Optimization and Validation of an In Vitro Blood Brain Barrier Permeability Assay Using Artificial Lipid Membrane

The blood-brain barrier is composed of non-fenestrated capillary endothelial cells and astrocytes which limit the brain penetration of most of the CNS drug candidates [1]. Large number of compounds enters the brain by transcellular passive diffusion, which is driven by concentration gradient between blood and the brain [2]. There are also two active processes involved in the BBB that influence penetration: active influx transporters (e.g. amino acid, peptides) and active efflux transporters (e.g. P-glycoprotein’s, multi-drug resistant proteins) [3,4]. In addition, plasma protein binding which reduces the free drug concentration available for BBB penetration and partial metabolism as well as renal excretion will also limit its penetration into the brain. Overall the passive diffusion is the major driving force moving most molecules into the brain.


Introduction
The blood-brain barrier is composed of non-fenestrated capillary endothelial cells and astrocytes which limit the brain penetration of most of the CNS drug candidates [1]. Large number of compounds enters the brain by transcellular passive diffusion, which is driven by concentration gradient between blood and the brain [2]. There are also two active processes involved in the BBB that influence penetration: active influx transporters (e.g. amino acid, peptides) and active efflux transporters (e.g. P-glycoprotein's, multi-drug resistant proteins) [3,4]. In addition, plasma protein binding which reduces the free drug concentration available for BBB penetration and partial metabolism as well as renal excretion will also limit its penetration into the brain. Overall the passive diffusion is the major driving force moving most molecules into the brain.
Several in vitro methods and in silico models have been employed for the prediction of blood-brain barrier permeability. The in vitro methods include cell cultures derived from cerebral (brain capillaries as well as primary, low passage or immortalized brain endothelial cells) and non-cerebral sources (MDCK cell lines) as well as non-cell based in vitro systems. Although, the prediction with brain endothelial cells gives the best scoring to the In vivo system [5] but difficulties in establishing and maintaining primary culture, as well as high time consuming methods, make the assay unfeasible as a high throughput screening assay.
Parallel artificial membrane permeability assay was originally reported with 10% (w/v) egg lecithin in dodecane and further its variants has been studied [6][7][8]. A comparison of the three most used PAMPA models viz., Hexadecane Membrane (HDM), Dioleyoylphosphatidylcholine (DOPC) and Double Sink (DS-PAMPA) were recently carried out by Avdeef and Tsinman [9] and explaining permeability differences. Because of the specific nature of the permeability, it was used mainly for the prediction of the gastrointestinal absorption [10]. Attempts to modify the monolayer to improve the prediction of BBB penetration were done using porcine polar brain lipids [11].
We have developed a modified approach for measuring steadystate drug permeability using a permeation barrier made of a tight layer of phospholipids (Porcine brain lipid and Phosphatidylcholine) on filter plate. The aim of this paper is to check the effect of incubation time in the in-vitro brain permeability which was not sufficiently investigated in the earlier findings. In this study we have tested the permeability at various time points and the methods has been validated using 16 structurally diverse commercial drugs covering a broad range of physicochemical properties and absorption properties upon oral administration in humans. P app values obtained from Porcine Brain Lipid (PBL) and Phosphatidylcholine lipid membrane were compared with literature reports. We have also run the same set of the 16 drugs for the log BB prediction using Qik Prop software and the predicted permeability values were compared with experimental log BB values.

Methods
Test compounds: The drugs chosen to validate the phospholipids vesicle based membrane model cover a wide range of physiochemical properties (molecular weight, Log P, Log D see Table 1). A set of 16 structurally diverse commercially drugs (which show effects on CNS with BBB penetration properties) were selected as a test candidates. Amongst them, 12 drugs belong to CNS + and 4 drugs to CNScategory [11]. The stock solutions (10 mm) of all standard drugs were prepared by dissolving the drug in DMSO. The maximum absorbance (λ max ) of these test standards was measured using a 96 well plate UV spectrophotometer (250-750 nm). The working solutions of the test standards were obtained by dissolving the drug in phosphate buffer pH 7.4 on a sonication bath (Branson 1510, Branson Ultrasonic B.V, The Netherlands) followed by filtration through 0.22 µm filter (Millex-GS, Millipore, USA). The concentration of the different drug solutions had to be high enough that the amount of drug in the acceptor chamber during permeation studies could be quantified by means of UVabsorbance, and still be below the solubility limits.

