Computational analysis and modeling of climate impact on Pteridium aquilinum (L.) populations

Abstract

Pteridium aquilinum is a medicinally important fern with a limited range in northern Iran, increasingly threatened by climate change. Using morphological, genetic, and environmental data, we assessed differentiation, adaptive capacity, and vulnerability across 11 populations. Factor analysis of mixed data (FAMD) identified stipe indument, pinnule shape, and pinnae number as key traits distinguishing populations. Redundancy and association analyses (RDA/CCA) revealed strong links between both morphological and genetic variation and climatic gradients, particularly temperature and humidity, indicating local adaptation. Several SCoT loci were detected as adaptive outliers. Spatial PCA showed that variation is shaped by both global and local spatial factors, forming clines and local variants. Populations varied in sensitivity and adaptive capacity; populations 2, 3, 7, and 8 exhibited the lowest adaptive indices and highest vulnerability. Connectivity modeling suggested that while some populations (e.g., 2, 4, and 6) may maintain or slightly improve connectivity, others risk isolation under future climates. Structural equation modeling (SEM) indicated a positive genetic contribution to adaptation, while differential equation modeling (DEM) predicted logistic growth with temporary instability and genetic decline in vulnerable groups. Overall, findings highlight spatially uneven adaptive responses and recommend targeted conservation through connectivity enhancement, assisted gene flow, and ex-situ preservation of adaptive genotypes.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-33035-1.

Introduction

Adverse effects of climate change, encompassing phenomena such as increased mean temperatures, altered precipitation regimes, and more frequent extreme weather events, represent one of the primary global threats to the extinction of plant species¹. These climatic shifts directly challenge species by disrupting the crucial relationship between genotypes and their environment. Depending on the species’ adaptive capacity, populations may respond by migrating to suitable habitats, undergoing phenotypic adjustments through plasticity or evolutionary changes, or facing localized extinction if adaptation is insufficient. Studies suggest that evolving phenotypic plasticity plays a prominent role in adaptive responses to climatic gradients, with divergence in environmentally responsive loci helping populations persist under changing conditions^2,3. Moreover, genome-level plasticity, such as epigenetic modifications and structural variation, can generate additional genetic variation and enhance adaptive potential under rapid climate change⁴. Empirical evidence further supports these mechanisms: for example, in Populus trichocarpa, significant phenotypic plasticity related to growth and phenology has been observed under temporal environmental fluctuations. However, adaptive capacity is not unlimited: constraints such as habitat fragmentation, limited dispersal, and low gene flow may prevent migration or in situ adaptation⁵.

The urgency is particularly high for taxa with restricted geographical distributions, which often lack connectivity and adaptive capacity to track shifting suitable conditions⁶. Therefore, conservation programs aimed at preserving these vulnerable taxa are of paramount importance. To formulate truly effective and predictive strategies, a thorough knowledge of a species’ intrinsic adaptive capacity, genetic structure, ecological sensitivity, and landscape connectivity is necessary^7,8. This need is particularly acute given that global assessments often overlook subpopulation vulnerabilities within widespread species⁹. Furthermore, advanced ecological and genetic modeling techniques, such as Differential Equation Modeling (DEM), are essential for quantifying the potential time-dependent impacts of climate change on population dynamics and genetic viability, thereby significantly informing proactive conservation planning for a particular species¹⁰.

Ferns are the second-largest group of vascular plants after angiosperms, contributing widely to global biodiversity with approximately 12,000 species¹¹. These vascular plants, which reproduce via wind-borne spores, are widely distributed globally but show high sensitivity to environmental factors^11,12. Abiotic factors are key drivers of fern species distribution at all scales¹³. Their dispersal is often more influenced by environmental conditions, such as water availability, light exposure, and soil properties, than by geographic distance^12,14. This close relationship between ferns and their environment makes them valuable bioindicators. Given the strong influence of environmental factors on fern distribution, it is important to consider how these conditions may shape genetic differentiation across landscapes, a concept central to Isolation by Environment (IBE)¹⁵.

The Bracken genus (Pteridium) is a globally widespread group of coarse, deciduous ferns (family Dennstaedtiaceae) found across all continents except Antarctica, noted for having the widest distribution of any fern in the world^11,12. Within this cosmopolitan group, P. aquilinum is the most recognized species, typically occupying temperate and subtropical regions.

Consequently, in global conservation assessments, this species is typically classified as Least Concern (LC) by global entities, such as the IUCN Red List, due to its ubiquitous presence and high ecological resilience¹⁶. In fact, due to its robust rhizomatous growth and allelopathic properties, P. aquilinum is often recognized internationally, and in regional zones like the Hyrcanian forests of Iran, as an aggressive weed that competes intensely with native flora and poses a management challenge in agricultural and pastoral landscapes¹⁷. However, this broad-scale resilience often masks critical local vulnerabilities. Despite the species’ overall success, studies focusing on peripheral or fragmented populations, such as those found across Iran’s geographical range, emphasize that local populations frequently exhibit significant genetic structure, differential adaptive capacity, and sensitivity to environmental shifts¹⁵. Therefore, while the species is not globally threatened, evaluating the long-term persistence and genetic viability of isolated populations in response to climate change is a critical conservation priority to prevent local extirpations and the loss of regional genetic diversity.

Such locally confined populations may face increased sensitivity to climatic variations, potentially experiencing demographic decline or even local extirpation, despite the species’ overall global security¹⁸. Therefore, predicting the persistence and resilience of these populations is crucial for effective conservation planning. This requires an assessment of their vulnerability, which represents the combined effect of a population’s sensitivity (its biological response to environmental conditions) and its exposure (the degree of climatic variation it experiences)¹⁸. Central to resilience is adaptive capacity, defined as the potential of species or populations to tolerate or adjust to change through mechanisms such as phenotypic plasticity^19,20. This capacity is shaped by the interplay of intrinsic traits and extrinsic factors such as habitat fragmentation, which can constrain gene flow and reduce evolutionary potential^19–21.

Landscape genetic studies are essential for understanding how spatial environmental variables influence the genetic structure of fern species, including P. aquilinum^22,23. These studies can reveal patterns of Isolation by Distance (IBD) and Isolation by Environment (IBE), gene flow, and local adaptation. The Landscape Adaptive Index (LAI) is a key tool for assessing a species’ adaptability and resilience to environmental changes such as temperature and precipitation variability^24,25. Adaptive capacity, influenced by phenotypic and genetic diversity, is a crucial component of a population’s response to environmental change¹⁸. Phenotypic plasticity, characterized by reaction norms, describes how a phenotype responds to environmental variation^26,27. Investigating these responses is important for predicting population distribution and conservation status under future environmental scenarios²⁸.

This research addresses a critical gap by providing the first integrative assessment of landscape genetics, phenotypic plasticity, and adaptive capacity in Iranian populations of P. aquilinum. The species was selected due to its medically significant status and high vulnerability as a fragmented taxon at the edge of its distribution range in Iran, making it an ideal model for studying climate change resilience in marginal populations. Crucially, our study moves beyond standard genetic approaches by employing a novel, multi-layered computational framework combining multivariate statistical methods (FAMD, sPCA, RDA, CCA) with advanced modeling (PLS-SEM and DEM). This robust approach allows us to not only quantify the effects of geographical and climatic factors on gene flow and adaptive traits but also to unravel the complex interplay that determines population survival. By establishing the Landscape Adaptive Index (LAI) and quantifying risk scores, we deliver specific, actionable conservation recommendations—a direct and non-generic outcome—that are vital for preserving this vulnerable species and serve as a crucial template for managing other rare, climate-sensitive ferns in the region.

The main objectives of this study are to provide a comprehensive assessment of eleven geographically isolated populations of P. aquilinum in Iran, focusing on their genetic connectivity, adaptive and landscape adaptive index, and phenotypic and genetic plasticity. Additionally, we aim to identify key morphological and genetic traits contributing to population differentiation, evaluate the potential impacts of environmental and climatic variables, and propose targeted conservation strategies to enhance the resilience and long-term survival of these rare and vulnerable populations.

