Genomic Vulnerability of a Threatened Possum in a Biodiversity and Climate Change Hotspot

ABSTRACT

Spatial variation in the magnitude and pace of climate change impacts means some species will be more affected than others. This includes those occurring within global biodiversity hotspots, regions with exceptional concentrations of endemic species that have often been shaped by long‐term climatic stability. Disproportionate levels of human‐induced pressure in these areas are threatening much of the biodiversity values that make these biodiversity hotspots exceptional. Southwest Western Australia is a highly fragmented temperate biodiversity hotspot experiencing rapid climate change and supporting many species of threatened mammals. One Critically Endangered arboreal folivore, the western ringtail possum (ngwayir; Pseudocheirus occidentalis), has contracted to ~10% of its range since European colonisation, has a history of translocation failure, and is being impacted by increasing aridity. Here, we used 9746 SNPs from 411 ngwayir distributed across southwestern Australia to assess climate adaptation and genomic vulnerability to climate change under three climate change scenarios and four time periods. We also compared the degree of adaptive mismatch imposed by a successful and a failed translocation to consider its contribution to translocation success. We found that ngwayir had low genetic diversity and high genomic vulnerability in the northern edge of their range, where threats such as urbanisation exert substantial pressure on populations. Genomic maladaptation did not appear to drive differences in translocation success, while genomic vulnerability scores identified areas most suitable for future translocations in the central southern portion of the species range. Collectively, our analyses reinforce the importance of urbanised areas for the conservation of ngwayir, and we suggest these areas are prioritised for adaptive management initiatives.

1. Introduction

Anthropogenic climate change is taking a multifaceted toll on biodiversity, from genes to biomes (Bellard et al. 2012). The capacity of species to respond to climate change relates to their potential to track suitable habitat via range shifts, and/or to adapt in situ (Aitken et al. 2008; Ruegg et al. 2018). Despite mounting evidence of adaptive responses to climate change amongst several taxonomic groups (e.g., Bi et al. 2019; Waldvogel et al. 2020), the adaptive potential of most species is uncertain (Catullo et al. 2019; Seaborn et al. 2021). As a result, adaptive potential is rarely included in climate vulnerability assessments, resulting in models and assumptions that may not reflect reality (Razgour et al. 2019; Ruegg et al. 2018) and incomplete information for management and conservation decisions.

The climatic stability of a region on an evolutionary timescale influences the capacity of contemporary biodiversity to adapt to rapid climate change (Harrison and Noss 2017; Trew and Maclean 2021). The amount of environmental disturbance a landscape experiences drives the evolution of traits in species to cope with such change; as a result, species from some regions are inherently more vulnerable (Harrison and Noss 2017). These regions include global biodiversity hotspots (Myers et al. 2000), where landscapes have often experienced relatively little change over geological timescales (Trew and Maclean 2021). In these areas, endemic species typically have ecological specialisations and life history traits that are not readily amenable to change (Buckley et al. 2024; Hopper 2009), whether it be through migration or adaptation. Many global biodiversity hotspots are also predicted to be hotspots for adverse impacts from anthropogenic climate change, particularly in drying Mediterranean‐type climates such as the southwest of Australia (Trew and Maclean 2021). Consequently, understanding the adaptive potential of species in biodiversity hotspots, where climate change is a novel yet significant threat, will be particularly informative.

Southwest Australia is a global biodiversity hotspot known for its exceptional floristic diversity and extensive loss of habitats due to land clearing (Hopper and Gioia 2004; Myers et al. 2000). This bioregion has been protected from extreme climate variation for millennia, creating old, climatically buffered mesic refugia that have contributed to prolific species richness and high endemism of both flora and fauna (Hopper 2009; Rix et al. 2015). Typical of biodiversity hotspots, many species are highly threatened, predominantly through habitat loss, impacts from invasive species and changed fire regimes (Monks et al. 2019; Wayne et al. 2017). Small mammals—particularly those in a ‘critical weight range’ of 35–5500 g—have experienced major declines throughout southwestern Australia over the past 50 years (Wayne et al. 2017) and now face one of the highest extinction risks globally driven by climate change (Trew and Maclean 2021; Urban 2015). Climate change in the region is evidenced by an increase in mean temperature of 1°C since 1990 (Department of Water and Environmental Regulation 2021) and a decrease in mean rainfall of 10%–15% since the 1970s (Ali et al. 2012). Under ‘business as usual’ climate change scenarios, temperatures are projected to increase by 4°C by 2100 (Department of Water and Environmental Regulation 2021), and winter rainfall to further decline by 50% (Dey et al. 2019). While some insights are emerging on evolutionary responses of marine species and plants to climate change in this region (e.g., Adam et al. 2022; Ahrens et al. 2020; Vranken et al. 2021), almost nothing is known about the adaptive potential of terrestrial vertebrates (but see Cummins et al. 2019). Such knowledge would greatly benefit fauna translocation planning and implementation, where Western Australia is a global leader (Mitchell et al. 2013; Morris et al. 2015).

The western ringtail possum (Pseudocheirus occidentalis) or ngwayir (pronounced n‐waar‐ear; Abbott 2001) is a small, arboreal marsupial endemic to southwestern Australia. Ngwayir have experienced an approximate 90% range contraction since European colonisation of southwestern Australia in the early 19th century (Jones 2004), with population estimates since 2015 ranging between 3400 and 18,000 mature individuals (Burbidge and Zichy‐Woinarski 2017; Teale et al. 2025). They currently persist within three isolated management zones: the Swan Coastal Plain (northernmost population), Southern Forest (inland population) and South Coast (southernmost population; Figure 1). Ngwayir are locally extinct within the city of Perth (Department of Parks and Wildlife 2017) but persist in other urban areas. Counterintuitively, these urban areas may become important strongholds for the species and hence are of high conservation value despite the risk of urbanisation impacts (Van Helden et al. 2021). Habitats vary across the management zones, ranging from peppermint (Agonis flexuosa) woodlands on the Swan Coastal Plain to tall jarrah ( Eucalyptus marginata ) forest in the Southern Forests and largely coastal heath and dense peppermint, jarrah and marri ( Corymbia calophylla ) forest near the South Coast.

FIGURE 1.

