Combined Effects of Isolation by Distance, Ecology and Resistance on Population Genetic Divergence in Rana kukunoris From Qinghai‐Tibetan Plateau

ABSTRACT

Investigation of the impact of ecological and landscape processes on genetic population structuring and differentiation is critical for understanding adaptation to different environments. Several hypotheses, including isolation by distance (IBD), isolation by environment (IBE) and isolation by resistance (IBR), have been proposed to explain the spatial patterns of genetic diversity and differentiation among populations. However, these hypotheses have rarely been tested together, and the relative importance of these processes is still unclear. Here, we employed genome‐wide SNP data to investigate the genetic structure of 32 Rana kukunoris populations from the Qinghai‐Tibetan Plateau and explored the patterns and extent of genetic divergence among these populations in relation to the three aforementioned isolation patterns. The results indicated that the species was genetically highly structured (F _st values ranging from 0.03 to 0.80) and composed of three distinct lineages, with a high level of gene flow between the eastern lineage and the other two lineages (southern and western). Geographic isolation, environmental isolation and landscape resistance have all significantly influenced genetic differentiation among populations, as indicated by Mantel tests, suggesting that all three processes (viz. IBD, IBE and IBR) have shaped the degree of genetic differentiation. Variance partitioning (Varpart4) showed that IBD accounted for most of the variance (16.8%), followed by IBR (4.1%) and IBE (3.2%). The three most important environmental factors influencing genetic divergence among R. kukunoris populations, identified through multiple regression on distance matrices (MRM), were the occurrence of saline–alkali soil, road proximity and altitude. Interestingly, the relative contributions of isolation hypotheses varied regionally: MRM and varpart4 analyses revealed that IBE dominated in the north (85.2%), shaped by rivers and precipitation; IBR was strongest in the east (23.0%), driven by rivers, road proximity and distance; while IBD prevailed in the south (8.5%). These findings provide novel insights into the different factors and their relative importance in shaping the degree of genetic differentiation in high‐altitude amphibians.

1. Introduction

Exploring the eco‐evolutionary processes that shape patterns of genetic diversity and differentiation is a central focus of contemporary evolutionary biology (Sunde et al. 2020; Jin et al. 2022). Apart from genetic drift and selection, geographical and ecological barriers that reduce the effective dispersal rate among populations also contribute to the spatial pattern of genetic differentiation (Slatkin 1993; Rousset 1997; Bolnick and Otto 2013). Strong genetic differentiation driven by isolation among populations can even lead to reproductive isolation and speciation (Sunde et al. 2020). Hence, understanding how geographic and landscape factors drive population genetic differentiation is crucial for both evolutionary and conservation biology (Sunde et al. 2020).

Several hypotheses related to geographic and landscape factors have been proposed to explain the formation of genetic diversity and differentiation patterns. The geographic isolation hypothesis (isolation by distance; IBD) proposes that increasing geographic distance leads to an increasing degree of genetic differentiation due to decreasing gene flow among populations (Wright 1943; Worsham et al. 2017). The environmental isolation hypothesis (isolation by environment; IBE) posits that genetic differentiation among populations increases with heightened environmental differentiation, independently of geographic distance (Sexton et al. 2014; Wang and Bradburd 2014). Furthermore, from a landscape genetics perspective, environmental heterogeneity across landscapes can impede species dispersal among populations due to high dispersal‐associated costs (e.g., energetic, time, risk and opportunity costs). This leads to a predicted positive correlation between the degree of genetic differentiation and resistance distance among populations, known as isolation by resistance (IBR; McRae 2006; McRae and Beier 2007; Spear et al. 2010). These three hypotheses—IBD, IBE and IBR—are not mutually exclusive but rather complementary, and they have been widely validated across various taxa (e.g., Wang and Bradburd 2014; Trense et al. 2021). Nevertheless, these hypotheses have rarely been tested together in a single study (but see Castilla et al. 2020; Jiao et al. 2024), leaving the relative contributions of each hypothesis to genetic differentiation still largely unclear.

The Qinghai‐Tibetan plateau (QTP) spans an area of 2.57 million square kilometres and exhibits diverse environmental conditions (Zhang et al. 2002). Renowned for its intense ultraviolet radiation, the QTP significantly reshapes the landscape features and climate patterns of central and eastern Asia (Norsang et al. 2021; Zhang et al. 2024). The northeastern region of the QTP is characterized by cold and arid conditions with relatively flat landscapes, whereas the southeastern region showcases typical alpine‐canyon topography and experiences a warm, moist climate. The eastern edge of the QTP features a mountainous environment with notable variations in altitude and temperature (Zhang et al. 2024). In addition, the saline–alkali environment of the Qaidam basin in Qinghai province, located between the southern and northern QTP, may play a crucial role in shaping the contemporary distribution patterns and genetic differentiation of amphibians. The extensive geographic area, diverse environmental conditions and varied landscape features of the QTP are expected to facilitate significant genetic differentiation among terrestrial vertebrates such as amphibians exhibiting low mobility and high natal philopatry (T. Beebee 1996). Hence, the QTP serves as an ideal geographic arena to investigate IBD, IBE and IBR hypotheses and to assess their contributions to genetic differentiation.

Amphibians, characterized by their limited dispersal ability and high philopatry, serve as ideal models for evaluating the relative contributions of IBD, IBE and IBR to contemporary patterns of genetic diversity and differentiation (Smith and Green 2005; Atlas and Fu 2019). R. kukunoris , a widely distributed amphibian in the QTP of China, inhabits a variety of habitats with distinct landscape characteristics (Fei et al. 2010; Chen et al. 2013; Zhou et al. 2013; Chen, Qin, et al. 2023; Chen, Chen, et al. 2023). Its populations endure extreme climatic conditions including significant temperature fluctuations, uneven precipitation and intense UV‐B radiation (Zhou et al. 2013; Chen, Chen, et al. 2023). The saline–alkali environment across its southern and northern distribution ranges leads to an inverted U‐shaped distribution of this species (Fei et al. 2010). Despite the availability of well‐developed genomic resources and a solid understanding of its phylogeography (Zhou et al. 2013; Wang et al. 2020; Chen, Chen, et al. 2023), the drivers of genetic differentiation among local populations have not been thoroughly investigated.

