The role of epistasis in determining the fitness landscape of HIV proteins

Abstract

The rapid evolution of HIV is constrained by interactions between mutations which affect viral fitness. In this work, we explore the role of epistasis in determining the fitness landscape of HIV for multiple drug target proteins, including Protease, Reverse Transcriptase, and Integrase. Epistatic interactions between residues modulate the mutation patterns involved in drug resistance with unambiguous signatures of epistasis best seen in the comparison of a maximum entropy sequence co-variation (Potts) model predicted and experimental HIV sequence "prevalences" when expressed as higher-order marginals (beyond triplets) of the sequence probability distribution. In contrast, the evidence for epistasis based on experimental measures of fitness such as replicative capacity is weak; the correspondence with Potts model "prevalence"-based predictions is obscured by site conservation and limited precision. Double mutant cycles provide in principle one of the best ways to probe epistatic interactions experimentally without reference to a particular background, and we find they reveal that the most strongly interacting mutations in HIV involve correlated sets of drug-resistance-associated residues, however the analysis is complicated by the small dynamic range of measurements. The use of correlated models for the design of experiments to probe viral fitness can help identify the epistatic interactions involved in mutational escape, and lead to better inhibitor therapies. Author summaryProtein covariation models provide an alternative to experimental measures for estimating the fitness of mutations in proteins from across a variety of organisms. Yet, for viral proteins, it has been shown that models including epistatic couplings between residues, or other machine learning models perform no better or even worse than a simpler independent model devoid of such epistatic couplings in estimating viral fitness measurements such as replicative capacities, providing weak or ambiguous evidence for epistasis. We show that the evidence for long-range epistasis is strong by the analysis of the high-order marginals of the MSA distribution (up to subsequences of length 14), which are accurately captured by a correlated Potts sequence-covariation model but not by an independent model. While double mutant cycles in principle provide well-established biophysical probes for epistatic interactions, we demonstrate that the analysis and comparison between model and experiment is difficult due to the much smaller dynamic range of the measurements, making them more susceptible to noise.