Delineating genetic management units of sambar deer (Rusa unicolor) in south-eastern Australia, using opportunistic tissue sampling and targeted scat collection

Keywords: Australia, DNA, invasive pest species, microsatellite, non-invasive sampling, opportunistic tissue sampling, Sambar deer, targeted scat collection.

Invasive species are recognised globally as major drivers of biodiversity loss (Allendorf and Lundquist 2003). Australia has a long history of damaging invasive species introductions (Hoffmann and Broadhurst 2016), including accidental arrivals to the continent and those brought intentionally –either as domestic pets and livestock for recreational hunting opportunities, or as biological control agents (Phillips et al. 2007). Some examples of the most damaging invasive species include the feral cat (Felis catus), red fox (Vulpes vulpes) and cane toad (Bufo marinus). These species have been associated with major declines in small mammals across the Australian continent (Woinarski et al. 2015). Larger species, including feral goats (Carpa hircus) and deer, have also caused significant damage to Australian ecosystems (Bayne et al. 2004; Davis et al. 2016), including those already under threat from anthropogenic climate change (Department of Environment 2015). Invasive species can have pronounced economic impacts (Perrings et al. 2000; Pimentel et al. 2005). For example, in Australia during 2001, the financial costs associated with economic loss and control of invasive species was estimated to be over A$9 billion, rising to over A$13 billion in 2012 (Hoffmann and Broadhurst 2016).

Wild deer, particularly sambar deer (Rusa unicolor), are currently causing significant environmental damage across south-eastern Australia, and pressure exists on land managers to improve deer management practices. Sambar are large tropical deer. The species has a native range in south-east Asia, including areas of India, Sri Lanka, Thailand, Nepal and China (Leslie 2011). It was first introduced into Australia during the 1800s to provide game for hunting (Bentley 1957). Sambar deer are generalist browsers and have established self-sustaining wild populations in Victoria, New South Wales (NSW), the Australian Capital Territory (ACT) and Northern Territory (NT) (Moriarty 2004). Surveys based on deer harvest by recreational hunters provide evidence of rapidly increasing sambar deer numbers in Victoria over the last decade, with estimates of over 30 000 sambar deer harvested in 2009, rising to almost 90 000 in 2017 (Moloney and Turnbull 2018) and more than 100 000 in 2018 (Moloney and Powell 2019). Increased sambar deer numbers have raised concerns regarding their social, economic and ecological impacts.

The largest sambar deer populations in Australia are known to occur in Victoria, where the species has been identified as a potentially threatening process (Parliament of Victoria 2017), and have been implicated in damage to native ecosystems (Peel et al. 2005; Bilney 2013). Sambar deer have been shown to browse a wide selection of trees, shrubs, forbs and grasses, including threatened native species in the Yarra Ranges National Park (Forsyth and Davis 2011), and their potential to permanently change vegetation composition and structure has been documented. They also consume several weed species, so are likely to disperse environmental weeds across Victorian landscapes (Eyles 2002). Antler rubbing by sambar deer has damaged stands of yellow wood (Acronychia oblongifolia) and shiny nematolepsis (Nematolepis wilsonii) in the Yarra Ranges and in East Gippsland (Bennett and Coulson 2011; Bilney 2013), linking sambar to the loss of rare and threatened plant species. In the Alpine National Park (ANP) sambar create wallows and game trails, and can damage sensitive subalpine peat communities, which are listed as threatened ecosystems (Department of Environment 2015). Another particularly serious impact of wild deer is their ability to carry and transmit endemic and exotic diseases to livestock (Cripps et al. 2019), wildlife (Ryan and Power 2012) and people (Ng et al. 2011).

