Claude Magnard (1), Pauline Traynard (1), Monika Twarogowska (1), Géraldine Ayral (1)
(1) Lixoft, Antony, France
Objectives: The design of first-in-human clinical trials relies on the extrapolation of preclinical animal data to human predictions. The ability to accurately predict the human pharmacokinetics (PK) of a drug candidate from the available preclinical and literature information is therefore a key challenge in drug discovery and development. For small molecules, simple allometric scaling of clearance and volume is often sufficient. However, for many biologics, their nonlinear PK imposes the use of more advanced methods to predict the PK in human. In this poster, we compare several model-based approaches for the first-in-human dose selection for the fully human IgG2 monoclonal antibody PF-03446962 targeting ALK1, using cynomolgus monkey data and literature information.
Methods: Drug concentration profiles after single and multiple doses of PF-03446962 for 14 monkeys are extracted from Luu et al [1]. Three different PK models are fitted on the data using a population approach in Monolix: (1) a linear 2-compartment model, (2) a simple target-mediated disposition (TMDD) model with linear and Michaelis-Menten elimination and (3) a more complex TMDD model, where the complex internalization rate and the binding affinity are fixed to literature values. All models are available in the Monolix TMDD library of models.
Each model is then used to simulate the human PK for a large range of doses using Simulx. Volumes and linear clearance parameters are scaled using classical allometric scaling. For model 2 (simple TMDD), Km and Vm are assumed identical in monkey and human. For model 3, the parameter values are replaced by the literature values for humans. The model predictions are finally compared with actual phase I human data.
Results: The linear model is not able to capture the monkey PK profiles, but both the simple and more complex TMDD models provide satisfactory diagnostic plots and good confidence on all estimated parameters. According to the parsimony principle, one would favor the simpler TMDD model with linear and non-linear clearance. However, the estimated Km and Vm parameters have no clear biological meaning and it is therefore difficult to decipher how they scale to humans. On the opposite, the more complex TMDD model has parameters with clear physiological interpretation, which in addition have been experimentally measured both in monkey and humans. This greatly facilitates the choice of parameter values for the prediction of human PK. When compared to the actual PK profiles in human from the phase I study, the more complex TMDD model shows the most accurate prediction, within 1 to 2-fold of observations.
Interestingly, because the more complex TMDD model includes a variable representing the free receptor, it is possible to use this model to simulate the free target (i.e receptor) relative to baseline and use this as surrogate of the biological effect. The simulation of the expected biological effect for a large range of doses allows to choose the first-in-human dose as the one corresponding to the minimal anticipated biological effect level (MABEL). For PF-03446962, the MABEL is predicted to be 0.005 mg/kg. This prediction would not be possible with the simpler TMDD model which only models the drug concentration.
Conclusions: This example illustrates how the incorporation of mechanistic information from the literature into a model-based approach can help to obtain more accurate first-in-human predictions and thus inform the choice of the first-in-human doses, in particular for biologics. While this workflow may at first seem complex compared to usual approaches, we show that its implementation using the MonolixSuite is straightforward.
References:
[1] Luu, K. T., Bergqvist, S., Chen, E., Hu-Lowe, D., & Kraynov, E. (2012). A Model-Based Approach to Predicting the Human Pharmacokinetics of a Monoclonal Antibody Exhibiting Target-Mediated Drug Disposition. Journal of Pharmacology and Experimental Therapeutics, 341(3), 702–708.
Reference: PAGE 29 (2021) Abstr 9613 [www.page-meeting.org/?abstract=9613]
Poster: Methodology - Model Evaluation