Microfluidic Blood-Brain Barrier Model Provides In Vivo-Like Barrier Properties for Drug Permeability Screening

Abstract

Efficient delivery of therapeutics across the neuroprotective blood-brain barrier (BBB) remains a formidable challenge for central nervous system drug development. High-fidelity in vitro models of the BBB could facilitate effective early screening of drug candidates targeting the brain. In this study, we developed a microfluidic BBB model that is capable of mimicking in vivo BBB characteristics for a prolonged period and allows for reliable in vitro drug permeability studies under recirculating perfusion. We derived brain microvascular endothelial cells (BMECs) from human induced pluripotent stem cells (hiPSCs) and cocultured them with rat primary astrocytes on the two sides of a porous membrane on a pumpless microfluidic platform for up to 10 days. The microfluidic system was designed based on the blood residence time in human brain tissues, allowing for medium recirculation at physiologically relevant perfusion rates with no pumps or external tubing meanwhile minimizing wall shear stress to test whether shear stress is required for in vivo-like barrier properties in a microfluidic BBB model. This BBB-on-a-chip model achieved significant barrier integrity as evident by continuous tight junction formation and in vivo-like values of trans-endothelial electrical resistance (TEER). The TEER levels peaked above 4000 Ω·cm² on day 3 on chip and were sustained above 2000 Ω·cm² up to 10 days, which are the highest sustained TEER values reported in a microfluidic model. We evaluated the capacity of our microfluidic BBB model to be used for drug permeability studies using large molecules (FITC-dextrans) and model drugs (caffeine, cimetidine, and doxorubicin). Our analyses demonstrated that the permeability coefficients measured using our model were comparable to in vivo values. Our BBB-on-a-chip model closely mimics physiological BBB barrier functions and will be a valuable tool for screening of drug candidates. The residence time based design of a microfluidic platform will enable integration with other organ modules to simulate multi-organ interactions on drug response.

Graphical abstract

Introduction

The blood brain barrier (BBB), mainly composed of highly specialized endothelial cells lining the cerebral capillaries and overlying astrocytic foot processes, forms a dynamic physical and metabolic barrier that strictly regulates molecular exchange between the blood and the brain. Inadequate brain penetration of therapeutic candidates across the BBB remains a major cause of failure in drug development for the majority of central neural system (CNS) disorders, including brain cancer, stroke, autism and Alzheimer’s disease (Pardridge, 2005). Development of high-fidelity in vitro models of the BBB will increase the efficiency of brain drug permeability screening and boost future growth of neurotherapeutics.

Most published in vitro models of the BBB are based on the transwell culture systems (Wolff et al., 2015), which have made it simple to setup cocultures of multiple neurovascular cells within the conventional cell culture dish platform, and thus allows moderate-high throughput permeability screening. Yet the large fluid-to- cell volume ratio and the static fluid condition in transwell-based models lead to uncontrollable and unstable biochemical gradients in the cell surroundings. Such models are thus limited in recapitulating the in vivo metabolism of the BBB as well as maintaining its phenotype. Novel BBB models recreating in vivo neurovascular microenvironment and representing BBB properties more completely are desired for high-content screening.

More recently, due to the prohibitive expense of current animal studies and their poor predictive power for human responses to drugs, there is an increasing trend in the pharmaceutical industry and research moving towards advanced biomimetic in vitro models of the human body, with the aim of replacing or reducing animal models. A series of global efforts have been initiated to build the next generation drug testing tool, tissue-engineered human cell-based multi-organ models, so-called Body-on-a-Chip (BOC) (Bhatia and Ingber, 2014; Esch et al., 2011; Frey et al., 2014; Maschmeyer et al., 2015; Sung et al., 2013; Sutherland et al., 2013). Integration of functional organ modules onto one single chip allows us to simulate multi-organ interactions and study the dynamics of drug activities and whole body response in vitro. Integration at this level requires transferring various organ models, including the BBB, onto more integratable platforms than traditional culture dish ones.

The emerging microfluidics-based BBB models have intrinsic advantages of precise control over microscale fluid delivery and versatile integration capabilities. Transferring BBB models onto microfluidic platforms allows for well-defined and controllable extracellular microenvironment, in situ functional analyses, and future integration into multi-organ or whole-body microsystems. Several proof-of-concept microfluidic BBB models have been proposed (Booth and Kim, 2012; Cho et al., 2015; Cucullo et al., 2011; Griep et al., 2013; Kim et al., 2015; Prabhakarpandian et al., 2013; Walter et al., 2016; Yeon et al., 2012). They have demonstrated continuous luminal perfusion, on-line monitoring of trans-endothelial electrical resistance (TEER) as a measure for the barrier tightness, and live cell imaging for permeability studies. Improved barrier integrity was also observed with laminar flow perfusion, as compared to the static controls. Novel microfluidic platforms that include microglia (Achyuta et al., 2013), pericytes and neurons (Brown et al., 2015a) in addition to the BBB, and combine 3D cell culture (Brown et al., 2015b) to recreate the neurovascular unit hold the potential to study the mutual interaction between the BBB and CNS.