Based on sambar deer occurrence records, habitat connectivity and ecology, a recent study investigating sambar deer distribution proposed four reproductively isolated populations in Victoria: eastern Victoria; Mount Cole; Timboon; and French Island (Forsyth et al. 2015). Of these, the proposed population in eastern Victoria has the largest distribution, estimated to cover over 66 300 km². This area encompasses most forested areas east of Melbourne through to the Victoria–NSW border and continues northwards into the ACT (Fig. 1; Forsyth et al. 2015). The eastern Victorian sambar deer population is thought to have been established by animals released by the Victorian Acclimatisation Society and by deer escaping from farms (Moriarty 2004). Known escapes and introductions occurred at Gembrook (undated), Kinglake (1863), Snake Island (1866) and Tooradin (1869–73) (Bentley 1967). The sambar deer populations at Mount Cole, French Island and Timboon (Fig. 1) are substantially smaller than the eastern Victorian population. Releases of sambar deer around Ercildoune in the 1870s and 1880s are thought to have founded the Mount Cole population (Forsyth et al. 2015). The origin of the French Island population is uncertain; it may have been established from a release during 1859 (Bentley 1967), or from animals originating from the Tooradin release swimming to the island (Forsyth et al. 2015). It is likely that the four sambar deer populations proposed by Forsyth et al. (2015) are genetically isolated from each other, because barriers to dispersal, including large expanses of cleared land, major roads and watercourses, exist among them (Forsyth et al. 2015). However, no previous studies have investigated the connectivity of sambar deer populations across Victoria.

Fig. 1.  Sampling locations of individual sambar deer genotypes (n = 105) in south-eastern Australia. The four ‘reproductively isolated populations’ suggested by Forsyth et al. (2015) are labelled (Eastern Victoria, French Island, Mount Cole and Timboon). Red points detail sambar deer sampled during the present study. Yellow points detail sambar deer occurrence data from the Atlas of Living Australia (ALA 2019).

Managing deer populations is a complex issue in Victoria. Despite clear evidence of the damage they cause to ecological and agricultural systems, they are also recognised as valuable game species (Davis et al. 2016). As such, recreational hunters expect healthy populations of deer to hunt. Some land managers have suggested that deer eradication is required where they are impacting sensitive vegetation communities (DEDJTR 2018). Current management strategies to reduce deer populations and their associated impacts in Australia include culling (DEDJTR 2018) and the use of fencing to exclude deer from sensitive areas (Bennett and Coulson 2008). A lack of understanding regarding deer distribution, abundance, connectivity and ability to spread to new areas makes the effective management of sambar deer in Victoria difficult. More research is required to investigate these aspects of sambar deer ecology and improve their management.

A molecular approach, where DNA is used to investigate the relatedness between sampled individuals, can be used to determine whether the putative populations are indeed genetically isolated and have the potential to be managed independently. Genetic data can be used to delineate population boundaries and assess connectivity between groups (Fraser et al. 2013). Investigating the genetic structure of animal populations, to identify individual management units, has previously been used to direct and improve eradication efforts for invasive species, including mink (Neovision vison), stoats (Mustela erminea) and feral pigs (Sus scrofa) (Hampton et al. 2004b; Veale et al. 2014; Mora et al. 2018). Small populations, with clear boundaries and no connectivity, represent the best opportunity for successful eradication because recruitment from other populations is unlikely to occur (Abdelkrim et al. 2005). Genetic data can also be used to detect dispersal pathways that could be subsequently targeted for ongoing control and surveillance (Adams et al. 2014). Subsequent to the implication of control actions, genetic data can also be used to determine the success of the program by identifying survivors and immigrants (Veale et al. 2013).

The ecology of sambar deer in Victoria is poorly understood, and no previous studies have incorporated a genetic approach for sambar deer research. Based on the ecology of sambar deer and habitat characteristics, a previous study has suggested that multiple, genetically isolated sambar deer populations exist in Victoria (Forsyth et al. 2015). Here, we aim to test this expectation using empirical genetic evidence. We used opportunistic sampling of tissue (sourced from hunters) and targeted non-invasive sampling of scats from three of the four proposed sambar deer populations (eastern Victoria, Mount Cole and French Island). Samples were not collected from Timboon due to difficulties accessing private properties and time constraints. Genetic samples were used to investigate the contemporary genetic structure of sambar deer in south-eastern Australia and attempt to delineate separate management units. This information will determine if sambar deer in Victoria form one large, homogenous population or are divided into distinct populations. Identifying genetically isolated populations will help land managers determine the feasibility of eradication and control efforts by determining the scale of connected sambar deer populations and likelihood of reinvasion after the implementation of management efforts.

