Natural protein structures have evolved exceptional robustness to mutations

Protein structures are often conserved across widely divergent sequences, suggesting high mutational robustness. However, how such robustness emerges through evolution, and how it relates to the underlying sequence-structure map, remains poorly understood. In contrast, the mutational profiles (distribution of structures upon point mutation) of RNA secondary structures are well characterised, exhibiting both high mutational robustness and high evolvability through mutational access to diverse folds. The recent revolution in protein structure prediction now enables analagous large-scale analyses for proteins. Here, we use the structure prediction algorithm ESMFold to systematically investigate the mutational profiles of natural, random, and de novo proteins. Unlike RNA, where functional and random sequences share similar mutational profiles, natural proteins are substantially more robust than random amino acid sequences, suggesting an evolutionary drive toward robustness. They also exhibit limited structural variation among close sequence neighbours, potentially constraining access to new folds. Interestingly, many de novo proteins do resemble random sequences in their mutational profiles, with low robustness relative to established proteins. These findings reveal how gene duplication and de novo gene birth follow distinct evolutionary trajectories toward functional proteins and highlight a potential role for large-effect mutations in the emergence of structural complexity.