Molecular Ecology (2002)

11

, 1285–1297

© 2002 Blackwell Science Ltd

Blackwell Science, Ltd

The genetic structure of natural and reintroduced roe deer (

Capreolus capreolus

) populations in the Alps and central Italy, with reference to the mitochondrial DNA phylogeography of Europe

C . VERNESI ,

*†

E . PECCHIOLI ,

†‡

D . CARAMELLI ,

†§

R . T IEDEMANN,

¶

E . RANDI

* *

and G. BERTORELLE

*

*

Dipartimento di Biologia, Università di Ferrara, Italy,

†

‘Biosfera — Biologia della Conservazione Ricerca e Didattica’, Firenze, Italy,

‡

Centro di Ecologia Alpina, Viote di Monte Bondone, Italy,

§

Dipartimento di Biologia Animale e Genetica, Università di Firenze, Italy,

¶

Institute of Biochemistry and Biology, University of Potsdam, Germany,

**

Istituto Nazionale per la Fauna Selvatica, Ozzano dell’Emilia, Italy

Abstract

The first hypervariable fragment (HVI) of the mitochondrial DNA control region wassequenced in 90 individuals of the European roe deer (

Capreolus capreolus

) from the Alps,central Italy and Spain. Pooling these data with 70 published sequences from several Euro-pean regions, we were able to identify patterns of divergence within the Italian peninsula,and in Europe in general. The results we obtained can be summarized as follows. First, thegenetic structure of European roe deer populations is substantial (

ΦΦΦΦ

ST

values around 0.6).Second, the divergence between some central Italian populations, the Alpine group (whichis genetically close to the French, the Spanish and the Norwegian samples) and the EasternEuropean populations seems to reflect Upper Pleistocene subdivisions, possibly related tothree southern European refugia. Third, a different group of central Italian individualsprobably diverged more recently from the Alpine group, and their attribution to the sub-species

C. c. italicus

does not appear justified. Fourth, the analysis of mitochondrial DNAin the roe deer can be used to identify recently reintroduced animals in the western Alpswhich clearly cluster within the Eastern European group, thus providing an important toolfor conservation and management strategies for this species.

Keywords

:

Genetic diversity, glaciation, mtDNA, population divergence, reintroduction, roe deer

Received 14 October 2001; revision received 10 April 2002; accepted 10 April 2002

Introduction

The roe deer is widespread in the Palaearctic and incontinental Asia, and includes two polytypic species(Grubb 1993; Danilkin 1996): the European roe deer(

Capreolus capreolus

) and the Siberian roe deer (

C. pygargus

).The two species have different body sizes, morphometrictraits and karyotypes (Sokolov & Gromov 1990; Danilkin1996). Mitochondrial DNA (mtDNA) sequences and theavailable fossil record suggest that Siberian and Europeanroe deer separated at the Pliocene/Pleistocene boundary

and have evolved independently for about 2–3 millionyears (Randi

et al

. 1998).The evolutionary history of the European roe deer dur-

ing the last 2 or 3 million years is not known. As for severalother animal and plant species in Europe, however, the lastPleistocene glaciation probably had a major impact on itspopulation history (Taberlet

et al

. 1998; Hewitt 1999). Ana-lysing the mtDNA variability in 40 individuals, Wiehler &Tiedemann (1998) suggested that the roe deer retreatedduring the last glacial period to a single western Mediter-ranean refugium, and subsequently colonized the contin-ent from west to east. Central-southern Italy representeda possible additional refugium during the last glaciation,and some isolated populations now found in this area,described as the endemic subspecies

Capreolus capreolus

Correspondence: G. Bertorelle, Dipartimento di Biologia,Università di Ferrara, Via Borsari 46, 44100 Ferrara, Italy. Fax:+ 39 0532249761; E-mail: ggb@unife.it

MEC_1534.fm Page 1285 Thursday, July 18, 2002 7:53 AM

1286

C . V E R N E S I

E T A L .

© 2002 Blackwell Science Ltd,

Molecular Ecology

, 11, 1285–1297

italicus

(Festa 1925), could represent the direct descendentsof populations from that refugium (Randi

et al

. 1998).In recent times, roe deer populations in the southwest of

Europe have been affected by habitat fragmentation andrestocking for hunting purposes. In Italy the roe deer,which was distributed in a nearly continuous pattern inplains, hills and mountain forests, was almost completelyeradicated in the Alps and in most of the northern and cen-tral Apennines by the mid-1900s. Overhunting and habitatchange to cultivated land were the main factors threaten-ing roe deer populations in Italy and western Europe(Perco & Calò 1994). Small remnant populations of puta-tively native roe deer survived in a few localities in theeastern Alps and in central-southern Italy.

During the last 30 years, roe deer have been reintro-duced in the western Italian Alps and the Apennines,mainly using stocks from the eastern Italian Alps, centralEurope and the Balkans (Lorenzini

et al

. 1997; Randi

et al

.1998). It is not always clear, however, if present day popu-lations descend only from reintroduced animals, from localancestors, or both. In contrast, roe deer in the eastern ItalianAlps seems to represent a native Alpine population. In thisarea, expansion of local populations and/or immigrationfrom neighbouring countries (Austria, Slovenia) contrib-uted to the establishment of a large population withoutintroductions.

Habitat fragmentation due to natural processes orhuman activities, together with restocking using distantlyrelated source groups, usually results in divergence amonglocal populations. Morphological, ethological and ecolo-gical variation has been observed among roe deer groups(Sokolov & Gromov 1990; Danilkin 1996), but the extent ofgenetic diversification and phylogeographical structuringis not known. Electrophoretic markers (Hartl & Reimoser1988; Lorenzini

et al

. 1997) and mtDNA data (Randi

et al

.1998; Wiehler & Tiedemann 1998) suggest some degree ofgenetic divergence between populations, but only smallsamples or limited geographical areas have been analysedso far.

In this paper we analyse the patterns of genetic varia-bility and differentiation in the largest available sampleof mtDNA sequences of the European roe deer, compiledby integrating published data with new sequences fromItalian and Spanish populations. From an evolutionary per-spective, we aim to understand the history of this species inEurope and the phylogenetic position of the populations ofthe Alps and central Italy, including the group identified asthe subspecies

C. c. italicus

. We also consider some practicalimplications of this analysis, such as the possibility of iden-tifying the impact of human hunting and reintroductionactivities on genetic variability and its geographical distri-bution. These implications are of primary interest whenimplementing conservation and management strategies(Caughley & Gunn 1996; King & Burke 2000).

Materials and methods

Samples

A total of 90 individuals were sampled and sequencedin this study. Tissue samples were collected by huntersduring the hunting season and kept at

−

20

°

C until used.We integrated our data with 70 sequences that werealready available (Randi

et al

. 1998; Wiehler & Tiedemann1998). Therefore, the data set analysed consists of 160sequences, with geographical origins shown in Fig. 1. Non-Italian samples, identified by country names and withsample sizes and sampling sites in parenthesis, were:Norway (five; Lier), Poland (five; Czempin), France (five;Vosges), Slovakia (five; Sitno), Hungary (one; Szentendre),Bulgaria (five; Silistra, Shume, Gabrovo, Sliven), Slovenia(five; Grosuplje), Austria (five; Steiermark) and Spain (11;Segovia, Toledo, Extremadura, Burgos, Cadiz). The Italiansamples are from the following areas: Liguria (17), Sondrio(two), Asiago (four), Monte Bondone (five), Trento (11),Tarvisio (five), Trieste (one), Siena (15), Florence (19),Castelporziano (17) and Arezzo (17).

