Distinguishing the victim from the threat: SNP‐based methods reveal the extent of introgressive hybridization between wildcats and domestic cats in Scotland and inform future in situ and ex situ management options for species restoration

Abstract The degree of introgressive hybridization between the Scottish wildcat and domestic cat has long been suspected to be advanced. Here, we use a 35‐SNP‐marker test, designed to assess hybridization between wildcat and domestic cat populations in Scotland, to assess a database of 295 wild‐living and captive cat samples, and test the assumptions of the test using 3,097 SNP markers generated independently in a subset of the data using ddRAD. We discovered that despite increased genetic resolution provided by these methods, wild‐living cats in Scotland show a complete genetic continuum or hybrid swarm structure when judged against reference data. The historical population of wildcats, although hybridized, clearly groups at one end of this continuum, as does the captive population of wildcats. The interpretation of pelage scores against nuclear genetic data continues to be problematic. This is probably because of a breakdown in linkage equilibrium between wildcat pelage genes as the two populations have become increasingly mixed, meaning that pelage score or SNP score alone is poor diagnostic predictors of hybrid status. Until better tools become available, both should be used jointly, where possible, when making management decisions about individual cats. We recommend that the conservation community in Scotland must now define clearly what measures are to be used to diagnose a wildcat in the wild in Scotland, if future conservation action is to be effective.

adaptive variation, which will be disadvantageous for the species' survival on an evolutionary timescale (Sgrò, Lowe, & Hoffmann, 2011;Spielman, Brook, & Frankham, 2004). In the extreme, management measures could drive population numbers down so severely, so as to threaten the population demographically. On the other hand, if the measures taken to eliminate hybrids are too weak/lenient, and animals with a large proportion of introgressed genome are allowed to breed, then costly conservation management actions are likely to have little impact on improving the status quo and all that will be conserved is a hybrid swarm. In either scenario, hybridization may impact on the population's fitness, so that it collapses in an extinction vortex (Caughley, 1994;Fagan & Holmes, 2006;Gilpin & Soulé, 1986). On the other hand, hybrids may have greater fitness than either parent species in anthropogenic landscapes making conservation of the native parent species more difficult (Lehman et al., 1991;Seehausen, Takimoto, Roy, & Jokela, 2008;Stelkens et al., 2014).

| The genetic management of hybridization
Management of hybridization requires tools to distinguish hybrids from non-hybrids. This is not straightforward, and our ability to do this is intrinsically linked to the methods chosen to measure hybridization. Molecular genetic tools have been commonly used to do this, primarily through the use of nuclear unlinked molecular genetic data (microsatellites and SNPs), followed by statistical assignment methods (most commonly STRUCTURE, (Pritchard, Stephens, & Donnelly, 2000;Falush, Stephens, & Pritchard, 2003;Falush, Stephens, & Pritchard, 2007); NewHybrids, (Anderson & Thompson, 2002); and BAPS (Corander, Waldmann, & Sillanpa, 2003)). However, a combination of other markers such as mtDNA, sex-linked markers and phenotypic traits may be used in addition to such nuclear marker data. The advent of genomewide (linked marker) data and whole genome data opens up possibilities of in-depth screening and reconstruction of ancestry and admixture (Hellenthal, Auton, & Falush, 2008;Lawson, Hellenthal, Myers, & Falush, 2012;Malinsky, Trucchi, Lawson, & Falush, 2016;Price et al., 2009). It also potentially opens up the possibility, through selective breeding, of the elimination of large tracts of introgression (Amador, Fernández, & Meuwissen, 2013;Amador, Hayes, & Daetwyler, 2014).
The basic principle of nuclear DNA hybrid testing is to survey the genome of an individual and estimate what proportion has been inherited from each parent species (its hybrid score, henceforth Q, (Pritchard et al., 2000)). At a conceptual level, this approach is relatively easy to understand, but in practice there are two issues which complicate matters and make hybridization a difficult genetic phenomenon, for which to assay.
1. By definition, hybridizing species are closely related (especially in the case of wild progenitor-domestic interactions (Randi, 2008)), and therefore much of their genome will be genetically indistinguishable from each other. Thus, if the approach is limited to a small number of markers, the first step is generally to try and find genetic markers that differentiate between the parent species and use this marker set to assay for hybridization. Therefore, the reliability of the marker set will be intrinsically linked to the quality of the reference data used to generate it. Since it is not always easy to find reliable reference individuals that do not have hybrid ancestry, this can be a complicating factor that introduces circularity and a level of uncertainty (Randi, 2008;Senn & Pemberton, 2009). The larger the number of sites in the genome (markers) that we can use to examine hybridization, the less reliant we are on any one marker and possible associated anomalies in the reference datasets, or effects of ancestral polymorphism. Approaches like those implemented within Bayesian assignment software (e.g., STRUCTURE and BAPS), which do not require the definition of reference individuals and which tolerate a degree of ancestral polymorphism due to the use of linkage disequilibrium as the primary model, also offer resilience to this (Bohling, Adams, & Waits, 2013;Putman & Carbone, 2014;Vähä & Primmer, 2006).

