Permeability Analysis of Neuroactive Drugs Through a Dynamic Microfluidic In Vitro Blood–Brain Barrier Model
Abstract—This paper presents the permeability analysis of neuroactive drugs and correlation with in vivo brain/ plasma ratios in a dynamic microfluidic blood–brain barrier (BBB) model. Permeability of seven neuroactive drugs (Ethosuximide, Gabapentin, Sertraline, Sunitinib, Traxopro- dil, Varenicline, PF-304014) and trans-endothelial electrical resistance (TEER) were quantified in both dynamic (micro- fluidic) and static (transwell) BBB models, either with brain endothelial cells (bEnd.3) in monoculture, or in co-culture with glial cells (C6). Dynamic cultures were exposed to 15 dyn/cm2 shear stress to mimic the in vivo environment. Dynamic models resulted in significantly higher average TEER (respective 5.9-fold and 8.9-fold increase for co-culture and monoculture models) and lower drug perme- abilities (average respective decrease of 0.050 and 0.052 log(cm/s) for co-culture and monoculture) than static mod- els; and co-culture models demonstrated higher average TEER (respective 90 and 25% increase for static and dynamic models) and lower drug permeability (average respective decrease of 0.063 and 0.061 log(cm/s) for static and dynamic models) than monoculture models. Correlation of the resultant logPe values [ranging from 24.06 to 23.63 log(cm/s)] with in vivo brain/plasma ratios (ranging from 0.42 to 26.8) showed highly linear correlation (R2 > 0.85) for all model conditions, indicating the feasibility of the dynamic microfluidic BBB model for prediction of BBB clearance of pharmaceuticals.
Keywords : —BBB, Central nervous system, Drug discovery, Endothelial cells, Microsystems, lBBB.
INTRODUCTION
Despite increasing demands for new treatments of disorders of the central nervous system (CNS) such as Alzheimer’s disease (AD),19 CNS drug research pro- gress has been significantly hindered by the prohibitive barrier from capillaries to brain tissue, the blood–brain barrier (BBB). Recent studies reported that AD was diagnosed in 1/3rd of senior deaths in the US,7 while a new case of AD is developed every 67 s.23 However, the clinical success rates for new CNS compounds (7%) remain lower than other healthcare areas such as cardiovascular disease (20%),36 while the average cost to develop a drug exceeded $1 billion.3 Such low suc- cess rates have been attributed partially to limited prediction capability in pre-clinical models to assess
the passage of drugs across the BBB.38 The BBB, mainly comprised of the capillary’s brain endothelial cells, is the key barrier restricting perfusion of nearly 100% of large (>500 Da) molecules and 98% of small molecules,37 complicating determination of effective dose concentrations of drugs targeting the CNS.
To potentially accelerate the development of new CNS-targeting pharmaceuticals, the high-throughput evaluation of trans-BBB properties can be achieved by developing massively-parallel, low-cost predictive models of BBB clearance43 either in vivo or in vitro. The BBB pre-clinical models allow the discovery of rejected compounds earlier and enable the reduction of attri- tion rates in clinical trials.27 They are capable of pre- dicting whether a compound’s interaction with the BBB will compromise its functionality or whether it reaches the CNS in significant amounts to have a pharmacodynamic effect.8 In vivo models provide similarly complex environments to human physiologi- cal conditions; however, they are subject to high cost, time, and ethical constraints. In vitro models, within the scope of cellular physiology, resolve such issues and enable feasible isolation and observation of indi- vidual physiological mechanisms in repeatable and controllable manners, resultantly emerging as a promising alternative or augmenting model for early drug screening (Fig. 1a).
In vitro BBB models recently incorporated dynamic flows, replicating more realistic in vivo conditions for higher prediction capability. The flow-based dynamic in vitro BBB models have exploited the mechano- transductive response of endothelial cells to wall shear stress (WSS) and its effect on BBB functions.9,10,15,46 For example, the authors’ previous dynamic models reported higher trans-endothelial electrical resistance (TEER) and lower permeability in comparison to tra- ditional transwell-based in vitro models, better repre- senting the cerebromicrovascular environment5,6 found to in vivo.
