Lucchini, V., Galov, A., and Randi, E. (2004). Evidence of ... · the wolf (Canis lupus) population...

Lucchini, V., Galov, A., and 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.

Keywords: Bayesian inference/Canis lupus/demography/genetic diversity/genetics/Malme/ microsatellites/mtDNA/population trend/wolf

Abstract: Historical information suggests the occurrence of an extensive human-caused contraction in the distribution range of wolves ( Canis lupus ) during the last few centuries in Europe. Wolves disappeared from the Alps in the 1920s, and thereafter continued to decline in peninsular Italy until the 1970s, when approximately 100 individuals survived, isolated in the central Apennines. In this study we performed a coalescent analysis of multilocus DNA markers to infer patterns and timing of historical population changes in wolves surviving in the Apennines. This population showed a unique mitochondrial DNA control-region haplotype, the absence of private alleles and lower heterozygosity at microsatellite loci, as compared to other wolf populations. Multivariate, clustering and Bayesian assignment procedures consistently assigned all the wolf genotypes sampled in Italy to a single group, supporting their genetic distinction. Bottleneck tests showed evidences of population decline in the Italian wolves, but not in other populations. Results of a Bayesian coalescent model indicate that wolves in Italy underwent a 100- to 1000-fold population contraction over the past 2000-10 000 years. The population decline was stronger and longer in peninsular Italy than elsewhere in Europe, suggesting that wolves have apparently been genetically isolated for thousands of generations south of the Alps. Ice caps covering the Alps at the Last Glacial Maximum ( c . 18 000 years before present), and the wide expansion of the Po River, which cut the alluvial plains throughout the Holocene, might have provided effective geographical barriers to wolf dispersal. More recently, the admixture of Alpine and Apennine wolf populations could have been prevented by deforestation, which was already widespread in the fifteenth century in northern Italy. This study suggests that, despite the high potential rates of dispersal and gene flow, local wolf populations may not have mixed for long periods of time.

Molecular Ecology (2004)

13

, 523–536 doi: 10.1046/j.1365-294X.2004.02077.x

© 2004 Blackwell Publishing Ltd

Blackwell Publishing, Ltd.

Evidence of genetic distinction and long-term population decline in wolves (

Canis lupus

) in the Italian Apennines

V . LUCCHINI ,

*

A . GALOV

†

and E . RANDI

*

*

Istituto Nazionale per la Fauna Selvatica (INFS), Ozzano Emilia (BO), Italy,

†

Department of Animal Physiology, Faculty of Science, University of Zagreb, Zagreb, Croatia

Abstract

Historical information suggests the occurrence of an extensive human-caused contractionin the distribution range of wolves (

Canis lupus

) during the last few centuries in Europe.Wolves disappeared from the Alps in the 1920s, and thereafter continued to decline inpeninsular Italy until the 1970s, when approximately 100 individuals survived, isolated inthe central Apennines. In this study we performed a coalescent analysis of multilocus DNAmarkers to infer patterns and timing of historical population changes in wolves survivingin the Apennines. This population showed a unique mitochondrial DNA control-regionhaplotype, the absence of private alleles and lower heterozygosity at microsatellite loci, ascompared to other wolf populations. Multivariate, clustering and Bayesian assignmentprocedures consistently assigned all the wolf genotypes sampled in Italy to a single group,supporting their genetic distinction. Bottleneck tests showed evidences of populationdecline in the Italian wolves, but not in other populations. Results of a Bayesian coalescentmodel indicate that wolves in Italy underwent a 100- to 1000-fold population contractionover the past 2000–10 000 years. The population decline was stronger and longer in peninsularItaly than elsewhere in Europe, suggesting that wolves have apparently been geneticallyisolated for thousands of generations south of the Alps. Ice caps covering the Alps at theLast Glacial Maximum (

c

. 18 000 years before present), and the wide expansion of thePo River, which cut the alluvial plains throughout the Holocene, might have providedeffective geographical barriers to wolf dispersal. More recently, the admixture of Alpineand Apennine wolf populations could have been prevented by deforestation, which wasalready widespread in the fifteenth century in northern Italy. This study suggests that,despite the high potential rates of dispersal and gene flow, local wolf populations may nothave mixed for long periods of time.

Keywords

: Bayesian inference,

Canis lupus

, coalescent, demographic history, microsatellites, mtDNAcontrol-region, population structure, wolf

Received 18 May 2003; revision received 11 September 2003; accepted 31 October 2003

Introduction

Wolves (

Canis lupus

) are highly adaptable and widelydistributed in ecosystems ranging from Arctic tundra toArabian deserts in the Old and New World (Mech 1970).Field observations and population and genetic studiesindicate that wolves may disperse rapidly over long dis-tances, either by recurrent post-reproductive dispersal orduring waves of population expansion (Fritts 1993; Forbes

& Boyd 1996; Vilà

et al

. 1999; Walton

et al

. 2001). Expandingwolf populations have rapidly recolonized suitable areasof their historical range in North America and Europe, andoccasional events of long-distance colonization have beendescribed (Forbes & Boyd 1996; Wabakken

et al

. 2001;Lucchini

et al

. 2002; Flagstad

et al

. 2003; Valière

et al

. 2003).Cycles of demographic contraction/expansion in the latePleistocene and extensive historical gene flow apparentlyled to widespread admixture of wolf populations world-wide. Hence, wolves in North America and Europe donot show any large-scale phylogeographical structure, sug-gesting that, before the recent human-caused population

Correspondence: Ettore Randi. Fax: +39 051796 628; E-mail:[email protected]

524

V . L U C C H I N I , A . G A L O V and E . R A N D I

© 2004 Blackwell Publishing Ltd,

Molecular Ecology

, 13, 523–536

fragmentation, wolves were not geographically differ-entiated (Vilà

et al

. 1999; Randi

et al

. 2000).However, historical processes and current population

dynamics might be more complex than previously sug-gested (Leonard

et al

. 2000; Carmichael

et al

. 2001), andphylogeographical inference from contemporary geneticdata may not reflect the complexity of historical processes(Barnes

et al

. 2002). Permanent physiographic traits oranthropogenic habitat fragmentation may limit individualdispersal and gene flow. Wolves do not expand in agricul-tural landscapes, which, in contrast, are commonly usedby other canids, like coyotes in North America and jackalsin Eurasia (Wayne

et al

. 1992). Wolves developed eitherstrictly territorial or migratory behaviours in westernCanada, adapting to the behaviour of their prevalent prey,the moose or the caribou (Carmichael

et al

. 2001). Adapta-tions to prey availability apparently constrained thepatterns of gene flow, thus driving to population patchinessand genetic differentiation (Carmichael

et al

. 2001).Wolves were presumably widespread almost every-

where in Eurasia throughout the Holocene (Boitani 1999).Human persecution, deforestation and the decrease ofnatural prey led wolf populations to decline in Europeduring the last centuries (Delibes 1990). Large populationssurvived in the Balkans and eastern Europe, while thespecies was eradicated in central Europe and Scandinavia,and survived only in fragmented populations in the Iberianand Italian peninsulas. Wolves disappeared from the Alpsin the 1920s and drastically declined in Italy in the twodecades after World War II. By 1973 there were approxi-mately 100 individuals surviving, isolated in the centralApennines (Zimen & Boitani 1975). Legal protection and

the expansion of natural prey populations contributed tocheck the wolf’s decline, and a census in 1983 suggestedthe presence of about 220 wolves (Boitani 1984). Thereafter,wolves expanded rapidly along the Apennine ridge,recolonizing the Western Italian and French Alps in 1992(Breitenmoser 1998; Lucchini

et al

. 2002; Valière

et al

. 2003)(Fig. 1). Ciucci & Boitani (1991) estimated an annualpopulation increase of 7% from 1973 to 1988, leading themto argue that wolves in Italy should now number about600 individuals.

