applyparastyle "body/p[1]" parastyle "Text_First"

Biological Journal of the Linnean Society, 2019, 126, 95–113. With 6 figures. Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 Inferring the biogeography and demographic history of an endangered in Europe from multilocus markers

LAURENCE DESPRÉS1*, CLÉMENT HENNIAUX1, DELPHINE RIOUX1, THIBAUT CAPBLANCQ1, SARA ZUPAN2, TATJANA ČELIK3, MARCIN SIELEZNIEW4, LUCIO BONATO5 and GENTILE FRANCESCO FICETOLA1,6

1Université Grenoble Alpes, LECA UMR5553, CNRS, F-38000 Grenoble, 2University of Primorska, Faculty of Mathematics, Natural Sciences and Information Technologies, Glagoljaška 8, SI-6000 Koper, 3Research Centre of the Slovenian Academy of Sciences and Arts, Jovan Hadži Institute of Biology, Novi trg 2, SI-1000 Ljubljana, Slovenia 4Laboratory of Evolutionary Biology and Ecology, Institute of Biology, University of Bialystok, 15- 245 Białystok, 5Department of Biology, Università degli Studi di Padova, Padova, 6Department of Environmental Science and Policy, Università degli Studi di Milano, Milano, Italy

Received 5 June 2018; revised 25 September 2018; accepted for publication 26 September 2018

The genetic structure of a is influenced by its history and by current gene flow. Using a population genomics approach, we infer the demographic history of the false ringlet ( oedippus) in Europe based on 1594 genome-wide double digest restriction site associated DNA sequencing loci from 96 individuals (32 localities) sam- pled throughout the fragmented species range. In contrast to the weak geographical structure in mitochondrial DNA, a clear nuclear differentiation was observed between the westernmost Atlantic localities, those from the western Alps and all other sampled localities. Mountain ridges were the main factor explaining population divergence at the European scale, while isolation by distance was found at a regional scale. We applied Approximate Bayesian Computation in a coalescent framework to infer past and contemporary demographic parameters. The best scenario suggested a first divergence between French and all other European populations around 66 000 years ago, such that the species survived the Last Glacial Maximum in at least two distinct areas separated by the Alps. This scenario fits species distribution modelling identifying variation of suitable areas with past climatic modifications. The Atlantic and western Alps lineages separated about 6000 years ago. Strong population decline was inferred in these lineages during historical time, in agreement with multiple records of recent decline of this species in Europe.

ADDITIONAL KEYWORDS: Coenonympha oedippus – ddRADseq – demographic history – genetic diversity – glacial refuges – mtDNA – population size – species distribution modelling.

INTRODUCTION conservation biology for decades. The Pleistocene cold periods in the northern hemisphere have influenced The genetic structure of a species reflects both its history the distribution of species with range fluctuations and ongoing gene flow. Characterizing demographic linked to climatic variations. During glaciations, many histories and identifying the main environmental temperate European taxa were restricted to southern factors shaping genetic variation at different spatial ice-free refuges (Taberlet et al., 1998). The present scales have been a major focus in evolutionary and distribution of most species in Europe has resulted from a northward recolonization from those southern *Corresponding author. E-mail: laurence.despres@univ-greno- refuges after the Last Glacial Maximum (LGM), about ble-alpes.fr 21 kya (Strandberg et al., 2011). Under this hypothesis,

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 95 96 L. DESPRÉS ET AL. the southern part of Europe should present the highest as a consequence of human activities (Lhonore & genetic diversity, in contrast to the recently recolonized Lagarde, 1999). Despite the wide distribution range of

northern part (Besold et al., 2008; Patricelli et al., 2013). this butterfly, from Atlantic to Pacific coasts, very little Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 However, genetic analyses have identified numerous information is available on the intraspecific pattern additional extra-mediterranean refuges, thus strongly of genetic diversity in C. oedippus, and on the genetic modifying the biogeographical view of Europe (Schmitt connections between populations. Coenonympha & Varga, 2012). This picture is complicated in species oedippus is generally considered to be a monotypic with a wide Eurasian distribution, where other potential species, despite many subspecific and infrasubspecific eastern refuges could have existed, with possible taxa proposed by different taxonomists (Bozano, 2002), admixture occurring between diverging lineages although genetic studies have often found cryptic during post-glacial recolonization (Grassi et al., 2008). genetic structure within butterfly species (Hebert Geographical distribution and genetic structure are et al., 2004; Dincă et al., 2011, 2015; Ritter et al., 2013; affected not only by the evolutionary history of species Lavinia et al., 2017). but also by dispersal abilities, present demographic We used double digest restriction site-associated characteristics – especially fluctuations in population DNA sequencing (ddRADseq) to identify thousands of size – and by habitat fragmentation (Keyghobadi, 2007; genetic markers without any prior knowledge on the Louy et al., 2013). Many lowland insect species have genome of Coenonympha (Peterson et al., 2012). We been particularly affected by human impact via the used nuclear genetic diversity to test for alternative intensification of agriculture (insecticide spraying, land demographic histories (splits, expansions, recent draining). Although they were abundant a few decades declines) of European populations by an Approximate ago, they now show highly fragmented populations Bayesian Computation (ABC) approach in a coalescent with high extinction risk (Hallmann et al., 2017). framework. Finally, we performed species distribution Analyses of genetic diversity within and between modelling (MaxEnt) to identify current and past populations provide key information for conservation climatically suitable areas for C. oedippus in Europe. of endangered species given that they allow us to infer important demographic parameters such as historical and contemporary effective population sizes, dispersal MATERIAL AND METHODS rates across populations and levels of consanguinity. This knowledge is necessary to guide conservation Study species actions such as the creation of corridors favouring the Coenonympha oedippus (Fabricius, 1787) (: natural re-colonization of suitable habitats, or the best ) is a univoltine Palearctic sedentary choice of individuals for a successful re-location. To species flying mainly in June and July (Čelik et al., date, most phylogeographical studies at the continental 2009; Verovnik et al., 2012; Bonato et al., 2014). It is scale have been based on mitochondrial DNA (mtDNA), a hygrophilous insect inhabiting mostly wet meadows and population genetic analysis have focused on a few and fens, where caterpillars feed on Carex spp. as well allozyme or microsatellite markers and required the as on Molinia caerulea, but at the southern range limit analysis of many individuals per population, which in north-eastern Italy and Slovenia it can be found also was not always possible for endangered species. With on drier grasslands (Habeler, 1972; Čelik et al., 2015). the development of high-throughput sequencing Coenonympha oedippus is one of the most endangered technologies, it is now possible to infer the genetic butterfly species in Europe and is listed in Annexes II diversity within and between populations with only and IV of the Habitats Directive as well in Appendix a few individuals per population, because the low II of the Bern Convention (Van Swaay et al., 2010). It number of individuals sampled is partly compensated became extinct in three of the 14 countries where it for by a very high number of loci genotyped (Nazareno had been recorded (Van Swaay & Warren, 1999), i.e. et al., 2017). in (Pastoralis & Reiprich, 1995), In this study we used high-throughput genotyping (Staub & Aistleitner, 2006) and (Dušej besides the classical mitochondrial barcode (partial et al., 2010). In most of other countries C. oedippus is CO1) to uncover the past and current factors involved declining and in the last 100 years it has disappeared in shaping the genetic structure of one of the most from many localities, e.g. in , where only endangered butterfly species in Europe, the false one meta-population is still present in Bavaria (Bräu ringlet (Coenonympha oedippus). Although the species et al., 2010), and France (Lhonore & Lagarde, 1999), is distributed across Eurasia from western France where the species went extinct in the Paris region, to (Bozano, 2002), its range is today highly and is currently present in only two disconnected and fragmented, especially throughout Europe (Kudrna distant regions: between the Atlantic coast and the et al., 2015), because its habitat (mainly wetlands) has Pyrenees (south-west France) and in the Rhône and been significantly reduced and is still disappearing Isère valleys in the Western Alps (eastern France). In

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 97 the former region, populations are locally abundant in Sample collection marshes (Poitou-Charente) and in managed maritime A total of 32 localities were sampled through most of the pine forests (Landes) (van Halder et al., 2008), while distribution range of C. oedippus in Europe, from the Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 in the latter region the species is restricted to three westernmost populations on the Atlantic coast (south- protected marshes (Lavours-Ain, Chautagne-Savoie west France) to those in the eastern Polish lowland, and Montfort-Isère) (Varin, 1964). The range of including many isolated localities around the Alpine C. oedippus in Slovenia also contracted, and it now has mountain range (Fig. 1; Table 1). Pairwise distances a disjunct distribution there (Čelik & Verovnik, 2010): between sampled localities ranged from 400 m to up to the predominantly limestone region of south-west 2600 km. The 32 sampling localities were categorized into Slovenia, and marshy areas in central Slovenia south five geographical regions based on the presence of natural of Ljubljana. In contrast, over 100 populations are barriers to dispersion (mountain range and distance): known to occur in northern Italy, but often restricted Atlantic (including populations from the Atlantic coast to small isolated areas (Bonelli et al., 2010). Knowledge to the Pyrenees foothills), Western Alps (including of the past and present distribution of C. oedippus in localities from the Rhône and Isère valleys), Southern eastern Europe is still inadequate. For example, in Alps (including localities from northern Italy to central Poland the species was considered extinct in the 1970s, Slovenia), Northern Alps (including localities from but over the last three decades several sites have been and Bavaria) and East European (including discovered in the eastern part of the country as a result Polish localities). Because of the endangered status of of intensification of inventory activities and therefore C. oedippus and in order to have the lowest impact as little is known about recent trends (Sielezniew et al., possible on the populations, only two to five males per 2010; Sielezniew, 2012). sampled locality were caught using entomological nets at In the last two decades, given the dramatic decline the end of the flying period (in July). They were kept dry of populations throughout the western part of its (<1 month) and, after wing removal, the body was kept European range, several studies have investigated in 70% ethanol at −20 °C for genetic analysis, except for the factors limiting population viability (for a review samples from Slovenia, which were kept at −80 °C. To test see Čelik et al., 2015). Current threats include land whether even less invasive sampling could be performed reclamation for agriculture, land drainage and on this endangered species, we used only two legs from urban expansion, but also natural reforestation of each of three specimens from Ger (Atlantic region). Legs grasslands. were kept at −80 °C until extraction.