Sites where ngwayir were sampled across southwestern Australia. Numbers represent the number of samples within close proximity, with single dots representing a single sample. Two translocation sites are shown with diamonds: Perup (blue, extant) and Karakamia (green, extirpated), and the dashed line represents the approximate range of the species prior to European colonisation. The boundaries of three ngwayir management zones are shown as solid coloured lines. Grey polygon indicates the extent of the Greater Perth metropolitan area.

Ngwayir are listed as Critically Endangered (IUCN, and in Australia and Western Australia; Burbidge and Zichy‐Woinarski 2017) and are predominantly threatened by invasive predators (cat, Felis catus and fox, Vulpes vulpes ), habitat fragmentation and inappropriate fire regimes (Department of Parks and Wildlife 2017; Wayne et al. 2005, 2006). Extreme temperatures and drought also directly impact ngwayir populations (Moore et al. 2026), resulting in mass mortality events, likely related to consumption of water‐stressed foliage containing toxic compounds that lead to dehydration and fatal bladder stones (Threatened Species Scientific Committee 2018) and a lower capacity to maintain water balance than other marsupial taxa (Cooper et al. 2020). Additionally, a previous range‐wide population genomic study indicated that genetic diversity is low throughout the range and identified up to eight distinct genetic clusters within the three management zones (White et al. 2021). These observations suggest that low genetic diversity and maladaptation under extreme climates may threaten the persistence of ngwayir, but this relationship has not been explicitly explored.

As ngwayir are clearly vulnerable to climate change, they are an ideal species for investigation of adaptive capacity: this could inform actions such as conservation translocations, which have largely been unsuccessful to date. There have been seven wild‐to‐wild translocations of ngwayir since 1991, with all but two translocations failing, likely due to interactions between predation, health, habitat preference and quality and competition (Department of Parks and Wildlife 2017; Palmer et al. 2023). Consequently, translocations are not considered a priority for this species until the reasons for failure are better understood, with current recovery actions directed towards reducing threats for extant populations in situ (Department of Parks and Wildlife 2017). Hence, understanding genomic vulnerability could reveal which parts of ngwayir's range are most vulnerable to climate change from an evolutionary viewpoint (Bay et al. 2018; Capblancq et al. 2020) and could guide targeted mitigation of its impacts.

Here, we sought to understand adaptive potential and vulnerability to climate change in ngwayir using genomic vulnerability mapping, as has recently been performed for plants (de Deus Vidal Junior et al. 2022), birds (Chen et al. 2022; Ericson et al. 2022), fish (Brauer et al. 2023; Booth et al. 2024), and mammals (McLennan et al. 2025, Razgour et al. 2019). We hypothesise that ngwayir vulnerability will be highest at range edges, where the magnitude of climate change is high, and where habitat fragmentation and other pressures are most impacting the species. We first use genotype‐environment associations (GEAs) of genomic data and bioclimatic variables to identify putative local adaptation, and map adaptive variation and gradients across southwestern Australia. We then calculate genomic vulnerability at four different future time points, across three realistic climate projections. We also evaluate potential climatic mismatches between the source and release sites used in ngwayir translocations to explore possible reasons for translocation failure. Collectively, our analyses can inform future conservation strategies for ngwayir across their range.

2. Materials and Methods

2.1. Sample Collection and DNA Sequencing

We aimed to obtain representative samples from within each management zone, as well as from one failed and one successful translocation site. Our dataset consisted of genetic data from 413 ngwayir individuals, of which 325 DNA samples were already sequenced (White et al. 2021). The remaining 88 tissue (ear clip) samples were collected during targeted monitoring or translocations by DBCA and the Australian Wildlife Conservancy (AWC; seven samples), or were collected opportunistically by local community conservation networks, predominantly from roadkill. Samples were well‐distributed across the current range of ngwayir and included individuals from two fenced sanctuaries: Karakamia (located north‐east of Perth, managed by AWC), and Perup (located in the Southern Forest management zone, managed by DBCA; Figure 1). A total of 55 ngwayir were translocated to Karakamia between 1995 and 2012 but failed to persist, with the last individual detected on a camera trap in 2017 (Palmer et al. 2023). Our representative Karakamia samples were collected from individuals during translocations in 2008 (n = 2) and 2012 (n = 5). Perup Sanctuary is one of only two successful, extant translocated populations. While ngwayir were historically present in Perup, they declined to undetectable numbers in many areas in the 2000s. Perup Sanctuary was established in late 2010 as a haven for other threatened and declining small mammal species. Translocations of ngwayir into Perup Sanctuary occurred as part of a mitigation translocation at Busselton Hospital on the Swan Coastal Plain (130 km away; n = 20 animals) in 2012 and from the Upper Warren (< 45 km away; n = 18 animals) in 2016–2017 (Wayne et al. 2023).

DNA was extracted from the tissue samples in the Conservation Genetics Lab of Western Australia's Department of Biodiversity, Conservation and Attractions (DBCA), using the salting‐out method (Sunnucks and Hales 1996). To eliminate RNA contamination, 3 μL 10 mg/mL of RNase was added to the TNES buffer. Genomic DNA was submitted to Diversity Arrays Technology Ltd. (DArT Pty Ltd., Canberra, Australia) for library preparation and reduced representation sequencing using their proprietary DArT‐seq pipeline. Raw demultiplexed sequences were returned for bioinformatic processing.

2.2. Bioinformatics

Bioinformatic processing and SNP calling were performed using the Setonix HPC at the Pawsey Supercomputing Research Centre, Perth. Sequencing quality was assessed using FastQC (Andrews 2010) and MultiQC (Ewels et al. 2016). Trimmomatic (Bolger et al. 2014) was used to remove restriction adaptor sequences and low‐quality (Phred < 20) bases from the start and end of reads, retaining only reads > 50 bp long. Alignment and SNP calling steps followed the GATK recommended pipeline (Van der Auwera et al. 2013). First, trimmed reads were aligned to the publicly available reference genome for P. occidentalis (DNA Zoo: https://www.dnazoo.org/assemblies/Pseudocheirus_occidentalis) using bowtie2 with local alignment and sensitive settings. Aligned reads were sorted and converted from SAM to BAM format using bcftools, before calling SNPs using bcftools mpileup and call commands.

