Mutation choice biases the structure of empirical fitness landscapes

Evolution occurs in a fitness landscape where each point represents a specific genetic makeup, mapped to a fitness value. Researchers have long endeavored to understand the statistical traits of fitness landscapes influenced by the level and type of interactions between genetic locations, collectively called epistasis. Recent high-throughput experimental methods can now measure the characteristics of thousands of genetic variants, bridging theory and observation.

However, these experiments still represent a small fraction of the overall genetic possibilities. In particular, many studies focus on a non-random subset of the fitness landscape. For instance, one common method generates all possible genetic intermediates between ancestral and evolved sequences differing by a modest number of mutations.

This study investigates the extent to which empirical fitness landscapes created under different selection methods reflect the overall landscape's statistical properties in a few well-studied random landscape models. The findings reveal that when mutations are chosen from an adaptive path, the variance explained by lower-order epistasis increases, local peaks decrease, and accessible paths expand compared to a random fitness landscape.

Conversely, selecting the most advantageous or harmful mutations leads to distortions in these statistics, with the degree of distortion depending more on the overall level of epistasis in the global landscape. Finally, the researchers utilized several large-scale mutagenesis datasets to simulate the impact of selection in more complex scenarios.