Materials and methods

The fern P. aquilinum was chosen as the focal species for this landscape genetics and adaptive capacity study for three critical and interconnected reasons. Firstly, its vulnerable and highly fragmented distribution in Iran, primarily restricted to isolated populations in the northern regions, makes it an excellent sentinel species for assessing the impacts of climate change and habitat loss on marginal populations¹⁵. Secondly, the species holds significant ethnomedicinal and ecological value in the region, underscoring the urgent need for robust conservation planning to prevent local extinction and the loss of genetic resources²⁹. Finally, as a widespread species in the world with a high capacity for phenotypic plasticity, P. aquilinum serves as an ideal model organism for employing multivariate computational modeling (such as FAMD and PLS-SEM) to unravel the complex interplay between geographical barriers, climatic variables, and local adaptation mechanisms²³.

Plant material

We examined 133 plants belonging to 11 geographical populations of P. aquilinum from the northern provinces of Iran (Fig. 1). Within each population, individuals were randomly sampled, maintaining a minimum distance of approximately 10–15 m between sampled plants to ensure genetic independence. The geographical and ecological characteristics of the eleven P. aquilinum populations are provided in Supplementary Table S1. Samples were collected from the Hyrcanian forests in the northern provinces of Gilan and Mazandaran, Iran. This region is characterized by a humid temperate climate with high annual precipitation (ranging from 1,200 to 2,000 mm) and mild temperatures, supporting dense broadleaf deciduous forests^30,31.

Fig. 1.

Distribution map of Pteridium aquilinum.

The study was conducted in Iran, covering populations across diverse ecological zones. Eleven populations were selected to represent the full geographical and environmental range of P. aquilinum in northern Iran. Selection criteria included variation in altitude, climatic conditions, habitat type, and accessibility for sampling. To ensure representativeness, we sampled individuals randomly within each population and recorded ecological variables such as altitude.

Genetic analysis

We used SCoT molecular markers for our genetic study, details of this have been given before³². Genomic DNA was extracted using the CTAB method, and DNA quality was assessed using gel electrophoresis. PCR amplification conditions and primers (SCoT1, SCoT2, SCoT41) were carefully optimized to ensure reproducible results, and clear, reproducible bands were scored as binary data. Genetic diversity and structure data, including results from AMOVA, UPGMA, DAPC, and LEA, were obtained from our previous study³² and subsequently used in the current analyses to explore population differentiation and adaptation. Full methodological details are provided in Alaeifar et al.³². Raw data were checked for missing values and scored consistently, and any ambiguous bands were excluded from further analyses.

Data analyses

Multicollinearity test

The multicollinearity of the studied variables was checked by using the Variance Inflation Factor (VIF) analysis in the ‘car’ package of R 4.3. A VIF value > 10 indicates multicollinearity issues³³. Prior to multicollinearity testing, all variables were standardized to ensure comparability, and any highly correlated variables were carefully examined. Variables with VIF > 10 were either removed or combined to avoid redundancy in subsequent analyses. This step ensures the reliability of regression-based models such as RDA and RRMM.

This procedure was performed to ensure that the predictor variables included in the RDA and RRMM models were statistically independent, thereby allowing these models to accurately evaluate the relationships addressed in the research questions.

Population differences in morphological characteristics

Factor Analysis of Mixed Data (FAMD) was performed on normalized morphological traits among P. aquilinum populations to assess the contribution of individual characteristics to geographic population differentiation. FAMD is specifically designed for datasets that contain both quantitative and qualitative variables and enables the extraction of shared structure while preserving the integrity of mixed data types^34,35. Prior to analysis, all continuous variables were standardized to zero mean and unit variance, and the dataset was screened for outliers to ensure analytical robustness following recommended procedures for multivariate analyses³⁶. The R packages FactoMineR and factoextra were used to implement the analysis.

The number of retained dimensions was determined based on eigenvalues greater than 1 and examination of the cumulative proportion of explained variance, consistent with best practices in multivariate ordination³⁷. Interpretation of FAMD dimensions was informed by variable loadings to identify traits with the strongest contribution to population differentiation. This approach provides a comprehensive assessment of morphological divergence across populations while accommodating the mixed nature of the dataset.

This analysis was conducted to address the research question regarding the extent to which morphological traits contribute to geographic population differentiation.

Landscape genetics

Association studies

We used Redundancy Analysis (RDA) and Canonical Correspondence Analysis (CCA) to assess the association between morphological traits and genetic loci with environmental and climatic variables. These constrained ordination methods are widely applied in ecological genomics because they enable direct quantification of how environmental gradients influence phenotypic and genetic variation^38,39.

All analyses were performed in PAST 4.3 using 999 permutations to assess model significance. Prior to analysis, environmental predictors were standardized, screened for multicollinearity using Variance Inflation Factors (VIF < 10), and missing values were imputed following standard ecological statistics procedures⁴⁰. RDA and CCA were selected because they are robust tools for partitioning variance attributable to environmental drivers, allowing identification of key climatic and ecological variables shaping morphological and genetic differentiation.

These analyses were conducted to address the research question concerning how environmental and climatic gradients contribute to patterns of morphological and genetic differentiation among populations.

Spatial principal components analysis (sPCA)

We performed sPCA to reveal the spatial structuring of genetic and morphological characteristics of P. aquilinum populations using the Adegenet package in R 4.3⁴¹. This multivariate statistical approach produces estimates of genetic or morphological variation and the spatial structures of the studied populations⁴¹.

sPCA was selected due to its ability to integrate spatial information into the analysis of genetic and morphological variation. Prior to sPCA, data were centered and scaled to ensure comparability. The significance of global and local structures was evaluated using Monte Carlo tests provided in Adegenet. Mantel tests and Moran’s I were applied to quantify spatial autocorrelation, with careful interpretation to avoid overestimating the effect of spatial proximity. This workflow ensures that both global trends and local spatial patterns are robustly detected and biologically meaningful.

This analysis was conducted to address the research question related to identifying spatial patterns underlying morphological and genetic variation across populations.

Identification of potentially adaptive outlier genetic loci

To identify genetic loci potentially associated with environmental, spatial, and climatic gradients, we applied a multi-method analytical framework incorporating complementary statistical approaches. Using multiple algorithms improves the reliability of detecting adaptive signals because each method operates under different assumptions regarding population structure and selection processes^42,43. In addition to RDA, CCA, and sPCA, we employed pcadapt and DAPC (Discriminant Analysis of Principal Components), enabling the detection of candidate loci based on both genotype–environment associations and population-level differentiation.

Prior to all analyses, genetic data underwent standard quality control, including filtering for missing data, retaining loci with minor allele frequency > 0.05, and removing loci exhibiting strong linkage disequilibrium to reduce false positives⁴⁴. Analyses were conducted using the Adegenet and LEA packages in R 4.3. Outlier loci identified by different methods were subsequently compared and integrated using a Venn diagram (VennDiagram package in R), and only loci supported by at least two independent approaches were retained as robust candidate adaptive markers. This integrative strategy allowed for increased confidence in distinguishing true adaptive loci from statistical artifacts.

These analyses were conducted to address the research question concerning the identification of genetic loci potentially under selection and associated with environmental and spatial gradients.

Phenotypic and genetic plasticity analysis

The phenotypic plasticity of morphological and genetic data (genetic diversity and Fst values) across geographical populations was determined using the R packages ggplot2 and ggfortify. Prior to analysis, all traits were standardized and checked for outliers. Genetic diversity metrics were calculated according to standard protocols, and Fst values were computed to quantify population differentiation.

Additionally, we used the Random Regression Mixed Model (RRMM) approach to describe population-level responses and variation among individuals²⁶ using the lme4 package in R 4.3. This model was chosen because it allows incorporation of covariates as fixed or random effects and estimates genotype-specific intercepts and slopes, enabling assessment of reaction norms across environments⁴⁵. Model assumptions, including normality of residuals and independence of random effects, were checked, and model fit was evaluated using AIC test.

The MCMCglmm and lme4 packages of R 4.3 were used for these analyses. MCMCglmm fits multivariate generalized linear mixed models using Markov Chain Monte Carlo techniques, providing robust estimates of variance-covariance structures among traits⁴⁶. Priors were carefully selected, and convergence diagnostics were performed to ensure reliable parameter estimation.

