Wilbert de Witte 1, Fouad Seriari 1, Luis David Jimenez Franco 1, Stephan Schaller
1 Esqlabs Gmbh (Gent, Belgium)
Background: The predictive performance of the PK-Sim large molecule biodistribution model has been publicly evaluated on a dataset from 8 different molecules, of which 6 were monoclonal antibodies, one was a domain antibody of 25 kDa, and one was the 5.5 kDa protein inulin.[1] Biodistribution data are mostly obtained from radiolabeled large-molecule studies, and these radiolabels can be either residualizing (i.e., retained in tissue cells after catabolism of the large molecule) or non-residualizing (i.e., released from tissue cells after catabolism of the large molecule). This makes the obtained radioactivity signal either the sum of intact and degraded large molecules (residualizing labels) or only intact large molecules (non-residualizing labels). In the initial PK-Sim model training dataset, novel modalities such as scFv, F(ab), nanobodies, and affibodies were not included, and only non-residualizing labels were used.
Objectives: We aimed to increase the variety of data used for biodistribution validation by expanding the dataset to include more therapeutic modalities and both residualizing and non-residualizing radiolabels.
Methods: Simulation models were created using the large molecule model in PK-Sim® v12.0 and were extended with renal clearance, organ-specific endosomal clearance compartments, and a residualizing radiolabel observer in MoBi® (v12.0). Data were collected from the literature, which was searched for studies with both residualizing and non-residualizing labels, and a large variety of modalities, and included 35 different molecules with data for mice, rats, and monkeys. A repeated sensitivity analysis was performed to identify the most sensitive parameters in the two-pore model (flow fraction via large pores, fluid recirculation flow proportionality factor, fraction endosomal, fraction
interstitial, fraction vascular, hydraulic conductivity, lymph flow proportionality factor, radius (large pores), radius (small pores)) for optimization of the model fit. Model fitting was performed in MoBi®, using the Levenberg-Marquardt algorithm and evaluated based on estimated parameter values and uncertainties as well as visual inspection of the goodness of fit.
Results: The literature dataset was well predicted for most molecules across species after fitting one of two molecule-specific parameters to the observed data for each molecule: renal clearance as fraction of the glomerular filtration rate was fitted for affibodies, diabodies, F(ab), scFv, and nanobodies, and FcRn affinity in endosomes was fitted for monoclonal antibodies. While the pharmacokinetic profiles differed greatly across modalities, no specific trends in model performance were observed with respect to modalities. The repeated sensitivity analysis revealed the dominant role of the vascular and interstitial fractions in determining the whole-organ concentrations relative to the plasma concentrations. Other parameters had reduced sensitivities or only showed sensitivity for the largest modality, which was the case for most of the lymph flow-related parameters. The model was further refined by estimating the interstitial and vascular fractions of the organs for which observations were available (liver, lung, muscle and spleen), and the endosomal fraction with the residualizing label data. While the model performance improved after estimation of interstitial and vascular fractions, the improvement was moderate (~20% reduction in total error) and the estimated parameter values were mostly similar and always within 3-fold of the original parameter values.
Conclusion: Our study successfully validated and refined the biodistribution model in PK-Sim and allowed us to further refine tissue-specific catabolism rates by using residualizing and non-residualizing radiolabeled biologics data from mice, rats, and monkeys, and a wide range of modalities. Further refinement of the human model will require a similar human biodistribution dataset. These results also highlight important differences between biodistribution studies using residualizing or non-residualizing labels, and the critical role of mechanistically incorporating the radiolabel’s fate into the model. Finally, our study showed the added value of biodistribution studies that use both types of radiolabels for the same compound, enabling precise separation of local catabolism from passive diffusion contributions to the final radioactivity signal.
References:
[1] C. Niederalt, L. Kuepfer, J. Solodenko, T. Eissing, H.-U. Siegmund, M. Block, S. Willmann, J. Lippert, A generic whole body physiologically based pharmacokinetic model for therapeutic proteins in PK-Sim, J. Pharmacokinet. Pharmacodyn. 45 (2018) 235–257. https://doi.org/10.1007/s10928-017-9559-4.
Reference: PAGE 34 (2026) Abstr 12150 [www.page-meeting.org/?abstract=12150]
Poster: Drug/Disease Modelling - Absorption & PBPK