Cell-Culture Models of the Blood–Brain Barrier

Monolayer Models

A monolayer of endothelial cells grown in the Transwell insert (Figure 2A) is used as a simple in vitro BBB model. The insert mimics the blood (luminal) side, whereas the well in which the insert sits mimics the parenchymal (abluminal) side. The microporous membrane support (0.4 μm) allows the exchange of small molecules and cell-secreted growth factors but prevents the migration of cells between the 2 compartments. To mimic the unique properties of BMECs (see Brain Microvascular Endothelial Cell section of this article), primary or low passage BMECs are used. Accumulating evidence shows that primary or low passage BMECs retain many of the biochemical and morphological properties that characterize the BBB in vivo, such as the presence of TJs and polarized expression of transporters, receptors, and enzymes.¹ Two big challenges, however, exist in isolating and culturing primary BMECs: the low yield and contamination by mural cells. Because the brain vasculature accounts only for 0.1% (v/v) of the brain, a large number of rodents are usually needed to generate enough BMECs.¹²⁰ Therefore, larger species, including bovine, porcine, and nonhuman primate, are used to increase the yield of BMECs. The problems with larger species, however, are that these animals are usually nontransgenic and there are no or few respective antibodies available for these species. For these reasons, rodents are still the most widely used animals in BBB research. Recently, temporary application of puromycin in the culture medium has been shown to dramatically improve the purity of primary BMECs.^34,121,122 This methodological breakthrough enables the use of BMECs at high purity and significantly advances BBB research.

Human cells may be needed when human-specific transporters/receptors or immunologic issues are involved. Primary human cells, however, are not usually available because of ethical reasons. To circumvent this issue, many immortalized human cell lines, such as human cerebral microvascular endothelial cell line (hCMEC/D3)¹²³ and immortalized human cerebral endothelial cells,¹²⁴ have been generated. One big benefit of the immortalized cell lines is their significantly enhanced ability to proliferate in culture. Because of this advantage, immortalized brain endothelial cell lines from rodents, such as rat brain endothelial cell line, mouse brain endothelial cell line, and mouse cerebral endothelial cell line, have also been generated. These immortalized cell lines, however, unlike primary or low passage BMECs, lose some of the unique properties that characterize the in vivo BBB.^125–129 It has been shown that some of the BBB-specific transporters and enzymes are expressed at much lower levels in the immortalized cell lines. For example, the expression of TJPs, including occludin and claudin-5, in hCMEC/ D3 cells is lower than that in primary cerebral endothelial cells.¹³⁰ RBE4 and hCMEC/D3 cells have been found to have less glucose transporter 1 compared with brain endothelium in situ.^130–132 Additionally, γ-glutamyl transpeptidase and alkaline phosphatase activity is dramatically reduced in RBE4 cells and other rat brain endothelial cell lines, compared with isolated rat brain capillaries.^133,134 Because of the above-mentioned limitations, these immortalized cell lines are not able to generate a tight monolayer and thus have inadequate barrier function. To overcome this problem, reagents that modulate BBB properties, such as cAMP and glucocorticoids, have been used to improve the tightness of the endothelial monolayer.^135–140

An alternative way to generate human BMECs is through human pluripotent stem cells (hPSC). Although previous studies showed that hPSC-generated endothelial cells do not have organ-specific properties,^141–143 Lippmann et al¹⁴⁴ successfully differentiated hPSC into endothelial cells with unique BBB properties. These hPSC-derived endothelial cells expressed TJPs, various nutrient receptors, amino acid and peptide transporters, and efflux transporters.¹⁴⁴

It should be noted that the monolayer model only has 1 major cell type (BMECs) and lacks the communication among different cell types. Accumulating evidence shows that the BBB properties are dictated by the microenvironment in the brain rather than intrinsic to BMECs, and cell–cell communications play a critical role in the maintenance of these properties.^76,145–147 Thus, this oversimplified monolayer model is not ideal for the study of BBB integrity. Because of the intrinsic simplicity,¹⁴⁸ however, monolayer models have been widely used in signaling pathway and transporter kinetic studies, binding affinity measurement, and high-throughput screening.

