A microfluidic model of human brain (μHuB) for assessment of blood brain barrier

Abstract Microfluidic cellular models, commonly referred to as “organs‐on‐chips,” continue to advance the field of bioengineering via the development of accurate and higher throughput models, captivating the essence of living human organs. This class of models can mimic key in vivo features, including shear stresses and cellular architectures, in ways that cannot be realized by traditional two‐dimensional in vitro models. Despite such progress, current organ‐on‐a‐chip models are often overly complex, require highly specialized setups and equipment, and lack the ability to easily ascertain temporal and spatial differences in the transport kinetics of compounds translocating across cellular barriers. To address this challenge, we report the development of a three‐dimensional human blood brain barrier (BBB) microfluidic model (μHuB) using human cerebral microvascular endothelial cells (hCMEC/D3) and primary human astrocytes within a commercially available microfluidic platform. Within μHuB, hCMEC/D3 monolayers withstood physiologically relevant shear stresses (2.73 dyn/cm2) over a period of 24 hr and formed a complete inner lumen, resembling in vivo blood capillaries. Monolayers within μHuB expressed phenotypical tight junction markers (Claudin‐5 and ZO‐1), which increased expression after the presence of hemodynamic‐like shear stress. Negligible cell injury was observed when the monolayers were cultured statically, conditioned to shear stress, and subjected to nonfluorescent dextran (70 kDa) transport studies. μHuB experienced size‐selective permeability of 10 and 70 kDa dextrans similar to other BBB models. However, with the ability to probe temporal and spatial evolution of solute distribution, μHuBs possess the ability to capture the true variability in permeability across a cellular monolayer over time and allow for evaluation of the full breadth of permeabilities that would otherwise be lost using traditional end‐point sampling techniques. Overall, the μHuB platform provides a simplified, easy‐to‐use model to further investigate the complexities of the human BBB in real‐time and can be readily adapted to incorporate additional cell types of the neurovascular unit and beyond.


