Biomolecular map of albumin identifies signatures of severity and early mortality in acute liver failure

Background & Aims: Acute liver failure (ALF) is associated with high mortality. Alterations in albumin structure and function have been shown to correlate with outcomes in cirrhosis. We undertook a biomolecular analysis of albumin to determine its correlation with hepatocellular injury and early mortality in ALF.Methods: Altogether, 225 participants (200 patients with ALF and 25 healthy controls [HC]) were enrolled. Albumin was purified from the baseline plasma of the training cohort (ALF, n = 40; survivors, n = 8; non-survivors, n = 32; and HC, n = 5); analysed for modifications, functionality, and bound multi-omics signatures; and validated in a test cohort (ALF, n = 160; survivors, n = 53; non -survivors, n = 107; and HC, n = 20).Results: In patients with ALF, albumin is more oxidised and glycosylated with a distinct multi-omics profile than that in HC, more so in non-survivors (p <0.05). In non-survivors, albumin was more often bound (p <0.05, false discovery rate <0.01) to proteins associated with inflammation, advanced glycation end product, metabolites linked to arginine, proline metabolism, bile acid, and mitochondrial breakdown products. Increased bacterial taxa (Listeria, Clostridium, etc.) correlated with lipids (triglycerides [4:0/ 12:0/12:0] and phosphatidylserine [39:0]) and metabolites (porphobilinogen and nicotinic acid) in non-survivors (r2 >0.7). Multi-omics signature-based probability of detection for non-survival was >90% and showed direct correlation with albumin func-tionality and clinical parameters (r2 >0.85). Probability-of-detection metabolites built on the top five metabolites, namely, nicotinic acid, L-acetyl carnitine, L-carnitine, pregnenolone sulfate, and N-(3-hydroxybutanoyl)-L-homoserine lactone, showed diagnostic accuracy of 98% (AUC 0.98, 95% CI 0.95-1.0) and distinguish patients with ALF predisposed to early mortality (log-rank <0.05). On validation using high-resolution mass spectrometry and five machine learning algorithms in test cohort 1 (plasma and paired one-drop blood), the metabolome panel showed >92% accuracy/sensitivity and specificity for prediction of mortality.Conclusions: In ALF, albumin is hyperoxidised and substantially dysfunctional. Our study outlines distinct 'albuminome' signa-tures capable of distinguishing patients with ALF predisposed to early mortality or requiring emergency liver transplantation.