Wolves in Italy have distinctive genetic traits showingone unique mitochondrial DNA control-region (mtDNACR) haplotype and microsatellite allele frequencies thatare sharply different from any other wolf population anddog breed typed so far (Randi

et al

. 2000; Randi & Lucchini2002). Random drift might have fixed one mtDNA haplo-type, assuming that the declining and isolated Italian wolfpopulation persisted at a very low effective size duringthe last century (Randi

et al

. 2000). However, the mtDNAgenome behaves as a single gene, which tracks the outcomeof a single (stochastic) coalescent process (Rosenberg &Nordborg 2002). More reliable inferences on past popu-lation dynamics should be obtained using multipleindependent genetic markers, and historical changes ineffective population size can now be inferred from micro-satellite allele frequency distributions (Wilson & Balding1998).

In this study we used a coalescent analysis of micro-satellite data (Beaumont 1999) to infer the extent andtiming of historical demographic changes in the Apenninewolves. We aimed to evaluate the likelihood of two alter-native scenarios of genetic diversification of wolves in

Fig. 1 Current distribution range (dark) ofthe wolf (Canis lupus) population in Italyand in the French Alps. The shaded areasindicate the approximate extensions of thePyrenean and Alpine ice caps at the lastglacial maximum. The grey line indicatesthe approximate coastline at the last glacialmaximum, when the Mediterranean Seawas about 120 m lower than present. Thearrows indicate the Po River and thelocation of the last wolf that was killed inthe Alps in 1912.

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

525


Molecular Ecology

, 13, 523–536

peninsular Italy, which might have been affected either bylate Pleistocene/early Holocene natural landscape changes,or by the more recent human-mediated habitat trans-formations. The ice caps covering the Alps and the wideexpansion of the Po River, which cut the alluvial plainsthroughout the Holocene, might have isolated wolvesin central-southern Apennines since the Last GlacialMaximum (

c

. 18 000 years before present). Alternatively,deforestation, which was already widespread in the fifteenthcentury in northern Italy, and direct human persecution,might have limited the rate of gene flow among wolves inthe Apennines and any other population in Europe duringthe last few centuries. The results from this study areaimed to provide new information on the dynamics of thehistorical processes that shaped the genetic diversity of thewolf population surviving in peninsular Italy.

Materials and methods

Sample collection and DNA analyses

Wolf and dog samples, 383 in all, were obtained from11 regions in Europe, Turkey, Israel and Saudi Arabia(Table 1). Presumptive wolves were identified morpho-logically by collectors, mainly using coat colour traits. Wolfsamples were obtained from found-dead, shot, or trappedwild-living individuals, except for some captive wolvessampled from Zoological Gardens in Spain and SaudiArabia (thanks to Carles Vilà, Uppsala University, Sweden).As far as was known, captive wolves originated from wild-living Spanish or Arabian populations and were used inthis study to compare the range of genotype diversityamong samples collected from distant geographical areas,not to assess the local population structure. Estimatesof genetic diversity obtained from small samples (Turkey,

Israel and Finland) should be evaluated with caution. Dogswere randomly sampled from 30 different breeds fromveterinary practices, and included some free-living dogs,which were collected in the central Italian Apennines.Previously identified hybrid wolves were included in thisstudy as reference genotypes for admixture analyses(Andersone

et al

. 2002; Randi & Lucchini 2002).Samples were individually stored at

−

20

°

C in 95%ethanol (tissues) or in a Tris/sodium dodecyl sulphatebuffer (blood). Total DNA was extracted from tissues usinga guanidinium–silica protocol (Gerloff

et al

. 1995) and fromblood using a salting-out procedure (Miller

et al

. 1988). Thediagnostic mtDNA CR haplotype of Italian wolves wasidentified using laboratory protocols described by Randi

et al

. (2000). All wolves and dogs were genotyped using apanel of 18 canine microsatellite loci, which were assignedto 16 different chromosomes (Neff

et al

. 1999), as describedby Randi & Lucchini (2002). DNA sequences and micro-satellite alleles were analysed in an ABI 3100 automatedsequencer, using the programs

genescan

3.7 and

genotyper

2.1 for microsatellites, and sequencing analysis 3.7 and

ntnavigator

for sequences.

Admixture analyses and identification of the hybrids

Andersone

et al

. (2002), Randi & Lucchini (2002) and Vilà

et al

. (2003) reported evidence of occasional wolf

×

dogcross-breeding in Italy and elsewhere in Europe. Wolvesand dogs have distinct multilocus microsatellite genotypesand hybrids can be identified by assignment procedures(Randi & Lucchini 2002; Vilà

et al

. 2003). In this study allthe genotypes were screened using a Bayesian admixtureprocedure implemented in

structure

(Pritchard

et al

.2000; http://pritch.bsd.uchicago.edu).

structure

wasused (with 10

6

iterations, following a burn-in period of

Table 1 Origin and number of wolf and dog samples used in this study

Group Origin Samples Collectors

Wolves Turkey 4 C. Vilà, Uppsala University, Uppsala, SwedenSpain 32 C. Vilà; J. C. Blanco, Madrid, SpainBulgaria 39 (5 dogs) P. Genov, Bulgarian Academy of Sciences, Sofia, BulgariaCroatia 28 (3 dogs; 1 hybrid) D. Huber, University of Zagreb, Zagreb, CroatiaIsrael 4 (1 hybrid) G. Kahila, Hebrew University, Jerusalem, IsraelFinland 13 I. Gade-Jorgensens and R. Stagegaard, Kopenhagen, DenmarkGreece 6 Y. Mertzanis, Thessaloniki, GreeceLatvia 47 (9 hybrids) Z. Andersone, Kemeri National Park, Jurmala, LatviaItaly 105 (2 hybrids) INFS, Bologna, ItalySaudi Arabia 10 C. Vilà

Dogs Switzerland 72 G. Dolf, University of Berne, Berne, SwitzerlandItaly 23 INFS

Presumptive wolves were phenotypically identified by collectors. Bayesian admixture analyses of microsatellite genotypes showed that some presumptive wolves (in brackets) were mislabelled dogs or hybrids with dogs.

526

V . L U C C H I N I , A . G A L O V and E . R A N D I


Molecular Ecology

, 13, 523–536

30 000 iterations) to identify the number of geneticallydistinct clusters that maximize the likelihood of the data,and to assign the individuals to the clusters, using onlygenetic information (for details see: Randi & Lucchini 2002).Thirteen presumptive wolf–dog hybrids were identified,and eight samples were found that were labelled as wolvesbut that were genetically assigned to the dog population(see Results). All these samples were excluded, and 267wolf and 95 dog samples were used for the followinggenetic analyses.

Genetic diversity within and among populations

Commonly used summary population genetic statisticswere computed for each locus and population using

genetix

4.04 (Belkir

et al

. 1996–2001; http://www.University-montp2.fr/

∼

genetix/genetix.htm). To avoid using smallsample sizes, the wolves from Greece (

n

= 6) were pooledwith those from Bulgaria (

n

= 34), and the wolves fromTurkey (

n

= 4) were pooled with those from Israel (

n

= 3).This should not blur the estimates of diversification amongregions, because the allele frequency distributions of thepooled groups were not significantly different (

P

> 0.05;Fisher’s exact test; Goudet

et al

. 1996).Allele numbers per locus and population were corrected

for sample size variation, computing the allelic richness

A

C

on a minimum of 19 individuals per sample, using

fstat

2.9.1 (Goudet 1995; http://www.unil.ch/izea/softwares/fstat.html). Moreover, 20 random subsamples wereconstructed, each one of 30 individuals, from the Italianwolf populations. Subsamples were used to estimate theexpected heterozygosity

H

E

in samples correspondingto the average size of the non-Italian wolves (

n

= 29.4,computed using the wolves from Spain, Bulgaria + Greece,Croatia, Finland and Latvia, and excluding Turkey, Israeland the captive wolves from Saudi Arabia).