DNA extraction, amplification and sequencing

DNA was extracted from hair (five or six roots per indi-vidual) or tissue samples (about 1–2 g) incubated overnightat 37

°

C in 500

µ

L of extraction buffer (Tris–HCl pH 8 10 m

m

,ethylenediaminetetraacetic acid 2 m

m

) with 0.125 mg ofProteinase K. Tissue remains were removed by centrifuga-tion and the DNA was extracted from the supernatant witha standard phenol/chloroform protocol (Sambrook

et al

.1989). DNA was purified and concentrated by centrifugaldialysis using Microcon-30 microconcentrators (Amicon).

The primer pair L-Pro/H-16493 (5’ CGTCAGTCTCAC-CATCAACCCCCAAAGC 3’, 5’ TGAGATGGCCCTGAA-GAAAGAACC 3’; Douzery and Randi 1997) was used toamplify, by polymerase chain reaction (PCR), one fragmentof approximately 400 base pairs belonging to the firsthypervariable domain of the mtDNA control region (D-loop). Amplification reactions were performed in a 20-

µ

Lvolume containing Tris–HCl, pH 8.8, 67 m

m

(NH

4

)

2

SO

4

16 m

m

, Tween-20® 0.01% (v/v), 1.5 m

m

MgCl

2

, 200

µ

m

ofeach dNTP, 25 pmol of each primer, 1 U

Taq

DNA Poly-merase (Polymed) and 2

µ

L template DNA. After an initialstep of 94

°

C for 3 min, 35 cycles of amplification wereperformed in a programmable thermal cycler (MJ ResearchPTC 150); each cycle being 94

°

C for 15 s, 55

°

C for 1 minand 72

°

C for 1 min.The amplified DNA product was separated in a 2% low-

melting agarose gel (Boehringer Mannheim Corporation).The appropriate band was excised and after elution (1 : 25)with distilled water, 1.5

µ

L of product was used in a secondamplification using the same primers. The reaction consisted

MEC_1534.fm Page 1286 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R

1287

© 2002 Blackwell Science Ltd,

Molecular Ecology

, 11, 1285–1297

of 30 cycles as follows: denaturation at 94

°

C for 45 s,annealing at 59

°

C for 45 s, and extension at 72

°

C for 45 s.Double-stranded products were purified by centrifuga-

tion with Sepharose gel CL-6B (Pharmacia AB) andsequenced using an automated ABI Prism 377 DNAsequencer (Perkin Elmer Corporation). The sequencingamplification was carried out in a 20-

µ

L volume reaction:8

µ

L of Big Dye Terminator mix (Perkin Elmer Corpora-tion), 90 ng of purified PCR product, 3.2 pmol of H-16493primer and 3.8

µ

L of distilled water. The cycle-sequencingprogramme consisted of a step of 94

°

C for 3 min and 25cycles at 94

°

C for 10 s, 56

°

C for 5 s, 60

°

C for 4 min. Thedye-labelled PCR products were purified with ethanol/sodium precipitation. Sequences were aligned using

Clus-tal X

(Thompson

et al

. 1994) and rechecked by eye.

Data analysis

Genetic relationships among sequences were summarizedby a statistical parsimony cladogram (Templeton

et al

. 1992).Recurrent mutations are not uncommon in the hypervari-able region of mtDNA, and the reconstruction of a single mostlikely or most parsimonious evolutionary tree is often impos-sible. Several methods have been developed to represent

different trees in a single phylogenetic network (Huber

et al

.2001). Using different criteria, each method tries to resolvelikely parallel events, but also retains character conflicts (inthe form of reticulations) when ambiguity remains. Majorgenetic clusters, as well as possible areas of ambiguity, cantherefore be identified. The statistical parsimony method(Templeton

et al

. 1992) joins all pairs of haplotypes havinga probability of parsimony (i.e. the probability of having nounobserved mutations) greater than 0.95.

The genetic divergence between populations wasanalysed by several different methods. The analysis ofmolecular variance (

amova

, Excoffier

et al

. 1992) subdividesthe genetic diversity into hierarchical components andestimates the indices

Φ

, which are molecular equivalents ofWright’s

F

statistics (Wright 1951). The statistical signi-ficance of variance components and

Φ

indices is evaluatedby randomization (see Excoffier

et al

. 1992). Pairwisecomparisons between all populations were also performedusing

Φ

ST

and the average net number of substitutions (Nei1987). The average net number of substitutions is a geneticdistance specific for population divergence events, becauseit considers the average number of pairwise substitutionsbetween groups purged by the estimated number of sub-stitutions occurring in the common ancestral group.

Fig. 1 Location map of the samples. Filled squares indicate the sampling sites (listed in the text). Pies represent the frequency of differentmtDNA clusters (see inset). The size of each pie is proportional to the sample size. The southern limit of the permafrost at the end of thelast ice age (18 000 before the present) is indicated by the dotted line. Individuals from Siena, Arezzo, Florence, Trento, Hungary and Spainwere typed in this study. Individuals from Sondrio, Monte Bondone, Asiago and Trieste were typed by Randi et al. (1998). Individuals fromTarvisio, Austria, Slovenia, France, Bulgaria, Norway, Slovakia and Poland were typed by Wiehler & Tiedemann (1998). Finally, someindividuals from Castelporziano and some from Liguria were typed in this study, and the others by Randi et al. (1998).

MEC_1534.fm Page 1287 Thursday, July 18, 2002 7:53 AM

1288

C . V E R N E S I

E T A L .

© 2002 Blackwell Science Ltd,

Molecular Ecology

, 11, 1285–1297

A neighbour-joining (Saitou & Nei 1987) population treewas reconstructed using the matrix of genetic distancesbetween populations. The effect of the sampling variance(probably very high in some samples with limited size) onthe tree topology was analysed following a bootstrap tech-nique. One thousand pseudo-data sets were generated byrandomly resampling individuals (with replacement) withineach sample. In this way, 1000 trees were reconstructed,and only the consensus tree with the bootstrap values wasreported.

Genetic distances between mtDNA haplotypes wereestimated assuming a Kimura two-parameter model, withdeletions weighted as transversions and the weightsestimated from the relative proportion of different muta-tions. The well-known heterogeneity of mutation ratesacross mtDNA sites was taken into account assuming agamma distribution of rates (Hasegawa

et al

. 1993), withparameter

α

= 0.28. The value for

α

was estimated by max-imum likelihood (Yang 1994), starting from a phylogenetictree reconstructed from the hypervariable region of themtDNA in eight cervids (data in Douzery & Randi 1997).

