Abstract
The Neotropical region has experienced a dynamic landscape evolution throughout the Miocene, with the large wetland Pebas occupying western Amazonia until 11-8 my ago and continuous uplift of the Andes mountains along the western edge of South America. Although the complex dynamics between the Andes and Amazonia may have strongly affected the trajectory of Neotropical biodiversity, there is little evidence for such an influence from time-calibrated phylogenies of groups that diversified during this period. Here, we generate one of the most comprehensive time-calibrated molecular phylogenies of a group of Neotropical insects: the butterfly tribe Ithomiini. Our tree includes 340 species (87% of extant species), spanning 26 million years of diversification in the Neotropics. We investigate temporal and spatial patterns of diversification, focusing on the influence of Miocene landscape tranformations on the dynamics of speciation, extinction and biotic interchanges at the Amazonia/Andes interface. We find that Ithomiini likely began diversifying at the interface between the Andes and the Amazonia around 26.4 my ago. Five subtribes with a very low extant diversity started diversifying early in western Amazonia, but a rapid decrease in diversification rate due to increased extinction rate between 20 and 10 my ago suggests a negative impact of the Pebas wetland system on these early lineages. By contrast, the clade containing the five most species-rich subtribes (85% of extant species) was characterized by a high, positive net diversification rate. This clade diversified exclusively in the Central Andes from 20 to 10 my ago. After the demise of the Pebas system (11-8 my ago), we found a sudden increase of interchanges with the Northern Andes and Amazonia, followed by local diversification, which led to a substantial renewal of diversification. In general, ecological turnovers throughout the Miocene strongly determined the dynamics of speciation, and extinction and interchanges, and appear as a key driving force shaping the region’s current extraordinary biodiversity.
INTRODUCTION
There has been a long fascination among biologists for the Neotropics and the origin of its intriguingly high biodiversity. The timing of Neotropical diversification, and therefore its major driving processes, is still controversial despite the large amount of publications that have addressed the question (e.g., 1, 2, 3).
Despite uncertainty about the precise timing and magnitude of surface uplift, the formation of the Andean cordilleras during the Cenozoic greatly shaped Neotropical landscapes and affected diversification in the Neotropics. As the Andes arose, they brought about new biotic and abiotic conditions along their slopes, modified the climate of the Neotropical region and deeply affected the formation of the Amazonian basin by bringing large amounts of sediments and modifying water drainage 1. There is increasing evidence that the Andes influenced the diversification of Neotropical lineages, primarily by increasing speciation rate, perhaps most spectacularly in the high altitude páramo habitat (e.g., 4). In parallel, the western part of the Amazon basin, which is connected to the Andes, has experienced major turnovers of ecological conditions. During the Oligocene, western Amazonia was occupied by a fluvial system flowing northward (paleo-Orinoco basin), which transformed ~23 million years (my) ago into an aquatic system of shallow lakes and swamps episodically invaded by marine conditions, known as the Pebas system 1, 5, 6, 7, 8. The Pebas was connected northward with the Caribbean Sea and likely also with the Pacific Ocean through the Western-Andean portal (“WAP”, 5), a low-altitude gap that separated the Central Andes and the Northern Andes until 13-11 my ago. During the late Miocene and during the Andean uplift, the accumulation of sediments combined with a sea level decrease initiated the eastward drainage of the Pebas, and by 10-8 my ago the region had changed into a fluvial system, which then turned into the modern configuration of the Amazon. More recently, climatic fluctuations during the Peistocene (2.50 my ago) may have led to episodic dryness affecting Amazonian forest habitats 9. The extent of the influence of Pleistocene events and their effects on Neotropical diversification, and even the importance of dryness episodes, are controversial (e.g, 2, 10, 11, 12).
In this study we purposely focus our attention mostly on the Miocene and Pliocene and how the interaction between the rise of the Andes and coincident large landscape modifications in western Amazonia have determined diversification and dispersal over 30 million years. The Pebas ecosystem covered up 1.1 million km2 at its maximum 7 and was probably not suitable for a terrestrial fauna. Therefore, between 23 and 10 my ago, diversification of terrestrial lineages may have been impeded in western Amazonia or restricted to its edges 7. By contrast, the uplift of the Central and the Northern Andes, also occurring throughout the Miocene and the Pliocene, and the ecological gradients present along this mountain chain probably constituted an important driver of diversification. In the last 10 to 8 my, the retreat of the Pebas may have provided opportunities for terrestrial lineages to radiate in western Amazonia. The Pebas may also have constrained dispersal, acting as a barrier between the Andes, the Guiana shield and western Amazonia. Thus, determining whether rates of interchanges have been constant throughout time since the origin of Ithomiini, or instead have increased after the Pebas’s retreat, would test the potential importance of this ecosystem in building the modern pattern of diversity 6.