Deviations from Hardy–Weinberg equilibrium for eachlocus and population were assessed using a Markov chainmethod (Guo & Thompson 1992) implemented in

genepop

3.2a (Raymond & Rousset 1995; http://www.cefe.cnrs-mop.fr). Deviations from allelic linkage equilibrium forall pairwise locus combinations were tested using theexact probability test in

genepop

. Critical levels insimultaneous tests of significance were adjusted usingHochberg’s procedure (Legendre & Legendre 1998: p. 18),as implemented in

adjusted p

-

values

(P. Legendre; http://www.fas.umontreal.ca/biol/legendre). The probability ofdifferent individuals sharing by chance an identicalgenotype (probability of identity; Waits

et al

. 2001), andthe expected probability of identity among fullsibs, wereestimated using the software

prob

-

id

3 (G. Luikart unpub-lished results).

Population differentiation was assessed by Fisher’s exacttest, analogues of pairwise mean

F

ST

(Weir & Cockerham

1984), and analysis of molecular variance (

amova

;Michalakis & Excoffier 1996), using

arlequin

2.0b8(Schneider

et al

. 2002; http://anthropologie.unige.ch/arlequin). Patterns of differentiation were visualized byfactorial correspondence analysis (Benzécri 1973) ofindividual multilocus scores computed using

genetix

.Various interindividual microsatellite genetic distanceswere computed using

microsat

1.5d (Minch

et al

. 1997),and clustered using the neighbour-joining procedure(Saitou & Nei 1987) as implemented in

phylip

3.5c (J.Felsenstein; http://evolution.genetics.washington.edu/phylip.html).

Bottleneck analyses and inferences on past population dynamics

Population bottlenecks can produce distinctive geneticsignatures in the distributions of allele size, expectedheterozygosity and in the genealogy of microsatelliteloci (Cornuet & Luikart 1996; Beaumont 1999; Garza &Williamson 2001). In this study three different procedureshave been used. The

M

-ratio test (Garza & Williamson2001; http://www.pfeg.noaa.gov/tib/carlos.htm) wasused to estimate the probability of the observed

M

-values,under the assumption of the two-phase mutation model (DiRienzo

et al

. 1994), with proportion of one-step mutations

ps

= 90% or 95%, average size of non one-step mutations

∆

g = 3.5, and

θ

= 4

N

e

µ

(where

N

e

= effective populationsize at equilibrium before the bottleneck;

µ

= mutation rate)equal to 5 or 10. Cornuet & Luikart’s (1996) Wilcoxon’ssigned-ranks and mode-shift tests, as implemented in

bottleneck

1.2.02 (Piry

et al

. 1999; http://www.ensam.inra.fr/URLB), were used under the two-phase mutation modelwith 90% or 95% single-step mutations.

Evidence for historical demographic changes was inferredusing the MSVAR procedure (Beaumont 1999; http://www.rubic.rdg.ac.uk/

∼

mab/software.html). Beaumont’smodel implements a Bayesian approach, where the dataand the parameters are random variables that have somejoint distribution that can be explicitly modelled. The priordistributions of the parameters and the conditional distri-butions of the data given the parameters allow estimationof the posterior distributions using Bayes theorem. MSVARassumes a strict stepwise mutation model (Kimura & Ohta1978), and estimates the posterior probability distributionsof the rate of population change

r

=

N

0

/N

1

(where

N

0

=current effective number of chromosomes;

N

1

= number ofchromosomes at the time of population decline or expan-sion), the time in generations when the population startedto expand or decline

tf

=

ta

/

N

0

(

ta

= number of generationssince the beginning of the expansion/decline), and thegenetic parameter

θ

= 2

N

0

µ

. The model assumes that thedemographic parameters are identical across loci, whilemutation rates are free to vary. Rectangular priors were

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

527


Molecular Ecology

, 13, 523–536

assumed for the parameters, with limits of (

−

5, +5) forlog

10

(

r

), log

10

(

tf ), and log10(θ), and performed the analysesfor a linear model of population change. Stability of theestimates was evaluated by five independent replicationsof the Markov chain simulations for each population, with20 000 thinned updates and a thinning interval of 10 000steps, leading to a total number of 208 updates (as sug-gested by Beaumont 1999). The first 10% of updates werediscarded to avoid bias in parameter estimation at startingconditions. The remaining data were used to obtain themedian (50%), and the lower (5%) and upper (95%)quantiles of the posterior distributions of the parameters.Consistency in shape of the posterior distributions fromthe individual runs was examined to evaluate convergenceof the output values. In all bottleneck tests only 17 loci wereused because one locus (CHP6) did not show a regularstep-wise mutation pattern.

Estimating Ne/N in Italian wolves

Census data suggested that the population size in the Italianwolf at the most recent bottleneck was approximatelyN = 100 individuals (Ciucci & Boitani 1991). Demographicparameters of the Italian wolf population are unknown,and the effective population size Ne cannot be estimateddirectly. Therefore, a measure of temporal changes in allelefrequencies has been used (Waples 1989). Genotypessampled from the wolf population in central Italy (that isfrom the areas where wolves survived in the 1960s andfrom where they started to recolonize the Apennines) weregrouped in two groups: (1) wolves which were born in1988–90 (n = 26); and (2) wolves which were born in 1994–96 (n = 24). Wolves in (1) were, on average, 6 years older

than the wolves in (2), that is, assuming that averagegeneration time in wolves is 3 years, there were twogenerations between (2) and (1). All DNA samples used inthese analyses were obtained from found-dead wolf bodies.Skulls are preserved at the INFS Museum, and the age ofwolf samples was estimated from canine tooth sections(Ballard et al. 1995). We have used two procedures. The firstone, implemented in software tm3.1 (Berthier et al. 2002;http://www.rubic.rdg.ac.uk/∼mab/software.html) is acoalescent approach which computes posterior distribu-tions of Ne assuming that allele frequencies were estimatedfrom a closed population sampled at two time points, withtime sufficiently short that mutations did not affect theobserved frequencies. The second procedure, implementedin mcleeps1.1 (Anderson et al. 2000; http://www.stat.washington.edu/thompson/Genepi/Mcleeps.shtml) is anoncoalescent maximum-likelihood approach.

Results

Admixture analyses and identification of hybrids

All the 18 microsatellites were polymorphic in wolves anddogs, with the exception of five monomorphic loci in thesmall Saudi Arabian sample. The individual multilocusgenotypes were unique, with values of probability ofidentity = 0.0, and probability of identity among full sibs =4.3 × 10−7 and 1.1 × 10−9 in wolves and dogs, respectively.These genotypes were screened for hybridization usingstructure (performed with n = 383, 18 loci, using onlygenetic information), which showed that the highestposterior probability of the data was obtained by splittingthe total sample into seven clusters (Table 2). All dogs

Population

Cluster

I II III IV V VI VII

WolvesTurkey 0.012 0.968 0.005 0.003 0.004 0.005 0.003Spain 0.004 0.005 0.011 0.006 0.004 0.006 0.964Bulgaria 0.016 0.036 0.749 0.131 0.005 0.029 0.033Croatia 0.745 0.011 0.073 0.116 0.011 0.035 0.009Israel 0.035 0.753 0.010 0.103 0.009 0.071 0.019Finland 0.017 0.876 0.023 0.008 0.006 0.061 0.010Greece 0.029 0.006 0.928 0.011 0.007 0.010 0.009Latvia 0.031 0.013 0.020 0.150 0.020 0.752 0.015Italy 0.003 0.002 0.002 0.007 0.980 0.003 0.003Saudi Arabia 0.980 0.004 0.003 0.003 0.004 0.004 0.002

Dogs 0.010 0.014 0.008 0.941 0.003 0.016 0.009

The posterior probability of the data was maximum with K = 7 clusters. The table shows the proportion of membership (qi) of each predefined sampled population in each of seven inferred clusters. Each population was assigned to a single cluster if its qi (i = I–VII) was ≥ 0.900, or to two or more clusters when the cumulative qi values were ≥ 0.900.

