Fitness seascapes facilitate the prediction of therapy resistance under time-varying selection

ABSTRACT

Predicting the evolution of drug resistance in infectious diseases may enable us to make rational drug choices to avoid resistance and exploit known molecular mechanisms of resistance. Fitness landscapes are commonly used in computational studies to model the genotype-fitness mapping. Canonical fitness landscapes do not intrinsically model varying selection pressure. However, a challenge to predicting the emergence of drug resistance is that disease agents in a patient will never experience a constant environment – the selection pressure imposed by a drug will vary as a result of the drug pharmacokinetic profile, dosing schedule, and spatial concentration gradient. Furthermore, because of evolutionary costs of resistance and different levels of genotype-specific resistance, different drug concentrations may be optimal for the fitness of different genotypes. Explicit modeling of these trade-offs and the resulting rank-order changes in fitness may be important to accurately predict the evolution of a disease population within a patient or other heterogeneous environments. Fitness seascapes extend the fitness landscape model by allowing the mapping to vary according to an arbitrary environmental parameter (e.g., drug concentration), allowing us to model the evolution of resistance with realistic pharmacological considerations. Here, we explore the importance and utility of fitness seascapes in predicting the emergence of resistance. First, we show how modeling genotype-specific dose response curves is necessary to accurately predict evolutionary outcomes in changing environments. Then, using an empirical fitness seascape measured in engineered E. coli, we performed computational experiments observing the impact of the rate of change in drug concentration and simulated patient nonadherence on the probability of evolutionary escape. We found that a greater rate of change in drug concentration resulted in a lower rate of resistance, or a lower rate of evolutionary escape. In simulated patients, higher rates of drug regimen nonadherence were associated with greater rates of resistance. Our work integrates an empirical fitness seascape into an evolutionary model with realistic pharmacological considerations. Future work may leverage this platform to optimize dosing regimens or design adaptive therapies to avoid resistance.