Combining movement and genetic data to assess a forest carnivore's response to forest fragmentation

Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation André Lourenço Mestrado de Biodiversidade, Genética e Evolução Departamento de Biologia 2013 Orientador Paulo Célio Alves, Professor Auxiliar, Faculdade de Ciências da Universidade do Porto Coorientador António Mira, Professor Auxiliar, Universidade de Évora Todas as correções determinadas pelo júri, e só essas, foram efetuadas. O Presidente do Júri, Porto, ______/______/_________ FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation i AGRADECIMENTOS Em primeiro lugar, e como não poderia deixar de ser, agradeço aos meus orientadores professor Paulo Célio Alves e professor António Mira que providenciaram não só os meios logísticos sem os quais este trabalho não seria possível, mas também pelos conhecimentos fornecidos e sugestões que ajudaram a melhorar significativamente este manuscrito. Gostaria de deixar também um agradecimento muito especial ao Filipe Carvalho que apesar de não ser oficialmente meu orientador, teve uma importância enorme neste trabalho ao fornecer os seus dados de telemetria da tese de doutoramento, pelos conhecimentos sobre genetas e também de análise estatística, e finalmente pelas suas sugestões que melhoraram significativamente a qualidade da minha tese. Ao staff do CTM, em especial à Susana Lopes por toda a ajuda que disponibilizou na leitura de microsatélites e por todos os ensinamentos laboratoriais que transmitiu. Deixo também um agradecimento especial à Patrícia Ribeiro e à Filipa Martins por terem conseguido aturar-me durante um ano, ensinando-me as técnicas laboratoriais fundamentais para a execução deste trabalho. Não poderia de deixar aqui uma nota especial ao Giovanni Manghi por toda a ajuda disponibilizada com softwares de Informação Geográfica, sendo crucial para executar as análises espaciais neste trabalho. Ao pessoal do CIBIO que compareceu ao meu treino da apresentação, pelas sugestões efectuadas para melhorar a mesma que foram essenciais para aumentar a qualidade. Como não poderia deixar de ser, deixo um agradecimento muito especial aos meus amigos pelo seu companheirismo, bom humor e com o qual passámos momentos espectaculares. Não irei estar a referir nomes em particular aqui porque esta secção ficaria enorme. À minha família que tanto estimo, deixo um grande agradecimento pela sua presença na minha vida. Aos meus irmãos, Manuela Lourenço, José Lourenço e Francisco Lourenço pelo companheirismo, pelas actividades e palhaçadas que costumamos fazer, pelas gargalhadas que partilhamos e também por alguns ensinamentos sobre a vida efectuados. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation ii Finalmente, agradeço às pessoas mais importantes na minha vida, os meus pais. São eles os responsáveis por aquilo que sou hoje e sem eles não teria chegado até aqui. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation iii RESUMO A modificação de habitats naturais é atualmente reconhecido como uma grande ameaça à biodiversidade. A perda e a fragmentação dos habitats são reconhecidas como as principais causas da alteração da paisagem natural. Especificamente, a fragmentação do habitat modifica a configuração da paisagem, criando várias áreas de habitats desfavoráveis (matriz) para as espécies. A divisão de grandes e contínuas “manchas” em fragmentos mais pequenos rodeados por habitat não favorável, provoca um decréscimo na conetividade ao longo de uma determinada área. Esta interrupção do movimento, entre outras consequências negativas, bloqueia o fluxo génico entre diferentes áreas. Posteriormente, este fenómeno vai conduzir a um aumento da consaguineidade e diminuição da diversidade genética populacional entre individuos na mesma população, aumento o risco de extinção. De modo a combater estes efeitos negativos, medidas de mitigação devem de ser implementadas. Para poderem ser eficientes, estas medidas devem de ser suportadas por uma investigação científica robusta, e o aparecimento de novas áreas de investigação tais como a genética da paisagem, podem ter um papel fundamental na biologia da conservação. Atualmente, paisagens Mediterrânicas são dominadas por sistemas agro-forestais, onde uma grande proporção da floresta Mediterrânica original foi transformada em campos agrícolas. Por isso, animais como carnívoros florestais podem ser particularmente afetados por estas mudanças. Para entender como é que o fluxo génico é moldado por determinados tipos de habitats, é importante determinar com precisão a resistência que estes oferecem. Neste estudo usaram-se dados de telemetria de geneta (espécie de hábitos florestais) previamente recolhidos no âmbito de outro estudo num sistema agro-florestal no sul de Portugal, para estimar uma função de seleção de recursos de modo a avaliar objetivamente os efeitos da paisagem na variação genética nesta população de genetas. Foram colocadas três hipóteses fundamentais: (1) os dados de telemetria revelariam que as genetas usam mais zonas florestais comparativamente com outros tipos de habitats; (2) dados de parentesco e de movimento revelarão resultados discordantes relativamente aos efeitos da auto-estrada como barreira impermeável ao fluxo génico; e (3) modelos que assumem a heterogeneidade do habitat como um factor crucial que influencia a variação genética entre individuos serão mais suportados que modelos mais simplistas (por exemplo, modelos de isolamento por distância ou por barreira). Dezassete microsatélites foram genotipados com sucesso e com uma baixa taxa de erro para 74 amostras, de modo a estimar as distâncias genéticas entre os indivíduos amostrados. A equação calculada por uma regressão logística condicional revelou resultados do uso de habitats que são semelhantes àqueles obtidos por estudos anteriores, demonstrando que as genetas selecionam positivamente áreas com grande disponibilidade de recursos ecológicos (florestas de montado e galerias ripícolas) e evitaram significativamente zonas agrícolas e zonas próximas de perturbação humana. Apesar de a auto-estrada limitar os movimentos (de acordo com dados de telemetria), FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation iv esta população particular não sofreu subestruturação genética, provavelmente devido à recente construção desta barreira, não dando tempo suficiente para a população responder geneticamente. Contrariamente às expetativas iniciais, o modelo de isolamento por distância foi mais suportado do que modelos alternativos, apesar de apresentar baixos valores de correlação (Mantel r=-0.07; p<0.001). A agregação de grande parte das amostras recolhidas numa zona de habitat favorável relativamente contínuo provavelmente afectou a robustez deste modelo. O facto de as amostras estarem próximas espacialmente (aumentando a probabilidade de os indivíduos serem aparentados) numa zona favorável, não há grandes variações a nível genético entre os indivíduos nessa zona particular de amostragem. Isto implica que as diferentes variáveis de habitat consideradas estarão mal representadas nos cálculos das distâncias ecológicas entre diferentes indivíduos. Dessa forma, nessa zona altamente favorável aonde a maioria das amostras foram recolhidas, é provável que o fator principal que influencia a variação genética seja a distância geográfica. Para além disso, o papel que as galerias ripícolas desempenham como corredores de dispersão numa matriz desfavorável habitats desfavoráveis e limitações relacionadas com as técnicas de modelação usadas neste estudo, poderão ter tido alguma influência nos resultados obtidos. Melhorar a metodologia usada aqui no futuro, através da consideração de escalas espaciais e temporais apropriadas e que descrevem realisticamente processos de conetividade de fluxo génico, serão fundamentais para obter resultados mais robustos. É bastante importante ter isto em conta, para as autoridades de conservação no futuro atuarem de modo a reduzir os efeitos de fragmentação em espécies Mediterrânicas. Palavras-chave: conetividade da paisagem; função de seleção de recursos; genética da paisagem; Genetta genetta; isolamento por resistência. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation v ABSTRACT Landscape modification is actually recognized as major threat to biodiversity. Habitat loss and habitat fragmentation per se are acknowledged as the main negative consequences caused by landscape changes. Especially, habitat fragmentation modify landscape configuration, creating several areas of lower quality (matrix). The division of large continuous patches into smaller habitat patches surrounded by a low permeable matrix, greatly disturbs connectivity across the landscape. This movement disruption, among other negative consequences, impedes gene flow across the landscape. Eventually, this phenomenon will lead populations to inbreeding depression and loss of genetic diversity, reducing overall population fitness. In order to counteract these negative effects, conservation measures should be implemented. Nevertheless, accurate measures should be supported by solid scientific knowledge, and new research fields, like the landscape genetics prove to play an important role in conservation. Currently, Mediterranean landscapes are dominated by agro-forestry systems, where a great proportion of the original Mediterranean forest was transformed into agricultural fields. Therefore, genetic connectivity of several forest specialists, such as forest carnivores, may be particularly affected by these landscape changes. To understand the role of specific landscape features in shaping gene flow, it is fundamental to accurately quantify the resistance that these features impose to gene flow. Here, taking advantage of previously radio-tracking data collected for common genets (which are forest species) in an agro-forestry system in southern Portugal, a resource selection function (RSF) was estimated to objectively assess how landscape variables influenced genetic relatedness. Here, three hypotheses were tested: (1) radio-tracking data will reveal that common genets use more forested areas when compared with other types of habitats; (2) parental analysis and movement data will not present concordant results; and (3) models which assume habitat heterogeneity as a crucial factor that influences genetic variation between individuals will be more statistically supported than simpler models (for example, isolation-by-distance and isolation-by-barrier models). Seventeen microsatellites were successfully genotyped with low genotyping error rates for 74 samples, in order to estimate genetic relatedness between individuals. The conditional logistic regression equation calculated in the RSF was in accordance with previous studies, demonstrating that common genets select positively areas with high availability of ecological resources (montado forests and riparian corridors) and avoided agricultural fields and areas near human disturbance. Despite the highway disrupt movements (accordingly with radio-tracking data), this particular population is not genetically substructured probably due to the recent construction of this feature. Thus, the population did not have enough time to respond to the construction of the highway. Contrary to the initial hypotheses, the IBD (isolation-by-distance) model was more supported than the competing models, despite presenting low correlation values (Mantel r=-0.07; p<0.001). Most samples are clustered (non-random sampling) in a small region holding suitable contiguous habitat FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation vi which probably hampered the analyses. Since these samples are spatially close (increasing the probability of familiar relationships) in a favourable area, there are not great variations at the genetic level between individuals in that particular sampled zone. This implies that the different landscape features that were analysed are poorly represented in ecological distances calculations between individuals. Thus, in that particular region the geographical distance is probably the main factor influencing gene flow. Additionally, the role that riparian corridors may play as dispersal enhancers in unsuitable habitats and limitations concerning the modelling techniques used here, may have also contributed for the observed results. Refining the methodology employed here in the future, by accounting spatial and temporal scales that likely describe more realistically how processes responsible for genetic connectivity operate, is fundamental to obtain more robust results. This is very important, if conservation authorities want to reduce the effects of fragmentation in Mediterranean species in a near future.. Keywords: landscape connectivity; resource selection function; landscape genetics; Genetta genetta; isolation-by-resistance. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 2 fragmented patches leads to inbreeding depression and loss of genetic diversity, reducing individual fitness and the ability for a particular population to adapt to environmental changes (Frankham 2005; Delaney et al. 2010; Struebig et al. 2011). Third, edge effects are expressed through changes of biotic (change in vegetation communities or alteration of species interactions such as competition or predation) and abiotic (such as microclimatic changes in insulation, moisture, wind patterns) conditions in the periphery of a patch (Murcia 1995; Ewers et al. 2007). The “new” boundary environment is generally hazardous for native species, reducing habitat quality and availability within a given area. These major threats can lead a population to a sharp decline, eventually reaching an irreversible critical threshold (also called extinction vortex) where environmental (eg: natural variation and catastrophic unpredictable events), genetic (eg: genetic drift) and demographic stochasticity (eg: natural oscilitation on yearly breeding success) acquire a significant importance as local drivers of extinction (Lande 1993; Dennis 2002; Blomqvist et al. 2010). Taxa intrinsic features also play an important role on determining extinction proneness. Despite being processes transversal to many taxonomical groups (Andrén 1994; Didham et al. 1996; Andrews & Gibbons 2005; Aguilar et al. 2006), the particular combination of biological characteristics exhibited by each taxon will determine a differential susceptibility to landscape modification (Crooks 2002; Cushman 2006; see also table 2 in Fischer & Lindenmayer 2007). Mammalian carnivores constitute a clear example of vulnerability to habitat loss and fragmentation, and the biological traits that make them susceptible are relatively well known (Sunquist & Sunquist 2001; Crooks 2002; Boitani & Powell 2012). Carnivore populations usually present low densities and slow growth rates derived from low reproductive outputs. Moreover, large area requirements and other anthropogenic pressures (eg: hunting) constitute additional factors that inflate the deleterious effects of landscape changes on this group. The great dispersal ability shown by carnivores (usually more pronounced on juveniles or sub-adults) may counter-balance or exacerbate these consequences. Patch isolation effects may be minimized by being able to move through different patches, but on the other hand, high mobility may imply a greater willingness to travel through unsuitable habitat, increasing energy costs and mortality risk (Bonte et al. 2012). 