Karyosystematics and molecular taxonomy of the anomalous blue butterflies (Lepidoptera, Lycaenidae) from the Balkan Peninsula

Abstract The Balkan Peninsula represents one of the hottest biodiversity spots in Europe. However, the invertebrate fauna of this region is still insufficiently investigated, even in respect of such well-studied organisms as Lepidoptera. Here we use a combination of chromosomal, molecular and morphological markers to rearrange the group of so-called anomalous blue butterflies (also known as ‘brown complex’ of the subgenus Agrodiaetus Hübner, [1822] and as the Polyommatus (Agrodiaetus) admetus (Esper, 1783) species group) and to reveal its cryptic taxonomic structure. We demonstrate that Polyommatus aroaniensis (Brown, 1976) is not as widespread in the Balkans as was previously thought. In fact, it has a dot-like distribution range restricted to the Peloponnese Peninsula in South Greece. Polyommatus orphicus Kolev, 2005 is not as closely related to the Turkish species Polyommatus dantchenkoi (Lukhtanov & Wiemers, 2003) as was supposed earlier. Instead, it is a Balkan endemic represented by two subspecies: Polyommatus orphicus orphicus (Bulgaria) and Polyommatus orphicus eleniae Coutsis & De Prins, 2005 (Northern Greece). Polyommatus ripartii (Freyer, 1830) is represented in the Balkans by an endemic subspecies Polyommatus ripartii pelopi. The traditionally recognized Polyommatus admetus (Esper, 1783) is shown to be a heterogeneous complex and is divided into Polyommatus admetus sensu stricto (the Balkans and west Turkey) and Polyommatus yeranyani (Dantchenko & Lukhtanov, 2005) (east Turkey, Armenia, Azerbaijan and Iran). Polyommatus nephohiptamenos (Brown & Coutsis, 1978) is confirmed to be a species with a dot-like distribution range in Northern Greece. Finally, from Central Greece (Timfristos and Parnassos mountains) we describe Polyommatus timfristos Lukhtanov, Vishnevskaya & Shapoval, sp. n. which differs by its haploid chromosome number (n=38) from the closely related and morphologically similar Polyommatus aroaniensis (n=47-48) and Polyommatus orphicus (n=41-42). We provide chromosomal evidence for three separate south Balkan Pleistocene refugia (Peloponnesse, Central Greece and Northern Greece/South Bulgaria) and stress the biogeographic importance of Central Greece as a place of diversification. Then we argue that the data obtained have direct implications for butterfly karyology, taxonomy, biogeography and conservation.


