Evolution of tRNA pool shapes variation in selection on codon usage across the Saccharomycotina subphylum

Across the major taxonomical domains, synonymous codons of an amino acid are found to be used in unequal frequencies. This codon usage bias – both in terms of the degree of bias and the identity of codons used – is highly variable, even among closely related species. Within a species, genome-wide codon usage bias reflects a balance between adaptive and non-adaptive microevolutionary processes. Variation in these microevolutionary processes results in across-species variation in codon usage bias. As codon usage bias is tightly linked to important molecular and biophysical processes, it is critical to understand how changes to these processes drive changes to the microevolutionary processes. Here we employ a population genetics model of coding sequence evolution to quantify natural selection and mutation biases on a per-codon basis and estimate gene expression levels across the budding yeasts Saccharomycotina subphylum. We interrogate the impact of variation in molecular mechanisms hypothesized to be driving the microevolution of codon usage. We find that natural selection and mutation biases evolved rapidly over macroevolutionary time, with high variability between closely related species. The majority (324/327) of yeasts exhibited clear signals of translational selection, with selection coefficients being correlated with codon-specific estimates of ribosome waiting times within species. Across species, natural selection on codon usage correlated with changes to ribosome waiting times, indicating that tRNA pool evolution is a major factor driving changes to natural selection on codon usage. We find evidence that changes to tRNA modification expression can contribute to changes in natural selection across species independent of changes to tRNA gene copy number, suggesting tRNA modifications also play a role in shaping natural selection on codon usage. Our work firmly establishes how changes to microevolutionary processes can be driven by changes to molecular mechanisms, ultimately shaping the macroevolutionary variation of a trait.