Figure 1. Map of Europe with the 32 sampled localities (coloured dots often overlapping), assigned to five geographical re- gions: Atlantic (orange), Western Alps (dark purple), Southern Alps (blue), Northern Alps (pink) and East European (green) (see Table 1 for details).

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 98 L. DESPRÉS ET AL. Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 du Marais de Lavours Provincia di Varese Provincia di Varese Friulano di Storia naturale Friulano Institute of Biology Institute of Biology Institute of Biology für Naturschutz und für Naturschutz Landschaftspflege Institution CEN Aquitaine CEN Aquitaine CEN Aquitaine CEN Aquitaine CEN Poitou-Charente CEN Poitou-Charente CEN Aquitaine CEN Aquitaine Réserve Naturelle CEN Savoie CEN Savoie CEN Savoie LECA LECA LECA University of Torino Servizi Agricoltura e Foreste, Agricoltura e Foreste, Servizi Servizi Agricoltura e Foreste, Agricoltura e Foreste, Servizi LECA Sezione Entomologica, Museo Sezione Entomologica, SRC SASA, Jovan Hadži Jovan SRC SASA, SRC SASA, Jovan Hadži Jovan SRC SASA, SRC SASA, Jovan Hadži Jovan SRC SASA, Amt für Umwelt, Vaduz Amt für Umwelt, Amt für Umwelt, Vaduz Amt für Umwelt, Bayerische Akademie Bayerische Collector R. Dupéré R. T. Le Moal T. N. Déjean N. N. Déjean N. M. Holthoff M. M. Holthoff M. V. Labourel V. V. Labourel V. C. Guérin C. P. Freydier P. P. Freydier P. P. Freydier P. L. Després L. L. Després L. L. Després L. S. Bonelli S. G. Forni G. D. Baratelli D. F. Ficetola F. P. Glerean P. T. Č elik T. Č elik T. Č elik U. Hiermann U. U. Hiermann U. M. Braü M. Number for CO1 1 2 1 0 1 0 1 0 3 1 0 1 3 2 1 1 1 2 2 0 2 2 3 1 1 2 Number for ddRADSeq 3 3 3 3 3 2 3 3 3 3 3 3 3 3 3 3 5 4 3 3 3 3 3 3 3 3 Number of individuals 3 3 3 3 3 2 3 3 3 3 3 3 3 3 3 3 5 4 3 3 3 3 3 3 3 3 Geographical region Atlantic Atlantic Atlantic Atlantic Atlantic Atlantic Atlantic Atlantic Western Alps Western Western Alps Western Western Alps Western Western Alps Western Western Alps Western Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Southern Alps Northern Alps Northern Alps Northern Alps Landes Atlantiques Administrative region Pays Basque/ Pays Pyrénées- Gironde Gironde Vienne Vienne Dordogne Dordogne Ain Savoie Savoie Savoie Isère Torino Torino Biella Varese Varese Treviso Udine Nova Gorica Sežana Ljubljana Liechtenstein Liechtenstein Oberbayern Country France France France France France France France France France France France France France Italy Italy Italy Italy Italy Italy Italy Slovenia Slovenia Slovenia Liechtenstein Liechtenstein Germany di Strada barje and number of individuals used for nuclear (ddRADSeq) and mitochondrial ( CO1 ) sequencing (see also (ddRADSeq) and mitochondrial Sample localities of Coenonympha oedippus and number individuals used for nuclear

Locality Mees Ger Bélin-Béliet Louchats Les Ardillasses Les Ragouillis Echourgnac Le Périer Lavours Chindrieux Nord Chindrieux Sud Prés-Crottis Montfort Caselette La Mandria Massazza Biandronno Villadosia Cornuda Castions Opatje Selo Gorjansko Ljubljansko Ruggell Schaan Munich ) Fig. 1 1. Table Code MEES Co BBL LOU PUY HOL ECH PES LV CNC CSC PCC MTF CSB LMD MAS BIA VIL COR TRBS COE CD LB RUG SCH MUN

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 99

DNA extraction DNA was extracted from the complete thorax of each

individual, with the exception of the three specimens Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 from Ger, using the DNeasy Blood and Tissue Kit (Qiagen, Germany) according to the manufacturer’s instructions and stored at −20 °C. For those three specimens, DNA was extracted from two legs using a cetyl trimethyl ammonium bromide (CTAB) chloroform/isoamyl alcohol protocol (Doyle & Doyle, 1987). University (UMCS), Lublin University (UMCS), University (UMCS), Lublin University (UMCS), University (UMCS), Lublin University (UMCS), University (UMCS), Lublin University (UMCS), Institution University of Bialystok University of Bialystok Maria Curie-Skłodowska Maria Curie-Skłodowska Maria Curie-Skłodowska Maria Curie-Skłodowska

ddRADseq library preparation and SNP calling A ddRADseq experiment was conducted on 104 samples (98 specimens and six replicates; Table 1) in Collector M. Sielezniew M. M. Sielezniew M. K. Palka K. K. Palka K. K. Palka K. K. Palka K. three libraries using a modified version of the protocol previously described (Peterson et al., 2012; Capblancq et al., 2015). Briefly, 200 ng of DNA template from each individual was double-digested with 10 units each of Number for CO1 0 0 1 0 2 1 SbfI–HF and MspI (New England Biolabs Inc., USA) at 37 °C for 1 h using the CutSmart buffer provided with the enzymes. Digestion was further continued together with the ligation of P1 (individually indexed) Number for ddRADSeq 3 2 2 3 3 3 and P2 adapters by adding 10 units of T4 DNA ligase (New England Biolabs), adapters P1 and P2 and 1 µL of 10 mm ribo-ATP (New England Biolabs) in each sample. The digestion-ligation was performed in a thermocycler (60 cycles of 2 min digestion at 37 °C Number of individuals 3 3 3 3 3 3 and 2 min ligation at 16 °C, followed by final heat inactivation of the enzymes at 65 °C for 10 min). An equal volume of all the digested-ligated individuals was pooled and purified with Agencourt AMPure XP beads (Beckman Coulter, France). After migration on a 1.6% agarose gel, fragments between 250 and 500 bp were Geographical region East European East European East European East European East European East European excised and purified by using QIAquick Gel Extraction Kit (Qiagen). Each ddRAD library was amplified in ten independent replicates of 15 PCR cycles (initial denaturation 10 min, 98 °C; 15 cycles of 98 °C for 10 s, 66 °C for 30 s and 72 °C for 1 min; followed by a final extension period of 10 min at 72 °C) in a final volume Administrative region Podlasie Podlasie Lublin Lublin Lublin Lublin of 20 µL with 1 µL of DNA template, 10 mm of dNTPs, 10 µm of each PCR primer (Peterson et al., 2012) and 2 U/µL of Taq Phusion-HF (New England Biolabs). The ten PCR products were pooled and purified with a QIAgen MinElute PCR Purification Kit (Qiagen). Country Poland Poland Poland Poland Poland Poland Each library was sequenced on an Illumina Hi-Seq 2500 Illumina sequencer (1/10 lane per library, paired- end 2 × 125 bp, Fasteris SA, Switzerland). Sequencing errors per lane (PhiX control) were very low (0.26, 0.82 and 0.27% for three libraries, respectively) which means that convergent sequencing errors (the same Continued

Locality Szorce Uhowo Zawadowka Kamien Antoniowka Swaryczow error occurring independently at the same nucleotide position in the same read) are very unlikely. Reads with a depth of coverage <5 were excluded from further Code SZO UHO ZAW KAM ANT SWA 1. Table analyses. Genotyping errors (locus and allelic dropout)

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 100 L. DESPRÉS ET AL. were estimated by comparing six replicate pairs (three assumption of correlated allele frequencies among inter-library and three intra-library replicate pairs). localities. For each K, ten runs were performed.

