A standardised method for interpreting the association between mutations and phenotypic drug resistance in Mycobacterium tuberculosis

Abstract

A clear understanding of the genetic basis of antibiotic resistance in Mycobacterium tuberculosis is required to accelerate the development of rapid drug susceptibility testing methods based on genetic sequence.Raw genotype-phenotype correlation data were extracted as part of a comprehensive systematic review to develop a standardised analytical approach for interpreting resistance associated mutations for rifampicin, isoniazid, ofloxacin/levofloxacin, moxifloxacin, amikacin, kanamycin, capreomycin, streptomycin, ethionamide/prothionamide and pyrazinamide. Mutation frequencies in resistant and susceptible isolates were calculated, together with novel statistical measures to classify mutations as high, moderate, minimal or indeterminate confidence for predicting resistance.We identified 286 confidence-graded mutations associated with resistance. Compared to phenotypic methods, sensitivity (95% CI) for rifampicin was 90.3% (89.6-90.9%), while for isoniazid it was 78.2% (77.4-79.0%) and their specificities were 96.3% (95.7-96.8%) and 94.4% (93.1-95.5%), respectively. For second-line drugs, sensitivity varied from 67.4% (64.1-70.6%) for capreomycin to 88.2% (85.1-90.9%) for moxifloxacin, with specificity ranging from 90.0% (87.1-92.5%) for moxifloxacin to 99.5% (99.0-99.8%) for amikacin.This study provides a standardised and comprehensive approach for the interpretation of mutations as predictors of M. tuberculosis drug-resistant phenotypes. These data have implications for the clinical interpretation of molecular diagnostics and next-generation sequencing as well as efficient individualised therapy for patients with drug-resistant tuberculosis.

FIGURE 1

Medium confidence values (MCVs) stratified by confidence value, drug susceptibility testing medium and antibiotic-resistance gene combination. In the three rows above the graph we show variants that were concordant on both media, the number of variants that had different confidence levels on liquid and solid (these are marked as “discrepant variants” and are listed in full in online supplementary material 9) and unique variants for which confidence levels were available on only one of the two media.

FIGURE 2

Comparison of the sensitivity and specificity of different groups of mutations. For each drug, two types of comparison were performed. First, the a) sensitivities and b) specificities were calculated with the associated 95% confidence levels for the “observed” phenotypic result. Specifically, the figures for high (Hi), high and moderate (Hi+Mo) and high and moderate and minimal (Hi+Mo+Mi) confidence interpretative best confidence values (iBCVs) were compared with using all mutations observed in the study. Moreover, genetic variants were included from a recent study by Farhat et al. [30]. Second, we conducted the same comparison using a “corrected” phenotype as reference (i.e. where we assumed that strains that were phenotypically susceptible but harboured either a Hi, Hi+Mo or H+Mo+Mi confidence iBCV mutation or mutations by Farhat et al. were false-susceptible results). For some drugs, such as capreomycin (CM), both percentages remained unchanged as all mutations were high confidence. Minimal target product profile (TPP) thresholds set by the World Health Organization for new molecular-based diagnostic tools compared to phenotypic drug susceptibility testing [31] are shown. These were intended for rifampicin (R), isoniazid (H), fluoroquinolones, kanamycin (KM), amikacin (AM) and CM only. However, in addition we included the threshold for the remaining drugs and overall results for comparison (for additional details see online supplementary material 11). MFX: moxifloxacin; OFX: ofloxacin; LFX: levofloxacin; S: streptomycin; ETO/PTO: ethionamide and prothionamide; Z: pyrazinamide.