Molecular phenotypes as key intermediates in mapping genotypes to fitness

We argue that the staggering complexity of the relationship between the organism's genomic sequence and its evolutionary success, as measured by organismal fitness, can become more tractable when viewed through the lens of phenotypic features. These phenotypic features can refer to molecular properties of a protein or protein complex or represent morphological characteristics such as body size and beak shape in birds. Using protein evolution as an example, we demonstrate that it is possible to describe phenotypic landscapes, in which every protein sequence is associated with a phenotypic value such as free energy of protein folding, using compact and interpretable models that can be learned from relatively small-scale datasets. The predicted phenotypic values then serve as explicit inputs to a model of organismal fitness, with the functional form of the fitness function given by biophysical considerations or learned from evolutionary data (fitness measurements for a collection of genotypes). Thus, instead of being a high-dimensional function of genotypes, fitness becomes a low-dimensional function of one or several phenotypes, making it much easier to visualize fitness landscapes and analyze their properties. Moreover, evolutionary dynamics on such landscapes can be decomposed into generation of phenotypic differences by mutations and subsequent motion on the low-dimensional fitness landscape. This two-tiered approach, made possible by recent advances in high-throughput molecular biology, may hold a key to better understanding of the evolutionary processes that have shaped, and continue to shape, all life on Earth.