Ecological factors and gene flow in wolves

Molecular Ecology (2006) 15, 4533– 4553

doi: 10.1111/j.1365-294X.2006.03110.x

Ecological factors influence population genetic structure of European grey wolves Blackwell Publishing Ltd

M A L G O R Z A T A P I L O T ,* W L O D Z I M I E R Z J E D R Z E J E W S K I ,† W O J C I E C H B R A N I C K I ,‡ V A D I M E . S I D O R O V I C H ,§ B O G U M I L A J E D R Z E J E W S K A ,† K R Y S T Y N A S T A C H U R A † and S T E P H A N M . F U N K ¶** *Museum and Institute of Zoology, Polish Academy of Sciences, Ul Wilcza 64, 00-679 Warszawa, Poland, †Mammal Research Institute, Polish Academy of Sciences, 17-230 Bialowie*a, Poland, ‡Institute of Forensic Research, Ul Westerplatte 9, 31-033 Kraków, Poland, §Institute of Zoology, National Academy of Sciences of Belarus, Akademicheskaya Str 27, 220072 Minsk, Belarus, ¶Institute of Zoology, Zoological Society of London, London RW1 4RY, UK

Abstract Although the mechanisms controlling gene flow among populations are particularly important for evolutionary processes, they are still poorly understood, especially in the case of large carnivoran mammals with extensive continuous distributions. We studied the question of factors affecting population genetic structure in the grey wolf, Canis lupus, one of the most mobile terrestrial carnivores. We analysed variability in mitochondrial DNA and 14 microsatellite loci for a sample of 643 individuals from 59 localities representing most of the continuous wolf range in Eastern Europe. We tested an array of geographical, historical and ecological factors to check whether they may explain genetic differentiation among local wolf populations. We showed that wolf populations in Eastern Europe displayed nonrandom spatial genetic structure in the absence of obvious physical barriers to movement. Neither topographic barriers nor past fragmentation could explain spatial genetic structure. However, we found that the genetic differentiation among local populations was correlated with climate, habitat types, and wolf diet composition. This result shows that ecological processes may strongly influence the amount of gene flow among populations. We suggest natal-habitatbiased dispersal as an underlying mechanism linking population ecology with population genetic structure.

Keywords: cryptic genetic structure, gene flow, genetic diversification, grey wolf, natal-habitatbiased dispersal, predator–prey interaction Received 4 April 2006; revision received 29 June 2006; accepted 25 July 2006

Introduction Understanding the micro evolutionary process that generates population genetic structure of large and highly mobile carnivoran mammals is crucial for improving our knowledge of the mechanisms of their adaptive divergence and speciation. Classical population genetics explains the population genetic structure by species behavioural traits (forming herds, flocks or colonies), geographical features limiting gene flow, such as spatial distance and topographic Correspondence: MaLgorzata Pilot, Fax: +48-22-6296302; E-mail: mpilot@miiz.waw.pl. **Present address: Nature Heritage Ltd., 145-157 St. John Street, London, UK, and Durrell Wildlife Conservation Trust, Les Augres Manor, Jersey JE3 5BP, UK. © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

barriers (Hartl & Clark 1997), or historical factors such as past colonization, range expansion or isolation in different glacial refugia (Hewitt 1996, 2000; Taberlet et al. 1998; Templeton 1998). However, besides geographical limitations and historical events, complex ecological processes may influence the amount of gene flow among populations.

Indeed, an increasing number of studies indicate cryptic genetic structures that cannot be explained either by geographical or historical factors (e.g. Sponer & Roy 2002; Spinks & Shaffer 2005). Strikingly, many of these studies concern large and medium-sized carnivoran mammals with extensive continuous distributions: grey wolf Canis lupus (Carmichael et al. 2001; Geffen et al. 2004), coyote Canis latrans (Sacks et al. 2004), lynx Lynx lynx and Lynx canadensis (Rueness et al. 2003a, b), puma Puma concolor (McRae et al.

2005), and arctic fox Alopex lagopus (Dalén et al. 2005). High

4534 M . P I L O T E T A L . mobility of these animals and their ability to cross most of potential topographic barriers (such as rivers or mountain ranges) minimize the influence of geographical factors on gene flow and reduce the effects of historical events, so that the effect of ecological factors may be more prominent.

The grey wolf is one of the most mobile terrestrial mammals that disperse rapidly over distances up to 900 km (Fritts 1983; Mech & Boitani 2003). Dispersing individuals were reported to successfully cross four-lane highways and circumvent large lakes and cities (Mech et al. 1995; Merrill & Mech 2000; Wabakken et al. 2001). The historical range of wolves covered nearly the entire Holarctic, from tundra to grasslands and deserts (Nowak 2003). Long-distance dispersal capabilities combined with the ability to occupy a variety of habitats imply high rates of gene flow that reduce genetic differentiation among local populations. Indeed, a study based on mitochondrial DNA (mtDNA) control region sequence data from a worldwide sample of grey wolves suggested an absence of a large-scale genetic structure, and indicated local, small-grained structure probably caused by the recent restricted gene flow (Vilà et al. 1999).

On the other hand, considerable morphological differentiation, that may be a result of genetic divergence, is observed within the species (Nowak 2003). Moreover, two studies on North American grey wolves reported nonrandom patterns of gene flow that may result from ecology and behaviour of the species. A study on microsatellite variability of grey wolves from the Canadian Northwest (Carmichael et al.

2001) revealed population genetic structure that corresponds with migration patterns of caribou Rangifer tarandus, the main prey of wolf in this region. On a larger scale, it was shown that vegetation types and climate influence genetic dissimilarities among grey wolf populations in North America (Geffen et al. 2004). Results of that study may have been restricted by small sample size. Therefore, we further investigated the problem of the effect of environmental and ecological factors on population genetic structure in large carnivores, based on an extensive sample of grey wolves from Eastern Europe. We analysed mtDNA control region sequences and 14 microsatellite loci for 643 individuals from 59 localities, distributed across a diversity of habitats and climatic zones. The analysis of both types of markers revealed nonrandom population genetic structure. We tested its dependence on historical, geographical and ecological factors, aiming to identify underlying mechanisms of genetic differentiation among wolf populations.

Materials and methods

Greece, and the European part of Turkey. These localities represent most of the area within the continuous range of the species in Europe (see Fig. 4 in Results). The number of samples in a locality varied from 2 to 37, with an average of 11. Most samples (97%) dated from the years 1995–2004.

Older samples (dated from the years 1958–1994) were pelts of wolves killed by hunters in Poland. A group of wolves from one locality will be referred to as a local population (we did not assume that a discrete population occurred in each locality; however, the definition of sample groups was necessary for population-based analyses).

Laboratory methods DNA extraction from soft tissues was performed using A & A Biotechnology extraction kit. QIAamp DNA Mini Kit (QIAGEN) was used for DNA extraction from pelts. DNA extraction from teeth was performed following the protocol of Yang et al. (1998) modified by Wandeler et al. (2003).

Amplification of 257 bp of the HV1 domain of the mtDNA control region was performed using the primers from Vilr et al. (1997). The polymerase chain reaction (PCR) mixture was made up of 1 U Taq polymerase, 200 µm dNTP, 2.0 µL 10 × concentrated PCR buffer, 1.5 mm MgCl2, 0.1 mm of primers and 4 µL of DNA for 20 µL reactions. The reaction conditions were as follows: 2 min at 94 °C of initial denaturation, 36 – 40 cycles of 20 s at 94 °C, 30 s at 69 °C, 40 s at 72 °C, and the final elongation step for 10 min at 72 °C. Negative controls were added to each set of samples during extraction as well as during PCR amplification to control for contamination. PCR products were purified using the QIAquick PCR Purification Kit (QIAGEN). Sequencing reactions were performed using BigDye Terminator Cycle Sequencing Kit (PerkinElmer) and detection of sequencing reaction products was carried out on ABI PRISM 3100 genetic analyser (Applied Biosystems). Sequencing results were analysed with ABI PRISM DNA Sequencing Analysis software, version 3.0, and alignments were performed using sequence navigator 2.0.

We also analysed 14 microsatellite loci: FH2001, FH2010, FH2017, FH2054, FH2079, FH2088, FH2096 (Francisco et al.

1996), C213, C250, C253, C466, C642 (Ostrander et al. 1993), AHT130 (Holmes et al. 1995) and VWF (Shibuya et al. 1994).

Microsatellites were amplified in five multiplexes, using Multiplex PCR Kit (QIAGEN) and the PCR conditions described in manufacturer’s instruction (with the annealing temperature 58 °C). PCR products were analysed on ABI PRISM 3100 genetic analyser. Allele lengths were determined using genescan 3.7 and genotyper 3.7 software.

Samples

Estimation of the total number of mtDNA haplotypes

We analysed 643 tissue samples of wolves from 59 localities situated in 10 countries: Poland, Lithuania, Latvia, Belarus, Ukraine, the European part of Russia, Slovakia, Bulgaria,

The fact that the number of haplotypes increases with the number of analysed samples was used to estimate the total number of wolf haplotypes in the study area and compare © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4535 it with the number of actually found haplotypes. We constructed a rarefaction curve plotting the cumulative number of haplotypes found with increasing sample size. The total number of haplotypes was estimated as the asymptote of this curve (Kohn et al. 1999; Leonard et al. 2005). As the sampling order affects the shape of the curve, the data set was randomised 1000 times (without replacement of haplotypes) using gimlet (Valière 2002) and 1000 rarefaction curves were generated using the r package (Ihaka & Gentleman 1996) and a script file produced by gimlet. The asymptote (a) for each curve was calculated from the equation: y = ax/(b + x), where y is the cumulative number of haplotypes, x is the number of sampled individuals, and b is the rate of decline in the slope of the curve (Kohn et al.

1999). The number of haplotypes was estimated as the mean value of the asymptote a for all iterations.

Analysis of mtDNA variability: phylogenetic analysis To test phylogenetic relationships among the haplotypes, we constructed phylogenetic trees in paup 4.0b10 (Swofford 1998) using a HKY + Γ model of nucleotide substitution with a shape parameter of the gamma distribution α = 0.0736, as estimated in modeltest 3.6 (Posada & Crandall 1998).

The phylogenies were rooted with two coyote sequences from GenBank (Accession nos AF008158, AF020700).

