Going with the flow: analysis of population structure reveals high gene flow shaping invasion pattern and inducing range expansion of Mikania micrantha in Asia

Abstract

Background and Aims

Mikania micrantha, a climbing perennial weed of the family Asteraceae, is native to Latin America and is highly invasive in the tropical belt of Asia, Oceania and Australia. This study was framed to investigate the population structure of M. micrantha at a large spatial scale in Asia and to identify how introduction history, evolutionary forces and landscape features influenced the genetic pattern of the species in this region.

Methods

We assessed the genetic diversity and structure of 1052 individuals from 46 populations for 12 microsatellite loci. The spatial pattern of genetic variation was investigated by estimating the relationship between genetic distance and geographical, climatic and landscape resistances hypothesized to influence gene flow between populations.

Key Results

We found high genetic diversity of M. micrantha in this region, as compared with the genetic diversity parameters of other invasive species. Spatial and non-spatial clustering algorithms identified the presence of multiple genetic clusters and admixture between populations. Most of the populations showed heterozygote deficiency, primarily due to inbreeding, and the founder populations showed evidence of a genetic bottleneck. Persistent gene flow throughout the invasive range caused low genetic differentiation among populations and provided beneficial genetic variation to the marginal populations in a heterogeneous environment. Environmental suitability was found to buffer the detrimental effects of inbreeding at the leading edge of range expansion. Both linear and non-linear regression models demonstrated a weak relationship between genetic distance and geographical distance, as well as bioclimatic variables and environmental resistance surfaces.

Conclusions

These findings provide evidence that extensive gene flow and admixture between populations have influenced the current genetic pattern of M. micrantha in this region. High gene flow across the invaded landscape may facilitate adaptation, establishment and long-term persistence of the population, thereby indicating the range expansion ability of the species.

INTRODUCTION

Biological invasion has become a severe problem due to the negative impacts of invasive alien species (IAS) on regional biota (Early et al., 2016; Iacarella et al., 2015). Among multiple hypotheses that have been proposed to explain invasion success, the influence of genetic diversity has been long recognized (Baker and Stebbins, 1965). Standing genetic variation has been found to result in rapid selection of IAS populations showing greater fitness, leading to successful establishment and range expansion in the introduced range (Barrett, 2015). On the other hand, introduction with few founder individuals often causes a reduction in genetic diversity, which may subsequently result in inbreeding and loss of fitness of the IAS during establishment and range expansion (Schrieber and Lachmuth, 2017). However, species with low genetic diversity have often been found to be successful invaders (Hagenblad et al., 2015).

This genetic paradox of invasion has been often explained by several mechanisms acting alone or in combination (Banerjee et al., 2019a). For example, multiple introductions have been found to increase the available gene pool when different lineages colonize geographically proximate locations, and subsequent gene flow across populations leads to increased genetic diversity for successful invasion (Roman and Darling, 2007; Kirk and Freeland, 2011). However, several geographical and environmental variables may influence gene flow among populations (Sork et al., 1999). The genetic pattern of a species across a landscape can be influenced if gene flow is impeded by geographical distance (IBD or isolation by distance; Wright, 1943), physical barriers (IBR or isolation by resistance; McRae, 2006) or and environmental dissimilarity resulting in local adaptation (IBE or isolation by environment; Wang and Bradburd, 2014). Recent studies have found that the negative effects of genetic depletion on population fitness may also be alleviated through advantageous inbreeding × environment interactions during early phases of invasions and towards the leading edges of expansion (Schrieber and Lachmuth, 2017).

Understanding the relative importance of these factors in a successful invasion is one of the central themes of invasive species research. The ability of IAS to successfully establish across a range of environmental conditions, despite being a subset of the species-wide gene pool and often undergoing loss of genetic diversity during introduction, provides a useful study system to investigate various evolutionary processes (Prentis et al., 2008). Molecular population genetics study can be instrumental in exploring the genetic components of invasiveness shaping the distribution of IAS in the introduced range.

Mikania micrantha, a climbing perennial weed of the family Asteraceae, is an invasive plant in its introduced range of Asia, Oceania and Australia (Ellison and Sankaran, 2017). Native to Mexico, Central America, the Caribbean and tropical South America (Ellison et al., 2008), M. micrantha is considered one of the hundred worst IAS (Lowe et al., 2000) due to its extensive economic and ecological impacts on natural forests, plantations and agricultural systems across its invasive range (Tripathi et al., 2012). The species reproduces through seeds and by rooting at stem nodes. Mikania micrantha is reported to be self-incompatible and thus requires insect pollination (Hong et al., 2007). The lightweight seeds can be dispersed long distance by wind whereas anthropogenic activities facilitate dispersal of stem fragments to new areas. High seed germination (Yang et al., 2005) and regeneration capacity from clonal fragments (Li et al., 2013) ensure the successful establishment of M. micrantha in newly invaded areas.

Previous molecular studies revealed high genetic diversity in part of its introduced range in southern China by using microsatellite (Geng et al., 2017), chloroplast simple sequence repeats (Wang et al., 2016) and intersimple sequence repeat (Wang et al., 2008) markers. However, for a comprehensive understanding of the role of genetic forces in invasion, studies at large spatial scale are also necessary, especially as the relative importance of factors driving genetic structure (e.g. gene flow, environmental heterogeneity) could vary along the spatial scale (Ward, 2006; Anderson et al., 2010). The present study was therefore framed to investigate the population structure of M. micrantha across a large area of its invasive range, particularly focusing on Asia, where multiple introductions and range expansion of the species have been documented (Day et al., 2016; Banerjee et al., 2019b). We expected the genetic pattern of the species to follow the ‘abundant centre hypothesis’ (Sagarin and Gaines, 2002; Guo, 2014), in which central populations near the introduction sites would harbour greater diversity than the populations at the leading edge. Given the long-distance dispersal ability of the species through human-mediated transfer of broken fragments and lightweight seeds, we also expected that genetic exchange between populations would influence the current genetic pattern of M. micrantha in this region. Based on the wide distribution of the species outside its native range, we hypothesized that genetic variation of M. micrantha would be influenced by the geography and environment of its invasive range.

