Bayesian Hierarchical Modeling of Receptor Occupancy in PET Trials
Renard, D.(1b), J. Witcher (1a), Y. Nie (2), F. Vandenhende (1b), A. Kumar (3), J. Miller (1a), J. Tauscher (1a), Y. Zhou (3), D. Wong (3)
(1) Eli Lilly and Co., a. USA, b. Belgium; (2) Limburgs University, Belgium; (3) PET center, John Hopkins University, USA
Objectives: Positron emission tomography (PET) is a well-established imaging technique to measure metabolic processes in living subjects. With specific radioligands, it can be used to determine the receptor occupancy (RO) of a drug, that is, the proportion of drug targets, in most cases neurotransmitter receptors or transporter proteins that are occupied by the drug. When applied early in drug development, this imaging technique can generate valuable information to support go/no go decisions and dosage selection to accelerate development of a drug. Quantification and dose-response analysis of RO is a fairly complex task that is often performed as a stepwise procedure, each step carrying over some assumptions. Uncertainty at each step is rarely carried forward in the analysis and this may result in RO values that are biased and/or overly precise. The objective of this work is to overcome those limitations by integrating all the steps in a single analysis.
Method: We build a hierarchical model that combines three components: a model quantifying RO based on time activity curves of the radiotracer in each brain region of interest, a model characterizing pharmacokinetics (PK) of the drug, and a model relating PK and RO. We use a Bayesian Markov chain Monte Carlo (MCMC) approach to fit the model.
Results: We apply the hierarchical modeling approach to a typical RO trial in which 12 subjects underwent a set of 4 scanning sessions. The first scan was performed at baseline, in drug-free conditions, and the other scans were conducted on three occasions following administration of a single dose of the drug. Four doses of the drug were administered in groups of three subjects. Blood samples were obtained in each scanning session to measure drug concentration in plasma. An informative prior was used to build the PK model, using data from trials that were previously conducted with this compound. We illustrate model fit, as well as the predictive value of this tool, in the tested example.
Conclusions: We present a Bayesian hierarchical model to estimate receptor occupancy and characterize the PK-RO relationship in ligand displacement PET studies. As a result, we obtain a predictive model for RO to assist in the dosage selection process. Unlike standard approaches to RO estimation, precision of the individual RO estimates can be easily quantified within the Bayesian framework. In addition, regional and overall brain occupancy can be estimated from the model and predicted for other dosage regimens.