We constructed trees using neighbour-joining, minimumevolution, maximum-likelihood, and maximum-parsimony algorithms. Confidence in estimated relationships was determined by calculating bootstrap values, which were obtained through 1000 replicates, using the heuristic search algorithm implemented in paup. Additionally, we constructed a Bayesian tree in mrbayes 3.1 (Huelsenbeck & Ronquist 2001) using the HKY + Γ model of nucleotide substitution, as estimated in mrmodeltest 2.2 (Nylander 2004). In the Markov chain Monte Carlo simulation, four chains were run simultaneously for 1 million generations.

Trees were sampled every 10 generations for a total of 100 000 trees in the initial sample. Stationarity of the process was determined to have occurred by the 10 000th trees and therefore ‘burn-in’ was completed by this stage. Thus, the tree and clade credibility values were obtained from 90 000 trees. The heterogeneity of mutation rates among lineages was tested by comparing the log-likelihoods of maximumlikelihood trees obtained with and without enforcing molecular clock, using the likelihood-ratio test of Shimodaira & Hagesawa (1999).

In order to estimate the coalescence time of Eastern European wolves, we calculated mean sequence divergence within wolves and net sequence divergence (corrected for ancestral within-species polymorphism) between wolves and coyotes, using the program mega 3.1 (Kumar et al.

2004). The standard error of these estimates was calculated with 1000 bootstrap pseudo-replicates. Because the HKY © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

model of nucleotide substitution is not implemented in mega, and because this model is a special case of the Tamura– Nei model (Nei & Kumar 2000), we used the Tamura–Nei model with a shape parameter of the gamma distribution α = 0.10, as estimated in tree-puzzle (Schmidt et al. 2000).

The minimum evolution tree constructed in mega using the Tamura–Nei model had similar topology as the trees constructed in paup using the HKY + Γ model.

The nested clade analysis (NCA) (Templeton 1998, 2004) of the geographical distribution of mtDNA haplotypes was performed to separate effects of a recurrent gene flow and historical factors, such as past fragmentation, colonization or range expansion. Statistical parsimony approach implemented in the software tcs (Clement et al. 2000) was used to construct the minimum spanning network, which was nested according to the rules described in Templeton et al. (1992) and Templeton & Sing (1993). The hypothesis of no geographical association of nested clades was tested using the program geodis (Posada et al. 2000). Results were interpreted using the inference key from Templeton (2004).

Additionally, we compared grey wolf haplotypes found in our study with those reported in previous studies (Vilr et al. 1999; Randi et al. 2000). This allowed us to identify haplotypes that have been previously found and to relate haplotypes from Eastern Europe to previously published phylogenetic trees of worldwide grey wolf haplotypes (Vilr et al. 1999; Leonard et al. 2005).

Analysis of mtDNA variability: frequency-based analysis To analyse population genetic structure, we used the spatial analysis of molecular variance implemented in the samova software (Dupanloup et al. 2002). This method defines groups of local populations that are geographically homogenous and maximally differentiated from each other. The method is based on a simulated annealing procedure that aims to maximize the proportion of total genetic variance due to differences between groups of populations, measured by ΦCT coefficient of the amova Φ-statistics (Excoffier et al.

1992). In contrast to classical tests of genetic structure (such as amova), in which groups of populations are defined a priori, the samova procedure finds a structure based solely on genetic data and geographical location of populations.

However, this approach requires the a priori definition of the number (K) of groups. Thus, we ran samova successively on our data set with different K, ranging from 2 to 20. An analysis with each K-value was performed twice to check whether results are consistent between runs. In each run, 100 simulated annealing processes were performed. The identification of the most probable number of groups was based on the pattern of changes in values of Φ-statistics parameters with K. Although the idea of the samova procedure is to ensure that the inferred groups are composed of adjacent populations,

4536 M . P I L O T E T A L . it can sometimes lead to the definition of groups in which all the populations are not geographically adjacent (Dupanloup et al. 2002). Thus, after identifying the most probable population genetic structure with samova, we modified it so as to receive geographically homogenous groups. These groups will be referred to as subpopulations of the total population of Eastern European wolves. We used amova procedure implemented in the arlequin software (Schneider et al. 2000) to calculate Φ-statistics for the inferred population genetic structure and test its significance. For a comparison, we also calculated Φ-statistics for a random grouping of local populations into 10 groups. Using the program contrib (Petit et al. 1998), we calculated haplotype diversity and allelic richness for the inferred subpopulations.

Allelic richness employs rarefaction method to control for the effect of sample size on the number of haplotypes, and thus allows comparing genetic diversity among groups of different size.

Additionally, we used the software spagedi (Hardy & Vekemans 2002) to calculate pairwise ΦST between local populations (an analogue of FST for haplotypic data). Next, we performed the Mantel test to check for the correlation between genetic distances (measured as linearized pairwise ΦST) and log-transformed geographical distances between local populations.

Analysis of genetic variability in microsatellite loci Population genetic structure in microsatellite loci was investigated using the geneland 1.0.5 software (Guillot et al.

2005b). geneland provides a Bayesian clustering method that allows making use of georeferenced individual multilocus genotypes for the inference of the number (K) and spatial distribution of subpopulations. In this software, all unknown parameters are inferred through MCMC computations. In our inference, we used a similar procedure as described by Coulon et al. (2006). At first, we ran the MCMC 10 times, allowing K to vary, with the following parameters: 200 000 MCMC iterations, maximum rate of Poisson process fixed to 500, uncertainty attached to spatial coordinates fixed to 0.5° (i.e. the minimal precision of our sample locations), minimum K fixed to 1, maximum K fixed to 20, maximum number of nuclei in the Poisson-Voronoi tessellation fixed to 200, and the Dirichlet model as a model for allelic frequencies. Next, we inferred the number of subpopulations from the modal K of these 10 runs, and ran MCMC 20 times with K fixed to this number and other parameters unchanged. We computed the posterior probability of subpopulation membership for each pixel of the spatial domain and the modal subpopulation for each individual for each of the 20 runs (with a burn-in of 20 000 iterations).

We also calculated the mean logarithm of posterior probability for each run. Finally, we checked the consistency of the results across these 20 runs. For each of these runs, we

Fig. 1 (a) Map of Europe indicating the study area. (b) Schematic map of 16 regions — spatial units used in the Mantel test and distance-based redundancy analysis for microsatellite data. Each circle represents one region and is situated in its centroid. Circle size reflects sample size. Darker grey area denotes the continuous wolf range in Europe, based on Sulkava & Pulliainen (1999), Jedrzejewski et al. (2002) and Boitani (2003), modified.

tested the significance of the inferred structure by performing a two-level amova (among and within subpopulations) with arlequin.

For population-based analyses of genetic variability in microsatellite loci, spatial units (sample groups) larger than local populations were desired to avoid potential biases (e.g. resulting from the presence of closely related individuals in the sample). Thus, we grouped local populations into 16 regions (Fig. 1), based on their geographical proximity, similarity of habitats, and genetic discontinuities revealed from the analysis of population genetic structure. Using the software spadedi (Hardy & Vekemans 2002), we calculated Nei’s standard genetic distance (DS) and pairwise FST between the regions. Next, we performed the Mantel test to check for © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4537 the correlation between genetic distances (measured either as DS or linearized pairwise FST) and log-transformed geographical distances between the regions.

Analysis of a dependence of genetic diversification on environmental variables To examine whether environmental factors may explain genetic differentiation among local wolf populations, we used a distance-based redundancy analysis, which is a form of multivariate multiple regression that can be performed directly on a genetic distance response matrix (Legendre & Anderson 1999; McArdle & Anderson 2001). Although partial Mantel test (Smouse et al. 1986) is the most common method of performing partial regression analyses for genetic distances (e.g. Carmichael et al. 2001; Sacks et al. 2004), the validity of this approach has been questioned (Raufaste & Rousset 2001; Rousset 2002). Therefore, following Geffen et al. (2004), the distance-based multivariate approach of McArdle & Anderson (2001) was used here instead of partial Mantel test.

In case of mtDNA, we used 59 localities as spatial units and pairwise ΦST as a measure of genetic differentiation.

In case of microsatellite markers, we used 16 regions as spatial units and Nei’s DS distances and pairwise FST values as measures of genetic differentiation. We tested for dependence of genetic differentiation among the sample groups on an array of predictor variables, grouped into five sets: (i) geographical distance (latitude and longitude); (ii) types of potential vegetation (lowland deciduous forests, lowland coniferous and mixed forests, mountain coniferous forests, forest-steppe, steppe, Mediterranean vegetation); (iii) temperature (mean annual temperature, mean January temperature and mean July temperature); (iv) mean annual rainfall; and (v) wolf diet composition (moose, Alces alces; red deer, Cervus elaphus; roe deer, Capreolus capreolus; and wild boar, Sus scrofa). All predictor variables except vegetation types were continuous. Vegetation types were presented as categorical variables, with two states: 1 if a sample group was located in a given vegetation category, and 0 if it was located in another vegetation category. Thus, each vegetation type was presented as a vector with values 0 and 1, and there were six such vectors corresponding to six vegetation types. All vegetation types were analysed as a set, i.e. they were combined in a single test.

Environmental variables were taken from the databases: WWF Terrestrial Ecoregions (data set provided by ESRI, www.esri.com) and WorldClimate (www.worldclimate.com).

The information about wolf diet composition (measured as a frequency of a given species in the total number of ungulates killed by wolves) in different localities of the study area was derived from published studies (Kerechun 1979; Vatolin 1979; Filonov 1989; Andersone 1998; Jedrzejewska & Jedrzejewski 1998; Sidorovich et al. 2003; Gula 2004; Nowak et al. 2005) © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

and unpublished master degree theses (Koniuch 2002; Kloch 2003; Nedzynska 2003; Wojtulewicz 2004) supervised by W. Jedrzejewski and J. Goszczynski. Only native and common ungulate species were considered.

Using the program distlm version 5 (Anderson 2004), we performed the marginal tests on individual sets of predictor variables, aiming to identify those variables that were correlated with genetic distance. The P values in these analyses were obtained using 9999 unrestricted, simultaneous permutations of the rows and columns of the distance matrix. Next, we performed the conditional tests, where latitude and longitude were included as covariables to individual sets of predictor variables or to multiple sets of predictor variables. The conditional tests allowed us to examine the extent to which any of the sets of predictor variables (or their combination) explains genetic diversification among wolf populations over and above that explained by geographical distance alone. The P values in these analyses were obtained using 9999 permutations of the rows and columns of the multivariate residual matrix under the reduced model (Anderson & Legendre 1999).

