Journal of Biogeography (J. Biogeogr.) (2016)

ORIGINAL Differentiation in the marbled white ARTICLE butterfly complex driven by multiple evolutionary forces Jan Christian Habel1*, Roger Vila2, Raluca Voda2,3, Martin Husemann4, Thomas Schmitt5,6 and Leonardo Dapporto2,7

1Terrestrial Ecology Research Group, ABSTRACT Department of Ecology and Ecosystem Aim Genetic and phenotypic data may show convergent or contrasting spatial Management, School of Life Sciences patterns. Discrepancies between markers may develop in response to different Weihenstephan, Technische Universit€at Munchen,€ D-85350 Freising, Germany, evolutionary forces. In this study we analyse inter- and intraspecific differentia- 2Institut de Biologia Evolutiva (CSIC- tion of closely related taxa in the marbled white butterfly species group. Based Universitat Pompeu Fabra), ESP-08003 on genetic and phenotypic characters we test for potential evolutionary drivers Barcelona, , 3Department of Life and propose a taxonomic revision. Sciences and Systems Biology, University of Location Western Palaearctic (including north-western ). Turin, I-10123 Turin, Italy, 4Entomology Department, Centrum fur€ Naturkunde, Methods We compared distributions of mitochondrial cytochrome c oxidase University of Hamburg, D-20146 Hamburg, subunit I gene (COI) sequences, of several allozyme loci, and of the shape of Germany, 5Senckenberg German wings and genitalia obtained by applying landmark-based techniques for the Entomological Institute, D-15374 Muncheberg,€ three butterfly species galathea (central and eastern Europe), Germany, 6Zoology, Institute for Biology, M. lachesis (Iberia) and M. lucasi (North Africa). Faculty of Natural Sciences I, Martin Luther University Halle-Wittenberg, D-06099 Halle, Results All studied markers showed a strong spatial structure, although dis- Germany, 7Department of Biology, University cordance among their patterns was detected. COI sequences, wing shape and of Florence, I-50019 Florence, Italy genitalia indicated a main split between M. galathea and M. lucasi. A lower dif- ferentiation between M. galathea and M. lachesis was found in wing shape and reflected in two mutations of the COI gene, while allozymes indicated a strong divergence. Within M. galathea, allozyme data and COI, but not morphology, revealed the existence of a slightly differentiated lineage in the Italian Penin- sula, France and Switzerland. Based on COI, Melanargia lucasi was split into two subgroups, a western and an eastern Maghreb lineage. Main conclusions Long-term isolation of Melanargia populations between North Africa and Europe led to divergence between M. galathea and M. lucasi. This was followed by a recent differentiation among populations isolated dur- ing the cold periods of the Pleistocene, such as M. lachesis in Iberia. These lin- eages are characterized by a tendency not to overlap in secondary sympatry. The different patterns of the four markers may arise from divergent evolution- *Correspondence: Jan Christian Habel, ary processes and pressures: wings may be mainly affected by natural selection, Terrestrial Ecology Research Group, genital structures by sexual selection, whereas long-term isolation and drift Department of Ecology and Ecosystem may have driven divergence of mitochondrial DNA and allozymes. Management, School of Life Sciences Weihenstephan, Technische Universit€at Keywords Munchen,€ Hans-Carl-von-Carlowitz-Platz 2, biogeography, geographic isolation, geometric morphometrics, parapatric D-85350 Freising, Germany. E-mail: [email protected] differentiation, refugia, reproductive isolation, sexual selection

flow typically established by dispersal constraints, climatic INTRODUCTION oscillations and/or tectonic events (Hewitt, 2004; Mittelbach Differentiation of genetic lineages and their diversification & Schemske, 2015). The persistence of such barriers is often into distinct species is a process predominantly occurring in only temporal and populations can disperse from their origi- allopatry and determined by the presence of barriers to gene nal areas, resulting in secondary contact between different

