Olivier Béquignon Abstract

Combined efforts of in vitro-derived data and computational chemistry for drug-induced liver injury prediction

Olivier J. M. Béquignon1, Steven Wink1, Gerard J. P. van Westen1, Bob van de Water1

1Division of Drug Discovery and Safety, LACDR, Leiden University, Leiden, The Netherlands
Drug-induced liver injury (DILI) is one of the main reasons of drug attrition during clinical trials and of drug withdrawal from the market1. This makes the early identification of hepatotoxicity of compounds a critical challenge. In silico hepatotoxicity prediction models make a cost-effective approach able to prioritize compounds for preclinical and clinical studies. Recent efforts have been made to create more accurate quantitative structure-property relationships (QSPR) models relating hepatotoxicity to chemical structure features2,3. However, to date only few have integrated in vitro quantitative concentration response mechanistic data to their models4; high content imaging allows for such detailed quantitative mode-of-action information5. This study aims at integrating in vitro activation signal of stress response pathways involved in DILI to standard molecular description of compounds to better identify sub-structures that contribute to the liability for DILI.

A library of 103 drugs with different liability for DILI was screened on previously established HepG2 BiP/Chop, p21/BTG2, HMOX1/SRXN1, ICAM1 and HSPA1B BAC-GFP reporter cell lines6. These cell lines are respectively associated with endoplasmic reticulum (ER), DNA damage, oxidative, inflammatory and heat-shock stress response pathways involved in DILI. Confocal microscopy images were obtained at 24, 48 and 72 hours after addition of the compounds (concentration ranging from 1 to 100 cmax). Additionally, propidium iodide (PI) and annexin V (AnxV) staining was performed to detect necrotic and apoptotic cells. Quantitative multiparameter image analysis was performed with CellProfiler7. GFP integrated signal above 1, 2, 3 and 4 times the median, as well as cmax, AnxV and PI values were used as descriptors of the compounds along with physicochemical and topological descriptors of the compound structures. Gradient Boosted Decision Tree classification models were built from 70% of compounds to relate both in vitro derived data and molecular structure description to the FDA DILI annotation converted to binary scale (Most and No-DILI-concern). The binary set was balanced using oversampling. Finally, models were tested on the internal set consisting of the remaining 30% of compounds. So far, the results show that stacked modelling built from in vitro-derived data combined with physicochemical and topological descriptors of the compound structures created highly predictive models. Sensitivity 0.96, Matthew’s correlation coefficient 0.88 and specificity 0.95.

1. Chen M, Vijay V, Shi Q, Liu Z, Fang H, Tong W. FDA-approved drug labeling for the study of drug-induced liver injury. Drug Discov Today. 2011;16(15-16):697-703. doi:10.1016/j.drudis.2011.05.007
2. Ekins S. Progress in computational toxicology. J Pharmacol Toxicol Methods. 2014;69(2):115-140. doi:10.1016/j.vascn.2013.12.003
3. Przybylak KR, Cronin MTD. In silico models for drug-induced liver injury — current status. Expert Opin Drug Metab Toxicol. 2012;8(2):201-217. doi:10.1517/17425255.2012.648613
4. Muster W, Breidenbach A, Fischer H, Kirchner S, Müller L, Pähler A. Computational toxicology in drug development. Drug Discov Today. 2008;13(7-8):303-310. doi:10.1016/j.drudis.2007.12.007
5. Tolosa L, Pinto S, Donato MT, et al. Development of a Multiparametric Cell-based Protocol to Screen and Classify the Hepatotoxicity Potential of Drugs. Toxicol Sci. 2012;127(1):187-198. doi:10.1093/toxsci/kfs083
6. Wink S, Hiemstra S, Herpers B, van de Water B. High-content imaging-based BAC-GFP toxicity pathway reporters to assess chemical adversity liabilities. Arch Toxicol. 2017;91(3):1367-1383. doi:10.1007/s00204-016-1781-0
7. Kamentsky L, Jones TR, Fraser A, et al. Improved structure, function and compatibility for cellprofiler: Modular high-throughput image analysis software. Bioinformatics. 2011;27(8):1179-1180. doi:10.1093/bioinformatics/btr095