Introduction

Global biodiversity is facing an anthropogenically-induced sixth mass extinction (Ceballos et al. 2015). One of the primary drivers of extinctions is the destruction and fragmentation of habitat, with effects compounded by other threats (Brook et al. 2008). Loss and fragmentation of native habitat typically reduces species’ population sizes and increases their isolation, impacting their genetic structure and diversity as movement of individuals through the landscape becomes limited.

The degree to which fragmentation impacts gene flow is influenced by species’ habitat requirements, their dispersal capability, and the remaining landscape matrix composition (Bender and Fahrig 2005). For example, for habitat generalists that can move through a more heterogeneous surrounding landscape, patch size and isolation are less important predictors of gene flow than for species with specialist habitat requirements that are unable to move through the matrix (Mossman and Waser 2001; Bender and Fahrig 2005). Whilst fragmentation reduces species abundance and gene flow, genetic health within populations can also decline as a result of loss of genetic diversity through inbreeding and genetic drift, with strong evidence these factors increase extinction risk (Spielman et al. 2004; Frankham 2005). Inbreeding may be unavoidable in small populations, with the increased homozygosity of deleterious alleles leading to reduced fitness (Frankham 2005), and decreased genetic diversity resulting in lowered adaptive capacity (Hoffmann et al. 2003; Frankham 2005; Allendorf et al. 2012), further exacerbating extinction risk, i.e. the ‘extinction vortex’ (Blomqvist et al. 2010). Many studies have found smaller and more isolated populations tend to have lower genetic diversity, consistent with theoretical expectations (e.g. Levy et al. 2010; Lino et al. 2019). Given anthropogenic habitat fragmentation is relatively recent in evolutionary time, the impact of reductions of gene flow, and thereby genetic diversity, on population health may not yet be fully realized for many long-lived species (Ewers and Didham 2006).

Conservation translocations are becoming an increasingly important tool in restoring gene flow between anthropogenically fragmented populations (Weeks et al. 2011; Frankham et al. 2017). Translocations include supplementing existing populations with new individuals or reintroducing a population into areas where they are locally extinct (IUCN/SSC 2013). Translocations can have both demographic benefits (increasing population size or distribution) and genetic benefits (increasing diversity or reducing inbreeding; Hufbauer et al. 2015). In small populations, genetic supplementation is often needed to maintain diversity (Lott et al. 2020; Nilsson et al. 2023). Translocations have variable success and can be financially costly, so require careful planning to ensure self-sustaining populations (Fischer and Lindenmayer 2000; Sheean et al. 2012). It is therefore important that genetic information is incorporated into the planning process and post-translocation evaluation to ensure conservation benefits are met.

The kenngoor (Phascogale calura; also known as the red-tailed phascogale), is one of the many Australian terrestrial mammals that has faced dramatic distribution declines since European settlement (Burbidge et al. 2009). The kenngoor is a short-lived (11–36 months) small (~ 40–60 g), semi-arboreal, carnivorous, and nocturnal marsupial (Bradley 1997; Stannard et al. 2010). Once widespread across western and central Australia (Fig. 1), its distribution is now limited to south-west Western Australia, an area less than 1% of its former range (Short and Hide 2011). As a result of these declines, it is currently listed as vulnerable under Australian federal legislation (Environment Protection and Biodiversity Conservation Act 1999), and as conservation-dependent in Western Australia (Biodiversity Conservation Act 2016). The kenngoor’s primary threats include habitat fragmentation, frequent fires, and predation by feral cats (Burbidge and Woinarski 2019), which are highly effective predators of mammals of this size. In the wild, the species persists in highly fragmented eucalypt woodland habitat within an agricultural area that has been extensively cleared (Fig. 1; Saunders 1989; Short and Hide 2011). In this landscape, Short et al. (2011) previously found that patch occupancy was not affected by patch size, and suggested that kenngoor can move across the landscape by utilising remnant roadside and farmland vegetation as ‘stepping stones’ to larger patches, although habitat patches isolated by > 5–10 km from the nearest occupied patch were unlikely to be colonised.