ª 2016 John Wiley & Sons Ltd http://wileyonlinelibrary.com/journal/jbi 1 doi:10.1111/jbi.12868 J. C. Habel et al. genetic lineages or taxa. The establishment of secondary sym- MATERIALS AND METHODS patry is a fundamental process accruing local diversity (Mit- telbach & Schemske, 2015). With growing availability of Study species large genetic data sets, it is becoming evident that the main- tenance of allopatry among species is not only determined The marbled white butterfly M. galathea is widely distributed by physical and ecological barriers and that most terrestrial in southern and central Europe. Its northern distribution taxa show a delay to secondary sympatry slowing down the margin runs through central England, Belgium, northern Ger- accumulation of species in communities (Pigot & Tobias, many and reaches the Baltic Sea in Poland. In Iberia, M. 2015). Different mechanisms have been advocated to explain galathea only occurs in a restricted area south of the Pyrenees, the lack of overlapping distributions among closely related while in the rest of the Iberian Peninsula, it becomes replaced species and lineages; these include density-dependent pro- by its close relative M. lachesis (Kudrna et al., 2015). Mela- cesses, competition for similar resources, reproductive inter- nargia lachesis occurs all across Iberia; North of the Pyrenees, ference and environmental preferences (Waters, 2011; Voda the taxon is restricted to the Mediterranean parts of the et al., 2015a,b). On the other hand, delay to secondary sym- French Roussillon (Bozano, 2002), where it locally overlaps patry may facilitate the last stages of the speciation process with M. galathea. In this potential area of syntopy, the distri- by reducing gene flow before reproductive barriers have been bution of the two taxa remains mostly micro-allopatric apart fully established. from some few areas where the two species hybridize (Dennis Speciation is also related to particular life history traits of et al., 1991; Fernandez-Rubio, 1991). Melanargia lucasi is species, such as philopatry and low dispersal ability (Clara- endemic to the hill and mountain areas of the Maghreb munt et al., 2012), hybridization and introgression (Mallet, region (, and ) (Bozano, 2002). 2007) and fast responses to particular ecological pressures and M. lucasi had recently been suggested (Wade, 2001). These factors may be differently reflected in to be classified as separate species based on the divergence in phenotypic and genotypic traits within and between species. mitochondrial markers (COI and 16S) and male genitalia Accordingly, the identification of a diversified population as (Nazari et al., 2010). Previously, these taxa were considered to a different species can be a difficult exercise. Hence, modern be distinct populations of M. galathea, supported by a high simi- integrative and biogeography should be based on larity in allozyme patterns (Habel et al., 2011) and mirrored by the comparison of results obtained from various markers to the high similarity of the wingless nuclear gene (Nazari et al., fully understand the evolutionary patterns and the degree of 2010). diversification in species groups. Based on DNA barcoding data, M. lachesis and M. galathea In this study, we perform molecular and morphological are only slightly diverged at the COI locus, and thus, species analyses on three closely related Nymphalid butterfly taxa status may not be justified for M. lachesis (Dinca et al., 2015). forming the Melanargia galathea (Linnaeus, 1758) species Nevertheless, these two taxa are phenotypically distinct (wing group in the Western Palaearctic. For these analyses, we col- patterns) and show a divergence of about 2.5% in the nuclear lected individuals from populations spread over major parts wingless gene (Nazari et al., 2010). Furthermore, molecular of its Western Palaearctic distribution range, encompassing analyses based on allozyme polymorphisms showed that differ- two different continents (North Africa and Europe) and a ent genetic groups of M. galathea exist in central Europe and large island (Sicily) located in the Mediterranean Sea. We the Balkans (Schmitt et al., 2006; Habel et al., 2011). Similarly, applied two molecular markers (mitochondrial DNA M. lucasi has been found to be subdivided into a western and (mtDNA) and allozyme polymorphisms) and two phenotypic an eastern lineage in the Maghreb (Habel et al., 2011). markers (wing and genitalia shape). This set of markers was selected to disentangle divergent selection and evolutionary regimes shaping inter- and intraspecific differentiation inde- Data sets pendently from each other: the divergence of mitochondrial All allozyme data were taken from previous studies (Schmitt DNA, a rather conservative marker is assumed to be mainly et al., 2006; Habel et al., 2011). Specimens for genitalia and driven by geographic isolation, while allozyme polymor- wing analyses were taken from the Siegbert Wagener collec- phisms may also be affected by environmental selection tion, access kindly provided by the Alexander Koenig (Voda et al., 2015a,b); the two phenotypic traits mainly Museum (Bonn, Germany), from the Museum of Zoology depend on divergent evolutionary drivers, with wing shapes and Natural History of the University of Florence (Italy) and being strongly affected by environmental conditions and gen- from Roger Vila’s tissue collection, which was further used italia structures are assumed to be mainly influenced by for newly generated COI sequences. sexual selection. These four markers were used to study potentially divergent evolutionary drivers (e.g. geographic, ecological and sexual isolation and selection), which may Cytochrome c oxidase subunit I (COI) have produced genetic and phenotypic variation across the M. galathea group. We further review the current taxonomic In total, 223 COI sequences were analysed, of which 132 rep- status of the M. galathea species, according to our results. resent newly generated data. Details on samples and

2 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex sampling localities are provided as Table S1 in Appendix S1. conservative burn-in of 500,000 generations was applied for Total genomic DNA was extracted for nine of these speci- each run after checking Markov chain Monte Carlo mens using Chelex 100 resin (100–200 mesh, sodium form; (MCMC) for convergence through graphically monitoring Bio-Rad Laboratories GmbH, CA, USA) under the following likelihood values in tracer 1.6 (available at: http://beast. protocol: one leg was removed and introduced into 100 lL bio.ed.ac.uk/Tracer). Independent runs were combined in of 10% Chelex solution; and 5 lL of Proteinase K logcombiner 1.8.0, as implemented in beast, and all (20 mg mL 1) was added. The samples were incubated over- parameters were analysed using tracer 1.6 to determine night at 55 °C and were subsequently incubated at 100 °C whether they had also reached stationarity. Tree topologies for 15 min. Samples were then centrifuged for 10 s at were assessed using treeannotator 1.8.0 in the beast 3000 rpm. A 658 bp fragment at the 50 end of the mitochon- package to generate a tree with median node heights. drial COI gene was amplified by polymerase chain reaction figtree 1.4.2 (available at: http://beast.bio.ed.ac.uk/FigTree) (PCR) using the primers LepF1 (50-ATTCAACCAATCATAA was used to visualize the tree along with node posterior AGATATTGG-30) and LepR1 (50-TAAACTTCTGGATGTCC probabilities and age deviations. AAAAAATCA-30) (Hebert et al., 2004). PCR was performed in 25 lL volume reactions containing 14.4 lL autoclaved Allozymes Milli-Q water, 5 lL59 buffer, 2 lL25mm MgCl2, 0.5 lL 10 mm dNTPs, 0.5 lL of each primer (10 lm), 0.1 lL Taq A total of 15 allozyme loci were analysed for 1158 individuals DNA Polymerase (Promega; 5 U lL 1, Madison, WI, USA) from 32 populations of the three taxa (23 populations of and 2 lL of extracted DNA. The typical thermal cycling profile M. galathea, 1 population of M. lachesis and 8 populations of was: denaturation at 92 °C for 60 s, followed by five cycles of M. lucasi). Additional information about sampling sites and 92 °C for 15 s, 49 °C for 45 s and 62 °C for 150 s, and by 35 sample sizes are given in Table S2 in Appendix S2. Further cycles of 92 °C for 15 s, 52 °C for 45 s and 62 °C for 150 s information on the analytical procedures (allozyme loci anal- and a final extension at 62 °C for 420 s. PCR products were ysed and respective running conditions) is given in Habel purified and sequenced by Macrogen Inc. (Amsterdam, the et al. (2011). Based on the raw data, we calculated pairwise