Introduction
The Balkan Peninsula is recognized as a European biodiversity hotspot, with high endemism found in animals and plants (Nicolić et al. 2014, Buj et al. 2015, Bregović and Zagmajster 2016. However, the invertebrate fauna of this region is still insufficiently investigated (Previšić et al. 2016), even in respect of such a well-studied group as Lepidoptera (butterflies and moths) (Sobezyk and Gligorović 2016).
Within Balkan Lepidoptera, the Agrodiaetus Hübner, [1822] blue butterflies are the most complicated group from the taxonomical point of view. The subgenus Agrodiaetus is a distinct monophyletic lineage within the species-rich genus Polyommatus Latreille, 1804 (Talavera et al. 2013a). Adult Agrodiaetus butterflies are small in size with wing span from 1.9 to 3.6 cm. Females are mostly warm brown on the upperside of the wings, whereas males can be either blue or brown. In the latter case, they resemble females. Thus, a species can be classified as either dimorphic or monomorphic depending on the wing color of the males. Most of the Agrodiaetus species have a white streak on the underside of hind wings, and this feature appears to be an apomorphic character of the subgenus Agrodiaetus. However, in a few species and populations this white streak is secondarily reduced or totally absent (Eckweiler and Bozano 2016).
The subgenus Agrodiaetus includes numerous species, subspecies and forms with uncertain taxonomic positions . It was estimated to have originated only about 3 million years ago (Kandul et al. 2004) and radiated rapidly in the Western Palaearctic (Kandul et al. 2007). The last published review of the subgenus includes 120 valid species (Eckweiler and Bozano 2016). Many of them have extremely local 'dot-like' distributions that are restricted to particular mountain valleys in the Balkan Peninsula, Asia Minor, Transcaucasus, Iran and Central Asia (Vila et al. 2010, Eckweiler andBozano 2016).
In most cases, species identification in Agrodiaetus is extremely difficult. The morphology of male genitalia is uniform throughout most of the species and, with a few exceptions (see Coutsis 1985Coutsis , 1986, at most it can help to separate groups of species, e.g. the Polyommatus dolus (Hübner, 1823) and P. admetus (Esper, 1783) species groups (see Kolev 2005), but not individual species. The differences in wing pattern and coloration between many Agrodiaetus species are very subtle or nearly lacking (Eckweiler and Bozano 2016).
Despite morphological similarity, the taxonomic and identification problems within the subgenus Agrodiaetus can be solved if chromosomal (de Lesse 1960a,b, Lukhtanov 1989 or molecular markers (Wiemers 2003, Kandul et al. 2004, Stradomsky and Fomina 2013, or their combination (Lukhtanov et al. 2006, Vila et al. 2010, Lukhtanov and Tikhonov 2015, Shapoval and Lukhtanov 2015a are applied. Although chromosome numbers are invariable in many groups of Lepidoptera (Robinson 1971, Lukhtanov 2014, Hernández-Roldán 2016, a few genera demonstrate chromosomal instability, a situation in which multiple closely related species differ drastically from each other by major chromosomal rearrangements, sometimes resulting in high variability in chromosome number (de Lesse 1960a,b, Talavera et al. 2013b). An unusual diversity of karyotypes is the most remarkable characteristic of the subgenus Agrodiaetus. Species of Agrodiaetus exhibit one of the highest ranges in chromosome numbers in the animal kingdom (Lukhtanov 2015). Haploid chromosome numbers in Agrodiaetus range from n=10 in A. caeruleus (Staudinger, 1871) to n=134 in A. shahrami (Skala, 2001) Dantchenko 2002a, Lukhtanov et al. 2005). Additionally, this subgenus demonstrates a high level of karyotypic differentiation with respect to chromosome size (Lukhtanov and Dantchenko 2002b) and variation in number of chromosomes bearing ribosomal DNA clusters (Vershinina et al. 2015). The karyotype is generally stable within species although differences between closely related taxa are often high and provide reliable characters for species delimitation, description and identification (de Lesse 1960a,b, Lukhtanov andDantchenko 2002a,b).
Molecular studies revealed that subgenus Agrodiaetus consists of 10 monophyletic clades: the P. transcaspicus (Heyne, 1895) group, the P. iphigenides (Staudinger, 1886) group, the P. ershoffii (Lederer, 1869) group, the P. poseidon (Herrich-Schäffer, 1844) group, the P. admetus group, the P. damone (Eversmann, 1841) group, the P. carmon (Herrich-Schäffer, 1851) group, the P. damon (Denis & Schiffermüller, 1775) group, the P. dolus group and the P. actis (Herrich-Schäffer, 1851) group (Kandul et al. 2002, Wiemers 2003. They also demonstrated that many species are clearly differentiated with respect to mitochondrial and nuclear DNA sequences. However, this is not a general rule, as the standard mitochondrial DNA barcodes are often identical or nearly identical between closely related taxa and even between morphologically distinct species (Kandul et al. 2004, Wiemers and Fiedler 2007. Generally, chromosomal characters in Agrodiaetus evolve more quickly than standard DNA barcodes, and because they are usually present as fixed differences, provide better markers for recently evolved taxa than nucleotide substitutions . Species delimitation is especially difficult within a group of so-called anomalous blue species (known also as 'brown complex' of the subgenus Agrodiaetus and as the Polyommatus admetus species complex). This group is composed of multiple species in which both male and female butterflies have similar brown coloration on the upperside of the wings (Lukhtanov et al. 2003).
The group of anomalous blue species includes taxa belonging to two clearly monophyletic and most probably sister clades: the P. admetus clade (comprises only monomorphic species -P. admetus, P. demavendi, P. khorasanensis. P. nephohiptamenos, P. ripartii, P. pseudorjabovi) and the P. dolus clade (comprises both monomorphic -P. alcestis, P. karacetinae, P. eriwanensis, P. interjectus, P. dantchenkoi, P. humedasae, P. aroaniensis, P. orphicus, P. timfristos sp. n., P. fabressei, P. violetae, P. valiabadi, P. rjabovianus; and dimorphic species -P. dolus, P. fulgens, P. menalcas). The anomalous blue butterflies represent a real stumbling block in the Agrodiaetus taxonomy (Lukhtanov et al. 2003). According to Eckweiler and Bozano (2016), the group is distributed in West Palearctic from Spain in the west to Mongolia in the east. The majority of the species have very localized distribution areas concentrated in (1) the Iberian Peninsula, (2) the Balkan Peninsula and (3) west Asia (mostly in the Middle East and Caucasus). Vila et al. (2010) studied in detail the European Agrodiaetus taxa distributed west of the 17th meridian, using a combination of molecular and chromosomal markers (Vila et al. 2010). Chromosomal and molecular markers were also applied to study the taxonomy of the Asian taxa . It is paradoxical that systematic studies based on combined analysis of molecular and chromosomal markers have never been applied to Balkan taxa of the P. admetus species complex. However, some DNA data can be found in GenBank (Wiemers 2003, Wiemers et al. 2007, 2009, Lukhtanov et al. 2009, Vila et al. 2010, Dincă et al. 2013) and chromosome numbers are known for a few Balkan populations (Coutsis and De Prins 2005, Kolev 2005.
The goal of the present study is a simultaneous investigation of chromosomal, molecular and morphological diversity in the anomalous blue butterflies from the Balkan Peninsula and interpretation of this diversity in terms of taxonomy. To achieve this goal, the following tasks were set: To collect specimens of all the taxa of the complex described from the territory of the Balkan Peninsula. To collect specimens from different populations of these taxa.
To study their karyotypes (chromosome number and structure) using standard protocols for staining.
To obtain data on the variability of molecular markers: mitochondrial DNA barcode (COI gene fragment) and nuclear internal transcribed spacer 2 (ITS2). These markers were selected because the usefulness of mitochondrial COI barcodes in taxonomic studies on species-level is generally recognized (Hebert et al. 2004, but see Wiemers and Fiedler 2007), and despite some limitations (Shapoval and Lukhtanov 2015c), internal transcribed spacer 2 was found to be a useful nuclear marker in butterfly taxonomy (Wiemers et al. 2009).
To study the variability of the wing pattern characters which can be potentially useful for delimitation of species and populations (presence/reduction/absence of the white streak on the underside of the hindwings, the development of the marginal marking on the underside of the wings, presence or absence of a white stroke on the underside of the forewings).
To interpret the discovered chromosomal, molecular and morphological diversity in terms of taxonomy using two original methodologies: (1) detecting and taxonomic interpretation of cryptic entities found in sympatry and allopatry using combined analysis of mitochondrial and chromosomal markers , and (2) critical evaluation of pre-existing morphology-based taxonomic hypotheses using DNA barcodes .

Taxon sampling
Butterflies for this study were collected in 2008 in the Balkan Peninsula by V.A. Lukhtanov, N.A. Shapoval and L. Rieppel, in 2016 in Hvoyna village (Bulgaria) by E.A. Pazhenkova and in the Tigirekskiy Reservation (the Altai Mountains, Russia) by M.S.Vishnevskaya in 2007 (Fig. 1, Table 1). We paid special attention to collecting the taxa in their type localities: mount Chelmos (Greece: Peloponnese) (type locality of Agrodiaetus alcestis aroaniensis Brown, 1976), mount Falakró near Granítis (Greece, Makedonía, Dráma district) (type locality of Agrodiaetus eleniae Coutsis & De Prins, 2005) and Hvoyna (south Bulgaria, the Rhodopi mts) (type locality of Polyommatus dantchenkoi orphicus). Unfortunately, in our research we did not have an opportunity to study the holotypes of these taxa. Taking into account a possibility of multiple cryptic species within a local area even in well-studied European butterflies (Dincă et al. 2011(Dincă et al. , 2013b, in each place we managed to collect (and then to study) as many individuals as possible paying special attention to the specimens with unusual or intermediate morphology.
Before processing butterflies were put in glassine envelopes and kept alive for less than one hour. Testes were removed and put into a vial with a fresh fixative (3:1, 96% Figure 1. Localities of the species collected for the study (the species list is presented in Table 1 ethanol: glacial acetic acid). The wings were removed and put into a glassine envelope, and the body was placed into a vial with 96% ethanol for further molecular analysis. All chromosome preparations, butterfly bodies in ethanol and wings in glassine envelopes are stored in the Department of Karyosystematics (Zoological Institute of the Russian Academy of Sciences, St. Petersburg).

