Transcriptomics: Blood Signature Separates MAC Lung Disease
- Aug 12
- 4 min read
Only about a quarter of people whose sputum grows Mycobacterium avium complex (MAC) have progressive lung disease. The rest are colonized, and separating the two rests on nonspecific cough and CT patterns that blur into the bronchiectasis these patients already carry. A two-hospital Taiwanese study took that problem to blood transcriptomics, sequencing RNA from peripheral blood mononuclear cells (PBMCs) in 120 culture-positive patients.
Key Takeaways
Blood transcriptomics separated MAC pulmonary disease from airway colonization at an external validation AUC of 0.78 across two independent medical centers.
The five-gene core signature (IGKV1D-39, IGKV6-21, OVCH1, PLAU, DMD) implicates B cell activation and matrix remodeling, not generic inflammation.
Support vector machines generalized to the second hospital; random forests trained on the same genes did not.
The design is cross-sectional, so this is a candidate diagnostic aid, not a tool for predicting progression.
What Blood Transcriptomics Measured Here
Cohort 1 in Kaohsiung supplied 74 patients (48 disease, 26 colonization) for discovery. Cohort 2 in Taipei added 46 more and was held back for validation, with models applied as trained and never refit. Libraries ran on a NovaSeq 6000 at paired-end 150 bp, so this is bulk transcriptomics of circulating immune cells rather than site-of-disease tissue. Patients averaged 68.2 years and three quarters had bronchiectasis.
Key Findings
147 differentially expressed genes, only 29 past FDR 0.05: the transcriptional gap between infection and colonization is real but narrow.
A five-gene core survived aggressive pruning: LASSO plus recursive feature elimination, run across four random seeds, cut 83 candidates to 18, and the same five genes recurred across top models.
DMD dropped hardest, at log2 fold change -3.15, while PLAU moved the other way, pointing toward fibrinolysis and matrix turnover in damaged airways.
Algorithm choice was not cosmetic: the three best models were all support vector machines, at validation AUCs of 0.78, 0.78, and 0.75.
Enrichment pointed at humoral immunity, led by antigen binding, immunoglobulin complex, and adaptive immune response.
Figure 1. Analysis workflow, running from differential expression and Gene Ontology analysis through LASSO and recursive feature elimination, model development with support vector machines and random forests, post-selection filtering, and external validation. Cohort 1 (n = 74) supported discovery; Cohort 2 (n = 46) served as the untouched validation set. Adapted from Lin et al. (2026), Emerging Microbes and Infections.
An Immunoglobulin Signal, Not an Interferon One
The gene list is the interesting part. Whole-blood work in mycobacterial disease has been dominated by interferon-inducible signatures since Berry and colleagues (2010) described one in active tuberculosis, so an immunoglobulin variable region signal is a departure. Chronic antigen exposure during slow infection plausibly drives that humoral expansion, and it suits the timescale of MAC disease better than an acute interferon burst would.
Reading an AUC of 0.78 Honestly
An AUC of 0.78 is useful, not decisive. It will not replace ATS/IDSA criteria, and the authors say so; what it can do is shift pretest probability for the ambiguous patient whose culture is positive and whose imaging is equivocal. The immunoglobulin genes carry a caveat, since they sit among the most polymorphic loci in the genome and their quantification is sensitive to alignment choices. Anyone reproducing this panel should expect batch correction across sequencing runs to matter as much as the modeling step, a pattern also visible in airway biomarker work on lung allograft decline.
Implications for Diagnostic Biomarker Programs
For teams building diagnostics in chronic infection, the practical takeaway is that a compact panel anchored in B cell repertoire and matrix genes travels across sites better than a broad inflammation score does. We see the same pattern in client work at Dalton, where RNA-seq and targeted expression panels hold up in validation mainly when the discovery cohort is phenotyped as tightly as the validation cohort. When one layer stalls near an AUC of 0.78, multi-omics biomarker discovery is usually the cheaper next move, since protein or metabolite evidence from the same draw costs less than recruiting another 200 patients.
Frequently Asked Questions
What is transcriptomics used for in clinical research?
Transcriptomics measures which genes are actively transcribed in a sample, usually by RNA-seq. Clinical studies use it to find expression signatures that separate disease states, stratify patients, or report on drug response. Blood is the common sample type because it can be drawn repeatedly using standard protocols.
Can blood transcriptomics distinguish MAC lung disease from colonization?
In this 120-patient study it did, at an external validation AUC of 0.78 from a five-gene panel. That is strong enough to inform a hard clinical call, not to make it alone. Prospective cohorts are needed before the signature could guide treatment.
How many patients does an RNA-seq biomarker study need?
This study is a reasonable floor: 74 for discovery, 46 for independent validation, with panel size capped near the N/10 rule to limit overfitting. Budgeting for a second, independently collected cohort matters more than adding sequencing depth to the first.
Conclusion
What the data support is that circulating immune cells carry a measurable, site-transferable trace of active MAC infection, and that the trace is humoral and structural rather than interferon-driven. The panel cannot yet decide who starts a multi-drug regimen, since nobody has shown that the signature moves with treatment. Its near-term value is as a tiebreaker for the equivocal culture-positive patient.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Lin, W.-Y., Huang, H.-L., Hsiung, C.-N., Huang, Y.-Y., Lee, M.-R., Cheng, M.-H., Chong, I.-W., & Wang, J.-Y. (2026). Blood transcriptomic signatures distinguish Mycobacterium avium complex pulmonary disease from colonization: a multicenter cohort study. Emerging Microbes & Infections, 15(1), 2678657. https://doi.org/10.1080/22221751.2026.2678657
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.