Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Swallowtail butterflies (Papilionidae) are a diverse and widespread group of insects that constitute a popular model system for ecological and evolutionary studies. We reconstruct the historical biogeography of Papilionidae to identify the dispersal or vicariance events that best explain their present-day distribution, and test several proposed biogeographical hypotheses about the processes that shape distribution patterns in cosmopolitan groups.
World-wide, with disjunct elements.
The phylogenetic relationships of 203 swallowtail species were determined by Bayesian inference using DNA data from mitochondrial (COI and COII) and nuclear (EF-1α) genes. Divergence time estimates were inferred using Bayesian relaxed clock approaches. To investigate competing biogeographical hypotheses, geographical range evolution was reconstructed using recently developed approaches: (1) a Bayesian empirical approach to dispersal–vicariance analysis that takes phylogenetic uncertainty into account, and (2) a likelihood approach implementing the dispersal–extinction–cladogenesis model that uses time-dependent stratified palaeogeographical matrices.
Our biogeographical results are congruent regardless of the biogeographical approaches or dating estimates used and support the importance of dispersal events in shaping swallowtail distributions. Contrary to common observations for other groups, the origins and diversification of northern taxa are better explained by range expansion through the Bering land bridge than by the Thulean or De Geer routes. We also stress that the seemingly Gondwanan biogeographical pattern in the Southern Hemisphere is more likely to have resulted from multiple, independent, long-distance dispersals than old vicariance events. The role of alternative colonization routes is also demonstrated for Madagascar, which facilitated multiple stepping-stone colonizations from India or Southeast Asia to Africa, and also for South America via the Caribbean land bridge.
Overall, the present geographical distributions of swallowtails can be better explained by dispersal events than by the long-held view of ancient vicariance events. This biogeographical study represents one of the most comprehensive phylogenetic and biogeographical studies on swallowtails. This work highlights the importance of using novel methodological approaches that provide the robust statistical frameworks needed to distinguish between competing biogeographical hypotheses. We emphasize the value of extensive taxonomic coverage for assessing the direction and frequency of supposedly rare events such as the multiple independent colonizations of Madagascar.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The historical biogeography of the Northern and Southern hemispheres has long been of interest to biologists (Lomolino et al., 2010). Since its inception, continental drift theory has provided exemplar systems in support of vicariance biogeography (Nelson & Platnick, 1981; Morrone & Crisci, 1995). However, the importance of vicariance in biogeography has been reconsidered in the last decade, as evidence for dispersal has accumulated from increasing numbers of molecular phylogenetic studies (e.g. Nothofagus spp.; Cook & Crisp, 2005). Indeed, numerous studies have revealed that vicariance is not always the dominant factor explaining patterns of distribution in widespread groups of organisms (e.g. Sanmartín et al., 2001; Sanmartín & Ronquist, 2004; Warren et al., 2010). Consequently, historical biogeography now tends to systematically investigate the relative roles of dispersal and vicariance events to better understand global patterns of biodiversity distribution by the integration of molecular dating approaches (Ree & Smith, 2008; Lomolino et al., 2010).
To explain the biogeographical history of the Northern Hemisphere, two major colonization routes have been hypothesized: (1) the North Atlantic land bridge (NALB) and (2) the Bering land bridge (BLB) (Tiffney, 1985; Milne, 2006; Fig. 1). The NALB is thought to have been used most during the early Cenozoic, connecting the biotas of North America and southern Europe via Greenland by a northern De Geer route, or by a southern Thulean route via Iceland and Great Britain (Tiffney, 1985; Fig. 1). The southernmost connection was severed during the early Eocene, approximately 50 million years ago (Ma), when the climate was mostly subtropical at these latitudes (Sluijs et al., 2006). The northernmost connection persisted until 40 Ma (Milne, 2006). The connection between Asia and North America via the BLB remained throughout most of the Cenozoic and was only severed at approximately 5.5–4.8 Ma (Marincovich & Gladenkov, 1999; Milne, 2006; Fig. 1). It is believed that the BLB was most important for inter-continental colonizations after the NALB disappeared (Tiffney, 1985; Tiffney & Manchester, 2001). The BLB reappeared during the Quaternary glacial epochs, when migration across the BLB was only possible for taxa of arctic or boreal affinities (Milne, 2006).
The Southern Hemisphere provides a prime example of a vicariance scenario, with disjunct trans-Pacific or trans-Indian distributions resulting from the sequential breakup of the southern supercontinent Gondwana starting 180 Ma (Lomolino et al., 2010). Biogeographical consequences of southern plate tectonics are manifold, most of them stemming from the effect of plate movements on the configuration of landmasses and marine basins, which then determine degrees of biotic isolation and opportunities for biotic exchange (Lomolino et al., 2010; Fig. 1) or more indirectly influence regional and global climates (Lomolino et al., 2010; Fig. 1). Thus, numerous dispersal routes have been revealed between continents (Sanmartín & Ronquist, 2004; Fig. 1). Several landmasses separated about 130–100 Ma, with South America drifting westwards from Africa as the Atlantic Ocean opened, and the Madagascar block breaking off from India (Blakey, 2008). Ancient groups generally harbour a disjunct biogeographical distribution caused by the creation of oceanic barriers to biotic dispersal (Lomolino et al., 2010). The continent of Australia–New Guinea began to gradually separate from Antarctica 80 Ma and actively moved north (55 Ma), but retained some connection with the remainder of Gondwana for about 10 million years (Beu et al., 1997; Sanmartín & Ronquist, 2004). After being separated from Madagascar (90 Ma), the Indian Plate collided with Asia about 45 Ma, forming the Himalayas and modifying climate as well as plate motion in Southeast Asia (Hall, 2002; Blakey, 2008; Fig. 1). At the same time, the southernmost part of Australia finally separated from Antarctica, closing the direct route to South America. During the Oligocene, South America finally separated from west Antarctica (Blakey, 2008). Afterwards, South America and Antarctica were not directly connected, but the multiple microplates of the Antarctic Peninsula remained near southern South America, acting as stepping stones and allowing biological interchange by the Drake Passage until its complete opening 23 Ma (Beu et al., 1997; Fig. 1). About 20 Ma, the collision between the Australian Plate and the south-western part of the Pacific Plate created new routes for biotic exchange with Southeast Asia (Hall, 2002; Fig. 1), also causing major climate changes that drastically modified weather patterns in Australasia. Finally, South America was connected to North America via the Isthmus of Panama, thereby allowing the Great American Faunal Interchange (3.5 Ma; Lomolino et al., 2010; Fig. 1).