Assuming introgression progresses through the generations by
backcrossing (as it will do in recently hybridizing populations (Goodman, Barton, Swanson, Abernethy, & Pemberton, 1999)), the proportion of the genome that has introgressed in any one individual reduces by, on average, ½ every generation (although there is considerable variation surrounding this (Boecklen & Howard, 1997)). This means that the more distant the hybrid ancestry of an individual is, the more difficult it is to detect. Very large numbers of genetic markers are required to detect distant hybrid ancestry reliably and estimate accurately the proportion of the genome that is introgressed. This means that it is much harder to understand situations where hybridization has happened between the parent species many generations ago or where a collapse into a hybrid swarm (Mayr, 1963) has occurred.

| Hybridization and the wildcat
Hybridization with domestic cat is an important threat to the wildcat, Felis silvestris. To date, many genetic studies have been conducted on wildcat hybridization worldwide using a variety of both molecular genetics and statistical assignment methods (e.g., | 401 SENN Et al. Grossen, Keller, & Wandeler, 2013;Nussberger, Wandeler, Weber, & Keller, 2014;O'Brien et al., 2009;Oliveira, Godinho, Randi, & Alves, 2008;Oliveira et al., 2015;Pierpaoli et al., 2003;Steyer, Tiesmeyer, Muñoz-Fuentes, & Nowak, 2018). In Britain, it is the only surviving native felid. Once widespread, it has been driven to near extinction by a combination of threats, which include habitat loss, persecution and hybridization with and disease transfer from the domestic cat (Felis catus; Macdonald, Daniels, Driscoll, Kitchener, & Yamaguchi, 2004;Macdonald & Loveridge, 2010). Now, its range is restricted to the Highlands of Scotland, north of the "Central Belt" running between Glasgow and Edinburgh Davies and Gray, 2010). The two species are estimated to have been separated from each other for at least 1.1my (Li et al., 2016), revising previous estimates of>250,000 years (Driscoll et al., 2007). Domestic cats have been present in Britain for over>2000 years (O'Connor & Kitchener, 2010), and so opportunity for hybridization with wildcat has existed since then, although it has been argued that its contribution would have been low until relatively recently (Kitchener, 1998), but see also Daniels, Balharry, Hirst, Kitchener, and Aspinall (1998).
The wildcat in Scotland, or Scottish wildcat, is a subpopulation of the European wildcat based on the current taxonomic consensus (Kitchener, Breitenmoser, Eizirik, & Werdelin, 2017), although it has been described as a subspecies Felis silvestris grampia Miller, 1912. Estimates of genetic divergence between wildcat in Scotland and continental Europe do not currently exist (see Neaves & Hollingsworth, 2013 for an initial exploration of haplotype differences). Although the European wildcat is classified as least concern (LC) globally according to the International Union for the Conservation of Nature [IUCN] Red List (assessment published in 2015), it is listed on Annex IV of the EU Birds and Habitats Directive, the assessment of the conservation status for the UK is "bad," and the trend is "declining" European Community Directive on the Conservation of Natural Habitats and of Wild Fauna and Flora (92/43/EEC).
In 2004, extrapolation from a variety of data available at the time led to a suggestion that of an estimate of 3,500 wild-living cats across Scotland, only 400 individuals would be likely to be considered phenotypically wildcat based on "classic pelage characteristics" . A decade later, extrapolations from camera-trapping data carried out across Northern Scotland over 23 sites by Kilshaw (2015), encompassing 32,732 trapping days resulted in an estimate of 115-314 cats which display wild or mostly wild pelage traits (a score of 14/21 or more on the 7PS pelage scoring system of (Kitchener, Yamaguchi, Ward, & Macdonald, 2005)). These numbers are framed against many wild-living feral cats or hybrids and were estimated from a subset of 15 of the 23 sites surveyed. Of these sites, 16% of cats caught on camera were deemed to be wildcats, 23% were hybrids, and 60% were feral/domestic Kilshaw (2015). The seminal genetic study of wildcats in Scotland by Beaaumont et al. (2001), based on analysis of 230 wild-living cat samples (collected mostly between 1989 and 1994) at nine microsatellite loci, estimated that 41% of cats sampled in the wild could be wildcats. A further 42% were deemed to be hybrid and 17% domestic . In surveys covering nine sites to scope out the priority areas for the Scottish Conservation Action Plan (SNH, 2014, of 45 individuals seen