The main aim of this study was to evaluate the relative importance of IBD, IBE and IBR explaining patterns of genetic differentiation among widely distributed R. kukunoris populations in a spatially explicit framework. Specifically, we analysed genome‐wide SNPs from 32 populations across the QTP to: (i) quantify the degree and spatial pattern of genetic divergence; (ii) simultaneously test IBD, IBE and IBR to determine their relative contributions to divergence and (iii) identify key predictors of population structure and divergence. Our results indicate that IBD, IBE and IBR jointly govern genetic diversity and divergence on the QTP, with IBD explaining most of the variance in genetic divergence. However, at regional scales, the relative importance of the three isolation patterns varies, emphasizing the scale dependency of the results.

2. Materials and Methods

2.1. Sample Collection

In total, 154 adult R. kukunoris individuals were collected from 32 sampling sites (Figure 1; Table S1), which cover the main geographic distribution of the species (Fei et al. 2009). Each sampling site was considered a population unit for subsequent genetic analyses. Sites with multiple individuals were analysed as populations, whereas sites represented by a single individual were retained in the dataset but interpreted cautiously due to their limited within‐site variance. Additionally, two individuals of Rana chensinensis were included as outgroups for phylogenetic analysis. After anaesthetising the frogs with an overdose of MS‐222, toe skin tissue samples were collected and preserved in 99% ethanol in the field. These samples were subsequently stored at −80°C in the laboratory until DNA extraction. Genomic DNA was extracted from each sample using a standard CTAB protocol. DNA concentration and purity were assessed using an ND‐1000 spectrophotometer (NanoDrop), and quality was further verified by electrophoresis on 1% agarose gels using lambda DNA as a standard. All experimental procedures were approved by the Animal Ethics Committee of Anhui University (ethical approval no. IACUC(AHU)‐2022‐007).

FIGURE 1.

Population genetic structure of Rana kukunoris based on SNPs. (a) map of the 32 sampling localities depicting altitudes, lakes, saline–alkali soil and the three genetic lineages of R. kukunoris . Admixture results with K values = 3 and the ovals delimit the three lineages. (b) PCA result based on SNPs. (c) NJ tree based on SNPs with two Rana chensinensis individuals as the outgroup. The values on the tree nodes indicate the bootstrap support of ≥ 60%. (d) Heterozygosity estimates for individuals across the three lineages. Differences among lineages were evaluated using Wilcoxon rank‐sum tests (***p < 0.001, **p < 0.01, *p < 0.05). (e) Relative migration network (N _m) among the three lineages.

2.2. High‐Throughput Sequencing and SNP Calling

To obtain a large set of high‐quality single nucleotide polymorphisms (SNPs), we employed the high‐resolution SLAF‐seq (Specific‐Locus Amplified Fragment Sequencing) method (Sun et al. 2013; Wei et al. 2020). An in silico digestion was performed using the R. kukunoris genome (GenBank accession: GCA_029574335.1) as a reference to select appropriate restriction enzymes. A combination of restriction enzymes, specifically HaeIII and HinCII, was selected to produce fragments ranging from 550 to 580 bp in length. The purified PCR products were individually indexed with sample‐specific barcodes to enable bioinformatic sample identification, then pooled and sequenced using the Illumina HiSeq sequencing platform (Illumina, San Diego, CA, USA).

The quality of the data were assessed using FastQC (Andrews 2010), and clean data were obtained by trimming primers, adapters and low‐quality reads with Fastp using default settings (Chen et al. 2018). The high‐quality clean reads were then mapped to the R. kukunoris reference genome (Chen, Qin, et al. 2023; Chen, Chen, et al. 2023) using BWA‐MEM (v.0.7.17; Li and Durbin 2009) with default parameters. The mapped reads were sorted using Samtools (v.1.3.1; Li et al. 2009), and duplicate reads were removed using Picard (v.1.67, http://broadinstitute.github.io/picard/) with default settings. To minimize unique biases associated with different variant calling pipelines (Clevenger et al. 2015), SNP calling was performed with Samtools and GATK (v.4.0.9; McKenna et al. 2010) using default settings. The concordant common sites identified by both methods were selected using the SelectVariants package in GATK with default parameters. Variant filtering was conducted according to the ‘best practices’ workflow established by the GATK team (McKenna et al. 2010). Sequencing depths and mapping ratios for each sample were calculated using the Samtools modules depth and flagstat, respectively.

2.3. Phylogenetic Inference, Genetic Structure and Genetic Diversity

For phylogenetic and population genetic analyses, we excluded SNPs with a minor allele frequency (MAF) of less than 0.05 and those with missing data exceeding 40% across all individuals. Only one biallelic SNP per locus was retained (Zhao et al. 2016). The SNPs were filtered using VCFtools with the parameters: ‐‐min‐alleles 2 ‐‐max‐alleles 2. The final filtered dataset comprised 11,291 informative biallelic SNPs.