Fig. 1
figure 1

Satellite images of the kenngoor distribution. A The estimated pre-European distribution of the kenngoor (white outlines; following Bradley et al. (2008), with reserves in this study shown as white dots. B A kenngoor (photo: Brad Leue/AWC). C The reserve locations included in this study (white dots in the southwest), which are generally representative of the current distribution of the species. D A close-up satellite image of the area around a single reserve showing the extent of habitat fragmentation, with treed nature reserves showing in dark green, within the lighter green areas cleared for agriculture

Full size image

Here we assess the genetic diversity of remnant wild kenngoor, to inform future management actions. Using reduced-representation sequencing, we assessed the genetic structure and genetic diversity of these remnant locations to determine the impact of habitat fragmentation on gene flow in the species. Following the predictions of Short and Hide (2011), we expected that closely located reserves would be more genetically similar than distant ones. Second, we predicted that geographically isolated locations will show reduced diversity due to inbreeding and genetic drift. Third, we investigated the genetic diversity of a wild-to-wild translocation (Kojonup) 10 years post-translocation, to determine whether additional supplementation may be necessary. Finally, we aimed to estimate global effective population size to provide an indication of future resilience of the species to diversity loss.

Materials and methods

Tissue samples and location information

Ear biopsies (1 mm) from wild-caught kenngoor were collected by the Australian Wildlife Conservancy (AWC), Bush Heritage Australia (BHA), and the Department of Biodiversity, Conservation and Attractions (DBCA) as part of routine monitoring and targeted trapping programs between 2007 and 2021. Survey areas included one farming property (Crossman) and 13 reserve locations within the Wheatbelt region of Western Australia (Fig. 1, Table 1). Collections were made under DBCA Animal Ethics Committee (AEC) approval numbers 2007-13, 2010-33, 2013-47, 2019-08A, 08-004480-1, 08-004480-2 and Queensland AEC permit CA 2018/04/1182. Sample sizes ranged from 1 to 28 between sites and years for a total of 209 samples (Table 1). No information on the true census size of the species for individual reserves is available.

Table 1 Kenngoor genomic DNA sample sizes across collection years and locations
Full size table

Kenngoors at the Kojonup reserve were established by a DBCA mixed-source, wild-to-wild conservation translocation and are managed by Bush Heritage. Founders were sourced from four wild locations in 2010 (12♀; 10♂) and 2011 (5♀; 3♂), for a total founder population size of 30. Source reserves were Boyagin (2010 = 7♀ 3♂; 2011 = 2♀ 1♂), East Yornaning (2010 = 2♀ 1♂; 2011 = 3♀ 2♂), Pingeculling (2010 = 4♂), and Tutanning (2010 = 3♀ 2♂).

DNA extractions and sequencing

DNA extractions were completed at the University of Sydney in 2020 for 74 samples and DBCA in 2021 for 147 samples. A standard salting out procedure was followed (Sunnucks and Hales 1996), with the addition of 3 μL 10 mg/mL RNase to the TNES buffer to remove RNA contamination. Genomic DNA was submitted to Diversity Arrays Technology Pty Ltd (Canberra) for library preparation and DArT-seq 1.0 high density sequencing following proprietary methods outlined in (Kilian et al. 2012; Sansaloni et al. 2011; Table S1). In brief, the Pstl-MspI restriction enzyme combination was used for DNA digestion and all other laboratory steps followed Georges et al. (2018). Demultiplexed raw sequencing data was obtained for 209 successfully sequenced individuals (Table 1; Table S1).

SNP calling and population genetic datasets

Different population genetic metrics and analyses require different approaches to calling and filtering SNPs to maximise accuracy of estimates (Schmidt et al. 2021). We employed varying approaches to SNP calling and filtering to match the theoretical assumptions and expectations of each analysis. These are outlined below and summarized in Table S2. Prior to all analyses, raw DArTseq data was quality checked and cleaned with process_radtags v2.53 to remove barcodes, discard low quality reads, and truncate reads to 68 bp (Catchen et al. 2013). trimmomatic v0.39 (Bolger et al. 2014) was used to quality trim reads and remove Illumina adaptor contamination.

Population structure analyses

To enable assessment of genetic structure and intraspecific population differentiation we utilized SNPs called across all natural populations. A reference-based pipeline in stacks v2.62 (Catchen et al. 2013) was used to assemble sequencing reads and identify single nucleotide polymorphisms (SNPs) with reference to the Antechinus stuartii genome (Brandies et al. 2020). Initially, cleaned reads were aligned to the indexed A. stuartii genome with bwa v0.7.17-r1188 (Li and Durbin 2009), before converting the files to.BAM format with samtools v1.11 (Danecek et al. 2021) to use as input to the stacks ref_map.pl to construct the SNP catalogue. Final filtering parameters are summarized in Table S2, and were implemented in stacks populations and in R v4.1.2 (R Core Team 2021) using custom scripts (Shaw et al. 2022; Wright et al. 2019; this study).