on 7,493 trap nights, only six could be wildcat based on strict phenotype (a score of 19/21 or more on the 7PS), 24 were hybrid, and 15 were domestic. Lowering the definition to the relaxed pelage criteria with a 7PS score of ≥14 also used by Kilshaw (2015) and Kilshaw, Johnson, Kitchener, and Macdonald (2015) changed the estimate to 22 wildcat individuals and eight hybrids. A clear issue with estimating wildcat numbers, aside from the challenges and costs involved with monitoring an elusive felid in the field, is the lack of consensus in standardizing a definition and difficulty of aligning the results of different survey methods (Neaves & Hollingsworth, 2013;Yamaguchi, Kitchener, Driscoll, Ward, & Macdonald, 2004).
This lack of standardization of methods and definitions makes comparisons between studies to demonstrate population and introgression trends impossible.
Conservation action for the wildcat in Scotland is now being coordinated according to a National Action Plan (https://www. Two primary means of distinguishing wildcats are currently used in the implementation of the Conservation Action Plan, the 7PS pelage scoring test of Kitchener et al. (2005), where 17/21 is taken as the cut-off for a wildcat individual, and a genetic test based on 35 nuclear SNP markers (Senn & Ogden, 2015). However, note that Kitchener et al. (2005) used a score of 19 and above to define a wildcat, so that some degree of introgression has been accepted for pragmatic management reasons, with only a limited impact on external morphology.
The pelage and SNP test are used in tandem for ex situ assessments (Senn & Ogden, 2015), whereas the pelage test alone is currently used for practical reasons when implementing in situ conservation measures (TNVR), although TNVR actions are being monitored retrospectively via the SNP test.
Here, we present the results of the screening of a sample of contemporary and historical wild-living animals and the entire ex situ population. This enables us to understand the current and historical situation for wildcat hybridization in Scotland as a basis for management decisions going forward. We also explore the robustness of the common approaches used to understand hybridization. We do this by comparison of hybrid scores obtained on the 35 SNP system of Senn and Ogden (2015) to data at >3,000 SNPs generated independently via ddRAD analysis. We also do this via exploration of the relationship of the 7PS pelage scoring test of Kitchener et al. (2005) to the 35 SNP genetic system. StepOne platform (Senn & Ogden, 2015). This dataset was divided into a variety of sub-datasets based on the different methodologies used to collect the samples. Since the focus of collection of wild-living cats has shifted over time, this has been done to assist with understanding how collection bias might influence conclusions drawn from the results (more later). The datasets (summarized in Table 1) are as follows:

35SNP_HISTORICAL_CATS
A dataset of 60 cats collected between 1895 and 1985 by National Museums Scotland, Natural History Museum (London) and the New Walk Museum Leicester. These cats are primarily cats identified as wildcats that were shot by gamekeepers. The samples taken from these animals consisted of fragments of dried or tanned skins taken with sterile scalpels, which were then extracted with Qiagen Investigator Kits (Qiagen) according to manufacturer's instructions.
Of these cats, 51 were scored for pelage characters, from the preserved skins by ACK.

35SNP_WILDLIVING_DEAD_CATS
A dataset of 125 cats collected from ~1990-2015 by National Museums Scotland with the assistance of a wide variety of partners.
The cats are primarily victims of road traffic accidents, although in some cases they were shot, or the fate of specimen is unclear. Note   variable SNPs typed in 76 cats: 20 captive, five domestic and 51 wild-living. The overall % missingness in the data matrix was 6.4%. This dataset is referred to as the 3097SNP_dataset. We consider this dataset to represent the most unbiased genetic data for the Scottish wildcat so far. The 35 SNP test was derived from European wildcat data, and thus although effort was made to minimize any impact of bias in subsequent re-design of the test for Scotland (Senn & Ogden, 2015), there is the potential that hidden issues with reference data or sub-structuring within the wildcat or domestic cat populations could have biasing consequences on this relatively small panel of markers. The purpose of this dataset here is primarily to verify the performance of 35SNP system.
A full sample list of all samples can be found in the Supporting Information (Table S1), and a summary of the datasets can be found in Table 1.