Paleontological studies have shown that the Pebas greatly contributed to the diversification of aquatic fauna such as molluscs 13, ostracods 14 and crocodilians 15. However, the fossil record also suggests a negative effect of the Pebas system on terrestrial fauna 16, 17. The hypothesis that the Pebas has shaped patterns of terrestrial diversification and dispersal in western Amazonia has grown over the years (e.g., 5, 6, 7,16, 18) but support from molecular phylogenies mostly stems from the observation that western Amazonian clades have diversified during the last 10-8 my and not before (19, 6 and reference therein). Yet, there is very little information on what happened before, when the Pebas was occupying western Amazonia, particularly on whether the presence of Pebas constrained diversification and interchange patterns in this region. A thorough assessment of the role of the Pebas ecosystem on diversification and dispersal requires phylogenies of large Neotropical clades that originated before the formation of the Pebas, i.e. clades older than 23 my old. Phylogenies of Neotropical clades meeting these conditions are surprisingly rare. In insects – which are among the most diverse terrestrial organisms - attempts to build phylogenies of Neotropical groups to test different drivers of Neotropical diversification have either suffered from a small size or a low sampling fraction (e.g. 19, 20, 21, 22, 23), and therefore from low statistical power and reliability.
Butterflies are among the best candidates for addressing the evolution of the Neotropical biota at such time scales. They are probably the best-known insects and Neotropical butterfly lineages have benefited from substantial phylogenetic research compared to other insects (e. g., 19, 20, 21, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35). Among the most emblematic Neotropical butterflies is the tribe Ithomiini (Nymphalidae: Danainae, 393 species), also referred to as the clearwing butterflies because of the transparent wings of the majority of species. Ithomiini are forest-dwellers distributed throughout the Neotropics, from sea level up into montane cloud forests (to 3000 m), where their larvae feed on plants of the families Solanaceae, Gesneriaceae and Apocynaceae 36. Species richness is primarily concentrated in the Andes, where about half of the species occur (mostly on the eastern slopes) and in western Amazonia. Ithomiini are chemically defended and they engage in Müllerian mimicry, whereby co-occurring species exhibit convergent wing colour patterns that advertise their toxicity to predators 37. Ithomiini butterflies represent a keystone group in Neotropical forests by numerically dominating mimetic butterfly communities and sharing wing colour patterns with a large number of other palatable and unpalatable Lepidoptera, such as the iconic Heliconius butterflies 38. For this reason, Ithomiini were used by both Bates 39 and Müller 37 in their original descriptions of deceptive (Batesian) and mutualistic (Müllerian) mimicry, respectively.
The diversity and the intriguing ecology of Ithomiini has generated a great interest and a broad and diverse literature on topics including life history 40, 41, 42, 43, chemical ecology 44, 45, 46, systematics 19, 21, 26, 32, 34, 36, 47, 48, cytogenetics 49, community ecology 50, 51, 52, 53, 54, 55, wing colour pattern evolution 56, and biogeography 19, 21, 26, 27. In this study, we build on existing molecular data and provide a large amount of novel DNA sequences for ithomiine species to generate the first species-level phylogeny of the entire tribe, providing a large and densely sampled (340 species included out of 393 currently recognized) phylogenetic dataset for a Neotropical insect clade that underwent diversification during the last ~30 million years 27, 57. This phylogenetic framework provides an ideal opportunity for investigating Neotropical diversification over a large evolutionary time-scale. Ithomiini originated before the Pebas system, thus offering the opportunity to investigate diversification before, during and after the environmental changes during the Miocene with a high statistical power. Importantly, contrary to many other large radiations of similar age, Ithomiini are endemic to the Neotropics. Their diversification therefore occurred without interaction with other biogeographic regions such as the Nearctic.
Here, we investigated the dynamics of speciation, extinction and dispersal rates in Ithomiini through space and time, using a combination of time- and trait-dependent models of diversification and historical biogeography. We focused on the interaction between the turnover of ecological conditions in western Amazonia and the Andean uplift during the Miocene, and we investigated whether geological and ecological events in both regions affected synergistically the diversification of Ithomiini. More specifically, an important role for Andean uplift and the Pebas would be supported if: (1) During the Pebas period, (a) Andean diversification largely exceeds Amazonian diversification, due to an increased diversification in the Andes driven by the evolving ecological gradient and uplift dynamics and/or a reduced diversification rate in Amazonia accompanying the loss of terrestrial habitats; (b) interchanges with western Amazonia are reduced; (c) interchanges between the Central and the Northern Andes are reduced, because of the existence of the WAP. (2) During the retreat of the Pebas: interchanges with western Amazonia and between the Central and the Northern Andes largely increase, as a result of new terrestrial habitats and the disappearance of the WAP, respectively. (3) After the Pebas period: Diversification rates in Amazonia globally increase and biotic interchanges are not constrained anymore. After the Pebas retreat, decrease of speciation rates through time suggest post-Pebas radiations in Amazonia, while increase of speciation rates through time may suggest a role of climatic fluctuations during the last 2.5 my.
