The ABCs of Experimental Evolution

Microbial evolution is complex and is influenced by many sources of variation. Experimental evolution is no exception, although it is more controlled, easily replicated, and typically devoid of interactions between species. Mathematical modeling of the evolutionary process can help in understanding the underlying mechanisms that drive outcome of such experiments. These models can be complex and parameter rich, limiting their feasibility for statistical inference. In this paper, we introduce the use of Approximate Bayesian Computation (ABC) as a tool for statistical inference in the study of experimental evolution. ABC is a fast and simple method for fitting complex models to data. We utilize this method, coupled with a mechanistic model of experimental evolution, to study the evolution process of bacteriophage ？X174 under benign selection pressure. Our results highlight three mutation-selection scenarios that could explain this process: high mutation/low selection pressure, low mutation/high selection pressure, and low mutation/low selection pressure, with posterior support of 19%, 9.5%, and 71.5% for each of these scenarios, respectively. Sequence data support the first candidate. Though surprising, this scenario was not improbable based on our analysis. 1. Introduction Mathematical modeling and statistical inference can aid in understanding the underlying evolutionary mechanisms that drive the outcome of evolution experiments (see, e.g., [1–3]). However, modeling and inference attempts are challenged by the complexity of the evolutionary process. This complexity stems from factors such as the stochastic nature of the experimental evolution process, the possibility of clonal interference (interactions between evolving genotypes), and experimental error. To address some of these challenges, we introduce a model of evolution that mimics the experimental evolution process in serial batch cultures. This model accounts for mutation, selection, drift and clonal interference. It also accounts for variability in the observed data resulting from experimental and sampling errors. Fitting the aforementioned model to estimate evolutionary parameters of interest is complicated by the need to integrate over all possible experimental evolution trajectories. This is a computationally expensive process similar to integrating over all possible genealogies to estimate evolutionary parameters in coalescent-based population genetics [4, 5]. We borrow methodology from the coalescent literature and propose an Approximate Bayesian Computation (ABC) framework to fit this model