Nanopore sequencing rivals Illumina for tuberculosis drug resistance and outbreaks

Tuberculosis remains a global public-health challenge, and knowing which drugs a patient’s Mycobacterium tuberculosis complex (MTBC) strain is susceptible to—and whether cases are linked in an outbreak—depends increasingly on whole-genome sequencing. For about a decade, short-read sequencing with Illumina has been treated as the reference standard for genomic drug susceptibility testing (gDST) and outbreak analysis. Long-read sequencing, represented by Oxford Nanopore Technologies (ONT), promises faster, simpler and portable workflows, but until now questions remained about whether its results are as trustworthy as Illumina’s. In the largest direct comparison to date, a team led by corresponding author Derrick W. Crook sequenced 508 clinical MTBC samples from South Africa and Vietnam on both short-read and long-read platforms. The samples were intentionally enriched for drug resistance and included several local outbreaks and multiple lineages. The study asked whether ONT data could predict resistance, assign lineages, and identify related cases as reliably as Illumina, and whether sequencing results from both platforms could be pooled for research and public-health use.

The investigators compared Illumina sequencing with ONT long-reads produced using the latest generation ONT R10.4.1 flow cells, V14 chemistry, 'super high accuracy' basecalling (dorado v4.3.0/5.0.0) and a bioinformatics pipeline built around the Clair3 deep-learning based variant caller. A total of 508 samples were sequenced by both technologies, with ONT runs carried out on GridION or PromethION; 425 samples (83.7%) had sufficient read depth for head-to-head comparison. Lineages matched for 407/425 (95.8%) samples, with the few disagreements explained by mixed lineages detected by one platform. Evidence of non-tuberculous mycobacterium (NTM) subpopulations appeared in nine samples. Using Illumina as the reference, ONT’s very major error (VME) rate across all 15 drugs was 1.0% (0.6-1.5%), major error (ME) was 1.7% (1.3-2.2%) and the unclassified rate was 6.9% (6.3-7.5%), all below thresholds set by CLSI. For individual drugs, VME and ME point estimates were ≤3% in 29/30 cases and ≤1.5% in 25/30. After filtering mixtures, 382 isolates remained and masking parts of the reference genome produced a mean SNP distance of 0.13 (median zero) between platforms; for 376/382 samples (98.4%, CI:96.6-99.4%) the difference was ≤1 SNP, keeping outbreak cluster calls highly concordant (43 putative clusters among 172 isolates showed few differences).

The study shows that, provided sequencing quality is high enough, the practical differences between modern ONT long-read data and Illumina short-read data for the key clinical outputs—lineage assignment, SNP-based relatedness and genomic drug susceptibility testing—are very small. Those differences fall within regulatory tolerances, meaning public-health laboratories and research groups can safely aggregate sequencing data produced on the two platforms for large-scale analyses. The portability and speed of ONT now become realistic advantages without sacrificing accuracy, which could simplify local outbreak response and expand access to genomic diagnostics. The results also broaden the market for sequence-based testing and suggest that, if replicated for other pathogens, long-read platforms could offer competitive, flexible options for routine surveillance and clinical decision-making.