SNP filtering steps followed the order and nature of recently published genotype‐environment studies (e.g., Brauer et al. 2023; Gates et al. 2023) using VCFtools (Danecek et al. 2011) and VCFfilter (Garrison et al. 2022). Indels, multiallelic SNPs, SNPs with > 20% missing data, and loci with a minor allele count < 3 were removed, as were individuals with > 30% missing data. SNP quality filtering was conditioned on read depth (QUAL/DP < 0.20), as described in O'Leary et al. (2018), and SNPs with a mapping quality < 30 and abnormally high depth (> mean depth per locus + 3SD) were removed. A subsampled dataset of only those localities represented by more than ten individuals (six localities) was used to test for deviations from Hardy–Weinberg equilibrium (HWE); SNPs out of HWE in > 50% of these localities were removed using a custom Perl script (GitHub: jpuritz/dDocent). Replicate pairs were used to calculate the genotyping error rate and remove genotype errors using a custom Python script (GitHub: CoBiG2/RAD_Tools). Linkage disequilibrium was calculated using a sliding window of 50 SNPs, moving five SNPs at a time, in Plink (Purcell et al. 2007), removing SNPs with a VIF > 5. To reduce bias introduced by related individuals, we calculated pairwise relatedness using the triadic likelihood estimator (Wang 2007) in the related package (Pew et al. 2015) in R and removed one individual from each pair with r > 0.7. Individual‐level observed SNP heterozygosity of this final filtered dataset was calculated using the gl.report.heterozygosity function in dartR (Gruber et al. 2018). We tested whether observed heterozygosity significantly differed between management zones and translocation sites using pairwise t‐tests with Bonferroni correction, and analysis of variance (ANOVA) with a Tukey's post hoc test using the stats and multcomp R packages (Hothorn et al. 2008).

2.3. Population Structure

We explored population structure across the ngwayir range using a combination of approaches. Firstly, we used a principal coordinate analysis (PCoA) to explore genetic differentiation across all individuals retained after filtering (n = 349) using the R package adegenet (Jombart 2008). We then used a sparse non‐negative matrix factorisation (sNMF) algorithm (Frichot et al. 2014) to estimate individual ancestry coefficients using LEA in R (Frichot and François 2015). We determined the most likely number of ancestry clusters (K) from 1 to 10 based on cross‐entropy, which was used to impute missing genotypes based on ancestry coefficients. Additionally, we used a spatially explicit modelling approach to assess geographic patterns of ancestry using Tess3R (Caye et al. 2016) and individuals with associated GPS coordinates and non‐imputed genotypes (n = 306). We determined the most likely number of ancestry clusters (K) from 1 to 10 based on cross‐entropy and root mean squared error scores using the projected least squares algorithm. We then interpolated the most likely ancestry clusters onto the landscape using kriging.

We further explored patterns of gene flow and identified priority areas for targeted habitat restoration across the ngwayir range using the Effective Estimated Migration Surfaces (EEMS) algorithm as implemented through the gl.run.eems function in dartR. We ran EEMS simulations using 300 demes for 5 million iterations, thinning every 99 iterations and discarding the first 10% of iterations as burn‐in. We visualised gene flow across the landscape using the log10 of effective estimate migration rate (m). We also explicitly tested for patterns of isolation‐by‐distance using a Mantel test of individuals with GPS coordinates using the gl.ibd function in dartR based on individual Euclidean genetic distance and 1000 permutations.

2.4. Environmental Data

Climate data were extracted from WorldClim (https://www.worldclim.org/) for 19 bioclimatic variables under current and future climate scenarios. We used projected climate for four time periods (2021–2040, 2041–2060, 2061–2080, 2081–2100) under three Shared Socioeconomic Pathways (SSPs; SSP126, SSP245 and SSP585) as defined by the Intergovernmental Panel on Climate Change (IPCC). SSPs represent different climate change emission and mitigation scenarios: SSP126 is best‐case, SSP585 is ‘business as usual’ and SSP245 is an intermediate of these (IPCC 2021). We used data projected under the General Circulation Model ACCESS‐CM2, as this model is accurate for the Australian climate (CSIRO and Bureau of Meteorology 2015).

Bioclimatic variables used in subsequent GEAs require environmental data to be uncorrelated (Rellstab et al. 2015). We extracted environmental data for individuals with reliable GPS coordinates that were retained during bioinformatic filtering (n = 332). We selected variables judged to be ecologically relevant and with a Pearson's correlation coefficient < |0.7|. We ensured these variables were significant in the GEA by using forwards selection (ordiR2step function; Blanchet et al. 2008) in the R package vegan (Oksanen et al. 2022). The four bioclimatic variables retained were mean diurnal range (mean of monthly maximum temperature minus minimum temperature; BIO2), maximum temperature of the warmest month (BIO5), minimum temperature of the coldest month (BIO6) and annual precipitation (BIO12).

The 332 samples with environmental data included samples from translocated ngwayir (Perup—46 samples; Karakamia—7 samples). Location of origin GPS coordinates were known for 18 individuals translocated from Busselton (Swan Coastal Plain) to Perup (Southern Forest), and so these samples were included in GEA using their Busselton coordinates. The remainder of the translocated individuals were excluded from GEAs given their GPS coordinates when sampled did not reflect their area of origin. This retained a dataset of 297 samples for GEAs.

2.5. Variance Partitioning

To disentangle the drivers of genetic variation of ngwayir across the landscape, we used Partial Redundancy Analyses (pRDAs) to explore the individual contributions of climate, geography and population structure to explainable genetic variance (Legendre and Legendre 2012). Variance partitioning is particularly important for species with isolation‐by‐distance genetic structure (Capblancq and Forester 2021), as is the case here (see Results). To represent climate, we used the four bioclimatic variables selected earlier (BIO2, BIO5, BIO6 and BIO12). The first two PC axes from the PCoA were used as proxies for neutral population structure, and the latitude and longitude of samples were used to characterise geography.