Analysis of karyotype
Testes were stored in the 3:1 fixative for several months at +4 °C and then stained with 2% acetic orcein for 30 days at 20 °C. We used a two-phase method of chromosome analysis following Lukhtanov and Dantchenko (2002b). In the first phase, stained testes were placed into a drop of 40% lactic acid on a slide where spermatocysts were dissected from testis membranes using entomological pins. Intact spermatocytes were transferred into a new drop of 40% lactic acid and covered with a coverslip. A Carl Zeiss Amplival light microscope was used for cytogenetic analysis. During the metaphase I stage, each spermatocyst was observed as a regular sphere consisting of 64 spermatocytes. In the second phase, different degrees of chromosome spreading were observed by gradually increasing pressure on the coverslip. The second phase was useful for studying the bivalent structure and counting the bivalent number. By scaling up the pressure on the coverslip, we were able to manipulate chromosomes, e.g. change their position and orientation on the slide, and consequently to resolve controversial cases of contacting or overlapping bivalents. Haploid chromosome numbers were counted in metaphase I (MI) and/or metaphase II (MII) of meiosis.

DNA extraction and sequencing
We used a 657-bp fragment within the mitochondrial COI gene and a 440-bp fragment within the ITS2 region. DNA was extracted using phenol-chloroform method according to the standard protocol (Sambrook and Russel 2006). The first two abdominal segments were homogenized in lysis buffer [25 mM EDTA, 75 mM NaCl, 10 mM Tris (pH 7.5)]. Then proteinase K (20 mg/ml) and 10% SDS were added and the samples were incubated for 2 h at 60 °C. DNA was extracted from lysate first with phenol/chloroform (1:1) and then with chloroform to remove any remaining phenol. DNA was precipitated with isopropyl alcohol in the presence of 0.1 M NaCl and pelleted by centrifugation. The pellets were washed with 70% ethanol, dried and dissolved in ddH2O. The extracted DNA was stored at -20 °C.
After amplification, PCR mix was loaded in 1% agarose gel and specific product was separated by gel electrophoresis (Fig. 2). Pieces of gel containing the DNA fragment of required length were cut out and then double-stranded DNA was purified using the method of 'DNA purification from agarose gels with MP@SiO 2 magnetic particles' according to the manufacturer's protocol (Sileks). Purified DNA fragments were extracted with ddH 2 O from magnetic particles pelleted with a magnetic rack and collected in a fresh tube. The concentration of purified DNA was estimated via gel electrophoresis (by comparing the brightness of the sample fragment to the brightness of the DNA marker (in our case 100 bp DNA Ladder, Thermo Fisher Scientific).
All the preparations for sequencing were held in "The Laboratory of Animal Genetics" of Saint-Petersburg State University and "Chromas" Core Facility, Saint-Petersburg State University Research Park. Sequencing was carried out in the Research Resource Center for Molecular and Cell Technologies. GenBank codes of the studied samples are provided in Tables 1 and 2.

Phylogenetic analysis
The analysis involved 221 COI sequences (169 GenBank sequences and 52 own material) and 117 ITS2 sequences (66 GenBank and 51 own data).
Sequences of different length (from 415 to 657 bp in case of COI and from 415 to 440 bp in case of ITS2) were included into the final dataset alignment. We used BioEdit 7.2.5 software (Hall 1999) to align the sequences and then edited them manually. The final COI alignment included 657 sites, with 137 variable sites and 112 parsimony-informative sites. The final ITS2 alignment included 440 sites, with 52 variable sites and 22 parsimony-informative sites.
Previously, no significant conflict was detected between the mitochondrial COI and nuclear ITS2 Agrodiaetus data sets (Vila et al. 2010). Thus, we combined mitochondrial and nuclear sequences to improve phylogenetic signal. This resulted in a concatenated alignment with a total of 1039 bp.
Phylogenetic relationships were inferred using Bayesian Inference (BI), maximum likelihood (ML) and maximum parsimony (MP) analyses. jModelTest was used to determine optimal substitution models for ML inference (Posada 2008).
Bayesian analyses were conducted using MrBayes, version 3.2 (Ronquist et al. 2012). Datasets were partitioned by codon position. Substitution models used for each partition were chosen according to jModelTest (Posada 2008): nst=2 and rates=invgamma for the first position, nst=2 and rates=gamma for the second position, and nst=6 and rates=gamma for the third position of COI barcodes. Substitution model nst=6 and rates=invgamm was chosen for ITS2. In evolution of ITS2 sequences, the mono, bi-and mullti-nucleotide insertions/deletions are frequent and contain phylogenetically important information. To account for this, each indel event was coded as a binary character (1/0, presence/absence of the gap independently of its length) and  then used in the Bayesian analyses of ITS2 and concatenated data sets. Two runs of 10 000 000 generations with four chains (one cold and three heated) were performed. Chains were sampled every 10 000 generations, and burn-in was determined based on inspection of log likelihood over time plots using TRACER, version 1.4 (available from http://beast.bio.ed.ac.uk/Tracer). The ML trees were inferred using MEGA6 under the GTR+G+I model. MP analysis was performed using a heuristic search as implemented in MEGA6 (Tamura et al. 2013). A heuristic search was carried out using the close-neighbor-interchange algorithm with search level 3 (Nei and Kumar 2000) in which the initial trees were obtained with the random addition of sequences (100 replicates). We used nonparametric bootstrap values (Felsenstein 1985) to estimate branch support for ML and MP trees. The bootstrap consensus tree was inferred from 500 replicates.

Haplotype network
Median network was constructed using the program Network 4.6.1.3. (Fluxus Technology, fluxus-engineering.com), with the Median Joining algorithm (Bandelt 1999). The algorithm picks close haplotype groups and finds hypothetical ancestors, to join the haplotypes in a common parsimony network. The program shows each haplotype with a colored circle. When the haplotypes are identical, they are united in one bigger circle under one name. Similar haplotypes then are combined in haplogroups ( Table 2). The network was constructed on the base of COI alignment, with 191 sequences. The length of the sequences was 612 bp with 116 parsimony-informative sites. The final alignment included only sequences of equal length. Short and ambiguous sequences were excluded. Karyotypes of the studied samples Table 3 Polyommatus admetus Fig. 3a-c The haploid chromosome number n=80 was found in MI and MII cells of two studied individuals from South and Central Greece. In two specimens (Greece, Smolikas Mt and Bulgaria) we counted approximately n=ca80 at MI. The last count was performed with an approximation due to the overlapping of some bivalents. The karyotype displayed one larger bivalent in the centre of the MI plate and one larger univalent in the centre of the MII plate.