Based on their different geological histories, several major differences are evident in biogeographical events between the Northern and Southern hemispheres. In the Northern Hemisphere, complex biogeographical histories involving old vicariance events or recent dispersals are generally recovered for plants and terrestrial animals (Sanmartín et al., 2001). By contrast, in the Southern Hemisphere, animal distributions are more congruent with the fragmentation of Gondwana whereas plant distributions are generally better explained by dispersal events (Sanmartín & Ronquist, 2004).
It is reasonable to expect that the existing distribution of biodiversity has been deeply influenced by the complex geological history of both hemispheres. To better understand the precise contribution of past geological events in shaping global biodiversity patterns, inference of the historical biogeography of cosmopolitan groups is essential. In this study, we use swallowtail butterflies (Lepidoptera, Papilionidae) as a model system to assess the role played by vicariance or colonization routes in shaping distribution patterns. Swallowtails are particularly suitable for testing the contribution of these processes because several groups exhibit disjunct distribution patterns in both hemispheres (Tyler et al., 1994; Scriber et al., 1995). Overall, the family comprises 32 genera and about 550 described species (Häuser et al., 2005), which are usually classified into three subfamilies: Papilioninae, Parnassiinae and Baroniinae (Simonsen et al., 2011). The subfamily Papilioninae is the largest with more than 485 species world-wide, reaching its greatest diversity in the tropics (Wallace, 1865; Zakharov et al., 2004). This subfamily harbours a disjunct distribution and is traditionally considered to exemplify vicariance (plate tectonic breakup of Gondwana), particularly within the Troidini (Braby et al., 2005) and the genus Papilio (Zakharov et al., 2004). The subfamily Parnassiinae is mainly distributed in the Palaearctic, with 67 species mostly occurring in central Asia and some specialized groups in high-elevation habitats (Weiss, 1991; Nazari et al., 2007; Michel et al., 2008). Several Parnassius species also occur in the Western Nearctic, a distribution considered to result from recent dispersal events (Michel et al., 2008; Todisco et al., 2012). Finally, the sole member of the subfamily Baroniinae, Baronia brevicornis, is endemic to Mexico and is considered a relict taxon (Tyler et al., 1994; Scriber et al., 1995).
To advance our understanding of the biogeographical history of a group, a robust and well-resolved phylogenetic tree is required (Ree & Smith, 2008). Taxon sampling is thus fundamental, as sparse taxon sampling may bias phylogenetic relationships as well as influencing the reliability of any estimates of ancestral areas of distribution (Nylander et al., 2008). Phylogeny should also be well connected with a temporal framework to allow relevant comparisons with known past geological events (e.g. changes in sea level, continental drift, orogenesis; Wiens & Donoghue, 2004; Lomolino et al., 2010). Although earlier studies have suggested an old origin – Late Jurassic – for swallowtails (e.g. Zeuner, 1943; Miller & Miller, 1997; Braby et al., 2005), most DNA-based studies have inferred a more recent origin between the Late Cretaceous and the early Cenozoic for various swallowtail groups (e.g. Zakharov et al., 2004; Nazari et al., 2007) as well as for the whole family (Simonsen et al., 2011; Condamine et al., 2012a). The more recent origin for swallowtails is not consistent with the hypothesis of a Gondwanan origin (Parsons, 1996a,b; Miller & Miller, 1997; Braby et al., 2005). If the diversification of swallowtails occurred during the Cenozoic, dispersal may be the main biogeographical process responsible for their present distribution. In this context it is essential to identify the different dispersal routes that the most recent common ancestors (MRCAs) of extant papilionid species may have used to colonize all continents in both hemispheres. Using comprehensive taxon sampling and two distinct biogeographical approaches, we specifically focus on assessing the timing and nature of these biogeographical events to investigate the use of known dispersal routes in the Northern and Southern hemispheres (Sanmartín et al., 2001; Sanmartín & Ronquist, 2004).
Materials and Methods
Taxon sampling and molecular data set
We used the data set of Condamine et al. (2012a), which includes a total of 203 swallowtail species, representing over one-third of the species in the family, sequenced for about 2.3 kilobases (kb) of two mitochondrial coding genes [cytochrome c oxidase subunit I (COI) and cytochrome c oxidase subunit II (COII)] and about 1.0 kb of the nuclear coding gene elongation factor-1α (EF-1α) (see Appendix S1 in Supporting Information). The taxon sampling comprises all described genera and subgenera (sensu Häuser et al., 2005), including: (1) the unique member of the Baroniinae, Baronia brevicornis; (2) 65 Parnassiinae species out of 68; and (3) 137 Papilioninae species out of 485. Voucher material and GenBank accession numbers are given in Appendix S1. Seven outgroups were chosen: three Pieridae species (Colias eurytheme, Pieris napi and Pieris rapae) and three Nymphalidae species (Coenonympha tullia, Libythea celtis and Vanessa cardui) that are closely related to Papilionidae (Wahlberg et al., 2009; Simonsen et al., 2011). Pyrgus communis (Hesperiidae) was used to root the tree.