Table 2 Bayesian clustering analysis in thepooled wolf and dog samples (383 indi-viduals, 18 loci, 11 sampled groups) per-formed using structure (Pritchard et al.2000)

528 V . L U C C H I N I , A . G A L O V and E . R A N D I

© 2004 Blackwell Publishing Ltd, Molecular Ecology, 13, 523–536

were assigned to a single cluster (IV) with proportion ofmembership qIV = 0.941, and were distinct from wolves.Wolves sampled from Turkey, Spain, Greece, Italy andSaudi Arabia were also assigned to distinct single clusters(cluster II, VII, III, V and I, respectively), each group withqi ≥ 0.928. In contrast, wolves sampled from Bulgaria,Croatia, Israel, Finland and Latvia were not assigned toany single cluster with qi ≥ 0.900, but they were split intotwo or more clusters.

Samples from Bulgaria, Croatia, Israel and Latvia werepartially assigned also to the dog cluster IV, with qi ≥ 0.103.Five presumptive wolves from Bulgaria and three fromCroatia were assigned to the dog cluster with individualqi ≥ 0.900. The mtDNA haplotypes of these samples joinedthe domestic dog mtDNA clusters named C1 and C2 inRandi et al. 2000). These samples, which had their ancestryonly in the dog gene pool, might be mislabelled as dogs.One sample from Croatia, one from Israel, nine from Latvia(a family of hybrid pups previously studied by Andersoneet al. 2002), two from Italy (which were previously studiedby Randi & Lucchini 2002), were jointly assigned to the dogand to the wolf clusters, showing admixed ancestry. All thewolves sampled in Italy, including the two admixed indi-viduals, showed the Italian wolf mtDNA CR haplotype(named W14 in Randi et al. 2000; see also Randi & Lucchini2002). The admixed wolves from Croatia and Israel showedtwo mtDNA haplotypes, which were found also in otherwolves sampled in eastern European regions. The familyof hybrid pups showed a unique mtDNA CR haplotype,which was identical to a haplotype found in other wolvesfrom Latvia (Andersone et al. 2002). All these samples,which had their ancestry only in the dog cluster, or thatwere admixed with dogs, were excluded from the followinganalyses.

Population genetic diversity

A total of 181 alleles were detected at 18 microsatellite lociin wolves and dogs. Wolves showed 166 alleles in total (9.2alleles per locus, on average), and dogs showed 143 alleles(7.9 alleles per locus, on average). Allelic diversity rangedfrom 2.4 alleles per locus in the Saudi Arabian wolves,to 6.8 in the samples from Greece–Bulgaria and Latvia(Table 3). One hundred Italian wolves showed 4.4 allelesper locus, on average. The correlation between sample sizeand allele number was positive, but not significant (r =0.19, P = 0.43; Z-test). Values of AC, estimated in sampleswith a minimum size of 19 diploid genomes, ranged from3.4 (Italy) to 6.1 (Latvia) alleles. Each wolf populationshowed from two to six private alleles (with frequencyf ≥ 0.01), except for the Italian wolves, which did not showprivate alleles. Private alleles, which are present at lowfrequencies in the studied wolf samples (the averagefrequency within populations of a private allele was f =0.065, SD = 0.081), may be lost by drift during populationbottlenecks.

Values of observed heterozygosity HO were in the rangeof 0.44 (Italy) to 0.71 (Latvia), and values of HE were in therange of 0.42 (Saudi Arabia) to 0.73 (Latvia). The averagevalue of HE, estimated in the total sample, or in 20 randomsubsamples of 30 Italian wolves was 0.49 (SD = 0.02).Therefore, wolves from Italy showed the lowest number ofalleles (adjusting for different sample size) and the lowestobserved heterozygosity (independent of the sample size).

Hardy–Weinberg and linkage equilibrium

The value of FIS estimated over all loci and populationswas significantly positive in wolves (FIS = 0.23; P ≤ 0.01),

Table 3 Genetic diversity in wolf and dog samples genotyped at 18 microsatellite loci

Origin of samples N A na HO (SD) HE (SD) FIS (95% CI)

WolvesSpain 32 4.7 4 0.50 (0.16) 0.60 (0.14) 0.17 (0.08 to 0.22)*Italy 103 4.4 0 0.44 (0.21) 0.49 (0.22) 0.10 (0.06 to 0.14)*Croatia 24 5.4 5 0.63 (0.12) 0.69 (0.11) 0.09 (−0.02 to 0.14)*Greece + Bulgaria 39 6.8 6 0.69 (0.13) 0.73 (0.10) 0.06 (−0.01 to 0.10)*Turkey + Israel 7 3.7 2 0.66 (0.21) 0.67 (0.10) 0.00 (−0.32 to 0.08)Saudi Arabia 7 2.4 2 0.48 (0.36) 0.42 (0.29) −0.14 (−0.45 to −0.11)Latvia 38 6.8 2 0.71 (0.09) 0.73 (0.08) 0.03 (−0.03 to 0.06)Finland 13 5.5 3 0.69 (0.13) 0.73 (0.07) 0.06 (−0.13 to 0.10)Total 263 9.2 0.56 (0.09) 0.72 (0.08) 0.23 (0.21 to 0.25)*Dogs 92 7.9 15 0.51 (0.14) 0.67 (0.17) 0.24 (0.19 to 0.28)*

Wolves from Greece and Bulgaria, and from Turkey and Israel were pooled to avoid low sample sizes. N = mean number of genotyped individuals per locus; A = mean number of alleles per locus; na = number of unique alleles found in a group when it is compared to all the other groups (an allele is considered as unique if it is present at frequency f ≥ 0.01 in a group). HO and HE = observed and expected heterozygosity (SD = one standard deviation). Values of the fixation index FIS are reported with their 95% confidence intervals (CI). The asterisks (*) indicate significant (P ≤ 0.01) departures from Hardy–Weinberg equilibrium, as estimated using genepop.

G E N E T I C I S O L A T I O N O F I T A L I A N W O L V E S 529


showing departures from Hardy–Weinberg equilibriumand suggesting a Wahlund effect, probably as a result ofthe pooling of genetically differentiated samples (Table 3).The average FIS was significantly positive in dogs (FIS =0.24; P ≤ 0.01). All the local wolf samples showed HOvalues slightly lower than HE, and positive FIS values(except in the enclosed wolves from Saudi Arabia). Themultilocus test showed that heterozygote deficit wassignificant (P ≤ 0.01) in wolves from Spain, Italy, Croatiaand Greece–Bulgaria (Table 3).

Pairwise values of linkage equilibrium were significantlydifferent from zero in 124 of 153 comparisons in wolves(P ≤ 0.01) but they were never significantly different indogs. After adjustment of the nominal probability levels,we found, respectively, 11, four, four and one allelic com-binations (involving different loci) that were not in linkageequilibrium in wolves from Spain, Greece–Bulgaria, Latviaand Italy (P ≤ 0.01). Wolves from Spain, which includeindividuals sampled from enclosures, Greece–Bulgariaand Turkey–Israel, which were pooled to avoid using lowsample sizes, do not represent random samples of the wild-living local populations. In contrast, the wolves sampledfrom Italy are representative of the source population. Theobserved departures from Hardy–Weinberg equilibriumand linkage equilibrium suggest that wolves in Italy arenot in mutation–drift equilibrium.

Divergence among wolf populations

Values of average multilocus FST = 0.23 showed thatgenetic diversity was significantly partitioned among theeight wolf groups listed in Table 3 (P ≤ 0.01; amova). Allpairwise FST values (Table 4), excluding Latvia vs. Greece–Bulgaria, and Finland vs. Turkey–Israel, were significantlydifferent (P ≤ 0.05).