The presence of highly related individuals and uneven sampling can bias genetic clustering analyses (Pew et al. 2015; Puechmaille 2016). Twenty-four highly related individuals (r > 0.4, methodology outlined below) were removed from the dataset (including six presumed duplicates). Highly sampled reserves (Boundain, Kojonup and Pingeculling) were subsampled down to n = 16 while keeping the sex ratio as equal as possible. Preliminary genetic structure analyses showed no clustering by sample collection year within reserves (data not shown), therefore within-reserve samples were pooled across years.

The relationship between genetic and geographic distance was assessed using an isolation-by-distance (IBD) model. IBD analysis was performed first with natural wild locations only, and second, including the Kojonup individuals. We used a mantel test to determine the strength of linear correlation between pairwise population divergence (FST) and straight-line geographic distance (gl.ibd function in dartR package (Gruber et al. 2018)).

Genetic clustering analysis was carried out using R package tess3r (Caye et al. 2016) using custom scripts (Shaw et al. 2022). tess3r analyses patterns of population genetic variation while controlling for spatial autocorrelation. As spatial information is incorporated, we excluded the translocated population due to its mixed source locations. To determine the number of genetic clusters (K), models of K between 1 and 10 were evaluated. Each K was tested with 100 repetitions, and 10% of the data masked to calculate cross-entropy values to determine the optimal value of K. All other parameters used default settings. To further visualise differences between locations, a principal component ordination analysis (PCOA) was conducted using the gl.pcoa.plot function in the dartR package (Gruber et al. 2018).

Genetic differentiation between locations was quantified using pairwise FST (Wright 1949; Weir and Cockerham 1984) and shared alleles. FST is a measure of genetic divergence between populations, and ranges from zero (no genetic differentiation) to one (completely diverged; Allendorf et al. 2012). FST was assessed in the R package stampp (Pembleton et al. 2013) with 1000 bootstraps performed across loci to calculate 95% confidence intervals. To assess similarities between locations, the proportion of shared alleles was calculated with the pairwise.propShared function in the popgenreport R package (Gruber and Adamack 2015). Only locations with > 5 individuals were included in the Fst and shared allele analyses.

Within-population diversity analyses

To assess the impact of habitat fragmentation on genetic diversity, we estimated relatedness, autosomal expected and observed heterozygosity (Schmidt et al. 2021), allelic richness, inbreeding, and private alleles for each location. For relatedness and allelic richness analyses, we used the same SNP dataset as outlined above. For heterozygosity analyses, autosomal estimates are presented (i.e. based on all sites) rather than SNP-based estimates as they are less biased by sample size and polyallelic sites (Schmidt et al. 2021). Raw sequence reads were assembled and polymorphic sites called using the de novo pipeline in stacks V2.62. The de novo pipeline allows for high confidence in heterozygosity estimates due to the ability to set the minimum allele read depth, which is not possible in the reference mapping pipeline. Cleaned reads (above) were processed using stacks denovo_map.pl, allowing three mismatches between stacks within individuals (-M) and between individuals (-n), a minimum allele depth (-m) of eight reads required to create a stack, and singletons were retained. Populations was rerun with filtering parameters set according to the theoretical requirements of each metric (Table S2).

The R package related (Pew et al. 2015) was used to assess within- and between- reserve relatedness, as sampling methods included using nest boxes (in which multiple individuals may shelter during the day) and concentrated trapping within grids. It is unknown whether related individuals tend to nest and/or live closer together compared to non-relatives, and unintentionally biased sampling of related individuals would impact diversity analyses. Relatedness was only assessed within years, as given three-year generation time of females means we were testing for parent/offspring and sibling relationships. The Wang estimator (Wang 2002) was selected to estimate relatedness values as it is more robust to small sample sizes (Wang 2017). The Wang estimator estimates pairwise relatedness (r) between individuals, ranging from zero to one, although negative estimates are possible due to sampling error (Wang 2002). Larger values indicate higher relatedness, with 0.5 indicating parent–offspring or sibling relationships. Where highly related individuals were identified (r > 0.4), one from each pair was removed prior to further analysis. A maximum relatedness threshold of 0.4 was selected to allow for the possibility that high relatedness could be an accurate reflection of fragmented populations while also removing siblings captured in the same trap area.

