Hybridization Promotes Speciation in Coenonympha Butterflies

Molecular Ecology (2015) 24, 6209–6222 doi: 10.1111/mec.13479

Hybridization promotes speciation in Coenonympha butterﬂies

THIBAUT CAPBLANCQ,*† LAURENCE DESPRES,*† DELPHINE RIOUX*† and JESUS MAVAREZ*† *LECA, Universite Grenoble Alpes, F-38000, Grenoble, France, †LECA, CNRS, F-38000, Grenoble, France

Abstract Hybridization has become a central element in theories of animal evolution during the last decade. New methods in population genomics and statistical model testing now allow the disentangling of the complexity that hybridization brings into key evolutionary processes such as local adaptation, colonization of new environments, species diversiﬁcation and extinction. We evaluated the consequences of hybridization in a complex of three alpine butterﬂies in the genus Coenonympha, by combining morphological, genetic and ecological analyses. A series of approximate Bayesian computation procedures based on a large SNP data set strongly suggest that the Darwin’s Heath (Coenonympha darwiniana) originated through hybridization between the Pearly Heath (Coenonympha arcania) and the Alpine Heath (Coenonympha gardetta) with different parental contributions. As a result of hybridization, the Darwin’s Heath presents an intermediate morphology between the parental species, while its climatic niche seems more similar to the Alpine Heath. Our results also reveal a substantial genetic and morphologic differentiation between the two geographically disjoint Darwin’s Heath lineages leading us to propose the splitting of this taxon into two different species.

Keywords: approximate Bayesian computation, ddRAD sequencing, geometric morphometrics, hybrid species, Lepidoptera, species diversiﬁcation Received 29 July 2015; revision received 9 November 2015; accepted 12 November 2015

have been challenged during the last decades, and Introduction interspecific hybridization is now accepted as a major Hybridization between well-established species is process with a profound impact on several important known to be frequent in nature (Rieseberg 1997; Mallet evolutionary phenomena such as adaptation (Gompert 2007; Mavarez & Linares 2008; Schwenk et al. 2008; Des- et al. 2008; Kim et al. 2008; Song et al. 2011; The Helico- cimon & Mallet 2009, Abbott et al. 2013), with around nius Genome Consortium 2012, Pardo-Diaz et al. 2012), 10% of animal species and 25% of plants involved in extinction (Rhymer & Simberloff 1996; Hohenlohe et al. interspecific crosses according to Mallet 2005. However, 2011; Vonlanthen et al. 2012) and diversification/specia- and particularly among zoologists, this process has for tion (Mallet 2007; Mavarez & Linares 2008; Nolte & long been considered either to be a genetic dead end Tautz 2010; Nice et al. 2013; Yakimowski & Rieseberg leading mainly to sterile hybrids or to have no major 2014). In particular, there is a recent interest in studying effect in the divergence of species, even when fertile the contribution of hybridization to speciation through hybrids are produced (Mayr 1963). Indeed, it is gener- the formation of new hybrid taxa. This has led to the ally accepted that gene flow through hybridization identification of hybrid species in diverse animal between diverging lineages can delay the achievement groups including insects (Mavarez et al. 2006; Salazar of reproductive isolation by homogenizing the gene et al. 2010; Kunte et al. 2011), birds (Brelsford et al. 2011; pools, working against the speciation process (Elgvin Hermansen et al. 2011), fishes (Nolte et al. 2005) and et al. 2011; Yakimowski & Rieseberg 2014). These views mammals (Amaral et al. 2014). These studies, mostly conducted during the last decade, have contributed to our understanding of hybrid speciation, which is Correspondence: Thibaut Capblancq, Fax: +33 4 76 51 42 79; E-mail: [email protected] defined as the emergence of a different, stable and iso-

© 2015 John Wiley & Sons Ltd 6210 T. CAPBLANCQ ET AL. lated population/species by hybridization between two living above 1500 m with a disjoint distribution includ- different taxa (Rieseberg 1997; Seehausen 2004; Gross & ing mainly the Alps, but also the French massif central, Rieseberg 2005; Mallet 2007; Mavarez & Linares 2008; Bosnia & Herzegovina, and Albania (Tolman & Lewing- Nolte & Tautz 2010; Abbott et al. 2013). A distinction ton 2008). These two species diverged 1.5–4 million has been made between ‘hybrid speciation’, where the years ago according to a phylogeny of the genus Coeno- emergence and isolation of the newborn lineage are the nympha based on mitochondrial COI sequences (Kodan- direct consequences of hybridization and ‘secondary daramaiah & Wahlberg 2009). The third taxon, the gene flow’, where a lineage already diverged from a Darwin’s Heath (Coenonympha darwiniana), shows an species undergoes introgression from another one intermediate morphology and completely replaces (Mavarez & Linares 2008; Schumer et al. 2014). The C. gardetta in two distinct and narrow areas: the Valais major difference is the role played by hybridization and Tessin regions in Switzerland and the southwest of during the initial divergence of the third species. How- the French Alps. Limited contact zones between pairs ever, even if the evolutionary steps differ, both proof the three taxa have been reported (Porter et al. 1994), cesses can result in the stabilization of a novel lineage but ongoing gene flow between them is unknown. The whose genome is a composite of two parental contribu- taxonomic status of C. darwiniana has been hotly tions and that can therefore be qualified as a ‘hybrid debated and specialists have variously classified it as a species’. subspecies of C. arcania, a subspecies of C. gardetta or as Hybrid species appear to be quite rare in nature, but hybrid between them (Porter et al. 1994; Wiemers 1998; the expanding literature on this topic is beginning to Schmitt & Besold 2010; Lafranchis 2000; Tolman & provide valuable information regarding the patterns Lewington 2008). However, most of these claims were associated with this phenomenon (Seehausen 2004; Mal- based on studies with limited geographical coverage, let 2007; Mavarez & Linares 2008; Abbott et al. 2013). without proper analyses of the ecology and morphology Some studies have highlighted a primordial role of of the species and with genetic analyses based exclu- intermediate morphology in the emergence of the new- sively on allozyme markers (Porter et al. 1994; Wiemers born taxon (Mavarez et al. 2006; Salazar et al. 2010), 1998; Schmitt & Besold 2010). To resolve the evolution- while others have focused more on the crucial ecologi- ary history of the complex, we first use a large SNPs cal adaptations brought by the genetic hybrid combina- data set to compare different scenarios through ABC tion (Nolte et al. 2005; Schwarz et al. 2005; Gompert procedure. We especially tested for the hybrid origin of et al. 2006; Kunte et al. 2011) or the intrinsic genomic C. darwiniana by confronting this hypothesis with alter- mechanisms which have allowed the isolation of the native evolutionary models of cladogenesis. As men- hybrid lineage from parental species (Lukhtanov et al. tioned above, a combination of evidence (i.e. 2015). Taken together, these studies suggest that even if phenotypes, behaviours, ecology, genetics) is helpful to population genetics provide definitely the best tools to support the hybrid origin of a taxon and understand identify hybrid species, multiple lines of evidence, the context of such speciation event. Yet this kind of including morphology, ecology and genetics, must be analysis has never been made for C. darwiniana.We taken into account when investigating the hybrid origin combined analysis based on population genetics, mor- of a taxon. Furthermore, most studies to date have pro- phometry of wing shapes and comparison of climatic posed the hybrid speciation hypothesis for a lineage niches to identify the genetic, morphological and eco- based on its patterns of genetic composition and/or logical patterns associated with the species and put phenotype intermediacy, without further statistical test- them in regard to the scenario retained by the ABC ing. Yet recent improvements in DNA sequencing (e.g. analysis. This set of analyses reveals new insights into genome projects, large SNP data sets) and model testing the crucial role that hybridization has played in the (e.g. approximate Bayesian computation procedures) diversification of this group of European butterflies. allow for more powerful testing of the specific evolutionary processes and the scenarios involved during a Material and methods hybrid speciation event (Nice et al. 2013; Sun et al. 2014). Butterfly sampling Here, we examine the role of hybridization in the evolutionary history of three European butterflies in the We sampled populations across the entire geographical genus Coenonympha (Nymphalidae: Satyrinae). Two are range of Coenonympha darwiniana in the south French well-differentiated species: (i) the Pearly Heath (Coeno- Alps and the Tessin region in Switzerland/Italy, as much nympha arcania, Linne 1761), a common lowland butter- as possible from the main range of Coenonympha gardetta fly present from Spain to Russia, and (ii) the Alpine (Alps), altogether with nearby populations of Coenonym- Heath (Coenonympha gardetta, Prunner 1798), a species pha arcania within these areas. Using entomological nets,