The c. 68 million DNA reads obtained were used to Estimated log probabilities (ln P(D)) were averaged Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 call single nucleotide polymorphism (SNP) genotypes across runs and compared to determine the posterior with the STACKS pipeline (Catchen et al., 2013) as probability of each K using Clumpak (Kopelman follows: the process_radtags function was first run et al., 2015). The best K was selected using the ΔK to demultiplex the data and filter the reads based method (Evanno et al., 2005) in Structure Harvester on their quality. We removed reads <100 nucleotides (Earl, 2012). As one single K value only provides an in length and cut all reads to this value, resulting incomplete picture of overall population structure, we in more than 92% of total reads being retained. On explored the pattern of population structure within average, we retained 89% of total reads by individual the main clusters detected (Janes et al., 2017). after removing reads of low quality or with uncalled To assess the respective roles of geographical bases (options -q, -c and -r). Individuals with <100 000 distance and orographic barriers in population genetic reads were discarded (N = 2, Table 1 and Supporting differentiation of this species, we performed a multiple Information, Table S1). Only SbfI reads were retained linear regression on distance matrix (MRM) in for de novo assembly on each individual using a package R ‘ecodist’ where genetic distance FST/(1 − FST) maximum of seven mismatches to merge two stacks was treated as a response matrix. The straight-line into a polymorphic locus (-M; ustacks function). This geographical distances (square-root transformed) and threshold was chosen after inspecting the effect of presence of mountain ranges higher than 1100 m a.s.l. increased values of M on the proportion of polymorphic were set as the explanatory matrices. The available data loci (see also Fig. S1). Highly repetitive stacks and over- on historical and present distribution of the species in merged tags were dropped using the ‘Removal’ (-r) and Europe showed that 700 m a.s.l. is the highest altitude the ‘Deleveraging’ (-d) options. A catalogue of the loci for most localities of the species (Verovnik et al., 2012; from all the individuals was built, with a maximum Bonato et al., 2014). Exceptions are scarce localities in of nine mismatches for merging two individual loci the southern foothills of the Alps where the species was (-n; cstacks function). Loci within each individual found on semi-open dry grasslands at 750–1100 m a.s.l. were searched against the catalogue (sstacks function) (Čelik & Rebeušek, 1996; Jugovic et al., 2018). and an SNP dataset was produced with the genotype For phylogenetic inference, we used every locus of each individual for every polymorphic position present in at least 59 individuals of the whole sampling (populations function). (N = 96), including invariant positions. Heterozygote positions were coded according to IUPAC. We used full sequences rather than just SNPs because full Genetic diversity and genetic structure sequences are preferable from the perspective of estimation branch length and topological accuracy (Leaché et al., For genetic diversity indices and analysis of population 2015). A maximum likelihood (ML) phylogenetic tree structure, only SNPs present in more than 60% of the was generated using RAxML (Stamatakis, 2014) with whole sampling were retained to avoid an excess of 100 rapid bootstrap inferences following search for missing data. SNPs with a minimum allele frequency the best ML tree using the GTR+G model for rate lower than 5% were removed from the dataset and heterogeneity. only one polymorphic site was kept for each RAD-tag (‘write_random_snp’ option) in order to analyse only unlinked polymorphisms. CO1 sequencing and phylogeographical Genetic diversity by individual (individual analysis heterozygosity, or observed heterozygosity Ho), within We sequenced also a mitochondrial marker for 38 each sampled locality (population genetic diversity, individuals representative of 24 localities from all the or expected heterozygosity He), and between all main regions studied (Table 1). CO1 was amplified population pairs was assessed using the R package using the primer pairs LCO–HCO and Jerry–Pat hierfstat. FIS and FST estimates were calculated (Wahlberg & Freitas, 2007) (PCR protocol: 95 °C for according to Weir & Cockerham (1984). Confidence 10 min; 35 cycles of 95 °C for 30 s, 50 °C for 30 s and intervals (95%) for Ho, He and FIS were assessed by 72 °C for 30 s; followed by a final extension period 1000 bootstraps across loci. of 72 °C for 7 min) and sequenced by Genewiz Clustering of individuals into homogenous genetic Company (UK). The resulting chromatograms were clusters ranging from K = 1 to K = 32 was tested using visualized in the software BIOEDIT v.7.2.5 (Hall, STRUCTURE 2.3.4 (Falush et al., 2003). For each 1999) and aligned using ClustalW and by eye. Five run, a burn-in period of 5000 steps was followed by sequences of C. oedippus available in BOLD (http:// 20 000 iterations under the admixture model and the www.barcodinglife.org/) were added to the multiple

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 101 alignment. They originate from the following Species distribution models localities: Ruggell, Liechtenstein (BOLD accession We used Maximum Entropy Modelling (MaxEnt) to code: PHLAF624-11); Munich, Oberbayern (GenBank build species distribution models (SDMs) relating Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 code: GU707147); Romano d`Ezzelino, Vicenza the distribution of C. oedippus to climatic variables, province, Italy (BOLD code: PHLSA390-11); Obluchye, and to assess potential distribution changes since the European Russia (GenBank code: EU920755); LGM. MaxEnt is a presence-background modelling Tavolzhanka, (BOLD code: LOWA191- tool; comparative analyses showed that MaxEnt is 06). Additionally, sequences from eight outgroup among the SDMs with the best predictive performance species were chosen based on the most recently (Elith et al., 2006, 2011). Models were calibrated on published phylogeny of Coenonymphina the basis of 453 presence records, obtained by carefully (Kodandaramaiah & Wahlberg, 2009), including six mining the literature and selectively extracting Coenonympha species [C. tullia, C. hero, C. glycerion, records from the Global Biodiversity Information C. nolckeni, C. phryne (previously under Triphysa) Facility (GBIF, 2016), and from our own surveys (see and C. myops (previously under Lyela); GenBank also Table S2). As climatic variables, we considered a codes EU920762, EU920750, EU920749, EU920754, set of variables that represent the climatic conditions EU920739, EU920741], and two species of strictly experienced by the species through the year: mean related genera (Heteronympha merope and Mydosama summer temperature, mean winter temperature, terminus; EU92073, DQ338765). An ML phylogenetic temperature seasonality, total precipitation during tree was generated after selecting for the best model summer, and total precipitation during winter. of molecular evolution using Mega7 v.7.0.14 (Kumar Variables were extracted from the Worldclim dataset et al., 2016). at the 10 arc-primes resolution (approx. 15 km within the study area) (Hijmans et al., 2005). For analyses, Demographic scenarios and population we only retained one presence record per cell. We built size inferences models with linear, quadratic and hinge features; we ran preliminary models with a range of different Competing hypotheses regarding population regularization multipliers (1, 2, 3, 4, 5, 7 and 10) and divergence at the European scale were compared selected the best regularization multiplier on the basis based on the nuclear data using ABC as implemented of the corrected Akaike’s information criterion (AICc) in DIYABC v.2.1 (Cornuet et al., 2014). Based on the (Warren & Seifert, 2011). We used a ten-fold cross- results from STRUCTURE, which identified three main validation to assess the predictive performance of genetic clusters (from west to east), we tested whether the best AICc model (Nogués-Bravo, 2009). Predictive the geographically intermediate lineage (Western performance was evaluated on the basis of the area Alps region) was more closely related to the western under the curve of the receiver operator plot of the (Atlantic region) or to the eastern lineage (remaining test data (AUC), averaged over the ten replicated runs regions). For each scenario we allowed population size (Manel et al., 2001). We assumed that a cell is suitable changes after each split time. The competing scenarios if its suitability value was higher than the 10% test were set using uniformly broadly distributed priors presence threshold (averaged over the cross-validated (102–107 individuals for population sizes and 1000– runs); we assumed a high suitability if suitability 700 000 years for divergence times). As C. oedippus was higher than 0.5 (Pearson et al., 2007; Elith et al., is a univoltine species (Čelik et al., 2009; Verovnik 2011). Models were then projected to the mid-Holocene et al., 2012; Bonato et al., 2014), divergence times were (6 kya) and LGM (21 kya) conditions, using the MPI- directly estimated in years. For each scenario, 100 000 ESM model. When projecting to past climates, we datasets were simulated and the posterior probability assessed whether models were projected into climatic was computed by performing a logistic regression conditions different from those found in the calibration on the 1% of simulated data closest to the observed climate using clumping and evaluating if climatic dataset (Cornuet et al., 2014). Summary statistics of variables are outside the training range (Elith et al., observed/simulated dataset comparisons were mean 2010). genetic diversity within populations, and FST and Nei’s distances among populations, using only SNPs with a minimum allele frequency >5%. We further estimated divergence time and tested for recent bottlenecks RESULTS within the western and intermediate lineages, and we tested alternative splitting hypotheses within the Nuclear genetic diversity eastern lineage. More details on alternative models, More than 60 million high-quality reads were obtained model checking and parameter estimates are given in with an average 600 000 reads per sample. A total of Figure S6. 102 samples (96 individuals and six replicates), with

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 102 L. DESPRÉS ET AL. an average of 7500 loci per individual (mean coverage adjusted R2 = 43.7%). Population diversity (He) ranged per locus: 60) were kept for genetic analysis (see also from 0.152 (in the locality MTF, Western Alps region)

Table S1). The three samples from Ger (Atlantic region) to 0.266 (in one locality from the Southern Alps region) Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 passed the filter, indicating that DNA extracted from (Table 2; Fig. 2B), and was again significantly lower two legs can be sufficient to successfully achieve the in the Western Alps region than in other regions 2 ddRADseq experiment. A total of 1594 loci (100 bp each, (F4,27 = 13.22, P < 0.01, adjusted R = 61%). Inbreeding including 126 monomorphic loci) present in >60% of the coefficients (FIS) ranged between 0.291 and 0.442 whole sampling (i.e. ≥59 individuals) were considered. and did not differ significantly between regions

