Background: Optimization of drug specific parameters in complex models, such as whole body physiologically-based pharmacokinetic (WBPBPK) models, by model fitting to observed data is challenging. This process is time-consuming and models are often unidentifiable/over-parameterized due to the large number of parameters and availability of data which are mostly limited to observations from plasma [1]. The current work investigates if simplified PBPK models in conjunction with preclinical in vivo data can provide suitable estimates of unbound tissue-partitioning coefficients (Kpu) for translation of drug distribution to human. The results from the current analysis will be compared to the gold-standard method for Kpu predictions, the Rodgers and Rowland (R&R) method [2]. To this end, mechanistic models with different levels of complexity were investigated using diazepam as a case study.
Methods: Criteria for compound selection were the availability of intravenous plasma concentrations data in rats and humans and relevant and sensitive in vitro information for Kpu-predictions (e.g., logP, fup and B:P [3]). Diazepam (DZP) was selected in this study due to the abundance of PK profiles available from the literature in humans (n=36) and rats (n=5). In order to limit the number of parameters for estimation, but retain the physiological structure of the PBPK model, tissues were clustered into 3 to 4 groups based on the similarity in tissue composition. For each of those groups of tissues, either a common Kpu value was estimated or a common Kpu scalar was determined. In the traditional PBPK strategy, a priori Kpus were calculated based on DZP physicochemical properties and human tissue composition data using R&R equations [2,4]. Blood flows and tissue volumes were taken from literature sources and scaled between species [5]. The different mechanistic models were fitted to the rat PK data and Kpu-values were estimated using FOCE-I methods in NONMEM v7.3.
Results: Compared to a kinetically lumped PBPK model, the current approach retains the tissue/organ structure of the WBPBPK model and model parameters are reduced on the basis of physiological similarities of tissue composition while blood-flows and organ volumes were maintained unchanged. Although the clustering analysis resulted in slightly different groups of tissues depending on the clustering method, kidney and liver were generally grouped together whereas bone, brain, muscle, pancreas were found more similar. Clustering into 4 tissue groups appeared more physiologically relevant in terms of tissue composition, with adipose as a separate group due to its particular composition. For DZP, these models described the rat concentration-time profiles well and allowed estimation of physiologically relevant partitioning coefficients. Prediction of human drug distribution using the Kpu values estimated by the investigated models using in vivo rat data, especially the model with 4 scalars using k-means clustering for determining similarity in tissue drug partitioning, considerably improved the distribution behavior of DZP in humans compared to the traditional PBPK approach.
Conclusions: PBPK models represent the gold-standard to predict human PK for first in human studies. Generating relevant information from in vitro and preclinical in vivo data might provide greater confidence particularly for drug distribution, as mechanisms of drug distribution are assumed similar between species. The current study provides a rationale and reproducible assessment of analyzing preclinical data to aid translation of drug distribution within a PBPK modelling framework. The work and models proposed therein may be extended to other compounds and species, for a more exhaustive evaluation.
References:
[1] Gueorguieva I et al. J Pharmacokinet Pharmacodyn (2006) ; 33(5):571-94
[2] Rodgers T & Rowland M. J Pharm Sci (2005); 94(6):1259-76.
[3] Yau E et al. AAPS J (2020); 22(41) :1-13
[4] Poulin P et al. J Pharm Sci (2011) ; 100(10):4127-57
[5] ICRP Publication 89. Ann ICRP (2002); 32 (3–4): 5–265