These analyses were conducted to address the research question regarding the extent and patterns of phenotypic and genetic plasticity among populations, and how individual and population-level variation respond to environmental gradients.

Populations’ ecological tolerance and adaptive capacity

Ecological tolerance

Ecological tolerance for each population was quantified as the mean of two components: Environmental Tolerance and Climate Tolerance, following standard approaches used in ecological niche theory and climate–trait variation studies^47,48.

Environmental Tolerance was calculated based on the range or standard deviation of key geographical variables (longitude, latitude, and altitude) across the sampled distribution of each population. These variables describe spatial heterogeneity and are commonly used as proxies for environmental gradients experienced by natural populations.

Climate Tolerance was estimated using the range or standard deviation of essential climatic variables, including temperature, precipitation, and humidity. These climatic descriptors represent major abiotic drivers influencing ecological and physiological limits of plant species⁴⁹.

All variables were standardized prior to analysis to ensure comparability across different units of measurement, and the dataset was screened for outliers to reduce potential bias. This combined approach provides an integrated estimate of each population’s tolerance breadth by capturing the variability of both geographic and climatic conditions within its distribution.

These calculations were conducted to address the research question regarding the ecological tolerance of populations and their capacity to persist under varying environmental and climatic conditions.

Adaptive capacity

The adaptive capacity was estimated by the weighted mean analysis⁵⁰. It was performed as follows:

Estimating Genetic Adaptation = Mean value of genetic diversity + Fst values.

Estimating Morphological Adaptation = Mean of morphological characters (Table 1). Detailed morphological trait data for all populations are provided in Supplementary Table S2.

Table 1.

Morphological characters were used in this study.

Number	Character
M1	Stipe indument
M2	Pinnule shape
M3	Pinnae number
M4	Leaf type
M5	Pinnule margin
M6	Pinnae size
M7	Pinnule size

The overall adaptive capacity (AC) was then computed as a weighted sum of these two components, assigning a weight of 0.3 to both Genetic and Morphological Adaptation. These weights were chosen based on previous studies emphasizing the important role of these components in plant adaptation, while acknowledging that they are not the sole determinants⁸. This approach also allows for consideration of other relevant factors, albeit indirectly, such as physiological plasticity and biotic interactions.

This analysis was conducted to address the research question regarding the overall adaptive capacity of populations, integrating both genetic and morphological components to evaluate their potential for adaptation to environmental and climatic challenges.

Phenotypic plasticity

Additionally, the phenotypic plasticity value for each character was determined as:

Phenotypic Plasticity= (Maximum value– Minimum value)/ Mean value.

This metric allows quantification of the range of trait variation within populations relative to the mean, providing a standardized measure of plasticity.

This calculation was conducted to address the research question regarding the extent of phenotypic plasticity for each trait within populations, allowing assessment of how traits vary in response to environmental conditions.

Estimating landscape adaptive index (LAI)

The LAI was estimated as follows:

Ecological tolerance was calculated as the mean of combined environmental (longitude, latitude, altitude) and climate variables (temperature and precipitation).

Prior to LAI estimation, multicollinearity among variables was assessed, and highly correlated variables were adjusted to reduce bias.

LAI was estimated using weighted mean and PCA analyses:

PCA was performed on combined ecological tolerance, genetic adaptation, morphological adaptation, and environmental tolerance data to evaluate the contribution of each factor.

Weights for the weighted mean LAI were derived from PCA loadings to provide a biologically informed estimate rather than arbitrary assignment.

In the PCA-based LAI method, the first PCA axis scores were used as representative values for the combined traits.

This workflow ensures that LAI reflects both environmental and genetic/ morphological contributions to population adaptability, providing a robust and reproducible measure of adaptive capacity across populations.

This analysis was conducted to address the research question regarding the overall landscape-level adaptive capacity of populations, integrating ecological, genetic, and morphological components to evaluate their ability to respond to environmental and climatic gradients.

Modeling and statistical analyses

Assessing species sensitivity, exposure, and vulnerability

For sensitivity, exposure, and vulnerability, we used components of NatureServe’s Climate Change Vulnerability Index (CCVI)⁵¹. The index considers two primary components of a species’ vulnerability to climate change:

1- Exposure to changes in climate.

2-Sensitivity of the species to changes in climate⁵².

To ensure comparability among variables, each sensitivity-related environmental factor was normalized using the min-max normalization method.

Sensitivity (S) calculation

Sensitivity (S) was estimated as the mean of two standardized sensitivity indices across populations: sensitivity to temperature (Temp-Sens) and sensitivity to humidity (Humidity-Sens):

Here, Temp-Sens and Humidity-Sens represent standardized biological responses of each population to changes in temperature and humidity, rather than raw environmental values. Standardization ensures comparability across populations.

Exposure (E) and vulnerability estimates

Exposure and vulnerability were calculated using the CENFA (Climate and Ecological Niche Factor Analysis) package in R 4.3⁵³.

To assess population exposure to climate change, absolute differences between current and projected future climatic conditions were computed for temperature and humidity:

Temp-Exp = abs (Temp-Future – Temp-Current).

Humidity-Exp = abs (Humidity-Future – Humidity-Current).

Overall exposure (E) was then calculated using the Euclidean norm (root mean square of the environmental change variables), providing a single quantitative measure of the intensity of climatic changes each population is expected to experience:

Vulnerability (V) calculation

Vulnerability of each population was quantified as the interaction between sensitivity and exposure, using the following formula:

This metric integrates both intrinsic biological sensitivity and the magnitude of environmental change, allowing a comprehensive assessment of population-level climate vulnerability. All calculations were carefully validated to ensure reproducibility.

These analyses were conducted to address the research question regarding the climate vulnerability of populations, integrating both species-specific sensitivity and exposure to environmental changes to evaluate potential risks under projected climate scenarios.

Predicting plant traits in response to climate change

We performed a random regression mixed model (RRMM) analysis to illustrate the phenotypic changes of populations in response to climate change. This method was chosen because it allows estimation of population-level variation in trait responses across environmental and climatic gradients, while quantifying both phenotypic plasticity and genetic differentiation in response to climate variables^26,45.

RRMM incorporates spatial and temporal environmental data, enabling predictions of population resilience under future climate scenarios⁵⁴. Prior to modeling, trait data were standardized, checked for outliers, and assumptions of normality and homogeneity of variances were verified. Model fit was evaluated using AIC test to ensure robustness of predictions.

This analysis was conducted to address the research question regarding the expected phenotypic responses of populations to projected climate change, allowing assessment of population resilience and trait variation under future environmental conditions.

Modeling the effects of climate change on population connectivity

We used two complementary models to study the impact of climate change on P. aquilinum populations’ connectivity:

1-Climate resistance model.

2-Genetic connectivity model^55,56.

The change in genetic connectivity of P. aquilinum populations due to climate change was assessed as a quantitative numerical index. This index was derived using Least Cost Paths and genetic resistance surfaces among populations, which identify the least-cost routes for gene flow and the likelihood of allele transfer between neighboring populations^55,56. Input data included the population resistance matrix and climatic resistance layers, and all calculations were performed in R using the raster and terra packages. Results are reported as numerical values of connectivity change for each population, and maps and plots (including Least Cost Path maps, Genetic Corridors, and population connectivity change plots) illustrate the magnitude of increase or decrease in connectivity.

Climate resistance model

Bioclimatic variables were obtained from WorldClim v2.1⁵⁷ at a spatial resolution of 30 arc-seconds (~ 1 km), including key descriptors such as mean annual temperature, temperature seasonality, and annual precipitation. Future climate projections for the year 2050 were derived from downscaled outputs of multiple General Circulation Models (GCMs) under the RCP 4.5 scenario, consistent with widely used climate-impact modeling approaches^1,57.

For each population, relevant climatic variables (temperature-, precipitation-, and humidity-related proxies where available) were extracted using precise geographic coordinates. All raster layers were resampled to a common projection, extent, and resolution and were normalized using z-score standardization to ensure equal weighting of variables. Variable selection followed ecological criteria and statistical screening, including Pearson correlation analysis (|r|< 0.7) to avoid multicollinearity, a standard practice in ecological niche and climate-resilience modeling⁵⁸.

