Global epistasis emerges from a generic model of a complex trait

Gautam Reddy, Michael M. Desai

TL;DR

This paper proposes a theoretical framework showing that "diminishing-returns" and "increasing-costs" epistasis, observed in evolutionary experiments, are a natural consequence of widespread microscopic epistasis in complex traits. They introduce a novel "connectedness" model that explains how predictable evolutionary trajectories can emerge despite the inherent randomness of mutations.

What Problem Does This Paper Address?

Evolutionary experiments, particularly long-term studies of microbial populations, often show remarkably similar patterns of adaptation across replicate lines, even though the specific mutations that arise are largely random. This "convergent evolution" is often attributed to "diminishing-returns epistasis," where beneficial mutations become less effective on fitter genetic backgrounds, and "increasing-costs epistasis," where deleterious mutations become more harmful on fitter backgrounds. While these global patterns are observed, their underlying cause – how widespread microscopic (between-mutation) epistasis leads to these global effects – has remained unclear. This paper aims to bridge this gap by providing a theoretical explanation for the emergence of global epistasis from microscopic interactions.

Methods

The authors employ a Fourier analysis of fitness landscapes. They represent the fitness of a genotype as a sum of terms, where each term corresponds to the contribution of individual loci (additive effects) and combinations of loci (epistatic effects). This framework allows them to mathematically derive the fitness effect of a mutation on a given genetic background. A key insight is that when a locus interacts with many other loci (widespread epistasis), the fitness effect of a mutation at that locus tends to decrease linearly with the background fitness. They also introduce a novel "connectedness" (CN) model for a fitness landscape. In this model, loci are involved in "pathways," and interactions occur within these pathways. This model allows for varying degrees of epistasis across loci and provides a way to generate fitness landscapes with specific properties, including tunable ruggedness and exponentially decaying correlations.

Datasets

The authors reanalyzed existing experimental data from:

  • Johnson et al., 2019: This dataset measured the fitness effects of 91 insertion mutations in yeast across approximately 145 background genotypes.
  • Wiser et al., 2013 and Lenski et al., 2015: These refer to the E. coli long-term evolution experiment (LTEE).
  • Kryazhimskiy et al., 2014: This dataset involved parallel evolution experiments in yeast.

No new datasets were introduced.

Key Insights and Ideas

  • Global Epistasis from Microscopic Interactions: The most significant insight is that the observed global patterns of diminishing-returns and increasing-costs epistasis are not special cases but rather generic consequences of widespread microscopic epistasis in complex traits.
  • Variance Fraction (VF) as a Key Predictor: The authors introduce the concept of "variance fraction" (VF) for a locus, which quantifies its contribution to the total genetic variance. They show that the VF directly predicts the slope of the diminishing-returns/increasing-costs trend and the magnitude of random fluctuations around this trend.
  • "Connectedness" Model for Landscape Generation: The CN model provides a biologically intuitive way to generate fitness landscapes where loci can have vastly different VFs, explaining why some loci might exhibit stronger epistatic effects than others. This model links the organization of biological pathways to evolutionary dynamics.
  • Memoryless Adaptation under Specific Conditions: Under the assumptions of the CN model and specific conditions, the evolutionary trajectory can become "memoryless." This means that the DFE of new mutations depends only on the current fitness, not the entire evolutionary history, leading to predictable adaptation despite stochastic mutations.
  • Universal Beneficial DFE: In the well-adapted regime, the distribution of beneficial fitness effects (DFE) takes on a universal exponential form, independent of the specific details of the fitness landscape (like the DVF), and depends only on the organism's "adaptedness."

Key Results

  • The theoretical model predicts a linear relationship between mutation fitness effect and background fitness, with a negative slope for diminishing-returns and increasing-costs epistasis. This prediction was confirmed by reanalyzing existing experimental data (Figure 3).
  • The model quantifies the relationship between the magnitude of global epistasis (slope) and microscopic epistasis (residual variance), finding them to be proportional.
  • Simulations using the CN model successfully recapitulated experimental patterns of declining adaptability (Figure 2c).
  • The theory predicts that in the well-adapted regime, adaptation follows a power-law gain in fitness (t¹/⁵), with the number of substitutions growing as t²/⁵. This is consistent with observations from the E. coli LTEE.

Limitations and Future Work

  • Assumptions of Normality: The derivation of key equations relies on the assumption that fitness and fitness effects are jointly normally distributed across genotypes. While justified for large numbers of independent epistatic interactions, this might not hold in all biological systems.
  • SSWM Assumption: Some analyses, particularly regarding adaptation dynamics, assume strong-selection-weak-mutation (SSWM) dynamics. The authors acknowledge that real evolutionary dynamics can deviate from this, especially due to clonal interference.
  • Modular Models: While the CN model is analyzed, the authors mention that other models with modular organization might exhibit different adaptation dynamics, suggesting this as a direction for future work.
  • Experimental Verification of Double Mutants: The paper proposes testable predictions regarding the additivity of fitness effects for double mutants, which would require further experimental validation.
  • Understanding Deviations from SSWM: Further work is needed to understand how global epistasis interacts with breakdowns in SSWM assumptions, such as clonal interference, which could influence adaptation dynamics.

Summary updated

Abstract

Epistasis between mutations can make adaptation contingent on evolutionary history. Yet despite widespread 'microscopic' epistasis between the mutations involved, microbial evolution experiments show consistent patterns of fitness increase between replicate lines. Recent work shows that this consistency is driven in part by global patterns of diminishing-returns and increasing-costs epistasis, which make mutations systematically less beneficial (or more deleterious) on fitter genetic backgrounds. However, the origin of this 'global' epistasis remains unknown. Here, we show that diminishing-returns and increasing-costs epistasis emerge generically as a consequence of pervasive microscopic epistasis. Our model predicts a specific quantitative relationship between the magnitude of global epistasis and the stochastic effects of microscopic epistasis, which we confirm by reanalyzing existing data. We further show that the distribution of fitness effects takes on a universal form when epistasis is widespread and introduce a novel fitness landscape model to show how phenotypic evolution can be repeatable despite sequence-level stochasticity.
BibTeX
@inproceedings{reddy2021global,
  author = {Gautam Reddy and Michael M. Desai},
  title = {{Global epistasis emerges from a generic model of a complex trait}},
  booktitle = {eLife},
  year = {2021},
  doi = {10.7554/elife.64740},
}

Loading related content...

Figure