In an effort to achieve in vivo-like BBB phenotype on chip, flow-induced shear stress has been exploited as a dominant design paradigm in microfluidic BBB models. Several on-chip models have shown that shear stress induces a 3–20 fold increase in TEER values as compared to static conditions (Booth and Kim, 2012; Griep et al., 2013). It is worth noting that the highest TEER values achieved in those models were still 1 to 2 orders of magnitude lower than the in vivo values (Wolff et al., 2015). Meanwhile, several recent transwell-based BBB models have presented significantly higher TEER values within the range of in vivo levels (Cantrill et al., 2012; Lippmann et al., 2012; Lippmann et al., 2014), suggesting that shear stress is not necessarily required for the establishment of in vivo-like barrier tightness in BBB models. More importantly, human BMECs are highly specialized that form 50 to 100 fold tighter barriers (Abbott, 2002) and respond to shear stress differently compared to peripheral microvascular endothelial cells, such as human umbilical vein endothelial cells (HUVECs). HUVECs typically transform into spindle-like morphology in response to shear stress and align in the direction of the flow (Chien, 2007). Such elongation and alignment have also been observed in microfluidic BBB models with low TEER values (Booth and Kim, 2014; Walter et al., 2016). In contrast, mature HBMECs retain cobble-stone morphology and resist elongation and alignment under shear stress (Reinitz et al., 2015; Ye et al., 2014). Although shear stress has been shown to modulate barrier tightness for HUVECs (Seebach et al., 2007) and some microfluidic BBB models at low levels, given the distinct responses of HBMECs to shear stress, it is uncertain whether shear stress would be effective in modulating HBMEC barrier tightness to achieve and maintain high levels of TEER (>1500 Ω·cm²). We hypothesized that a microfluidic BBB model can achieve and maintain in vivo-like BBB properties without applying shear stress.

Major hurdles to establishing an authentic microfluidic model of human BBB are the lack of high-fidelity human brain microvascular endothelial cells (BMECs) and the difficulties of using human cells and establishing their BBB phenotype in sealed microfluidic devices. Most microfluidic BBB models use human (Griep et al., 2013; Walter et al., 2016) or murine (b.End 3, Booth and Kim, 2012; Kim et al., 2015; RBE4, Achyuta et al., 2013; Cho et al., 2015; Prabhakarpandian et al., 2013) brain endothelial cell lines, probably due to the simplicity and scalability of culture. Some utilized primary human BMECs (Brown et al., 2015b; Cucullo et al., 2011). Immobilized brain endothelial cells often exhibit poor barrier properties (Weksler, 2005; Wolff et al., 2015) and inevitable species difference in terms of barrier tightness, transport and metabolic activities (Shawahna et al., 2013), while primary BMECs may lose many of their BBB characteristics quickly in culture (Shawahna et al., 2013), and the cell number is often limited by tissue availability. Besides, it is always challenging for microfluidic devices to use human cells, which are delicate and sensitive to agitation, while preserving their phenotype. Microfluidic cell seeding into sealed devices is challenging and often inefficient with regards to cell viability and placement (Young and Beebe, 2010), both of which are critical to establishing the BBB phenotype.

To address the limitations of current models, meet the requirement for future integration and test our hypothesis, we developed a pumpless microfluidic BBB model that has been demonstrated to mimic in vivo BBB integrity and permeability with minimized shear stress applied. Our model utilized human BMECs derived from an expandable source of human induced-pluripotent stem cells (iPSCs) using a biased spontaneous differentiation approach developed by Shusta and Palecek groups (Lippmann et al., 2012; Lippmann et al., 2014). The microfluidic platform consists of a cell insert and a microfluidic housing, by which we were able to prepare BBB constructs in traditional culture dish setup thus ensure efficient cell seeding and minimize agitation. The microfluidic housing was designed to provide continuous perfusion while minimize the wall shear stress. It also involves integrated electrodes for on-line TEER monitoring. The model operates based on matching the blood residence time in the human brain, allowing for physiologically realistic transport of nutrients and exogenous substances across the BBB without the need for an external pump or tubing. This model can easily be adapted for further integration onto multiple-organ-on-a-chip platforms.

Materials and Methods

1). System construction

The BBB-on-a-chip system (BBBoC) consists of a stand-alone insert that carries BBB constructs and a microfluidic culture platform that provides physiologically relevant luminal perfusion and supports BBB construct maintenance.

The microfluidic platform is composed of three layers (Fig. 1A): (i) a chamber layer that holds two reservoirs with a total volume of 450 μl and a 6.5 mm diameter neuronal chamber at the center, (ii) a medium perfusion layer that accommodates the BBB cell insert in a pocket and has 4 parallel microchannels of 300 μm × 160 μm (width × height) connecting the luminal chamber with the reservoirs, and (iii) a lid layer that covers the reservoirs and the neuronal chamber, allowing for gas exchange while minimizing medium evaporation. Both the lid and the perfusion layer have a central circular opening housing the electrodes. All three layers were designed in Inventor (Autodesk Inc., San Rafael, California), and fabricated with a 3D object printer (Objet 30Pro, Stratasys Ltd., Rehovot, Israel) using stereolithography. Objet VeroClear photopolymer was chosen as the printing material for its rigidity and transparent appearance. 3D printed parts were then coated with a 4 μm thick conformal layer of parylene-C using a vapor deposition system (PDS-2010 LABCOTER® 2, Specialty Coating Systems, Indianapolis, IN) as described previously (Esch et al., 2015) to ensure chemical resistance and biocompatibility.

Figure 1.

Design of the BBB-on-a-Chip. (A) Schematic exploded view of the microfluidic platform. The device consists of a cell insert and three 3D printed plastic layers: i) a bottom perfusion layer with microchannels and bottom electrodes, ii) a middle layer that forms reservoirs and the neuronal chamber, and iii) a top lid layer with top electrodes that covers the neuronal chamber and the reservoirs minimizing evaporation. The cell insert was made from two silicone sheets and a sandwiched porous polycarbonate membrane, and was assembled between the bottom and the middle layers. (B) The assembled device, with or without the lid. A red dye was used for visualization of microchannels, the neuronal chamber and reservoirs. (C) Side view showing the fluid pathway, electrode wiring connected to a Millicell-ERS Volt-Ohm Meter and the BBB co-cultural orientation. The zoom-in panel showing the cross section was drawn to scale except for cells and the porous membrane. A step chamber with a height of h_sc was introduced to minimize shear stress on the BMEC surface.