We first utilized MEGA X software (Kumar et al. 2018) to construct a neighbour‐joining (NJ) tree based on a set of maximum composite likelihood estimates with 1000 bootstrap replicates, employing the p‐distance model with 11,291 SNPs and two R. chensinensis individuals as the outgroup. Additionally, we investigated the genetic structure among R. kukunoris populations using ADMIXTURE (Alexander et al. 2009) and Principal Component Analysis (PCA). ADMIXTURE was employed to infer historical lineages with K‐values ranging from 1 to 10 (Alexander et al. 2009). The lowest K value, determined through cross‐validation, is considered the most appropriate number of clusters, and the web application Pophelper (Francis 2017) was used to visualize population structure. Furthermore, PCA was conducted to approximate population structure among all study samples using smartPCA (Price et al. 2006). Genetic divergence was estimated among 31 sites with sufficient sample sizes for population‐level F _st calculation. Because QDBT was represented by only a single individual (Table S1), this site was excluded from pairwise F _st estimation but retained for analyses where individual‐level data could be included. Unbiased pairwise F _st values were calculated with the diveRsity R package (Keenan et al. 2013; Sundqvist et al. 2016) using the sample size corrected approach of Weir and Cockerham (1984), with 1000 bootstrap replicates and a significance level of 0.05. Genome‐wide heterozygosity was calculated from the filtered SNP matrix using VCFtools as the number of heterozygous sites divided by the total number of heterozygous and homozygous sites across each genome. Differences in heterozygosity among groups were assessed using Wilcoxon rank‐sum tests. The f ₄‐statistic (Reich et al. 2009) was calculated as f ₄ (Pop1, Pop2; Pop3, Pop4) using qpDstat (version 1152) in ADMIXTOOLS v.5.1 (Patterson et al. 2012). This statistic measures the correlation of allele frequency differences between two pairs of populations and is widely used to detect signals of admixture or introgression. The deeply diverged population R. chensinensis was used as the outgroup (Pop4). Directional relative migration was inferred in R using divMigrate (diveRsity; Alcala et al. 2014; Sundqvist et al. 2016). We supplied a diploid GENEPOP file and used the Alcala N _m estimator (stat = ‘N _m’), which yields directional, relative migration strengths normalized to the maximum edge (= 1). Network edges were plotted and only links with relative strength ≥ 0.05 were displayed (filter_threshold = 0.05).

We utilized the SNAPP (v.1.4.1) plugin of BEAST (v.2.4.4) to estimate divergence times among populations using a molecular clock model (Bryant et al. 2012; Bouckaert et al. 2014; Stange et al. 2018). Due to our limited computational resources, we employed a smaller dataset comprising 71 individuals, generated by randomly sampling two individuals from each site and the outgroup, except for one site where only a single individual was available. Furthermore, given the sensitivity to missing data for SNAPP model, we filtered the SNPs with more than 5% missing data across these 71 individuals. This refined SNAPP‐specific dataset included 2183 biallelic SNPs. To ensure the reliability of our results, we repeated the analysis by randomly substituting individuals within each population, thereby verifying the consistency of the results (Figure S2). The most recent divergence time of 7.8 Ma between R. kukunoris and R. chensinensis served as a calibration node (Zhou et al. 2013). In SNAPP, we conducted three independent analyses with 1,000,000 MCMC iterations. Each chain was thinned by sampling every 1000 trees to reduce serial correlation, and we verified the convergence of the MCMC and effective sample sizes, which were above 200, using TRACER v.1.7 (Rambaut et al. 2018). The results from the three independent chains were combined in LOGCOMBINER v.2.4.4 (Bouckaert et al. 2014). We employed DensiTree v.2.2.6 (Bouckaert et al. 2014) to visualize the SNAPP trees after discarding the first 10% of each MCMC chain as burn‐in. Finally, we used TreeAnnotator v.2.4.4 to explore the maximum credibility trees with median heights (Drummond and Rambaut 2007).

2.4. Isolation by Distance, Environment and Resistance

Geographic distances were calculated using ArcMap, implemented in ArcGIS Desktop v.10.3, based on latitude and longitude data from the sampling sites. Slatkin's linearized F _st (T = F _st/(1 − F _st)) (Slatkin 1995) was compared with geographic distance to investigate whether the observed patterns of genetic differentiation conform to the isolation by distance (IBD) model. Additionally, the correlations between genetic and geographic distance matrices were analysed using the Mantel test with the vegan package (v.2.6.4.) in R software (Oksanen et al. 2022).

The Estimated Effective Migration Surface (EEMS) algorithm, as outlined by Petkova et al. (2016), was utilized to identify the spatial configuration of dispersal corridors and barriers, thereby illustrating the spatial patterns of genetic diversity and gene flow. This analysis identifies geographical regions that deviate from isolation by distance, which can be attributed to variations in gene flow. We set the deme numbers to 100 and 400 and conducted three independent runs using function RunEEMS_SNPS in EEMS, following the methodology detailed by Ferrer Obiol et al. (2025), using 10,000,000 Markov Chain Monte Carlo iterations, 1,000,000 burn‐in iterations and 1000 thinning intervals.

To analyse habitat suitability, the distribution of R. kukunoris was modelled using MaxEnt v.3.4.1 (Phillips et al. 2017). We utilized species occurrence data from 76 sites, comprising 68 locations from our field investigations and 8 records obtained from GBIF and iNaturalist (Table S4). We incorporated bioclimatic variables and landscape features, including altitude, ultraviolet irradiance, slope degree, slope aspect, vegetation type, as well as the presence of roads, rivers and saline–alkali soils, into our models at a resolution of 30 arc sec. To avoid multicollinearity (|r| > 0.7, Table S3), we conducted pairwise Pearson correlations among the 19 bioclimatic variables and landscape features. Consequently, the climate factors employed in our model included the mean diurnal range (Bio2), mean temperature of the driest quarter (Bio9) and precipitation of the driest month (Bio14), alongside landscape features such as altitude, slope degree/aspect, vegetation types, roads, rivers and saline alkali soils. The raster data for altitude and slope degree were extracted from the Chinese Digital Elevation Model using ArcGIS (v.10.3) (http://srtm.csi.cgiar.org/, SRTM30, v.2.1), whereas vegetation type data were obtained from the Resource and Environment Science and Data Center (https://www.resdc.cn/data.aspx?DATAID=133). Ultraviolet data were sourced from glUV (https://www.ufz.de/gluv/) and processed using the costdistance function in ArcGIS. To integrate the bioclimatic and landscape data, these variables were standardized to ensure identical resolution, extent and projection using the raster and rgdal packages in R. The workflow involved three sequential steps: (1) Projection transformation of all input rasters to the WGS 1984 coordinate system (EPSG:4326); (2) resolution resampling to a uniform 30‐arcsecond grid (~1 km); (3) spatial cropping of processed rasters to the extent of China's DEM layer, ensuring consistent spatial boundaries for analysis.

