The use of geostatistics and GIS for evolutionary history ...webpages.icav.up.pt/ptdc/bia-bic/118151/2010/tomovic...LJILJANA TOMOVIC´1,2*, JELKA CRNOBRNJA-ISAILOVIC´2,3 and JOSÉ

The use of geostatistics and GIS for evolutionaryhistory studies: the case of the nose-horned viper(Vipera ammodytes) in the Balkan Peninsula

LJILJANA TOMOVIC1,2*, JELKA CRNOBRNJA-ISAILOVIC2,3 and JOSÉ CARLOS BRITO4

1Institute of Zoology, Faculty of Biology, University of Belgrade, Studentski trg 16, 11000 Belgrade,Serbia2Institute for Biological Research, University of Belgrade, Bulevar Despota Stefana 142, 11000Belgrade, Serbia3Department of Biology and Ecology, Faculty of Sciences and Mathematics, University of Niš,Višegradska 33, 18000 Niš, Serbia4CIBIO – Centro de Investigação em Biodiversidade e Recursos Genéticos da Universidade do Porto,Instituto de Ciências Agrárias de Vairão, R. Padre Armando Quintas, 4485-661 Vairão, Portugal

Received 8 March 2010; revised 29 May 2010; accepted for publication 31 May 2010bij_1513 651..666

Geostatistics and geographical information system (GIS) procedures are novel techniques helpful for the identifi-cation of environmental correlates sustaining contact zones among subspecies or closely related species. In this paper,we tried to infer evolutionary scenarios for Vipera ammodytes across the European part of its distribution area usinggeostatistics and ecological niche-based models, hence trying to solve several biogeographical questions thatremained unclear after the application of classical morphological tools and genetic analyses. Eleven morphologicaltraits from 871 vipers were analysed with geostatistics and ecological niche-based modelling. Interpolation by krigingwas used to generate surfaces of morphological variation, which were combined with spatial principal componentsanalysis (SPCA). SPCA maps were used to test putative morphological differentiated groups with discriminantfunction analysis (DFA). Maximum entropy modelling and seven environmental variables were used to identifyfactors limiting the distribution of groups and areas for the potential occurrence of such groups. Three patterns ofmorphological variation were observed: a north-west/south-west cline, transition zones with steep clines of variationin a west–east arc, and particular character traits that disturbed the general cline. SPCA identified between threeand nine putative population groups, of which three were supported by DFA. Areas of potential occurrence of thesegroups were coherent with the range of the three subspecies of V. ammodytes currently recognized. The distributionof all subspecies was mostly related to precipitation in the driest month. Areas of probable sympatry betweensubspecies are generally small and restricted. The main patterns of geographic variation of morphological charactersfor V. ammodytes were similar to the patterns obtained for Vipera latastei and Vipera monticola; the sameenvironmental factors limit the distribution of differentiated groups of vipers in the Balkans and the IberianPeninsula. The influence of humidity on the variation of morphological traits in spatially separated viper taxa fromthe two European peninsulas coincides with their phylogenetic relatedness. Geostatistics and GIS procedures weresuccessful in the identification of environmental correlates sustaining contact zones among V. ammodytes subspeciesin the Balkans. The same techniques should be applied for studying other parapatric forms and refugia regions.© 2010 The Linnean Society of London, Biological Journal of the Linnean Society, 2010, 101, 651–666.

ADDITIONAL KEYWORDS: Balkans – biogeography – evolution – morphological variability.

*Corresponding author. E-mail: [email protected]

Biological Journal of the Linnean Society, 2010, 101, 651–666. With 5 figures

© 2010 The Linnean Society of London, Biological Journal of the Linnean Society, 2010, 101, 651–666 651

INTRODUCTION

The dynamic and complex palaeogeographic andpalaeoclimatic history of Europe undoubtedly influ-enced the population dynamics of many taxa, espe-cially those susceptible to changes in temperatureand humidity, such as the squamate species. Contem-porary phylogeographic studies prove that the BalkanPeninsula was one of the most important Europeanrefugia (together with the Iberian Peninsula and theApennines) during the Pleistocene Ice Ages (Taberletet al., 1998; Crnobrnja-Isailovic, 2007).

Palaeogeographical, hydrological, and palaeoclima-tological changes during the Quaternary must havehad significant influence on the evolutionary historyof many recent reptile species. European vipers rep-resent a good model system for analyses of the impactof these factors, as their phylogeny largely coincidewith the timescale presented: they differentiated fromother Eurasian vipers during the early Miocene (Lenket al., 2001; Garrigues et al., 2005), and the ancestralform of the Vipera aspis complex was found in theMiocene fossil records (Szyndlar & Böhme, 1993). Onthe other hand, basal intraspecific genetic differentia-tion within two of the three species belonging tothe Vipera aspis complex (Vipera ammodytes, Viperaaspis, and Vipera latastei) was dated to the middlePliocene [V. aspis, 4.5–2.8 Mya (Ursenbacher et al.,2006); V. ammodytes, 4.1–3.6 Mya (Ursenbacheret al., 2008)]. Finally, further genetic differentiationsof the nose-horned viper (V. ammodytes) were dated tothe Pleistocene (Ursenbacher et al., 2008).

Several studies used advanced statistical multivari-ate analysis, such as principal component analysis(PCA) or canonical variate analysis (CVA), to illustrategeographic patterns of variation in morphologicalcharacters (Thorpe, 1987a and references therein).These studies stressed the relationship between sharptransition zones and steep clines in morphologicaltraits with the occurrence of hybrid zones resultingfrom secondary contact (Thorpe, 1987b), and tried todisentangle the effects of ecological adaptation andphylogeographic processes in current morphologicalpatterns (e.g. Thorpe et al., 1991). The combination ofPCA and CVA with interpolation algorithms furtherenhanced the robustness of analyses, and emphasizedthe usefulness of geostatistical approaches to theanalyses of geographic patterns of variation in mor-phological characters (e.g. Báez & Brown, 1997).Recently, the combination of geostatistics with geo-graphical information systems (GIS) has increased theimportance of geography in evolutionary biology (Kidd& Ritchie, 2006; Swenson, 2008). GIS are powerfultools to analyse geographic-related processes, andprovide novel insights into morphological and geneticpatterns of variation (Kidd & Ritchie, 2000; Hoffmann

et al., 2003; Brito et al., 2008; Martínez-Freiría et al.,2009), the location of hybrid zones (Swenson, 2006;Brito et al., 2008; Martínez-Freiría et al., 2008), andthe dynamics of gene flow therein (Spear et al., 2005).The combination of ecological niche modelling withmolecular phylogenies expanded even more evolution-ary biology studies by linking geographic patternsof ecological and genetic variation in evolutionaryprocesses (Hugall et al., 2003; Knouft et al., 2006;Knowles, Carstens & Keat, 2007).

