Selection bias in microbial mutation accumulation studies and the impact of colony growth

In microbial mutation-accumulation (MA) studies, it is widely thought that natural selection is silenced by repeated single-cell bottlenecks. In support of this claim, 40 published tests for selection bias failed to detect a deficit of non-synonymous relative to synonymous mutations in wild-type microbes. Here we show that this most likely reflects selective reporting and lack of power. Our meta-analysis of 10,856 mutations from wild-type microbial MA reveals a clear signal of selection: non-synonymous mutations are observed 7.7\% less often than synonymous mutations. However, our inference of the bias is hampered by a widespread failure to consider the mutation spectrum. To overcome this, we provide a multinomial-logit model that jointly estimates the mutation spectrum and selection. By applying this to a 194-line Escherichia coli MA experiment and five previous E. coli datasets (869 mutations) we reveal a deficit of non-synonymous mutations, although the reduction is not significant. While approaches do exist for correcting for selection bias, all currently assume that microbial MA lines are grown in well-mixed liquid culture rather than as surface colonies where competition is spatially structured. Although existing theory suggests selection bias should be stronger under colony growth, using agent-based simulations we show that this actually depends on the scale over which neighbouring cells compete and how unevenly they divide: it can be weaker, equivalent to, or stronger than in homogeneous growth. While our preliminary assessment is that it is considerably stronger, quantitative predictions will require the empirical details of colony growth to be better resolved.