Females know better: Sex‐biased habitat selection by the European wildcat

Abstract The interactions between animals and their environment vary across species, regions, but also with gender. Sex‐specific relations between individuals and the ecosystem may entail different behavioral choices and be expressed through different patterns of habitat use. Regardless, only rarely sex‐specific traits are addressed in ecological modeling approaches. The European wildcat (Felis silvestris silvestris) is a species of conservation concern in Europe, with a highly fragmented and declining distribution across most of its range. We assessed sex‐specific habitat selection patterns for the European wildcat, at the landscape and home range levels, across its Iberian biogeographic distribution using a multipopulation approach. We developed resource selection functions in a use‐availability framework using radio‐telemetry data from five wildcat populations. At the landscape level, we observed that, while both genders preferentially established home ranges in areas close to broadleaf forests and far from humanized areas, females selected mid‐range elevation areas with some topographic complexity, whereas males used lowland areas. At the home range level, both females and males selected areas dominated by scrublands or broadleaf forests, but habitat features were less important at this level. The strength of association to habitat features was higher for females at both spatial levels, suggesting a tendency to select habitats with higher quality that can grant them enhanced access to shelter and feeding resources. Based on our results, we hypothesize that sex‐biased behavioral patterns may contribute to the resilience of wildcats’ genetic integrity through influencing the directionality of hybridization with domestic cats. Our study provides information about European wildcats’ habitat use in an Iberian context, relevant for the implementation of conservation plans, and highlights the ecological relevance of considering sex‐related differences in environmental preferences.