The factorial correspondence analysis plotting of theindividual scores showed that all wolves from Italy weredistinct and grouped separately from all the other wolves(Fig. 2). The Italian wolves were completely distinct fromall the other wolves on the first factorial axis (FA-I, explain-ing 6.6% of the total genetic diversity), while the wolvesfrom Spain were almost completely distinct from the other

wolves on FA-III (explaining 2.6% of the total geneticdiversity). The neighbour-joining tree of interindividualdistances (1 − ps, the proportion of shared alleles; Bowcocket al. 1994) supported the factorial correspondence analysisresults, showing that all the Italian wolves joined into aunique cluster, distinct from all the other wolf and dog

Population 1 2 3 4 5 6 7 8

1 Spain 0.32 0.17 0.14 0.20 0.35 0.13 0.162 Italy 0.22 0.25 0.23 0.29 0.33 0.21 0.293 Croatia 0.16 0.15 0.07 0.12 0.20 0.07 0.114 Greece + Bulgaria 0.19 0.11 0.07 0.12 0.20 0.05 0.105 Turkey + Israel 0.31 0.16 0.10 0.09 0.23 0.10 0.066 Saudi Arabia 0.41 0.12 0.24 0.18 0.26 0.20 0.247 Latvia 0.18 0.12 0.06 0.00 0.03 0.18 0.088 Finland 0.30 0.21 0.09 0.11 0.03 0.25 0.04

Table 4 Pairwise FST (above the diagonal)and RST (Slatkin 1995; below the diagonal)values among eight wolf populations (seeTable 3)

Fig. 2 Factorial correspondence analysis (Benzécri 1973), computedusing genetix, showing relationships among the multilocusgenotypes of individual wolves. FA-I, FA-II (a), and FA-III (b) are,respectively, the first, second and third principal factors ofvariability.



clusters (Fig. 3). The neighbour-joining identified six mainclusters in wolves: Spain, Greece + Bulgaria, Saudi Arabia,Italy, Croatia, and a mixed cluster including wolves fromLatvia, Finland and Turkey + Israel. Three individual wolvesfrom Bulgaria were also included in this cluster. However,there was no bootstrap support to any of these clusters.

structure (performed excluding all dogs and hybrids)showed that the wolf sample (n = 267) can be subdividedinto six distinct clusters (not shown). Population andindividual assignments to the clusters showed the samepatterns reported in Table 2. In particular, wolves fromLatvia, Italy and Spain were assigned to three distinctclusters with qi ≥ 0.900. The other groups were assigned togeographically admixed clusters: Turkey with Israel andFinland, Bulgaria with Greece, Croatia with Saudi Arabia.All wolves from Turkey and Italy and almost all wolvesfrom Spain were assigned to their respective clusters,while samples from the other locations were assigned tomore than one cluster.

structure, factorial correspondence analysis andneighbour-joining procedures have different resolution

powers. Factorial correspondence analysis and neighbour-joining are explorative methods, which are used to describethe extent of genetic diversification. In contrast, structureallows an identification of the number of distinct geneticclusters (based on likelihood), and a probabilistic assign-ment of the individuals to the clusters. Factorial corre-spondence analysis showed that Italian wolves form themost distinct group among all the studied samples. Resultsfrom neighbour-joining were concordant with structure,because both procedures identified seven main clusters,that is, dogs and six clusters of wolves.

Inference on past demography in the Italian wolves

Garza & Williamson (2001) suggested that reliable M-valuesrequire sample sizes ≥ 25. Cornuet & Luikart (1996) sug-gested a minimum of 15 individuals for the Wilcoxon’s testwhen 10 or more loci are available, and a minimum of 30individuals in the mode-shift test. Therefore, for the bottle-neck tests we used samples with n = 20, except for Finland(n = 13), and we excluded the captive populations. To avoid

Fig. 3 Unrooted neighbour-joining tree computed by phylip using (1 − ps) genetic distances (Bowcock et al. 1994) among individualmicrosatellite multilocus genotypes. The dog and wolf clusters, and the main recognizable wolf clusters are indicated by a thick line or byellipses (dotted). These clusters are not supported by bootstrap values larger than 50%.



biases because of unequal sample sizes, we used also 20random subsamples each one of 30 wolves from Italy.Results from each subsample were evaluated independently.

The M-test did not indicate evidence of recent bottle-necks in wolf populations, except in the Italian sample (n =103) and in each one of the random subsamples (n = 30),which showed values of average M significantly smallerthan MC (P < 0.05), using ps = 90% or 95%, and θ = 5 or 10.Wolves from Croatia and Finland showed also significantvalues of the M-test (P < 0.05) using ps = 95% and θ = 5.None of the Wilcoxon’s signed-ranks and mode-shift testswere significant, except for one replicate of Italian sub-samples (one Wilcoxon’s test with P < 0.05, and one shifteddistribution). Thus, these procedures showed instances ofpopulation bottlenecks almost only in the Italian wolves.

The MSVAR procedure was used to analyse the totalsample (n = 103) and 20 random subsamples of the Italianwolves (n = 30). We analysed also the samples collectedfrom Croatia (n = 24), Bulgaria (n = 34), Latvia (n = 38) andFinland (n = 13), a group obtained pooling all easternEuropean wolves (n = 115), and a group including onlywolves from the Balkan regions (Bulgaria, Greece, Croatia;n = 64). Five independent simulations for each sampleshowed concordant results, except in samples fromBulgaria, Latvia and Finland, which showed SD values ofthe 50% quantiles up to six times higher than in the othersamples (Table 5). Results showed that the 5% and 95%quantiles of the posterior distributions of the two demo-graphic parameters log10(r) and log10(tf ) were smaller thanthe value of the priors (Table 5). The posterior log10(r)distributions from the Italian wolf samples have 5% and95% quantiles in the negative region of the plots, indicatinga sharp population decline (Fig. 4a). The 50% log10(r)quantiles were negative either in the complete sample set

(n = 103) or in each one of the random subsamples of 30Italian wolves (Table 5). However, the sample size affectedthe posterior values: the estimates of population declinewere higher in the subsamples (50% log10(r) quantile =−3.03) than in the total Italian wolf sample (50% log10(r)quantile = −2.09) (Table 5).

Posterior values of log10(tf ) were strongly positive,indicating an ancient decline of wolves in Italy (Table 5;Fig. 4a). The sample size affected also the values of log10(tf ),indicating population declines more ancient in the sub-samples than in the total Italian sample (Table 5). A valueof r = 0.008, computed as the antilog of the 50% quantile,indicated a 125-fold decline (range = 33.3–500.0) in current(N0) to ancestral (N1) population size in Italian wolves.Population contraction started at ta = 15.8N0 (range = 5.2–57.5) generations ago.

Ciucci & Boitani (1991) suggested that 50% of the wolvesare in the breeding pool in the Italian population. Thus, theeffective size of the Italian wolf population at the mostrecent bottleneck could have been Ne = 50 individuals, andN0 = 100 (measured in chromosomes). However, geneticdata suggested lower Ne values. The observed mtDNAmonomorphism in the Italian wolf population could havebeen generated by random drift during 120–160 years ofisolation in a stable population with female effective popu-lation size Ne f as low as 15–20 (Randi et al. 2000). If adult sexratio is 1 : 1 (Ciucci & Boitani 1991), Ne f = 15–20 correspondsto Ne = 30–40 and N0 = 60–80. The median quantile of theposterior distribution of Ne (and the 5% and 95% quantiles)was Ne = 283 (79–475) using tm3.1 (with 20 000 updatesand maximum population size of 500). The maximum-likelihood estimate was Ne = 170, using mcleeps1.1 (with1000 Monte Carlo replicates and maximum populationsize of 500). These estimates were obtained sampling the

Table 5 Posterior distributions of the demographic parameters used in the MSVAR procedure (Beaumont 1999), described by the 5%, 50%,95% quantiles, with values of one standard deviation (SD) computed across five replicates

Sample (size)

