(180f) DILI-Sim: A Mechanistically Based Computational Model for Predicting Drug Induced Liver Injury (DILI)

Drug-induced liver injury (DILI) is the adverse drug event that most frequently leads to termination of clinical development programs, as well as to regulatory actions on drugs, including failure to gain approval for marketing, restrictions in labeling, and withdrawal from the market. Of particular concern is the rare or ?idiosyncratic? DILI that may only become evident late in clinical development or after a drug is marketed. This concern is driving increases in the size and duration of many Phase 3 clinical trials and creating a bottleneck in the delivery of new medicines to patients.

DILI-sim is a predictive model based on physiological processes involved in the initiation and propagation of DILI. The Hamner-UNC Institute for Drug Safety Sciences and Entelos, Inc., have partnered to develop DILI-sim in the PhysioLab® biosimulation platform. The Food and Drug Administration (FDA) has endorsed the effort and plans to use DILI-sim as a tool for recognizing drugs with a high propensity for DILI. Some of the key aspects of DILI to be addressed with DILI-sim include species differences in susceptibility to DILI, pharmacologic class versus structure-specific DILI, distinction between benign liver chemistry changes and those that indicate serious DILI potential, genetic and environmental factors (e.g. diet and disease) that may increase DILI susceptibility, and biomarkers that may identify subsets of patients who are susceptible to DILI. Acetaminophen (APAP), isoniazid, and valproic acid were chosen as exemplar drugs around which to construct a multiple pathways model. Multiple species, including but not limited to mice, rats, and humans are included. In vivo and in vitro data from rodents and humans at therapeutic doses and doses known to cause DILI were used for parameter optimization and model validation. Following development, DILI-sim will be tailored to predict idiosyncratic DILI during the clinical trial phase and for novel compounds during the early development phase. The characterization of APAP induced DILI was the primary task during phase I of development.

A physiologically based pharmacokinetic (PBPK) model was developed to describe APAP absorption, distribution, metabolism, and excretion. The metabolic pathways of glucuronidation, sulfation, and CYP oxidation to the reactive metabolite N-acetyl-p-benzo-quinone imine (NAPQI) were included. Multiple dosing routes, including intravenous, intraperitoneal, and oral were also incorporated. The PBPK model was linked to a glutathione depletion model to account for oxidative stress from NAPQI. The cell life cycle was modeled, including the ability of cells to recover from stress or suffer apoptosis or necrosis. Cell regeneration rates have been proposed as key factors in liver recovery from DILI versus acute liver failure, and were therefore also considered. Biomarkers such as alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (AP; ALP), gamma-glutamyl transferase (GGT), albumin, prothrombin time, lactate dehydrogenase (LDH), bilirubin, (direct/conjugated and total), as well as several novel markers, are or will be represented as model outputs.

The model accurately reproduced APAP pharmacokinetic measures for rats, mice, and humans. Several key findings concerning APAP induced DILI have already been identified. The depletion of 3'-phosphoadenosine-5'-phosphosulfate (PAPS) is crucial to accurately reproducing the sulfation pathway in rats. Metabolism differences across species explain much of the variation in species specific susceptibility to APAP induced DILI. In-depth comparisons were also made between rats and mice with regard to their ability to re-synthesize GSH after GSH depletion. The use of N-acetyl-cysteine (NAC) as a treatment for APAP overdose was incorporated into the model, and treatment times were analyzed and compared to standard clinical practices to predict optimal NAC use. Finally, the model was used to successfully predict the species differences in hepatotoxicity of methapyrilene using in vitro to in vivo extrapolation. The absorption, distribution, and metabolism of methapyrilene were estimated from the physical characteristics of the compound and in vitro metabolism. The model correctly predicted that rats would be less sensitive to acetaminophen than mice or humans, but most sensitive to methapyrilene. A parametric search for in silico rats, mice, and humans susceptible to methapyrilene DILI predicted that methapyrilene would cause very little hepatotoxicity in humans. This finding agrees with clinical reports.