Inter-species and in vitro-in vivo scaling for quantitative modeling of whole-body drug pharmacokinetics in patients: Application to the anticancer drug oxaliplatin
Simona Catozzi (1), Roger Hill (2), Xiao-Mei Li (3), Sandrine Dulong (1,3), Elodie Collard (4), Annabelle Ballesta (1)
(1) Institut Curie, Inserm U900, MINES ParisTech, CBIO - Centre for Computational Biology, PSL Research University, Saint-Cloud, France (2). EPSRC and MRC Centre for Doctoral Training in Mathematics for Real-World Systems, University of Warwick, Coventry, UK. (3) UPR “Chronotherapy, Cancers and Transplantation,” Faculty of Medicine, Université Paris-Saclay, Villejuif, France. (4) CEA, CNRS, NIMBE, Université Paris- Saclay, Gif-sur-Yvette, France.
Objectives: We designed a physiologically based pharmacokinetic (PBPK)model of the anticancer drug oxaliplatin, and developed a scaling methodology that shows how to accurately integrate preclinical results into whole-body models, to ultimately conduct in silico clinical trials.
Methods: We built two compartmental models for oxaliplatin PK - one for the whole blood and one for the whole body PK, - representing the amount of drug's platinum (Pt) in the plasma, in the red blood cells (RBCs), in the liver and in other non-eliminating organs. The whole-blood model was calibrated using ex-vivo data from Mouse, Rat and Human [1, 2]. The whole-body model was calibrated using data from patients treated with oxaliplatin as a chemotherapy [3]. Parameter estimation was performed with least squares minimization with the CMAES algorithm (Matlab). Parameter identifiability was analyzed with the profile likelihood method for practical identifiability [4].
Results: With the whole-blood oxaliplatin PK model, we investigated various methodologies for parameter scaling to faithfully calibrate the human model using preclinical datasets. For each of the three species, the minimal model-to-data errors were retrieved from the independent calibration on each individual dataset. This error got remarkably high for the Human model, when its parameters were obtained by adjusting the ones of the combined Mouse and Rat models using PB scaling. Hence, such simple approach for inter-species translation revealed to be inadequate to reproduce Human observations. We thus looked for the parameter that was driving the failure of this strategy, by using all the three datasets and allowing one parameter to be different by species while keeping the other five computed by PB scaling laws. This was repeated for all six parameters and the best fit was obtained for the species-specific estimation of Pt plasma protein binding rate, yielding model errors as good as the minimal ones. Further, we considered the more realistic case of partial access to Human data, and we integrated to the preclinical datasets, only the total Pt plasma levels, for Human model calibration, which still guaranteed accurate fits and similar parameter identifiability. Finally, we extended our whole blood PK model, to a whole body semi-mechanistic Human model of oxaliplatin PK. All the parameters but three were inferred from the ex-vivo model, and the remaining three were estimated from the time-concentration profiles of total plasma Pt. The model achieved good fit to clinical data, and was validated on unseen measurements, which endorsed the legitimacy of our ex-vivo to in-vivo scaling methodology. Moreover, the model provided predictions on the time-resolved concentrations of free and bound Pt in the liver, such data being unavailable in patients due to technical constraints.
Conclusions: This work presents a rational approach towards reliable inter-species translation from bench to bedside, providing a good balance between precision and computational cost. It built upon the formalism of physiologically based modeling that allow for mechanistic understanding of a drug’s pharmacokinetics and predictions of unreachable quantities in the clinical context.
References:
[1] F R Luo et al. J. Biochem. Mol. Toxicol. 1999.
[2] L Pendyala and PJ Creaven. Cancer Res. 1993.
[3] C Massari et al. Cancer Chemother. Pharmacol. 2000.
[4] A Raue et al. Bioinformatics 2009.