Log(r) Log(tf ) r = N0/N1(range)

tf = ta/N0(range)5% 50% 95% 5% 50% 95%

Italy (103) −2.78 (0.43) −2.09 (0.29) −1.52 (0.18) +0.72 (0.13) +1.20 (0.19) +1.76 (0.29) 0.008 (0.002/0.030) 15.8 (5.2/57.5)Italy (30) −3.77 (0.24) −3.03 (0.30) −2.30 (0.26) +1.32 (0.21) +1.90 (0.25) +2.54 (0.22) 0.001 (0.000/0.005) 79.4 (20.9/346.7)East Europe (115) −1.78 (0.38) −1.29 (0.23) −0.96 (0.19) −0.64 (0.21) −0.34 (0.15) −0.07 (0.18) 0.051 (0.016/0.110) 0.46 (0.23/0.85)Balkans (64) −1.32 (0.11) −0.99 (0.10) −0.64 (0.09) −0.52 (0.16) −0.13 (0.06) +0.28 (0.04) 0.102 (0.048/0.229) 0.74 (0.30/1.90)Bulgaria (34) −3.16 (0.61) −1.98 (0.50) −1.16 (0.23) −0.07 (0.15) +0.50 (0.28) +1.34 (0.44) 0.010 (0.001/0.069) 3.16 (0.85/21.9)Croatia (24) −3.92 (0.11) −2.83 (0.22) −0.96 (1.66) +0.44 (0.13) +1.31 (0.18) +2.14 (0.11) — —Latvia (38) −3.98 (0.28) −1.16 (0.36) −1.33 (0.45) −0.17 (0.30) +0.77 (0.39) +1.98 (0.31) — —Finland (13) −4.10 (0.27) −2.79 (0.63) −1.64 (0.48) +0.24 (0.33) +1.12 (0.45) +2.12 (0.26) — —

Values for r = N0 /N1 (N0 = current effective number of chromosomes; N1 = number of chromosomes at the time of population decline or expansion), and tf = ta/N0 (ta = number of generations since the beginning of the expansion/decline) were not estimated in samples from Croatia, Latvia and Finland because of their failure ot converge. Values for the Italian wolf subsamples, Italy (30), are the average values (SD) of 20 replicated samples each one of 30 randomly selected individuals.



Italian wolf population in the 1990s. Assuming that wolvesnumbered 100 in 1973, and that the population increasedat a rate of 7% per year (Ciucci & Boitani 1991), we couldhave had c. 400 wolves in 1993 in Italy, that is Ne/N =170/400 = 0.42, as estimated using mcleeps1.1, or Ne/N = 283/400 = 0.71 (range: 79/400 = 0.20; 1.0), as estimatedusing tm3.1.

Based on these values, we use Ne = 20–40 (N0 = 40–80),and a generation time of 3 years to estimate ta. The popu-lation decline in Italian wolves might date back to 1896(range 624–6900) years ago, computed with Ne = 20, or to3794 (range 1248–13 800) years ago, computed with Ne =40. Posterior values were higher when estimated from thesubsamples, corresponding to a 1000-fold decline in N0 attime ta = 79.4N0 = 9528/19 056 years ago, computed withNe = 20/40 (Table 5).

The pooled wolf groups (eastern Europe, Balkans),which have sample sizes comparable to, or smaller thanthe total Italian wolf sample, did not show such strongsignals of population decline (Table 5). The posteriorvalues of the parameters indicate a 19.6-fold (range = 9.1–62.5), and a 9.8-fold (range = 4.4–20.8) population decline,at ta = 0.46N0 (range = 0.23–0.85) and 0.74N0 (range = 0.30–1.90) generations ago in eastern Europe and Balkan wolfgroups, respectively. Similar results were obtained from the

Bulgarian sample, which has a size similar to the Italianwolf subsamples (n = 34). The 50% quantiles of log10(r) andlog10(tf ) were lower than values computed from the Italiansubsamples, and correspond to a 100-fold population con-traction (with a very wide range = 14.5–1000), at ta = 3.16N0(range = 0.85–21.9) generations ago (Table 5). The MSVARsimulations of samples from Croatia, Latvia and Finlandapparently failed to converge and produced either unimodalor bimodal distribution of the parameters (Fig. 4b).

Discussion

Mitochondrial and microsatellite DNA data showed thatwolves in peninsular Italy have lower variability than, andare genetically distinct from, other wolf populations inEurope and elsewhere. Random drift can drive isolatedpopulations to lose gene diversity and differentiate eitherthrough protracted demographic declines and/or strongbottlenecks. In this study we aimed to evaluate the likeli-hood of two scenarios explaining population decline anddiversification across different timescales in wolves isolatedin peninsular Italy. Wolves might have been geneticallyisolated in the Apennines: (i) thousands of years ago, as aconsequence of the natural landscape changes caused bythe last Pleistocene glaciation; or (ii) more recently, as a

Fig. 4 Histograms showing the distribu-tion frequency of the posterior values of thedemographic parameters log10(r) (upperpanels), and log10(tf ) (middle panels), andthe bivariate plots of log10(tf ) vs log10(r)(lower panels), as computed by MSVAR(Beaumont 1999). (a) Italian wolves (n =103), (b) a sample of the bimodal distribu-tions that were obtained in Croatian wolves(n = 24).



consequence of the human-caused deforestation anderadication of the species from the Alps, during the lastfew centuries. Results of the coalescent analysis (Beaumont1999) of microsatellite genotypes provide evidence thatwolves in peninsular Italy underwent a strong and long-lasting decline in effective population size, thus supportingthe first scenario.

A bottleneck test based on frequency distributions ofmicrosatellite allele size (Garza & Williamson 2001) showedevidence of departure from mutation–drift equilibrium inwolves in Italy, but not elsewhere in Europe. The statisticM, which is used in this test, should achieve new equilib-rium values rapidly, i.e. within a few hundred generations(Garza & Williamson 2001). Thus, these results may describethe genetic consequences of the most recent human-causeddecline of the wolf population in the central and southernApennines during the last few centuries. In contrast, thebottleneck tests by Cornuet & Luikart (1996) did not showsignals of population decline in any wolf population. TheM- and HE-tests use different summary statistics to detectsignals of population declines, and could have differentpowers to detect bottlenecks in different time windows.Moreover, the results of bottleneck tests are sensitive tosample size, number of loci used and the microsatellitemutation patterns.

Bayesian analyses of the microsatellite data support withhigh probability a pattern of strong population decline orisolation for wolves in Italy, which was much more ancientthan a few hundred generations. Given the available dataset, the assumptions of the model and the values of the priors,results suggest that the wolf population in Italy declinedmore than 100-fold in size during the last few thousandyears. These results are incompatible with a single bottle-neck in the last century. A population contraction at ta =15.8N0 generations ago would be congruent with historicalinformation only if N0 = 2 chromosomes, corresponding toone individual at the bottleneck event (i.e. 15.8 × 2 × 3years generation time = 94.8 years). Also if it is assumedthat wolves in the Apennines were effectively isolated forthe last two to four centuries, as a result of the decliningpopulation in the Alps, we need to postulate extremely loweffective population size of Ne = 2–4 for centuries, which isextremely unrealistic. According to historical information(Zimen & Boitani 1975), the most recent bottleneck wasshort (about 10 years, corresponding to about three wolfgenerations, from the 1960s to the early 1970s), and appar-ently not strong, because approximately 100 wolves sur-vived. Census data (Ciucci & Boitani 1991) and mtDNAinference (Randi et al. 2000) suggest that a value of Ne/N = 0.2–0.4 could be appropriate for the Italian wolvesat the most recent bottleneck. However, demographic para-meters of the Italian wolf population are unknown, andreliable estimates of Ne cannot be obtained directly, usingindependent nongenetic data.

Results of coalescent simulations should be, however,interpreted with caution. We observed a sharp sample sizeeffect on the posterior values of the parameters. A sampleof n = 103 indicates a 125-fold decline in current to ances-tral population size in Italian wolves, starting at ta = 15.8N0generations ago, which may correspond to about 2000–4000 years ago. In contrast, random subsamples of n = 30indicate a 1000-fold decline in N0 at time ta = 79.4N0, corre-sponding to about 9000–19 000 years ago (Table 5). Theeffect of the sample size for MSVAR is unkown, there is nostudy showing the pattern of correlation between samplesize and posterior values of parameters. However, theresults obtained using the smaller sample sizes furthersupport a model of long-term population decline in penin-sular Italian wolves. Explaining the observed distributionof microsatellite diversity in the current wolf population inItaly in terms of the most recent isolation (about one century)and bottleneck (about 100 individuals for about three gen-erations), is certainly less likely than alternative scenariossuggesting more ancient isolation and/or bottlenecks.