In this study, geostatistics and GIS were combinedto examine morphological variability in a viper fromthe Eastern Mediterranean basin. Among Europeanvipers, V. ammodytes is one of the most widespread inSouthern Europe. It is a moderately sized viper, upto 1 m long, with a typical ‘horn’ on the snout, justabove the rostral scale. Its range extends from centralnorthern Italy, southern Austria, through the Balkansand southern Romania, to north-eastern Turkey andTranscaucasia (Arnold & Ovenden, 2002). In previoussystematic revisions, up to seven subspecies have beendescribed, but a recent comprehensive morphologicalstudy by Tomovic (2006) proved the reliability of onlythree of them: V. a. ammodytes (northern Italy, south-ern Austria, Slovenia, Croatia, Bosnia and Herze-govina, Montenegro, northern Albania, most of Serbia,both north-western FYR Macedonia and Bulgaria,and western Romania), V. a. montandoni (south-eastern Romania, most of Bulgaria, western Turkey,north-eastern, northern, and north-western Greece,most of FYR Macedonia, the southernmost part ofSerbia, and southern and central Albania), and V. a.meridionalis (central Greece, Peloponnesus, andCyclades), with several transition zones betweenthem. The most intriguing result of that study was theapparently increasing gradient of morphological dif-ferentiation from the north-west to the south-east,suggestting higher degrees of diversity in south-eastern populations.

A recent phylogeographic study of V. ammodytes(Ursenbacher et al., 2008) confirmed that the basalsplitting within this species happened between 4.1and 3.6 Mya. In addition, population groups from thesouthern Balkans displayed higher genetic differen-tiation in comparison to the north-western groups:overall genetic divergence (e.g. number of clades)seemed to be higher than phenotypic divergence.

The present morphological study aims to inferevolutionary scenarios for V. ammodytes across itsdistribution area using geostatistics and GIS-basedniche models, and tries to solve several biogeographi-cal questions that remained unclear after the appli-cation of classical morphological analytical tools. Thespecific questions stated in this study are: (1) arethere any geographic trends in the distribution ofindividual morphological traits; (2) are there coherent

652 L. TOMOVIC ET AL.

© 2010 The Linnean Society of London, Biological Journal of the Linnean Society, 2010, 101, 651–666

morphological groups with consistent spatial dis-tributions; (3) is the distribution of morphologicalgroups related to environmental factors; (4) where arecontact zones between morphological groups located;(5) is there any congruence in distribution betweenmorphological groups and genetic clades defined inprevious phylogenetical studies? The linking ofgeostatistics, ecological niche modelling, and GISis expected to identify morphologically coherentgroups, environmental sources of adaptation for suchgroups, and areas of secondary contact and probablehybridization among groups.

MATERIAL AND METHODSDATA

Specimens and morphological charactersThe study area includes the European range of thespecies (Fig. 1), thus excluding the populations from

the Asian part of Turkey. A total of 871 specimens(434 males and 437 females) with clear geographicassignment were examined from the collections ofseveral institutions (details in Tomovic, 2006). Thegeographic location of the specimens examined inmuseum collections was determined manually fromGoogle Earth and latitude/longitude was inscribed ina georeferenced database. The locations of specimenswere displayed in GIS ArcMap 9.2 (ESRI, 2006) onthe WGS84 datum (Fig. 1).

For each specimen, a total of 11 meristic andmorphometric characters were recorded: BPC, bodypattern complexity, coded in four categories relatedto the increasing levels of complexity of the dorsalcoloration pattern; DES_F and DES_M, distancebetween the eye and supralabial scales (mean valueof the both sides) in females and males; HHE_F andHHE_M, horn height from the rostral plate to thetop of the horn in females and males; HWI_F and

Figure 1. Study area, major toponomies, and location of specimens of Vipera ammodytes used in the analysis (WGS84projection). The analysis mask delimits the area for which continuous morphological trait surfaces were derived. The thicklines delimit the subspecies separation proposed by Tomovic (2006): 1, Vipera ammodytes ammodytes; 2, Viperaammodytes montandoni; 3, Vipera ammodytes meridionalis; ALB, Albania; AUS, Austria; BHZ, Bosnia and Herzegovina;BUL, Bulgaria; CRO, Croatia; GRE, Greece; ITA, Italy; MAC, Macedonia; MON, Montenegro; ROM, Romania; SER,Serbia; SLO, Slovenia; TUR, Turkey.

MORPHOLOGY OF THE NOSE-HORNED VIPER 653


HWI_M, head width across the widest part ofthe head in females and males; NSR, number ofnasorostral scales in contact with other scales;RHE_F and RHE_M, rostral plate height in femalesand males; RWI_F and RWI_M, rostral plate width infemales and males; SC_F and SC_M, number of sub-caudal scales in females and males; SHE_F andSHE_M, snout height from edge of the upper lip tothe canthus rostralis in females and males; SPR,number of suprarostral scales; VEN, number ofventral scales, excluding preventrals and anal, follow-ing the method used by Saint-Girons (1978). It rangedfrom broken marks or complete marks with discon-tinuous large spots to broad marks in a zigzag shapewith very broad unions of dorsal marks. These char-acters were reported to present geographic variationin previous morphological analyses (Tomovic &Džukic, 2003; Tomovic, 2006). Analyses were con-ducted separately for males and females in order toavoid sexual dimorphism-related biases. Althoughonly adult specimens (i.e. males with a snout–ventlength, SVL > 40 cm and females with a SVL > 45 cm)were included in analyses, all biometric traits werecorrected for SVL using residuals from regression ofeach trait against SVL in order to avoid body-sizebiases.

Environmental factorsA set of seven low to slightly correlated (r < 0.700 inall cases) environmental factors or ecogeographicalvariables (hereafter EGVs), known to affect the dis-tribution of European viperid snakes, were selectedfor the analyses (Santos et al., 2006; Brito et al.,2008; Martínez-Freiría et al., 2008). Two types ofEGV were considered: (1) topographical – a digitalelevation model with altitude (30″ resolution; USGS,2006), from which a slope was generated using the‘Slope’ function of the GIS; and (2) climatic – a setof five climate layers (0.0083 degrees resolution;Hijmans et al., 2005), representing seasonal andextreme trends of temperature and precipitation(Table 1). The ambiguity and imprecision in locationdescription in some museum collections precludedthe use of finer scales of analysis (i.e. 5 ¥ 5- or1 ¥ 1-km grid cells). Thus, although the size of thegrid cells (pixel) of the EGV was around 1 km2, thelocation of specimens was available at a 10 ¥ 10-kmresolution. Therefore, in order to combine both mor-phological and environmental data, the EGV wereresampled to a coarser resolution (10 km) using the‘Aggregate’ function of ArcMap GIS. In the newEGV, each output pixel contains the mean value ofthe input pixels that are encompassed by the extentof the output pixel. All EGV were quantitative. Allvariables were projected in the WGS84 datum.

MODELLING PROCEDURES

Spatial patterns in individualmorphological charactersThe specimens examined did not cover all pixels ofthe study area (Fig. 1). Therefore, it was necessary tointerpolate trait values at unsampled locations (Kidd& Ritchie, 2000; Brito et al., 2008; Martínez-Freiríaet al., 2009). For each trait, a continuous surface wascreated with ‘Kriging’ interpolation method (Olivier,1990), implemented in ArcMap GIS in the ‘Geostatis-tical Analyst’ extension (Johnston et al., 2001). Whenmore than one specimen was examined with the samegeographic coordinates, the mean for the trait wasused. Each standard trait map was reclassified intoequal intervals between their respective maximumand minimum values for the trait, and then convertedto raster format. Accurate interpolation modelsshould have the mean error close to 0, the smallestpossible root-mean-square error and average stan-dard error, and the root-mean-square standardizederror close to 1 (for details see Johnston et al., 2001).

Correlation coefficients between continuous mor-phological trait surfaces, and with latitude and lon-gitude, were used to identify geographic trends inmorphological traits, and were calculated with the‘Band Collection Statistics’ tool of ArcMap GIS.

