Allopatric speciation is thought to occur in the absence of gene flow, thereby suggesting that widespread vagile species might be less likely to generate restricted sister taxa because of a lack of isolation. The butterfly genus Vanessa provides an ideal test of this concept, as it contains some of the most cosmopolitan and vagile species of butterflies on the planet, as well as some highly restricted taxa. Given the age of these groups, this arrangement offers a special opportunity to examine the relationship between vagility and phylogeny in generating novel taxa; specifically, does the vagility of some lineages impede allopatric speciation, leaving restricted clades more speciose? A phylogenetic hypothesis is proposed for all species belonging to the butterfly genus Vanessa based on DNA sequences from one mitochondrial and eight nuclear gene regions. The resulting topology shows very little conflict among gene regions, with five well-supported clades corresponding to morphologically consistent species groups. The data very strongly indicate a polyphyletic genus Antanartia, and thus to preserve monophyly two species previously assigned to Antanartia are transferred to Vanessa, Vanessa hippomenecomb.n. and Vanessa dimorphicacomb.n., resulting in a total of 22 species placed in Vanessa. A biogeographical analysis shows that in many cases the most geographically restricted species are sister to geographically widespread species, suggesting dispersal and allopatric speciation. Surprisingly, in almost all cases the divergences between widespread and restricted species are quite old (>5 Ma), suggesting long-term isolation and stability of both vagile and sedentary species, despite the high (even intercontinental) vagility of many extant species and, by extension, ancestral species. The biogeography of Vanessa suggests that species vagility and allopatry do not fully explain the forces governing cladogenesis in this remarkable genus.
The cosmopolitan butterfly genus Vanessa (Lepidoptera: Nymphalidae) contains some 20 species, occurring on all continents except Antarctica, with six species having ranges that span several thousand kilometres. Two species are familiar to many people in the Holarctic region: the Painted Lady [Vanessa cardui (Linnaeus, 1758)], which has a global distribution; and the Red Admiral [Vanessa atalanta (Linnaeus, 1758)]. Both are known to be long-distance migrants and can occasionally be found moving north in large numbers during spring (Stefanescu, 1997; Stefanescu, 2001; Mikkola, 2003). Vanessa cardui is particularly well known for long-distance flights, as individuals have been recorded from the North Atlantic islands of Iceland and Svalbard, Norway, where they are unable to overwinter, representing a flight of at least 1000 km from the nearest known breeding areas. In sharp contrast, several other species are not nearly as mobile, and several are restricted island endemics, such as Vanessa tameamea Eschscholtz, 1821 from the Hawaiian Islands and Vanessa vulcania Godart, 1819 on Madeira and the Canary Islands. This is a clear paradox: how can a clade of butterflies containing some of the world's most mobile and widespread insects also have closely related species that are such narrow endemics? Furthermore, does this mix of widespread and restricted species suggest an evolutionary pattern by which the restricted species might arise from more mobile ancestors, or are there separate clades of vagile and restricted species, as highly vagile species would be too mobile to permit localized specialization or speciation (Kodandaramaiah, 2009)? Finally, localized taxa would seem more vulnerable to extinction events than highly mobile species: do phylogenetic relationships suggest that the most geographically restricted species are also the youngest lineages, and vice versa?
Vanessa was extensively revised by Field (1971) 40 years ago, and since then a few of his subspecies have been elevated to species level (e.g. Wiemers, 1995), one new species has been described (Hanafusa, 1992), and Field's resurrected genera Bassaris and Cynthia have been resynonymized with Vanessa (Wahlberg et al., 2005; Vane-Wright & Hughes, 2007). It is this broader view of Vanessa that we follow here. The species-level taxonomy is thus quite stable, and any new results are likely to just fine-tune the currently accepted classification. The phylogenetic relationships among some of the species have been examined, but not in detail. A few molecular studies including species of Vanessa have been published (Wahlberg et al., 2005; Otaki et al., 2006), with the most significant result being that the superficially divergent species Vanessa abyssinica (Felder & Felder, 1867) is, in fact, a species of Vanessa (Wahlberg et al., 2005; Vane-Wright & Hughes, 2007), and does not belong in the genus Antanartia, as was long believed (Howarth, 1966). This result raises broader questions about relationships between the genera: specifically, are previously unsampled species of Antanartia [Antanartia dimorphicaHowarth, 1966, Antanartia hippomene (Hübner, 1823) and Antanartia borbonica (Oberthür, 1880)] more closely related to Antanartia delius (Drury, 1782) and Antanartia schaeneia (Trimen, 1879) or to Vanessa?