Netherlands). The remaining 123 sequences generated for this FST values with Arlequin v. 3.0 (Excoffier et al., 2005). study were obtained from the Biodiversity Institute of Ontario, Canada. In this case, a glass fibre protocol (Ivanova et al., Wing shape 2006) was employed to extract DNA; PCR and DNA sequenc- ing were carried out following standard DNA barcoding proce- We analysed the wing patterns of 347 male individuals (M. dures for (deWaard et al., 2008). To examine galathea, N = 248; M. lachesis, N = 41; M. lucasi, N = 58) patterns of genetic variation, we constructed a haplotype net- from 66 sites covering major parts of the Western Palaearctic work with PopART (http://popart.otago.ac.nz) using the TCS range (see Table S3 in Appendix S3). Standardized digital algorithm (Clement et al., 2000). Several loops indicating images were taken of all butterflies’ fore and hind wings ambiguous connections were broken taking into account fre- using a Canon M1 digital camera (50 mm lens). Shapes of quency and genetic distance criteria (Excoffier & Langaney, the left fore and hind wings were quantified by 13 homolo- 1989). gous landmarks on each wing (26 in total). To ensure repeatability and to minimize measurement errors, land- marks were exclusively placed on wing vein intersections or Phylogenetic inference and dating locations where a wing vein meets the edge of the wing. Bayesian inference (BI) on the basis of COI haplotypes was Landmarks were digitized using the program Tps-dig 2.12 employed to infer phylogenetic relationships and estimate (Rohlf, 2008). A schematic overview of all landmarks on the divergence times, with beast 1.8.0 (Drummond et al., 2012). wings is given as Fig. S1 in Appendix S4. As outgroup, we used published COI sequences of four other Generalized procrustes analysis (GPA) was applied to the species in the : M. larissa, M. russiae, M. occitanica and landmark data in order to remove variation in scale and ori- M. ines. jModeltest 2.1.4 (Darriba et al., 2012) was entation and to superimpose the objects in a common coor- employed to select the best-fitting DNA substitution model dinate system (Adams et al., 2004). Partial warps were according to the Akaike information criterion (AIC). As a calculated using the shape residuals from GPA. By applying result, the GTR+I+G model was used. The gamma distribu- principal components analyses (PCA) to partial warps, rela- tion was estimated automatically from the data using six rate tive warps (PCs) were obtained and used as variables in a categories. A fixed substitution rate prior of 0.0115 per site partial least squares discriminant analysis (PLSDA) (Mit- per My was used, based on that estimated for the entire teroecker & Bookstein, 2011). mitochondrial genome of several (Brower, 1994). An uncorrelated relaxed clock (Drummond et al., 2012) and Genitalia a constant population size under a calibrated Yule model were established as priors. Two independent chains were run A total of 123 males (M. galathea, N = 75; M. lachesis, for 50 million generations each, sampling every 1000 steps. A N = 25; M. lucasi, N = 23) were examined. An overview of

Journal of Biogeography 3 ª 2016 John Wiley & Sons Ltd J. C. Habel et al. all specimens and sample sites is given as Table S4 in the other markers. Single specimens for each marker were Appendix S5. Genitalia were dissected using standard proce- finally rotated by using the same parameters resulting in a dures and the tegumen and valves were photographed using model based on shared populations. Procrustes analyses were a Nikon Coolpix 4500 camera, mounted on a binocular carried out by using the ‘recluster.procrustes’ function of the microscope. The number of distal processes on the valves R package recluster (Dapporto et al., 2014). Note that the (cornuti) was counted. Furthermore, a combination of land- selection of COI as a reference configuration for all Pro- marks and sliding semi-landmarks (Bookstein, 1997) was crustes analyses does not affect the overall results; moreover, applied to the tegumen (13) and valva (12) outline: 4 points this analysis does not change the diversity patterns inside on the outlines of both tegumen and valves that could be each configuration, but only rotates and translates them to precisely identified in all individuals were considered as land- minimized arbitrary differences in location and orientation. marks (type II and type III landmarks, Bookstein, 1997), The correlations between the population configurations of all whereas the other points were allowed to slide along the out- pairs of markers were tested with the ‘protest’ function of line trajectory. GPA and PLSDA were applied in similar ways the vegan package for R. Finally, the bi-dimensional configu- as described for the wing shape. rations for specimens were projected in the RGB colour space by using the recluster package (Dapporto et al., 2013). This procedure attributes red, yellow, green and blue colours Overall representation of distribution patterns to each corner of a bi-dimensional representation and the In order to establish whether the species boundaries are cor- colour of each point in the graph is interpolated among rect, we tested for correspondence among the COI structure, these extreme values. As a consequence, each point in a RGB allozyme patterns, taxonomic classification and phenotypic colour space receives a unique colour according to its posi- patterns. For each marker, we obtained bi-dimensional ordi- tion with more similar cases receiving more similar colours nations of the variation among specimens. For the dissimi- (Kreft & Jetz, 2010). Specimens belonging to the same spe- larity matrices based on p-distances for COI and on FST for cies and the same area of 2 9 2 degrees for latitude and lon- allozymes, we applied principal coordinate analysis (PCoA) gitude were grouped and their individual RGB colours were reducing the dimensionality of the original matrices to two plotted on a map as pie charts. For COI, in order to high- dimensions. To the relative warps obtained by the analyses light possible patterns of mutual exclusion to a smaller spa- of wings and genitalia, we applied partial least squares dis- tial scale, we also aggregated the specimens to 0.5 9 0.5 criminant analysis (PLSDA) using species as grouping vari- degree rectangles. able that returned a bi-dimensional representation of the variation among specimens. PLSDA also identified which RESULTS shape variables are responsible for the differentiation among taxa; these differences were visualized by thin plate splines. Genotypes For wings and genitalia, we evaluated the degrees of diversifi- cation among species as a percentage of specimens that can We detected two main groups based on the COI sequences: be blindly attributed to their taxon by applying a Jackknife a European cluster with M. galathea and M. lachesis and a algorithm classifying each specimen individually. PLSDA and North African cluster with M. lucasi (Fig. 1a–e), separated by Jackknife analyses were carried out with the ‘plsda’ and ‘pre- 25 mutations (3.8% divergence) (Fig. 1e). The TCS haplo- dict’ functions of the mixOmics R package (L^e Cao et al., type network showed a clear geographic structure within 2011). The bi-dimensional representations obtained for dif- each of the two main groups (Fig. 1e). However, when the ferent markers are not directly comparable as they have arbi- PCoA configuration was projected in the RGB space, the trary orientation and scale. Moreover, the cases (specimens) genetic structure within clades was less evident (Fig. 1a). are not the same for different markers. To allow a direct Therefore, we analysed the two groups separately (Fig. 1b– comparison among the patterns of different markers, we d). Melanargia lucasi was subdivided into two lineages sepa- applied the method described by Dapporto et al. (2014). We rated by a minimum of eight mutations (1.2%): the first lin- aggregated specimens into populations on the basis of their eage was represented by specimens from Morocco, while the membership to the same species and to the same square of second lineage contained specimens sampled in Algeria and 2 9 2 degrees of latitude and longitude and computed the Tunisia (Fig. 1b,e). The M. galathea-lachesis group showed a population configurations by calculating, for each marker, more complex pattern (Fig. 1b–f): three main groups were the barycenter of the specimens belonging to the same spe- identified, separated by just one or two mutations; M. lach- cies/square. Subsequently, we minimized the differences esis was confined to one of these lineages (Fig. 1e). Despite between population configurations for different markers by the low genetic differentiation, the three groups had clear applying a series of Procrustes analyses. We used the COI spatial and taxonomic patterns. All specimens identified as data, for which most populations were sampled as a refer- M. lachesis based on wing patterns shared similar haplotypes ence. The configurations for genitalia, wing shape and allo- and clustered together. No individuals identified as M. zymes were rotated and scaled based on the average galathea clustered within this group (Fig. 1e). The two addi- configurations of the shared populations between COI and tional lineages consisted solely of individuals identified as M.

