Untargeted metabolomics as a diagnostic tool in NAFLD: discrimination of steatosis, steatohepatitis and cirrhosis

• Published in: Metabolomics Vol. 17; no. 2; p. 12
• Main Authors: Masarone, Mario, Troisi, Jacopo, Aglitti, Andrea, Torre, Pietro, Colucci, Angelo, Dallio, Marcello, Federico, Alessandro, Balsano, Clara, Persico, Marcello
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2021; Springer Nature B.V
• Subjects: Biochemistry; Biomedical and Life Sciences; Biomedicine; Cell Biology; Cirrhosis; Developmental Biology; Disease; Fatty liver; Glutathione; Hepatocellular carcinoma; Learning algorithms; Life Sciences; Liver cancer; Liver cirrhosis; Liver diseases; Mathematical models; Metabolites; Metabolomics; Molecular Medicine; Original Article; Pathophysiology; Phenylalanine; Steatosis; Taurocholic acid; NAFLD; NASH; Untargeted metabolomics; Ensemble machine learning diagnostics
• ISSN: 1573-3882; 1573-3890; 1573-3890
• DOI: 10.1007/s11306-020-01756-1

Introduction Non-Alcoholic Fatty Liver Disease encompasses a spectrum of diseases ranging from simple steatosis to steatohepatitis (or NASH), up to cirrhosis and hepatocellular carcinoma (HCC). The challenge is to recognize the more severe and/or progressive pathology. A reliable non-invasive method does not exist. Untargeted metabolomics is a novel method to discover biomarkers and give insights on diseases pathophysiology. Objectives We applied metabolomics to understand if simple steatosis, steatohepatitis and cirrhosis in NAFLD patients have peculiar metabolites profiles that can differentiate them among each-others and from controls. Methods Metabolomics signatures were obtained from 307 subjects from two separated enrollments. The first collected samples from 69 controls and 144 patients (78 steatosis, 23 NASH, 15 NASH-cirrhosis, 8 HCV-cirrhosis, 20 cryptogenic cirrhosis). The second, used as validation-set, enrolled 44 controls and 50 patients (34 steatosis, 10 NASH and 6 NASH-cirrhosis).The “Partial-Least-Square Discriminant-Analysis”(PLS-DA) was used to reveal class separation in metabolomics profiles between patients and controls and among each class of patients, and to reveal the metabolites contributing to class differentiation. Results Several metabolites were selected as relevant, in particular:Glycocholic acid, Taurocholic acid, Phenylalanine, branched-chain amino-acids increased at the increase of the severity of the disease from steatosis to NASH, NASH-cirrhosis, while glutathione decreased (p < 0.001 for each). Moreover, an ensemble machine learning (EML) model was built (comprehending 10 different mathematical models) to verify diagnostic performance, showing an accuracy > 80% in NAFLD clinical stages prediction. Conclusions Metabolomics profiles of NAFLD patients could be a useful tool to non-invasively diagnose NAFLD and discriminate among the various stages of the disease, giving insights into its pathophysiology.