Divergence times between groups were estimated usingthe method proposed by Gaggiotti & Excoffier (2000). Thismethod, based on the net number of substitutions betweenpopulations, is able to remove the possible effects of bottle-necks and unequal population sizes. For the translation ofmutational units into years, we omitted the sites withindels and we assumed minimum and maximum substitu-tion rates per site per million of years of 0.04 and 0.08,respectively. These values bracket several calibrated esti-mates for the most variable mtDNA regions in mammals(Randi

et al

. 1998).Finally, two neutrality tests, Tajima’s D (Tajima 1989)

and Fu’s

F

S

(Fu 1997), were applied. Under the assumptionof neutrality, these tests can identify the effects of demo-graphic changes. Statistical significance is evaluated simu-lating random samples using the coalescent approach(Hudson 1990; Schneider

et al

. 2000). For both statistics, ademographic expansion produces large negative values.The

F

S

test seems to be particularly sensitive to this phe-nomenon (Fu 1997).

Several computer packages have been used for the com-putations:

tcs

(Clement

et al

. 2000),

arlequin

(Schneider

et al

. 2000),

Phylip

(Felsenstein 1989), and

paml

(Yang 1997).In addition, a program was written to resample individualswithin localities and compute bootstrapped distancesbetween populations.

Results

Sequence analysis

A total of 49 different haplotypes (see Table 1) and 41polymorphic sites were found among the 160 mtDNA

sequences considered (length of the consensus region:342 base pairs), with 41 substitutions (35 transitions, sixtransversions) and four indels.

The statistical parsimony network contains some loopsbetween haplotypes (Fig. 2), indicating that recurrentmutations occurred at some nucleotide sites. Nevertheless,this analysis suggests the presence of a strong phylogeo-graphic structure in the roe deer. Two major groups ofsequences separated by at least three substitutions can beidentified: cluster A, mainly associated with all westernand central European populations, and cluster B, typical ofthe easternmost populations. On the basis of their geo-graphical origin and their genetic similarity, haplotypes incluster A can be further subdivided in three subclusters,called A1, A2 and A3.

The network group A1, which is identified by its star-like shape, is geographically associated with the Florenceand Arezzo areas. Almost all (10/11) haplotypes in thiscluster are found in that region (eight of them occur onlythere), and all 36 individuals sampled had sequences clas-sified as cluster A1. None of the remaining populations orregions in Europe analysed here is as clearly characterizedby cluster A1.

Cluster A2, which is directly connected through a tran-sition and an insertion to the haplotype most frequentlyfound in the Alps (haplotype 16 in Fig. 2), correspondsalmost perfectly to the Siena and the Castelporzianoreserve samples. This group of sequences is importantbecause it includes all 17 individuals from the Castelpor-ziano reserve, usually classified as the subspecies

Capreoluscapreolus italicus

, but also 12 out of 15 individuals in theSiena sample.

Cluster A3 is the most geographically widespread groupof sequences in our sample. The reticulations within thiscluster prevent the identification of a clear internal structure,but it appears evident that several haplotypes in this cluster,separated only by a few mutations, are found in areas as farapart as Spain, France, Norway, Poland and the Alps. Allthe sequences sampled in Trento, Monte Bondone, Sondrio,Asiago, France, Norway, Spain, and four out of five in Poland,comprising a total of 35 individuals, cluster together ingroup A3.

Finally, cluster B includes 13 out of 16 sequences sampledin the more eastern countries of Slovenia, Slovakia, Bulgariaand Hungary. Five other individuals within this clustercome from Austria (three), Tarvisio (one) and Poland (one).Interestingly, all the 17 sequences sampled in the Liguriaarea, where reintroductions have been extensive anddocumented, fall in cluster B. Ligurian sequences (indicatedby an asterisk in Fig. 2) constitute a group of privatehaplotypes, well within cluster B but not sampled in otherareas.

The evolutionary relationship between groups A1, A3and B cannot be identified (whereas A2 is clearly related to

MEC_1534.fm Page 1288 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R

1289

© 2002 Blackwell Science Ltd,

Molecular Ecology

, 11, 1285–1297

Table 1

List of haplotypes and segregating sites

1111111111111111111111111111111111111111155555555556666666666666666666666666666666888999999900000000000001111111111222222229990134788344445555678801124667790112223356740755596047847899923006035867502658904

1 1

CTAATGTGTGAACTATAATTTAAATTATCAATTGCTCAAAT

Florence2 13

...............-...................C.....

Florence3 1

..........G...............G........C.....

Florence4 1

..................................TC.....

Florence5 2

G..................................C.....

Florence6 1

.................TG................C.....

Florence7 15

...................................C.....

Arezzo(11), Austria(2), Tarvisio(2)8 4

........C..........................C.....

Arezzo(3), Trieste(1)9 1

.................................A.C....G

Arezzo10 2

.................................A.C.....

Arezzo11 27

......-A.............G.....C.......C.....

Castelporziano(17), Siena(10)12 1

......-A...........C.G.....C.......C.....

Siena13 1

.....................G.............C.....

Siena14 3

.............G.......G.............C.....

Siena(2), Spain(1)15 1

......-A.............G.....C.......C..GG.

Siena16 18

.....................G.....C.......C.....

Trento(10), Asiago(4), Montebondone(3), Sondrio(1)17 1

G....................G.....C......––T....

Trento18 2

...........G..G..G...G..C..........C.....

Montebondone19 6

....C..A......G............C.......C.....C

Slovenia(3), Austria(2), Bulgaria(1)20 1

...GC..A......G..........C.C.......C.G...

Liguria21 5

....CA.A....T.G........GC..C.......C.G...

Liguria22 1

....C..A......T.........C..C.......C.G...

Liguria23 6

....C..A......G............C.......C.G...

Liguria24 1

.C..CA.A....T.G........GC..C.......C.G...

Liguria25 1

..G.C..A......T.........C..C.......C.G...

Liguria26 1

....C..A......T.........C..C....C..C.G...

Liguria27 1

....C..A......G............C....C..C.G...

Liguria28 9

..............G......G.....C.......C.....C Trieste(1), Norway(5), France(1), Spain(2)29 1 ....C..A......G....................C..... Bulgaria30 4 ..............G......G..C..........C..... Poland(3), Slovakia(1)31 2 ...................................C.G... Tarvisio(1), Slovakia(1)32 3 ....C..A......G....................C.G... Slovakia33 2 ....C..A......G.........C..C.......C.G... Austria(1), Tarvisio (1)34 2 ...GC..A......G............C.....C.C..... Bulgaria35 1 .......A......G......G..C..........C..... France36 1 ..............G......GG.C..........C..... France37 1 .......A......G.T...CG.............C..... France38 1 ............T.G......G.............C..... France39 1 ..............G......G..C..........C.G... Tarvisio40 1 ....C..A...G..G.......G....C.......C.G... Slovenia41 1 ............T........G.....C.......C..... Slovenia42 1 ....C..ACA...CG....................C..... Poland43 1 ..............G......G.....C.......C.G... Poland44 1 ...GC..A......G............CT....C.C..... Bulgaria45 1 ....C..A......G.......G....C.......C.G... Hungary46 1 .......A......G.....C........GG....C..... Spain47 1 .......A......G.....C...C..C..G....C..... Spain48 5 .......A......G.........C..C.G.....C..... Spain49 1 .......A......G.....C...C....GG....C..... Spain

The numbering of nucleotide sites refers to the Bos taurus sequence (GenBank, Accession No. J01394). The absolute frequency of each haplotype is indicated in the second column. The last column reports the geographical origin and the frequency distribution of the haplotypes.

