Yuchen Guo, Tingjie Guo, Laura B. Zwep, J. G. Coen van Hasselt
1Division of Systems Pharmacology and Pharmacy, Leiden Academic Center for Drug Research, Leiden University
Introduction In physiologically based pharmacokinetic (PBPK) modelling, the expression levels of drug metabolizing enzymes and transporters (DMET) are essential parameters in predicting the drugs’ pharmacokinetics [1]. DMET expression values are often correlated, and a drug’s metabolism typically involves multiple enzymes and transporters. Properly accounting the dependency structure between the DMETs is crucial yet often insufficiently considered in current practice, which may result in inadequate capture of interindividual variability. Copulas, a unique type of joint distribution models, are well-suited for modelling such correlations but generally require large datasets for a robust estimation [2,3]. DMET expression data are typically obtained from biopsies or post-mortem samples, limited by small sample sizes, which is a challenge for modelling the correlation structures of DMET expression. The aim of this study is (1) to develop and evaluate joint distribution models for DMETs, (2) to demonstrate the incorporation of DMETs data in a typical PBPK workflow, using selegiline as a case study. Methods Model development: We used a publicly available DMET database Genotype-Tissue Expression (GTEx) and extracted gene expression data of CYP, UGT, ABC and SLC families [4]. We modeled the joint distribution of these data using three methods: 1) direct modelling assuming a multivariate normal (MVN) distribution, 2) modelling using transformed MVN (transMVN), where the marginal distribution was first transformed to normal scales, and 3) copula modelling [2]. Candidate copulas, including full and reduced copulas were estimated, with the lowest BIC model selected for the comparison with MVN and transMVN. To assess the performance of the developed models, virtual DMET datasets simulated from each model were compared to the observed DMET data using visual inspection and quantitative metrics. Selegiline case study: To illustrate DMET integration in PBPK models, selegiline, a Parkinson’s treatment metabolized by CYP2B6 and CYP2C19 enzymes, was used for a case study [5]. DMET gene expression data served as the surrogate for hepatic protein abundance, which, along with liver weight, was used to scale in vitro CYP enzyme activity to in vivo hepatic clearance (CLhep) using a well-stirred model. The area under the concentration-time curve (AUC) and its coefficient of variation (CV) were calculated following a 10 mg intravenous dose to assess the inter-individual variability in selegiline exposure. To provide a mechanistic understanding of the metabolism of selegiline, the contribution for each metabolic pathway and enzyme was quantified based on the percentage of their respective hepatic clearance to total hepatic clearance. To assess the influence of different joint models on PBPK predictions, CLhep, AUC and CV calculated from observed DMET data were used as reference. The errors for these predictions were calculated for virtual DMET data generated using MVN, transMVN and copula models. Results Model development: The DMETs datasets for joint distribution model development consist of 13 CYP enzymes, 6 UGT enzymes, 4 ABC transporters and 17 SLC transporters from 226 adult individuals. MVN, transMVN and copulas were successfully developed for all DMETs. Compared to the MVN model, the transMVN and copula models exhibited comparable and better description on both marginal distributions and the dependency structure across DMET families. Selegiline case study: Based on the observed DMET expression data, the median CLhep and AUC of selegiline in a 70 kg adult were predicted to be 90.6 L/h and 110.4 ng*h/mL, respectively. Variability in DMET gene expression led to substantial inter-individual variability in drug exposure (CV 154%). In terms of metabolic contributions, CYP2B6 was responsible for approximately 89% of selegiline’s hepatic clearance, while the desmethylselegiline pathway accounted for 47% of the metabolism. The transMVN and copula models exhibited similarly low errors in predicting CLhep and AUC compared to MVN, with the copula model achieving the lowest error for the CV of AUC. Conclusions The novel transMVN method and copulas could well capture DMET expression dependencies. Virtual DMET data generated from joint distribution models allow for prospective prediction of clearance and exposure in a PBPK workflow. This approach can be relevant not only for genomics- and proteomics-informed PBPK analysis but also for broader applications in QSP modelling.
1. Ladumor, Mayur K., et al. Scientific reports 9.1 (2019): 9709. 2. Zwep, Laura B., et al. Clinical Pharmacology & Therapeutics 115.4 (2024): 795-804. 3. Guo, Yuchen, et al. of Pharmacokinetics and Pharmacodynamics (2024): 1-12. 4. Lonsdale, John, et al. Nature genetics 45.6 (2013): 580-585. 5. Hidestrand, Mats, et al. Drug metabolism and disposition 29.11 (2001): 1480-1484.
Reference: PAGE 33 (2025) Abstr 11460 [www.page-meeting.org/?abstract=11460]
Poster: Methodology - New Modelling Approaches