Polyommatus ripartii pelopi
The haploid chromosome number was determined to be n=90 in MI and MII cells of seven studied individuals from different localities (Greece, Bulgaria). At MI, two bivalents were especially large and were situated in the centre of the metaphase plates. Bivalent 1 was 1.4-1.6 times larger than bivalent 2. The sizes of the remaining 88 bivalents decreased more or less linearly. At MII, two univalents were especially large and were situated in the centre of the metaphase plates. Chromosome 1 was 1.4-1.6 times larger than chromosome 2. The sizes of the remaining 88 chromosomes decreased more or less linearly. In three specimens we counted approximately n=ca 90 at MI. The last count was an approximation due to the overlapping of some bivalents. In three specimens, the diploid chromosome number was estimated as 2n=ca180 in male asynaptic meiosis. Fig. 3h The haploid chromosome number was determined to be n=90 in MI and MII cells of two studied individuals. At MI, two bivalents (one big and one medium-sized) were larger than the others. At MII, two univalents (one big and one mediumsized) were larger than the rest. The sizes of the remaining 88 bivalents and univalents decreased more or less linearly. In four specimens we counted approximately n=ca90 at MI. The last count was an approximation due to the overlapping of some bivalents. . Polyommatus (Agrodiaetus) karyotypes. Bar =10 µ. a-b P. admetus, sample LR08D109, Greece, MI, n=80. One large bivalent in the centre of the plate can be seen c P. admetus, sample LR08D109, Greece, MII, n=80. One large chromosome in the centre of the plate can be seen d P. ripartii pelopi, sample LR08D249, Greece, MI, n=90. Two large bivalents in the centre of the plate can be seen e P. ripartii pelopi, sample LR08D144, Greece, MI, n=90. Two large bivalents in the centre of the plate can be seen f P. ripartii pelopi, sample LR08D145, Greece, MI, n=90. Two large bivalents in the centre of the plate can be seen g P. ripartii pelopi, sample LR08D92, Greece, MII, n=90. Two large chromosomes in the centre of the plate can be seen h P. nephohiptamenos, sample LR08D494, Northern Greece, MI, n=90. All the bivalents are situated in a plane with the largest elements in the centre of the circular metaphase plate. Bivalents are clearly separated from each other by gaps. Two bivalents are larger than the rest. i P. aroaniensis, sample LR08D102, Greece, MI, n=47. Fig. 3i In the single studied specimen collected in the type locality (Greece, Mt. Chelmos) haploid chromosome number n=47 was found in MI cells. Bivalents were fairly well The haploid chromosome number was determined to be n=38 in prometaphase, MI and MII cells of the holotype and six studied paratypes. Bivalents at MI and prometaphase and univalents at MII were fairly well differentiated with respect to their size; however, it was difficult to subdivide them objectively into size groups because the sizes of the 47 elements decrease more or less linearly.

Polyommatus orphicus orphicus
Two different haploid chromosome numbers (n=41 and n=42) were observed in MI and MII cells of the four specimens studied. This variation was most likely caused by polymorphism for one chromosome fussion/fission. This polymorphism resulted in three types of MI karyotype: n=41 (homozygous for chromosomal fusion/fission, one pair of fused chromosomes), n=42 (homozygous for chromosomal fusion/fission, two pairs of unfused chromosomes) and n=41 (heterozygous for chromosomal fusion/fission, 40 bivalents and one trivalent). Bivalents at MI and univalents at MII were fairly well differentiated with respect to their size; however, it was difficult to subdivide them objectively into size groups because the sizes of the elements decrease more or less linearly.

Polyommatus orphicus eleniae
Chromosome numbers (n=41 and n=42) were observed in MI and MII cells of the four specimens studied. This variation was most likely caused by polymorphism for one chromosome fussion/fission. This polymorphism resulted in three types of MI karyotype: n=41 (homozygous for chromosomal fusion/fission, one pair of fused chromosomes), n=42 (homozygous for chromosomal fusion/fission, two pairs of unfused chromosomes) and n=41 (heterozygous for chromosomal fusion/fission, 40 bivalents and one trivalent). Bivalents and univalents were fairly well differentiated with respect to their size; however, it was difficult to subdivide them objectively into size groups because the sizes of the elements decrease more or less linearly.

Phylogenetic reconstruction
Bayesian analysis of the 657-bp region of COI gene resulted in a phylogram, showing a high level of posterior probability for the majority of the revealed clades. Analysis  of the 221-specimen dataset recovered the P. admetus and P. dolus species groups as distinct monophyletic lineages. This is consistent with the previous conclusions (Wiemers 2003, Kandul et al. 2004, 2015, Vila et al. 2010, Dincă et al. 2013a). The tree divided into two parts (P. admetus and P. dolus groups) is shown in Figures 6-8.  Within the P. admetus group, the species P. ripartii appeared as a polyphyletic assemblage consisting of four monophyletic lineages: the "Balkan" clade, including specimens from Greece and Bulgaria, "West-European" clade, including butterflies from France, Italy and Spain, "mixed" (or Eurasian) clade, including butterflies distributed from Spain to Mongolia, and Turkish-Transcaucasian clade, including butterflies from Turkey and Armenia. The last clade formed an independent lineage, sister to the species P. demavendi (Pfeiffer, 1938) from east Turkey, Transcaususus and Iran.
P. admetus sensu auctorum formed two independent clades: one consisting of European and west Turkish specimens and another consisting of specimens from east Turkey, Armenia and Azerbaijan. Polyommatus nephohiptamenos appeared on the Bayesian tree as a paraphyletic group consisting of nine weakly differentiated individuals. On the Figure 6. Fragment of the Bayesian tree of P. admetus and P. dolus complexes based on analysis of COI barcodes and focused on P. nephohiptamenos, P. admetus and P. ripartii pelopi. Polyommatus pseudorjabovi clade is not shown in details, for its composition see Lukhtanov et al. (2015a). The West-European and the "mixed" (Eurasian) clades of P. ripartii are shown in Fig. 7. Polyommatus dolus group is shown in Fig. 8. Numbers at nodes indicate Bayesian posterior probability. MP and ML trees (Figs 18 and 21 in Appendix 2), P. nephohiptamenos tended to form a monophyletic clade, but the bootstrap support of this clade was very low.
The P. dolus group is interesting for its Balkan species position. P. aroaniensis formed an independent clade separate from P. timfristos sp. n., which formed a monophyletic clade as well. Specimens of P. orphicus orphicus and P. orphicus eleniae were closely related and formed together a paraphyletic cluster.
Because of low variability, it was difficult to use ITS2 as a single marker to construct the phylogeny of Agrodiaetus. Therefore, we decided to combine the sequence   data on COI and ITS2 and constructed a tree on the base of these two markers (Fig. 9). We used 75 specimens for which we had data on both markers. Total length of the combined sequence was 1039 bp. The Bayesian tree constructed on the base of the concatenated alignment revealed generally the same topology as in the case of COI tree, however with a higher support for few clades, and P. orphicus orphicus + P. orphicus elenia formed a monophyletic clade with a posterior probability value 77.