A total of 1314 independent SNPs were retained by (F4,27 = 2.146, P > 0.05; Fig. 2C). selecting one random SNP per locus with minimum allele frequency >5%. Genotyping errors ranged from 1 to 10% for locus dropout (absence of a locus in the replicate) Population structure but allelic dropout (heterozygous position genotyped as The STRUCTURE Bayesian assignment approach homozygous in the replicate) was always ≤1.5%. showed that C. oedippus populations are genetically Observed heterozygosity of individuals (Ho) ranged differentiated across Europe. The highest likelihood from 0.109 (Western Alps region) to 0.169 (Atlantic was for K = 7 and ΔK was maximum for K = 3, with region) (Table 2; Fig. 2A), and was significantly a secondary maximum for K = 8 (Fig. 3 and see also lower in the localities from the Western Alps region Fig. S2). At K = 3, a primary separation was found compared to other regions (F4,27 = 7.01, P < 0.01, between the following three groups of populations, from

Table 2. Genetic diversity indices per sampled locality, based on the ddRADSeq dataset

Geographical region Code Ho 95% CI Ho He 95% CI He FIS 95% CI Fis

Atlantic MEES 0.1520 0.1377–0.1671 0.2448 0.2278–0.2617 0.3958 0.3277–0.4306 Atlantic Co 0.1385 0.1244–0.1522 0.2170 0.2005–0.2349 0.3300 0.2999–0.4248 Atlantic BBL 0.1512 0.1367–0.1661 0.2467 0.2298–0.2654 0.4171 0.3326–0.4471 Atlantic LOU 0.1658 0.1512–0.1802 0.2480 0.2310–0.2656 0.3663 0.2758–0.3887 Atlantic PUY 0.1606 0.1475–0.1755 0.2573 0.2407–0.2739 0.4277 0.3236–0.4286 Atlantic HOL 0.1450 0.1289–0.1611 0.2336 0.2124–0.2559 0.4250 0.3068–0.4500 Atlantic ECH 0.1623 0.1463–0.1781 0.2437 0.2254–0.2629 0.3215 0.2715–0.3905 Atlantic PES 0.1691 0.1549–0.1835 0.2370 0.2201–0.2531 0.2983 0.2267–0.3426 Western Alps LV 0.1183 0.1052–0.1321 0.2029 0.1838–0.2216 0.3510 0.3482–0.4834 Western Alps CNC 0.1408 0.1275–0.1542 0.2023 0.1853–0.2203 0.3081 0.2422–0.3645 Western Alps CSC 0.1293 0.1160–0.1449 0.1924 0.1763–0.2095 0.2922 0.2612–0.3923 Western Alps PCC 0.1335 0.1200–0.1470 0.1950 0.1788–0.2113 0.3195 0.2541–0.3823 Western Alps MTF 0.1094 0.0959–0.1228 0.1525 0.1380–0.1667 0.2915 0.2040–0.3544 Southern Alps CSB 0.1538 0.1400–0.1678 0.2386 0.2198–0.2561 0.3390 0.2937–0.4119 Southern Alps LMD 0.1355 0.1220–0.1505 0.2325 0.2132–0.2504 0.4419 0.3599–0.4753 Southern Alps MAS 0.1574 0.1443–0.1721 0.2520 0.2349–0.2693 0.3701 0.3188–0.4289 Southern Alps BIA 0.1470 0.1342–0.1595 0.2274 0.2123–0.2432 0.3380 0.2988–0.4021 Southern Alps VIL 0.1626 0.1498–0.1758 0.2660 0.2503–0.2826 0.3773 0.3376–0.4368 Southern Alps COR 0.1314 0.1169–0.1462 0.2263 0.2069–0.2443 0.4090 0.3545–0.4772 Southern Alps TRBS 0.1420 0.1189–0.1663 0.2473 0.2162–0.2788 0.4372 0.3332–0.5147 Southern Alps COE 0.1572 0.1417–0.1726 0.2333 0.2152–0.2504 0.3490 0.2657–0.3835 Southern Alps CD 0.1525 0.1382–0.1673 0.2425 0.2241–0.2603 0.3747 0.3135–0.4317 Southern Alps LB 0.1534 0.1394–0.1676 0.2251 0.2079–0.2419 0.3300 0.2553–0.3782 Northern Alps SCH 0.1488 0.1360–0.1633 0.2484 0.2307–0.2654 0.3707 0.3446–0.4553 Northern Alps MUN 0.1400 0.1264–0.1545 0.2282 0.2113–0.2461 0.4173 0.3280–0.4442 Northern Alps RUG 0.1526 0.1381–0.1672 0.2266 0.2095–0.2428 0.3015 0.2636–0.3854 East European SZO 0.1383 0.1222–0.1543 0.2157 0.1954–0.2353 0.4038 0.2857–0.4333 East European UHO 0.1631 0.1463–0.1805 0.2404 0.2166–0.2647 0.2979 0.2367–0.4017 East European ZAW 0.1651 0.1481–0.1823 0.2350 0.2149–0.2562 0.3918 0.2214–0.3683 East European KAM 0.1517 0.1377–0.1655 0.2374 0.2181–0.2539 0.3656 0.3050–0.4154 East European ANT 0.1518 0.1358–0.1659 0.2536 0.2326–0.2737 0.4060 0.3363–0.4579 East European SWA 0.1500 0.1360–0.1634 0.2315 0.2149–0.2487 0.3766 0.2911–0.4080

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 103 Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019

Figure 2. Boxplots for (A) observed heterozygosity (Ho), (B) expected heterozygosity (He) and (C) FIS, within localities sampled in each of the five European geographical regions: Atlantic, Western Alps, Southern Alps, Northern Alps and East European. west to east: (1) all localities from the Atlantic region; genetic distances (P = 0.42) of sampled localities. (2) all localities from the Western Alps region, with the Genetic differentiation between localities separated by possible exception of MTF (Isère valley); and (3) all mountain ranges higher than 1100 m was significantly localities from the remaining regions, i.e. Northern Alps, higher (P < 0.01) compared to other localities. At the Southern Alps and East European region. The locality regional scale, a strong and significant pattern of MTF remained uncertainly assigned to one or the other isolation by distance (IBD) was found across the of the two latter groups. At K = 4 all localities from the localities from the Western Alps region (R2 = 0.98). Southern Alps region formed a distinct group with Moderate and significant IBD was found across the exception of the easternmost one (LB, Ljubljansko localities within the Atlantic region (R2 = 0.55), and barje, Slovenia), which remained uncertainly assigned. low but significant IBD was found across those from At K = 5 the locality from the Isère valley (MTF) was the Southern Alps region (R2 = 0.11). Instead, no IBD separated from all the others. At K = 6 the localities from was observed across populations in the East European the Northern Alps region were separated from those region (see also Fig. S3). IBD was not tested across the in the East European region. At K = 7, some evidence three localities from the Northern Alps. of admixture was retrieved between the latter groups. In the ML tree based on the nuclear dataset (Fig. S4), At K = 8, the two localities from south-west Slovenia the two replicates for each replicated individual (N = 6) (CD, COE), which are the only sampled localities from grouped together with 100% bootstrap support (BS), the dry ecotype (Jugovic et al., 2018), separated clearly whereas individuals from a single locality grouped from the remaining localities of the Southern Alps together only for some localities, especially those from region, which showed some differentiation between a the Northern Alps region and East European region. western group and an eastern group. Relationships between localities were overall poorly

Pairwise FST values ranged from 0 to 0.36 (see supported, with only a few well-supported groups, Table S3). At the entire European scale, there was from west to east: (1) all localities from the Atlantic no significant correlation between geographical and region (99% BS), within which the single locality from

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 104 L. DESPRÉS ET AL. Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019

Figure 3. Results of Bayesian genetic clustering (STRUCTURE) based on 1314 unlinked SNPs, for the two most likely numbers of clusters K = 3 and K = 8 (see also Fig. S2 where the probability of assignment to a given cluster is indicated for each individual). the Pyrenees was well separated (Co; 100% BS); (2) not in the Atlantic populations. It differed by only one all localities from the Western Alps region (87% BS), mutation from the second-most represented haplotype with two subgroups, i.e. the locality from Isère valley (Hap_3), which was found in the Atlantic and in the (MTF; 100% BS) and all others from the Rhône valley Southern Alps regions. Most other haplotypes were (100% BS); (3) the two localities from Liechtenstein very similar (one to five mutations from either Hap_1 (RUG, SCH; 99% BS); (4) both localities of the dry or Hap_3), including the haplotype of a previously ecotype from the Southern Alps region (CD, COE; sequenced individual from European Russia (Hap_17), 100% BS); and (5) two localities in the East European but with the remarkable exception of the haplotype region (KAM, UHO; 80% BS). Of the five geographical of the single individual sampled from central Asia regions, only the Atlantic and Western Alps were (Hap_16). The latter had 3.5–3.7% divergence from strongly supported as monophyletic groups in the ML all the other haplotypes, while pairwise divergence tree (BS ≥ 87%). within Europe did not exceed 0.44% (Fig. 4B). Many haplotypes were found in the Southern Alps region (ten haplotypes for 16 sampled specimens) while only Mitochondrial diversity the most common haplotype was found in the Western A total of 17 haplotypes were found for the CO1 Alps region. fragment sequenced from 43 individuals: 38 individuals Mitochondrial nucleotide diversities (π and θ; in our sample (Table 1) and five specimens from the Fig. 4B) were highest in the Southern Alps and the BOLD database (Fig. 4A). The most common haplotype Northern Alps regions (>0.0035), moderate in the (Hap_1) was shared by 16 individuals out of a total of 43 Atlantic region and East European region (0.0015– and was found across central and eastern Europe but 0.0020), and zero in the Western Alps region.

