Protein structure shapes natural genetic variation and <i>de novo</i> adaptation

Protein structural context strongly influences the functional outcomes of mutations, but exploiting this intuition to interpret the consequences of natural variants has been limited by the paucity of structural information. Here, we show that simple molecular heuristics based on predicted protein structures-solvent accessibility and the local complexity of the protein fold-are strong predictors of protein-coding variation in genomes from yeast to plants to humans. Polymorphisms in structured and unstructured regions alike were subject to these constraints, revealing fine-grained selection across amino acid residues and secondary structures. The same analyses distinguished deleterious variants in datasets from existing deep mutational scans. Extending the metrics to <i>de novo</i> adaptation, we developed a massively parallel directed evolution platform that isolated hundreds of independent adapted lineages within days. Optimized experimental conditions revealed dozens of rapamycin-resistant alleles of the <i>FPR1</i>/FKBP gene in a single experiment-resistance mutations that matched predictions from our structural heuristics. Together, these data demonstrate the power of a biochemical perspective to contextualize the functional landscape of both existing and newly arising genetic variation.