Integração de abordagens ômicas na investigação de alterações metabólicas associadas à hepatite D

Abstract

Background and aims: Hepatitis D virus (HDV) is a single-stranded circular RNA satellite virus with hepatocyte-specific tropism that occurs in the presence of hepatitis B virus (HBV), being functionally dependent on it to complete its infectious cycle. Transmission occurs predominantly via the parenteral route and may present as simultaneous coinfection or as superinfection in individuals chronically infected with HBV, the latter being associated with more severe clinical outcomes, including early cirrhosis and an increased risk of hepatocellular carcinoma. Given the central role of the liver in metabolic regulation, HDV infection is expected to induce alterations in systemic metabolism. However, to date, comprehensive characterization of these alterations, particularly through omics-based approaches, remains limited. In this context, this study aimed to investigate serum metabolomic and lipidomic signatures capable of distinguishing HDV infection from HBV monoinfection, as well as to explore their discriminative potential. Materials and methods: This exploratory study employed untargeted metabolomic and lipidomic approaches based on high-resolution liquid chromatography–mass spectrometry (LC–MS) and gas chromatography–mass spectrometry (GC–MS), respectively. Residual anonymized serum samples from individuals with confirmed HDV infection and HBV-monoinfected controls from the Brazilian Amazon were provided by the Central Laboratory of Public Health of Acre and analyzed. Exploratory analyses were performed, including principal component analysis (PCA), as well as volcano plots and heatmaps to identify differentially abundant features. Predictive modeling using supervised machine learning algorithms was applied to evaluate discriminative performance. Results: Metabolomic analysis identified six metabolites significantly altered in HDV infection, predominantly related to lipid and phospholipid metabolism. Linoleic acid, heptadecanoic acid, 3-hydroxyicosanoic acid, pregnanolone, and choline showed higher abundance, whereas LysoPC(18:0/0:0) showed lower abundance in HDV-positive serum samples. Machine learning models demonstrated good discriminative performance, with Gradient Boosting achieving an area under the curve (AUC) of 0.943, with sensitivity and specificity of 86%. In contrast, lipidomic analysis revealed four lipid species with significantly lower abundance in HDV-positive samples, particularly cholesterol-derived compounds. Random Forest showed the best performance in this dataset, with an AUC of 0.854, sensitivity of 84.8%, and specificity of 100%. In both approaches, model interpretability, assessed using Shapley additive explanations (SHAP) and permutation-based feature importance, identified the same metabolites and lipids as the main contributors to model performance. Conclusions: HDV infection is associated with distinct metabolomic and lipidomic alterations in the context of HBV infection, suggesting a disruption of metabolic homeostasis, possibly associated with the severity of liver injury. These findings provide preliminary evidence supporting the use of metabolomic and lipidomic signatures, integrated with artificial intelligence analyses, for biomarker investigation, with potential applications in improving diagnosis and in characterizing alterations associated with hepatitis D. Further studies with larger and longitudinal cohorts are required to validate these findings.