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 105 Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019

Figure 4. Mitochondrial variability (partial CO1 gene; 630 bp). A, minimum spanning network for the 17 haplotypes found in 43 specimens of Coenonympha oedippus. Each colour represents a different geographical region, and the size of each pie represents the number of specimens sharing the same haplotype. B, within-region genetic diversity expressed as π (pair- wise nucleotide divergence), θ and hdiv (haplotype diversity). N is the number of sequenced samples and h the number of haplotypes.

A phylogenetic analysis of the CO1 sequences population size change (Figs 5B, S6B). An impressive (Fig. S5) did not recover well-supported relationships decline was detected in all populations during the last between different localities of C. oedippus. 2000 years, with those in the Atlantic region and in the Rhône valley declining from ~106 to 104 individuals, and the single population of the Isère valley (MTF) Historical demographic scenarios and from 106 to ~500 individuals. When focusing on population size inference the remaining European populations, the analyses The most likely scenario (Figs 5A, S6A) was a first were not able to distinguish between alternative divergence between the western groups of populations scenarios for the splitting and/or admixture among the (Atlantic region and Western Alps region) and the other populations of the East European region, those from European populations, around 66 kya [95% confidence the Northern Alps region and those in the Southern interval (CI): 30–95 kya], followed by a much more Alps region (results not shown). recent divergence between the Atlantic group of populations and the Western Alps group, around 6 kya (95% CI: 1–10 kya). These divergence events were Species distribution models associated with moderate population size changes, MaxEnt models showed excellent performance in and the inferred population sizes for the three lineages describing present-day distributions in central and ranged between 105 and 106 individuals at splitting western Europe (Fig. 6A); average AUC across the times; a strong population decline was observed in the cross-validated runs was 0.94 (SD = 0.028). Summer Western Alps lineage during the last 1000 years with temperature and summer precipitation were the an estimated current median effective population size variables with the strongest contribution to the of only around 8000 individuals (Figs 5A, S6A). model (35% and 29%, respectively). Suitability in When focusing on the western populations (Atlantic the mid-Holocene (6 kya) was similar to the present- and Western Alps regions), the scenario involving a day situation, with broader highly suitable areas population bottleneck was much more likely than a (Fig. 6B). The situation was very different in the LGM scenario involving only population divergence without (21 kya). In this period, the model suggested three

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 106 L. DESPRÉS ET AL. Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019

Figure 5. Results of the ABC demographic analysis. A, analysis within Europe (N = 96 individuals; 1314 SNPs), for the split between the three main genetic lineages: Atlantic, Western Alps and remaining regions corresponding to the ‘East lineage’. Only the best scenario is shown (posterior probability 0.997; 95% CI: 0.995–0.999). B, analysis within France (N = 38 individuals; 1123 SNPs), for the split between the Atlantic region, Rhône valley (LV, PCC, CNC and CNS localities) and Isère valley (MTF locality). Only the best scenario (with bottlenecks) is shown (posterior probability 0.77; 95% CI: 0.401–1.000). N: estimated median of the effective population size; CI95%: 95% confidence interval of the effective popula- tion size; t: 95% confidence interval of the time since splitting or since bottleneck in years ago (ya). For more details on the ABC models see Figure S6. separate suitable areas. Two small suitable areas 2017). In contrast to the low mitochondrial variation were at opposite ends of the Pyrenean chain (Fig. 6C). detected by sequencing the cytochrome oxidase 1 Furthermore, a broader suitable area was present mitochondrial (CO1) gene, which is routinely used from northern Prealps to western Pannonian Plain, as a barcode in butterflies, we found large variation southern Prealps, Padan Plain and north-western part in nuclear genetic diversity across Europe, and of the Balkan peninsula. identified localities where loss of genetic diversity poses threats to species persistence. Furthermore, we show that two butterfly legs provide enough material DISCUSSION to apply this technique to highly endangered species without killing specimens. Indeed, leg removal Potential of using ddRADseq genotyping on was shown to have no effect on butterfly survival endangered species or behaviour (Koscinski et al., 2011). Finally, the We demonstrate here for the first time the potential application of new statistical techniques (ABC and of a recently developed genotyping next-generation coalescent inferences, niche modelling) on this NGS sequencing (NGS) strategy (ddRADseq) to identify dataset representative of the current distribution of thousands of genetic markers from an endangered the species in Europe with sampling of only three butterfly species, without any prior knowledge individuals per locality allowed us to infer complex on the genome, at reasonable cost. We show that demographic scenarios. genotyping only three individuals per locality allowed us to infer robust genetic population indices and phylogeographical patterns, at a spatio-temporal Biogeographical history of C. oedippus scale suitable for conservation purposes. This result The analysis of nuclear and mitochondrial variation of is concordant with other recent studies which C. oedippus specimens collected throughout most of the have demonstrated that working with large NGS European range of the species, together with species datasets provides sufficient robustness in population distribution modelling, suggests that the ancestors of genetics analysis to use only a few individuals per all current European populations survived the last population (Willing et al., 2012; Nazareno et al., Pleistocene glacial period in at least two refuges, west

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 107 Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019

Figure 6. Results of species distribution models: suitability for Coenonympha oedippus under (A) present-day, (B) mid-Hol- ocene (6 kya) and (C) Last Glacial Maximum (21 kya) climatic conditions. Here, 0.0883 is the 10% test presence threshold; 0.342 is the maximum test sensitivity plus specificity threshold; and values >0.5 indicate very high suitability, as 0.5 is the typical suitability of presence points used by MaxEnt for calibration (Elith et al., 2011).

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 108 L. DESPRÉS ET AL. and east of the Alps. Species distribution modelling glacial period (Atlantic and Adriatic–Mediterranean) based on current occurrence of the species suggests that are also the most likely origin for one other satyrine

only three areas were potentially suitable for the species butterfly, Maniola jurtina (Schmitt et al., 2005). Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 during the LGM: two very small coastal areas west and The distinct, highly divergent haplotype found east of the Pyrenees, and one much larger eastern area in Kazakhstan suggests that there was at least one from the northern Prealps, through the Adriatic region, other more eastern refuge for the species during to the north-western part of the Balkan peninsula the Pleistocene glaciations, but this refuge did not and northern Italy. The ABC demographic inferences contribute to the recolonization of Europe after the suggest that two distinct lineages survived the LGM, LGM. More samples from central and eastern Europe one most probably in an area west of the Alps, and (e.g. , , Belarus, ) and from another somewhere north-east and/or south-east of the Asia would be necessary to reconstruct the post-glacial Alps. The Atlantic and Western Alps lineages diverged biogeographical history of C. oedippus throughout its only after the LGM, about 6 kya. During mid-Holocene whole distribution range. warming, the climatically suitable area increased The current estimated effective population size is dramatically northwards especially in Western Europe far higher for the lineage distributed in central and (Bell & Walker, 1992), initially probably allowing gene eastern Europe (106 individuals) and that surviving in flow between populations in different French regions. the Atlantic region (105) compared to the populations of However, the subsequent forest expansion in this part the Western Alps (103). Furthermore, the strong decline of Europe is the most likely reason for the separation observed in the Western Alps is recent, with dramatic between the populations of the Atlantic regions and population decline estimated from 106 down to 103 in those of the Western Alps region. Indeed, simulations of recent centuries. This scenario based on nuclear markers potential land cover after LGM in Europe consistently is also supported by the highest mtDNA haplotype suggest that extensive forests occupied large areas diversity and divergence found in the Southern Alps of Europe, particularly north and west of the Alps and Northern Alps regions, suggesting that different (Strandberg et al., 2011). In contrast, weak genetic haplotypes were randomly lost during/following structure was observed within the ‘East’ lineage fragmentation in the Atlantic and Western Alps regions. (STRUCTURE and ML analyses), which is concordant Indeed, although in both latter regions the haplotypes with the MaxEnt prediction of a large, climatically found were common haplotypes in Europe, none was suitable area north, south and east of the Alps and with shared between the two regions. The lack of mtDNA the large population size inferred throughout time for variability across all populations from the Western Alps this lineage. region, where only one haplotype was found, supports a The absence of geographical structure in the dramatic population decline in this region. variation of the mitochondrial CO1 marker, with closely related haplotypes present from western France to Russia, and the most common haplotype found from Genetic differentiation across populations the Western Alps to eastern Poland, also suggests In contrast to the mitochondrial marker, the ddRADseq rapid expansion of the species throughout Europe multilocus analysis allowed us to differentiate the after the last glacial period. The current locations samples according to their geographical location, with a in the Southern Alps region account for most of the clear E–W and N–S population genetic differentiation. mitochondrial haplotype diversity, but comparable However, in accordance with the analysis of mtDNA diversity persists in the small locations still surviving haplotypes, we found little support for highly in the Northern Alps region, suggesting that the two diverging lineages in Europe: there were only few regions were interconnected throughout the Würm informative sites (i.e. differently fixed nucleotides glaciation, without strong population bottlenecks, across populations), and the relationships between as suggested by the large climatically suitable area localities were overall poorly resolved. Of the main during LGM, with four small highly suitable areas genetic groups well supported both in phylogenetic predicted by MaxEnt analysis along the northern, (ML tree) and in population genetic (STRUCTURE) eastern and southern Prealps. The star-like patterns analyses, three correspond to subspecies previously in the mitochondrial haplotype network suggest described based on wing coloration pattern variation, two distinct expansion events, presumably from the i.e. aquitanica Varin, 1952 in Atlantic region (including Northern–Eastern–Southern Alps region for Hap_1 Charente-Maritime, Landes and Pyrenees), rhodanica (with unique derived haplotypes in the Southern Alps, Varin, 1964 in the Rhône valley and herbuloti Varin, Northern Alps and East European regions) and from 1952 in the Isère valley. Further morphological the Atlantic region for Hap_3 (with some derived analysis would be required to test whether these haplotypes in this region only). Interestingly, two distinct genetic groups can indeed be distinguished allopatric centres of differentiation during the last based on phenotypic traits.