Specifically, we aimed to assess the population genetic structure of M. micrantha and to investigate the influence of (1) multiple introductions, (2) gene flow between populations and (3) landscape features, on the observed large-scale genetic pattern of this species. We used nuclear simple sequence repeats (SSRs) to characterize the molecular genetics of populations sampled across a latitudinal gradient from its invasive range. The co-dominant and multi-allelic SSR markers have been found to be useful in analysing genetic diversity and gene flow and in evolutionary studies for several wild species (Mason, 2015). Although whole genome-scale data are increasingly used in molecular characterization and a reference genome of M. micrantha has recently become available (Liu et al., 2020), SSR markers were used in this study due to their abundance in the genome, high levels of polymorphism, higher mutation rate and cost-effectiveness to analyse genetic patterns involving large numbers of samples (Vieira et al., 2016). We also discussed how the findings of this study could be used to frame management strategies for this species.

MATERIALS AND METHODS

Sample collection, DNA extraction and SSR amplification

A total of 1052 samples of M. micrantha Kunth were collected from 46 populations across its invasive range in Asia (Fig. 1; Supplementary Data Table S1). The populations were separated by more than 30 km. For each population, location data were recorded using a Garmin GPS navigator. We collected eighth nodal leaves from five to ten healthy individuals, separated by at least 10 m, to ensure plant maturity and uniformity of developmental stages across all the populations. The leaf tissues were dried by silica gel and preserved in zip-lock plastic pouches at −20 °C until further use. Voucher specimens for each population were deposited in the herbarium of Sun Yat-sen University (SYS).

Fig. 1.

Map of the study area showing locations of the 46 populations of Mikania micrantha. The populations were clustered into four groups based on estimated posterior membership probability implemented in GENELAND. Abbreviations of populations are given in Supplementary Data Table S1.

Total genomic DNA was extracted from dried leaf samples using a HiPure SF Plant DNA Mini Kit (Magen Technology Inc., Guangzhou, China) following the manufacturer’s protocol. The samples were genotyped at 12 fluorescently labelled nuclear microsatellite markers (SSRs) which demonstrated good quality in M. micrantha (Hong et al., 2008; Yan et al., 2011) (Supplementary Data Table S2). Optimization of the polymerase chain reaction (PCR) conditions was conducted using two randomly selected individuals from each population. Final amplifications were carried out in a 20-µL reaction mix containing 2 µL of cDNA, 2 µL of Easy-Taq buffer, 0.2 µL of Easy-Taq enzyme (TransGen Biotech, Beijing, China), 1.6 µL of dNTP, 1 µL of each primer and 12.2 µL of ddH₂O. The optimized thermal profile consisted of a 4-min initial denaturation step at 94 °C followed by amplification for 31 cycles of denaturing for 40 s at 94 °C, annealing for 40 s at 53 °C and extension for 1 min at 72 °C, and a final cycle at 72 °C for 10 min. Amplification products were first inspected in 1 % agarose gel with ethidium bromide and photographed under UV light. The amplified products were finally run on an ABI 3130xl automated sequencer with the GeneScan-500 LIZ size standard and analysed using GENEMAKER version 2.2.0 (Softgenetics, LLC., State College, PA, USA) and the peak value was determined according to the length of each combined fragment.

Analysis of genetic diversity

We used GENALEX version 6.5 (Peakall and Smouse, 2012) to estimate the parameters of genetic variation for each microsatellite locus and population: total number of alleles (N_a), the effective number of alleles (N_e), mean expected heterozygosity (H_e), mean observed heterozygosity (H_o) and the Shannon information index (I). FSTAT version 2.9.3.2 (Goudet, 2002) was used to estimate the allelic richness (A_R) and the unbiased estimates of Wright’s F-statistics: overall population inbreeding coefficient (F_it), interpopulation genetic differentiation coefficient (F_st) and intrapopulation inbreeding coefficient (F_is). Gene flow between populations (N_m) was calculated from the pairwise F_st values using the formula: N_m = (1 − F_st)/4F_st (Wright, 1969). Because the presence of null alleles may impede the proper inference of population genetic structure, we checked the maximum-likelihood estimates of the null allele frequencies for each locus × population combination using the expectation maximization algorithm (Dempster et al., 1977) as implemented in FREENA (Chapuis and Estoup, 2007). The F_is was further estimated using the Bayesian approach (IIM) in INEST version 2.2 (Chybicki and Burczyk, 2009). We used 300 000 cycles, sampled every 100 steps and discarded the first 30 000 steps as burn-in. To infer the statistical significance of inbreeding, we compared the Deviance Information Criterion (DIC) between the full model (nfb, where n: null alleles, f: inbreeding coefficient, and b: genotyping failures), the model including null alleles and inbreeding coefficient (nf), and the model having null alleles and genotyping failures (nb). Deviation from Hardy–Weinberg equilibrium (HWE) at the population level was tested using GENEPOP version 4.7.2 (Rousset, 2008). To test the significance of heterozygote excess and deficiency for each population (at P = 0.05), we used the score test as implemented in GENEPOP. For this analysis, a Markov chain algorithm was employed with 10 000 dememorization steps and the algorithm was run for 20 batches with 10 000 iterations per batch.

Analysis of genetic structure

The Bayesian program STRUCTURE version 2.3.4 (Evanno et al., 2005) was used to investigate the population structure without a priori information of sampling locations and to group the individuals under subpopulations. We considered admixture models with putative numbers of populations (K) ranging from 1 to 10. For each K, we iterated the models ten times with 100 000 burn-in periods followed by 100 000 Markov chain Monte Carlo (MCMC) replicates. STRUCTURE HARVESTER (Earl and vonHoldt, 2012) was used to determine the optimum K. The results of the replicates at the optimum K were further processed using CLUMPP 1.1.2 (Jakobsson and Rosenberg, 2007) to assess the membership proportions for clusters identified in STRUCTURE. We used BAPS version 6.0 (Corander et al., 2008) to analyse the population structure further at the level ‘group of individuals’ for the non-spatial model with optimum K-values obtained from STRUCTURE analysis.

To integrate the spatial information in determining the genetic clusters, we used a spatially explicit Bayesian clustering algorithm in the program GENELAND version 4.0.7 (Guillot et al., 2008) in R. For each of the 46 populations, the MCMC simulations were run for 100 000 iterations under the spatial model with the assumption of uncorrelated allele frequencies between samples and each of every 1000^th iteration was saved. These MCMC outputs were post-processed with a horizontal and vertical discretization of the study domain in 100 pixels and a burn-in of 500 saved iterations to visualize the posterior distribution and membership probability of these populations.