Wakeley (2000) showed that population subdivision intoa large number of demes connected by migration increasesthe coalescence timescale of a genealogy. Coalescence timesdepend on the effective population size, which is affectedby migration in fragmented populations. When migrationrates are large, the population is effectively panmictic, andthe history of the samples conforms to the standard single-population coalescent. But, when migration rates are small,the coalescence times are much longer in fragmented thanin panmictic populations of similar total effective size (Rayet al. 2003).

The posterior values of the demographic parameterssupport estimates of population declines of about 10- to100-fold in eastern European and Balkan wolf groups, andin the Bulgarian sample, from ta = 0.46N0 to ta = 3.16N0generations ago. Assuming that the effective populationsize in eastern European wolves is at least one order ofmagnitude larger than in Italy, that is N0 = 1000, populationdeclines might date back to 1380 (in the eastern Europeangroup) or 9480 (in the Bulgarian wolves) years ago. How-ever, these results may be biased by model assumptions.Beaumont’s procedure assumes a closed unstructuredpopulation, which probably does not hold for the pooledgroups (eastern Europe, Balkans), or for the Bulgarianwolves. Structured or open populations may show apattern of decline simply because the allele coalescenttimes are longer in fragmented populations with migration(Wakeley 2000; Ray et al. 2003). Population structure andmigration would predictably inflate the estimated ratesand times of population decline in Bulgarian and easternEuropean wolves. These conclusions support a scenarioof comparatively stronger long-term population declinein populations from peninsular Italy than in other wolfpopulations in Europe. A signal of population decline



generated by ancient geographical substructuring, arisingfrom restricted migration, still lends support to a model ofancient isolation of the Italian wolves south of the Alps,also if the population has been demographically stable inthe last thousands of generations.

A scenario of long-lasting isolation of wolves in the ItalianApennines is supported by multivariate morphometricdata, showing that skulls from Italian Apennine wolves aresharply distinct from other Eurasian wolves (Nowak &Federoff 2002). To stress the importance of the observedmorphological diversification, Nowak & Federoff con-firmed a distinct subspecific rank for wolves in peninsularItaly (Canis lupus italicus). This morphometric data setshowed also that wolf and dog skulls do not overlap in themultivariate space, thus showing no apparent instanceof hybridization. Morphometric and genetic resultsconcordantly confirmed that the occurrence of wolf × dogcross-breeding is not frequent in Europe (Vilà & Wayne1999; Randi et al. 2000).

Conclusions

Wolves in the Apennines could have been, at least partially,genetically isolated from any other wolf population inEurope for some thousands of years, and not just for afew decades, as suggested by information on the species’historical distribution range. The Alpine ice caps at the lastglacial maximum might have provided a geographicalbarrier that isolated wolves in refuge areas south of theAlps. Deglaciation and the expansion of extant ecosystemswere completed only after the Younger Dryas cold spell(c. 10 000 years ago; Dawson 1996). Thus, the admixture ofwolf populations expanding from different glacial refugescould have been relatively recent. Moreover, the Po River,which cut the plain from the western Alps to the AdriaticSea, was much more expanded during the last glaciation,because of the lower sea level and the presence of a northAdriatic land-bridge (Fig. 1; Dawson 1996). For thousandsof years in the Holocene, the Po River basin was flanked byextensive flooded alluvial plains and marshes, which werepartially drained only in the last 2000 years (Sereni 1961).Admixture of Alpine and Apennine wolf populationscould have been prevented also by deforestation and theconcomitant eradication of wild ungulate populations,which were already widespread during the fifteenthcentury in northern Italy as a result of expanding share-cropping agricultural systems (Sereni 1961).

The results of this study suggest that, despite the highpotential rates of dispersal and gene flow, local wolfpopulations may not mix for long periods of time. Wolvesfrom the Apennines are currently expanding, recolonizingparts of their historical range in the western Italian andFrench Alps (Lucchini et al. 2002). Meanwhile, from theeast, wolves with distinct genotypes are moving from

Slovenia towards the Italian border in the eastern Alps. Itwill be interesting to observe if and how wolves expandingfrom the west (bearing Apennine genotypes) and from theeast (with Balkan genotypes) will mix during the ongoingprocess of natural recolonization of the Alps.

Acknowledgements

We deeply appreciated the collaboration of everybody who kindlyprovided wolf and dog samples, and in particular: L. Boitani, P.Ciucci, G. Dolf, F. Francisci, V. Guberti, A. De Marco and C. Vilà.M. Bolognini kindly provided information on methods to age wolfteeth. We thank L. Boitani and C. Vilà for constructive discussionon conservation and genetics of wolves in Europe, and M.Beaumont for advice in the use of Bayesian models in populationgenetic analyses. Comments by the subject editor and twoanonymous referees stimulated us to focus better the meaningof the results and improve this study. The Italian Ministry ofEnvironment (Nature Conservation), the Italian Ministry ofForeign Affairs and the Istituto Nazionale per la Fauna Selvaticasupported this project.

References

Anderson EC, Williamson EG, Thompson EA (2000) Monte Carloevaluation of the likelihood for Ne from temporally-spaced data.Genetics, 156, 2109–2118.

Andersone Z, Lucchini V, Randi E, Ozolins J (2002) Relationshipsbetween wolves and dogs in Latvia as documented using mito-chondrial and microsatellite DNA markers. Mammalian Biology,67, 79–90.

Ballard WE, Matson GM, Krausman PR (1995) Comparison oftwo methods to age gray wolf teeth. In: (eds Carbyn LN, Fritts SH,Seip DR), Ecology and Conservation of Wolves in a Changing World.Occasional Publications, no. 35, pp. 455–460. Canadian CircumpolarInstitute, University of Alberta Edmonton, Alberta, Canada.

Barnes I, Matheus P, Shapiro B, Jensen D, Cooper A (2002)Dynamics of Pleistocene population extinctions in Beringianbrown bears. Science, 295, 2267–2270.

Beaumont MA (1999) Detecting population expansion and declineusing microsatellites. Genetics, 153, 2013–2029.

Belkhir K, Borsa P, Chikhi L, Raufaste N, Bonhomme F (1996–2001) GENETIX 4.04 Logiciel sous Windows TM pour la Génétique desPopulations. Laboratoire Génome, Populations, Interactions: CNRSUMR. 5000. Université de Montpellier II, Montpellier.

Benzécri JP (1973) L’analyse Des Donnèes: T. 2, L’analyse Des Corre-spondances. Dunoud, Paris, France.

Berthier P, Beaumont MA, Cornuet J-M, Luikart G (2002) Likelihood-based estimation of the effective population size using temporalchanges in allele frequencies: a genealogical approach. Genetics,160, 741–752.

Boitani L (1984) Genetic considerations on wolf conservation inItaly. Bollettino Di Zoologia, 51, 367–373.

Boitani L (1999) Final Draft Action Plan for the Conservation of Wolves(Canis lupus) in Europe. Council of Europe, Strasbourg.

Bowcock AM, Ruiz-Linares A, Tomfohrde J, Minch E, Kidd JR,Cavalli-Sforza LL (1994) High resolution human evolutionarytrees with polymorphic microsatellites. Nature, 368, 455–457.

Breitenmoser U (1998) Large predators in the Alps: the fall and riseof man’s competitors. Biological Conservation, 83, 279–289.



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

Ciucci P, Boitani L (1991) Viability assessment of the Italian wolfand guidelines for the management of the wild and a captivepopulation. Ricerche di Biologia della Selvaggina. Istituto Nazionaledi Biologia della Selvaggina, Ozzano Dell’emilia (Bologna), 89, 1–58.