1.2-Maintaining landscape connectivity 1.2.1-Overview Counteracting landscape changes effects has been a major challenge for conservation authorities. To implement effective conservation measures, one must have knowledge about several landscape features such as patch size and isolation, characteristics of the surrounding matrix and ecological requirements of the target species (Fahrig 2003; Fischer & Lindenmayer 2007). Preventing fragmentation and habitat loss is probably the best solution to maximize FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 3 conservation efforts (Crooks & Sanjayan 2006). This is rarely accomplished and, in most situations, conservation authorities are faced already with post-disturbance scenarios. Restoring habitat quality and the original amount of area would be the ideal solution to this problem (Mortelliti et al. 2010; Brückmann et al. 2010; Hodgson et al. 2011). However, logistical and budget constraints hamper the viability of many conservation projects and alternative solutions must be considered in order to mitigate the loss of pristine habitat. One viable option to reduce the fragmentation effects on species population (especially isolation effects) is to increase landscape connectivity between habitat patches. Landscape connectivity is defined as a context/species-specific concept that quantifies how movement of a particular entity (eg: pollen, seeds, genes, individuals or species) is facilitated on a particular environment (Crooks & Sanjayan 2006). The concept can be decomposed in two components: structural and functional (see review in Baguette et al. 2013). Structural connectivity describes the physic properties of the landscape such as the arrangement of patches, isolation degree and topography. Functional connectivity assesses the ecological and biological responses of a particular species or entity (individual, genes, seeds, etc.) to the structural characteristics of the landscape. Studies focusing on landscape connectivity started to increase on early 1990’s, denoting the importance that this field acquired in conservation biology (Crooks & Sanjayan 2006). Complementary research fields such as metapopulation theory and landscape ecology, largely contributed for increasing the knowledge of important landscape processes (eg: matrix permeability, immigration patterns and dynamics of colonization and extinction patterns in patches) (Moilanen & Hanski 2006; Taylor et al. 2006). Additionally, development of GIS (Geographic Information System) and better modelling tools, as well the development of higher performance computers have allowed the generation of connectivity maps with finer resolution, contributing with more detailed information to researchers about spatio-temporal patterns of species response to fragmentation (Adriaensen et al. 2003; McRae 2006; Saura & Torné 2009). During these last two decades, theoretical and empirical studies addressed the potential benefits of enhancing linkage between areas (Noss 1987; Beier & Noss 1998; Prevedello & Vieira 2010). Improving connectivity reduces movement costs (foraging movements, juvenile dispersal, migration), helps to prevent inbreeding depression, diminishes the risk of extinction in small isolated recipient populations and promotes ecological processes stability, such as natural disturbances, nutrient cycles or vegetation succession (Vilà et al. 2003; Crooks & Sanjayan 2006; Moilanen & Hanski 2006). Conversely, increasing connectivity in some situations can have biodiversity negative effects on biodiversity by promoting disease and exotic species spread and also can act as ecological traps, since they may increase exposure to several threats (eg: predator attraction) (Hess 1994; Crooks & Suarez 2006; McCallum & Dobson 2006). Outweighing the benefits and deleterious effects of such effort is a procedure that should be situation-specific; however, generally the advantages are obvious and cannot be disregarded. Despite its positive FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 4 effects on biodiversity, setting defragmentation measures into practice had lead to much controversy. Especially, corridor’s use and efficiency for maintaining long-term populations’ viability was questioned in the past (Simberloff et al. 1992; Beier & Noss 1998; Gilbert-Norton et al. 2010). The main criticisms concerned the small set of organisms that have been tested and the poorly implemented experimental designs (Beier & Noss 1998). However, Gilbert-Norton and colleagues (2010) on a review based on recent studies concluded, that in fact corridors are used by several species form different taxonomic groups. However, it is still not certain if corridors, as other conservation measures, such as stepping stone patches and management of sub-optimal matrix habitat (Fischer & Lindenmayer 2002; Baum et al. 2004; Prevedello & Vieira 2010) are efficient for maintaining long-term viability in populations (Hodgson et al. 2011). Improving not only modelling techniques, but also adopting robust management and assessment frameworks will enable scientists to effectively address connectivity issues. This is crucial to restore connectivity in fragmented landscapes. 1.2.2-Landscape genetics as a tool to assess landscape functional connectivity Landscape functional connectivity can be measured through studies on individual movement behaviour (e.g. radio-telemetry) to assess landscape resistance to dispersal, migration and daily movements (Kindlmann & Burel 2008). Although these metrics give information about permeability to movement, they fail essentially in resolving one key aspect – they do not provide information concerning successful reproduction of migrants (Mills & Allendorf 1996; Vilà et al. 2003; Jaquiéry et al. 2011). To answer questions such as “Does a corridor enables enough gene flow to prevent inbreeding depression in the recipient isolated population?” or “Does a road creates population substructuring?”, the obvious approach is to gather information regarding gene flow. Hence, indirect gene flow assessment can act as surrogate measure of landscape permeability, providing at the same time information about genetic variability among subpopulations (Cushman et al. 2006; Pérez et al. 2009; Frantz et al. 2010). In order to help addressing these types of questions, the new research field of landscape genetics arose. Following technological advances on molecular techniques, landscape genetics emerged as a promising research field tool that integrates landscape ecology, population genetics and spatial analyses. It is mainly concerned on investigating the impact of landscape features on species microevolutionary processes such as gene flow, genetic drift and adaptive genetic variation (Manel et al. 2003). The spatial and temporal scales involved on landscape genetics studies are smaller than other population genetic studies, such as phylogeography (Wang 2010). Thus, although holding a great potential to be applied to other research areas (Balkenhol et al. 2009), its applicability to address contemporary connectivity conservation issues has been recognized, leading to an increase of published papers on the last decade concerning this subject (Storfer et al. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 5 2010). Landscape genetics statistics and its limitations have been addressed, leading to the improvement of spatial and genetic models employed at population and individual levels (Dyer et al. 2010; Cushman et al. 2013). All this research contributed for the establishment of a standard statistical framework. Basically, it comprises the correlation (eg: Mantel test, partial Mantel test) between pairwise (individual or populations) matrices of genetic and ecological distances (the term “effective distances” is also used in the literature) for the detection of genetic discontinuities and/or detection of the influence of particular landscape features on gene flow (eg: Coulon et al. 2004; Stevens et al. 2006; Graves et al. 2012). Following the causal modelling framework developed by Cushman et al. (2006), three types of models are usually tested: isolation-by-distance (IBD), isolation-by-barrier (IBB) and isolation-by-resistance (IBR). The only difference between the models relies on ecological distances calculations, since different distance metrics are employed (see below). The first two are considered as null hypotheses where correlation values are confronted with the alternative hypothesis represented by the IBR model. The latter is usually translated into several models that contain different combinations of landscape variables (Shirk et al. 2010; Garroway et al. 2011). In these tests, an adequate sampling that realistically describes the spatio-temporal processes operating in the landscape of interest, and additionally, the ability to accurately estimate genetic distances and ecological distances are crucial steps to robustly assess model performance. The use of adequate molecular markers is the first step to obtain reliable differentiation measures between individuals or populations. In their review, Storfer and colleagues (2010) identified microsatellites (short tandem sequence repeats found in the nuclear genome) as the preferred molecular markers, at least in studies where the target species were animals (encompassing 70% of the reviewed papers). The fine temporal window (few to dozens of generations) that researchers face on landscape genetics studies requires markers with high mutation rates and hence, microsatellites constitute suitable candidates (Selkoe & Toonen 2006; Wang 2010). Genetic markers with higher mutation rates exhibit high allelic diversity within an evolutionary short period of time, retaining enough resolution to detect population microevolutionary responses to changes in landscape (Cushman et al. 2006; Garroway et al. 2011; Amos et al. 2012). Markers with lower mutation rates, such as sequences of mitochondrial and nuclear DNA, are more suitable to analyse events from a distant past such as the assessment of postglacial colonization genetic patterns or the evaluation of genetic isolation effects of a particular historical natural barrier (Shafer et al. 2010; Bryson et al. 2011). Despite their high resolution power, microsatellite genotyping is prone to a variety of errors such as: (1) the systematic nonamplification of an allele generally due to a point mutation on marker’s primer binding regions – null alleles; (2) allele stochastic amplification failure – allele dropout; and (3) allele misgenotyping due to human factors or to the generation of PCR artefacts Taq polymerase slippage on early cycles of PCR - false alleles (reviewed for example in Pompanon et al. 2005; Hoffman & Amos 2005; Selkoe FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 6 & Toonen 2006). Several factors can contribute to the increase of a particular type of errors (see table 2 in Pompanon et al. 2005). For example, low DNA quality and quantity is a common cause of allele dropout, constituting a relevant issue on non-invasive genetics (Waits & Paetkau 2005; Beja-Pereira et al. 2009; but see allele dropout for high quality samples in Soulsbury et al. 2007). Errors are inevitable and the best solution is to establish a solid quality control system, either on the laboratory procedures and on data analysis (Goossens et al. 1998; Piggott et al. 2004; Dakin & Avise 2004; Johnson & Haydon 2007; Chybicki & Burczyk 2009). Failure to account these errors may lead, among others, to a false increase on observed number of genotyped homozygotes (caused by null alleles and allele dropout), miscalculations of allelic frequencies and interference with Hardy-Weinberg and parentage analysis (Viard et al. 1998; Dakin & Avise 2004; Van Oosterhout et al. 2006; Johnson & Haydon 2007). Consequently, spurious scientific conclusions may be extrapolated. Landscape genetics studies rely heavily on a free bias genotyping to accurately estimate genetic distances between genetic units (eg: Pérez-Espona et al. 2008; Braunisch et al. 2010; Shirk et al. 2010). Phenomena such as null alleles or allele dropout, as stated above, lead to miscalculations of genetic distances and may obscure true marker polymorphism. Polymorphism is acknowledged as an important feature on this research field, providing more resolution power to microsatellites to detect landscape effects on genetic structuring (Holderegger & Wagner 2008; Wang 2010; Landguth et al. 2012). Thus, when theses errors are not taken into account, false landscape gene flow relationships can be obtained, having serious repercussions at conservation planning and management. Calculation of ecological distance matrices for IBD and IBB models is straightforward. The IBD model simply assumes that genetic differentiation between individuals is only dependent of geographical distance (Wright 1943). Hence, the model predicts that one particular individual is more related with geographically closer individuals than the ones far apart. A pairwise distance (distance expressed in geographic or map cell units) matrix is then constructed and correlated with a pairwise matrix of genetic distances (eg: Murphy et al. 2010; Phillipsen & Lytle 2012; Quaglietta et al. 2013). The IBB model is used to test the effects of a particular barrier (eg: highway or a river) on gene flow (Cushman et al. 2006; Shirk et al. 2010). Panmixia on either side of the barrier with no gene flow between different sides of the barrier is assumed by the IBB model. No distance costs are considered between individuals on the same side of the barrier, while it is assigned a disproportionate maximum cost to cross the barrier. Hence, it is expected a higher relatedness among individuals on the same side of the barrier than among samples from different sides. These models are fairly unrealistic for most of the times. The IBD model assumes that an animal perceives the surrounding environment homogeneously. This is false for most species (see Frantz et al. 2010 for opposite results), since there is a hierarchical habitat selection where suitable habitats are selected over hostile environments (McRae 2006; Broquet et al. 2006). IBB model is also rather too simplistic since it disregards completely landscape features that may influence gene FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 7 flow, besides the barrier itself. Additionally, not all barriers constitute completely impermeable features (Coulon et al. 2006; Frantz et al. 2010). Therefore assigning maximum resistance values to the barrier may be untruthful. The IBR model accounts for the heterogeneity of the landscape, being much more realistic in describing the underlying spatial processes that regulate gene flow (eg: Cushman et al. 2006; Wang et al. 2008; Garroway et al. 2011; Apodaca et al. 2012). To accurately estimate a pairwise ecological distance matrix from a resistance surface (raster or vectorial layer representing the different landscape features with varying permeability), two common algorithms are usually employed: least cost path (LCP) and circuit theory (Adriaensen et al. 2003; McRae 2006). LCP algorithm allows the calculation of a single optimal path between a pair of individuals (i.e., the path that holds a minimum value of cumulated resistance), being especially used on the last decade (eg: Cushman et al. 2006; Braunisch et al. 2010). The minimal values of ecological distances that were calculated are then translated into a pairwise matrix. Criticism concerning LCP increased on the last years, especially concerning the limitation of only accounting for one single path in distance calculations. This scenario is many times considered unrealistic, since it is unlikely that a particular animal has the knowledge to choose a single path that is necessarily the best one. To circumvent this problem, McRae (2006) developed Circuitscape software which uses an algorithm that borrows much of the mathematical foundations from circuit theory. Given that electricity has properties of a random walk in an electric circuit, then resistance parameters can be expressed as the probability of a random individual travelling through the cells that connect nodes (individuals or populations). Unlike LCP, circuit theory has the advantage of accounting for multiple possible pathways. Pairwise resistances are then calculated by averaging the cumulated resistance of each processed path among nodes. However, whether one algorithm is chosen over other, one of the biggest challenges that researchers face in landscape genetics still remains present: after selecting the variables of interest, one must assign specific resistance to movement values to each environmental variable (also called in the literature as parameterization of resistance values) (Spear et al. 2010; Zeller et al. 2012). So, one question must be posed “What criteria should be used to assign resistance scores?”