Abstract
Early diagnosis of tuberculosis (TB) in children is hampered by non-specific symptoms, low bacillary loads, and difficulties in obtaining respiratory samples. We evaluated urinary metabolomics using high-resolution nuclear magnetic resonance (HR-(1)H-NMR) spectroscopy as a non-invasive tool to discriminate TB disease, TB infection (TBI), and healthy controls (HC). Urine samples from 101 children enrolled in the Spanish Paediatric TB Network (pTBred) were retrospectively analysed, including 62 TB cases (26 microbiologically confirmed [cTB]), 17 TBI, and 22 HC. Metabolic fingerprints were generated via(1) H-NMR spectroscopy. Group discrimination was assessed using partial least squares discriminant analysis (PLS-DA) with cross-validation. Metabolites were selected using VIP > 2 and p < 0.05. Validated PLS-DA models (TB vs. HC, cTB vs. HC and cTB vs. TBI) achieved ≥ 70% accuracy (AUC-ROC 0.867-0.971), with sensitivities ranging from 69.1% to 90.0% and specificities from 77.3% to 86.7%. Discrimination was strongest for cTB vs. TBI (AUC-ROC 0.971; sensitivity 90.0%; specificity 86.7%). A set of discriminatory metabolites was identified across validated comparisons. Compared with HC, TB disease was characterized by increased levels of phenylalanine and two unidentified NMR signals, together with reductions in several energy and nitrogen-related metabolites; benzoate and histidine showed borderline differences. In contrast, TBI was associated with higher levels of isoleucine, N-acetylglutamine, glutamine, creatinine and 2-hydroxyvalerate. Urine HR-(1)H-NMR metabolomics suggests potential distinct metabolic signatures across TB states. While not yet suitable as a confirmatory test, these preliminary findings show potential as a triage tool to help prioritise children for microbiological testing and may eventually reduce the need for invasive sampling.