A slight mismatch between a gene's codon usage and the cellular tRNA supply is beneficial.

Each amino acid except two is encoded by multiple synonymous codons, but at unequal frequencies. Such codon usage bias (CUB) is observable in almost all species, and commonly assumed as the result of natural selection towards an optimal CUB that matches the cellular tRNA supply. Here we hypothesize instead that the optimal CUB of a gene should slightly mismatch the tRNA supply to avoid excessive translational costs, while ensuring adequate functional payoff. By modifying the CUB of a resistance gene expressed in bacteria under antibiotic selection, we demonstrate that a small mismatch with the tRNA supply confers faster bacterial growth than those with minimized or large CUB-tRNA mismatches. Intriguingly, the optimal degree of CUB-tRNA mismatch increases as the resistance gene becomes less important in media with lower antibiotic concentrations, which is explainable by our model as a shift in the balance between the gene's functional payoff and translational cost. Furthermore, genomic analyses in model organisms suggest that the optimal degree of CUB-tRNA mismatch is larger for endogenous genes with lower functional importance and higher mRNA abundance, respectively supporting the impact of functional payoff and translational cost. Finally, we find that mutations increasing or decreasing the CUB-tRNA mismatch of native genes are both predominantly deleterious, such that the CUB-tRNA mismatch is likely selectively maintained rather than minimized to that achievable in the presence of genetic drift and mutational bias. These results challenge the commonly assumed unidirectional selection on CUB and highlight the CUB-modulated balance between functional payoff and translational cost.