Clinical microbiology often assumes a sample is resistant or susceptible. Making such a classification relies on applying a threshold (usually called a cutoff) to quantitative data, such as minimum inhibitory concentrations (MICs). If the MICs are strongly bimodal, then this is trivial and reproducibility is guaranteed. If the MICs are unimodal, then one is left with the unsatisfactory situation whereby the clinical test does not appear to be reproducible since natural variation can cause individual results to “flip-flop” over the cutoff.
Tim Davies, who led this work (open access), shows that whilst amoxicillin, a beta-lactam antibiotic, has a bimodal MIC distribution, when taken with the beta-lactamase inhibitor clavulanate in the combined therapy co-amoxiclav, the MIC distribution becomes unimodal with all the attendant problems of a lack of reproducibility. This is further complicated by the US and European bodies (CLSI and EUCAST) adopting incompatible testing approaches.
The original aim of this work was the test how well knowing the genetics of an E.coli infection could be used to predict its susceptibility to standard treatments, like co-amoxiclav. Simple presence/absence of beta-lactamase genes is not sufficient; instead the predictive model has to include knowledge of promoter mutations and copy number. Even with these improvements, the above reproducibilty problem ensures there is an upper threshold to the sensitivity/specificity possible.
In other words, translating genetics into clinical microbiology for E.coli is going to be a whole lot harder than it was for M. tuberculosis!