The largest ever global study of tuberculosis has identified all the mutations conferring resistance to approved antimicrobials, setting the scene for more rational use of drugs, increasing mechanistic understanding of how Mycobacterium tuberculosis develops resistance to antibiotics and pointing to new drug targets.

The Comprehensive Resistance Prediction for Tuberculosis International Consortium (CRyPTIC) analyzed 15,211 M. tuberculosis samples from 27 countries on 5 continents. Using a new quantitative test for drug resistance and a new approach to identifying all the mutations in a sample of drug-resistant TB bacteria, the project generated a dataset that was then used to quantify how each mutation reduced the effectiveness of 13 antimicrobial drugs.

"This innovative, large-scale, international collaboration has enabled us to create possibly the most comprehensive map yet of the genetic changes responsible for drug resistance in tuberculosis," said Derrick Crook, professor of microbiology at the University of Oxford, who led the project.

A series of nine new preprint manuscripts, each documenting a different aspect of the CRyPTIC project, was published on the bioRxiv and medRxiv preprint servers on October 19.

The CRyPTIC dataset is expected to improve treatment of tuberculosis through a precision medicine approach that uses a genomics-based method for assessing susceptibility of an individual infection to a specific drug.

That would replace the culture-based testing, which currently is used to assess minimum inhibitory concentrations (MIC) at which a drug will be effective. Although still the reference standard for most drugs, it can take over a month to complete culture-based testing, which requires expensive, complex laboratory capacity.

Working from the catalogue of resistance-associated mutations, the researchers developed a machine learning system for predicting susceptibility from whole genome sequences for each of the 13 drugs.

"The ultimate goal is to achieve a sufficiently accurate genetic prediction of resistance to most antituberculosis drugs, so that whole genome sequencing can replace culture-based drug susceptibility testing for TB," Crook said.

The dataset, which is now publicly available, can be used to scan the M. tuberculosis genome for mutations that were not previously known to cause drug resistance. It also makes it possible to assess how individual mutations and combinations of mutations are related, not just to broad mechanisms of resistance -- or of susceptibility -- to a particular drug, but also to minor changes in how a drug works.

While the resistance mechanisms are already well understood for some drugs, others were not, particularly in the case of new and repurposed drugs.

The researchers assessed the MIC at which the 13 drugs inhibited the growth of the large collection of M. tuberculosis samples and then performed genome-wide association studies on resistant samples to plot resistance phenotypes.

The research found previously uncataloged variants associated with resistance for all the drugs.

The 13 drugs were the first-line therapies ethambutol, isoniazid and rifampicin; second-line drugs amikacin, ethionamide, kanamycin, levofloxacin, moxifloxacin and rifabutin; and the new and repurposed drugs bedaquiline, clofazimine, delamanid and linezolid.

The resistant phenotype distributions differed between the drugs, with low numbers of resistant isolates for the new and repurposed drugs, which have not yet been widely used in tuberculosis treatment. The GWAS uncovered 66 isolates resistant to bedaquiline, 97 resistant to clofazimine, 77 resistant to delamanid and 67 resistant to linezolid.

Some variants were identified in novel genes, some were novel variants in known genes, and some were known variants. For example, among genes that are known to confer resistance to antimicrobial drugs, the researchers identified uncataloged variants in gyrB, the subunit of DNA gyrase, associated with levofloxacin and moxifloxacin resistance.

The machine learning system made it possible to identify combinations of mutations that are correlated with different forms of drug resistance.

Previously uncataloged variants are important because they could improve resistance prediction and shed light on underlying resistance mechanisms. The mutations could be de novo, or it may be that they have previously been implicated in resistance but there was not enough evidence for them to be catalogued.

The researchers have drawn up a list of the 20 genes that are most relevant to the development of resistance for each of the 13 drugs. They say the CRyPTIC dataset covers every resistance-causing mutation in M. tuberculosis and that it will be kept updated as new mutations emerge.