| INTRODUCTION
The blood-brain barrier (BBB) is the prominent barrier at the interface of the blood stream and the central nervous system (CNS) and is primarily responsible for maintaining brain homeostasis and protecting the CNS from harmful foreign entities. 1 As the brain's first line of defense against solutes and particulates in the blood, the brain microvascular endothelial cells form a tight barrier that limits the transport of nutrients and other molecules into and out of the CNS space. Combined with pericytes and astrocytes, these cells collectively form a neurovascular unit, contributing to the overall BBB phenotype. The brain endothelium is characterized by expression of tight junction complexes lack of fenestrations, and low pinocytic activity. 2,3 Although these characteristics are imperative for normal brain function, the BBB limits the penetration of therapeutics into the brain. 4 As a result, there is a clear need for the development of adequate models to further investigate the mechanisms of transport across the BBB in order to design better brain delivery strategies.
Assessing transport of nanoparticles, proteins, and other therapeutics across the BBB can be challenging; nonetheless, researchers have designed various in vivo models to investigate this transport in both heathy and diseased BBB. [5][6][7][8] Animal models inherently include all contributing factors that dictate the transport across the BBB.
However, translating findings from rodent models to humans remains a challenge. 9,10 Further, the complexity of the in vivo environment also poses a challenge for interpreting the results. For example, transport of nanoparticles into the brain in vivo is a combined outcome of immune clearance and permeation across the BBB, thus making it difficult to deconvolute the contributions of each factor from the measured experimental outcome. Common techniques to investigate BBB transport of therapeutics in vivo include single carotid injections, internal carotid artery perfusion, and intravenous injections. 11,12 Using intravenous injections can be disadvantageous for investigating BBB transport due to the potential rapid metabolism of the therapeutic, resulting in metabolism-induced artifacts and greater likelihood of clearance before reaching the brain microcirculation. Alternatively, single carotid injections and internal carotid artery perfusions can reduce the likelihood of clearance while also limiting metabolic events within the brain microcirculation. Unfortunately, these techniques are labor-intensive, requiring significant training and expertise to properly implement. 13 As a result, there continue to be strong interests in developing simple yet physiologically relevant in vitro models of the human BBB that are also highly tunable and customizable to be used as tools to further investigate brain-related phenomena.
To date, the primary in vitro tool of choice for researchers studying human BBB permeability is the static transwell migration assay, also referred to as the Boyden chamber assay. These assays offer the flexibility to conduct both monolayer 14 and coculture experiments, [15][16][17] can noninvasively estimate barrier permeability using transendothelial electrical resistance (TEER) measurements, and are convenient for acquiring permeability information across a monolayer, including disease models. [18][19][20] However, transwell inserts can be subject to increased, artificial paracellular diffusion at the monolayer perimeter by a phenomenon known as "edge effects," especially for highly hydrophilic compounds. 21 This erroneous effect results from incomplete coverage of the porous inserts at the monolayer perimeter due to the inability of the endothelial cells to form tight junctions along the inner wall of the apical chamber. 22 Typically, analyte concentrations are sampled from the apical or basolateral chamber over time without the ability to actively monitor transport.
Additionally, depending on the cell culturing conditions and experimental setup, TEER values can vary significantly. 23 To confound these reported values further, reports often misrepresent the TEER value by describing it in terms of total resistance or area-dependent resistance.
Hemodynamic shear stress experienced by endothelial cells is an important mechanotransduction regulator not present in static transwell migration assays. Depending on the blood vessel geometry and condition, endothelial cells can experience a range of shear stresses.
In vitro studies report shear stresses between <1 and 85 dyn/cm 2 induce a variety of biological responses. 24 For instance, shear stress acts as a pleiotropic modulator of the endothelial cell physiology, regulating genes involved in cell division, differentiation, migration, extracellular matrix protein secretion, cell-cell adhesion, and apoptosis. 25 As a result, shear stress contributes to an overall polarized brain endothelium, influencing such properties as asymmetric expression of localized enzymes and carrier-mediated transport systems, production of vasoactive substances and cell adhesion molecules, cell survival, and energy metabolism. [26][27][28] The maintenance of brain microvascular endothelial cells is directly impacted by this hemodynamic shear stress, influencing tight junction formation and multidrug resistance transporter expression. 29 Unlike endothelium in other organs of the body, brain microvascular endothelial cells resist elongation in response to both curvature and shear stress. [30][31][32] Interestingly, a report by Garcia-Polite et al. demonstrates cerebrovascular function (i.e., expression of tight junction proteins ZO-1, Claudin-5, and efflux pump P-gp) can be directly correlated to the magnitude and nature of shear stress. Higher than physiologically relevant (40 dyn/cm 2 ) and pulsatile shear stresses resulted in downregulation of ZO-1, Claudin-5, and P-gp; however, tight junction marker expression recovered when physiological shear was reestablished, 33 further suggesting the importance of maintaining hemodynamic shear stress among in vitro systems to more accurately represent the BBB microenvironment.