RESULTS
Time-calibrated phylogeny
Tree topology and time-calibration
We generated a time-calibrated phylogeny that comprised 340 out of 393 Ithomiini species (Supporting Information S1-S2-S3), based on three mitochondrial and six nuclear gene fragments. The tree topology was obtained under maximum likelihood inference, and branch lengths were estimated by Bayesian inference, using six secondary calibration points from Wahlberg et al. (57). The tree topology was generally well supported, including deep nodes (Supporting Information S4). We found a crown age of Ithomiini of 26.4 (CI=22.75-30.99) my ago (Figure 1, Supporting Information S5) and a divergence time from its sister clade Tellervini of 42.1 (CI=39.50-48.44) my ago. All subtribes (10 in total) diverged in the first 10 million years, in the following order (Figure 1): (1) Melinaeina (26.4, CI=22.75-30.99 my ago), (2) Mechanitina (24.2, CI=21.00-28.59 my ago), (3) the clade consisting of Tithoreina and the Methonina (23.6, CI=20.40-27.71 my ago), (4) Athesitina (22.1, CI=19.09-26.25 my ago) and (5) a large clade that comprises the five most species-rich subtribes (Ithomiina, Napeogenina, Oleriina, Dircennina and Godyridina), hereafter called the “core-group”. The relationships between tribes were similar to those recovered in a recent higher-level phylogeny of Ithomiini based on a combination of 3 gene regions and morphological characters 34, except that Brower et al. (34) recovered Mechanitina as sister to Tithoreina+Methonina. Lineage ages in our phylogeny were generally younger than those inferred in Wahlberg et al. (57), but older than those inferred in Garzon-Orduña et al. (58) (see De-Silva et al. 48 for further discussion of such differences).
Diversification rates
Time-dependent diversification
We combined three methods to assess patterns of diversification through time and across clades. We first used MEDUSA 59, which automatically detects shifts of diversification processes across a phylogeny. The analysis detected two significant shifts from the background process of diversification (Figure 1, Supporting Information S6). One was at the root of the core-group accounting for ~85% of present-day diversity of Ithomiini. The second shift was at the root of a subgroup within the genus Melinaea, which appears to have diversified rapidly during the last one million years (hereafter “ Melinaea-group”).
We then used the method developed by Morlon et al. 60 to fit time-dependent models of speciation and extinction on the different partitions based on the MEDUSA results. This method does not detect shifts automatically, but it allows both speciation and extinction rates to vary through time and across lineages. The results confirmed that the partitioned models (either or both of the two shifts detected by MEDUSA) had a significantly better fit and that the two-shift model was significantly better than the one-shift models (Table 1). For the core-group, no model of time-dependent diversification had a significantly better fit than the null model of constant speciation rate without extinction (0.255 lineage-1my-1, Figure 2, Table 1, Supporting Information S6). Under the null model, the core-group diversity increased exponentially during the last 20 million years and reached its current diversity (334 extant species). For the Melinaea-group, the best fitting model was an exponentially decreasing speciation rate without extinction, with a very high initial speciation rate (7.62 lineage-1my-1 at the root, 0.342 lineage-1my-1 at present, Figure 2, Table 1, Supporting Information S6). The Melinaea-group radiated into eight species during a time lapse of only one million years. On the remaining background lineages, the best fitting model involved a time-dependency of both speciation and extinction rates. The resulting diversification rate was high during the initial stages of diversification (0.75 at the root), but decreased rapidly and became negative around 19 my ago. The background diversification rate then started a slow recovery, but remained negative (-0.038 lineage-1my-1 at present, Figure 2, Table 1, Supporting Information S6). Consequently the background tree diversity reconstructed from this model shows a pattern of diversity that increased during the first 10 my up to ~60 species before slowly declining toward its current diversity (51 species) during the last 15 my. This signal of high relative extinction rate was not affected by initial parameters for maximum likelihood search or by diversification rate heterogeneity potentially remaining from the background (see discussion in Supporting Information S7). Clades within the background tree showing positive net diversification rates during the last 5 my supported the recovery trend described above (Supporting Information S7).
We also used TreePar 61 to fit models of constant diversification rate in 5 my time-bins for each partition identified by MEDUSA, as a second and independent assessment of diversification through time. TreePar accommodates constant birth-death models within time-intervals and allows those birth-death rates to vary between time intervals. The results were congruent with those of the time-dependent models of diversification obtained with Morlon et al. (60)’s method (Figure 2, Supporting Information S7). The diversification rates estimated for the core-group within 5 my time-bins remained relatively constant through time and the turnover rate was close to 0, supporting a null or extremely low relative extinction. For the background tree, we found that diversification rate was highest between 25 and 20 my ago, declining toward 0 during the last 15 my, in agreement with the results obtained with the method of Morlon et al. (60). Turnover largely increased and reached a maximum during the last 15 million years, supporting a very high relative extinction rate, similarly to our results of the time-dependent diversification models of Morlon et al. (60)’s method.