The top and bottom electrodes were both made of a 0.8 mm diameter Ag/AgCl pellet electrode (A-M systems, Sequim, WA), an electrical insulating PEEK tubing (1 mm ID, 1.8 mm OD, IDEX Health & Science, Oak Harbor, WA), and a silver tubing (2 mm ID, 2.5 mm OD, Otto Frei, Oakland, CA). They were concentrically assembled and installed in the housings on the lid and the perfusion layer (Fig. 1A). Liquid polydimethylsiloxane (PDMS) pre-polymer mixed with curing agent (Sylgard 184, Dow Corning, Corning, NY) at a 10:1 (w/w) ratio was applied to fill gaps and was cured at 50°C for 24 h to hold the electrodes in place and ensure leakage-free installation. All 3D printed parts and electrodes were sterilized in 70% ethanol before use.

The cell insert (Fig. 1A) accommodating cocultures of BMECs and astrocytes was constructed from two 0.5 mm thick silicone sheets (Grace Bio-Labs, Bend, OR) and a sandwiched porous polycarbonate membrane with 0.4 μm pore size (Whatman Cyclopore, GE Healthcare, Wilkes-Barre, PA). The silicone sheets were patterned with a 6.5 mm diameter circular cutout in the center and were irreversibly bonded with the polycarbonate membrane at room temperature using oxygen plasma activation and aminopropyltriethoxysilane modification as previously described (Abaci et al., 2015; Sunkara et al., 2011). The inserts were autoclaved before use for cell culture.

2). Derivation of BMECs from human iPSCs

Human iPSCs IMR90–4 (WiCell, Madison, WI; Yu et al., 2007) were maintained and differentiated to brain microvascular endothelial cells (BMECs) based on the protocol developed by Lippmann et al. (2014). Briefly, iPSCs were expanded on Matrigel (Corning Inc., Corning, NY) in mTeSR1 medium (STEMCELL Technologies, Vancouver, BC, Canada) till colonies approached borders of their neighbors. Cells were then switched to a differentiation medium DMEM/F12 medium with HEPES containing 20% KnockOut Serum Replacement, 1× MEM Non-Essential Amino Acids Solution, 1 mM L-glutamine and 0.1 mM β-mercaptoethanol (Sigma, St. Louis, MO). After 6 days of differentiation, cells were switched to Human Endothelial Serum Free Medium supplemented with 1% Platelet Poor Derived Serum (Biomedical Technologies, Baltimore, MD), 20 ng/mL bFGF (R&D Systems, Minneapolis, MN), and 10 μM retinoic acid (Sigma), and cultured for two more days before passaging them for BBB cocultures. All media were replenished daily throughout the culture. All reagents not specified were from Life Technologies (Carlsbad, CA).

3). BBBoC assembly and operation

The cell inserts of BMEC-astrocyte cocultures were first prepared in culture dishes. The porous membranes of the inserts were coated on one side with a mixture of collagen IV (400 μg/mL) and fibronectin (100 μg/ mL) in DPBS, and incubated at 37 °C for at least 4 hours. The coating solution was then removed and the membranes were rinsed with DPBS. Rat primary astrocytes (Life Technologies) at passage 2 were seeded onto the non-coated side of the membranes and cultured in astrocyte growth medium (85% D-MEM with high glucose + 15% fetal bovine serum) for 48 h. Human iPSC-derived BMECs were then seeded onto the coated side of the membranes. Differentiated cells from a 35 mm diameter culture well (9.6 cm² culture area) could seed 12 collagen I/fibronectin coated cell inserts (4 cm² total). The cell inserts were then maintained in a coculture medium (Human Endothelial Serum Free Medium supplemented with 1% Platelet Poor Derived Serum (Biomedical Technologies), 10 μM retinoic acid, and 1× Penicillin-Streptomycin) for 12 h prior to transferring to the on-chip system. All reagents not specified were from Life Technologies or Sigma.

The BBBoC was assembled by sandwiching a cell-loaded insert between the chamber and the perfusion layers, with BMEC side facing the microchannels (Fig. 1A, C). The silicone frame also functioned as sealing gaskets between the two layers. Both reservoirs and the neuronal chamber were filled with coculture medium and capped with the lid. The assembled chip (Fig. 1B) was placed on a rocking platform (Next Advance, Averill Park, NY) tilted at ±12° with a quick tilt direction change every 30 s. The whole system was placed inside a 5% CO₂ cell culture incubator.

4). Fluid dynamics and computational simulation

The fluid flow in BBBoC is driven by gravity. The flow rate (Q) can be calculated from equation (1), where ΔP, R and L are the pressure drop, the hydrodynamic resistance and the distance between the inlet and the outlet, respectively; θ is the tilting angle of the rocker platform; ρ and g are fluid density and gravity constant. R comes largely from those four rectangular cross-section microchannels in parallel. The resistance of each channel (R_ch) can be estimated by equation (2), where μ is the dynamic fluid viscosity; l, w, and h are the length, width and height of the channel, respectively. For the desired Q, a set of parameters that fit equations (1) and (2) were selected in the draft design.

$$Q=\frac{\Delta P}{R}=\frac{\rho gL\text{sin}\theta }{R}$$ (1)

$$R_{ch}=\frac{12\mu l}{w{h}^3}{\left[1-\frac{192h}{{\pi }^{5}w}tanh\left(\frac{\pi w}{2h}\right)\right]}^{-1}$$ (2)

The fluid dynamics in BBBoC was simulated in 3D using COMSOL Multiphysics to optimize the microchannel design for desired perfusion rate and wall shear stress. The Laminar Flow interface was used. Gravity was applied as the only volume force. The steady state incompressible Navier-Stokes equations were used to model the fluid flow. The flow rate and the shear stress were derived from the velocity results. μ = 0.78×10⁻³ Pa·s was used for culture medium at 37°C (Wang et al., 2013).

