Physiologically-Based Predictions of Monoclonal Antibody Pharmacokinetics: Insights from a Large-Scale Data Analysis
Salih Benamara1,3, Dr Erik Sjögren2, Dr Florence Gattacceca3, Dr Marylore Chenel2, Dr Antoine Deslandes1, Dr Laurent Nguyen1, Dr Donato Teutonico1
1Translational Medicine Unit, Quantitative Pharmacology, Sanofi, 2Pharmetheus AB, 3Computational Pharmacology and Clinical Oncology (COMPO) Team, Inria Sophia Antipolis-Méditerranée, Cancer Research Center of Marseille, Inserm UMR1068, CNRS UMR7258, Aix Marseille University UM105
Background Monoclonal antibodies (mAbs) have become a cornerstone of oncology and immunology therapies, traditionally administered intravenously over extended treatment periods, sometimes lasting several years (1). To enhance patient convenience and quality of life, there is increasing interest in less invasive, faster, and more flexible administration methods, such as subcutaneous (SC) delivery (2). Nearly half of the mAbs approved by the FDA in recent years include a SC administration option (3), underscoring the growing significance of this route. Whole-body physiologically-based pharmacokinetic (WB-PBPK) models are valuable tools in drug development, enabling pharmacokinetic (PK) predictions across species and populations (4). However, predicting mAb human PK and bioavailability after SC administration remains challenging due to the lack of standardized methods for mechanism-based absorption and elimination modeling. Among the biological processes incorporated into PBPK models, endosomal clearance and FcRn-mediated salvaging are the primary mechanisms governing mAb elimination (5). However, the absence of established reference methods to inform the FcRn dissociation constant (KdFcRn) reduces the possibility for high confidence PK predictions. Similarly, the translatability of data from current in vitro and pre-clinical models to clinical SC absorption is still poor (6). As a result, predicting human PK and bioavailability after SC administration remains challenging, especially in cases where intravenous (IV) PK data are unavailable. Objectives The overall objective of this study was to establish a model-based strategy for predicting mAbs PK in humans following SC administration with and without prior data on IV PK. This involved the following goals: 1.The construction of a reference database used for planned modeling investigations and evaluation. 2.Evaluate a recently developed physiologically-based SC absorption module implemented in the Open Systems Pharmacology (OSP) Suite to predict mAbs SC absorption and bioavailability, in the context of use for an IV to SC administration switch (8). 3.Establish an approach for naïve PBPK model-based predictions of clinical PK of mAbs after IV administration. 4.Establish an approach for naïve PBPK model-based predictions of clinical PK of mAbs after SC administration, in a context of use for First-in-Human trials (FIH) of a mAbs administrated subcutaneously without prior IV PK data. Methods PBPK modeling was conducted in PK-Sim and MoBi from the open-source platform of OSP Suite (7). Data processing, NCA analysis, and plotting were performed in R. An extension of the SAEM algorithm (9) available in the R-package Saemix (10), and coupled with the WB-PBPK model, was applied to investigate inter-mAbs variability for key processes in the SC absorption module. At each step, the predictive performance of the models was assessed by overlaying the predicted profiles with the observed data. Each model performance was considered acceptable if the predicted exposure PK parameters were within two-fold of the corresponding observed PK metrics. 1.On-line sources were searched to identify suitable clinical studies reporting PK profiles after IV and SC administration of mAbs. Reported data was digitized, collated and curated for the construction of a reference database. Drugs showing target-mediated drug disposition behavior were excluded. Each drug was modeled using the generic WB-PBPK model for large molecules in PK-Sim® and for those with available IV reference data a drug specific value of the KdFcRn was estimated through model fitting to observed data. 2.MAbs in the database with both IV and SC reference data (n=30) were included in the evaluation of a recently developed SC module. For each drug, a disposition model was established using one drug specific parameter (i.e., Molecular weight) by estimating the KdFcRn, and leveraging clinical concentration-time data after IV administration. The estimated KdFcRn was subsequently used to predict absorption and bioavailability applying the mechanistic SC physiologically based absorption model. Briefly, this model describes the local injection site by drug dispersion, dynamic distribution and plasma-lymphatic uptake to the whole-body circulatory system (8). 3.To assess naïve predictions of systemic PK disposition for mAbs after IV administration, the median of estimated human KdFcRn values from mAbs with available IV data (n=50) was evaluated. To achieve this, the database was randomly divided into training (n=35) and validation (n=15) datasets. The median of the estimated drug-specific KdFcRn values from the training dataset was then used for predictions of the validation dataset. 4.The identified median KdFcRn value was applied to perform naïve PK predictions after SC administration using the generic WB-PBPK model in OSP Suite and the new SC absorption module. The predictive performance of this approach was assessed using an external dataset (n=12). Results 1.A reference database including 62 mAbs was constructed from clinical studies reporting PK profiles after IV and/or SC administration. Among these mAbs, 50 were identified with IV reference data, 42 with available SC reference data, and 30 were identified with both IV and SC reference data. 2.The generic implementation of the SC module predicted area under the curve (AUC), maximum concentration (Cmax), and time to maximum concentration (Tmax) after SC delivery within a 0.50-2.00-fold range for 98% of the mAbs identified with both IV and SC reference data. 3.Using the median of estimated human KdFcRn from the training dataset, plasma exposure and half-life were accurately predicted within a 0.50–2.00-fold range for 100% of the validation dataset. This median value was considered as a reference value. 4.Using the reference KdFcRn value, the generic WB-PBPK model including the SC absorption module, predicted AUC, Cmax, and Tmax within a 0.50-2.00-fold range for 94% of the external dataset. Furthermore, the lymphatic flow and endosomal uptake in the injection site were identified as sensitive parameters. In order to integrate variability in the absorption phase, these parameters with their inter-mAbs variability were estimated using SAEM algorithm coupled with the WB-PBPK model (9). This hybrid pop WB-PBPK approach indicated that an improved overall predictive performance could be achieved by increasing the default endosomal uptake rate from 0.294 to 0.741 (1/min). Conclusion The reported workflow, which integrates the WB-PBPK model, the SC absorption module, parameter estimates and inter-mAbs variability with pop WB-PBPK approach, along with the reference value for human KdFcRn, provides a robust framework for naïve predictions of mAbs PK in humans following SC administration. This approach offers an efficient alternative to traditional preclinical models, for guiding FIH trials in the absence of PK data after IV administration.