. For resistance value assignment to a particular landscape, there are methods that are more suitable than others. Expert opinion is the easiest way to parameterize resistances surfaces. One or more researchers with experience or taking advantage of previous published papers regarding the biology of a particular organism, assign differential resistance to the landscape variables (Coulon et al. 2004; Murray et al. 2009; Spear & Storfer 2010). When empirical data about presence or dispersal is either absent or hard to obtain, this approach may be useful. However, this method is subjective and possibly inaccurate, specially due to possible species and habitat differences across regions (Spear et al. 2010). To improve on expert opinion resistance parameterization, some authors relied on model optimization (Cushman et al. 2006; Pérez-Espona et al. 2008; Shirk et al. 2010). Briefly, a resistance model including the same variables is tested FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 8 within a range of resistance values. The resistances models that best match with genetic data are objectively selected through model selection procedures such as AIC (Akaike Information Criterion) based multimodel inference. Optimization offers more power than the previous approach where a set of resistance hypotheses is tested instead of only one. Nevertheless, it suffers from the same type of bias, once the range of resistance values tested is limited and it is dependent of researchers’ choice (Spear et al. 2010). The use of non-genetic field empirical data (point counts, mark-recapture studies or radio/GPS telemetry for example) to estimate ecological distances, can be one way to avoid subjectivity. Indeed, using this data in habitat suitability modelling via resource selection functions (RSF) constitutes a valuable tool to address landscape connectivity (Coulon et al. 2008; Chetkiewicz & Boyce 2009; Sawyer et al. 2011). The underlying principle is based on the fact that habitat preference and movement are intimately linked. In this approach, landscape permeability scores for each variable are estimated through the habitat suitability models developed for species presence or intensity of use of each landscape unit. Thus, each map unit or pixel has a suitability or RSF score associated. Higher RSF values represent more permeable areas and lower otherwise. These scores can be easily converted to resistance values by simply inverting the RSF scale (eg: [RSF score] -1 ), where now higher values represent more resistant to movement areas (Chetkiewicz & Boyce 2009; Shafer et al. 2012). At conservation management level, this can aid conservation managers to implement important decisions, such as the location of corridors in areas that maximize connectivity (Chetkiewicz & Boyce 2009; Pullinger & Johnson 2010; Squires et al. 2013). However, obtaining field data of species presence can be difficult for many elusive organisms (eg: nocturnal carnivores), require intensive sampling effort and are financially expensive, driving many researchers to choose other parameterization approaches. Additionally, there is not only uncertainty regarding the choice of the landscape variables that truly affect genetic variation in a particular study area, but there may be also a disconnection between spatial and temporal scales where/when field data was collected and the genetic processes that contributed for the observed genetic structure of a population(s) (Spear et al. 2010). Those reasons are likely explanations for the fact that few landscape genetic studies took advantage of empirical data to parameterize resistance surfaces (Wang et al. 2008; Cushman et al. 2011; Shafer et al. 2012; Reding et al. 2013). Assigning resistance values to landscape variables for effective distances calculation can be performed using genetic data itself. Until now, probably only one study accomplished successfully this approach. By using standardized landscape attributes as predictive variables and pairwise genetic distances as response variable, Garroway et al. (2011) used multiple regression on distance matrices (MRDM) to construct a resistance surface. The authors found that the final multivariate surface provided high statistical power to explain population genetic differentiation across the landscape. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 9 Choosing the most appropriate approach is not straightforward. Factors such as availability of empirical information, species nature or budget constraints may influence this choice (Spear & Storfer 2010; Zeller et al. 2012). Nevertheless, while it seems that ideally one must avoid expert opinion approaches, other methods such as genetic parameterization employed by Garroway et al. (2011) or the use of indirect gene flow predictors, such as habitat selection studies still require more studies to evaluate its performance. 1.3-Study species The present study addresses the landscape effects on gene flow for common genets (Genetta genetta, Linnaeus 1758; Fig. 1). This species is a small arboreal carnivore that is mainly characterized by a long cat-like body with a yellow pale coat pattern, exhibiting longitudinal rows of dark spots and also a long tail with dark rings (Livet & Roeder 1987; Calzada 2007). The only exceptions are the rare melanic (grayish fur color) and albine (individuals with white fur coloration) phenotypes which were only detected in Europe (Gaubert & Mézan-Muxart 2010; Delibes et al. 2013). Intersex differences are little evident, although males are in general slightly bigger and heavier than females (Calzada 1998; Larivière & Calzada 2001). Common genets belong to the Viverrinae subfamily (Mammalia, Carnivora, Viverridae) which includes, besides Genetta spp. genus, other groups such as civets and African linsangs (Gaubert et al. 2004b; Gaubert et al. 2005b). In Africa, Genetta genetta may be misidentified with other sympatric species (namely Genetta felina), but skull biometric measurements and recent molecular studies have helped to resolve systematic issues among these cryptic Genetta species (Gaubert et al. 2004a; Gaubert et al. 2005a). This taxonomic confusion between Genetta genetta and Genetta felina led scientists to create ambiguous distribution maps, especially for the former (Gaubert et al. 2004a). It is well accepted now that the small-spotted genet occurs in the Arabian Peninsula, and Fig.1-Genetta genetta adult individual. http://www.regiaodeleiria.pt/wp-content/uploads/2013/09/gineta.jpg FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 10 sub-Saharan Africa, excepting for areas with dense rain forests (central-west Africa; see Fig. 2) (Larivière & Calzada 2001; Gaubert et al. 2005a). Contrarily to other species of the genus, Genetta genetta is the only viverrid where its distribution range extends to Europe, occupying the most extensive geographical area observed in Genetta species (Larivière & Calzada 2001). However, they are not native in Europe since individuals from the Maghreb region were introduced by humans at multiple places in Southern Europe (at least in Andalusia and Catalonia), probably during Muslim invasions (Gaubert et al. 2009; Gaubert et al. 2011). Following initial humanmediated colonization, the relatively similar bioclimatic conditions exhibited by Maghreb and Iberian Peninsula (Dobson & Wright 2000) likely contributed for the successful European expansion. Low temperatures seem to be the main limiting environmental variable that limits its distribution in Europe (Virgos & Casanovas 1997; Virgós et al. 2001). Accordingly, current distribution encompasses the southern region of Europe, namely Iberian Peninsula (including Mallorca, Ibiza and Cabrera islands) and the southern region of France (Calzada 2007). Recent evidence points that common genets are spreading beyond France. Observation records in northern Italy are increasing, suggesting a natural spread in that territory (Gaubert et al. 2008b). There are also sporadic observations recorded in Germany, Belgium and Switzerland, but were likely animals used as pets that were abandoned or escaped (Livet & Roeder 1987). Literature about biology and ecology of this species is poorly available for Africa (Rosevear 1974). Since the present work deals with genets in Portugal (section “2.1 Study area”), much of the information about these topics will be provided for the European range. Genetta genetta, like many carnivores species, is a solitary and nocturnal species with two major peaks of activity during the night – one after sunset and other before sunrise (Livet & Roeder 1987; Palomares & Delibes 1994). The exhibited solitary behaviour demands that olfactory signals (ano-urogenital secretions, latrines) play a major role on inter-individual communication, territory delimitation and reproduction (Roeder 1980; Barrientos 2006). Reproduction on common genets is well documented (Livet & Roeder 1987; Larivière & Calzada 2001). Breeding season spans from January to September with intensifying mating activity during February-March. Gestation period lasts for 10-11 weeks and cubs stay with the mother for two-four months. After that, juveniles start dispersing to establish their own territory, reaching sexual maturity at two years of age. Common genets have an euryphagous diet (feed on a wide variety of food) which includes small mammals, arthropods (despite their low contribution on biomass) and birds as predominant prey groups, although other elements such as amphibians, reptiles, plants and fruits are often found in their scats (Delibes et al. 1989; Virgós et al. 1999; Rosalino & Santos 2002). Despite the variety of food items that they can intake, genets are labelled as small mammal specialist with the facultative ability to change its feeding habits towards different types of preys (Virgós et al. 1999). FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 11 This species demonstrates a great flexibility regarding space use. They were observed in a varied habitat types including holm and cork oak montado, pine forests, riparian woodlands, olive groves, shrublands and rocky areas (Palomares & Delibes 1994; Larivière & Calzada 2001; Camps & Alldredge 2013). This flexibility in habitat use range it is not a synonym of habitat preference. There is a clear hierarchical habitat selection process, where forested areas with dense shrub cover (eg: holm oak forests, pine forests with dense underbrush) are primarily selected over other habitats. Two key features likely explain this preference: (1) trees and dense understory vegetation offer several potential places for resting sites (thickets, hollow trees, branches, dead trunks on the ground) and protection against predators; and (2) shrubby areas constitute a suitable habitat for small mammals (like Apodemus sylvaticus), guaranteeing high availability of food (Livet & Roeder 1987; Galantinho & Mira 2009; Camps 2011; Rosalino et al. 2011). Riparian woodlands assume also a great importance for genets (as for several other species), especially in Mediterranean environments for at least 3 reasons (Virgós 2001; Matos et al. 2009; Santos et al. 2011): (1) water is a limiting resource during the Mediterranean dry season, being confined to larger water bodies and major riparian streams. (2) associated trees and shrub cover provide shelter; and (3) riparian ecosystems can act as important dispersal corridors. Among the unsuitable habitats, it is known that genets actively avoid farmland areas and urban environments since they lack proper vegetation conditions or present high disturbance levels (Galantinho & Mira 2009; Pereira & Rodríguez 2010; Camps & Alldredge 2013). The differential habitat use exhibited by genets is crucial for the establishment of the home range, once habitat quality greatly influence the availability of food and shelter resources (Camps & Alldredge 2013). In general, there are little Fig.2-Worldwide distribution of the common genet. Adapted from Herrero & Cavallini (2008). FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 18 Table 1Characterization of 20 microsatellite loci selected to genotype Genetta genetta samples. Information regarding optimized multiplex sets and fluorescent labels used in this study is also provided. F - Forward; R – Reverse; 6-FAM (5' TGT AAA ACG ACG GCC AGT 3’); VIC (5' TAA TAC GAC TCA CTA TAG GG 3’); NED (5' TTT CCC AGT CAC GAC GTT G 3’); PET (5' GAT AAC AAT TTC ACA CAG G 3’); V (µl) – Volume added by each forward primer, reverse primer and fluorescent dye to the multiplex panel mix (pure H 2 O was added to make up a total volume of 100 µl). * Marker not included on further genetic analysis due to amplification inconsistency. Locus Repeat motif Primer sequence (5'-3') Label Multiplex V (µl) Reference A104 (CA) 23 F:TGAAAGAATTGCTTGGTATGG R:GCATGGTTGGTGAACATTC VIC Panel 1 0.8 Gaubert et al. (2008a) A108 (CA) 16 (TGCACACG CACGCG) (CA) 12 F:TGCATTACAATCACTCACTCTC R:TAGGTGGAAATCAATCTGTTG 6-FAM Panel 1 0.8 Gaubert et al. (2008a) C101 (ATGG) 12 F:TCCCACAGAAGGAACAGTC R:GCTTGTCCCATCAGAGTGT VIC Panel 1 1.6 Gaubert et al. (2008a) D111 (TAGA) 14 F:TGCTTTTTCTTTAATCCCTCTC R:TATCCTCAGCAGTCCTCAGAG 6-FAM Panel 1 0.8 Gaubert et al. (2008a) D4 (TATC) 15 F:TTGGAGAGGATTTCACTGAC R:TAGGCTTAGGAGATTTAGCAAG NED Panel 1 0.8 Gaubert et al. (2008a) Ggen 2.1 (CTTT) 27 F: CCACATAATAGCTGCTGT R: CAAAGGAGCTGAACACGT PET Panel 1 1.2 Fernandes et al. (2009) Ggen 2.A16 (TAGA) 17 F: TCCCAGATTCATTCAGTC R: TTATGGGCCTCTCTCCACGA NED Panel 1 0.8 Fernandes et al. (2009) Ggen 4.10 (CTTT) 4 CATT (CTTT) 3 (CT) 2 (CTTT) 13 F: CTCTGTTGGCCTTTCGTA R: GGTTCCTAAAACAGCTAC VIC Panel 1 1.6 Fernandes et al. (2009) Ggen 4.12 (TAGA) 13 F: GTGAGCTTCCATAATAGC R: GCTTTTCCAGAGAAACAG PET Panel 1 1.2 Fernandes et al. (2009) A110 (AC) 22 F:TCGTGCTGACGTGTTTAGC R:TTTGCCTTCCACAAAGAGG 6-FAM Panel 2 1.1 Gaubert et al. (2008a) A5 (GT) 16 F: GAACTCGGGGCTTAGATGTC R:CTGGAAAGATGAGGGGACTT PET Panel 2 1.2 Gaubert et al. (2008a) B105 (GA) 18 F:CGTGTATGTGTGTGGTGTGTG R:CCCCTACCTTCTTCATCCAAC VIC Panel 2 0.8 Gaubert et al. (2008a) Ggen 2.A13 (TCTA) 14 F: TAGGCCCCCAATCACATG R: ACTAGTCAGGTTCTCCAG 6-FAM Panel 2 0.8 Fernandes et al. (2009) Ggen 3.3 (TCTA) 3 TCA (TCTA) 18 F: CCTGTATATATTTATGGC R: TGAAAAATAGCTTTAGAC NED Panel 3 5.0 Fernandes et al. (2009) A112 (GT) 21 F:CCAACTGCCTCTGTGACTC R:CCAAAACCTATCCGAGAATG VIC Panel 3 1.6 Gaubert et al. (2008a) B104 (AG) 20 F:ATCTGCTACTGGCAAGTCAAC R:GCCTGTTTCAGTTTCTGTGTC NED Panel 3 3.2 Gaubert et al. (2008a) Ggen 2.A15 (TCTA) 14 TCA (TCTA) 3 F: TATACCCCTCATAGCTCA R: CGAATCATATCAGGCTAG VIC Panel 3 1.6 Fernandes et al. (2009) Ggen 1.30* (TTTC) 17 F: ACATTATTAACTAAGCTA R: GTGGTGATTACATCAGTC PET Fernandes et al. (2009) Ggen 2A2* (TCTA) 14 F: GGACTCATAATCCACGGA R: CGAGTCACTAAACTCTAC 6-FAM Fernandes et al. (2009) Ggen 2.A25* (CTTT) 17 (CT) 8 (CTTT) 2 (CT) 2 (CTTT) 4 F: TCTGGAGACTCCAATTTG R: GGCTTCTCAAGAAAACCT 6-FAM Fernandes et al. (2009) FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 19 2.3.2-Sample genotyping Extraction of genomic DNA was carried out from tissue, hair and blood samples using DNeasy Blood & Tissue Kit (QIAGEN), following the manufacturer’s protocol. Once hairs present low DNA quantity and quality, about 10 hairs with visible root bulbs were used to increase probability of amplification success (Goossens et al. 1998; Beja-Pereira et al. 2009). The quality and quantity of extracted DNA were assessed by gel electrophoresis. Three µl of bromophenol blue were added to two µl of extracted DNA and then loaded into a 0.8% agarose gel containing GelRed (DNA fluorescent dye; BioTarget). Gels containing DNA were run at 300V and extracted DNA was visualised in a UV transilluminator device (Bio-Rad). PCR was performed using the Multiplex PCR kit (QIAGEN) which is adequate for multiplex reactions. This kit contains a Multiplex PCR Master Mix (HotStarTaq DNA Polymerase), Multiplex PCR Buffer and Q-Solution. Each PCR reaction included five µl of Multiplex PCR Master Mix, three µl of pure H 2 O, one µl of a multiplex panel mix (table 1) and one µl - three µl of genomic DNA, depending on the DNA concentration and quality. One – two µl of extracted DNA from tissue and blood samples were added into the PCR reaction while two – three µl of DNA were added when extracted from hair samples. A negative control was also used to identify possible DNA contaminations. The optimized PCR cycling conditions were equal across the three sets. First, an initial activation step at 95ºC for 15 min, followed by nine cycles of denaturation at 95ºC for 30 s, annealing at 58ºC with a temperature decrease of 0.5ºC each cycle and extension at 72ºC for 30 s. Following this, 28 cycles of denaturation at 95ºC for 30 s, annealing at 54ºC during one minute and extension at 72ºC for 30 s, finishing with eight additional cycles of denaturation at 95ºC for 30 s, annealing at 53ºC for 45 s and extension at 72ºC for 30 s. The protocol was completed with a final extension at 60ºC during 30 min. Amplification success was also evaluated