Vane-Wright and Hughes (2007) review several biogeographic hypotheses for Vanessa, particularly with regard to the Vanessa indica (Herbst, 1794) species complex. They suggest rather controversially that the V. indica group originated in North America, although all extant species occur in the Old World. They also suggest that vicariance has played an important role in the historical biogeography of the complex, and rightly state that knowledge of the age of the clade would help resolve questions of provenance and diversification across the lineage.
Here, we present a phylogenetic hypothesis for all species in the genus Vanessa, and use this phylogenetic hypothesis to study the temporal framework of divergences across the genus. We base our phylogenetic hypotheses on DNA sequences of nine gene regions (one mitochondrial and eight nuclear protein coding genes).
Material and methods
All species of Vanessa were studied (Table S1), including the relatively recently described Vanessa dilectaHanafusa, 1992, which is restricted to West Timor in the Indonesian Archipelago, for which cytochrome c oxidase subunit I (COI) sequences were taken from the study by Otaki et al. (2006). When possible, more than one individual was sampled per species. Out-groups were taken from the tribe Nymphalini, including seven species of the putative sister genus Hypanartia (Wahlberg et al., 2005; Wahlberg, 2006), one species from each of the genera Polygonia, Nymphalis, Kaniska, Aglais, Mynes, Araschnia and Symbrenthia, as well as two species of Antanartia that have not been included in phylogenetic analyses previously (A. dimorphica and A. hippomene) and two species of Antanartia s.s. (Table S1). Between five and nine gene regions, known to be informative at the species and genus level (Kodandaramaiah & Wahlberg, 2007, 2009; Peña & Wahlberg, 2008; Wahlberg et al., 2009b), were sequenced from a selection of individuals (between one and three individuals per species; Table S1): from the mitochondrial genome COI, and from the nuclear genomes elongation factor 1α (EF-1α), wingless (wgl), ribosomal protein S5 (RpS5), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), arginine kinase (ArgKin), carbamoylphosphate synthase domain protein (CAD), isocitrate dehydrogenase (IDH) and cytosolic malate dehydrogenase (MDH), for a total of 7010 bp. Primers and laboratory protocols were taken from Wahlberg & Wheat (2008). The direct sequencing of polymerase chain reaction (PCR) products was performed on a Beckman Coulter (Stockholm, Sweden), Applied Biosystems 3170xl Genetic Analyzer (Turku, Finland) or an Applied Biosystems 3130xl Genetic Analyzer (Hawaii). Intraspecific variation in COI was assessed by sequencing several individuals from as widely separate collecting localities as possible, also including COI sequences deposited in GenBank from the study by Otaki et al. (2006).
The resulting chromatograms were examined by eye in bioedit (Hall, 1999), and any heterozygous positions (two equally sized peaks observed at one position) in the nuclear genes were coded with International Union of Pure and Applied Chemistry (IUPAC) ambiguity codes (no heterozygous positions were observed in the mitochondrial gene). All sequences are from protein-coding genes, and thus alignment was trivial.
Data were analysed using Bayesian inference and maximum parsimony. The Bayesian analysis was performed with mrbayes 3.1 (Ronquist & Huelsenbeck, 2003) on the combined dataset, with all specimens with sequence data included. Missing gene sequences were coded as ‘?’. The dataset was partitioned into gene regions and analysed simultaneously under the GTR+Γ model for each partition. Parameter values were estimated separately for each gene region using the ‘unlink’ command, and the rate prior (ratepr) was set to ‘variable’. Because preliminary analyses became stuck in the ‘land of long trees', we set the branch length prior to 0.01 using the command brlenspr = unconstrained:Exp(100.0) (Marshall, 2010). The analysis was run twice simultaneously for 10 million generations with four chains (one cold and three heated), and every 1000th generation was sampled. The first 1000 sampled generations were discarded as burn-in from both runs (based on a visual inspection of when the log likelihood values reached stationarity and the standard deviation of split frequencies was below 0.01), leaving 2 × 9000 sampled generations for the estimation of posterior probabilities. Results of the two simultaneous runs were compared for convergence using tracer v1.4.6 (Drummond & Rambaut, 2007).
