The combined dataset included 3139 base pairs (bp) from 125 samples including seven outgroups. The MP (Maximum Parsimony) analysis of the combined dataset resulted in 80 equally most parsimonious trees, the strict consensus of which is shown in Fig 1. The ML (Maximum Likelihood) analysis on the same dataset resulted in a topology that differed at several nodes (Fig 2). However, these differences were mainly in the basal nodes that were weakly supported in both analyses, and in the ML tree, preceded by short branches. Nodes that were well supported (>80% bootstrap values) in the ML analysis also figured in the MP topology, albeit with lower support in general. Both analyses supported the monophyly of Mycalesina (MP: 98% bootstrap; ML: 100% bootstrap). The BI (Bayesian Inference) tree was different to both MP and ML, but incongruence was restricted mainly to the basal part of the tree (Fig 3). Nodes supported strongly in the ML analysis were recovered with strong posterior probabilities in the BI tree. Results that are consistent across the three tree-building methods are elaborated below.

Bicyclus, Hallelesis and Nirvanopsis emerged as monophyletic groups in all three analyses whereas the remaining genera were either paraphyletic or polyphyletic. The monophyly of Bicyclus was strongly supported in the model-based analyses but only moderately so in MP. Hallelesis and Nirvanopsis, each represented by two species, were monophyletic with strong support. Nirvanopsis was nested within Lohora, rendering the latter paraphyletic. Relationships among Heteropsis species were broadly consistent with subgeneric groupings in 7 . A group consisting of M. adolphei, M. janardana, M. sangaica, M. mamerta and M. malsara was nested within Heteropsis. The remaining Mycalesis species clustered into two clades, but Mycalesis as a whole was polyphyletic.

Six higher clades within Mycalesina were common to all three analyses, each with moderate to strong support in the model-based analyses, but weaker support in MP. These will be referred to as the stable clades and have been named either with novel or available names to facilitate discussion (Figs 1, 2 and 3). Relationships within these stable clades were similar, if not congruent, between analyses (an exception is the unresolved nature of Mycalesis clade II in MP). Some previously recognised species groups (Additional file 1) in Mycalesis were also recovered as well-supported clades. The Mineus group of Evans (2-species group of Aoki and colleagues 29 ; approximating to Calysisme of Moore plus Jatana) was one such group, which includes M. mynois, M. perseus, M. mineus, M. intermedia, M. perseoides and M. visala. M. perseus is the most widely distributed Mycalesis species, ranging from the Indian subcontinent to NE Australia. The other members of the Mineus group are restricted to the Oriental region. M. patnia ('4-species' group of Aoki and colleagues 29 ) was either nested within or sister to the '1-species' group (including M. gotama, M. patnia, M. orseis, M. francisca and M. anaxias; equivalent respectively to Moore's Sadarga, Nissanga, Suralaya, Gareris and Virapa). Members of these three species groups together formed a stable clade, which will be referred to as Mycalesis clade I.

The Australasian species (M. barbara, M. aethiops, M. mucia, M. phidon, M. mehadeva, M. cacodaemon, M. sirius, M. duponchelii, M. discolobus, M. elia, M. terminus, M. mulleri, M. splendens, M. interrupta, M. biliki, M. richardi and M. sara) in addition to the Oriental M. dohertyi, M. maianeas, M. itys, M. fuscum and M. anapita formed another stable clade, hereafter referred to as Mycalesis clade II (equivalent to Mydosama of Moore plus Savanda, Nebdara, Satoa and possibly Culapa). The Oriental genera within the clade were basal members, while the Australasian species were a strongly supported clade nested within. The other stable clades within Mycalesina are Lohora (Lohora + Nirvanopsis), Hallelesis, Bicyclus and Heteropsis (including some species of Mycalesis; Fig 1, 2 &3).

