Using a simplified Rough Mount Fuji model to disentangle how multi-peaked fitness landscapes can be highly navigable

Abstract

Evolving populations, especially in the strong-selection-weak-mutation limit, can be modelled as adaptive walks on fitness landscapes, moving in fitness-increasing mutational steps until reaching a fitness peak--a local optimum. Simulations of such adaptive walks--on a multi-peaked empirical landscape of the folA gene and on landscapes generated by the Rough Mount Fuji (RMF) model-- have shown that some landscapes are highly navigable, meaning that the highest x% of peaks are reached by >> x% of adaptive walks. This prompts the question of how adaptive walks can be so successful despite the local, myopic rules behind each adaptive step. Here, we investigate this question using simulations and mathematical approximations of random adaptive walks on a simplified RMF landscape. The landscape has a low-to-intermediate fitness region, whose size reconciles a low peak density with a high peak number. Despite the high number of peaks, walkers are likely to exit this region without terminating at a peak because the probability of a peak transition at each step is low and a fitness gradient guides walkers to the high-fitness region in few steps. Thus, three features are sufficient to explain why adaptive walks in the simplified RMF landscape are likely to reach a small fraction of top-ranking peaks: a low-to-intermediate fitness region with a high number of peaks, a low peak-transition probability, and which is crossed in few steps. We find that these three features are also present in the empirical folA landscape, suggesting that similar principles may apply.