Cornuet JM, Luikart G (1996) Description and power analysis oftwo tests for detecting recent population bottlenecks from allelefrequency data. Genetics, 144, 2001–2014.

Dawson AG (1996) Ice Age Earth. Late Quaternary Geology and Cli-mate. Routledge, New York.

Delibes M (1990) Status and conservation needs of the wolf in theCouncil of Europe member States. Nature and EnvironmentSeries, Strasbourg, 47, 1–46.

Di Rienzo A, Peterson AC, Garza JC, Valdes Slatkin M, Freimer NB(1994) Mutational processes of simple-sequence repeat loci inhuman populations. Proceedings of the National Academy of Sci-ences of the USA, 91, 3166–3170.

Flagstad O, Walker CW, Vilà C et al. (2003) Two centuries of theScandinavian wolf population: patterns of genetic variabilityand migration during an era of dramatic decline. MolecularEcology, 12, 869–880.

Forbes SH, Boyd DK (1996) Genetic variation of naturally colon-izing wolves in the central Rocky Mountains. ConservationBiology, 10, 1082–1090.

Fritts SH (1993) Record dispersal by a wolf from Minnesota.Journal of Mammology, 64, 166–167.

Garza JC, Williamson EG (2001) Detection of reduction in popula-tion size using data from microsatellite loci. Molecular Ecology,10, 305–318.

Gerloff U, Schlotterer C, Rassmann K et al. (1995) Amplification ofhypervariable simple sequence repeats (microsatellites) fromexcremental DNA of wild living Bonobos (Pan paniscus).Molecular Ecology, 4, 515–518.

Goudet J (1995) fstat, Version 1.2. A computer program tocalculate F-statistics. Journal of Heredity, 86, 485–486.

Goudet F, Raymond M, de Meeus T, Rousset F (1996) Testingdifferentiation in diploid populations. Genetics, 144, 1933–1940.

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

Kimura M, Ohta T (1978) Stepwise mutation model and distribu-tion of allelic frequencies in a finite population. Proceedings of theNational Academy of Science of the USA, 75, 2868–2872.

Legendre P, Legendre L (1998) Numerical Ecology, 2nd edn.Elsevier, Amsterdam.

Leonard JA, Wayne RK, Cooper A (2000) Population genetics ofIce Age brown bears. Proceedings of the National Academy ofSciences, USA, 97, 1651–1654.

Lucchini V, Fabbri E, Marucco F, Ricci S, Boitani L, Randi E (2002)Non-invasive molecular tracking of colonizing wolf (Canislupus) packs in the western Italian Alps. Molecular Ecology, 11,857–868.

Mech D (1970) The Wolf: the Ecology and Behavior of an EndangeredSpecies. University of Minnesota Press, Minneapolis, MN.

Michalakis Y, Excoffier L (1996) A generic estimation of popula-tion subdivision using distance between alleles with specialreference to microsatellite loci. Genetics, 142, 1061–1064.

Miller SA, Dykes DD, Poleski HF (1988) A simple salting outprocedure for extracting DNA from human nucleated cells.Nucleic Acid Research, 16, 12–15.

Minch E, Ruiz-Linares A, Goldstein DB, Feldman MW, Cavalli-Sforza LL (1997) microsat: A computer program for calculatingvarious statistics on microsatellite allele data. Stanford University,Palo Alto, CA.

Neff MW, Broman KW, Mellersh CS et al. (1999) A second-generation linkage map of the domestic dog, Canis familiaris.Genetics, 151, 803–820.

Nowak RM, Federoff NE (2002) The systematic status of the Italianwolf Canis lupus. Acta Theriologica, 47, 333–338.

Piry S, Luikart G, Cornuet JM (1999) bottleneck: a computerprogram for detecting recent reductions in the effectivepopulation size using allele frequency data. Journal of Heredity,90, 502–503.

Pritchard JK, Stephens M, Donnelly PJ (2000) Inference of popula-tion structure using multilocus genotype data. Genetics, 155,945–959.

Randi E, Lucchini V (2002) Detecting rare introgression ofdomestic dog genes into wild wolf (Canis lupus) populationsby Bayesian admixture analyses of microsatellite variation.Conservation Genetics, 3, 31–45.

Randi E, Lucchini V, Christensen MF et al. (2000) MitochondrialDNA variability in Italian and east European wolves: detectingthe consequences of small population size and hybridization.Conservation Biology, 14, 464–473.

Ray N, Currat M, Excoffier L (2003) Intra-deme molecular diver-sity in spatially expanding populations. Molecular Biology andEvolution, 20, 76–86.

Raymond M, Rousset F (1995) genepop Version 1.2.: populationgenetics software for exact tests and ecumenicism. Journal ofHeredity, 86, 248–249.

Rosenberg NA, Nordborg M (2002) Genealogical trees, coalescenttheory and the analysis of genetic polymorphisms. NatureReviews. Genetics, 3, 380–390.

Saitou N, Nei M (1987) The neighbor-joining method: a newmethod for reconstructing phylogenetic trees. Molecular Biologyand Evolution, 4, 406–425.

Schneider S, Roessli D, Excoffier L (2002) ARLEQUIN version 2.0.2. ASoftware for Population Genetics Data Analysis. Genetics andBiometry Laboratory, Department of Anthropology andEcology, University of Geneva, Geneva, Switzerland.

Sereni E (1961) Storia Del Paesaggio Agrario Italiano. Laterza, Bari.Slatkin M (1995) A measure of population subdivision based on

microsatellite allele frequencies. Genetics, 139, 457–462.Valière N, Fumagalli L, Gielly L et al. (2003) Long-distance wolf

recolonization of France and Switzerland inferred fromnon-invasive genetic sampling over a period of 10 years. AnimalConservation, 6, 83–92.

Vilà C, Wayne RK (1999) Hybridization between wolves and dogs.Conservation Biology, 13, 195–198.

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

Vilà C, Walker C, Sundqvist A-K et al. (2003) Combined use ofmaternal, paternal and bi-parental genetic markers for the iden-tification of wolf-dog hybrids. Heredity, 90, 17–24.

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

Waits LP, Luikart G, Taberlet P (2001) Estimating the probabilityof identity among genotypes in natural populations: cautionsand guidelines. Molecular Ecology, 10, 249–256.



Wakeley J (2000) The effects of subdivision on the geneticdivergence of populations and species. Evolution, 54, 1092–1101.

Walton LR, Cluff HD, Paquet PC, Ramsay MA (2001) Movementpatterns of barren-ground wolves in the central CanadianArctic. Journal of Mammology, 82, 867–876.

Waples RS (1989) A generalized approach for estimating effectivepopulation size from temporal changes in allele frequency.Genetics, 121, 379–391.

Wayne RK, Lehman N, Allard MW, Honeycutt RL (1992)Mitochondrial DNA variability of the gray wolf: genetic con-sequences of population decline and habitat fragmentation.Conservation Biology, 6, 559–569.

Weir BS, Cockerham CC (1984) Estimating F-statistics for theanalysis of population structure. Evolution, 38, 1358–1370.

Wilson IJ, Balding DJ (1998) Genealogical inference from micro-satellite data. Genetics, 150, 499–510.

Zimen E, Boitani L (1975) Number and distribution of wolves inItaly. Zeitschrift Fur Säugetierkunde, 40, 102–112.

This study is part of a long-term project on the population andconservation genetics of wolves in Italy. E. Randi is head ofconservation biology and genetics at INFS. V. Lucchini, researcherat INFS, is involved in some projects on conservation and forensicgenetics in a number of avian and mammalian species. A. Galovis a researcher at the Univesity of Zagreb, and is working on thepopulation and conservation genetics of wild and domesticatedmammals.

Lucchini, V., Galov, A., and Randi, E. (2004). Evidence of ... · the wolf (Canis lupus) population...

Documents

Transcript of Lucchini, V., Galov, A., and Randi, E. (2004). Evidence of ... · the wolf (Canis lupus) population...