To examine which subset of predictor variables will provide the best model explaining genetic differences among wolf populations, we performed the forward selection procedure on all sets of variables, using the program distlm forward (Anderson 2003). The forward selection procedure consists of sequential tests, fitting each set of variables one at a time, conditional on the variables that were already included in the model. Most pairs of the tested predictor variables were correlated; for example, habitat types were correlated with temperature. However, the forward selection procedure allowed us to control for the correlations between the predictor variables. Similarly, as in the previous analysis, the P values were obtained using 9999 permutations of the rows and columns of the multivariate residual matrix under the reduced model. Results of the above tests allowed us to identify sets of variables that were important in explaining the genetic differentiation among wolf populations, while controlling for effects of geographical distance and other analysed variables.

Results Genetic variability of Eastern European wolves We found 21 haplotypes of mtDNA control region among analysed samples. Fourteen of these haplotypes have been known from previous studies (Vilr et al. 1999; Randi et al.

2000; Jedrzejewski et al. 2005) and seven were found for the first time (see Table 1 for GenBank Accession numbers).

Most localities (83%) had more than one haplotype and neighbouring localities frequently had the same haplotypes.

The number of haplotypes found in a locality was correlated with the number of analysed samples (r = 0.60, P < 0.001).

4538 M . P I L O T E T A L .

Table 1 Percentage frequencies of mtDNA haplotypes among subpopulations of wolves in Eastern Europe. Numbers of samples, different haplotypes, and unique haplotypes, as well as haplotype diversity and allelic richness are indicated for each subpopulation. The most common haplotypes in each subpopulation are marked in bold. For haplotypes found in earlier studies, ID numbers previously assigned are given: ‘lu’ denotes haplotype names from Vilr et al. (1999), and ‘RW’ from Randi et al. (2000). AF344300 and AF098123 are GenBank Accession numbers of haplotypes from Jedrzejewski et al. (2005) and an unpublished study of B. F. Koop and coworkers, respectively.

Accession numbers of new haplotypes found in this study are denoted by asterisks Subpopulations Haplotype w1 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 w13 w14 w15 w16 w17 w18 w19 w20 w21 N samples N different haplotypes N unique haplotypes Haplotype diversity Allelic richness

S1 67.1 11.3 10.3 1.1 0.7 6.4 1.1 0.3 0.3 0.3

S2 20.2 5.6 15.7 2.3 2.3 1.1 44.9 3.4

S3

S4

6.5 20.3 64.0 2.0 2.0

20.0

4.6

5.0

S5

S6

S7

18.2 9.1 63.6

15.0 10.0

S8

S9

S10

4.9 2.5 11.1

27.8 85.4

25.0

5.6

100

16.7

50.0 1.1 3.4

9.1 22.2 75.0 88.9 2.4

5.5 22.2

2.4 0.6 2.4 1.1 283 11 2 0.52 2.01

89 10 0 0.73 3.01

153 7 1 0.55 1.94

20 5 0 0.71 2.76

The total number of haplotypes within the study area was estimated from the rarefaction curve at 23 (mean estimate: 23.3 ± 1.4, median: 23.1, the range: 20.0–29.9).

Genetic variability in nuclear markers was assessed for 545 samples that were successfully genotyped in at least 11 of 14 analysed loci (most of samples for which the genotyping failed were tanned pelts). Mean number of alleles per locus in the total population was 11 (range 5–18). Observed heterozygosity estimated at 0.71 (SD = 0.10) was lower than expected heterozygosity estimated at 0.78 (SD = 0.08), and heterozygote deficiency was significant (P < 0.0001; see Appendix I).

Phylogenetic relationships among mtDNA haplotypes Nucleotide diversity among grey wolf haplotypes was 0.017 (SD = 0.009), and mean within-species sequence divergence was 0.032 (SE = 0.013). The net sequence divergence between wolves and their closest wild relatives, coyotes, was 0.334

11 1 0 0 0

11 4 0 0.60 2.40

9 2 1 0.22 0.89

41 6 2 0.27 1.14

8 2 1 0.43 1.00

ID lu12, RW8 AF344300 lu8 lu7, RW4 lu13, RW13 RW16 lu17 AY842293* AF098123 lu3, RW9 RW17 DQ421802* lu10, RW5 lu6 DQ421803* RW1 DQ421804* DQ421805* RW18 DQ421806* DQ421807*

18 6 2 0.84 3.58

(SE = 0.209). The likelihood-ratio test of Shimodaira & Hagesawa (1999) failed to reject the hypothesis of clock-like evolution of analysed sequences (P = 0.11). The phylogenetic relationships among haplotypes revealed the presence of two main clades (Fig. 2a). Most individuals (87%) had haplotypes from the clade 4-1. Bootstrap support values were low, most likely due to the small number (15) of parsimonyinformative sites between wolf sequences. However, all methods of tree construction provided similar topologies and supported these two main clades. Moreover, the minimum-spanning network approach that is considered to reflect intraspecific phylogenetic relationships better than phylogenetic trees (Crandall et al. 2000) also supported the subdivision of haplotypes into two main clades (Fig. 2b).

There was no clear geographical pattern in the distribution of haplotypes: the ranges of both clades extended over most of the study area. It indicates that the Eastern European wolf population does not have geographically distinct subunits that would be reciprocally monophyletic for mtDNA © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4539

Fig. 2 Phylogenetic relationships among mtDNA haplotypes of Eastern European wolves, based on 257 bp of control region sequence. (a) Minimum-evolution tree with maximum-likelihood distances. Bootstrap support is indicated at nodes if found in more than 50% of 1000 bootstrap trees. Additionally, bootstrap support is indicated for two main clades, which are named the same as the respective clades from the minimum-spanning network. (b) Minimum-spanning network of the haplotypes. Big circles represent the haplotypes and small circles indicate interior nodes that were absent from the sample because of insufficient sampling or extinct haplotypes. Each line represents a single mutational change. Similar haplotypes are grouped into nested clades, which are denoted by rectangles. The clades for which restricted gene flow with isolation by distance (RGF) has been indicated are marked in light grey, and the clades for which past events of the range expansion (RE) have been indicated are marked in dark grey.

haplotypes, suggesting there are no evolutionarily significant units (sensu Moritz 1994b). Haplotypes previously reported from Western Europe fall into the two main clades of Eastern European haplotypes: a haplotype from the Apennine Peninsula (lu5 in Vilr et al. 1999; W14 in Randi et al. 2000) fall into the clade 4-2, and haplotypes from the Iberian Peninsula (lu1, lu3, lu4 in Vilr et al. 1999; W19, W20 in Randi et al.

2000) fall into the clade 4-1. The Scandinavian haplotype (lu12 in Vilr et al. 1999) is identical with our haplotype w1 (see Table 1). Thus, also in the scale of the entire Europe, there are no evolutionarily significant units. In the trees of worldwide wolf haplotypes (Vilr et al. 1999; Leonard et al.

2005), haplotypes from both clades of Eastern European Š 2006 The Authors Journal compilation Š 2006 Blackwell Publishing Ltd

wolves do not form monophyletic branches, but are intermixed with haplotypes from Asia and North America.

Although the differentiation of the Eastern European wolf population is not strong enough to form reciprocally monophyletic subunits, haplotype frequencies substantially differ between local populations. A permutation categorical contingency analysis of the whole network rejected the null hypothesis of no association with geographical location (P < 0.0001), which indicated population differentiation.

For three clades, the NCA indicated restricted gene flow with isolation by distance. Of the remaining significant results, the NCA showed past events of range expansion in five clades of different levels (Fig. 2b; Appendix II).

4540 M . P I L O T E T A L .

Table 2 Φ-statistics parameters for different groupings of local wolf populations in Eastern Europe based on mtDNA: (1) The grouping revealed in samova with 10 groups assumed; (2) The grouping (1) modified so as to obtain spatially homogenous subpopulations (see Fig. 4); (3) The grouping (2) modified by pooling subpopulations S1 and S2; (4) Random grouping of local populations into 10 groups. For definitions of parameters ΦSC, ΦST, ΦCT, see Fig. 3 Subdivision

ΦSC

ΦST

ΦCT

P

(1) samova, 10 groups (2) Homogenous, 10 groups (3) Homogenous, 9 groups (4) 10 random groups

0.071 0.117 0.181 0.397

0.440 0.441 0.463 0.381

0.398 0.367 0.344 −0.021

< 0.00001 < 0.00001 < 0.00001 0.659

substantially differed in haplotype composition (Table 1) and in environmental characteristics, as indicated by mean values of analysed environmental variables for localities within each subpopulation (Table 3).

Population genetic structure inferred from microsatellite loci

Fig. 3 The pattern of changes in values of Φ-statistics parameters with the assumed number of groups (K), revealed using samova. ΦSC measures the proportion of the variance among local populations within groups. ΦST measures the proportion of the variance among local populations within the total population. ΦCT denotes the fraction of the total variance that is explained by the grouping.

Population genetic structure inferred from mtDNA The results of the spatial analysis of molecular variance (samova) indicated significant population genetic structure for each assumed number of groups, from 2 to 20 (P < 0.00001 in each case). In a graph of changes in values of Φ-statistics parameters with K, the highest increase in ΦCT value occurred between K = 9 and K = 10, and all parameters of Φ-statistics stabilized beginning from K = 10 (Fig. 3).

Thus, we identified K = 10 as the most probable number of groups. In the subdivision into 10 groups inferred by the samova procedure, some groups were not geographically homogenous, as some single localities were placed within the area of other groups. However, after modifying this subdivision so as to receive geographically homogenous subpopulations, we still received a highly significant subdivision (ΦCT = 0.37, P < 0.00001; Table 2 and Fig. 4) that was assumed to be the most probable population genetic structure. The subpopulations identified in this way

Out of 10 geneland runs with varying K, seven gave a modal number of 3 subpopulations, and three gave a modal number of 4 subpopulations. We then preformed 20 runs with K = 3 and compared the distribution of subpopulations inferred in subsequent runs. The results of these runs showed globally good consistency. In five independent runs, individuals were assigned in the same way: two subpopulations (A and B) were modal subpopulations for 298 and 245 individuals, respectively (Fig. 5a). The third subpopulation (C) was modal for two individuals only and the majority of the area of this subpopulation corresponded to the part of the study area with no sampled individuals (Appendix IIIa). In other five runs, subpopulation C was modal for none of the individuals. In the remaining 10 runs it was modal for 1– 21 individuals (depending on the run) from nine locations that did not constitute a geographically homogenous group (Fig. 5b). This suggests that subpopulation C is a ‘ghost population’, as defined by Guillot et al. (2005a), rather than a real subpopulation.