The collection of samples for population genetic studies from wild animals can be challenging. Deer are large but cryptic animals, inherently difficult and expensive to trap, which limits the ability to collect large numbers of tissue or blood samples for genetic analyses (Hampton et al. 2018). DNA from tissue can be provided by recreational hunters, but this requires a good relationship with hunters and restricts sampling to areas where hunting is permitted. Alternatively, deer DNA can be sourced non-invasively through the collection of faecal pellets, often referred to as scats (Davies et al. 2020). Because scats are continuously deposited in the environment, animals do not need to be culled or caught to be sampled. Scat collection therefore allows for targeted rather than opportunistic DNA collection.

Sambar deer tissue (ear or liver) and scat samples were collected from three of the four Victorian sambar populations proposed by Forsyth et al. (2015): eastern Victoria (EV); French Island (FI); and Mount Cole (MC) (Fig. 1). Sampling was conducted over a period of 3 years, from 2015 until 2018. All sampling was performed under the provisions of the Wildlife Act 1975 and National Parks Acts 1975 from the Department of Land Water and Planning (Permit no. 1000 7699).

All sambar deer tissue samples from EV (n = 56) were supplied by licenced recreational hunters and Parks Victoria. Deer scats, presumed to be individual sambar deer, were collected from EV (n = 33), FI (n = 23) and MC (n = 23). DNA was sampled from scats using a swabbing method described in Davies et al. (2020). Attempts were made to prevent swabbing scats from the same individual by employing an exclusion zone of ~100 m between scat samples. Geographic locations were recorded for scat samples using a handheld GPS device (Garmin, Schaffhausen, Switzerland), or georeferenced by recording the nearest road, track or town. DNA was isolated from tissue samples using DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany) and from scat samples using the Qiagen QIAamp DNA Mini kit, following protocols outlined in Davies et al. (2020).

DNA from the outer surface of scats, as targeted in the present study, can be degraded by environmental conditions such as rain. As such, DNA isolates from all samples collected were screened for DNA quality and quantity, as outlined in Davies et al. (2020).

DNA isolates from EV (n = 69), FI (n = 5) and MC (n = 11) were amplified for a ~600-base-pair (bp) section of the mitochondrial (mt)DNA control region using primers CervCRH and CervtPro (Balakrishnan et al. 2003), following methods described in Davies et al. (2020). PCR products were sequenced using Sanger sequencing by the Australian Genome Research Facility (AGRF) in Melbourne, Australia. Sequences were trimmed using Sequence Scanner Software v2.0 (Applied Biosystems, Waltham, MA, USA) and the software MEGA7 (Kumar et al. 2016) was used to align sequences using CLUSTAL W (Thompson et al. 1994). To understand the number of sambar deer lineages introduced to Victoria and to identify mitochondrial differences that may contribute to population structure, aligned sequences were used to produce a haplotype network via the R packages pegas (Paradis 2010) and ape (Popescu et al. 2012). All new haplotypes were submitted to GenBank. Due to a lack of DNA template, mitochondrial DNA amplification and sequencing was not possible for all samples that were genotyped using microsatellite markers.

To gain information regarding the geographical origin of Victorian sambar deer, we compared sequence data produced by the present study with mt(DNA) control region sequence data reported by Gupta et al. (2015) and Martins et al. (2018), representing sambar deer from their native range in south-east Asia. We compared a 139-bp section of control region sequence in common between the present study and those of Gupta et al. (2015) and Martins et al. (2018). The 139-bp segment of control region sequence was also extracted from mitochondrial genome data (NC_020745.1, Hassanin et al. 2012) of the Javan rusa deer (Rusa timorensis), which was used as an outgroup. A rooted neighbour joining tree was produced using the R package ape (Paradis and Schliep 2019).