© 2015 John Wiley & Sons Ltd HYBRIDIZATION AND SPECIATION IN ALPINE BUTTERFLIES 6211 twelve populations of C. darwiniana (seven in south digestion–ligation reaction was performed on a ther- French Alps and five in Tessin) were sampled; we mocycler using a succession of 60 cycles of 2 min at also sampled eleven populations of C. arcania and 37 °C for digestion and 2 min at 16 °C for ligation, nineteen populations of C. gardetta (Fig. 1). Sample sizes after which the enzymes were heat-inactivated at per population depend on the analysis (genetics or 65 °C for 10 min. An equal volume mixture of all the morphology) and are listed in Table S1 (Supporting digested–ligated fragments was purified with Agen- information). court AMPure XP beads (Beckman Coulter, France). Fragments were size-selected between 250 and 500 bp by gel extraction (agarose gel: 1.6%) and purified with Genetic data acquisition QIAquick Gel Extraction Kit (Qiagen). The ddRAD DNA was extracted from the complete thorax of each library obtained was amplified using the following individual using the DNeasy Blood and Tissue Kit PCR mix: a volume of 20 lL with 1 lL of DNA tem- (QIAgen, Germany) according to the manufacturer’s plate, 10 mM of dNTPs, 10 lM of each PCR probe instructions and stored at À20 °C. A double-digested (Peterson et al. 2012) and 2 U/lLofTaq Phusion-HF RAD (restriction site-associated DNA) experiment was (New England Biolabs Inc.), and the following PCR conducted on 174 individuals using a modified ver- cycles: an initial denaturation at 98 °C for 10 min; 15 sion of the protocol described in Peterson et al. (2012). cycles of 98 °C for 10 s, 66 °C for 30 s and 72 °C for In short, 300 ng of DNA template from each individ- 1 min; followed by a final extension period at 72 °C ual was double-digested with 10 units each of SbfI– for 10 min. The amplified ddRAD library was puri- HF and MspI (New England Biolabs Inc.) at 37 °C fied with QIAgen MinElute PCR Purification Kit (Qia- during one hour in a final volume of 34 lL using the gen) and sequenced on an Illumina Hi-Seq 2000 (4/10 CutSmart buffer provided with the enzymes. Diges- lane, paired-end 2 9 100 bp, Fasteris SA, Switzerland). tion was further continued together with the ligation The DNA reads (120 million reads) from the library of P1 and P2 adapters (see Peterson et al. 2012) by were used to call SNP genotypes with the STACKS adding 10 units of T4 DNA ligase (New England Bio- pipeline (Catchen et al. 2013) to create the SNPs data set labs Inc.), adapters P1 and P2 in 10-fold excess (com- used in the ABC procedure (see section ‘Test of evolu- pared to the estimated number of restriction tionary history scenarios with approximate Bayesian fragments) and 1 lLof10mM ribo-ATP (New Eng- computation (ABC) method’) and multivariate analysis land Biolabs Inc.) in each sample. This simultaneous (see section ‘Multivariate analysis’).

48 Innsbruck Bern

46 Lyon

Latitude Milano 45

43 Marseille

81216 Longitude

Fig. 1 Map of the European Alps with the sampling locations for the three taxa in the complex: Coenonympha arcania (squares), Coenonympha gardetta (circles) and Coenonympha darwiniana (triangles). Right: picture of C. arcania (bottom) and C. gardetta (top) with the putative hybrid species (C. darwiniana) in the middle. (Pictures from http://www.digitale-naturfotos.de and http:// www.lepiforum.de).

Scenario 1 Scenario 2 Scenario3 Fig. 2 The 5 scenarios tested in the ABC

A A A analysis with model involving only cladogenesis events in the top and models involving hybridization in the bottom. Time A corresponds to the ancestral population, P1 to the current Coenonympha arcania population, P2 to Coenonympha gardetta population, P3 to French Coeno- P1 P3 P4 P2 P1 P3 P4 P2 P1 P3 P4 P2 nympha darwiniana population and P4 to Swiss C. darwiniana. Scenario 4 Scenario 5

A A

Time

P1 P3 P4 P2 P1 P3 P4 P2

Test of evolutionary history scenarios with especially for the origin of C. darwiniana. To compare approximate Bayesian computation (ABC) method them and find the most plausible one, we used the SNP data set with the approximate Bayesian computation We used the STACKS pipeline to treat the ddRAD (ABC) approach implemented in the software DIYABC sequencing data as follows: the process_radtags function (Cornuet et al. 2014). In short, a large number of simu- was first run to demultiplex the data and filter the reads lated data sets were produced under each tested scenario on their quality with a sliding window of 15% of the read by sampling parameter values into predefined prior dis- length and a Phred score limit of 10. Reads with low tributions, and an estimation of each scenario posterior quality or uncalled bases were removed. A de novo probability was calculated as a function of the similarity assembly was then performed on each individual read between several summary statistics of simulated data pool using the ustacks function, the minimum coverage sets and observed data using a logistic regression proce- for creating a stack of identical reads (-m) was fixed to 5, dure. These posterior probabilities enable a ranking of and a maximum of 4 different nucleotides were imposed scenarios and allow for the identification of the most to merge two different stacks in a polymorphic locus probable evolutionary history. We only considered five (-M). Highly repetitive stacks and overmerged tags were evolutionary scenarios (Fig. 2), based on previously pub- dropped using both the ‘removal algorithm’ (-r) and the lished studies dealing with the evolution of the complex ‘deleveraging algorithm’ (-d). A data set of the loci from (Porter et al. 1994; Wiemers 1998; Schmitt & Besold 2010) all the individuals’ loci data sets was then built using the and our own hypothesis based on the morphometric and cstacks function in order to find all the alleles in the phylogenetic analyses (see Results). libraries and to produce a catalog of consensus alleles for each locus. A maximum of 5 mismatches were allowed Scenario design and simulations. A uniform distribution for considering two individual tags as the same locus with a large interval was chosen for each prior increas- and to merge them in the catalog (-n). The sets of loci ing the need in computation time but overcoming the constructed within each individual read pool were then lack of knowledge on population sizes, divergence searched against the newly produced catalog (sstacks times and admixture rate (see Table S2 Supporting function), and a SNP data set was produced with the information). Only one approximate C. arcania / genotype of each individual for every polymorphic C. gardetta divergence time of 1.5–4 million of years was position (populations function). Only tags present in more available from Kodandaramaiah & Wahlberg (2009). For than 60% of the sampling were used to find the SNPs in each scenario, a total of one million data sets were sim- order to avoid an excess of missing data. Finally, SNPs ulated and the posterior probability was computed by with a frequency lower than 5% were removed from performing a logistic regression on the 1% of simulated the data set, and only one polymorphic site was kept for data closest to the observed data set (Cornuet et al. each RAD-tag. This genetic matrix was then used to test 2008, 2010). Summary statistics used for observed/sim- different models of evolution. ulated data sets comparisons are the mean gene diver- Several scenarios have been proposed for the evolu- sity and Fst across all loci and Nei’s distances among tionary history of these three Coenonympha species and