We additionally performed an analysis with BAMM v.2.5.0 62, which also detects shifts in diversification dynamics using a Bayesian framework. The results supported the presence of a shift at the root of the core-group with an increasing diversification rate compared to the lineages diverging earlier. More details about this analysis can be found in Supporting Information S8.
Diversification in the Andes
We compared the pattern of diversification in the Andes and in the rest of the Neotropics and we assessed the rate of interchanges between these regions using character state-dependent diversification models (ClaSSE, 63). Such models estimate character state-dependent rates of speciation, extinction and cladogenetic state transitions (i. e., occurring at nodes). Species were classified into two biogeographical states, Andean and non-Andean. We compared 10 models to test whether rates of speciation, extinction or transitions were different or not between regions. For both the whole Ithomiini and the core-group we failed to identify a single model that fitted the data significantly better than other models (Table 2). For the whole Ithomiini, the model with the lowest AIC score had a higher speciation rate in the Andes (λ222=0.230) than in non-Andean regions (λ111=0.118) and those speciation rates were higher than the colonization rates (λ112=λ212=0.079). The second best model had, again, a higher speciation rate in the Andes (λ222=0.231) than in non-Andean regions (λ111=0.108) and the colonization rate out of the Andes was higher (λ212=0.095) than into the Andes (λ212=0.047). In the core-group, there were four models within an AIC difference of 2. All models had extremely low extinction rates. Only one of them inferred different speciation rates and three of them inferred a higher colonization rate into the Andes than out of the Andes (Table 2). Therefore, there is no support for a general pattern of increasing diversification in the Andes across the entire core-group (~85% of the extant diversity). This interpretation was confirmed by additional analyses in which we tested the presence of a hidden character using the model HiSSE 64. The results supported the conclusion that the Andes are not directly associated with higher diversification rates (see Supporting Information S9), and that diversification was also fast in some non-Andean clades and slow in some Andean lineages.
Ancestral state inference is not implemented in ClaSSE. Instead, we used the BiSSE 65 model, in which transitions occur only along branches. To infer ancestral states on the whole tree, we fitted the BiSSE models corresponding to the best-fitting ClaSSE models (i.e. model1: different speciation rates, model2: different speciation and transition rates). In both cases the most likely state of the root was non-Andean but with high uncertainty (probability of 0.508 and 0.543 respectively) and there was uncertainty at the nodes leading to the core-group (Figure 1, Supporting Information S10). The most recent common ancestor (hereafter MRCA) of the core-group was inferred as most likely Andean (model1: probability of 0.558, model2: probability of 0.625). In both models the MRCA of all background subtribes were inferred to be non-Andean with a strong support (except for Athesitina, Figure 1, Supporting Information S10). In the core-group, the best model (with different speciation rates) inferred an Andean origin for subtribes Napeogenina, Dircennina and Godyridina, whereas the second best model (different speciation and transition rates) inferred an Andean origin for all five subtribes (Supporting Information S10).
Diversification in Amazonia
We also investigated the pattern of Amazonian diversification during the post-Pebas period (8-0 my ago). We fitted a model of time-dependent speciation rate using Morlon et al.’s method 60 (no extinction, based on our previous results) to assess whether speciation rates decreased through time (supporting radiations accompanying the post-Pebas recolonizations) or increased through time (supporting a recent diversification potentially caused by Pleistocene climatic fluctuations). We identified four Amazonian diversification events – clades whose nodes were inferred to be almost all Amazonian (see historical biogeography results below) – in the core-group and three in the background subtribes. All core-group clades showed decreasing speciation rates through time, suggesting an early diversification, perhaps following the appearance of new forest habitats accompanying the Pebas retreat. Among the three clades from the background lineages only the genus Methona followed a trend of decreasing speciation rate. The other two clades, Mechanitis + Forbestra and Melinaea (the whole genus) supported an increasing speciation rate through time, which is consistent with a potential effect of Pleistocene climatic fluctuations in driving diversification.
Historical biogeography
We divided the Neotropical region into 9 areas (Supporting Information S11) and we assigned each ithomiine species to these areas according to their current distribution. We performed historical biogeographic reconstruction using BioGeoBEARS 66, under two models (DEC and DIVALIKE), using a three-step procedure outlined below. We first performed a “null” model with uniform dispersal multipliers. Based on the results of the null model, we computed rates of dispersal between specific regions for all 1-my intervals. We then implemented those rates in a time-stratified model.