MEC_1534.fm Page 1289 Thursday, July 18, 2002 7:53 AM

1290 C . V E R N E S I E T A L .

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

A3). The robustness of these four groups of sequences,however, is confirmed by different tree or networkreconstruction algorithms (maximum likelihood, maximumparsimony, neighbour-joining and the reduced mediannetwork). These results are not reported, since all the treesthey generate are included in the network presented inFig. 2.

Population analysis

Areas or localities with a sample size equal to or greaterthan four have been considered in this analysis. Thesamples of Sondrio, Trieste and Hungary were thereforeexcluded, and the data set was reduced to 156 individualsand 17 localities.

Genetic diversity. No genetic diversity was observed in theNorwegian, Asiago and Castelporziano samples (Table 2).For Norway and Asiago, however, the sample size wasvery small (five and four individuals, respectively), and theprobability of finding a monomorphic sample of this size isnot negligible even when the population is polymorphic. Ifwe assume for example that the allelic counts have a simplemultinomial distribution, and that the frequency of themost common allele in the population is 0.6, between 8and 9% of the samples of size five will be monomorphic.On the other hand, the 17 identical sequences found in the

Castelporziano reserve more clearly suggest that thispopulation is genetically very homogeneous.

In the other localities, gene diversity H ranges between0.18 (Trento) and 1.0 (France), and nucleotide diversityπ ranges between 0.22% (Arezzo) and 1.1% (Poland) (seeTable 2). These differences should not be taken at theirface value, since standard deviations are high. However, ageneral pattern can be identified. Excluding the Liguriasample, where reintroduced individuals probably have dif-ferent origins, genetic variation in several Italian samplesseems lower than in other populations. This is probably aconsequence of the isolation and/or smaller populationsizes in central Italy. In addition, some non-Italian samples(Spain and Bulgaria) included individuals from differentregions or from different sites in the same region (Poland),whereas Italian samples refer to more limited areas.

Finally, we note that nucleotide diversity πi computedassuming a Kimura two-parameters model and a gammadistribution for the mutation rates, is only slightly largerthan π, suggesting only a small effect of homoplasy on thisstatistic.

Variance components. In the general analysis of the Europeanpopulations, the genetic variance was subdivided intotwo components: within populations, and between popula-tions. The results obtained including or excluding themonomorphic samples (first two lines in Table 3) confirm

Fig. 2 The network of haplotypes (numbered as in Table 1). The size of the circles is proportional to the number of individuals found withthat haplotype. Small black circles indicate missing haplotypes, whereas double lines correspond to indels. Asterisks identify the sequencesof Ligurian origin. Four main clusters of haplotypes (A1, A2, A3 and B) are separated by dotted lines.

MEC_1534.fm Page 1290 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R 1291

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

the presence of a strong genetic structure in the Europeanroe deer: most of the genetic diversity reflects differencesbetween populations, with ΦST values around 0.6.

When the same type of analysis was repeated for thenumerous Italian samples (lines 3 and 4 in Table 3), asimilar degree of divergence between populations emerges,even when the monomorphic samples and the presumablyreintroduced Ligurian samples are excluded.

We then analysed the Italian samples (excluding mono-morphic samples and Liguria) subdividing the geneticvariance into three components: within populations,between populations of the same geographical region, andbetween different regions (regions here are central Italyand the Alps). Partitioning the total variation in threehierarchical levels also allows the estimation of the relativedivergence between populations of the same region (meas-ured by ΦSC) and between different regions (measured byΦCT). The percentage of the variability within populationsis comparable to the previous analyses (line 5 in Table 3).Different populations within the same region are clearlydifferentiated (ΦSC = 0.58), but different regions are not(ΦCT not significantly different from 0). This result, how-ever, does not indicate the absence of divergence betweencentral Italy and Alps, which is evident from the networkin Fig. 2. Rather, it seems to be the consequence of the factthat in central Italy there are, in fact, two different genepools, which are connected only through the most frequentAlpine haplotype: one is found in Florence and Arezzo, theother in Siena (and also in the monomorphic sample formCastelporziano). The heterogeneity of the central Italy groupsubstantially reduces the ΦCT value, and this view is con-firmed by the high ΦCT value obtained after subdividing

Table 3 Results of hierarchical analysis of genetic variance

Type of subdivisionNo. of populations

Percentage of variance components Φ statistics

Within population

Between population

Between pop./within group

Between groups ΦST ΦSC ΦCT

General analysis1. All populations 17 37.1 62.9 — — 0.63 — —2. Monomorphic samples excluded 14 41.9 58.1 — —

0.58— —

Italian samples3. All populations 9 29.7 70.3 — — 0.70 — —4. Monomorphic samples and

Liguria excluded6 41.5 58.7 — — 0.58 — —

5. As analysis 4, but with two groups (CeIt, ItAl)

6 41.4 — 58.2 0.4 0.59 0.58 0.0

6. As analysis 4, but with three groups (SI, FL + AR, ItAl)

6 39.9 — 22.3 37.8 0.62 0.37 0.40

In the analysis of the Italian localities, after excluding the monomorphic (Castelporziano and Asiago) and reintroduced (Liguria) samples, two groups of populations are identified: central Italy (CeIt), and Italian Alps (ItAl). The central Italian group is additionally subdivided in the last analysis into Florence + Arezzo and Siena. Significance for Φ statistics (P < 0.05, 1000 permutations) is indicated by bold type.

Table 2 Genetic diversity in 17 localities with sample size equal toor greater than four individuals

Locality n k H (SD) π (SD) π i (SD)

Castelporziano 17 1 0.0 0.0 0.0Siena 15 5 0.56 (0.14) 0.49 (0.05) 0.51 (0.05)Arezzo 17 4 0.57 (0.12) 0.22 (0.01) 0.22 (0.01)Florence 19 6 0.54 (0.13) 0.38 (0.03) 0.38 (0.03)Liguria 17 8 0.82 (0.07) 0.87 (0.11) 0.92 (0.12)Monte Bondone 5 2 0.60 (0.18) 0.88 (0.53) 0.92 (0.59)Trento 11 2 0.18 (0.14) 0.21 (0.02) 0.22 (0.02)Asiago 4 1 0.0 0.0 0.0Tarvisio 5 4 0.90 (0.16) 0.99 (0.65) 1.06 (0.73)Austria 5 3 0.80 (0.16) 0.93 (0.59) 0.99 (0.65)Slovenia 5 3 0.70 (0.22) 0.93 (0.59) 1.00 (0.67)France 5 5 1.0 (0.13) 0.93 (0.59) 0.97 (0.63)Bulgaria 5 4 0.90 (0.16) 0.58 (0.27) 0.60 (0.28)Norway 5 1 0.0 0.0 0.0Slovakia 5 3 0.70 (0.22) 0.82 (0.47) 0.86 (0.51)Poland 5 3 0.70 (0.22) 1.11 (0.79) 1.21 (0.92)Spain 11 6 0.80 (0.11) 0.86 (0.18) 0.91 (0.20)All 160 49 0.94 (0.01) 1.28 (0.02) 1.38 (0.02)

n = sample size; k = number of haplotypes; H = gene diversity (Nei 1987); π = nucleotide diversity (%), with deletions weighted as substitutions; π i nucleotide diversity computed assuming a Kimura two-parameters model with deletions weighted as transversion and a gamma distribution for the mutation rate with α = 0.28. Sampling standard deviations (SD) for H are estimated following Nei (1987), whereas we used the derivations by Tajima (1983) to estimate the SD of π and π i. The last line refers to the total sample of 160 sequences. The expected number of alleles in a population is strongly affected by the sample size; the value of k as a measure of genetic variability should therefore be interpreted with caution.