© 2015 John Wiley & Sons Ltd HYBRIDIZATION AND SPECIATION IN ALPINE BUTTERFLIES 6213 populations. Gene diversity was chosen because allelic Missing values were replaced by the mean frequency of richness is directly influenced by hybridization; the the concerned allele in all the sampling. genetic differentiation (Fst) and distance (Nei’s) among taxa were used because these indexes reflect the degree Geometric morphometry analysis of differentiation between populations and are highly relevant to compare evolutionary histories. We performed a geometric morphometric analysis including wing shape, patterns and venation, with a particular Assessing the ‘goodness of fit’ of the model. Once the most interest on traits that have been previously considered as probable scenario was identified, different posterior anal- indicative of intermediacy in C. darwiniana such as the yses were carried out to evaluate the trustfulness of the form and position of eyespots, the shape of the hindwing procedure. First, to be sure of the choice of the scenario, white band and the global size of the wings. We identified 500 new data sets were simulated with each historical 22 and 18 homologous landmarks in the fore- and hind- model creating pseudo-observed data sets for which pos- wings, respectively, corresponding to either (a) ‘pure’ terior probabilities were re-estimated with the same pro- wing shape when using vein intersections and termini or cedure described above. Type I and type II errors were (b) wing patterns when using eyespots and the white evaluated by measuring, respectively, the fraction of data band outline (see Fig. S1 Supporting information). sets simulated under the best scenario that were assigned The wings were scanned with a resolution of 1200 dpi, to other scenarios and the fraction of data sets simulated and the geometric morphometric analysis was performed under other scenarios that were assigned to the best sce- using the tps (thin-plate spline) series of programs (avail- nario. Second, we wanted to know whether our selected able at http://life.bio.sunysb.edu/morph/) for file cre- model is able to produce data sets similar to the observed ation (TPSUTIL, ver 1.58, Rohlf 2006b) and landmarks one. This confidence in the model is evaluated by esti- positioning (TPSDIG 2, ver 2.17, Rohlf 2006a). A Procrustes mating the similarity between simulated (1000 simula- superimposition was then applied to the landmarks coor- tions) and real data sets using summary statistics dinates in order to remove nonshape variation in loca- different than the ones used for model choice (an impor- tion, scale and orientation. After a dimensionality tant precaution, see Cornuet et al. 2010). For each sum- reduction to correct for the effect of using a large number mary statistic, a P-value is estimated by ranking the of variables relative to the number of specimens was observed value among the values obtained with simu- applied (see Merot et al. 2013), we tested shape differ- lated data sets, and a principal component analysis was ences among species using a one-way MANOVA on the also performed to check visually the position of the remaining principal components (PCs) with shape as observed data in relation to the data sets generated from dependent variable and species as factor. A canonical the posterior predictive distribution. variate analysis (CVA) allowed for the investigation of species discrimination (Merot et al. 2013), and a dendro- Posterior distribution of parameters and their precision. It gram was produced with morphometric distances was possible to evaluate the posterior distributions of (Mahalanobis distances) among groups to allow for a parameters for our selected scenario using a local linear clearer visualization of population morphological posi- regression on the 1% of simulations closest to the tion in the three species complex. The generalized Pro- observed data set (Cornuet et al. 2010). These distribu- crustes analysis and the different statistics were tions gave an idea of the most probable value (median) performed with R and the RMORPH (M. Baylac, personal and the ‘approximation’ (width of the distribution) for communication), ADE4 (Chessel et al. 2004) and APE (Par- each historical model. A precision of these estimations adis et al. 2014) libraries. In parallel to the Procrustes was evaluated by computing the median of the absolute superimposition and shape analysis, a wing size index error divided by the true parameter value of the 500 was obtained using the log-transformed centroid size pseudo-observed data sets simulated under the selected directly from the wing landmarks. Difference in size scenario using the median of the posterior distribution among species was tested through an analysis of variance as point estimate (Cornuet et al. 2010). (ANOVA) with size as a dependent variable and species as factor. Multivariate analysis Climatic niche comparisons We investigate individuals grouping and differentiation without any assumptions through a principal compo- Occurrence data of the three species in the Alps were nent analysis (PCA) carried out on the same SNPs data collected from numerous institutions data sets: FLAVIA set. We used R 3.1.0 (R Development Core Team) and (French association of entomology, http://www.fla the library ADEGENET (Jombart 2008) to achieve the PCA. via.ade.free.fr), MNHN (French national museum of