Haplotype network analysis
The complicated relationships between species of P. admetus and P. dolus groups were also reflected by a haplotype network (Figs 10 and 11) constructed on the base of COI. To construct the network we used 191 specimens that were collapsed in 96 haplotypes representing 26 haplogroups (Table 2): 10 haplogroups for P. admetus group and 16 haplogroups for P. dolus group.
Polyommatus ripartii was represented by 82 specimens divided in 38 haplotypes and four haplogroups which corresponded completely with the four clades revealed on the Bayesian tree (Fig. 10). Polyommatus admetus sensu auctorum was found to include two haplogroups. One haplogroup was represented by specimens from the Balkan and west Turkey (P. admetus admetus), and the other haplogroup was represented by specimens from Armenia and Azerbaijan (P. admetus yeranyani + P. admetus malievi). These two haplogoups were clearly distinct from one another as can be seen in the number of Figure 11. Haplotype network of P. dolus species group. Colored circles represent different taxa. Each line segment represents a mutation step, and white small circles represent "missing" haplotypes. nucleotide substitutions between them. Polyommatus nephohiptamenos was represented by a distinct haplogroup most close to P. ripartii ripartii haplogroup.
As a by-product of our study, we also discovered that within our samples P. demavendi comprised two haplogroups. One haplogroup was represented by specimens of P. demavendi belovi, whilst the other was represented by P. demavendi lorestanus. Polyommatus pseudorjabovi was represented by a single differentiated haplogroup. A distinct haplogroup represented by a single haplotype was found within P. khorasanensis.
Concerning P. dolus group (Fig. 11) we would like to mention that all recognized species, except for P. fulgens and P. fabressei, were represented by clearly distinct COI haplogroups. Polyommatus fulgens and P. fabressei were closely related and even shared one haplotype, despite clear differences in butterfly wing color and karyotypes.
Despite presumed conspecifity (Kolev 2005), Polyommatus orphicus and P. dantchenkoi were found to be in the opposite parts of the recovered net, being separated by a number of other species (P. alcestis, P. violetae, P. aroaniensis, P. timfristos). The chromosomally distinct taxa P. alcestis and P. karacetinae were found to be also distinct with respect to their COI haplotypes. These two taxa were already treated as different species by Wiemers et al. (2009).

Butterfly morphology
One of the main characteristic features of the anomalous blue butterflies is the upperside wing color. All males and females have brown upper side of the wings, and therefore the group is also called "brown" complex. As for the underside (Fig. 12), there are some differentiated characters of the wing pattern that allow the defining of seven morphological types.

Species level monophyly, paraphyly and polyphyly
The studied taxa were found to demonstrate a relatively low level of COI and ITS2 differentiation in terms of genetic distances between species and numbers of evolutionary steps between the taxa on haplotype network (Figs 10 and 11). This result is not unexpected in light of our previous knowledge of this group (Wiemers and Fiedler 2007). The low genetic differentiantion results in relatively low support for some recovered clades (e.g. for P. timfristos, Figs 8 and 9) and in non-monophyly of some taxa (P. nephohiptamenos, P. orphicus) with respect to COI gene or to combination of COI and ITS2. Species-level non-monophyly in DNA barcode gene trees can have multiple explanations (Mutanen et al. 2016). In our case, combination of low interspecific differentiation with low level of intraspecific variation indicates that preservation of ancestral polymorphism and incomplete lineage sorting (rather than interspecific hybridization) is the most likely mechanism explaining the pattern observed. This finding is also in agreement with the previous conclusion that the subgenus Agrodiaetus itself and its species represent young evolutionary entities (Kandul et al. 2004). We should also stress that despite the obvious paraphyly, the taxa P. nephohiptamenos and P. orphicus are distinct with respect to the COI barcodes, and this can be seen on both Bayesian tree (Figs 6-8) and haplotype network (Figs 10 and 11).
An entirely different situation was found in P. ripartii and P. admetus sensu auctorum. In these taxa polyphyly in COI trees arises as a result of deep intraspecific divergence. There are two theoretically possible explanations for this kind of non-monophyly. First, each taxon can be a mix of unrecognized multiple species (Dincă et al. 2011(Dincă et al. , 2013b. Second, a profound irregularity in barcodes can be caused by reasons other than speciation resulting in extraordinary intra-specific barcode variability (Pazhenkova and Lukhtanov 2016). Among these reasons, interspecific mitochondrial introgression ) and blending of deeply diverged mitochondrial lineages which evolved in allopatry in different Pleistocene refugia (Pazhenkova and Lukhtanov 2016) are most likely ones. The first explanation could be applied to P. admetus sensu auctorum which most probably comprises two allopatric species, P. admetus sensu stricto and P. yeranyani (see the section Taxonomy below). The situation with P. ripartii sensu lato seems to be much more complicated. A combination of the first and the second explanations could be applied to P. ripartii sensu lato, and West-European and Eurasian clades could represent sympatric (parapatric?) intraspecific lineages (Dinca et al. 2013) whereas Turkish-Transcaucasian clade could represent an allopatric species. Additioanl studies are required to solve this problem.

P. admetus
The chromosome number of P. admetus was first established by H. de Lesse who discovered n=80 in populations from Bulgaria (Kalotina) and W Turkey, and n=78-80 (with predominance of n=79) in populations from the eastern part of Turkey (de Lesse 1960a,b). The last count (n=78-80 with predominance of n=79) was later confirmed for populations from Armenia (Lukhtanov and Dantchenko 2002a), Turkey and Azerbaijan (Dantchenko andLukhtanov 2005, Lukhtanov et al. 2015a). Here we confirm the haploid chromosome number n=80 for Dragoman near Kalotina (Bulgaria) and demonstrate that this karyotype occurs in other localities in Greece. The karyotype of the European samples (with predominance of n=80) seems to be similar, but not completely identical to the karyotype of samples from east Turkey, Armenia and Azerbaijan (with predominance of n=79).

P. ripartii
This transpalearctic species has been known to have a stable karyotype (n=90, including one large, one medium and 88 small elements) throughout its whole distribution range from Spain in the west to the Altai in the east (de Lesse 1960a,b, Kandul 1997, Lukhtanov and Datchenko 2002, Vila et al. 2010, Vershinina and Lukhtanov 2010, Przybyłowicz et al. 2014). The number n=90 was also found in P. ripartii pelopi (Coutsis et al. 1999), and we confirmed this count for samples from South and Central Greece and from Bulgaria.

