Phylogenetic Clustering of Extinction: Insights from Empirical and Simulated Data

UoM administered thesis: Master of Philosophy

  • Authors:
  • Adam Dickson


Extinction often demonstrates a phylogenetically non-random pattern, and extinction of closely related taxa can be quantified by treating extinction as a binary trait. Phylogenetic clustering of extinction has been observed with regards to stochastically simulated phylogenies, animals at risk of extinction (Fritz and Purvis, 2010), and in the fossil record (Hardy et al., 2012; Soul and Friedman, 2017). Nonetheless, there remains a paucity of research into how phylogenetic clustering acts in non-stochastic simulations of evolution, and how this compares to empirical data. Here, D is used (Fritz and Purvis, 2010) to assess the clustering of extinct taxa in simulated and empirical phylogenies, treating extinction as a binary trait. TREvoSim, a novel, selection-based method for simulating the evolution of binary traits, is used to simulate evolutionary lineages, and explore how alterations to the model parameters, including rate of environmental mutation, amount of selection, and number of taxa affect the phylogenetic clustering of extinction within the simulated trees. An empirical dataset of 30 published phylogenies was collected from a diverse array of animal groups, for comparison to the simulated data. Packages within RStudio were utilised to measure the phylogenetic clustering of extinction (D) and tree symmetry (Colless' Index; Ic) of both the simulated and empirical data. The results of these tests were used as a comparison between the simulated and empirical data, in order to understand the similarities and differences between the two data types. The conducted experiments show that TREvoSim produces phylogenies prone to high clustering of extinct taxa (~ -2 D), compared to the empirical phylogenies that demonstrate clustering close to a Brownian Motion (stochastic) assumption of extinction (~ 0 D). The phylogenies produced by TREvoSim are more asymmetric than the empirical phylogenies, demonstrated by high values of tree imbalance (~ 1 Ic). It is hypothesised that the high values of extinction clustering and tree asymmetry in the TREvoSim data is an artefact of rapid phyletic evolution over a short time scale, when compared to the deep-time empirical phylogenies, which show a more cladogenetic pattern. The empirical phylogenies may also be affected by sampling biases, an issue that TREvoSim does not face. Furthermore, as the simulated and empirical data operate on different time scales, it may be the case that we are observing the same patterns, but at different time scales. The tendency for TREvoSim to produce extremely clustered extinctions and asymmetrical trees may also be explained by the biologically unrealistic assumptions of TREvoSim; exclusively asexual reproduction and sympatric speciation, lack of ecology and lack of mass extinction events. This may imply that it is unsuitable to compare phylogenetic clustering of extinction in TREvoSim to that of empirical phylogenies, due to the biologically unrealistic assumptions of the model. Future work should focus on simulating for longer periods of time, and incorporating artificial sampling biases, to explore the impact of sampling bias and time scales on tree shape and the phylogenetic clustering of extinction. Furthermore, comparison of empirical data to selection-based models of evolution that incorporate sexual reproduction, allo/parapatric speciation, ecologies, and mass extinctions could alleviate the restrictions within TREvoSim.


Original languageEnglish
Awarding Institution
Award date1 Aug 2021

Related information