Transforming the discovery of targeted protein degraders: the translational power of predictive PK/PD modeling
Robin Thomas Ulrich Haid (1,2), Andreas Reichel (1)
(1) Bayer AG, Germany, (2) ETH Zurich, Switzerland
Introduction: Targeted protein degraders (TPDs) are attracting considerable interest with the promise to address disease-related proteins not druggable with inhibitors due to lack of an amenable active site [1, 2]. Instead of altering the activity of their targets, degraders employ the cell’s own ubiquitin proteasome system to break them down completely. This mechanism of action (MOA) requires the drug to form a ternary complex with its target protein and an E3 ligase enzyme, which marks the target for removal [3]. The ensuing degradation has been observed to persist beyond the time frame of detectable TPD levels in vivo suggesting protein resynthesis is a slow process [4].
Despite their novel MOA, degrader PK/PD is still approached with a mindset deeply rooted in inhibitor drugs, which impedes the proper interpretation of experimental data [5]. Researchers have thus resorted to serendipity and testing unsustainably large numbers of compounds directly in vivo [2]. This trial-and-error based avenue is inefficient, time-consuming, and expensive, urging the clear need for predictive PK/PD modeling specifically tailored for TPDs to improve decision making.
Meanwhile, the modeling and simulation community has been fascinated with the unique MOA of TPDs, spawning ever more sophisticated and complex mathematical models [6–10]. While there are also a few publications pursuing a more fit-for-purpose approach, none of them feature practical use cases for how to support drug discovery [11–13].
Our comprehensive modeling framework, in contrast, emerged from the close collaboration with interdisciplinary project teams and addresses the questions raised there while relying exclusively on readily available data.
Objectives:
- Establish and validate a quantitative systems pharmacology (QSP) model for TPDs to better understand their MOA.
- Leverage these insights to identify potential use cases for model-informed drug discovery and development in the preclinical setting.
- Apply the developed modeling framework to in-house TPD projects to assess its practical utility and impact on decision making.
Methods: The experience with in-house projects showed that different models were required to address the questions raised by different team members. The focus of medicinal chemists, for example, was mainly on how to optimize the TPD’s binding affinities, while DMPK scientists wanted to identify the PK/PD driver and pharmacologists were primarily interested in statements regarding downstream pharmacodynamic readouts. Thus, we propose a set of three models.
1) Our QSP framework builds on earlier models describing ternary complex formation in biochemical assays [14]. It assumes rapid equilibrium for drug binding and treats the monoubiquitination reaction catalyzed by the E3 ligase as rate limiting for protein degradation. Furthermore, the model only allows for saturation of the cell’s degradation machinery by the drug but not by the target protein. Its parameters can be classified as either compound-specific (three binding affinities) or system-related (expression levels of E3 ligase as well as target protein baseline half-life and rate of induced target degradation by the UPS) with all of them being observable in orthogonal assays.
2) Model lumping affords a separate set of four intuitive parameters (baseline protein half-life as well as degradation potency, extent of maximal degradation and concentration of maximal degradation) to describe protein degradation over time, both in vitro as well as in vivo. When simulating in vivo studies, the drug’s PK is described with a suitable compartmental model, while the PD parameters are informed from in vitro assays.
3) Going beyond degradation, the model directly links protein levels to downstream pharmacodynamic effects, while also accounting for potential target inhibition by the degrader drug.
Results: To validate the focused QSP model, protein degradation was accurately predicted for a set of nine different compounds in three different cell lines [15]. This confirmed that the key compound-specific and system-related parameters had been correctly identified and that the MOA is described well.
The lumped mechanistic model was then used to predict in vivo protein degradation from in vitro data for twelve compounds covering three different targets. Drug PK was informed from prior exposure studies, while protein degradation was studied in cell culture. For predictions to be accurate, degradation potency observed in vitro had to be corrected for incubation time using mechanistic modeling. Moreover, non-specific binding in plasma as well as in culture media had to be considered.
Twice, the predictions for an oncology project were slightly off, which led to two important insights. First, it was observed that plasma concentrations were not representative of tumor exposure for one of the compounds. Second, the rate of target protein resynthesis in tumor cells seemed to increase under long-term treatment, presumably due to the selection pressure exerted by TPD treatment.
Our modeling activities also led to useful insights about the TPD mechanism of action itself, such as identifying AUC as the main PK/PD driver. However, infrequent dosing regimens (e.g., monthly) were found to require sustained release formulations, as protein resynthesis takes place on the time frame of days rather than months [16].