MEC_1534.fm Page 1291 Thursday, July 18, 2002 7:53 AM

1292 C . V E R N E S I E T A L .

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

Tab

le 4

Pair

wis

e Φ

ST v

alue

s (b

elow

the

dia

gona

l) a

nd n

et n

umbe

r of

sub

stit

utio

ns b

etw

een

popu

lati

ons

(abo

ve th

e d

iago

nal)

Cas

telp

orzi

ano

Sien

aA

rezz

oFl

oren

ceL

igur

iaM

onte

bond

one

Tre

nto

Asi

ago

Tar

visi

oA

ustr

iaSl

oven

iaFr

ance

Bul

gari

aN

orw

aySl

ovak

iaPo

land

Spai

n

Cas

telp

orzi

ano

0.10

4.26

4.71

6.01

2.66

2.05

2.04

3.49

2.93

2.93

3.18

4.95

3.09

4.49

3.70

3.44

Sien

a0.

113.

123.

575.

511.

641.

331.

312.

412.

142.

342.

014.

322.

093.

382.

452.

47A

rezz

o0.

920.

720.

516.

221.

852.

132.

120.

611.

373.

072.

414.

683.

201.

961.

953.

61Fl

oren

ce0.

870.

700.

336.

702.

262.

542.

551.

041.

803.

532.

865.

143.

642.

402.

484.

06L

igur

ia0.

790.

680.

760.

755.

536.

156.

102.

721.

541.

164.

662.

395.

111.

773.

705.

02M

onte

bond

one

0.82

0.46

0.62

0.59

0.63

0.55

0.53

0.98

1.69

2.33

0.56

4.68

0.69

2.54

0.50

0.90

Tre

nto

0.88

0.50

0.74

0.69

0.73

0.32

0.00

1.81

2.14

2.59

1.93

5.06

1.02

3.71

2.03

1.92

Asi

ago

1.00

0.45

0.76

0.68

0.68

0.19

−0.1

31.

782.

112.

561.

915.

021.

013.

672.

001.

90T

arvi

sio

0.84

0.54

0.38

0.41

0.45

0.22

0.56

0.45

0.12

1.41

1.03

3.11

1.98

−0.0

60.

301.

89A

ustr

ia0.

830.

520.

550.

530.

320.

340.

600.

510.

03−0

.18

1.45

0.93

1.87

0.18

0.87

2.10

Slov

enia

0.83

0.54

0.72

0.68

0.27

0.41

0.64

0.56

0.28

−0.0

51.

990.

661.

871.

081.

782.

57Fr

ance

0.84

0.50

0.67

0.64

0.59

0.15

0.58

0.49

0.23

0.30

0.37

3.70

0.82

1.49

−0.2

50.

50B

ulga

ria

0.92

0.70

0.82

0.78

0.44

0.63

0.82

0.80

0.52

0.25

0.19

0.57

3.88

2.22

3.30

4.44

Nor

way

1.00

0.59

0.84

0.76

0.66

0.31

0.64

1.00

0.52

0.53

0.52

0.33

0.79

2.99

0.90

1.03

Slov

akia

0.89

0.63

0.64

0.61

0.36

0.45

0.73

0.68

−0.0

10.

060.

250.

320.

460.

670.

762.

93Po

land

0.83

0.54

0.61

0.60

0.52

0.12

0.56

0.45

0.08

0.19

0.32

−0.0

70.

510.

310.

180.

64Sp

ain

0.75

0.52

0.68

0.67

0.61

0.22

0.50

0.41

0.37

0.39

0.44

0.14

0.60

0.28

0.48

0.17

Mos

t (84

%) o

f the

ΦST

val

ues

are

sign

ifica

nt if

test

ed in

depe

nden

tly a

t the

5%

leve

l. Si

gnifi

cant

ΦST

val

ues

at th

e ta

ble-

wid

e le

vel o

f 5%

(seq

uent

ial B

onfe

rron

i tec

hniq

ue, H

olm

197

9) a

re in

dica

ted

in b

old. the central Italian samples into two groups, Siena and

Florence–Arezzo (see last line in Table 3).Finally, the pairwise comparison between all 17 popu-

lations (Table 4) again suggests high levels of geneticdivergence. About 80% of the ΦST values are larger than 0.3(40% are larger than 0.6), but significance after sequentialBonferroni correction (Holm 1979) is only reached whenlarge samples are involved in the comparison. The ΦSTand net number of substitutions for the Ligurian sampleindicate a clear similarity with the Austrian, Slovenian andSlovakian samples.

Population tree. The population tree in Fig. 3 supports theresults obtained from the network and the genetic varianceanalyses. The Ligurian sample clusters with the Bulgarianone and close to three other eastern populations (Slovakia,Slovenia and Austria). Castelporziano and Siena form ahighly supported group, as well as Trento and Asiago.We note however, that the inferred divergence of theTrento + Asiago group is mainly due to a strong drift orsampling effect, since 14 out of 15 sequences in this groupare identical. The Florence and Arezzo populations seemalso very close genetically, but very different from theirgeographical neighbours from Siena and Castelporziano.Finally, a branch supported by 70% of the bootstrappedtrees suggests an additional separation between twoheterogeneous groups, one constituted by the easternmostsamples (excluding Poland), Tarvisio, Liguria, Florenceand Arezzo, and the other including the remainingsamples. These two groups correspond approximately tothe clusters A1 + B and A2 + A3 in the haplotype networkof Fig. 2, respectively.

Divergence times and neutrality tests. Divergence times wereonly estimated between five main groups of localities, inorder to avoid major sampling effects. Groups here aregenetically distant and well-defined clusters in the popu-lation tree. The results (Table 5) can be summarized asfollows: (i) the shortest divergence time, only a fewthousand years, is estimated between two populations inthe Alps (Trento and Asiago) and the group of populationsfrom France, Spain and Poland; (ii) these two groups ofpopulations appear separated from the Central Italianpopulations of Siena and Castelporziano by around10 000–20 000 years; (iii) all the other comparisons suggestmuch larger divergence times, with the highest values(between 50 000 and 100 000 years depending on themutation rate) estimated between the group of eastern andreintroduced samples of Bulgaria, Slovenia and Liguriaand two groups of Italian samples, one in the Alps (Trentoand Asiago) and the other in central Italy (Florence andArezzo).

Tajima and Fu neutrality tests were separately applied tothe five groups defined above. The results of these 10 tests

MEC_1534.fm Page 1292 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R 1293

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

produced nine negative deviations from the neutralityexpectations (five of them statistically significant). Thesingle positive value was not statistically significant. Onlyfor the Florence–Arezzo group were both statistics(Tajima’s D and and Fu’s Fs) negative and significant. Thisresult is not unexpected from a simple visual inspection ofthe network in Fig. 2: the cluster A1, which includes all thesequences sampled in the Arezzo and Florence areas, has atypical star-like shape.