via gel electrophoresis: three µl of bromophenol blue mixed with two µl of PCR product into a 2% agarose gel to ensure a good resolution power to discriminate small fragments of interest (150 bp - 300 bp). A 100–1000 bp DNA ladder (NZYTech) was added to the gel for product size comparison. Gels were run at 300V and amplified DNA was visualized in a UV transilluminator device (Bio-Rad). PCR products were run on an ABI 3130 capillary sequencer (Applied Biosystems), using a size standard LIZ 725 (Nimagen), and scored using GeneMapper version 4.0 (Applied Biosystems). A singleplex PCR was performed whenever a particular marker from a multiplex set failed to amplify in order to increase amplification probability. In this case, PCR was performed using five µl Multiplex PCR Master Mix, 2.8 µl of distilled H 2 O, 0.4 µl of forward primer (diluted 1:100), 0.4 µl of reverse primer (dilueted 1:10) and 0.4 µl of marker specific fluorescent dye (diluted 1:10). Singleplex PCR cycling conditions were similar to those described for multiplex sets. Laboratory procedures to minimize genotyping errors were carried out. Two different protocols were performed to assess allele dropout and false allele rates: one for high quality samples (tissue and blood) and other for non-invasive hair samples. The reason to employ different methodologies FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 20 lies essentially on the fact that the latter is much more affected by stochastic errors than the former (eg: Gagneux et al. 1997; Bonin et al. 2004; Hoffman & Amos 2005). Independent PCRs were employed a second time on a subset of high quality samples. This approach has been recognized as a valid one to estimate microsatellite errors (eg: Bonin et al. 2004; Dewoody et al. 2006). Twenty three samples (about 34% of the whole dataset) were re-genotyped. Mismatches between the two genotype datasets were dealt differently according with allele scoring results. If a homozygote (AA or BB) and a heterozygote (AB) were scored for a particular locus between different screenings, then the heterozygote would be considered as the true genotype. Favouring heterozygotes over homozygotes follows the assumption that that dropout is more likely to occur than the presence of false alleles (Broquet & Petit 2004). A heterozygote would also be considered as the definitive genotype if one obtained AA for the first replicate and BB for the second. For different heterozygotes (AB and AC), the PCR would be repeated using the marker individually. The third replicate (if in accordance with one of the previous repetitions) would be accepted for further analysis. All hair samples were re-genotyped twice to estimate dropout and false allele rates. Procedures to determine the true genotype were similar to the ones employed for blood and tissue samples, excepting on one situation. Since allele dropout rates are generally higher for this type of samples, a third replicate was performed for homozygote loci to increase confidence on results. For both protocols, when both alleles failed to amplify for a given locus, individual markers PCRs were carried out to increase PCR performance. Hair samples that consistently failed to give consensus genotypes were removed from further analysis. Sex from each individual was genetically determined to improve parentage analysis (see section “2.4Microsatellites data analysis”). Published primer sets that amplify conserved regions of X and Y chromossomes in mammals - DBY intron 7 (Hellborg & Ellegren 2003) and DBX intron 5 (Hellborg & Ellegren 2004) - were tested on a set of eight high-quality samples (five known males and three known females). PCR reactions were performed as follows: five µl Multiplex PCR Master Mix, 3.2 µl of distilled H 2 O, 0.4 µl of forward primer (diluted 1:10), 0.4 µl of reverse primer (diluted 1:10) and 1 µl of DNA sample. Optimized PCR program consisted of an initial activation step at 95ºC for 15 min, followed by 12 cycles of denaturation at 95ºC for 30 s, annealing at 61ºC with a temperature decrease of 0.5ºC each cycle and extension at 72ºC for 45 s, followed by 23 cycles of denaturation at 95ºC for 30 s, annealing at 56ºC during 30 s, extension at 72ºC for 45 s and a final extension step at 60ºC during 5 min. Sex determination was performed by gel electrophoresis as described above. Although PCR products were obtained from DBX and DBY for some samples (product size superior to 300 bp and 400 bp respectively), they failed to amplify on samples of lower quality. To circumvent this problem, these PCR products were sequenced in order to redesign new primers – genX5 and genY7 (table 2). It is expected that primers that generate shorter amplified fragments will increase PCR efficiency. Successful amplifications were purified using ExoSAP (constituted by a mix of exonuclease I and Shrimp Alkaline Phosphatase enzymes; FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 21 Applied Biosystems). Sanger sequencing reactions were carried out on 10 µl reaction volumes with the following composition: 0.4 µl of termination reaction reagent BigDye (Applied Biosystems), one µl of BigDye buffer (Applied Biosystems), 0.5 µl of forward primer (diluted 1:10) and 7.1 µl of distilled H 2 O. Sequence reaction profile was as follow: initial denaturation at 94ºC for three min, followed by 24 cycles of denaturation at 96ºC during 10 s, annealing at 50ºC for five seconds, finishing with a elongation step at 60ºC for four minutes. Reaction sequence products were sequenced on an ABI 3130 capillary sequencer (Applied Biosystems). Sequenced fragments were screened on BioEdit (Hall 1999). From these fragments, primers genX5 and genY7 were designed using Primer 3 (Untergasser et al. 2012). Primers were tested using the same samples in a single reaction. PCR reaction consisted in 5 µl of Multiplex PCR Master Mix, 1.5 µl of genY7 forward primer (diluted 1:100), 1.5 µl of genY7 reverse primer (diluted 1:10), 0.1 µl of genX5 forward primer (diluted 1:100), 0.1 µl of genX5 reverse primer (diluted 1:10), 1.6 µl of 6-FAM fluorescent dye (diluted 1:10) and 1-3 µl of extracted DNA. Optimized PCR cycling conditions consisted of an initial activation step at 95ºC for 15 min, followed by 12 cycles of denaturation at 95ºC for 30 s, annealing at 58ºC with a temperature decrease of 0.5ºC each cycle and extension at 72ºC for 30 s, followed by another set of 28 cycles of denaturation at 95ºC for 30 s, annealing at 53ºC during 30 s, extension at 72ºC for 30 s and a final extension step at 60ºC for 5 min. These primer sets proved to be more efficient than those initially tested for DBX intro 5 and DBY intron 7. Sex determination was performed via visual inspection using a UV transilluminator device (Bio-Rad) (Fig. 6). Table 2Characterization of the primers re-designed for sexing genets. Primer Primer sequence (5'-3') Tm (ºC) G/C% Product size genX5 F: TGT AAA ACG ACG GCC AGT AGCCTGGGGATTGGTTTTCT R: TCCCATCTCAACATCGCTGA 59.21 58.81 50.00 50.00 189 genY7 F: TGT AAA ACG ACG GCC AGT AGTTGTTGGCATAAAATGTTTGA R: GGCGTCCGTATCTTCCATTT 55.83 58.05 30.43 50.00 250 Tm - Melting temperature; G/C% - Percentage of guanines and cytosines; 2.4-Microsatellite data analyses Allele dropout and false allele rates were estimated using software Pedant (Johnson & Haydon 2007). The software uses a maximum likelihood method to estimate jointly ε 1 (allele dropout rate) and ε 2 (false allele rate). Due to different error proneness, high quality samples and hair samples were assessed separately. Following author’s recommendation, 10000 steps were set to perform maximum likelihood search. To estimate null allele frequency accurately, it is important to account for inbreeding since both measures are correlated. Both contribute to observed homozygosity excess and failure to account them simultaneously can introduce important bias on analysis (Dakin & Avise 2004; Chapuis & Estoup 2007). Undetected null alleles may inflate estimation of inbreeding coefficient (F), whereas disregarding inbreeding can lead to an overestimation of null FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 22 allele frequencies (Van Oosterhout et al. 2006; Campagne et al. 2012). Considering that most sampled individuals in the present study are concentrated in a small area, it is possible that related individuals are present in the dataset. Thus, it is important to account for the possibility of inbreeding. Here, the newly developed INEst software (Chybicki & Burczyk 2009) was used. INEst employs a maximum likelihood method that estimates jointly the inbreeding coefficient (F) and null allele frequencies (r). Population Inbreeding Model (PIM) with a jackknife procedure was used to estimate F and r for each locus. The output provides parameters estimates and the standard errors by locus. A standard one tailed z-test was then performed to test if the estimated r was significantly higher than zero (α=0.05). After accounting for errors, markers must be tested for possible deviations of Hardy-Weinberg (HW) equilibrium and linkage disequilibrium (LD). The presence of related individuals in the dataset may bias the analysis, increasing the risk of committing type I errors (accepting erroneously the alternative hypotheses of Hardy-Weinberg disequilibrium and linkage disequilibrium). The HardyWeinberg model assumes that sampled individuals are not related on a random mating scenario (Robertson & Hill 1984; Allendorf and Luikart 2007). Genotypes between related individuals are much more similar than among unrelated ones, causing deviations from allelic frequencies predicted by the model (Robertson & Hill 1984; Bourgain et al. 2004). The nonrandom association of alleles between different loci (linkage disequilibrium) is inflated when analyzing samples with high levels of relatedness. Linkage equilibrium assumes genotypic independence among samples (Allendorf and Luikart 2007). This assumption is violated because genotypes from individuals that share familiar relationships are correlated, increasing covariance between alleles at different loci (Weir et al. 2006; Slatkin 2008). To avoid false positives, related individuals must not be included when checking for marker equilibrium departures. Taking advantage from genetic profiles, sexing, age information from trapped individuals and data about their home ranges (Carvalho et al. in prep.), familiar relationships could be inferred with a reasonable level of confidence, especially for cage-trapped individuals. COLONY software (Jones & Wang 2010) was used for this purpose. It applies a maximum likelihood method to infer jointly parent/offspring relationships and sibships. Markers deviating from Hardy-Weinberg equilibrium and linkage equilibrium expectations may lower the power of the analysis. However, the method is relatively robust if a reasonable number of polymorphic markers are used (Wang 2004). Three runs using the full likelihood method (with high likelihood precision) were performed. Only parent/offspring, full sibling and half sibling relationships with probabilistic values superior to 0.95 were considered. However, only parent/offspring and full sibling pairs were considered to filter the data in order to perform equilibrium tests. One must note that this data filtering meant to minimize familiar relationships with highest values of relatedness (parent/offspring and full sibling pairs). If two genets shared a parent/offspring or full-sibling relationship, one of them would be arbitrarily removed from equilibrium analyses. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 23 Exact tests of Hardy-Weinberg equilibrium and linkage disequilibrium were performed in GENEPOP 4.2 (parameters were set to dememorisation=10000, batch length=50000 and batch number=2000; Rousset 2008). Once multiple comparisons among loci pairs are carried out by these exact tests, the chance of making type I errors increase. To correct the level of significance (α) for multiple tests, the False Discovery Rate (FDR) method was used assuming α=0.05 (Benjamini & Hochberg 1995). This method is more accurate than the classical Bonferroni correction (Verhoeven et al. 2005). Number of alleles, observed and expected heterozygosities were calculated in GenAlEx 6.5 (Peakall & Smouse 2006; Peakall & Smouse 2012). Fig.6Sex identification of four samples in gel electrophoresis. Two females (1) and two males (2) are illustrated. 2.5-Spatial analyses 2.5.1-Environmental spatial variables A vectorial layer containing ten classes of land cover for 2010 (CCDRA, 2012-2014) was reclassified into five cover types (Fig. 5). Especially, classes regarding hypothesized unsuitable areas such as farmlands (cultures, exotic plantations, pasturelands) and urban areas (roads and settlements) were merged in order to avoid model over-fitting (Burnham & Anderson 2002). The land cover map was converted into a vectorial polygon map with a grid cell size of 100 m. The grid resolution choice was a compromise between ecological accuracy and computing times. Kernel density estimates showed that step size (distance between two consecutive biangulations/triangulations; see section “2.5.2-Resource selection function”) of 100 m constituted the maximum value of the probability density function for a period of 30 minutes. This means that in the available radio-telemetry dataset (Carvalho et al in prep.), a genet would move often a distance of about 100 m, every 30 minutes. Taking into consideration this distance and previous studies regarding movement (Palomares & Delibes 1988; Palomares & Delibes 1994), 100 m seemed a valid resolution unit that represents well the fine spatial scale at which common genets 2 2 1 1 FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 24 perceive the surrounding environment while travelling through the landscape. Within each grid cell, four landscape variables (table 3) were calculated to develop the habitat suitability model and posteriorly estimate the RSF scores (see section “2.5.2-Resource selection function”). These four predictors were chosen based on its biological meaningfulness influence on genets’ movement (eg: Galantinho & Mira 2009; Camps & Alldredge 2013; see also section "1.3-Study species"). For the habitat categorical variable, a pixel was assigned with 1 or 0 if the cell grid presented more than 50% of forest area or agricultural area, respectively. Distance to water bodies with standing water all year (here only water bodies with an area higher than 10000 m 2 were considered) such as reservoirs, ponds or dams was included in the analysis since they are additional water suppliers (besides riparian corridors) during the dry season (Rosalino et al. 2005). Similarly, only major urban areas (with an area superior to 30000 m 2 ) and roads were represented, given that they may constitute the major disturbance sources (Ditchkoff et al. 2006; Fahrig & Rytwinski 2009). All calculations were performed using QGIS version 2.1.0 (QGIS Development Team 2013) and accessory GIS tools, including GRASS version 6.4.3 (GRASS Development Team 2013) and PostGIS version 2.1.0 (PostGIS Project Steering Committee 2013). The centroids of each grid cell were used to calculate distance metrics variables. Table 3Description of the predictive landscape variables. 