| INTRODUC TI ON
Human activities have widespread impacts across the environment and generally have disruptive effects in the ecosystems causing substantial threats to species diversity and conservation (Pimm et al., 2014). Therefore, a thorough understanding of how species respond to these changes is at the core of ecology and conservation biology.
These studies can be addressed across temporal and/or spatial dimensions, and the latter is usually tackled through geographical (i.e., home ranges; Moorcroft, 2012) and carrying capacity approaches (i.e., resource selection; Manly, McDonald, Thomas, McDonald, & Erickson, 2004). Studying species' spatial requirements contribute not only to the advancement of our understanding of their ecology but may also inform adequate management actions and allow the adoption of efficient conservation policies (Van Moorter, Rolandsen, Basille, & Gaillard, 2016).
Sexual differences in spatial behavior, particularly in the natal and breeding dispersal processes in mammals and birds, have been deeply investigated for decades, and the underlying causes for these differences have long been the subject of careful attention (see Dobson, 1982 andGreenwood, 1980;reviewed by Dobson, 2013). Less consideration, however, has been given to differences in habitat selection patterns between genders in established populations. Regardless, significant intersexual differences in habitat use patterns have been demonstrated in several mammal species.
In northern Botswana, Stokke and Du Toit (2002) found that, during the dry season, family groups of African elephants (Loxodonta africanus) selected habitats closer to perennial drinking water than bull groups. In a species deeply affected by predation, the white-tailed deer (Odocoileus virginianus), Kie and Bowyer (1999) found that females with young made greater use of habitats with dense cover than did males, where preferred herbaceous forage was less abundant. Carnivore species may also show distinct social behaviors and spacing patterns between genders (Crook, Ellis, & Goss-Custard, 1976). Studies performed in several species of the Felidae family also suggest that females tend to prefer habitats that grant them higher protective cover and access to feeding resources (see Chamberlain, Leopold, & Conner, 2003;Conde et al., 2010 andRamesh, Kalle, &Downs, 2015). Taken together, these studies are indicative of a general strategy across mammalian species by which females maximize their reproductive output and survival of the young. However, this topic remains largely underexplored in the scientific literature, potentially undermining a broader perception of the mechanisms driving social structure and demographic patterns of many species, with potential consequences for their management and conservation.
Despite being classified as "Least Concern" at global scale, the European wildcat (Felis silvestris silvestris, Schreber, 1777) is a protected flagship species of special conservation concern in Europe, where it is strictly protected through the Bern Convention and Habitats Directive (Yamaguchi, Kitchener, Driscoll, & Nussberger, 2015). A recent study reported five main genetic clusters in Europe, corresponding to distinct biogeographic units (BGUs, Figure 1), and all except the Iberian BGU exhibit eroded within-cluster genetic diversity, most likely due to recent bottlenecks (Mattucci, Oliveira, Lyons, Alves, & Randi, 2016). These results support not only the conservation importance of the Iberian wildcat metapopulation but also suggest it was a stronghold in the recent history of the wildcat in Europe. However, declines across several wildcat populations in Iberia have been reported (Lozano, Virgós, Cabezas-Díaz, & Mangas, 2007;Sarmento, Cruz, Eira, & Fonseca, 2009;Sobrino, Acevedo, Escudero, Marco, & Gortázar, 2008), which prompted its conservation status to "Vulnerable" in Portugal (Cabral et al., 2005), and "Near Threatened" in Spain (López-Martín et al. 2007  . Another threat to the conservation of European wildcats is hybridization with con-specific domestic cats (F. silvestris catus), which is documented throughout Europe (Mattucci et al., 2016;Yamaguchi et al., 2015). These recent population declines, prey scarcity and widespread hybridization, coupled with habitat degradation, depict a concerning conservation scenario for the Iberian wildcat BGU. Therefore, a robust understanding of European wildcats' ecological requirements and threats is association to habitat features was higher for females at both spatial levels, suggesting a tendency to select habitats with higher quality that can grant them enhanced access to shelter and feeding resources. Based on our results, we hypothesize that sex-biased behavioral patterns may contribute to the resilience of wildcats' genetic integrity through influencing the directionality of hybridization with domestic cats.
Our study provides information about European wildcats' habitat use in an Iberian context, relevant for the implementation of conservation plans, and highlights the ecological relevance of considering sex-related differences in environmental preferences.
Moreover, most habitat selection studies still ignore sex-related differences, potentially underestimating important environmental factors for one sex and overemphasizing them for the other (Conde et al., 2010). Such differences could entail important consequences for the European wildcat as they may reveal sex-specific behavioral responses to human-induced environmental change, deepening our understanding of this species' adaptive and nonadaptive responses, and potentially unraveling key aspects related to animal's ability to cope with novel environments (Sih, 2013;Sih, Ferrari, & Harris, 2011). For example, if females have a narrower spatial niche breadth than males, then their capacity to adapt to habitat changes (e.g., changes in land use) might be limited.
Another fundamental aspect of wildcats' behavior with high relevance for the species' conservation is the contact with domestic cats, potentially leading to hybridization and disease transmission (such as Feline leukemia virus, FeLV, and feline immunodeficiency virus, FIV). Different space use patterns could influence hybridization directionality (defined as the relative skewness in the frequency of interbreeding events between two groups under reciprocal hybridization; Bettles, Docker, Dufour, & Heath, 2005), and consequently the incorporation of hybrids into the wild population. This situation was reported for canid species-grey wolf x dog (Godinho et al., 2011;Leonard, Echegaray, Randi, & Vilà, 2013), or red wolf x coyote (Bohling & Waits, 2015)-, where matings between female wolves and male dogs, and between female red wolves and male coyotes, were overwhelmingly more abundant than the opposite.
These examples illustrate how the behavior and interbreeding processes can have important consequences for the conservation of fragile populations. Therefore, considering sex-specific behavioral patterns across a broad geographic scale is of utmost importance for accurately defining conservation strategies, particularly for the European wildcat in the Southwestern part of its range, where it is currently of high conservation concern (Cabral et al., 2005;López-Martín et al. 2007).
With this study we evaluate sex-specific patterns of space use by European wildcats. Specifically, we estimate home range sizes and develop sex-specific resource selection functions (RSFs) for the European wildcat, to test a hypothesis of sex-biased habitat selection. We hypothesize that females should select higher quality habitats to secure adequate breeding grounds and resource availability within smaller areas, whereas males should be more habitatflexible so as to gain access to multiple females, occupying larger territories. We develop RSFs at the landscape and home range levels across several populations within the range of the Iberian wildcat BGU. With this broad scale population-structured approach, we provide new insights into the spatial ecology of the European wildcat  The "available" sample for home range scale selection is the same as the "used" sample for landscape scale selection (see methods section).
in southwestern Europe, which can inform conservation strategies aimed at reversing the declining trends of this small felid.

