Switches in genomic gc content drive shifts of optimal codons under sustained selection on synonymous sites

The major codon preference model suggests that codons read by tRNAs in high concentrations are preferentially utilized in highly expressed genes. However, the identity of the optimal codons differs between species although the forces driving such changes are poorly understood.We suggest that these questions can be tackled by placing codon usage studies in a phylogenetic framework and that bacterial genomes with extreme nucleotide composition biases provide informativemodel systems. Switches in the background substitution biases fromGC toAT have occurred in Gardnerella vaginalis (GC=32%),and fromAT toGC inLactobacillus delbrueckii (GC=62%) and Lactobacillus fermentum (GC= 63%). We show that despite the large effects on codon usage patterns by these switches, all three species evolveunder selectiononsynonymous sites. InG. vaginalis, the dramatic codon frequency changes coincide with shifts of optimal codons. In contrast, the optimal codons have not shifted in the two Lactobacillus genomes despite an increased fraction of GC-ending codons.We suggest that all three species are in different phases of an on-going shift of optimal codons, and attribute the difference to a stronger background substitution bias and/or longer time since the switch in G. vaginalis. We showthat comparative and correlativemethods for optimal codon identification yield conflicting results for genomes in flux and discuss possible reasons for themispredictions.Weconclude thatswitches in the direction of the background substitution biases can drivemajor shifts in codon preference patterns even under sustained selection on synonymous codon sites.