Recent developments in this field have resulted in a diversity of three-dimensional cell culture models and several dynamic systems with the ability to incorporate hemodynamic shear. 34,35 Still, simultaneous visualization of the BBB and the associated transport through the barrier in real-time remains a challenge. [36][37][38][39] Direct visualization at a cellular level provides real-time monitoring of the cellular morphology and can be used as a proxy for cell behavior. This allows for measurement of protein localization information in addition to expression levels. With the ability to directly capture transport, one can collect more complex information, such as the precise interactions of a particulate of interest (e.g., monocyte, virus, nanoparticle) before, during, and after interacting with the BBB, which otherwise would be impossible. This capability also simplifies the measurement of transport kinetics while simultaneously offering higher temporal resolution than would be possible using a traditional sampling-type approach.
Some models have attempted to visualize transport across the BBB in real time; 40,41 however, the shear stresses applied in these experiments (3.8 × 10 −3 to 0.15 dyn/cm 2 ) are often orders of magnitude lower than what are considered physiologically relevant within the brain microvasculature (1-30 dyn/cm 2 ). 24,[42][43][44] Maintaining the culture under higher shear stress for prolonged periods of time in a microfluidic environment poses a significant challenge. 45 This limitation is especially significant given that previous reports indicate low shear stresses may be insufficient to induce the proper morphological and biochemical changes. For example, studies performed using a bovine aortic endothelial cell model have shown that expression of p53, a tumor suppressing protein, was upregulated when the cells were subjected to 3 dyn/cm 2 but not 1.5 dyn/cm 2 . The mechanotransduction effects of shear stress are believed to mediate several cellular functions, including the inhibition of cellular proliferation by the activation of p53 expression with the potential of arresting endothelial cell apoptosis. 46 Furthermore, in the absence of laminar flow, static monolayers can be subject to uncontrolled growth, resulting in formation of multiple layers, if allowed to proliferate. 47 Therefore, a model with the ability to incorporate physiologically relevant shear stresses is essential to effectively capture biologically relevant transport across any barrier in direct contact with the bloodstream. Additional limitations of existing models to probe human brain permeability include the use of rodent brain endothelial cells, 36,40 which do not exhibit the same anatomical and molecular complexities as their human counterparts. 48,49 Alternatively, while the use of primary human brain endothelial cells may have significant advantages, 37,50 these cells can be difficult to acquire, variable in nature, and challenging to culture and maintain, especially in a microfluidic environment. 51 Herein, we report the development of a microfluidic human BBB model (μHuB) with the ability to directly monitor both the barrier and associated transport in the presence of physiologically relevant shear conditions. This model leverages a commercially available chip with low required volumes and a well-characterized, immortalized cell line to provide a convenient and effective research tool for investigating the human BBB and its permeability. Because of the transparent nature of the glass and polydimethylsiloxane (PDMS) μHuB structure, temporal and spatial permeability data across the BBB can be easily acquired using a conventional or confocal fluorescent microscope. We further demonstrate that μHuB is modular and can be readily adapted for more complex, coculture experiments to further bridge the gap between existing tools for investigating the human BBB and underlying biology.  Figure 1). This design was chosen to facilitate comparisons with other transwell models with a pore size of 3 μm, which are commonly used to study transport across static in vitro models. 52 Initially, devices were coated with a variety of basement membranes, including rat tail collagen Type 1, human fibronectin, and laminin, which have been used to promote cell adhesion in the literature. 40,53,54 Optimal cell adhesion was observed with a thin coating of human fibronectin and was used for all studies reported. Consistent cell attachment to the upper portion of the PDMS channel proved challenging using standard injection techniques. Therefore, we adopted a two-step seeding protocol as described by Herland  Within the μHuB, the magnitude of the expression of these proteins, however, increased dramatically in response to fluid flow as compared to its static counterpart.
The impact of shear stress on cell viability was investigated with a live/dead assay. Cell viability was measured by the reduction of C 12resazurin to red-fluorescent C 12 -resorufin. SYTOX Green was used as a counterstain to identify cells with compromised cell membranes. This green-fluorescent nucleic acid stain cannot penetrate intact cell membranes and remains non-fluorescent until bound to the nucleus. Relative