DNA isolates from tissues (n = 56) and scats (n = 79) were genotyped using 11 fluorescent-dye-labelled cervid microsatellites: BL42; BM757; INRA121; IDVGA55; TGLA53; TGLA57 (Bonnet et al. 2002); Ca18; Ca43 (Gaur et al. 2003); CelJP38; OarFCB5; and RT7 (Pérez-Espona et al. 2008), previously described in Davies et al. (2020). The power of the microsatellite suite to identify individuals was high, with a probability of identity (P_ID) of 2.7 × 10⁻⁷ for unrelated individuals and 1.0 × 10⁻³ for full siblings (Davies et al. 2020). Genotyping was carried out on the Applied Biosystems 3730 DNA analyser and GENEMAPPER 3.7 software (Applied Biosystems) by AGRF. All DNA isolates derived from scat samples were genotyped in triplicate as recommended by Davies et al. (2020). Consensus genotypes from replicates were generated using ConGenR (Lonsinger and Waits 2015). The R package allelematch (Galpern et al. 2012) was used to identify identical genotypes, which were removed from the dataset.

Genetic structuring was assessed using three methods. First, the Bayesian clustering approach in the program STRUCTURE (Pritchard et al. 2000) was used to identify the most likely number of genetic clusters. The software assigns individuals to clusters by minimising deviations from Hardy–Weinberg (HW) proportions and Linkage Disequilibrium (LD). STRUCTURE was run with admixture and correlated allele frequencies, with 3 000 000 Markov Chain Monte Carlo (MCMC) iterations and a burn-in of 1 000 000 iterations for K from 1 to 10. The most likely number of clusters was chosen based on the value of K, with the highest posterior probability from values that have plateaued (Pritchard et al. 2000). To assess if missing genotype data influenced clustering, STRUCTURE analyses were repeated on the same set of individuals (n = 105) with a reduced number of loci (six in total), as well as a subset of individuals (n = 64) with complete genotypes (individuals missing data at more than one locus were omitted from analyses).

Second, GENELAND, which implements spatial Bayesian clustering methods (Guillot et al. 2005) was used to investigate fine-scale population structure for a subset of samples with coordinate data from across Victoria (n = 46) and EV alone (n = 27). GENELAND analyses were performed using the spatial model and run with both correlated and uncorrelated allele frequencies. The correlated allele frequency model is more powerful at detecting subtle population structure; however, it is more sensitive to departures from model assumptions, such as the presence of isolation by distance (Guillot 2008). For both correlated and uncorrelated allele frequency models, the maximum number of populations (K) was set to 10, with 1 000 000 iterations and a burn in of 50 000. The thinning parameter was set to 1000 and 10 independent runs were conducted. The most likely number of clusters inferred by GENELAND was chosen according to the run with the highest posterior probability. Convergence was assessed in line with the GENELAND manual (Guillot et al. 2005).

Third, Discriminant Analysis of Principal Components (DAPC) was used to assess population structure using the R package adegenet version 2.0.2 (Jombart 2008). DAPC is based on genetic distances rather than the minimisation of HW proportions (Jombart 2008). We used the ‘find.cluster’ function to infer groups. The optimal number of clusters was chosen according to the lowest BIC value. The number of principal components included in the DAPC analysis (n = 11) was determined using the ‘optim.a.score’ function.

GenAlEx version 6.5 (Peakall and Smouse 2006) was used to calculate the mean number of alleles (N_A), mean expected (H_E) and observed heterozygosity (H_O) and number of private alleles (P_A) across all loci for clusters assigned by STRUCTURE. Deviations from Hardy–Weinberg equilibrium (HWE) and genotypic disequilibrium for each cluster were calculated using GENEPOP 4.2 (Rousset 2008). The R package diveRsity (Keenan et al. 2013) was used to calculate allelic richness (A_R, mean number of alleles corrected for differences in sample size), F_ST (Weir and Cockerham 1984) and D_jost (Jost 2008) in order to estimate genetic differentiation between clusters. Mantel tests were performed in GenAlEx to test for correlation between geographical and genetic distances for all samples with coordinate data in defined clusters. We also tested for evidence of fine-scale isolation by distance, by conducting spatial autocorrelation analyses in GenAlEx for all individuals of each cluster, for distances of 15–150 km. MICRO-CHECKER 2.2.3 (Van Oosterhout et al. 2004) was used to check for the presence of null alleles, stuttering and allelic dropout for each defined cluster (EV, FI and MC).