ML analysis on the COI dataset recovered the six stable clades (Additional file 2). Mycalesis clade II was nested within Mycalesis clade I in the EF-1α tree, while the remaining four stable clades were retrieved (Additional file 3). The wingless tree was generally poorly-supported with very short branches, but recovered the Heteropsis and Lohora clades (Additional file 4). Hallelesis and Bicyclus were represented by a single species, whereas Mycalesis clades I and II were not monophyletic.

The likelihood score of the ML topology in Fig 2 (In L = -38610.95571) was significantly higher than the "genus monophyly" topology (ln L = -38702.28514), wherein each genus was constrained to be monophyletic (P = 0.002, Shimodaira-Hasegawa test, Table 1). On the contrary, there was no significant difference in likelihood scores between the ML topology and the tree where Mycalesis clades I and II were reciprocally monophyletic.

Each of the six stable clades was successively removed and the resulting datasets subjected to RaxML analyses. The resulting trees are summarized in Additional file 5. We found that deleting any of the clades did not increase support for basal nodes; nodes with <50% bootstrap values did not receive >50% support.

The dating analysis in BEAST indicates that the ancestors of the six stable clades diverged between 26 and 21 mya (Fig 3) Initial splits within the Lohora, Bicyclus, Mycalesis groups I and II began ca. 18-20 mya. The two species of Hallelesis started diverging ca. 1.5 mya.

We propose that the group of Mycalesis species nested within Heteropsis should henceforth be classified under Heteropsis. Mycalesis clades I and II were never sister to each other in the combined analyses, although a topology with Mycalesis clades I and II constrained to be sister to each other was not significantly worse than the ML topology (where they are not sister to each other). Nevertheless, we propose that members of clade I and II should be classified under two respective genera. The genetic divergence between the clades is comparable to the divergence between other genera in our study, supporting our re-classification scheme. Furthermore, the two clades almost completely allopatric, with clade I restricted to South and SE Asia, whereas clade 2 is predominantly Australasian. Moreover, even under the unlikely event that these two clades indeed turn out to be each others' sisters with strong support, our proposed classification system would still remain valid. Clade I includes the type species of the genus (M. francisca) and we propose that members of this clade should be classified under Mycalesis sensu stricto. We transfer the species in the Mycalesis Clade II to Mydosama (type species, Dasyomma fuscum Felder & Felder, 1860). We also subsume Nirvanopsis under Lohora.

It is important to bear in mind that our divergence time estimates are based on a single secondary calibration point (from 64 ) and as such are a first approximation. We here present our interpretation of the biogeographic history of the group based on these timing estimates and the strongly supported monophyletic groupings. The age of the most recent common ancestor of Mycalesina suggested by the dating analysis precludes a Gondwanan origin of the group. The most likely sister clade of the group is composed of a group of genera distributed predominantly in the Oriental and the easternmost parts of Palaearctic region 59 64 . Miller's hypothesis 11 was that the ancestor of mycalesines was of Oriental origin and the group diversified through dispersals to the Afrotropical and Australasian regions. Mycalesis clade I emerged sister to remaining mycalesine clades in the ML tree, in support of Miller's hypothesis. However, the MP and BI trees retrieved the African Bicyclus as sister to the rest of the mycalesines, although without strong support for the latter group. This would point to an African origin, as already shown in two other nymphalid groups, Junonia and Charaxes 6 65 . In this scenario, mycalesines started evolving in Africa, with subsequent colonization of Madagascar where there was explosive radiation of the Heteropsis group, and this group subsequently invaded Asia. With the poorly supported basal relationships, however, we are unable to distinguish between hypotheses of African and Asian origin. Nonetheless, it is clear that there has been at least one dispersal event between the Afro-Malagasy and Asian regions in the early evolution of the group. This colonization(s) has most likely occurred across the Arabian Peninsula, which is thought to have been covered by forests at some point during the Miocene 66 67 . The Arabian Peninsular region appears to have been important for butterflies as a corridor of 'geodispersal' (concordant dispersal of several lineages over the same route) 68 between Africa and Asia.