© 2015 John Wiley & Sons Ltd 6214 T. CAPBLANCQ ET AL. natural history, http://www.mnhn.fr), CSCF (Centre and Coenonympha gardetta at the origin of an ancestral Suisse de Cartographie de la Faune, http:// population of Coenonympha darwiniana (scenario 4) is far www.cscf.ch), Mercantour National Park (http:// superior to any other scenario. All scenarios involving www.mercantour.eu), Asters CEN (Conservatoire only cladogenesis without hybridization appeared d’Espaces Naturels de haute Savoie, http://www.aster- highly improbable given the observed data set and s.asso.fr), LPO (Ligue pour la protection des oiseaux, were thus completely rejected by the analysis. Scenario http://www.lpo.fr), GBIF (Global Biodiversity Informa- 4 supposes a unique ancestral population for C. darwini- tion Facility, http://www.gbif.org) and personal data ana and thus a common origin for the taxon, which from Michel Savourey and Roger Vila. After deleting all would have split later into the two geographical clus- points with too low precision of coordinates or with ters present today (Fig. 3B). It is worth mentioning that obvious mistakes, around 6900 points remained for the genetic differentiation and distinct geographical C. arcania, 4000 points for C. gardetta and 1000 for range of these two C. darwiniana clusters prompted us C. darwiniana distributed all along the Alps and adja- to test a scenario with two independent hybridization cent regions. These occurrence data were associated events for the origins of these two entities (scenario 5). with the 19 ‘bioclimatic’ variables available in the However, it also showed a posterior probability close to WORLDCLIM database (Hijmans et al. 2005, http://world zero and was therefore rejected. clim.org). These variables represent a range of metrics It is relevant to observe that scenario 4 is the one (mean value, variance and extremes) on global tempera- with the best fit (40% of the summary statistics calcu- ture and precipitations. For more precision, we used the lated on data sets simulated with scenario 4 are very 30-s resolution grids ( 1km2), and we kept only one close to the summary statistics calculated with the real occurrence per species in each pixel in order to avoid genetic data set) in comparison with the other four sce- overweighed values. Importation and manipulation of narios tested (only 19–33%). Our model produces simu- the data were carried out in R (R Development Core lations slightly distant from the observed genetic data Team) using dismo and raster libraries (Hijmans et al. set in the summary statistics space, but given the sim- 2015). plicity of designed scenarios, it is not surprising that An ordination method ‘outlying mean index’ (OMI, they do not perfectly match the real history of our pop- Doledec et al. 2000) was used to compare each species ulations and do not produce simulations in complete environmental niches obtained from occurrence data concordance with observed data. Moreover, type I and climatic variable correspondence. This approach (11.4%) and type II (0.05%) errors were very low in gives the position and the breadth of each species niche comparison with similar studies (Cornuet et al. 2010; along main axes calculated from a set of environmental Barres et al. 2012). The scenario 4 was thus greatly dis- variables. The niche position corresponds to the mean tinguishable from the others. Taken together, these location of the species in the environmental space, and scores in post-ABC procedure verifications strongly the breath represents the variability of the used envi- support the evolutionary scenario 4, that is a hybridiza- ronment (SD of the species values in the OMI axes, tion event, at the origin of C. darwiniana. Thuiller et al. 2004). The method is appropriate to eval- All the parameter estimates under the selected sce- uate ecological niche proximity among taxa and gives nario show quite realistic values (Fig. 3), with a diver- an idea of the environmental capacities of each species. gence time between C. arcania and C. gardetta of The analysis was performed in R using the ADE4 library. 1.7 Æ 0.05 Myr (t3 in Fig. 3C). The time for the hybridization event that gave origin to C. darwiniana (t2: 12000 Æ 100 years) and the time for the splitting Results between the two geographical populations currently observed in this species (t1: 10000 Æ 150 years) appear Genetic evidence of hybridization nonetheless quite recent. However, these two times cor- The double-digested RAD libraries produced a mean of respond to a suitable period for numerous C. arcania/ 3200 fragments (RAD-tags) with a mean coverage of 35 C. gardetta contacts and species range upheaval (see reads/fragment for the 174 individuals analysed. The Discussion). Our analysis also indicates that the initial polymorphism in these RAD-tags allowed the scoring parental contributions in the genetic composition of the of 851 independent SNPs. These markers were first ancestral C. darwiniana population (i.e. parameters ra used for the evolutionary model testing with the and 1-ra) were not 50/50. Indeed, there is evidence for approximate Bayesian computation procedure, and the larger contribution of a C. arcania genetic background results obtained are summarized in Fig. 3. The poste- (75%), suggesting multiple events of hybridization and rior probabilities (Fig. 3A) show that the scenario backcrossing between the hybrid and the parental involving hybridization between Coenonympha arcania species.

0 25 000 50 000 75 000 Populations divergence time (t1) Density

0 10 000 30 000 Admixture time (t2) t3 A

ra 1 – ra Density t2 P1 HP P2

1 500 000 2 750 000 4 000 000 Parental speciation time (t3)

t1 HP Density

P1 P3 P4 P2 0.25 0.50 0.75 Initial admixture prop. (ra)

Fig. 3 Results of the ABC procedure with the posterior probabilities of the different tested scenarios (top left), the retained scenario (bottom left, A corresponds to the ancestral population, P1 to Coenonympha arcania population, P2 to Coenonympha gardetta population, HP to an ancestral hybrid population, P3 to French Coenonympha darwiniana and P4 to Swiss C. darwiniana) and the posterior distribution of parameters corresponding to this scenario (right).

PCA identifies four well-differentiated clusters, corre- Morphological proximities among the complex sponding to the putative parental species C. arcania and A total of 510 specimens had both right forewings and C. gardetta, and the two geographical isolates of C. dar- right hindwings in good conditions and were included winiana (Fig. 4). The C. arcania and C. gardetta groups in the morphological analysis. A major result of the exhibit the highest interspecies differentiation along the morphometric analysis is the good concordance they most informative principal component (PC1 19.3%), show with the genetic clustering analysis (Fig. 5A). We while the two geographical groups of C. darwiniana take found highly significant differences in size (ANOVA: intermediate scores along the axis. The second PC P-value <2eÀ16, d.f. = 3, F-value = 172.2) and shape (7.3%) differentiates these two populations of C. dar- (MANOVA: P-value <2eÀ16, d.f. = 3, approx. F-value winiana from the other two species and points out the = 14.5) among the four previously established genetic genetic variability intrinsic to these lineages. Finally, we clusters. For these two traits, the strongest differentia- observe that the C. darwiniana individuals are clustered tion is visible between C. arcania and C. gardetta in two distinct groups along the third PC (3.7%), corre- individuals with almost no overlap in wing centroid sponding to the populations from the Tessin area and size values (Fig. 5B) and in scores of the discriminant the south France Alps, respectively. A substantial analysis first axis (Fig. 5A). This first axis of the CVA genetic structure within C. darwiniana is thus high- combines the majority of the discriminant variability lighted by this analysis and matches perfectly the dis- (80%) and separates clearly the four groups, with both continuous geographical distribution of this species.

parental species, C. darwiniana from Switzerland shows a closer phenotypic proximity with C. gardetta as shown by the large overlap between these two groups in the discriminant analysis result (Fig. 5A). Finally, the relative homogeneity of each group, in terms of wing shape and size, points out their morphological integrity. The second axis of the CVA corresponds to C. darwiniana speciﬁc variability, which emphasizes the differentiation within this taxon (two distinct genetic lineages) even if it concerns only a little part of the discriminant information (LD2, 14%).

−10 −5 0 5 10 Climatic niches analysis The climatic niches of the four populations can be visually inspected in Fig. 6. The first two axes of the OMI analysis explained almost all the variability of climatic niches and represented, respectively, 96 (OMI1) and 3% (OMI2) of the total inertia. The first one, negatively correlated with the rainfall variables and positively correlated with the temperature variables, mostly corresponds to altitudinal effects with a part of local specificities. It thus logically separates C. arcania and C. gardetta with great significance for each climatic vari- −5 0 5 10 able. On this axis, both French and Swiss C. darwiniana appear to have a climatic niche much closer to C. gardetta, with only French C. darwiniana showing a −10 –5 0 5 10 15 slight overlap with C. arcania. This is concordant with the fact that C. darwiniana completely replaces Fig. 4 Principal component analysis obtained from the SNPs C. gardetta at slightly lower elevations in southern lati- data set. The first three PCs are represented with the percent- tudes. On the other hand, the fact that Swiss C. darwini- age of information kept on each of them. The four main groups ana shows a higher degree of overlap with C. gardetta are represented by different symbols and colours: Coenonympha for climatic conditions supports, again, a closer proxim- arcania ( ), Coenonympha gardetta ( ), French Coenonympha dar- ity between these two taxa. Although the second axis winiana ( ) and Swiss C. darwiniana ( ). Ellipses contain about 2/3 of the individuals for each group. (OMI2) only explains 3% of the variability, it distin- guishes the humid continental climate influence in the central and east Alps with cool and rainy summers C. darwiniana populations taking intermediate values (high mean T°C values during the wettest quarter) and between the two putative parental species (Fig. 5A–C). the oceanic influence in the western flank of the Alps Other analyses of landmarks corresponding only to with more precipitation during the winter and spring either venation, the shape of the hindwing white band (high mean T°C values during the driest quarter). This or only the positions of hindwing eyespots confirm this axis mainly differentiates the two C. darwiniana taxa, intermediacy (not shown). Our results robustly support with the Swiss population in continental climatic condi- the intermediate position of C. darwiniana in regard to tions and the French population in mountain ranges the putative parental species, not only for easily visible with lower summer rainfall. characters in the phenotypes such as the eyespots’ col- our or their alignment, as previously highlighted by Discussion some experts, but also for most of the wing structures. The analysis also showed a significant distinction Hybrid species hypothesis for Coenonympha between the two C. darwiniana populations (MANOVA, darwiniana P-value = 1.4eÀ07, approx. F-value = 4.6), in agreement with the results from the genetic differentiation analy- By comparing different evolutionary scenarios based on ses. Furthermore, we observe that although in general numerous genetic markers, we validated the hybrid ori- C. darwiniana populations appear intermediate between gin hypothesis for the Coenonympha darwiniana lineages