The subpopulations inferred in the five independent runs were separated by narrow border zones, indicating steep genetic discontinuities (Fig. 5a and Appendix IIa). Other runs differed from this modal result in the assignment of individuals situated near the border zones between subpopulations (Fig. 5b), and — as a result — the borders between subpopulations were less straight (Appendix IIIb). The runs that inferred the most complicated pattern of the distribution of subpopulations, with subpopulation C that was not spatially homogenous, had the highest mean posterior probability. However, repeatability of the © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4541

Fig. 4 (a) Subpopulations of wolves in Eastern Europe delimited based on frequencies of mtDNA haplotypes, against the background of the continuous wolf range (darker grey area). Each symbol represents one local population and is situated in its centroid. The size of a symbol reflects sample size. Different symbols represent local populations assigned to different groups by the samova procedure. Arrows indicate local populations for which the original samova assignment has been changed to receive geographically homogenous groups. (b) Frequency of haplotypes belonging to different 2-step clades of the minimum-spanning network (see Fig. 2) in each subpopulation.

Table 3 Climate, potential vegetation and mean wolf diet composition in the areas of wolf subpopulations S1–S10. Climatic variables (temperature and rainfall) were calculated as means from the studied localities within respective subpopulations. Similarly, mean wolf diet composition in a subpopulation was calculated as a mean from localities where wolf diet composition was known (see the main text for references). Only common ungulate prey species were considered: moose (A.a.), red deer (C.e.), roe deer (C.c.), and wild boar (S.s.). Symbols of potential vegetation are as follows: BF, boreal forest; TF, temperate deciduous and mixed forest; FS, forest-steppe; ST, steppe; MF, temperate mountain forest; MW, Mediterranean woodlands and shrubs

Subpopulation

Mean annual temperature (°C)

Mean temperature of January (°C)

Mean temperature of July (°C)

Mean annual rainfall (mm)

Potential vegetation

S1 S2 S3 S4 S5 S6 S7 S8 S9 S10

5.4 4.4 6.7 1.8 4.7 7.7 9.8 9.2 8.1 13.0

−6.8 −8.2 −5.9 −14.2 −13.6 −6.5 −1.5 −1.7 −3.5 1.9

17.4 16.5 18.3 18.1 22.5 21.0 21.6 19.3 18.1 21.3

610 604 625 563 363 472 450 663 687 517

BF, TF BF, TF TF, FS BF, TF ST ST ST MF MF MW

© 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

Mean wolf diet composition (%) A.a.

C.e.

C.c.

S.s.

39 26 3

21 0 41

24 14 43

16 60 13

0 0

47 46

41 50

12 4

4542 M . P I L O T E T A L .

Fig. 5 Subpopulations of wolves in Eastern Europe, delimited based on allele frequencies of microsatellite loci, against the background of the continuous wolf range (darker grey area). (a) Modal assignment of individuals, i.e. assignment that was inferred in five independent runs using the software geneland. (b) Recapitulation of the assignment results from all 20 runs. The maps are based on 545 individuals that were successfully genotyped. Each point represents a wolf or several wolves from the same location, and different symbols represent subpopulations.

result among runs may be a better indicator of the true assignment to subpopulations than the mean posterior probability (see Coulon et al. 2006). The general location of two main subpopulations was consistent among all 20 runs, and 453 (83%) individuals were assigned in the same way in all runs (259 individuals to subpopulation A and 194 individuals to subpopulation B). Out of the remaining 92 individuals, 67 were assigned either to subpopulation A or B, 23 either to subpopulation A or C, and two individuals were assigned to subpopulations A, B or C, depending on the run (Fig. 5b).

Because of the high consistency among the 20 performed runs, we decided that performing more runs with K = 3 is unnecessary. However, we performed three additional runs with K = 4. These runs indicated two subpopulations corresponding to subpopulations A and B revealed from runs with K = 3. The third subpopulation, corresponding to subpopulation C, was modal for none of the individuals (one run) or for 13 individuals from four locations that were not spatially grouped (two runs). Fourth subpopulation overlapped spatially with the third one and was not modal for any individual. It confirmed that K = 3 was the proper number of subpopulations.

For the genetic structure inferred from each of the 20 runs with K = 3, we performed the analysis of molecular variance (amova). amova confirmed the significance of the structure inferred by geneland (P < 0.00001 in each case), although genetic differentiation among subpopulations was low (FST ranged from 0.014 to 0.024 depending on the run).

Isolation by distance: mtDNA and microsatellites Spatial differentiation in haplotype frequencies, measured as linearized pairwise ΦST between 59 localities, was significantly higher than expected for a panmictic population and followed isolation by distance (Mantel test, r = 0.149, P = 0.007). Spatial differentiation in frequencies of microsatellite alleles, measured as Nei’s standard genetic distance between 16 regions, also followed isolation by distance (r = 0.241, P = 0.036). However, when genetic distance was measured as linearized pairwise FST, its dependence on geographical distance was insignificant (r = 0.186, P = 0.078).

Dependence of genetic diversification on environmental variables: mtDNA A test on the influence of latitude and longitude (treated as covariables) on genetic differentiation among local wolf populations showed that pairwise ΦST measures between localities strongly depended on latitude (P = 0.0001), but not on longitude (P = 0.48). As in Europe many environmental factors change along the north–south axis, we analysed an array of environmental variables to evaluate whether they may explain genetic differentiation among local wolf populations over and above the influence of geographical distance. In marginal tests, two sets of environmental variables were significantly correlated with genetic distance: vegetation types and temperature (Table 4a). When geographical coordinates were taken into account in a form of © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4543 Table 4 Effects of environmental factors on genetic differentiation of Eastern European wolves based on mtDNA and microsatellite analysis. Marginal and conditional tests of individual variable sets as well as sequential tests of the forward selection procedure are reported (see Materials and methods for the description of the tests). P indicates probability values and ‘%var’ the percentage of the genetic variation explained by the particular variable. In the case of sequential tests, ‘%var’ indicates the percentage of the genetic variation explained by a cumulative effect of variables. The top-down sequence of variables corresponds to the sequence that was indicated by the forward selection procedure Marginal tests Variable set

P

Conditional tests %var

P

Sequential tests P

%var

0.002 0.112 0.995 0.036

22.5 29.4 100.0 —

0.023 0.134 0.026 0.208 0.013

47.5 60.9 63.5 84.3 85.5

53.2 — 0.6 4.8

0.091 0.126 0.402 0.003

52.2 67.5 68.1 75.9

26.9 19.8 24.1 6.2 —

0.006 0.034 1.000 — —

82.8 98.4 100.0 — —

— 47.1 10.3 11.7

0.004 0.042 0.492 0.924

36.4 83.5 93.4 93.7

26.0 19.5 25.2 5.8 —

0.010 0.021 0.866 — —

82.1 98.8 100.0 — —

%var

(a) Tests for mtDNA and genetic distances measured as pairwise ΦST All local populations, without considering the prey composition Vegetation 0.002 22.5 0.038 24.0 Temperature 0.029 11.7 0.379 4.6 Coordinates < 0.001 17.1 — — Rainfall 0.196 3.1 0.159 2.8 17 local populations, for which the prey composition was known Temperature 0.023 47.5 0.553 9.5 Rainfall 0.011 28.8 0.138 9.8 Prey 0.057 20.1 0.417 2.9 Vegetation 0.011 46.9 0.018 33.8 Coordinates 0.013 42.6 — — (b) Tests for microsatellite loci and Nei’s standard genetic distance All regions, without considering the prey composition Vegetation 0.091 52.2 0.005 Coordinates 0.002 43.1 — Temperature 0.109 33.0 1.000 Rainfall 0.058 17.7 0.323 Eight regions, for which the prey composition was known Temperature 0.006 82.8 0.481 Prey 0.099 53.7 0.393 Vegetation 0.179 51.8 0.249 Rainfall 0.165 25.4 0.472 Coordinates 0.058 57.2 — (c) Tests for microsatellite loci and genetic distances measured as pairwise FST All regions, without considering the prey composition Coordinates 0.004 36.4 — Vegetation 0.247 43.2 0.042 Temperature 0.089 33.1 0.676 Rainfall 0.023 20.8 0.099 Eight regions, for which the prey composition was known Temperature 0.010 82.1 0.518 Prey 0.158 49.6 0.402 Vegetation 0.326 18.3 0.228 Rainfall 0.148 27.9 0.495 Coordinates 0.062 56.8 —

covariables in the multiple regression analysis, vegetation types were still significantly correlated with genetic distance, but temperature was not (Table 4a). However, when a combined influence of vegetation types and temperature was taken into account, these variables were significantly correlated with genetic distance (P = 0.01) and explained 43% of the genetic variation over and above the influence of geographical distance. Additionally, the forward selection procedure (Anderson 2003) that classifies variables according to the proportion of explained variation, fitted vegetation © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

types and temperature prior to geographical distance in the multiple regression model (Table 4a).

For 17 localities situated in the northwestern part of the study area (within subpopulations S1–S3, S8 and S9, see Fig. 4), we were able to analyse the influence of yet another factor: wolf diet composition. For these 17 localities, four sets of environmental variables were correlated with genetic distance: geographical coordinates, temperature, rainfall, and vegetation types (Table 4a). Wolf diet composition was not significantly correlated with genetic distance. However,

4544 M . P I L O T E T A L . when only the frequency of one prey species, red deer, in the wolf diet was taken into account, the correlation was marginally significant (P = 0.057) and 20% of genetic variation was explained by this factor. When geographical coordinates were taken into account in a form of covariables in the multiple regression analysis, only vegetation types were significantly correlated with genetic distance (Table 4a).

However, we found that a combination of temperature, rainfall, vegetation types and a frequency of red deer in the wolf diet explained 54% of the genetic variation among these localities (P = 0.008), when considered together in the conditional test where geographical coordinates were included as covariables. Moreover, all of the above variables were classified as more important than geographical distance by the forward selection procedure (Table 4a).

For eight regions from the northwestern part of the study area (see Fig. 1), we also analysed the correlation of genetic distance with wolf’s diet composition. For these eight regions, only temperature was correlated with genetic distance in a marginal test (P = 0.006 for Nei’s genetic distance and P = 0.01 for FST). Wolf diet composition was not significantly correlated with genetic distance in a marginal test, but the forward selection procedure fitted it just after temperature in the multiple regression model both in the case of Nei’s genetic distance (Table 4b) and pairwise FST (Table 4c). In this case, both temperature and wolf diet composition were significantly correlated with genetic distance and explained together 98% and 99% of genetic variation, respectively, while other variables were insignificant.

