Computational protocol: HOST PLANT UTILIZATION, HOST RANGE OSCILLATIONS AND DIVERSIFICATION IN NYMPHALID BUTTERFLIES: A PHYLOGENETIC INVESTIGATION

Similar protocols

Protocol publication

[…] The taxonomy of Nymphalidae has changed drastically over the past few years, due to a much improved phylogenetic understanding (especially Wahlberg et al. , , ; Brower et al. , ; Pena et al. ; Simonsen et al. ; Willmott and Freitas ; Silva-Brandao et al. ). We closely followed the species lists and delimitations of higher butterfly taxa suggested by the Tree of Life Website (Maddison and Schulz ), as this section of the website is well curated and provides a “point of stability” in the taxonomy. We used the program Mesquite (Maddison and Maddison ) to create the phylogeny, starting with a list of all nymphalid genera recognized in Tree of Life, deleting taxa without host plant information, presenting the remaining taxa as a nonresolved bush, and then manually resolving the phylogeny by moving branches.For the Nymphalidae phylogeny itself we also incorporated well-supported changes suggested more recently (in particular in Wahlberg et al. ), but not yet reflected in Tree of Life. Additional sources were: Kodandaramaiah et al. (, b, Mycalesina, Coenonymphina); Ohshima et al. (, Apaturinae); Pena et al. (, , Euptychiina, Satyrini); Price et al. (, Dirini); Ortiz-Acevedo and Wilmott (, Preponini), and Penz et al. (, Brassolini). It should be noted that to make full use of this information, we resolved a few sections of the phylogeny while leaving taxa not included in the studies listed above as unresolved at the base of the respective clades, rather than collapsing the entire clades as would have been the strictly correct procedure. However, we also tested the effects of using a more conservative phylogeny in accordance with the Tree of Life site (as of 2012). The results were very similar and are not shown here.Furthermore, we obtained some information on branch lengths in the complete phylogeny by incorporating ages of all dated nodes in Wahlberg et al. (), which corresponded to nodes in our phylogeny, using the tool “Node age constraints” in the program Mesquite (Maddison and Maddison ). Ages of nodes were added as “fixed,” but branch lengths were then transformed using the option Enforce Minimum Node Age Constraints. Using this method it was possible to obtain branch length estimates also for sections of the phylogeny containing taxa with host plant information not included in Wahlberg et al. () or where more recently proposed topologies are inconsistent with this study. However, to avoid zero-length branches we had to arbitrarily give all splits between two most closely related terminal taxa for which age information was lacking an arbitrary minimum age of 10 Mya (most dated splits of this kind in the phylogeny are in the range 12–20 Mya).Our final database of the Nymphalidae host records (Appendix S1) contains 551 butterfly taxa. In the large majority of cases these taxa are at the genus level, but a few are species that are currently unplaced. In the phylogenetic analyses, moreover, the diverse satyrine tribe Pronophilini, with 66 neotropical genera only recorded to feed on Poales, was for simplicity treated as a single taxon (underestimating slightly the very conserved nature of use of Poales in the Satyrinae). All available host plant records were investigated at the species level, and the 551 taxa in the database thus represents about 6000 species in the family (Maddison and Schulz ) to the extent that there are host plant records at the species level.For purposes of overview and illustration, we also made use of a simplified phylogeny of Nymphalidae at the level of tribes (or subfamilies for some smaller taxa; Figs. , ). [...] Most of the taxa in our butterfly database (in most cases genera) completely lack species that have been recorded to feed on more than one host plant order (note that the orders listed for each higher taxon is the complete list of orders reliably recorded from this taxon, and can be from a number of specialists rather than from a species feeding on several orders). We defined a “polyphagous” species as one having larvae that feed on at least two orders (character “2 orders”) or at least three orders (character “3 orders”), and the database (Appendix S1) lists all and only such species (along with some where the host records do not currently permit us to decide whether they should be deemed polyphagous according to these criteria or not).To quantify the phylogenetic distribution of these states of polyphagy, we used both methods based on parsimony and on maximum likelihood (Pagel ), contrasting the results with those obtained from analyzing the distribution of host character states (the use of each order across the phylogeny) with the same methods. We also applied the same analyses to a data set from the bark beetle genus Dendroctonus (Kelley and Farrell ), which is of interest because it has been subject to discussion regarding whether it shows mostly transitions from a generalized to a more specialized host use or rather the reverse (Kelley and Farrell ; Nosil ; Stireman ).First, we investigated the phylogenetic signal in these characters, that is whether closely related species are more likely to share a character state than expected from chance. This was done in two ways: performing a randomization test (parsimony method) and calculating Pagel's λ (likelihood method). The randomization test was performed in Mesquite (Maddison and Maddison ), comparing the number of steps in each character for the given tree with 999 random trees created by reshuffling of terminal taxa. Lambda was estimated for each character using the fitDiscrete function in the Geiger package (Harmon et al. ) in R 2.13 (R Development Core Team ). Because branch lengths are not available for the complete phylogeny, we tested the effects of either rendering the tree ultrametric by transforming branch lengths using the “arbitrarily ultrametricize” function in Mesquite, or enforcing minimum node age constraints as described earlier. The results were qualitatively very similar and only results from the enforced minimum node ages are shown here.Second, as a complementary analysis, we investigated the degree of homoplasy and apicality in the characters, that is whether the same character state tends to evolve only once or rarely (more likely to result in the same state being shared by relatives and being optimized as going “deep” into the phylogeny), or repeatedly and convergently (more likely to result in an apical distribution of the state in the phylogeny). This was done in two ways: calculating the retention index (parsimony method) and the delta estimate (likelihood method). The retention index (Farris ) was calculated for each character as well as for the complete character matrix in Mesquite. Delta was estimated in the same way as Pagel's λ. Delta describes if change mostly occurs in the tips or deep in the tree (where values below 1 represents change mainly occurring toward the base of the tree and above 1 change tends to occur more at the tips of the tree).Third, we compared the frequencies of reconstructed transitions from monophagy to polyphagy or vice versa (using the “Summarize state changes in tree” option in Mesquite), reasoning that a higher frequency of transitions to polyphagy is indicative of an apical distribution of this character state. […]

Pipeline specifications

Software tools Mesquite, PHYSIG, GEIGER
Application Phylogenetics