Comparison of a typical PK/PD model versus a mechanistic QSP model to predict the Phase II of a PSCK9 inhibitor, using MonolixSuite
Géraldine Cellière (1), Pauline Traynard (1)
(1) Simulations Plus, Lixoft division, Antony, France
Objectives: Compared to PK/PD models, QSP models incorporate more mechanistic details and model entities which have not been measured experimentally. By putting more emphasis on the biological relevance, QSP models are believed to be more capable of extrapolating from preclinical to clinical or from healthy volunteers to patients. To investigate the benefit of QSP models, we have compared the predictions for phase II obtained from a typical PK/PD, a mechanistic PK/PD and a QSP model fitted on the phase I data of a PSCK9 inhibitor for cholesterol lowering.
Methods: Phase I data for the anti-PCSK9 mAb has been presented in [1]. Here we use the individual profiles for total PSCK9, total drug and cholesterol to develop three models of different complexities. For all three models, the PK is described by a TMDD model that captures the drug and PSCK9 target concentrations. The PD is described by:
- Typical PK/PD model (A): Cholesterol is described by a turnover model in which degradation is inhibited by free PSCK9.
- Mechanistic PK/PD (B): LDLr, a receptor which internalizes cholesterol into hepatocytes is added. Cholesterol degradation is mediated by LDLr and LDLr degradation depends on PSCK9.
- QSP model (C): presented in [1], the model in addition includes hepatic cholesterol and its feedback on PSCK9 and LDLr via SREBP2.
The three models were fitted to the phase I data with Monolix and literature information on parameter values was used when necessary. The three models were then used to simulate with Simulx several phase II designs. As individuals in phase II are already statin-treated (while phase I patients were untreated), the parameter values were modified to capture PSCK9 and cholesterol levels in these patients. Patients with LDLr mutations leading to familial hypercholesterolemia were also considered. The model’s predictions were then compared to the actual phase II data [2].
Results: Calibration of the parameters of model A could be done using the phase I data only, while model B and C required respectively a little and extensive literature search to fix the parameters not directly related to the measured species. Parameter estimation run time for each model was: A 1 min, B 5 min and C 60 min.
When simulating phase II statin-treated patients receiving the anti-PCSK9 mAb with several dosing regimens, all 3 models gave very similar predictions which were in accordance with the actual phase II cholesterol data. The simulation of patients with LDLr mutations was only possible with model B and C, as model A didn’t include LDLr in the model. These simulations showed that these patients would only have a minor benefit.
Conclusions: In this example, all three models allowed to correctly predict phase II efficacy data, with the typical PK/PD model (A) requiring the least development efforts. However, more specific questions related to familial hypercholesterolemia could only be investigated with more mechanistic models (B and C) which incorporate intermediate chemical species between the drug target and PD effect. Model development and simulation of typical PK/PD and QSP models was very efficient thanks to the built-in tools of the MonolixSuite.
References:
[1] Gadkar et al. (2014), CPT: PSP 3(11).
[2] Baruch et al. (2017), Am Jour Cardiology 119(10)