Spatial patterns in multivariatemorphological variabilityThe identification of areas with multivariate clinesand with coherent morphological variability followedseveral steps.

1. A spatial principal components analysis (SPCA)was undertaken using the continuous morphologi-cal trait surfaces derived in the previous step(Kidd & Ritchie, 2000; Brito et al., 2008; Martínez-Freiría et al., 2009). The SPCA maps were created

Table 1. Environmental factors used to test for relation-ships with spatial variation in the morphological variabil-ity of Vipera ammodytes

Code Variable Range and units

ALTI Altitude From 0 to 3547 m a.s.l.SLOP Slope From 0 to 82%TANR Temperature annual

rangeFrom 18.0 to 34.2 °C

TMIN Minimum temperatureof coldest month

From -16.9 to 9.8 °C

PDRY Precipitation of driestmonth

From 0 to 121 mm

PWET Precipitation of wettestmonth

From 34 to 297 mm

PSEA Precipitation seasonality From 10 to 86 mm



using the ‘Principal Components Analysis’ exten-sion of ArcMap GIS. Trait surfaces were unstand-ardized and used variance/covariance matrices.For each PC map, the loading scores were used asa measure of the association between surface traitmaps, and the total variance accounting for eacheigenvalue was used to evaluate the level of expla-nation of the analysis.

2. The individual PC maps were combined in a singlemap with the ‘Composite Bands’ tool of ArcMapGIS, and resulted in a multiple-band map display-ing the individual PC maps together as a Red-Green-Blue composite.

3. The study area was partitioned into geographicalareas (groups) according to the main subdivisionsobserved in the SPCA map. The ‘IsoCluster’ and‘Maximum Likelihood Classification’ tools ofArcMap GIS (ESRI, 2006) were used to identifythe range of each group. To avoid ambiguities inthe decision of how many groups could be found inthe composite SPCA map, seven distinct classifi-cation maps were produced, each one assuming thepresence of between three and nine groups. Theminimum and maximum number of groups wasselected according to previous analyses: there arethree subspecies in the study area defined accord-ing to morphological variation (Tomovic, 2006),and nine well-supported genetic lineages wereidentified with molecular markers (Ursenbacheret al., 2008). Application of the maximum likeli-hood classification function to determine the rangeof each group, and independently testing each pos-sible classification scheme of between three andnine groups, removed subjectivity from the processof selecting how many groups are present andwhat is their distribution.

4. The significance of group number and rangeproposed by the classification function was testedwith discriminant function analysis (DFA). DFAwas performed to clarify the relative importance ofmorphological variables as discriminators betweena priori groups and the relative positions of thecentroids of those groups (Tomovic, 2006). Malesand females were analysed separately because ofsignificant sexual dimorphism (Tomovic et al.,2002). A stepwise classification procedure was usedto evaluate population membership with STATIS-TICA 6.0 (StatSoft Inc., 2003). Reallocation withcross-validation was used to assess the distinctnessof the specimen grouping, and the percentage ofcorrect assignment of specimens to each populationwas taken as a measure of model robustness.

Biogeographic patterns in morphological groupsThe examined specimens were assigned into geo-graphic areas in the ArcMap GIS, following the same

groups tested with DFA (see above) that werehypothesized from the SPCA maps. The accuratedetermination of the absence of V. ammodytes in aregion of study was not possible because of bothdaily and seasonal variations in the activity patternsof the species (Crnobrnja-Isailovic, Ajtic & Tomovic,2007), as well as because of logistic constraints.Therefore, to detect biogeographic patterns in thedistribution of morphological variability, themaximum entropy principle was used (Phillips,Anderson & Schapire, 2006, Phillips & Dudík, 2008).This modelling technique requires only presencedata as input, but consistently performed well incomparison with other methods (Elith et al., 2006),especially for low samples sizes (Hernandez et al.,2006). A matrix with specimen localities for eachgroup and the set of EGVs was imported intoMAXENT 3.3 (Phillips et al., 2006). A total of 20replicates were run for each model type with randomseed, which allows a different random 20% test/80%train data partition in each run. Presence data foreach replicate were chosen by bootstrap allowingsampling with replacement. Models were run withauto-features mode (Phillips et al., 2006), with amaximum of 1000 interactions and regularization setto 1.0, and the area under the curve (AUC) of thereceiver operating characteristic (ROC) plot wastaken as a measure of the overall fit of the models(Fielding & Bell, 1997). The importance of each EGVfor explaining the distribution of vipers was deter-mined by the average percentage contribution ofeach EGV for the models (Phillips et al., 2006). Therelationship between occurrence of vipers and eachEGV was determined by the examination of the pro-files of response curve plots from univariate models(Martínez-Freiría et al., 2008; Brito et al., 2008,2009). Similar profiles between two groups for agiven EGV were taken as an indication of identicalrelationships between the occurrence of these speciesand the range of variation of the EGV. A distinctprofile of a group in relation to the other groups wastaken as an indication of a divergent relationship,and possible exclusion of that group within therange of values of the EGV selected exclusively(Martínez-Freiría et al., 2008; Brito et al., 2008,2009).

In order to identify potential contact zones betweengroups in the Balkans, the average Maxent modelswere imported into ArcMap GIS. The maximumentropy models were reclassified using the tenth per-centile training presence, given that this thresholdassumes that 10% of specimen localities might haveerrors in geographic referencing: a possible problemwhen dealing with museum data sets where speci-mens were collected by different researchers overlong time periods (Raes et al., 2009). Models were



reclassified to display the areas of probable absenceand of probable presence for each group. To identifyareas of probable contact zones between groups, thesemaps were intersected in the GIS and the pixelsof probable presence common to both geographicgroups were taken as representing probable sympatry(Martínez-Freiría et al., 2008; Brito et al., 2008,2009).

RESULTSSPATIAL PATTERNS IN INDIVIDUAL

MORPHOLOGICAL CHARACTERS

The Kriging interpolation method produced accuratecontinuous surfaces of morphological variation(Table 2). Three basic patterns of spatial variation inmorphological traits could be observed (Fig. S1).

1. A geographic cline from the north-western region(Italy, Austria, Slovenia, and Croatia) to the south-ern region (Greece and the European part ofTurkey). The cline was either for decreasing orincreasing values of traits (decreasing, VEN,SC_M, NSR, and RWI; increasing, HWI and RHE)with decreasing latitude.

2. Transition zones with steep clines of morphologicalvariation in a west–east arc along the Prokletije,Šar Planina, Osogovo, and Stara Planina moun-tain chains (see Fig. 1 for toponymies). This

pattern was mostly apparent in the traits VEN,SC_F, and RHE.

3. Particular character traits that disturbed thegeneral north-western to southern cline occurredin the chain of mountains ranging from southernMacedonia to the Pirin, Rila, and WesternRhodope mountains of Bulgaria. This pattern wasmostly evident in traits HHE, SHE, and DES.

The surface traits of HHE_F, HWI_F, NSR, RHE_M,RWI_F, and SHE_M exhibited high correlations(r > 0.700) with at least one other trait, and wereexcluded from further analyses. Latitude was posi-tively correlated with VEN, HWI_F, NSR, SCM, andSHE_M, whereas longitude was negatively correlatedwith HHE_F, NSR, RHE_M, RHE_F, and HHE_M.