All DNA isolates from scats and tissue had DNA concentrations greater than 0.05 ng µL⁻¹ and produced PCR products for the quality control amplification step, so were retained for genetic analysis, as recommended by Davies et al. (2020).

Mitochondrial control region sequencing of tissue (n = 56) and scat samples (n = 32) identified two different haplotypes (R.u1 and R.u2). The first haplotype (R.u1, GenBank accession number: MK473445) was found in individuals sampled from EV and FI, and has been previously reported by Davies et al. (2020). The second haplotype (R.u2, GenBank accession number: MK473444) was unique to sambar deer from MC (Fig. 2).

Fig. 2.  Haplotype network based on the 536-bp partial mitochondrial DNA control region. Sampling locations included: Eastern Victoria, French Island and Mount Cole.

Phylogenetic analysis indicated that Victorian sambar deer are likely to have originated from Sri Lanka. The 139-bp Victorian mtDNA sequence from the present study was identical to that of sambar deer (n = 4) from Sri Lanka, whereas differences of 4–8 bp were observed between Victorian sambar deer and those sampled from different locations across south-east Asia (Supplementary material Fig. S1). This result is in agreement with historical records (Zoological and Acclimatisation Society of Victoria 1872) that indicate the importation of sambar deer from Sri Lanka to Victoria, Australia, during the 1800s.

When collecting DNA from discarded sources, multiple scats of the same individual can be inadvertently sampled. Matching of genotypic data derived from scat samples (n = 79) revealed that 52 individual sambar deer had been sampled, with 27 individuals being sampled more than once (four individuals in EV, 12 at FI and 11 at MC). After removal of matching genotypes, 105 individual sambar deer (n = 52 originating from scats and n = 53 originating from tissues) were identified from the regions sampled, i.e. EV (n = 82), FI (n = 11) and MC (n = 12).

Inspection of the mean log-likelihood inferred by STRUCTURE indicated the most likely number of populations to be K = 3 (Supplementary material Fig. S2). STRUCTURE clustered most individuals strongly (Q > 0.8) to their location of origin (Fig. 3). Two individuals were identified as not originating from their sampled region, one individual sampled in EV was assigned to the MC cluster and one individual from FI was assigned to the EV cluster. Cluster assignment for K = 3 appeared highly similar when only six loci were included and for the subset of individuals (n = 64) with complete genotypes (Supplementary material Fig. S3).

Fig. 3.  Genetic structure and assignment of individual sambar deer (n = 105) using STRUCTURE. Each sambar deer is represented by a single vertical bar. Individuals are grouped according to their population of origin. Clusters are displayed as different colours and the fraction of each individual colour represents the probability of assignment to the cluster of that colour.

Using the uncorrelated allele frequency model across all regions, GENELAND detected K = 3, matching the results of STRUCTURE, whereas the correlated allele frequency model estimated K = 4, identifying an extra cluster (Kanumbra, n = 2) within the EV population (Fig. 4, Supplementary material Fig. S4). Using eastern Victorian samples alone, GENELAND estimated K = 1 using the uncorrelated allele frequency model and K = 3 using the correlated allele frequency model (Supplementary material, Fig. S5), possibly suggesting further substructure of sambar deer present within the EV population.

Fig. 4.  Population structure of sambar deer in Victoria inferred using GENELAND with correlated and un-correlated allele frequency models. Points indicate clusters (K = 4) inferred using correlated allele frequency model. Lines surround clusters (K = 3) inferred using uncorrelated allele frequency model. Grey shading indicates forested areas.

In total, five genetic clusters were inferred using DAPC (FI, MC and three clusters in EV), with all individuals grouping to their sampling location except one individual from FI that was assigned to EV (Fig. 5, Supplementary material Fig. S6).

Fig. 5.  Scatterplot of Discriminant Analysis of Principal Components (DAPC) based on five clusters inferred in adegenet. Circles represent individuals and colours correspond to assigned clusters.

