<p>Phenotypic mutations include non-heritable sequence changes arising from synthetic errors such as mistranscription or mistranslation. Despite wide potential impact of both error types, their relationship remains underexplored. Using Circ-Seq and mass spectrometry data, we perform a genome-wide analysis of mistranscriptions and mistranslations in five model organisms, and find that genes with frequent mistranslations exhibit lower mistranscription rates. We hypothesize that this pattern is explainable by a negative epistasis between the two error types. We assess our hypothesis with systematic experimental measurements that confirm the prevalence of negative epistasis, as well as in silico evolutionary simulations that suggest the extra deleterious effects resulting from negative epistasis are sufficient to create selection against mistranscription in highly mistranslated genes. Moreover, empirical genomic analyses indicate that genes with frequent mistranslation purge nonsynonymous mistranscriptions more efficiently, and that highly translated transcripts show fewer mistranscriptions. Combined, our results reveal an unrecognized interaction between mistranscription and mistranslation.</p>