Bayesian phylogenetic analyses
Phylogenetic analyses were carried out using Bayesian inference (BI), a probabilistic method routinely used to conduct analyses of multigene data sets (Felsenstein, 2004; Edwards, 2009). For each gene, the best-fit model of sequence evolution was selected using jModelTest (Posada, 2008) based on the Bayesian information criterion (Brown & Lemmon, 2007). The data set was further combined and partitioned according to five different strategies: (1) a single data partition; (2) one partition for the mitochondrial genes and one for the nuclear gene; (3) one partition per gene; (4) one partition per codon position for mitochondrial genes and one per codon position for the nuclear gene; and (5) one partition per codon position for each gene.
Bayesian inference analyses were carried out with MrBayes 3.1.2 (Ronquist & Huelsenbeck, 2003) using the following settings: two independent runs with eight Markov chains, one cold and seven incrementally heated, that ran for 10 × 106 generations, sampling the trees every 100 cycles. A conservative burn-in of 25% was applied after checking for stability on the log-likelihood curves and the split frequencies of the runs. All trees sampled prior to this threshold were discarded, with the remaining trees being used to generate a 50% majority rule consensus tree. Posterior probability (PP) values ≥ 0.95 were considered to indicate strong support for a clade (Felsenstein, 2004).
Two separate analyses were then conducted for each partitioning strategy. Alternative partitioning strategies were compared using Bayes factors (Brown & Lemmon, 2007). Estimates of the harmonic mean of the likelihood values, obtained with the sump command in MrBayes (Ronquist & Huelsenbeck, 2003), were used to approximate the Bayes factors (Brown & Lemmon, 2007). Bayes factor values > 10 were considered to be indicative of strong support for one hypothesis over another (Brown & Lemmon, 2007).
Fossil calibrations for dating
Five fossil calibrations were included as minimum age constraints. The oldest swallowtail fossils are from the early Eocene (48 Ma; Green River formation in Colorado), and consist of two species in the genus Praepapilio (Simonsen et al., 2011). The fact that Praepapilio species do not share all the specific apomorphic characters that characterize the Baroniinae, Parnassiinae or Papilioninae indicates that these fossils may only represent a minimum age for Papilionidae (see Ho & Phillips, 2009; for the rationale of this constraint). Therefore we calibrated the Papilionidae crown group with a minimum age of 48 Ma. The two other unequivocal fossils of Papilionidae are Thaites ruminiana from the early Oligocene (30 Ma), and Doritites bosniackii from the late Miocene (15 Ma). In recent morphological cladistic analyses, Thaites was sister to Luehdorfiini or Zerynthiini, while Doritites was the sister to Archon (Luehdorfiini; Nazari et al., 2007). We have conservatively chosen to use 30 Ma as the minimum age for the Parnassiinae crown group whereas the crown of Luehdorfiini was confined to a minimum age of 15 Ma.
External fossil calibrations were also used for outgroup nodes with two other non-Papilionidae fossils. We calibrated the Pieridae crown group with a minimum age of 34 Ma (oldest fossil of Pieridae, Stolopsyche libytheoides, from the late Eocene in the Florissant Formation). Nymphalidae were well diversified in the late Eocene (e.g. Doxocopa wilmattae, Jupitellia charon, Libythea florissanti or Nymphalites obscurum from the Florissant Formation) and we consequently used a minimum age of 34 Ma for the crown age of this family following Wahlberg et al. (2009). Finally, a specific constraint was set for the root, consisting of a maximum age of 183 Ma, based on the assumption that all but the earliest clades of the order Lepidoptera diversified in association with angiosperm plants. Papilionoidea should thus be younger than the origin of angiosperms about 183 Ma (Bell et al., 2010).
Bayesian estimates of divergence times
We have preferentially employed a Bayesian relaxed clock approach that accounts for rate variation across lineages (Drummond et al., 2006) and uses Markov chain Monte Carlo (MCMC) procedures to approximate the posterior distribution of rates and divergence times and simultaneously infer their credibility intervals. The Bayesian relaxed clock approach implements flexible techniques for taking calibration uncertainties into account (Yang & Rannala, 2006; Ho & Phillips, 2009). In particular, it allows the use of ‘soft’ bounds, where the probability that the true divergence time lies outside the bounds is small but non-zero, and ‘hard’ bounds that have an equally probable distribution (i.e. uniform distribution) of ages between the bounds (Yang & Rannala, 2006). To take into account the uncertainty level of the fossil ages, all calibrations were treated as ‘soft’ bounds using a lognormal distribution (Yang & Rannala, 2006; Ho & Phillips, 2009). For comparison purposes, we also performed analyses with ‘hard’ bounds using a uniform distribution to cross-validate our results.
