People with multidrug-resistant tuberculosis had a less diverse community of microbes in their lower airways than people with drug-sensitive tuberculosis, according to a small observational study. The groups also differed in which microbes were most prominent, suggesting that tuberculosis drug-susceptibility categories were associated with distinct lung microbiome profiles.
The study included 30 patients with active pulmonary tuberculosis: 11 with drug-sensitive disease and 19 with multidrug-resistant disease. They were selected from 150 patients screened prospectively at Wenzhou Central Hospital between January 2021 and December 2023. Drug-susceptibility classification used GeneXpert MTB/RIF, mycobacterial culture and drug-susceptibility testing, and treatment before the lung-fluid sample was collected did not exceed 14 days.
Researchers examined bronchoalveolar lavage fluid, or BALF, collected by bronchoscopy from the affected lung lobe. At least 15 millilitres was collected, kept on ice and stored at minus 80 degrees Celsius. The DNA was sheared to about 350 base pairs and paired-end sequenced, with negative controls processed alongside the patient samples.
A clear separation between the groups
The study's four reported measures of alpha-diversity, which summarise the microbial mix within each sample, were all higher in the drug-sensitive group. The differences were statistically significant for the Chao1, ACE, Shannon and Simpson indices, with all p-values below 0.05. The supplied results do not report the exact index values or the size of each difference.
The overall mix of microbes also separated the two groups. A statistical test called PERMANOVA estimated that drug-susceptibility category accounted for 55% of the variation in community composition, with p below 0.05. A second test, ANOSIM, produced an R value of 0.808 and a p-value of 0.0001. These figures describe a strong association in this cohort, not proof that drug resistance caused the microbial differences.
Streptococcus and Prevotella were the most prominent organisms reported in the drug-sensitive samples, with relative-abundance summaries of 14.13% plus or minus 8.05% and 12.78% plus or minus 6.75%, respectively. In the multidrug-resistant samples, Staphylococcus, Cutibacterium and Mycobacterium were the leading organisms, at 13.69% plus or minus 1.45%, 12.24% plus or minus 1.30% and 11.87% plus or minus 1.54%. These are relative proportions, not measurements of the total amount of microbes in the lungs.
What the sequences suggested about microbial activity
The researchers used the sequence data to predict what the microbial community might be capable of doing. The predicted functional profiles differed between the groups, with a PERMANOVA R-squared value of 0.54 and p below 0.05. Multidrug-resistant tuberculosis samples were relatively enriched for annotations linked to infectious disease and immune-system functions. Drug-sensitive samples showed broader enrichment for carbohydrate, amino-acid, nucleotide and lipid metabolism, as well as translation, DNA repair, transport and energy metabolism. Because these were predicted annotations, the analysis does not directly show that those activities were taking place.
The study also examined CAZymes, enzymes involved in processing carbohydrates. Sixty-eight CAZyme families were identified. Forty-five families were enriched in the drug-sensitive group, while CBM48 and GT25 were enriched in the multidrug-resistant group. The overall CAZyme composition differed between groups, with a PERMANOVA R-squared value of 0.52 and p below 0.05. This was an exploratory comparison based on relative abundance.
Correlation-based networks painted a similar picture of different community structures. The drug-sensitive network contained 25 microbial nodes and 176 connections, compared with 24 nodes and 95 connections in the multidrug-resistant network. Its clustering coefficient was 0.77 versus 0.45, and its density was 0.59 versus 0.34. Mycobacterium had the highest number of connections in the integrated network, with a degree of 9. The links represent statistical co-occurrence patterns, however, not experimentally demonstrated biological interactions.
A candidate signature with a large caveat
An exploratory random-forest analysis selected six candidate genera: Mycobacterium, Jimgerdemannia, Escherichia, Enterococcus, Bacillus and Acinetobacter. In internal five-fold cross-validation, a model built from those six genera had an area under the curve of 0.957 and an accuracy of 0.900. A model using 12 core taxa had a lower area under the curve of 0.885 and an accuracy of 0.867. These estimates came only from internal testing within the small study and may be optimistic.
The authors describe the six-genus panel as a preliminary candidate signature rather than a validated diagnostic test. The study was cross-sectional, so it cannot establish whether the microbiome differences came before drug resistance, followed it or reflected another factor. It was also conducted at a single centre with a small selected cohort, and the functional, network and classification analyses were exploratory. No independent validation group or formal a priori power calculation was reported.
The findings therefore point to a possible lower-airway microbiome pattern associated with multidrug-resistant tuberculosis, but they do not show that the microbes cause drug resistance or that changing the microbiome would alter treatment outcomes. Larger multicentre studies that follow patients over time will be needed to test whether the six-genus pattern can be reproduced and whether the predicted functional differences correspond to measured microbial activity or clinically meaningful outcomes.
Paper data and sources
Original title: Metagenomic insights into the pulmonary microbiome of drug-sensitive and multidrug-resistant tuberculosis patients
Authors: Lin Z, Shi J, Qiu C et al.
Journal/Repository: Annals of medicine
Status: Peer-reviewed
First online: 2026-10-19
DOI: Not available
Original paper · Full text