Fixation and absorption in a fluctuating environment

TL;DR

This paper develops a mathematical model to analyze how fluctuating environmental conditions affect the probability and timing of mutant fixation or extinction in finite populations, extending classical neutral and constant selection models.

Contribution

It introduces a generic model incorporating demographic noise and fluctuating selection, providing asymptotic formulas for fixation probability and times that interpolate between known limits.

Findings

Formulas for fixation probability and times under fluctuating selection

Model captures transition between neutral and constant selection regimes

Asymptotic results valid for large populations

Abstract

A fundamental problem in the fields of population genetics, evolution, and community ecology, is the fate of a single mutant, or invader, introduced in a finite population of wild types. For a fixed-size community of $N$ individuals, with Markovian, zero-sum dynamics driven by stochastic birth-death events, the mutant population eventually reaches either fixation or extinction. The classical analysis, provided by Kimura and his coworkers, is focused on the neutral case, [where the dynamics is only due to demographic stochasticity (drift)], and on \emph{time-independent} selective forces (deleterious/beneficial mutation). However, both theoretical arguments and empirical analyses suggest that in many cases the selective forces fluctuate in time (temporal environmental stochasticity). Here we consider a generic model for a system with demographic noise and fluctuating selection. Our system is characterized by the time-averaged (log)-fitness $s_0$ and zero-mean fitness fluctuations. These fluctuations, in turn, are parameterized by their amplitude $\gamma$ and their correlation time $\delta$. We provide asymptotic (large $N$) formulas for the chance of fixation, the mean time to fixation and the mean time to absorption. Our expressions interpolate correctly between the constant selection limit $\gamma \to 0$ and the time-averaged neutral case $s_0=0$.