Despite these advantages, dynamic in vitro models have not been widely accepted for BBB permeability screening in the pharmaceutical industry yet. Accep- tance of in vitro models first requires a standard for translation of model results to the in vivo condition,39 in terms of its predictive ability of permeability to new compounds. Currently, none of the existing dynamic in vitro BBB models utilizing microfluidics2,5,22,42 or hollow fibers14,15 have been assessed for a large num- ber (>2) of CNS drugs to elucidate a translational standard for permeability. The successful assessment of permeability to multiple CNS drugs would establish quantitative correlation with in vivo permeability, and demonstrate the high-throughput potential for such a dynamic model.
FIGURE 1. Microfluidic blood–brain barrier models. (a) The pre-clinical drug screening process would benefit from more innovative in vitro models. Though in vitro models are advantageous due to their low cost, time and ethical con- straints, high experimental control over isolation of individual mechanisms, and allow a more repeatable and high-through- put approach, they lack the complexity of the in vivo envi- ronment. Microfluidic in vitro models allow higher model complexity by introducing a dynamic environment, while maintaining experimental control. (b) The illustration of the previously developed dynamic lBBB system that recreates the micro-cerebrovascular environment with dynamic flows and co-culture of endothelial and glial cells. Also included in the system is two sets of resistance-measuring AgCl elec- trodes. Graphic modified from previously published graphic.5
We previously developed a dynamic in vitro micro- fluidic BBB model (lBBB, Fig. 1b), and characterized the effects of chemical/pH modulation5 and WSS6 on BBB functions, such as cell morphology, fluorescent tracer permeability, TEER measurement by integrated electrodes, and BBB protein expression.6 To establish quantitative correlation with in vivo permeability, this paper reports the permeability measurement of seven CNS drugs (Ethosuximide, Gabapentin, Sertraline Hydrochloride, Sunitinib Malate, Traxoprodil Mesy- late, Varenicline Tartrate, PF-3084014) across the dynamic lBBB model as well as static in vitro transwell platforms prepared with both mono- and co-cultured BBB layers (endothelial and glial cells). Due to the abundance of evidence that drugs of small molecular weights (<500 Da) better cross the BBB, the selected drugs were limited to similarly small molecular weight drugs. Then, to ensure inclusion of a diverse sample set, the selection preferences were given to drugs that cover a wide range of hydrophilicity (between 21.27 and 5.15 logPo/w), which generally dictates the compound’s ability to cross the plasma membrane, and have a wide range of relevant clinical applications including treat- ments for depression, epilepsy, pain, nicotine addiction, and tumors. Additionally, concentration-dependent cytotoxicity on brain endothelial cells for each drug was analyzed. For quality control of prepared cultures, TEER was monitored before and after each permeability test to remove outliers and define cell contiguity. Finally, permeability coefficients were correlated to brain/plasma ratios from previous studies of each drug.
STRUCTURE AND FABRICATION OF THE MICROFLUIDIC BBB MODEL
The previously developed dynamic in vitro micro- fluidic BBB model has been utilized due to proven validation on TEER and permeability properties.5 The microfluidic structure of the lBBB system (Fig. 2a) comprised multiple layers of polydimethylsiloxane (PDMS) and glass substrates to form two crossing channels (4 mm). At the junction of the crossing channels in different layers, a free-standing porous polycarbonate (PC) membrane11 was embedded allowing the diffusion path from the top to the bottom fluidic channels. The PC membrane also provided the co-culture surface for both endothelial and glial cells, enabling the permeability testing through the co-cul- tured BBB cell layers. On both sides of the co-culture membrane, two sets of thin-film electrodes were microfabricated by depositing and patterning AgCl layers on a glass substrate. The distance from the membrane to each electrode, equivalent to the channel height, was only 200 lm, allowing high-accuracy TEER measurement. Each channel layer was con- nected to separate pairs of fluidic inlets and outlets for individual permeability analysis.