4 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex

Figure 1 Spatial patterns of differentiation for the three species Melanargia galathea, M. lachesis and M. lucasi based on COI sequences in the Western Palaearctic region, including North Africa. Colours indicate their similarity; size of circles reflect sampling sizes. (a) Distribution of PCoA configurations projected in a RGB space for all samples (and species) analysed; (b–d) PCoA configurations independently for M. galathea+M. lachesis and for M. lucasi, as indicated by the blue line in (b). (e) haplotype network revealing remarkable genetic differentiation with clear geographic structure within each of the two main groups; (f) with higher spatial resolution detecting three distinct distribution patterns.

Journal of Biogeography 5 ª 2016 John Wiley & Sons Ltd J. C. Habel et al. galathea displaying a clear geographic structure: the first nent (Fig. 5b). The main pattern of diversification was pro- group occurs in Italy and southern France (red), and the sec- duced by the number of cornuti, lower in M. lucasi ond in central and eastern Europe, as well as western Asia compared to M. galathea and M. lachesis. Furthermore, the (green) and northern Iberia (dark green) (Fig. 1b,e). When valvae were less elongated in M. lucasi compared to M. examined in a higher spatial resolution (0.5 9 0.5 degrees, galathea and also less elongated compared to M. lachesis i.e. c. 40 9 55 km), the distribution pattern of the four lin- (valva PC1: 48.7% of variance); the tegumen showed a eages became more evident and the three lineages appeared longer uncus in M. galathea (tegumen PC2: 17.5% of vari- allopatric in most of the collection sites (Fig. 1f). ance). Jackknifing resulted in several wrong assignments of BI based on COI haplotypes recovered a phylogenetic tree M. galathea individuals as M. lachesis based on genital shape with a similar topology to the one previously published for (table in Fig. 5b). the genus Melanargia by Nazari et al. (2010) (Fig. 2). Mela- nargia lucasi formed a well-differentiated clade with strong DISCUSSION support (posterior probability: 1.0). The estimated time to the most recent common ancestor for the three taxa was In this study we performed a comparative biogeographic around 2 Ma. analysis of the M. galathea species complex based on two Allozyme analyses revealed a clear split between M. genetic (COI and allozymes) and two phenotypic markers galathea and M. lachesis, but the populations of M. lucasi (wing and genitalia shape) for populations sampled across clustered within M. galathea. However, we detected two dis- Europe and North Africa. All markers revealed a strong spa- tinct genetic clusters in Europe: a western European cluster tial differentiation, but in some cases their patterns were dis- that includes Italy, France and the North African populations cordant. COI sequences and the morphological markers and a south-eastern-central European cluster (Fig. 3a–b). displayed mostly congruent differentiation patterns that were The protest analysis revealed that the geographic pattern of partly in contrast to data provided by allozymes and the variation of COI, wings and genitalia are significantly corre- nuclear wingless gene previously assessed by Nazari et al. lated with each other (Fig. 3c), while allozyme variation (2010). showed no significant spatial correlation with the three other Morphological and mitochondrial data gave strong evi- markers. dence for an inter-specific split between the European M. galathea and the North African M. lucasi, which was not clearly reflected by allozymes and the gene wingless. Con- Phenotypes versely, the differentiation between the two European taxa, Wing shape analyses produced a total of 44 relative warps M. galathea and M. lachesis, was particularly strong accord- (22 for each wing). PLSDA using species membership as a ing to allozymes and wingless data. For these two taxa, diver- grouping variable produced three distinct clusters (Fig. 4a– gence in mitochondrial and morphological markers was less c). Three relative warps (PCs) explained most of the variance pronounced, but nevertheless evident, hence justifying the (hind wing PC1: 31.5%; fore wing PC2: 21.9%; hind wing acceptance of M. lachesis as a valid species. COI sequences PC3: 10.5%). The inspection of thin plate splines for these and allozymes suggested an additional split within the Magh- variables revealed that M. lucasi is characterized by more reb region into a western and an eastern lineage. However, elongated fore and hind wings compared to M. galathea and for the allozyme data, the eastern Maghreb populations clus- M. lachesis, while M. lachesis has a more elongated discoidal tered together with populations from Sicily and the Italian cell (Fig. 4b). Jackknifing showed that most of the specimens Peninsula, while they formed an independent group within could be blindly assigned to their species on the basis of M. lucasi for COI. Furthermore, at least two additional allo- wing shape (Fig. 4b). The distribution of wing shape in geo- zyme clusters, an Italian and a Balkan one, were detected. graphic space was largely congruent with the differentiation These showed similarities with the COI haplo-groups, but pattern of COI (Fig. 4a). they were not detected by wing and genital morphology. The analyses of genitalia produced a total of 42 relative The incongruence we detected among some of the markers warps (22 for the tegumen and 20 for the valva). PLSDA may have several non-exclusive reasons, which we discuss using the 42 relative warps and the number of cornuti as below. Specifically, we focus on the biogeographic history of variables and species membership as a grouping factor pro- the populations and species and highlight potential evolu- duced two distinct clusters (M. galathea + lachesis versus tionary drivers, which may lead to such diverging signatures. M. lucasi) for the first component, but showed a tendency to Finally, we revise the current taxonomic status of the three separate M. galathea from M. lachesis for the second compo- taxa we analysed in this study.