| Inference of hybrid scores
Hybrid scores (Q) were assigned to individual cats in both the 35SNP dataset of all wildcats (35SNP_global_dataset) and the ddRAD dataset (3097SNP_dataset) using STRUCTURE 2.3.4 (Falush et al., 2003;Pritchard, 2010;Stephens, Smith, & Donnelly, 2001 alone, in the absence of any other data, using identical analysis parameters to those given above. Linear models (GLM) were fitted to the data using the glm function of R. The dependent variable was Q_35, which was fitted as a logit-transformed value following Beaumont et al. (2001).

| Analysis of hybrid scores
The explanatory variables fitted were as follows: (a) 7PS pelage score (7PS), a continuous measure which was were centred on the mean value (14.947) prior to inclusion in the analysis, so that in the presence of interactions, the coefficients for linear variables were evaluated at the mean level of the interacting term. Both the linear and quadratic terms were fitted; (b) dataset, a factor with four levels: 35SNP_domestic_cat; 35SNP_captive_cats; 35SNP_historical_cats; and 35SNP_wildliving_dead_cats. The interaction between 7PS and dataset was also fitted, once it was established that that the terms were significant in the absence of the interaction.
The significance of terms in the model was evaluated through tstatistics for each term. The significance of change in log-likelihood (deviance) between the new and old models was evaluated against the chi-squared distribution, at the exclusion of each term.

| Mapping
Cats with known locality (Supplementary Material 1) either had an associated grid reference, which was converted to a Northing and Easting, or only had a locality descriptor, in which case an approximate location was estimated with the aid of Google Maps. These were plotted using ARC GIS, overlapping points were manually jittered to facilitate ease of viewing, and no cats were moved across the boundaries of Wildcat Priority Areas.

| Pelage scoring
7PS scores were estimated for cats where good quality photographs were available, or where cat skins had been preserved, following the method of Kitchener et al. (2005), with a minor modification. Instead of scoring presence/absence of broken stripes, and presence of spots, on flanks and hindquarters together, the presence/absence of broken stripes and spots was scored separately for flanks and hindquarters. In practice, these characters (broken stripes and spots) are correlated with each other, so that scores did not vary between these ways of scoring these factors and it was practically easier to score each sector of the pelage separately, especially as the score for each sector (flanks and hindquarters) may be different.  Figure 1c).

| Distribution of hybrid scores in the different sample populations of cats
Structure analysis of the datasets 35SNP_wildliving_dead_cats and 35SNP_wildliving_survey_cats revealed a continuum of hybrid scores for cats sampled in the wild when judged against cats in captivity, 35SNP_captive_cats and the historical wild dataset 35SNP_ historical_cats (Figure 2). The results are summarized by hybrid score in Table 1. To understand whether the distribution of historical cats scores differed significantly to that found in the wild in contemporary samples, the following tests were run: A two-tailed Kolmogorov-Smirnov test revealed that the distribution of Q_35 in the 35SNP_historical_cats dataset was significantly different to that of 35SNP_wildliving_dead_cats (D = 0.740, p-value < 0.001).
Fisher's exact testing of 35SNP_wildliving_dead_cats against 35SNP_historical_cats for wildcats (Q_35 ≥ 0.75) versus nonwildcat (Q_35 < 0.75) was highly significant (odds ratio = 0.023, To understand whether the distribution of scores for captive cats differed significantly to that found in contemporary and historical samples, the following tests were run: The distribution of Q_35 scores in the 35SNP_captive_cats dataset was also significantly different to that of 35SNP_wildliv-ing_dead_cats (D = 0.75833, p-value <0.001). The distribution of Q_35 scores in 35SNP_historical_cats and 35SNP_captive_cats was somewhat different from each other (D = 0.256, p-value = 0.0279).
The 167 cats with geographical localities were mapped along with their hybrid scores (Figure 3).

| Pelage and hybrid score
The association between pelage and hybrid score in a dataset of 130 individuals can be found in Figure 4.
The final model of logit (Q_35) contained 7PS as a linear term and the factor dataset. Fitting the interaction between 7PS and dataset generated a marginal improvement in the model (Chisq = 10.905, p = 0.02352, df = 3), with the interaction 7PS*35SNP_historical_cats being marginally significant (est = 0.228 ± 0.11, t = 2.00, p = 0.048).
The final model excluding the interaction is presented in Table 2.

