Computational protocol: The abiotic and biotic drivers of rapid diversification in Andean bellflowers (Campanulaceae)

Similar protocols

Protocol publication

[…] Phylogenies were inferred using maximum likelihood (ML) and Bayesian inference as implemented in RAxML 8.0 (Stamatakis, ) and Beast 2.1.3 (Bouckaert et al., ). Four calibration points were used to estimate divergence times: the fossil seed of †Campanula paleopyramidalis as a minimum age constraint of 16 Myr for the most recent common ancestor of C. pyramidalis and C. carpatica (Cellinese et al., ; Crowl et al., ); a geological maximum age constraint of 29.8 Myr, which corresponds to the age of the Kure atoll, the oldest emerged island of the Hawaiian Ridge (Clague, ), for the crown group of the endemic Hawaiian clade, encompassing c. 125 spp. in six genera represented in our sampling by Brighamia, Cyanea, Clermontia and Delissea (Antonelli, ); and two secondary age constraints from Bell et al. (): 41–67 Myr ago for crown group Campanulaceae, and 28–56 Myr ago for crown group Campanuloideae. The fossil record for Campanulaceae is very poor, and †C. paleopyramidalis, described from Miocene deposits in the Nowy Sacz Basin in the West Carpathians of Poland (Łancucka‐Srodoniowa, ), is the only fossil appropriate for calibration of divergence time estimates in this family (Antonelli, ; Cellinese et al., ). It is assigned as a close relative of C. pyramidalis on the basis of their shared reticulate seed coats, a feature uncommon in the genus (Łancucka‐Srodoniowa, ; Cellinese et al., ).The optimal RAxML tree was dated using penalized likelihood in treePL (Smith & O'Meara, ) with hard minimum and maximum age constraints. Confidence intervals were generated using 1000 RAxML bootstrap trees. We then simultaneously inferred phylogenetic relationships and divergence times using both relaxed uncorrelated lognormal and exponential clock models in Beast. A lognormal prior (mean of 1.5 and SD of 1.0) was assigned to the fossil calibration age of 16 Myr, a uniform prior was assigned to the geological age constraint at 29.8 Myr (maximum hard bound), and normal priors were placed on the two secondary age constraints (mean age of 56.0 Myr and SD of 8.3 Myr for Campanulaceae; mean age of 43.0 Myr and SD of 8.0 Myr for Campanuloideae). The dated tree from treePL was specified as the starting tree in each of eight separate Beast runs, which were each conducted for 10 million generations of Markov chain Monte Carlo (MCMC). Convergence was assessed using effective sample size (ESS) values of the runs in Tracer 1.6 (Rambaut et al., ), applying a cutoff value of 200. The maximum clade credibility tree, including credibility intervals (CIs) for ages and posterior probabilities (PPs) for node support, was then assembled using TreeAnnotator (Bouckaert et al., ).Robustness of age estimates was assessed by removing one or a series of calibration points. The following sets of calibration points were used for this purpose: the Campanulaceae secondary constraint only; the fossil and the Campanulaceae secondary age constraint; and the fossil and both secondary age constraints. Divergence time estimation can be sensitive to branch length variation, which is potentially influenced by growth form (Smith & Donoghue, ). As they are generally herbaceous and have longer internal branches than their woody relatives, we suspected this may be the case for Lysipomia. We thus removed Lysipomia species entirely, and re‐estimated molecular divergence times to determine if these differences had an impact on results. We find no evidence that the results are affected by dating method, calibration strategy, or branch length heterogeneity: the 95% CIs of the re‐estimated ages overlap with the 95% CIs from the Beast analysis that we use for diversification analyses (see also Table S2). [...] We modeled the impact of traits on the diversification of Neotropical bellflowers by concurrently estimating their impact on speciation, extinction, and transition rates using BiSSE (Maddison et al., ). For each of the four traits (Andean occurrence, elevation, pollination syndrome, and fruit type), we evaluated eight BiSSE diversification models of increasing complexity in which speciation, extinction, and transition rates were allowed to either vary or remain equal between traits (Table S5). Analyses were performed using the R package diversitree 0.7‐6 (FitzJohn, ). Once the best‐fitting model was selected, CIs for each parameter were estimated for the tree. We used an exponential prior following FitzJohn (), and began the MCMC with the ML estimates. We ran MCMC for 20 000 generations and applied a burn‐in of 2000 steps. We then computed the net diversification rates for each trait. Finally, to determine if these traits were correlated, pairwise trait comparisons were performed across a random sampling of 500 trees from the posterior distribution in both a ML and Bayesian framework in BayesTraits 2.0 (Pagel & Meade, ). The significance of binary trait dependence was assessed against a rival model where these traits evolved independently using likelihood ratio tests (ML) and Bayes factors of the harmonic mean of likelihood values.Binary state‐speciation and extinction models have received recent criticism, including high type I error rates (Maddison & FitzJohn, ; Rabosky & Goldberg, ), especially for trees with fewer than 300 terminals and for traits present in < 10% of taxa (Davis et al., ). The distribution of characters states is also important; for example, rapid diversification rates in a region of a phylogeny can be erroneously attributed to a particular character state when the trait is characterized by high transition rates (Rabosky & Goldberg, ). However, despite these known issues, SSE models remain a viable framework to test the effect of particular traits on species diversification, particularly when these caveats have been mitigated (Ng & Smith, ), as we have attempted to do here. Where possible, we also compare the trait‐dependent results with the inferences made with nontrait‐dependent models (i.e. BAMM and RPANDA) to show that that they are consistent. […]

Pipeline specifications

Software tools RAxML, BEAST, Diversitree, BayesTraits
Application Phylogenetics