MICRO-CHECKER indicated homozygote excess at locus BL42 in the EV region, consistent with the presence of null alleles or Wahlund effect. These findings were not consistent across all regions sampled – therefore BL42 was retained for further analysis. Mean observed heterozygosity (H_O) and expected heterozygosity (H_E) across all regions was 0.42 (s.d. = 0.14) and 0.43 (s.d. = 0.13) respectively (Table 1). Deviation from HWE (P < 0.05) was observed at locus RT7 within the EV cluster. Significant genotypic disequilibrium (P < 0.05) was detected between locus pairs CelJP38 and IDVGA55, BL42 and INRA121 in the EV cluster, and between Ca18 and OarFCB5 and BL42 and TGLA53 in the MC cluster. Both private alleles and genetic differentiation tests (F_ST and D_jost) showed that the three clusters of sambar deer are likely to be separate, with little mixing between groups. Private alleles within a sampled group can provide a measure of genetic distinctiveness, but estimates are dependent upon sample size. The EV cluster had the highest number of private alleles (n = 15). Five private alleles were detected for the MC cluster, which is relatively high considering the small number of individuals sampled (n = 12); in comparison, for FI, where a similar number of individuals were sampled (n = 11), only one private allele was detected. All three population clusters were substantially differentiated using F_ST and D_jost (Table 2). Genetic differentiation was high and ranged from 0.19 (FI–EV) to 0.44 (FI–MC) for D_jost and from 0.24 (FI–EV) to 0.48 (FI–MC) for F_ST (Table 2). Mantel tests showed no correlation between geographical and genetic distance within each individual region (EV: r = 0.068; P = 0.220, FI: r = 0.006; P = 0.400, MC r = –0.002; P = 0.500).

Table 1.  Genetic diversity metrics for the three sambar deer population clusters identified by STRUCTURE
n, sample size, N_A, mean number of alleles, A_R, mean allelic richness (rarified), P_A, number of private alleles, H_O, observed heterozygosity and H_E, expected heterozygosity

Table 2.  Genetic differentiation of clusters identified by STRUCTURE (D_jost below the diagonal and F_ST above the diagonal), bootstrapped 95% confidence intervals shown in parentheses
F_ST, Weir and Cockerham’s fixation index estimator (Weir and Cockerham 1984); D_jost, estimator of Jost’s D (Jost 2008)

Sambar deer have established themselves as a problematic invasive species in south-eastern Australia, and a lack of understanding of the connectivity between populations has complicated their management. For the first time, we use microsatellite genotyping to investigate the population structure of sambar deer in Victoria and report three distinct population clusters (EV, FI and MC). Genetic clustering of individuals inferred in the pesent study agree with previous assumptions made by Forsyth et al. (2015), who suggested that sambar deer from EV, FI and MC are reproductively isolated. The F_ST and D_jost values observed in the present study show that all clusters displayed high levels of genetic differentiation. The highest genetic differentiation between clusters was observed between FI and MC, and the lowest genetic differentiation between FI and EV. The MC cluster exhibited a separate mitochondrial lineage (haplotype, R.u2), whereas FI and EV shared a single haplotype (R.u1). This suggests that either (1) the FI sambar deer population was established from the same stock as other releases in EV, or (2) the area was colonised by animals swimming to the island from Tooradin, as suggested by Forsyth et al. (2015). The separate mitochondrial haplotype identified at MC also suggests a lack of gene flow between MC and other sambar deer populations (FI and EV).

Several different factors may have contributed to the genetic differentiation observed in Victorian sambar deer, including barriers to dispersal, founder effect and maintenance of genetic differentiation from source populations. Although sambar deer are recognised as strong swimmers (Leslie 2011), the waters of Westernport Bay are likely to impede movement of individuals between the EV and FI populations. Sambar deer commonly display preferences for areas of thick vegetation cover (Bentley 1957), so large areas of cleared land are likely to be perceived as a barrier by this species. Expanses of cleared land exist between the forested areas inhabited by sambar deer in EV and FI, and are likely to have impeded movement of individual animals (and hence gene flow) between these areas. Likewise, large areas of inhospitable habitat exist between MC and the occupied range of sambar deer at EV and FI (Fig. 1). These areas would have also likely acted as a barrier to dispersal and gene flow. An additional element to consider is that even if the three populations were established from the same source, with limited or in the absence of gene flow, the MC and FI sambar populations are likely to be affected by high levels of genetic drift, driven by the low number of individuals in some of these populations.