Discussion

The genetic structure of the European roe deer, as inferredfrom the mtDNA first hypervariable region, appearssubstantial. Populations differ significantly, with only 40%of the genetic variability shared between them and ΦSTvalues around 0.6. These levels of divergence are amongthe highest values observed among cervids (e.g. Gonzalezet al. 1998; Goodman et al. 2001; Hundertmarmark et al.2002) or other terrestrial ungulates (e.g. Templeton 1999;

Birungi & Arctander, 2000; Van Hooft et al. 2002; Nerstingand Arctander, 2002). Female philopatry, which is welldocumented in the roe deer in central Europe (Linnell et al.1998), and perhaps recent habitat fragmentation mostprobably played a role in differentiating local populations.However, when the genetic divergence between popula-tions is analysed considering their geographical location,some interesting patterns seem to emerge, and additionalexplanations are needed.

In general, samples from geographically close areas tendto be more similar than samples from more distant regions.However, when all populations are simultaneouslyanalysed with the Mantel procedure (Mantel 1967), the cor-relation between genetic and geographical distances is notsignificant (P > 0.05 regardless of the measure of geneticdistance considered). The model of isolation by distanceappears therefore to be insufficient to explain the pattern ofgenetic variation in the roe deer, and at least three situ-ations at three different geographical scales can justify thisresult. We will discuss them in turn.

1 0 0

A sia goM on te B o nd o n e

N o rw a yP ola n d

F ra n ce

A rezzo

F lorence

S lov ak iaA u str ia

S lov en ia L igu ria

B u lga ria

9 1

7 35 3

4 0 5 7

5 0

4 4

7 0

4 1

3 95 0

9 8

Ta rv is io

Tren to

S ien a

C a ste lp orzian o

S p a in

2 9

Fig. 3 Consensus neighbour-joining treeamong populations. Numbers indicatethe percentage of boostrapped trees thatproduced the clusters obtained in theconsensus tree. Bootstrapped trees areobtained by resampling individuals withinpopulations.

Table 5 Divergence times (in thousand of years) between five main groups of localities

Florence + Arezzo Siena + Castelporziano Trento + AsiagoBulgaria + Slovenia + Liguria

Siena + Castelporziano 79–40Trento + Asiago 60–30 19–9Bulgaria + Slovenia + Liguria 107–53 66–33 105–53Spain + France + Poland 42–21 14–7 3–2 76–38

Intervals are estimated considering two different mutation rates (0.04 and 0.08 substitutions per site per million years).

MEC_1534.fm Page 1293 Thursday, July 18, 2002 7:53 AM

1294 C . V E R N E S I E T A L .

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

Patterns and models of genetic variation in Europe

The genetic distances between the western, northern andcentral European populations, including Spain, France,Norway, Poland, and part of the Alps, are much smallerthan expected from the large geographical distancesseparating them. Most of the mtDNA sequences found inthese areas are genetically close and constitute a group ofhaplotypes called A3 in the statistical parsimony network.Two additional results support this view: (i) the averagepairwise ΦST values between these populations (whenthe Alpine populations of Trento, Monte Bondone andAsiago are considered) is 0.33, about half of the globalvalue observed in the general analysis; (ii) the estimateddivergence times within this group are much lower than inany other comparison.

A possible explanation for these results can be related toa colonization process from an Iberian refugium. Beforediscussing this hypothesis, we note, however, that probablytwo additional groups of genetically similar sequences canbe identified, which appear essentially restricted to twoareas. A distinct group of haplotypes, called B in the networkanalysis, is mainly found in the easternmost populationsanalysed here (with the exception of the presumablyreintroduced individuals sampled in the Italian Liguriasample). This cluster is separated by at least three muta-tions from all the other sequences. A second group of 11haplotypes with an evident star-like shape, called A1 in thenetwork, is restricted to the central Italian areas of Florenceand Arezzo: 10 haplotypes belonging to cluster A1 arefound in these samples, eight of them only there. Again, thepopulation tree identifies a group of eastern samples andthe central Italy group of Florence and Arezzo, and thisresult is confirmed by the pairwise ΦST analysis: theaverage value amongst the easternmost samples ofBulgaria, Slovenia and Slovakia, and between Florence andArezzo, is 0.30 and 0.33, respectively. Including the Liguriasample, where reintroduced individuals probably havedifferent origins, increases the first figure only to 0.33.

The pattern of relatively low genetic divergence withinthree major groups of populations in Europe is counterbal-anced by the higher or much higher average ΦST valuesbetween these groups. When the western-central Europeangroup is compared to the eastern and the Florence +Arezzo groups, average ΦST becomes 0.45 and 0.59, respect-ively, whereas ΦST = 0.71 in the comparison between theeastern and the Florence + Arezzo groups. The analysis ofthe net number of substitutions between populationsinstead of ΦST as a measure of population divergenceprovides very similar results.

Several analyses point therefore to three major groups ofmtDNA lineages distributed in three European areas, anda specific method (Dupanloup and Excoffier, unpublished)designed to recover the population grouping associated

with the highest divergence between groups confirms thisresult. We therefore suggest that an ancient process ofsubdivision due to the Upper Pleistocene glaciations andthe subsequent re-colonization from three Southern refugia(Iberia, Italy, Balkans) could have shaped the present geneticstructure of the roe deer in Europe. The estimated diver-gence times between these three groups of populations arein the order of magnitude of several tens of thousand ofyears, depending on the mutation rate and assuming nosubsequent gene flow. These dates are obviously approx-imate. Still, they give us the impression that the subdivisionof these groups can be traced back to some intermediateice ages after the Eamian interglacial (135 000 years ago),probably not the most recent glaciation around 20 000 yearsago. The neutrality tests provide some evidence of a possibledemographic expansion associated with the re-colonizationprocess.

The hypothesis that the roe deer expanded from putativerefugia in Iberia, Italy and the Balkans into northern Europeafter each glacial maximum agrees with the reconstructedevolutionary history of several plant and animal species inEurope (Taberlet et al. 1998; Hewitt 1999). In particular, theroe deer expansion resembles the so called ‘bear pattern’(Hewitt 1999), in that individuals from the Italian refugiumdid not spread over the Alps, and populations of Iberianorigin prevail now in northern Europe. Consistent withthis hypothesis, the genetic divergence (ΦST) between thethree roe deer populations more clearly related to theputative southern refugia (Spain, Bulgaria and the centralItaly area of Florence and Arezzo) falls in the 5% right tailof the frequency distribution of ΦST empirically obtainedby resampling with replacement 1000 groups of threenonmonomorphic populations from the complete data set(data not shown).

