Application of a Bayesian population approach to physiologically-based modelling and simulation of mavoglurant pharmacokinetics
Thierry Wendling (1, 2), Swati Dumitras (2), Ralph Woessner (2), Etienne Pigeolet (3), Kayode Ogungbenro (1) and Leon Aarons (1)
(1) Manchester Pharmacy School, The University of Manchester, Manchester, United-Kingdom, (2) Drug Metabolism and Pharmacokinetics, Novartis Institutes for Biomedical Research, Basel, Switzerland, (3) Pharmacometrics, Novartis Pharma AG, Basel, Switzerland
Objectives: Mavoglurant (MVG) is an antagonist at the metabotropic glutamate receptor 5 currently under clinical development at Novartis Pharma AG for the treatment of CNS diseases. The aim of this work was to develop and optimise a population physiologically-based pharmacokinetic (PBPK) model for MVG to predict the impact of drug-drug interaction and age on its PK.
Method: A whole-body PBPK model for drug disposition was first developed and optimised with data from a Phase-I study of intravenously administered MVG using a Bayesian approach. We developed a three-stage hierarchical model to describe both uncertainty and inter-individual variability (IIV) in the drug-specific parameters. Prior information on the system-specific parameters was extracted from the physiology literature. For drug-specific parameters, prior distributions were constructed based on the results of in vitro and animal experiments. A sensitivity analysis was performed prior to model fitting to identify the parameters that could be updated just by plasma data. Parameters’ posterior distributions were approximated by random draws using MCMC simulations in NONMEM. Three chains of 10^6 iterations were computed. Convergence to the equilibrium distribution was monitored using the potential scale reduction statistic . The optimised model was then used together with a mechanistic absorption model to predict MVG PK when orally co-administered with ketoconazole in adults or administered alone in children. The predictive performance of the model was evaluated using data from three other clinical studies.
Results: The population PBPK model allowed good description of MVG plasma PK data following IV administration in healthy adults. Prediction of the MVG-ketoconazole interaction was consistent with results of an in-house non-compartmental analysis of the clinical data (3-fold increase in systemic exposure). Finally, scaling of the PBPK model allowed reasonable extrapolation of MVG PK from adults to 3 to 11 year-old children.
Conclusions: Population PBPK modelling and simulation for MVG provided further insight into its PK, including the source and magnitude of IIV. The Bayesian approach allowed uncertainty in some of the drug-specific parameters to be reduced. The model can be used to predict plasma and brain (target site) PK profiles following oral administration of various immediate-release formulations of MVG alone or when co-administered with a perpetrator, in adults as well as in children.
 Gelman A, Rubin DB. Inference from Iterative Simulation Using Multiple Sequences. Statistical Science 1992; 7: 457-72.