Fabrication of the BBB model was similar to the processes described previously,5,6 with some modifica- tions. First, top and bottom glass substrates were sput- tered with layers of Cr/Au/Ag (20/80/800 nm) that were patterned utilizing a lift-off process, forming TEER electrodes. Then the Ag surface was chlorinated chemi- cally by dipping the substrate into 30 mM FeCl3 for 50 s to form an oxidized surface of AgCl, which is non-toxic with high long-term stability.41 Second, following 30 s O2 plasma oxidation at 125 W glass substrates were pressed into uncured PDMS pre-polymer (10:1 elastomer:curing agent ratio), which was then cured on a 110° hot-plate for 30 min. The replica mold was constructed on a Si sub- strate by lithographically patterning SU-8 2075 into defining top and bottom micro-channel structures. For effective bonding, top and bottom PDMS channel layers (with embedded glass layers) and PC membrane were plasma oxidized for 30 s, and the PC membrane was treated for 30 min with 5% 3-aminopropyltriethox- ysilane (APTES) on a hotplate set to 80 °C.4 Third, all three substrates were pressed together at room temper- ature, sandwiching the PC membrane in an irreversible bond. The APTES-based bonding method minimized occurrences of leaks compared with the PDMS pre- polymer ‘‘mortar’’ method.11
FIGURE 2. Microfluidic blood–brain barrier chip for perme- ability assays. (a) Multi-layered channel structure made from patterned PDMS substrate with embedded glass electrode layers. Luminal (top, yellow) and abluminal (bottom, blue) channels are both 4 mm wide and are seeded with bend.3 endothelial cells and C6 astrocytes, respectively, on either side of the free-standing PC membrane. Electrodes allow non- invasive TEER measurement. (b) To test permeability through the lBBB system, the drug is run of constant concentration CL is run through the luminal channel for time Dt, and per- meability is computed from the measured abluminal concen- tration DCA in sample volume VA, and the area A.
MATERIALS AND CELL CULTURE
CNS-Targeting Compounds
Seven brain-targeting commercially-available drugs were utilized for permeability screening of the in vitro BBB models. The seven drugs were provided by the Compound Transfer Program (CTP) by Pfizer Inc., including Varenicline (PF-3430574), Gabapentin (PF-345043), Traxoprodil (PF-1486212), Sertraline (PF-579897), Ethosuximide (PF-344988), Sunitinib (PF-262192), and an unnamed compound (PF-3084014). The compounds have been proven to cross the in vivo BBB to some extent, and represent a wide range of applications, including treatment of pain, depression, seizures, nicotine addiction, and tumor suppression (Table 1). The compounds held small and comparable molecular weights between 141.17 and 489.65 Da, within the typically known size ranges (<500 Da) for efficient diffusion through BBB, and ranged widely in hydrophobicity from 21.27 to 5.15 logPo/w. The com- pounds were provided in pure powder substances and were re-constituted in dimethylsiloxane (DMSO), etha- nol, or methanol below the maximum solubility. The re-constituted solutions were kept at 220 °C away from direct light to prepare stock solutions.
Cell Culture
Two immortalized cell lines were utilized for this permeability study, which have been widely accepted for BBB co-culture models, including the brain endothelial cell line bEnd.3,29,35,52 and the glial cell line C6,31,32 each of which was respectively derived from a mouse and a rat and was obtained from the ATCC. Both the cell lines were cultivated in DMEM/F12 media (Lonza) supple- mented with 10% fetal bovine serum (Hyclone), with 1% Penicillin/Streptomycin and Amphotericin B (EMD) for contamination control. Media was buffered to 7.35 in all cases, and the cells for seeding experiments were taken from recently confluent sub-cultures only.
The cultivated cells were seeded in 6-well transwells in both the mono-culture and co-culture, serving as static control for comparison to dynamic experiments with the lBBB chips. For high adhesion cell seeding, both the transwells and the lBBB chips were coated overnight with poly-lysine (100 lg/mL) and Collagen IV/Fibronectin (100 lg/mL each) respectively to facilitate attachment of C6 and bEnd.3 cells. C6 astrocytes were seeded first on the underside of the membrane in both models at a density of 6 9 104 cells/ cm2. After 2 days, bEnd.3 cells were seeded on the top surface of the membrane at the same density. The seeded cells reached stable confluence on the day 6 for co-culture and day 4 for mono-culture, when perme- ability experiments were performed.
