top of page


Reproducible algorithmic generation of resistance catalogues improves resistance prediction for bedaquiline in M. tuberculosis

D Adlard(1) D Eyre(1) D W Crook(1) T EA Peto(1) S V Omar(2) P W Fowler(1)

1:University of Oxford; 2:National Institute for Communicable Diseases, Johannesburg

Bedaquiline (BDQ) is a recent addition to the WHO recommended treatment regimen for multi-drug resistant tuberculosis, however rising levels of resistance threaten to reduce its efficacy against Mycobacterium tuberculosis (Mtb). Catalogues of mutations associated with resistance to bedaquiline are key to detecting resistance genetically. However, the recent second edition of the WHO resistance catalogue, which was built using considerable domain knowledge, assumes a high level of genetic homogeneity, uses complex grading rules and is not reproducible. We will show that (i) catalogues can be reproducibly and automatically constructed from clinical datasets and (ii) by using a less stringent approach overall performance is improved. We algorithmically applied a less statistically conservative method to the whole genome sequencing data and phenotypic drug susceptibility testing measurements of a dataset of 11,867 Mtb isolates, thereby reproducibly cataloguing genetic variants in known resistance genes and reliably predicting BDQ resistance. Unlike previous approaches, no variant in the mmpL5 gene is associated with resistance. Our catalogue therefore only considers variation in Rv0678, pepQ, and atpE, achieving a cross-validated sensitivity of 81.8 ± 3.8% for the 97.5 ± 0.3% of samples where a definite prediction can be made. Our results also show that minor Rv0678 variants are clinically relevant, mirroring published evidence for the fluoroquinolones and analysis done by the WHO, suggesting that bioinformatics thresholds must be lowered to catch them.

bottom of page