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 109

At the European scale, the pairwise genetic 0.33) between this locality and the neighbouring ones differentiation (FST) was moderate (0.04–0.15 on suggests that gene flow has been interrupted between

average) given the wide geographical range sampled. MTF and other Western Alps localities (LV, PCC, CNC, Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 The main barrier to gene flow appeared to be mountain CSC). The latter have similar levels of genetic diversity ranges rather than geographical distance, with the (around 0.20, all 95% CIs overlapping): LV is a protected Alps and Massif Central separating the three main site, only a few hectares in size, but geographically genetic lineages. This is in line with the known biology close (~5 km) to a larger habitat in Savoie (several of this lowland species, which has never been recorded hundred hectares, three populations sampled: PCC, at an elevation above 1100 m, although it is often CNC, CSC). Genetic differentiation between these two found in mountain foothills (e.g. Pyrenées, Alps). areas is low, suggesting that ongoing gene flow has The weak genetic structure and the IBD patterns probably helped to maintain a relatively high genetic observed within geographical regions suggest that diversity, which could reflect the legacy of formerly populations were presumably more connected in large and interconnected populations. Alternatively, the recent past. Indeed, historical records from the sufficiently large populations might have persisted beginning of the 20th century suggest a much larger over time to maintain the high genetic diversity found distribution throughout France, Switzerland and in this protected area of the Rhône valley. Germany. Low-altitude wetlands and oligotrophic In comparison to the Western Alps region, localities grasslands are the habitats that have suffered the from the Atlantic region are far more diversified and most from intensive agriculture development, land connected, with pairwise FST usually not exceeding 0.10, draining and urbanization since the early 20th century except for the southernmost locality from the Pyrenees throughout Europe, especially in western Europe (Co), which is more than 100 km from the closest sampled (Levers et al., 2016). For some other butterfly species, localities (see Table S3), and is also the least diverse their current genetic structure may be explained population within the region (He = 0.217, Table 2). This better based on past rather than present distribution suggests that populations in the Atlantic region are still (Orsini et al., 2008; Sielezniew et al., 2012). large and genetically connected, in accordance with a large climatically suitable area in this region, and with the persistence of several mitotypes. Genetic erosion and drift The same pattern of IBD is observed among Lowest genetic diversity was found in the Rhône localities throughout northern Italy to central Slovenia and Isère valleys (Western Alps region), with < 0.20. (Southern Alps region). However, locality LB from The locality MTF (Isère valley) showed significantly central Slovenia was found admixed with localities of less diversification than any other locality, while the both East European and Southern Alps regions, but highest diversity was observed in localities from the not with the nearby localities CD and COE, which Atlantic, Southern Alps and East European regions, formed a distinct genetic cluster (Fig. 3). Locality LB with > 0.25. The low genetic diversity observed in comprises wet grasslands, while CD and COE are in some localities, especially MTF (0.15), suggests allele a distinct karstic habitat, sub-mediterranean dry loss through genetic drift in isolated populations with grasslands in different successional stages up to light low effective size. This genetic erosion observed only woods (Čelik & Verovnik, 2010), which are drier and in the Western Alps populations is concordant with from phytosociological aspect different from the typical the observation of a single mitotype throughout this wet grasslands where C. oedippus is mostly found region, as opposed to all other regions, where several in Europe. Interestingly, a recent analysis of wing mitotypes were found, indicating the persistence of morphology found ecotype differences for wing size large populations. and shape (Jugovic et al., 2018). The distinctiveness For the Western Alps region, population size has of this dry habitat might have limited gene flow and/ been estimated to range from a few hundred in the or promoted local adaptations, but a larger sampling Isère valley (MTF) to several thousand individuals in (in terms of both individuals from the two contrasted the Rhône valley (PCC) by capture–mark–recapture habitats and SNPs across the genome) would be (unpublished data). These direct estimates from the necessary to test these hypotheses. field fit well with population size estimates from gene Our results also show that the fragmentation of coalescence simulations, suggesting that our previous C. oedippus populations in France started long before distributions and model selection through the ABC intensive agriculture and urbanization, as the split procedure are realistic. The available habitat consists between Atlantic and Western Alps lineages was of several hundred hectares in the Rhône valley, while it estimated to >15 kya. This fragmentation into two is restricted to 6 ha of protected area in the Isère valley lineages in France is unlikely to be due only to the (MTF locality). Despite the nearest populations being ecological barrier of the Massif Central, although we only about 60 km apart, the high FST values (around found that mountain ranges are relevant barriers to

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 110 L. DESPRÉS ET AL. gene flow in this species. During Holocene warming, We thank all the collectors and their respective institu- natural reforestation could also explain the population tions/administrations for giving permission to collect

fragmentation of this open-land butterfly. In Europe, specimens of C. oedippus in protected areas for this Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 human populations expanded as early as 11 kya, study: in France, T. Le Moal, N. Déjean and V. Labourel linked to revolutions in Neolithic agriculture that (CEN Aquitaine), M. Holthoff (CEN Poitou-Charentes), sustained substantial population growth (Barker, P. Freydier (CEN Savoie), CEN Isère, and C. Guérin 2009). By maintaining semi-natural open habitats, (Réserve Naturelle du Marais de Lavours, Ain); in human activities might have mitigated the negative Germany, M. Braü; in Liechtenstein, U. Hiermann and impact on C. oedippus in terms of land monopolization the Amt für Umwelt in Vaduz; in Italy, the Ministero for agriculture and urbanization. Most of the current dell’Ambiente e della Tutela del Territorio e del Mare, and European populations of the species are found in more particularly D. Baratelli and G. Forni (Provincia protected areas that are managed in order to maintain di Varese), J. Guidici, R. Filipello (Parco La Mandria), an open environment. In addition to openness of the P. Glerean (Sezione Entomologica, Museo Friulano di habitat, oligotrophic soil favouring grasses and sedges Storia Naturale) and S. Bonelli (University of Torino); in (i.e. larval host plants with erect leaf orientation) over Poland, K. Pałka (Maria Curie-Skłodowska University, other herbs (with plane leaf orientation) appears to be Lublin); G. Nève (University Marseille, France) for giving a key factor for pre-adult stages (Čelik et al., 2015), as access to preserved biological material; and P. Dupont it creates microhabitats with herb vegetation structure (Museum National d’Histoire Naturelle, France) and providing suitable microclimatic conditions. Yann Baillet (Association FLAVIA) for sharing informa- Therefore, forest extension after the LGM is a more tion on past and present species distributions in France. likely factor than early agriculture in explaining the This work was supported by the Conseil Départemental fragmentation of C. oedippus during the Neolithic. In de l’Isère (Pôle Biodiversité). Poland and Belarus the butterfly is restricted almost exclusively to some fen communities (Sielezniew, 2012; REFERENCES Kulak & Yakovlev, 2018), which could have been relatively stable open ecosystems before recent human-induced Barker G. 2009. The agricultural revolution in prehistory: drainage and eutrophication (Jabłońska et al., 2014). why did foragers become farmers? Oxford: Oxford University Currently they have to be managed to prevent ecological Press. succession and one population went extinct before that Bell M, Walker M. 1992. Late Quaternary environmental need was realized (Sielezniew et al., 2010). However, many change: human and physical perspectives. Harlow: Longman historical population extinctions have been recorded in Scientific and Technical. France (Lhonore & Lagarde, 1999), Switzerland (Dušej Besold J, Schmitt T, Tammaru T, Cassel‐Lundhagen A. et al., 2010), Germany (Bräu et al., 2010), Italy (Bonelli 2008. Strong genetic impoverishment from the centre of dis- et al., 2010; Bonato et al., 2014), Slovenia (Čelik et al., 2015), tribution in southern Europe to peripheral Baltic and iso- Slovakia (Pastoralis & Reiprich, 1995) and Bulgaria (Staub lated Scandinavian populations of the pearly heath butterfly. & Aistleitner, 2006) during the last century, indicating that Journal of Biogeography 35: 2090–2101. Bonato L, Uliana M, Beretta S. 2014. Farfalle del Veneto: the population decline of this butterfly species is ongoing atlante distributivo. Padua: Marsilio. due to intensive agriculture and urbanization. Bonelli S, Canterino S, Balletto E. 2010. Ecology of Coenonympha oedippus (Fabricius, 1787)(Lepidoptera: Nymphalidae) in Italy. Oedippus 26: 25–30. CONCLUSION Bozano G. 2002. Guide to the butterflies of the Palearctic region: Satyridae Part 3: Tribe , Subtribes Melanargiina Despite a highly fragmented distribution in Europe, and Coenonymphina, Genera Melanargia, Coenonympha, populations of C. oedippus still conserve a high level of Triphysa and Sinonympha. Milan: Omnes Artes. genetic diversity, except in a few locations (e.g. MTF) Bräu M, Dolek M, Stettmer C. 2010. Habitat requirements, where there is evidence for genetic erosion and lack larval development and food preferences of the German of connectivity. This high genetic diversity appears population of the False Ringlet Coenonympha oedippus to be a legacy of previously large and interconnected (Fabricius, 1787)(Lepidoptera: Nymphalidae) – research on populations that expanded after the LGM from at least the ecological needs to develop management tools. Oedippus two distinct refuges probably separated by the Alps. 26: 41–51. Capblancq T, Despres L, Rioux D, Mavarez J. 2015. Hybridization promotes speciation in Coenonympha butter- ACKNOWLEDGMENTS flies. Molecular Ecology 24: 6209–6222. Catchen J, Hohenlohe PA, Bassham S, Amores A, We acknowledge three anonymous reviewers for help- Cresko WA. 2013. Stacks: an analysis tool set for popula- ful comments on a previous version of the manuscript. tion genomics. Molecular Ecology 22: 3124–3140.