Primary brain endothelial cells were extracted from Sprague–Dawley rats_ENREF_27 for morphological comparisons to bEnd.3 cells.30 Rats were euthanized with CO2, and forebrains were removed, diced, and digested in 1 mg/mL collagenase II and 15 lg/mL DNAse I in DMEM with 50 lg/mL gentamycin for
1.5 h at 37 °C under 250 RPM rotation. Following centrifugation in 20% BSA for 20 min at 1000 g, the re-suspended pellet was again digested in 1 mg/mL collagenase-dispase and 6.7 lg/mL DNAse I in DMEM for 1 h at 37 °C under 200 RPM rotation. Following the 2nd centrifugation for 20 min at 1000 g, the further digested cells were separated on a 33% Percoll gradient and centrifuged a 3rd time for 10 min at 1000 g. Finally, isolated brain endothelial cells were plated for 2 days in DMEM supplemented with 10% plasma-derived ser- um, 1.5 ng/mL bFGF, 5 lg/mL insulin, 5 lg/mL transferrin, 5 ng/mL sodium selenite, 50 lg/mL genta- mycin, and 4 lg/mL puromycin. 500 nM hydrocorti- sone was added on day 1 to improve the tightness of the cell–cell junctions; and puromycin, added initially to selectively kill non-endothelial cells, was removed on day 3; and cells were imaged on day 5. All supplements were obtained from Sigma-Aldrich.
TESTING METHODOLOGY
Fluorescent Imaging of Endothelial Cell Morphology
To compare morphological properties of the bEnd.3 cell line to primary brain endothelial cells, the presence of tight junctions were visualized through
immunostaining of ZO-1 expression, and all cell images were obtained in the models without electrodes. Fol- lowing rinsing with phosphate buffered saline (PBS), confluent cells were fixed with 4% paraformaldehyde (Avantor) for 10 min at room temperature. Cell mem- branes were permeabilized with 0.1% Triton X-100 in PBS for 10 min and blocked with 5% goat serum in permeabilization buffer for 1 h. Cultures were incubated with primary mouse anti-ZO-1 antibody (Santa Cruz) overnight at 4 °C, then incubated with secondary Alexa-fluor goat anti-mouse antibody (Invitrogen) for 1 h at room temperature. Following counterstaining with 1 lg/mL 4¢,6-diamidino-2-phenylindole (DAPI, Enzo), the cells were imaged with a Nikon microscope.
Dynamic Flow Experiments
The lBBB chips were prepared for permeability experiments under fluid flow in a similar process to that previously described.5,6 The fabricated lBBB chips were sterilized with 70% EtOH, and connected to a 205S cartridge pump (Watson–Marlow) via gas- permeable marprene tubing, and the entire setup, including the pump, chips, and media reservoirs, was placed in a humid incubator (37 °C, 5% CO2). Cells were seeded as described in the previous section. Chips were perfused at ~10 lL/min until day 2 after bEnd.3 seeding, and flow on the luminal side was increased to 2 mL/min, providing a level of WSS relevant to the BBB (3–20 dyn/cm2).18 WSS (sA) applied to the endothelial cells was ~15 dyn/cm2 according to the equation for WSS in a rectangular channel where Q is flow, l is dynamic viscosity of the media (0.012 dyn s/cm2), and h and w are the channel height and width, respectively. The high aspect ratio (20:1) of the channel ensures that most of the cells experience a uniform WSS, except the high-drag regions near the side-walls.6
Cytotoxicity Testing
To evaluate cytotoxicity levels of each drug and to establish permeability assay concentration limit, lac- tate dehydrogenase (LDH) assay (Pierce) was per- formed on bEnd.3 cells following exposure to various concentrations of each drug for 24 h. Cells that undergo apoptosis release LDH in proportion to the level of toxicity, and the LDH assay reaction produces red formazan that can be measured via absorbance readings. To accurately quantify cytotoxicity, conflu- ent cultures of bEnd.3 cells with a constant cell seeding number (9000/well) were respectively exposed to 1, 10,100, and 1000 lM drug concentrations in DMEM/F12 media with controlled pH of 7.35 for 24 h in a 96-well plate. Negative controls (untreated) and positive con- trols (lysis buffer) were added to each plate to establish a basepoint and maximum levels for LDH release, respectively. Following the LDH assay, absorbance values were recorded at wavelengths of 490 and 680 nm. LDH activity was derived from these mea- surements according to: LDH = abs(490 nm) — abs (680 nm), (2) yielding responses proportional to the LDH expression in the cells. Results were reported as ‘‘% toxicity’’, as the ratio between sample LDH response and positive control LDH response (100%).