SPATIAL PATTERNS IN MULTIVARIATE

MORPHOLOGICAL VARIABILITY

The variance accounted by the three most explicativedimensions on the SPCA was 80%, and practically alltraits explained an important level of variation in thePC1, PC2, or PC3 axes (Table 3). The PC1 summarizedthe major cline, with north-western to southern orien-tation, observed in some individual surface trait maps:low values in Italy, Austria, Slovenia, Croatia, andBosnia and Herzegovina, and high values in Greeceand Turkey (Fig. 2; PC1). The PC1 also illustrated

Table 2. Statistical measures to assess the performance of the Kriging interpolation of morphological traits fromspecimens of Vipera ammodytes

Trait Mean

Root-mean-square

Averagestandarderror

Meanstandardized

Root-mean-squarestandardized N

VEN 0.030 3.555 3.318 0.009 1.070 871SC_M 0.015 2.499 2.680 0.006 0.934 432SC_F -0.014 2.316 2.354 -0.006 0.985 433SPR 0.001 0.677 0.607 0.002 1.112 863NSR -0.002 0.607 0.654 -0.003 0.932 868HWI_M -0.011 1.637 1.630 -0.006 1.004 417HWI_F -0.007 1.723 1.602 -0.004 1.071 423HHE_M -0.005 0.581 0.572 -0.009 1.015 417HHE_F 0.002 0.579 0.585 0.003 0.989 423SHE_M 0.001 0.415 0.305 0.002 1.425 417SHE_F -0.002 0.397 0.373 -0.004 1.064 423RHE_M 0.001 0.396 0.386 0.005 1.019 417RHE_F 0.002 0.451 0.401 0.004 1.122 423RWI_M -0.001 0.321 0.320 -0.002 1.003 417RWI_F 0.000 0.287 0.283 -0.001 1.015 423DES_M -0.001 0.297 0.283 -0.004 1.045 417DES_F -0.001 0.290 0.293 -0.005 0.986 423BPC -0.004 0.744 0.681 -0.006 1.092 871

N = number of specimens analysed. Trait codes are explained in the Material and methods section.



steep clines along the Prokletije and Šar Planinamountains, and along the Južna Morava River valley(Serbia). PC2 summarized the high variability alongthe mountains of Macedonia, northern Greece, andsouth-western Bulgaria, which was also observed inseveral individual PC maps (Fig. 2; PC2). The combi-nation of the highest loading scores exhibits a complexpattern of morphological variation (Fig. 2, SPCA).Nevertheless, it is clearly very complicated to decideunambiguously the number of groups present andtheir range. The ‘IsoCluster’ and ‘Maximum LikelihoodClassification’ functions allowed the independent pro-duction of maps depicting bewteen three and nineputative groups (their respective range and results arepresented in Fig. S2). The validity of these putativesubdivisions was assessed with a discriminant func-tion analysis. The average total correct classification ofspecimens between males and females consistentlydecreased with increasing numbers of subdivisionsessayed: it started with 84% assuming three groups(males, 83.6%; females, 83.8%) and ended with 53%assuming nine groups (males, 55.3%; females, 51.3%).

Table 3. Loading scores and percentage of total varianceexplained by the first three principal componentsextracted according to the spatial principal componentsanalysis of individual surfaces of variation of morphologi-cal traits from Vipera ammodytes

PC1 PC2 PC3

% of variance 45.8 17.9 16.2VEN 0.37 -0.10 -0.22SC_M 0.36 -0.37 -0.02SC_F 0.21 -0.45 0.29SPR -0.08 0.28 0.61HWI_M -0.31 -0.25 0.32HHE_M 0.19 0.42 0.41SHE_F -0.25 -0.16 -0.05RHE_F -0.40 -0.22 -0.04RWI_M 0.36 -0.05 0.25DES_M -0.26 -0.38 0.14DES_F 0.12 -0.33 0.37BPC 0.35 -0.11 -0.07

Figure 2. Spatial principal component maps for the first (PC1) and second (PC2) most informative axes, spatial principalcomponents analysis (SPCA) map depicting morphological variability accounted for (80%) by the first three axes of Viperaammodytes, and individual subdivision tested with discriminant function analysis assuming three groups that correspondto presently recognized subspecies. Subdivisions were based on a maximum likelihood classification function using asignature file defined by the clustering of individual PC maps (see Material and methods for details). Each symbol in the‘3 groups’ map represents a locality for which morphological data were available.



In fact, the major decrease in average correct classifi-cation occurred from three to four groups (from 84 to68%). The distribution of the three groups was coher-ent, and divided the study area into three mainregions: north-western, central-eastern, and southernareas (Figs 2, 3 groups). These results clearly supportthe case that the scenario with three morphologicalgroups is the putative subdivision that provides thehighest correct classification of specimens. Within thethree-groups scenario, the average correct classifica-tion of specimens was very high for the north-westerngroup (above 95%), but was relatively low for thecentral-eastern group (Table 4). On average, therewere 16% of misclassifications of specimens (sexescombined), which was especially marked in thecentral-eastern group (Table 4). Interestingly, the dis-tribution of the three groups roughly follows the pro-posed distribution for the three subspecies of V.ammodytes currently recognized by Tomovic (2006)using classical morphological analyses (cf. Figs 2, 3

groups with Fig. 1). Therefore, to avoid duplication ofdesignations (subspecies and regions), the three mor-phologically coherent regions are hereafter designatedaccording to the subspecific taxonomy: the north-western group corresponds to V. a. ammodytes;the central-eastern group corresponds to V. a. montan-doni; and the southern group corresponds to V. a.meridionalis.

BIOGEOGRAPHIC PATTERNS IN

MORPHOLOGICAL GROUPS

The maximum entropy models identified a set oftopographical and climatic EGVs explaining the dis-tribution of subspecies (Table 5). The distribution ofsome subspecies was influenced by common EGVs,such as precipitation in the wettest month for V. a.ammodytes and V. a. meridionalis, and, remarkably, allsubspecies are mostly related to precipitation in thedriest month. The analysis of EGV response curves

Figure 3. Response curves for the environmental factors most related to the distribution of subspecies of Viperaammodytes. These variables add, on average, at least 75% of contribution to each individual model.

Table 4. Number and percentage of Vipera ammodytes specimens classified in each group according to the canonicaldiscriminant functions of morphological traits

Group

Correct classification Misclassification Total N

Males Females Average Average (males – females)

North-western 274 (95%) 298 (96%) 96% 4% 288–311Central-eastern 56 (57%) 47 (49%) 53% 47% 98–97Southern 33 (70%) 21 (72%) 71% 29% 47–29Average 84% 84% 84% 16%

Analyses were conducted for males and females separately.



(Fig. 3) by pairs of subspecies reveals that: (1) allpopulations occur in areas with distinct precipitationin the driest month, V. a. ammodytes above 40 mm, V.a. montandoni 20–50 mm, and V. a. meridionalis below10 mm; (2) populations of V. a. ammodytes and V. a.meridionalis occur in areas with high and low precipi-tation in the wettest month, respectively; (3) popula-tions of V. a. montandoni and V. a. meridionalis occurin areas with intermediate and high precipitationseasonality, respectively; and (4) populations of V. a.montandoni but especially V. a. meridionalis occur inareas with low annual temperature range. Therefore,the distribution of subspecies and their contact zonesshould be mostly related to precipitation in the driestmonth (see Fig. S3 for a representation of subspeciesdistribution according to geographical variation inprecipitation in the driest month).