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 111

Čelik T, Brau M, Bonelli S, Cerrato C, Vreš B, Balletto E, GBIF. 2016. GBIF Occurrence Download. doi:10.15468/ Stettmer C, Dolek M. 2015. Winter-green host-plants, lit- dl.ywhpmz. Available at: https://www.gbif.org/ Grassi F, De Mattia F, Zecca G, Sala F, Labra M. 2008. ter quantity and vegetation structure are key determinants Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 of habitat quality for Coenonympha oedippus in Europe. Historical isolation and Quaternary range expansion of Journal of Insect Conservation 19: 359–375. divergent lineages in wild grapevine. Biological Journal of Čelik T, Rebeušek F. 1996. Atlas ogroženih vrst dnevnih met- the Linnean Society 95: 611–619. uljev Slovenije. Ljubljana: Slovensko entomološko društvo Habeler H. 1972. Zur Kenntnis der Lebensraume von Štefana Michielija. Coenonympha oedippus F. (Lep. Satyridae). Nachrichtenblatt Čelik T, Verovnik R. 2010. Distribution, habitat preferences der Bayerischen Entomologen 21: 51–54. and population ecology of the False Ringlet Coenonympha Hall TA. 1999. BioEdit: a user-friendly biological sequence oedippus (Fabricius, 1787)(Lepidoptera: Nymphalidae) in alignment editor and analysis program for Windows 95/98/ Slovenia. Oedippus 26: 7–15. NT. Nucleic Acids Symposium Series 41:95–98. Čelik T, Vreš B, Seliškar A. 2009. Determinants of Hallmann CA, Sorg M, Jongejans E, Siepel H, Hofland N, within-patch microdistribution and movements of endan- Schwan H, Stenmans W, Muller A, Sumser H, Horren T, gered butterfly Coenonympha oedippus (Fabricius, 1787) Goulson D, de Kroon H. 2017. More than 75 percent (Nymphalidae: ). Hacquetia 8: 115–128. decline over 27 years in total flying insect biomass in pro- Cornuet J-M, Pudlo P, Veyssier J, Dehne-Garcia A, tected areas. PLoS One 12: e0185809. Gautier M, Leblois R, Marin J-M, Estoup A. 2014. Hebert PD, Penton EH, Burns JM, Janzen DH, DIYABC v2.0: a software to make approximate Bayesian Hallwachs W. 2004. Ten species in one: DNA barcoding computation inferences about population history using sin- reveals cryptic species in the neotropical skipper but- gle nucleotide polymorphism, DNA sequence and microsatel- terfly Astraptes fulgerator. Proceedings of the National lite data. Bioinformatics 30: 1187–1189. Academy of Sciences of the United States of America 101: Dincă V, Lukhtanov VA, Talavera G, Vila R. 2011. 14812–14817. Unexpected layers of cryptic diversity in wood white Leptidea Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. butterflies. Nature Communications 2: 324. 2005. Very high resolution interpolated climate surfaces for Dincă V, Montagud S, Talavera G, Hernández-Roldán J, global land areas. International Journal of Climatology 25: Munguira ML, García-Barros E, Hebert PD, Vila R. 1965–1978. 2015. DNA barcode reference library for Iberian butter- Jabłoń ska E, Falkowski T, Chormań ski J, flies enables a continental-scale preview of potential cryptic Jarzombkowski F, Kłosowski S, Okruszko T, diversity. Scientific Reports 5: 12395. Pawlikowski P, Theuerkauf M, Wassen MJ, Kotowski W. Doyle J, Doyle J. 1987. Genomic plant DNA preparation from 2014. Understanding the long term ecosystem stability of a fresh tissue-CTAB method. Phytochemical Bulletin 19: 11–15. fen mire by analyzing subsurface geology, eco-hydrology and Dušej G, Wermeille E, Carron G, Ziegler H. 2010. nutrient stoichiometry–case study of the Rospuda Valley Concerning the situation of the False Ringlet Coenonympha (NE Poland). Wetlands 34: 815–828. oedippus (Fabricius, 1787)(Lepidoptera: Nymphalidae) in Janes JK, Miller JM, Dupuis JR, Malenfant RM, Switzerland. Oedippus 26: 38–40. Gorrell JC, Cullingham CI, Andrew RL. 2017. The K= 2 Earl DA. 2012. STRUCTURE HARVESTER: a website conundrum. Molecular Ecology 26: 3594–3602. and program for visualizing STRUCTURE output and Jugovic J, Zupan S, Buzan E, Čelik T. 2018. Variation implementing the Evanno method. Conservation Genetics in the morphology of the wings of the endangered grass- Resources 4: 359–361. feeding butterfly Coenonympha oedippus (Lepidoptera: Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Nymphalidae) in response to contrasting habitats. European Guisan A, Hijmans RJ, Huettmann F, Leathwick JR, Journal of Entomology 115: 339–353. Lehmann A. 2006. Novel methods improve prediction of spe- Keyghobadi N. 2007. The genetic implications of habitat cies’ distributions from occurrence data. Ecography 29: 129–151. fragmentation for . Canadian Journal of Zoology 85: Elith J, Kearney M, Phillips S. 2010. The art of modelling 1049–1064. range‐shifting species. Methods in Ecology and Evolution 1: Kodandaramaiah U, Wahlberg N. 2009. Phylogeny and 330–342. biogeography of Coenonympha butterflies (Nymphalidae: Elith J, Phillips SJ, Hastie T, Dudík M, Chee YE, Yates CJ. Satyrinae)–patterns of colonization in the Holarctic. 2011. A statistical explanation of MaxEnt for ecologists. Systematic Entomology 34: 315–323. Diversity and Distributions 17: 43–57. Kopelman NM, Mayzel J, Jakobsson M, Rosenberg NA, Evanno G, Regnaut S, Goudet J. 2005. Detecting the number Mayrose I. 2015. Clumpak: a program for identifying of clusters of individuals using the software STRUCTURE: a clustering modes and packaging population structure simulation study. Molecular Ecology 14: 2611–2620. inferences across K. Molecular Ecology Resources 15: Falush D, Stephens M, Pritchard JK. 2003. Inference 1179–1191. of population structure using multilocus genotype data: Koscinski D, Crawford LA, Keller HA, Keyghobadi N. linked loci and correlated allele frequencies. Genetics 164: 2011. Effects of different methods of non‐lethal tissue sam- 1567–1587. pling on butterflies. Ecological Entomology 36: 301–308.

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 112 L. DESPRÉS ET AL.