Discussion Dependence of genetic diversification on environmental variables: microsatellite loci Similarly as in the case of mtDNA, pairwise genetic distance between regions calculated for microsatellite loci depended on latitude (P = 0.002 for DS distance and P = 0.003 for FST), but not on longitude (P = 0.236 and 0.156, respectively). In marginal tests, only rainfall was significantly correlated with genetic distance measured as pairwise FST (P = 0.023). It was also marginally correlated with Nei’s genetic distance (P = 0.058), while other variables were not correlated with it (Table 4b, c). When geographical distance was taken into account in a form of a covariable in the multiple regression analysis, vegetation types were correlated with Nei’s genetic distance and with pairwise FST (P = 0.005 and 0.042, respectively), but other variables were not correlated with any genetic distance measure (Table 4b, c). An influence of vegetation types explained 53% of the genetic variation measured as Nei’s genetic distance and 47% of the genetic variation measured as pairwise FST, over and above the influence of genetic distance. As opposite to genetic variation measured for mtDNA, if vegetation types and temperature were analysed together, they were not correlated with genetic distance.

In the case of Nei’s genetic distance, the forward selection procedure fitted vegetation types before geographical distance and other variables in the multiple regression model (Table 4b, c). In this test, neither geographical distance nor temperature was significant, when the influence of vegetation types was taken into account. Both the results of the conditional test and the forward selection procedure indicated that the influence of vegetation types explain substantial proportion of genetic variation over and above the influence of genetic distance. In the case of pairwise FST, geographical distance was fitted before vegetation types (Table 4b, c). However, in both cases the influence of vegetation types explained more genetic variation than the influence of geographical distance.

We found that most local wolf populations in Eastern Europe had more than one mtDNA haplotype and most haplotypes were widely distributed. This result is contradictory to previous studies, based on less extensive sampling, which suggested that the majority of extant populations in Eurasia have unique haplotypes (Wayne et al. 1992; Vilr et al. 1999).

This discrepancy can be explained by the fact that the number of haplotypes found in a locality depends on the number of analysed samples. It may be also important that samples from strongly fragmented western populations prevailed within the data analysed by Wayne et al. (1992) and Vilr et al. (1999), whereas those from the continuous species range in Eastern Europe were limited. Indeed, Randi et al. (2000) showed that wolves from southeastern and northeastern Europe are more differentiated (9 haplotypes were found among 29 individuals from Bulgaria and 3 haplotypes among five individuals from Finland), which is in agreement with our results. In our study, the number of detected haplotypes is close to the expected total number of haplotypes, estimated from the rarefaction curve. This result indicates that our sample is representative for the studied population, as most haplotypes were detected.

The distribution of haplotypes may result from both current and historical processes. Thus, an insight into population history is needed for the proper inference on the contemporary factors shaping population genetic structure.

According to fossil records, grey wolves and coyotes diverged about 2 million years ago (Nowak 2003). Given this divergence date, a net sequence divergence between these species of 33.4% and a mean sequence divergence in Eastern European grey wolves of 3.2%, a coalescence of Eastern European wolf haplotypes can be roughly estimated at about 200 000 years ago. Such coalescence time is consistent with that estimated for all known wolf lineages at 290 000 years ago (Vilr et al. 1999), taking into account that European wolves represent most of these lineages (Vilr et al. 1999). As this coalescence time substantially pre-dates © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4545 the Last Glacial Maximum (17 000–21 000 years ago), it implies that most haplotypes diverged prior to the last glaciation. Thus, the current genetic pattern may potentially result from processes that occurred during and after the Pleistocene glaciations. One of the most important historical processes that strongly influenced patterns of genetic differentiation of many species was isolation in different glacial refugia (Hewitt 1996, 2000; Taberlet et al. 1998).

However, the NCA provided no evidence for any past fragmentation events within the Eastern European wolf population, and indicated past events of range expansion as the main historical factor that influence current distribution of haplotypes. It was impossible to indicate the direction of this range expansion, as the results for different nested clades indicated different directions and some of them were opposite one another. It suggests nondirectional spread of haplotypes throughout the continent. Thus, it can be concluded that the differentiation of the wolf population in Eastern Europe is a result of past admixture (i.e. range expansions of different haplotypes in different directions) and present restricted gene flow. As current restrictions in gene flow were identified as a factor shaping the distribution of haplotypes for the clade 4-1 that included haplotypes of most (87%) Eastern European wolves, this indicates the ability of mtDNA to reveal contemporary patterns in this case.

The restrictions in gene flow are reflected in the distinct population structure revealed from the frequencies of mtDNA haplotypes by the samova analysis. This structure has two unusual features. First, subpopulation S8, connected with the mountains, is presented as having the noncontinuous range (see Fig. 4). This is due to the lack of samples from the Romanian Carpathian Mountains located between the two parts of this subpopulation. Second, subpopulation S2 is located within the range of subpopulation S1, which is an unusual pattern. The general pattern of subdivision into subpopulations can be explained by the influence of ecological factors, but significant ecological differences between areas of subpopulations S1 and S2 are not obvious.

However, there may be other causes of the existence of a distinct subpopulation S2 within subpopulation S1. For example, a recent appearance of a haplotype that is new for a given area (e.g. as a result of immigration) may result in a temporary appearance of an ‘island’ of its high concentration that may be recognized as a separate subpopulation. The fact that subpopulation S2 was not recognized as a distinct subpopulation based on microsatellite analysis is consistent with this explanation. If subpopulation S2 was not distinguished within subpopulation S1, the population structure was still highly significant (ΦCT = 0.34, P < 0.00001; see Table 2).

Such distinct population genetic structure points to restrictions in gene flow, despite long dispersal ranges of wolves. Extensive and overlapping ranges of the main clades of the minimum-spanning network indicate the © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

absence of geographical features in the landscape that would constitute efficient barriers for wolf dispersal. Thus, the restrictions in gene flow must be caused by factors other than geographical barriers. High distinctiveness of subpopulations may be to some extent an artefact of the analysis method. Changes in haplotype frequencies may be more clinal, not as discrete as the samova analysis suggests.

However, the results show explicitly that the population is not panmictic. Moreover, it is unlikely that the population differentiation results from the isolation by distance alone, as genetic differentiation between wolves from different localities was correlated with latitude, but not with longitude.

This reflects the influence of environmental factors, as in Europe the north–south axis is a direction of the most prominent changes in many environmental variables, such as vegetation types, temperature and depth of snow cover.

Variation in morphology, size and colour of Old World wolves also is the greatest along the north–south axis (Bibikov 1985; Nowak 2003). The dependence of genetic differentiation on clinally changing environmental variables is in agreement with the theoretical model of Doebeli & Dieckmann (2003), showing that processes of evolutionary diversification may lead to sharp geographical differentiation along environmental gradients. As the environmental gradients that we consider have existed for an extended period of time, they are likely to be reflected in mtDNA despite its relatively low variability.

Importantly, the same result — correlation of genetic distance between wolf populations with latitude and with environmental variables — was obtained based on microsatellite loci analysis. It indicates that this pattern is independent on the type of markers and on the way of delimiting sample groups. However, the population genetic structure in microsatellite loci was less pronounced and fewer subpopulations were delimited than in the case of mtDNA analysis. Differences in the degree of structuring in mitochondrial and nuclear DNA were also observed in North American grey wolf populations, which were grouped into clusters when restriction fragment length polymorphism (RFLP) profiles of mtDNA were taken into account, but did not show similar groupings based on microsatellite analysis (Geffen et al. 2004). The contrasting mitochondrial and nuclear DNA patterns were also reported for other animals, e.g. green turtle, Chelonia mydas (Karl et al. 1992), Oregon slender salamander, Batrachoseps wrighti (Miller et al. 2005), brown bear, Ursus arctos (Waits et al. 2000), and wolverine, Gulo gulo (Chappell et al. 2004). They are usually explained by differences in male and female dispersal (Karl et al. 1992; Avise 1994; Moritz 1994a): male-biased gene flow implies low introgression of mtDNA haplotypes from neighbouring populations, and therefore greater structuring in mtDNA as compared with nuclear markers. It is also a possible explanation for the results obtained in our study, as some data bring evidence that long-distance dispersal may be

4546 M . P I L O T E T A L . male-biased in wolves (Wabakken et al. 2001; Flagstad et al.

2003; Jedrzejewski et al. 2005). Differences in mitochondrial and nuclear DNA patterns may also result from the fact that the effective population size of mtDNA is four times smaller than that of nuclear DNA, and therefore mtDNA variability is more sensitive to random drift than variability of nuclear DNA (Avise et al. 1984). Thus, genetic diversification resulting from a limited, but not totally prevented gene flow may be more pronounced in frequencies of mtDNA haplotypes than in the variability of microsatellite loci.

It is also possible that small number of samples from southern and eastern part of the study area prevented detection of distinct subpopulations there (Fig. 5). On the contrary, the northwestern part of the study area was extensively sampled, and thus the genetic discontinuity between subpopulations A and B was highly supported by the data. It is important that the location of this border zone is similar to the location of the border between two biggest subpopulations, S1 and S3, delimited based on mtDNA analysis. In the area of this genetic discontinuity, there are no obvious barriers to gene flow, and the fact that it is situated in the horizontal axis suggests that it is caused by environmental variables changing along the south-north gradient. The influence of gradually changing environmental variables would also explain why the location of the border zones inferred from mtDNA and microsatellite analyses is not exactly the same: in the absence of an absolute barrier some level of admixture must occur, and the real discontinuity is probably not as sharp as the results of both analyses indicate.

The dependence of genetic differentiation on climatic and ecological variables suggests a link between the ecological and genetic spatio-temporal processes, which was previously suggested for another large, mobile carnivore, Canadian lynx (Rueness et al. 2003b; Stenseth et al. 2004a, b). It has been shown that both demographic and population genetic patterns may be influenced by the interaction between a carnivore and its prey (wolf: Carmichael et al. 2001; lynx: Stenseth et al. 2004b). A possible mechanism that would explain how the composition of ungulate community may influence wolf dispersal (and therefore patterns of gene flow) is differential prey selection. In multiprey systems, certain prey species may be preferred to others (Carbyn 1983; Potvin et al. 1988; Dale et al. 1994; Kunkel et al. 2004).