The average AUC for both training and testdata was relatively high for all subspecies (averagetraining, 0.929; average test, 0.885), suggesting areasonable fit for all ecological models (Table 5). Areasof potential occurrence of subspecies (Fig. 4) wereidentified for the three subspecies.

1. Vipera ammodytes ammodytes was found in rela-tively continuous areas along the north-westernregion (from 0 up to 2000 m a.s.l.), ranging fromItaly/Austria to the west–east arc formed along theProkletije and Šar Planina mountain chains, andalong the Južna Morava River valley (Serbia).Suitable and isolated areas are also identified alongthe Stara Planina Mountain chain in Serbia andBulgaria.

2. Vipera ammodytes montandoni was found in frag-mented areas (from 0 up to 1700 m a.s.l.) covering

most of Macedonia, Bulgaria (except the aforemen-tioned mountains), Romania, Turkey, and Greece(including the Olympic and Óssa mountains), withthe exception of coastal areas to the Aegean Sea. Asuitable and isolated area is also suggested for theAlbanese populations coastal to the Adriatic Sea.

3. Vipera ammodytes meridionalis was found in con-tinuous areas (from 0 up to 1100 m a.s.l.) restrictedto Greece, but was also found in fragmented areascovering the Olympic, Óssa, and Calcidic moun-tains along the Aegean Sea.

Areas of probable sympatry between subspecies(Fig. 5) are generally small and restricted, and wereidentified for: (1) V. a. ammodytes–V. a. montandoni inthe west–east arc formed along the Prokletije and ŠarPlanina mountain chains, along the Južna MoravaRiver valley (3.4% of study area), as well as along thecoastal regions of Montenegro, northern Albania andsouthern Croatia; (2) V. a. montandoni–V. a. meridiona-lis in the Olympe, Óssa, and Calcidic mountains ofGreece (1.6% of study area); (3) V. a. ammodytes–V. a.meridionalis in only one pixel on the western coast ofGreece, Jonian Sea (0.0002% of study area). Most ofthe misclassified specimens by the DFA (Table 4) arelocated within areas of probable sympatry or in thevicinities of these areas (Fig. 5): (1) 66% of misclassi-fied specimens between V. a. ammodytes and V. a.montandoni (from a total of 105 misclassified); (2) 76%of misclassified specimens between V. a. montan-doni and V. a. meridionalis (from a total of 25 misclas-sified). The 12 misclassified specimens between V. a.ammodytes and V. a. meridionalis were locatedfar from the two pixels predicted with potentialsympatry.

Table 5. Sample sizes, average and standard deviation (SDs) of training, and test area under the curve (AUC) for the20 maximum entropy models, and average and standard deviation of percentage contribution of each environmental factorfor the models (see Table 1 for variable acronyms)

Vipera ammodytesammodytes

Vipera ammodytesmontandoni

Vipera ammodytesmeridionalis

N training samples 173 80 29N test samples 43 20 7Training AUC (SD) 0.904 (0.009) 0.914 (0.013) 0.968 (0.008)Test AUC (SD) 0.863 (0.022) 0.846 (0.038) 0.947 (0.031)% contribution (SD)ALTI 7.6 (3.1) 8.7 (4.8) 2.7 (1.1)SLOP 6.3 (2.8) 6.9 (3.8) 8.8 (4.8)TANR 3.6 (2.0) 11.4 (4.5)* 12.6 (9.5)*TMIN 8.0 (2.5)* 10.8 (4.3)* 6.5 (3.1)PDRY 53.5 (6.8)* 39.4 (7.5)* 47.0 (16.5)*PWET 13.7 (5.7)* 9.8 (5.0) 11.3 (4.8)*PSEA 7.4 (2.8) 13.0 (5.6)* 11.1 (6.1)*

*Variables that sum at least 75% of contribution on average for each subspecies.



Figure 4. Average probability of occurrence for each subspecies of Vipera ammodytes identified by spatial principalcomponents analysis (SPCA; see Figure 2), and supported by discriminant function analysis, derived by maximum entropymodelling (20 bootstrap models for each subspecies) on a 10 ¥ 10-km scale. Standard deviation of predictions representedin smaller insets.



Figure 5. Location of potential contact zones between subspecies of Vipera ammodytes on a 10 ¥ 10-km scale accordingto maximum entropy modelling and location of misclassified specimens by the discriminant function analysis.



DISCUSSIONSPATIAL PATTERNS IN MORPHOLOGICAL

CHARACTER VARIATION

Geostatistics and GIS procedure was recently success-fully applied for inferring evolutionary scenarios in V.latastei and Vipera monticola from the Iberian Pen-insula and the Maghreb, respectively (Brito et al.,2008), as well as for the identification of environmen-tal correlates that sustain a contact zone between V.latastei, V. aspis and Vipera seoanei (Martínez-Freiríaet al., 2008, 2009).

The main patterns of geographic variation of mor-phological characters for V. latastei and V. monticola(Brito et al., 2008) were in some way surprisinglysimilar to the patterns obtained for V. ammodytes inthis study, e.g. the same environmental factors limitthe distribution of differentiated groups of vipers inboth peninsulas. Meristic traits (e.g. number of scalesin reptiles) show geographic variation along an envi-ronmental factor gradient, and could result fromadaptation for optimal heat and water exchange (e.g.Soule, 1966; Horton, 1972; Malhotra & Thorpe, 1997;Sanders, Malhotra & Thorpe, 2004). Also, morphologi-cal similarity in dorsal colour pattern could be theresult of convergent evolution, e.g. adaptive evolutionto similar environments, and reflect selection pres-sures such as camouflage (Sanders et al., 2004).Undoubtedly, the influence of humidity (e.g. precipi-tation of the driest month) on the variation of theanalysed set of meristic traits in spatially separatedviper taxa from the Iberian and Balkan Peninsulascoincides with their phylogenetic relatedness, butcould be a general phenomenon among snakes, andthus requires more detailed research.

DISCREPANCY BETWEEN GENETIC AND

MORPHOLOGICAL SPATIAL PATTERNS

In this particular case, the overall phenotypic differ-entiation in V. ammodytes (Tomovic, 2006; this study)is not completely congruent with genetic differentia-tion (Ursenbacher et al., 2008), i.e. the pattern ofphenotypic variation does not reflect in detail theevolutionary history of the taxon.

The first splitting of V. ammodytes in the Balkansprobably happened in the early to mid Pliocene (4.1–3.6 Mya), when the ancestral group separated intoseven clades: Montenegrin, north-eastern, north-western, south-western, Cyclades, Peloponnese, andsouth-eastern (Ursenbacher et al., 2008).

Except divergence between the Peloponnesus andthe Cyclade clades, which took place 3.5 Mya, and thebasal splitting within the south-eastern clade thathappened c. 2.7 Mya, all splits within each clade andsubclade took place in the Pleistocene (mainly duringthe last 0.7 Mya).