In our study, the Bayesian relaxed clock analyses were carried out with the program beast 1.6.2 (Drummond & Rambaut, 2007), using the lognormal model of Thorne et al. (1998), which is uncorrelated in beast. The XML file for the beast analysis was created using BEAUti (in the beast package) with the following non-default settings and priors: the site model was set to the GTR + Γ distribution with default parameters, the clock model was set to a relaxed clock with uncorrelated rates, the tree model was set to a Yule process of speciation, and the MCMC parameters were set to 5 × 107 generations with sampling every 1000 generations and the first 25% discarded as burn-in. To minimize the size of the parameter space, we enforced the topology obtained from MrBayes in all beast analyses. The remaining parameters were based on the model used in the original MrBayes analysis (GTR + Γ). A lognormal distribution for each fossil constraint, with the 95% interval bounded by the minimum age of the fossil (2.5% quartile) and a maximum age (97.5% quartile) of 183 Ma, corresponding to the origin of the angiosperms (Bell et al., 2010). The lower limit of the prior distribution for the root time (‘root height’) was set to 183 Ma.
Historical biogeography: parsimony-based versus likelihood-based approaches
In our study, several considerations were taken into account to define the palaeogeological model. First, the Earth was divided into six classical biogeographical regions (Lomolino et al., 2010). Second, these regions were further split into smaller biogeographical units to maximize congruence with previous biogeographical studies, which were delineated using: (1) palaeogeological and palaeoclimatic criteria (e.g. Raven & Axelrod, 1974; Meijer & Wortel, 1999; Hall, 2002; Blakey, 2008), (2) the presence of one (or more) extant biodiversity hotspots within a biogeographical region (Mittermeier et al., 2004), and (3) the known distribution of extant swallowtail species (e.g. Weiss, 1991; Tyler et al., 1994; Parsons, 1996b). The resulting palaeogeological model comprises the following 11 areas: A, West Palaearctic (WP; from western Europe to the Ural Mountains, excluding North Africa); B, East Palaearctic (EP; from the Urals to Japan); C, West Nearctic (WN; from Alaska to the eastern Rocky Mountains); D, East Nearctic (EN; from the eastern Rocky Mountains to the Atlantic coast); E, Afrotropics (the entire African Plate); F, the Madagascan region (Madagascar and surrounding archipelagos); G, Amazonian Neotropics (South America); H, the Mexican region (Central America and the Caribbean islands); I, India (Indian Plate and Iranian and Tibetan plateaus); J, Southeast Asia (Malaya and the Sunda Shelf); and K, Australasia (Sahul + southwest Pacific and Melanesian islands). Species distributions were coded as presence or absence in each area and marginal distributions or human introduction events were excluded (Hines, 2008; Nylander et al., 2008).
Dispersal–vicariance analysis (DIVA)
DIVA is a parsimony-based method that optimizes the mapping of ancestral areas onto phylogenetic nodes by minimizing the number of dispersal and extinction events using a three-dimensional cost matrix (Ronquist, 1997). Although widely used, this method has some known limitations (Kodandaramaiah, 2010), but to address these concerns, the DIVA approach was recently refined and upgraded (Nylander et al., 2008). Uncertainty in phylogenetic relationships and ambiguities in DIVA optimizations can now be accounted for by the Bayes-DIVA approach of Nylander et al. (2008), in which DIVA analyses are carried out using random trees from post burn-in samples.
The program rasp 1.1 (Yu et al., 2011) was used to perform DIVA analyses because it implements new refinements of Bayes-DIVA. All analyses were based on the post burn-in tree samples (with 10,000 randomly selected trees among the 75,000 post burn-in trees). We performed Bayes-DIVA analyses using rasp, with distribution parameters set to three maximum areas (since at least one species harbours a wider area of distribution that encompasses four areas) and the root distribution fixed to null (i.e. root not fixed).
This approach describes ancestor–descendant transitions between geographical ranges by processes of dispersal (range expansion), local extinction (range contraction) and cladogenesis (range subdivision/inheritance) and estimates the likelihood of species range data arrayed at the tips of a phylogenetic tree as a function of the rates of dispersal and local extinction (Ree & Smith, 2008). One of the major advantages of the DEC method is the fact that it is possible to generate stratified models, in which distinct biogeographical models may be used for specific time slices of interest. Using likelihood algorithms, this method estimates global rates of dispersal and extinction given all possible ancestral states at internal nodes, and then uses those rates to estimate node-by-node relative probabilities (RP) for ancestral areas and range inheritance scenarios without conditioning on any other range inheritance in the remaining nodes (Ree & Smith, 2008).
To define the stratified biogeographical model, we used the 11 defined areas, which corresponds to a set of 211 = 2048 theoretically possible geographical ranges (area subsets). Then, we imposed some geographical structure to the model by constraining allowed states in the matrix to combinations of three areas (plus those species ranges larger than three areas that were present in the terminals). Many of the 2048 possible ranges were also excluded from consideration based on the biological implausibility of their spatial configurations (Ree & Smith, 2008). Third, following the principles described in Ree & Smith (2008), we added temporal constraints on the dispersal rates between areas, based on palaeogeographical reconstructions of area position through time (e.g. Raven & Axelrod, 1974; Meijer & Wortel, 1999; Hall, 2002; Blakey, 2008). Specific constraints on dispersal rates were set for a series of six time slices (of 10 million years each). For each time slice, we constructed a matrix of scaling factors (between 0.0 and 0.5) for dispersal rates between areas according to their geographical position, interpreting greater distances and/or the extent of geographical barriers (sea straits, mountain chains) as being inversely proportional to the expected rate. In the absence of any barriers, the dispersal rate between adjacent areas was fixed to 0.5 whereas a dispersal rate of 0.25 was specified whenever a geographical barrier had to be crossed (e.g. between Africa and Madagascar). A simple statistical process was used to arbitrarily fix dispersal rates between areas separated by only one area, with the corresponding dispersal rates being the result of the product of two dispersal rates. For instance, dispersing from India to Australasia implies a passage through Southeast Asia, so the rate is equal to 0.5 (dispersal rate between India and Southeast Asia, in the absence of barriers) × 0.25 (dispersal rate between Southeast Asia and Australasia, across a sea strait) = 0.125. Finally, we allowed long-distance dispersal (for instance between Africa/Madagascar and Australasia) with a dispersal rate equal to 0.01; dispersals were disallowed between geographical areas separated by two or more areas. All DEC analyses were carried out using maximum clade credibility trees resulting from beast dating analyses that implemented hard- and soft-bound calibrations.