The grouping of some Asian mycalesines with the Afro-Malagasy genus Heteropsis is well supported by all combined analyses and independently by the COI and EF-1α datasets (note that the close relationship of M. adolphei with Heteropsis is also supported by morphological data, in particular the form of the male genitalia; 7 18 ). This grouping suggests at least one dispersal event from the Afrotropical to the Oriental Region, as the Oriental members are derived within the Heteropsis group. Our trees also suggest tentatively (although lacking support for the placement of Admiratio + Masoura) that the Madagascan mycalesines comprise two independent radiations (subgenera Admiratio+Masoura, and the clade including subgenera Heteropsis+Henotesia). Greater taxon and gene sampling (note that Admiratio and Masoura are only represented in our tree by COI) is therefore needed to establish a robust topology for Heteropsis sensu lato that can lead to a stronger inference of the biogeographic history of the group. It is very probable that two species (H. comorana and H. narcissus) have colonised the Comoros and Mascarenes from Madagascar 18 26 . However, only one recent colonisation event is suggested from Africa; the secondary colonization of the species B. anynana of Comoros [256].

Sulawesi has been colonised at least twice independently early in the evolution of Mycalesina (between ca. 24-19 mya), by the ancestors of the Lohora clade and of M. itys. Interestingly, while Lohora has radiated into 19 species, the ancestor of M. itys seems not to have undergone in situ speciation within the island (M. itys is the only Mycalesis species endemic to Sulawesi).

The endemic Australasian fauna share a common ancestor that was derived from the Oriental region. This ancestor diversified rapidly in New Guinea, with the first splits occurring ca. 15 mya. The endemics in the Solomon Islands (biliki-splendens-richardi-interrupta-sara) descend from an ancestor derived from New Guinea in the Late Miocene. M. terminus and M. sirius (Mycalesis II clade) expanded their range into Australia from New Guinea while M. perseus (Mycalesis I clade), has colonized Australia and the Solomons between 1 and 2 mya.

Although some mycalesine faunas, such as that now represented as Clade I of SE Asia, tend to be rather brownish or homogeneous in colour pattern, this is not the case everywhere. The endemics in the Australasian region and the Lohora clade of Sulawesi include more colourful members, and may even be mimetic; in the case of Mycalesis drusillodes so spectacularly that they had been described as different species, the male mimicking a Tellervo and the female a Taenaris 13 14 . The Malagasy fauna, in particular, is exceptionally diverse in wing pattern and wing shape. There is also evidence for mimicry between different clades of Heteropsis in Madagascar and between Heteropsis (Masoura) masoura (in this case sexually monomorphic) and the aposematic pierid Mylothris phileris 7 . Dead leaf mimicry is also evident in the Malagasy Heteropsis drepana. The significance of these regional differences in colour pattern and form in different faunas perhaps relates to aspects of the environment, notably differing suites of predators 69 , but there is to date no experimental evidence of their role. In Madagascar, mycalesines occurring in open areas or on margins of forests are markedly more orange rather than brown 7 .

This remarkable inter-specific diversity in wing patterns may also have been mediated, at least partially, by sexual selection. It has been shown experimentally that eyespots on the dorsal surface of B. anynana are used by females to choose mates 47 48 . Pronounced sexual dimorphism occurs in some African Bicyclus, and especially within the Heteropsis subgenus Henotesia in Madagascar, another case extreme enough that females had not been been matched to males, neither had different seasonal stages of males been correctly synonymised (but see 7 70 ). Wing patterns are known to have sex-specific evolutionary rates within Bicyclus 50 . All these lines of evidence point towards sexual selection as an additional mediator of morphological diversification. Results in 50 also suggested that mate signalling selects for varying dorsal wing patterns while ventral characters are selected upon by predation pressure. Thus, a complex interaction of differential selective forces along with the heterogeneity of environments inhabited by these species seems to have driven the morphological diversification of wing patterns in this group, and much needs to be learnt about the systems.