| Study areas and European wildcat data
European wildcats were captured and radio-tracked in five study areas across the Iberian Peninsula ( Figure  Domestic cats occur in all study areas (except in CNP, where their occurrence is reduced), mainly associated to human settlements and its close vicinities. However, all individuals were analyzed using genetic markers and revealed no trace of admixture with domestic cats (see Supporting Information Appendix S1). Information regarding capture and monitoring procedures of European wildcats for each study area is provided in Supporting Information Appendix S1. All applicable institutional and/or national guidelines for the care and use of animals were followed.

| Home range estimation
We estimated individual wildcats' home ranges using Kernel Density Estimators (KDEs) at the 90% isopleth using a fixed reference scaled bandwidth, which was estimated manually by reducing the reference bandwidth by steps of 0.1 until the home range became fragmented or presented lacuna (Kie, 2013). The minimum number of fixes required for reliable home range estimation was determined with a bootstrap analysis, by estimating the minimum convex polygon area (MCP), increasing the sample size logarithmically until the MCP achieved an asymptote (Kenward, 2000). All home range analyses were performed using R software v3.2.5 (R Core Team 2017) with the rhr v1.2.906 package (Signer & Balkenhol, 2015). A default output grid of 100 × 100 was used to perform home range analyses. Bootstrap analyses were performed using move v2.1.0 Rpackage (Kranstauber & Smolla, 2016), with 100 repetitions per step.
Differences between male and female home range sizes were investigated with type II ANOVA tests, using stats v3.4.3 package (R Core Team 2017). Data on home range size was square-root transformed to achieve normality prior to analysis. The normality of the response variable and ANOVA residuals was checked using the Shapiro-Wilk test.

| Predictor covariates
We considered three types of covariates potentially important for the European wildcat: landcover, topography, and disturbance-  (Palomares et al., 2000). Specific landscape features, such as human structures (human settlements and roads) and water sources, might have disproportionally large effects in shaping the spacing patterns of the European wildcat (e.g., Klar et al., 2008;Monterroso et al., 2009). Therefore, we included the distance from each telemetry fix to the nearest feature of each of these covariates as potential explanatory covariates (