Parsimony analyses were conducted using a heuristic search algorithm in tnt (Goloboff et al., 2008) on the equally weighted combined dataset, including only specimens with five or more gene sequences, so as to avoid fallacious taxon placement as a result of missing data. The data were subjected to 100 random addition rounds of successive Sectorial, Ratchet, Drift and Tree Fusing searches (Goloboff, 1999; Moilanen, 1999; Nixon, 1999). We evaluated the character support for the clades in the resulting cladograms using Bremer support (Bremer, 1988, 1994) and Partitioned Bremer Support (PBS; Baker & DeSalle, 1997; Gatesy et al., 1999). The scripting feature of tnt was used to calculate these values (see Peña et al., 2006), and the PBS values were summarized using the partition congruence index (PCI; Brower, 2006).
Bayesian inference of phylogeny and times of divergence were carried out using beast v1.5.2 (Drummond & Rambaut, 2007). The analysis was only performed with a selection of Vanessa samples that had at least five of the nine gene regions sequenced. The dataset was partitioned into the gene regions and analysed simultaneously under the GTR+Γ model for each partition separately, and with a relaxed clock allowing branch lengths to vary according to an uncorrelated Lognormal distribution (Drummond et al., 2006). The tree prior was set to the birth–death process. A fossil Vanessa is known from the late Eocene, from ca 34 Ma (Miller & Brown, 1989), and it has been used previously as a minimum age constraint for the split between the lineages leading to extant Vanessa and Hypanartia (Wahlberg, 2006; Wahlberg et al., 2009a). Based on the results of the latter two studies, the age of the root (the first split between extant species of Vanessa) was constrained to be 28 million years (with a standard deviation of 3 million years). All other priors were left to the defaults in beast. Parameters were estimated using three independent runs of 10 million generations each (with a pre-run burn-in of 100 000 generations), with parameters sampled every 1000 generations. Convergence was checked in tracer v1.4.6 and summary trees were generated using treeannotator v1.5.3, which are both part of the beast package.
The evolution of vagility was investigated by dividing the species into those known to be highly vagile (records from areas very distant to the normal range of a species, e.g. oceanic islands) and those that have not demonstrated vagility. Highly vagile species are Vanessa carye (Hübner, 1812) (range South America, found on Easter Island and the Tuamotu Archipelago), Vanessa itea (Fabricius, 1775) (range Australia and New Zealand, found on Norfolk and Rapa islands), V. cardui (cosmopolitan, found occasionally on Iceland and Svalbard), Vanessa virginiensis (Drury, 1773) (North America, found on Caribbean, Azores, Madeira and Canary islands), V. atalanta (Holarctic, found occasionally on Iceland) and V. indica (Asia, found on Okinawa and Luzon). All other species are not known to be particularly vagile. The occurrence of high vagility was mapped onto the phylogenetic hypothesis using parsimony.
A dispersal vicariance analysis (diva; Ronquist, 1997) was performed to determine the most likely ancestral state reconstruction of distributions. We used the topology found in the beast analysis in this exercise. The program diva assigns a cost of one for dispersal and extinction, and a cost of zero for vicariance and within-area speciation. The least-cost ancestral reconstruction is derived based on this cost matrix (Ronquist, 1997). The distributions of all Vanessa species were defined according to the major biomes of the world, i.e. Neotropical, Nearctic, Palaearctic, Afrotropical, Oriental and Australasian. We analysed the dataset unconstrained by the maximum number of ancestral areas, as well as by constraining the maximum ancestral areas from six to two using the ‘maxareas’ option, with the latter being the maximum number of biomes inhabited by an extant species (other than the cosmopolitan V. cardui).
Combined analyses of the multigene dataset found similar topologies for species of Vanessa, differing only at one node (for Bayesian results, see Fig. 1). There was very little conflict between the different gene regions according to the PBS and PCI analyses (Fig. 1). Vanessa was found to be paraphyletic with regard to A. dimorphica + A. hippomene. The clade including these two and all Vanessa was strongly supported. However, the position of A. dimorphica + A. hippomene differed between the parsimony and Bayesian analyses, with parsimony placing the clade as sister to Vanessa gonerilla (Fabricius, 1775) + V. itea, whereas in the Bayesian analysis the clade is sister to the rest of Vanessa, except V. carye + Vanessa annabella (Field, 1971) and V. gonerilla + V. itea (Fig. 1). The V. atalanta species group, including V. abyssinica, is a well-supported monophyletic group, as is the V. cardui-species group (without the previously erroneously included V. carye and V. annabella). Within the V. atalanta species group, the Hawaiian endemic V. tameamea is the sister species to V. atalanta with good support. The position of V. vulcania, often considered a subspecies of V. indica, is unresolved with regard to the rest of the V. indica species complex. All species for which more than one specimen was sampled are monophyletic (Fig. 1).