In northeastern Europe, three cervids — red deer, moose, and roe deer — dominate wolf diet (Okarma 1995). A positive selectivity for red deer and strong functional response to an increase in red deer densities have been observed (Okarma 1995; Jedrzejewski et al. 2000), and the abundance of this species in wolf diet was correlated with genetic structure in our tests. This suggests that differences in hunting strategy for prey of different sizes may lead to local prey specialization, and thus to genetic differentiation in wolves.

A connection between population genetic structure and prey specialization was previously suggested for wolves in north-western Canada (Carmichael et al. 2001). According to that study, each wolf pack is connected with a particular herd of caribou, and thus migratory routes of caribou are reflected in population genetic structure of wolves. Another case of a connection between genetic differentiation and prey specialization is known for the arctic fox. Population genetic structure of this species is consistent with the subdivision into two ecotypes: ‘lemming’ ecotype that feeds mainly on lemmings and ‘coastal’ ecotype that feeds mainly on eggs, birds and carrion from the coastal ecosystem (Dalén et al. 2005). Similarly, significant genetic differentiation was found between two groups of killer whales Orcinus orca that occurred in the same area, but specialized in foraging on fish or on marine mammals (Hoelzel et al. 1998).

As the composition of the ungulate community strongly depends on the habitat type, the same mechanism may lead to the dependence of genetic differentiation of wolves on these two factors. As suggested by Geffen et al. (2004), wolf dispersal may be habitat-biased. Young individuals often stay in their natal packs for a long time (Mech & Boitani 2003), learning to prey on animals characteristic for the habitat where they live. The fact that young wolves observed or assisted in hunting of particular species may result in their subsequent willingness to choose these species for prey and to choose habitats where these particular species are abundant, which will increase the wolves’ chances for survival (Gese & Mech 1991). Indeed, some authors suggest that experience and learning help wolves to successfully attack their prey and to avoid being harmed (Mech & Peterson 2003; Peterson & Ciucci 2003). Similarly, studies on Canadian lynx suggested that individuals familiar with prey conditions in a certain area would stay within an area of similar conditions when dispersing, because there is a cost to the process of learning how to use a new habitat, which reduces the probability of reproductive success (O’Donoghue et al. 2001). Differences in reproduction and mortality in areas with known vs. unknown prey species may lead to the connection between habitat types, diet composition and dispersal patterns.

The hypothesis that wolf dispersal may be habitat-biased is additionally supported by the fact that one of the most common processes of pack formation known as ‘budding’, which results in the establishment of a new pack close to the parental pack (Mech & Boitani 2003; Jedrzejewski et al.

2004), promotes a selection of similar habitats by related individuals. Natal-habitat-biased dispersal was also suggested for coyotes as the most probable explanation of population genetic structure that corresponded to habitatspecific breaks (Sacks et al. 2004). Thus, natal-habitat-biased dispersal together with prey specialization may induce an association of habitat types and diet composition with population genetic structure. © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4547 Additionally, as studies on Canadian lynx showed, snow conditions are the climatic factor that directly influences predator–prey interactions and thus may influence population genetic patterns through nonrandom dispersal (Stenseth et al. 2004b). Snow conditions may play the same role in genetic differentiation of other large predators found in habitats with long, snowy winter seasons (Stenseth et al.

2004b). Indeed, our study showed that besides habitat types and prey composition, wolf population genetic structure in Eastern Europe is influenced by temperature, which is one of the main factors regulating snow conditions. Although none of these environmental and ecological factors constitute absolute barrier to gene flow, their combined influence may lead to population differentiation.

Based on an example of a highly mobile and widely distributed carnivore species, we showed that a distinct genetic structure may occur in regions where environmental changes are gradual and do not prevent gene flow. Our results support the growing body of literature that demonstrates the influence of ecological factors such as habitat types and diet composition on geographical patterns of genetic variation within the species. It indicates the importance of further studies aimed at understanding the direct mechanism that links population ecology and population genetic structure.

Acknowledgements We are grateful to Y. Iliopoulos from the Greek Society for the Protection and Management of Wildlife ARCTUROS, Z. Andersone, A. N. Bunevich, O. Buzbas, I. Dikiy, V. Dumenko, J. Goszczynski, T. KaKamarz, A. Kloch, M. Nedzynska, S. Nowak, A. Olczyk, M. Shkvirya, W. )mietana, E. Tsingarskaya, V. Tokarskiy, M. Wojtulewicz and S. Zhyla, who helped in collecting the samples. We thank G. Guillot for the consultation on the use of the geneland software. We are grateful to J. Goszczynski, O. Liberg, B. N. Sacks, R. Van Den Bussche, R. K. Wayne, and two anonymous reviewers for helpful comments on earlier drafts of the manuscript. This project was funded by the former Polish State Committee for Scientific Research (Grant no. 6P04F 09421), the budget of Mammal Research Institute, European Nature Heritage Fund — Euronatur, and the UK Wolf Conservation Trust.

References Anderson MJ (2003) DISTLM Forward: a FORTRAN Computer Program to Calculate a Distance-Based Multivariate Analysis for a Linear Model Using Forward Selection. Department of Statistics, University of Auckland, New Zealand. Available at: http:// www.stat.auckland.ac.nz/∼mja.

Anderson MJ (2004) DISTLM Version 5: a FORTRAN Computer Program to Calculate a Distance-Based Multivariate Analysis for a Linear Model. Department of Statistics, University of Auckland, New Zealand. Available at: http://www.stat.auckland.ac.nz/∼mja.

Anderson MJ, Legendre P (1999) An empirical comparison of permutation methods for tests of partial regression coefficients in a linear model. Journal of Statistical Computation and Simulation, 62, 271–303. © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

Andersone Z (1998) Summer nutrition of wolf (Canis lupus) in the Slitere Nature Reserve, Latvia. Proceedings of the Latvian Academy of Sciences, Section B, 52, 79–80.

Avise JC (1994) Molecular Markers, Natural History and Evolution, pp. 1–510. Chapman & Hall, New York.

Avise JC, Neigel JE, Arnold J (1984) Demographic influences of mitochondrial DNA lineage survivorship in animal populations.

Journal of Molecular Evolution, 20, 99–105.

Bibikov DI, ed. (1985) The Wolf. History, Systematics, Morphology, Ecology. Izdatelstvo Nauka, Moskva. [in Russian].

Boitani L (2003) Wolf conservation and recovery. In: Wolves: Behavior, Ecology, and Conservation (eds Mech LD, Boitani L), pp. 317–340. The University of Chicago Press, Chicago.

Breen M, Jouquand S, Renier C et al. (2001) Chromosome-specific single-locus FISH probes allow anchorage of an 1800-marker integrated radiation-hybrid/linkage map of the domestic dog genome to all chromosomes. Genome Research, 11, 1784–1795.

Carbyn LN (1983) Wolf predation on elk in Riding Mountain National Park, Manitoba. Journal of Wildlife Management, 47, 963–976.

Carmichael LE, Nagy JA, Larter NC, Strobeck C (2001) Prey specialization may influence patterns of gene flow in wolves of the Canadian Northwest. Molecular Ecology, 10, 2787–2798.

Chappell DE, Van Den Bussche RA, Krizan J, Patterson B (2004) Contrasting levels of genetic differentiation among populations of wolverines (Gulo gulo) from northern Canada revealed by nuclear and mitochondrial loci. Conservation Genetics, 5, 759–767.

Clement M, Posada D, Crandall KA (2000) tcs: a computer program to estimate gene genealogies. Molecular Ecology, 9, 1657–1660.

Coulon A, Guillot G, Cosson J-F et al. (2006) Genetic structure is influenced by landscape features: empirical evidence from a roe deer population. Molecular Ecology, 15, 1669–1679.

Crandall KA, Bininda-Emonds ORP, Mace MM, Wayne RK (2000) Considering evolutionary processes in conservation biology.

Trends in Ecology & Evolution, 15, 290–295.

Dale BW, Adams LG, Bowyer RT (1994) Functional response of wolves preying on barren-ground caribou in a multiple-prey ecosystem. Journal of Animal Ecology, 63, 644–652.

Dalén L, Fuglei E, Hersteinsson P et al. (2005) Population history and genetic structure of a circumpolar species: the arctic fox.

Biological Journal of the Linnean Society, 84, 79–89.

Doebeli M, Dieckmann U (2003) Speciation along environmental gradients. Nature, 421, 259–264.

Dupanloup I, Schneider S, Excoffier L (2002) A simulated annealing approach to define the genetic structure of populations.

Molecular Ecology, 11, 2571–2581.

Excoffier L, Smouse PE, Quattro JM (1992) Analysis of molecular variance inferred from metric distances among DNA haplotypes: application to human mitochondrial DNA restriction data.

Genetics, 131, 479–491.

Filonov KP (1989) Ungulates and Large Predators in Wildlife Reserves.

Izdatelstvo Nauka, Moskva. [in Russian with English abstract].

Flagstad O, Walker C, Vila C et al. (2003) Two centuries of the Scandinavian wolf population: patterns of genetic variability and migration during an era of dramatic decline. Molecular Ecology, 12, 869–880.

Forbes SH, Boyd DK (1997) Genetic structure and migration in native and reintroduced Rocky Mountain wolf populations.

Conservation Biology, 11, 1226–1234.

Francisco LV, Langston AA, Mellersh CS, Neal CL, Ostrander EA (1996) A class of highly polymorphic tetranucleotide repeats for canine genetic mapping. Mammalian Genome, 7, 359–362.

4548 M . P I L O T E T A L .

Fritts SH (1983) Record dispersal of a wolf from Minnesota. Journal of Mammalogy, 64, 166–167.

Geffen E, Anderson MJ, Wayne RK (2004) Climate and habitat barriers to dispersal in the highly mobile grey wolf. Molecular Ecology, 13, 2481–2490.

Gese EM, Mech LD (1991) Dispersal of wolves (Canis lupus) in northeastern Minnesota. Canadian Journal of Zoology, 69, 2946–2955.

Guillot G, Estoup A, Mortier F, Cosson JF (2005a) A spatial statistical model for landscape genetics. Genetics, 170, 1261–1280.

Guillot G, Mortier F, Estoup A (2005b) geneland: a computer package for landscape genetics. Molecular Ecology Notes, 5, 708–711.

Gula R (2004) Influence of snow cover on wolf Canis lupus predation patterns in Bieszczady Mountains, Poland. Wildlife Biology, 10, 17–23.

Guo SW, Thompson EA (1992) Performing the exact test of Hardy– Weinberg proportion for multiple alleles. Biometrics, 48, 361–372.

Hardy OJ, Vekemans X (2002) spagedi: a versatile computer program to analyse spatial genetic structure at the individual or population levels. Molecular Ecology Notes, 2, 618–620.

