A computational framework to in silico screen for drug-induced hepatocellular toxicity

Abstract

Drug-induced liver injury (DILI) is the most common trigger for acute liver failure and the leading cause of attrition in drug development. In this study, we developed an in silico framework to screen drug-induced hepatocellular toxicity (INSIGHT) by integrating the post-treatment transcriptomic data from both rodent models and primary human hepatocytes. We first built an early prediction model using logistic regression with elastic net regularization for 123 compounds and established the INSIGHT framework that can screen for drug-induced hepatotoxicity. The 235 signature genes identified by INSIGHT were involved in metabolism, bile acid synthesis, and stress response pathways. Applying the INSIGHT to an independent transcriptomic dataset treated by 185 compounds predicted that 27 compounds show a high DILI risk, including zoxazolamine and emetine. Further integration with cell image data revealed that predicted compounds with high DILI risk can induce abnormal morphological changes in the endoplasmic reticulum and mitochondrion. Clustering analysis of the treatment-induced transcriptomic changes delineated distinct DILI mechanisms induced by these compounds. Our study presents a computational framework for a mechanistic understanding of long-term liver injury and the prospective prediction of DILI risk.

Keywords: drug-induced liver injury; hepatotoxicity; prediction; toxicogenomics; transcriptomics.

Fig. 1.

INSIGHT: the computational framework to in silico screen for drug-induced hepatocellular toxicity. A) Predicting DILI at day 29 and identifying DILI signature genes using post-treated rat liver profiles at 24 hour by EN model in the Open TG-GATEs database. B) In silico screen for DILI risk of 185 compounds by assessing the enrichment of DILI signature expression alterations in compound treated gene expression profiles in PHH from CMap L1000 database. C) Mechanistic characterization of predicted high DILI-risk compounds by cell morphology changes and functional pathway analysis.

Fig. 2.

DILI prediction using post-treated transcriptomic data and biochemistry markers. A) Overview of the EN model process. B) ROC curves of DILI predictions using post-treated expressions at low, middle, and high dosages. C) DILI prediction accuracy of EN model and biochemistry markers in training set with cross-validation, leave-out test set, and all the samples.

Fig. 3.

EN model identified DILI signature genes with hepatotoxic drug-induced expression alterations in a dose-dependent manner and enriched in DILI-associated functional pathways. A) Pairwise Pearson correlation coefficient between the expressions of the DILI signature genes across all the samples. B) Significantly increased expression of Dnm1l (Rho = 0.28, P = 6.8 × 10^-13) and decreased expression of Slc27a5 (Rho = -0.38, P = 1.3 × 10^-13) after hepatoxic drug treatment in a dose-dependent manner. The significant dose-dependent up-/down-regulation in the expression of Dnm1l/Slc27a5 was not observed in non-hepatotoxic drug-treated samples. C) PS of the DILI signature genes and the dose-dependent expression changes of the DILI signature genes after the hepatotoxic drug treatments. The heatmap showed the correlation between the gene expression and dose level of each drug treatment. D) DILI signature enriched with Reactome pathways.

Fig. 4.

DILI risk prediction of 185 compounds using INSIGHT. A) The INSIGHT prediction for 185 compounds and the DMSO treatment in the CMap L1000 database. Each dot represents one individual treatment. The INSIGHT scores (x-axis) and their log-transformed P-values (y-axis) are shown for each prediction. B) Twenty-seven compounds are predicted to be with high DILI risk. C) Emetine killing effect of HepG2 cell lines after treatment for 24, 32, and 40 hour.

Fig. 5.

Predicted high DILI-risk compounds induced dysregulation of endoplasmic reticulum and mitochondrial function. A) Differential morphological alterations between predicted high DILI-risk compound and low DILI-risk compound treatments. The x-axis represents the difference in morphology features between the two groups, and the y-axis represents log-transformed P-values of difference. B) Individual morphology feature values for compounds. C) Hierarchical clustering analysis of the predicted high DILI-risk compounds based on the treatment-induced pathway changes measured by the normalized enrichment score (NES) in GSEA. Two clusters of compounds were identified. D) Significantly upregulated pathways by a cluster of the predicted high DILI-risk compounds. The dot represents the significance of the enrichment from GSEA analysis. E) Different oxidative stress signature scores between the two clusters of compounds.