P. nephohiptamenos
The haploid chromosome number was erroneously given for this taxon as n=8-11 by Brown and Coutsis (1978), and later corrected by Coutsis and De Prins (2007) who established the chromosome number with an approximation due bivalents overlaps as n=ca84-88. Here we were able to make a precise count of chromosome elements in this taxon and to demonstrate that n=90, exactly as in P. ripartii. We do not confirm the proposed difference between P. nephohiptamenos and P. ripartii in number of large chromosomes (Coutsis and De Prins 2007). In our squash preparations, both species demonstrate one big and one medium-sized element in the haploid chromosome set.

P. aroaniensis
The haploid chromosome number for this taxon was erroneously given as n=15-16 by Brown (1976a), and later corrected to be n=48 in few studied metaphase plates by Coutsis et al. (1999). In the single studied sample we were able to make a precise count of chromosome elements and found the haploid chromosome number to be n=47. Both counts (previous n=48 and n=47 in this study), are essentially different from those found in closely related P. timfristos and P. orphicus (Kolev 2005, this work) and P. humedasae (Troiano et al. 1979, Vila et al. 2010).

P. orphicus and P. eleniae
The chromosome number of P. orphicus was first established by Kolev (2005) who discovered n=41-42 in population from Hvoyna (Bulgaria), thus, similar to the karyotype found in P. dantchenkoi from remote east Turkey (Lukhtanov et al. 2003).
The chromosome number of P. eleniae was established first by Coutsis and De Prins (2005) who discovered n=41 in population from Falakro Mt near Granitis (Greece). Coutsis and De Prins reported that despite identical chromosome number, karyotypes of P. orphicus and P. eleniae were different in respect to their structure. Karyotype of P. eleniae was reported to be more asymmetrical than karyotype of P. orphicus (that is, the chromosomes were more differentiated with respect to their size).
Here we reinvestigated the karyotypes of P. orphicus and P. eleniae originating directly from their type-localities. Our data confirm previous chromosome number counts, but do not confirm the differences in karyotype structures. In our opinion, the presumed differences could appear because of differences in staining techniques used by Kolev (2005) for P. orphicus and Coutsis and De Prins (2005) for P. eleniae (see Wiemers and De Prins 2004). In our study, we used the same technique for both taxa, and we did not find any differences in the karyotype structure.

P. timfristos
The haploid chromosome number of this taxon is established first here as n=38 and thus differs by at least three fixed chromosome fussions/fixions from P. orphicus orphicus and P. orphicus eleniae (n=41-42). This number is similar (but not identical) to that found in P. humedasae (n=39, Vila et al. 2010). We are not sure that the karyotypes of P. timfristos and P. humedasae are related in their origin because they are not found in proximity and separated by an area where P. orphicus with n=41-42 is distributed.

P. admetus
The Balkan and west Turkish populations of Polyommatus admetus have a unique hindwing underside pattern (Polyommatus admetus type, Fig. 13a, b, c, d) and can be easily separated on the basis of morphology from other species. However, some taxonomic and identification problems appear if oriental populations of P. admetus sensu lato are considered. In 2004, P. admetus yeranyani from Armenia and P. admetus malievi from Azerbaijan were described . The two last taxa differ from the nominative subspecies morphologically. They usually have a distinct white streak on the underside of the hindwing, and the marginal pattern of the wing underside is not as prominent as in P. admetus admetus. In fact, P. admetus yeranyani and P. admetus malievi are phenotypically similar to P. ripartii and P. demavendi, and their identification is not always easy. Karyological analysis revealed a minor difference between the western and oriental forms (see above), and molecular analysis demonstrated that they were differentiated with respect to COI barcodes and did not constitute together a monophyletic entity. This barcode distinctness is especially clearly expressed in the haplotype network (Fig. 10). Therefore, in accordance with the criterion of avoiding non-monophyletic groups in taxonomy (Vila et al. 2013), they should be treated as distinct species P. admetus and P. yeranyani.

P. ripartii
The distribution of COI haplotypes in P. ripartii demonstrates a very complex picture. This taxon is represented by several clades on the phylogenetic reconstructions. The West-European clade includes butterflies from France, Italy and Spain. Another clade (a "mixed", or Eurasian clade) includes butterflies from the whole Western Palaearctic region from Spain to Mongolia. Eastern Turkish-Caucasian clade (P. ripartii paralcestis) is strongly differentiated and appears as a group close to P. demavendi. Complicated taxonomy and phylogeography of P. ripartii have recently been topics of several specific studies and publications (Vila et al. 2010, Vodolazhsky et al. 2011, Dincă et al. 2013a and are out of the focus of the present paper. The sequences obtained in our study confirm that Balkan samples represent one of the major clades within P. ripartii populations, thus P. ripartii pelopi is confirmed as a valid subspecies.

P. nephohiptamenos
Taxonomic interpretation of this local Balkan endemic is difficult since it is morphologically very similar and chromosomally seems to be identical to the close species P. ripartii. However, distinct COI barcodes in combination with ecological differentiation (P. nephohiptamenos is a high altitude species, whereas P. ripartii pelopi can be found usually at middle and low elevations) do not allow us to reject the pre-existing taxonomic hypotheis that P. nephohiptamenos represents a distinct taxonomic entity. The fact that P. nephohiptamenos retains its homogeneity with respect to COI being surrounded by closely related P. ripartii is additional indirect evidence for a presence of genetic boundaries between them. Further molecular and genetic studies are required to understand the real taxonomic status of P. nephohiptamenos.