Hartl DL, Clark GC (1997) Principles of Population Genetics, 3rd edn.

Sinauer Associates, Sunderland, Massachusetts.

Hewitt GM (1996) Some genetic consequences of ice ages, and their role in divergence and speciation. Biological Journal of the Linnean Society, 58, 247–276.

Hewitt GM (2000) The genetic legacy of the Quaternary ice ages.

Nature, 405, 907–913.

Hoelzel AR, Dahlheim M, Stearn SJ (1998) Low genetic variation among killer whales (Orcinus orca) in the eastern North Pacific and genetic differentiation between foraging specialists. Journal of Heredity, 89, 121–128.

Holmes NG, Dickens HF, Parker HL (1995) Eighteen canine microsatellites. Animal Genetics, 26, 132–133.

Huelsenbeck JP, Ronquist F (2001) mrbayes: Bayesian inference of phylogeny. Bioinformatics, 17, 754–755.

Ihaka R, Gentleman R (1996) R: a language for data analysis and graphics. Journal of Computational and Graphical Statistics, 5, 299– 314.

Jedrzejewska B, Jedrzejewski W (1998) Predation in Vertebrate Communities. The Bialowie*a Primeval Forest as a Case Study.

Springer Verlag, Berlin.

Jedrzejewski W, Jedrzejewska B, Okarma H et al. (2000) Prey selection and predation by wolves in BiaKowie%a Primeval Forest, Poland. Journal of Mammalogy, 81, 197–212.

Jedrzejewski W, Nowak S, Schmidt K, Jedrzejewska B (2002) The wolf and the lynx in Poland — results of a census conducted in 2001. Kosmos, 51, 491–499. [In Polish with English summary].

Jedrzejewski W, Schmidt K, Jedrzejewska B et al. (2004) The process of wolf pack splitting in BiaKowie%a Primeval Forest, Poland. Acta Theriologica, 49, 275–280.

Jedrzejewski W, Branicki W, Veit C et al. (2005) Genetic diversity and relatedness within packs in an intensely hunted population of wolves Canis lupus. Acta Theriologica, 50, 3–22.

Karl SA, Bowen BW, Avise JC (1992) Global population genetic structure and male-mediated gene flow in the green turtle (Chelonia mydas): RFLP analyses of anonymous nuclear loci.

Genetics, 131, 163–173.

Kerechun SF (1979) Impact of predators on ungulate numbers in the Carpathian zone. In: Ecological Fundamentals of Protection and Rational Utilisation of Predatory Mammals (ed. Sokolov VE), pp. 43–44. Izdatelstvo Nauka, Moskva. [in Russian].

Kloch A (2003) Distribution, abundance and food preferences of wolves (Canis lupus) from Lasy Napiwodzko-Ramuckie and Puszcza Piska.

MSc Thesis, Mammal Research Institute PAS and University of Warsaw [in Polish].

Kohn MH, York EC, Kamradt DA et al. (1999) Estimating population size by genotyping faeces. Proceedings of the Royal Society of London. Series B, Biological Sciences, 266, 657–663.

Koniuch J (2002) Population dynamics, structure and food composition of wolf Canis lupus in Puszcza Knyszy˜ska. MSc Thesis, University of BiaKystok and Mammal Research Institute PAS [in Polish].

Kumar S, Tamura K, Nei M (2004) mega3: Integrated software for Molecular Evolutionary Genetics Analysis and sequence alignment. Briefings in Bioinformatics, 5, 150–163.

Kunkel KE, Pletscher DH, Boyd DK, Ream RR, Fairchild MW (2004) Factors correlated with foraging behavior of wolves in and near Glacier National Park, Montana. Journal of Wildlife Management, 68, 167–178.

Legendre P, Anderson MJ (1999) Distance-based redundancy analysis: testing multispecies responses in multifactorial ecological experiments. Ecological Monographs, 69, 1–24.

Leonard JA, Vilà C, Wayne RK (2005) Legacy lost: genetic variability and population size of extirpated US grey wolves (Canis lupus). Molecular Ecology, 14, 9–17.

Lucchini V, Galov A, Randi E (2004) Evidence of genetic distinction and long-term population decline in wolves (Canis lupus) in the Italian Apennines. Molecular Ecology, 13, 523–536.

McArdle BH, Anderson MJ (2001) Fitting multivariate models to community data: a comment on distance-based redundancy analysis. Ecology, 82, 290–297.

McRae BH, Beier P, Dewald LE, Huynh LY, Keim P (2005) Habitat barriers limit gene flow and illuminate historical events in a wide-ranging carnivore, the American puma. Molecular Ecology, 14, 1965–1977.

Mech LD, Boitani L (2003) Wolf social ecology. In: Wolves: Behavior, Ecology, and Conservation (eds Mech LD, Boitani L), pp. 1–34.

University of Chicago Press, Chicago, Illinois.

Mech LD, Fritts SH, Wagner D (1995) Minnesota wolf dispersal to Wisconsin and Michigan. American Midland Naturalist, 133, 368– 370.

Mech LD, Peterson RO (2003) Wolf — prey relations. In: Wolves: Behavior, Ecology, and Conservation (eds Mech LD, Boitani L), pp. 131–157. University of Chicago Press, Chicago, Illinois.

Merrill SB, Mech LD (2000) Details of extensive movements by Minnesota wolves. American Midland Naturalist, 144, 428–433.

Miller MP, Haig SM, Wagner RS (2005) Conflicting patterns of genetic structure produced by nuclear and mitochondrial markers in the Oregon slender salamander (Batrachoseps wrighti): Implications for conservation efforts and species management.

Conservation Genetics, 6, 275–287.

Moritz C (1994a) Applications of mitochondrial DNA analysis in conservation: a critical review. Molecular Ecology, 3, 401– 411.

Moritz C (1994b) Defining ‘Evolutionary Significant Units’ for conservation. Trends in Ecology & Evolution, 9, 373–375.

Nedzynska M (2003) Wolf (Canis lupus Linnaeus, 1758) in Roztocze and Kotlina Sandomierska. Abundance, distribution and food preferences. MSc Thesis, Mammal Research Institute PAS and A. Mickiewicz University of Poznan. [in Polish].

Nei M, Kumar S (2000) Molecular Evolution and Phylogenetics.

Oxford University Press, Oxford.

Nowak RM (2003) Wolf evolution and taxonomy. In: Wolves: Behavior, Ecology, and Conservation (eds Mech LD, Boitani L), pp. 239–258. The University of Chicago Press, Chicago, Illinois. © 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4549 Nowak S, MysKajek RW, Jedrzejewska B (2005) Patterns of wolf Canis lupus predation on wild and domestic ungulates in the Western Carpathian Mountains (S. Poland). Acta Theriologica, 50, 263–276.

Nylander JAA (2004) MRMODELTEST Version 2. Program Distributed by the Author. Evolutionary Biology Centre, Uppsala University.

O’Donoghue M, Boutin S, Murray DL et al. (2001) Coyotes and Lynx. In: Ecosystem Dynamics of the Boreal Forest (eds Krebs CJ, Boutin S, Boonstra R), pp. 275–323. Oxford University Press, Oxford.

Okarma H (1995) The trophic ecology of wolves and their predatory role in ungulate communities of forest ecosystems in Europe.

Acta Theriologica, 40, 335–386.

Ostrander EA, Sprague GF Jr, Rine J (1993) Identification and characterization of dinucleotide repeat (CA)n markers for genetic mapping in the dog. Genomics, 16, 207–332.

Peterson RO, Ciucci P (2003) The wolf as a carnivore. In: Wolves: Behavior, Ecology, and Conservation (eds Mech LD, Boitani L), pp. 104–130. The University of Chicago Press, Chicago.

Petit RJ, El Mousadik A, Pons O (1998) Identifying populations for conservation on the basis of genetic markers. Conservation Biology, 12, 844–855.

Posada D, Crandall KA (1998) modeltest: testing the model of DNA substitution. Bioinformatics, 14, 817–818.

Posada D, Crandall KA, Templeton AR (2000) geodis: a program for the cladistic nested analysis of the geographical distribution of genetic haplotypes. Molecular Ecology, 9, 487–488.

Potvin F, Jolicoeur H, Huot J (1988) Wolf diet and prey selectivity during two periods for deer in Quebec: decline versus expansion. Canadian Journal of Zoology, 66, 1274–1279.

Randi E, Lucchini V, Christensen MF et al. (2000) Mitochondrial DNA variability in Italian and East European wolves: detecting the consequences of small population size and hybridization.

Conservation Biology, 14, 464–473.

Raufaste N, Rousset F (2001) Are partial Mantel tests adequate?

Evolution, 55, 1703–1705.

Raymond M, Rousset F (1995) genepop (version 1.2): population genetics software for exact tests and ecumenicism. Journal of Heredity, 86, 248–249.

Rousset F (2002) Partial Mantel tests: reply to Castellano and Balletto. Evolution, 56, 1874–1875.

Roy MS, Geffen E, Smith D, Ostrander E, Wayne RK (1994) Patterns of differentiation and hybridization in North American wolf-like canids revealed by analysis of microsatellite loci.

Molecular Biology and Evolution, 11, 553–570.

Rueness EK, Jorde PE, Hellborg L et al. (2003a) Cryptic population structure in a large, mobile mammalian predator: the Scandinavian lynx. Molecular Ecology, 12, 2623–2633.

Rueness EK, Stenseth C, O’Donoghue M et al. (2003b) Ecological and genetic spatial structuring in the Canadian lynx. Nature, 425, 69–72.

Sacks BN, Brown SK, Ernest HB (2004) Population structure of California coyotes corresponds to habitat-specific breaks and illuminates species history. Molecular Ecology, 13, 1265–1275.

Schmidt HA, Strimmer K, Vingron M, von Haeseler A (2000) TREE-PUZZLE, Version 5.0: Maximum Likelihood Analysis for Nucleotide, Amino Acid and Two-state Data. Available at: http:// www.tree-puzzle.de.

Schneider S, Roessli D, Excoffier L (2000) ARLEQUIN, Version 2.000.

A Software for Population Genetic Data Analysis. Genetic and Biometry Laboratory, University of Geneva.

Shibuya H, Collins BK, Huang TH-M, Johnson GS (1994) A poly© 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

morphic (AGGAAT)n tandem repeat in an intron of the canine von Willebrand factor gene. Animal Genetics, 25, 122.

Shimodaira H, Hagesawa M (1999) Multiple comparisons of log-likelihoods with applications to phylogenetic inference.