Estimated times of divergence, given the calibration point used, suggest that the major lineages diverged from each other during the late Oligocene, and that the extant species have diverged from their sister species/groups mainly during the late Miocene (Fig. 2). The divergence of V. tameamea and V. atalanta from each other is estimated to have happened about 8 Ma, which is much earlier than the emergence of the current high Hawaiian islands (Kauai is approximately 5 million years old), but present-day atolls, like Gardner Pinnacle, would have been large islands, over 1000 m a.s.l., at that time (Price & Clague, 2002). The late Miocene is also the estimated time of divergence for the species pair V. itea + V.gonerilla, as well as V. cardui + Vanessa kershawi (M’Coy, 1868). In contrast, the V. indica species complex appears to have diversified in the Pliocene, making these island endemics much younger than the islands upon which they are found.
Unconstrained analysis in diva gives highly ambiguous results, with widespread ancestors found at all deeper nodes (Fig. 2). Constraining maximum ancestral areas to two suggests that the common ancestor of Vanessa was found in South America and Africa or North America. Africa is found to be the ancestral area of the basal nodes, with colonizations of Australia/New Zealand (V. itea species group and V. cardui species group), South America (V. cardui species group) and the Palaearctic (V. atalanta species group) from Africa. The Oriental region appears to have been colonized from the Palaearctic (the V. indica species complex). Based on this pattern, North America or the Palaearctic may have been the source for V. tameamea.
The most parsimonious reconstruction of vagility implies that highly vagile species have evolved six times independently (Fig. 2), with the sister species of highly vagile species being relatively sedentary in all cases. An alternative, nonparsimonious reconstruction suggested by the diva analysis would be that nonvagile species evolved from vagile ancestors 12 times.
The phylogenetic relationships between Vanessa species appear to be very robust, with very little conflict between the nine gene regions sequenced for this study. This is in stark contrast to a closely related genus Polygonia, where strong conflict between mitochondrial and nuclear genes was discovered (Weingartner et al., 2006; Wahlberg et al., 2009b). The only unstable and poorly supported relationship was that of the A. hippomene + A. dimorphica clade, which was either sister to the V. itea species group in the parsimony and beast analyses (Fig. 2), or was sister to the V. cardui + V. atalanta species groups in the mrbayes analysis (Fig. 1). This appears to be the result of relatively rapid divergence of the three lineages; thus, despite sequence data from nine gene regions, we do not have enough information to definitively resolve their relationships.
The placement of A. hippomene and A. dimorphica within Vanessa was somewhat of a surprise, but appears to be strongly supported (the monophyly of Vanessa receives strong support with no conflict between gene regions). We thus move them to the genus as Vanessa hippomenecomb.n. and Vanessa dimorphicacomb.n. We note that even if future studies show that this species pair is the sister to the rest of Vanessa, the name Antanartia could not be applied to them, as the type species of the genus is A. delius, and Antanartia s.s. does not appear to be closely related to Vanessa (Wahlberg & Nylin, 2003; Wahlberg et al., 2005; Wahlberg, 2006). The phylogenetic placement of the unsampled A. borbonica is still unresolved. According to Howarth (1966), A. borbonica shares a barbed phallus in male genitalia and a missing signum on the corpus bursa of the female with A. delius and A. schaeneia, whereas V. hippomene, V. dimorphica and V. abyssinica share an unbarbed phallus in the males and a well-developed signum in the females. Interestingly, the features of the latter three species are shared with species of Vanessa in general. This suggests that A. borbonica may belong in Antanartia, making these morphological characters phylogenetically useful in distinguishing the genera.
The 22 recognized species of Vanessa can be placed in five well-supported species groups (Fig. 1): the New World V. carye species group (V. carye and V. annabella), the African V. hippomene species group (V. hippomene and V. dimorphica), the Australasian V. itea species group (V. itea and V. gonerilla), and the cosmopolitan V. cardui and V. atalanta species groups (including V. abyssinica). The position of the V. carye species group as sister to the rest of Vanessa is not strongly supported in the parsimony analysis, but is stable across the methods of analyses used here. Both the parsimony and the beast analyses place the African V. hippomene species group as sister to the Australasian V. itea species group, but this is not strongly supported, and may result from long-branch attraction. Within the V. cardui species group, the Australian endemic V. kershawi is sister to the cosmopolitan V. cardui, and these two are sister to the rest of the V. cardui species group, the species of which are found mainly in South America (one species, V. virginiensis, is also found in North America). Relationships in the V. atalanta species group are also well supported, with V. abyssinica being sister to the rest. Within the V. atalanta species group, the position of the island endemic V. vulcania is not clear within the V. indica complex, but it does appear to be a separate lineage, although it was long considered a subspecies of V. indica (Wiemers, 1995).