Specific analyses were also carried out to reconstruct the ancestral area for the root using three distinct combinations of range sizes (one to three ancestral areas). All possible combinations of ancestral areas were enforced for the root and their global likelihood scores were further compared to determine the most likely ancestral area. For all nodes (including the root), a given area was treated as significantly supported when it had a difference in likelihood score greater than or equal to two log-likelihood units (Ree & Smith, 2008).
Overall, the molecular matrix comprised 210 taxa and 3286 nucleotides (COI, COII and EF-1α). Based on Bayes factor comparisons, the best partitioning strategy for our data set had nine partitions (one per codon position for each of the three genes). Results from the best-fit partitioning strategy using MrBayes are presented in Fig. 2. Overall, we recovered strong PP (≥ 0.95) for 80% (162 out of 202) of ingroup internodes (Appendices S2 & S3). These analyses also recovered the monophyly of Papilionidae and all subfamilies and tribes with PP = 1. The monotypic genus Baronia is sister to the rest of Papilionidae, while Parnassiinae and Papilioninae are strongly supported as sister clades (Fig. 2). Tribe Papilionini (the genus Papilio) is sister to tribe Teinopalpini (Teinopalpus and Meandrusa). Troidini is sister to Papilionini + Teinopalpini, and tribe Leptocircini is sister to the rest of Papilioninae (Fig. 2). For the genus Papilio, almost all internodes are highly supported (64 out of 75 internodes with PP ≥ 0.95), and the Old World and New World subgenera are sister clades (PP = 1; Fig. 2).
Estimates of divergence times
Results of fossil-based dating analyses with hard and soft bounds are given in Appendix S2 for all internal nodes, with their median ages and 95% confidence intervals (CI). Results of analyses using hard and soft bounds were consistent, only differing by the slightly older ages that were usually recovered for analyses with hard bounds (even though the upper estimates of CI are similar for all nodes). For clarity, and because analyses with soft bounds better account for age uncertainties (Yang & Rannala, 2006), only estimates of divergence times obtained with soft bounds are presented (Fig. 2). Overall, a Cenozoic origin of swallowtails was recovered in the Eocene c. 52 Ma (CI 46–62.5 Ma) around the early Eocene Climatic Optimum. Subfamilies Parnassiinae and Papilioninae arose in the middle Eocene c. 41 Ma (CI 34–51 Ma) and in the early Eocene c. 47 Ma (CI 41–57.5 Ma), respectively. The three extant Parnassiinae tribes appeared in the Oligocene: 25 Ma for Luehdorfiini (CI 19–33 Ma), 30 Ma for Zerynthiini (CI 24–37 Ma) and 25 Ma for Parnassiini (CI 19–31 Ma) (Fig. 2). The most species-rich genus of Parnassiini, Parnassius, emerged around 17 Ma (CI 13–22 Ma). The four extant Papilioninae tribes originated around the late Eocene–early Oligocene boundary: 35 Ma for Leptocircini (CI 28–43 Ma), 38 Ma for Troidini (CI 32–47 Ma), 35 Ma for Teinopalpini (CI 27–44 Ma) and 31 Ma for Papilionini (CI 26–38 Ma), the most species-rich tribe of the family (Fig. 2).
The results of DEC analysis are presented in Fig. 3, with likelihood scores (L) for each node (with their relative probabilities, RP) in Appendix S2, and the results of Bayes-DIVA analysis presented in Appendix S3. For the root, Bayes-DIVA analyses reconstruct an origin in the EP, whereas DEC optimizations recover an optimal range area encompassing the EP and the WN regions. All other combinations of range areas (with up to three ancestral areas) were significantly less well supported. Similar ancestral areas (EP for Bayes-DIVA; EP and WN for DEC) were found using chronograms obtained with hard or soft bounds. Overall, no southern landmasses were reconstructed at deeper nodes under either method.
In spite of some differences, the two methods agree on the global biogeographical pattern. They both infer an origin for the Papilionidae in the northern region around the EP. Moreover, the results of ancestral area reconstructions unambiguously support a dynamic dispersal pattern (only a few vicariance events were reconstructed between the EP and WN) (Fig. 3). More precisely, ancestors of the Papilionidae have progressively dispersed towards Australasia or South America in the Southern Hemisphere through different landmasses in the Northern Hemisphere. All biogeographical scenarios suggest that Southeast Asia and the Mexican region have been intensely used as dispersal routes over the tropics for swallowtails, as revealed by the number of reconstructions for these areas at intermediary nodes (Fig. 3). Few reverse dispersal events are observed in our biogeographical reconstruction (e.g. from Southeast Asia or India to the EP).
