Large-scale analyses of synonymous substitution rates can be sensitive to assumptions about the process of mutation.

A popular approach to examine the roles of mutation and selection in the evolution of genomes has been to consider the relationship between codon bias and synonymous rates of molecular evolution. A significant relationship between these two quantities is taken to indicate the action of weak selection on substitutions among synonymous codons. The neutral theory predicts that the rate of evolution is inversely related to the level of functional constraint. Therefore, selection against the use of non-preferred codons among those coding for the same amino acid should result in lower rates of synonymous substitution as compared with sites not subject to such selection pressures. However, reliably measuring the extent of such a relationship is problematic, as estimates of synonymous rates are sensitive to our assumptions about the process of molecular evolution. Previous studies showed the importance of accounting for unequal codon frequencies, in particular when synonymous codon usage is highly biased. Yet, unequal codon frequencies can be modeled in different ways, making different assumptions about the mutation process. Here we conduct a simulation study to evaluate two different ways of modeling uneven codon frequencies and show that both model parameterizations can have a dramatic impact on rate estimates and affect biological conclusions about genome evolution. We reanalyze three large data sets to demonstrate the relevance of our results to empirical data analysis.