The Pleistocene glaciations influenced the contrac-tion of the species range mostly in northern parts, soit could be supposed that the previously formed cladesdiverged during the Pleistocene, and that some refu-gial populations persisted in middle Dalmatia, south-ern Montenegro, lowland and coastal Albania, and inGreece. The possible borders for subsequent spread-ing of haplotypes during interglaciation periods andunification of genetic structure of V. ammodytes popu-lations could be the high mountain chains of thecentral part of the Balkans (e.g. Prokletije, ŠarPlanina, Korab, and western part of the StaraPlanina mountains). As most of the individuals thatwere analysed here for external morphology were notexamined for mtDNA haplotypes, and vice versa, itcan be said that only one of the previously definedgenetic groups recognized by Ursenbacher et al.(2008) shows phenotypic plasticity: the southern sub-clade (within the south-eastern clade) includes bothV. a. meridionalis and V. a. montandoni phenotypes.On the contrary, three clades (Montenegrin, north-western, and north-eastern) express the same, V. a.ammodytes, phenotype. The remaining main geneticclades show presence of either V. a. meridionalis(Cyclade and Peloponnese clades) or V. a. montandoniphenotypes (south-western clade and eastern sub-clade within the south-eastern clade), despite theirphylogenetic relatedness, thereby suggesting conver-gent evolution.

BIOGEOGRAPHIC PATTERNS IN

MORPHOLOGICAL GROUPS

The analysed clinal pattern of morphological varia-tion in V. ammodytes is concordant with proposedsubspecies number and distribution (Tomovic, 2006).It is also concordant with the clinal pattern of present(modern) variation in precipitation of the driestmonth (this study). This is expected, as analysedphenotypic traits reflect local adaptation. However,there are some discrepancies that may best beexplained by patterns in the Quaternary of Europeand the Balkans: the climate had been severelychanging from the Tertiary to the end of the Pleis-tocene (Issar, 2003). The climate of the early Pliocenewas warm and humid, but deterioration began in themiddle Pliocene (3.1 Mya), heralding the Quaternarycooling: the most severe climatic oscillations occurredduring the last 700 000 years, with cyclic glacial andinterglacial periods (e.g. Hewitt, 1999, 2004).

The predicted sympatry between V. a. ammodytesand V. a. montandoni phenotypes could be the reflec-tion of the climatic oscillations during the period ofthe last deglaciation and post-Pleistocene period inthe western and central Balkans. During the transi-tion from the Pleistocene to the early Holocene, the



whole Mediterranean region was characterized byhigh or rising lake levels, indicating a wet climatedistinctly different from that of the present day(Roberts et al., 2001). At that time, V. a. ammodytesphenotypes could prevail in the aforementionedareas. But, three periods of aridity occurred duringthe transition from the late-Pleistocene to theHolocene (18.5–16 Kya, 14.4–12.7 Kya, and 11.1–9.7 Kya in the central Mediterranean Sea;Combourieu-Nebout et al., 1998), as well as in themid Holocene (Digerfeldt, Sandgren & Olsson, 2007).Thus, the frequency of V. a. montandoni phenotypesbetter adapted to conditions of low precipitation couldincrease towards the north-west. On the contrary,during the periods of more humid climates (i.e. at thebeginning of the Holocene; Rohling & Hilgen, 1991),V. a. ammodytes phenotypes overwhelmed the areaswhere predicted sympatry with V. a. montandonioccurs (coastal areas of Eastern Adriatic, northernAlbania, the Valley of Južna Morava River in Serbia,and along the Stara Planina Mountain chain inSerbia and Bulgaria).

According to geographic variation of environmentalvariables, the valley of the lower Vardar River inMacedonia has suitable habitat conditions for V. a.meridionalis phenotypes, but currently they are notpresent there. However, four specimens that camefrom eastern Macedonia were classified by the DFAas V. a. meridionalis. Similarly, during the earlyHolocene, the climate of the southern part of theBalkan peninsula (Peloponnesus and mainlandGreece) had altered several times, from cold andhumid to warm and dry, and vice versa (Issar, 2003,and references therein). It could be supposed thatduring those dry and warm periods, V. a. meridionalisphenotypes, being more adapted to conditions ofextremely low precipitation, spread towards thenorth; but, during the more humid and cold periods,they were overwhelmed by V. a. montandoni pheno-types. Areas of predicted sympatry between these twophenotypes (Olympe–Óssea and the Calcidic Moun-tains of Greece) probably harbour diverse microhabi-tats that have been suitable for the coexistence ofboth V. a. meridionalis and V. a. montandoni since theonset of the Holocene (Tzedakis et al., 2002).

The most intriguing zone is the central part of theBalkans (i.e. Macedonia). It is a highly differentiatedzone in PC2 and SPCA, and breaks the cline fromnorth-west to south (Fig. 2). Also, it has the predictedoccurrence of V. a. montandoni phenotypes, but a verylow occurrence of V. a. meridionalis: there are noareas of sympatry predicted in the region (Fig. 5).Macedonia might have been part of the historicalrange of all three forms. It is likely that V. a.ammodytes phenotypes entered the area from thewest (mountainous part) during periods of wet

climate (i.e. the beginning of the Holocene, and againfrom the late Holocene to the present day; Rohling &Hilgen, 1991; Roberts et al., 2001). This phenotypecurrently inhabits the western, mountainous part ofthe country, characterized with moist climatic condi-tions. Possibly the V. a. montandoni phenotype occu-pied Macedonia in the periods of drier climate (midHolocene; Roberts et al., 2001; Wright et al., 2003);currently it inhabits the central and eastern parts ofthe country, which are characterized by a dry climatetoo. The V. a. meridionalis phenotype, which isadapted to extreme dryness, currently does notinhabit Macedonia. Historically, this phenotype couldonly prevail in the area in periods of extreme dryness,as it has been proven that the north-western part(coastal area, west of Pindus Mountain) of Greece hada more humid climate, even during the mid Holocene,compared with the eastern part (continental part,east of Pindus Mountain) (Tzedakis, 1993; Tzedakiset al., 2002; Lawson et al., 2004). Thus, the Ionian Seaand the Pindus Mountain would act as barriers, lim-iting gene flow that might have originated from rangeexpansions of the clades, and preserving accumulateddifferences that originated during the periods of iso-lation (Tzedakis et al., 2002).

The results of this study do not provide enoughinformation about the extent of species genetic poten-tial for observed phenotypic plasticity in externalmorphology. But, the discrepancies in the recent dis-tribution of adaptive phenotypes in Macedonia couldbe explained by the dynamic action of selectiveregimes in different directions, coupled with popula-tion bottlenecks where some portion of genetic vari-ability was lost. As a result, today there are ‘relict’phenotypes that obviously cope with contemporaryclimatic conditions. Although V. ammodytes is activefrom early spring to late autumn, it has been proventhat there are differences in the activity pattern ofmales and females in different seasons (i.e. in springand summer): males tend to be more active in spring(because of their greater energetic costs for reproduc-tive activities), whereas females are more active inthe summer season during the gestation period(Crnobrnja-Isailovic et al., 2007).

Future work should focus on research of environ-mental patterns of both the genetic and the morpho-logical variability of individuals from populations inzones of predicted sympatry, in order to confirm orreject the hypothesis suggested in this study.

CONCLUSIONS

This study, based on geostatistics and the GIS proce-dure used for the identification of environmental cor-relates that sustain a contact zone among differentsubspecies, indicated that the same environmental



factors limit the distribution of differentiated groupsof related taxa that geographically more-or-lessreplace each other (for comparison see Brito et al.,2008; Martínez-Freiría et al., 2008, 2009).

The same techniques should be applied for studyingother parapatric forms and regions (e.g. the ApenninePeninsula), to test whether the same environmental(and possibly palaeoenvironmental) factors couldhave an influence on the evolutionary history, mor-phological differentiation, intraspecific taxonomy, andgeographic distribution of different animal groups inrefugial regions of Europe (Hewitt, 1999).