Finally, the 46 populations were grouped into regions based on the clusters identified by non-spatial and spatial algorithms. ARLEQUIN version 3.1 was used for analysis of molecular variance (AMOVA; Excoffier et al., 1992) in which genetic differentiation was estimated at three levels: among regions, among populations and among populations within regions. Based on the genetic distance (F_st), the genetic relationship between populations was evaluated through a principal coordinate analysis (PCoA) using GENALEX version. 6.5. Nei’s genetic distance among populations was further estimated using POPULATIONS version 1.2 (Langella, 2000), and was used to construct an unrooted neighbour-joining (NJ) tree with 1000 bootstrap replicates.

Identifying determinants of genetic pattern

We tested the effects of four predictor variables that may influence genetic differentiation in M. micrantha in its invaded range:geographical distance (IBD), bioclimatic variables as a proxy for environmental distance (IBE) and two potential resistance surfaces (IBR). To detect patterns of IBD among all populations, the relationship between pairwise genetic distances [F_st/(1 − F_st)] and geographical distances (log-transformed values) was examined using Mantel tests (Mantel, 1967), as implemented in PASSAGE version 2 (Rosenberg and Anderson, 2011), with 10 000 random permutations. In addition, to identify biogeographical boundaries exhibiting the largest genetic discontinuities between population pairs, we used Monmonier’s maximum difference algorithm on the pairwise F_st matrix, as implemented in BARRIER version 2.2 (Manni et al., 2004).

For environmental factors (IBE), a principal components (PC) analysis was performed on 19 standardized climatic variables, downloaded from the WorldClim database version 1.4 (http://www.worldclim.org/) (Hijmans et al., 2005) at a spatial resolution of 5 arc minutes, using the package FactoMineR (Lê, 2008) in R, and environmental values for 46 populations were estimated along the first three axes explaining maximum variance. The Euclidean distance was calculated between populations with these values using the ecodist package (Goslee and Urban, 2007) in R. We further estimated climate analogy of our sampling locations with the native range of the species. We used multivariate environmental similarity surface (MESS) analysis (Elith et al., 2010) implemented in the dismo package version 1.1-4 (Hijmans et al., 2017) in R, and compared the environment of grid cells occupied by the 46 populations with that of the entire native range. Grid cells having a positive value would indicate a similar environment between two ranges whereas grid cells with a dissimilar environment for at least one variable would receive negative values.

We used altitude and anthropogenic activities as two resistant surfaces (IBR). A raster of altitude data (30-second-resolution grid) was obtained from https://research.cip.cgiar.org/gis. To characterize the anthropogenic activities, we used global human footprint data version 2.0 (Wildlife Conservation Society - WCS, Center for International Earth Science Information Network - CIESIN - Columbia University, 2005). The nearest-neighbour resampling approach was adopted to ensure uniform resolution for both the rasters, and the global files were clipped with the extent of our study area and converted to ascii format. These operations were conducted in ArcMap 10.2. Finally, we used CIRCUITSCAPE 4.0 (McRae, 2006) to estimate pairwise resistance distances between all pairs of 46 populations for altitude (resistance) and anthropogenic activities (conductance) (Supplementary Data Fig. S1).

Both linear and non-linear regression techniques were used to analyse the relationship of genetic differentiation [F_st/(1 − F_st)] with geographical, environmental and resistance distances. The genetic distance matrix was used as the dependent variable and nine independent univariate, bivariate and multivariate models were generated with different sets of predictor variables. For linear regression analysis, we used the MRM (multiple regression on distance matrices) function implemented in the ecodist package in R. Each model was run 10 000 times and the regression coefficient and significance of each predictor variable were estimated. For non-linear regression analysis, we used generalized dissimilarity modelling (GDM; Ferrier et al., 2007) implemented in the gdm package (Manion, 2018) in R. The gdm.varImp function was used with 500 permutations to estimate the variable importance in the individual models. We used a backward elimination approach in which the least important variable was removed per run and the importance of the remaining variable(s) was reassessed.

We used BOTTLENECK version 1.2.02 (Piry et al., 1999) to estimate expected heterozygosity at mutation-drift equilibrium under the stepwise mutation model (SMM) and the infinite allele model (IAM). The significance of heterozygosity excess was evaluated using the two-tailed Wilcoxon test and Sign test. To determine the relative influence of mutation and genetic drift on genetic differentiation among populations, we used the allele size permutation test using SPAGEDI version 1.5 (Hardy and Vekemans, 2002). The null hypothesis of R_st = F_st (genetic differentiation under SMM = genetic differentiation under IAM) was tested by randomly permuting alleles 5000 times, which provided a 95 % confidence interval of the simulated distribution of R_st values.

RESULTS

Genetic diversity

Analysis of genetic variation across 46 natural populations revealed the presence of 84 alleles from 12 SSR loci ranging from four to 16 alleles per locus (Table 1). Most of the alleles (n = 55) were shared among populations, and 29 private alleles were found distributed among 15 populations. One population from China (GX) was found to contain the most loci (n = 6) with private alleles. Average allelic richness (A_R), observed number of alleles (N_a) and effective number of alleles (N_e) ranged from 1.85 to 3.17, 2.17 to 7, and 1.62 to 3.4 respectively. The average observed and expected frequency of heterozygotes ranged from 0.2 to 0.6 and 0.34 to 0.67 respectively. One population from Japan (JPA) showed the lowest genetic diversity (H_o = 0.3; H_e = 0.34) whereas populations from China (GX, SZ, DG), Hong Kong and Macao (HKD, MCB), and India (NED, KEC) showed high genetic diversity (H_o and H_e ≥ 0.5).

Table 1.

Genetic variability for the 12 SSR markers within 46 populations of Mikania micrantha