| The status of wildcat hybridization in Scotland
The study we present here confirms that hybridization between wild-living wildcats and domestic cats in Scotland is extensive (Figure 2). Using a 35-locus SNP test, whose resultant hybrid scores correlate highly with those generated independently from over >3,000 unbiased loci obtained via ddRAD analysis, we demonstrate that contemporary wild-living cat populations within Scotland consist of a genetic continuum between Felis silvestris and Felis catus and that this was not the historical situation. The study of hybridization is constantly mired in issues surrounding certainty of baseline and thus potentially circularity (discussed in the introduction of Estoup, Cornuet, Rousset, & Guyomard, 1999). Hybridization between wildcats and domestic cats in Scotland appeared to be pervasive in previous studies using nine microsatellite loci  and 14 SNP markers (SNH, 2014, Commissioned Report No. 768.) and has long been apparent in phenotypic measures (Daniels et al., 1998;Yamaguchi et al., 2004), but discussion around historical baselines and the consequent power of the methods used to determine hybridization has remained (e.g., Neaves & Hollingsworth, 2013) and thereby dogged effective implementation of management action.
This study demonstrates that contemporary populations of wildliving cats can be considered to be a hybrid swarm (Mayr, 1963) of genetically intermediate types and do not appear to display the more bimodal distribution of hybrid scores attributably to systems where hybridization is rare (e.g., as displayed in Cervus elaphus and C. nippon in Scotland with some localized exceptions (Goodman et al., 1999;Senn & Pemberton, 2009;Smith et al., 2018). A question to be resolved in more detail is what the spatial and temporal patterns of the progression of hybridization have been.
Clearly, it is very hard to unpick the effects of sampling bias, since the motivations for the collection of samples have varied over the years. The 35SNP_historical_cats are cats that were shot or trapped in deliberate attempts to target wildcats by gamekeepers, whereas 35SNP_wildliving_dead_cats mostly represent road traffic accident victims. Perhaps, on crossing any given stretch of road wildcats and domestic cats are equally likely become a fatality, but cats with wildcat-like tabby pelage may be more likely to be collected by the volunteers that have handed the samples in (biasing the sample towards wildcats and hybrids). Conversely more hybridized and domestic cats may be more likely to occur near busier urban roads, accentuating the bias in the opposite direction (note though that in the Scottish Highlands, many roads cross relatively uninhabited areas). The 35SNP_wildliving_survey_cats sample represents a contemporary sample of cats that were deliberately targeted for their presence in areas of supposed optimal wildcat habitat (SNH, 2014, Commissioned Report No. 768.), and therefore one presumes might be subject to the same bias as dataset 35SNP_historical_cats.
The 35SNP_historical_cat's dataset is drawn from a more restricted geographical range, with several cats being collected from the region of the Southern Highlands, thus potentially introducing bias in an unknown direction as there will undoubtedly be both a spatial and temporal component to hybridization.
With these issues in mind, the data are nevertheless strongly indicative that there has been a recent acceleration in hybridization. Hybridization has presumably been occurring to some de- The 35SNP_wildliving_survey_cats, whilst representing a targeted and recent sampling of animals from areas thought to contain F I G U R E 3 A map of Scotland with the locations of cats samples within the datasets 35SNP_historical_cats (circles), 35SNP_wildliving_ dead_cats (triangles) and 35SNP_wildliving_survey_cats (squares). Where only a verbal location was given, the approximate location was chosen with the aid of Google Maps (e.g., the blue cluster in the map centre). Overlapping points were separated manually for ease of viewing, whilst respecting the boundaries of the Wildcat Priority Areas which are given in red. Points are coloured using the same genetic categories used for Figure 2 F I G U R E 4 (a) The relationship between Q_35 and the 7PS pelage score. (b) Prediction from the model in Table 2  It is worth noting that of the 21 35SNP_wildliving_dead_cats that pass the wildcat criteria LBQ_35≥0.75, only two were sampled in the last ten years (since 2008)-two further animals do not have exact dates-the remainder were sampled between 1990 and 2007.
More extensive sampling of cats in the wild is currently underway through Scottish Wildcat Action (www.scottishwildcataction.org).