| Expansion of the μHuB model with astrocytes
As the neurovascular unit comprising the BBB contains additional cell types beyond the brain endothelium, μHuB can be expanded by cocul-

| Discussion
We have presented the design and characterization of a realistic yet simple in vitro model of the human BBB: μHuB. Importantly, the μHuB recapitulates several of the most critical aspects of the in vivo BBB, specifically the incorporation of appropriate brain endothelial cells 59 into a vessel-like architecture that exposes the cells to shear. 66 Moreover, by combining a commercially available, immortalized cell line with a straightforward, commercially available microfluidic chip, we have developed a highly accessible model that can be readily adopted and utilized as an experimental tool and analysis method for dynamically visualizing particulates of interest in future studies. An essential, functional participant of the neurovascular unit is the basement membrane. Basement membrane in the brain is primarily composed of laminins and collagen IV. 67 We found, however, that hCMEC/D3 cellular morphology and adherence to the internal glass and PDMS surfaces were optimal when coated with human fibronectin. This may be partially attributed to the structural support provided by the chip itself, as collagen IV has been implicated to have a primarily structural, scaffold-like function. 68  The functional properties of our model were investigated by conducting permeability experiments using dextrans of varying molecular weights. These and other tracer compounds, like Evans blue and horseradish peroxidase, are commonly used to assess the permeability of the BBB. 72,73 The tight intercellular junctions between brain endothelial cells has been shown to exclude passive transport of molecules having Stokes' radii >10 Å. [74][75][76] As in vitro models do not fully recapitulate all of the necessary components for such a "tight" BBB, researchers often use dextrans with varying Stokes' radii to determine the relative "leakiness" due to passive diffusion around the endothelial cells. As expected, a size-dependent trend was observed in the permeability (P e ), where molecules with a larger Stokes' radii crossed the barrier at a reduced rate ( Figure 6).
Overall, transport of the tracers reported were comparable in magnitude to those measured in prior experimental work using neonatal rat brain endothelial cells on a similar scaffold design (15 × 10 −6 cm/s for a 10 kDa dextran reported here versus 40 × 10 −6 cm/s for a larger 40 kDa dextran reported by Deosarkar and coworkers). 40 Our permeability data also agree with mathematical modeling to calculate permeability values for macromolecules with similar Stokes' radii across an endothelial barrier. 65 Yuan et al. measured permeabilities for FITC-dextrans (10 and 70 kDa) in vivo. Both dextrans were found to exhibit low, but detectable permeabilities of 0.31 × 10 −6 cm/s for 10 kDa and 0.15 × 10 −6 cm/s for 70 kDa. 77 These values are much lower than our reported findings as well as for other in vitro models.
One explanation for this could be that as hCMEC/D3 cells are an immortalized cell line, tight junction expression may be reduced as compared to their primary counterparts. Researchers have developed a variety of different human brain endothelial cell lines, including BB19, hBMEC, hCMEC/D3, and TY10. Eigenmann and coworkers report dramatic differences between the tight junction protein expressions between these immortalized cell types. 63 Theoretically, the use of primary human brain microvascular endothelial cells in the μHuB model would lead to a reduction in the permeability. Inclusion of additional cellular components (e.g., astrocytes and pericytes) may also enhance the barrier properties. Sajja and coworkers 78 as well as Herland and coworkers 50 have shown that the addition of these other cell types caused a reduction in the permeability values. Modeling by Li and coworkers suggests that the astrocytes contribute significantly to the diffusive barrier properties of the BBB. 65 The permeabilities reported in our study were calculated based on our current understanding of small macromolecule translocation across the BBB, namely that the transport of dextran tracers through the BBB should remain constant with time. The transport data acquired using the μHuB can also be used to investigate potential temporal differences in permeability. As seen in Figure 6f, different temporal regions of a single experiment can have apparent permeabilities that differ over twofold but are still comparable to previously reported literature. To our knowledge, these differences are unlikely to be captured using other tools. With the dynamic visualization capability of the μHuB, heterogeneities originating from spatial biological variability can also be assessed in a single experiment by analyzing the local permeability at different azimuthal locations along the semipermeable barrier. To our knowledge, investigations into this type of variability have not been reported to date. As a result, the μHuB can be a powerful tool for developing a deeper mechanistic understanding of any type of particulate transport through the BBB both in time and space.
As the blood-brain barrier consists of various cell types in addition to brain endothelial cells, including astrocytes, pericytes, and glial cells, a coculture of primary human astrocytes and hCMEC/D3 was successfully cultured using a similar protocol for the hCMEC/D3 only models to achieve complete lining of the central compartment with primary astrocytes. Different cell types can easily be incorporated into the central compartment to further investigate the functional roles of BBB components and how specific cell-to-cell interactions affect transport of molecules across the brain endothelium. Additionally μHuB can be easily expanded to incorporate additional components of interest, including the use of differentiation factors (e.g., 8-CPT-cAMP and Ro 20-1,724), 79 primary human brain endothelial cells instead of the immortalized line, modification of cell type ratios to represent different regions of the brain, 80 and modulation of the applied shear stress, to create a holistic model of a healthy BBB. μHuB can also be readily modified to further investigate how transport is affected in a diseased state, such as when there is inflammation caused by a traumatic brain injury or as the result of an invasive glioblastoma.

| Conclusions
We have reported the development of μHuB, an easy-to-use human microfluidic blood-brain barrier model. The ability of endothelial monolayers in the μHuB to mimic the lumen of the BBB depends critically on a newly developed protocol to condition the cells to physiologically relevant shear conditions. Using this conditioning protocol, monolayers can be maintained at physiologically relevant shear stresses to spatially and temporally resolve the transport of particulates across the BBB in real-time. We anticipate that experiments in the μHuB can easily be expanded to quantify and mechanistically investigate transport of molecular and particulate species across various states of the BBB.  For coculture experiments, primary human astrocytes (Catalog #1800) were obtained from ScienCell and maintained astrocyte medium (Catalog #1801) also obtained from ScienCell. Cells were cultured on poly-L-lysine coated tissue culture flasks (2 μg/cm 2 ), which were allowed to coat in the incubator overnight prior to use. Cells were incubated at 37 C, 95% humidity and 5% CO 2 until confluent.

| Culture of hCMEC/D3 and primary astrocytes in μHuB
To facilitate endothelial cell attachment, human fibronectin (300 μg/ mL) was injected in the outer compartment and allowed to incubate for 1 hr at 37 C and 5% CO 2 . The entire device was perfused with complete cell culture media. To devoid the device from any residual entrapped air, the device was primed using inert N 2 gas at 6 PSI for 30 min. Devices were placed inside cell culture incubator prior to use.
For coculture experiments, the device was first perfused with a thincoating of Matrigel (1:5) in the central compartment for 1 hr at 37 C and 5% CO 2 prior to coating the outer channels with human fibronectin (300 μg/mL) as described previously.
hCMEC/D3 grown to 70 to 80% confluency were trypsinized and resuspended in cell culture media with increased serum concentration (10%). Cell suspension at~5 × 10 7 cells/mL was injected into the outer compartment at 6 μL/mL using a Harvard Apparatus Pump For coculture seeding, after replenishing media in the outer compartments containing endothelial cells, primary human astrocytes were injected into the central compartment and allowed to attach.
To condition cells to physiological shear stresses, 10% FBS containing media was injected according to a linear ramp profile (100 μL/min-5 μL/min) over 12 hr using a Harvard Apparatus PHD ULTRA™ with a 6 × 10 MultiRack attachment for multi-syringe perfusion. Constant 5 μL/min injection rate was maintained for at least 6 hr prior to use. Devices were inspected for any bubble formation and immediately used for further studies.