Molecular Biology and Evolution, 16, 1114–1116.

Sidorovich VE, Tikhomirova LL, Jedrzejewska B (2003) Wolf Canis lupus numbers, diet and damage to livestock in relation to hunting and ungulate abundance in northeastern Belarus during 1990–2000. Wildlife Biology, 9, 103–111.

Smouse PE, Long JC, Sokal RR (1986) Multiple regression and correlation extensions of the Mantel test of matrix correspondence.

Systematic Zoology, 35, 627–632.

Spinks PQ, Shaffer HB (2005) Range-wide molecular analysis of the western pond turtle (Emys marmorata): cryptic variation, isolation by distance, and their conservation implications.

Molecular Ecology, 14, 2047–2064.

Sponer R, Roy MS (2002) Phylogeographic analysis of the brooding brittle star Amphipholis squamata (Echinodermata) along the coast of New Zealand reveals high cryptic genetic variation and cryptic dispersal potential. Evolution, 56, 1954–1967.

Stenseth NC, Ehrich D, Rueness EK et al. (2004a) The effect of climatic forcing on population synchrony and genetic structuring of the Canadian lynx. Proceedings of the National Academy of Sciences, USA, 101, 6056–6061.

Stenseth NC, Shabbar A, Chan K-S et al. (2004b) Snow conditions may create an invisible barrier for lynx. Proceedings of the National Academy of Sciences, USA, 101, 10632–10634.

Sulkava S, Pulliainen E (1999) Canis lupus Linnaeus, 1758. In: The Atlas of European Mammals (eds Mitchell-Jones AJ, Amori G, Bogdanowicz W et al.), pp. 314–315. T&AD Poyser, London.

Swofford DL (1998) PAUP*: Phylogenetic Analysis Using Parsimony (*and Other Methods). Sinauer Associates, Sunderland, Massachusetts.

Taberlet P, Fumagalli L, Wust-Saucy AG, Cosson JF (1998) Comparative phylogeography and postglacial colonization routes in Europe. Molecular Ecology, 7, 453–464.

Templeton AR (1998) Nested clade analysis of phylogenetic data: testing hypothesis about gene flow and population history.

Molecular Ecology, 7, 381–397.

Templeton AR (2004) Statistical phylogeography: methods of evaluating and minimizing inference errors. Molecular Ecology, 13, 789–809.

Templeton AR, Crandall KA, Sing CFA (1992) A cladistic analysis of phenotypic associations with haplotypes inferred from restriction endonuclease mapping and DNA sequence data. III. Cladogram estimation. Genetics, 132, 619–633.

Templeton AR, Sing CFA (1993) A cladistic analysis of phenotypic associations with haplotypes inferred from restriction endonuclease mapping. IV. Nested analysis with cladogram uncertainty and recombination. Genetics, 134, 659–669.

Valière N (2002) gimlet: a computer program for analysing genetic individual identification data. Molecular Ecology Notes, 2, 377–379.

Vatolin BA (1979) On the wolf population and regulation of its numbers in the Bryansk region. In: Ecological Fundamentals of Protection and Rational Utilisation of Predatory Mammals (ed. Sokolov VE), pp. 91–93. Izdatelstvo Nauka, Moskva. [in Russian].

Vilà C, Amorim IR, Leonard JA et al. (1999) Mitochondrial DNA phylogeography and population history of the grey wolf Canis lupus. Molecular Ecology, 8, 2089–2103.

Vilà C, Savolainen P, Maldonaldo JE et al. (1997) Multiple and ancient origins of the domestic dog. Science, 276, 1687–1689.

4550 M . P I L O T E T A L .

Wabakken P, Sand H, Liberg O, Bjärvall A (2001) The recovery, distribution, and population dynamics of wolves on the Scandinavian Peninsula, 1978–98. Canadian Journal of Zoology, 79, 710– 725.

Waits L, Taberlet P, Swenson JE, Sandegren F, Franzen R (2000) Nuclear DNA microsatellite analysis of genetic diversity and gene flow in the Scandinavian brown bear (Ursus arctos). Molecular Ecology, 9, 421–431.

Wandeler P, Smith S, Morin PA, Pettifor RA, Funk SM (2003) Patterns of nuclear DNA degradation over time — a case study in historic teeth samples. Molecular Ecology, 12, 1087–1093.

Wayne RK, Lehman N, Allard MW, Honeycutt RL (1992) Mitochondrial DNA variability of the gray wolf: Genetic consequences of population decline and habitat fragmentation.

Conservation Biology, 6, 559–569.

Wojtulewicz M (2004) Population dynamics, migrations and diet of wolf (Canis lupus) in the region of Biebrza. MSc Thesis, Warsaw Agricultural University and Mammal Research Institute PAS. [in Polish].

Yang DY, Eng B, Waye JS, Dudar JC, Saunders SR (1998) Improved DNA extraction from ancient bones using silica-based spin columns. American Journal of Physical Anthropology, 105, 539– 543.

M. Pilot uses molecular genetic methods in studies on ecology and evolution of mammals. Professors W. Jedrzejewski and B. Jedrzejewska are specialists in predator-prey relationships, animal population dynamics, and biogeographical patterns in carnivore ecology. W. Branicki is interested in application of molecular methods to human and animal forensic studies. V. Sidorovich studies the ecology of carnivorous mammals in northeastern Belarus. K. Stachura is interested in applying GIS methods to study patterns of animal distribution. S. M. Funk uses molecular and ecological tools to address questions of conservation management and evolutionary biology with an emphasis on carnivores.

© 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4551

Appendix I Heterozygosity deficit in microsatellite genotypes of Eastern European wolves and its effect on the results of the analysis of population genetic structure We analysed deviations from Hardy–Weinberg equilibrium (HWE) and their direction (heterozygote deficit or excess) using the exact test of Guo & Thompson (1992) implemented in genepop (Raymond & Rousset 1995). In the total population of Eastern European wolves, we found a deficit in average observed heterozygosity relative to HWE (P < 0.0001), and positive FIS (0.09). This deviation from HWE was due to 11 loci. We also found significant linkage disequilibrium for 59 of 91 pairs of loci (P < 0.05, after correcting for multiple tests). Among analysed loci, only two pairs were located on the same chromosomes and in both cases the loci were located in distant parts of these chromosomes (Breen et al. 2001), so disequilibrium was unlikely to be due to physical linkage.

We also tested for HWE in two main subpopulations, A and B, presented in Fig. 5a (two individuals assigned by geneland to subpopulation C were included into subpopulation B). In both subpopulations, heterozygote deficit was significant (P < 0.0001 in each case), and FIS was positive (0.10 in subpopulation A and 0.06 in subpopulation B).

Additionally, we performed multipopulation tests for HWE for each analysed locus, using 16 regions as spatial units that defined sample groups (see Fig. 1). Out of 14 analysed microsatellite loci, seven showed significant heterozygote deficit (P < 0.001 in each case, after correcting for multiple tests). For these seven loci, the number of sample groups showing significant heterozygote deficit was as follows: 2 at the loci FH2079, FH2088, C253, and AHT130; 5 at the locus FH2017; 9 at the locus C213, and 10 at the locus C642. At the remaining loci, significant heterozygote deficit was observed in at most one sample group. None of the analysed loci showed significant heterozygote excess. FIS was positive in 15 out of the 16 sample groups and ranged from 0.01 to 0.20.

The observed deficit of heterozygotes may due to several reasons: (i) the presence of null alleles; (ii) the existence of an underlying genetic structure (Wahlund’s effect); (iii) inbreeding in local wolf populations; (iv) the presence of closely related individuals (members of the same packs) in the sample. Null alleles are likely to occur at the loci showing significant heterozygote deficit, especially at the loci C213 and C642 that show heterozygote deficit in more than half of the sample groups. However, other population genetic studies on Eurasian and North American wolves (that used different sets of microsatellite loci) also showed significant heterozygote deficit and positive values of FIS, which was explained by moderate inbreeding, the presence of closely related individuals in the analysed sample, or the presence of genetic structuring (Roy et al. 1994; Forbes & Boyd 1997; Lucchini et al. 2004).

Although we cannot exclude the presence of null alleles in our data, they are unlikely to be a problem for the analysis of population genetic structure, as simulations have shown that geneland is robust with regard to null alleles (unpublished data of G. Guillot reported in Coulon et al. 2006). The Wahlund’s effect cannot be excluded, either. However, it would not negate the existence of the genetic structure detected but only imply the existence of an additional, undetected structure. The presence of closely related individuals in the sample is unlikely to lead to the improper inference of population genetic structure by geneland as well (see Coulon et al. 2006). On the other hand, the departure from the model assumptions (HWE and linkage equilibrium) may create ghost populations (Coulon et al. 2006). Ghost populations, as defined by Guillot et al. (2005a), are not modal for any individual and thus can be easily identified. Subpopulation C inferred in our study in most geneland runs might be a ghost population, because it was modal for only several individuals (1–21 depending on the run), and these individuals did not form spatially homogenous groups (see Results section). The inference of this subpopulation may be the result of the departure from the model assumptions.

© 2006 The Authors Journal compilation © 2006 Blackwell Publishing Ltd

4552 M . P I L O T E T A L .

Appendix II Results of nested clade analysis of phylogenetic relationships among mtDNA haplotypes of wolves from Eastern Europe

Clade distances (Dc) and nested clade distances (Dn) are calculated for each clade within the nested group. In the row labelled I-T the average differences in distances between interior and tip clades are given. Interior clades are shaded. At the bottom of those boxes in which one or more of the geographical distance measures for nested clades was significantly large (L) or small (S) is a line with the biological inference.

The numbers refer to the sequence of questions in the inference key (Templeton 2004) that the pattern generated, followed by the answer to the final question in the key. RE, range expansion; CRE, continuous range expansion; LDC, long distance colonization; RGF, recurrent restricted gene flow with isolation by distance; LDD, long distance dispersal. Š 2006 The Authors Journal compilation Š 2006 Blackwell Publishing Ltd

E C O L O G I C A L F A C T O R S A N D G E N E F L O W I N W O L V E S 4553

Appendix III Results of the analysis of population genetic structure using geneland: maps of the posterior probability to belong to each subpopulation (a) for the geneland assignment presented on Fig. 5a, (b) for the geneland assignment with the highest mean posterior probability. Lighter colours denote higher assignment probabilities to a given subpopulation. Units of axis are geographical coordinates.

Š 2006 The Authors Journal compilation Š 2006 Blackwell Publishing Ltd