Coculture Models

To better mimic the anatomic structure of the in vivo BBB, BMECs are cocultured with other cells that directly contribute to the barrier properties of BBB.^{4,26,77–83,98,99} Based on the evidence that interaction between BMECs and astrocytes increases the expression of transporters and TJs in BMECs, induces the formation of cell polarity in BMECs, and promotes a phenotype more closely mimicking the in vivo BBB,^26,149–158 BMECs are cocultured with astrocytes to improve their barrier properties. In this coculture model, BMECs are seeded in the Transwell insert and astrocytes are grown either on the underside of the Transwell insert or at the bottom of the well in which the insert sits (Figure 2A). Consistent with previous reports, higher TEER and lower permeability to molecular tracers are observed in this BMEC–astrocyte coculture model.^{135,159–161} Cells of different origins have been used in this model. For example, mouse BMECs cocultured with astrocytes generated a TEER of 200 to 300 Ω·cm^2,162 whereas bovine BMECs cocultured with astrocytes produced a higher TEER around 500 to 600 Ω·cm^2.159–161 Additionally, a TEER of 900 Ω·cm² was achieved when porcine BMECs were used.¹⁶³ When hPSC-derived endothelial cells were cocultured with rat astrocytes, the TEER reached 860 Ω·cm^2.144 The different TEER values and thus the barrier properties of BMECs are probably because of more efficient BMEC isolation/purification methods rather than species difference, given that equally high TEER values (600–1000 Ω·cm²) were observed in the BMEC–astrocyte coculture model independent of species.^16,164–166 However, to answer the question clearly whether there is a species difference in BMEC tightness, comparative studies using the same methods and conditions within the same laboratory are needed. It should be noted that the TEER values obtained from the BMEC–astrocyte coculture model are still lower than that in in vivo conditions, which is estimated to be more than 1000 Ω·cm^2.14,167 suggesting that other players also contribute to the integrity of BBB.

The BMEC–astrocyte coculture model has been widely used to study stroke, the leading cause of disability and the third most common cause of death in the United States.¹⁶⁸ Hypoxia and hypoglycemia, 2 pathological changes in stroke, can be easily mimicked in vitro by lowering oxygen and glucose levels.¹⁶⁹ This in vitro stroke model has been applied to investigate endothelial pathophysiology,^30,170 endothelial barrier permeability,^171,172 cell–cell communication,¹⁷³ and free radicals¹⁷⁴ during stroke. Additionally, it has also been used in large-scale screening for compounds that have protective function in hypoxia-reoxygenation induced injury.¹⁷⁵ The advantages of using this model to study stroke include¹⁶⁹ (1) the severity of stroke can be easily controlled by regulating oxygen and glucose levels, (2) reperfusion-induced injury can be replicated by reoxygenation, (3) the contribution of hypoxia/hypoglycemia alone or together can be investigated, (4) the contribution of each cell type can be easily identified, (5) the molecular mechanism underlying stroke can be dissected out because of the intrinsic simplicity of this system, and (6) it allows large-scale screening of compounds that have therapeutic potentials for stroke. One major disadvantage of this in vitro model is that it lacks shear stress (see Dynamic Models section of this article for details).

Because of the key role of pericytes in BBB regulation,^4,98,99 a BMEC–pericyte coculture model has also been developed. The cells are grown in Transwell insert as described above. Pericytes dramatically enhanced the TEER in this coculture model, compared with the monolayer model.¹⁷⁶ In addition, multidrug resistance-related protein 6 expression and matrix metalloproteinase secretion in BMECs were increased in this model.^177,178 This model is particularly useful in the study of BEMC–pericyte interaction/communication. Although pericytes enhance BMEC tightness, astrocytes are absent in this BMEC–pericyte coculture model, suggesting that this model may be further improved by incorporating astrocytes.

To replicate the anatomic structure of the in vivo BBB fully, a BMEC–pericyte–astrocyte triple coculture system was built and characterized. In this model, BMECs are plated in the Transwell insert with pericytes on the underside of the insert and astrocytes at the bottom of the well (Figure 2A). The direct contact of BMECs and pericytes mimics the anatomic structure of neurovascular unit in vivo. Although direct cell–cell communication between astrocytes and BMECs/ pericytes is missing in this triple-culture model, indirect effect of astrocytes in BBB regulation via secreted soluble factors is included. As expected, this triple coculture model has significantly higher TEER and lower permeability,^94,179 indicating that both pericytes and astrocytes contribute to the integrity of BBB. These data suggest that the BMEC–pericyte–astrocyte triple coculture model is a more reliable in vitro BBB model. Therefore, this model has been widely used in many BBB permeability studies.^180–185 Like the BMEC–astrocyte coculture model, this triple-culture model can also be used as an ischemia-reperfusion model for pathophysiological studies or large-scale screenings.