Northern Hemisphere biogeography
Our divergence time estimates support an Eocene origin (about 52 Ma) for Papilionidae. This time frame does not exclude the potential role of either of the two colonization routes (i.e. NALB and BLB) that were present during the diversification of the group and could account for the present-day pattern of distribution (Sanmartín et al., 2001). However, our study provides more support for the role of the BLB in swallowtail dispersal because the biogeographical analyses unequivocally recovered the Beringian region (WN + EP) as the most likely ancestral area of origin for the Papilionidae (Fig. 3). This result is robust and does not change even when phylogenetic uncertainty and estimates of divergence times are considered. According to the classification made for the Northern Hemisphere by Sanmartín et al. (2001), papilionids may be considered a ‘medium’ time-class group, typical of the early–mid Cenozoic. Such groups originated as warm-temperate groups associated with boreotropical mixed mesophytic forests (Sanmartín et al., 2001). Fifty-two million years ago, the climate was warmer in the Beringian region than today due to the early Eocene Climatic Optimum (Sluijs et al., 2006). The dispersals of swallowtails may therefore have mostly occurred through the BLB coincidently with the development of boreotropical biota (Tiffney, 1985; Sanmartín et al., 2001; Tiffney & Manchester, 2001; Donoghue, 2008).
Pattern within the Palaearctic and Nearctic regions
Enghoff (1995) presented a general area cladogram for the Holarctic that reflects the present continental configuration: ((WN,EN),(WP,EP)). In this study, we have used the same divisions of geographical areas for two reasons: (1) to compare our results with those obtained in the meta-analysis of Sanmartín et al. (2001), who also used this area cladogram, and (2) to implement this palaeogeographical model in biogeographical analyses in order to disentangle the directionality of the colonization routes.
Enghoff (1995) proposed that dispersals between WN and EN were more frequent than those between WP and EP. He attributed this pattern to the difference in duration of the epicontinental seaways, which were presumably important dispersal barriers. The Mid-Continental Seaway divided the Nearctic from 100 to 70 Ma, while the Turgai Sea divided the Palaearctic for much longer, from 180 to 30 Ma. Based on the examination of time-calibrated phylogenies, Sanmartín et al. (2001) demonstrated that faunal exchanges have been significantly more extensive in the Palaearctic than in the Nearctic. This does not mean that the Turgai Sea was not an effective barrier, but rather that almost all of the Palaearctic exchanges occurred after the closing of the Turgai Sea in the Oligocene.
According to our results, dispersals were more intense within the Palaearctic. Interestingly, the frequency of Palaearctic dispersals also corresponds to the palaeogeographical reconstructions (Blakey, 2008). The Turgai Sea, which apparently formed an effective biogeographical barrier, closed 30 Ma and made range expansion possible for Asian groups towards Europe, leading to an increase in WP–EP distributions during the Miocene (Sanmartín et al., 2001). Within swallowtails, we have confirmed this trend, because an increase in dispersals between the two Palaearctic regions after 30 Ma is clearly demonstrated (Fig. 3).
The Nearctic pattern is more difficult to explain. Two predicted dispersal waves, first after the closing of the Mid-Continental Seaway about 70 Ma and then after the erosion of the early Rocky Mountains in the Oligocene (30 Ma), are not evident for the early–mid Cenozoic time class (Sanmartín et al., 2001). According to Sanmartín et al. (2001), neither the second uplift phase of the Cordilleran range in the late Oligocene–Miocene (25–20 Ma), nor the Quaternary glaciations were significant in preventing faunal exchange within the Nearctic, since WN–EN distributions are common in both the late Cenozoic and Quaternary (20–0 Ma). For swallowtails, we recovered eastward dispersals after 25 Ma that confirm the predictions of Sanmartín et al. (2001), as these dispersal events have not been constrained by the Cordilleran uplift (Fig. 3). The strong directional asymmetries in continental dispersals show that, during the Cenozoic and Quaternary, WN and EP acted as the main source areas of continental dispersals in the Nearctic and Palaearctic for swallowtails, in agreement with the results of previous studies (Sanmartín et al.,2001; Milne, 2006; Hines, 2008).
Bering land bridge: the main northern inter-continental Cenozoic dispersal route
Both our dating and biogeographical results suggest that the BLB was the main colonization route between the Palaearctic and Nearctic. This result is plausible because of the short duration of the Thulean route, which closed almost immediately after the origin of Papilionidae (Fig. 3). Although the Northern Hemisphere was often colonized via this route (Tiffney, 1985; Sanmartín et al., 2001; Tiffney & Manchester, 2001), the NALB was not followed by swallowtails. Neither the Thulean route nor other available pathways such as the De Geer route are recovered in our biogeographical reconstructions. The De Geer route continued to exist after the disappearance of the Thulean route but was located far to the north and was probably restricted to cold-adapted groups (Sanmartín et al., 2001).
Dispersal over the BLB was primarily controlled by the climatic and biotic conditions that prevailed through time. Sanmartín et al. (2001) recognized three periods of exchange between EP and WN. The first BLB (early–mid Cenozoic), which partly coincided with the Thulean route, was covered with mixed-mesophytic forest and was presumably dominated by dispersal of faunal elements associated with this forest (Tiffney, 1985). The second bridge (late Cenozoic) was covered by taiga forest and inhabited by associated faunal elements, whereas the third bridge (Quaternary) was dominated by tundra habitats (Sanmartín et al., 2001). Numerous studies have stated that the first BLB was relatively unused for inter-continental faunal exchange because these early Cenozoic dispersals appeared to have mostly been realized through the NALB (e.g. Marincovich & Gladenkov, 1999; Sanmartín et al., 2001; Tiffney & Manchester, 2001). Biotic exchanges were more important across the second BLB (20–3 Ma), and even more frequent during the third BLB in the Quaternary (Sanmartín et al., 2001).
