The authors have declared that no competing interests exist.
Conceived and designed the experiments: CP ME. Performed the experiments: CP ME. Analyzed the data: CP ME. Contributed reagents/materials/analysis tools: CP ME. Wrote the paper: CP ME.
Current address: McGuire Center for Lepidoptera and Biodiversity, Florida Museum of Natural History, University of Florida, Gainesville, FL, US.
The species rich butterfly family Nymphalidae has been used to study evolutionary interactions between plants and insects. Theories of insect-hostplant dynamics predict accelerated diversification due to key innovations. In evolutionary biology, analysis of maximum credibility trees in the software MEDUSA (modelling evolutionary diversity using stepwise AIC) is a popular method for estimation of shifts in diversification rates. We investigated whether phylogenetic uncertainty can produce different results by extending the method across a random sample of trees from the posterior distribution of a Bayesian run. Using the MultiMEDUSA approach, we found that phylogenetic uncertainty greatly affects diversification rate estimates. Different trees produced diversification rates ranging from high values to almost zero for the same clade, and both significant rate increase and decrease in some clades. Only four out of 18 significant shifts found on the maximum clade credibility tree were consistent across most of the sampled trees. Among these, we found accelerated diversification for Ithomiini butterflies. We used the binary speciation and extinction model (BiSSE) and found that a hostplant shift to Solanaceae is correlated with increased net diversification rates in Ithomiini, congruent with the diffuse cospeciation hypothesis. Our results show that taking phylogenetic uncertainty into account when estimating net diversification rate shifts is of great importance, as very different results can be obtained when using the maximum clade credibility tree and other trees from the posterior distribution.
Hostplant shifts have been invoked to be responsible for a great part of the biodiversity of herbivorous insects [
The butterfly family Nymphalidae has been an important taxon for developing some of the mentioned hypotheses. Nymphalidae contains around 6000 species [
Although it has been suggested that part of the great diversity of Nymphalidae butterflies is a result of hostplant-insect dynamics, it is necessary to use modern techniques to investigate whether the diversification patterns of Nymphalidae are in agreement with the theoretical predictions. It is necessary to test whether the overall diversification pattern of Nymphalidae is congruent with events of sudden diversification bursts due to hostplant shift or climatic events [
In this study, we used a time-calibrated genus-level phylogenetic hypothesis for Nymphalidae butterflies [
For analyses, we used the phylogenetic trees from the study of Wahlberg et al. (2009) [
Rate shifts were estimated for the following nodes (besides the background rate): 1) root, 2) Limenitidinae + Heliconiinae, 3)
Species richness data for Nymphalidae genera were compiled from several sources including the specialist-curated lists [
Hostplant data for Nymphalidae species were compiled from several sources including [
We used the statistical software R version 3.0.1 [
Patterns of diversification in Nymphalidae were analyzed by using MEDUSA version 093.4.33 [
As the MEDUSA and MultiMEDUSA approaches estimated an increase in net diversification in the clade Ithomiini, we tested whether this pattern can be explained by an increase in the birth-rate due to hostplant use and performed analyses using the BiSSE model [
The MEDUSA analysis on the MCC tree in combination with richness data estimated 18 significant changes in the tempo of diversification in Nymphalidae history (
Shift N° | Shift.Node | Model | r | LnLik.part | AICc | Taxa |
---|---|---|---|---|---|---|
1 | 399 | yule | 0.092459 | -1055.957 | Nymphalidae (root) | |
2 | 691 | bd | 0.054129 | -406.3703 | Limenitidinae + Heliconiinae | |
3 | 299 | yule | 0.311199 | -6.3058 | ||
4 | 224 | yule | 0.290989 | -6.2601 | ||
5 | 750 | yule | 0.186913 | -147.4146 | Oleriina + Ithomiina + Napeogenina + Dircennina + Godyrina | |
6 | 405 | yule | 0.116252 | -555.0276 | Satyrinae | |
7 | 495 | yule | 0.064656 | -124.9143 | Coenonymphina | |
8 | 609 | yule | 0.240562 | -35.0908 | Phyciodina in part | |
9 | 787 | bd | 0.042332 | -43.7819 | Danaini in part | |
10 | 231 | yule | 0.209416 | -4.7955 | ||
11 | 478 | yule | 0.311684 | -9.2218 | ||
12 | 659 | yule | 0.219253 | -11.2686 | ||
13 | 444 | yule | 0.220615 | -43.7812 | Satyrina | |
14 | 524 | yule | 0.190754 | -26.0651 | Mycalesina | |
15 | 355 | yule | 0.234041 | -6.2013 | ||
16 | 714 | yule | 0 | 0 | ||
17 | 377 | yule | 0.311671 | -4.1986 | ||
18 | 688 | yule | 0.024724 | -17.5256 | Pseudergolinae | |
19 | 583 | yule | 0 | 0 | 5090.492 |
Shift.Node = node number, Model = preferred diversification model by MEDUSA, r = net diversification rate, LnLik.part = log likelihood value.