Based on STRUCTURE results, one individual from EV showed ancestry from MC and one individual from FI displayed EV ancestry. The individual from EV displaying MC ancestry did not have the MC mitochondrial haplotype (R.u2) and had missing data at six of the 11 loci (Supplementary Data S1). Missing data can negatively affect population structure inference (Reeves et al. 2016) because Bayesian assignment, based on data from a small number of microsatellite loci genotyped in individuals with missing data, is potentially unreliable. The individual from FI displaying ancestry from EV had missing data at only one of the 11 loci and had alleles at six loci that were not present in other FI individuals. This finding potentially indicates the recent migration of individuals into the FI population; however, the small sample size obtained from FI limits confidence in the result, requiring further sampling to ensure that the different alleles identified in this individual are not present in deer located elsewhere on the island.

Sequencing of the mitochondrial control region revealed two sambar deer haplotypes (Fig. 2), providing evidence of at least two female founders of the Victorian sambar deer population. One haplotype was shared among deer from EV and FI (R.u1), with the second unique to sambar deer from MC (Ru.2). The presence of the different haplotypes among regions may have resulted from a range of different scenarios. One explanation is that the individual animals originally brought to Australia were from a single population (likely from Ceylon, now Sri Lanka) containing a limited number of haplotypes, and by chance only individuals with the R.u2 haplotype were released at MC. This is in line with reports from Bentley (1967) and records from the Zoological and Acclimatisation Society of Victoria (1872), which indicate four adult sambar deer were imported from Ceylon in 1861. However, to decisively conclude that the Sri Lankan haplotype is not present elsewhere throughout the native range of sambar deer, given the lack of phylogeographic structure evident from the available data, more detailed sampling would be required. Alternatively, the sambar deer introduced to MC may have been sourced from a different area of the native range.

Sambar deer are inherently difficult to capture and sample for genetic studies. Our approach, which combined opportunistic sampling of tissue sourced from hunters or land management agencies (Parks Victoria) with targeted scat sampling, has enabled the collection of DNA from 105 individual sambar deer across a wide geographical area. Using this approach, we were able to describe the population structure of sambar deer across south-eastern Australia. DNA isolated from tissue (including blood) samples is typically higher in both yield and quality, and data derived from such samples are less prone to errors compared with those generated from degraded DNA. Although scat samples are relatively easy to collect, they are more likely to contain degraded DNA.

Although it is preferable to use tissue samples over scat samples for genetic studies, tissue sampling for deer can be limited to areas that allow hunting and/or locations where deer control efforts are undertaken. For example, such sampling would not usually be possible in national parks (where hunting is prohibited) unless control operations are being undertaken, where deer pose the greatest threat to biodiversity. Scat collection offers a valid alternative to tissue sampling in protected areas and, as in the present study, sampling can be focused in areas in which deer activity has been detected. Furthermore, deer scats can be easily differentiated from the scats of native mammals so non-experts, volunteers and citizen scientists could be utilised to conduct field sampling and increase sample sizes. Because differentiation among scats of individuals of a single species can be problematic, DNA profile matching is necessary to ensure that individuals are not represented more than once in a dataset.

Increasing the sample size, geographical coverage of samples and number of microsatellite loci may help elucidate more subtle differences in population structure (Landguth et al. 2012) and possibly clarify the different estimate of K observed using GENELAND. However, the differences in K estimates we observed in the EV sambar population using the correlated and uncorrelated allele frequency models in GENELAND could have arisen for several reasons. The additional cluster inferred from the correlated allele frequency model may accurately represent further population structuring within the geographically large EV population. This explanation would also help explain deviations from HWE and genotypic disequilibrium observed in the EV cluster. Alternatively, the additional cluster may have been identified due to gaps in the sampling distribution (large areas of the EV population were not sampled). Previous studies have reported that the effect of isolation by distance (IBD) can cause spurious results when using GENELAND and other spatially explicit Bayesian clustering techniques (Frantz et al. 2009). Despite no evidence of IBD in the EV population, there may be other factors at play. For example, the large distribution and rapid expansion of sambar deer populations in Victoria may violate model assumptions.