In contrast to these other studies, we have demonstrated that the first BLB was an important pathway for colonizing the Northern Hemisphere by swallowtails that also promoted their diversification. Indeed, 15 million years after their origin, swallowtails diversified in the Beringian region and the major extant lineages arose (Fig. 3). The relatively high frequency of trans-Beringian distributions in the early Cenozoic is unexpected because EP and WN were only connected at high latitudes (Marincovich & Gladenkov, 1999). In addition, we found little evidence for exchanges in the second BLB except for two temperate-adapted groups in the genus Papilio. These results are congruent because the increasingly cooler climate of the Miocene is likely to have restricted the dispersal of warm-adapted groups (Sanmartín et al., 2001). Finally, the two cold-adapted groups have clearly recently used the third BLB during the Quaternary (Fig. 3).
Biogeographical implications of an intriguing distributional pattern in swallowtails
Previous publications on the phylogeny and biogeography of the genus Papilio have hypothesized that these swallowtails originated 55–65 Ma in EN–WP (Zakharov et al., 2004). The genus produced two lineages corresponding to New World and Old World clades. The diversification of the New World lineage is a complex biogeographical puzzle because its phylogenetic structure comprises not only New World species (subgenus Pterourus) but also two Old World lineages: Papilo alexanor and the subgenus Papilio (Chilasa), found respectively in WP and Southeast Asia (Fig. 4a). To explain their disjunct distribution, Zakharov et al. (2004) proposed that they appeared c. 46 Ma, at a time of favourable climatic conditions across the Thulean route. They further assumed that Papilio (Chilasa) species might have spread to Asia soon after the drying of the Turgai Sea about 30 Ma, which opened the route between Europe and Asia (Zakharov et al., 2004). Our analyses are not in agreement with this hypothesis, because diversifications of the Papilo alexanor, Chilasa and Pterourus lineages are dated around 24 Ma (20–30 Ma) and therefore conflict with the Thulean route hypothesis. Instead, our biogeographical analyses suggest that the MRCA appeared in EP + WN and used the BLB to disperse from EP to WP for the Papilo alexanor lineage, congruent with the closure of the Turgai Sea (30 Ma). The MRCA dispersed from EP to Southeast Asia for subgenus Papilio (Chilasa). The Papilio (Pterourus) clade remained isolated in the Nearctic and diversified (Fig. 4a). The apparent disjunct distribution pattern is thus attributable to BLB dispersal and a Cenozoic cooling climate that fragmented the ancestral range in the northern areas.
Southern Hemisphere biogeography
Northern Gondwana pattern
Interestingly, our results disagree with the general rule proposed by Sanmartín & Ronquist (2004) in which the present-day distribution of taxa should conform to a vicariance pattern caused by the Gondwana breakup. Nonetheless, swallowtails have long been considered an old group that originated in Gondwana because of the apparently disjunct distribution exhibited by several groups. Numerous biogeographical hypotheses have been put forward, such as an African origin and subsequent vicariance events with South America and India (Parsons, 1996a,b; Miller & Miller, 1997; Braby et al., 2005). This scenario corresponds to the well-known ‘southern Gondwana pattern’ and predicts that African taxa diverged basally, followed by a New Zealand clade and then a South America–Australia clade. Although it has been modified to include other southern landmasses, this model is widely accepted to explain distributions of other old animal groups (Sanmartín & Ronquist, 2004). However, our analyses did not support this pattern.
On the contrary, our results better correspond to the ‘northern Gondwana pattern’ (Sanmartín & Ronquist, 2004). Biogeographical relationships among the landmasses that form the south-eastern tropics (Africa, Madagascar, India, Southeast Asia, Australia and New Guinea) are typically explained as the result of recent dispersal along the coasts of the Indian Ocean. These areas share a tropical biota, and exchanges between the eastern landmasses have been frequent since the collision of the Australian and Asian plates in the Miocene (Raven & Axelrod, 1972). Long-distance dispersal over water barriers is the likeliest explanation for several groups that radiated long after the separation of Africa and Australia (Warren et al., 2010). Indian Ocean bathymetry and past sea levels support the repeated existence of sizeable islands across the western Indian Ocean, greatly reducing the isolation of Madagascar from Asia. A chain of stepping-stone islands probably acted as a land bridge between Asia/India and Madagascar–Africa during the Cenozoic. The existence of this land link is supported by the trans-Indian distribution of many terrestrial groups (Warren et al., 2010).
Moreover, our analyses reveal strong concordant dispersal patterns in the Southern Hemisphere. In other words, dispersals tend to occur between certain areas and in certain directions. Indeed, we recovered at least three independent dispersals between Asia/India and Madagascar (Fig. 3). In addition, these dispersals occurred during the early–mid Miocene, after the late Oligocene Warming Event and before the middle Miocene Climatic Optimum, which implies that sea level was relatively low. The timing of these long-distance dispersals is highly congruent with geological evidence, hereby confirming that islands were present between the two continents (Warren et al., 2010). Together, this evidence indicates that several ancestral swallowtails crossed the Indian Ocean from Asia/India to reach the Madagascan region using stepping-stone islands. It is noteworthy that this dispersal scenario corresponds to the general rule recovered for plants (Sanmartín & Ronquist, 2004). Swallowtails are highly specialized phytophagous insects (e.g. Tyler et al., 1994; Scriber et al., 1995; Braby et al., 2005; Michel et al., 2008) and thus it may be hypothesized that their biogeographical history is partially congruent with that of the plants they probably followed during their evolution (Condamine et al., 2012a).
Trans-oceanic dispersal explains the disjunct distribution of swallowtails
Within the genus Papilio, the MRCA of Old World swallowtails originated in the late Oligocene 25 Ma in EP and Southeast Asia. Two southern lineages are identified.