Biogeographic null model
We started with a “null” biogeographic model, which restricted the area adjacency but set all dispersal probabilities to 1, and we compared the models DEC and DIVALIKE. The null DEC model had a better fit than the null DIVALIKE model (likelihoods: DECnull: −1335.802, DIVALIKEnull: - 1347.869), hence we used the DEC model in all subsequent analyses.
In both models, the ancestral area of the Ithomiini MRCA was unclear (Central Andes + Western Amazonia for the highest probability). The areas where the two first divergences occurred, which led to the Melinaeina and the Mechanitina subtribes, were also unclear (Figure1, Supporting Information S12). The ancestor of the remaining ithomiine lineages was recovered to be only occupying the Central Andes. Following this node (23.6 my ago) all the divergences occurred in the Central Andes until 9.4 my ago, when the first colonization event out of the Central Andes occurred (MRCA of Oleria, which dispersed into Western Amazonia). Hence, our null biogeographic reconstruction found that all subtribes except Melinaeina and Mechanitina originated and started diversifying in the Central Andes. Interchanges between regions appeared to have increased during the last 10 my. However, all node reconstructions at the basal nodes of the background lineages were highly uncertain.
Biogeographic diversification of the core-group
Using this null model, we investigated more specifically the biogeographic pattern of the core-group by computing rates of dispersal among different regions. We applied the null model to 100 trees randomly sampled from the BEAST posterior distribution and extracted the state with the highest probability at each node. Then, for each 1-my interval, we computed the number of specific transitions divided by the number of lineages existing during this interval and fitted a spline line on the distribution of points.
As observed in the ancestral state reconstruction on the maximum clade credibility tree (MCC), no dispersal event occurred during the initial Central-Andean phase of diversification in the core-group (Figure 3). Between 13-8 my ago a major peak of interchanges between the Andes and Amazonia occurred, followed by a second peak between 4-0 my ago (Figure 3). The first peak was almost entirely driven by colonization from the Andes toward the Amazonia, whereas the second peak involved many reverse colonizations toward the Andes. We also recorded a large peak of colonizations from the Central Andes toward the Northern Andes between 13-8 my ago, also followed by a second peak 4-0 my ago (Figure 3, Supporting Information S11-S12). Colonization of Central America may have started 8 my ago, but interchanges largely increased during the last 4 my (Supporting Information S11). Colonizations of the Atlantic Forest also started early (around 13 my ago), but the rate of interchanges between the Atlantic Forest and the remaining Neotropical regions remained relatively constant during the last 10 my (Supporting Information S11-S12).
We also used the biogeographic reconstruction to estimate local diversification, namely the cumulative number of divergences inferred to have occurred exclusively in a given region. As described above, until ~10 my ago speciation events occurring in the Central Andes fully account for the core-group diversification (no dispersal events). During the last ~10 my, we observed a dampening of speciation events in the Central Andes (Figure 3). At the same time, following the peaks of dispersal identified above, Northern Andean and Amazonian lineages started diversifying, although the latter diversified at a slower pace than the former. This reflects the large number of dispersal events into the Northern Andes that were followed by important local diversification, for example in the genera Hypomenitis (17/20 species in the phylogeny occur in the Northern Andes) and Pteronymia (30/45 species in the phylogeny occur in the Northern Andes), or in subclades of the genera Oleria or Napeogenes. We also identified some important transitions to lowland Amazonia, for example at the origin of the Brevioleria-clade, during early divergence in the genus Oleria, or in the genus Hypothyris (see results on Diversification in Amazonia).
Time-stratified biogeographic model
We used the results highlighted above to refine the biogeographic model by incorporating the variations of dispersal rates identified into a model accounting for time-stratified dispersal multipliers. This time-stratified model designed from rates of colonization computed above led to a significant improvement of the model (likelihoods: DECnull: −1335.802, DECstrat: −1321.805). Both ancestral state reconstructions were very congruent but the time-stratified model increased the resolution of several nodes throughout the tree (Supporting Information S11-S12). We identified one major difference in the ancestral states. From the null model the ancestral state of the subtribe Melinaeina, the first lineage to diverge, was highly unclear and the first nodes within Melinaeina were identified as Central Andean, although this was not strongly supported. Likewise, in the null model, Mechanitina, the second lineage to diverge, was inferred to have diversified in the Atlantic Forest but this was poorly supported (Supporting Information S11-S12). The time-stratified model greatly increased the resolution of all these deep nodes, inferring that both Mechanitina and Melinaeina likely initially occupied Western Amazonia. For Melinaeina and Mechanitina this result was in agreement with the BiSSE ancestral state reconstruction. The ancestral state reconstruction inferred that the Ithomiini occupied the Andes from the root, but this was very weakly supported. By contrast, all reconstructions, either BioGeoBEARS or BiSSE, inferred an Andean origin for the core-group.