| Habitat selection analyses
We investigated habitat selection at two levels-landscape and home range-following the scalar hierarchy proposed by Johnson (1980), and examined selection at both levels using resource selection functions (RSFs; Manly et al., 2004). At the landscape level (Johnson's 2nd order selection), we compared two randomly generated sets of points: one generated within each animal's home range ("used" sample) and the other generated within the available landscape ("availability" sample). We considered the available landscape as the minimum convex polygon taken from outermost fixes for each studied population, to which we added a buffer equal to the mean home range radius of our radio-tracked animals, r = 2.11 km (Dillon & Kelly, 2008). At the home range level (Johnson's 3rd order selection), we compared radio-tracking locations from each animal within its respective home range ("used" sample) with the set of random points within its home range generated in the previous step ("availability" sample), therefore, applying a two-level hierarchical approach. The number of random points generated to represent availability was ten times the number of fixes used for each individual and population, for both selection levels of analysis (Koper & Manseau, 2012;Northrup, Hooten, Anderson, & Wittemyer, 2013).
We developed RSFs for males and females independently, at both levels, to account for potentially different habitat selection patterns. We developed generalized linear mixed-effects models (GLMMs) with a dummy response variable representing used and available locations (1 and 0, respectively), fitted with a logit link function. Individual wildcats were considered as random effect to accommodate for inter-individual and uneven sample size variation (Gillies et al., 2006). Although we considered that study areas could be a potential source of variation, preliminary analyses showed that including individuals wildcats as the sole random variable provided the most parsimonious models. We developed RSFs following a sequential stages approach. First, we assessed the univariate association between the response variable and each covariate in their linear and quadratic forms. The complete set of univariate models for each covariate class was ranked according to their Akaike's information criteria corrected for small sample size (AICc) (Burnham & Anderson, 2002), and we retained the covariates in the model with the lowest AICc value. We then conducted a Spearman rank correlation test on all retained covariate pairs. Whenever a pair of covariates was correlated (ρ > 0.5), we kept that which had a stronger univariate relationship with our response variable (Hosmer & Lemeshow, 2000). On a second stage, we developed a full-effects model which included all covariates retained in the previous step. Then, we built a set of models with all possible combinations between covariates, which were then ranked following AICc criteria (Burnham & Anderson, 2002). We considered models with ΔAICc values ≤2 units of the lowest AICc to have substantial support for being best models (Burnham & Anderson, 2002). The coefficients of each variable included in the top models' set were assessed following a model averaging procedure (Burnham & Anderson, 2002). All continuous variables were standardized to z-scores prior to modeling. Models that failed to converge were excluded from the analyses.
Model fit was assessed using k-fold cross-validation whereby 80% of the data was used in the modeling procedures, which were then used to predict the probability of use in the remaining 20% (Boyce, Vernier, Nielsen, & Schmiegelow, 2002

| Habitat selection at landscape and home range levels
We considered 2,671 locations (1,181 for males and 1,490 for females) obtained within home range limits to evaluate habitat selection. The RSF that best explained landscape-level habitat selection by females included both linear and quadratic terms for agricultural and scrubland areas, for distance to broadleaf forests, and for slope and elevation ( Table 2). The following model, within a ΔAICc ≤ 2, included also both the linear and quadratic terms for distance to roads and to permanent water sources (  Figure 2).
Interestingly, the strength of association/avoidance (i.e., coefficient estimates) with specific habitat features were higher for females than for males (Supporting Information Appendix S4, Table D3).
The cross-validation procedure revealed a poor and a good model fit for females and males final RSFs, respectively (ρ females = 0.41; ρ males =0.96; Table 2).
The best RSF model explaining habitat selection at the home range level by female wildcats included the linear term for scrublands, broadleaf forests, and distance to human settlements and to permanent water sources. The next three models, with a ΔAICc ≤ 2, also included the quadratic term for human settlements, scrublands, and slope (Table 2). A set of 10 top-ranked models emerged for males at the home range level, including combinations of the linear and quadratic terms of agricultural areas, distance to broadleaf forests, to scrublands and to permanent water bodies, and slope (Table 2). Females favored the use of areas with higher vegetation cover (scrublands and broadleaf forests) and closer to human settlements within their home ranges, whereas male wildcats selected steeper areas closer to water bodies and to broadleaf forests, but avoided agricultural fields (Figure 3). Although the strength of association with specific habitat features within home ranges was higher for females than for males, their importance appears to be lower at the home range than at the landscape level (Supporting Information Appendix S4,

