Limiting fitness distributions in evolutionary dynamics

Darwinian evolution can be modeled in general terms as a flow in the space of fitness (i.e. reproductive rate) distributions. In the diffusion approximation, Tsimring et al. have showed that this flow admits “fitness wave” solutions: Gaussian-shape fitness distributions moving towards higher fitness values at constant speed. Here we show more generally that evolving fitness distributions are attracted to a one-parameter family of distributions with a fixed parabolic relationship between skewness and kurtosis. Unlike fitness waves, this statistical pattern encompasses both positive and negative (a.k.a. purifying) selection and is not restricted to rapidly adapting populations. Moreover we find that the mean fitness of a population under the selection of pre-existing variation is a power-law function of time, as observed in microbiological evolution experiments but at variance with fitness wave theory. At the conceptual level, our results can be viewed as the resolution of the “dynamic insufficiency” of Fisher’s fundamental theorem of natural selection. Our predictions are in good agreement with numerical simulations.

💡 Research Summary

This paper presents a novel theoretical framework for understanding Darwinian evolution by modeling it as a dynamical flow in the space of fitness distributions. The central finding is that evolving fitness distributions are universally attracted to a specific one-parameter family of distribution types, characterized by a fixed parabolic relationship between skewness and kurtosis.

The authors begin by critiquing the established “fitness wave” theory by Tsimring et al., which describes evolution as a traveling Gaussian wave of fitness. They argue that this theory, derived under a “diffusion approximation” requiring infinitely frequent mutations with infinitesimal effects, is limited. It primarily describes phases of rapid positive selection and fails to encompass negative (purifying) selection, a fundamental evolutionary force.

To develop a more general theory, the authors start from the replicator-mutator equation in continuous fitness space (Eq. 1), which incorporates selection and mutation with a given distribution of fitness effects (DFE). A key innovation is reformulating this equation in terms of the cumulant generating function (CGF) ψ_t(s) of the fitness distribution. This yields an exact, closed-form solution (Eq. 2), interpretable as transport of the initial CGF along characteristics in (t,s)-space, plus a mutational source term (Fig. 1).

This solution naturally separates two evolutionary regimes: the “selection of pre-existing variation” (Eq. 3), dominant when mutations are rare or initial diversity is high; and the “selection of new mutations” (Eq. 4), dominant when mutations are frequent.

The paper’s core contribution is a set of four limit theorems derived from asymptotic analysis of these regimes:

1. Theorem 1 (Positive selection of pre-existing variation): If the initial fitness distribution has an unbounded right tail (enabling positive selection), the fitness distribution converges to a normal type. The mean fitness (μ_t) and variance (σ²_t) scale as power-laws in time: μ_t ~ t^(α-1) and σ²_t ~ t^(α-2), where α is related to the tail thickness of the initial distribution.
2. Theorem 2 (Negative selection of pre-existing variation): If the initial distribution has a finite maximum fitness, it converges in type to a “reversed” Gamma distribution with shape parameter β. This family interpolates between a reversed exponential (β=0) and a normal distribution (β→∞), unifying positive and negative selection within a continuous spectrum (Fig. 2). The variance decays as σ²_t ~ t^(-2).
3. Theorem 3 (Positive selection of new mutations): If beneficial mutations exist (Δ+ > 0) and evolution is driven by new mutations, the fitness distribution converges to the normal type independently of the detailed shape of the DFE, demonstrating strong universality.
4. Theorem 4 (Mutation-selection balance): If all mutations are deleterious or neutral (Δ+ ≤ 0), the distribution reaches a mutation-selection balance. The equilibrium mean fitness is μ_∞ = x_+ - U, independent of the DFE’s shape (a result noted by Eshel). The equilibrium distribution becomes normal only in the limit of high mutation rates or very small fitness effects.

Theoretical implications are profound. The authors frame their results as resolving the “dynamic insufficiency” of Fisher’s fundamental theorem of natural selection: while Fisher’s theorem gives the rate of change of mean fitness, this work specifies the dynamical attractors for the full distribution of fitness. Empirically, the power-law increase of mean fitness predicted by Theorem 1 aligns with observations in microbial evolution experiments but contrasts with the constant speed predicted by fitness wave theory.

The paper validates all theoretical predictions with numerical simulations of the Wright-Fisher process and a simple genetic algorithm, showing excellent agreement. In summary, this work uncovers universal statistical patterns in evolutionary dynamics that transcend specific biological details, providing a general and unifying limit theory for the dynamics of fitness distributions under selection and mutation.