The optimal feature class transformations (LQH) and regularization multiplier value (β = 5) were determined using ENMeval (Muscarella et al. 2014; Table S5), based on minimal change in the corrected Akaike Information Criterion (ΔAICc = 0) (Phillips et al. 2017; Zhao et al. 2021). These parameters were subsequently utilized to optimize the MaxEnt algorithm (Gong et al. 2020; Kou et al. 2020). Following this, the MaxEnt model implemented in the Biomod2 package was used to predict the potential habitat suitability of the species and to assess the relative importance of environmental variables (Zhao et al. 2021). For the modelling process, 10 replicates were run using a 5‐fold cross‐validation strategy (CV.k = 5) and 75% of data were used for training in each replicate (CV.perc = 0.75). Ultimately, the optimized model was chosen based on the True Skill Statistic (TSS = 0.985, Table S6), and the potential suitability (x) derived from this model was utilized in subsequent analyses. The ENMeval and Biomod2 packages are integrated within R software v.4.3.0. We employed ArcGIS 10.3 to visualize the outputs from the ecological niche modelling (Figure 3). According to the habitat suitability index (HSI) derived from MaxEnt predictions, the potential range of R. kukunoris was classified using the natural breaks method (Zhu et al. 2021) into four suitability levels: unsuitable (HSI < 0.1), low suitability (0.1 < HSI ≤ 0.35), moderate suitability (0.35 < HSI ≤ 0.7) and highly suitable (HSI > 0.7). The resulting map illustrates the potential habitat distribution of R. kukunoris across China.

FIGURE 3.

Map depicting the cumulative current flow density between sampling sites, as inferred from the Circuitscape model. Brighter colours indicate higher modelled current flow. Sampling localities are shown as symbols, with squares representing the northern lineage (N), circles representing the eastern lineage (E) and diamonds representing the southern lineage (S), and fill colour indicating mean heterozygosity.

For the IBE analysis, we calculated the matrices of suitability difference (x) for potentially isolating features, including Bio2, Bio9, Bio14, altitude, slope degree, rivers, vegetation types, saline–alkali soils and ultraviolet radiation, using the vegdist function with Euclidean distance. For the IBR analysis, we utilised the inverse suitability value (1/x) of these features (Trense et al. 2022) and applied Circuitscape v.5.0 to generate resistance surfaces and distances through the pairwise resistance method (Anantharaman et al. 2019). Circuit theory, analogous to random walk models, provides a robust framework for understanding movement and gene flow across complex landscapes (McRae and Beier 2007). We employed overall predictors, encompassing geographic, environmental and landscape characteristics, to investigate the relationships among genetic, geographic, environmental and landscape distance matrices through Mantel tests with 9999 permutations (Oksanen et al. 2022). Additionally, we conducted a partial Mantel test to examine the relationships between genetic distance and geographic/environmental/landscape distance matrices, utilising 9999 permutations to control for the effects of the other two distance matrices (i.e., IBD, IBR or IBE). These functions were implemented in the vegan package v.2.6.4.

To investigate the effects of isolation by distance (IBD), isolation by environment (IBE) and isolation by resistance (IBR) on the genetic differentiation of R. kukunoris , we conducted multiple regression on distance matrices (MRM) using the ecodist package (v.2.0.9). To reduce multicollinearity among predictors, all environmental and geographic variables were pre‐screened using the VIF function in the car package (v.3.1.3), and only those with a variance inflation factor (VIF) < 10 were retained for subsequent analyses (Quinn and Keough 2002; Table S8). Following Dormann et al. (2013), VIFs were calculated separately for the overall dataset and for each identified lineage. Because environmental correlations varied among regions (lineages), the sets of retained predictors differed slightly across lineages (Table S8). This approach ensured that predictors within each MRM were mutually independent within their environmental contexts, thereby improving model stability and interpretability. Prior to modelling, all continuous predictors were centred and scaled (z‐standardized to mean 0, SD 1) to reduce scale effects and facilitate comparability across variables. The relative contributions of IBD, IBE and IBR were then quantified using the varpart4 function in the RFunctions.R script (https://github.com/csdambros/BioGeoAmazonia), which partitions the variance into unique and shared fractions. All analyses were conducted in R v.4.3.0.

3. Results

3.1. Sequencing and SNP Calling

A total of 154 individuals of R. kukunoris and two individuals of R. chensinensis were sequenced, generating approximately 1.84 billion paired‐end reads in total (Table S2). 91.37% bases show a quality score above 30 (Q30), and the guanine‐cytosine content was 44.57%. The number of SLAF tags varied from 152,610 to 433,097 across individual samples, and 53,214,731 SLAFs in total were obtained, with an average sequencing depth of 11.50×. In total, 4,834,847 SNPs were obtained, and SNP number varied from 236,382 to 1,945,400 per individual. After sequential filtering, a total of 11,291 high quality biallelic SNPs were retained.

3.2. Genetic Diversity, Genetic Structure and Phylogenetic Inference

Multiple analyses, including NJ tree, ADMIXTURE and PCA, consistently revealed a congruent genetic structure in R. kukunoris (Figure 1; Figure S1). The NJ tree based on 11,291 SNPs identified three major genetic clusters corresponding to geographic regions (Figure 1): a northern cluster (N) comprising eight populations along the northern margin of the Qinghai–Tibet Plateau (QTP); an eastern cluster (E) including 16 populations from the eastern edge of the plateau; and a southern cluster (S) encompassing eight populations from the southeastern QTP. SNAPP‐based divergence estimates indicated that the eastern and northern lineages diverged around 4.71 million years ago (95% HPDI: 3.54–5.92 Ma), whereas the southern lineage separated from the eastern lineage approximately 2.65 Ma (95% HPDI: 1.98–3.32 Ma; Figure 1a; Figure S2).