2.6. Genotype‐Environment Associations

Redundancy Analysis (RDA) is one of the best performing and robust GEA tools for genomic vulnerability analyses due to the multivariate nature reducing the number of false positives of putatively adaptive loci (Capblancq and Forester 2021; Forester et al. 2018). All individuals with GPS coordinates from their original habitat were included in the RDA (297 samples). Using the R package vegan, we performed a pRDA regressing the full SNP dataset (9746 SNPs) against the four bioclimatic variables (BIO2, BIO5, BIO6 and BIO12) to test for associations. The first two PCs from the PCoA were used as a conditioning variable to control for neutral population structure. To identify candidate outlier loci from the pRDA, we used the ‘rdadapt’ function described in Capblancq and Forester (2021), retaining loci with a significant association with at least one bioclimatic variable (p < 0.01 after Bonferroni correction). We then repeated the pRDA using only outlier candidate loci; this ‘adaptively enriched’ model was the basis for genomic vulnerability scores. We extrapolated the adaptively enriched pRDA across the landscape to visualise adaptive turnover and identify where changes in local adaptation were the greatest or most rapid using the ‘adaptive_index’ function from Capblancq and Forester (2021).

2.7. Testing for Adaptive Mismatch in Ngwayir Translocations

Given that the reasons for translocation failure in ngwayir are unclear, we explored the potential role of adaptive mismatch to translocated environments by using the above GEA to model the ‘ideal’ genotypes suited to the translocation location. We calculated a mismatch score between the predicted genotypes for Karakamia and Perup with associated translocated individual genotypes based on the summed differences in adaptive genotypes using a custom R script. To compare mismatch between Perup and Karakamia, a two‐sample t‐test with equal variance was conducted in R. As the subset of ngwayir translocated from Busselton to Perup (18 individuals) had both original and new GPS coordinates, we compared mismatch to their original environment with mismatch to their translocated environment using the same method and tested for significance using a pairwise t‐test.

2.8. Genomic Vulnerability

Each raster cell in the ngwayir range had a projected genotype that reflected the GEA from the adaptively enriched pRDA under the current climate. Using the current GEA as a baseline, genotypes were projected under climate change for three different SSPs and four time periods. The Euclidian distance between the genotypes reflecting current climate and those predicted under future climate is the genomic vulnerability score, that is, the degree of change in adaptive variation associated with the projected amount of climate change. We used the ‘genomic_offset’ function developed by Capblancq and Forester (2021) to calculate genomic vulnerability scores and then mapped these across the ngwayir range to visualise genomic vulnerability across the landscape under each climate change scenario.

We also assessed whether future projections of each bioclimatic variable would occur outside of the current range of conditions experienced by ngwayir. We converted each projected raster (per bioclimatic variable, SSP and time period) into binary format using the minimum and maximum value of each variable across our contemporary samples, assigning cells as either within or outside the current range. All calculations and conversion were performed using the terra v1.9.27 package in R.

3. Results

3.1. Bioinformatics and Filtering

The sequenced, called and aligned SNP dataset contained 850,229 SNPs and indels from 411 individuals. After filtering for high quality loci and individuals, 9746 SNPs from 349 individuals remained (Table S1). The SNP genotyping error rate calculated from replicate samples was 0.037. Missing SNP data averaged 3.6% per individual before genotype imputation. Nearly all pairwise comparisons of observed heterozygosity between management zones and translocation sites were significantly different using both statistical tests (p < 0.05), excluding between Karakamia and the Swan Coastal Plain (p = 1, both tests), and Perup and the Southern Forest (p = 0.394, Tukey's post hoc test). Heterozygosity estimates indicated lower genetic diversity in the Swan Coastal Plain compared to other management zones, with the greatest diversity in the Southern Forest (including individuals translocated to Perup; Figure 2A).

FIGURE 2.

Genetic diversity and structure of ngwayir. (A) Individual observed heterozygosity calculated using 9746 SNPs. Columns and error bars represent mean and 95% confidence intervals per management zone or translocation site, with the number of samples indicated above each column. (B) Principal coordinate analysis (PCoA) of all samples (n = 349). Points are coloured by management zone (circles) or translocation site (diamonds). (C) Interpolated spatial explicit ancestry coefficients using Tess3R and K = 2. Colours indicate interpolated assignment to one cluster (white) or the other (dark grey) and black lines indicate boundaries of bioregions defined in the Interim Biogeographic Regionalisation for Australia dataset (DCCEEW). Coloured points indicate genetic samples and associated management zone. (D) Density plot of Mantel test for isolation‐by‐distance. Cells colours denote density of points, with greater number of points in cells with brighter colours.

3.2. Population Structure

All approaches supported a largely isolation‐by‐distance pattern of population differentiation across the ngwayir range. A mostly linear spread of points and high variance within populations was evident within the PCoA, with some overlap between the three management zones (Figure 2B). Most individuals translocated to Karakamia likely originated from the Swan Coastal Plain management zone, whereas individuals translocated to Perup show mixed ancestry from both the Swan Coastal Plain and Southern Forest management zones. Cross‐validation estimates for both model‐based approaches (sNMF and Tess3R) followed an exponential decay pattern, with the greatest improvement in cross‐validation when increasing from K = 1 to K = 2, typical of isolation‐by‐distance and provide weak evidence for an optimal K of 2 (Figure S1). Spatial interpolation under K = 2 highlighted this gradient of differentiation across the landscape, ranging from the South Coast to the Swan Coastal Plain (Figure 2C). At K = 3, individuals largely separated into their respective management zones, although some Swan Coastal Plain individuals further from the coast and in the southern extreme of the region were more closely aligned to Southern Forest ancestry (Figure S2A). At K = 4, individuals from the Southern Forest population were separated into two separate clusters (Figure S2B). The Mantel test supported an overall pattern of isolation‐by‐distance, showing a statistically significant association between geographic and genetic distance (Figure 2D; Mantel r = 0.54, p < 0.001).

Effective migration rates from EEMS indicated coldspots in gene flow throughout patches of the Swan Coastal Plain, between the Swan Coastal Plain and Southern Forests regions and between the Porongorup Ranges and other South Coast individuals (Figure S3). Hotspots of gene flow within the Swan Coastal Plain were localised to some semi‐urban areas such as Margaret River, Busselton and Bunbury.

3.3. Variance Partitioning

The full model indicated that climate, geography and population structure together explain 18.3% of the total genetic variation in the dataset (Table 1). The independent (or ‘pure’) effects of climate, geography and population structure were each highly significant (p ≤ 0.001). Of these three categories, population structure accounted for the greatest proportion of explainable variance (16.9%), followed by climate (14.8%) and geography (8.4%). The remaining explainable variance was confounded amongst variables (59.9%) and reflects collinearity amongst the predictors that is impossible to separate statistically (Capblancq and Forester 2021).

