Unravelling the complex inhibition network of voriconazole and its metabolites using a middle-out PBPK approach
Ayatallah Saleh (1,2), Josefine Schulz (1), Jan Schlender (3), Linda B.S. Aulin (1), Franziska Kluwe (1,2), Gerd Mikus (1,4), Wilhelm Huisinga (5), Charlotte Kloft (1)* and Robin Michelet (1)*
(1) Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universität Berlin, Germany, (2) Graduate Research Training program PharMetrX, Germany, (3) Novartis Institute of Biomedical Research, Basel, Switzerland, (4) Department of Clinical Pharmacology and Pharmacoepidemiology University Hospital Heidelberg, Heidelberg, Germany, (5) Institute of Mathematics, University of Potsdam, Germany *Shared senior authorship
Introduction: The broad-spectrum antifungal voriconazole (VRC) exhibits large inter- and intraindividual PK and PD variability, nonlinear PK and complex metabolism. The large PKPD variability leads to a high risk of therapeutic failure or adverse events (AEs) [1]. VRC N-oxide (NO), the main metabolite in plasma, has been linked to some of these AEs. In vitro analyses have shown that VRC, NO and hydroxyvoriconazole (OH-VRC) reversibly and time-independently inhibit all three relevant CYP enzymes responsible for VRC metabolism [2]. Although multiple VRC PBPK models have been developed [3,4], the contribution of each metabolic pathway to VRC clearance (CL) and the impact of metabolite-mediated inhibition and autoinhibition remain unclear.
This work aimed to develop a parent-metabolites PBPK model to (i) elucidate the complex PK of VRC and (ii) quantify its metabolite-mediated inhibition.
Methods: A coupled parent-metabolites PBPK model for VRC, NO and OH-VRC was developed by expanding a VRC PBPK model using PK-Sim® and MoBi®,v. 11 [5]. NO was assumed to be formed by CYP2C19, CYP3A4 and CYP2C9 [2], and OH-VRC via CYP3A4 only [6]. Both metabolites were excreted via glomerular filtration, and NO was also eliminated via hepatic CL [7]. Inhibition equations [8] were implemented to describe the complex interaction network of VRC autoinhibition and metabolite-mediated inhibition on each CYP enzyme.
A middle-out approach was used to develop the model, utilising data from 3 clinical studies for subtherapeutic (50 mg) and therapeutic (400 mg) single 2 h i.v. infusion, and mean VRC plasma PK profiles from 2 clinical studies (digitised from [9]) for i.v. multiple-dosing regimens (MD). Model parameters not substantiated by literature or in vitro analyses were estimated using a Monte Carlo approach. Influential parameters identified from local sensitivity analyses were inferred stepwise using clinical data. First, the CYP2C19 poor metaboliser (PM) data was used to estimate the CYP3A4-based metabolism of VRC. Then the competitive inhibition constant (Ki) of CYP2C19 by NO was inferred in subsequent steps based on data of single and multiple i.v. dose in CYP2C19 intermediate (IM), normal (NM and rapid (RM) metabolisers. Model evaluation was performed by comparing predicted to observed PK profiles and maximum plasma concentrations (Cmax) and by calculating the average fold error (AFE); AFE closer to 1 indicates accurate prediction.
Results: The a priori model accurately predicted the subtherapeutic single dose (AFE50mg/2h=0.98) but not the therapeutic dosing regimens. It overpredicted VRC PK profiles for CYP2C19 PM (AFEPM=1.06) and underpredicted them for CYP2C19 IM (AFEIM=0.96), NM (AFENM=0.81), and RM (AFERM=0.95). Predictions of OH-VRC PK profiles were adequate (range AFEOH-VRC=0.95-1.48), and NO PK profiles were underpredicted (AFENO=0.03-0.67). The model accurately predicted the first VRC Cmax of MD regimens but underpredicted PK profiles during maintenance dosing (AFEMD=0.67).
The CYP3A4 formation rates for NO and OH-VRC were estimated to be 0.09 1/min and 0.06 1/min, which improved the predictive performance for VRC (AFEPM=0.99) and OH-VRC (AFEPM=0.86), but not for NO (AFEPM=0.15). The Ki of CYP2C19 by NO was estimated to be 0.54 µM, leading to improved predictions of VRC (AFEIM=0.98, AFENM=0.90, AFERM=1.10) and OH-VRC (AFEOH-VRC=0.63-0.97). The persistent underprediction of NO (AFENO=0.38-0.58) led to the hypothesis of NO CL autoinhibition, which was tested using an empirical inhibition equation on NO’s total hepatic CL, driven by its hepatic intracellular concentration. The half-maximal inhibitory concentration (IC50) was estimated to be 0.85 µM, which resulted in good prediction of VRC, NO and OH-VRC across all dose levels (VRCSD: AFEIM=1.00, AFENM=0.94, AFERM=1.16; AFENO=0.51-0.77; AFEOH-VRC=0.72-1.11; VRCMD: AFEMD=1.09).
Conclusion: The reversible autoinhibition of VRC and the incorporation of metabolite-mediated inhibition successfully described the clinical data, supporting the hypothesis that VRC’s primary metabolites play a crucial role in the overall inhibitory effect of VRC on its own metabolism. This model can be used to design experiments to test these hypotheses, allowing further model qualification and subsequently treatment individualisation to reduce therapeutic failure and AEs.
References:
[1] J. Schulz et al. Drug Metab. Rev. 51: 247–265 (2019).
[2] J. Schulz et al. Pharmaceutics 14: 477 (2022).
[3] X. Li et al. Clin. Pharmacokinet. 59: 781–808 (2020).
[4] J. Dong et al. Pharm. Res. 39: 1921–1933 (2022).
[5] A. Saleh et al. 30th Population Approach Group Europe (PAGE). (2022).
[6] N. Murayama et al. Biochem. Pharmacol. 73: 2020–6 (2007).
[7] S. Li et al. Front. Pharmacol. 12: 730826 (2021).
[8] I.H. Segal. Wiley-Interscience. (1993).
[9] L. Purkins et al. Br. J. Clin. Pharmacol. 56 Suppl 1: 2–9 (2003).