| Visualization and inner lumen characterization of μHuB with actin stain
After flow conditioning of model, DPBS was perfused to replace the cell culture media. 4% PFA was injected into all device compartments and allowed to remain at room temperature for 15 min. The device was again perfused with DPBS to move any residual PFA. Fixed cells permeabilized using 0.2% Triton X-100 in DPBS for 10 min. The device was again perfused with DPBS to move any residual Triton X-100. Thermofisher ActinRed™ 555 ReadyProbes™ Reagent was used to stain for cytoskeleton, using two drops per mL of DPBS for 30 min at room temperature. The device was perfused with DPBS one final time prior to imaging.
For coculture μHuBs, the same actin staining procedure described above was used with slight modifications. ThermoFisher ActinGreen™ 488 ReadyProbes™ was used to stain hCMEC/D3 cytoskeleton in the vascular compartment and Thermofisher ActinRed™ 555 Ready-Probes™ Reagent was used to stain primary human astrocyte cytoskeleton in the tissue compartment. For each dye solution, two drops per mL of DPBS was used and allowed to remain in the respective compartment for 30 min at room temperature prior to perfusing with DPBS and imaging.

| Tight junction protein characterization in μHuB (ZO-1, Claudin-5)
After flow-conditioning, μHuB was perfused with DPBS to replace the cell culture media. 4% PFA was injected into all device compartments and allowed to remain at room temperature for 15 min. The device was again perfused with DPBS to remove any residual PFA. Fixed cells were then permeabilized using 0.2% Triton X-100 in DPBS for 10 min.
The device was again perfused with DPBS to move any residual Triton X-100. The device was blocked with 5% donkey serum and 5% goat serum for 30 min at room temperature. ZO-1 (1:100) and Claudin-5 (1:200) primary antibodies were diluted in antibody diluting buffer (0.1% Tween-20 and 0.1% BSA) at 4 C overnight. Corresponding fluorescently labeled secondary antibodies Anti-Goat and Anti-Donkey (1:1000) was allowed to incubate for 1 hr at room temperature prior to perfusing with DPBS and was immediately imaged.

| Acquisition of transport information in μHuB
Following flow-conditioning, 312.5 nM of FITC-Dextran (10 and 70 kDa) was injected into the apical channel at 5 μL/min over 2 hr.
Device was maintained humidified and at 37 C and 5% CO 2 using a Zeiss environmental enclosure. Images were acquired using a 5X objective in 1 min intervals for the duration of the experiment.

| Quantification of FITC-dextran permeation using fluorescent microscopy
Acquired fluorescent image stacks from transport experiments were imported into MATLAB and analyzed using a custom code. Briefly, the average pixel intensity and standard deviation within the apical channel and the basolateral chambers were calculated for each frame.
Intensity in the basolateral chamber was normalized to the equilibrium intensity of the apical channel, resulting in a normalized intensity profile (Figure 6c). Frames collected prior to the apical chamber reaching an equilibrium intensity were excluded from the analysis. Permeability was calculated from the normalized intensity profiles using: where V/S is the ratio of apical volume to surface area. The linear portion of the resulting intensity over time curve was fit to a line using the MATLAB fit function and weighting with the standard deviations of the intensity. The slope of this line was then used to calculate the permeability as shown in Equation 1 and as described in previous work. 40,81 Stationary and inflection points were identified using quadratic and cubic fits, respectively, with identical weighting. The permeability of the analyte was assessed by using frames acquired before the intensity profile plateaued. For example, Figure 6d shows a normalized intensity profile for 70 kDa dextran. As before, frames collected prior to the apical chamber reaching its equilibrium value are not included. The profile plateaus between t = 50 min and t = 100 min. Based on the fitting inflection points, this curve changes slopes at t = 60 min. Only frames before t = 60 min were used for the permeability calculations. Permeability of the acellular scaffold (P scaffold ) was subtracted from the overall permeability observed (P total ) to calculate the true permeability of the endothelial cell barrier (P e ) for a given tracer (Equation 2). 82

| Statistical analysis
Experiments were run in triplicate, and permeability error bars represent a 95% confidence interval based on the linear fitting.