Some of the 18 changes in diversification correspond to rate increases in very species-rich genera:
We used MEDUSA to find out whether taking into account the phylogenetic signal from the random sample of 1000 trees from the posterior distribution can return similar estimates of diversification to the values obtained from the MCC tree. We ran MultiMEDUSA on the random sample of 1000 trees (
We found that the analysis by MultiMEDUSA on the 1000 trees estimated lower median net diversification rates for the diversification shifts found by MEDUSA on the MCC tree derived from the random sample of trees (
Shift.Node | rate by MEDUSA | Median rate by MultiMEDUSA | probability of being recovered |
---|---|---|---|
1 | 0.092 | not found | 0.000 |
2 | 0.055 | -0.030 | 0.864 |
0.184 | |||
0.166 | |||
0.111 | |||
6 | 0.119 | 0.039 | 0.131 |
7 | 0.066 | -0.052 | 0.195 |
8 | 0.232 | 0.166 | 0.319 |
9 | 0.042 | -0.049 | 0.897 |
10 | 0.058 | 0.149 | 0.619 |
11 | 0.311 | 0.208 | 0.831 |
0.219 | |||
13 | 0.082 | 0.117 | 0.276 |
14 | 0.099 | 0.115 | 0.379 |
15 | 0.113 | 0.127 | 0.825 |
16 | 0.222 | -0.005 | 0.659 |
17 | 0.243 | 0.248 | 0.785 |
18 | 0.192 | -0.064 | 0.024 |
19 | 0.064 | -0.087 | 0.381 |
We also compared the results from MultiMEDUSA (derived from the sample of 1000 trees) with the shifts found by MEDUSA on the MCC tree derived from this random sample. In the summary statistics, MultiMEDUSA reports the frequency of the diversification shifts found in the trees (parameter
For the diversification shifts found in both the MCC tree and most of the samples of 1000 trees (frequency more than 90%;
There were four net diversification rate increases found in the trees from the random sample (
MultiMEDUSA provided mean and standard deviation statistics for the diversification values on the shifts on the 1000 trees (
It is also evident that not all the diversification shifts estimated on the MCC tree are consistently recovered in most of the 1000 trees. Some of the shifts in the MCC tree are recovered in very few trees, for example the shift for the clade Satyrinae is recovered with a probability of 0.18 (
The MEDUSA analyses, taking into account phylogenetic uncertainty, estimated a net diversification rate increase in part of the clade Ithomiini across more than 95% of the trees. Our BiSSE analysis found a positive effect of the character state “feeding on Solanaceae” on the net diversification rate on part of Ithomiini (Oleriina + Ithomiina + Napeogenina + Dircennina + Godyridina) (
Df | lnLik | AIC | ChiSq | p | |
---|---|---|---|---|---|
full | 6 | -1613.3 | 3238.5 | ||
equal.lambda | 5 | -1619.4 | 3248.9 | 12.3 | 0.00045 |
Speciation and net diversification rates are significantly higher in Solanaceae feeders (speciation rate =
The MEDUSA method has been used to infer changes in net diversification rates in a phylogenetic tree. Since its publication [
In this study, the effect of phylogenetic uncertainty on the inferred diversification shifts by MEDUSA is amplified because some Nymphalidae taxa appear to be strongly affected by long-branch attraction artifacts [
If there is strong phylogenetic signal for increases or decreases in net diversification rates for a node, it is expected that these shifts would be inferred by MEDUSA in most of the posterior distribution of trees. However, weak phylogenetic signal for some nodes can cause some clades to be absent in some trees and MEDUSA will be unable to estimate any diversification shift (due to a non-existent node). The relatively low phylogenetic support for many nodes in the Nymphalidae tree is likely the reason why MEDUSA estimated net diversification rate shifts with a probability higher than 0.90 in the sample of trees for only four shifts: the genus
Keith Brown suggested that feeding on Solanaceae was an important event in the diversification of Ithomiini butterflies [
We investigated whether the strong signal for an increase in net diversification rate for Ithomiini (found by MEDUSA) can be explained due to the use of Solanaceae plants as hosts during larval stage. For this, we used a Bayesian approach [
Our BiSSE analysis, extended to take into account missing taxa and phylogenetic uncertainty, shows a significantly higher net diversification rate for Ithomiini taxa, which can be attributed to the trait “feeding on Solanaceae hostplants” (