Figure 2 Bayesian inference phylogentic tree based on COI haplotypes. Bayesian posterior probabilities higher than 0.5 are shown next to the recovered branches. Horizontal error bars indicate uncertainty in estimated divergence times. The numbers following the country names represent the number of COI sequences sharing the same haplotype (the haplotype codes for all specimens used are indicated in Table S1). The codes for haplotype HMG19 correspond to the following countries: IT – Italy, FR – France, MK – Macedonia, DE – Germany, GR – , IR – , AM – Armenia and RU – Russia.

6 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex

HML1 M. lucasi (Morocco 2) 0.68 HML2 M. lucasi (Morocco 1) HML3 M. lucasi (Morocco 1)

1 HML4 M. lucasi (Morocco 1)

0.63 HML5 M. lucasi (Morocco 1)

0.97 HML6 M. lucasi (Morocco 1)

1 HML7 M. lucasi (Morocco 1) 1 HML8 M. lucasi (Morocco 1)

0.86 HML9 M. lucasi (Tunisia 1) HML10 M. lucasi (Tunisia 1) HML11 M. lucasi (Algeria 4, Tunisia 3) 1 HML12 M. lucasi (Tunisia 1) HML13 M. lucasi (Tunisia 4) HMLa1 M. lachesis (Spain 1) HMLa2 M. lachesis (Spain 1) 0.99 HMLa3 M. lachesis (Spain 1) HMLa4 M. lachesis (Spain 1) HMLa5 M. lachesis (Spain 1) HMLa6 M. lachesis (Spain 31, France 5, 3) HMLa7 M. lachesis (Spain 1) HMLa8 M. lachesis (Spain 1) HMG1 M. galathea (Italy 1)

0.72 HMG2 M. galathea (France 1) 0.53 HMG3 M. galathea (France 2) 0.85 HMG4 M. galathea (Italy 1) HMG5 M. galathea (Italy 1)

0.7 HMG6 M. galathea (Italy 1) HMG7 M. galathea (Italy 38, France 16) HMG8 M. galathea (Italy 1)

0.65 HMG9 M. galathea (Italy 2, France 1) HMG10 M. galathea (Italy 1)

0.87 HMG11 M. galathea (Germany 1) HMG12 M. galathea (Germany 1) HMG13 M. galathea (Italy 1) HMG14 M. galathea ( 1) HMG15 M. galathea (Macedonia 2, Greece 2, Turkey 1) HMG16 M. galathea (Greece 2) HMG17 M. galathea (Russia 1)

0.85 H18MGMG M. galathea (Italy 1) 0.85 1 H19MG M. galathea (IT 6, FR 4, MK 2, DE 1, GR 1, IR 1, AM 1, RU 1) HMG 20 M. galathea (Spain 1) H21MG M. galathea (France 3) H22MG M. galathea (Italy 2) H23MG M. galathea (Italy 1) H24MG M. galathea (Italy 1) H25MG M. galathea (Italy 4) H26MG M. galathea (Italy 1) H27MG M. galathea (Spain 1) H28MG M. galathea (Spain 6) 0.56 0.74 H29MG M. galathea (Spain 1) H30MG M. galathea (Spain 1) H31MG M. galathea (Armenia 1) H32MG M. galathea (Italy 1, Greece 1) H33MG M. galathea (Italy 1) 0.93 1 1 0.67 1

3.5 3 2.5 2 1.5 1 0.5 0 My

Journal of Biogeography 7 ª 2016 John Wiley & Sons Ltd J. C. Habel et al.

Figure 3 Spatial patterns of differentiation for the three species Melanargia galathea, M. lachesis and M. lucasi based on allozyme polymorphisms; colours indicate their similarity. (a) PCoA configurations projected on a map (size of circles reflect sampling sizes). (b) results from PCoA in a RGB space, (c) inlet table with the results of a protest analysis incorporating all four characters analysed. ***P ≤ 0.001.

which we attributed all the specimens to M. galathea or Concordant and divergent patterns M. lachesis, is rather surprising, especially considering that Our analysis of COI revealed that M. galathea and M. we analysed several populations from the contact zone of lachesis are represented by one major haplo-group with both taxa (Fig. 1f). This suggests that, despite rare reports three geographic sublineages separated by a comparatively of specimens with intermediate wing patterns between M. low number of mutations. In European butterflies, a diver- galathea and M. lachesis, gene flow appears to be mostly gence of 0.3% in COI, similar to what we found between absent. the Iberian lineage of M. galathea and M. lachesis, is usu- Strict spatial segregation and chequered distribution pat- ally associated with divergence into interfertile and not terns among lineages were not only recorded between M. always allopatric lineages often showing little or no pheno- galathea and M. lachesis but also appear to be a general fea- typic differentiation (Dinca et al., 2015). Introgression has ture of the M. galathea species group. The three main COI been shown to occur even among well-recognized species lineages, comprising M. galathea and M. lachesis, detected in of European butterflies (Mallet, 2005), leading to discor- Europe and in the Middle East show clear geographic dances between species-specific wing patterns and genetic boundaries and have been found to coexist in the same differentiation [e.g. the cases of Lysandra (Talavera et al., 0.2 9 0.2 degree rectangle only in two areas (for M. galathea 2013) and Iphiclides (Dinca et al., 2015)]. From this per- and M. lachesis and between two M. galathea lineages, spective, the congruence of COI sequences, wing shape Fig. 1f). The geographic distributions of the COI and allo- and allozyme with the wing colour pattern on the basis of zyme lineages strongly support repeated contractions of

8 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex

Figure 4 Spatial patterns of differentiation for the three species Melanargia galathea, M. lachesis and M. lucasi based on wing shape structures; colours indicate their similarity. (a) Results from partial least square discriminant analysis projected on a map (size of circles reflect sampling sizes) and (b) PCs based on three relative warps; the inlet table presents results from Jackknifing.