Diversity metrics were calculated excluding highly related individuals. Individual autosomal observed heterozygosity (HO) and population autosomal diversity metrics (observed and expected heterozygosity, inbreeding coefficients and private alleles) were calculated by stacks populations. For individual estimates, two outliers were removed as previous analyses had identified a low call rate in these individuals (West Ashby samples with observed heterozygosity of 0.00049 and 0.00054). In addition, as allelic variation is more sensitive to bottlenecks than heterozygosity (Allendorf et al. 2012), allelic richness was calculated in the R package hierfstat (Goudet 2005). The allelic.richness function calculates the average number of alleles per loci, standardised for sample size using rarefaction techniques (El Mousadik and Petit 1996; Petit et al. 2008).

A one-way ANOVA was used to test for statistically significant differences in relatedness and individual observed heterozygosity across populations. For relatedness, this was conducted only for years that had multiple locations with sample sizes of five or greater. For observed heterozygosity, it was conducted only for populations with sample sizes of eight or greater (Nazareno et al. 2017). A Bartlett test was used to test the assumption of equal variance, and a heteroskedastic covariance matrix applied if necessary. An alpha level of 0.05 was used for all statistical tests.

Genetic drift analyses

treemix V1.1 (Pickrell et al. 2012; Pickrell and Pritchard 2012a) was used to test for evidence of genetic drift. Given a set of allele frequencies, Treemix generates a maximum likelihood tree by inferring patterns of population splitting based on Gaussian approximations of genetic drift (Pickrell and Pritchard 2012a). Firstly, stacks de novo was rerun from cstacks with four Antechinus argentus samples as the outgroup. Antechinus are in the same subfamily and tribe as Phascogale (Krajewski et al. 2000) and the two form a monophyletic group (Van Dyck 2002). The output was filtered according to the parameters in Table S2. Genetically similar reserves with small sample sizes were merged to maximise sample sizes (Dryandra and Crossman with Contine Block; Quinn’s Block with West Ashby). The translocation to Kojonup reserve was specified to TreeMix as a known migration event (-cor_mig) assuming an equal contribution of the four source locations. Sample size correction was disabled (-noss), as this can overcorrect and underestimate genetic drift (Pickrell and Pritchard 2012b). Bootstrap replicates were generated to judge confidence in tree topology (–bootstrap). To determine the most likely number of migration events, we tested zero to ten migration events (-m) and assessed the best result through the residuals (Pickrell et al. 2012; Pickrell and Pritchard 2012a).

Effective population size

Global effective population size was estimated using the single-sample linkage disequilibrium method (LDNE) assuming random mating (Hill 1981; Waples 2006; Waples and Do 2010; Jones et al. 2016) as implemented in NeEstimator V2.1 (Do et al. 2014). This method is based on the random linkage disequilibrium that exists each generation in a finite population (Hill 1981; Waples and Do 2010). The reference-based pipeline was used as above to generate SNPs, with a minimum minor allele frequency of 0.05 (Table S2; Marandel et al. 2020). Ne was estimated for the two years with the highest sample sizes, 2021 (n = 56) and 2019 (n = 36). Additionally, as there were minimal differences in allele frequencies detected between years, samples from 2019 to 2021 (n = 106) were pooled to maximize sample sizes for a third estimate.

Results

The total number of variant sites after filtering varied among datasets, from 2331 (SNPs in the Ne 2019 dataset) to 9169 (total variant sites in the individual HO dataset) (Table S3). All datasets had more than 35 individuals (36–185), although sample sizes per reserve varied (Table S3).

There was a significant positive IBD relationship between genetic divergence and geographic distance across wild reserves (Fig. 2C), which explained only ~ 12% more variation without the translocated Kojonup population included (without: p = 0.001, R2 = 0.515, with: p = 0.001, R2 = 0.396). All Kojonup comparisons were more closely related than would be expected based on distance (Fig. 2D, blue dots).