TABLE 1.

Variance partitioning of climate (clim.), geography (geog.) and population structure (struct.) using Partial Redundancy Analyses (pRDAs).

Partial RDA models	Inertia*	Proportion total variance	P (> F)	Proportion explainable variance
Full model: F ~ clim. + geog. + struct.	355.83	0.183	0.001	1.000
Pure climate: F ~ clim. | (geog. + struct.)	52.84	0.027	0.001	0.148
Pure structure: F ~ struct. | (clim. + geog.)	60.20	0.031	0.001	0.169
Pure geography: F ~ geog. | (clim. + struct.)	29.72	0.015	0.001	0.084
Confounded climate/structure/geography	213.07	0.109		0.599
Total unexplained	1593.10	0.820
Total inertia	1948.93

^*Inertia values are analogous to variance. Significant p values (p < 0.05) appear in bold.

3.4. Genotype‐Environment Associations

The pRDA controlling for neutral population structure was a significant model (p ≤ 0.001). RDA1 accounted for 40.5% of the explainable variation and was most strongly correlated with BIO2 (mean monthly diurnal temperature range; R ² = 0.41). The second and third RDA axes accounted for 22.9% and 22.5% of the explainable variation, respectively. RDA2 was most closely associated with BIO12 (annual precipitation; R ² = 0.66), and BIO2 was the most influential for RDA3 (R ² = 0.27). Across all three RDA axes, a total of 481 SNPs were identified as putative outliers (Figure 3).

FIGURE 3.

(A–B) Putatively adaptive outliers identified from the pRDA are shown in orange against a background of neutral loci shown in grey. Arrows show the direction and strength of correlation between all loci and bioclimatic variables (BIO2, BIO5, BIO6, BIO12).

The adaptively enriched pRDA using only candidate loci was also highly significant (p ≤ 0.001; Figure S4). RDA1 accounted for 54.4% of the explainable variation and was most strongly associated with BIO5 (maximum temperature of the warmest month; R ² = −0.85) and BIO6 (minimum temperature of the coldest month; R ² = −0.81). RDA2 and RDA3 accounted for 21.1% and 15.7% of explainable variation, respectively, and were both most strongly correlated with BIO2 (R ² = −0.63 and R ² = −0.76, respectively).

Maps of adaptive indices across the landscape highlighted adaptive gradients in response to bioclimatic variables associated with RDA axes (Figure 4). RDA1 was associated with BIO5 and BIO6, largely differentiating coastal and inland areas and reflecting the more stable and less extreme temperatures of coastal regions. RDA2 was associated with BIO2 and BIO12 and differentiated northern coastal and inland areas from southern coastal and high rainfall areas. RDA3 was also associated with BIO2 and showed similar spatial patterns to RDA2.

FIGURE 4.

Adaptive turnover across the ngwayir range. Changes in colour represent different positions along an adaptive gradient—darker colours (purple) and lighter colours (yellow) indicate adaptation to opposite ends of an environmental gradient. Masking indicates the approximate distribution of ngwayir before European colonisation.

3.5. Translocation Adaptive Mismatch

Adaptive mismatch scores did not significantly differ between ngwayir translocated to Karakamia or Perup (p = 0.20). A sub‐analysis of Busselton ngwayir translocated to Perup revealed they were significantly mismatched to their new environment (p < 0.001; Figure 5B), with their mean mismatch score for Busselton (252) increasing to 319.

FIGURE 5.

(A) Mismatch scores for ngwayir translocated to either Karakamia (northeast of Perth, Swan Coastal Plain) or Perup (Southern Forest). Diamonds represent individuals translocated from Busselton to Perup, while circles represent individuals of unknown provenance. (B) Mismatch scores for the subset of 18 translocated individuals of known provenance, comparing their genomic mismatch to their source environment (Busselton) and translocation site (Perup).

3.6. Genomic Vulnerability

Mapped projections of genomic vulnerability across the landscape showed that vulnerability varies across time and space (Figure 6). Across all three climate change scenarios, vulnerability is projected to increase to some degree with time. Under the mildest climate change scenario, SSP126, genomic vulnerability is relatively low. Under the most extreme climate change pathway (SSP585), genomic vulnerability is higher, particularly in the north‐eastern edge of the range. The pattern of higher genomic vulnerability in the northern range edge compared to southern regions was consistent across all SSPs.

FIGURE 6.

Maps of genomic vulnerability under four time periods and three shared socioeconomic pathways leading to different severity of climate change. Columns represent four time periods (2021–2040, 2041–2060, 2061–2080, 2081–2100) and rows represent Shared Socioeconomic Pathways (SSP126, SSP245, SSP585). Darker colours (red) represent higher genomic vulnerability, and scores are relative across scenarios. Masking indicates the approximate distribution of ngwayir before European colonisation.

Across most projections, most bioclimatic variables were predicted to remain within the range of conditions experienced by ngwayir throughout their distribution (Figures S5–S7). However, in the northern Swan Coastal Plain both bio5 (maximum temperature of the warmest month) and bio6 (minimum temperature of the coldest month) were predicted to be outside the range of conditions currently experienced by ngwayir anywhere, with greater extent and rate of novel conditions under more extreme climate change scenarios.

4. Discussion

Given the pace of climate change globally, and especially in southwestern Australia, it is critical to understand evolutionary potential and the ability of species such as ngwayir to adapt. We found that ngwayir exhibit putative adaptive variation in response to at least four climatic variables, and we could extrapolate these relationships over time and space to determine indices of adaptive turnover and vulnerability to climate change. Our results show that genomic vulnerability will increase in the future across all climate change scenarios, with ngwayir from northern regions (primarily the Swan Coastal Plain) being more vulnerable than those from southern regions. However, we did not find evidence that adaptive mismatch linked to climate could explain variation in the success of ngwayir translocations. Realistically, management of current threats to existing populations is the best way to support ngwayir surviving longer‐term threats such as climate change. However, our genomic vulnerability assessment can now be used to guide managers aiming to bolster ngwayir survival during extreme weather events, and those planning translocations.

