module was used to demultiplex and filter the raw sequencing reads. Because there is no reference genome, the reads were assembled de novo in stacks 2.5.9 with the default parameters. Coverage was also calculated using stacks 2.5.9. Additionally, we applied quality filters, including a minor allele frequency (MAF) threshold of 0.05 and a missing data threshold of 0.2.

2.2.4. Genetic diversity and inbreeding coefficient

We estimated observed heterozygosity (H_O), expected heterozygosity (H_E), nucleotide diversity (π), and inbreeding coefficient (F_IS) using the populations module in stacks 2.5.9 (Rochette et al., 2019). Comparisons of genetic parameters (H_E, H_O, π, and F_IS) between different groups of Myricaria laxiflora (i.e., wild vs. ex situ, upstream (WU1–WU5) vs. downstream (WD1–WD3), upstream vs. ex situ, and downstream vs. ex situ populations) were performed using independent samples t-tests. Kolmogorov–Smirnov tests were performed to assess the normality of the data before performing these comparisons.

2.2.5. Population genetic structure

Population genetic structure was determined using the full set of SNP markers. We first inferred genetic clusters with the program ADMIXTURE 1.3.0, which employs a Bayesian clustering method (Alexander et al., 2009). This analysis was conducted with K values ranging from 1 to 10, and the K value with the lowest cross-validation (CV) error was selected as the optimal number of clusters. We used Genome-wide Complex Trait Analysis (GCTA 64 v.1.93.2) software (Yang et al., 2011) to perform PCA by calculating eigenvectors and eigenvalues. Pairwise genetic differentiation (F_ST) values among populations were calculated using stacks 2.5.9, and the results were visualized with the R package “gplots” (Warnes et al., 2016).

3. Results 3.1. Global synthesis

The 201 cases from the 176 articles were distributed across all inhabited continents and spanned 79 countries or regions (Fig. S1). After 2000, the number of studies increased rapidly, though with inter-annual variability (Fig. S2). Of the 12 types of molecular markers recorded, the most commonly used (45%) were nuclear simple sequences repeats (nSSR) (Fig. S2). Except for the 11 cases in which sample size was unavailable, the average sample size was < 30 in 140 cases, while it was ≥ 30 in the remaining 51 cases.

In most case studies (166/201; 82.6%), ex situ populations failed to capture all the genetic clusters of the wild populations (Fig. 2). Similarly, excluding the 84 cases in which geographic coverage information was unavailable, initial material collection for ex situ conservation covered only part of (< ca. 75%) the natural distribution range of the target species in most cases (92/117; 78.6%); most of the distribution range was covered in only 25 cases (21.4%) (Fig. 2). As expected, failure to cover all wild genetic clusters in ex situ populations was more likely to occur when spatially biased sampling was used for ex situ conservation (χ² = 7.841, df = 1, p = 0.005).

3.2. Case study of Myricaria laxiflora 3.2.1. Genotyping and quality control

We genotyped 222 individuals of Myricaria laxiflora (wild: 186; ex situ: 36) (Table 1). After demultiplexing, we retained an average of 2,803,923 incorporated reads per individual (range = 736,024–20,349,135) (Table S2). The stacks 2.5.9 populations module identified 2,031,031 SNPs with an average read depth per individual of 23.2× (SD = 15.0×, min = 8.92×, max = 107.3×) (Table S2). After filtering, the genomic data retained 68,600 SNPs for the following analysis.

3.2.2. Genetic diversity and inbreeding coefficient

H_E ranged from 0.021 (WU3 and WU4) to 0.265 (ES3), H_O ranged from 0.017 (WU3) to 0.250 (ES3), and π varied from 0.022 (WU3 and WU4) to 0.278 (ES3) (Table 1). Genetic diversity (i.e., H_E, H_O, and π) differed between ex situ and wild (upstream + downstream) populations (H_E, p = 0.005; H_O, p = 0.006; π, p = 0.004) (Fig. 4), but not between downstream and ex situ populations (Fig. 4). The genetic diversity of both ex situ (H_E, p < 0.001; H_O, p < 0.001; π, p < 0.001) and downstream (H_E, p < 0.001; H_O, p < 0.001; π, p < 0.001) populations was significantly higher than that of upstream populations (Fig. 4). Pairwise genetic differentiation (F_IS) ranged from 0.012 (WU4) to 0.072 (ES3) (Table 1). F_IS was significantly higher in ex situ populations than in both wild and upstream populations, but not between ex situ and downstream populations, nor between upstream and downstream populations (Fig. 4).

