In Situ, In Vitro, and In Silico Permeability Values as Inputs for Blood-Brain Barrier Penetration Prediction: Impact on Brain Exposure for Passively Diffusing Compounds, with Ethanol as a Case Study
Aleksandr Petrov1,2, Elena Righetti3,4, Michael Rapp5, Charlotte Kloft1,6, Andreas Reichel7, Wilhelm Huisinga1,2
1Graduate Research Training Program PharMetrX, 2Institute of Mathematics, University of Potsdam, 3Fondazione the Microsoft Research-University of Trento Centre for Computational and Systems Biology (COSBI), 4Department of Cellular, Computational and Integrative Biology (CIBIO), University of Trento, 5Department of Social and Preventive Medicine, University of Potsdam, 6Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, 7Preclinical Modeling and Simulation, Preclinical Development, Bayer, Bayer AG
Introduction: For compounds acting on the central nervous system (CNS), predicting brain concentrations is crucial, as it often differs significantly from plasma due to the protective properties of the blood-brain barrier (BBB). Physiologically based pharmacokinetic (PBPK) modeling is widely used to predict brain exposure based on compound properties and physiological data [1-3]. One of the key input parameters in these models is the permeability surface area product (PS), which determines permeability across the BBB. There is, however, no consensus in published PBPK CNS models on selecting PS values. Some models use permeability values from in situ brain perfusion experiments [1,2], while others rely on data from various in vitro techniques, including cell line-based (Caco2, MDCK, LLC-PK1) and Parallel Artificial Membrane Permeability Assay (PAMPA) methods [2, 4]. Additionally, some models employ in silico predicted PS values obtained through regression models based on in situ brain perfusion or PAMPA data [2, 3]. For a given compound, permeability values can span several orders of magnitude, making its selection for PBPK model inputs challenging. Objectives: •To investigate the variability in PS values obtained through in situ, in vitro and in silico methods for compounds crossing the BBB via passive diffusion. •To demonstrate the impact of PS values on brain exposure in humans using ethanol—a well-known compound crossing the BBB via passive diffusion that is also relevant to our research on alcohol addiction development. Methods: The assessment of PS variability focused on compounds crossing the BBB via passive diffusion (0.3 < Kp,uu,brain < 3), avoiding variability from receptor-mediated processes like active transport. Experimental permeability values from in situ brain perfusion, in vitro cell lines and PAMPA methods were collected from the literature [5,6]. Additionally, permeability values predicted by in silico models [7,8] were considered, following the approach of previously published PBPK CNS models [2,3]. For scaling, human brain mass [9] was used for in situ brain perfusion and BBB vasculature surface area [10] for in vitro permeability values. To predict ethanol concentration in the brain, we used a model previously developed for humans [1,3]. Validation of ethanol brain exposure prediction was performed using human brain data [11], and a 2-fold error margin was applied to assess prediction accuracy. Results: Permeability data were collected for 10 compounds, with each compound having 1 to 3 measurements from each of the experimental techniques and 2 predicted values from in silico models. On average, PS differences across all techniques varied by a factor of 230, ranging from a 3-fold difference for lidocaine to a 1385-fold for ethanol. For half of the compounds, the highest PS values came from in situ perfusion, while the lowest came from in vitro PAMPA. In vitro cell lines generally provided intermediate values with the least variability and maximum PS of 40 L/h, compared to 522 L/h and 2357 L/h for in situ perfusion and in vitro PAMPA, respectively. Despite ethanol exhibiting the highest variability in PS values, brain exposure predictions remained within a 2-fold error margin for all PS values, although they slightly overpredicted the observed data. The only exception was a PS value predicted by an in silico model based solely on logP. The effect of permeability on brain exposure to ethanol became evident only for PS values below 5 L/h, while all experimentally based PS values were within the range of 17 to 277 L/h. Conclusions: The analysis showed that even for compounds crossing the BBB primarily via passive diffusion, PS values can vary by up to four orders of magnitude, highlighting the need for careful justification when selecting input values for PBPK CNS models. However, using ethanol as a case study, brain exposure predictions remained insensitive to PS values differing by several orders of magnitude. This is likely because the distribution of such compounds is blood flow-limited rather than permeability-limited, making PS differences particularly critical for the latter. Meanwhile, using in silico models for permeability prediction should be used with caution, as failing to account for their limitations may lead to unrealistic PS values. Future research should focus on improving PS input selection for permeability-limited compounds to enhance the accuracy of brain exposure predictions.
[1] Gaohua L et al. Drug Metab Pharmacokinet. 2016;31(3):224–33. [2] Saleh, M et al. J Pharmacokinet Pharmacodyn. 2021;48:725–741 [3] Bowman C et al. Biopharm Drug Dispos. 2023;44(1):60–70. [4] van Valkengoed et al. J Pharmacokinet Pharmacodyn. 2025;52(16). [5] Avdeef A. John Wiley & Sons. 2012:575-680. [6] Smith Q R et al. Fluid Barriers CNS. 2024;21(1):100. [7] Luco J et al. Current Computer-Aided Drug Design 2006;2(1):31–55. [8] Grumetto L et al. Mol Pharm. 2016;13(8):2808-2816. [9] International Commission on Radiological Protection (ICRP). ICRP Publication 2002;89. [10] Abbott N J et al. Neurobiol Dis. 2010;37(1):13-25. [11] Thierauf-Emberger A et al. Int J Mol Sci. 2023;24(17):13499.