A C

B LD2 (14%) −6 −4 −2 0 2 4

−6 −4 −2 0 2 4 6 LD1 (80%)

Fig. 5 Individual wing shape and patterns discrimination (A) and relative wing size differences (B) with four groups: Coenonympha arcania ( ), Coenonympha gardetta ( ), French Coenonympha darwiniana ( ) and Swiss C. darwiniana ( ). The centroid size (relative wing size index) calculated during morphometric analysis shows significant differences among the four studied groups, and a Tukey post hoc test highlighted size difference between each pair of population. A MANOVA realized on PCs scores after the Procrustes superimposition shows high differences among populations (P-value <2eÀ16), and only a discriminant analysis performed on these PCs scores is represented in the figure. And (C) is a dendrogram showing morphological distances between groups. and were able to partially reconstruct the history of the evolutionary steps that could have led to the currently complex this taxon forms with its parental species Coe- observed genetic structure among the three species. The nonympha arcania and Coenonympha gardetta. The ddRAD results, completely rejecting the scenarios involving sequencing technique allowed us to genotype a large only cladogenesis events, provide strong evidence in number of codominant markers (851 SNPs) which are favour of the hypothesis that hybridization was likely widely distributed over the species genomes (one involved in the evolutionary events that gave rise to of the advantages of the RADseq and similar tech- C. darwiniana and its resulting patterns of genetic, mor- niques, see Baird et al. 2008). This property of the mark- phological and somewhat ecological intermediacy. ers was ideal for quantifying the posterior probability Moreover, it was possible to distinguish between two of different evolutionary histories through approximate relatively close alternatives: a single hybridization event Bayesian computation methods. This procedure clearly implying a unique origin and a double hybridization favoured the hybrid species scenario giving it by far the event implying two independent origins for the two highest probability among the five evolutionary models C. darwiniana lineages. The unique hybrid origin was tested. ABC methods have already been used to com- strongly preferred during the analysis with a differenti- pare different types of evolutionary scenarios (Estoup ation of the two populations coming only later. We et al. 2004, Csillery et al. 2010; Barres et al. 2012), includ- acknowledge, however, that the simplicity of the tested ing the study of hybridization events in another butter- scenarios does not permit for a more precise reconstruc- fly species (Nice et al. 2013). But keeping in mind that tion of all the steps in the evolutionary history of the no abstract models can fully account for the inherent complex. Such a detailed reconstruction was not the complexity of species evolutionary history (Csillery objective of this work. et al. 2010) and that ABC approaches are sensitive to The results provided by the multiproxy approach are complexity differences in the compared models, we in agreement with the retained scenario and offer sub- decided to use this approach only to choose between stantial information about the consequences of three pure cladogenesis scenarios and two simple sce- hybridization on the phenotypes and ecology of the narios combining cladogenesis and hybridization (see C. darwiniana lineages. The potential implications of Fig. 2). These scenarios correspond to the most credible hybridization in species evolution remain little known

individuals. This result increases our confidence in our conclusions and adds support to the view that ddRAD sequencing can be a powerful technique for genetic clustering at a fine scale (Rubin et al. 2012; Wagner et al. 2013; Cruaud et al. 2014). If we accept a common and unique origin for the two C. darwiniana populations, which is supported by the ABC method, it is surprising to have two distinct groups in the PCA (Fig. 3). Although both taxa show intermediate patterns when compared with the parental species, French C. darwiniana is closer to C. arcania, whereas Swiss C. darwiniana is closer to C. gardetta (similar patterns in morphology and climatic conditions). This observation first led us to suppose two independent hybrid lineages as shown in North American Lycaeides butterflies (Nice et al. 2013). But the ABC mod- elling clearly rejected this possibility and converged towards a unique and common origin for the two C. darwiniana taxa. Given the recent origin of the taxa (<20 000 years), the divergence observed between them seems large to be due only to independent evolution Fig. 6 Climatic niche separation in the OMI ordination analy- after geographical splitting. A possible explanation sis. The first axis corresponds to most of the precipitation/temperature gradients (see section Material and Method for more would be a differential backcrossing rate with the par- details). The second axis corresponds to mean temperature ental species for each of the two C. darwiniana lineages. during the wettest (top of the graph) and driest (bottom) per- Under this hypothesis, independent parental genetic iod of the year. contributions subsequent to the lineage splitting observed within C. darwiniana would explain the substantial separation. The genetic stability evident now in (Mallet 2005; Abbott et al. 2013), yet the genetic, mor- the populations would be the result of their repeated phological and ecological patterns associated with a interactions with one or both parental species, possibly hybrid species can provide new insights to better driven by different environmental pressures. However, understand the possible outcomes of the phenomenon. this hypothetical scenario remains uncertain and needs to be tested. It could provide an interesting case of suc- Genetics. PCA provides an idea of genetic clustering cessive hybridization events leading to differential without making any assumptions and supports the adaptations and illustrate the tenuous equilibrium hybrid origin scenario. The two C. darwiniana lineages between backcrossing due to incomplete reproductive have intermediate scores between C. arcania and isolation of the newborn lineage and success of hybrid C. gardetta along the first PC which reflects the highest population establishment. genetic structure within the complex. Although not a formal proof of hybrid origin, this result greatly con- Morphology. Morphology can play an important role in cords with that hypothesis. The successive differentia- species isolation for butterflies and other animals tion shown by the first three PCs completely because of visual preferences associated with mating corresponds to the successive events of the hybrid spe- behaviours. This is one of the reasons why it has been cies scenario (parental differentiation, hybridization investigated for most animal hybrid species. In some event and hybrid lineages differentiation, see Fig. 3). As cases, the hybrid lineage shows extreme traits compared stated above, the first PC shows the ancestral diver- to the parental species (e.g. a sculpin lineage in Nolte gence of the lowland and alpine species with intermedi- et al. 2005 or the bat Artibeus schwartzi in Larsen et al. acy of the two lineages of C. darwiniana. The second 2010), and these extreme phenotypes are potentially axis corresponds to the distinction of the hybrid lin- advantageous in species fitness. But most hybrid species eages on the one side and the parental species on the present an intermediate morphology (e.g. the hybrid other side, showing the intrinsic genetic variability of butterfly Heliconius heurippa in Mavarez et al. 2006 and these taxa and their history since the speciation event. Salazar et al. 2010; the birds Passer italiae in Hermansen Finally, PC3 discriminates the two C. darwiniana popu- et al. 2011; Dendroica auduboni in Brelsford et al. 2011 or lations, consistent with the geographical location of the the butterfly Papilio appalachiensis in Kunte et al. 2011).