Country/region	Population	N	A R	N a	N e	N p	H o	H e	F is	I
Mainland China	YN	35.42	2.30	4.33	2.04	2	0.37	0.46	0.196	0.87
	GH	52.58	2.63	5.50	2.58	1	0.48	0.57	0.169	1.08
	HN	37.83	2.52	3.92	2.48	0	0.43	0.56	0.235	0.99
	GX	38.42	3.17	7.00	3.40	5	0.42	0.67	0.390	1.40
	ZH	10.92	2.60	3.33	2.50	1	0.43	0.55	0.269	0.96
	SZ	32.83	2.99	5.58	3.08	1	0.43	0.65	0.350	1.27
	DG	15.92	2.33	2.92	2.27	0	0.50	0.52	0.083	0.86
Hong Kong	HKA	12.00	2.02	2.25	1.91	0	0.54	0.42	−0.240	0.66
	HKB	12.00	2.08	2.41	1.89	0	0.54	0.44	−0.174	0.69
	HKC	12.75	2.41	3.33	2.17	0	0.48	0.52	0.117	0.89
	HKD	13.92	2.42	3.08	2.32	0	0.60	0.55	−0.048	0.91
Macao	MCA	9.00	2.24	2.83	1.97	0	0.51	0.48	0.005	0.79
	MCB	11.00	2.41	3.17	2.23	0	0.52	0.51	0.029	0.87
Taiwan	PD	34.33	2.60	5.17	2.41	1	0.30	0.52	0.426	1.05
	ML	12.00	1.93	2.50	1.67	0	0.36	0.36	0.028	0.59
	TW	23.08	2.13	3.50	1.99	1	0.38	0.41	0.098	0.74
	JY	11.42	2.42	3.25	2.27	1	0.45	0.48	0.127	0.86
	TY	10.83	2.29	3.08	2.13	0	0.51	0.46	−0.066	0.80
India	KEA	24.17	2.65	4.83	2.56	3	0.46	0.56	0.186	1.06
South-west	KEB	52.00	2.78	5.75	2.96	4	0.46	0.59	0.233	1.75
	KEC	18.83	2.63	3.92	2.59	1	0.51	0.56	0.115	1.02
	KED	13.58	2.27	3.08	2.13	0	0.47	0.47	0.030	0.80
	KEE	15.83	1.89	2.83	1.66	0	0.30	0.34	0.155	0.59
North-east	NBA	26.33	2.44	3.58	2.37	0	0.48	0.53	0.125	0.93
	NBB	38.83	2.77	5.92	2.69	3	0.45	0.59	0.245	1.17
	NEA	10.25	2.82	3.75	2.77	0	0.40	0.60	0.388	1.08
	NEB	43.67	2.39	4.33	2.19	0	0.47	0.52	0.113	0.93
	NEC	71.50	2.54	5.67	2.39	3	0.44	0.56	0.218	1.05
	NED	23.58	2.40	3.33	2.30	0	0.50	0.53	0.063	0.90
East	KOL	32.83	2.63	5.33	2.51	3	0.48	0.57	0.176	1.08
Bangladesh	BL	18.08	2.41	3.92	2.21	1	0.46	0.49	0.083	0.90
Vanuatu	VA	20.00	2.12	2.75	2.01	1	0.49	0.44	−0.106	0.73
Sri Lanka	SL	18.00	2.34	3.25	2.27	0	0.51	0.50	0.017	0.86
Malaysia	MYA	12.33	2.43	3.42	2.39	0	0.51	0.49	−0.005	0.88
	MYB	4.50	2.17	2.33	2.05	0	0.43	0.42	0.113	0.68
	MYC	10.92	1.91	2.42	1.72	0	0.42	0.34	−0.175	0.57
	MYD	7.92	1.85	2.17	1.74	0	0.43	0.34	−0.208	0.54
Indonesia	INA	21.83	2.43	3.83	2.28	2	0.45	0.52	0.159	0.93
	INB	71.75	2.36	3.25	2.18	2	0.40	0.51	0.248	0.87
	INC	18.83	1.99	2.50	1.89	0	0.43	0.40	−0.056	0.65
Philippines	PL	17.42	2.47	4.00	2.28	0	0.47	0.52	0.117	0.94
Singapore	SG	8.67	2.43	2.92	2.26	0	0.54	0.53	0.043	0.88
Thailand	TH	21.92	2.09	3.17	1.92	0	0.38	0.39	0.071	0.72
Japan	JPA	9.33	1.86	2.17	1.62	0	0.20	0.34	0.456	0.55
	JPB	9.67	2.02	2.33	1.83	0	0.23	0.39	0.476	0.64
	JPC	11.42	2.04	2.50	1.85	0	0.24	0.40	0.456	0.66
	Mean ±SE	22.61±2.35	2.36±0.04	3.62±0.17	2.24±0.06	0.78±0.18	0.44±0.01	0.49±0.01	0.13±0.03	0.88±0.04

N: mean number of alleles; A_R: allelic richness; N_a: mean observed number of alleles; N_e: effective number of alleles; H_o: observed frequency of heterozygotes; H_e: expected frequency of heterozygotes; F_is: coefficient of inbreeding; I: Shannon index; N_p: number of private alleles. Bold numbers indicate significant heterozygote deficiency at P < 0.05.

For each locus, the observed heterozygosity (H_o) and expected heterozygosity (H_e) values ranged from 0.25 to 0.73 and 0.24 to 0.64 respectively (Supplementary Data Table S3). All loci, except locus 6, were found to contain private alleles. The Bayesian analyses from INEST revealed higher DIC values for the nb model (52 973.9) in comparison to that obtained for the nf (52 645.6) and nfb (52 539.3) models. The F-statistics analysis across populations over all loci revealed that inbreeding coefficient values at the intrapopulation level (F_is) ranged from −0.206 to 0.381, at the total population level (F_it) ranged from −0.068 to 0.511 and at the interpopulation level (F_st) ranged from 0.135 to 0.274 (Table 2). Gene flow (N_m) between populations varied from 0.661 to 1.598. The distribution of null alleles calculated for each locus × population combination revealed a median of 0.037 with 25 % and 75 % quartiles of zero and 0.132 respectively. The global F_st values before and after correction for null alleles were 0.139 and 0.134 respectively (Supplementary Data Table S3). After correction for null alleles, global gene flow increased marginally (N_m = 1.62) over that calculated with null alleles (N_m = 1.55).

Table 2.

Analysis of molecular variance (AMOVA) for 46 populations of Mikania micrantha under spatial and non-spatial models

Model	Levels of variance	d.f.	Sum of squares	Variance	% of variation	F-statistics
Non-spatial (BAPS)	When K = 3
	Among groups	2	139.55	0.088	6.73	F st = 0.173
	Among populations within groups	43	310.45	0.138	10.57	F sc = 0.113
	Within populations	2058	2221.34	1.079	82.70	F ct = 0.067
	When K = 8
	Among groups	7	281.45	0.145	11.11	F st = 0.170
	Among populations within groups	38	168.55	0.077	5.90	F sc = 0.066
	Within populations	2058	2221.34	1.079	82.99	F ct = 0.111
Spatial (GENELAND)	When K = 4
	Among groups	3	176.48	0.103	7.90	F st = 0.173
	Among populations within groups	42	273.53	0.122	9.36	F sc = 0.102
	Within populations	1997	2221.34	1.079	82.74	F ct = 0.079