Consistent with this deep divergence, genetic diversity varied markedly among lineages (Kruskal–Wallis test: χ ² = 25.251, df = 2, p < 0.001), with observed heterozygosity being highest in the northern lineage (H = 0.072) and the lowest in the southern lineage (H = 0.049; Figure 1d; Figure S3). Based on the ADMIXTURE‐defined subdivision of the eastern lineage (E1–E3; Figure 1a), f ₄ tests revealed pervasive but uneven gene flow among lineages (Table 1). We used the topology ((P1, P2), P3; Out) and report f ₄ (the BABA‐ABBA numerator; sign‐equivalent to D); thus negative and significant (|Z| > 3, estimate via block‐jackknife) f ₄ indicated excess ABBA and gene flow between P1 and P3. At the lineage level, strong signals involved the northern lineage with both eastern and southern lineages—for example, f ₄ (N, E; S, R. chensinensis ) = −0.046, Z = −17.69 and f ₄ (N, S; E, R. chensinensis ) = −0.039, Z = −12.72. Within the eastern sublineage, significant gene flow was detected between E3 and E1 (f ₄ (E3, E2; E1, R. chensinensis ) = −0.010, Z = −4.99; f ₄ (E1, E2; E3, R. chensinensis ) = −0.006, Z = −3.23), as well as S–E connections involving E1 and E3 (f ₄ (E2, E1; S, R. chensinensis ) = −0.026, Z = −11.43; f ₄ (E3, E1; S, R. chensinensis ) = −0.031, Z = −11.79). The strongest single contrast was N versus E1 (f ₄ (N, E1; S, R. chensinensis ) = −0.062, Z = −18.85). Collectively, these analyses indicate widespread but spatially heterogeneous gene flow, with pronounced connectivity involving the northern lineage, coupled with exchanges both within the eastern lineage and between the eastern and southern lineages, as well as detectable allele sharing between the eastern and northern lineages.

TABLE 1.

Results of f ₄‐statistics using Rana chensinensis as the outgroup.

Pop1	Pop2	Pop3	Outgroup	D_stat	Z_score	BABA	ABBA	nSNPs
S	E	N	R. chensinensis	−0.007	−3.631	209	244	4903
N	E	S	R. chensinensis	−0.046	−17.685	209	434	4903
N	S	E	R. chensinensis	−0.039	−12.716	244	434	4903
S	E2	N	R. chensinensis	−0.008	−3.198	213	252	4668
S	E3	N	R. chensinensis	−0.010	−3.705	217	264	4668
E3	E2	E1	R. chensinensis	−0.010	−4.994	233	279	4668
E1	E2	E3	R. chensinensis	−0.006	−3.225	233	262	4668
N	E1	S	R. chensinensis	−0.062	−18.848	172	463	4668
N	E2	S	R. chensinensis	−0.036	−12.594	213	383	4668
N	E3	S	R. chensinensis	−0.032	−11.144	217	365	4668
E2	E1	S	R. chensinensis	−0.026	−11.431	190	312	4668
E3	E1	S	R. chensinensis	−0.031	−11.788	194	337	4668

Note: Significant (|Z| > 3) D values indicate gene flow between Pop2 and Pop3.

3.3. Topographic and Environmental Barriers Shape Spatial Genetic Structure and Connectivity

Genetic differentiation (F _st) among populations ranged from 0.03 to 0.80 (Table S7), whereas geographic distances spanned 15.94–936.84 km. The MaxEnt model showed high predictive performance (TSS = 0.985), with the predicted distribution closely matching the known range of R. kukunoris . Elevation was identified as the most influential variable shaping habitat suitability, followed by presence of saline–alkali soils (Figure S4).

Building on this habitat model, the EEMS analysis revealed pronounced spatial heterogeneity in effective migration across the species' range. Effective migration was highest between the southern and eastern populations but low between the northern and other lineages (Figure 2). The relative migration network supported these findings, revealing asymmetric gene flow among lineages (Figure 1e; Figure S5). The strongest migration occurred from the southern lineage to the eastern lineage (N _m = 1.00), followed by moderate exchange between the northern and eastern lineages (N _m = 0.53) and minimal exchange between the northern and southern lineages (N _m = 0.12). Within the ADMIXTURE‐defined eastern sublineages, migration was also strongly asymmetric: the dominant edge was from E2 to E3 (relative strength = 1.00), with a nearly reciprocal but slightly weaker flow from E3 to E2 (0.98). Additional inflow into E3 originated from E1 to E3 (0.66), whereas movement from E1 to E2 was moderate (0.77). The reverse directions were weaker: from E2 to E1 (0.46) and from E3 to E1 (0.44). Landscape connectivity modelling using Circuitscape also indicated weak connectivity both within and between the northern and eastern lineages, but strong internal connectivity within the southern lineage (Figure 3). Together, these spatial patterns highlight the influence of topographic and environmental heterogeneity in shaping population connectivity and genetic structure across the QTP.

FIGURE 2.

Posterior mean migration rates inferred by EEMS are visualized, with blue indicating regions of elevated gene flow (dispersal corridors), white representing neutral areas consistent with isolation by distance, and orange denoting zones of reduced gene flow (dispersal barriers). Sampled demes are marked as black dots. The black boundary line corresponds to the provincial boundaries of China. (a) All population; (b) populations from southern lineage; (c) populations from eastern lineage and (d) populations from northern lineage.

3.4. Importance of Geographical, Environmental and Landscape Factors in Explaining Genetic Differentiation

Before evaluating the relative effects of IBD, IBE and IBR, we first examined the spatial relationship between genetic and geographic distances across populations of R. kukunoris (Figure S6). A significant positive correlation was observed, indicating a clear pattern of isolation by distance at both the species and lineage levels (Table 2).

TABLE 2.

Results of simple and partial Mantel tests for isolation by environment (IBE), isolation by distance (IBD), and isolation by resistance (IBR) in 31 Rana kukunoris populations.