Weights for climatic variables were assigned based on their ecological relevance and their reported influence on plant physiological thresholds and climate sensitivity in the literature⁴⁹. All data extraction, processing, and raster computations were performed in R (version 4.3) using the raster and terra packages.

Genetic connectivity model

The genetic connectivity model integrates climate resistance with genetic distance among populations. Genetic distances were calculated using molecular markers and the Jaccard similarity coefficient, and connectivity was modeled using the Cost-Distance approach in the gdistance package of R 4.3⁵⁹. Combining climate resistance and genetic distance enables a comprehensive assessment of connectivity under future climate scenarios⁶⁰.

All modeling and related analyses were performed using multiple R packages, including raster, rgdal, gdistance, ggplot2, reshape2, sf, osmdata, rnaturalearth, and rnaturalearthdata^61,62. Assumptions of the models were checked, including spatial independence of residuals and appropriate scaling of input layers, to ensure reproducibility and biological relevance.

The genetic corridor Estimation and map

We constructed a map of the genetic corridor under the influence of the landscape context to compare it with climate resistance and genetic distance patterns. This analysis aimed to identify potential dispersal routes facilitating gene flow among P. aquilinum populations under current and future climate conditions.

First, the geographic coordinates of populations were converted into a Spatial Points object using the WGS84 geographic coordinate system. Then, a resistance surface was generated based on proximity to population points and adjusted using climatic and topographic layers to better reflect landscape heterogeneity.

A transition layer was constructed from the resistance raster using the mean of neighboring cell values, followed by geographic correction (geoCorrection) to ensure spatial accuracy and minimize edge effects. From this, a resistance distance matrix was computed to quantify potential movement costs between populations. Finally, accumulated cost maps were generated to identify the most feasible pathways and corridors for gene flow, representing regions of low resistance and high potential connectivity.

These analyses were performed in R 4.3 using the raster, gdistance, and sf packages. Model assumptions—such as isotropy of resistance values and spatial resolution consistency—were verified, and sensitivity analyses were conducted by varying cell size and resistance weighting to test model robustness.

These analyses were conducted to address the research question regarding how climate change is expected to alter population connectivity, identifying potential barriers and corridors for gene flow, and evaluating the capacity of populations to maintain genetic exchange under future environmental scenarios.

Modeling the effects of Climatic factors on plant traits

Random regression mixed models (RRMM) were used to predict plant trait responses to environmental variables (longitude, future temperature, and future humidity) using the MCMCglmm package in R 4.3. This modeling framework was selected because it allows simultaneous estimation of fixed environmental effects and random population-level variation, making it suitable for hierarchical ecological data.

The prior distributions for the models were defined as follows:

prior<- list (R = list (V = 1, nu = 0.002), G = list (G1 = list (V = 1, nu = 0.002))).

For each trait, the model was fit using the following formula:

formula <- as. Formula (paste (trait, “~ Longitude + Temperature + Humidity”)).

Morphological traits were modeled to predict their potential changes under future climate scenarios. Populations were treated as random effects to account for spatial autocorrelation and intraspecific variability.

All predictors were standardized prior to analysis, and diagnostic checks were performed to ensure model convergence and stability.

This analysis was conducted to address the research question regarding how climatic factors are expected to influence plant trait variation across populations, allowing prediction of morphological responses under future environmental scenarios.

Risk score calculation for population vulnerability assessment

To assess the vulnerability of P. aquilinum populations to climate change, we calculated a composite risk score based on three key components: genetic connectivity, climate change impact, and adaptive capacity^63,64. This integrative approach allows for a comprehensive evaluation of both environmental exposure and intrinsic biological resilience.

The genetic connectivity matrix was derived from genetic distance measures estimated using circuit theory⁶¹ and landscape resistance models⁶⁵. Connectivity values were standardized (z-scores) to ensure comparability across populations, and inverse connectivity values were used to emphasize populations with reduced gene flow.

The climate change impact component was obtained from the projected differences between future and current climatic variables (temperature and humidity), while adaptive capacity values were derived from previously calculated genetic and morphological indices. All variables were normalized using min–max scaling prior to risk score estimation to minimize unit and range bias.

The final Risk Score was calculated as follows:

According to this formula, populations with lower genetic connectivity, higher projected climate change impact, and lower adaptive capacity received higher risk scores, indicating greater vulnerability.

Model sensitivity was evaluated by varying the weighting of each component (± 10%) to ensure that no single factor disproportionately influenced the results. All computations were performed in R 4.3 using custom scripts and the packages raster, gdistance, and dplyr.

This analysis was conducted to address the research question regarding the relative vulnerability of populations to climate change, integrating genetic connectivity, projected environmental impacts, and adaptive capacity to identify populations at highest risk.

Structural equation modeling (SEM)

We performed structural equation modeling (SEM) to explore the interrelationships among latent ecological, genetic, and morphological factors influencing population differentiation and survival of P. aquilinum. SEM was selected because it enables simultaneous testing of multiple dependent and independent relationships, capturing both direct and indirect pathways that influence population structure and adaptation⁶⁶.

The latent factors studied included:

• Genetic: Genetic diversity and Fst.

• Environment: Longitude and latitude.

• Bioclim: Temperature and humidity.

• Tolerance: Ecological and environmental tolerance.

• Adaptation: Genetic adaptation, morphological adaptation, environmental adaptation, and landscape adaptive index (LAI).

• Morphology: Stipe indument, pinnae size, and pinnule size.

This framework was developed based on prior theoretical and empirical evidence suggesting that geographic and climatic gradients shape genetic variation, which in turn influences morphological differentiation and adaptive capacity.

For model estimation, bootstrapped datasets (n = 500 resamples) were used to improve the stability of parameter estimates. Prior to SEM, all variables were standardized (z-scores) to minimize scale effects and ensure convergence.

We used Partial Least Squares Structural Equation Modeling (PLS-SEM), also known as composite-based SEM²², implemented in R 4.3 using the packages plspm and semPLS. This method was preferred over covariance-based SEM because it is more robust to small sample sizes and non-normal data distributions.

Model evaluation included assessing the reliability and validity of constructs using multiple criteria:

• Indicator reliability was examined via loading values (> 0.7 considered acceptable).

• Internal consistency was assessed using Cronbach’s alpha (α) and etc., all expected to exceed 0.7.

• Convergent validity was confirmed if the average variance extracted (AVE) exceeded 0.5.

Path coefficients and their significance (p < 0.05) were interpreted to identify the most influential relationships among genetic, environmental, and adaptive factors.

This analysis was conducted to address the research question regarding the interdependent effects of ecological, genetic, and morphological factors on population differentiation and adaptive capacity, allowing identification of direct and indirect drivers of population survival under environmental gradients.

Differential equation modeling (DEM)

To explore the dynamic responses of P. aquilinum populations over time and under changing climatic conditions, we applied Differential Equation Modeling (DEM). This approach was used to simulate temporal changes in key biological parameters (morphological, genetic, environmental, and climatic) and to understand how they jointly influence population trajectories under projected scenarios.

DEM was selected because it can represent non-linear feedbacks and time-dependent interactions among ecological and genetic variables, which are not easily captured by static statistical models. These analyses were conducted using the deSolve package in R 4.3⁶⁷.

Model formulation

We assumed that the system dynamics could be described by a set of coupled ordinary differential equations (ODEs) representing the temporal evolution of biological variables. The model incorporated the following key considerations:

Spatial influence

Geographic coordinates (longitude and latitude) were included as spatial covariates influencing the rate of change of biological parameters, capturing potential gradients in growth and adaptation.

Biological interactions

Morphological and genetic parameters were modeled as interacting nonlinear terms, representing direct and indirect influences among biological traits. For example, genetic diversity may enhance morphological stability under environmental stress, while morphological adaptation can influence survival and dispersal potential.

Population dynamics

Each population was modeled as an independent unit with population-specific parameters (α, β, γ) to account for variability in environmental response. This allowed each differential equation system to reflect the ecological and genetic uniqueness of the population while maintaining a shared overall structure.

Numerical simulation and parameter Estimation

The system of ODEs was solved using the Runge–Kutta integration method, which provides numerical stability and accuracy in simulating dynamic systems⁶⁸. Parameter estimation was performed using the Levenberg–Marquardt algorithm (LMA)—also known as the damped least squares (DLS) method—to minimize the residual sum of squares between observed and predicted data^69,70.

