A systematic framework for the incorporation of preclinical species data into physiologically-based pharmacokinetic models using a middle-out approach to improve the translation of drug distribution in humans
Estelle Yau (1,2,3), Michael Gertz (2), Kayode Ogungbenro (1), Leon Aarons (1), Andrés Olivares-Morales (2)
(1) Centre for Applied Pharmacokinetic Research (CAPkR), University of Manchester, UK; (2) Roche Pharma Research and Early Development (pRED), Roche Innovation Center Basel, Switzerland; (3) Current affiliation: Sanofi R&D, DMPK France, France
Objectives:
Optimization of parameters in complex models, such as the whole-body physiologically-based pharmacokinetic (PBPK) model by model fitting to observed data is a challenging and time-consuming process, given the large number of parameters and availability of data which are mostly limited to plasma observations [1]. Additionally, the decision on which parameters to fix or estimate is subjective and, therefore the final model and parameters may vary significantly between different investigators [2]. In this study, a systematic approach was developed to address this issue. The objectives were to investigate if simplified PBPK models in conjunction with preclinical in vivo data can provide suitable estimates of unbound tissue-partitioning coefficients (Kpu) and to evaluate the performance of these models for translation of drug distribution from animals to humans.
Methods:
Diazepam was selected due to the availability of intravenous plasma studies in rats (n=5), monkeys (n=1) and humans (n=7) as well as relevant and sensitive in vitro information for Kpu-predictions (e.g., logP, fup and B:P [3]). To reduce the complexity of the PBPK model structure and/or limit the number of parameters for estimation, two approaches were considered. Firstly, using established kinetic lumping methods based on tissue time constants [4] and secondly, using cluster analysis to identify tissues sharing common composition which in turn can be clustered by assuming either common Kpus or Kpu scalars. The different mechanistic models were fitted to the rat and monkey PK data and Kpu-values were estimated using FOCE-I and SAEM methods [OA{1] in NONMEM v7.3. Model performance was assessed based on physiological plausibility, visual and numerical predictive checks. The best models in preclinical species were then used to predict the human PK profile and volume of distribution at steady state (Vss). For comparison, the traditional PBPK strategy, whereby a priori Kpus were calculated based on diazepam physicochemical properties and human tissue composition data using the Rodgers and Rowland equations, was also evaluated [5,6,7]. The model system-related parameters (i.e., blood flows and tissue volumes) were taken from literature sources and scaled between species [8].
Results:
Compared to a kinetically lumped PBPK model, the use of common Kpus or Kpu scalars by clustering analysis retains the tissue/organ structure of the whole-body PBPK model while model parameters are reduced on the basis of physiological similarities of tissue composition.
Several PBPK models with either 3 or 4 common Kpus or scalars (different possible tissue groupings) were successful at describing the rat IV data comparable to empirical models (reasonable fits, good parameter precision). Considering the different criteria (precision of estimates, physiologically relevance of partitioning coefficients and estimated Vss value), the best model for fitting the rat PK were the models with 3 or 4 common Kpu scalars using k-means clustering. Consequently, prediction of human drug distribution was performed using the Kpu values estimated by fitting these best models to rat or monkey in vivo data. The prediction of the distribution behavior of diazepam in humans was considerably improved compared to the traditional PBPK approach. The blood Vss in human was predicted within 1.1 to 3.1-fold error (47 L to 132 L) of the observed blood Vss (152 L) after optimisation of either rat or monkey data, respectively. These predictions were better compared to the blood Vss estimated from the traditional PBPK modelling approach (41 L, corresponding to 3.7-fold error).
Conclusions:
PBPK models represent the gold-standard to predict human PK for first-in-human studies. This work proposed a systematic strategy to integrate preclinical data and fit simplified PBPK models to successfully predict the distribution of small molecules in human. Using the approach of simplified PBPK models with common scalars, PK profiles could be well described in preclinical species and using the best models, plasma profiles were successfully predicted in human for diazepam. For an exhaustive evaluation, the work and models proposed herein may be extended to different drug classes and more compounds. This strategy for PBPK models of drug distribution could also be extended to translation within species e.g., from adult to paediatric population.
References:
[1] Gueorguieva I et al. J Pharmacokinet Pharmacodyn (2006) ; 33(5):571-94
[2] Margolskee, A. et al. Eur J Pharm Sci (2017); 96 : 610-625
[3] Yau E et al. AAPS J (2020); 22(41) :1-13
[4] Nestorov, I. A. et al. J Pharmacokinet Biopharm (1998); 26 (1) : 21-46
[5] Rodgers T & Rowland M. J Pharm Sci (2005); 94(6):1259-76.
[6] Rodgers T & Rowland M. J Pharm Sci (2005); 95(6):1238-57.
[7] Poulin P et al. J Pharm Sci (2011) ; 100(10):4127-57
[8] ICRP Publication 89. Ann ICRP (2002); 32 (3–4): 5–265