New tool speeds tuberculosis genome surveillance and resistance detection

Public health teams need fast, reliable ways to read the genomes of Mycobacterium tuberculosis complex (MTBC) bacteria to follow how the disease spreads and to spot drug resistance. Mireia Coscollá led a team to build Pathotypr, a software tool designed to make those tasks simpler and more standardised. The researchers started by reconstructing an MTBC family tree from 26,813 genomes, using 609,003 polymorphic sites to define a refreshed set of lineage markers. Pathotypr works from both genome assemblies and raw sequencing reads, and it was created to assign bacteria to the recognised MTBC lineages consistently while also reporting resistance-associated variants. Importantly, Pathotypr supports all 14 currently recognised MTBC lineages (L1-L10, A1-A4), meaning it can place samples into the full, up-to-date lineage framework. The tool combines marker-based calls with an alignment-free approach, so it can give harmonised lineage labels quickly and without the need for slow whole-genome alignments. This groundwork was intended to let labs and public health agencies compare results across studies and borders with the same consistent lineage definitions.

To make assignments fast and robust, the team implemented a k-mer/Random Forest framework that uses the marker backbone for lineage calling and the WHO catalogue for resistance-associated variant reporting. They tested Pathotypr on multiple datasets: 498 complete RefSeq assemblies, 88,071 samples typed by UShER-TB, 162 clinical read sets for closest-reference matching, and 7,148 CRyPTIC isolates that had phenotypic drug susceptibility data. On the 498 complete genomes, marker-based and alignment-free lineage calls were 100% concordant, and prediction accuracy stayed 100% on a set of 254 independent assemblies. In the large UShER-TB collection of 88,071 non-ambiguous samples, root-lineage concordance with TB-Profiler was 100%, while Pathotypr also identified lineage 10, A1 and A2 where present. For resistance screening, overall genotype-phenotype concordance was 85.0%. Performance was especially high for rifampicin (95.8% sensitivity, 95.0% specificity) and isoniazid (93.0% sensitivity, 97.9% specificity). The tool runs quickly — about 1 second per sample — allowing roughly 88,071 samples to be analysed in about 24 hours on four threads.

The combination of rapid lineage assignment and confident resistance screening is intended to support near real-time and cross-border tuberculosis surveillance. In an MDR-enriched subset from the CRyPTIC collection, Pathotypr helped reconstruct 135 probable introduction events into Germany, Italy and Ukraine, and revealed that 33.7% of introduction-associated isolates carried MDR/pre-XDR genotypes. That kind of information can help public health teams detect where resistant strains are arriving and spreading, and to prioritise follow-up actions. Because Pathotypr provides harmonised lineage names and leverages the WHO catalogue for resistance calls, it can make datasets comparable across different labs and nations. Fast, standardised outputs mean surveillance networks could identify concerning trends sooner and feed actionable information into contact tracing, treatment guidance and regional responses. Overall, the tool aims to make genomic surveillance more scalable, interoperable and useful for controlling tuberculosis in different settings.