DISCUSSION
We generated one of the largest species-level phylogenies to date for a tropical insect group, the emblematic Neotropical butterfly tribe Ithomiini. With 340 out of 393 species included and a crown age of 26.4 my, this phylogeny offers a unique opportunity to investigate the dynamics of diversification of an insect group throughout the Neotropical region during the major geological and ecological events that have occurred since the Miocene. We discuss our findings below and we propose that the dynamics of multiple landscape transformations during the Miocene, and more specifically the interactions between the Andes and the Pebas system, have strongly affected the dynamics of speciation, extinction and biotic interchanges of Ithomiini butterflies in the Neotropical region.
Early diversification at the interface of the Pebas and Central Andes: has the Pebas driven extinction?
The Ithomiini probably originated along the early Andean foothills at the transition with western Amazonia. The onset of the uplift of the eastern cordillera of the Central Andes during late Oligocene coincides with the origin of Ithomiini 67 and our reconstruction of the ancestral biogeographic area for the MRCA of the tribe was unable to distinguish between Central Andes or western Amazonia. The Pebas ecosystem replaced the previous western Amazonian terrestrial ecosystem from 23 to 10 my ago. Wesselingh et al. (7) described the Pebas as an ecosystem “which was permanently aquatic with minor swamps and fluvial influence, and was connected to marine environments”, and may have reached a maximum size of 1.1 million km2. The presence of fossil marine fishes 68 and molluscs 13 testifies to the presence of saline waters. More recently, Boonstra et al. 69 found evidence from foraminifera and dinoflagellate cysts that marine incursions reached 2000 km inland from the Caribbean sea during the early to middle Miocene during periods of high sea levels. The extent and duration of these marine influences is controversial (see 70 and references therein). Yet, it is undeniable that the Pebas system was not suitable for terrestrial fauna and flora, and therefore was likely to affect diversification and dispersal of the terrestrial fauna, including early Ithomiini lineages.
The timing of diversification of background lineages reveals a fast early diversification, perhaps following the colonization of South America during the pre-Pebas period – the sister clade of Ithomiini, Tellervini, is found in Australia and Papua New Guinea. Diversification was perhaps partly facilitated by an early shift to a new and diverse hostplant family, the Solanaceae, which is particularly diverse in the Neotropics 36, but the possible effects of new hostplants on early diversification will be difficult to distinguish from the effects of newly available landscapes. Yet, diversification rate rapidly decreased through time, driven by an increasing relative extinction rate and at a time corresponding to the replacement of the terrestrial habitats by the Pebas ecosystem, i.e., ca. 23 my ago. Although the ancestral area of Ithomiini is ambiguous, the two first diverging Ithomiini lineages (Melinaeina and Mechanitina) were clearly endemic to western Amazonia (time-stratified model in BioGeoBEARS and BiSSE reconstruction) and therefore were likely to be affected by the dramatic landscape modifications of the Miocene. However, there are uncertainties surrounding the other deep nodes and also the time when first colonization of the Andes occurred. Two scenarios can be envisioned: (1) The remaining lineages (Tithoreina, Methonina, Athesitina and the core-group) became endemic to the Central Andes (supported by the BioGeoBEARS reconstruction) and we do not know what has driven the shift of diversification at the root of the core-group; and (2) All background lineages were ancestrally western Amazonian (supported by BiSSE ancestral state reconstruction and the sister-clade Tellervini being a group restricted to lowlands) and central Andean endemicity occurred at the root of the core-group only, but extinction events in the background lineages (potentially higher in western Amazonian lineages) may have falsified the BioGeoBEARS ancestral state reconstruction. Indeed, asymmetrical extinction across different geographical regions, as suspected here, may lead to inaccurate inferences of past geographic ranges 71, 72. In our case, if western Amazonian lineages were more prone to extinction than Andean lineages due to the presence of the Pebas, ancestral reconstruction of the distribution areas based on a phylogeny of extant taxa, i. e. those that survived extinction, will be biased towards Andean regions. Such a scenario, where background lineages were ancestrally western Amazonian, would explain the common pattern of high relative extinction rate in the background lineages and a shift of diversification process at the root of the core-group.
The idea that the Pebas may have driven extinction is well supported by a recent evaluation of an Amazonian fossil record, which pointed at a major decline of diversity in western Amazonia during the early and middle Miocene 16. This study concludes that mammalian diversity dropped from 11 orders, 29 families and 38 species during late Oligocene down to 1 order, 2 families and 2 species during middle Miocene (see also 17). These results and the pattern of extinction we found in (at least some) early Amazonian Ithomiini, which occurred during the Pebas period, strongly suggest that the late Oligocene fauna occupying western Amazonia suffered from extinction during the Pebas period. The progressive recovery of these background lineages toward the present, including positive diversification rates in some recent lineages, also concurs with the idea that the retreat of the Pebas released the constraints on diversification during the last 10 my (Supporting Information S7).