| The status of captive wildcats
The higher quality of the captive population could also be taken as evidence to support the temporal trend apparent in the wild samples, as most of the potential founding animals (with known acquisition dates) to the studbook were taken into captivity in 1960s and  (Amador et al., 2014;Amador, Toro, & Fernández, 2012;Hellenthal et al., 2008;Lawson et al., 2012;Malinsky et al., 2016;Price et al., 2009). However, if the level of inbreeding is found to be high, then this will be an equally concerning issue for the Scottish wildcat population in captivity (Crnokrak & Roff, 1999;Frankham, 2010) and genetic rescue (Frankham, 2015) of the population with wildcats from mainland Europe may need to be considered as a long-term population recovery option. Clearly, any attempt to remove introgressed portions of the genome through captive breeding may also remove important adaptive variation. Inbreeding will also be a threat for populations in the wild as they shrink (Brook, Tonkyn, O'Grady, & Frankham, 2002). These genetic factors will all need to be considered within future restoration programmes for the species.

| Measuring hybridization
The interpretation of pelage scores against genetic data is clearly difficult. There is a significant relationship between Q_35 (logit-transformed) and 7PS pelage (Table 2, Figure 4b); however, the intercept of the relationship varies by dataset (Figure 4), and there is some evidence that the slope may also vary as the interaction of 7PS with 35SNP_historical_cats is marginally significant (Est=0.228 ± 0.11, t = 2.00, p = 0.048). Any discussion of the relationship between 7PS pelage and Q_35 score should be caveated by the statement that the dataset is small (n = 104) and therefore we should be wary of over-interpretation of the available evidence. There is no reason why we should expect a simple relationship between Q_35 and pelage scores, and it is in fact likely that this relationship will be complex (here, fitted quadratic terms to the model were not significant).
Phenotypic traits, such as coat colour and the stripe patterning as measured by the 7PS score, are likely to be under the control of a small number of genes (Cieslak, Reissmann, Hofreiter, & Ludwig, 2011;Eizirik et al., 2010), some of which may exert dominant effects on phenotype and which will likely interact to produce pelage traits in a non-additive manner. As hybridization proceeds to intro- TA B L E 2 The model of logit (Q_35) with 7PS fitted as a linear term and dataset fitted as a factor. The explanatory variable 7PS was centred on it mean, and the factor levels are evaluated against the level DATASET35SNP_captive_cats making of rapid management decisions in the field or during quarantine, although they may become relevant within future management of the captive breeding programme (see above).
The relationship of 7PS and logit Q_35 displays so much variance that the fine-scale predictive value of 7PS on Q_35 is low. It seems, however, that using a cut-off of 17 on the 7PS score broadly sections the least hybridized cats from those that are most hybridized. Of 75 genetically identified wildcats using LBQ_35, 58 would have also been identified by pelage score, with the remaining 17 having a pelage score below 17 (Figure 4c). There were five cats that would pass the pelage cut-off for wildcat but score below the genetic cut-off.
The 7PS and Q_35 are currently used jointly for ex situ management decisions in a decision matrix where the genetic data carry greater weight (Senn & Ogden, 2015), and it seems that, where logistically feasible, that would be a more appropriate approach to adopt across the board. Currently, only 7PS is used for most in situ management decisions, and by doing so, it seems possible that high genetic-scoring cats may be missed ( Figure 4). Advances in point-of-use DNA technology may make this easier over the coming years (Morrison, Watts, Hobbs, & Dawnay, 2018). It should be noted, however, that very few contemporary wild-living cats have met either the genetic or pelage criteria (Figure 4a). We recommend that the conservation community in Scotland must now define clearly what measures are to be used to define a wildcat living in the wild in Scotland, if future conservation action is to be effective.

ACK N OWLED G EM ENTS
The authors would like to thank the many people who have handed roadkill wildcat samples to the National Museum of Scotland over the years; without these repeated individual efforts, these types of study are not possible. We are extremely grateful for your assistance.
We are also extremely grateful for the assistance of the wildcat captive holding community in the UK, for their participation in collecting

AUTH O R CO NTR I B UTI O N S
HS conceived and designed the study, analysed the data and wrote the paper. DB and AK analysed the data and wrote the paper. MG and JK performed wet laboratory analysis. BH carried out mapping.
RC participated in fieldwork. DWM coordinated survey work. All authors critically reviewed the paper.

DATA ACCE SS I B I LIT Y
All SNP data available from the Dryad Digital Repository: https://doi.