The work addresses a persistent blind spot in tuberculosis diagnostics, because the large-scale genetic changes that long reads can detect are systematically missed by the short-read sequencing platforms that dominate routine whole-genome sequencing. According to figures cited in the study from the World Health Organization, the bacterium caused an estimated 10.7 million new cases and 1.2 million deaths in 2024. Of particular concern is multidrug-resistant tuberculosis, defined as resistance to the two first-line drugs rifampicin and isoniazid.

Roughly 390,000 people were diagnosed with multidrug- or rifampicin-resistant tuberculosis in 2024, and the treatment success rate for these patients stands at only 68 percent, compared with 88 percent for drug-susceptible disease. The gap is largely attributed to resistance being detected only after empirical therapy has failed, which makes rapid, mechanism-based diagnosis a central pillar of the WHO's End-TB strategy. Whole-genome sequencing has transformed the detection of resistance mutations in tuberculosis bacteria, but the technology most widely used in clinical laboratories relies on short reads, fragments of DNA a few hundred base pairs long.

Short reads excel at finding single-nucleotide changes and small insertions or deletions, yet they struggle to resolve repetitive genomic regions, large insertions and deletions, and insertions of mobile genetic elements. These larger alterations, known as structural variants and defined as changes of at least fifty base pairs, include insertions, deletions, tandem duplications, inversions and translocations. Recent long-read studies have begun to catalogue them systematically, but functional proof that a given structural variant actually causes resistance has remained scarce.

The research team, led by Qiongdan Wang, Yaya Geng, Xiaoqun Zheng and Guangxin Xiang, sequenced eleven clinical isolates: three drug-susceptible strains and eight multidrug-resistant strains, all collected from distinct patients in the Wenzhou region of Zhejiang Province between July 2023 and June 2024. The fully susceptible laboratory reference strain H37Rv served as an internal control. Phenotypic drug susceptibility testing on the BACTEC MGIT 960 system confirmed that all eight resistant isolates were resistant to at least rifampicin and isoniazid, and six of them, R2 through R6 and R8, showed additional resistance to the fluoroquinolones levofloxacin and moxifloxacin, meeting the WHO definition of pre-extensively drug-resistant tuberculosis.

On the sequencing side, the researchers ran each genome through both Oxford Nanopore long-read sequencing on a PromethION instrument and Illumina short-read sequencing. The long-read assemblies covered on average 99.14 percent of the H37Rv reference genome with a mean of only 2.17 contigs, a dramatic improvement in contiguity over the Illumina-only assemblies, which averaged 132.5 contigs. Polishing the long-read assemblies with short reads did not significantly improve genome coverage, and variant calls from the two approaches agreed in more than 90 percent of cases, giving the team confidence that long reads alone could reliably detect structural variants in these genomes.

Variant calling with BCFtools identified 3,233 mutations unique to the multidrug-resistant isolates, including 132 structural variants. Rather than attempting another genome-wide catalogue, the team deliberately narrowed its focus to the 65 genes listed in the WHO's 2023 mutation catalogue of tuberculosis drug resistance. Within that restricted set, they found 88 mutations across 14 genes exclusive to the resistant isolates, of which 65 had already been classified by the WHO and 23 were unclassified.

Crucially, only two structural variants fell inside these catalogued resistance genes: an IS6110 insertion of 1,358 base pairs in Rv0678, designated c.59_60ins1358bp, and a 131-base-pair deletion in ethA, designated c.135_265del. An independent structural variant caller, SVIM-asm, flagged 101 multidrug-specific structural variants, 90 of which overlapped the BCFtools calls, and the only structural variants consistently detected by both methods within WHO-catalogued genes were precisely these two. Sanger sequencing confirmed all candidate mutations, and phenotypic testing established a genotype-phenotype link for each.

The Rv0678 insertion, found in isolate R8, was associated with resistance to bedaquiline, a key later-generation drug. IS6110-mediated disruption of Rv0678 is an established resistance mechanism, but the insertion site reported here, between codons 59 and 60, is a new allelic variant of that mechanism. The ethA deletion, found in isolate R6, was associated with resistance to ethionamide and, to the authors' knowledge, had never been described before.

