A Cross-Compartment Five-Gene Host Transcriptomic Signature Distinguishes Active Tuberculosis in Whole Blood and Plasma Cell-free RNA

This is the authors' abstract. We don't add key points for this paper.

Background Tuberculosis (TB) remains one of the deadliest infectious diseases worldwide, with gaps in timely, accurate diagnosis sustaining transmission and disproportionate burden in low- and middle-income countries. Current sputum-based tests show limited sensitivity and poorly distinguish active TB (ATB) from latent infection (LTBI), highlighting the need for reliable non-sputum biomarkers reflecting host disease status. Objective To investigate the conservation of tuberculosis-associated host transcriptomic signals between plasma cell-free RNA and whole blood, and to identify a minimal gene signature for accurate disease classification. Methods We applied an integrated transcriptomic and machine learning framework to discover a biologically plausible minimal gene signature associated with ATB. RNA sequencing data of whole blood, and plasma cell-free RNA available publicly were analysed independently by DESeq2 to determine differentially expressed genes among ATB, LTBI, and healthy controls. Core disease-associated pathways were determined using functional enrichment and protein-protein interaction network analyses. Genes were prioritized based on biological relevance, pathway convergence, and cross-dataset consistency. The resulting gene signature was used to train the machine learning classifiers to evaluate diagnostic performance. Results Transcriptomic and network analyses revealed that ATB is characterized by interferon-driven immune activation, neutrophil-mediated inflammation, and dysregulated immune regulation, detectable in both whole blood and cfRNA. A five-gene host signature ( GBP5 , GBP1 , ANKRD22 , CCR5AS , and KLF2 ) demonstrated robust diagnostic performance in both whole blood and cfRNA, capturing shared host-response biology despite compartment-specific expression differences. Machine learning models trained on this gene set achieved high diagnostic accuracy, with area under the receiver operating characteristic curve values of 0.98 in whole blood and 0.94 in cfRNA. Notably, CCR5AS , a long non-coding RNA, emerged as a previously underexplored component of host transcriptional responses in TB. Conclusions A five-gene host-response signature, conserved in whole blood and plasma cfRNA, reliably distinguishes active tuberculosis from latent infection and healthy controls, enabling minimal, cross-platform diagnostics.