Bayesian Total-Evidence Dating Reveals the Recent Crown Radiation of Penguins

Abstract

The total-evidence approach to divergence time dating uses molecular and morphological data from extant and fossil species to infer phylogenetic relationships, species divergence times, and macroevolutionary parameters in a single coherent framework. Current model-based implementations of this approach lack an appropriate model for the tree describing the diversification and fossilization process and can produce estimates that lead to erroneous conclusions. We address this shortcoming by providing a total-evidence method implemented in a Bayesian framework. This approach uses a mechanistic tree prior to describe the underlying diversification process that generated the tree of extant and fossil taxa. Previous attempts to apply the total-evidence approach have used tree priors that do not account for the possibility that fossil samples may be direct ancestors of other samples, that is, ancestors of fossil or extant species or of clades. The fossilized birth–death (FBD) process explicitly models the diversification, fossilization, and sampling processes and naturally allows for sampled ancestors. This model was recently applied to estimate divergence times based on molecular data and fossil occurrence dates. We incorporate the FBD model and a model of morphological trait evolution into a Bayesian total-evidence approach to dating species phylogenies. We apply this method to extant and fossil penguins and show that the modern penguins radiated much more recently than has been previously estimated, with the basal divergence in the crown clade occurring at ∼12.7 ∼12.7 Ma and most splits leading to extant species occurring in the last 2 myr. Our results demonstrate that including stem-fossil diversity can greatly improve the estimates of the divergence times of crown taxa. The method is available in BEAST2 (version 2.4) software www.beast2.org with packages SA (version at least 1.1.4) and morph-models (version at least 1.0.4) installed.

Keywords: Birth–death process; calibration; divergence times; MCMC; phylogenetics.

Figure 1.

Posterior probabilities of fossils to be sampled ancestors for eight models summarized in Table 2. In the legend, P stands for the partitioned model, G for gamma variation across characters, Mkv for conditioning on variable characters, R for relaxed clock model, dns for , , and tree prior parameterization, lmp for , , and tree prior parameterization, and the numbers correspond to analyses in Table 2.

Figure 2.

The posterior and posterior predictive distributions for the tree length, , and genealogical statistics on the left and tree imbalance statistic and Colless’s tree imbalance index, , on the right for model 8 in Table 2. The plots do not show obvious discrepancy in the posterior and posterior predictive distributions of these statistics. The posterior predictive distribution for the branch length related statistics ( and ) is more diffuse than the posterior distribution although both distributions concentrate around the same area. The distributions of the tree imbalance statistics almost coincide.

Figure 3.

An MSACC tree for the total-evidence analysis. The numbers at the bases of clades show the posterior probabilities of the clades. The filled circles represent sampled ancestors. Fossils with positive evidence of being sampled ancestors are shown in red (gray in printed version). Fossils Paraptenodytes antarcticus and Palaeospheniscus patagonicus both appear around the same time and have the same prior probabilities of 0.42 of being sampled ancestors but the morphological data provides positive evidence for the former to belong to a terminal lineage and for the latter to be a sampled ancestor. Penguin reconstructions used with permission from the artists: fossil species by Stephanie Abramowicz and extant penguin species by Barbara Harmon.

Figure 4.

The evidence for fossils to be sampled ancestors. The samples above the shaded area (i.e., with log Bayes Factors greater than 1) have positive evidence to be sampled ancestors and below the shaded area (log Bayes factors lower than ) have positive evidence to be terminal nodes.

Figure 5.

The maximum clade credibility tree of extant penguins with common ancestor ages. The bars are the 95% HPD intervals for the divergence times. The mean estimates and 95% HPD intervals are summarized in Supplementary Table 4 available on dryad. The numbers at the bases of clades show the posterior probabilities of the clades (after removing fossil samples).

Figure 6.

The ages of the MRCAs of the extant penguin taxa for the eight analyses (Table 2) of morphological data, total-evidence analysis with all fossils under relaxed (8DNA R) and strict (8DNA S) molecular clocks, total-evidence analysis under the birth–death model without sampled ancestors (8BDDNA R), and total-evidence analysis with crown fossils only. Abbreviations in the names of analyses are the same as in Figure 1.