Folding without glutamine: Proteome-scale removal of a canonical amino acid

All known life uses a canonical alphabet of 20 amino acids, yet which residues are structurally essential remains untested at proteome scale. Isoleucine was recently eliminated from ribosomal proteins in Escherichia coli , establishing that amino acid alphabet reduction is feasible. Here we ask whether an entire amino acid can be eliminated from a proteome, and develop a computational framework to identify the best target. Using two independent approaches — ProteinMPNN structure-based probabilities and multiple sequence alignment conservation — we define “keep-score” as the fraction of positions where the native residue is optimal. Both methods converge on an unexpected candidate: glutamine has the lowest keep-score, with proline and glycine the highest — counter to substitution-matrix intuition. Glutamine’s low keep-score is not explained by a single obvious substitute; glutamate is preferred in only 14% of replacements. Removing this target proteome-wide, we replaced every glutamine across 4,148 proteins with ProteinMPNN-optimal alternatives and modeled the resulting sequences with AlphaFold2. The recoded proteome maintained native-like structures: 85.5% of models exceeded a TMscore of 0.8, indistinguishable from the 85.9% baseline for rebuilt native sequences. Orthogonal validation — ESMFold predictions, Rosetta energy calculations, and conformational ensemble comparisons — revealed no evidence of systematic misfolding attributable to glutamine removal. As a consistency check, a glutamine-free TEM-1 β-lactamase designed with ProteinMPNN conferred ampicillin resistance in E. coli , indicating fold preservation can extend to biological function. Evolutionary evidence further supports these findings: 24% of E. coli proteins have glutamine-free orthologs in prokaryotic genomes, and glutamine ranks lowest in keep-score in Saccharomyces cerevisiae . These results show that an entire amino acid can be removed from a proteome with folds broadly preserved, establishing keep-score as a general framework and glutamine as the primary candidate for recoding. Crucially, this reflects a computational analysis of structural replaceability, not a demonstration of biological dispensability.