Polyommatus orphicus
Polyommatus dantchenkoi orphicus was described (Kolev 2005) and later considered (Tshikolovets 2011, Eckweiler and Bozano 2016) as a subspecies of P. dantchenkoi, a species known from east Turkey, because P. d. dantchenkoi and P. d. orphicus shared a similar phenotype and number of chromosomes (Lukhtanov et al. 2003, Kolev 2005. At times, P. orphicus has been considered as a distinct species (e.g. Van Swaay et al. 2010); however, its species level status was not justified.
Our molecular data demonstrate that, despite similarity in number of chromosomes, P. d. dantchenkoi and P. d. orphicus are not closely related as was previously thought. In the haplotype network, these taxa were found to be placed in the opposite parts of the recovered net, being separated by a number of other species (Fig. 11). Their merging would result in a polyphyletic assemblage (Fig. 8). Avoiding non-monophyletic groups is a preferable option in practical taxonomy (Talavera et al. 2013a). Therefore, P. dantchenkoi and P. orphicus should be considered as two distinct species. We should also note that the COI barcodes alone (as in our study) can provide weak evidence for monophyly or non-monophyly of taxa since trees inferred from single markers sometimes display relationships that reflect the evolutionary histories of individual genes rather than of the species being studied. In case of Agrodiaetus, COI barcodes showing such a discrepancy between species and gene trees may be a result of interspecific mitochondrial introgression (Lukhtanov et al. 2008. Despite this limitation, we argue that monophyletic clusters resulting from the DNA barcode analysis are better primary taxonomic hypotheses than para-or polyphyletic ones . Polyommatus eleniae was described from a place located 80 km south-west from the type locality of P. orphicus. Polyommatus orphicus and P. eleniae have the same number of chromosomes, but it was supposed that they were different in karyotype structure (Coutsis and De Prins 2005). Additionally, it was supposed that P. eleniae differed from P. orphicus by the constant lack of a white postdiscal streak on the forewing underside (character 1 on Fig. 12) and by strong reduction or total lacking of a white streak on the hindwing underside (character 2 on Fig. 12) (Coutsis and De Prins 2005). In P. orphicus these streaks are supposed to be always sharply defined (Kolev 2005). Our study does not support the difference in karyotypes (see above). Our analyses showed that the supposed differences in morphology disappeared if individual variations were taken into account. Although the "typical" phenotype of P. orphicus (Figs 12 and 14a) often present in in Hvoyna, the individuals with different level of reduction of white streak on both fore-and hindwing underside are very common (Fig. 13h, i, j). These individuals with confidence can be identified as P. orphicus as they have the same karyotype and do not differ in mitochondrial haplotypes. Thus, the morphological difference between individuals from Hvoyna (P. orphicus) and Falakro Mt (P. eleniae) is not clear and is not based on fixed characters. The difference in karyotypes was also not confirmed in our analysis (see the section Chromosomal diversity above). Therefore, we conclude that the population from Falakro Mt is most probably conspecific with P. orphicus and can be treated as a subspecies P. orphicus eleniae.

P. aroaniensis
This taxon was first described by Brown (1976a) as a subspecies of P. alcestis and two years later was raised to species rank (Brown and Coutsis 1978). Despite its similarity to other taxa of the brown complex, especially with P. humedasae, P. orphicus orphicus and P. o. eleniae, it differs by its karyotype and COI barcodes. Its species distinctness confirmed by chromosomal analysis (Coutsis et al. 1999) has never been questioned. Thus, there has been no problem with treatment of P. aroanisnsis as a separate species. However, there are numerous identification problems associated with P. aroaniensis because several populations from Central and Northern Greece, as well as from other countries of the Balkan Peninsula were identified as P. aroaniensis (see the section Distribution areas below), but their karyotypes were not studied. In our work, we discovered that two of these populations (from Timfristos Mt and Parnassos Mt) represented a previously unrecognized species. Below we name it and provide its formal description.
Underside: ground color light brown with yellowish coffee-milk tint. Greenish blue basal suffusion very slight, nearly lacking. One basal black spot is present only on hindwings. Discoidal black spot is present on the forewings, but can be slightly seen on the hindwings (absent or vestigial). Postdiscal black ocelli are encircled by a whitish border. They are prominent on the forewings, forming a strongly curved row. Postdiscal black ocelli on the hindwing small. Submarginal and antemarginal mark- ing is absent on the forewings, and absent or vestigial on the hindwings. White streak on hindwings clearly visible. In one specimen the white streak is vestigial, in one the white streak is almost absent (can be slightly distinguished), and in one specimen there is an additional short streak between postdiscal and submarginal areas of the wing, straight under the main white streak. Fringe brown, slightly darker than the underside ground color.
Genitalia: the male genitalia have a structure typical for other species of the subgenus Agrodiaetus (Coutsis, 1986).
Females (Fig. 17a-g). Forewing length 15.8-17.5 mm.Upperside: ground color as in males, but lighter dark brown and without sex brand and scaletuft. Fringe greyish brown. Underside: ground color and general design as in males but fringes lightercolored. Greenish blue basal suffusion almost invisible. White streak on hindwing underside is present in all paratypes and demonstrates a variable level of reduction.
Diagnosis. Polyommatus timfristos (n=38) differs by at least three fixed chromosome fusions/fissions from the most closely related and allopatric P. orphicus orphicus and P. orphicus eleniae (n=41-42). P. timfristos (n=38) differs by at least nine fixed chromosome fusions/fissions from allopatric P. aroaniensis (n=47). From the closely related P. orphicus and P. aroaniensis, P. timfristos differs also by a number of nucleotide substitutions within the studied 657-bp fragment of the mitochondrial COI gene.
The chromosome number in P. timfristos (n=38) is similar (but not identical) to that found in P. humedasae (n=39, Vila et al. 2010). However, we are not sure that these karyotypes are related in their origin because they are not found in proximity and separated by an area where P. orphicus with n=41-42 is distributed. With respect to COI barcodes, the pair P. timfristos/P. humedasae is more differentiated than pairs P. timfristos/P. aroaniensis and P. timfristos/P. orphicus.
From sympatric and syntopic P. ripartii pelopi the new species can usually be distinguished by the absence of submarginal marking and strong reduction of greenish blue basal suffusion. These characteristics are usually (but not always) better expressed in P. r. pelopi specimens. In doubtful cases, the separation is only possible on the base of chromosomal and molecular markers since these species are different: the chromosome number of P. r. pelopi is n=90; they also have fixed differences in 33 positions within the studied 657-bp fragment of COI gene.
Etymology. Timfristos is a mountain in the eastern part of Evrytania and the western part of Phthiotis in Central Greece. The name is a noun.

P. admetus Figs 24-26
This species is widespread in the Balkan Peninsula. It is local in the northern part of Hungary (Ilonczai andBálint 2001, Bálint et al. 2006) and recorded in Slovakia (Kulfan and Kulfan 1992, Eckweiler andBozano 2016). It has been shown to be widely distributed in the western part of Romania, but no exact localities were provided (Eckweiler and Bozano 2016). It is common in Greece and found in Croatia, Bosnia and Herzegovina, Montenegro, Serbia, Bulgaria, The Republic of Macedonia, Albania and European Turkey (Sijaric and Mihljevic 1972, Hesselbarth et al. 1995, Abadjiev 2001, Tolman and Lewington 2008, Eckweiler and Bozano 2016. Fig. 27 Polyommatus ripartii is widespread in the southern part of the Balkan Peninsula (Greece and Bulgaria); however, it is more local in the north. It is known from Albania, the Republic of Macedonia, south Serbia (Kudrna et al. 2011, Eckweiler andBozano 2016), Bosnia and Herzegovina (Koren 2010). It was mentioned for European Turkey (Hesselbarth et al. 1995, Tolman andLewington 2008) and recently found in Croatia (Koren 2010, Dincă et al. 2013a.

