Modeling PK and PK/PD in of anti-VEGF agents in a preclinical rabbit model of VEGF-induced retinal hyperpermeability
Carsten Terjung
Bayer AG, Pharmaceuticals, Preclinical Modeling & Simulation, Germany
Objectives: In animal experiments, the study design should always aim at striking a balance between providing meaningful results and limiting the number of animals used. For VEGF-induced retinal leakage in rabbits, this is particularly challenging, because the desired anti-VEGF efficacy changes over time, and only limited time points can be assessed in each individual, as the compartment of interest – the vitreous humor – is only accessible after animal sacrifice. Here, we present a PK/PD modeling approach to design and evaluate a preclinical study of aflibercept, brolucizumab and ranibizumab in rabbits [1].
Methods: The pharmacokinetics of VEGF and the anti-VEGF agents after intravitreal injection were calculated using a minimal physiology-based model, based on the assumed predominant elimination of biologics by the anterior route, via diffusion through the vitreous and elimination from the anterior chamber through turnover of aqueous humor. Ocular elimination half-life estimates were calculated considering the geometry of the rabbit eye and the molecular weight-based hydrodynamic radii of the VEGF protein, the respective anti-VEGF agents, and the VEGF/anti-VEGF complexes using the modeling framework established by Hutton-Smith et al. [2]. Initial estimates for the binding affinity of the anti-VEGF agents were taken from literature references [3]. The pharmacological effect was predicted based on a semi-mechanistic model of VEGF-induced hyperpermeability in rabbits [4].
The outcomes of the in-silico modeling were used to design the in vivo rabbit VEGF-A165–induced retinal vascular permeability study using aflibercept, brolucizumab, and ranibizumab. The concentration–time profile data for each of the anti-VEGF agents were evaluated using noncompartmental analysis (NCA). A post-hoc comparison of the predicted vitreous concentration–time profiles for anti-VEGF agents with actual total concentrations measured by a ligand-binding assay was performed. For the PK/PD evaluation, leakage scores, which were assessed by the extent of fluorescein visualized in the optical coherence tomography fluorescein angiography imaging, were categorized into two groups depending on the leakage score; these categories were assigned a probability of 0 or 1 for each individual leakage assessment. The model-predicted area under the concentration–time curve of free exogenous VEGF in the vitreous between VEGF injection and leakage observation was used as the basis for fitting a logistical model to describe the resulting probability of moderate to severe retinal leakage in the treated animals.
Results: The post-hoc comparison revealed that the half-life predictions based on the in-silico model framework closely matched the observed in vivo PK results. A logistic model was employed to describe the relationship between the predicted areas under the concentration–time curve of free, that is, not blocked, exogenous injected VEGF and the probability of observing moderate to severe leakage grading in rabbits for the three anti-VEGF agents. Overall, the model performed with an accuracy of 85%. Based on the model, the necessary fraction of injected VEGF that needs to remain unblocked to observe moderate to severe leakage grades may be calculated for a certain probability to observe retinal leakage. The estimated probability for moderate to severe leakage increases significantly once more than approximately 40% of the injected VEGF is not captured by the respective anti-VEGF agent. This finding is consistent with our initial assessment of the data reported by Edelman et al. [4] for this rabbit leakage model.
Conclusions: A modeling-based approach was employed to accompany a preclinical in vivo study from beginning to end. The intravitreal PK profiles of aflibercept, brolicizumab and ranibizumab after injection into the vitreous humor of rabbits were simulated considering diffusion characteristic of the different agents and the geometry of the rabbit eye. A PK/PD hypothesis based on literature data was used to select relevant time-points for an intravitreal VEGF challenge in rabbits. The collected intravitreal exposure data of the three anti-VEGF agents confirmed the initial PK simulations and were used to refine the PK/PD model. This model can be used for the planning of in vivo animal experiments for future assessment of new anti-VEGF modalities or new dosing schemes.
References:
[1] Schubert, W., Terjung, C., Rafique, A., Romano, C., Ellinger, P., & Rittenhouse, K. D. (2022). Evaluation of Molecular Properties versus In Vivo Performance of Aflibercept, Brolucizumab, and Ranibizumab in a Retinal Vascular Hyperpermeability Model. Translational Vision Science & Technology, 11(10), 36-36.
[2] Hutton-Smith LA, Gaffney EA, Byrne HM, Maini PK, Schwab D, Mazer NA. A mechanistic model of the intravitreal pharmacokinetics of large molecules and the pharmacodynamic suppression of ocular vascular endothelial growth factor levels by ranibizumab in patients with neovascular age-related macular degeneration. Mol Pharm. 2016; 13: 2941–2950
[3] Stewart MW. The study of intravitreal drug pharmacokinetics: does it matter? and if so, how? Expert Opin Drug Metab Toxicol. 2018; 14: 5–7.
[4] Edelman JL, Lutz D, Castro MR. Corticosteroids inhibit VEGF-induced vascular leakage in a rabbit model of blood-retinal and blood-aqueous barrier breakdown. Exp Eye Res. 2005; 80: 249–258.