Trans-Endothelial Electrical Resistance (TEER)
To evaluate layer contiguity of the bEnd.3 cells, and to act as quality control for prepared BBB cultures, TEER was monitored prior to and following each permeability measurement. For the transwell-based static model, the transwells with seeded cells were moved to an EndOhm chamber (WPI) connected to an EVOM2 epithelial voltohmeter (WPI). For the lBBB dynamic model, the outputs of voltages and currents from the embedded electrodes were connected through an electrode adapter (WPI) to an EVOM2 epithelial voltohmeter. To calculate TEER, initial background resistances of the blank chip/membrane Rblank were subtracted from the measured resistance. The differ- ence value was multiplied with the cell culture area A (0.16 or 4.67 cm2 for the chips and transwells, respec- tively), resulting in the TEER values of only the endothelial cells in X cm2 from the following equa- tion.29
Drug Permeability
To measure the BBB permeability of each drug, aforementioned seven drugs in various concentrations were perfused through the dynamic BBB models (Fig. 2b) or pipetted in the static models, and con- centration change on the receiving side was measured. To minimize the presence of media serum components in downstream LC–MS measurements, drugs were diluted in PBS. Each drug was injected into the luminal (top) side of the membrane dissolved in PBS with pH of 7.4. The permeated drugs were collected at the abluminal side while being perfused with pure PBS. Transwell permeability assays were conducted for 20 min, and the 1.5 mL abluminal fluid was collected.
Dynamic permeability assays for all replicates were conducted simultaneously at 10 lL/min, with assay times of 20 min, yielding 200 lL of sample. All the collected samples were stored at 220 °C until con- centration measurements. The permeability coefficient P was obtained from the measurement following the equation:The measured endothelial permeability Pe repre- sents the rate of drug diffusion through the BBB layer, and was compared to the in vivo clearance from literature. For this correlation with in vivo results, measured in vitro Pe values were compared to brain/ plasma ratios (B/P or Kp). A drug’s B/P is an optimal metric for correlation with BBB permeability, because B/P calculation is typically part of the standard pharmacokinetic (PK) profiling of compounds,1 and was thus consistently available for the drugs used in this study.34,40,44,45,48,50 These values were calculated from the brain and plasma area under the curves (AUCs) from time/concentration profiles of in vivo biodistribution studies.45 For consistency, all refer- enced B/P values are from PK studies in the rat (Sprague–Dawley), with the exception of Sunitinib (mouse, strain unspecified) and PF-3084014 (guinea pig).
Sample Compound Quantification (HPLC–UV/ LC–MS)
To accurately quantify the concentrations from the tested samples, their quantitative concentrations (molarity) were measured by performing analysis with liquid chromatography (LC) and mass spectrometry (MS). First, the samples were mixed with mobile phases of either 0.1% formic acid in H2O or 0.1% formic acid in acetonitrile (ACN). The samples in the mobile phases were injected into a coupled LC–MS equipment setup: Agilent 1290 LC system, Agilent 6550 iFunnel Q-TOF mass spectrometer. The LC sys- tem utilized an Agilent Eclipse reversed-phase C-18 column. The separated compositions of the samples were then fed into the MS system where the samples were ionized (positive ion mode) through electro-spray ionization for detection. All the ionized compounds were detected as [M+H] + species, intact compounds plus one proton. The obtained mass spectra were analyzed with Agilent Mass Hunter Quantitative Analysis software, where the AUC were calculated and utilized for quantification. Due to the structural closeness to an amino acid, the quantification of Gabapentin was performed utilizing a high-perfor- mance liquid chromatography (HPLC) method.24 The HPLC-based protocol utilized the proportional fluo- rescence emission from o-phthalaldehyde (OPA) to the amount of Gabapentin under the incident light with wavelengths of 340/455 nm. The protocol analyzed the samples containing Gabapentin through reverse phase analysis utilizing an Insertsil PH 5 lm column (4.0 9 150 mm2, GL Sciences) in an Agilent 1100 HPLC system. The samples were mixed with mobile phases that contain Buffer A (40 mM NaPhosphate, pH 7.8) and Buffer B (ACN:MeOH:H2O, 45:45:10).