All parameters (α, β, γ) were constrained within biologically plausible ranges based on prior empirical observations to prevent overfitting and to maintain ecological realism.

Mathematical representation

The population dynamics of P. aquilinum were modeled using a system of first-order differential equations derived from logistic population growth principles, extended to incorporate genetic and environmental feedbacks. The general model can be represented as:

where:

= population size (or abundance index) of population i at time t.

= intrinsic growth rate of population i.

, , and =Standardized values representing genetic, morphological, environmental, and climatic influences, respectively.

, , and = Coefficients describing the relative contribution of each factor.

All parameters were optimized using the Levenberg–Marquardt algorithm to minimize residuals between observed and predicted trajectories. This formulation captures both logistic population dynamics and the additive effects of key biotic and abiotic drivers.

The model was implemented and solved numerically using the Runge–Kutta method in the deSolve package⁶⁷, allowing simulation of both short-term (0–10 times steps) and long-term (100 times steps) dynamics.

Model validation and sensitivity analysis

Model performance was evaluated using mean squared error (MSE) and R² values. Good model fit was indicated by lower MSE and higher R² values (> 0.8). A global sensitivity analysis was also performed to assess the relative contribution of each parameter to system dynamics, ensuring that the modeled interactions were not dominated by any single variable.

Graphical outputs of the ODE models were generated to visualize the predicted trajectories of biological parameters across time and space, allowing for comparison among populations and between current and future climate scenarios.

All DEM simulations were cross-validated using subsets of data (80% training, 20% testing) to ensure reproducibility and robustness.

This analysis was conducted to address the research question regarding the temporal dynamics of population responses under changing climatic conditions, integrating genetic, morphological, environmental, and climatic factors to predict population trajectories and assess resilience over time.

Results

Population differences in morphological characteristics

Factor analysis of mixed data (FAMD), based on squared cosine values, identified stipe indument (M1), pinnule shape (M2), and pinnae number (M3) as the most influential morphological characteristics contributing to the differentiation among P. aquilinum populations. These three traits exhibited the highest factor loadings, while other variables, such as M4 and M6, contributed to differentiation with lower factor loadings (Fig. 2). This indicates that M1–M3 are the primary drivers of morphological variance among the studied populations.

Fig. 2.

Factor analysis of mixed data (FAMD) showing the complex differentiation of P. aquilinum populations based on morphological and environmental variables. (A) Variable Contribution Plot. This plot illustrates the correlation and contribution of variables (Morphological: M, Soil: S, and General Adaptations) to the first two dimensions (Dim 1 and Dim 2). Dim 1 and Dim 2 explain 26.7 and 17.9% of the total variance, respectively. Variables with longer vectors and warmer colors (e.g., M1, M2, M3) show a higher contribution to population differentiation. The color scale (contrib) indicates the relative contribution score of each variable to the inertia of the dimensions. (B) Individual Grouping Plot. The projection of the eleven studied populations (1 to 11) onto the Dim 1-Dim 2 plane. This plot demonstrates the relative clustering and differentiation patterns among the geographical populations based on the combined dataset.

Landscape genetics results

Association study

Morphological data analysis confirmed a significant association with environmental and climatic variables. Redundancy analysis (RDA) performed on the morphological data against environmental and climatic variables (with 999 permutations) revealed a highly significant association (p = 0.001). This finding underscores that the morphological differentiation in the studied populations is under a strong influence of environmental factors.

Patterns of Morphological Correlation: The RDA (Fig. 3A) demonstrated a significant and contrasting relationship between the morphological characters and two key environmental factors. Temperature showed a significant positive correlation (p < 0.05) with several key morphological traits, including M1 through M3 (Stipe indument, Pinnule shape, and Pinnae number), as well as M7 (pinnule size). This suggests a common morphological response that potentially adapts to elevated temperatures in these traits. In contrast, humidity exhibited a notable negative correlation with M1-M3, M6 (Pinnae size), and M7. This differential pattern of correlation highlights the relative and contrasting importance of these two factors (temperature and humidity) in shaping the morphology of the studied populations.

Fig. 3.

Association analyses between morphological characters, SCoT genetic loci, and environmental variables. (A) Redundancy analysis (RDA) of morphological data. This plot illustrates the correlation of seven morphological characters (M1-M7) with geographical and climatic variables. Vectors show the direction and strength of environmental factors (Temp, Longitude, Humidity) influencing the morphological traits. Axis 1 and Axis 2 represent the two canonical axes explaining the maximum cumulative variance in the data. (B) Canonical correspondence analysis (CCA) of SCoT genetic loci. This plot demonstrates the association between SCoT loci and environmental variables. The blue points labeled with x and numbers (e.g., X1) represent the individual SCoT genetic loci.

Strong evidence was also observed for the association of specific genetic loci (SCoT loci) with the environment. RDA and Canonical Correspondence Analysis (CCA) conducted to examine the relationship between genetic loci (SCoT) and environmental variables also yielded significant results (p < 0.01). These findings clearly suggest that several specific loci are potentially under environmental selective pressure: SCoT Loci 1, 3, and 6 were distinctly associated with longitude. SCoT Locus 5 showed a strong correlation with the humidity variable. SCoT Loci 16 and 17 were significantly associated with temperature. These results compellingly indicate that environmental factors impact not only the species’ morphology but also their genetic structure, identifying candidate loci for environmental adaptations across the study populations (Fig. 3A,B).

The outlier’s genetic loci with adaptive potentials

A combined analysis identified a set of robust candidate adaptive loci by integrating results from four distinct detection methods (Fig. 4). To pinpoint genetic loci with the highest adaptive potential, we constructed a Venn diagram (Fig. 4) integrating the outlier results from Redundancy Analysis (RDA, Blue), Spatial Principal Component Analysis (sPCA, Green), Discriminant Analysis of Principal Components (DAPC, Yellow), and pcadapt (Pink). The overlap within this diagram provides strong, convergent evidence, suggesting that these shared loci are less likely to be false positives and are therefore highly robust candidates for environmental adaptation.

Fig. 4.

Identification of candidate adaptive genetic loci using a combined computational approach.

Robust and Shared Loci: Crucially, several SCoT loci were consistently identified across multiple methods, indicating high confidence in their adaptive significance. For instance, bands 16 and 17 were commonly flagged by both RDA and sPCA, methods that specifically account for environmental association and spatial structure, respectively. These loci represent particularly strong candidates for further study.

Method-specific findings

While consensus was high for several loci, some outliers were unique to a single method. For example, band 23 was uniquely identified by the DAPC method. The presence of these unique loci suggests that different analytical approaches capture distinct facets of selective pressure and population structure, which may warrant specialized investigation (SCoT band numbers were derived from Alaeifar et al.,³².

In summary, the combined approach confirmed a core set of genetic loci likely involved in adaptation, providing robust targets for future functional validation.

Phenotypic plasticity, genetic plasticity, and Spatial structuring

The P. aquilinum populations exhibit critical phenotypic and genetic plasticity, a mechanism likely enabling adaptation across heterogeneous environments (Fig. 5). The observed co-occurrence of phenotypic and genetic plasticity across P. aquilinum geographical populations (Fig. 5) suggests a crucial mechanism for the species’ persistence. Specifically, the plasticity observed in characters previously identified as having adaptive potential is essential for populations to rapidly cope with localized environmental changes. This phenotypic variation appears to facilitate local adaptive responses across the landscape, a hypothesis that is strongly supported by the subsequent spatial analyses.

Fig. 5.

Representative spatial distribution maps illustrating phenotypic plasticity across P. aquilinum populations. Each map displays the geographical distribution of a specific morphological trait (M1-M7 and genetic diversity). Each data point on the map represents a sampled geographical population. The color gradient of these points (as detailed in the map legend) indicates the local magnitude of the trait value. Warmer colors (yellow) signify a higher value. The numerical value below each map (e.g., plasticity = 0.99) represents the overall phenotypic plasticity index (PPI) calculated for that trait across all studied populations. This PPI value is a single measure of the trait’s total capacity for plastic response. Full trait names (M1-M7) are available in (Table 1).