Most species had multiple specimens collected from different parts of the world. In all cases, intraspecific COI haplotypes clearly form clades, to the exclusion of other species. This is particularly interesting for V. kershawi and V. cardui, as the two have sometimes been considered to be the same species. Vanessa cardui haplotypes from across its global range are very similar to each other, but are quite divergent from the V. kershawi haplotypes, providing good evidence that the two taxa represent cladistically distinct species. Among the Indonesian endemics, Vanessa dejeanii (Godart, 1823), Vanessa samani (Hagen, 1895), Vanessa buana (Fruhstorfer, 1898) and Vanessa dilecta (Hanafusa, 1992), the haplotypes examined are very close to each other, but still form species-specific monophyletic groups. V. buana and V. dilecta have almost identical COI haplotypes, suggesting that they are perhaps one species with allopatric distributions on Sulawesi and West Timor. In contrast our specimen of V. dejeanii was collected on Bali, whereas the specimen from Otaki et al. (2006) is from Java, and thus the genetic distance between the two suggests that they have diverged from each other. However, our sampling is too sparse to make any strong conclusions, especially based on just COI data (e.g. Wahlberg et al., 2003; Elias et al., 2007).
The resolution of most of the species clades is probably related to the age of most species in Vanessa. Almost all sister species have diverged from each other in the Miocene (over 5 Ma), with the exceptions being V. samani, V. dejeanii and V. buana (Fig. 2). This is again in contrast to the closely related genus Polygonia, where many sister species have diverged in the Pliocene (2–4 Ma) (Wahlberg et al., 2009b). Of note here is that the times of divergence are based on the same fossil calibrations from a previous study on the whole subfamily (Wahlberg, 2006).
The phylogenetic relationships and calibration data do reveal some interesting species-level patterns in Vanessa that are broadly relevant. Although the number of species in Vanessa is too small for compelling statistical analysis, the genus does speak to the themes of lineage age, diversification and distribution. The relationships between widespread and restricted species is surprising: the three most widespread species in Vanessa, V. cardui, V. virginiensis and V. atalanta, all have much more restricted sister taxa, V. kershawi, Vanessa altissima (Rosenberg & Talbot, 1914) and V. tameamea, respectively. In addition, the widespread V. indica is closely related to a number of narrow endemics, i.e. V. vulcania, V. samani, V. dejeanii, V. buana and V. dilecta. Furthermore, V. itea (widespread in the Australasian region) is sister to and partially sympatric with the New Zealand endemic V. gonerilla, whereas the widespread V. carye (South America and some Eastern Pacific islands) is sister to the western North American V. annabella. Finally, V. dimorphica is widespread across Africa and proximate archipelagos, whereas V. hippomene is restricted to South Africa and Madagascar, and is, relatively speaking, a restricted species. Yet none of these sister taxa appear to be recent divergence events, suggesting that these widespread species have maintained broad genetic continuity over relatively long time periods during which other, equally old Vanessa lineages became genetically isolated and speciated. The comparable branching patterns and age of the lineages with sedentary species suggests that they have survived for equally long periods as vagile species, although in relatively restricted ranges, and that the vagile lineages are not constantly generating sedentary sister taxa. Assuming all sedentary taxa were not once much more widespread, we can infer that sedentary habits do not confer a predilection for extinction in Vanessa. If this were the case we would expect the sedentary species to be consistently younger, branching more apically, relative to the vagile species. For example, V. tameamea would be a younger, recently derived lineage, rather than one that has apparently resided in the Hawaiian islands for approximately 8 million years after diverging from a common ancestor with V. atalanta.
How or why particular species of Vanessa have evolved extremely vagile characteristics, while their sister taxa became insular endemics, may be important in understanding the broader dynamics of speciation and dispersal (Kodandaramaiah, 2009). Vagility is not a synapomorphy, nor does vagility in one species inhibit speciation in sister taxa. Were this the case we would expect sedentary species to have sedentary sister taxa. Regardless of which sister species became more or less vagile, the consistent divergence in this character between sister species is remarkable across Vanessa. This suggests that speciation may have occurred between sister taxa in the face of gene flow from the vagile species, and that factors beyond allopatry are important in speciation.