The samples were first mixed with fluoraldehyde OPA reagent in a ratio of 1:2 in the auto-injector system prior to the column injection. The mobile phase con- sisted of a gradient elution from 0% buffer B/100% buffer A to 65% buffer B/35% buffer A over 20 min, and the detector was set at the absorption wavelength of 338 nm.
RESULTS AND DISCUSSION
Chromatographic Analysis
To establish defined linear ranges for chromato- graphic quantitation methods, standard solutions of each drug in known concentrations in PBS (pH buf- fered 7.4) were analyzed to provide an interpolation standard for sample quantification (Fig. 3). The resultant absorbance curves defined very highly linear ranges (R2 > 0.99) and were utilized to interpolate concentrations of each sample (Fig. 3). In cases where a sample was found to be above the defined linear range, samples were diluted and run again to ensure accurate quantification.
We believe that the effects of serum components on downstream measurements were negligible. Downstream LC–MS measurements of each of these compounds in serum/plasma have been conducted in previous studies by several research groups resulting in negligible serum component interactions. Furthermore, the presence of media serum components was diluted by using phosphate-buffered saline (PBS) during permeability assays.
FIGURE 3. Linear standard curves for chromatographic detection. Standards were used to define a linear range, and as a quantitative standard for interpolation of sample results. Samples which were measured to fall above the defined linear range were diluted and re-run to ensure accurate interpolation. All standard curves defined a linear range with R2 of higher than 0.99. Analysis methods for each drug were LC–MS, except for Gabapentin, which was HPLC–UV.
Morphology
The optical images of the immortalized bend.3 cells, obtained on day 4 of endothelial culture did not show any significant dissimilarities in morphology from those of the primary rat brain endothelial cells (Fig. 4). Particularly, cells from both sources clearly showed full confluence of highly elongated cells and strong tight junction expression of ZO-1 (green) among all adjacent cells. The images also showed that both the cell groups held comparable sizes of their highly elongated shape, typically ranging in 10–30 lm width, and 50–80 lm length. Since the monolayer of endo- thelial cells mainly determines the BBB permeability, such similarities in morphology validate the use of bEnd.3 cells to examine the diffusion properties of the BBB, such as TEER and permeability. As the zonal occludins are localized exclusively at the interface between cell membranes and tight junctions,49 we suspect that the background in these images are of secondary antibody, either non-specifically bound or in unbound globules. Despite this background, the dis- tinct boundaries where tight junctions are expressed are sufficiently clear from these images to conclude expression of tight junctions and observe cell shapes.
Cytotoxicity
The LDH measurement results (Fig. 5) showed that all the seven drug compounds did not cause toxic effects to the brain endothelial cells in the dynamic BBB model up to 10 lM, defining the maximum range of testing in this study. Four compounds (Traxoprodil, Gabapentin, Ethosuximide, and Varenicline) did not induce any increased LDH expression over the nega- tive control at any measured concentrations between 10 lM and 1 mM. Compound PF-3084014 induced toxic response of 39% of positive control at high concentrations of 1000 lM or higher, while Sertraline and Sunitinib respectively induced toxic response of 62% and 64% at the concentrations of 10 lM or higher. Thus, these corresponding maximum accept- able concentration ranges were selected for the per- meability assays to avoid the unrelated errors such as the loss of functioning cells due to toxicity. Only the selected concentrations were used as CL during per- meability assays. Note that some in vivo toxicological information was available from MSDS documentation for the seven drugs utilized in this study. However, these toxicological tests, such as LD50 (50% lethal dose) represent the potential concentrations to poison the individual entity of animals, thus not providing direct translation to cytotoxicity in the cellular envi- ronments concerned by this study.