3.2.3. Population genetic structure

Admixture proportions were estimated from K = 2 to K = 10 (Fig. S3). The optimal clustering number was K = 10 for all Myricaria laxiflora individuals (Fig. S4). At K = 10, a high proportion of genetic admixture was detected between ex situ and wild populations, without a clear pattern. The second-lowest CV error occurred at K = 8 (Fig. S4). Nevertheless, our primary focus was on K = 2, where all individuals from upstream populations (WU1–WU5) were assigned to one cluster, while the remaining individuals from the downstream and ex situ populations were assigned to the other cluster (Fig. 3B). PCA results showed a strong separation between the upstream populations and all other populations, with the downstream and ex situ populations clustering into one group together (Fig. 3C). A large span of ex situ collections (yellow dots) does not sort with any wild populations, which may represent the living collections from the submerged populations (Fig. 3C).

Pairwise genetic differentiation (F_ST) was high among all populations, with an average of 0.3127, ranging from 0.0204 to 0.5420 (Table S3). The heatmap clearly highlights the distinction between upstream populations and the remaining populations (Fig. 5). ADMIXTURE (K = 2), PCA, and F_ST analyses showed a clear segregation between the upstream and other populations.

4. Discussion 4.1. Spatially biased sampling and failure to cover all wild genetic clusters: a global synthesis

Previous genetic assessments of ex situ plant populations have included both genetic diversity and the delineation of genetic clusters. Furthermore, several studies have recommended that population genetic structure should be qualitatively and quantitatively considered during material sampling for ex situ conservation (Hoban and Schlarbaum, 2014; Guerrant et al., 2015; Guja et al., 2015). However, many studies have solely focused on comparing levels of genetic diversity in ex situ and wild populations. Few studies have determined whether all wild genetic clusters of one species are being collected for ex situ conservation (Zumwalde et al., 2022). Our global synthesis of 201 case studies revealed that 82.6% of ex situ plant populations from around the world have failed to cover all genetic clusters of wild populations. Our analysis also suggests that the failure to cover all wild genetic cluster(s) is more likely when materials are collected from only part of the natural distribution range of the target species.

Our finding that spatially biased sampling is common in ex situ conservation of wild plants is important. An earlier global meta-analysis study reported that spatially biased sampling caused lower genetic diversity in ex situ populations (Wei and Jiang, 2021). Furthermore, Thus, we recommend that future ex situ conservation genetic studies should prioritize both genetic diversity and the coverage of wild genetic clusters, which are closely related to effective population size and geographic coverage (Rosenberger et al., 2021; Schumacher et al., 2024). In addition, to improve the representativeness of genetic diversity and wild genetic clusters in plant populations under ex situ conservation, plant materials must be collected from most of the distribution ranges of target species (Zumwalde et al., 2022; Schumacher et al., 2024).

4.2. Higher genetic diversity but a failure to cover all wild genetic clusters: a case study

Our case study provides rare empirical evidence that a spatially biased sampling strategy for living collections can result in higher genetic diversity but fail to cover all wild genetic clusters in ex situ population of an endangered riparian plant species. First, we found that the genetic diversity of Myricaria laxiflora in ex situ populations was significantly higher than that in both the extant wild populations (upstream + downstream) and the upstream populations, but comparable to that of the downstream populations (Fig. 4). Consistent with the unidirectional dispersal hypothesis (Honnay et al., 2010) and a previous conservation genetic study of this species (Liu et al., 2006), we found that the downstream populations of M. laxiflora had significantly higher genetic diversity than the upstream populations. Because the ex situ populations were collected only from submerged and downstream populations, our results indicated that the spatially biased sampling from source populations with high genetic diversity caused the elevated genetic diversity observed in the ex situ populations. It is worth noting that at least one of the three ex situ populations of M. laxiflora (e.g., E1) was established with materials from more than one wild population (Liu et al., 2006), and this is an effective sampling strategy to enhance genetic diversity in ex situ populations (Wei and Jiang, 2021).