During the present study, three of the four Victorian sambar deer populations identified by Forsyth et al. (2015) were sampled. Further collection of genetic data from the population not sampled for the present study (Timboon, shown in Fig. 1) would help determine whether deer within this area also form a distinct management unit. Using genetic data to determine whether eradication, control or containment is feasible at Timboon is a priority for future research due to the potential for this population to spread into neighbouring areas, including the Great Otway National Park.

Analyses of genetic data using a landscape genetic approach could be used to determine the influence of landscape features on gene flow and connectivity across Victoria. Such an approach could also identify leading edges of range expansion, which could be targeted for prioritised eradication or control (Rollins et al. 2009). Additionally, further genetic analyses will allow estimates of effective population size for each of the clusters identified in the present study. This will provide an indication of the size of each cluster and determine if each population is expanding or contracting. By collecting genetic information from scats before and after control operations, the effectiveness of the control operations can be assessed by estimating abundance of the target species using genetic mark–recapture models (Berry et al. 2012), or through the detection of genetic bottlenecks (Hampton et al. 2004a; Rollins et al. 2006). Incorporating genetic approaches into sambar deer research will provide important information regarding distribution and population dynamics and is therefore recommended.

It is important to note that habitat connectivity exists across the Victoria–NSW border, especially along densely forested expanses of the Great Dividing Range. It is therefore expected that geneflow would occur across State boundaries. The identification of sambar deer management units that occur across state boundaries would highlight the need for cooperation between Victorian and NSW management agencies and underscore the need to align deer management objectives across state borders.

Current management strategies of deer in Victoria are often ad hoc, because little is known about the ecology of this invasive species (Davis et al. 2016). Sambar deer management interventions require more empirical data to help assign resources and improve management outcomes (Parliament of Victoria 2017). The findings of the present study have important implications for sambar deer management in Victoria. We have identified three genetically isolated sambar deer populations in Victoria, which should be applied as distinct management zones. Establishing deer management zones in Victoria will help guide where to conduct surveillance, monitoring, control and eradication efforts to meet localised deer management objectives (DEDJTR 2018).

The work of Hone et al. (2010) suggested that the removal of around 40% of sambar deer would be required to stop annual population growth. If sambar deer formed a single homogenous population across Victoria, removing this number of deer would be a daunting and unfeasible process, exacerbated by the rugged and inaccessible terrain preferred by the species. Here we provide evidence that multiple sambar deer populations exist in Victoria. With this information land managers can focus control and eradication efforts to the areas identified (EV, FI and MC) and make evidence-based decisions regarding whether control and eradication efforts are feasible and cost effective. We show sambar deer on French Island are genetically isolated from deer on the mainland, suggesting that eradication of sambar deer from French Island is possible and control operations undertaken on the island have a low risk of reinvasion. In contrast, the observed connectivity between sambar deer across eastern Victoria suggests that eradication attempts performed at geographical scales smaller than the entire EV region are not feasible because reinvasion is likely to occur quickly, unless further population structuring is shown by follow up studies with a higher resolution power as we suggest above. As such, an asset protection approach to deer impacts across EV that focuses on protecting vegetation communities assessed to be at the highest levels of risk (e.g. alpine peatlands and warm temperate rainforest communities) may be practical in the short-term. We suggest that incorporating a genetic approach into future studies will greatly improve the management of sambar deer in Victoria. The sampling methodology and genetic approach applied here could also be used to delineate management units for other introduced deer species in Victoria.

The authors declare no conflicts of interest.

The data that support the present study will be shared upon reasonable request to the corresponding author.

Christopher Davies was supported by an Australian Government Research Training Program (RTP) Stipend and RTP Fee-Offset Scholarship through Federation University Australia. This research was also funded by the Holsworth Wildlife Research Endowment – Equity Trustees Charitable Foundation and Federation University Australia’s School of Health and Life Sciences.