This study is concordant with convergent evolutiondriving morphological similarity rather than phyloge-netic proximity (Sanders et al., 2004). Thus, the originsof similarity in external morphology, especially in scalenumber and/or dorsal colour pattern in squamate taxa,should be analysed with caution in order to avoidmisinterpretations in taxonomic revisions.

ACKNOWLEDGEMENTS

This study was partially supported by the projectPOCTI/BIA-BDE/55596/2004 from Fundação para aCiência e Tecnologia (FCT, Portugal). J.C.B. has acontract (Programme Ciência 2007) from FCT. J.C.I.was funded by DAAD grant ref. 324/jo-Yu. J.C.I. andL.J.T. were partially funded by MNTR Republic ofSerbia grant no. 143040. We would like to thank thecurators and staff of the following natural historymuseums and institutes for their permission toanalyse the specimens of their collections: Zoologis-ches Forschungsinstitut und Museum ‘AlexanderKoënig’ (Bonn), Naturhistorisches Museum (Vienna),Museums of Natural History (Ljubljana, Belgrade,Skopje), Land Museum (Sarajevo), Institute for Bio-logical Research (Belgrade), Institute of Biology, Uni-versity of Priština, National Museum of the BulgarianAcademy of Sciences (Sofia). We give special thanks toI. Krizmanic and R. Ajtic for providing specimens fromtheir private collections.

REFERENCES

Arnold E, Ovenden D. 2002. A Field Guide to the Reptilesand amphibians of Britain and Europe, 2nd edn. London:Collins.

Báez M, Brown RP. 1997. Testing multivariate patterns ofwithin-island differentiation in Podarcis dugesii fromMadeira. Journal of Evolutionary Biology 10: 575–587.

Brito JC, Acosta AL, Álvares F, Cuzin F. 2009. Biogeog-raphy and conservation of taxa from remote regions: anapplication of ecological-niche based models and GIS toNorth-African Canids. Biological Conservation 142: 3020–3029.

Brito JC, Santos X, Pleguezuelos JM, Sillero N. 2008.Inferring Evolutionary Scenarios with Geostatistics andGeographical Information Systems (GIS) for the viperidsnakes Vipera latastei and V. monticola. Biological Journalof the Linnean Society 95: 790–806.

Combourieu-Nebout N, Paterne M, Turon J-L, Siani G.1998. A high-resolution record of the last deglaciation in thecentral Mediterranean Sea: palaeovegetation and palaeohy-drological evolution. Quaternary Science Review 17: 303–317.

Crnobrnja-Isailovic J. 2007. Cross-section of a refugium:genetic diversity of amphibian and reptile populations inthe Balkans. In: Weiss S, Ferrand N, eds. Phylogeography ofSouthern European Refugia: Evolutionary perspectives onthe origins and conservation of European biodiversity. Neth-erlands: Springer, 327–337.

Crnobrnja-Isailovic J, Ajtic R, Tomovic L. 2007. Activitypatterns of the sand viper (Vipera ammodytes) from thecentral Balkans. Amphibia-Reptilia 28: 582–589.

Digerfeldt G, Sandgren P, Olsson S. 2007. Reconstructionof Holocene lake-level changes in Lake Xinias, centralGreece. Holocene 17: 361–367.

Elith J, Graham CH, Anderson RP, Dudyk M, Freer S,Guisan A, Hijmans RJ, Huettmann F, Leathwick JR,Lehmann A, Li J, Lohmann LG, Loiselle BA, ManionG, Moritz C, Nakamura M, Nakazawa Y, McOverton J,Peterson AT, Phillips S, Wisz MS, Zimmermann NE.2006. Novel methods improve prediction of species’ distri-butions from occurrence data. Ecography 29: 129–151.

ESRI. 2006. Arcmap 9.2. USA: Environmental SystemsResearch Institute, Inc.

Fielding AH, Bell JF. 1997. A review of methods for theassessment of prediction errors in conservation presence/absence models. Environmental Conservation 24: 38–49.

Garrigues T, Dauga C, Ferquel E, Choumet V, FaillouxA-B. 2005. Molecular phylogeny of Vipera Laurenti, 1768and the related genera Macrovipera (Reuss, 1927) andDaboia (Gray, 1842), with comments about neurotoxicVipera aspis aspis populations. Molecular Phylogenetics andEvolution 35: 35–47.

Hernandez PA, Graham CH, Master LL, Albert DL. 2006.The effect of sample size and species characteristics onperformance of different species distribution modelingmethods. Ecography 29: 773–785.

Hewitt GM. 1999. Post-glacial re-colonization of Europeanbiota. Biological Journal of the Linnean Society 68: 87–112.

Hewitt GM. 2004. Genetic consequences of climatic oscilla-tions in the Quaternary. Philosophical Transactions of theRoyal Society, London, Series B 359: 183–195.

Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A.2005. Very high resolution interpolated climate surfaces forglobal land areas. International Journal of Climatology 25:1965–1978. Available at http://www.worldclim.org/

Hoffmann MH, Glab AS, Tomiuk J, Schmuths H, FritschRM, Bachmann K. 2003. Analysis of molecular data ofArabidopsis thaliana (L.) Heynh. (Brassicaceae) with Geo-graphical Information Systems (GIS). Molecular Ecology 12:1007–1019.



Horton DR. 1972. Lizard scales and adaptation. SystematicZoology 21: 441–443.

Hugall A, Moritz C, Moussalli A, Stanisic J. 2003. Recon-ciling paleodistribution models and comparative phylogeog-raphy in the Wet Tropics rainforest land snail Gnarosophiabellendenkerensis (Brazier 1875). Proceedings of theNational Academy of Sciences of the United States ofAmerica 99: 6112–6117.

Issar AS. 2003. Climate changes during the Holocene andtheir impact on Hydrological Systems. Cambridge: Cam-bridge University Press.

Johnston K, Ver Hoef JM, Krivoruchko K, Lucas N.2001. Using arcgis Geostatistical Analyst. EnvironmentalSystems Research Institute, Inc.

Kidd DM, Ritchie MG. 2000. Inferring the patterns andcauses of geographic variation in Ephippiger ephippiger(Orthoptera, Tettigoniidae) using Geographical InformationSystems (GIS). Biological Journal of the Linnean Society 71:269–295.

Kidd DM, Ritchie MG. 2006. Phylogeographic informationsystems: putting the geography into phylogeography.Journal of Biogeography 33: 1851–1865.

Knouft JH, Losos JB, Glor RE, Kolbe JJ. 2006. Phyloge-netic analysis of the evolution of the niche in lizards of theAnolis sagrei group. Ecology 87: S29–S38.

Knowles LL, Carstens BC, Keat ML. 2007. Couplinggenetic and ecological-niche models to examine how pastpopulation distributions contribute to divergence. CurrentBiology 17: 940–946.

Lawson I, Frogley M, Bryant C, Preece R, Tzedakis P.2004. The Lateglacial and Holocene environmental historyof the Ioannina basin, north-west Greece. QuaternarySciences Review 23: 1599–1625.

Lenk P, Kalyabina S, Wink M, Joger U. 2001.Evolutionary relationships among the True Vipers(Reptilia: Viperidae) inferred from mitochondrial DNAsequences. Molecular Phylogenetics and Evolution 19:94–104.

Malhotra A, Thorpe RS. 1997. Microgeographic variationin scalation of Anolis oculatus (Dominica, West Indies):a multivariate analysis. Herpetologica 53: 49–62.