Solanaceae plants contain chemical compounds and it has been suggested that the high diversity of Ithomiini is consistent with the “escape-and-radiate scenario” due to a shift onto Solanaceae [
The increase in diversification rate inferred by MEDUSA occurred after the probable shift from Apocynaceae to Solanaceae, as the Solanaceae feeders in the subtribes Melinaeina and Mechanitina are not included in the diversification shift (shift number 5 in
Although the Solanaceae genera used by the Ithomiini clades are well known [
The diffuse cospeciation hypothesis predicts almost identical ages of insects and their hostplants, while the “resource abundance-dependent diversity” and the “escape-and-radiate” hypotheses posit that insects diversify after their hostplants [
Our MultiMEDUSA approach showed a significant slowdown in net diversification rate in the subtribe Danaina of the Danini. Both Danaina and the sister clade Euploeina feed mainly on Apocynaceae and thus a hostplant shift should not be responsible for the observed slowdown of diversification in the Danaina. As expected, our BiSSE analysis of Apocynaceae feeders shows that there is no effect of feeding on this plant family on the net diversification rates of Nymphalidae lineages. Many of the Danaina are large, strong fliers, highly migratory and involved in mimicry rings, including the best-known migratory butterfly, the monarch (
Lineages in the diverse family Satyrinae radiated simultaneously with the radiation of their main hostplant, grasses, during the climatic cooling in the Oligocene [
We found that even though MEDUSA estimated several diversification shifts in the maximum clade credibility tree of Nymphalidae, only a few of these shifts were found in more than 90% of the trees from the posterior distribution. In the literature, it is common practice that conclusions are based on the shifts estimated on the maximum clade credibility tree. However, by using a MultiMEDUSA approach, we found that for this Nymphalidae dataset some of these shifts might be greatly affected by phylogenetic uncertainty. Moreover, some of these shifts can be recovered either as increases or decreases in net diversification rate depending on the tree from the posterior distribution that was used for analysis. This means that contradictory conclusions would be made if only the maximum clade credibility tree was used for analysis. We recommend that all datasets should be analyzed using the MultiMEDUSA approach in order to test whether the results are robust when phylogenetic uncertainty is taken into account.
MEDUSA appears to be sensitive to the number of nodes with high posterior probability and width of age confidence intervals. For our data, it would be necessary to obtain a posterior distribution of trees with no conflicting topology, and very similar estimated ages for nodes in order to consistently recover most of the diversification shifts on the posterior distribution of trees that were inferred by MEDUSA on the MCC tree.
Our MultiMEDUSA approach to perform analyses on the posterior distribution of trees found strong support for an increase in net diversification rate in the tribe Ithomiini, the genus
(PDF)
The same pattern is recovered, speciation and net diversification rates are significantly higher for Solanaceae feeders
(PDF)
(PDF)
(PDF)
Speciation and net diversification rates are similar.
(PDF)
(PDF)
(PDF)
(PDF)
(ZIP)
This script removes the outgroup taxa and loads the richness data for the tree terminals.
(TXT)
(TXT)
(ZIP)
(CSV)
(CSV)
(CSV)
(CSV)
(TXT)
(TXT)
(CSV)
(TXT)
(CSV)
(CSV)
(TXT)
(TXT)
(TXT)
We are thankful to Mark Cornwall for help with the script to extend MEDUSA to include phylogenetic uncertainty, Niklas Wahlberg for commenting on the manuscript and giving us the posterior distribution of trees, Luke Harmon for commenting on the manuscript, Jeffrey C. Oliver and anonymous reviewers for their comments, which greatly improved the manuscript, Jessica Slove Davidson and Niklas Janz for access to their hostplant data. The study was supported by a Kone Foundation grant (awarded to Niklas Wahlberg), Finland (C. Peña) and the Research Council of Norway (grant no. 204308 to M. Espeland). We acknowledge CSC–IT Center for Science Ltd. (Finland) for the allocation of computational resources.