| Home range size
We observed a high variability in home range size for European wildcats in the southwestern range of its distribution, both at the intraspecific and intragender levels. While the flexibility in male home range size reflects a pattern observed throughout its distribution range, we detected much higher variability than ever reported for females. Male home ranges are known to vary between 3.03 km 2 (Anile et al., 2017) and 53.30 km 2 (Liberek, 1999). Conversely, females' have been reported to range between <1.5 km 2 (Germain, Benhamou, & Poulle, 2008) and 6.23 km 2 (Jerosch, Götz, & Roth, 2017  Home range size of carnivores is related to metabolic needs, diet, habitat characteristics, and distribution of feeding resources, as well as interactions at both inter-and intraspecific levels (Gittleman & Harvey, 1982;Gompper & Gittleman, 1991). While having sampled European wildcats from five areas with intrinsically distinct characteristics provides a broader perception of the species requirements and ecological flexibility, it predictably entails the high variability that characterizes this species. Therefore, it is likely that the heterogeneity in habitat conditions, demographic parameters, and prey availability among study areas would be reflected in the observed variability in home range size, even among individuals of the same gender (Mattisson et al., 2013;Nilsen, Herfindal, & Linnell, 2016).

| Sex-biased habitat selection
Male and female European wildcats exhibited distinct habitat selection patterns, with several habitat components presenting a different importance for the establishment of both genders' home ranges, thereby supporting our initial hypotheses. Our results support that females settle their home ranges in areas with protective cover and low disturbance, but that grant them access to predictably high prey availability habitats (i.e., scrubland-agriculture mosaics). A reduced level of contact with anthropogenic features and permanent water sources (associated with lower elevations) revealed to be important, as disclosed by their high coefficient values. At the same biogeographic region, Sarmento et al. (2006) found that female wildcats selected autochthonous broadleaf forests and avoided dense scrubland habitats in a mountainous area with low rabbit availability. The authors argued that broadleaf forests provide shelter and the highest availability of rodents, the most consumed food resources locally (Sarmento, 1996). The apparent lower importance of these habitat variables for male wildcats may suggest that they are less influenced by habitat quality, hence, potentially more tolerant to habitat fragmentation and human contact.
Within home ranges, the strength of association to the considered covariates was lower than observed at the landscape level, which suggests a pattern of use closer to the covariates' availability. Nevertheless, females retained a stronger association to habitat features than males. Females also selected areas that provided vegetation cover, although there was a tendency to use areas closer to human settlements, where they can gain access to agricultural fields (as interpreted by the high correlation between these two covariates: ρ = 0.60). This behavior could potentially be related to high prey availability in these areas. The lower strength of association at the home range level could be indicative of either habitat selection occurring primarily at the higher level, that is, landscape, or the existence of unaccounted covariates with higher importance for wildcats at this level. Prey availability has been widely reported to influence the distribution of this species (Klar et al. 2008, Monterroso et al., 2009Silva, Kilshaw, Johnson, Macdonald, & Rosalino, 2013  distribution range, where rodents act as alternative prey (Lozano, Moleon, & Virgos, 2006). European rabbits benefit from a mosaic habitat structure that maximizes scrubland-pastureland ecotone (Fernández, 2005;Lombardi, Fernández, Moreno, & Villafuerte, 2003); therefore, these areas provide the most profitable hunting grounds for European wildcats in Mediterranean areas. However, as rabbit availability decreases in more temperate or alpine climates, rodents take place as European wildcats' main prey (Lozano et al., 2006). Rodents are also usually associated with mosaic habitats, namely at forest edges, although they are also present in agricultural areas, especially near streams (Osbourne, Anderson, & Spurgeon, 2005;Sullivan & Sullivan, 2006). Although we could not consider prey availability in our models, the habitat structure preferentially selected by European wildcats matches where prey abundance is predictably highest.
Sex-biased habitat selection where females display stronger selection for better quality habitats is consistent with those observed in the other wild felids. Conde et al. (2010) observed that while both male and female jaguars (Panthera onca) showed selection for tall forests, females strongly avoided the disturbed habitats whereas males used them in proportion to their availability. Also, Broomhall, Mills, and du Toit (2003) found that female cheetahs (Acinonyx jubatus) used dense woodlands, which provided the greatest prey availability and protective cover against dominant competitors, more frequently than males. Chamberlain et al. (2003) found that male and female bobcats (Lynx rufus) selected and used habitats differently. Female bobcats consistently selected <8 year-old pine stands, where prey is more abundant, and other habitats were even least selected during breeding and kitten-rearing. This "prudent female" behavior is consistent with the sex-biased selection pattern we observed in European wildcats. However, our habitat selection models for females at the landscape level performed poorly, with correlation values below 0.50, which may indicate some level of variability in habitat selection within this level. The inclusion of other covariates in our model, such as prey availability, could also provide a clearer picture about females' home range selection process across the landscape.

