The Metabolomics of Non-Alcoholic Fatty Liver Disease: Of Networks and Biomarkers

• Published in: Journal of hepatology 2021, Vol.75 (Suppl. 2), p.S579
• Main Authors: McGlinchey, Aidan J, Geng, Dawei, Govaere, Olivier, Ratziu, Vlad, Bugianesi, Elisabetta, Schattenberg, Joern M., Daley, Ann, Hyötyläinen, Tuulia, Anstee, Quentin, Oresic, Matej
• Format: Article
• Language: English
• ISSN: 1600-0641; 0168-8278

Background and aims : Non-alcoholic fatty liver disease (NAFLD), the leading cause of chronic liver disease, affects 25%+ of people worldwide. Detailed understanding of the metabolomics of NAFLD, and non-invasive diagnostic techniques for the stages of NAFLD are unavailable. We identify specific serum molecular lipid signatures to these ends. First, we leverage lipidomic and polar metabolomic data (n = 643) subjects, to produce a clear, meaningful interaction map, linking lipids, metabolites, clinical factors and disease outcomes. We find non-spurious associations therein, as features of interest, and for downstream analysis. Third, NAFLD fibrosis biomarker identification was performed using machine learning, with our candidate lipids/metabolites to be forwarded to a successor project; the LITMUS project, towards clinically-applicable, non-invasive, sensitive and specific classification of NAFLD patients. Method : Serum lipids and polar metabolites were measured by mass spectrometry in the EPoS cohort of patients (n = 176 lipids and n = 36 polar metabolites), combined with clinical data from (n = 643 subjects), followed by model-based clustering, giving 10 lipid clusters (LCs). Correlations were calculated pairwise between (1) all LCs, (2) “input” clinical data (height, weight, BMI, blood platelet count) and (3) outcomes (fibrosis, steatosis, NAS score, etc.). Non-rejection rates (NRRs) were calculated for relationships, remove spurious associations (NRR > 0.4). We project the remaining associations as a network; a novel metabolomic overview NAFLD. ANOVA and Tukey’s Honest Significant Differences (Tukey HSDs) revealed detailed metabolic signatures across NAFLD, fibrosis and steatosis stages. Random forest machine learning was used to classify NAFLD patients: LOW (0-1 fibrosis grade) or HIGH (2–4 fibrosis grade), using individual lipids and metabolites, identifying putative biomarkers. Results : In linewith our previous findings, many lipids associate with steatosis and fibrosis in NAFLD. Our novel overview network revealsas sociations between specific LCs and clinical variables, such as TGs (LC3), and a subgroup of TGs of lowest and highest carbon numbers (LC9) along with PC (O)s (LC7) positively associating with NAFLD score and fibrosis. Conversely, LPCs (LC4), particularly sphingomyelins (SMs, LC6), negatively associated with these variables. Many other metabolites changing across NAFLD stages beg further discussion. Conclusion : In addition to genera