5). TEER measurement and validation

TEER was measured daily using the 4-point probe method (Srinivasan et al., 2015). Voltage-sensing Ag/AgCl pellets and annular current electrodes were connected to a Millicell-ERS Volt-Ohm Meter (EMD Millipore, Billerica, MA; Fig. 1C). The ERS meter applies an AC square-wave current of ± 10 μA at 12.5 Hz and records the electrical resistance (R). That for a blank porous membrane coated with collagen- fibronectin (R_m) was also measured and subtracted from each recording. TEER values were normalized to unit area of the endothelium using equation (3),

$$TEER=\left(R-R_m\right)\times A,$$ (3)

where A is the endothelial culture area.

Comparison on the resistance readings was made between BBBoC electrode system and the traditional “chopstick” electrodes in transwell setting using a transwell-like calibration device (Supplementary Fig. SI B). The 3D printed calibration device was designed to accommodate the cell inserts from the BBBoC model while allowing for the same placement of the chopstick electrodes as in a transwell. The dimensional parameters which may affect current density distribution, such as the lower chamber diameter d and the clearance below the membrane h, were all kept exactly the same as a 24-well transwell. A series of cell inserts were coated with different hydrogels (collagen, gelatin and silicone hydrogels) to serve as testing samples with a range of transmembrane resistance. Silicone hydrogels with different crosslinking densities and ion permeability were prepared as described in Kim et al., 2008. Each cell insert was assembled and measured for transmembrane resistance in both the calibration device and the BBBoC using a Millicell-ERS Volt-Ohm Meter.

6). Immunofluorescence

Cells were analyzed by immunofluorescence staining for zonula occludens protein-1 (ZO-1), claudin-5 and actin filaments (F-actin) on the designated day of culture on chip. For claudin-5 staining, cells were fixed and permeabilized on chip in ice-cold methanol, blocked at room temperature in DPBS containing 40% goat serum and 0.1% Triton X-100 (40% DPBS-G, Sigma) for 30 min, and incubated with mouse anti-Cladudin-5 monoclonal antibody (1:50) in 40% DPBS-G at 4°C overnight. The samples were then washed, incubated with Alexa Fluor 488-labled goat anti-mouse secondary antibody (1:200) in 40% DPBS-G at room temperature for 1 hour. For ZO-1 and F-actin, staining was carried out at room temperature. Cells were fixed with 4% paraformaldehyde (Boston Bioproducts, Ashland, MA) for 10 min, washed with DPBS, permeabilized with 0.1% Triton X-100 in DPBS for 10 min, blocked with 5% bovine serum albumin (BSA, Sigma) and 0.1% Triton X-100 in DPBS for 1 hour, and then incubated with fluorescein isothiocyanate (FITC)-conjugated ZO-1 monoclonal antibody (10 μg/mL, Thermo Fisher Scientific, Waltham, MA) or Cruzfluor 555 conjugated phalloidin (1 μg/mL, Santa Cruz Biotechnology, Dallas, TX) in 1% BSA for 1 hour. All samples were mounted on slides with Fluoroshield™ with DAPI (Sigma) for nuclear counterstain. Images were captured with a Zeiss LSM 710 confocal microscope and analyzed in ImageJ. All reagents not specified were from Life Technologies.

7). Permeability studies

Permeability studies were performed after the TEER values reached plateau, typically at around day 4–6 of culture on chip, using large fluorescent tracers and three model drugs. Testing solutions include fluorescein isothiocyanate (FITC)-dextrans of 70kDa, 20kDa, and 4kDa at 0.5 mg/ml; caffeine at 10 μM; cimetidine at 1 μg/ml; and doxorubicin at 2.5 and 10 μM. They were added to the reservoirs while control culture medium filled the neuronal chambers. Both reservoirs and neuronal chambers were sealed with breathable polyurethane membranes (Sigma) to prevent evaporation. BBBoC were immediately put back on the rocker and the luminal perfusion resumed. Both abluminal and luminal solutions were collected at 10 min for caffeine, 2 hr for cimetidine, and 24 hr for doxorubicin and FITC-dextrans, to determine the amount of compounds transported across the BBB. All reagents not specified were from Sigma.

