Metabolic heterogeneity, networks, and biomarkers of drug-induced liver injury

TL;DR

This study explores metabolic differences in drug-induced liver injury (DILI) and identifies specific biomarkers to improve diagnosis and treatment.

Contribution

The study introduces a 10-metabolite classifier and drug-specific metabolic networks to better understand and diagnose DILI subtypes.

Findings

Metabolic disruptions in glutathione, fatty acid, and carnitine metabolism are significant in DILI progression.

Elevated long-chain acylcarnitines like C18:1 Car and C16:2 Car are linked to herb-induced DILI through Cpt2 suppression.

A machine learning-based 10-metabolite model accurately distinguishes DILI subtypes with high accuracy.

Abstract

Drug-induced liver injury (DILI) represents a major adverse drug reaction with significant clinical implications. The diversity of causative drug agents, incomplete understanding of pathogenic mechanisms, and absence of specific diagnostic biomarkers pose substantial challenges for DILI diagnosis and clinical management. This study aimed to characterize the metabolic heterogeneity across different types of DILI and identify high-specificity metabolic biomarkers for DILI classification. A multicenter targeted metabolomics study was conducted on 516 serum samples collected from 200 patients with DILI and 221 healthy controls. We characterized the metabolic dynamics throughout DILI progression, with significant disruptions presented in glutathione, fatty acid, and carnitine metabolism. By characterizing and comparing the metabolic profiles among antibiotics-, herbs-, non-steroidal anti-inflammatory drugs-, and statins-DILI patients, we constructed four drug-specific metabolic networks of DILI based on the metabolic coordination between metabolites. Notably, the elevated long-chain acylcarnitines (such as C18:1 Car and C16:2 Car) distinctively underlie herb-DILI's pathological progression. In monocrotaline-induced liver injury mouse models, hepatic carnitine acyltransferase II (Cpt2) mRNA expression was suppressed. Further, two-sample Mendelian randomization supported a causal relationship between C18:1 Car and total bilirubin levels. Finally, we developed a 10-metabolite classifier to distinguish between different DILI subtypes using machine learning algorithms, yielding accuracies of 0.915 and 0.904 on two independent test sets. These findings enhance the understanding of the metabolic heterogeneity in DILI and provide evidence supporting the use of responsive metabolic traits for the clinical diagnosis and treatment of DILI.

Image 1

•Metabolites' change trajectory and dynamics throughout DILI's disease progress were illustrated.•Metabolic networks of antibiotics-, herbs-, NSAIDS-, and statins-DILI were constructed based on metabolic coordination.•Elevated long-chain acylcarnitines potentially contribute to progression of herbs-DILI via Cpt2.•Mendelian randomization revealed the casual effect between C18:1 Car and TBIL.•A 10-metabolite model for DILI identification was developed with accuracies of 0.915 and 0.904.

Metabolites' change trajectory and dynamics throughout DILI's disease progress were illustrated.

Metabolic networks of antibiotics-, herbs-, NSAIDS-, and statins-DILI were constructed based on metabolic coordination.

Elevated long-chain acylcarnitines potentially contribute to progression of herbs-DILI via Cpt2.

Mendelian randomization revealed the casual effect between C18:1 Car and TBIL.

A 10-metabolite model for DILI identification was developed with accuracies of 0.915 and 0.904.

Figures