Parallel to the events occurring in western Amazonia during the Pebas period, the core-group MRCA (19.1-22.1 my old) occupied the Central Andes. This event was of major importance in shaping the diversification of Ithomiini since it is the origin of 85% of the current Ithomiini diversity. Firstly, from this event until ~10 my ago, all core-group lineages exclusively diversified in the Central Andes, meaning that from 19.1-22.1 to ~10 my ago not a single dispersal event occurred out of the Central Andes. Secondly, the core-group corresponds to a shift of diversification dynamics, characterized by a low (or zero) extinction rate and a constant speciation rate, which greatly contrasts with the slow and even negative diversification dynamics of the background lineages during the same period. Consequently, the Central Andes hosted most of the diversification during the first half of Ithomiini history. A two-fold higher Andean diversification rate was found across the whole Ithomiini, which may be mainly the consequence of the diversification rate shift found at the root of the core-group. By contrast, when considering only the core-group, Andean and non-Andean lineages had similar diversification rates. The lack of support for a general increase in diversification rate in the Andes within the core-group is also supported by analyses performed independently on different core-group subtribes. For example, in both Oleriina 73 and Godyridina 19, radiations occurred in both Andean and Amazonian genera.
Dispersal out of the Central Andes at the demise of the Pebas
Gentry (74) pointed at a dichotomy observed in the geographic distribution of Neotropical plant diversity, showing that groups could be divided into Andean-centred versus Amazonian-centred patterns. Clades tend to be species-rich in one of these centres and relatively species-poor in the other. Antonelli & Sanmartín (6) coined this observation the “Gentry-pattern”. They also suggested that in the absence of a barrier between the Andes and the Amazon basin we should observe continuous interchanges between these regions. Antonelli & Sanmartín (6) proposed that the Pebas could be this “missing long-lasting barrier needed for creating the disjunction between Andean-centred and Amazonian-centred groups”. Therefore, in addition to the constraints on diversification discussed above, we predicted that the Pebas ecosystem should have influenced interchanges toward or across western Amazonia.
Our results conform surprisingly well to the scenario proposed by Antonelli & Sanmartín (6). Ithomiini are Andean-centered with more than a half of their current diversity occurring in the Andes (see also 75). Here we show that interchanges have been virtually absent during the Pebas period, with a period as long as 9-12 my without interchanges. However, rates of interchanges from the Central Andes toward the Northern Andes and Amazonia suddenly peaked ~10 my ago (between 13-8 my ago) and more recently (4-0 my ago). The Western Andean Portal (WAP) is a low-altitude gap that separated the Central Andes and the Northern Andes until 13-11 my ago, and which may have connected the Pebas system and the Pacific Ocean (6). The closure of the WAP may have allowed multiple colonizations of the Northern Andes facilitated by the presence of connecting higher altitude habitats. In parallel, between 10-8 my ago the Pebas system was drained eastward, leading to the formation of the present-day configuration of Amazonian drainage basin. It was accompanied by the expansion of terrestrial forest habitats in western Amazonia. This corresponds precisely to the timing at which core-group lineages colonized western Amazonia and then diversified.
Diversification across the whole Neotropics following the demise of the Pebas
We found a strong dampening of local speciation in the Central Andes during the last 10 my. However, colonizations following the retreat of the Pebas system were followed by large local bursts of diversification within the Northern Andes and Amazonia. As an illustration, from our biogeographic reconstruction, 69 divergence events occurred strictly in the Central Andes in the coregroup from 20 my ago until present-day. However, multiple independent dispersal events followed by local diversification lead to the exact same number of divergences occurring strictly in the Northern Andes during the last 9 my only. The genera Hypomenitis and Pteronymia, for example, diversified extensively within the Northern Andes with 23 and 53 valid species respectively. We also identified four Amazonian radiations in the core-group and three in the background lineages. Lineages that dispersed into the Northern Andes and Amazonia after the demise of the Pebas system probably benefited from a large range of free ecological niches, including a diversity of host-plants that had already diversified or that radiated concomitantly. Two of the background Amazonian radiations, the genus Melinaea and the clade Mechanitis + Forbestra, showed increasing speciation rate toward the present, due in the former case to the shift detected by MEDUSA (Melinaea-group). The recent and dramatic increase in diversification rate of the Melinaea-group, which produced at least eight species and 50 subspecies 43 in just 1 my, is particularly intriguing. This radiation may be interpreted as support for an effect of recent climatic fluctuations during the Pleistocene on the diversification of this group (as well as the Mechanitis + Forbestra clade), although ecological drivers of speciation classically invoked in mimetic butterfly diversification, such as colour pattern and hostplant shifts,cannot be ruled out 42, 43. Five other Amazonian radiations showed diversification rates decreasing through time, meaning that diversification was highest just after the retreat of the Pebas. Recent radiations in western Amazonia that post-date the Pebas period have been repeatedly reported. For example, in the genus Astrocaryum (Arecaceae), the upper Amazonian clade started diversifying only ~6 my ago 76. In Taygetis butterflies, Amazonian lineages show rapid diversification during the last 7-8 my 35. Such convergent timing of diversification in western Amazonia strongly supports the scenario of a post-Pebas recovery of terrestrial habitats, which triggered dispersal followed by local diversification.