The sample concentrations of FITC-dextrans or doxorubicin were determined by quantifying the fluorescence levels (Ex = 492 nm, Em = 518 nm for FITC-dextrans; Ex = 480 nm, Em = 580 nm for doxorubicin) using a microplate spectrofluorometer (Molecular Devices, Sunnyvale, CA). Caffeine and cimetidine concentrations were measured at Cornell University Biotechnology Resource Center Proteomics and Mass Spectrometry Facility using liquid chromatography-tandem mass spectrometry (LC-MS/MS) with multiple reaction monitoring (MRM) mode. The apparent permeability (P_app) of a testing solute was calculated from equations (5)–(6) (Wong et al., 2013),

$$C_{al}=C_l\left(1-exp\left(-\frac{P_{app}A}{V_{al}}\text{t}\right)\right)$$ (4)

$$P_{app}=\frac{V_{al}{C}_{al}}{A{C}_{l}t},when\mathrm{t}\ll \frac{V_{al}}{A·P_{app}}$$ (5)

where V_al and C_al are the abluminal (neuronal) fluid volume and solute concentration, respectively; A is the endothelial culture area; C_l is the luminal solute concentration; and t is perfusion period. Permeability from a blank porous membrane with collagen-fibronectin coating (P_M) was also determined using the same approach and was used to calculate BBB permeability (P_BBB ) according to equation (6),

$$\frac{1}{P_{app}}=\frac{1}{P_{BBB}}+\frac{1}{P_M}$$ (6)

8). Statistical analysis

Data are presented as mean ± SEM. Multiple groups were analyzed by ANOVA or two groups compared using Student’s t -test (GraphPad Prism). P <0.05 was considered significant. Correlations between groups were assessed using Pearson’s correlation coefficient (r).

Results and discussion

1. Microfluidic system design and operation

The BBBoC microfluidic system is an integratable platform that was designed using a pumpless scheme, as previously described (Sung et al., 2010), to allow robust, long-term maintenance of BBB constructs for drug permeability studies. Fluids were delivered to the luminal compartment through microchannels via gravity driven flow achieved by placing BBBoC on a tilted rocking platform, and were recirculated between the two reservoirs by reversing the tilt direction periodically. By utilizing gravity as the driving force and thus eliminating the need for external pumps and tubing, we were able to run multiple BBBoC units (n = 6 to 12) in parallel on a single rocking platform for more than 10 days with no failure from air bubbles, a notorious issue that often occurs in microfluidic cell culture systems and is particularly devastating to endothelial cells.

To actually mimic the in vivo transport at the BBB interface, we designed the neuronal chamber and the microchannels using residence-time based scaling (Abaci and Shuler, 2015). The neuronal chamber volume and the luminal perfusion rate were proportionally scaled down (by a factor of 1/12350) from the tissue volume and the blood flow rate for a human adult brain (Graf et al., 2012) to achieve the same residence time as in vivo (Supplementary Table SI). Such residence time based design provides the base for future integration of our BBBoC with other organ modules to build Multi-Organ-on-a-Chip (MOC) and ultimately Body-on-a-Chip (BOC) systems.

2. Optimization of fluid dynamics for minimized wall shear stress

The reciprocating flow induced by gravitational force and periodic redirection of the rocking platform enables medium recirculation at the luminal side but also expose BMECs to oscillatory shear stress (OSS), which has been shown to affect aortic and umbilical vein endothelial cell functions (Chappell et al., 1998; Guo et al., 2007; De Keulenaer et al., 1998). It should be noted that brain capillary endothelial cells in vivo experience frequent flow fluctuations, prolonged stalls and even reversals of direction under physiological conditions (Kleinfeld et al., 1998). To test our hypothesis on the necessity of applying shear stress in establishing in-vivo like barrier properties in microfluidic BBB models, we aimed to minimize the wall shear stress on the cell surface. We also hypothesized that by reducing the magnitude of OSS that BMECs experience to below signal inducing levels, we could avoid potential adverse effects from OSS exposure while still providing physiologically realistic perfusion to the BBB constructs.

We proposed a “step chamber” design to minimize wall shear stress magnitude while providing the same desired flow rate. The porous membrane where BMECs reside was elevated above the fluid delivering channel plane by a small distance h_sc, the height of the step chamber (Fig. 1C). Finite element analysis of the fluid velocity with varying h_sc showed that over the range of h_sc from 0 to 0.5mm, there was only ± 2% variation in the resulting flow rate (53.54 ± 1.08, Mean ± S.D., n = 11). The introduction of the step chamber had little effects on the overall hydrodynamic resistance, and thus we were still able to maintain the desired perfusion rate. Meanwhile, both the average (τ_avg) and the maximum (τ_max) values of wall shear stress that BMECs experience dropped dramatically with increasing h_sc (Fig. 2A–B). The regions close to entrances and exits to the microchannels experienced the highest shear stress, whereas the central area showed uniform shear stress pattern with low magnitude (Fig. 2A). The introduction of a 0.5 mm thick step chamber reduced τ_avg by 94% from 0.250 to 0.014 dyne/cm², and τ_max by 95% from 1.800 to 0.023 dyne/cm², supressing both τ_avg and τ_max considerably below the signal induction levels (Jo et al., 1997; Olesen et al., 1988). In addition, simulated τ_avg showed a linear correlation (R² = 0.9973, P < 0.0001) with h^–2, where h is the sum of the microchannel height (h_microchannei) and h_sc (Fig. 2C). Although our system is not a typical parallel-plate flow system (PPFC), we found that with the current configuration with four evenly distributed microchannels connected to the luminal circular chamber, h and τ_avg follow equation (7), which is typically used for PPFC platforms.

Figure 2.

Simulation results for the shear stress on BMEC surface. The maximum wall shear stress occurred near the entrance/exit to the microchannels, while relatively lower shear stress levels were uniformly present near the center (A). Both the average and the maximum levels of shear stress on BMEC surface decreased dramatically when the step chamber height increased from 0 to 0.5 mm (A-B). The average shear stress levels showed an inversely proportional relationship with the step chamber height squared (C).

$${\tau }_{avg}=\frac{6\mu Q}{w}{h}^{-2}$$ (7)

where Q is the volumetric flow rate, and w is substituted with the chamber diameter. The linear regression slope (6.49×10⁻⁵ dyne) in Fig. 2C closely approaches the calculated coefficient (6.42×10⁻⁵ dyne) from equation (4). Therefore, we can easily determine the step chamber height for desired levels of average shear stress. We chose h_sc = 0.5 mm in the final design to allow effective convection and diffusion and to minimize the wall shear stress induced by the reciprocating flow.