The first comprises Papilio (Eleppone) anactus, endemic to Australia, sister to the Afrotropical Papilio (Princeps) clade plus an Asian Papilio (Menelaides) clade 1 (Fig. 2). The position of Papilio anactus raises the question of whether Australia was colonized from Africa or Southeast Asia. Zakharov et al. (2004) assumed that the Papilio (Eleppone) lineage reached Australia through the Tasmanian land bridge around 40–47 Ma. Instead, our results suggest that the ancestral area of origin is Southeast Asia and India for Papilio (Eleppone) and Papilio (Menelaides) clade 1, and India and Madagascar for Papilio (Princeps) clade 1. They also indicate that their MRCA dispersed via Southeast Asia to Australia, which at that time drifted northwards and connected with Southeast Asia via intervening islands (Hall, 2002), and then evolved into Papilio (Eleppone) (23 Ma). This same MRCA also left a Southeast Asian lineage that corresponds to Papilio (Menelaides) clade 1 and further dispersed to the Madagascan region from India by stepping stones in the Indian Ocean to give Papilio (Princeps) clade 1 (Fig. 4b).
The second lineage of Old World swallowtails produced a diverse group of modern Papilio (Princeps) distributed throughout Southeast Asia, India and Madagascar around 21 Ma (Fig. 4b). From this ancestral region, Papilio (Princeps) clade 2 dispersed to Madagascar, constituting a possible vicariance event caused by climatic change and sea-level fluctuation (Warren et al., 2010). Subgenus Papilio (Achillides) evolved in situ 18 Ma, and further dispersed in Australasia (Fig. 4b). Papilio (Princeps) clade 3 represents a third example of a colonization event from India to Madagascar. Finally, Papilio (Menelaides) clade 2 appeared 14 Ma in Southeast Asia, and separated into two groups: one staying in the ancestral area and another dispersing southwards to Australasia.
A long-discussed aspect of the biogeography of Troidini is the distribution pattern of genera within the Australasian lineage (Parsons, 1996b; Braby et al., 2005). The Madagascan endemic genus Pharmacophagus is sister to the genus Cressida found in Australasia, and the origin of this clade is dated to 26 Ma, while both are sister to the birdwing genera (Ornithoptera, Trogonoptera and Troides) occurring in Southeast Asia and Australasia (Fig. 2). However, the present-day distribution of many organisms occurring in these regions is more likely to be the result of trans-oceanic dispersal from Asia/India to the Madagascan region (Warren et al., 2010). Our analyses suggest that the MRCA of these two genera was distributed in EP and Southeast Asia, but do not reject the hypothesis of an origin in EP, Southeast Asia and Australasia. The MRCA dispersed from Asia/India through the Indian Ocean and to Australia, evolving into Pharmacophagus and Cressida, respectively (Fig. 4c). This situation suggests the existence of past connections between Madagascar and the Australia/Oriental regions around that time. In the Cenozoic, connections between Asia/India and Madagascar through palaeo-islands situated in the Indian Ocean demonstrate the possibility of stepping-stone dispersal events in the middle of the Indian Ocean, making it easier to infer Miocene dispersals between these regions (Warren et al., 2010).
The present-day mixed distribution pattern of Papilionidae is the result of intense diversification in tropical areas, which was favoured by mainland proximity and numerous island colonization events with subsequent speciation (Wallace, 1865; Zeuner, 1943). Sea-level fluctuations between the late Miocene and the Pliocene–Pleistocene are likely to have speeded up the diversification of the Papilionidae by the isolation of populations, but also complicate the understanding of biogeographical processes in the Indo-Pacific region, especially for disentangling dispersal and vicariance events (Condamine et al., 2012b).
This study advances the understanding of the global biogeographical processes that shaped current distribution patterns. It is only through our extensive taxonomic coverage, fossil-based chronograms and a variety of biogeographical methods that we were able to reconstruct a sufficiently resolved biogeographical history of this group. Overall, the results of our study fit well with general predictions proposed for the evolution of widespread groups in the Southern and Northern hemispheres (e.g. Sanmartín et al., 2001; Sanmartín & Ronquist, 2004). However, several major discrepancies were found, such as our recovery of numerous dispersals through the BLB rather than through the NALB. We also found evidence that swallowtails originated in the Northern Hemisphere (Beringia), contradicting the hypothesis of a current distribution pattern caused by Gondwanan vicariance.
We thank E. Jousselin and F. Delsuc for their helpful comments on early presentations and drafts. Partial funding for this study was obtained through the program ANR ‘Biodiversité’ of the French National Agency for Research (project ANR BIONEOCAL) and by a grant of the graduate school of Montpellier II University (SIBAGHE). F.S. also acknowledges a Natural Sciences and Engineering Research Council of Canada, Discovery Grant.
Fabien Condamine was a PhD student at the Centre de Biologie pour la Gestion des Populations and is now a post-doc at the Centre Appliquées de Mathématiques with Hélène Morlon. He is interested in historical biogeography, global patterns in biodiversity such as latitudinal diversity gradients and fine-scale patterns such as island biogeography. He aims to decipher the main evolutionary and ecological processes that have shaped current patterns of biodiversity.
Felix Sperling is interested in processes of evolution ranging from the diversification of major insect lineages to the formation of species boundaries.
Gael Kergoat is a research scientist who has focused on understanding the evolution of phytophagous insects and their host plants.
Author contributions: F.L.C. conceived the ideas, collected and analysed the data and wrote the first draft; F.A.H.S. and G.J.K. revised and corrected the manuscript.