populations to the three main southern European peninsulas locus may not reflect neutral patterns. Regardless of this, (i.e. Iberia, Italy, Balkans) during cold stages of the Quater- the differentiation at the nuclear level was supported by nary followed by interglacial expansions to central Europe the wing shape pattern, according to which 97.6% of all (Fig. 6). The occurrence of contact zones along the main M. lachesis specimens could be blindly assigned to the cor- mountain chains of Europe (i.e. Alps and Pyrenees) and a rect species. It must be noted that, we only analysed a strong contribution of south-eastern European populations single population of M. lachesis for allozymes and only to central Europe, facilitated by the absence of mountains few wingless sequences were available (Nazari et al., 2010); isolating the Balkan Peninsula, is one of the most recurrent more detailed analyses of these markers could reveal differ- paradigms in the phylogeography of European species ent patterns of variation. (Hewitt, 1996, 2004; Schmitt, 2007). A completely different pattern exists between the taxa liv- The comparatively low levels of divergence displayed by ing in allopatry in North Africa and Europe. Melanargia COI and limited diversification in genital shape support a lucasi displayed a high COI divergence (minimum relatively recent split between M. lachesis and M. galathea. p-distance: 3.8%) with respect to M. galathea and M. lachesis The high variation in allozymes and the strong differentia- suggesting that M. lucasi represents a distinct species, in con- tion of the wingless gene was unexpected in this context cordance with Nazari et al. (2010). Additionally, the genitalia (but see below), as it is well known that mtDNA tends to and wing shape of M. lucasi differ from the two European diverge faster than nuclear DNA (Avise, 2009). In fact, the taxa. The wingless gene, allozymes and wing colour patterns high divergence found for the wingless gene is unusual in are relatively similar to M. galathea, although patterns of closely related taxa and could suggest that evolution at this high divergence between European and North African species

Journal of Biogeography 9 ª 2016 John Wiley & Sons Ltd J. C. Habel et al.

Figure 5 Spatial patterns of differentiation for Melanargia galathea, M. lachesis and M. lucasi based on genitalia structures; colours indicate their similarity. (a) Results from partial least square discriminant analysis projected on a map (size of circles reflect sampling sizes) and (b) PCs based on three relative warps; the inlet table presents results from Jackknifing. and lineages are the rule and not the exception (Husemann of Sicily, separating Tunisia and Sicily by a current mini- et al., 2014). This differentiation has likely evolved during mum distance of 140 km (Habel et al., 2011). However, due long geographic separation between the two continents, to the lowering of the sea level during cold stages of the which were only connected during the Messinian salinity cri- Quaternary, North Africa and Sicily were geographically con- sis 5.9–5.3 million years before present (Manzi et al., 2013). siderably closer than at present, and reconstructions of coast- Genetic imprints of that period, or even older, were detected lines estimated that this channel was only 50 km (Manzi in many groups (Habel et al., 2012 with references et al., 2013). Thus, one explanation for the similarity in allo- therein). The minimum current distance between Africa and zyme and wing pattern between M. galathea and M. lucasi Europe is about 15 km at the Strait of Gibraltar, a relatively might be a relatively recent exchange of individuals and short sea barrier considering the dispersal abilities of many hence of genetic information across the Strait of Sicily with organisms, in particular flying (Fig. 6). In fact, recent subsequent introgression that could explain the discordance colonizations (i.e. < 100,000 years) between North Africa with mitochondrial markers. and Iberia have been detected for different taxa (e.g. Cosson et al., 2005) and many Atlanto-Mediterranean butterfly spe- Drivers leading to discordant differentiation cies are found on both sides of this strait (Tolman & Lew- ington, 2008). However, this is not the case of the group The different patterns shown by the four markers used in studied here as the opposed sides of the Gibraltar strait host this study suggest that several mechanisms have been either M. lucasi (North Africa) or M. lachesis (Iberia), which involved in the process of differentiation for these species. were distinct in all studied markers. MtDNA is mostly advocated as a neutral marker evolving The genetic and morphological differences in the M. in a clock-wise manner (Hewitt, 1996; Avise, 2009). We galathea species group are less pronounced across the Strait found the strongest differentiation for COI between

10 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex

Figure 6 Spatial patterns of differentiation for the three species Melanargia galathea, M. lachesis and M. lucasi based on COI sequences; colours indicate their similarity, using PCoA configurations projected in a RGB space for all samples (and species) analysed. Potential glacial refugial regions (R) are indicated with cycles; post-glacial range expansions are shown by arrows.

M. galathea and M. lucasi. This split occurred more than shown along altitudinal and latitudinal, as well as other 2 Ma and is most probably the first one in the M. galathea spe- environmental gradients (e.g. for Pararge aegeria: Vande- cies group, reflecting the fact that Europe and Africa represent woestijne & Van Dyck, 2011). The genitalia structures are two distinct geographic entities constantly being separated by not under thermal, but likely under sexual selection, poten- sea since the Pliocene. In contrast, M. galathea and M. lachesis tially explaining differences in the observed patterns. How- seem to represent a different case: although separated only by ever, more detailed and experimental studies are necessary two mutations in COI (p-distance: 0.3%), M. lachesis is consis- to disentangle these aspects. tently distinguished from its congeners by COI sequences as well as wing coloration and allozyme patterns. The taxonomy of the M. galathea species group These differences may be explained by several non-exclu- revisited sive factors, including climatic and environmental prefer- ences, restricted dispersal behaviour and density-dependent The phenotypic and genetic data revealed significant splits phenomena (Pigot & Tobias, 2015; Voda et al., 2015b). among the three Melanargia taxa and the commonly used However, one might speculate that the diverging ecological taxonomy is incongruent with our results: in the past, M. preferences (M. galathea and M. lucasi in southern Europe lucasi was mostly treated as a subspecies of M. galathea (e.g. and North Africa avoid extremely warm and dry environ- Tolman & Lewington, 2008) and only recently taxonomists ments, whereas M. lachesis is present under such condi- accepted this taxon as a distinct species (Nazari et al., 2010; tions; personal observations of the authors) may be one Tshikolovets, 2011). All markers, except allozymes, support major driver of this inconsistency among markers. M. lucasi as a valid species endemic to the Maghreb region. Although allozymes are known as a suitable neutral, bio- The two taxa have likely been separated since the Pliocene, geographic markers, several examples have suggested that but with possible events of introgression later on. they may be under thermal selection (e.g. Karl et al., 2009). In contrast, M. lachesis was broadly accepted as a distinct Hence, the differential habitat use might have evoked a species (Habel et al., 2011), although recent results chal- strong selective pressure on allozymes, enhancing the diver- lenged this hypothesis (Dinca et al., 2015). The compara- gence between these two groups. tively weak split based on genitalia, mtDNA, and wing shape, As the ratio of dark and light coloration on butterfly but strong differentiation in allozymes, and nuclear markers wings can strongly influence their thermoregulation, wing and the strong tendency for chequered distribution patterns colour pattern differences between M. lachesis on the one indicate locally restricted gene flow between M. lachesis and hand and the two other taxa on the other hand might also M. galathea. Therefore, all three entities should be accepted be triggered by the differential habitat use of these two as distinct species, but with different evolutionary histories groups. This hypothesis is supported by a previous study on and ages. the evolution of wing shapes in butterflies, revealing that Our study provides further support that multiple marker phenotypic differentiation may occur fast and at small geo- studies yield a more comprehensive understanding of biogeo- graphic scales (Habel et al., 2013). Such shifts have been graphic dynamics and support the robustness of the resulting