In addition to the above-mentioned cell types, neurons and other cells have also been cultured with BMECs. It has been shown that neurons induce the expression of BBB-related enzymes and occludin in BMECs.^111,112 Neurons also significantly decreased the leakage of sucrose in the BMEC–astrocyte coculture model.^108–110 Additionally, macrophages when cocultured with BMECs significantly increased the TEER.¹⁸⁶ Recently, stem cells and stem cell-derived cells have been used in in vitro BBB models. For example, embryonic neural progenitor cells have been shown to induce BBB properties in BMECs.^187–189 Neural progenitor cell-derived cell mixtures, including neurons, astrocytes, oligodendrocytes, and proliferating neural progenitors, have also been shown to induce the BBB phenotype in rat BMECs.¹⁹⁰

The coculture models incorporate important players other than BMECs and thus better mirror the anatomic structure of the in vivo BBB. These coculture models generate a tighter barrier and are ideal for permeability studies, cell–cell interaction, and lead compound identification/optimization in new-drug R&D. The most significant disadvantage of these coculture models is the lack of shear stress, which plays a critical role in the induction and maintenance of the BBB phenotype.^86,191 Therefore, it is advisable to validate the results generated from the static models in dynamic models and in vivo.

Cone-Plate Apparatus

The cone and plate viscometer was first used to build shear force.¹⁹² In this model, a rotating cone generates shear force, which is transmitted to the endothelial monolayer via the medium (Figure 2B). The angular velocity and the cone angle determine the shear stress it generates. Because the shear stress is not evenly distributed along the radius of the plates, the endothelial monolayer receives different shear stress depending on its location in the plates. In addition, other cell components of the BBB, such as astrocytes and pericytes, are not included in this model, which limits its application in BBB research and diminishes the reliability and significance of data it generates.

Dynamic In Vitro BBB Model

To incorporate both shear stress and other cell types, microporous hollow fibers are used (Figure 2C).^193–196 In this model, BMECs and astrocytes are plated in the inner (luminal) and outer (abluminal) sides of the porous hollow fibers, respectively (Figure 2D). The culture medium is then pumped into the system via a variable-speed pump to generate shear stress comparable with that seen in physiological conditions in vivo (5–23 dynes/cm²).^197,198 To maintain the stable microenvironment, a gas-permeable tubing system is used for the exchange of O₂ and CO₂. Using bovine aortic endothelial cells and C6 glioma cell line, a TEER value of ≈600 Ω·cm² was achieved in this dynamic model.¹⁹⁹ Further side-by-side comparative studies showed that this dynamic model generated a TEER >10× higher than that in the Transwell coculture model and that its permeability to sucrose and phenytoin was, respectively, 10× and 5× less than that in the coculture system,²⁰⁰ suggesting a tighter barrier function of this model. Cucullo et al¹⁹⁶ also compared the barrier function of this model using different endothelial cells and found that brain endothelial cells (human BMECs and hCMEC/D3) generated a much higher TEER (1200 Ω·cm²) and less permeability to sucrose or phenytoin than endothelial cells of peripheral origin (human umbilical vein endothelial cells), rationalizing the use of brain endothelial cells to construct in vitro BBB models. Additionally, the expression of transporters, ion channels, and efflux proteins was dramatically induced in BMECs in this model.^199,201,202 Altogether, these data suggest that this dynamic model better mimics the in vivo BBB by replicating its anatomic and physiological properties.^{86,167,194,196,199,200}

This dynamic in vitro BBB model has been used to study the pathophysiology of various CNS diseases, including ischemia-reperfusion-induced injury and epilepsy.^167,196,202 The molecular events during ischemia include flow disturbance (loss of shear stress) and induction of inflammatory response (release of reactive oxygen species and cytokines by white blood cells).²⁰³ These changes have been replicated in the dynamic BBB model by flow cessation and reperfusion in the presence of white blood cells.²⁰⁴ It has been shown that flow cessation for 1 hour in this system induces a biphasic opening of the BBB^167,196 and ibuprofen pretreatment partially prevents BBB failure and decreases the duration and degree of BBB breakdown,¹⁹⁶ suggesting that the inflammation immediately after ischemic stroke is one of the major causes of BBB failure. In another study, endothelial cells and astrocytes from normal or drug-resistant epileptic human brain tissue were cultured in this dynamic system.²⁰² The permeability to phenytoin, a substrate for P-glycoprotein, was significantly reduced when endothelial cells from epileptic brain were used, although TEER and permeability to sucrose were not affected.²⁰² This effect was independent of the origin of astrocytes and could be reversed by P-glycoprotein blocker XR9576,²⁰² suggesting that the drug-resistant BBB phenotype in epileptic patients can be replicated in this dynamic BBB model.