Spatial Principal Component Analysis (sPCA) confirmed that genetic and morphological diversity are shaped by both broad-scale geographic clines and fine-scale local factors. The sPCA analysis of both genetic and morphological characteristics (Figs. 6 and 7) yielded highly significant results after 999 permutations (p = 0.02). This analysis effectively partitioned the spatial variance into two key components:

Fig. 6.

Spatial principal component analysis (sPCA) plots illustrating the spatial structure of genetic data. (A) Connection Network Map showing the spatial proximity and relationships among the sampled populations (represented by points) used to define the spatial weights for the sPCA analysis. (B) Genetic Cline Map (Spatial Visualization) showing the continuous global spatial variation in genetic data. This map represents the primary global genetic structure (most significant positive Eigenvalue). The red contour lines (isoclines) indicate the spatial gradient or rate of change in the genetic data. (C,D) Genetic variant maps (spatial visualization) showing the formation of distinct genetic variants (local groupings) within the P. aquilinum distribution area. These panels represent the first two significant spatial eigenvectors that capture the local genetic structure (negative Eigenvalues). The squares in these panels represent the sampled populations, and their color variation shows the magnitude of the genetic score (local genetic variance) for each population along that specific spatial axis. (E) Eigenvalue Bar Plot. The black bars indicate positive eigenvalues (significant global structure) and the grey/ white bars indicate negative eigenvalues (significant local structure). (F) Eigenvalue Decomposition Plot showing the relationship between spatial autocorrelation (I) and variance. The overall sPCA model was significant (p-value = 0.02).

Fig. 7.

Representative spatial principal component analysis (sPCA) plots illustrating the spatial structure of morphological data in P. aquilinum populations. (A) Connection Network Map showing the spatial proximity and relationships among the sampled populations (represented by squares) used to define the spatial weights for the sPCA analysis. (B) Morphological cline map (spatial visualization) showing the continuous global spatial variation in morphological data. This map represents the primary global morphological structure (most significant positive Eigenvalue). The red contour lines (isoclines) indicate the spatial gradient or rate of change in the morphological data. (C,D) Morphological variant maps (spatial visualization) showing the formation of distinct morphological variants (local groupings) within the P. aquilinum distribution area. These panels represent the first two significant spatial eigenvectors that capture the local morphological structure (negative Eigenvalues). The squares in these panels represent the sampled populations, and their color variation shows the magnitude of the morphological score (local morphological variance) for each population along that specific spatial axis. (E) Eigenvalue Bar Plot. The black bars indicate positive eigenvalues (significant global structure) and the white/ grey bars indicate negative eigenvalues (significant local structure), with the negative eigenvalues showing a stronger influence as noted by their magnitude. (F) Eigenvalue decomposition plot showing the relationship between spatial autocorrelation (I) and variance. The overall sPCA model was significant.

Global Spatial Impact: Significant positive eigenvalues indicated the presence of a global spatial structure. This suggests that large-scale geographic factors, such as longitude or altitude gradients, drive the formation of genetic clines and morphological variants across the entire study area (Figs. 6 and 7).

Local Spatial Effect: Significant negative eigenvalues confirmed a simultaneous, strong local effect. This indicates that finer-scale, localized spatial variables (e.g., microclimate, specific habitat patches) are also influencing both genetic and morphological variation within specific populations. Therefore, the diversity within P. aquilinum populations is not simply random but is structured hierarchically by both global and local spatial factors. This complex spatial structuring explains the maintenance of both the broad genetic diversity and the locally adapted morphological plasticity observed.

Populations response to climate change

The analysis of Random Regression Mixed Models (RRMM) reveals significant population-level divergence in morphological responses to climate shifts, indicating variable adaptive capacities (Fig. 8). The RRMM analysis, treating populations as a random effect, demonstrated significant morphological differentiation in response to climate shifts (Fig. 8). This result moves beyond a simple global response, highlighting that not all populations respond uniformly to changing environmental conditions.

Fig. 8.

Representative plots of RRMM analyses for morphological characteristics in P. aquilinum populations, showing response to simulated climate shift. Left Panels (Per Trait): Scatter plots showing the relationship between the future predicted environmental scenario (X-axis: e.g., Future Temperature + Future Humidity) and the predicted change in morphological trait value (Y-axis: e.g., Future score). Each colored point represents a different population (refer to the accompanying legend). Right Panels (Per Population): Plots illustrating the random effect value of each population (Population No. 1 to 11) for the corresponding morphological trait. Positive values on the X-axis indicate a positive deviation (increase) from the mean population response, while negative values indicate a negative deviation (decrease). This highlights which populations show the strongest phenotypic plasticity in response to the climate shift.

Evidence of Divergent Responses: The random effect values for key morphological traits exhibited both substantial positive and negative deviations from the overall mean population value across the geographic range. This divergence signifies that populations are following distinct evolutionary trajectories. For example, in the case of stipe indumentum (a trait critical for water retention), populations No. 1, 2, 4, 9, and 11 exhibited a slight positive deviation above the mean. This positive increase suggests that these populations have plastic or adaptive mechanisms that amplify the trait under climate pressure. Conversely, the remaining populations showed a decrease in this trait. This finding is critical, as it suggests these populations may either be maladapted to the climate impact or be under pressure from different, localized environmental variables. Similar patterns of heterogeneous response (positive vs. negative deviation from the mean) were consistently observed for other key morphological traits, including pinnule shape, pinnae number, and pinnule size. The differential nature of these shifts confirms that P. aquilinum possesses highly variable adaptive potential across its range, with some populations better positioned to cope with climate change impacts than others (Fig. 8).

Sensitivity, exposure, and vulnerability

The overall vulnerability assessment reveals significant heterogeneity in risk profiles across P. aquilinum populations, identifying specific units of high conservation concern (Fig. 9). Analysis of the vulnerability components revealed differential population responses to environmental change (Fig. 9A, B). Notably, populations 1 and 8 exhibited relatively lower estimates for both sensitivity and vulnerability. While low sensitivity is generally a positive indicator (suggesting a buffered response to environmental stress), the overall conservation risk must be considered in light of adaptive potential. Despite some low vulnerability estimates, a core group of populations displayed dangerously low adaptive capacity, placing them at high risk of decline. The assessment of the Landscape Adaptive Index (LAI), which integrates multiple factors related to the ability to cope with change (Fig. 9C), identified a key vulnerability. Populations 2, 3, 7, and 8 showed the lowest estimates for both adaptive capacity and the LAI. Low adaptive capacity implies that even if current sensitivity is low, these populations lack the genetic or ecological flexibility to respond to future, severe climate changes. This low LAI directly translates to a significantly higher risk of extinction or severe population size reduction for these four populations. This interpretation is strongly supported and confirmed by the elevated risk scores specifically estimated for populations 2, 3, 7, and 8 (Fig. 9D).

Fig. 9.

Assessment of sensitivity, exposure, vulnerability, and risk across P. aquilinum populations in response to climate change. (A) Comparison of Sensitivity, Exposure, and Vulnerability. Bar plot illustrating the estimated values for the three key metrics across the sampled populations (Pop 1 to Pop 11). (B) Trends in sensitivity, exposure, and vulnerability across populations. Line plot showing the spatial trends and relative differences of the three metrics among populations. (C) Comparison of populations for landscape adaptive index (LAI) based on weighted mean. Box plot showing the distribution and weighted mean estimates of the LAI across the different populations. (D) Population priority based on Risk score after climate change. Bar plot illustrating the calculated risk score for the populations.

Least cost path and genetic corridors

Analysis of genetic corridors identified critical pathways for gene flow, revealing a clustering of connectivity that informs conservation strategies (Fig. 10A–C). Least cost path and genetic corridor analyses successfully delineated the easiest and most plausible routes for gene flow across P. aquilinum populations (Fig. 10A–C). These pathways confirm a strong pattern of spatial connectivity. Gene flows primarily occur within distinct geographic clusters: between populations 2, 4, and 5, 6; and between populations 1 and 7–11. This grouping suggests that the immediate landscape facilitates localized dispersal, while barriers likely impede gene flow between these larger clusters. The adjacent populations within these clusters exhibit a wide range of genetic polymorphism, from 37% to 81.40%. The significant genetic differences detected by AMOVA³² among these populations underscore those mixing individuals from adjacent, connected populations would be highly beneficial for conservation efforts, as it is expected to result in a moderate increase in overall genetic diversity by introducing beneficial alleles.