Significance (1023 permutations): all significant (P < 0.001) between Vc (among group component of variance) and F_st, Vb (among population (within group) component of variance) and F_sc, and Va (within population component of variance) and F_ct

Genetic structure

STRUCTURE analysis revealed that ΔK was highest at K = 3 and K = 8 (Fig. 2) and BAPS analysis also revealed a pattern (Supplementary Data Fig. S2) consistent with STRUCTURE. When K = 3, the populations from China (except YN, GX and SZ), Hong Kong and Macao, Japan, Vanuatu and one population from Malaysia (MYC) formed one cluster. A high level of admixture was detected in the Indian populations. Roughly, the north-east and east Indian populations could be clustered with populations from Bangladesh and the Philippines (cluster 2) whereas the south-west Indian populations were clustered with the Southeast Asian (Malaysia, Indonesia, Singapore, Thailand) and Taiwan populations (cluster 3). When K = 8, the southern China populations formed one cluster with Hong Kong and Macao (cluster 1), consistent with the pattern observed with K = 3, except that the populations from Japan and Vanuatu formed a separate gene pool with the Philippines population (cluster 2). The north-east and east Indian populations formed a cluster with two populations from China (GX and SZ) (cluster 3) whereas three south-west Indian populations were found to belong to a separate cluster (cluster 4). Three populations from Taiwan (TW, JY and TY) showed similarity with the populations from Malaysia, Singapore and Sri Lanka (cluster 5). The Indonesia and Thailand populations formed a separate gene pool with two other Taiwan populations (cluster 6). One population from China (YN) formed a group with the Bangladesh population (cluster 7) whereas the final cluster (cluster 8) was formed with one south-west India population.

Fig. 2.

Population structure of 1052 individuals of Mikania micrantha in the study area based on 12 microsatellite loci using the assignment result inferred from STRUCTURE analysis (K = 3 and K = 8). A single vertical bar represents the membership coefficient of each individual in K clusters, with population locations labelled. Abbreviations of populations are given in Supplementary Data Table S1.

When integrating spatial information in the clustering algorithm implemented in GENELAND, the genetic structure was not identical to that obtained from non-spatial algorithms as the 46 populations were grouped under four genetic groups (Fig. 1; Supplementary Data Fig. S3A). Populations from China (except YN and GX), Hong Kong and Macao formed one genetic group. The YN population belonged to a separate cluster with Japan, Vanuatu, Bangladesh and Philippine populations (cluster 2) whereas the GX population belonged to the cluster formed by the north-east, east and south-west Indian populations (cluster 3). The Southeast Asian (Malaysia, Indonesia, Singapore, Thailand) populations formed a cluster with populations from Taiwan and Sri Lanka (cluster 4) (Fig. S3B).

The first three axes of the PCoA accounted for only 55.68 % variation and no definite genetic clustering was observed along these three axes (Fig. 3A). Along the x-axis, the populations clustered roughly into two groups in which populations from Southeast Asian countries and Taiwan were separated from the rest of the populations, this being partially consistent with the spatial and non-spatial genetic clustering algorithms. The NJ tree was nearly consistent with the findings of the spatial genetic clustering (Fig. 3B). The southern China populations (except YN, GX and SZ) formed a branch with Hong Kong and Macao populations. The YN population belonged to the branch formed with Japan, Vanuatu and Bangladesh populations whereas the GX and SZ populations formed a cluster with north-east and east Indian populations. The south-west India populations formed a separate branch, which showed close proximity to the branch formed by the Southeast Asian and Taiwan populations.

Fig. 3.

Relationships among 46 populations of Mikania micrantha as inferred from genetic distance. (A) Principal coordinate analysis based on F_st in which the first, second and third principal coordinates account for 29.55 %, 13.71 % and 12.42 % of the variation respectively. (B) Unrooted neighbour-joining tree based on Nei’s genetic distance. Abbreviations of populations are given in Supplementary Data Table S1.

We observed high gene flow (>75 % quartile N_m) between populations of different clusters such as southern China and north-east India, whereas low gene flow (< median N_m) was observed between populations of southern China and Taiwan, Indonesia and Malaysia (Fig. 4). Pairwise analysis of genetic differentiation showed significant F_st values between all except eight population pairs (at P = 0.05), showing low genetic differentiation among populations (Supplementary Data Table S4). AMOVAs based on the variable number of clusters (K = 3 and 8) identified by spatial and non-spatial clustering algorithms also revealed that genetic differentiation within populations accounted for > 80 % of the variation (82.7 % when K = 3 and 82.99 % when K = 8 under the non-spatial model; and 82.74 % when K = 4 under the spatial model) (Table 2).

Fig. 4.

Pairwise population matrix of gene flow between 46 populations of Mikania micrantha. The colour bars represent the distribution of gene flow values across five classes centred around the median. Abbreviations of populations are given in Supplementary Data Table S1.

Determinants of genetic structure

The Mantel test revealed a significant positive correlation between geographical and genetic distances (r = 0.2, P = 0.01; Supplementary Data Fig. S4). Monmonier’s algorithm identified those geographical barriers that could potentially disrupt gene flow between populations of Malaysia and Indonesia (barrier 1), and Taiwan and southern China (barrier 2), and isolate the populations of TH and KEE (Fig. S5). When considering the bioclimatic variables, the PCA revealed that the first three axes could explain 81.2 % of the variation (Table S5). Bio6 (Min temperature of the coldest month) was the highest contributory variable for PC1, whereas Bio5 (Max temperature of the warmest month) and Bio13 (Precipitation of wettest month) were found to contribute maximum to PC2 and PC3 respectively. The analysis of climate analogy between native and invasive ranges revealed that all populations received positive MESS values ranging from 0.69 (in population KOL of east India) to 48.36 (in population HN of southern China), and high climate analogy (MESS > 75 % quartile) was detected for 12 populations (Table S6).

Using genetic distance as the response variable, linear regression analysis using the MRM function revealed significant regression coefficients (r²) for all nine models (Table 3). The full model (model 1) with all predictor variables showed the maximum r² value (r² = 0.119) followed by model 7 (geographical distance and anthropogenic influence; r² = 0.109). Models computed only with geographical (model 2) and environmental (model 3) variables had the minimum r² values. Across the three models in which geographical distance was considered as one of the predictor variables, a significant effect of geographical distance on genetic differentiation was found (Table 3). None of the three bioclimatic variables had a significant association with genetic differentiation. The non-linear regression explained a maximum of 28.45 % of the deviance in the genetic distance among populations (Table 3). Like the regression models, the full GDM model explained the maximum variation followed by models with resistance surfaces as predictor variables (28.38 %). Backward elimination of the non-significant variables from the individual model revealed that altitude was the only significant contributory variable (at P = 0.05) across all models that considered the resistance surfaces as explanatory variables.

