Senior Principal Scientist Pfizer Inc, United States
Disclosure(s):
Eric Greenwald, PhD: No relevant disclosure to display
Objective: Developing a drug requires balancing several factors that define potency, efficacy and pharmacokinetic (PK) properties. Identifying the magnitude and duration of target engagement early in drug development can help focus the chemistry design criteria and accelerate the advancement of therapeutics to patients. To assess this early in the drug development process an in vitro experimental method was developed and coupled with a mathematical model to elucidate which PK metric (Cmax, Cave, or threshold concentration) best correlates with efficacy. This initial effort was streamlined by the development of a MATLAB based app to assist in the model-based analysis of experimental data. Additionally, given the need for robust translation of the in vitro findings, an in silico analysis of the PK driver models was completed to validate their connection to clinical PK driver assessments.
Methods: Previously an in vitro washout assay was developed to discriminate between 3 general PK drivers of efficacy: Cave, Cmax and threshold concentrations. By exposing cells to a drug for varying durations and concentrations such that the AUC is consistent across all the samples, the relationship between the duration of exposure and an end-point readout, in this case cell proliferation, the PK driver for a drug mechanism can be determined. 3 different PK driver computational models were defined to analyze the data. With a standardized experimental design, the work was further expanded through the development of a MATLAB app to facilitate fitting these models to the data and assessing the goodness of fit to identify the most likely PK driver. In addition, to assess the compatibility of these computational models with clinical PK driver correlation analysis, in silico clinical trials were carried out where each of the 3 PK driver computational models were assumed to be the ground truth for anti-tumor efficacy and PK variability of a hypothetical compound was applied to these models. The simulated efficacy from each of these clinical trials was then correlated against Cave, Cmax and Cmin using a logistic regression.
Results: The analysis methods were successfully deployed in the MATLAB app and applied to several compounds with different PK drivers. Furthermore, the app has sufficient flexibility to address unique effects such as compound instability and incomplete washout, resulting in a robust identification of the correct PK driver. Finally, the in silico exploration of our computational PK driver models confirmed that each model was correctly aligned to correlation outcomes consistent with those commonly performed clinically. This analysis supports the use of the in vitro methodology coupled with the computational models to correctly discern clinically relevant PK drivers early in preclinical drug discovery.