Fig. 10.

Analysis of spatial connectivity, least cost paths, and genetic corridors across P. aquilinum populations. (A) Least Cost Path Map. The spatial map illustrates the identified easiest and most probable paths for gene flow connecting adjacent populations. The map’s axes are defined by Latitude (X-axis) and Longitude (Y-axis). (B) Connectivity Matrix. This matrix provides the quantitative cost or resistance values between population pairs. The darker colors in the matrix indicate higher potential for gene flow (lower resistance/ cost) between those two populations. (C) Genetic Corridor Map. This map displays the landscape resistance surface and potential corridors for gene flow. The map’s axes are defined by Longitude (X-axis) and Latitude (Y-axis). The white color indicates areas of low resistance suitable for genetic passage (genetic corridors), while the brown color indicates areas of high resistance due to natural barriers. (D,E) Population Connectivity Change Plots. These plots illustrate the magnitude of change in population connectivity based on two different methodologies: D typically shows connectivity change using the Least Cost Distance method, and E shows connectivity change using the Buffered map method. The change in the color of the circles (where warmer colors, like yellow, indicate higher connectivity) represents the intensity of this change.

Change in genetic connectivity due to climate change

The results of the least cost and genetic diversity model used to assess changes in connectivity across P. aquilinum populations are presented in (Fig. 10D, E). The scales beside the maps indicate the magnitude of connectivity change, showing increases or decreases due to climate shifts. Populations 2, 4, 6, and 7 are expected to experience a slight increase in connectivity, likely as a result of climate and environmental changes.

Structural equation modeling (SEM) results

Structural Equation Modeling (SEM) confirmed a significant and complex network of causal relationships among genetic, environmental, and morphological factors influencing P. aquilinum survival and adaptation (Fig. 11A). The path diagram generated by the SEM analysis (Fig. 11A) indicated that all latent factors exerted significant influences on population adaptability and survival, as well as on each other (p < 0.01). The Partial Least Squares SEM (PLS-SEM) model demonstrated a good fit (Fig. 11B), lending high confidence to the complex interplay detected among the variables that drive P. aquilinum population differentiation and survival.

Fig. 11.

Results of the partial least squares structural equation modeling (PLS-SEM) analysis. (A) Path diagram illustrating the interplay among latent factors and variables. The model shows the effects of environmental, genetic, and bioclimatic factors on the tolerance, adaptation, and morphological differentiation of P. aquilinum. (Thin Gray Lines: These lines connect the latent factors (e.g., ‘Bioclim’, ‘Genetic’) to their observed indicator variables (e.g., ‘Temperature’, ‘Humidity’, ‘Genetic diversity’). They represent the measurement model and the factor loadings that define each latent construct. Black Path Lines: These lines represent the structural model and the causal relationships (path coefficients, beta) among the latent factors. Line thickness is directly proportional to the absolute magnitude of the path coefficient (beta value). Line Type: Solid lines represent significant positive associations (P < 0.05); Dashed lines represent significant negative associations (P < 0.05)). (B) Reliability test plot. The plot displays the cross-validated redundancy values, assessing the model’s predictive power and reliability. The Y-axis represents the reliability indices (Rho-A and Rho-C values) for each latent factor, with the dashed blue line indicating the minimum acceptable threshold for reliability. Values above this line confirm a good fit of the constructed model.

The genetic and environmental latent factors were consistently and positively associated with all other latent factors. This result highlights their paramount importance as the fundamental drivers shaping morphological diversification, physiological tolerance, and overall adaptation of P. aquilinum in the studied areas. Despite the general positive trend, the SEM revealed critical, contrasting relationships that suggest biological trade-offs under environmental stress: Tolerance vs. Bioclimate: The bioclimatic factor (Bioclim) showed a significant negative association with plant tolerance (beta= -0.08, p = 0.001), even though Bioclim positively influenced other factors. Genetics vs. Morphology: When comparing the ultimate relationship with adaptation, genetic factors were positively associated, whereas morphological factors were negatively associated. The model further revealed non-uniform relationships within the latent factors, demonstrating nuanced biological responses: Within the morphology factor, the indument variable exhibited a different relationship compared to the pinnae and pinnule sizes, indicating its potentially unique role in stress mitigation. Similarly, morphological adaptation was negatively related to its latent construct, while the other adaptation variables were positive. As anticipated and confirmed by the model, internal variables like higher temperatures correlated with lower humidity levels within the Bioclim factor (which was constructed exclusively using temperature and humidity variables). Longitude and latitude also exhibited differing influences, reflecting their complex roles in structuring genetic and environmental variation across the landscape.

Differential equation modeling (DEM)

The dynamics of P. aquilinum populations were assessed through Differential Equation Modeling (DEM) analysis, utilizing the integrated genetic, environmental, climatic, and morphological datasets. The results are summarized in Fig. 12.

Fig. 12.

Modeling results for population dynamics and genetic diversity spread. The figure presents five quantitative panels illustrating the outcomes from the differential equation modeling (DEM) analysis, based on integrated genetic, environmental, climatic, and morphological data for 11 local populations. (A) Long-term population dynamics (100 Time Steps): Illustrates the overall logistic growth and the predicted carrying capacity (K) achieved by 11 modelled local populations (‘Row 1’ through ‘Row 11’). (B) Short-term population dynamics (0 to 10 Time Steps): Shows the initial fluctuations in population size (N), emphasizing differential stability among the 11 populations in the short term. (C) Population dynamics of high-volatility population (Pop 1): A focused view highlighting the short-term oscillatory behavior for a representative population. (D) Relative growth rate (r) of Population over Time: Plots the change in the relative growth rate, illustrating the differential short-term instability and variability in population performance. (E) Simulated gene diversity spread: Plots the change in mean genetic diversity (Y-axis) over time (X-axis) for 11 distinct populations, showing the divergent genetic response with some populations facing predicted genetic loss.

Population Dynamics and Stability (Figs. 12A,B,D): Long-term simulations (over 100-time steps, Fig. 12A) indicate that all 11 local populations follow a logistic growth pattern, successfully stabilizing near their carrying capacity (K). The estimated K values range, suggesting a general capacity for sustained survival in their current local areas. However, this overall trend masks significant short-term variation. The short-term dynamics (0-to-10-time steps, Fig. 12B) and the analysis of the Relative Growth Rate (r) (Fig. 12D) both show considerable fluctuations and differential responses across the populations, with some (e.g., Population1) exhibiting greater volatility in growth rate (r) compared to others. This highlights that while survival is predicted across the studied time frame, the sensitivity to short-term environmental changes varies substantially among populations. The detailed dynamics of a highly fluctuating example are shown in Fig. 12C. Simulated Genetic Viability (Fig. 12E): Gene Diversity Spread provides a critical insight into the genetic response over time. The analysis reveals a clear divergence in genetic viability among the populations. Those with higher initial genetic diversity (e.g., Pop1, Pop2) are predicted to maintain or slightly increase their diversity. In stark contrast, populations with lower initial diversity (e.g., Pop10, Pop11) show a significant, steady decline in diversity, dropping to negative indices by the end of the simulation. This finding strongly suggests that existing differences in genetic structure will lead to differential long-term persistence in the face of climatic or environmental shifts, even if immediate survival is predicted.

While DEM simulations indicate logistic growth and stabilization near carrying capacity (K) for all populations, integration with previous analyses of adaptive capacity (LAI), sensitivity, and risk scores highlights that some populations (e.g., Pop2, Pop3, Pop7, and Pop8) are more vulnerable. These populations exhibit lower genetic and morphological adaptive potential, higher sensitivity to environmental fluctuations, and elevated risk scores. Therefore, the predicted “survival difficulties” arise from both the intrinsic vulnerability of specific populations and short-term environmental fluctuations captured by DEM, rather than from overall demographic trends alone. This combined approach allows a more comprehensive assessment of long-term population risk in response to climate change.

