(M1530-11-64) Use of Quantitative Systems Toxicology (QST) to Identify Potential Intrinsic Mechanisms of Toxicity

Purpose: BAY1128688, a selective inhibitor of aldo-keto reductase family 1 member C3 (AKR1C3), was in clinical trials as a potential therapy to provide pain relief for women with endometriosis.  Although no liver toxicity was observed in rats and cynomolgus monkeys, clinical development was prematurely terminated due to post-treatment hepatotoxicity¹ following 30mg QD or 60mg BID for 12 weeks. This finding prompted use of DILIsym®, a QST model, to determine possible mechanisms contributing to the observed drug-induced liver injury (DILI)^2,3. Although bilirubin elevations >2x upper limit of normal (ULN) and alanine aminotransferase (ALT) elevations >2xULN were observed in some individuals while on treatment, severe ALT elevations >3xULN were mostly observed at the end-of-treatment visit (12 weeks after starting treatment) or after treatment cessation. The timing for hepatotoxicity presents an opportunity to explore mechanisms that may contribute to or lead to increased susceptibility for delayed hepatotoxic events in DILIsym.

Methods: Mechanistic in vitro data and a physiologically based pharmacokinetic (PBPK) model developed with clinical exposure data were utilized to predict the hepatotoxic liability of BAY1128688 in a simulated population (SimPops™) that represents inter-patient variability. A sensitivity analysis was performed to determine whether oxidative stress, mitochondrial toxicity, or bile acid transporter inhibition was the largest contributor to predicted ALT elevations. Following initial simulations and mechanistic sensitivity analysis, uncertainties in the BAY1128688 parameterization around mode of bile acid transporter inhibition and liver partition coefficient were explored to bring simulations into better alignment with the clinically reported bilirubin and ALT elevations. Specifically, the mode of bile salt export pump (BSEP) inhibition was modulated. After optimization with the highest clinical dose, 60mg BID for 12 weeks, all lower clinical doses were simulated and compared to clinical observations for ALT and bilirubin.

Results: Initial simulations used a quantitative structural activity relationship (QSAR) predicted liver partition coefficient. The default mode of bile acid transporter inhibition (mixed inhibition type with α=5) was applied. These initial simulations overpredicted the incidence of ALT and bilirubin elevations. Sensitivity analyses indicated that the effect of BAY1128688 on bile acid transport was the dominant toxicity mechanism leading to simulated ALT elevations, and BSEP inhibition was shown to be the most impactful for ALT elevations. For optimization, first the BAY1128688-induced alterations in bilirubin disposition and known effects of BAY1128688 on bilirubin transporters (e.g., UGT1A1) were used to optimize the liver partition coefficient and bring predicted bilirubin elevations into alignment with clinical data. Upon matching bilirubin elevations (Table 1), the mode and potency of BSEP inhibition were optimized to bring simulations into alignment with magnitude and delayed timing for ALT elevations. Optimization revealed mixed inhibition for BSEP with α=14 captured the frequency of ALT elevations while on BAY1128688 treatment (Table 1). Additionally, subsequent simulations at lower doses showed good alignment with observed magnitude of liver enzyme elevations while on treatment (Figure 1).

Conclusion: While DILIsym is not currently capable of directly predicting post-treatment ALT elevations as were observed with BAY1128688, simulations indicated dose-dependent intrinsic toxicity that is believed to be a necessary initial step in what was likely an adaptive immune response⁴. The approach taken herein can be used to inform predictions of drug-mediated intrinsic hepatotoxicity not evident in current preclinical testing, and that may be a prerequisite for delayed toxicity mediated by adaptive immune mechanisms.

References: 1. ClinicalTrials.gov. Nat’l Lib Med (US) 2021; ID:NCT03373422.
2. Woodhead. Pharm Res 2019; 36(3):48.
3. Battista. Toxicol Sci 2018; 166(1):123.
4. Mosedale. Clin Pharmacol Ther 2017; 101(4):469.

Acknowledgments: C.B., G.T.G., and L.K.M.S. are employees of Simulations Plus, Inc. E.G. W., A.R., and M.R. are employees of Bayer AG. P.B.W. has served as a consultant to Bayer AG and serves as chair of the Scientific Advisory Board for the DILI-sim Initiative that is the producer of the DILIsym® software. This study was funded by Bayer AG, and Bayer AG was responsible for the study design, data collection and analysis, and overall conduct of the study.

Table 1. Summary of on-treatment clinical observations and optimized simulation results for plasma total bilirubin (TBIL) and alanine aminotransferase (ALT) across 12-week dosing regimens.



Figure 1. Clinical observations (circles) and simulated values (triangles) for a) plasma total bilirubin fold change from an individual’s baseline, and b) maximum plasma alanine aminotransferase (ALT). The patient with highest recorded ALT is indicated by the green box on both plots.