Second, current ex situ populations of Myricaria laxiflora failed to cover all wild genetic clusters, because the upstream genetic cluster was missed. Individuals from the ex situ and downstream populations clustered into the same genetic group, while individuals from the upstream populations formed a distinct genetic cluster (Figs. 2 and 4). This confirmed that individuals of M. laxiflora under ex situ conservation originated from the submerged and remnant downstream wild populations (Table 1). In other words, spatially biased sampling caused the genetic composition in the current ex situ populations to be biased towards the downstream populations, while failing to capture that of the upstream populations.

In general, our case study demonstrates that collecting materials solely from wild plant populations with high genetic diversity (i.e., core or downstream populations) can result in ex situ populations with high levels of genetic diversity. However, this strategy may lead to the omission of germplasm from peripheral, or upstream populations, and then fail to capture locally adapted genotypes across the entire range of a species (Guerrant et al., 2014; Chacón-Vargas et al., 2020; Clugston et al., 2022; Liu et al., 2024). Therefore, effective sampling strategies for plant ex situ conservation must collect genetically representative samples not only from wild populations with high genetic diversity but also from those with unique genetic composition (Guerrant et al., 2014; Hoban et al., 2018).

5. Conclusions

In this study, we investigated whether wild genetic clusters are adequately covered in ex situ populations of wild plants, and if not, why not. Our global synthesis of previous studies revealed that ex situ populations rarely cover all wild genetic clusters of a target species, primarily due to spatially biased sampling strategies with low geographic coverage. Our own case study of an endangered riparian plant species revealed that a spatially biased sampling strategy was able to constitute an ex situ population with a high level of genetic diversity, but failed to cover an upstream genetic cluster. This challenges the common perception that spatially biased sampling typically creates ex situ populations with lower genetic diversity than that found in wild populations (Wei and Jiang, 2021). Importantly, our findings indicate that sampling strategies for the ex situ conservation of plant species must consider both the level of genetic diversity and the coverage of wild genetic clusters.

To increase the representativeness of both genetic diversity and wild genetic clusters, we recommend designing initial sampling strategies. For species with no genetic structure in the wild (e.g., those with only one wild population), collection with an appropriate number of materials (e.g., ≥ 30 or all available individuals) is a suitable sampling strategy. For species with range-wide population genetic structure in the wild, we recommend considering both geographic coverage (e.g., all genetic clusters) and effective population size (e.g., ≥ 30) when collecting materials to establish ex situ populations (Hoban and Schlarbaum, 2014; Wei and Jiang, 2021). To avoid outbreeding depression, ex situ populations should be established separately for each genetic cluster or conservation unit. Within each unit, a mixing strategy (i.e., collecting materials from multiple source populations) can be used to increase the genetic diversity of ex situ populations (Wei and Jiang, 2021).

Acknowledgements

We thank Ms. Lvjuan Wang for her assistance in the fieldwork. This work was financially supported by National Key Research and Development Program of China (2024YFF1307400), Hubei Provincial Natural Science Foundation and Three Gorges Innovation Development Joint Fund (Grant No. 2023AFD195), and China Three Gorges Corporation (NBZZ202300130).

CRediT authorship contribution statement

Zhiqiang Xiao: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Writing-original draft, Writing-review & editing, Visualization, Project administration, Funding acquisition. Hui Liu: Methodology, Formal analysis, Investigation, Resources. Guiyun Huang: Conceptualization, Methodology, Formal analysis, Investigation, Resources. Di Wu: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Funding acquisition. Liwen Qiu: Formal analysis, Investigation, Resources. Jinhua Wu: Formal analysis, Investigation, Resources. Xinzeng Wei: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Review & editing. Mingxi Jiang: Conceptualization, Methodology, Project administration, Funding acquisition.

Data availability statement

The data presented in this study are included within the article. Additional datasets used for visualizations are available from the corresponding author upon reasonable request.

Declaration of competing interest

The authors of this study declare that they have no conflict interest to report.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2025.09.001.