More than just being a consequence of hybridization, geographical distributions of these species. Temperature intermediate traits could play a major role in species and precipitation regimes were probably not the mecha- emergence and rise, as was the case for Heliconius heur- nism behind species isolation. Nonetheless, the climatic ippa species (Mavarez et al. 2006; Salazar et al. 2010). In niche does appear intermediate for the French C. dar- our case, we found no transgressive traits, but a strong winiana lineage, supporting the view of some degree of morphological intermediacy for most of the wing ele- adaptation to intermediate climatic conditions. That ments, including patterns and venation shape, position- makes sense considering the southern exposure and ing and size. This morphology of C. darwiniana, easily modest altitudes of the mountain ranges where they are visible when looking at specimens of each species, has found in France. Further analyses of ecological differen- been the principal argument behind the hypothesis of tiation, such as fine-scale habitat use and relationships its hybrid origin (Porter et al. 1994). The possibility that with host plants, remain necessary to paint a clear pic- intermediate values in wing traits are independent from ture of the role of ecology in the differentiation, isola- hybridization for this taxon seems unlikely when taking tion and coexistence of both C. darwiniana lineages vis a all the evidence into account. Although it could be only vis their parental species. a neutral consequence of hybridization, the homogeny Our study provides strong evidence of hybridization of taxa in terms of wing patterns led us to suspect that between two distinct Coenonympha species in the com- the intermediacy could be involved in species isolation. position of C. darwiniana and a substantial reproductive Intermediate phenotype preferences by hybrids have isolation as suggested by the genetic, morphologic and been recorded for fish species (Selz et al. 2014) and climatic clustering of the taxa. However, an important could have been involved in C. darwiniana rise. Further- additional aspect needs to be evaluated: the evidence more, the two C. darwiniana lineages differ from each that reproductive isolation between the hybrid and par- other, but within each population, the morphology is ental lineages is a direct consequence of hybridization quite homogenous. Intra- C. darwiniana morphological (Mavarez & Linares 2008; Nolte & Tautz 2010; Schumer variability is not as large as the morphological variabil- et al. 2014). It is difficult with our results to determine ity we encountered in the parental species taken whether hybridization has been the key mechanism of together. From all the phenotypic combinations poten- the speciation process (as shown for example for a scul- tially brought by the mixture of the parental entities, pin hybrid species in Nolte et al. 2005; or for a hybrid the newborn lineage probably only rose and established butterfly species in Mavarez et al. 2006) or whether a from a restricted pool of them. lineage already diverged from one of the parental species has undergone a strong introgression from the Ecology. Ecology is known to play a major role in other. However, the current stability, homogeneity and hybrid species emergence (Mallet 2007; Mavarez & specificities of C. darwiniana in its two geographical Linares 2008; Abbott et al. 2013). The mix of different populations together with the complete exclusion of genetic backgrounds leads to novel combinations that parental species in these regions suggest an important can favour the rise of a hybrid lineage by multiple role of hybrid patterns in the adaptation and persis- means (Seehausen 2004; Mallet 2007; Abbott et al. 2010, tence of this taxon. 2013) including intermediate phenotypes (Mavarez et al. 2006; Hermansen et al. 2011; Kunte et al. 2011) and/or Structure of the complex and taxonomic considerations transgressive phenotypes (Nolte et al. 2005; Gompert et al. 2006). For example, one of the main mechanisms The genetic and morphological analyses differentiated by which new hybrid lineages achieve reproductive iso- four entities within this species complex with complete lation vis-a-vis their parental species seems to be the col- concordance and in agreement with the ‘four-unit onization of habitats not used by them (Rieseberg 1997; complex’ hypothesis first proposed for this group by Nolte et al. 2005; Gompert et al. 2008; Krehenwinkel & Schmitt & Besold (2010). For each of the analyses used, Tautz 2013; Nice et al. 2013). We investigated the posi- the clustering pattern was similar, even if the climatic tioning of the hybrid species in terms of climatic condi- analysis failed to clearly discriminate the three alpine tions to determine whether an extreme climatic niche taxa (C. gardetta and the two C. darwiniana popula- was a key parameter in the hybrid species’ emergence. tions). The main differentiation was observed between However, we found that the climatic niches of C. arcania and C. gardetta, with no overlap between C. gardetta and C. darwiniana strongly overlap, espe- these two species regardless of which marker was cially for the Swiss C. darwiniana lineage, although the used. Contrastingly, the C. darwiniana individuals clus- two species only share narrow sympatric zones in the tered in two groups corresponding to the two distinct wild. This indicates that large-scale climatic conditions geographical ranges of the taxon, one in the Tessin are unlikely to be responsible for the differences in the area (mostly in Switzerland) and the other in the