Mantel test			Partial mantel test
Feature	r	p	Feature	Partialled out feature	r	p
All_IBR	0.7475	0.0001	All_IBR	All_IBE	0.7433	0.0001
All_IBD	0.7609	0.0001	All_IBR	All_IBD	0.4502	0.0005
All_IBE	0.1725	0.0752	All_IBD	All_IBE	0.7700	0.0001
E_IBR	0.6886	0.0004	All_IBD	All_IBR	0.4946	0.0001
E_IBD	0.5073	0.0001	All_IBE	All_IBD	0.2086	0.0459
E_IBE	0.2696	0.1107	All_IBE	All_IBR	0.0694	0.2519
N_IBR	0.2637	0.1966	E_IBR	E_IBD	0.5416	0.0214
N_IBD	0.2627	0.2175	E_IBR	E_IBE	0.6703	0.0019
N_IBE	0.6958	0.0125	E_IBD	E_IBE	0.5568	0.0001
S_IBD	0.7439	0.0022	E_IBD	E_IBR	0.0442	0.4276
S_IBR	0.7809	0.0006	E_IBE	E_IBD	0.3722	0.0787
S_IBE	−0.1309	0.6857	E_IBE	E_IBR	0.1702	0.1789
			N_IBR	N_IBD	0.1714	0.2333
			N_IBR	N_IBE	0.1540	0.2016
			N_IBD	N_IBE	0.3566	0.1777
			N_IBD	N_IBR	0.1698	0.1692
			N_IBE	N_IBD	0.7186	0.0099
			N_IBE	N_IBR	0.6773	0.0177
			S_IBR	S_IBE	0.7885	0.0006
			S_IBR	S_IBD	0.3625	0.0973
			S_IBD	S_IBE	0.7393	0.0010
			S_IBD	S_IBR	−0.077	0.6197
			S_IBE	S_IBR	−0.2171	0.8278
			S_IBE	S_IBD	−0.0447	0.4989

To further explore the underlying causes of spatial genetic differentiation and connectivity, we performed Mantel and partial Mantel tests. The results indicated that both IBD and IBR significantly influenced genetic differentiation among populations (p < 0.001; Table 2). However, after controlling for IBD, partial Mantel tests revealed that IBE remained significantly correlated with genetic differentiation (p < 0.05). Variation partitioning analysis (varpart4) showed that IBD, IBR and IBE together explained 66.4% of the total variance in genetic differentiation. Among these, IBD made the largest independent contribution (16.8%), followed by IBR (4.1%) and IBE (3.2%) (Figure 4).

FIGURE 4.

Venn diagram of variation partitioning results based on distance matrices, illustrating the relative contributions of three explanatory variable groups. Negative fractions were retained in the figure to show the full results of variation partitioning; these negative values indicate fractions of explained variation that are smaller than expected by random predictors and should be interpreted as zero or negligible explanatory power (Legendre 2008). (a) All populations; (b) populations from the southern lineage; (c) populations from the eastern lineage and (d) populations from the northern lineage.

Additionally, MRM indicated that the overall model explained a significant proportion of the genetic variation (R ² = 0.686, p < 0.001). IBD (β = 0.818, p < 0.001), roads (IBR, β = 0.271, p = 0.019; IBE, β = 0.153, p = 0.018), saline–alkali soils (IBE, β = 0.111, p = 0.007) and altitude (IBE, β = 0.127, p = 0.035) were also identified as significant predictors. In contrast, rivers, vegetation, slope and climatic factors (bio2, bio9, bio14) had no significant effects (Table 3).

TABLE 3.

Results of the multiple regression on dissimilarity matrices (MRM) for isolation by environment (IBE) and Isolation by resistance (IBR) in 31 Rana kukunoris populations.

Variable	Regression coefficient	p
IBD	0.818	< 0.001
IBR_bio14	−0.180	0.114
IBR_bio2	−0.096	0.395
IBR_saline–alkali soils	−0.191	0.061
IBR_road	0.271	0.019
IBE_river	−0.015	0.709
IBE_elevation	0.127	0.035
IBE_bio14	−0.001	0.967
IBE_bio9	0.020	0.681
IBE_bio2	0.025	0.583
IBE_aspect	0.026	0.591
IBE_saline–alkali soils	0.111	0.007
IBE_vegetation	−0.061	0.218
IBE_slope	−0.085	0.251
IBE_road	0.153	0.018
	R 2 = 0.686	p < 0.001

Across the three lineages, distinct patterns of spatial genetic differentiation were observed (Figure 4; Table S9). In the northern lineage, IBE was the sole significant driver of genetic differentiation according to both Mantel and partial Mantel tests (p < 0.05), and it independently explained 85.2% of the variance in variation partitioning analysis. MRM also confirmed a significant model fit (R ² = 0.742, p = 0.017), with rivers and precipitation under the IBE framework identified as significant predictors. In the eastern lineage, both IBD and IBR showed significant effects in Mantel and partial Mantel tests, together accounting for 58.0% of the variance, led by IBR (23.0%) and IBD (13.0%). MRM confirmed significance (R ² = 0.633, p = 0.004) with geographic distance, roads and rivers as key predictors. In the southern lineage, IBD and IBR were significant in both tests, collectively explaining 59.1% of variance, though IBD contributed most (8.5%). MRM results remained significant (R ² = 0.709, p = 0.006), with IBD as the only significant predictor.

4. Discussion

4.1. Contemporary and Historical Patterns and Processes

Our findings indicate that R. kukunoris comprises three distinct lineages. The northern and eastern lineages correspond to the two major groups described in earlier studies (Zhou et al. 2013), whereas the southern lineage constitutes a previously unrecognized lineage revealed by the expanded sampling and genomic resolution of the present study, with significant gene flow occurring between the eastern lineage and the other two. We also inferred that this species diverged from R. chensinensis approximately 6.46 million years ago (95% HPD interval: 4.83–8.06 million years), a timeline that predates Zhou et al. (2013)'s estimate of around 1.3 million years (95% CI: 0.86–1.71 million years) but aligns more closely with the divergence time estimated by Zhou et al. (2012) using secondary calibration points (7.8 million years, 95% CI: 3.1–13.8 million years). Based on geological data, the QTP moved northward during the Miocene (5.33 million years ago), and the resulting intracontinental compression led to the formation of foreland basins and the Qilian orogenic belt (Du et al. 2020). Hence, it seems plausible that the formation of the Qilian orogenic belt contributed to the divergence between R. chensinensis and R. kukunoris around 6.46 million years ago. We estimated that R. kukunoris split into the eastern and northern lineages approximately 4.71 million years ago (95% HPDI: 3.54–5.92 Ma), whereas the southern lineage diverged from the eastern lineage approximately 2.65 million years ago (95% HPDI: 1.98–3.32 Ma). This divergence time is consistent with the findings of Zhou et al. (2012), who reported that the radiation of R. kukunoris occurred around 3.5 million years ago (95% CI: 0.74–8.4 Ma), correlating with the late Cenozoic uplift of the QTP (1.76–3.76 Ma; Li and Fang 1999).