Table 3.

Results from multiple regression on distance matrices (MLM) and generalized dissimilarity modelling (GDM) analyses demonstrating the proportion of genomic variation explained by geographical distance, environmental distance (bioclimatic variables) and two resistance surfaces

Models	Linear (MRM)		Non-linear (GDM)
	R 2	Significant variable (coefficient)	Deviance	Percentage of variance explained	Important variable (importance)
1 (Full model: GEO + ENV + ALT + ANT)	0.119	GEO (0.176)	72.73	28.45	PC1 (21.62)* ALT (82.67)*
2 (GEO)	0.042	GEO (0.204)	98.05	3.53	GEO
3 (ENV)	0.059	NA	96.96	4.61	PC1 (70.42) PC3 (29.91)
4 (ALT + ANT)	0.156	NA
5 (GEO + ENV)	0.046	GEO (0.206)	92.89	8.61	PC1 (48.30) GEO (53.21)*
6 (GEO + ALT)	0.098	GEO (0.169) ALT (−0.269)	80.33	20.97	ALT (33.27)* ANT (2.24)
7 (GEO + ANT)	0.109	GEO (0.175) ANT (−0.309)
8 (ENV + ALT)	0.076	ALT (−0.305)	72.79	28.38	PC1 (21.62) ALT (87.67)*
9 (ENV + ANT)	0.087	ANT (−0.343)

*Variable significance at P < 0.05.

GEO: geographic distance; ENV: three PC axes of the bioclimatic variables – PC1 (Minimum temperature of the coldest month), PC2 (Maximum temperature of the warmest month), PC3 (Mean diurnal range); ALT (altitude), ANT (human global footprint).

The Wilcoxon test and sign test indicated that seven populations appeared to have a bottleneck with the SMM model, whereas bottlenecks were detected for 17 populations under the IAM model (Table 4). A mode shift in allele frequency distribution was observed for 14 populations. The global F_st (0.141) was higher than R_st (0.125), and fell within the 95 % confidence interval of the R_st (0.089–0.155) values, suggesting an influence of mutation on genetic differentiation.

Table 4.

Results of bottleneck analyses for each of the 46 populations of Mikania micrantha

Country	Population	IAM		SMM		Mode shift
		Sign test	Wilcoxon test (two-tailed)	Sign test	Wilcoxon test (two-tailed)
Mainland China	YN	0.518	0.970	0.005	0.034	L-shaped
	GH	0.059	0.110	0.017	0.006	L-shaped
	HN	0.004	0.001	0.309	0.301	L-shaped
	GX	0.071	0.006	0.004	0.001	L-shaped
	ZH	0.035	0.008	0.153	0.151	Shifted
	SZ	0.175	0.013	0.059	0.266	L-shaped
	DG	0.019	0.003	0.122	0.204	Shifted
Hong Kong	HKA	0.005	0.005	0.008	0.054	Shifted
	HKB	0.029	0.002	0.057	0.102	Shifted
	HKC	0.044	0.034	0.185	0.791	L-shaped
	HKD	0.001	0.000	0.061	0.013	L-shaped
Macao	MCA	0.030	0.110	0.124	0.266	L-shaped
	MCB	0.259	0.052	0.351	0.910	L-shaped
Taiwan	PD	0.220	0.970	0.003	0.003	L-shaped
	ML	0.335	0.470	0.562	0.569	L-shaped
	TW	0.532	0.910	0.025	0.110	L-shaped
	JY	0.207	0.067	0.460	0.898	Shifted
	TY	0.375	0.147	0.470	0.765	L-shaped
India	KEA	0.532	0.151	0.018	0.064	L-shaped
	KEB	0.168	0.233	0.166	0.129	L-shaped
	KEC	0.007	0.034	0.387	0.622	Shifted
	KED	0.048	0.009	0.525	0.700	L-shaped
	KEE	0.401	0.850	0.274	0.110	L-shaped
	NBA	0.005	0.005	0.408	0.850	L-shaped
	NBB	0.379	0.380	0.018	0.008	L-shaped
	NEA	0.008	0.005	0.346	0.266	Shifted
	NEB	0.118	0.129	0.080	0.266	L-shaped
	NEC	0.154	0.052	0.016	0.003	L-shaped
	NED	0.004	0.000	0.452	0.380	Shifted
	KOL	0.178	0.339	0.003	0.005	L-shaped
Bangladesh	BL	0.424	0.413	0.122	0.147	L-shaped
Vanuatu	VA	0.031	0.005	0.415	0.206	L-shaped
Sri Lanka	SL	0.065	0.007	0.508	0.520	L-shaped
Malaysia	MYA	0.149	0.021	0.339	0.898	L-shaped
	MYB	0.035	0.007	0.530	0.232	Shifted
	MYC	0.360	0.193	0.506	1.000	L-shaped
	MYD	0.326	0.250	0.314	0.652	Shifted
Indonesia	INA	0.313	0.077	0.076	0.233	L-shaped
	INB	0.111	0.005	0.598	0.791	L-shaped
	INC	0.025	0.009	0.361	0.320	L-shaped
Philippines	PL	0.365	0.519	0.059	0.233	L-shaped
Singapore	SG	0.008	0.001	0.153	0.042	Shifted
Thailand	TH	0.404	0.301	0.251	0.380	L-shaped
Japan	JPA	0.096	0.206	0.599	0.831	Shifted
	JPB	0.048	0.010	0.315	0.322	Shifted
	JPC	0.030	0.067	0.201	0.278	Shifted

IAM: infinite allele model of mutation; SMM: somatic model of mutation.

Bold entries indicate significance (P < 0.05) using both Sign and Wilcoxon tests.

DISCUSSION

In this study, we have found high genetic diversity of M. micrantha across a large part of its invasive range. High gene flow between populations was found to alleviate the impacts of bottleneck and inbreeding on population fitness, leading to range expansion of the species in this region.