Journal of Biogeography 11 ª 2016 John Wiley & Sons Ltd J. C. Habel et al. taxonomic conclusions. However, at the same time, our data Dapporto, L., Voda, R., Dinca, V. & Vila, R. (2014) Compar- underline that delineating a specific driver leading to specific ing population patterns for genetic and morphological structures remains difficult. markers with uneven sample sizes. An example for the butterfly Maniola jurtina. Methods in Ecology and Evolu- tion, 5, 834–843. ACKNOWLEDGEMENTS Darriba, D., Taboada, G.L., Doallo, R. & Posada, D. (2012) Funding for this research was provided by the project ‘Bar- jModelTest 2: more models, new heuristics and parallel 9 coding Italian Butterflies’, by the European Union’s Seventh computing. Nature Methods, , 772. Framework programme for research and innovation under Dennis, R.L.H., Williams, W.R. & Shreeve, T.G. (1991) A the Marie Skłodowska-Curie grant agreement No 609402- multivariate approach to the determination of faunal 2020 researchers: Train to Move (T2M) postdoctoral fellow- structures among European butterfly species (Lepidoptera: ship to R. Voda, by European Union’s Horizon 2020 Rhopalocera). Biological Journal of the Linnean Society, 101 – research and innovation programme under the Marie Sklo- ,1 49. dowska-Curie grant (project no. 658844 Eco-PhyloGeo) to L. deWaard, J.R., Ivanova, N.V., Hajibabaei, M. & Hebert, P.D.N. Dapporto and by the Spanish Ministerio de Economıa y (2008) Assembling DNA barcodes: analytical protocols. Competitividad CGL2013-48277-P and PRX15/00305 to R. Methods in molecular biology: environmental genetics (ed. by – Vila. We are grateful to numerous colleagues who provided C. Martin), pp. 275 293. Humana Press, Totowa, NJ. samples and to two anonymous referees for constructive Dinca, V., Montagud, S., Talavera, G., Hernandez-Roldan, J., comments on a draft version of this article. Munguira, M.L., Garcıa-Barros, E., Hebert, P.D.N. & Vila, R. (2015) DNA barcode reference library for Iberian but- terflies enables a continental-scale preview of potential REFERENCES cryptic diversity. Scientific Reports, 5. Adams, D.C., Rohlf, F.J. & Slice, D.E. (2004) Geometric Drummond, A.J., Suchard, M.A., Xie, D. & Rambaut, A. morphometrics: Ten years of progress following the ‘revo- (2012) Bayesian phylogenetics with BEAUti and the BEAST 29 – lution’. Italian Journal of Zoology, 71,5–16. 1.7. Molecular Biology and Evolution, , 1969 1973. Avise, J.C. (2009) Phylogeography: retrospect and prospect. Excoffier, L. & Langaney, A. (1989) Origin and differentia- Journal of Biogeography, 36,3–15. tion of human mitochondrial DNA. American Journal of 44 – Bookstein, F.L. (1997) Landmark methods for forms without Human Genetics, ,73 85. landmarks: morphometrics of group differences in outline Excoffier, L., Laval, G. & Schneider, S. (2005) Arlequin ver. shape. Medical Image Analysis, 1, 225–243. 3.0: an integrated software package for population genetics 1 – Bozano, G.C. (2002) Guide to the butterflies of the Palearctic data analysis. Evolutionary Bioinformatics, ,47 50. region Satyridae, Part 3: Subtribes Melanargiina and Coeno- Fernandez-Rubio, F. (1991) Guia de mariposas diurnas de la nympha, Genera Melanargia, , Triphysa and Peninsular Iberica, Baleares, Canarias, Azores y Madeira Sinonympha. Omnes Artes, Milano. (Papilionidae, Pieridae, Danaidae, Satyridae y Hesperiidae). Brower, A.V. (1994) Rapid morphological radiation and conver- Ediciones Piramide, Madrid. € gence among races of the butterfly Heliconius erato inferred Habel, J.C., Rodder, D., Lens, L. & Schmitt, T. (2011) From from patterns of mitochondrial DNA evolution. Proceedings Africa to Europe and back: refugia and range shifts cause of the National Academy of Sciences USA, 91,6491–6495. high genetic differentiation in the European Marbled Claramunt, S., Derryberry, E.P., Remsen, J.V. & Brumfield, White butterfly Melanargia galathea. BMC Evolutionary 11 R.T. (2012) High dispersal ability inhibits speciation in a Biology, , 215. continental radiation of passerine birds. Proceedings of the Habel, J.C., Husemann, M., Schmitt, T., Zachos, F.E., Hon- Royal Society B: Biological Sciences, 279, 1567–1574. nen, A.-C., Petersen, B., Parmakelis, A. & Stathi, I. (2012) Clement, M., Posada, D. & Crandall, K.A. (2000) TCS: a Microallopatry caused diversification in Buthus scorpions computer program to estimate gene genealogies. Molecular (Scorpiones: Buthidae) of the Atlas Mountains (NW 7 Ecology Notes, 9, 1657–1659. Africa). PLoS ONE, , e29403. Cosson, J.F., Hutterer, R., Libois, R., Sara, M., Taberlet, P. & Habel, J.C., Husemann, M., Schmitt, T., Dapporto, L., € Vogel, P. (2005) Phylogeographical footprints of the Strait Rodder, D. & Vanderwoistijne, S. (2013) A forest butterfly of Gibraltar and Quaternary climatic fluctuations in the in the oasis of the Sahara desert: isolation does not matter. 2 – western Mediterranean: a case study with the greater Journal of Heredity, , 234 247. white-toothed shrew, Crocidura russula (Mammalia: Sorici- Hewitt, G.M. (1996) Some genetic consequences of ice ages, dae). Molecular Ecology, 14, 1151–1162. and their role in divergence and speciation. Biological Jour- 58 – Dapporto, L., Ramazzotti, M., Fattorini, S., Talavera, G., nal of the Linnean Society, , 247 276. Vila, R. & Dennis, R.L.H. (2013) recluster: an unbiased Hewitt, G.M. (2004) Genetic consequences of climatic oscil- clustering procedure for beta-diversity turnover. Ecography, lation in the Quaternary. Philosophical Transactions of the 359 – 36, 1070–1075. Royal Society B: Biological Sciences, , 183 195.