Martínez-Freiría F, Santos X, Pleguezuelos JM, LizanaM, Brito JC. 2009. Geographical patterns of morphologicalvariation and environmental correlates in contact zones: amulti-scale approach using two Mediterranean vipers.Journal of Zoological Systematics and EvolutionaryResearch 47: 357–367.

Martínez-Freiría F, Sillero N, Lizana M, Brito JC. 2008.GIS-based niche models identify environmental correlatessustaining a contact zone between three species of Europeanvipers. Diversity and Distributions 14: 452–461.

Olivier MA. 1990. Kriging: a method of interpolation forgeographical information systems. International Journal ofGeographic Information Systems 4: 313–332.

Phillips SJ, Anderson RP, Schapire RE. 2006. Maximumentropy modelling of species geographic distributions.Ecological Modelling 190: 231–259.

Phillips SJ, Dudík M. 2008. Modelling of species distribu-

tions with Maxent: new extensions and a comprehensiveevaluation. Ecography 31: 161–175.

Raes N, Roos MC, Slik JWF, Van Loon EE, ter Steege H.2009. Botanical richness and endemicity patterns of Borneoderived from species distribution models. Ecography 31:180–192.

Roberts N, Reed JM, Leng MJ, Kuzucuoglu C, FontugneM, Berttaux J, Woldring H, Bottem S, Black S, Hunt E,Karabtytkoglu M. 2001. The tempo of Holocene climaticchange in the eastern Mediterranean region: new high-resolution crater-lake sediment data from central Turkey.Holocene 11: 721–736.

Rohling EJ, Hilgen FJ. 1991. The eastern Mediterraneanclimate at times of sapropel formation: a review. Geologie enMijnbouw 70: 253–264.

Saint-Girons H. 1978. Morphologie externe comparée etsystématique des vipéres d’Europe (Reptilia, Viperidae).Revue Suisse de Zoologie 85: 565–595.

Sanders KL, Malhotra A, Thorpe RS. 2004. Ecologicaldiversification in a group of Indomalayan pitvipers (Trim-eresurus): convergence in taxonomically important traitshas implications for species identification. Journal of Evo-lutionary Biology 17: 721–731.

Santos X, Brito JC, Sillero N, Pleguezuelos JM, LlorenteGA, Fahd S, Parellada X. 2006. Inferring habitat-suitability areas with ecological modelling techniques andGIS: a contribution to assess the conservation status ofVipera latastei. Biological Conservation 130: 416–425.

Soule M. 1966. Trends in the insular radiation of a lizard.American Naturalist 100: 47–64.

Spear SF, Peterson CR, Matocq MD, Storfer A. 2005.Landscape genetics of the blotched tiger salamander (Amby-ostoma tigrinum melanostictum). Molecular Ecology 14:2553–2564.

StatSoft Inc. 2003. STATISTICA (data analysis softwaresystem), version 6.0. Tulsa: StatSoft Inc.

Swenson NG. 2006. GIS-based niche models reveal unifyingclimatic mechanisms that maintain the location of avianhybrid zones in a North American suture zone. Journal ofEvolutionary Biology 19: 717–725.

Swenson NG. 2008. The past and future influence of geo-graphic information systems on hybrid zone, phylogeo-graphic and speciation research. Journal of EvolutionaryBiology 21: 421–434.

Szyndlar Z, Böhme W. 1993. Die fossilen Schlangen Deut-schlands: geschichte der Faunen und ihrer Erforschung.Mertensiella 3: 381–431.

Taberlet P, Fumagalli L, Wust-Saucy A-G, Cosson CJ-F.1998. Comparative phylogeography and postglacial coloni-zation routes in Europe. Molecular Ecology 7: 453–464.

Thorpe RS. 1987a. Geographic variation: a synthesis ofcause, data, pattern and congruence in relation to subspe-cies, multivariate analysis and phylogenesis. Bolletino diZoologia 54: 3–11.

Thorpe RS. 1987b. Complex clines: the predictivity of com-plicated patterns of geographic variation portrayed by mul-tivariate analysis. Biological Journal of the Linnean Society31: 75–88.



Thorpe RS, Brown RP, Malhotra A, Wüster W. 1991.Geographic variation and population systematics: distin-guishing between ecogenetics and phylogenetics. Bolletinodi Zoologia 58: 329–335.

Tomovic L. 2006. Systematics of the nose-horned viper(Vipera ammodytes Linnaeus, 1758). Herpetological.Journal 16: 191–201.

Tomovic L, Džukic G. 2003. Geographic variability andtaxonomy of the nose-horned viper, Vipera ammodytes (L.1758), in the central and eastern parts of the Balkans: amultivariate study. Amphibia-Reptilia 24: 359–377.

Tomovic L, Radojicic J, Džukic G, Kalezic ML. 2002.Sexual dimorphism of the sand viper (Vipera ammodytes L.)from the central part of Balkan Peninsula. Russian Journalof Herpetology 9: 69–76.

Tzedakis P. 1993. Long-term tree populations in northwestGreece through multiple Quaternary climatic cycles. Nature364: 437–440.

Tzedakis PC, Lawson IT, Frogley MR, Hewitt GM,Preece RC. 2002. Buffered tree population changes in aQuaternary refugium: evolutionary implications. Science297: 2004–2047.

Ursenbacher S, Conelli A, Golay P, Monney J-C, ZuffiMAL, Thiery G, Durand T, Fumagalli L. 2006. Phylo-geography of the asp viper (Vipera aspis) inferred frommitochondrial DNA sequence data: evidence for multipleMediterranean refugial areas. Molecular Phylogenetics andEvolution 38: 546–552.

Ursenbacher S, Schweiger S, Tomovic L, Crnobrnja-Isailovic J, Fumaggali L, Mayer W. 2008. Molecularphylogeography of the nose-horned viper (Viperaammodytes, Linnaeus (1758)): evidence for high geneticdiversity and multiple refugia in the Balkan Peninsula.Molecular Phylogenetics and Evolution 46: 1116–1128.

USGS. 2006. Shuttle Radar Topography Mission (SRTM):mapping the world in 3 dimensions. United States Geo-logical Survey. Available at http://srtm.usgs.gov/index.html

Wright HE, Ammann B, Stefanova I, Atanassova J,Margalitadze N, Wick L, Blyakharchuk T. 2003. Late-glacial and early-holocene dry climates from the Balkanpeninsula to Southern Siberia. In: Tonkov S, ed. Aspects ofpalynology and palaeoecology. Sofia-Moscow: Pensoft, 127–136.

SUPPORTING INFORMATION

Additional supporting information may be found in the online version of this article:

Figure S1. Spatial variation patterns of morphological traits from Vipera ammodytes.Figure S2. Spatial principal component map of morphological variability of Vipera ammodytes (SPCA) com-bining the three most explicative axes and population subdivision tested with discriminant function analysis(DFA).Figure S3. Distribution of the Vipera ammodytes subspecies according to the variation in the averageprecipitation of the driest month in the study area.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materialssupplied by the authors. Any queries (other than missing material) should be directed to the correspondingauthor for the article.



Documents

The use of geostatistics and GIS for evolutionary history ...webpages.icav.up.pt/ptdc/bia-bic/118151/2010/tomovic...LJILJANA TOMOVIC´1,2*, JELKA CRNOBRNJA-ISAILOVIC´2,3 and JOSÉ