High genetic diversity of M. micrantha

Genetic diversity in the introduced range has been found as an important factor for invasion success of many invasive species, such as Lantana camara (Ray and Quader, 2014) and Ambrosia artemisiifolia (Meyer et al., 2017). Comparing the genetic diversity parameters with other invasive species, our study found higher expected and observed heterozygosity in M. micrantha, although with lower allelic richness (Supplementary Data Table S7). These estimates are comparable to those from regional studies of M. micrantha in its invasive range (Geng et al., 2016), and higher than those reported from its native range (A_R = 1.52–1.77; H_e = 0.175–0.342) (Bravo-Monzón et al., 2018). However, drawing a comparative inference of genetic diversity between native and invasive ranges of this species is difficult because the sampled populations were collected from the northernmost limit of its native range that may not represent the source locations of our studied populations.

Most of the populations (33 out of 46) showed heterozygote deficiency, which may be due to the Wahlund effect, the presence of null alleles or inbreeding (Chun et al., 2010). In this case, the Wahlund effect is unlikely to be a cause of heterozygote deficiency because our genetic structure analysis showed no indication of clear subpopulation structure. We also checked for deviation from HWE after correction for null alleles and noted the same observation. Bayesian analysis using the DIC indicate that inbreeding may have contributed to the observed heterozygosity deficiency. Because M. micrantha is a self-incompatible species (Hong et al., 2007) and local spread occurs primarily through vegetative reproduction (Tripathi et al., 2012), inbreeding appears to be the most likely explanation for the observed heterozygote deficiency. We observed significant linkage disequilibrium for a majority of the populations (37 out of 46), suggesting that asexual reproduction is the predominant mating system in the studied populations.

In addition to heterozygote deficiency, a large number of populations (24 out of 46) also showed evidence of genetic bottlenecks. In the case of intentional introduction as with M. micrantha, founder populations are usually small because a subset of native populations is introduced which are adapted to particular ecosystems of the introduced range (Zenni et al., 2016). These populations often undergo demographic bottlenecks (Arredondo et al., 2018), as has been found for our samples collected from the presumed introduction locations of Hong Kong, Indonesia and north-east India. From the initial introduction locations, long-distance dispersal of M. micrantha occurs through lightweight pappus-bearing seeds and/or human-mediated transfer of broken stolon fragments (Tripathi et al., 2012). During range expansion, the leading-edge populations are expected to experience loss of genetic diversity due to genetic drift operating through repeated bottlenecks or founder events (Piaggio et al., 2017). Similar observations were also recorded for M. micrantha populations in southern China, north-east India, Taiwan and Japan.

High genetic variation within populations due to multiple introductions and gene flow

The findings of our study indicate that the detrimental effects of bottlenecks and heterozygote deficiency on population fitness might have been alleviated by extensive gene flow between M. micrantha populations. The pairwise F_st comparisons and AMOVA showed a low level of genetic differentiation among populations whereas clustering algorithms revealed the presence of admixtured populations in this region. Admixture and low genetic differentiation among populations may be caused by multiple introductions and/or high gene flow (Kočiš Tubić et al., 2015; Li et al., 2019). In our study, both spatial and non-spatial clustering algorithms revealed the presence of multiple clusters and genetic admixture in these populations and is in accordance with the genetic structure obtained from the transcriptome analysis of M. micrantha populations from this region (Yang et al., 2017). In addition, invasive populations originating from multiple introductions often carry private alleles from divergent geographical sources (Ray and Quader, 2014). The presence of 29 private alleles in 15 populations of its invasive range indicated that these populations might have generated from different geographical sources and subsequent gene flow between these populations might distribute these private alleles, as evident from their scattered distribution across the study area. In the case of a single introduction contributing these private alleles, the founding population would have to be very large. This appeared to be unlikely for intentional introductions of M. micrantha for different purposes and in different times in this region (Day et al., 2016).

While we could not infer the exact introduction history of M. micrantha into this region, from the genetic structure found in this study it is reasonable to deduce that multiple introductions followed by high gene flow between different genetic groups shaped the present distribution of the species in this region. Indeed, high gene flow was observed between genetic clusters (e.g. between populations of south China and north-east India). The populations of north-east and south India, whose introductions were separated spatially and temporally, were found to have high admixture coefficients due to high gene flow between these regions. Similar genetic patterns arising from multiple introductions and gene flow were observed for other invasive species such as Lonicera maackii (Barriball et al., 2015), Arundo donax (Tarin et al., 2013) and Ambrosia artemisiifolia (Chun et al., 2010). This evidence of multiple introductions is in accordance with previous studies that demonstrated that M. micrantha was introduced in the Philippines through the Pacific trade routes, for its medicinal values in the Hong Kong botanical garden, as a ground cover crop for rubber plantations through Bogor botanical garden in Indonesia and as a weed of cinchona plantations in north-east India (Banerjee et al., 2019b). The populations collected from the introduced locations in Hong Kong (HKC) and Indonesia (INC) demonstrated outbreeding and the associated bottleneck strength of these founding populations was weakened probably due to high gene flow with the surrounding populations. The founder effects may also be weakened when additional propagules are introduced from the native range, possibly from distinct populations, followed by coalescence of these populations over time (Dlugosch and Parker, 2008). The detrimental effects of inbreeding and bottlenecks of the founder population in north-east India (NBB) thus might have been mitigated by high gene flow with the adjacent region (NEB) where the species was reintroduced during World War II to camouflage airfields (Ellison and Sankaran, 2017). Similarly, the detrimental effects of inbreeding in the Philippines population might be countered by high gene flow with populations from south China and India. Overall, these findings indicate that multiple introductions and gene flow between genetic clusters might aid the founder populations of M. micrantha to establish in this region.

High gene flow and environmental analogy influencing range expansion

From Hong Kong, M. micrantha was introduced to mainland China through Shenzhen (SZ) where the population showed higher genetic diversity compared to other south China populations except Guangxi (GX). This reduction of genetic diversity towards the edge of the species range was probably due to founder events experienced by these populations. However, high gene flow between leading-edge populations of southern China ensured maintenance of genetic diversity and contributed to successful establishment and range expansion of M. micrantha in this region. A similar pattern was observed for range expansion of M. micrantha in north-east India. These regions constitute one of the largest tea-growing regions of the world and are well connected through multiple tea trading routes. Being a weed of tea plantations (Puzari et al., 2010), gene flow of M. micrantha in this region is probably influenced by related anthropogenic activities.