Trans-Endothelial Electrical Resistance
For this study, cells were cultured in the device only until they had reached confluence and steady-state TEER was reached. As a result, cell cultures have been maintained in the device only for up to 7 days, during which any observed changes in electrode performance or background resistance potentially caused by protein fouling was not observed. TEER measurement results (Fig. 6) showed that the change in average TEER before and after permeability was within 5 X cm2 for all model conditions, and that outliers were observed with occurrence rates of 12.5 and 37.5% in 8 cases, respectively for the dynamic BBB model and the static Transwell model, with an average standard deviation of 19% of the total steady-state TEER values. The outliers were defined as when data points were deviated from the average by more than 29 standard devia- tions. It was hypothesized that outliers were caused by pinholes in the bEnd.3 cell layer or by apoptosis due to cytotoxicity by the tested drugs. The TEER measure- ment results were utilized for quality pre-control for permeability assays by enabling the exclusion of outliers.
FIGURE 5. Cytotoxicity of each drug tested in this study as measured by LDH expression following 24 h exposure to dif- ferent concentrations. Data is reported as a ratio to LDH levels expressed by cells exposed to lysis buffer (100% toxicity). Also included is the negative control, or the LDH expression of untreated cells, indicating a baseline of negligible cyto- toxicity. Standard deviations displayed with error bars. Con- ditions significantly higher than negative control denoted with *. All n 5 4.
FIGURE 4. Immunostaining of the brain endothelial cell line bEnd.3 cell line used for the BBB models in this study (a) and extracted primary brain endothelial cells from the rat (b) for reference. Cells were fixed and permeabilized, and stained with antibodies targeting the ZO-1 and conjugated to Alexa- fluor 488 (green), and counterstained with DAPI nuclear stain (blue). Cell morphologies were qualitatively similar, with strong expression of tight junctions as indicated by ZO-1 expression, and with similar shape and size, suggesting cor- relation with the in vivo physiology.
As shown previously,5 co-culturing bEnd.3 cells with astrocytes results in significantly elevated TEER levels of BBB models in both static (~90% increase) and dynamic (~25% increase) conditions. These TEER levels indicate more fully contiguous cell layers and more strongly expressed tight junctions,21 though they are not indicative of cell transcytotic activity,16 which is the primary path for compounds that cannot pass through tight junctions.
Finally, TEER levels were measured to be signifi- cantly higher for dynamic lBBB compared with static transwells for both co-cultured (5.9 fold increase) and mono-cultured (8.9 fold increase) BBB models. A consensus has been reached regarding BBB models that a minimum TEER level of 150 X cm2 is required for BBB models to achieve reasonably representative and consistent permeability characteristics,47 and this threshold was consistently reached for both mono- cultured (223 X cm2) and co-cultured (280 X cm2) embodiments of the dynamic lBBB model, but not for their static transwell counterparts (47 and 25 X cm2). This indicates that the dynamic model represents a significant improvement in terms of monolayer tightness.
Drug Permeability
The feasibility of the BBB model as a predictive platform for drug screening was tested with peremability measurement of the seven drugs (Table 2; Fig. 7). Comparison of the logPe averages across all seven drugs appear to suggest two trends in permeability: (1) lower permeability in co-cultured models than in mono-cultured models, and (2) lower permeability through dynamic lBBB models than static transwells. First, the drug logPe coefficients of mono-cultured models were, on average, 0.063 and 0.061 log(cm/s) lower than for co-cultured models in static and dynamic conditions, respectively, while the average LogPe coefficients were lower in the dynamic in vitro BBB models than static Transwell models by 0.050 and 0.052 log(cm/s) for co-cultured and mono- cultured models, respectively (Fig. 8). These trends indicate that optimal model conditions in regards to meability coefficients (logPe) between static and dynamic models, where dynamic logPe corresponds to the x-axis, and static logPe corresponds to the y-axis. In the case of both co-culture (a) and mono-culture (b) versions of the models, drug logPe of static BBB models with otherwise similar cul- ture conditions were higher than their corresponding logPe (dotted line), with an average offset of 0.050 and 0.052 for co-cultured and mono-cultured versions of the model, respectively. These results indicate that dynamic models provide higher barrier activity and better model performance, in agreement with the higher model TEER values. 95% confi- dence limits for the linear regression are displayed.