Modeling activities were then also extended to a downstream readout of efficacy, again using in vitro data as input. In doing so, fundamental differences between conventional inhibitors and degraders were identified, demonstrating that the simple concept of functional potency does not apply to the latter.
Conclusion: Our modeling demonstrates that the focused QSP framework can be used to 1) guide medicinal chemistry during compound optimization by identifying target values for the individual degrader’s binding affinities and 2) translate target degradation across different cell types and species by leveraging data on baseline protein half-life and E3 ligase expression levels. The lumped mechanistic model, in turn, allows to identify suitable compounds for efficacy studies and informs experimental design with regards to the selection of 1) dose schedules, 2) measurement time points and 3) number of animals.
Predictive modeling thus supports project teams in reducing the amount and size of preclinical PK/PD studies in order to save time, resources, and animals. Moreover, by making drug discovery more rational it allows to flag potential challenges early on and can also be employed to derive risk mitigation strategies. Going forward, this approach also lays out a straightforward path for clinical translation, where it is expected to result in more accurate human dose predictions. Thus, we believe that the framework presented here has the potential to profoundly transform degrader drug discovery.
References:
[1] Teng, M. & Gray, N.S. The rise of degrader drugs. Cell. Chem. Biol. 30, 864–878 (2023).
[2] Kong, N.R. & Jones, L.H. Clinical translation of targeted protein degraders. Clin. Pharmacol. Ther. 114, 558–568 (2023).
[3] Lai, A.C. & Crews, C.M. Induced protein degradation: an emerging drug discovery paradigm. Nat. Rev. Drug Discov. 16, 101–114 (2017).
[4] Mares, A. et al. Extended pharmacodynamic responses observed upon TPD-mediated degradation of RIPK2. Commun. Biol. 3, 140 (2020).
[5] Watt, G.F., Scott-Stevens, P. & Gaohua, L. Targeted protein degradation in vivo with proteolysis targeting chimeras: current status and future considerations. Drug. Discov. Today Technol. 31, 69–80 (2019).
[6] Bartlett, D.W. & Gilbert, A.M. A kinetic proofreading model for bispecific protein degraders. J. Pharmacokinet. Pharmacodyn. 48, 149–163 (2021).
[7] Guzzetti, S. & Morentin Gutierrez, P. An integrated modelling approach for targeted degradation: insights on optimization, data requirements and PKPD predictions from semi- or fully-mechanistic models and exact steady state solutions. J. Pharmacokinet. Pharmacodyn. 50, 327–349 (2023).
[8] Zhao, H. & Narjes, F. Kinetic modeling of PROTAC-induced protein degradation. ChemMedChem 18, e202300530 (2023).
[9] Park, D., Izaguirre, J., Coffey, R. & Xu, H. Modeling the effect of cooperativity in ternary complex formation and targeted protein degradation mediated by heterobifunctional degraders. ACS Bio Med Chem Au 3, 74–86 (2022).
[10] Chaudhry C. Mathematical model for covalent proteolysis targeting chimeras: thermodynamics and kinetics underlying catalytic efficiency. J. Med. Chem. 66, 6239–6250 (2023).
[11] O'Brien Laramy, M.N., Luthra, S., Brown, M.F. & Bartlett, D.W. Delivering on the promise of protein degraders. Nat. Rev. Drug Discov. 22, 410–427 (2023).
[12] Bartlett, D.W. & Gilbert, A.M. Translational PK–PD for targeted protein degradation. Chem. Soc. Rev. 51, 3477–3486 (2022).
[13] Wang, A.F. & Ayyar, V.S. Pharmacodynamic models of indirect effects and irreversible inactivation with turnover: applicability to mechanism-based modeling of gene silencing and targeted protein degradation. J. Pharm. Sci. 113, 191–201 (2024).
[14] Han, B. A suite of mathematical solutions to describe ternary complex formation and their application to targeted protein degradation by heterobifunctional ligands. J. Biol. Chem. 295, 15280–15291 (2020).
[15] Haid, R.T.U. & Reichel A. A mechanistic pharmacodynamic modeling framework for the assessment and optimization of proteolysis targeting chimeras (PROTACs). Pharmaceutics 15, 195 (2023).
[16] Harling, J.D., Scott-Stevens, P. & Gaohua, L. Developing pharmacokinetic/pharmacodynamic relationships with TPDs. In Protein Degradation with New Chemical Modalities: Successful Strategies in Drug Discovery and Chemical Biology (eds. Weinmann, H. & Crews, C.) 75–93 (Royal Society of Chemistry, London, UK, 2021).