Sifting Through the Noise: Minority Variant Detection in Mycobacterium tuberculosis and Clinical Implications in Bedaquiline Resistance

Abstract

Bulk whole-genome sequencing (WGS) is usually applied to study genetic variation that differentiates distinct cell populations. However, bulk WGS can also capture recently-evolved diversity emerging within a single population. The latter requires the accurate detection of low-frequency (minority) or sub-consensus variants present at mapped read frequencies typically 75–95%. Low-frequency variants are relevant for understanding within-host pathogen evolution, somatic oncogenesis in cancers, and microbiome diversity, among other applications. Specifically for Mycobacterium tuberculosis (Mtb), the causative agent of tuberculosis (TB), patient-derived Mtb isolates from bacterial culture are not pure clones. Within-host mixtures of drug-resistant and drug-susceptible Mtb strains during TB infection (heteroresistance) have been shown to precede drug resistance development through subsequent fixation. Genomic sequencing of bulk Mtb sequencing data from patient-derived samples has identified resistance-associated minority variants, and these minority variants have been shown to improve resistance prediction in a number of cases. However, minority variant detection in the field is currently not standardized and the reliability of some methods is debatable. Further, there is uncertainty around the within-sample allele frequency (AF) threshold at the presence of a minority variant will have a clinical impact. With these two factors working together—minority variant detection that is too permissive, and resistance phenotype association models that consider variants down to allele frequencies of 1%—the field is not well-equipped to glean real signal from the as yet limited analysis of minority variants in Mtb. To this end, this thesis focuses on the benchmarking the detection of minority variants in bulk short-read Mtb sequencing data derived from patients, and use these findings to investigate the association between high-confidence minority variants in candidate bedaquiline resistance genes and bedaquiline resistance phenotype in a sample of clinical Mtb isolates. In Chapter 1, we benchmark seven variant callers on precision, recall and false positive characteristics for detecting minority variants using simulated short-read WGS data for 700 Mtb strains. We identify FreeBayes to be the top-performing tool genome-wide and find reference bias and low genomic sequence complexity to be major sources of false variant calls. Our evaluation of these widely-used and versatile variant callers is applicable to many areas of genomic research including somatic mutation detection in cancers, microbiome genetic diversity analysis and within-host pathogen evolution. Further, we show false variant calls can be mitigated by the exclusion of low mappability regions defined in our study and imposing a minimum allele frequency cutoff. These recommendations will enable further study on the threshold for the clinical impact of minority variants. In Chapter 2, our work from chapter 1 enables us to describe the prevalence of high-confidence minority variants in clinical Mtb strains and investigate the association of these variants with bedaquiline resistance. We find an enrichment of minority variants in Rv0678, a primary candidate bedaquiline resistance gene, and in bedaquiline-resistant isolates. Using a set of nested regression models we find that minority variants at sub-consensus allele frequencies down to 15% offer added explanatory power for bedaquiline MIC determination.