Abstract

with Sam F. Greenbury and Ard A. Louis In the biological literature fitness landscapes are often described in intuitive terms, as low-dimensional landscapes of the kind encountered in everyday experience with ‘peaks’ and ‘valleys’. Genotype space, however, is extremely high dimensional, resulting in counter-intuitive structural properties of genotype-phenotype maps. We present results showing that these structural properties, such as the existence of large neutral networks, make fitness landscapes navigable. For three biologically realistic models of genotype-phenotype maps — RNA secondary structure, protein tertiary structure, and protein complexes — we find that fitness maxima can be reached from almost any phenotype without passing through fitness valleys, even if phenotypes are assigned random fitnesses. This in turn indicates that fitness landscapes only contain very few true valleys. We also consider evolutionary simulations between pairs of functional RNA sequences and show that paths of monotonically increasing fitness are also likely to be used under evolutionary dynamics. Our findings can inform evolutionary predictions as well as directed evolution experiments.