3. Validation of on-line TEER measurement

BBBoC system is equipped with on-line electrodes to monitor the TEER of BMEC-astrocyte cocultures. In assembled devices, the top and bottom electrodes sit 3 mm and 0.66 mm, respectively, away from the culture membrane (Fig. 1C zoom-in). The relatively large clearance between the top electrodes and the culture membrane, approximately equivalent to the membrane radius (r = 3.25mm), and the circular current electrode design help create a relatively uniform electrical current density across the membrane. To investigate if the electrode system of the current design provides valid resistance readings, we compared our system with the two most commonly used electrode systems for transwell-based barrier models: the Endohm chamber and the “chopstick” electrodes (Supplementary Fig. SI A, Srinivasan et al., 2015). Our system adopted a similar electrode geometry as the Endohm system, which is considered to provide accurate and reproducible results consistent with those from a more sophisticated Ussing Chamber (Srinivasan et al., 2015). The chopstick electrode system has been criticized for the potential unequal distribution of current flow throughout the cell layer due to spatial constrains and shows 20% to 40% higher reading of resistance for large transwells (eg. 6-well transwells) compared to the Endohm chambers. However, it gives approximately the same readings for smaller transwells such as 12 well or 24-well transwells (with the same growth area as the cell insert in BBBoC) as the Endohm system, indicating that it creates reasonably uniform electrical field across the culture membrane when used for small transwells. We then made a comparison on the resistance readings between BBBoC electrode system and the chopstick electrodes in 24-well transwell setting using a transwell-like calibration device. The resistance data from the two devices showed excellent consistency with each other, with a linear correlation slope of 1.001 (Pearson r = 0.9989, P < 0.0001) and a lower baseline reading from BBBoC (Supplementary Fig. I C). These results demonstrate that the resistance readings from our BBBoC are accurate and valid.

4. Barrier integrity well established and maintained on chip

A prominent feature of the BBB phenotype is the unique tightness between BMECs that restricts paracellular transport across the BBB. Such barrier tightness is supported by an elaborated network of intercellular junctions, mainly the tight junctions. To obtain high-fidelity human BMECs with BBB properties, we took the approach of directing human iPSCs toward neuroectoderm via biased spontaneous differentiation and subsequently purifying the BBB-like endothelial population on a selective matrix as previously described (Lippmann et al., 2012; Lippmann et al., 2014). The resulting endothelial cells respond to astrocytic cues and acquire substantial attributes of the BBB (Lippmann et al., 2012; Lippmann et al., 2014). To analyze the barrier tightness of our BBBoC model, we examined the expression of tight junction (TJ) proteins: 1) claudins, a family of transmembrane proteins that provide the backbone of the TJs (Cooper et al., 2011), particularly, claudin-5, the isoform of claudins most commonly found in the BBB (Hewitt et al., 2006); 2) ZO-1, a membrane-associated cytoplasmic protein that is pivotal in maintaining TJ stability and functionality (Itallie et al., 2010) through interacting with the TJ transmembrane components and coupling to the actin cytoskeleton (Fanning et al., 1998). The immunostaining revealed a high-level protein expression of claudin-5 and ZO-1 on both day 3 and day 10 (Fig. 3A). The fluorescent images showed continuous networks of claudin-5 (red) and ZO-1 (green) outlining the contours of BMECs, indicating that well-organized tight junctions had formed as early as day 3 on chip, and such continuous cell-cell contacts were sustained at least until day 10 in the BBBoC model.

Figure 3.

BBB characteristic barrier integrity established and maintained on chip. (A) Cocultures of hiPSC-derived BMECs and astrocytes were maintained on chip and examined by immunostaining for the tight junction proteins, ZO-1 and claudin-5. Representative fluorescence images reveal continuous networks of claudin-5 (red) and ZO-1 (green) on both day 3 and day 10. Nuclear staining with DAPI in blue. Scale bar, 50 μm. (B) TEER values from BBBoC increased remarkably within 2–3 days on chip and sustained at high levels up to 10 days. Controls include BMEC (green) and astrocyte (red) monocultures. Values are means ± SEM; n = 10 – 16.

We further investigated if the tight junction networks we observed above rendered high TEER, a hallmark of BBB integrity (Srinivasan et al., 2015). The TEER was measured at a baseline level of 55 ± 6 Ω·cm² after 12 h in coculture with astrocytes and right after the constructs were transferred to the microfluidic platform. The TEER curve increased by nearly eighty fold to the peak level of 4399 ± 242 Ω·cm² within the first 48 h on chip, indicating quick tightening of BMEC cell-to-cell contacts. The TEER values then fell back slightly to a plateau of around 3000 Ω·cm². Such pullback has also been observed in transwell-based BBB using iPSC derived BMECs (Lippmann et al., 2012). The TEER then sustained at that high level for at least up to day 10 on chip (Fig. 3B, blue). By comparison, the TEER from BMEC monoculture only reached a peak of 368 ± 60 Ω-cm² (Fig. 3B, green), while the TEER for astrocyte monoculture showed slight increase with time in culture yet never reached 15 Ω·cm² (Fig. 3B, red). To our best knowledge, this is the first time that a microfluidic BBB model has achieved TEER levels within the range of reported in vivo values between 1500 and 8000 Ω·cm² (Wolff et al., 2015). The sustained TEER values in our system are nearly 10 – 20 fold of the highest values reported from the existing microfluidic BBB systems (Booth and Kim, 2012; Griep et al., 2013). Together, these results demonstrate that an in vivo-like barrier tightness was established and well maintained for a prolonged period in our BBBoC model.