Beyond the deep split among the three major lineages, ADMIXTURE further indicated substructuring within the eastern lineage (E1–E3). This fine‐scale subdivision is consistent with geographical isolation imposed by the highly dissected alpine‐valley topography of the Hengduan Mountains at the eastern margin of the QTP. The pronounced environmental heterogeneity, driven by steep elevational gradients, has likely facilitated this divergence, and these processes were probably amplified by Quaternary climatic oscillations, which periodically altered habitat connectivity and produced alternating phases of isolation and reconnection (Wei et al. 2020; Jiao et al. 2024). Looking to the future, combining genome‐wide data with palaeoclimatic reconstructions in spatially explicit demographic frameworks will be valuable for reconstructing the demographic history underlying these patterns.

The spatial patterns of gene flow in R. kukunoris , inferred from EEMS, f ₄‐statistics, and the relative migration network, indicate extensive and bidirectional gene flow among lineages. However, these exchanges are not equal in magnitude: the combined evidence suggests a net downslope bias, strongest from south to east and then to north, broadly following the present‐day elevational contrast (S > E > N). The Circuitscape results independently indicate a higher level of connectivity along south–north corridors, supporting the overall south‐to‐east‐to‐north flux. A probable explanation is that historical isolation followed by secondary contact among elevationally differentiated habitats contributed to asymmetric gene flow. Similar dynamics of alternating isolation and connectivity have been reported in other montane taxa affected by Pleistocene climatic fluctuations (Qu et al. 2011).

Interestingly, the eastern lineage may represent a secondary contact zone, as suggested by its intermediate geographic position between the northern and southern lineages, together with significant allele‐sharing signals detected by f ₄‐statistics and admixture patterns revealed by ADMIXTURE. This pattern likely stems from historical isolation during glacial periods, when alpine glacier advances fragmented populations into separate refugia, promoting localized gene flow (Wei et al. 2020; Zhou et al. 2013; Hewitt 2000). At the same time, within the eastern lineage, gene flow between E3 and E2 was markedly stronger than that between E1 and E2, likely because the E1–E2 connection is impeded by multiple barriers in the Hengduan region including the Qionglai Mountains, the Bayan Har range of the Kunlun system and the Dadu River. Collectively, these findings demonstrate that both historical orogenic events and contemporary landscape heterogeneity have jointly shaped the spatial genetic architecture and gene flow dynamics of R. kukunoris across the QTP.

4.2. Isolation by Distance, Environment and Resistance

Across the range, IBD, IBE and IBR all contributed to genetic differentiation, with IBD predominant overall (~16.8%). This predominance is expected given the macrogeographic extent of the QTP—which lengthens pairwise distances and constrains dispersal (Wright 1943; Slatkin 1993) as well as due to the life history of amphibians, which is characterized by low vagility and high natal philopatry. Together, these factors result in a clear pattern of distance‐dependent genetic differentiation (Wei et al. 2020). Beyond distance, environmental and landscape context also matter. Signals attributed to IBE—notably elevation, precipitation and presence of saline–alkali soils—indicate that environmental dissimilarity promotes divergence, consistent with ecological specialization and incipient local adaptation (Cun and Wang 2010; Wang and Bradburd 2014; Wei et al. 2020). Field surveys show that saline–alkali habitats largely exclude R. kukunoris, plausibly via dehydration, ion imbalance and osmoregulatory stress given their permeable skin (Christy and Dickman 2002; Hopkins and Brodie 2015). In parallel, IBR captures movement resistance from terrain and infrastructure: roads can impose thermal/desiccation stress, increase traffic mortality and add noise/hydrological disturbance, jointly reinforcing isolation (McRae and Beier 2007; Spear et al. 2010; Jackson and Fahrig 2011; T. J. C. Beebee 2013; Dornas et al. 2019; Grenat et al. 2023). Although rivers often act as barriers for amphibians, their marginal effect in our regressions was weak after accounting for other covariates, likely because many streams in the study area are narrow and slow‐flowing (Zhao et al. 2009).

The relative importance of IBD, IBE and IBR varied among regions, reflecting local topoclimatic conditions and the extent of human disturbance. In the northern region, IBE exerts the strongest influence, with rivers and precipitation emerging as key predictors; this pattern accords with the area's arid climate, limited water availability, and the isolating effect of the eastern Qilian Range (Du et al. 2020; Feng et al. 2020). In the east, IBR predominated: roads increase resistance, whereas rivers reduce effective resistance by providing riparian corridors, rendering geographic distance comparatively uninformative; this matches the region's dense infrastructure, rugged relief and multi‐branch headwaters of the Yellow River (Trense et al. 2021; Brierley et al. 2022; Haugen et al. 2023; Tu et al. 2023). By contrast, in the south a simpler IBD pattern prevails: alpine meadow–wetland complexes form broad, permeable corridors with gentle environmental gradients and few hard barriers, promoting stepwise dispersal and greater genetic differentiation with increasing geographic distance (Wright 1943; Slatkin 1993).

