Patients with pediatric cirrhosis-sepsis (PC-S) attain early mortality. Plasma bacterial composition, the cognate metabolites, and their contribution to the deterioration of patients with PC-S to early mortality are unknown. We aimed to delineate the plasma metaproteome-metabolome landscape and identify molecular indicators capable of segregating patients with PC-S predisposed to early mortality in plasma, and we further validated the selected metabolite panel in paired 1-drop blood samples using untargeted metaproteomics-metabolomics by UHPLC-HRMS followed by validation using machine-learning algorithms. We enrolled 160 patients with liver diseases (cirrhosis-sepsis/nonsepsis [n=110] and noncirrhosis [n=50]) and performed untargeted metaproteomics-metabolomics on a training cohort of 110 patients (Cirrhosis-Sepsis/Nonsepsis, n=70 and noncirrhosis, n=40). The candidate predictors were validated on 2 test cohorts-T1 (plasma test cohort) and T2 (1-drop blood test cohort). Both T1 and T2 had 120 patients each, of which 70 were from the training cohort. Increased levels of tryptophan metabolites and Salmonella enterica and Escherichia coli-associated peptides segregated patients with cirrhosis. Increased levels of deoxyribose-1-phosphate, N5-citryl-d-ornithine, and Herbinix hemicellulolytic and Leifsonia xyli segregated patients with PC-S. MMCN-based integration analysis of WMCNA-WMpCNA identified key microbial-metabolic modules linked to PC-S nonsurvivors. Increased Indican, Staphylobillin, glucose-6-phosphate, 2-octenoylcarnitine, palmitic acid, and guanidoacetic acid along with L. xyli, Mycoplasma genitalium, and Hungateiclostridium thermocellum segregated PC-S nonsurvivors and superseded the liver disease severity indices with high accuracy, sensitivity, and specificity for mortality prediction using random forest machine-learning algorithm. Our study reveals a novel metabolite signature panel capable of segregating patients with PC-S predisposed to early mortality using as low as 1-drop blood.
Read full abstract