5. Selective permeability to fluorescent tracers and drugs

We next evaluated the barrier function of BBBoC by investigating their permeability (P_BBB) to FITC-dextran tracers and small molecule drugs that are representative of different permeability (Avdeef, 2012) and transport mechanisms.

Various sized FITC-dextran tracers are commonly used in vivo and in vitro to assess paracellular permeability (Kuntz et al., 2014; Matter and Balda, 2003; Yao et al., 2014). In this study, we find that the porous polycarbonate membrane with matrix coating alone are highly permeable to 4K~70KDa FITC-dextrans with a permeability of 2~2.5×10⁻⁴ cm/s, while the BBB constructs built on the membrane were nearly impermeable to those large tracer molecules. The measured P_BBB was on the order of magnitude of 10⁻⁷~10⁻⁸ cm/s, and was inversely correlated with Stokes radius of FITC-dextran (Fig. 4A). The permeability to FITC-dextran decreased by nearly 90% when the corresponding Stokes radius increased from 1.4 nm (4KDa) to 6 nm (70KDa). strong size selectivity towards hydrophilic non-ionic tracers indicates the appropriate formation of tight junction complexes between adjacent endothelial cells (Matter and Balda, 2003). The overall low and size-dependent permeability to FITC-dextrans further confirmed the barrier integrity of the BBB constructs in our model. Remarkably, we achieved extremely low permeabilities to FITC-dextrans that were comparable to previously reported in vivo values (Shi et al., 2014; Yuan et al., 2009) and the in vitro data from transwell models with high TEER values (Cramer, 2014; Matthes et al., 2011). Our results were around two to three orders of magnitude lower than those reported from existing microfluidic BBB models (Booth and Kim, 2012).

Figure 4.

In vivo-like permeability to fluorescence tracers and drugs. (A) BBB permeability to FITC-dextrans inversely correlates with Stokes-Einstein radius. (B) TEER values, an index of BBB integrity, decreased by doxorubicin treatment in a dose-dependent manner, while (C) the apparent permeability to doxorubicin remained the same for different doses. (D) Permeability to FITC-dextrans and small molecule drugs measured using our BBB-on-a-Chip model correlates with in vivo data (Avdeef, 2012; Shi et al., 2014). All data presented as mean (± standard error of the mean). n = 3. *p<0.05, #p<0.005 compared to pre-treatment; $ compared to 2.5μM group. NS, non-significant.

We further examined the permeability of our BBB model to three small molecule drugs: i) caffeine, a small lipophilic molecule that rapidly pass through the BBB in vivo through simple transendothelial diffusion and saturable carrier mediated transport (McCall et al., 1982); ii) cimetidine, a H2-receptor antagonist and prototypical organic cation that shows moderate permeability across the BBB via efflux transporters and organic cation transporters (OCTs) (Koepsell et al., 2007); and iii) doxorubicin, an antineoplastic agent that is widely used to treat many different cancers yet suffers poor penetration across the BBB into brain tumors, due to its back-transport into blood by active efflux transporters, such as P-glycoproteins (Ohnishi et al., 1995). Our BBBoC model exhibited differential permeability covering 3 orders of magnitude towards those three drugs: 4.85 ± 1.84 ×10⁻⁴ cm/s for caffeine, 1.11 ± 0.09×10⁻⁶ for cimetidine, and 1.54 ± 0.66×10⁻⁷ cm/s for doxorubicin. These results closely resemble the in vivo situations. In addition, we observed a dose-dependent decrease of TEER values from ~3500 Ω·cm² to as low as ~800 Ω·cm² after 24 h treatment with 0~10 μm doxorubicin (Fig. 4B), suggesting that doxorubicin treatment compromised the BBB integrity, consistent with the vascular toxicity of doxorubicin reported in vitro and in vivo (Kotamraju et al., 2000; Murata et al., 2001). Yet the permeability to doxorubicin measured under 2.5 μM or 10 μM treatment remained the same (Fig. 4C), which implies that either the efflux transporter functionalities were not affected or the remaining transporters were sufficient to maintain the barrier for doxorubicin against the concentration gradient. This suggestion is in consistent with the abundant expression of the active efflux transporters observed in hiPSC-derived BMECs (Lippmann et al., 2012).

The permeability of the BBBoC (P_BBB and LogP_BBB) to various sized fluorescent tracers and three small molecule drugs is summarized in Supplementary Table SII, and compared with in vivo permeability data derived from rodent in situ brain perfusion studies (Avdeef, 2012; Shi et al., 2014). Our LogP_BBB results tightly correlate with the reported in vivo data (Pearson r = 0.9520, P = 0.0034, Fig. 4D), indicating that an in vivo-like, highly selective barrier is established in our model.

6. Effects of dynamic flow condition on BMEC morphology and barrier properties

We also investigated the effects of the flow condition in BBBoC on actin filament formation and cytoskeleton organization in BMECs. Fluorophore-conjugated phalloidin staining showed random orientated actin filaments across the field, although the actin filaments within each cell were mostly aligned (Supplementary Fig. SII A). The dominant direction of actin filaments for each cell was analyzed using an ImageJ directional analysis plugin – OrientationJ (Fonck et al., 2009), and the results for 1000 cells were summarized in Fig. SII B. No preferential alignment to the flow direction or any direction was observed. Such results were expected due to the low magnitude of shear stress in our system and the distinct response of BMECs to shear stress. A similar pattern of actin filament organization has been observed in BMECs even under much higher shear stress (Reinitz et al., 2015).