P. nephohiptamenos Figs 28-30
Polyommatus nephohiptamenos has a dot-like distribution area and is known from the high altitudes of north-east Greece (Mt Pangeon, Mt Phalakro and Mt Orvilos) and south-west Bulgaria (Mt Orvilos, also known as Mt Slavyanka, Mt Alibotush and Kitka Planina) (Kolev 1994, Tolman and Lewington 2008, Eckweiler and Bozano 2016. Fig. 31 Polyommatus aroaniensis has been considered as a relatively widespread species (Kolev and van der Poorten 1997). Apart from its type-locality (South Greece, Peloponnese), it has been recorded in different parts of Central and Northern Greece (Brown 1976a, Wakeham-Dawson and Spurdens 1994, Wakeham-Dawson 1998, Pamperis 2009, from a few areas in south Macedonia van der Poorten 1997, Melovski andBozhinovska 2014) and from some localities in south-west Bulgaria and one isolated place in the central part of the country (Abadjiev 2001, Kolev 1994, Kolev and van der Poorten 1997. Tshikolovets (2011) and Eckweiler and Bozano (2016) show its distribution extending into Albania, although the species has not been recorded from this country in recent surveys (Verovnik and Popović 2013). Verovnik et al. (2015) recorded it in Bosnia and Herzegovina. Koren and Laus (2015) recorded it in Croatia; however, this record was not confirmed by molecular data (Lovrenčić et al. 2016).

P. aroaniensis
Our chromosomal data confirm P. aroaniensis in South Greece (Peloponnese), but cannot confirm it in Central and Northern Greece and in Bulgaria where it is replaced by the closely related allopatric species P. timfritos and P. orphicus. In the light of the data obtained, the occurrence of P. aroaniensis in Bulgaria, Albania, Macedonia and Bosnia and Herzegovina seems to be doubtful and requires a confirmation based on chromosomal analysis. We cannot exclude that the populations from Albania, Macedonia and Bosnia and Herzegovina could represent P. orphicus or even undescribed taxa of the subgenus Agrodiaetus.

P. timfristos Figs 32-35
This species is known from Timfristos and Parnassos Mts in Central Greece only.

P. orphicus Figs 36-39
This species is known from South Bulgaria and Northern Greece only. However, its occurrence in other countries in the northern Balkan is theoretically possible (see above).

An alternative classification and conservation
Theoretically, the main groupings in the P. humedasae -P. orphicus -P. timfristos -P. aroaniesnsis subcomplex can be interpreted as subspecies-level taxa, if the polytypic species concept is applied. None of them appears to be sympatric in distribution, and taken together they form a moderately supported monophyletic lineage on the COI+ITS2 tree (Fig. 9). Under this scenario, this subcomplex would be considered a diverse array of allopatric populations, each of which possesses unique genetic attributes (karyotypes and molecular markers) and is distributed in a particular area within the Alp-Balkan region. As possible theoretical support for this alternative classification, one can argue that differences in chromosome numbers in Agrodiaetus do not necessarily result in complete reproductive isolation, and, at least in some particular cases, do not prevent interspecific hybridization and genetic introgression ).
However, even if the last statement is true, it does not mean that chromosomal rearrangements are irrelevant to formation of genetic barriers between populations. Chromosome changes have been shown to be important in speciation in the blue butterflies , Kandul et al. 2007, Talavera et al. 2013b. Even a weak decrease in fertility in heterozygotes for multiple chromosomal rearrangenments can result in selection against them and in formation of a boundary between chromosomally diverged homozygous populations. Additional studies are required to shed light on this topic. Recent studies have treated P. humedasae, P. aroaniensis and P. orphicus as species-level taxa (Eckweiler and Bozano 2016), which our study suggests is a reasonable interpretation although distribution areas of P. aroaniensis and P. orphicus should be corrected. Based on our current knowledge, if P. humedasae, P. aroaniensis and P. orphicus are considered species-level taxa, P. timfristos should be treated as a species-level taxon as well.
Regardless of its taxonomic status as a species or subspecies, P. timfristos represents a unique entity within the genus Polyommatus that deserves additional study. A better understanding of its evolutionary history may be helpful in understanding mechanisms of chromosomal diversification within the genus, and may further elucidate the biogegraphy of the south Balkan and Aegean regions. As a distinct taxonomic entity occupying a very restricted area in Central Greece it should be considered a candidate on the list of protected species in Greece and the whole of Europe.

Biogeography
Analysis of distribution areas and phylogeny of the P. dolus lineage shows that the phylogeograpic history of this complex involved a combination of dispersal and vicariance events with a clear general trend of dispersal from the East (Iran), where the group most likely arose, to the West: to the Mediterranean area and to the Iberian Peninsula (Vila et al. 2010). The Europe was estimated to be colonized approximately 1.24 Mya (range 0.88-1.64 Mya). Approximately 1.15 Mya (range 0.80-1.51 Mya), the Euro-pean lineage was divided into three subclades located (1) in the Balkan Mountains and Alps (P. aroaniensis sensu auctorum: the Balkans; P. humedasae: the Alps), (2) southern Spain (P. violetae), and (3) the Iberian-Italian region (P. fabressei + P. dolus), respectively (Vila et al. 2010).
Three chromosomal sublineages discovered in our study (P. aroaniensis sensu stricto + P. orphicus +P. timfristos) represent late Pleistocene splits of the Balkan subclade that evolved in allopatry within the Balkan refugium. Given the deep level of chromosomal divergence between these sublineages, we assume that there was a long period of allopatric differentiation when they were separated by geographic or/and ecological barriers. In our opinion, this is evidence for presence of three separate Balkan subrefugia in the past (Pelonnese, Central Greece and Northern Grecee/South Bulgaria).
Greece, as a part of the Balkan Peninsula, has been already reported to harbor genetically differentiated lineages from the rest of the Balkans for a number of animal species as a result of evolution in multiple separate refugia (Kasapidis et al. 2005, Alexandri et al. 2012, Karaiskou et al. 2014. Thus, our data provide a chromosomal evidence for this refugia-within-refugia concept (Gòmez andLunt 2007, Karaiskou et al. 2014), and the discovery of a new, chromosomally diverged species P. timfristos stresses the biogeographic importance of Central Greece as a separate Pleistocene refugium within the Balkans.

Taxonomic conclusion
We propose the following taxonomic arrangement of the P. dolus and P. admetus lineages (chromosome numbers are in parentheses when known, the Balkan taxa are in bold):  (Lukhtanov et Wiemers, 2003) (Vila et al. 2010) or separate subspecies of Polyommatus ripartii (Eckweiler and Figure 19. Fragment of the ML tree of P. admetus and P. dolus complexes based on analysis of COI barcodes and focused on details in the structure of the West-European and the "mixed" (Eurasian) clades of P. ripartii. Numbers at nodes indicate bootstrap support.