Abstract

A protein’s genetic architecture – the set of causal rules by which its sequence determines its function – also determines the effects of mutations and thus the possible evolutionary routes a protein may take. Prior work suggests that the genetic architectures of proteins are complex, with large amounts of high order epistasis that constrains evolution and limits the ability to predict protein function from sequence. However, prior work may have overstated both the extent and impact of epistasis by analyzing genetic architecture from the perspective of a single reference genotype and failing to fully account for global nonlinearities – both of which can artificially inflate estimates of epistasis – and by considering only a single protein function and direct evolutionary paths between pairs of proteins – both of which make epistasis a constraint on evolution. Here I will describe a reference-free method for inferring protein genetic architecture from combinatorial deep mutational scanning datasets that accounts for global nonlinearities. Applying this approach to 20 previously collected datasets reveals that the genetic architecture of most proteins studied to date are simple: main and pairwise interactions among amino acids, along with a simple nonlinear correction, explains a median of 96% of phenotypic variance (>92% in every case). We further used this approach to dissect the genetic architecture and evolution of a transcription factor’s specificity for DNA by combining deep mutational scanning with ancestral protein reconstruction. As before, the genetic architecture was simple, with few high order interactions and many main and pairwise interactions instead. However, these pairwise interactions massively expanded the number of opportunities for single-residue mutations to switch specificity from one DNA element to another. By bringing variants with different specificities close together in sequence space, pairwise epistatic interactions can thus facilitate the evolution of new molecular functions. By reorienting how we estimate epistasis, reference-free analyses can reveal simple and intelligible protein genetic architectures and thus provide an experimentally and analytically tractable route forward for understanding protein genetic architecture and its evolution.