Codon usage bias—the preferential use of certain synonymous codons—is a fundamental feature of all genomes. Codon usage has a key role in determining gene-expression levels in all organisms that have so far been studied1–3. Nearly all human-infecting viruses show patterns of codon usage that are distinct from those of human genes—yet they express their proteins efficiently in host cells to cause diseases and pandemics. The mechanism behind this evasion of codon usage control by viral RNA translation is unknown. Here we show that viral proteins are subject to strong codon usage control when they are translated like human genes, but that they can evade this control when translated from viral replicons. This evasion is mediated by viral 5′ untranslated regions (UTRs) in diverse human viruses, which support codon-usage-insensitive translation. Canonical mRNA translation depends on codon usage, requiring the 5′ cap, 3′ polyA tail and their associated proteins, which suggests that mRNA looping has a role in the effect of codon usage on translation. Notably, RNA circularization for mRNAs with viral 5′ UTRs restores codon-usage-dependent translation, owing mainly to non-optimal codon-usage-mediated repression. These results suggest that mRNA circularization is crucial for initiating codon-usage-dependent translation, and that viral RNAs bypass this mechanism by blocking circularization, allowing efficient translation despite their poor codon usage profiles.
