New preprint: looking at rifampicin-resistant subpopulations in clinical samples

Since clinical samples are usually grown in a MGIT tube for a while before some "crumbs" are harvested for DNA extraction, they are metagenomic in the sense that they can and do contain multiple colonies. This means we should expect subpopulations in our analysis but most bioinformatics tools and file formats inherently assume a homogenous sample with a single genome.

In this preprint Viki Brunner examines the small proportion of samples with a rifampicin-resistant subpopulation in a dataset of 35,538 samples which have been both whole genome sequenced and tested for rifampicin susceptibility. The sensitivity of resistance prediction is increased from 94.3% to 96.3% if you allow samples with 5% or more of reads supporting a rifampicin (RIF) resistant associated variant (RAV) to call resistance, as opposed to the more usual 75% or 90%.

Drawing on her earlier work she shows that these samples with a RIF RAV are less likely to have a compensatory mutation elsewhere in the RNA polymerase and, interestingly, if you then look at the distribution of minor alleles you can infer that resistance arose from a secondary infection in at least a third of these samples.