Phylogenetic analysis based on core single-nucleotide polymorphisms showed that most isolates belonged to Lineage 2, the dominant clade circulating in China and one frequently associated with heightened transmission and drug resistance. The variants were distributed across multiple isolates, consistent with independent acquisition rather than a single clonal expansion. To test whether these structural variants truly cause resistance, the team turned to Mycobacterium smegmatis, a fast-growing, non-pathogenic relative of the tuberculosis bacterium that is widely used as a surrogate for functional studies of mycobacterial genes.

They constructed recombinant strains overexpressing either the wild-type or the structural-variant versions of Rv0678 and ethA, confirmed robust expression by quantitative PCR, and used Western blotting with anti-His-tag antibodies to check protein production. Wild-type strains produced full-length proteins of the expected sizes, but in both structural-variant strains the predicted truncated peptides were undetectable, possibly because of rapid degradation, failed translation, or limitations of the gel system. All strains reached comparable cell densities by 36 hours, ruling out growth defects as a confounder.

The resistance phenotypes were unambiguous. The strain carrying the variant Rv0678 allele showed a bedaquiline minimum inhibitory concentration of 4 micrograms per millilitre, compared with less than 0.016 micrograms per millilitre for the controls, the lowest concentration tested, so the fold-change is censored. The strain carrying the variant ethA allele showed an ethionamide minimum inhibitory concentration of 256 micrograms per millilitre, sixteen-fold higher than the wild-type strain at 16 micrograms per millilitre and eight-fold higher than the empty-vector control at 32 micrograms per millilitre.

AlphaFold 3 structural modelling then supplied a mechanistic rationale. The Rv0678 frameshift introduces a premature stop codon at position 22, truncating the protein and eliminating the C-terminal DNA-binding repressor domain; loss of this repressor is thought to constitutively upregulate the MmpS5-MmpL5 efflux pump, reducing intracellular bedaquiline accumulation. The ethA deletion likewise introduces a premature stop codon, removing a substantial portion of the C-terminal domain of EthA, a monooxygenase required to convert ethionamide into its active form, the ETO-NAD adduct that normally inhibits InhA and blocks mycolic acid biosynthesis.

The authors are careful to note the limits of the work. The heterologous expression system does not fully reproduce the native regulatory and metabolic environment of Mycobacterium tuberculosis, whose endogenous orthologues MSMEG_1380 and MSMEG_6440 may retain partial functionality, and confirmation in the tuberculosis bacterium itself, ideally through gene knockout and complementation, would further solidify the causal role of the novel deletion. The mechanism by which the structural-variant strains show elevated minimum inhibitory concentrations relative to the empty-vector control also remains unresolved, and the small sample of eleven isolates means the prevalence and broader clinical significance of the new ethA variant across different geographic settings is unknown.

Even so, the study demonstrates that long-read sequencing can uncover both known and novel resistance mechanisms that short-read platforms systematically miss, and the authors argue that integrating long-read sequencing into routine diagnostic workflows would improve detection of resistance-associated structural variants and help guide more effective treatment regimens for patients with drug-resistant tuberculosis.

Subject of Research: A novel ethA structural variant conferring ethionamide resistance in multidrug-resistant Mycobacterium tuberculosis identified by long-read sequencing and functional validation

Article Title: Long-read sequencing of clinical multidrug-resistant Mycobacterium tuberculosis reveals a novel ethA structural variant conferring ethionamide resistance

Article References: Wang, Q., Geng, Y., Wu, L., Xia, D., Li, R., Hu, B., Gao, S., Lu, J., Dai, J., Xu, F., Zheng, X., & Xiang, G. (2026). Long-read sequencing of clinical multidrug-resistant Mycobacterium tuberculosis reveals a novel ethA structural variant conferring ethionamide resistance. Frontiers in Microbiology, 17, 1929307. https://doi.org/10.3389/fmicb.2026.1929307

Image Credits: AI Generated

DOI: 10.3389/fmicb.2026.1929307

Keywords: tuberculosis, multidrug resistance, long-read sequencing, Oxford Nanopore, structural variants, ethA, ethionamide, Rv0678, bedaquiline, Mycobacterium smegmatis, AlphaFold 3, WHO mutation catalogue