Although the decline in genetic diversity at range limits is supported by many numerical studies (Guo, 2014), contrasting examples are also found, especially at broad geographical scales (Shi and Chen, 2012; Lecompte et al., 2016). A similar pattern of high genetic diversity was observed in leading-edge populations of Singapore (SG), Malaysia (MYA–MYD), south India (KEA and KEB) and Guangxi (GX). Our results indicate that high level of outbreeding might have ensured maintenance of heterozygosity in three Malaysian populations (MYA, MYC and MYD). On the other hand, in spite of experiencing recent bottleneck events, as evident from the mode shift in SG and MYB populations, high gene flow between populations (SG and MYA, MYB) might have contributed to the genetic diversity of the leading-edge populations. Similarly, although having a high inbreeding coefficient and experiencing recent bottleneck events, the high genetic diversity of the GX population might be due to extensive gene flow with surrounding populations (e.g. SZ and HN). The leading-edge populations of south India (KEA and KEB) also showed high heterozygosity, possibly due to high gene flow and consequent admixture between populations, as evident from cluster analysis.

The influence of gene flow between populations and subsequent admixture on the spatial genetic pattern of M. micrantha in this region was further confirmed from the weak relationship between genetic distance and geographical distance (IBD), environmental variables (IBE) and resistance surfaces (IBR). Monmonier’s algorithm, as implemented in the BARRIER program, also revealed that geographical barriers could only hinder gene flow between a few populations in Taiwan, south India, Thailand and Malaysia. With multiple introductions, IBD can explain the genetic differentiation at the early stages of invasion when the populations introduced separately are geographically separated. With increasing gene flow during range expansion and migration, this effect of IBD on genetic differentiation becomes weaker. The long history of M. micrantha in this region (introduction c. 1900 and continuous range expansion thereafter) could therefore explain the observed weak relationship between IBD and genetic differentiation. The similar weak relationship between geographical and genetic distances has been observed for other invasive plant species, such as Lantana camara (Ray and Quader, 2014) and Heracleum mantegazzianum (Henry et al., 2009), and these patterns are often associated with the introduction history of the species concerned.

Our study found that 17 out of 37 inbred populations and 11 out of the 24 bottlenecked populations invaded sites with high environmental suitability (>50 % quartile of the MESS distribution) and possessed high genetic diversity. This indicates that environmental suitability might mitigate the negative effects of inbreeding and founder effects and allow the populations to persist, reproduce and expand. Similar beneficial aspects of the environment have been experimentally proven for other invasive species, but involving environmental factors other than climate (Schrieber et al., 2019). Interestingly, the majority of these populations are in the margin of the species’ distribution (e.g. HN in southern China, JPB in Japan), thereby indicating further range expansion ability of M. micrantha in this region.

Analysis of environmental analogy also revealed that some populations (NBA, NBB and NEA from north-east India, GX from southern China) invaded heterogeneous environments and showed demographic bottlenecks and high inbreeding. In accordance with recent studies reporting a tendency for climatic niche shift of M. micrantha towards cold and dry areas of its invasive range (Banerjee et al., 2017), the findings of this study indicate that the minimum temperature of the coldest month and altitude might contribute to genetic variation in this species. The distribution of M. micrantha has been recorded across a latitudinal gradient ranging from sea level up to 4000 ft. Previous studies have found that altitudinal gradient and related climate conditions can influence genetic differentiation through the impediment of gene flow by reproductive isolation, asynchronous flowering, decreased pollination success and increased resource allocation to clonal reproduction (Hahn et al., 2012). Because our sampling locations also included high-altitude areas of its distribution range (NBA, NBB and NEA in north-east India), it seemed likely that the genetic differentiation was influenced by altitude and temperature. However, anthropogenic drivers often counteract potential effects of natural gradients on population genetic diversity by influencing gene flow between populations (Hahn et al., 2012; Arredondo et al., 2018). A similar pattern was also evident from our study: a > 75 % quartile N_m was observed between NBA and NBB with other populations of north-east India and between GX and other southern China populations. High allelic richness and observed heterozygosity in these marginal populations indicated that gene flow could provide beneficial genetic variation and might facilitate maintenance of genetic diversity and subsequent adaptation of these populations to novel environments (Ramos et al., 2018; Bontrager and Angert, 2019).

Management implications

To the best of our knowledge, this is the first study that has investigated the population structure of M. micrantha at a large spatial scale in Asia. We found multiple genetic groups in this region, and extensive gene flow between populations contributing to high genetic diversity and establishment of the species in heterogeneous environments. Given of its ability for long-distance dispersal primarily through human-mediated transfer of broken fragments (Ellison and Sankaran, 2017) and the extent of anthropogenic activities in this region (e.g. the proposed Belt and Road Initiative; Normile, 2017), further spread of the species in novel regions can be anticipated (Liu et al., 2019). In this context, the findings of our study emphasize continuous surveillance of anthropogenic activities to prevent new introductions and further spread of M. micrantha in this region. Site-specific management strategies, in the form of early detection, rapid response and effective control, should be implemented with inter-regional collaborations (Clements et al., 2019). Our study found preliminary evidence of multiple introductions of M. micrantha in the region. However, further studies with samples from its native range and using additional markers are required to infer the introduction history of the species in the region. This may provide valuable insights into the biological control programme for this species and be used to implement quarantine measures to safeguard this region from new introductions of existing and/or novel lineages from the native range.

SUPPLEMENTARY INFORMATION

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: Details of sampling locations of the 46 populations of Mikania micrantha. Table S2: Information for the 12 microsatellites used in the study. Table S3: Genetic diversity at the 12 microsatellite loci and estimates of genetic differentiation and gene flow. Table S4: Pairwise F_st values between 46 populations of Mikania micrantha. Table S5: Loadings on three axes of the principal component analysis (PCA) of the bioclimatic variables. Table S6: Multivariate environmental similarity surface (MESS) values. Table S7: Examples of recent experimental studies investigating genetic diversity and differentiation parameters of plant species in their invasive range. Fig. S1: Current maps used to identify areas contributing most to connectivity between sampling locations. Fig. S2: Population structure using the assignment result inferred from the BAPS analysis. Fig. S3: Population structure inferred from GENELAND with uncorrelated allele frequencies. Fig. S4: Scatterplot of Mantel test results showing the relationship between pairwise genetic and geographical distances. Fig. S5: Potential gene flow barriers between populations in the study area.

FUNDING

This study was supported by grants from the National Natural Science Foundation of China [grant numbers 41776166 and 31700178]; the Natural Science Foundation of Guangdong Province [grant numbers 2017A030313159 and 2017A030313189]; Guangdong Basic and Applied Research Foundation [grant number 2019A1515012221]; the programme of Guangdong Key Laboratory of Plant Resources [grant number PlantKF05]; the Fundamental Research Funds for the Central Universities; and the Chang Hungta Science Foundation of Sun Yat-sen University.