It has been a dominant design paradigm for a microfluidic BBB model to utilize flow induced shear stress to modulate BBB integrity. Yet it has not been shown that shear stress is either necessary or effective in establishing in vivo-like barrier properties in microfluidic BBB models. In our study, we designed the microfluidic platform to offer physiologically relevant perfusion while minimizing shear stress on the cell surface. We demonstrated that our microfluidic BBB model using iPSC derived BMECs was able to establish and maintain in vivo-like barrier tightness and compound permeability in the absence of unidirectional laminar shear stress. We also showed that high TEER levels can be sustained for up to at least 10 days on chip, which is longer than other microscale BBB models and static transwell models using the same cell source (Lippmann et al., 2012; Lippmann et al., 2014). One of the plausible explanations for sustained barrier functions is that oxygen and nutrient transportation and waste product clearance are limited in static culture wells while continuous perfusion provides sufficient nutrient and waste exchange.

7. Limitations and future work

As with any existing BBB model, the fluid to endothelial surface area ratio in BBBoC is higher than in vivo conditions. The BBBoC was designed to be readily adaptable for use in whole body microphysiological systems. The volume of recirculating medium was scaled down from whole body blood volume in mimicking human blood circulation (Supplementary Table SI), which makes the fluid-to-tissue ratio larger in the absence of other tissues. Yet the BBBoC still represents the smallest fluid to tissue ratio among existing transwell and microfluidic BBB systems (Supplementary Table SIII). Based on the comparison on daily medium consumption per unit area of endothelial cells, we found that although microfluidic systems hold the potential to achieve in vivo-like fluid-to-tissue volume ratio, most current microfluidic models actually have much larger ratios, even compared to transwell models, due to single-pass perfusion and complex tubing system. Our gravity-driven pumpless platform enables medium recirculation with much lower fluid-to-tissue ratio than other flow systems. We recognize that the ratio is still higher than in vivo values, but we believe that our BBBoC model can serve as a good platform for testing drug permeability across the BBB or effects from drugs or metabolites targeting the BBB. However, it may not be as effective in modelling the BBB as a source of metabolites due to the higher fluid to cell ratio. In the future, we will consider 3D vascular structure to improve the fluid to endothelial surface area ratio, although the current model benefits from simple setup and convenient drug administration and sampling.

BBBoC has demonstrated in vivo-like permeability to various sized dextran tracers as well as small molecule drugs. The application of our model can go beyond small molecule drug screening. Cross-BBB transport, except for passive paracellular transport and transendothelial diffusion, is mediated by assorted receptors and transporters. A variety of efflux transporters and amino acid and peptide transporters have been detected in iPSC derived BMECs, as well as various receptors, including transferrin receptor (TfR), lipoprotein receptor-related protein (LRP-1), insulin receptor; receptor for advanced glycation end products (RAGE) and leptin receptor (Lippmann et al., 2012). Recent research in CNS drug delivery has made great progress in penetrating the BBB by utilizing TfR or LRP-1 mediated transcytosis and bispecific antibodies to those receptors (Jones and Shusta, 2007; Tian et al., 2015; Yu et al., 2011). Our BBBoC model alone or integrated with other organ modules will be a valuable tool for studying the pharmacokinetics and pharmacodynamics of those “shuttle” antibodies.

The current model used primary rat astrocytes, which are generally smaller and less complex than human counterparts. The rat astrocytes are not necessarily problematic in their role of promoting the BBB phenotype. Our system and Shusta group’s transwell model (Lippmann et al., 2012) have both shown that BMECs can response to rat astrocytic cues to develop advanced barrier properties. We plan to switch to human-sourced astrocytes to create a fully human BBB model in the future, but believe that this chimeric model is still useful.

Conclusions

We demonstrated a microfluidic BBB model that mimics in vivo BBB integrity and compound permeability, is suitable for brain drug permeability studies and can be integrated into multi-organ microphysiological systems. The fact that we used an expandable source of human iPSCs to derive BMECs renders our model closer to human physiology, not limited by cell availability, and potentially to be patient- specific. We present sustained TEER levels above 2000 Ω·cm² which are the highest levels reported in a microfluidic BBB model, and for the first time fall into the range of in vivo values. The corresponding in vivo-like compound permeability further demonstrates that our model captures in vivo BBB barrier function and will be a valuable platform for drug permeability studies. In addition, our pumpless microfluidic platform that maintained the BBB constructs was geared for drug permeability screening. The stand-alone cell inserts, separating premature BBB construct preparation from long-term maintenance of mature BBB, simplify dual-side cell seeding and enables uniform cell coverage on membrane. The pumpless design using gravity driven flow for recirculating luminal perfusion allows for bubble-free, hassle-free long-term maintenance of BBB barrier function. Our data as well as some prior data demonstrate that an effective BBB model can be formed in the absence of high levels of fluid shear stress. The integrated electrodes enables on-line TEER measurement and thus barrier integrity monitoring; Open access to both luminal and abluminal compartments also allows convenient drug administration and sampling. Such a system model can be adapted easily to other barrier tissue model, such as GI tract and skin tissues. Notably, our BBB-on-a-chip platform was dimensioned based on blood residence time in human brain tissue, which allows for physiologically relevant mass transport at the BBB interface, making it easily adaptable for next level integration into multi-organ or body-on-a-chip systems.