4.1. Population Structure and Diversity

Both model‐free (PCoA) and model‐based (Tess3R, Mantel test) analyses suggested that ngwayir exhibit a largely isolation‐by‐distance pattern of genetic differentiation across their range, as was shown in previous population genetic analyses (White et al. 2021). The large range contraction of ngwayir (~90% loss; Jones 2004) and their split into isolated management zones would be expected to be reflected in elevated population structure (Rogan et al. 2023). However, the relatively limited structure we detected suggests that, despite short generation lengths in ngwayir (Burbidge and Zichy‐Woinarski 2017; Wayne et al. 2005), the effects of recent habitat fragmentation are lagging (Pinto et al. 2024).

Across the three management zones, the significantly lower individual observed heterozygosity in the Swan Coastal Plain management zone relative to the Southern Forest and South Coast zones likely reflects two drivers. First, this zone encompasses the trailing edge of the ngwayir's northern distribution; and second, this zone is subject to ongoing extensive habitat clearing for urban infrastructure that began after European colonisation in the early 19th century. The Swan Coastal Plain is the most heavily human‐populated zone, with larger cities, including Perth and Bunbury, and smaller regional towns intersecting with habitat used by ngwayir. Continued habitat fragmentation in this region could further impact observed heterozygosity by causing genetic isolation, potentially leading to genetic bottlenecks and erosion—as highlighted by coldspots of gene flow identified in our landscape genomics analyses throughout the region.

4.2. Climate as a Driver of Local Adaptation

While the historically widespread distribution of ngwayir (de Tores 2000) suggests a degree of adaptability, our analyses suggested that there has been local adaptation in ngwayir, particularly in response to temperature and rainfall. We found 481 loci putatively under selection in response to one of four bioclimatic variables, and adaptive gradients primarily separated either inland and coastal areas or higher and lower rainfall areas. Southwestern Australia is characterised by a predominantly Mediterranean climate, but local variation in rainfall and temperature (e.g., mesic refugia) has led to a proliferation of biodiversity (Hopper and Gioia 2004; Rix et al. 2015). Our results suggest these environmental differences have shaped adaptive turnover across the ngwayir's range. For example, RDA1 shows adaptive turnover between coastal and inland areas, reflecting habitat differences between the Swan Coastal Plain (a climate‐stable coastal, peppermint‐dominated region) and the Southern Forest (cooler, taller, jarrah‐dominated forest). In RDA2, adaptive turnover was associated with the landscape transition from a coastal ‘higher rainfall province’ (600–1500 mm annual rainfall) and the further inland and drier ‘transitional rainfall province’ (300–600 mm annual rainfall; Hopper and Gioia 2004). An adaptive response to changes in rainfall would logically correspond with the need for ngwayir to obtain water from foliage to maintain water balance, being a heat‐sensitive species (Cooper et al. 2020).

4.3. Spatial Variation in Vulnerability to Climate Change

Globally, local extinction is highest where the magnitude of climate change is greatest (Román‐Palacios and Wiens 2020). Southwestern Australia is projected to experience the largest contraction of Mediterranean climate worldwide under climate change (Klausmeyer and Shaw 2009): under SSP585, rainfall is predicted to decline by up to 45% and days above 35°C will increase by 125% by 2090 (Department of Water and Environmental Regulation 2021). As a result, the future climate of southern areas will likely resemble the current climate further north, with more frequent and intense hot days, less winter rainfall and more extreme weather events (Department of Water and Environmental Regulation 2021). These spatial patterns likely underpin our predictions of greater climate vulnerability in the northern extent of the range, with adaptive pressure increasing under more severe climate change scenarios.

The impacts of climate change have already been observed or projected in plant species from southwestern Australia, particularly in more northern regions (Klausmeyer and Shaw 2009; Monks et al. 2019). A similar pattern is seen in ngwayir, with the northern and inland extents of the range, including most of the Swan Coastal Plain management zone, projected to be most severely impacted, particularly under SSP585. The higher vulnerability projected under SSP585 suggests that there may be limited evolutionary potential and possible maladaptation in these populations. Areas such as Bunbury, Dwellingup State Forest, Harvey and Collie may become unsuitable for ngwayir, leading to a southwards contraction of the species' distribution. As ngwayir are heat‐sensitive (Cooper et al. 2020), elevated heat‐related stress and mortality are also expected to occur. For example, heatwaves and an exceptionally dry 2023–2024 summer (80%–99% below average summer rainfall in the Greater Perth region; Bureau of Meteorology 2024) resulted in widespread drought and death of habitat‐forming vegetation (described in Fontaine et al. 2024), leading to ngwayir mortality throughout the Swan Coastal Plain (D. Huxley, pers. comm., 10 September 2025). The ubiquitous pressure of climate change will likely exacerbate and interact with other existing threats, such as feral predators, altered fire regimes and habitat loss (Department of Parks and Wildlife 2017). Accordingly, protecting extant populations in situ across intact, natural, fragmented and urban habitats remains critical, particularly given that translocations are not currently recommended for this species.

Occasionally, climate change will create novel ecosystems that accommodate range shifts (Hobbs et al. 2009), with poleward contractions occurring in numerous taxa around the globe (Pecl et al. 2017; Wiens 2016). However, contraction to cooler climates is often impossible because suitable habitat is either absent or disconnected (Trew and Maclean 2021). Ngwayir face both challenges: for example, individuals in the South Coast management zone cannot move poleward because no terrestrial habitat exists in that direction, while fragmented landscapes and the limited mobility of ngwayir (Yokochi et al. 2015), particularly throughout the Swan Coastal Plain (as indicated by our estimated migration rates), will also prevent southward migration without human assistance.

The Swan Coastal Plain is also highly urbanised (Hiller et al. 2013). While conservation is needed across all areas where ngwayir persist, contraction to residential areas (Wintle et al. 2019) presents additional conservation challenges. Urban areas, as seen in the South Coast management zone (Van Helden et al. 2021), may provide alternative food sources but also increase threats such as roadkill or predation by domestic dogs and cats (Van Helden et al. 2020). These patterns highlight opportunities for urban conservation alongside continued management of threats in natural habitats. Based on the genomic vulnerability and connectivity patterns identified here, management actions such as providing cool shelter sites, irrigating vegetation to maintain foliage water content and creating possum‐friendly gardens and streetscapes could be prioritised, particularly in the Swan Coastal Plain. Such approaches have already shown to be effective around Albany in the South Coast management zone (Van Helden et al. 2021).