Fig. 2
figure 2

Genetic similarity between wild kenngoor individuals using principal coordinates analysis, coloured by reserve location from A all locations and B only the translocated Kojonup reserve location and the four source locations. Isolation-by-Distance (IBD), where each point is a pairwise comparison of geographic versus genetic distance between C all natural wild locations and D all wild locations plus the translocated Kojonup reserve individuals (shown by blue points, all below the line indicating lower genetic distance than predicted by geographic distance)

Full size image

Spatial population structure analysis identified K = 2 as the optimal number of populations (Fig. 3). The cross-entropy criterion plot did not produce a plateau or asymptote, but had the largest change in gradient at K = 2 (Caye et al. 2017), consistent with weak structure corresponding to a north–south division. The northern-most reserves, Pingeculling and Boyagin (and to a lesser extent Tutanning), were more divergent from the southern reserves than expected based on geographic distance alone. The southern cluster was most strongly represented by Boundain, West Ashby, Contine Block, Dongolocking, Jaloran and Yackrikine, with the remainder of the reserves identified as having mixed ancestry. Secondarily, there was an east–west division within the southern group. The eastern reserves of Dongolocking, Yackrikine and Jaloran were more genetically distinct than expected compared to western reserves, particularly Boundain, West Ashby and Quinn’s Block.

Fig. 3
figure 3

TESS3R genetic clustering results. A Bar plot of ancestry proportions and B ancestry coefficients plotted against geographic location of reserves for K = 2. C Bar plot of ancestry proportions and D ancestry coefficients plotted against geographic location of reserves for K = 3. Darker colour bands on the maps indicate a stronger assignment to the corresponding cluster. The translocated Kojonup population has been excluded due to having mixed source locations which results in expected and uninformative allele frequency differences. Abbreviations follow Table 1

Full size image

The PCOA indicated limited population structure, consistent with the tess3r and IBD results, with only 4.1% and 2.1% of the genetic variation represented on the first two principal component (PC) axes, respectively (Fig. 2A). When considering the translocated Kojonup population and its four source locations, Kojonup was identified as a distinct cluster, albeit with similarly low levels of variation explained on PC axes (PC1 5.5%, PC2 3.9%; Fig. 2B).

Reserve areas showed weak differentiation in pairwise FST and shared allele analyses. Pairwise FST was low overall (global mean FST = 0.0597; Table 2), but every comparison was statistically significant (p < 0.001), possibly due to small sample sizes. FST ranged from 0.023 (Yackrikine and Dongolocking; 95% CI 0.018–0.029) to 0.104 (Pingeculling and West Ashby; 95% CI 0.094–0.115). The proportion of shared alleles were similarly high overall (Table 2) ranging from 88.0% (Boyagin and West Ashby) to 92.6% (East Yornaning and Boundain). Reserve areas identified as the same geographic group in TESS3R analysis tended to be more similar in both FST and shared alleles than those in different geographic groups; this was more evident with north–south clusters than north-southeast-southwest clusters.

Table 2 Summary of pairwise FST and shared alleles across reserves
Full size table

The translocated Kojonup population was most similar to the East Yornaning source location (FST = 0.039; 92.3% shared alleles) and least similar to the Boyagin source location (FST = 0.072; 89.8% shared alleles). Interestingly, Kojonup showed greater affinity to other southern reserves than to two of its northern source locations (Pingeculling and Boyagin).

Overall within-year relatedness within a reserve was low (most medians below zero; Table 3; Fig. S1), and there was no evidence of related individuals (r > 0.2) across reserves. When examining related pairs (r > 0.2), a high proportion were from the Boundain 2021 sample (36% of the 64 comparisons) (Fig S1). The overall median relatedness value for this reserve remained low (r = 0.013 in 2019).

Table 3 Population diversity summary statistics
Full size table

Genetic diversity was uniformly distributed across reserves. Individual autosomal heterozygosity estimates ranged from 0.0010 (from West Ashby) to 0.0026 (from Jaloran; Fig. S2). For reserves with sample sizes of at least eight, population-level autosomal heterozygosity estimates indicated a 20% difference between the least (West Ashby; HO = 0.00158) and most diverse (East Yornaning; HO = 0.0019) and no significant difference between reserves overall (ANOVA with heteroskedastic covariance matrix applied; p = 0.071). Similarly, there was no significant difference in allelic richness (1.16–1.18; 95% CIs overlap). All reserves had unique diversity in the form of private alleles, highest in Boyagin and Tutanning (21 alleles per individual) and lowest in Yackrikine and Quinn’s Block (four alleles per individual). No reserve had significant inbreeding estimates (all 95% CIs overlapped zero).

