Exposure-response (E-R) analyses are an integral component in the development of oncology products. Characterizing the relationship between drug exposure metrics and response allows the sponsor to use modeling and simulation to address both internal and external drug development questions (e.g., optimal dose, frequency of administration, dose adjustments for special populations). This white paper is the output of an industry-government collaboration among scientists with broad experience in E-R modeling as part of regulatory submissions. The goal of this white paper is to provide guidance on what the preferred methods for E-R analysis in oncology clinical drug development are and what metrics of exposure should be considered.

Tusamitamab ravtansine (SAR408701) is an antibody-drug conjugate (ADC), combining a humanized monoclonal antibody (IgG1) targeting carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5) and a potent cytotoxic maytansinoid derivative, DM4, inhibiting microtubule assembly. SAR408701 is currently in clinical development for the treatment of advanced solid tumors expressing CEACAM5. It is administered intravenously as a conjugated antibody with an average Drug Antibody Ratio (DAR) of 3.8. During SAR408701 clinical development, four entities were measured in plasma: conjugated antibody (SAR408701), naked antibody (NAB), DM4 and its methylated metabolite (MeDM4), both being active. Average DAR and proportions of individual DAR species were also assessed in a subset of patients. An integrated and semi-mechanistic population pharmacokinetic model describing the time-course of all entities in plasma and DAR measurements has been developed. All DAR moieties were assumed to share the same drug disposition parameters, excepted for clearance which differed for DAR0 (i.e. NAB entity). The conversion of higher DAR to lower DAR resulted in a DAR-dependent ADC deconjugation and was represented as an irreversible first-order process. Each conjugated antibody was assumed to contribute to DM4 formation. All data were fitted simultaneously and the model developed was successful in describing the pharmacokinetic profile of each entity. Such a structural model could be translated to other ADCs and gives insight of mechanistic processes governing ADC disposition. This framework will further be expanded to evaluate covariates impact on SAR408701 pharmacokinetics and its derivatives, and thus can help identifying sources of pharmacokinetic variability and potential efficacy and safety pharmacokinetic drivers.

OBJECTIVE: The aims of this study are (i) to assess the predictive reliability of the physiologically based software PhysPK versus the well-known population approach software NONMEM for the cited semi-mechanistic PK model, (ii) to determine whether these modelling approaches are interchangeable and (iii) to compare acausal with causal modelling approaches in the framework of semi-mechanistic PK models.
METHODS: A semi-mechanistic model was proposed, which assumed oral administration of a solid dosage form with a peripheral compartment and two active metabolites. The model incorporates intestinal transit, dissolution limited by solubility, variable efflux transporter expression along the gut and linear and non-linear metabolism in the gut and liver. Four different approximations to the theoretical model were developed in order to validate both the new software and modelling methodology.
RESULTS: Plasmatic concentrations correlation plots as well as relative errors in AUC0-48 and Cmax predictions revealed the accuracy of PhysPK in the prediction of these exposition parameters. Physiological and acausal object oriented version systematically under-estimated AUC0-48 and Cmax of the parent drug, whereas metabolites were over-estimated when taking the semi-mechanistic and extraction-based metabolism version as the reference.
CONCLUSIONS: PhysPK has been properly validated, where differences are related to numerical precision of integrators and solvers. A systematic bias for the parent drug and active metabolites was predicted when a semi-mechanistic approach including extraction-based metabolism was compared to the physiologic and acausal approach, showing that interchangeability might be possible when intrinsic-clearance metabolism is implemented in the semi-mechanistic approach. The acausal and object-oriented methodology allows for defining the semi-mechanistic model through its local mechanisms and relationships among entities, without the need to build the final set of Ordinary Differential Equations.

Accurate prediction of the human pharmacokinetics (PK) of a candidate monoclonal antibody from nonclinical data is critical to maximize the success of clinical trials. However, for monoclonal antibodies exhibiting nonlinear clearance due to target-mediated drug disposition, PK predictions are particularly challenging. That challenge is further compounded for molecules lacking cross-reactivity in a nonhuman primate, in which case a surrogate antibody selective for the target in rodent may be required. For these cases, prediction of human PK must account for any interspecies differences in binding kinetics, target expression, target turnover, and potentially epitope. We present here a model-based method for predicting the human PK of MAB92 (also known as BI 655130), a humanized IgG1 κ monoclonal antibody directed against human IL-36R. Preclinical PK was generated in the mouse with a chimeric rat anti-mouse IgG2a surrogate antibody cross-reactive against mouse IL-36R. Target-specific parameters such as antibody binding affinity (KD), internalization rate of the drug target complex (kint), target degradation rate (kdeg), and target abundance (R0) were integrated into the model. Two different methods of assigning human R0 were evaluated: the first assumed comparable expression between human and mouse and the second used high-resolution mRNA transcriptome data (FANTOM5) as a surrogate for expression. Utilizing the mouse R0 to predict human PK, AUC0-∞ was substantially underpredicted for nonsaturating doses; however, after correcting for differences in RNA transcriptome between species, AUC0-∞ was predicted largely within 1.5-fold of observations in first-in-human studies, demonstrating the validity of the modeling approach. Our results suggest that semi-mechanistic models incorporating RNA transcriptome data and target-specific parameters may improve the predictivity of first-in-human PK.