Bacterial gene 5' ends have unusual mutation rates that can mislead tests of selection.

Despite early assumptions of neutrality, numerous mechanisms are now thought to cause selection on synonymous mutations, commonly supported by a low evolutionary rate at synonymous sites (Ks). This has been best evidenced in the first ~10 codons of genes in E. coli, where Ks is less than around half that of the gene body. Diverse lines of evidence support the hypothesis that these first ~10 codons are under selection for high AT content which causes low mRNA stability that in turn enables ribosomal initiation. There remains one enigmatic discrepancy, however, namely that the low Ks domain extends far beyond the first 10 codons. Here we ask why this is. As we see no evidence that the zone influencing protein levels has been misestimated, we consider three further hypotheses: that reduced Ks is a) owing to overlapping genes, b) reflects an extended slow translational "ramp," and c) is mutational. We reject the first two as in both E. coli and Bacillus sp. the extended low Ks domain persists on analysis of non-overlapping genes and in Bacillus, where fast optimal codons tend to be A/T-ending, a fast-to-slow codon trend is seen. We fail to falsify the third hypothesis. Employing mutation accumulation data for E. coli we show that the 5' end has a lower mutation rate, with the first 10 codons having a rate around half that of the gene body, this then steadily increasing following the trend seen for Ks. Compositional variation is likely to explain some of the difference, the 5' end lacking GC-rich runs while these are most mutagenic. We conclude that even a highly reduced Ks is not always adequate to substantiate selection on synonymous mutations. This result has broad implications for inference of the causes of evolutionary rate variation.