Mechanistic PKPD modelling of targeted protein degradation to improve compound selection in drug discovery and translation to the clinic
Peter Ashcroft, Allison Claas, Domenico Bullara, Javier Estrada Diez, Birgit Schoeberl, and Suzanne Gaudet
Novartis Institute of Biomedical Sciences; Basel Switzerland and Cambridge MA USA.
Introduction: Protein degradation via PROTACs (proteolysis targeting chimeras) or molecular glue degraders is an attractive new therapeutic approach. It provides a way to intervene on targets that were previously considered “undruggable”, or without a clear catalytic pocket. Additionally, for any target, degradation can provide a pharmacodynamic advantage over inhibition, particularly if the turnover of the target protein is slow, leading to the degradation of the target outlasting the presence of the drug. This in turn can lead to potential advantages with respect to dosing, side effects, or possible resistance. Examples of targets for degradation include BRD9 for AML [1] and IKZF2 (Helios) to disrupt regulatory T-cell activity [2].
Objectives: During a drug discovery effort for a degrader, we can seek to optimize many compound characteristics, such as the pharmacokinetic properties of compounds or the binding kinetics to the target and/or to the ligase that will tag it for degradation. But where are the sweet spots and limitations for different targets? And how can we better translate in vitro and in vivo observations of protein degradation to predict degradation in the clinical setting?
Methods: To answer these questions, we have developed multi-scale dynamical systems PKPD models of different modes of protein degradation. These mechanistic models are built around state variables for target concentration, drug concentration, and ligase concentration, and include kinetics for target protein turnover, as well as binary and ternary complex formation, ubiquitination, and degradation. Using judiciously chosen and justifiable assumptions, we arrived at a simplified model that best captures currently-observable data across multiple modes of protein degradation. These assumptions include constant total ligase concentration, rapid binding kinetics, and irreversible dissociation of ubiquitylated target. Further assumptions around pharmacokinetic timescales and stationary states can be imposed to further reduce the model and improve tractability.
Results: Under steady-state conditions, the target degradation can be described by just two non-dimensional parameters: one describing the biological system properties and another for compound-specific properties. This allows for tractable visualizations of the degrader space, which in turn leads to suggested paths for compound optimization. This mapping can also be used to predict clinical response based on what we know about human cell and protein biology. For example, a higher target turnover rate in human compared to preclinical models would lead to less protein degradation. Through the simplified model we can then readily assess if it is possible for this decreased pharmacodynamic effect to be compensated by an increased drug concentration.
Conclusion: This PKPD model and subsequent analyses provide contextualized solutions to ranking targeted protein degrading compounds and predicting outcomes in relevant clinical scenarios.
References:
[1] Weisberg, E., Chowdhury, B., Meng, C. et al. BRD9 degraders as chemosensitizers in acute leukemia and multiple myeloma. Blood Cancer J. 12, 110 (2022).
[2] Bonazzi, S., d’Hennezel, E., Beckwith, R.E.J. et al. Discovery and characterization of a selective IKZF2 glue degrader for cancer immunotherapy. Cell Chemical Biology, online 0.1.03.2023 (2023).