2.5.2-Resource selection function A resource selection function (RSF) derived from the habitat suitability model where the four landscape variables constituted the predictor set, was used to parameterize resistance surfaces and calculate ecological distances between samples. Conditional logistic regression was employed here within a match case-control design framework in order to compare habitat use (presence points or observed paths) with available habitat (points or paths randomly generated) (Boyce et al. 2002; Johnson et al. 2006). This method differs from standard logistic regression in which each specific used location (scored as 1) is matched to a group of available locations (scored as 0), allowing the comparison between empirically observed locations to random locations that represent habitat availability. When animals use a particular habitat type with higher or lower Variable Code Type Description Habitat Hab categorical Presence (1) or absence (0) in forested areas. Forested areas include riparian vegetation and montado (with arboreal cover >30%). Non-forested areas are composed by agricultural lands (cultures, pastures, exotic plantations); Distance to human disturbance dist_human Continuous Distance to the nearest human disturbance (urban areas and all types of roads) with an area higher than 30000 m 2 ; Distance to riparian vegetation dist_rip Continuous Distance to the nearest riparian corridor; Distance to water bodies dist_water Continuous Distance to the nearest water body with an area higher than 10000 m 2 (excluding riparian ecosystems); FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 25 proportion than what is present in the available dataset, then the animal is selecting or avoiding that particular habitat type. The conditional logistic regression equation takes the exponential form (Johnson et al. 2006) showed on equation 1: w(x)=exp(β 1 x 1 + β 2 x 2 +…+ β i x i ) (1) where β i is the coefficient estimate for variable i, x i is the measured variable i in a particular map unit (eg: raster pixel or a 100 m grid cell in our case) and w(x) is the RSF or suitability score associated to that map unit, weighting all analyzed variables. As demonstrated by Johnson et al. (2006), adapting the equation 1 to calculate probabilities of use (also called RSPF – Resource Selection Probability Function) is a valid approach and it can be accomplished by using equation 2: w*(x)= exp(β 1 x 1 + β 2 x 2 +…+ β i x i ) 1+exp(β 1 x 1 + β 2 x 2 +…+ β i x i ) where w*(x) corresponds to the RSPF scores. Here, the used dataset constituted by VHF (very high frequency) nocturnal radio-telemetry biangulations (two bearings taken in less than 5 minutes), triangulations (three bearings taken in less than 10 minutes) and diurnal resting sites (homing-in technique) locations (data gathered by Carvalho et al. in prep.), was compared with an available points dataset composed of randomly generated points (see below). This use-available design is performed at a scale within home ranges, being categorized as third/fourth-order habitat selection (Johnson 1980). Given the smooth topography of the study area, accuracy of locations determined by VHF radio-telemetry was relatively high (error inferior to 100 m) (Carvalho et al. in prep.). Few locations presented low spatial accuracy and were removed from the dataset. Movement data from 21 adult genets (9 males and 12 females) with stable home ranges (calculated through 100% Minimum Convex Polygon technique) was used to represent used locations sample set (Fig. S1). Radio-telemetry data is spatially and temporally autocorrelated, violating the independence assumption of most statistical methods. Violation of independence may introduce significant inference bias on habitat selection models, possibly leading to spurious conclusions (Nielsen et al. 2002; Fieberg et al. 2010). Independence among locations was assumed here, by considering only locations separated within a 4 hour interval (Palomares & Delibes 1994). Given the role that they play on genet’s ecology and biology, resting sites provide important information about habitat selection. The choice of particular resting sites is largely conditioned by the surrounding habitat (Camps 2011). Ideally, the habitat surrounding resting sites must fulfil foraging (food and water intakes), reproductive and low disturbance requirements, and thus it is likely that the location of resting sites represent areas highly permeable (see also Camps & Alldredge 2013). Therefore, resting sites were also included (2) FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 26 as used locations, which were also highly accurate contributing for the overall location accuracy. Buffers with 800 m radius were created around each used location using ArcGis 10 (ESRI 2011). The 800m represented the mean distance between independent biangulations/triangulations among all genets. To represent availability, software Geospatial Modelling Environment (Beyer 2012) was used to generate 20 random points within each buffer. To better represent availability inside a buffer, random points were set apart at least 100m following the same criteria used to set grid resolution. Usually, the criterion employed to choose the number of points is arbitrary (eg: Klar et al. 2008; Shafer et al. 2012). Here, the number of random locations was decided upon simulations performed by Northrup et al. (2013). Their simulations showed that a minimum of 20 points per buffer was enough to provide reliable coefficient estimates. Predictive landscape variables were measured on each used and random point. Prior to statistical analyses, skewed variables were transformed to approach normality and to reduce the influence of extreme values, using logarithmic transformations. Given the possible range of distance values exhibited by continuous variables, it is expected that they have a greater influence on w(x) scores on equation 1 or 2 than the forest predictor. To minimize this unbalanced scale effect between a categorical and continuous predictors, all variables were standardized using a ztransformation (Quinn & Keough 2002). Multicollinearity between predictors was assessed with Spearman correlation and variance inflation factors (VIF). Predictive variables were considered correlated and eliminated if they presented values of | r |>0.5 and VIF>3 (Zuur et al. 2009). The best combination of variables (best model) was determined through an information-theoretic approach (ITA), following criteria established by Burnham & Anderson (2002). First, univariate conditional logistic regression was applied to remove unimportant predictors (p>0.25; Hosmer and Lemeshow 2000). Models including the remaining important landscape variables (i.e. p<0.25) were considered in the candidate model set, if they presented a ∆AIC<2 (difference of AIC between a particular model and the best model). The best global model would be chosen if it presented an Akaike weight (w i )>0.9 alone (probability of being the best model), otherwise model averaging would be applied to all models in the set (Akaike weight sum (w i )>0.9), in order to obtain the best average parameter estimates across all variables (Burnham & Anderson 2002). Conditional logistic regression was fitted using the “clogit” function from survival package in R, while model selection was accomplished using MuMIn package (R Development Core Team 2012). The final logistical equation (in the form of equation 2) was applied to each grid cell to estimate RSF scores. Given that there are not records of genets crossing large water bodies, a RSF score of 0 was assigned to each cell containing water bodies. Predictive capacity of the top model was evaluated using 5-fold cross validation (Boyce et al. 2002). The data from the 21 individuals (used and available locations) was divided into five bins (classes) where each bin contained approximately 20% of the total dataset. Data across bins was assumed independent because points belonging to a single individual were not scattered into FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 27 different bins (Hirzel et al. 2006). Conditional logistic coefficient estimates were recalculated using 80% (four bins) of the dataset as training data while withholding the remaining bin as test data. This procedure was repeated until all bins were used as testing data. Five logistical equations were estimated and re-applied to each cell. Hence, five new suitability maps expressing probability of use were created. To examine top model performance, the approach proposed by Boyce et al. (2002) was employed. For each one of the five maps, RSF probability scores were divided into four equal intervals (from a probability of 0 to 1.0) and ranked from 1 (interval with lowest RSF values) until 4 (interval with highest probability use values). Using R software (R Development Core Team 2012), Spearman-rank correlation test was performed between rank categories and the respective area adjusted frequencies of used points. The area adjusted frequencies are calculated by simply dividing the number of used points falling in grid cells exhibiting a particular rank interval by the total area (expressed in number of cells) that a particular rank/category occupies in the suitability map. If correlation between ranks and frequencies is high, it means that used locations fall within higher ranked areas (more suitable areas) indicating that the best model calculated using all data has a good predictive performance. Usually, the number of chosen rank intervals is arbitrary (eg: Pullinger & Johnson 2010; Kunkel et al. 2013). Here, only four intervals were chosen since testing datasets were relatively small, especially compared to other studies employing GPS telemetry. If several classes of RSF scores were created, there would be a higher risk of having categories with no observations (even if they were suitable) due to data sparseness (Wiens et al. 2008). Accordingly, keeping four suitability classes (classes can be interpreted as low, medium, high and very high probabilities of use) was considered as reasonable choice that conciliates model accuracy and simplicity. 2.5.3-Landscape genetics analyses The suitability map calculated with the top model’s conditional logistic equation was converted into a resistance vectorial map. Resistance scores in each grid cell were calculated by applying a simple formula to RSF probability scores: R=[1-w*(x)]100 (3) where R is the resistance score and w*(x) is the RSPF score. The resistance values range from 0 to 100. The vectorial IBR map was rasterized with ArcGis 10 (ESRI 2011) and exported to Circuitscape version 3.5.8 (McRae 2006) using ArcGis tool “Export to Circuitscape” (Jenness Enterprises 2010). Circuit theory based algorithms such as the one implemented in Circuitscape were chosen due to the ability to account for multiple possible paths to estimate resistance between nodes (here they represent individuals; see section “1.2.2Landscape genetics as a tool FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 34 Fig.10Resistance map constructed using conditional logistic equation from the top model. Whiter areas present higher resistance values and are mainly correspondent to urban areas and roads. Darker areas represent suitable areas that have low resistance values such as montado forests and riparian corridors. 4-DISCUSSION 4.1-Microsatellite performance Several studies have reported a broad interval of rates in microsatellite genotyping (Pompanon et al. 2005; Beja-Pereira et al. 2009). Given the lower quality and quantity of DNA, it is expected that non-invasive genetics samples (such as hairs) will present higher error rates. Despite being recorded occasionally low error rates such as 2% (Bonin et al. 2004), genotyping errors superior to 10% (Broquet et al. 2007) and 20% (Gagneux et al. 1997; Johnson & Haydon 2007) can be frequently registered. This issue is less noticeable when using high quality samples (Bonin et al. 2004; Pompanon et al. 2005), despite that great dropout rates have been registered on special cases (Soulsbury et al. 2007). Consensus regarding the establishment of a critical error threshold has not been reached, where some argue that even low genotyping errors levels such as 1% may introduce significant biases (Hoffman & Amos 2005), while others advocate that an error <2% is acceptable (Bonin et al. 2004). Trying to establish a critical acceptable threshold constitutes a dubious approach once the severity of the effects that errors cause is conditioned by the scientific goals of FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 35 Table 7Model 5-fold cross validation using Spearman-rank correlation test (rho). Table 8-Mantel and partial Mantel correlation results. Ggenetic distances; The | symbol separates the genetic and model matrices from the covariate matrix. For example, G~IBR | IBD means that a partial Mantel test was performed between G and IBR model, while removing the effects of the IBD matrix. Model Rank frequencies Rho p Partition 1 1 0.00000 1.0 0.04 2 0.00050 3 0.00100 4 0.00383 Partition 2 1 0.00071 0.8 0.17 2 0.00064 3 0.00081 4 0.00385 Partition 3 1 0.00000 0.8 0.17 2 0.00045 3 0.00120 4 0.00081 Partition 4 1 0.00081 0.6 0.21 2 0.00057 3 0.00101 4 0.00092 Partition 5 1 0.00000 1.0 0.04 2 0.00039 3 0.00114 4 0.00230 Model topology Mantel r P G ~ IBD -0.129 <0.001 G ~ IBB -0.002 0.915 G ~ IBR -0.110 <0.001 G ~ IBD | IBR -0.070 <0.001 G ~ IBR | IBD 0.010 0.642 FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 36 Fig.11Current map created by Circuitscape for the IBR model. Dark blue areas represent higher values of current (i.e., areas highly permeable to movement) and lighter areas represent low values of current (i.e., with lower probability of being crossed by a random walker). Fig. 12– Current map created by Circuitscape for IBD model. Similarly to Fig. 11, darker areas correspond to high current zones, while lighter pixels represent low current areas. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 37 a particular study (Taberlet et al. 1999). Consequently, each particular case must be careful analyzed. In this work, hair samples presented high allele dropout rates, where eight markers exhibited more than 10% (table 4). This highlights the importance of conducting quality procedures to minimize errors in low quality samples. The protocol used here to control hair genotyping errors was similar to the ones employed in another studies (Frantz et al. 2004; Mullins et al. 2010). Given that three replicates were performed for homozygotes, allied with the removal of problematic hair samples and the use of protocols that diminish dropout (Goossens et al. 1998), it is believed that the presence of dropout in hair multilocus genotypes was minimal. A third allele was scored on seven occasions among replicates, increasing false allele rates. Scored genotypes were completely distinct, i.e. did not involve adjacent alleles, which allows to discard miscoring due to stutter patterns (Dewoody et al. 2006). Since third replicates unambiguously corroborated all the second replicates, it is possible that a human factor (possibly contamination) during the first genotyping contributed for the observed “extra alleles” (Hoffman & Amos 2005). It is hard to ascertain the possible cause, but third replicates likely reduced the possibility of false alleles. High quality samples showed little genotyping error rates. This confirms the expectations that for these types of samples, less rigorous quality control protocols such as re-genotyping of a subset of samples can constitute a valid tool when logistical resources are limited (Dewoody et al. 2006). Nevertheless, two markers presented error rates higher than zero in high-quality samples: Ggen 2.1 and A108 (table 4). The latter marker A108 had a concerning dropout rate >5% for this kind of samples. From the 23 high quality samples re-genotyped, only one had a genotype not concordant with the first scoring. The reason for the relative high rate of allele dropout present by this marker may be related to the way that error rates were estimated. Allele dropout per heterozygote calculations are weighted by the proportion of available heterozygotes (Wang 2004; Johnson & Haydon 2007). Accordingly, if there is a low proportion of heterozygotes (A108 presented a Ho=0.333), ε 1 errors will be inflated by each heterozygote where an allele fails to amplify. Additionally, given the high quality sample and the small product size generated by A108 (Sefc et al. 2003; Hoffman & Amos 2005), here it was considered that this marker had an insignificant impact in the multilocus dataset. Initially, marker Ggen 2.1 exhibited very high levels of dropout in multiplex for tissue and blood samples. Performing singleplex reactions twice for all samples allowed the discovery of problematic alleles, decreasing the dropout rate to about 2%. However, and given the high expected heterozygosity presented by Ggen 2.1, there was a homozygote excess that caused deviations from HW equilibrium, even if related individuals were experimentally removed from equilibrium analysis. A significantly different from zero null allele frequency (accounting for inbreeding using software INEst) was detected, constituting another possible explanation for the observed homozygote excess. In this particular case, knowing if this marker was only affected by dropout (and consequently null allele presence is considered as a “false FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 38 positive” result caused by high rates of dropout) or if it was affected by both processes is hard to ascertain. Soulsbury et al. (2007) reported similar problems regarding allele dropout for specific loci amplified on high quality samples, supporting the idea that more research is needed concerning this problematic. Clearly, the most viable option was to remove this marker from posterior statistical procedures. Marker polymorphism constitutes a key characteristic to retain genetic signatures caused by recently emerged landscape features that negatively affect dispersal (Wang 2010). Here, markers exhibited a moderately high heterozygosity levels and moderate allele numbers (mean allele number disregarding marker Ggen 2.1 was 5.69; number of alleles ranged from three to ten), considering the sample size and the small spatial scale that most individuals were sampled. Simulation analysis performed by Landguth et al. (2012) revealed that allele number is fundamental to increase statistical power for partial Mantel tests. However, it can be compensated by employing a high number of markers. Other landscape genetics studies conducted at similar spatial scales and using genetic markers with comparable diversity values (mean number of alleles ranged from 5.4 to 7.5 among the cited studies), had resolution enough to assess the effects of landscape features on gene flow (Hepenstrick et al. 2012; Apodaca et al. 2012; Koen et al. 2012). Common genets have great dispersal ability and short generation times (2 years), which are biological features that increase the power to reliably detect recent genetic signals (Landguth et al. 2010; Landguth et al. 2012). Accordingly, based on the procedures used here to minimize genotyping errors and on the comparable diversity values found on previous studies, it is unlikely that landscape genetics analysis was compromised by a high genotyping error rate or due to hypothetical insufficient polymorphism levels. Fig. 13 – Estimated cluster membership in Geneland. X and Y axis represent UTM coordinates. Given that the highway cross the area in a west-east axis, it is visible by this figure that the population is not structured by the highway. Small black dots represent samples locations. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 39 4.2Landscape genetics analyses 4.2.1-Best model and RSF validity The match case-control design employed here detected landscape variables that were positively associated with the presence of genets, constituting a result in accordance with previous studies conducted in Mediterranean environments (eg: Larivière & Calzada 2001; Galantinho & Mira 2009; Matos et al. 2009). Genets significantly selected for forest cover over agricultural areas, which confirms the previous importance given by other researchers to forest habitats (including riparian corridors) on meeting food, shelter, reproductive and hydric requirements (Galantinho & Mira 2009; Santos et al. 2011; Camps & Alldredge 2013). On the contrary, it has been advocated that agricultural habitat types such as vine yards, pastureland and farmland in general lack proper native shrub and tree cover, presenting low availability of resting sites and food resources (Pereira & Rodríguez 2010; Camps 2011). A low distance to the nearest urban features was a variable that was associated with less used locations. Again, this result meets previous expectations that common genets avoid urban areas (Galantinho & Mira 2009; Camps & Alldredge 2013). It is known that anthropogenic areas with high disturbance levels (eg: roads, villages) may repel carnivores within a “buffer-disturbance zone”, reducing the functional available habitat in the vicinity of these areas (eg: Forman & Deblinger 2000; Crooks 2002; Ditchkoff et al. 2006). Finally, the variable “distance to water bodies” was the only predictor that was not significantly related to genet’s presence. Water in riparian ecosystems are restricted to small pools during the summer (Matos et al. 2009; Santos et al. 2011). Considering the harsh conditions faced by species in Mediterranean areas during the dry season, additional sources of a limiting resource such as water can be valuable. Two reasons could have contributed for the non-significant results obtained for this variable. First, if water reservoirs are indeed selected by genets (or other carnivores), then a positive selection would be expected only during the dry season. Some studies managed to analyse their data by constructing different models for each season (eg: Shafer et al. 2012; Squires et al. 2013). Confronting different temporal models (eg: summer vs winter) is appropriate to assess habitat selection in regions where there is a relevant temporal contrast of resource availability (McLoughlin et al. 2010). However, in the current work, the number of observed locations on dry season only comprised about 30% of the total dataset used in the logistic equation. Accordingly, and given the small dataset used, partitioning the data into two seasonal small unequal subsets could have introduced important bias on equation’s coefficients (Rice et al. 2013). The second reason is concerned with an intimate connection between human water use and agricultural practices. Water reservoirs in Mediterranean areas play an important role in agriculture, where water use is intrinsically associated with culture irrigation and water source for cattle. The presence of unsuitable habitat such as irrigated cultures nearby water reservoirs and cattle herding, will likely difficult the access of genets to these landscape elements (Brotons et al. 2004; Mestre et al. 2007; Pita et al. 2009). FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 40 Model evaluation was performed using a common approach: k-fold cross validation. This approach is especially useful when one does not possess independent data to validate the estimated landscape model (Boyce et al. 2002; Chetkiewicz & Boyce 2009; Kunkel et al. 2013). In this validation three out of five data subsets failed to give significant results despite presenting relatively high correlation coefficients (table 7). The third partition did not presented significant correlations because areas with high use probabilities (rank 3) presented a higher frequency of used points than more suitable zones (rank 4). Nevertheless, the higher frequencies were detected at the most suitable ranks and thus, it was considered that this bin was acceptable (it also presented a Spearman-rank correlation of 0.8). The other two partitions (partition 2 and partition 4) exhibited non-significant correlations due to high frequencies on rank 1 (unsuitable areas). In both cases, results were largely affected by one single animal which represented special cases. The testing dataset from partition 2 contained one subadult individual that comprised about 70% of the total observations in rank 1 habitat, contributing for the observed high frequency on this rank. Additionally, the animal held a home range where 70% of area has probability suitability scores inferior to 0.5 (rank 1 and rank 2). Due to intraspecific competition, it is common that sometimes subadults are forced by adults to establish their territories in suboptimal areas (Mergey et al. 2011). This particular individual also represented 73% of the observations in rank 4 habitat. Accordingly, despite being surrounded by unsuitable habitat, its territory also comprised highly suitable areas, probably being fundamental for the subadult’s survival. In partition 4, the majority (about 70%) of observed records in unsuitable habitat (rank 1) belong to an adult female. Indeed, this female foraged often in agricultural areas during the radio-tracking period, probably taking advantage during the winter of a greater abundance of migratory birds such as lapwings (Vanellus vanellus) in farmland habitats (Moreira et al. 2005; Sánchez et al. 2008; Carvalho pers. observation). Doing a simple modelling exercise, by removing these two cited individuals, predictive performance in partition 2 and 4 rises to significant maximum values (rho=1). Accordingly, it was considered that the general model had a good predictive ability in the study area. 4.2.2-Isolation-by-barrier Matching initial expectations, the barrier model presented the lowest fit with genetic data. The results presented by the IBB model are corroborated by parental analysis (Fig. 7) and Geneland results (Fig. 13), reinforcing the little genetic effects caused by this linear infrastructure. From radio-tracking data, it was observed that all home ranges from animals living nearby were bounded by the highway, with almost none locations obtained in the opposite side of the highway, reinforcing at least the behavioural barrier effect (see Fig. 4 and Fig. S1; Carvalho et al. in prep.). This observation gives some clues about the permeability of this particular feature, being apparent that it interferes with movement to some degree. Disrupting movement and gene flow is fundamentally dependent of species and barrier attributes (Holderegger & Giulio 2010). The FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 41 behaviour exhibited towards roads (road avoidance, car avoidance) plays an important part in determining species ecological and genetic sensitivity towards this phenomenon (Fahrig & Rytwinski 2009; Jackson & Fahrig 2011). On the other hand, the width of the physical structure, traffic volumes, barrier’s age and presence of movement conductors (eg: culverts) may determine the overall effects that a given linear feature may exert on a species (Jaeger et al. 2005; Jaarsma et al. 2006). Empirical studies showing the genetic effects of particular barriers have been published (eg: Riley et al. 2006; Zalewski et al. 2009; Frantz et al. 2010). However, there are important differences regarding infrastructure characteristics. The highway analyzed in this study presents a mean monthly daily traffic of about 6400 vehicles (InIR 2011) and a width of 30 m and thus, these characteristics are not comparable with previous research: (1) it has lower width and considerably lower traffic volume (for example daily traffic volume is 15 times lower than the motorway assessed in Riley et al. 2006); (2) it is crossed by a large number of culverts and viaducts and it does not have tall fences, which increase permeability for wildlife movement (Ascensão & Mira 2007; Hepenstrick et al. 2012); (3) the lag time (time between barrier construction/formation and the genetic response) is also an important factor to be considered (Landguth et al. 2010). Since its construction, only about 7-8 generations in common genets have experienced highway effects, which may be not sufficient time for the population to genetically respond to the new disturbance feature. This highlights the temporal disconnection between movement data and population genetic responses (Anderson et al. 2010; Spear et al. 2010). Once movement data describes patterns during a specific period of time where they were measured, population genetic structure has a lag time response towards the changing landscape. Hence, the combination of the three factors previously mentioned has likely contributed for the absence of a genetic response, suggesting that immigration and genetic connectivity between northern and southern areas are not seriously affected by this highway, at least at short term (Mills & Allendorf 1996; but see also Vucetich & Waite 2000). Nevertheless, the eventual small effects on genets of this particular barrier may not be applicable to other carnivore species. Other medium-sized carnivores presenting smaller population sizes such as polecats (Mustela putorius; Cabral et al. 2005), may experience more rapidly the barrier effects caused by this particular structure. On these species, overall population genetic diversity may decrease much faster, not only due to the barrier effects mediated by road mortality, but also through home ranges pile-up which increases the difficulties for dispersing individuals to establish a territory, and consequently the mortality rate may grow (Riley et al. 2006; Holderegger & Giulio 2010). Reduction in genetic diversity may decrease population fitness and cause a loss of evolutionary potential (Frankham 2005). Future monitoring of highway effects may be advisable for these sensible species. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 42 4.2.3-Isolation-by-distance and IBR model performance Despite IBD and IBR models being significantly supported by Mantel tests, partial Mantel analysis depicted a higher independent support for a pattern of isolation-by-distance. The partial Mantel results go against to what was initially hypothesized. Direct spatial measurements of movement obtained in telemetry studies only document that an individual travelled a particular distance through particular habitat(s). Spatial and temporal scales at which both processes operate can be completely different, possibly leading to incongruence between gene flow and dispersal (Anderson et al. 2010; Spear et al. 2010). For example, movement reflects contemporaneous patterns of dispersal, where gene flow represents historical dispersal events in the landscape during multiple generations. Thus, assessing connectivity with different types of data (genetic or movement) may yield different results. Additionally, biological features such as sex, life stage and other specific traits from a given species may present different patterns of dispersal and gene flow, and failing to integrate that data could hamper the results obtained (Coulon et al. 2004; Fedy et al. 2008). Therefore caution must be taken when using empirical movement data. This issue is fairly discussed for example by Spear and colleagues (2010) in the scope of resistance surface parameterization. So far, studies that managed to combine landscape genetics and empirical dispersal data assumed that the latter is an indirect surrogate of gene flow (Shafer et al. 2012; Reding et al. 2013). This assumption is legit several times, once both metrics are generally correlated (Bohonak 1999) and the landscape variables selected for foraging and other ecological activities likely represent in most cases, suitable area that is used by a species for reproductive purposes (being informative of landscape permeability to gene flow) (Beier et al. 2008; Cushman & Lewis 2010; Baguette et al. 2013). Based on previous studies in Mediterranean environments, there is no reason to reject the hypothesis that the most suitable habitats, such as forest and riparian corridors, are in general habitats used by genets (males and females) for reproductive and survival purposes (Larivière & Calzada 2001; Barrientos 2006; Camps & Alldredge 2013; Filipe et al. in prep; but see below). However, if the RSF represents realistically habitats conductive or resistant to gene flow, why the IBR model performed worse than the IBD model? Several explanations or their combination could explain this result. The most likely contributing factor that explains lack of support of the landscape resistance model is derived from sampling scheme. The fact that samples locations were obtained from locations not specifically adjusted for this particular study possibly hampered a proper investigation of the role of habitat on gene flow. The majority of samples are clustered in a northwest-southeast strip of relatively contiguous montado forest with about 20 km of length (Fig. 5). Accordingly, pairwise node calculations between those individuals involve mainly suitable forested areas (montado), don’t giving the chance to test the effects of other low permeable landscape features. Recently, this was shown by Oyler-McCance et al. (2013) where they performed simulations and detected that clustered sampling designs are not adequate for landscape genetics studies. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 43 Additionally, by joining the current map from IBR model (highly permeable areas are concordant with the corridor; Fig. 9) and genet’s great dispersal ability, as corroborated by radio-tracking data (in the study area mean daily dispersal was about 2.5 km, while maximum dispersal distance was about 10 km; Carvalho et al. in prep) and the distance between related individuals (Fig. 7 and Fig. 8), this issue in the sampling design becomes more evident. Thus, the combination of these factors likely contributed for highest support for the IBD model. Other studies also found more support for the IBD model (Broquet et al. 2006; Koen et al. 2012). In those studies, despite the distribution of samples was not clustered, sampling was performed across areas strongly covered by suitable habitat, promoting high gene flow rates. In those cases, the population genetic differentiation becomes mainly dependent of the geographical distance separating individuals, partially being similar to what happened in this study. Another feature that may have contributed for the results obtained is the importance of riparian ecosystems as enhancers of dispersal. The food and water supplies provided and the availability of resting sites, make riparian areas important areas for carnivore survival, even if surrounded by an inhospitable matrix (Virgós 2001; Matos et al. 2009; Santos et al. 2011). Probably the best study that empirically demonstrates the value of linear corridors for dispersal in agricultural mosaics was conducted in Spain by Pereira & Rodríguez (2010). Despite the higher density and use of other linear habitat features such as hedgerows, their study illustrates well the importance that riparian corridors may present in assuring functional connectivity. The large use of riparian corridors by the genet may allow them to perceive fragmented habitat as functionally continuous if forest areas are connected by these elements. If this is the case, then genes “movement” is mainly determined by geographical distance in patches separated by farmlands but also connected by riparian ecosystems. Riparian habitat is extensive and well preserved in the study area. These areas are also accounted as highly permeable habitat features by the RSF model, emphasizing their biological and ecological importance. Furthermore, in the study dataset, two dispersers walked significant distances (approx. 3km and 7km, respectively) in these corridors surrounded by farmlands, constituting additional evidence of the importance of linear elements in Mediterranean agro-forestry systems (Carvalho et al. in press.). However, this connectivity pattern may be poorly represented by the multi-path algorithm implemented in software Circuitscape. The level of pairwise resistance assigned by the software is proportional to the number of pathways connecting two nodes (McRae 2006). Assuming that both nodes (individuals) are best connected by a single optimal path (riparian corridor) integrated in an inhospitable matrix, the overall effective distance between them may be overestimated since the number of possible paths providing inter-node connectivity are limited to the cells containing low resistance (i.e. represented by riparian vegetation) (see also Spear et al. 2010). Hence, circuit theory can perform worse than LCP on these particular cases, contributing for a lower performance of the IBR model (McRae 2006). FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 50 Bryson RW, García-Vázquez UO, Riddle BR (2011) Phylogeography of Middle American gophersnakes: mixed responses to biogeographical barriers across the Mexican Transition Zone. Journal of Biogeography, 38, 1570–1584. Burnham K, Anderson D (2002) Model selection and multimodel inference - a practical informationtheoretic approach. Springer-Verlag, New York, USA. Cabral M, Almeida J, Almeida P et al. (2005) Livro vermelho dos vertebrados de Portugal. Instituto da Conservação da Natureza, Lisboa. Calzada J (1998) Gineta común, Genetta genetta (Linnaeus, 1758). Galemys, 10, 3–11. Calzada J (2007) Genetta genetta (Linnaeus, 1758). In: Atlas y Libro Rojo de los Mamíferos Terrestres de España (eds Palomo L, Blanco J, Gisbert J), pp. 330–332. Dirección General para la Biodiversidad, SECEM-SECEMU, Madrid. Campagne P, Smouse PE, Varouchas G, Silvain J, Leru B (2012) Comparing the van Oosterhout and Chybicki-Burczyk methods of estimating null allele frequencies for inbred populations. Molecular Ecology Resources, 12, 975–982. Camps D (2011) Resting site selection, characteristics and use by the common genet Genetta genetta (Linnaeus 1758 ). Mammalia, 75, 23–29. Camps D, Alldredge JR (2013) Multi-Scale habitat use and selection of common genet Genetta genetta (Viverridae, Carnivora) in a Mediterranean environment. Mammalia, 1–11. Carvalho F, Carvalho R, Mira A, Beja P (in prep.) Flexibility of resting site use by a Mediterranean forest carnivore. CCDRA (2012-2014) Biodiversity conservation in a changing world. IC&DTn1/SAESCTN/ALENT07-0224-FEDER-001755). Universidade de Évora and CIBIO-UE. Centro de Geofísica de Évora (2013). http://cge. uevora.pt. Last acessed on September 2013. Chapuis M-P, Estoup A (2007) Microsatellite null alleles and estimation of population differentiation. Molecular Biology and Evolution, 24, 621–631. Chetkiewicz C-LB, Boyce MS (2009) Use of resource selection functions to identify conservation corridors. Journal of Applied Ecology, 46, 1036–1047. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 51 Chybicki IJ, Burczyk J (2009) Simultaneous estimation of null alleles and inbreeding coefficients. The Journal of Heredity, 100, 106–113. Coulon A, Cosson JF, Angibault J-M, Cargnelutti B, Galan M, Morellet N, Petit E, Aulagnier S, Hewison AJM (2004) Landscape connectivity influences gene flow in a roe deer population inhabiting a fragmented landscape: an individual – based approach. Molecular Ecology, 13, 2841– 2850. Coulon A, Guillot G, Cosson J-F, Angibault J-M, Aulagnier S, Cargnelutti B, Galan M, Hewison AJM (2006) Genetic structure is influenced by landscape features: empirical evidence from a roe deer population. Molecular Ecology, 15, 1669–1679. Coulon A, Morellet N, Goulard M, Cargnelutti B, Angibault J-M, Hewison AJM (2008) Inferring the effects of landscape structure on roe deer (Capreolus capreolus) movements using a step selection function. Landscape Ecology, 23, 603–614. Craiu R V, Duchesne T, Fortin D, Baillargeon S (2011) Conditional logistic regression with longitudinal follow-up and individual level random coefficients: a stable and efficient two-step estimation method. Journal of Computational and Graphical Statistics, 20, 767–784. Crooks KR (2002) Relative sensitivities of mammalian carnivores to habitat fragmentation. Conservation Biology, 16, 488–502. Crooks KR, Burdett CL, Theobald DM, Rondinini C, Boitani L (2011) Global patterns of fragmentation and connectivity of mammalian carnivore habitat. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 366, 2642–2651. Crooks KR, Sanjayan M (2006) Connectivity conservation: maintaining connections for nature. In: Connectivity Conservation (eds Crooks KR, Sanjayan M), pp. 1–28. Cambridge University Press, New York, NY. Crooks J, Suarez A (2006) Hyperconnectivity, invasive species, and the breakdown of barriers to dispersal. In: Connectivity Conservation (eds Crooks KR, Sanjayan M), pp. 451–478. Cambridge University Press, New York, NY. Cushman S (2006) Effects of habitat loss and fragmentation on amphibians: A review and prospectus. Biological Conservation, 128, 231–240. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 52 Cushman S, Lewis JS (2010) Movement behavior explains genetic differentiation in American black bears. Landscape Ecology, 25, 1613–1625. Cushman S, McKelvey KS, Hayden J, Schwartz MK (2006) Gene flow in complex landscapes: testing multiple hypotheses with causal modeling. The American Naturalist, 168, 486–99. Cushman S, Raphael MG, Ruggiero LF, Shirk A, Wasserman TN, O'Doherty EC (2011) Limiting factors and landscape connectivity: the American marten in the Rocky Mountains. Landscape Ecology, 26, 1137–1149. Cushman S, Wasserman T, Landguth E, Shirk A (2013) Re-Evaluating causal modelling with Mantel tests in landscape genetics. Diversity, 5, 51–72. Dakin EE, Avise JC (2004) Microsatellite null alleles in parentage analysis. Heredity, 93, 504–509. Delaney KS, Riley SPD, Fisher RN (2010) A rapid, strong, and convergent genetic response to urban habitat fragmentation in four divergent and widespread vertebrates. PloS One, 5, 1–11. Delibes M, Mézan-Muxart V, Calzada J (2013) Albino and melanistic genets (Genetta genetta) in Europe. Acta Theriologica, 58, 95–99. Delibes M, Rodriguez A, Parreno F (1989) Food of the common genet (Genetta genetta) in northern Africa. Journal of Zoology, London, 218, 321–326. Dennis B (2002) Allee effects in stochastic populations. Oikos, 96, 389–401. Dewoody J, Nason JD, Hipkins VD (2006) Mitigating scoring errors in microsatellite data from wild populations. Molecular Ecology Notes, 6, 951–957. Didham RK, Ghazoul J, Stork NE, Davis AJ (1996) Insects in fragmented forests: a functional approach. Trends in Ecology & Evolution, 11, 255–260. Ditchkoff SS, Saalfeld ST, Gibson CJ (2006) Animal behavior in urban ecosystems: modifications due to human-induced stress. Urban Ecosystems, 9, 5–12. Dobson M, Wright A (2000) Faunal relationships and zoogeographical affinities of mammals in north-west Africa. Journal of Biogeography, 27, 417–424. Driscoll D, Banks SC, Barton PS, Lindenmayer DB, Smith AL (2013) Conceptual domain of the matrix in fragmented landscapes. Trends in Ecology & Evolution, 28, 605–613. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 53 Duchesne T, Fortin D, Courbin N (2010) Mixed conditional logistic regression for habitat selection studies. The Journal of Animal Ecology, 79, 548–555. Dyer RJ, Nason JD, Garrick RC (2010) Landscape modelling of gene flow: improved power using conditional genetic distance derived from the topology of population networks. Molecular Ecology, 19, 3746–3759. ESRI (2011). ArcGis Desktop: Release 10. Environmental Systems Research Institute, Redlands, California, USA. Ewers RM, Thorpe S, Didham RK (2007) Synergistic interactions between edge and area effects in a heavily fragmented landscape. Ecology, 88, 96–106. Fahrig L (2001) How much habitat is enough? Biological Conservation, 100, 65–74. Fahrig L (2003) Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution, and Systematics, 34, 487–515. Fahrig L, Rytwinski T (2009) Effects of roads on animal abundance: an empirical review and synthesis. Ecology and Society, 14, 1–21. Fedy BC, Martin K, Ritland C, Young J (2008) Genetic and ecological data provide incongruent interpretations of population structure and dispersal in naturally subdivided populations of whitetailed ptarmigan (Lagopus leucura). Molecular Ecology, 17, 1905–1917. Fernandes C, Rodrigues M, Hajji GM, Vercammen P, Bruford MW, Santos-Reis M (2009) Isolation and characterisation of 11 tetranucleotide microsatellite loci in the common genet (Genetta genetta). Conservation Genetics, 10, 1931–1934. Fieberg J, Matthiopoulos J, Hebblewhite M, Boyce MS, Frair JL (2010) Correlation and studies of habitat selection: problem, red herring or opportunity? Philosophical Transactions of the Royal Society of London. Series B, Biological sciences, 365, 2233–44. Fischer J, Lindenmayer DB (2002) The conservation values of paddock trees for birds in a variegated landscape in southern New South Wales. 2. Paddock trees as stepping stones. Biodiversity & Conservation, 11, 833–849. Fischer J, Lindenmayer DB (2007) Landscape modification and habitat fragmentation: a synthesis. Global Ecology and Biogeography, 16, 265–280. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 54 Foley J, Defries R, Asner GP Carol Barford, Bonan G, Carpenter SR, Chapin FS, Coe MT, Daily GC, Gibbs HK, Helkowski JH, Holloway T, Howard EA, Kucharik CJ, Monfreda C, Patz JA, Prentice IC, Ramankutty N, Snyder PK (2005) Global consequences of land use. Science, 309, 570–574. Forman RTT, Deblinger RD (2000) The ecological road-effect zone of a Massachusetts (USA) suburban highway. Conservation Biology, 14, 36–46. Frankham R (2005) Genetics and extinction. Biological Conservation, 126, 131–140. Frantz AC, Pope LC, Etherington TR, Wilson GJ, Burke T (2010) Using isolation-by-distancebased approaches to assess the barrier effect of linear landscape elements on badger (Meles meles) dispersal. Molecular Ecology, 19, 1663–1674. Frantz AC, Schaul M, Pope LC, Fack F, Schley L, Muller C, Roper T (2004) Estimating population size by genotyping remotely plucked hair: the Eurasian badger. Journal of Applied Ecology, 41, 985–995. Gagneux P, Boesch C, Woodruff DS (1997) Microsatellite scoring errors associated with noninvasive genotyping based on nuclear DNA amplified from shed hair. Molecular Ecology, 6, 861–868. Galantinho A, Mira A (2009) The influence of human, livestock, and ecological features on the occurrence of genet (Genetta genetta): a case study on Mediterranean farmland. Ecological Research, 24, 671–685. Garroway CJ, Bowman J, Wilson PJ (2011) Using a genetic network to parameterize a landscape resistance surface for fishers, Martes pennanti. Molecular Ecology, 20, 3978–3988. Gaubert P, Del Cerro I, Palomares F, Godoy J (2008a) Development and characterization of 11 microsatellite loci in a historically introduced carnivoran, the common genet (Genetta genetta). Molecular Ecology Resources, 8, 1375–1377. Gaubert P, Fernandes CA, Bruford MW, Veron G (2004a) Genets (Carnivora , Viverridae) in Africa: an evolutionary synthesis based on cytochrome b sequences and morphological characters. Biological Journal of the Linnean Society, 81, 589–610. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 55 Gaubert P, Godoy J, Cerro I, Palomares F (2009) Early phases of a successful invasion: mitochondrial phylogeography of the common genet (Genetta genetta) within the Mediterranean Basin. Biological Invasions, 11, 523–546. Gaubert P, Jiguet F, Bayle P, Angelici FM (2008b) Has the common genet (Genetta genetta) spread into southeastern France and Italy? Italian Journal of Zoology, 75, 43–57. Gaubert P, Machordom A, Morales A, López-Bao JV, Veron G, Amin M, Barros T, Basuony M, Djagoun CAMS, San E, Fonseca C, Geffen E, Ozkurt S, Cruaud C, Couloux A and Palomares F (2011) Comparative phylogeography of two African carnivorans presumably introduced into Europe: disentangling natural versus human-mediated dispersal across the Strait of Gibraltar. Journal of Biogeography, 38, 341–358. Gaubert P, Mézan-Muxart V (2010) Where have the “black genets” gone? A likely restriction of melanistic cases of the common genet (Genetta genetta) to its introduced range. Mammalian Biology, 75, 353–357. Gaubert P, Taylor PJ, Veron G (2005a) Integrative taxonomy and phylogenetic systematics of the genets (Carnivora, Viverridae, Genetta): a new classification of the most speciose carnivoran genus in Africa. In: African Biodiversity (eds Huber B, Sinclair B, Lampe K), pp. 371–383. Springer, New York. Gaubert P, Tranier M, Delmas A, Colyn M, Veron G (2004b) First molecular evidence for reassessing phylogenetic affinities between genets (Genetta) and the enigmatic genet-like taxa Osbornictis , Poiana and Prionodon (Carnivora , Viverridae). Zoologica Scripta, 33, 117–129. Gaubert P, Wozencraft WC, Cordeiro-Estrela P, Veron G (2005a) Mosaics of convergences and noise in morphological phylogenies: what’s in a viverrid-like carnivoran? Systematic Biology, 54, 865–894. Gehring TM, Swihart RK (2003) Body size, niche breadth, and ecologically scaled responses to habitat fragmentation: mammalian predators in an agricultural landscape. Biological Conservation, 109, 283–295. Gilbert-Norton L, Wilson R, Stevens JR, Beard KH (2010) A meta-analytic review of corridor effectiveness. Conservation Biology, 24, 660–668. Goossens B, Waits LP, Taberlet P (1998) Plucked hair samples as a source of DNA: reliability of dinucleotide microsatellite genotyping. Molecular Ecology, 7, 1237–1241. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 56 GRASS Development Team (2013) Geographic Resources Analysis Support System (GRASS) Software. Open Source Geospatial Foundation. http://grass.osgeo.org. Graves T, Beier P, Royle JA (2013) Current approaches using genetic distances produce poor estimates of landscape resistance to interindividual dispersal. Molecular Ecology, 22, 3888–3903. Graves T, Wasserman TN, Ribeiro MC, Landguth E, Spear S, Balkenhol N, Higgins CB, Fortin MJ, Cushman S, Waits L (2012) The influence of landscape characteristics and home-range size on the quantification of landscape-genetics relationships. Landscape Ecology, 27, 253–266. Guillot G, Estoup A, Mortier F, Cosson JF (2005) A spatial statistical model for landscape genetics. Genetics, 170, 1261–1280. Hall T (1999) BioEdit: a user-friendly biological sequence alignment editor and analysis program for Windows 95/98/NT. Nucleic Acids Symposium Series, 41, 95–98. Hanski I (2010) Landscape fragmentation, biodiversity loss and the societal response. EMBO reports, 6, 388–392. Hellborg L, Ellegren H (2003) Y chromosome conserved anchored tagged sequences (YCATS) for the analysis of mammalian male-specific DNA. Molecular Ecology, 12, 283–291. Hellborg L, Ellegren H (2004) Low levels of nucleotide diversity in mammalian Y chromossomes. Molecular Biology and Evolution, 21, 158–163. Hepenstrick D, Thiel D, Holderegger R, Gugerli F (2012) Genetic discontinuities in roe deer (Capreolus capreolus) coincide with fenced transportation infrastructure. Basic and Applied Ecology, 13, 631–638. Hess G (1994) Conservation corridors and contagious disease: a cautionary note. Conservation Biology, 8, 256–262. Hirzel AH, Le Lay G, Helfer V, Randin C, Guisan A (2006) Evaluating the ability of habitat suitability models to predict species presences. Ecological Modelling, 199, 142–152. Hodgson J, Moilanen A, Wintle B, Thomas C (2011) Habitat area, quality and connectivity: striking the balance for efficient conservation. Journal of Applied Ecology, 48, 148–152. Hoffman J, Amos W (2005) Microsatellite genotyping errors: detection approaches, common sources and consequences for paternal exclusion. Molecular Ecology, 14, 599–612. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 57 Holderegger R, Giulio M Di (2010) The genetic effects of roads: A review of empirical evidence. Basic and Applied Ecology, 11, 522–531. Holderegger R, Wagner HH (2008) Landscape Genetics. BioScience, 58, 199–207. Holt R, Robinson G, Gaines M (1995) Vegetation dynamics in an experimentally fragmented landscape. Ecology, 76, 1610–1624. ICN (2006) Plano sectorial da Rede Natura 2000. Instituto para a Conservação da Natureza, Lisboa. InIR (2011) Relatório de tráfego na rede nacional de auto-estradas. 1º trimestre de 2011. Instituto de Infra-estruturas Rodoviárias, Lisboa. Jaarsma CF, van Langevelde F, Botma H (2006) Flattened fauna and mitigation: Traffic victims related to road, traffic, vehicle, and species characteristics. Transportation Research Part D: Transport and Environment, 11, 264–276. Jackson ND, Fahrig L (2011) Relative effects of road mortality and decreased connectivity on population genetic diversity. Biological Conservation, 144, 3143–3148. Jaeger JAG, Bowman J, Brennan J, Fahrig L, Bertc D, Bouchardc J, Charbonneauc N, Frankd K, Gruberd B, von Toschanowitz K (2005) Predicting when animal populations are at risk from roads: an interactive model of road avoidance behavior. Ecological Modelling, 185, 329–348. Jaquiéry J, Broquet T, Hirzel H, Yearsley J, Perrin N (2011) Inferring landscape effects on dispersal from genetic distances: how far can we go? Molecular Ecology, 20, 692–705. Jeness Enterprises (2010) Export to Circuitscape tool for ArcGis 10. Flagstaff, Arizona, USA. Johnson DH (1980) The comparison of usage and availability measurements for evaluating resource preference. Ecology, 61, 65–71. Johnson PCD, Haydon DT (2007) Maximum-likelihood estimation of allelic dropout and false allele error rates from microsatellite genotypes in the absence of reference data. Genetics, 175, 827– 842. Johnson C, Nielsen SE, Merrill E, McDonald T, Boyce MS (2006) Resource selection functions based on use-availability data: theoretical motivation and evaluation methods. Journal of Wildlife Management, 70, 347–357. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 58 Jones O, Wang J (2010) COLONY: a program for parentage and sibship inference from multilocus genotype data. Molecular Ecology Resources, 10, 551–555. Kindlmann P, Burel F (2008) Connectivity measures: a review. Landscape Ecology, 23, 879–890. Klar N, Fernández N, Kramer-Schadt S, Herrmanne M, Trinzenf M, Buttnerf I, Niemitz C (2008) Habitat selection models for European wildcat conservation. Biological Conservation, 141, 308– 319. Koen EL, Bowman J, Garroway CJ, Mills SC, Wilson PJ (2012) Landscape resistance and American marten gene flow. Landscape Ecology, 27, 29–43. Kunkel KE, Atwood TC, Ruth TK, Pletscher DH, Hornocker MG (2013) Assessing wolves and cougars as conservation surrogates (M Gompper, A Vanak, Eds.). Animal Conservation, 16, 32– 40. Lande R (1993) Risks of population extinction from demographic and environmental stochasticity and random catastrophes. The American Naturalist, 142, 911–927. Landguth EL, Cushman SA, Schwartz MK, McKelvey K, Murphy M, Luikart G (2010) Quantifying the lag time to detect barriers in landscape genetics. Molecular Ecology, 19, 4179–4191. Landguth EL, Fedy BC, Oyler-McCance SJ, Garey A, Emel S, Mumma M, Wagner H, Fortin MJ, Cushman S (2012) Effects of sample size, number of markers, and allelic richness on the detection of spatial genetic pattern. Molecular Ecology Resources, 12, 276–284. Larivière S, Calzada J (2001) Genetta genetta. Mammalian Species, 680, 1–6. Livet F, Roeder J (1987) La genette (Genetta genetta). In: Encyclopèdie des Carnivores de France (eds Artois M, Delattre P). Socité Française pour l’Etude et la Protection des Mammifères. Manel S, Schwartz MK, Luikart G, Taberlet P (2003) Landscape genetics: combining landscape ecology and population genetics. Trends in Ecology & Evolution, 18, 189–197. Mantel N (1967) The detection of disease clustering and a generalized regression approach. Cancer Research, 27, 209–220. Markoulatos P, Siafakas N, Moncany M (2002) Multiplex Polymerase Chain Reaction: A Practical Approach. Journal of Clinical Laboratory Analysis, 16, 47–51. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 59 Matos HM, Santos MJ, Palomares F, Santos-Reis M (2009) Does riparian habitat condition influence mammalian carnivore abundance in Mediterranean ecosystems? Biodiversity and Conservation, 18, 373–386. McCallum H, Dobson A (2006) Disease and connectivity. In: Connectivity Conservation (eds Crooks KR, Sanjayan M), pp. 479–501. Cambridge University Press, New York, NY. McLoughlin PD, Morris DW, Fortin D, Vander Wal E, Contasti AL (2010) Considering ecological dynamics in resource selection functions. Journal of Animal Ecology, 79, 4–12. McRae (2006) Isolation by resistance. Evolution, 60, 1551–1561. Mergey M, Helder R, Roeder J-J (2011) Effect of forest fragmentation on space-use patterns in the European pine marten (Martes martes). Journal of Mammalogy, 92, 328–335. Mestre F, Ferreira J, Mira A (2007) Modelling the distribution of the european polecat Mustela putorius in a Mediterranean agricultural landscape. Revue d'Écologie, 62, 35-47. Mills L, Allendorf FW (1996) The one-migrant-per-generation rule in conservation and management. Conservation Biology, 10, 1509–1518. Moilanen A, Hanski I (2006) Connectivity and metapopulation dynamics in highly fragmented landscapes. In: Connectivity Conservation (eds Crooks KR, Sanjayan M), pp. 44–71. Cambridge University Press, New York, USA. Moreira F, Beja P, Morgado R, Reino L, Gordinho L, Delgado A, Borralho R (2005) Effects of field management and landscape context on grassland wintering birds in Southern Portugal. Agriculture, Ecosystems & Environment, 109, 59–74. Mortelliti A, Amori G, Boitani L (2010) The role of habitat quality in fragmented landscapes: a conceptual overview and prospectus for future research. Oecologia, 163, 535–547. Mullins J, Statham MJ, Roche T, Turner PD, O’Reilly C (2010) Remotely plucked hair genotyping: a reliable and non-invasive method for censusing pine marten (Martes martes, L. 1758) populations. European Journal of Wildlife Research, 56, 443–453. Murcia C (1995) Edge effects in fragmented forests: implications for conservation. Trends in Ecology & Evolution, 10, 58–62. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 66 Vallone PM, Butler JM (2004) AutoDimer: a screening tool for primer-dimer and hairpin structures. BioTechniques, 37, 226–231. Verhoeven KJF, Simonsen KL, Mcintyre LM (2005) Implementing false discovery rate control: increasing your power. Oikos, 108, 643–647. Viard F, Franck P, Dubois MP, Estoup A, Jarne P (1998) Variation of microsatellite size homoplasy across electromorphs, loci, and populations in three invertebrate species. Journal of Molecular Evolution, 47, 42–51. Vilà C, Sundqvist A-K, Flagstad Ø, Seddon J, Bjornerfeldt S, Kojola I, Casulli A, Sand H, Wabakken P, Ellegren H (2003) Rescue of a severely bottlenecked wolf (Canis lupus) population by a single immigrant. Proceedings of the Royal Society Series B: Biological Sciences, 270, 91–97. Virgós E (2001) Relative value of riparian woodlands in landscapes with different forest cover for medium-sized Iberian. Biodiversity and Conservation, 10, 1039–1049. Virgós E, Casanovas JG (1997) Habitat selection of genet Genetta genetta in the mountains of central Spain. , 42, 169–177. Virgós E, Llorente M, Cortés Y (1999) Geographical variation in genet (Genetta genetta L.) diet: a literature review. Mammal Review, 29, 119–128. Virgós E, Romero T, Mangas JG (2001) Factors determining “gaps” in the distribution of a small carnivore, the common genet (Genetta genetta), in central Spain. Canadian Journal of Zoology, 79, 1544–1551. Virgós E, Tellería J, Santos T (2002) A comparison on the response to forest fragmentation by medium-sized Iberian carnivores in central Spain. Biodiversity and Conservation, 11, 1063–1079. Vucetich JA, Waite TA (2000) Is one migrant per generation sufficient for the genetic management of fluctuating populations? Animal Conservation, 3, 261–266. Waits LP, Paetkau D (2005) Noninvasive genetic sampling tools for wildlife biologists: a review of applications and recommendations for accurate data collection. Journal of Wildlife Management, 69, 1419–1433. Wang J (2004) Sibship reconstruction from genetic data with typing errors. Genetics, 166, 1963– 1979. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 67 Wang IJ (2010) Recognizing the temporal distinctions between landscape genetics and phylogeography. Molecular Ecology, 19, 2605–2608. Wang Y-H, Yang K-C, Bridgman CL, Lin L-K (2008) Habitat suitability modelling to correlate gene flow with landscape connectivity. Landscape Ecology, 23, 989–1000. Weir BS, Anderson AD, Hepler AB (2006) Genetic relatedness analysis: modern data and new challenges. Nature Reviews Genetics, 7, 771–780. Wiegand T, Revilla E, Moloney K (2005) Effects of habitat loss and fragmentation on population dynamics. Conservation Biology, 19, 108–121. Wiens TS, Dale BC, Boyce MS, Kershaw GP (2008) Three way k-fold cross-validation of resource selection functions. Ecological Modelling, 212, 244–255. Wright S (1943) Isolation by distance. Genetics, 28, 114–138. Zabala J, Zuberogoitia I, Garin I, Aihartza J (2001) Small carnivore trappability: seasonal changes and mortality. A case study on European mink Mustela lutreola and spotted genet Genetta genetta. Small Carnivore Conservation, 25, 9–11. Zalewski A, Piertney S, Zalewska H, Lambin X (2009) Landscape barriers reduce gene flow in an invasive carnivore: geographical and local genetic structure of American mink in Scotland. Molecular Ecology, 18, 1601–1615. Zeller K, McGarigal K, Whiteley AR (2012) Estimating landscape resistance to movement: a review. Landscape Ecology, 27, 777–797. Zuur A, Ieno E, Walker N, Saveliev A, Smith G (2009) Mixed effects models and extensions in ecology with R. Springer, New York, USA. FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 68 6-SUPPLEMENTARY MATERIAL Table S1-Information about samples used on genetic analysis. ID Sex Type Source Circuit_coord UTM X UTM Y MG2 male muscle Roadkill roadkill site 575787 4281626 MG3 female muscle roadkill roadkill site 585927 4271838 MG4 female muscle roadkill roadkill site 583250 4273209 MG5 male muscle roadkill roadkill site 588657 4265063 MG6 male muscle roadkill roadkill site 585842 4271007 MG8 female muscle roadkill roadkill site 585210 4275953 MG9 male muscle roadkill roadkill site 575172 4281302 MG10 male muscle roadkill roadkill site 578896 4282932 MG11 male muscle roadkill roadkill site 582620 4273270 MG12 female muscle roadkill roadkill site 576720 4276266 MG13 female muscle roadkill roadkill site 585902 4271851 MG14 male muscle roadkill roadkill site 578797 4282902 MG15 male muscle roadkill roadkill site 578359 4275613 MG16 female muscle roadkill roadkill site 579546 4274900 MG17 female muscle roadkill roadkill site 585399 4272022 MG18 male muscle roadkill roadkill site 588741 4269885 MG19 male muscle roadkill roadkill site 585962 4271821 MG20 female muscle roadkill roadkill site 583489 4273166 MG21 female muscle roadkill roadkill site 575555 4281501 MG22 female muscle roadkill roadkill site 560103 4277637 MG23 male muscle roadkill roadkill site 560132 4277851 MG24 female muscle roadkill roadkill site 577489 4282490 MG25 male muscle roadkill roadkill site 591783 4268153 MG26 male muscle roadkill roadkill site 579127 4275118 MG27* male muscle roadkill roadkill site 573899 4277628 MG28 female muscle roadkill roadkill site 590799 4269336 MG29 female muscle roadkill roadkill site 586268 4276537 MG30 female muscle roadkill roadkill site 572875 4277923 MG32 male muscle roadkill roadkill site 575509 4281480 MG33 female muscle roadkill roadkill site 601777 4281015 MG34 male muscle roadkill roadkill site 598265 4274124 MG35 male muscle roadkill roadkill site 577195 4278937 MG38 male muscle roadkill roadkill site 563039 4295087 MG40 female muscle roadkill roadkill site 580605 4253356 MG57 male muscle roadkill roadkill site 584135 4272853 MG58 female muscle roadkill roadkill site 576221 4281931 MG60 male muscle roadkill roadkill site 587086 4271686 MG61 female muscle roadkill roadkill site 580642 4274274 SG1 male blood trapping trapping site 584766 4267180 SG2 female blood trapping home range centroid 585076 4267688 SG3 female blood trapping trapping site 584766 4267180 SG4* female blood trapping trapping site 585265 4271789 SG5 male blood trapping home range centroid 575898 4277504 SG6 female blood trapping home range centroid 581904 4274659 SG7* male blood trapping home range centroid 581822 4274885 SG8* male blood trapping home range centroid 581145 4277207 SG9 male blood trapping home range centroid 581682 4271162 SG10 female blood trapping home range centroid 582484 4276702 SG11 female blood trapping home range centroid 581971 4271354 FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 69 SG12 male Blood trapping home range centroid 581336 4277925 SG13* male Blood trapping trapping site 581987 4277597 SG14 female Blood trapping home range centroid 583152 4276679 SG15 male blood trapping home range centroid 583101 4277281 SG16 male blood trapping trapping site 575728 4275303 SG17 male blood trapping trapping site 575782 4275863 SG18 female blood trapping trapping site 576102 4275343 SG19 male blood trapping home range centroid 577903 4274289 SG20 female blood trapping home range centroid 577603 4274137 SG21 female blood trapping trapping site 577238 4275090 SG22 female blood trapping home range centroid 577367 4283813 SG23 male blood trapping home range centroid 574078 4285081 SG24 female blood trapping home range centroid 575567 4284219 SG25 female blood trapping home range centroid 573715 4285095 SG26* female blood trapping home range centroid 582991 4282132 SG27* male blood trapping trapping site 582094 4282721 SG28 female blood trapping home range centroid 580312 4283003 SG29 female blood trapping trapping site 581240 4279801 PG1 1 male hair trapping roadkill site 575441 4265795 PG2 male hair trapping home range centroid 584335 4271526 PG3 female hair roadkill roadkill site 560425 4261346 PG4 female hair roadkill roadkill site 554934 4293825 PG5 female hair trapping home range centroid 576723 4283693 PG6 female hair trapping trapping site 575480 4284369 PG7 female hair roadkill roadkill site 587724 4276095 PG11 1 undefined hair roadkill roadkill site 604531 4284471 PG13 undefined hair roadkill roadkill site 560906 4267437 Circuit_coord – Criterion employed to define UTM X and UTM Y coordinates. These coordinates were set as nodes in Circuitscape. 1 Samples not included in genetic analysis. *Sample not included in HW and LD tests. Table S2-Spearman correlation matrix. Table S3VIF scores of landscape variables. hab dist_rip dist_water dist_rip 0.299 dist_water 0.121 0.208 dist_human 0.131 -0.012 0.086 Variable VIF hab 1.103 dist_rip 1.083 dist_water 1.037 dist_human 1.022 FCUP Combining movement and genetic data to assess a forest carnivore’s response to forest fragmentation 70 Fig.S1Small portion of the study area illustrating the home ranges calculated for 21 genets. The home ranges that overlap belong to individuals form different sex.