The most striking aspect of diversification that may be related to speciation in mycalesine butterflies is in the scent organs, which are exceptionally diverse, including male androconia on forewing underside, hindwing upperside, abdominal underside and upperside, specialized wing veins, and specialized polished areas that may have protective function 7 . These androconia are known to release male sex pheromones during courtship, which are crucial determinants of reproductive isolation between species 71 72 . Even to humans, mycalesine scent brushes differ in odour 7 . A denser sampling is needed to test whether the exceptional diversification of androconia is related to elevated levels of speciation in this group. Our phylogeny provides a starting point to identify groups that can be used for further comparative analyses that seek to understand the underlying processes that might have generated these diverse characters.

From a biogeographic perspective, the moderate to weak dispersive powers of mycalesine butterflies explains some of their diversity, because poor dispersive power is expected to be conducive towards higher rates of allopatric speciation 73 . Dispersal into a novel area is more likely to result in allopatric speciation compared to a group with high vagility that can maintain regular geneflow 73 . Vicariance events are also more likely to result in speciation when the ancestor in question has weak dispersive powers. For instance, the aridification of central India during the Pliocene may have resulted in vicariant events that left descendant sister species pairs disjunct in South-West India and North-East India 74 . Unfortunately, we are unable to test this hypothesis with our limited taxon sampling.

The radiation of Satyrini (Satyrinae) butterflies, to which Mycalesina belongs, is thought to have closely followed the diversification of C₄ grasses during the Oligocene 33 . Host-plant mediated speciation can occur due to episodes of host-expansion and specialization during the history of the group 75 76 . The degree of specialization of satyrines on their grass host plants is unknown, but Mycalesis is one satyrine group where host-plant preference hierarchy and specialization has been reported 77 . A rigorous test of the hypothesis that mycalesines radiated by co-evolving with their grass hostplants is, however, hindered by the lack of reliable host plant records for most mycalesines.

We surmise that a combination of several factors - androconial diversification, sexual selection on wing patterns, moderate dispersive powers, and perhaps also co-evolution with grasses - has resulted in the high diversity of species in the group. In summary, this group of butterflies presents an exciting opportunity to understand patterns and processes of diversification in insects, especially in unravelling the complex interactions among various selective forces and developmental aspects of wing patterns. The phylogeny encompassing the entire radiation and nomenclatural rationalisation at a generic level presented here is the first step that will eventually permit large-scale analyses to explore specific hypotheses within a comparative framework.

Specimens of 42 species of Mycalesis, 28 of Heteropsis, four of Lohora and two of Nirvanopsis were collected either by the authors or collaborators at different times between 2001 and 2008. 13 widespread species were represented by more than one sample. DNA was preserved either through dessication or by immersing two legs in alcohol. Once the samples reached the lab, DNA was extracted from two legs using the DNEasy extraction kit (QIAgen). DNA was amplified from three gene regions - 1450 bp (base pairs) of COI (cytochrome oxidase subunit I), a mitochondrial gene, and two nuclear genes - EF-1α (Elongation Factor 1 alpha; 1240 bp) and wingless (400 bp). The gene trio has been successful in resolving relationships of species within a genus in several nymphalid studies 65 78 79 80 81 .

Readers are referred to 82 for a list of primer sequences used here. COI was amplified using the primer pairs LCO-HCO and Jerry-Pat. Three primer pairs were used for EF-1α - Starsky-Luke, Cho-Verdi and EF51.9-EFrcM4, while LepWing1 and LepWing2 or Wingnut 1A (5'-GAA ATG CGN CAR GAR TGY AA-3') and Wingnut-3 (5'-ACY TCR CAR CAC CAR TGR AA-3') were used for wingless. The PCR protocol used for Starsky-Luke was as follows -95°C for 7 min, 40 cycles of 95°C for 30 s, 55°C for 30 s and 72°C for 1 min followed by a final extension period of 72°C for 10 min. For Wingnut 1A-Wingnut 3, conditions were 80°C for 1 min, 40 cycles of 94°C for 1 min, 46-52°C for 2 min and 72°C for 1-2 min followed by a final extension period of 72°C for 10 min. For the remaining five primer pairs, we used the following protocol: 95°C for 7 min, 40 cycles of 95°C for 30 s, 50°C for 30 s and 72°C for 1 min followed by a final extension period of 72°C for 10 min. Successfully amplified PCR products were sequenced with a Beckmann-Coulter CEQ8000 automated sequencer. The resulting chromatograms were visualized in BioEdit v7.0.5.3 83 and aligned by eye. We also included sequence data from the genera Bicyclus, Heteropsis and Hallelesis that were available on Genbank. Outgroup data were taken from 32 . Additional file 5 lists the samples used in this study with their collection localities and Genbank accession numbers for respective sequences.