The TreeMix model incorporating two migration events provided the best model fit based on residual optimisation (Fig. 4). Pingeculling and Boyagin formed a slightly drifted group. There was no evidence of genetic drift in any other reserves. Evidence for migration primarily occurred from Pingeculling and Boyagin into the southern reserves, with weaker evidence for migration in the opposite direction.

Fig. 4
figure 4

Optimum TreeMix tree and residual results with two migration events. A Maximum Likelihood tree for kenngoor inferred by TreeMix when two migration events are incorporated. The scale bar shows ten times the average standard error of the entries in the sample covariance matrix. The length of the branch indicates magnitude of genetic drift. B Residual fit of the tree. Dark red and dark blue (as defined by the colour scale on the right) indicate reserves that have been least well modelled

Full size image

The global Ne estimate for 2021 was 97 individuals (95% individual jackknifed CI: 71–148). The estimate for 2019 was 48 individuals (95% CI: 34–78). When samples 2019–2021 were pooled the global estimate was 117 individuals (95% CI: 94–151).

Discussion

Extensive loss and degradation of natural habitat due to human land use can cause dramatic reductions in the size and connectivity of remnant populations, which may impact on a species’ long-term genetic diversity. The kenngoor was previously widely distributed across much of Australia but since European occupation has become restricted to an extensively fragmented area that represents less than 1% of their former range (Short and Hide 2011). For a small species (< 60 g) with a short generation time (3 years; Bradley 1997), there is a surprising lack of strong genetic differentiation given the habitat has been fragmented for ~ 75 years (Saunders 1989). We find that the species retains similar genetic diversity across the 13 highly fragmented reserve areas sampled, with little evidence of increased inbreeding or within location relatedness. Our landscape-scale analysis identified fine-scale genetic structure consistent with isolation-by-distance, suggesting that connectivity has been adequate to maintain some gene flow. We also showed that a wild-to-wild translocation of the species, founded from four source locations, was effective in retaining the genetic diversity present in the broader landscape at ten years post-translocation. A deeper insight into the genetic structure, diversity, and size of remnant wild kenngoor will assist in developing more effective conservation strategies for the species, including in the design of future translocations.

Population structure

A lack of gene flow among small and isolated populations can result in increased population structure (Levy et al. 2010; Allendorf et al. 2012; Junker et al. 2012). Surprisingly for this species, we found weak genetic structure that was primarily explained by geographic distance. The significant positive IBD values indicate a gradual gradient of genetic differentiation consistent with persistent habitat connectivity and gene flow, rather than abrupt changes over short distances due to dispersal barriers for a forest-dependent species in such a highly fragmented landscape. This was further supported by the high proportion of shared alleles between all reserves, indicating that most genetic variation occurs at the individual level rather than the reserve level.

Our analyses identified fine-scale structure and isolation of northern reserve areas (primarily Pingeculling and Boyagin), and further structuring of the southern reserves across an east–west gradient. However, broad patterns of admixture suggest this is primarily an effect of IBD rather than barriers to migration between regions. The TreeMix model detected a signal of genetic drift in Pingeculling and Boyagin, and these populations had high estimates of FST relative to other populations consistent with the change in allele frequencies due to drift (Allendorf et al. 2012). However, the divergence was not particularly strong (max. FST = 0.104), and a loss of diversity associated with genetic drift (Allendorf et al. 2012) was not observed in either reserve. This may reflect the recent isolation of a historically admixing population. This pattern of emerging genetic structure may require future monitoring to determine whether augmenting gene flow through genetic supplementation or enhancing connectivity through habitat restoration is needed to retain connectivity in the landscape.

The genetic connectivity identified here is consistent with the apparent ability of kenngoor to move between remnant habitat patches. Distances between reserves in this study ranged from 7 to 22 km, with the translocated Kojonup population located 64 km away from the closest population. While these sampling locations broadly represent the known geographic range, the species has been recorded in other patches not sampled here (Short et al. 2011). Short et al. (2011) suggested kenngoor have the capacity to disperse across the landscape, citing their presence in patches smaller than the reported home range, community sightings in and around buildings (Short and Hide 2011), and evidence of movements up to 800 m over 24 h (Bradley 1997). Indeed, the Australian Wildlife Conservancy recently detected an individual at Paruna Wildlife Sanctuary, ~100 km from any known population (J. Pierson). The ability to use a range of habitat features likely assists in moving through the landscape (Bender and Fahrig 2005) and the former wide distribution of kenngoor suggests they historically have had a wide habitat tolerance (Short et al. 2011; Short and Hide 2011). Other forest-dependent native species seem to have been more significantly impacted by fragmentation than this species (e.g. Lada et al. 2008; Lancaster et al. 2011; Li et al. 2015).