The divergence times between the three main groupsof populations analysed here are shorter than the timesestimated between refugium populations in other species(Taberlet et al. 1998; Hewitt 1999). Two complementaryexplanations can account for this difference. First, previousestimates are based on haplotype divergence and not onmethods specifically designed for the analysis of popula-tion divergence. The method we employed here (Gaggiottiand Excoffier, 2000) is specific for the analysis of popula-tions. Removing the predivergence differentiation betweenalleles should avoid overestimation of divergence times.The second explanation is related to the demographicphenomena associated with the glaciations. Taberlet et al.(1998) suggest that divergence times may be long, comparedto the age of the glaciations, because populations in therefugium areas were unaffected by glaciations. Estimateddivergence time should therefore reflect the much earlierdate of arrival of a species in the European continent. It ispossible, however, that at least for the roe deer the refugiapopulations acted not only as source populations for

MEC_1534.fm Page 1294 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R 1295

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

postglacial expansions, but also received genetic contribu-tions from other European populations during the glacialperiods of range contractions. This hypothesis impliessome level of genetic homogenization between refugiapopulations, possibly reflected in the shorter divergencetimes estimated in this study.

Finally, we note that Wiehler & Tiedemann (1998)proposed the hypothesis of a single west-Mediterraneanrefugium for the roe deer. Our results, based on a muchlarger sample, appear easier to reconcile with the three-refugia hypothesis.

Population structure in central Italy

A second important situation where a simple model ofisolation by distance does not apply is represented by thecentral Italian samples. Despite short geographical distances,the Florence–Arezzo and the Siena–Castelporziano groupsare genetically very different, with an estimated divergencetime among the highest values observed in this study.

A possible explanation for this result is that the Siena–Castelporziano group represents a second Italian refugium,separated between 10 000 and 20 000 years ago from theancestors of the Alpine populations. This hypothesiswould be consistent with the view that individuals fromCastelporziano, sometimes assigned to the putativesubspecies of Capreolus capreolus italicus, descend from apopulation isolated in refugium areas during the Pleis-tocene (Randi et al. 1998). We note, however, that geneticdivergence within the Siena–Castelporziano group isstrongly affected by the monomorphism of the Castelpor-ziano sample, probably due to the founder effect related tothe artificial establishment of this population in historicaltimes. As a consequence, estimated divergence times donot appear robust, and only two safe conclusions can besuggested. First, defining the Castelporziano population asa subspecies is not supported by the mtDNA data, unlessone is willing to identify at least three or four subspecies ofCapreolus capreolus. The genetic analysis of the isolatedgroups in Southern Italy (such as the Gargano and theOrsomarso populations) would hopefully clarify whether,at least in those areas, an Italian subspecies of roe deer canbe recognized. Second, a very efficient barrier to gene flowmust have prevented short-range migrations between twoCentral Italy regions. The Arno river seems a plausible can-didate, as it has isolated the Florence and the Siena areassince the Pliocene (Bartolini & Pranzini 1981).

What about restocking in central Italy? Different authorshave postulated a substantial or negligible (Lorenzini et al.1996) component of reintroduced individuals in theseareas. As far as our samples are concerned, a substantialreintroduction of females can be excluded. The geneticanalysis actually suggests that the populations from Firenze,Arezzo, Siena and Castelporziano have distinct genetic

features, and the large number of mtDNA haplotypesfound only in these areas is consistent with this view. Theanalysis of specimens conserved in museums since thebeginning of the century (i.e. before the time whenintroductions are described) will hopefully clarify thispoint further.

The Liguria sample

The Liguria sample in northwestern Italy represents thethird situation where genetic and geographical distancesare clearly not correlated. In this case, however, theexplanation is very simple: reintroduction of individuals iswell documented in this area, where the local populationsprobably became extinct between the two World Wars. Thenetwork analysis and the population tree clearly identifythe individuals sampled in Liguria as belonging to theEastern European group. This result confirms historicalrecords (provided by local authorities) of restoration ofwestern Alpine populations with animals from EasternEuropean countries. The pairwise ΦST values are alsoconsistent with the suggested sources of reintroducedindividuals in Austria and Slovenia, but do not exclude amore eastern origin in Slovakia. We note, however, thatnone of the Ligurian individuals share the same haplotypewith specimens included by the network analysis in theEastern Europe cluster B. The limited sample sizes couldexplain this finding (see for example Helgason et al. 2000).It is also likely that the exact place, or places, of originof reintroduced animals does not correspond exactly tothe areas covered by our Eastern European samples.As a matter of fact, the substantial genetic variabilityobserved in the Ligurian population is compatible withreintroduction from different source populations.

Final remarks: conservation and management implications

The data and the analyses we presented suggest that thehypervariable region of the mtDNA of the roe deer isgeographically structured not only in Europe but also inItaly, which allows the identification of reintroductionevents and more generally the description of geographical‘management units’ (Moritz 1994). This marker thereforeshould be regarded as an important tool for developingconservation and management policies in the roe deer,since its analysis could provide helpful information beforeany decisions concerning selective hunting and/or reintro-duction of animals are made.

Acknowledgements

We thank Paolo Casanova for stimulating discussions in the earlyphases of this research, Martina Lari, Alberto Cavallaro, Alessandro

MEC_1534.fm Page 1295 Thursday, July 18, 2002 7:53 AM

1296 C . V E R N E S I E T A L .

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

Romei for technical help, Guido Barbujani, Mike Bruford andIsabelle Dupanloup for helpful comments, and Juan Carranza forproviding the Spanish samples. We are also grateful to Bob Wayneand three anonymous reviewers for useful suggestions thatimproved the paper, and to Krisztina Vasarhelyi and MilviaChicca for editing the text. This work has been supported by agrant from the Consiglio Regionale Toscano of the FederazioneItaliana della Caccia.

References

Bartolini C, Pranzini E (1981) Plio-Quaternary evolution of theArno basin drainage. Zeitung fur. Geomorphologie N. F.(Suppl.)Bd, 40, 77–91.

Birungi J, Arctander P (2000) Large sequence divergence ofmitochondrial DNA genotypes of the control region withinpopulations of the african antelope, kob (Kobus kob). MolecularEcology, 9, 1997–2008.

Caughley G, Gunn A (1996) Conservation Biology in Theory andPractice. Blackwell Science, Cambridge, MA.

Clement M, Posada D, Crandall KA (2000) TCS: a computerprogram to estimate gene genealogies. Molecular Ecology, 9,1657–1619.

Danilkin A (1996). Behavioural Ecology of Siberian and European RoeDeer. Chapman & Hall, London.

Douzery E, Randi E (1997) The mitochondrial control region ofCervidae: evolutionary patterns and phylogenetic content.Molecular Biology and Evolution, 14, 1154–1166.

Excoffier L, Smouse PE, Quattro JM (1992) Analysis of molecularvariance inferred from metric distances among DNA haplo-types: application to human mitochondrial DNA restrictiondata. Genetics, 131, 479–491.

Felsenstein J (1989) phylip — Phylogeny Inference Package(Version 3.2). Cladistics, 5, 164–166.

Festa E (1925) Il capriolo dell’Italia centrale. Bollettino MuseoZoologia Anatomia Comparata, Torino, XL, Nuova Serie, N 37, 1–2.

Fu YX (1997) Statistical tests of neutrality of mutations againstpopulation growth, hitchhiking and background selection.Genetics, 147, 915–925.

Gaggiotti OE, Excoffier L (2000) A simple method of removing theeffect of a bottleneck and unequal population sizes on pairwisegenetic distances. Proceedings of the Royal Society of London, B:Biological Sciences, 267, 81–87.