Our scale‐explicit synthesis shows that, at the macro scale, R. kukunoris exhibits a pronounced IBD signal indicative of distance‐limited dispersal and strong site fidelity, whereas at regional scales the balance shifts contextually—steep environmental gradients elevate IBE and high resistance (e.g., dense roads, dissected river networks) amplifies IBR, with rivers locally acting as riparian corridors—together producing lineage‐specific evolutionary trajectories across the Qinghai–Tibetan Plateau and revealing incipient local adaptation in environmentally extreme habitats (e.g., saline–alkali zones). This pattern of divergent evolution—where conspecific populations adapt along distinct ecological and geomorphological pathways—aligns with broader evidence from amphibians and other freshwater taxa (Schneider 2005; Good et al. 2006; Cummins et al. 2019; Shen et al. 2019; Liu et al. 2025). Accordingly, our results underscore the combined influence of environmental, landscape and historical factors on genetic differentiation (Jiao et al. 2024) and emphasize conservation strategies tailored to discrete evolutionary lineages and to the specific barriers and corridors governing their connectivity in complex plateau landscapes (Chhina et al. 2024).

Importantly, EEMS and Circuitscape captured system‐wide variation in effective migration, whereas regression‐based approaches (e.g., MRM) quantify the marginal effects of specific predictors; these methods therefore operate at different spatial and analytical scales and should be viewed as complementary rather than conflicting. Integrating results across these scales provides a coherent framework for identifying context‐dependent processes such as local adaptation, which can be further tested using genome‐wide scans and explicit landscape‐genomic models. We acknowledge several limitations. First, because gene flow weaves lineages together, a simple tree view can be misleading‐boundaries based only on phylogenetic trees should be treated with caution, as introgression can distort inferred branching and distances (Leaché et al. 2014; Hibbins and Hahn 2022). Second, inference is constrained by uneven sampling, small per‐population sample sizes and SNP density, which can reduce the spatial resolution and robustness of EEMS estimates (Petkova et al. 2016). Looking ahead, parallel macro‐ and regional‐scale work should integrate genome‐wide selection scans with landscape‐genomic modelling in contact zones and adopt comparative designs (multi‐species within regions or cross‐regional contrasts within species) to test generality and to disentangle effects of selection from demography.

5. Conclusions

We show that integrating macro‐ and regional‐scale perspectives clarifies how distance, environment and resistance jointly structure genetic differentiation on the QTP. We also identify two putative secondary contact zones involving the eastern lineage—at its interfaces with the southern and northern lineages—highlighting ongoing connectivity amid deep subdivision. Together, our results provide a practical framework for plateau amphibian phylogeography and point to actionable next steps: date secondary contact and quantify direction and magnitude of gene flow using genome‐wide data and test the permeability of major barriers with hypothesis‐driven landscape models. These steps will sharpen lineage boundaries and clarify the spatial organization of gene flow.

Author Contributions

The study was conceived by W.C. and J.M. with significant later contributions from X.F. and S.W.; samples were collected by H.C., H.X., Y.W., X.L., Z.Z., H.Q., M.T., L.P., J.L., H.X. and S.W. contributed to the analyses. H.X., W.C., L.J., X.F., B.F. and J.M. wrote the original draft, and all authors reviewed it.

Funding

This work was supported by the National Science Foundation of China (32571739, 32270457, 31872216 and 31670392).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Cross‐validation error from ADMIXTURE analyses across K values. The K value with the lowest cross‐validation error was used to infer the most likely number of genetic clusters.

Figure S2: Sensitivity analysis of SNAPP species‐tree estimates using alternative individual subsampling. (a) Primary SNAPP species tree shown as Figure 1a in the main text. (b) SNAPP species tree inferred after randomly substituting individuals within populations to evaluate the stability of divergence‐time estimates.

Figure S3: Individual genome‐wide heterozygosity estimates for Rana kukunoris populations. Heterozygosity was calculated from the filtered SNP matrix as the number of heterozygous sites divided by the total number of called homozygous and heterozygous sites for each individual.

Figure S4: Habitat suitability map for Rana kukunoris generated using MaxEnt. Habitat suitability index (HSI) values were classified as unsuitable (HSI < 0.1), low suitability (0.1 < HSI ≤ 0.35), moderate suitability (0.35 < HSI ≤ 0.7) and high suitability (HSI > 0.7). The black boundary line indicates the international boundary of China, and boxplots show the relative importance of environmental variables in the MaxEnt model. The current species range was downloaded from the IUCN Red List (http://www.iucnredlist.org/).

Figure S5: Relative migration network (N _m) among eastern sublineages of Rana kukunoris. Network edges represent directional relative migration strengths normalized to the maximum inferred value.

Figure S6: Isolation‐by‐distance analyses for Rana kukunoris. Relationships between genetic and geographic distances are shown for (a) the overall dataset, (b) the southern lineage, (c) the eastern lineage and (d) the northern lineage. Colours indicate the density of population pairs.

Table S1: Sampling localities, elevation, geographic coordinates, sample sizes and lineage assignments for Rana kukunoris.

Table S2: Sequencing statistics for 156 individuals: 154 Rana kukunoris and two Rana chensinensis individuals (ZGLM1 and ZGLM2).

Table S3: Pairwise Pearson correlation matrix among environmental and landscape variables used for habitat suitability and landscape genetic analyses.

Table S4: Occurrence records of Rana kukunoris used for MaxEnt habitat suitability modelling.

Table S5: ENMeval evaluation of MaxEnt feature classes and regularization multipliers for model selection.

Table S6: Performance of MaxEnt models implemented in biomod2, evaluated using the True Skill Statistic (TSS) across pseudo‐absence and cross‐validation replicates.

Table S7: Pairwise F _st values among 31 Rana kukunoris sampling sites.

Table S8: Variance inflation factor (VIF) values of predictors retained after collinearity screening for multiple regression on distance matrices (MRM) analyses.

Table S9: Multiple regression on distance matrices (MRM) results for isolation by distance (IBD), isolation by environment (IBE) and isolation by resistance (IBR) predictors.

Data Availability Statement

Raw sequence data for both the species have been made available on NCBI Sequence Read Archive (SRA) database and can be found under the BioProject accession numbers PRJNA1153062 and PRJNA1153066 for Rana kukunoris and Rana chensinensis , respectively, and the script used in the analyses is available at https://github.com/mysfxh/Rana‐kukunoris_IBD‐IBE‐IBR.git.