Despite a lack of strong fragmentation effects on kenngoor, detectable genetic structure across populations suggests some resistance to gene flow, even between close reserves. The importance of canopy density and dependence on suitable nesting hollows for kenngoor (Short et al. 2011) means it is likely that they primarily move through agricultural land rather than utilise it. Small remnant habitat on private land or roadside/riverine vegetation is likely crucial for connectivity, with an occupancy analysis suggesting a dispersal limit of ~5–10 km across agricultural land in the current network of corridors (Short et al. 2011). Future work should aim to conduct trapping on private agricultural land and roadside/riverine vegetation. It will be important to determine which landscape features are important for the species and most promote connectivity to enable their protection and/or rehabilitation.

Within-population diversity

No reserve areas showed signs of inbreeding, extreme genetic drift, or comparatively lower levels of diversity that are expected to contribute to increased extinction risk in small populations. Consistent with the weak genetic structure detected, this suggests landscape connectivity has thus far facilitated adequate gene flow between reserves to retain diversity.

While genetic diversity is spread uniformly across reserves, it remains difficult to quantify whether overall diversity is high enough to prevent future declines. A benefit of the autosomal heterozygosity method presented here is the ability to compare across studies and species (Schmidt et al. 2021). Kenngoor values (population HO = 0.0015 – 0.0018) fall within the upper range of published autosomal heterozygosity estimates, although these estimates are rare, in very different taxa, and/or are limited to comparisons of single genomes (Hohenlohe et al. 2010; Gopalakrishnan et al. 2017; Schmidt et al. 2021). The taxonomically closest comparable estimate available for kenngoor is the boodie (Bettongia lesueur), with most populations having a median heterozygosity of ~ 0.00075, and one mixed source translocated population reporting ~ 0.00125 (Nistelberger et al. 2023). Further studies incorporating autosomal heterozygosity have been recommended (Willi et al. 2022), and comparisons to species with similar life history traits will be useful for evaluating relative patterns of diversity in kenngoor (Romiguier et al. 2014).

Establishing genetic baselines for threatened species is challenging, but historical samples can help contextualise modern diversity. Future work could use protocols specialised for degraded DNA and museum specimens (e.g. Roycroft et al. 2021) to investigate whether current kenngoor diversity reflects historic levels. For example, Pacioni et al. (2015) found woylies (Bettongia penicillata ogilbyi) underwent a 20% decline in nuclear diversity and a 46–91% decline in mtDNA diversity over the past 2000–4000 years. Given that kenngoor has undergone similar dramatic distributional declines to the woylie (Burbidge et al. 2009; Wayne et al. 2013), it seems unlikely that current diversity represents historic diversity.

Long-term genetic outcomes of a mixed wild-to-wild translocation

Genetic considerations should be an integral part of translocation strategies, and genetic management is critical for long-term success (Weeks et al. 2011). Our 10-year post-translocation genetic assessment found that Kojonup retains the broader genetic diversity of the species and is not showing signs of inbreeding or genetic drift. Furthermore, despite being established with only 30 founders, there is no evidence of a genetic bottleneck. Unfortunately, due to a lack of access to founder samples we were unable to assess the contribution of founder groups directly. Regardless, we observed the contemporary Kojonup gene pool best represents the East Yornaning source location and least represents the Boyagin source location, despite Boyagin contributing more founders overall and in the initial release. While genetic distances are consistently smaller than predicted by geographic distance (Fig. 2), Kojonup largely fits into the broader IBD pattern for the species, despite representing a translocation of individuals from four northern reserves to the southern end of kenngoor distribution.

Interestingly, despite the establishment of the Kojonup population from locations across the north of the species distribution, the Kojonup population 10 years post-translocation showed the most genetic affinity to reserves on the southern edge of the range, suggesting gene flow from surrounding areas may have contributed to retention of genetic diversity. Additional unsampled populations may be nearby (e.g. Short et al. (2011) trapped individuals near Katanning, ~ 35 km from Kojonup reserve) and a vegetated waterway visible in satellite imagery between Katanning and Kojonup reserve might support gene flow with natural populations. Temporal analyses have shown translocated populations may lose diversity relative to their source populations with time. For example, all translocated populations of Leporillus conditor lost diversity relative to the source populations between 5 and 26 years post-translocation (White et al. 2020b), and translocated Lagostrophus fasciatus populations showed signs of genetic bottlenecks and inbreeding 5–13 years after serial translocations (White et al. 2020a). Our findings highlight the value of wild-to-wild translocations in contexts where animals can disperse freely and maintain admixture with surrounding reserves.

