ABSTRACT
Contact tracing requires reliable identification of closely related bacterial isolates. When we noticed the reporting of artefactual variation between M. tuberculosis isolates during routine next generation sequencing of Mycobacterium spp, we investigated its basis in 2,018 consecutive M. tuberculosis isolates. In the routine process used, clinical samples were decontaminated and inoculated into broth cultures; from positive broth cultures DNA was extracted, sequenced, reads mapped, and consensus sequences determined. We investigated the process of consensus sequence determination, which selects the most common nucleotide at each position. Having determined the high-quality read depth and depth of minor variants across 8,006 M. tuberculosis genomic regions, we quantified the relationship between the minor variant depth and the amount of non-Mycobacterial bacterial DNA, which originates from commensal microbes killed during sample decontamination. In the presence of non-Mycobacterial bacterial DNA, we found significant increases in minor variant frequencies of more than 1.5 fold in 242 regions covering 5.1% of the M. tuberculosis genome. Included within these were four high variation regions strongly influenced by the amount of non-Mycobacterial bacterial DNA. Excluding these four regions from pairwise distance comparisons reduced biologically implausible variation from 5.2% to 0% in an independent validation set derived from 226 individuals. Thus, we have demonstrated an approach identifying critical genomic regions contributing to clinically relevant artefactual variation in bacterial similarity searches. The approach described monitors the outputs of the complex multi-step laboratory and bioinformatics process, allows periodic process adjustments, and will have application to quality control of routine bacterial genomics.