Kudrna O, Pennerstorfer J, Lux K. 2015. Distribution Peterson BK, Weber JN, Kay EH, Fisher HS, Hoekstra HE. atlas of European butterflies and skippers. Schwanfeld: 2012. Double digest RADseq: an inexpensive method for de Wissenschaftlicher Verlag Peks. novo SNP discovery and genotyping in model and non-model Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 Kulak A, Yakovlev R. 2018. Peculiarities of biology and species. PLoS One 7: e37135. the current state of the populations of the False Ringlet Ritter S, Michalski SG, Settele J, Wiemers M, Fric ZF, Coenonympha oedippus (Fabricius, 1787) on the territory of Sielezniew M, Šašić M, Rozier Y, Durka W. 2013. Belarus. Ukrainian Journal of Ecology 8: 342–349. Wolbachia infections mimic cryptic speciation in two para- Kumar S, Stecher G, Tamura K. 2016. MEGA7: Molecular sitic butterfly species, Phengaris teleius and P. nausithous Evolutionary Genetics Analysis version 7.0 for bigger data- (Lepidoptera: Lycaenidae). PLoS One 8: e78107. sets. Molecular Biology and Evolution 33: 1870–1874. Schmitt T, Robert S, Seitz A. 2005. Is the last glaciation Lavinia PD, Bustos EON, Kopuchian C, Lijtmaer DA, the only relevant event for the present genetic population García NC, Hebert PD, Tubaro PL. 2017. Barcoding the structure of the Meadow Brown butterfly Maniola jurtina butterflies of southern South America: Species delimitation (Lepidoptera: Nymphalidae)? Biological Journal of the efficacy, cryptic diversity and geographic patterns of diver- Linnean Society 85: 419–431. gence. PLoS One 12: e0186845. Schmitt T, Varga Z. 2012. Extra-Mediterranean refugia: The Leaché AD, Banbury BL, Felsenstein J, de Oca AN-M, rule and not the exception? Frontiers in Zoology 9: 22. Stamatakis A. 2015. Short tree, long tree, right tree, wrong Sielezniew M. 2012. Strzępotek edypus Coenonympha oedip- tree: new acquisition bias corrections for inferring SNP phy- pus. In: Makomaska-Juchiewicz M. Baran P, ed. Monitoring logenies. Systematic Biology 64: 1032–1047. gatunków zwierząt. Przewodnik metodyczny. Part II. Levers C, Butsic V, Verburg PH, Müller D, Kuemmerle T. Warszawa: GIOŚ, 258–273. 2016. Drivers of changes in agricultural intensity in Europe. Sielezniew M, Palka K, Michalczuk W, Bystrowski C, Land Use Policy 58: 380–393. Holowinski M, Czerwinski M. 2010. False Ringlet Lhonoré J, Lagarde M. 1999. Biogeography, ecol- Coenonympha oedippus (FABRICIUS, 1787)(Lepidoptera: ogy and conservation of Coenonympha oedippus (Fab., Nymphalidae) in Poland: state of knowledge and conserva- 1787) (Lepidoptera: Nymphalidae: Satyrinae). Annales de la tion prospects. Oedippus 26: 20–24. Société Entomologique de France 35: 299–307. Sielezniew M, Rutkowski R, Ponikwicka-Tyszko D, Louy D, Habel JC, Abadjiev S, Schmitt T. 2013. Genetic Ratkiewicz M, Dziekanska I, Švitra G. 2012. Differences legacy from past panmixia: high genetic variability and low in genetic variability between two ecotypes of the endangered differentiation in disjunct populations of the Eastern Large myrmecophilous butterfly Phengaris (= Maculinea) alcon– Heath butterfly. Biological Journal of the Linnean Society the setting of conservation priorities. Insect Conservation 110: 281–290. and Diversity 5: 223–236. Manel S, Williams HC, Ormerod SJ. 2001. Evaluating pres- Stamatakis A. 2014. RAxML version 8: a tool for phylo- ence–absence models in ecology: the need to account for genetic analysis and post-analysis of large phylogenies. prevalence. Journal of Applied Ecology 38: 921–931. Bioinformatics 30: 1312–1313. Nazareno AG, Bemmels JB, Dick CW, Lohmann LG. 2017. Staub R, Aistleitner U. 2006. Das Moor-Wiesenvögelschen Minimum sample sizes for population genomics: An empir- (Coenonympha oedippus)—oder worauf es im grenzüber- ical study from an Amazonian plant species. Molecular schreitenden Artenschutz ankommt. Alpenrheintal—eine Ecology Resources 17: 1136–1147. Region im Umbau. Analysen und Perspektiven der räumli- Nogués‐Bravo D. 2009. Predicting the past distribution of chen Entwicklung. Schaan, Verlag der Liechtensteinischen species climatic niches. Global Ecology and Biogeography Akademischen Gesellschaft, 245–254. 18: 521–531. Strandberg G, Brandefelt J, Kjellström E, Smith B. 2011. Orsini L, Corander J, Alasentie A, Hanski I. 2008. Genetic High‐resolution regional simulation of last glacial maximum spatial structure in a butterfly metapopulation correlates climate in Europe. Tellus A 63: 107–125. better with past than present demographic structure. Taberlet P, Fumagalli L, Wust-Saucy AG, Cosson JF. Molecular Ecology 17: 2629–2642. 1998. Comparative phylogeography and postglacial coloniza- Pastoralis G, Reiprich A. 1995. Zoznam Motylov vyskytu- tion routes in Europe. Molecular Ecology 7: 453–464. jucich sa na uzemi Slovenska. Komarno: Spišska Nova Ves. van Halder I, Barbaro L, Corcket E, Jactel H. 2008. Patricelli D, Sielezniew M, Ponikwicka-Tyszko D, Importance of semi-natural habitats for the conservation Ratkiewicz M, Bonelli S, Barbero F, Witek M, Buś MM, of butterfly communities in landscapes dominated by pine Rutkowski R, Balletto E. 2013. Contrasting genetic struc- plantations. Biodiversity and Conservation 17: 1149–1169. ture of rear edge and continuous range populations of a Van Swaay C, Cuttelod A, Collins S, Maes D, López parasitic butterfly infected by Wolbachia. BMC Evolutionary Munguira M, Šašić M, Settele J, Verovnik R, Verstrael T, Biology 13: 14. Warren M, Wiemers M, Wynhoff I. 2010. European Red Pearson RG, Raxworthy CJ, Nakamura M, Townsend List of Butterflies. Luxembourg: Publications Office of the Peterson A. 2007. Predicting species distributions from European Union. small numbers of occurrence records: a test case using cryptic Van Swaay C, Warren M. 1999. Red data book of European geckos in Madagascar. Journal of Biogeography 34: 102–117. butterflies (Rhopalocera). Strasbourg: Council of Europe.

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113 FALSE RINGLET POPULATION GENOMICS IN EUROPE 113

Varin G. 1964. Contribution à l’étude des Satyridae Warren DL, Seifert SN. 2011. Ecological niche modeling in (Lepidoptères) Coenonympha oedippus Fabricius. Sous- Maxent: the importance of model complexity and the per- espèce rhodanica Varin nova en Savoie. Bulletin mensuel de formance of model selection criteria. Ecological Applications Downloaded from https://academic.oup.com/biolinnean/article-abstract/126/1/95/5214084 by Divisione Coordinamento Biblioteche Milano user on 30 January 2019 la Société Linéenne de Lyon 2: 92–93. 21: 335–342. Verovnik R, Rebeušek F, Jež M. 2012. Atlas of butter- Weir BS, Cockerham CC. 1984. Estimating F‐statistics flies (Lepidoptera: Rhopalocera) of Slovenia. Miklavž na for the analysis of population structure. Evolution 38: Dravskem polju: Center za kartografijo favne in flore. 1358–1370. Wahlberg N, Freitas AV. 2007. Colonization of and radiation Willing E-M, Dreyer C, Van Oosterhout C. 2012. Estimates

in South America by butterflies in the subtribe Phyciodina of genetic differentiation measured by FST do not necessarily (Lepidoptera: Nymphalidae). Molecular Phylogenetics and require large sample sizes when using many SNP markers. Evolution 44: 1257–1272. PLoS One 7: e42649.

SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article at the publisher’s web-site. Figure S1. Variation of the proportion of polymorphic loci de novo reconstructed when increasing the number of mismatch M allowed between two reads to be merged. The threshold M = 7 within an individual, and 9 between individuals, was chosen. Figure S2. Results of Bayesian genetic clustering (STRUCTURE) based on 1314 unlinked SNPs, with number of clusters ranging from K = 3 to K = 8. For each K the mean log likelihood is indicated (ten replicates), and Evanno’s ΔK is shown. Individuals (N = 96) are represented by vertical bars and grouped by locality (N = 32).

Figure S3. Relationship between genetic [FST/(1 − FST)] and geographical distance (square root transformed) for the main groups of populations obtained with the STRUCTURE analysis, except for the three localities of the Northern Alps region (sample size too small for regression analysis). Figure S4. Maximum likelihood tree based on 1594 concatenated 100-bp ddRADseq fragments. Bootstrap values are shown next to the branches (100 replicates). The tree is drawn to scale, with branch lengths representing the number of substitutions per site, and is rooted at midpoint as the outgroup is unknown. The analysis involved 96 individuals and six replicates. Of the five geographical regions, only two (Atlantic and Western Alps) form well- supported clades (BS > 99%), and are highlighted in colour. Figure S5. Maximum likelihood tree under the General Time Reversible model allowing for invariable sites [(GTR+I), 62% sites] based on mitochondrial sequences (partial CO1). Bootstrap support is indicated on each node (500 replicates). Branch lengths at scale represent the number of substitutions per site. The analysis involved 51 sequences, including 43 sequences from C. oedippus and eight outgroups. Figure S6. Alternative demographic histories (A) for the three main genetic lineages of C. oedippus in Europe (Atlantic, Western Alps and Eastern) and (B) for the French populations; posterior probabilities (and 95% CI) for each scenario are indicated, with the posterior predictive error of the retained scenario (the probability of assign- ing scenario 2 when scenario 1 was simulated). The 1000 (10%) simulations closest to the observed summary statistics for each scenario (red or green dots) were plotted in a principal components analysis together with the observed summary statistics (yellow dot) to check (1) that the two alternative hypotheses can be distinguished and (2) that the retained scenario adequately represents the observed dataset (Cornuet et al., 2014). A, Atlantic; EE, Eastern Europe; NA,: Northern Alps; SA, Southern Alps; WA, Western Alps. Table S1. Results of the ddRADseq experiment on 104 samples (98 individuals, three intra-library replicates and three inter-libraries replicates); two samples with fewer than 100 000 reads were excluded from analysis (in italic). After quality filtering, the mean number of loci per individual was 7552 and the mean coverage per locus was 60. Table S2. Presence records of C. oedippus in Europe used for MaxEnt inferences.

Table S3. Pairwise genetic distances (Weir–Cockerham FST, below diagonal) and geographical distances (in km, above diagonal) between sampled localities. Distances between localities from the same geographical region are indicated in italic.

SHARED DATA The demultiplexed ddRAD sequencing data are available from the European Nucleotide Archive database (project PRJEB29148), and the mitochondrial CO1 sequences in BOLDsystems (project OEDIP).

© 2018 The Linnean Society of London, Biological Journal of the Linnean Society, 2019, 126, 95–113