Effective population size

Effective population size has been recommended as a standard metric for genetic monitoring of threatened species (Hoban et al. 2020). Although estimating Ne in wild populations is challenging and our study design was limited by small sample sizes (and in some comparisons, assumption of random mating), we obtained a global Ne estimate for kenngoor of 34–151 individuals, highlighting the need for ongoing management of the species to ensure long term viability. Based on the guidelines of Frankham et al. (2014) and Hoban et al. (2020), and assuming gene flow between reserves, kenngoor likely has an effective population size that will allow short-term persistence without accumulating inbreeding. However, it is far too low to avoid negative genetic consequences in the long-term. This is an issue shared with other at-risk Australian mammals. White et al. (2020b) estimated an Ne of 140 and 82 individuals for the two extant island populations of the vulnerable hare-wallaby (Lagostrophus fasciatus). Ne estimates for six populations of the threatened greater stick-nest rat (Leporillus conditor) ranged from 6 to 47 individuals (White et al. 2020b). Ne for the critically endangered (but extensively translocated) woylie (Bettongia penicillata) ranges from 10 to 269 individuals in thirteen populations (Farquharson et al. 2021). While there are animals in captive and enclosure populations that will also be a part of kenngoor’s conservation plan, the persistence of a long-term wild population without additional management seems far from assured.

Management recommendations

Our genetic analyses indicated that none of the wild reserves included here appear to require immediate conservation action. All reserves show a history of connectivity and contain diversity representative of the broader species’ gene pool. As all locations are in the same narrow region, there is little benefit to managing reserves separately (Weeks et al. 2016; Farquharson et al. 2021), and the species should be considered a single management unit (Coates et al. 2018).

While we did not identify a strong signal of genetic drift in any reserves to date, it is nonetheless important that connectivity between reserves is maintained to promote gene flow and allow population expansion. The work of Short et al. (2011) and the results of this study suggest the kenngoor likely utilises small patches and corridors to move between reserves. Short et al. (2011) specify riverine corridors and remnant bush on private land as highly important landscape elements to facilitate movement between larger remnant patches of vegetation. Rising salinity along riverine vegetation and widening of roads threatens habitat corridors in this region (Short et al. 2011). A landscape genetic approach (e.g. Skey et al. 2023) is needed to determine which habitat variables are of highest priority for protection, including reserve size, distance between reserves, density of small patches and corridors, vegetation type, effect of waterways, roads, and extent of cleared agricultural land. Promoting landscape connectivity in this region will also likely benefit many other species, such as the critically endangered Bettongia penicillata (Farquharson et al. 2021) and black-flanked rock wallaby Petrogale lateralis (Willers et al. 2011).

Future genetic supplementation or reintroduction programs should aim to maximise representation from different reserves to maintain species-level genetic diversity. Individuals should be sourced from more geographically distant and genetically divergent locations, and effort should be taken to not source highly related individuals. Practically, this means mixing individuals from reserves with higher FST values in different geographic regions (north, south-east, and south-west).

Conclusion

Despite massive distributional declines (Short and Hide 2011), remnant wild kenngoor in a fragmented agricultural landscape retain genetic connectivity and genetic diversity, potentially utilising components of the fragmented landscape, such as riverine vegetation and remnant bush on farmland, to move between reserves (Short et al. 2011). There has been no robust estimate of wild population size of kenngoor, particularly as they can be difficult to detect and trap, thus traditional methods such as mark-recapture to estimate census population size (Nc) are challenging to implement. Currently the IUCN lists a stable population size of 9000 (Burbidge and Woinarski 2019). As we document here however, the genetically effective population size of kenngoor in this fragmented landscape is significantly lower than the estimates required to ensure long-term persistence of genetic diversity and avoidance of deleterious inbreeding (Frankham et al. 2014; Hoban et al. 2020). Conservation action to preserve ongoing landscape connectivity for the species will be required into the future, as well as continuing management of other threats such as invasive predators. Genetically informed conservation management will aid in ensuring the long-term viability of kenngoor in this landscape.