Abstract
INTRODUCTION: Submassive hepatic necrosis (SMHN, defined as necrosis of 15-90% of the entire liver on explant) is a likely characteristic pathological feature of ACLF in patients with hepatitis B cirrhosis. We aimed to comprehensively explore microbiome and bile acids patterns across enterhepatic circulation and build well-performing machine learning models to predict SMHN status. METHODS: Based on the presence or absence of SMHN, 17 patients with HBV-related end-stage liver disease who received liver transplantation were eligible for inclusion. Serum, portal venous blood, and stool samples were collected for comparing differences of BA spectra and gut microbiome and their interactions. We adopted the random forest algorithm with recursive feature elimination (RF-RFE) to predict SMHN status. RESULTS: By comparing total BA spectrum between SMHN (-) and SMHN (+) patients, significant changes were detected only in fecal (P = 0.015). Compared with the SMHN (+) group, the SMHN (-) group showed that UDCA, 7-KLCA, 3-DHCA, 7-KDCA, ISOLCA and α-MCA in feces, r-MCA, 7-KLCA and 7-KDCA in serum, γ-MCA and 7-KLCA in portal vein were enriched, and TUDCA in feces was depleted. PCoA analysis showed significantly distinct overall microbial composition in two groups (P = 0.026). Co-abundance analysis showed that bacterial species formed strong and broad relationships with BAs. Among them, Parabacteroides distasonis had the highest node degree. We further identified a combinatorial marker panel with a high AUC of 0.92. DISCUSSION: Our study demonstrated the changes and interactions of intestinal microbiome and BAs during enterohepatic circulation in ACLF patients with SMHN. In addition, we identified a combinatorial marker panel as non-invasive biomarkers to distinguish the SMHN status with high AUC.