© 2015 John Wiley & Sons Ltd 6220 T. CAPBLANCQ ET AL. southwest of the Alps (mostly in France). Furthermore, Acknowledgements each of the four ‘units’ shows intrinsic homogeneity in We are very grateful to Flavia entomologist association, Mer- their genetic composition, wing patterns and ecological cantour National Parc, French National Natural History (climatic) conditions. Regarding taxonomy, C. arcania Museum and Asters conservatory for their help in data collec- and C. gardetta have long been considered as different tion. We would also like to thank Yann Baillet, Pascal Dupont, species (Lafranchis 2000; Descimon & Mallet 2009; Marie-France Leccia, Gregory Guichert and Eric Drouet for Kodandaramaiah & Wahlberg 2009; Schmitt & Besold helping us collecting samples and giving us helpful advice on 2010), and our combination of approaches confirms the project. Finally, we thank Luis Daniel Llambi, Eric Bazin and Frederic Boyer for comments and revision of the manu- this taxonomic arrangement. The status of C. darwini- script. This project was funded by the ‘Conseil general de ana has been more debated (see Porter et al. 1994; Wie- l’Isere’ and a ‘CNRS INEE – APEGE’ grant. mers 1998; Schmitt & Besold 2010; Descimon & Mallet 2009). The genetic positioning of this taxon within the complex was previously examined based on protein References electrophoresis and conclusions varied from a sub- Abbott RJ, Hegarty MJ, Hiscock SJ, Brennan AC (2010) Homo- species of C. gardetta (Porter et al. 1994; Wiemers 1998) ploid hybrid speciation in action. Taxon, 59, 1375–1386. to a subspecies of C. arcania (Schmitt & Besold 2010). Abbott R, Albach D, Ansell S et al. (2013) Hybridization and The small number of markers used prevented these speciation. Journal of Evolutionary Biology, 26, 229–246. studies from detecting the hybrid composition of dar- Amaral AR, Lovewell G, Coelho MM, Amato G, Rosenbaum winiana’s genetic background, and their divergent con- HC (2014) Hybrid speciation in a marine mammal: the cly- clusions are consistent with the fact that hybridization mene dolphin (Stenella clymene). PLoS ONE, 9, e83645. can lead to incongruence in the coalescence patterns of Baird NA, Etter PD, Atwood TS, et al. (2008) Rapid SNP discovery and genetic mapping using sequenced RAD markers. genetic markers (Leache & McGuire 2006; Meng & PLoS ONE, 3, e3376. Kubatko 2009; Roos et al. 2011). Despite these consid- Barres B, Carlier J, Seguin M et al. (2012) Understanding the erations, the taxon is now mostly considered as a dif- recent colonization history of a plant pathogenic fungus ferent species (Lafranchis 2000; Tolman & Lewington using population genetic tools and Approximate Bayesian 2008). The observed differentiation of C. darwiniana in Computation. Heredity, 109, 269–279. regard to its two parental species effectively agrees Brelsford A, Mila B, Irwin DE (2011) Hybrid origin of Audu- 20 – with the ‘cluster’ definition of a species (Coyne & Orr bon’s warbler. Molecular Ecology, , 2380 2389. Catchen J, Hohenlohe PA, Bassham S, Amores A, Cresko WA 1998; Mallet 2007). Even so, one question remains (2013) Stacks: an analysis tool set for population genomics. unanswered: how should we treat the two geographi- Molecular Ecology, 22, 3124–3140. cal populations of C. darwiniana? Although they are Chessel D, Dufour AB, Thioulouse J (2004) The ade4 package-I- genetically, morphologically and climatically close and One-table methods. R News, 4,5–10. appear to have a relatively recent common origin, we Consortium THG (2012) Butterfly genome reveals promiscuous are definitely able to separate them in two distinct exchange of mimicry adaptations among species. Nature, 487, – groups. For this reason, we propose to go further than 94 98. Cornuet J-M, Santos F, Beaumont MA et al. (2008) Inferring Schmitt & Besold (2010) by giving species status to population history with DIY ABC: a user-friendly approach each of the two currently recognized subspecies of to approximate Bayesian computation. Bioinformatics, 24, C. darwiniana. Their disjunct geographical ranges 2713–2719. together with their differences in genetic composition, Cornuet J-M, Ravignie V, Estoup A (2010) Inference on popula- wing structure and patterns and, to a lesser extent, tion history and model checking using DNA sequence and their climatic niches seem to us as enough evidence to microsatellite data with the software DIYABC (v1.0). BMC 11 justify their species status. We propose to raise the Bioinformatics, , 401. Cornuet J-M, Pudlo P, Veyssier J et al. (2014) DIYABC v2.0: a names previously used for the subspecies (Schmitt & software to make approximate Bayesian computation infer- Besold 2010) to a species level: Coenonympha darwiniana ences about population history using single nucleotide poly- for the Tessino population and Coenonympha macromma morphism. DNA sequence and microsatellite data. for the southwest Alps population. Bioinformatics, 30, 1187–1189. Further studies are needed now to investigate the Coyne JA, Orr HA (1998) The evolutionary genetics of specia- mechanisms (ecological, behavioural or others) that tion. Philosophical Transactions of the Royal Society of London B: 353 – maintain reproductive isolation among these four pop- Biological Sciences, , 287 305. Cruaud A, Gautier M, Galan M et al. (2014) Empirical assess- ulations into the wild. Such information could greatly ment of RAD sequencing for interspecific phylogeny. Molecu- help to better understand the factors which have lar Biology and Evolution, 31, 1272–1274. favoured the hybrid lineages rise and provide evi- Csillery K, Blum MGB, Gaggiotti OE, Francßois O (2010) dence concerning the potential adaptive role of Approximate Bayesian Computation (ABC) in practice. hybridization. Trends in Ecology & Evolution, 25, 410–418.

Descimon H, Mallet J (2009) Bad species. In: Ecology and Evolu- chondrial data: evidence for a misleading mitochondrial tion of European Butterflies (eds Shreeve TG, Dennis RHL), pp. gene tree. Molecular Phylogenetics and Evolution, 39, 628–644. 219–249. Cambridge University Press, Cambridge. Lukhtanov VA, Shapoval NA, Anokhin BA, Saifitdinova AF, Doledec S, Chessel D, Gimaret-Carpentier C (2000) Niche sepa- Kuznetsova VG (2015) Homoploid hybrid speciation and ration in community analysis: a new method. Ecology, 81, genome evolution via chromosome sorting. Proceedings of the 2914–2927. Royal Society of London B: Biological Sciences, 282, 20150157. Elgvin TO, Hermansen JS, Fijarczyk A et al. (2011) Hybrid spe- Mallet J (2005) Hybridization as an invasion of the genome. ciation in sparrows II: a role for sex chromosomes? Molecular Trends in Ecology & Evolution, 20, 229–237. Ecology, 20, 3823–3837. Mallet J (2007) Hybrid speciation. Nature, 446, 279–283. Estoup A, Beaumont M, Sennedot F, Moritz C, Cornuet JM Mavarez J, Linares M (2008) Homoploid hybrid speciation in (2004) Genetic analysis of complex demographic scenarios: animals. Molecular Ecology, 17, 4181–4185. spatially expanding populations of the cane toad, Bufo mari- Mavarez J, Salazar CA, Bermingham E et al. (2006) Speciation by nus. Evolution, 58, 2021–2036. hybridization in Heliconius butterflies. Nature, 441, 868–871. Gompert Z, Fordyce JA, Forister ML, Shapiro AM, Nice CC Mayr E (1963) Animal Species and Evolution. Harvard University (2006) Homoploid hybrid speciation in an extreme habitat. Press, Cambridge, MA. Science, 314, 1923–1925. Meng C, Kubatko LS (2009) Detecting hybrid speciation in the Gompert Z, Forister ML, Fordyce JA, Nice CC (2008) Wide- presence of incomplete lineage sorting using gene tree incon- spread mito-nuclear discordance with evidence for introgres- gruence: a model. Theoretical Population Biology, 75,35–45. sive hybridization and selective sweeps in Lycaeides. Merot C, Mavarez J, Evin A et al. (2013) Genetic differentiation Molecular Ecology, 17, 5231–5244. without mimicry shift in a pair of hybridizing Heliconius spe- Gross BL, Rieseberg LH (2005) The ecological genetics of cies (Lepidoptera: Nymphalidae). Biological Journal of the Lin- homoploid hybrid speciation. Journal of Heredity, 96, 241–252. nean Society, 109, 830–847. Hermansen JS, Sæther SA, Elgvin TO et al. (2011) Hybrid speci- Nice CC, Gompert Z, Fordyce JA et al. (2013) Hybrid speciation ation in sparrows I: phenotypic intermediacy, genetic admix- and independent evolution in lineages of alpine butterflies. ture and barriers to gene flow. Molecular Ecology, 20, 3812– Evolution, 67, 1055–1068. 3822. Nolte AW, Tautz D (2010) Understanding the onset of hybrid Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) speciation. Trends in Genetics, 26,54–58. Very high resolution interpolated climate surfaces for global Nolte AW, Freyhof J, Stemshorn KC, Tautz D (2005) An inva- land areas. International Journal of Climatology, 25, 1965–1978. sive lineage of sculpins, Cottus sp. (Pisces, Teleostei) in the Hijmans RJ, Phillips S, Elith JL, Elith J (2015) dismo: Species Rhine with new habitat adaptations has originated from Distribution Modeling. Available from http://cran.r-projec- hybridization between old phylogeographic groups. Proceed- t.org/web/packages/dismo/index.html. ings of the Royal Society of London B: Biological Sciences, 272, Hohenlohe PA, Amish SJ, Catchen JM, Allendorf FW, Luikart G 2379–2387. (2011) Next-generation RAD sequencing identifies thousands Paradis E, Blomberg S, Bolker B et al. (2014) ape: Analyses of of SNPs for assessing hybridization between rainbow and west- Phylogenetics and Evolution. Available from http://cran.r-pro- slope cutthroat trout. Molecular Ecology Resources, 11,117–122. ject.org/web/packages/ape/index.html. Jombart T (2008) adegenet: a R package for the multivariate Pardo-Diaz C, Salazar C, Baxter SW et al. (2012) Adaptive analysis of genetic markers. Science, 24, 1403–1405. introgression across species boundaries in Heliconius butter- Kim M, Cui M-L, Cubas P et al. (2008) Regulatory genes con- flies. PLoS Genetics, 8, e1002752. trol a key morphological and ecological trait transferred Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE between species. Science, 322, 1116–1119. (2012) Double digest RADseq: an inexpensive method for de Kodandaramaiah U, Wahlberg N (2009) Phylogeny and novo SNP Discovery and genotyping in model and non- biogeography of Coenonympha butterflies (Nymphalidae: Satyri- model species. PLoS ONE, 7, e37135. nae)–patterns of colonization in the Holarctic. Systematic Ento- Porter AH, Schneider RW, Price BA (1994) Wing pattern and mology,34,315–323. allozyme relationships in the Coenonympha arcania group, Krehenwinkel H, Tautz D (2013) Northern range expansion of emphasising the C. gardetta -darwiniana contact area at Bell- European populations of the wasp spider Argiope bruennichi wald, Switzerland (Lepidoptera, Satyridae). Nota Lepidoptero- is associated with global warming–correlated genetic admix- logica, 17, 155–174. ture and population-specific temperature adaptations. Molec- Rhymer JM, Simberloff D (1996) Extinction by hybridization ular Ecology, 22, 2232–2248. and introgression. Annual Review of Ecology and Systematics, Kunte K, Shea C, Aardema ML et al. (2011) Sex chromosome 27,83–109. mosaicism and hybrid speciation among tiger Swallowtail Rieseberg LH (1997) Hybrid origins of plant species. Annual butterflies. PLoS Genetics, 7, e1002274. Review of Ecology and Systematics, 28, 359–389. Lafranchis T (2000) Les Papillons de jour de France, Belgique Roos C, Zinner D, Kubatko LS et al. (2011) Nuclear versus et Luxembourg et leurs chenilles. Collection Parthenope, edi- mitochondrial DNA: evidence for hybridization in colobine tions Biotope. monkeys. BMC Evolutionary Biology, 11, 77. Larsen PA, Marchan-Rivadeneira MR, Baker RJ (2010) Natural Rubin BER, Ree RH, Moreau CS (2012) Inferring phylogenies hybridization generates mammalian lineage with species from RAD sequence data. PLoS ONE, 7, e33394. characteristics. PNAS, 107, 11447–11452. Salazar C, Baxter SW, Pardo-Diaz C et al. (2010) Genetic evi- Leache AD, McGuire JA (2006) Phylogenetic relationships of dence for hybrid trait speciation in Heliconius butterflies. horned lizards (Phrynosoma) based on nuclear and mito- PLoS Genetics, 6, e1000930.

