Abstract
A large proportion of the global tuberculosis (TB) burden is asymptomatic and not detectable by symptom-based screening, driving the TB epidemic through continued M. tuberculosis transmission. Currently, no validated tools exist to diagnose incipient and subclinical TB. Nested within a large prospective study in household contacts of pulmonary TB cases in Southern India, we assessed 35 incipient TB and 12 subclinical TB cases, along with corresponding household active TB cases (n=11), and household controls (n=39) using high throughput methods for transcriptional and protein profiling. We split the data into training and test sets and applied a support vector machine classifier followed by a Lasso regression model to identify signatures. The Lasso regression model identified an 11-gene signature (ABLIM2, C20orf197, CTC-543D15.3, CTD-2503O16.3, HLADRB3, METRNL, RAB11B-AS1, RP4-614C10.2, RNA5SP345, RSU1P1, and UACA) that distinguished subclinical TB from incipient TB with a very good discriminatory power by AUCs in both training and test sets. Further, we identified an 8-protein signature comprising b-FGF, IFNγ, IL1RA, IL7, IL12p70, IL13, PDGF-BB, and VEGF that differentiated subclinical TB from incipient TB with good and moderate discriminatory power by AUCs in the training and test sets, respectively. The identified 11-gene signature discriminated well between the distinct stages of the TB disease spectrum, with very good discriminatory power, suggesting it could be useful for predicting TB progression in household contacts. However, the high discriminatory power could partly be due to over-fitting, and validation in other studies is warranted to confirm the potential of the immune biosignatures for identifying subclinical TB.