Gonzalez S, Maldonado JE, Leonard JA et al. (1998) Conservationgenetics of the endangered Pampas deer (Ozotoceros bezoarticus).Molecular Ecology, 7, 47–56.

Goodman SJ, Tamate HB, Wilson R et al. (2001) Bottlenecks, driftand differentiation: the population structure and demographichistory of sika deer (Cervus nippon) in the Japanese archipelago.Molecular Ecology, 101, 1357–1370.

Grubb P (1993) Order Artiodactyla. In: Mammal Species of theWorld. A Taxonomic and Geographic Reference (eds Wilson DE,Reeder DM), pp. 377–414. Smithsonian Institution Press,Washington.

Hartl GB, Reimoser F (1988) Biochemical variation in roe deer(Capreolus capreolus L.): are r-strategists among deers geneticallyless variable than K-strategists? Heredity, 60, 221–227.

Hasegawa MA, Di Rienzo A, Kocher TD, Wilson AC (1993)Toward a more accurate time scale for the human mitochondrialDNA tree. Journal of Molecular Evolution, 37, 347–354.

Helgason A, Sigurthardottir S, Nicholson J et al. (2000) EstimatingScandinavian and Gaelic ancestry in the male settlers of Iceland.American Journal of Human Genetics, 67 (3), 697–717.

Hewitt GM (1999) Post-glacial re-colonization of European biota.Biological Journal of the Linnean Society, 68, 87–112.

Holm S (1979) A simple sequentially rejective multiple test pro-cedure. Scandinavian Journal of Statistics, 6, 65–70.

Huber KY, Watson EE, Hendy MD (2001) An algorithm forconstructing local regions in a phylogenetic network. MolecularPhylogenetics and Evolution, 19 (1), 8.

Hudson RR (1990) Gene genealogies and the coalescent approach.In: Oxford Survey in Evolutionary Biology (eds Futuyma D,Antonovics J), pp. 1–44. Oxford University Press, New York.

Hundertmark KJ, Shields GF, Udina IG, Bowyer RT, Danilkin AA,Schwartz CC (2002) Mitochondrial phylogeography of moose(Alces alces): late Pleistocene divergence and populationexpansion. Molecular Phylogenetics and Evolution, 22, 375–387.

King TL, Burke T (2000) Special issue on gene conservation:identification and management of genetic diversity. Introduc-tion. Molecular Ecology, 8, S1–S3.

Linnell JDC, Wahlström K, Gaillard JM (1998) From birth to inde-pendence: Birth, growth, neonatal mortality, hiding behaviourand dispersal. In: The European Roe Deer: the Biology of Success(eds Andersen R, Duncan P, Linnell JDC), pp. 257–283, Scandi-navian University Press, Oslo.

Lorenzini R, Mattioli L, Rustioni M, Patalano M (1996) Allozymeand craniometric variability in the roe deer (Capreolus capreolus)from central Italy. Zeitschrift Fur Säugertierunde, 61, 7–24.

Lorenzini R, Burrini L, Mazzoni della Stella R (1997) Biochemicalgenetic differentiation in some roe deer populations of Tuscany,central Italy. Italian Journal of Zoology, 64, 239–244.

Mantel N (1967) The detection of disease clustering and a general-ized regression approach. Cancer Research, 27, 209–220.

Moritz C (1994) Defining ‘evolutionary significant units’. Trends inEcology and Evolution, 9, 373–375.

Nei M (1987) Molecular Evolutionary Genetics. Columbia UniversityPress, New York.

Nersting LG, Arctander P (2001) Phylogeography and conser-vation of impala and greater kudu. Molecular Ecology, 10,711–719.

Perco F, Calò CM (1994) The status of roe deer in Italy. In: Proceed-ings of the 2nd European Roe Deer Meeting. Brixen, South Tyrol /Italy, October 27–30 (ed. Wotschikowoky U), pp. 43–49. Wild-biologische Gesellschaft, Munich.

Randi E, Pierpaoli M, Danilkin A (1998) Mitochondrial DNA poly-morphism in populations of Siberian and European roe deer(Capreolus pygargus and C. capreolus). Heredity, 80, 429–437.

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

Sambrook J, Fritsch EF, Maniatis T (1989). Molecular Cloning. ALaboratory Manual. Cold Spring Harbor Laboratory Press, NewYork.

Schneider S, Roessli D, Excoffier L (2000) Arlequin, A software forpopulation genetic data analysis, Version 2.0. University ofGeneva, Switzerland.

Sokolov VE, Gromov VS (1990) The contemporary ideas on roedeer (Capreolus Gray, 1821) systematization: morphological,ethological and hybridological analysis. Mammalia, 54, 431–444.

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

MEC_1534.fm Page 1296 Thursday, July 18, 2002 7:53 AM

G E N E T I C S T R U C T U R E O F E U R O P E A N R O E D E E R 1297

© 2002 Blackwell Science Ltd, Molecular Ecology, 11, 1285–1297

Tajima F (1983) Evolutionary relationship of DNA sequences infinite populations. Genetics, 105, 437–470.

Tajima F (1989) Statistical method for testing the neutral mutationhypothesis by DNA polymorphism. Genetics, 123, 585–595.

Templeton AR (1999) Human races: a genetic and evolutionaryperspective. American Anthropologist, 100, 632–650.

Templeton AR, Crandall KA, Sing CF (1992) A cladistic analysis ofphenotypic associations with haplotypes inferred fromrestriction endonuclease mapping and DNA sequence data. III.Cladogram estimation. Genetics, 132, 619–633.

Thompson JD, Higgins DG, Gibson TJ (1994) ClustalW: improvingthe sensitivity of progressive multiple sequence alignmentthrough sequence weighting, position-specific gap penaltiesand weight matrix choice. Nucleic Acids Research, 22, 4673–4680.

Van Hooft WF, Groen AF, Prins HT (2002) Phylogeography of theAfrican buffalo based on mitochondrial and Y-chromosomalloci: Pleistocene origin and population expansion of the Capebuffalo subspecies. Molecular Ecology, 11, 267–279.

Wiehler J, Tiedemann R (1998) Phylogeography of the Europeanroe deer Capreolus capreolus as revealed by sequence analysis of

the mitochondrial Control Region. Acta Theriologica Supplement,5, 187–197.

Wright S (1951) The genetical structure of populations. Annals ofEugenetics., 1, 323–334.

Yang Z (1994) Maximum likelihood phylogenetic estimation fromDNA sequences with variable rates over sites: approximatemethods. Journal of Molecular Evolution, 39, 306–314.

Yang Z (1997) PAML: a program package for phylogenetic ana-lysis by maximum likelihood. CABIOS, 13, 555–556.

This work arose from an interest of Cristiano Vernesi, ElenaPecchioli and David Caramelli, co-founders of the Association‘Biosfera-Biologia della Conservazione’, in the management of theroe deer in Italy. All of them contributed to the sample collectionand the laboratory analyses. Ralph Tiedemann and Ettore Randiprovided samples and experience regarding this species in otherparts of Europe. Giorgio Bertorelle is interested in the populationgenetics of several species, and together with Cristiano Vernesianalysed and interpreted the data.

MEC_1534.fm Page 1297 Thursday, July 18, 2002 7:53 AM