| Implications for the contact with domestic cats: hybridization and disease transmission
Paired with the reduction of suitable habitat, hybridization with domestic cats is one of the main conservation threats to European wildcats (Yamaguchi et al., 2015). Some studies (see Germain et al., 2008;Gil-Sanchez, Jaramillo, & Barea-Azcon, 2015;Sarmento et al., 2009) suggest that behavioral differences between wild and domestic cats could mediate a barrier between the two counterparts, which could be weakened by a depression of the wildcat population, with concomitant increased contact between these two subspecies.
Our results support that female wildcats have stricter habitat requirements than males, which appear to exhibit higher toler-  (Mattucci et al., 2016). It is possible that such hybridization directionality may be driven by the stricter habitat requirements of female wildcats. Although our results do not allow testing the causational link between habitat quality and the rate of hybridization between wild and domestic cats, they certainly raise this hypothesis and pave way for future research directions about these eco-evolutionary processes.
The dynamics of contact between domestic and wild counterparts also affects disease transmission, with downstream consequences for the wildlife conservation (Smith, Sax, & Lafferty, 2006). Domestic cats are thought to be the main reservoirs of viral pathogens that have been reported in several wildcat populations in Europe (e.g., Artois & Remond, 1994;Duarte, Fernandes, Santos, & Tavares, 2012;Millán & Rodríguez, 2009). However, the spillover of diseases from domestic to wildcat populations, largely neglected, deserve urgently to be studied as they may also have important consequences for the conservation of the European wildcat.

| CON CLUDING REMARK S
Using a multipopulation approach, we explicitly handle the sexspecific behavioral choices of the European wildcat expressed through differential spatial ecology and hypothesize that such specificities could entail important eco-evolutionary consequences.
Our results suggest that females have stricter habitats requirements and are less tolerant to humans at the landscape level than males, which have important implications for the conservation and management of European wildcat populations. This prudent behavior of females provides them with the best reproductive and foraging options by reducing energetic costs, increasing accessibility to prey, and securing adequate breeding grounds (Gittleman & Harvey, 1982;Gompper & Gittleman, 1991). Several authors propose the restoration/preservation of ecological corridors to increase gene flow within biogeographic regions as a main conservation priority for the European wildcat (Klar et al., 2008;Mattucci et al., 2016;Say, Devillard, Léger, Pontier, & Ruette, 2012;Steyer, Johnson, Kitchener, & Macdonald, 2016), to promote the genetic rescue of small and inbred populations (Whiteley, Fitzpatrick, Funk, & Tallmon, 2015). Our study adds to these recommendations and indicate that conservation actions should target mid-elevation areas with some degree of topographic complexity, where patches of broadleaf forests and scrublands should be promoted, and be large enough to accommodate a viable number of females. This would ensure the long-term persistence of healthy and resilient source populations of European wildcats. Contact between populations should then be secured through the establishment and maintenance of effective ecological corridors, namely through the Natura 2000 network, to allow the dispersal of individuals.
Additionally, our study highlights the ecological relevance of considering sex-related differences in environmental preferences to accurately address conservation issues. To our knowledge, this is the first study to cope with sex-specific habitat selection patterns at a multipopulation scale of the European wildcat in the humandominated landscapes of Western Europe.

ACK N OWLED G M ENTS
We thank the GVNP for the logistic support during the planning and field sampling stages of this work and to the ICNB (Portuguese

DATA ACCE SS I B I LIT Y
The data used for this study are archived in Dryad Digital Repository by Oliveira et al. (2018).