BACKGROUND: Pyrazinamide is one of four first-line antibiotics used to treat tuberculosis; however, antibiotic susceptibility testing for pyrazinamide is challenging. Resistance to pyrazinamide is primarily driven by genetic variation in pncA, encoding an enzyme that converts pyrazinamide into its active form. METHODS: We curated a dataset of 664 non-redundant, missense amino acid mutations in PncA with associated high-confidence phenotypes from published studies and then trained three different machine-learning models to predict pyrazinamide resistance. All models had access to a range of protein structural-, chemical- and sequence-based features. RESULTS: The best model, a gradient-boosted decision tree, achieved a sensitivity of 80.2% and a specificity of 76.9% on the hold-out test dataset. The clinical performance of the models was then estimated by predicting the binary pyrazinamide resistance phenotype of 4027 samples harbouring 367 unique missense mutations in pncA derived from 24 231 clinical isolates. CONCLUSIONS: This work demonstrates how machine learning can enhance the sensitivity/specificity of pyrazinamide resistance prediction in genetics-based clinical microbiology workflows, highlights novel mutations for future biochemical investigation, and is a proof of concept for using this approach in other drugs.

The primary platform used for pyrazinamide (PZA) susceptibility testing of Mycobacterium tuberculosis is the MGIT culture system (Becton Dickinson). Since false-resistant results have been associated with the use of this system, we conducted a multicenter evaluation to determine the effect of using a reduced cell density inoculum on the rate of false resistance. Two reduced inoculum densities were compared with that prescribed by the manufacturer (designated as "BD" method). The reduced inoculum methods (designated as "A" and "C") were identical to the manufacturer's protocol in all aspects with the exception of the cell density of the inoculum. Twenty genetically and phenotypically characterized M. tuberculosis isolates were tested in duplicate by ten independent laboratories using the three inoculum methods. False-resistant results declined from 21.1% using the standard "BD" method to 5.7% using the intermediate ("A") inoculum and further declined to 2.8% using the most dilute ("C") inoculum method. The percentages of the resistant results that were false-resistant declined from 55.2% for the "BD" test to 28.8% and 16.0% for the "A" and "C" tests, respectively. These results represent compelling evidence that the occurrence of false-resistant MGIT PZA susceptibility test results can be mitigated through the use of reduced inoculum densities.

The emergence of multi and extensively drug-resistant tuberculosis is a significant impediment to the control of this disease because treatment becomes more complex and costly. Reliable and timely drug susceptibility testing is critical to ensure patients receive effective treatment and become non-infectious. Molecular methods can provide accurate and rapid drug susceptibility results. We used DNA sequencing to detect resistance to the first-line antituberculosis drugs, isoniazid (INH), rifampin (RIF), pyrazinamide (PZA), and ethambutol (EMB), and the second-line drugs, amikacin (AMK), capreomycin (CAP), kanamycin (KAN), ciprofloxacin, (CIP) and ofloxacin (OFX). Nine loci were sequenced: rpoB for resistance to RIF, katG and inhA (INH), pncA (PZA), embB (EMB), gyrA (CIP and OFX), rrs, eis, and tlyA (KAN, AMK, and CAP). A total of 314 clinical M. tuberculosis complex isolates, representing a variety of antibiotic resistance patterns, genotypes and geographical origins were analyzed. The molecular data were compared to the phenotypic data and the accuracy values were calculated. Sensitivity and specificity values (as percentages) for the first-line drug loci were rpoB (97.1, 93.6), katG (85.4, 100), inhA (16.5, 100), katG and inhA together (90.6, 100) pncA (84.6, 85.8), and embB (78.6, 93.1). The values for the second-line drugs were also calculated. The size and scope of this study, in numbers of loci and isolates examined, and the phenotypic diversity of those isolates, support the use of DNA sequencing to detect drug resistance in the M. tuberculosis complex. Further, the results can be used to design diagnostic tests utilizing other mutation detection technologies.

The current study describes the development of a unique real-time PCR assay for the detection of mutations conferring drug resistance in Mycobacterium tuberculosis (Mtb). The Rifampicin Resistance Determinant Region (RRDR) of rpoB and specific regions of katG and the inhA promoter were targeted for the detection of rifampicin (RIF) and isoniazid (INH) resistance, respectively. Additionally, this assay was multiplexed to discriminate Mycobacterium tuberculosis complex (MTBC) strains from Nontuberculous Mycobacteria (NTM) strains by targeting the IS6110 insertion element. High resolution melting (HRM) analysis following real-time PCR was used to identify Mtb strains containing mutations at the targeted loci, and locked nucleic acid (LNA) probes were used to enhance the detection of strains containing specific SNP transversion mutations. This method was used to screen 252 Mtb clinical isolates including 154 RIF resistant strains and 174 INH resistant strains based on the agar proportion method of drug susceptibility testing (DST). Of the 154 RIF resistant strains, 148 were also resistant to INH and therefore classified as multidrug resistant (MDR). The assay demonstrated a sensitivity and specificity of 91% and 98%, respectively, for the detection of RIF resistance, and 87% and 100% for the detection of INH resistance. Overall, this assay showed a sensitivity of 85% and a specificity of 98% for the detection of MDR strains. This method provides a rapid, robust, and inexpensive way to detect the dominant mutations known to confer MDR in Mtb strains and offers several advantages over current molecular and culture-based techniques.