FIGURE 6. TEER levels of prepared BBB models, 4 days after endothelial cell seeding as quality control. TEER was measured before and after permeability assays, and outliers were selected and removed as unacceptable for permeability testing. TEER for co-cultures (a, c) were significantly higher than for mono-cultures (b, d), and TEER for the lBBB cultures (a, b) were significantly higher than static transwells (c, d). Standard deviations displayed with error bars. All n 5 8.
FIGURE 7. Permeability coefficients of each compound used in the study. Data is presented in Fig. 5 as logPe according to convention. Dynamic conditions significantly different than transwell controls denoted with *. All n 5 4.
For all model conditions, there was a strong corre- lation with in vivo B/P ratios with linear regression accuracy of R2 > 0.85. As the B/P ratio increased from 0.42 to 26.8, the corresponding average logPe values of each drug proportionally increased from 24.06 to 23.63 log(cm/s) (Fig. 9). Though multiple-drug cor- relation of brain clearance results between transwell- based in vitro models and in vivo animal models has been shown previously,39,53 this reports the first dem- onstration of in vivo correlation for pharmaceutical drug clearance in a dynamic microfluidic model of the BBB. These confirmed correlation results, in addi- tion to the practical advantages of the lBBB (high-throughput, material conservation, integrated sensing, controlled delivery), demonstrate that micro- fluidic models are a promising tool for pharmaceutical drug screening.
FIGURE 9. In vivo correlation of average permeability coef- ficients (see Table 2). Data is displayed as logPe according to convention. Brain/plasma ratios were referenced from literature.34,40,44,45,48,50 In the case of both co-culture (a) and mono-culture (b) versions of the models, permeabilities were consistently lower for dynamic lBBB than static transwells, indicating increased barrier function. All cases showed a highly correlated positive trend with brain/plasma ratio, indi- cating that the BBB model is feasible for prediction of in vivo brain clearance.
No correlative trend was deterministically exhibited between permeability profiles or B/P and logPo/w (octanol/water coefficient) or molecular weight, implying the potential influence of other physico- chemical properties on diffusive properties of the sample drugs through the BBB. The lack of public information on some physicochemical properties of the tested compounds as well as requirement of a larger dataset of compounds tested in this study has currently limited more comprehensive, multi-descriptor quanti- tative structure–activity response (QSAR) analysis.54 However, within the limited testing, Sertraline, which exhibited the best BBB clearance, showed both the highest logPo/w (octanol/water coefficient), thus the highest lipophilicity, and the highest logPch (alkane/ water coefficient), indicating a lack of polar interac- tions,17 likely explaining its comparatively excellent brain penetration in the test group. This is because capacity factors (polar interactions per surface unit) have been shown a significant decreasing correlation with BBB permeability of compounds.13
CONCLUSIONS
We have demonstrated the permeability analysis of neuroactive drugs and correlation with in vivo brain/plasma ratios in a dynamic microfluidic BBB model. Seven neuroactive drugs, including Ethosuxi- mide, Gabapentin, Sertraline, Sunitinib, Traxoprodil, Varenicline, PF-3084014, were analyzed in terms of TEER and permeability in both dynamic (microfluidic) and static (transwell) BBB models either with brain endothelial cell line bEnd.3 in monoculture, or in co-culture with glial cell line C6. For all seven drugs, dynamic and co-culture models respectively resulted in lower permeability, and significantly higher TEER, than static and mono-culture models, providing the justification for the dynamic co-culture microfluidic BBB model utilized in this study. Correlation of the resultant logPe values [ranging from 24.06 to 23.63 log(cm/s)] with in vivo brain/plasma ratios (ranging from 0.42 to 26.8) showed highly linear correlation (R2 > 0.85) for all model conditions, indicating the feasibility of the dynamic microfluidic BBB model for prediction of BBB clearance of pharmaceuticals. Within our knowledge, Nirogacestat this is the first reported drug clearance study in a microfluidic BBB model.