12 Journal of Biogeography ª 2016 John Wiley & Sons Ltd Diversification in a butterfly species complex

Husemann, M., Schmitt, T., Zachos, F.E., Ulrich, W. & Habel, Talavera, G., Lukhtanov, V.A., Rieppel, L., Pierce, N.E. & J.C. (2014) Palaearctic biogeography revisited: evidence for Vila, R. (2013) In the shadow of phylogenetic uncertainty: the existence of a North African refugium for Western the recent diversification of Lysandra butterflies through Palaearctic biota. Journal of Biogeography, 41,81–94. chromosomal change. Molecular Phylogenetics and Evolu- Ivanova, N.V., Dewaard, J.R. & Hebert, P.D.N. (2006) An tion, 69, 469–478. inexpensive, automation-friendly protocol for recovering Tolman, T. & Lewington, R. (2008) Collins butterfly guide: high-quality DNA. Molecular Ecology Notes, 6, 998–1002. the most complete field guide to the butterflies of Britain and Karl, I., Schmitt, T. & Fischer, K. (2009) Genetic differentia- Europe. Collins, London. tion between alpine and lowland populations of a butterfly Tshikolovets, V. (2011) Butterflies of Europe and the Mediter- is related to PGI enzyme genotype. Ecography, 32, 488–496. ranean area. Vadim Tshikolovets, Pardubice. Kreft, H. & Jetz, W. (2010) A framework for delineating bio- Vandewoestijne, S. & Van Dyck, H. (2011) Flight morphol- geographic regions based on species distributions. Journal ogy along a latitudinal gradient in a butterfly: do geo- of Biogeography, 37, 2029–2053. graphic clines differ between agricultural and woodland Kudrna, O., Pennerstorfer, J. & Lux, K. (2015) Distribution landscapes? Ecography, 34, 876–886. atlas of European butterflies and skippers. Peks, Schwanfeld. Voda, R., Dapporto, L., Dinca, V. & Vila, R. (2015a) Cryptic L^e Cao, K.-A., Boitard, S. & Besse, P. (2011) Sparse PLS dis- matters: overlooked species generate most butterfly beta- criminant analysis: biologically relevant feature selection diversity. Ecography, 38, 405–409. and graphical displays for multiclass problems. BMC Voda, R., Dapporto, L., Dinca, V. & Vila, R. (2015b) Why Bioinformatics, 12, 253. do cryptic species tend not to co-occur? A case study on Mallet, J. (2005) Hybridization as an invasion of the genome. two cryptic pairs of butterflies. PLoS ONE, 10, e0117802. Trends in Ecology and Evolution, 20, 229–237. Wade, M.J. (2001) Evolution: infectious speciation. Nature, Mallet, J. (2007) Hybrid speciation. Nature, 446, 279–283. 409, 675–677. Manzi, V., Gennari, R., Hilgen, F., Krijgsman, W., Lugli, S., Waters, J.M. (2011) Competitive exclusion: phylogeography’s Roveri, M. & Sierro, F.J. (2013) Age refinement of the ‘elephant in the room’? Molecular Ecology, 20, 4388–4394. Messinian salinity crisis onset in the Mediterranean. Terra Nova, 25, 315–322. SUPPORTING INFORMATION Mittelbach, G.G. & Schemske, D.W. (2015) Ecological and evolutionary perspectives on community assembly. Trends Additional Supporting Information may be found in the in Ecology and Evolution, 30, 241–247. online version of this article: Mitteroecker, P. & Bookstein, F. (2011) Linear discrimina- Appendix S1 Sampling localities for mtDNA analyses. tion, ordination, and the visualization of selection gradi- Appendix S2 Sampling localities for allozyme analyses. ents in modern morphometrics. Journal of Evolutionary Appendix S3 Sampling localities for wing vein analyses. Biology, 38, 100–114. Appendix S4 Landmarks selected for wing vein analyses. Nazari, V., Hagen, W.T. & Bozano, G.C. (2010) Molecular Appendix S5 Sampling localities for genitalia analyses. systematics and phylogeny of the Marbled Whites (Lepi- doptera: Nymphidae, , Melanargia Meigen). BIOSKETCH Systematics Entomology, 35, 132–147. Pigot, A.L. & Tobias, J.A. (2015) Dispersal and the transition Jan Christian Habel is assistant professor at the Department to sympatry in vertebrates. Proceedings of the Royal Society of Terrestrial Ecology at the Technical University Munich, 282 B: Biological Sciences, , 20141929. Germany. He is working in the field of conservation biology, Rohlf, J. (2008) TPSDIG, version 1.39. Department of Ecol- molecular ecology and biogeography, on various taxa. ogy and Evolution, State University of New York, Stony Brook, NY. Author contributions: J.C.H., T.S. and L.D. created the study Schmitt, T. (2007) Molecular biogeography of Europe: Pleis- setup; J.C.H., R.Vo., R.V., M.H. and L.D. collected and anal- tocene cycles and postglacial trends. Frontiers in Zoology, ysed the data; all authors interpreted the data and wrote the 4, 11. manuscript. Schmitt, T., Habel, J.C., Zimmermann, M. & Muller,€ P. (2006) Genetic differentiation of the Marbled White but- Editor: Aristeidis Parmakelis terfly, Melanargia galathea, accounts for glacial distribu- tion patterns and postglacial range expansion in southeastern Europe. Molecular Ecology, 15, 1889–1901.

Journal of Biogeography 13 ª 2016 John Wiley & Sons Ltd