The combined dataset was analyzed under the maximum parsimony (MP) criterion in TNT v 1.1 84 . Heuristic searches including traditional TBR branch swapping procedures and 'New Technology' searches were performed on 1000 random addition replicates. Support for respective clades was estimated using bootstrap values calculated from 1000 pseudo-replicates with 10 replicates each. Maximum likelihood (ML) analyses were performed in RAxML III 85 with default heuristic search algorithms. The GTR+G model, which was chosen by jModelTest 86 under the Akaike Information Criterion, was imposed on the three gene partitions independently with the gamma parameter estimated in 4 discrete rate categories. Bootstrap values were calculated from 1000 pseudo-replicates. Since some samples did not have a complete three-gene dataset (Additional file 6), ML analyses were performed on individual gene datasets excluding samples missing data for respective genes. Individual gene analyses also allow us to assess nodal support from each gene.

We performed Shimodaira-Hasegawa tests 87 to test whether specific topologies resulting from the above analyses were significantly better than competing topologies. The program MacClade 88 was used to construct two constraint trees; in the first tree each genus were forced to be monophyletic and in the second tree only species in Mycalesis clades I and II (see results) were placed in a monophyletic group, while the position of other species remained unconstrained. These constraint trees were used in PAUP* 89 to derive a "genus monophyly" tree and "Mycalesis monophyly" tree through a likelihood heuristic search. To examine support for the above hypotheses the likelihood scores of these trees were compared with the unconstrained ML tree using the one-tailed Shimodaira-Hasegawa log-likelihood test as implemented in PAUP*, using the re-sampling estimated log-likelihood (RELL) technique approximation with 10,000 bootstrap replications. These analyses were implemented on the combined dataset.

We also conducted a 'rogue' analysis to test whether there was one or more 'rogue' clades that was leading to the observed low basal support values. Each of the six 'stable' clades identified in the Results section was successively removed and the datasets analysed in RaxML with bootstrapping to test whether deletion of one of these clades resulted in stronger support.

We used the software BEAST v 1.4.8 90 for Bayesian inference (BI) of phylogenetic relationships and divergence times simultaneously. The analysis was carried out without outgroups since BEAST does not rely on outgroups to root the tree; instead it uses a relaxed molecular clock. The "treeModel.RootHeight" prior (i.e., the age at the root of the tree) was set to a normal distribution with a mean of 27 million years and a standard deviation of 3. This date was taken from 64 which studied the divergence times within Nymphalidae based on a 10-gene dataset from >400 genera (including Mycalesis, Bicyclus, Hallelesis and Heteropsis), with both minimum and maximum calibration points. The dataset was partitioned into nuclear (EF-1α and wingless combined) and mitochondrial (COI) genes, with parameter values estimated independently for each partition. The GTR+G model was imposed with a relaxed clock where branch lengths were allowed to vary according to an uncorrelated lognormal distribution 91 . The tree prior was set to the Birth-Death process, while all other priors were left to their defaults in BEAST. The analysis was run twice for 10,000,000 generations of MCMC analyses in BEAST and the chains were sampled at every 1,000 generations, yielding a total of 10,000 samples for each run. Whether the parameter estimates and tree topology were at equilibrium was determined by using the program Tracer 91 . The first 1,000,000 generations (or 1000 trees) were discarded as burn-in.