Large-scale re-engineering of synonymous sites is a promising strategy to generate vaccines either through synthesis of attenuated viruses or via codon optimized genes in DNA vaccines. Attenuation typically relies on de-optimisation of codon pairs and maximization of CpG dinucleotide frequencies. So as to formulate evolutionarily-informed attenuation strategies that aim to force nucleotide usage against the direction favoured by selection, here we examine available whole-genome sequences of SARS-CoV-2 to infer patterns of mutation and selection on synonymous sites. Analysis of mutational profiles indicates a strong mutation bias towards U. In turn, analysis of observed synonymous site composition implicates selection against U. Accounting for dinucleotide effects reinforces this conclusion, observed UU content being a quarter of that expected under neutrality. Possible mechanisms of selection against U mutations includes selection for higher expression, for high mRNA stability or lower immunogenicity of viral genes. Consistent with gene-specific selection against CpG dinucleotides, we observe systematic differences of CpG content between SARS-CoV-2 genes. We propose an evolutionarily-informed approach to attenuation that, unusually, seeks to increase usage of the already most common synonymous codons. Comparable analysis of H1N1 and Ebola finds that GC3 deviated from neutral equilibrium is not a universal feature, cautioning against generalization of results.

Owing to a lag between a deleterious mutation’s appearance and its selective removal, gold-standard methods for mutation rate estimation assume no meaningful loss of mutations between parents and offspring. Indeed, from analysis of closely related lineages, in SARS-CoV-2 the Ka/Ks ratio was previously estimated as 1.008, suggesting no within-host selection. By contrast, we find a higher number of observed SNPs at 4-fold degenerate sites than elsewhere and, allowing for the virus’s complex mutational and compositional biases, estimate that the mutation rate is at least 49-67% higher than would be estimated based on the rate of appearance of variants in sampled genomes. Given the high Ka/Ks one might assume that the majority of such intra-host selection is the purging of nonsense mutations. However, we estimate that selection against nonsense mutations accounts for only ∼10% of all the “missing” mutations. Instead, classical protein-level selective filters (against chemically disparate amino acids and those predicted to disrupt protein functionality) account for many missing mutations. It is less obvious why for an intracellular parasite, amino acid cost parameters, notably amino acid decay rate, are also significant. Perhaps most surprisingly, we also find evidence for real time selection against synonymous mutations that move codon usage away from that of humans. We conclude that there is common intra-host selection on SARS-CoV-2 that acts on nonsense, missense and possibly synonymous mutations. This has implications for methods of mutation rate estimation, for determining times to common ancestry and the potential for intra-host evolution including vaccine escape.

The nucleotide composition, dinucleotide composition, and codon usage of many viruses differs from their hosts. These differences arise because viruses are subject to unique mutation and selection pressures that do not apply to host genomes; however, the molecular mechanisms that underlie these evolutionary forces are unclear. Here, we analysed the patterns of codon usage in 1,520 vertebrate-infecting viruses, focusing on parameters known to be under selection and associated with gene regulation. We find that GC content, dinucleotide content, and splicing and m6A modification-related sequence motifs are associated with the type of genetic material (DNA or RNA), strandedness, and replication compartment of viruses. In an experimental follow-up, we find that the effects of GC content on gene expression depend on whether the genetic material is delivered to the cell as DNA or mRNA, whether it is transcribed by endogenous or exogenous RNA polymerase, and whether transcription takes place in the nucleus or cytoplasm. Our results suggest that viral codon usage cannot be explained by a simple adaptation to the codon usage of the host – instead, it reflects the combination of multiple selective and mutational pressures, including the need for efficient transcription, export, and immune evasion.

Errors throughout gene expression are likely deleterious, hence genomes are under selection to ameliorate their consequences. Additional stop codons (ASCs) are in-frame nonsense ‘codons’ downstream of the primary stop which may be read by translational machinery should the primary stop have been accidentally read through. Prior evidence in several eukaryotes suggests that ASCs are selected to prevent potentially-deleterious consequences of read-through. We extend this evidence showing that enrichment of ASCs is common but not universal for single cell eukaryotes. By contrast, there is limited evidence as to whether the same is true in other taxa. Here, we provide the first systematic test of the hypothesis that ASCs act as a fail-safe mechanism in eubacteria, a group with high read-through rates. Contra to the predictions of the hypothesis we find: there is paucity, not enrichment, of ASCs downstream; substitutions that degrade stops are more frequent in-frame than out-of-frame in 3’ sequence; highly expressed genes are no more likely to have ASCs than lowly expressed genes; usage of the leakiest primary stop (TGA) in highly expressed genes does not predict ASC enrichment even compared to usage of non-leaky stops (TAA) in lowly expressed genes, beyond downstream codon +1. Any effect at the codon immediately proximal to the primary stop can be accounted for by a preference for a T/U residue immediately following the stop, although if anything, TT- and TC- starting codons are preferred. We conclude that there is no compelling evidence for ASC selection in eubacteria. This presents an unusual case in which the same error could be solved by the same mechanism in eukaryotes and prokaryotes but is not. We discuss two possible explanations: that, owing to the absence of nonsense mediated decay, bacteria may solve read-through via gene truncation and in eukaryotes certain prion states cause raised read-through rates.

In bacteria stop codons are recognized by one of two class I release factors (RF1) recognizing TAG, RF2 recognizing TGA, and TAA being recognized by both. Variation across bacteria in the relative abundance of RF1 and RF2 is thus hypothesized to select for different TGA/TAG usage. This has been supported by correlations between TAG:TGA ratios and RF1:RF2 ratios across multiple bacterial species, potentially also explaining why TAG usage is approximately constant despite extensive variation in GC content. It is, however, possible that stop codon trends are determined by other forces and that RF ratios adapt to stop codon usage, rather than vice versa. Here, we determine which direction of the causal arrow is the more parsimonious. Our results support the notion that RF1/RF2 ratios become adapted to stop codon usage as the same trends, notably the anomalous TAG behavior, are seen in contexts where RF1:RF2 ratios cannot be, or are unlikely to be, causative, that is, at 3′untranslated sites never used for translation termination, in intragenomic analyses, and across archaeal species (that possess only one RF1). We conclude that specifics of RF biology are unlikely to fully explain TGA/TAG relative usage. We discuss why the causal relationships for the evolution of synonymous stop codon usage might be different from those affecting synonymous sense codon usage, noting that transitions between TGA and TAG require two-point mutations one of which is likely to be deleterious.