Conclusion
Our research shows that the timing of diversification and biogeographic interchanges in Ithomiini butterflies are tightly associated with the turnover of ecological conditions that occurred during the Miocene. Our findings suggest that the ecological turnover that first accompanied the expansion of the Pebas system has led to a decline of diversification, potentially driven by increasing extinction, in early lineages adapted to the ecological conditions that existed during the Oligocene in western Amazonia. Such a decline of diversity has also been documented in the fossil record 16, which calls for further investigations on the role of the Pebas in driving extinction during the Miocene. By contrast, lineages that colonized the Central Andes 20 my ago rapidly diversified. However, during the entire existence of the Pebas, these lineages remained trapped in the Central Andes (at least 9-12 my without dispersal events out of the Andes). The closure of the West Andean Portal, connecting the Central and North Andes, and the associated demise of the Pebas (10-8 my ago), apparently released these long-lasting barriers, allowing interchanges with the Northern Andes and Amazonia and opening new opportunities for diversification. As a result of these multiple events, major differences appear between the different faunas. Central Andean lineages started diversifying early (at least 19 my ago), allowing species to accumulate over a long period of time, but diversification slowed down during the last 10 my. In contrast, the Northern Andean fauna is recent (13-11 my old at most), driven by multiple colonization events sometimes followed by important bursts of diversification. In parallel, some Amazonian lineages may be old (late Oligocene), but modern diversity almost entirely arose during the last 8-10 my, after the demise of the Pebas ecosystem. Taken together, all this information points to a robust scenario for Neotropical diversification, which highlights the role of Miocene ecosystem turnover in determining the timing of interchanges, speciation and extinction in the world’s most biologically diverse region.
Author contributions
NC and ME conceived the study, with contribution from KRW, GL and AVLF. All co-authors provided specimens and sequences. NC, ME, FPP, CFA, DLDS performed the labwork. NC performed the analyses. NC wrote the paper with major contributions from ME, and contributions from all co-authors.
List of supporting information:
S1. List of all individuals and species used in this study, and biogeographic distribution of species.
S2. Gene fragment partitions and substitution models associated obtained by Partition Finder v.1.1
S3. Node constraints used for time-calibration of the tree.
S4. Maximum likelihood tree obtained with IQ-tree, with bootstrap support indicated at the nodes.
S5. Bayesian time-calibrated tree obtained with BEAST v.1.8 with median node ages and 95%HPD indicated at the nodes.
S6. Full results of diversification analyses.
S7. Testing the effect of diversification rate heterogeneity within the background on diversification rate estimates
S8. Diversification analysis with BAMM.
S9. Trait-dependent diversification using HiSSE.
S9. Ancestral state reconstruction performed using trait-dependent diversification models
S10. Rates of colonization between different region, computed on the core-group and used to design a time-stratified biogeographic model.
S11. Results of biogeographic ancestral state reconstruction obtained with BioGeoBEARS and using the “null” or time-stratified model.
Acknowledgements
This project was funded by an ATIP (CNRS, France) grant awarded to ME, with LDS as a postdoc. NC was funded by a doctoral fellowship from Ecole Doctorale 227 (France). ME acknowledges additional funding by the ANR grant SPECREP (ANR-14-CE02-0011-01). We thank the authorities of Peru, Ecuador and Brazil (SISBIO n° 10438-1) for providing research and collection permits, as well as many assistants for their help in the field. Molecular work was performed at the GenePool (University of Edinburgh, UK), CBMEG-Unicamp (Brazil) and the Service de Systématique Moléculaire UMS2700 of the MNHN (France). We are grateful to Niklas Wahlberg for providing unpublished sequences of Greta diaphanus. We thank Fabien Condamine and Hélène Morlon for constructive discussions about time-dependent diversification analyses. AVLF thanks CNPq (grant 303834/2015-3), RedeLep-SISBIOTA-Brasil/CNPq (563332/2010-7), National Science Foundation (DEB-1256742), FAPESP (grants 2011/50225-3 and 2012/50260-6) and USAID (Mapping and Conserving Butterfly Biodiversity in the Brazilian Amazon).
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