4.4. Implications for Translocations

Conservation translocations and establishment of feral predator‐free sanctuaries (havens) are common practice in Australia due to the unprecedented decline in native mammals (Legge et al. 2018; Woinarski et al. 2015). While havens effectively protect species from feral predators, they do not eliminate other challenges, such as adapting to novel environments (Bubac et al. 2019). Potential maladaptation following translocation has received little attention in southwestern Australian fauna, with the notable exception of the Western Swamp Tortoise (Mitchell et al. 2013, 2016). However, the genetic implications of translocations, particularly local adaptation and outbreeding depression, are increasingly recognised (e.g., Nistelberger et al. 2023; Rick et al. 2023; Weeks et al. 2011; White et al. 2020). Here, individuals translocated from both the Busselton (Swan Coastal Plain management zone) to Perup Sanctuary (Southern Forest management zone) translocation and the translocation into Karakamia were significantly genomically mismatched to their new environment.

The observed mismatch is unsurprising given that both translocations crossed different climatic regions. Although we could not infer a precise location of origin for the Karakamia individuals, genetic ancestry indicated they likely originated in the Swan Coastal Plain, but of uncertain location. This translocation would thus have involved a transition to slightly hotter and drier climates than their source, whereas the translocation from Busselton to Perup Sanctuary resulted in a transition to cooler, wetter forest—these contrasting directions of climatic mismatch may have contributed to their contrasting outcomes (one success, one failure). Regardless, our results indicate that adaptive mismatch alone was unlikely to determine translocation success. Previous translocations have generally failed despite considerable management, with failures attributed to interacting effects of predation, habitat quality, competition and health (Department of Parks and Wildlife 2017; Palmer et al. 2023). For example, the initial 2012 Perup release experienced low survivorship (< 30% of collared animals surviving for more than 4 months), primarily due to predation by native birds of prey and altered behaviour, including animals frequently resting on the ground instead of in hollows (Wayne et al. 2023). In contrast, the 2016–2017 translocation achieved higher survivorship, successful recruitment and representative ancestry (and admixture) from both the Swan Coastal Plain and Southern Forest individuals (Wayne et al. 2023). Recent monitoring suggests that the Perup population has likely reached carrying capacity (500–600 animals; Wayne et al., in prep.), with the translocation generally achieving all pre‐defined success criteria (> 50% of animals surviving > 1 month post‐release; > 30% of animals surviving > 12 months post‐release, except for the Busselton cohort; evidence of females producing young within 24 months of release). Together, these findings indicate short‐term dynamics are more likely to affect translocation success of ngwayir than long‐term climatic maladaptation.

Nevertheless, the similar mismatch scores between Perup and Karakamia indicate that adaptive mismatch remains relevant to translocation planning. Karakamia lies near the northern limit of the pre‐European ngwayir range, where genomic vulnerability is predicted to be greatest. Combined with frequent heatwaves and drought in the Greater Perth region (BOM 2024), this suggests that Karakamia may now be climatically unsuitable for ngwayir and that thorough assessment of local site suitability should precede any future translocations. While several factors may contribute to translocation failure before adaptive mismatch becomes a concern (Berger‐Tal et al. 2020; Bubac et al. 2019), incorporating adaptive mismatch into planning could improve long‐term outcomes by identifying individuals best suited to recipient environments (Aitken and Bemmels 2016; Booth et al. 2024).

Patterns of adaptation and genomic vulnerability may also inform population restoration efforts. Historically, biodiversity in southwestern Australia has contracted to and expanded from mesic refugia during periods of climatic change (Byrne 2008; Rix et al. 2015; Trew and Maclean 2021). Under SSP585, the region between Margaret River and Walpole‐Denmark is predicted to have the lowest genomic vulnerability, coinciding with historical mesic refugia (Rix et al. 2015) and the ‘higher rainfall province’ (Hopper and Gioia 2004). This area was also predicted to remain suitable for ngwayir under future climate scenarios using species distribution modelling (Molloy et al. 2014). Consequently, if additional insurance populations are established, this region appears to offer the greatest climatic suitability and adaptive potential. Based on current and future climatic conditions as modelled here, founding individuals for a potential translocation should consider being sourced from the drier and warmer regions to the north rather than the climatically cooler South Coast population. However, despite sporadic reports of ngwayir from Walpole to Denmark (Department of Parks and Wildlife 2017), further investigation is needed to assess ecological suitability and identify threatening processes that could compromise translocation success.

4.5. Future Directions

To validate the use of genomic vulnerability and adaptive mismatch for ngwayir, and potentially other species in the southwestern Australian biodiversity hotspot, future studies should verify adaptive indices and putative local adaptation. However, it is important to note that the genomic vulnerability approach utilised here can only provide a relative score of vulnerability and further ground‐truthing is required to operationalise estimates (Lotterhos 2024). Fitness‐related data, such as reproductive success, would help with this ground‐truthing, while annotation of putatively adaptive genes identified through our GEA could provide insight into the genetic basis of adaptation and strengthen links between genomic vulnerability and climate adaptation, although such associations remain tentative (Fitzpatrick and Keller 2015; Ruegg et al. 2018). This annotation would also require the development of further genomic resources for the species, as the most closely related lineage with an annotated transcriptome is the brushtail possum ( Trichosurus vulpecula ), to which ngwayir last shared an ancestor 25–30 million years ago (Ruedas and Morales 2005).

Genomic vulnerability is an emerging tool that has been integrated with ecological niche modelling (Chen et al. 2022; Razgour et al. 2018, 2019), hybridisation (Brauer et al. 2023), and historical population declines (Turbek et al. 2023) to improve conservation assessments. Consistent with the original rationale for incorporating adaptation into climate risk assessments (Razgour et al. 2018), genomic vulnerability should be considered alongside other measures of species viability (DeSaix et al. 2022). Integrating genomics with physiological data, species distribution or niche modelling, predation, disease and threats from extreme events would provide a more comprehensive assessment of vulnerability. For ngwayir, much of this information is already available (Cooper et al. 2020; Molloy et al. 2014; Wayne et al. 2023), making an integrated vulnerability assessment a research priority.