Schmitt T, Besold J (2010) Upslope movements and large scale expansions: the taxonomy and biogeography of the Coenonym- pha arcania – C. darwiniana –C. gardetta butterfly species com- J.M., L.D. and T.C. designed the study. T.C. and D.R. plex. Zoological Journal of the Linnean Society, 159, 890–904. performed ddRAD experiments. T.C. performed bioin- Schumer M, Rosenthal GG, Andolfatto P (2014) How common formatics treatment of the resulted sequences, as well – is homoploid hybrid speciation? Evolution, 68, 1553 1560. as the genetic, morphologic and climatic analysis. T.C. Schwarz D, Matta BM, Shakir-Botteri NL, McPheron BA (2005) wrote the manuscript, and all authors contributed sub- Host shift to an invasive plant triggers rapid animal hybrid speciation. Nature, 436, 546–549. stantially to revisions. Schwenk K, Brede N, Streit B (2008) Introduction. Extent, processes and evolutionary impact of interspecific hybridization in animals. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 363, 2805–2811. Data accessibility Seehausen O (2004) Hybridization and adaptive radiation. Trends in Ecology & Evolution, 19, 198–207. All the following data are available at Dryad Digital Selz OM, Thommen R, Maan ME, Seehausen O (2014) Beha- Repository (doi:10.5061/dryad.j4389) vioural isolation may facilitate homoploid hybrid speciation in cichlid fish. Journal of Evolutionary Biology, 27, 275–289. Demultiplexed ddRAD sequencing data Song Y, Endepols S, Klemann N et al. (2011) Adaptive introgres- The stacks catalog of loci is available sion of anticoagulant rodent poison resistance by hybridiza- The variants files – both the unfiltered variants and the tion between old world mice. Current Biology, 21, 1296–1301. variants used in the ABC analyses Sun Y, Abbott RJ, Li L et al. (2014) Evolutionary history of Pur- Morphologic data in the form of the coordinates of the ple cone spruce (Picea purpurea) in the Qinghai-Tibet Plateau: 40 morphometric landmarks for each used individual. homoploid hybrid origin and Pleistocene expansion. Molecu- A table with occurrence locations used for climatic lar Ecology, 23, 343–359. Thuiller W, Lavorel S, Midgley G, Lavergne S, Rebelo T (2004) niche comparisons Relating plant traits and species distributions along biocli- Digitalized pictures of the butterfly wings are freely matic gradients for 88 leucadendron taxa. Ecology, 85, 1688– available upon request from T. Capblancq. 1699. Tolman T, Lewington R (2008) Guide des Papillons D’Europe et D’Afrique du Nord. Delachaux et Niestle, Paris, France. Vonlanthen P, Bittner D, Hudson AG et al. (2012) Eutrophica- Supporting information tion causes speciation reversal in whitefish adaptive radia- tions. Nature, 482, 357–362. Additional supporting information may be found in the online ver- Wagner CE, Keller I, Wittwer S et al. (2013) Genome-wide sion of this article. RAD sequence data provide unprecedented resolution of Table S1 Table of the different studied populations with geo- species boundaries and relationships in the Lake Victoria graphical coordinates, country, species and number of used – cichlid adaptive radiation. Molecular Ecology, 22, 787 798. individuals for the genetic and the morphological analysis. Wiemers M (1998) Coenonympha darwiniana - a hybrid taxon? New insights through allozyme electrophoresis (Lepidoptera, Table S2 Table of the different prior parameter ranges and dis- Nymphalidae, Satyrinae). tributions included in the ABC analysis. Yakimowski SB, Rieseberg LH (2014) The role of homoploid hybridization in evolution: a century of studies synthesizing Fig. S1 Landmarks used for geometric morphometrics, shown genetics and ecology. American Journal of Botany, 101,1–12. on a ventral view of Coenonympha arcania wings.