Discussion

Contrary to the paradigm of high gene flow traditionally assumed for spore-dispersed ferns, our study reveals that P. aquilinum at its Iranian range margin exhibits pronounced spatial genetic and morphological structure, indicating that local adaptation—not dispersal—dictates its response to climate change.

Our analyses demonstrate strong spatial structuring across populations of P. aquilinum in Iran, with both genetic and morphological variation being tightly linked to local environmental conditions^71,72. This pattern suggests that the species, which has a restricted distribution in Iran, follows a “persist-in-place” strategy²⁴, relying on localized adaptive mechanisms rather than range shifts. The Landscape Adaptive Index (LAI) (representing combined climatic and environmental tolerances) and the Adaptive Capacity (integrating genetic and morphological contributions) both exhibited substantial variation among populations, with low-LAI populations showing higher conservation risk scores^73,74, while high-LAI populations represent critical genetic refugia for future conservation planning^73,75.

Multiple lines of evidence from RDA, CCA, sPCA, and Isolation-by-Environment jointly establish that environmental heterogeneity is the primary driver of genetic and morphological divergence in P. aquilinum.

RDA and CCA results showed significant associations between specific genetic loci or key morphological traits and climatic/geographical variables, reinforcing the central role of environmental filtering in shaping divergence²³. Strong genetic structure (AMOVA, Fst) and the IBE patterns detected by sPCA⁷⁵ provide a mechanistic explanation for this differentiation, with drivers such as soil moisture and temperature gradients aligning precisely with adaptive genomic signals. These findings mirror global patterns in fragmented fern populations, where micro-environmental selection outweighs the homogenizing influence of spore dispersal^62,63, and highlight that edge populations of Pteridium (despite the genus’s global ecological breadth) may be highly sensitive to marginal climatic variation^76,77. Similar environmentally driven divergence is reported for other regionally restricted Iranian taxa, especially in the heterogeneous landscapes of the Zagros^78–80, reinforcing the role of environmental filtering as a dominant evolutionary force.

Our results indicate that adaptation in P. aquilinum arises from the combined influences of genetically based local adaptation and high phenotypic plasticity, each contributing at different spatial and functional scales.

Candidate adaptive loci detected through RDA, DAPC, sPCA, and pcadapt provide a robust genetic foundation for local adaptation (Fig. 4), while the very high phenotypic plasticity index (PPI ≈ 0.99; Fig. 5) demonstrates the species’ strong short-term buffering capacity against environmental variation. Spatial structuring in both genetic and morphological traits (sPCA; Figs. 6 and 7) reflects the interplay of global and local selective pressures. SEM results further show that the Environmental latent factor is a major positive driver of overall adaptability, while the negative association between morphological and adaptive latent factors reflects expected evolutionary trade-offs rather than performance constraints^19,20,49. Together, these findings highlight that populations may follow distinct adaptive trajectories depending on their inherent genetic variation and trait-level plasticity^81,82.

Climate-response modeling (RRMM and PLS-SEM) indicates that future climatic shifts will generate population-specific functional trait transitions, producing an adaptive syndrome focused on enhanced water-use efficiency.

RRMM projections show heterogeneous shifts in traits such as trichome density and pinnule margin shape, forming a coordinated drought-tolerance strategy aligned with rising temperatures and declining humidity⁸³. For example, trichome density is predicted to increase markedly in population 1 (consistent with known drought-mitigation mechanisms)^84–86 while population 3 lacks this response, indicating vulnerability. Similarly, transitions toward entire pinnule margins, a known adaptation to warmer and drier conditions^87–90, vary across populations. Reductions in pinnae number in certain populations^91–94 further indicate structural adjustments aimed at reducing evaporative surface area. PLS-SEM clarifies that many of these responses are mediated by environmental factors such as soil moisture, highlighting the causal pathways underlying observed climate-trait interactions^19,20. These heterogeneous, trait-specific trajectories illustrate that only certain populations possess the appropriate plastic or genetic potential to cope with future climate regimes.

Because populations are locally adapted and climate change is accelerating, the predicted decline in future gene flow represents an imminent threat to the persistence of P. aquilinum in Iran.

Our connectivity modeling shows that several populations will experience reduced genetic connectivity and increased bottleneck risk under future climate scenarios. Importantly, under the projected RCP 4.5 conditions, increasing aridity and ongoing habitat fragmentation elevate environmental resistance, directly leading to the predicted decline in future gene flow. This pattern aligns with connectivity declines reported for other geographically restricted plant species facing climate-induced habitat desiccation^95–97. In Iran’s arid and semi-arid regions, where P. aquilinum already has low abundance, specialized microhabitats, and limited dispersal capacity⁹⁸, the combination of climate-driven fragmentation and reduced long-distance spore dispersal poses a substantial conservation concern.

Taken together, our findings emphasize that conserving P. aquilinum requires strategies that simultaneously maintain genetic diversity, preserve locally adapted lineages, and mitigate climate-driven isolation.

Based on genetic structure, LAI profiles, and climate-connectivity projections, we recommend the following conservation actions for Iranian P. aquilinum.

• Prioritizing vulnerable populations (especially populations 7 and 8) for immediate protection due to their low genetic variability and high-risk scores.

• Restoring ecological corridors to enhance local and regional connectivity, coupled with continued monitoring to prevent genetic drift and inbreeding⁶¹.

• Implementing adaptive genetic management, while avoiding maladaptive gene transfers that may induce outbreeding depression; local adaptation must guide assisted gene flow.

• Ex-situ conservation through spore banking and propagation of elite locally adapted genotypes to support vulnerable populations.

Collectively, these actions reflect the urgent need for a targeted, evolutionarily informed conservation strategy tailored to the unique adaptive landscape of P. aquilinum in Iran.

Conclusion

This study provides the first integrative assessment of landscape genetics and climate change vulnerability for the peripheral populations of P. aquilinum in Iran. Our findings conclusively demonstrate that the global resilience often attributed to this cosmopolitan species does not translate to resilience at its marginal, locally-adapted Iranian range limit.

The key findings are: (1) A strong Isolation by Environment (IBE) pattern exists, driven primarily by temperature and humidity, which has led to pronounced genetic and morphological divergence among populations. (2) Adaptation in P. aquilinum relies on a synergistic interplay between phenotypic plasticity (e.g., predicted shifts in trichome density and pinnule shape for water-use efficiency) and adaptive genetic loci. (3) The Landscape Adaptive Index (LAI) and risk modeling reveal a critical conservation paradox: despite global persistence predictions (IUCN Least Concern), the vulnerability at the local population level is extremely high, particularly for those with low genetic variability (Populations 7 and 8). Furthermore, our models predict a concerning decrease in genetic connectivity in several populations dueising environmental resistance, severely limiting long-term evolutionary potential. These results necessitate an action-oriented conservation strategy.

Ultimately, our research underscores the importance of fine-scale, genetic-based vulnerability assessments for widespread species, offering a crucial model for adaptive management under rapidly changing climatic conditions.

Limitations and future directions

Despite its comprehensive and integrative methodological framework, this study has some limitations that should be acknowledged to contextualize the findings and guide future research. First, although the sampling design covers the full known geographical and ecological range of P. aquilinum in northern Iran, the study is inherently limited to a regional scale. Consequently, the observed patterns of genetic structure, environmental associations, and adaptive potential reflect landscape-level processes within the Hyrcanian range and cannot be directly generalized to the species’ global distribution. Broader sampling across other continents and bioclimatic zones would help clarify whether the adaptive signatures detected in Iranian marginal populations reflect local specialization or part of global evolutionary trajectories.

Second, the genetic analyses rely on SCoT markers, which, despite their usefulness in detecting population structure and adaptive associations, provide limited genomic resolution compared with high-density approaches such as RADseq, GBS, or whole-genome resequencing. Genome-wide methods would allow more precise detection of selection signatures, linkage patterns, and the functional relevance of adaptive loci, and would also improve identification of fine-scale demographic processes such as inbreeding, bottlenecks, and historical migration.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

M. Sh. and F. K. Conceptualization of the project, designed the research, analysis and wrote the manuscript and M. A. collected the samples and lab work. All authors reviewed the manuscript.

Data availability

The datasets used and/ or analyzed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.