More broadly, genomic vulnerability represents only one component of realised vulnerability. It is a relatively static metric that does not account for evolutionary processes such as phenotypic plasticity, genetic drift, gene flow, mutation, inbreeding, or changes in population size (DeSaix et al. 2022). It is also important to distinguish between chronic (‘press’) and acute (‘pulse’) climate stressors (Harris et al. 2018). While press forces drive gradual changes in temperature and rainfall, pulse events such as droughts, heatwaves and bushfires can cause immediate mortality. Managing these pulse threats should remain the priority for ngwayir conservation, particularly given the species' limited dispersal capacity (Yokochi et al. 2015). If populations cannot tolerate pulse events, local extinctions may occur before adaptation to long‐term climate change is realised. Further research into the genetic basis of tolerance to extreme heat and drought, including phenotypic plasticity and gene expression, would therefore complement genomic vulnerability analyses. Although genomic vulnerability does not capture immediate extinction risk, it remains an important component of long‐term conservation planning.

5. Conclusions

This study contributes to a growing body of literature demonstrating that genomic vulnerability is relatively high in populations that are collapsing, reinforcing climate change as a real and serious threat to biodiversity (Bay et al. 2018; Ruegg et al. 2018). Analysis of genomic vulnerability of ngwayir shows that populations on the Swan Coastal Plain are the most threatened by climate change and need additional conservation effort, particularly when considered in combination with existing threats in this highly altered and extensively urbanised management zone. Integrated management of existing threats, which likely interact synergistically with climate forces, is needed across the entire ngwayir range to improve population resilience to climate change. For example, reduction of predation by invasive species and improving habitat quality could support larger populations that are more resilient to stochastic forces and have higher potential to adapt to climate change. If climate change continues apace, without threat reduction ngwayir are likely to undergo substantial population declines and risk local extinctions across their range. Nevertheless, adaptive gradients and genomic vulnerability analyses suggest that the projected future climate between Margaret River and Denmark may provide a suitable climatic refugium, provided that suitable habitat and ecological conditions persist there and that other major threats are mitigated.

Author Contributions

Grace I. Marsh contributed to all sections of data analysis and drafting of the manuscript. Sean James Buckley designed and supervised the study, obtained resources, contributed to population genomics analyses and assisted with manuscript writing. Daniel J. White collected the original tissue samples from across Southwest Western Australia and extracted DNA for sequencing (n = 325), providing the genomic data as well as guidance around project design and supervising this study. Adrian F. Wayne provided additional samples (Perup) and provided guidance around implications and lessons from conservation management actions. Bryony J. Palmer provided samples (Karakamia) and provided guidance around implications and lessons from conservation management actions. Nicola J. Mitchell designed and supervised the study and assisted with manuscript drafting. All authors contributed to the interpretation of results, writing of the manuscript and critically revised the manuscript.

Funding

Genomic sequences for this project were supported by the Threatened Species Recovery Hub of the Australian Government's National Environmental Science Program (NESP) to DW.

Ethics Statement

No live animals were handled as part of this study. Samples were sourced from previous research and monitoring projects, translocations and through opportunistic sampling of deceased animals (largely roadkill). Appropriate ethics approvals were granted to the Western Australian Department of Biodiversity, Conservation and Attractions (DBCA); formerly Department of Environment and Conservation (DEC) or Department of Parks and Wildlife (DPaW) or Australian Wildlife Conservancy (AWC) during these previous activities. Initial translocation and subsequent monitoring and sampling of the Perup Sanctuary animals were conducted under DEC AEC 2012/20, DPAW AEC 2014/09, DPAW AEC 2015/45, DPAW AEC 2015/32, DBCA AEC 2018/22F and DBCA AEC 2021/25G. Opportunistic tissue sampling and cage trapping in the Southern Forest were conducted under DEC AEC 2009/51, DEC AEC 2009/52, DEC AEC 2012/56, DEC AEC 2012/57, DPAW AEC 2015/43 and DBCA AEC 2022/33D.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Record of SNPs and individuals retained at each stage of filtering.

Figure S1: Cross‐validation estimates for Tess3R followed an exponential decay pattern, with the greatest improvement in cross‐validation when increasing from K = 1 to K = 2, typical of isolation‐by‐distance.

Figure S2: Interpolated spatial explicit ancestry coefficients using Tess3R and (A) K = 3 and (B) K = 4. Colours indicate assignment to a cluster, and black lines indicate boundaries of bioregions defined in the Interim Biogeographic Regionalisation for Australia dataset (DCCEEW). Coloured points indicate genetic samples and associated management zone.

Figure S3: Effective migration rate (m) across the ngwayir range in log10 scale using the Estimating Effective Migration Surfaces (EEMS) algorithm. Brighter (yellow) colours indicate regions of higher gene flow and darker (purple) colours indicate regions of reduced gene flow. Black lines indicate boundaries of bioregions defined in the Interim Biogeographic Regionalisation for Australia dataset (DCCEEW). Coloured points indicate genetic samples and associated management zone.

Figure S4: Putatively adaptive outliers identified from the adaptively enriched pRDA. Arrows show the direction and strength of adaptive loci with bioclimatic variables.

Figure S5: Projection of bioclimatic variables and their relationship to the current range of conditions experienced by ngwayir under SSP126. Binary maps were based on the minimum and maximum values of current samples for each variable. Regions in red indicate areas outside of the current range of conditions experienced by ngwayir and areas in blue indicate areas within them.

Figure S6: Projection of bioclimatic variables and their relationship to the current range of conditions experienced by ngwayir under SSP245. Binary maps were based on the minimum and maximum values of current samples for each variable. Regions in red indicate areas outside of the current range of conditions experienced by ngwayir and areas in blue indicate areas within them.

Figure S7: Projection of bioclimatic variables and their relationship to the current range of conditions experienced by ngwayir under SSP585. Binary maps were based on the minimum and maximum values of current samples for each variable. Regions in red indicate areas outside of the current range of conditions experienced by ngwayir and areas in blue indicate areas within them.

Data Availability Statement

Individual genotype data for population genomic and adaptive genomic analyses, and environmental data used for genotype‐environment association analyses, are available on Figshare (https://doi.org/10.6084/m9.figshare.31524436). Code used for analyses is available in the Github repository https://github.com/ConGenSean/Ngwayir_GenomicVulnerability.