Our geographically broad sampling allows us to make some comparative inferences about the relative vagility of even widespread species. The low levels of genetic substructuring within taxa like V. atalanta and V. cardui across continents suggests that these species are the most mobile in the genus. Other taxa, like V. virginiensis, although apparently widespread, exhibit much greater genetic structuring, implying dispersal but not consistent genetic exchange and panmictic populations, as with V. atalanta and V. cardui. In fact, the Dominican Republic V. virginiensis sample suggests near species-level divergence, and bears further examination. Thus vagility appears to trump time in the context of lineage divergence, as the older V. atalanta has less geographic structure than the younger V. virginiensis.
The historical biogeography of the genus is highly ambiguous, yet the common theme from the deeper nodes is of a widespread ancestral distribution and a series of vicariance or dispersal (in the case of several insular species) events leading to the distributions of extant species. None of the biomes can be singled out as a centre of endemism. Also, despite the importance of vicariance in explaining some of the extant distributions, the unconstrained analysis still requires 13 dispersal events. However, the unconstrained analysis suggests that Hawaii was part of the distribution of the ancestral Vanessa + Hypanartia, which is not possible as the islands did not exist 40–50 Ma. Constraining the ancestral distributions to five or fewer areas results in Hawaii being colonized by the ancestor of V. atalanta + V. tameamea, and the ancestral distribution of the V. atalanta group excluding V. abyssinica being the Palaearctic region. This suggests that Hawaii was colonized from the eastern Palaearctic region, although North America as the source cannot be excluded, as the results suggest that North America was colonized at the same time. It is worth noting that the only other endemic Hawaiian butterfly, Udara blackburni (Tuely, 1878) (Lycaenidae) is Asian in origin (Edwards et al., 2001).
Price & Clague (2002) suggest that much of Hawaii's fauna arrived in the past 5 million years, coinciding with the emergence of the existing group of high islands. However, calibrations of endemic insect lineages (Jordan et al., 2003; Bonacum et al., 2005; Rubinoff & Schmitz, 2010) indicate much older arrivals, pre-dating our estimate for Vanessa in the archipelago. Of course, our estimate of the age of the V. tameamea (8 million years) simply tells us that the lineage diverged from the extant V. atalanta lineage then, but not whether this divergence was the result of colonization of Hawaii, or if a more recent ancestor colonized the islands at a later date, and subsequently became extinct in its mainland location. The other inferred island colonizations (viz. the V. indica species complex) are all younger than the estimated ages of the islands themselves. Nevertheless, these restricted, insular species appear to have persisted for millions of years, suggesting that a lack of vagility may not condemn such lineages to relatively rapid extinction when compared with widespread, vagile sister species.
A currently intractable mystery is the absence of the cosmopolitan species from the regions supporting insular taxa. There is every indication that these cosmopolitan species could disperse to many of the islands, yet they either do not disperse or cannot survive once they have arrived. It is possible that the occasional cosmopolitan species does reach the range of an insular species, and then becomes subsumed in a much more abundant resident species, leaving little trace. However, this model suggests a lack of reproductive barriers between species, and might not fully explain the conspicuous absence of sympatric cosmopolitan taxa.
Our biogeographic analysis of the global genus Vanessa suggests periods of rapid expansion and patterns of inconsistent speciation, in which some lineages diverged while, in other regions, geographically far more disparate populations maintained genetic contact and conspecific status. A deeper understanding of population-level phenomena in Vanessa may be important in unravelling the constraints on speciation in some lineages, whereas sister taxa diverge into insular endemism (Kodandaramaiah, 2009). Unfortunately, because these processes have apparently been ongoing for the past 30 million years, ecological or geographic conditions essential to the pattern we see today may be long gone, although historically crucial in the evolution of the genus.
We are grateful to the African Butterfly Research Institute and Steve Collins, Andrew Brower, Tim Davenport, André Freitas, Zdenek Fric, Lauri Kaila, Jaakko Kullberg, Constantí Stefanescu, Andrew Warren and Keith Willmott for providing the specimens used in this study. We thank Johannes J. Le Roux and Franziska Bauer for assistance in the lab. Both Bayesian analyses were run on the CBSU BioHPC facility at Cornell University, U.S.A. We also thank Ullasa Kodandaramaiah, Jadranka Rota and an anonymous referee for stimulating discussion and comments on the article. The study was funded by the Swedish Research Council and the Academy of Finland (grant number 118369) to N.W. and by Hawaii Agricultural Experiment Station Hatch Funds (HAW00956-H) to D.R.