Evaluating the impact and detectability of mass extinctions on total-evidence dating

Background

Biological diversity is continually generated as new species originate, and continually lost through the process of extinction. By representing the order and timing of branching events in the tree-of-life, dated phylogenetic trees underpin knowledge of when, how quickly, and in what environment, different lineages originated. Often, dated phylogenetic trees are estimated from molecular sequence data. This is incredibly powerful given the quantity of molecular data we can generate and the effective ways in which we can model its evolution. However, molecular data cannot be extracted from extinct organisms. What therefore can dated phylogenetic trees tell us about extinction?

This topic has been discussed extensively, and there is good evidence that signatures of extinction can be detected in dated phylogenetic trees, even when only extant species are sampled. Further, the fossilised-birth-death process (FBDP) now enables analyses that explicitly integrate fossils, including the process by which fossils are sampled, with phylogenetic trees. Numerous studies have used the FBDP to analyse both extinction and mass extinction in dated phylogenetic trees. However, previous studies that use the FBDP to test for mass extinction use dated phylogenetic trees that have already been estimated.

Often these dated phylogenetic trees have been estimated with a simpler version of the FBDP (or an even simpler birth-death branching process) that assumes speciation and extinction rates are constant. In this preprint, Du and colleagues investigate whether using methods that assume constant extinction when estimating dated phylogenetic trees impacts our ability to detect mass extinction in subsequent analyses. This has the potential to fundamentally impact how we interpret some of the most catastrophic events in evolutionary history.

Key findings

Du and colleagues analysed simulated and empirical datasets to assess how the method for estimating a dated phylogenetic tree impacts our ability to detect mass extinction events. First, they simulated fossils, molecular data, and morphological data on birth death branching processes with 50 species (Fig. 1). These birth death branching processes either had a constant extinction rate or a mass extinction event 30 million-years-ago (Ma). With the simulated data, they then estimated dated phylogenetic trees. They compared several different approaches for estimating the dated phylogenetic trees, but the most important contrast was between two main types of models; 1) a model that explicitly incorporates a mass extinction event (in the tree prior), and 2) a model that has constant extinction rates.

Figure 1. Both a depiction of a mass extinction event and a summary of the type of simulation performed by the authors. A birth-death branching process is shown by the black lines. There are extant species i.e. those branches that reach the present (50 in the actual simulation), and extinct species i.e. those for which the lineage ends at the mass extinction event (indicated by the grey bar). Molecular and morphological data are simulated along the branches of the branching process. Fossils (pink circles) are sampled along the branches of the branching process. The simulated molecular data, morphological data, and fossils, are then used to estimate a dated phylogenetic tree. The authors perform an alternative simulation where the extinction rate is constant.

When they compare these models for estimating the dated phylogenetic tree, the model that matches how the data was simulated is consistently favoured over alternative models i.e. when the underlying data is simulated with a mass extinction event there is support for the model with the mass extinction event, and when data is simulated without a mass extinction event there is support for the model without the mass extinction event. This result is consistent across different simulations, although support for the model with mass extinction (in datasets simulated with mass extinction) increased when there were more lineages present before the mass extinction event. This makes sense, when there are more lineages, there is more data (in the form of branches leading to the mass extinction event) upon which to detect mass extinction.

However, whether or not the dated phylogenetic tree was estimated with a model containing or lacking a mass extinction event had little impact on the estimated tree topology (the order of branching events) or divergence times (when, in the past, the branching events are estimated to have happened). This was the case regardless of whether the data was simulated with a mass extinction event.

The authors also investigated how the method of estimating the dated phylogenetic tree (i.e. whether or not it was estimated with a model that included a mass extinction event) affected whether a mass extinction event could subsequently be detected on the dated phylogenetic tree. In other words, they compared the fit of models with or without a mass extinction event to dated phylogenetic trees that were already estimated. Interestingly, the method for estimating the tree had little impact on whether or not a mass extinction event was detected.

These findings were reflected in analyses of tetraodontiform fish and crinoids. Although both clades are likely to have been affected by mass extinctions, including mass extinction in the model when estimating dated phylogenetic trees had negligible impact on the estimated topology or divergence times.

Importance

Estimates of dated phylogenetic trees are, by their very nature, extremely sensitive to the assumptions we make about evolution and the fossil record. This dependence is important and creates circularity in how we perform macroevolutionary analyses with dated phylogenetic trees i.e. dated trees are highly sensitive to our assumptions, but a key way we aim to become better informed about evolution is through analysing dated phylogenetic trees. Du et al.’s study is therefore a striking contribution that highlights our ability to estimate whether a clade is impacted by mass extinction is only minimally influenced by the initial assumptions that we make about mass extinction when estimating the dated phylogenetic tree.

Future directions and questions for the authors

However, this study does not overturn the fact that divergence time estimation is extremely sensitive to assumptions. Character data, whether molecular or morphological, only tells us how different things are from each other. Therefore, our estimates of dated phylogenetic trees must at some level be sensitive to our assumptions about molecular/morphological evolutionary rates, speciation or extinction rates (which govern our expectation of how evolutionary branching events are distributed through time), and the rate that fossils are sampled. How then are we getting divergence time estimates that are the same regardless of the model used? Do our assumptions about molecular rates override the assumptions we make about mass extinction (perhaps because there is so much molecular data)? Perhaps, in these relatively small simulations (50 species), the assumptions we make about speciation/extinction rates (including mass extinction) when estimating the dated phylogenetic tree don’t really matter? In that vein, it’s interesting that when there are more lineages at the time of the mass extinction there is a stronger signal for mass extinction. If you increased the number of lineages further (for example by simulating bigger branching processes) would the signal for the mass extinction event become even stronger? And in such situations would the resulting dated phylogenetic tree show more sensitivity to whether or not a mass extinction event was included in the model used to estimate the tree?

 

doi: https://doi.org/10.1242/prelights.41969

Read preprint (No Ratings Yet)