Abstract
Background
Tuberculosis (TB) remains a leading global cause of infectious disease mortality, while current diagnostic methods face substantial limitations. Extracellular vesicle (EV)–encapsulated microRNAs (miRNAs) demonstrate significant potential as diagnostic biomarkers. This study aimed to evaluate their diagnostic utility for TB by identifying expression disparities between TB patients and healthy controls (HCs) and exploring EV miRNA diagnostic performance.
Methods
HCs were healthy individuals excluded from pulmonary infectious diseases, and TB patients were clinically and laboratory‐confirmed cases without other concurrent pulmonary infections. Small RNA sequencing analysis was performed on plasma EVs from a discovery cohort (
n
= 40) to identify differentially expressed (DE) miRNAs. Findings were validated via quantitative real‐time PCR (qRT‐PCR) in an independent cohort (
n
= 24). Subsequently, an EV miRNA combinatorial diagnostic model was developed using random forest classifiers in a modeling cohort (
n
= 58), incorporating feature selection based on importance scores and correlation coefficients. Hyperparameter optimization and model stability verification were performed via grid search with five‐fold cross‐validation. Diagnostic performance was evaluated through receiver operating characteristic (ROC) analysis. Bioinformatics analysis included target gene prediction using miRNA databases, followed by Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses.
Results
Sequencing revealed 67 significantly DE EV miRNAs, with 92.5% demonstrating marked downregulation in TB patients. This pattern was confirmed by qRT‐PCR validation. Eleven validated miRNAs were used for model construction. Feature selection identified three optimal miRNAs (miR‐19b‐3p, miR‐181b‐5p, and miR‐98‐5p), generating seven combinatorial models. Individual miRNAs exhibited area under the curves (AUCs) > 0.85, while combinations achieved AUCs > 0.89. Notably, the combinatorial signature of miR‐19b‐3p and miR‐181b‐5p demonstrated outstanding diagnostic discrimination (AUC = 1.000; sensitivity and specificity both 100%), with a five‐fold cross‐validated score of 0.964 indicating robust model stability. Bioinformatics analysis revealed that the target genes of these two miRNAs were enriched in immune response pathways, including the Notch signaling pathway and the cyclic adenosine monophosphate (cAMP) signaling pathway.
Conclusion
This EV miRNA signature demonstrates remarkable diagnostic accuracy for TB, offering a clinically actionable solution to diagnostic challenges.