PAMPA-PBL procedure:
A PAMPA-PBL was performed in a 96 well sandwich plate format according to Schmidt and Lynch [12] with slight modification. The porcine brain lipid membrane was constituted by adding the 4 µL (20 mg/mL in dodecane) PBL membrane preparation on the PVDF filter in each well of 96 well plates. The test solutions (300 µL of 250 µM drugs) were added to each donor well while the acceptor wells were filled with 300 µL of PBS. The donor plate was placed on the top of the acceptor plate to create a sandwich. The assembly was incubated for different time intervals (2,5,8,16 and 24 hours) at 25°C. After completion of incubation, the sandwich was disassembled and the acceptor solutions were transferred to a 96-well UV transparent plates (Corning) and the concentration of the drugs was measured spectrophotometrically (Spectramax 190; Molecular Device Corporation, California, USA) at the wavelength most appropriate for each drug. The experiments were performed at least in triplicate for every compound. The mean value and standard error mean are reported.
Where accepter accepter equlibrium equlibrium Where, V D is the donor solution volume (μL), V A is the acceptor solution volume (μL), A is the surface area of the filter (cm 2 ) and t is the incubation time.
Effect of time on permeability: Diffusion of the tested compounds in the 96 well plate, from the donor to acceptor compartments for the PAMPA-PBL and PAMPA-Phosphatidylcholine lipid assays was monitored at various time points (2,5,8,16, 24 hours). Drugs were added to 96 well plates in three replicates for a set time points. After completion of the incubation, the acceptor plate was separated and the diffused drug was analyzed by UV spectrophotometer. The P app calculated for each time points are listed in Table 3.
In-Silico studies: All 16 drugs were evaluated for properties predictions by using QikProp V 1.3 software. Monte Carlo statistical mechanics simulation [13] was used to generate descriptors like Solute-Water Coulomb and Lennard-Jones energy, Solute internal energy, Dipole moment, Solute Accessible Surface Area (SASA), Hydrophobic (FOSA) and Hydrophilic (FISA) component, Donor and acceptor hydrogen bonds, non conjugated amines and amides and No. of rotatable bonds. There were total 5 significant descriptors to calcute Log BB (brain/blood concentration ratio) value vis., FOSA, FISA, amine, dipole moment and rotatable bond. The FOSA and non conjugated amine increase the concentration of drug in brain. The increased polarity of drug will reflect as an increase in the hydrophilic surface area, dipole moment, and flexibility of drug.
Hence, it will lead to increase the concentration of the drug in blood. The QPlogBB values generated from an In-silico method were compared with the experimental log BB [14] values are shown in Table  2,

Shorter Incubation time
Incubation time with PBL previously reported was 18 hours [11] and later its decrease has been set for other solvents down to 2 hours by using the constant agitation at 200 rpm [15]. In this present study we have optimized the BBB permeability assay an In vitro higher throughput model for the determination of the permeability in two different lipids viz., PBL and Phosphatidylcholine. We have attempted permeability studies at various time points like 2, 5, 8, 16 and 24 hours (Table 3), negligible difference was observed in the permeability values (P app ) after 5 hours of incubation for both (PAMPA-PBL and PAMPA-Phosphatidylcholine) models. A set of 16 structurally diverse, commercially available drugs were used to validate both PAMPA models and both are capable to identify CNS + (High brain penetration) and CNS -(Low brain penetration) drugs. Each model characterized all the compounds as per the literature [16] classification. The P app value of the drugs was separated into two P app ranges. The compounds which have P app value greater than 4 × 10 -6 Cm/S are CNS + and the compounds which have P app value lower than 2 × 10 -6 Cm/S are CNS -. Moreover, the only difference in the experimental setup between two models is the lipid membrane. While the PAMPA-PBL model uses a more complex porcine brain extract to mimic the blood-brain barrier, the membrane barrier associated with the PAMPA-Phosphatidylcholine model, consists of 2% (w/v) L-α-Phosphatidylcholine dissolved in dodecane.
The correlation of the P app value of the PAMPA-PBL and PAMPA Phosphatidylcholine assays with the literature reports showed good linearity of r 2 of 0.9487 and 0.930 respectively; Figures 1 and 2. In-silico permeability prediction: Penetration of BBB, into the CNS, is a complicated procedure involving a number of physiochemical properties (17)(18)(19). For permeability through the BBB, QikProp predicts QPlogBB; the brain blood partition coefficient. For the assessment of QPlogBB, a set of 16 drug molecules, in which 12 drugs pass through the BBB and enter the CNS and 4 Non CNS drugs as negative control was used.  The physicochemical properties and permeability values of the drugs used in the validation of the PAMPA Porcine brain lipids and Phosphatidylcholine vesicle membrane models. In Figure 3 all of the investigated drugs are lie within the indicated limits of -3< QPlogBB <1.2; the usual limits given for an experimentally derived BBB penetration ranges between -2.0 and +1.0. Drugs with log BB greater that 0.3 are characterized as excellent, whereas drugs with log BB less than -1.0 are considered poor (20). These limits are denoted in Figure 3. Fifty percent of the compounds are above the 0.3 threshold, twenty five percent of the compounds are in between 0.3 and -1 and final 25% have values of less than -1. The compounds which have less than -1, QPlogBB value are used as antibiotics and for cardiovascular disease and are not specifically designed to penetrate BBB. The correlation of the logBB value of the both predicted with the experimental reports showed good correlation of r 2 of 0.91

Conclusion
Both the PAMPA-PBL and PAMPA Phosphatidylcholine models have been successfully developed and optimized in high throughput format. We observed that the differences between the artificial lipids (PBL and Phosphatidylcholine) on the BBB classification of the tested compounds seem to be negligible. With the same experimental conditions, the Porcine Brain Lipids and Phosphatidylcholine displayed very good correlation for the selection of CNS + and CNScompounds. The comparison of two different set of data gives a better classification of the compounds. The In-silico data also showed fair correlation with the experimental log BB values.
Although, PAMPA can't be used as a surrogate assay for a cellular permeability assay, but this assay can be used for the prediction of BBB penetration in a more robust fashion. Our model gives near to 95% of the predictions which are in good accordance with the literature.