Recently, a revised model with hollow fibers with trans-mural microholes of 2 to 4 μm has been developed to allow transmigration/trafficking studies.¹⁶⁷ Extravasation of white blood cells was observed when the BBB is breached by flow cessation and reperfusion.¹⁶⁷ It should be noted that the space between BMECs and astrocytes (the thickness of the hollow fiber) is 150 μm¹⁶⁷ compared with the immediate contact of these cell types in vivo. This extremely thick layer may affect the extravasation of immune cells from the luminal side to the abluminal side. Thus, cautions should be taken when interpreting data generated from this dynamic in vitro BBB model. In addition, this dynamic in vitro BBB model also has three other disadvantages. First, it does not allow direct visualization of the endothelial morphology in the luminal compartment. Although cells can be characterized after trypsin treatment, the harvesting process may change the morphology and physiological characteristics of the cells. Second, the cell number (>1×10⁶) and technical skills required to build this model are relatively high. Third, the time required to reach steady-state TEER is longer (usually 9–12 days)^167,199 compared with that in coculture models (usually 3–4 days). These shortcomings prevent the use of this dynamic in vitro BBB model in large-scale screens. This model, however, may be useful in lead compound validation/optimization in new-drug R&D.

Microfluidic-Based BBB Models

To address the limitations of the dynamic in vitro BBB model, microfluidic device-based in vitro BBB models have been developed.^205,206 The microfluidic BBB (μBBB) model is one of the first such models. It contains 2 perpendicularly crossing channels, which allow dynamic flow and generate shear stress; a polycarbonate porous membrane at the intersection of these channels, which enables coculture of BMECs and astrocytes; and multiple built-in Ag/AgCl electrodes for TEER measurement^205,206 (Figure 3A). These channels have a height of 200 μm and widths of 2 mm (luminal) and 5 mm (abluminal). In this model, BMECs and astrocytes are seeded on the luminal and abluminal sides of the membrane, respectively (Figure 3B). A pump and a gas-permeable tubing system are used to generate shear stress and allow O₂–CO₂ exchange, respectively. Under a shear stress of 0.023 dynes/cm² (much lower than that in physiological conditions), this μBBB model significantly improved the TEER of bEND.3 cells to 140 Ω·cm² after 3 days of culture, in contrast to 15 Ω·cm² in the static model.^205,206 When cocultured with astrocytes, the TEER was further increased to ≥250 Ω·cm^2.205,206 Additionally, the permeability increased on histamine exposure, followed by recovery, suggesting the stability of this model.^205,206 This μBBB model was further improved by replacing the oxidation sensitive AgCl electrodes with inert platinum ones and decreasing the cross-sectional area.²⁰⁷ These modifications allow accurate measurement of TEER and reduce the amount of cells needed. Without shear stress, hCMEC/D3 cells reached a TEER of ≈40 Ω·cm² within 3 days of culture in this improved μBBB model.²⁰⁷ With only 18 hours of physiological shear stress (5.8 dynes/cm²), the TEER was increased to 120 Ω·cm^2.207 Tumor necrosis factor-α, a known inflammatory cytokine, dramatically decreased the TEER to 12 Ω·cm^2.207 These data suggest that this improved model is able to respond to mechanical and biochemical stimuli and thus can be used in permeability studies. The different TEER values in these studies may be because of different cell types (bEND.3 versus hCMEC/D3 and monolayer versus coculture), shear stress (0.023 versus 5.8 dynes/cm²), and time (3 days versus 18 hours).

Another microfluidic-based BBB model originally designed to study the BBB permeability of drugs involves a microhole structure.²⁰⁸ This model is composed of 2 horizontally aligned chambers connected by a microhole structure (Figure 3C).²⁰⁸ Endothelial cells suspended in medium will be infused in the brain chamber and trapped in the microholes because of pressure gradient. By using human umbilical vein endothelial cells and astrocyte-conditioned medium with a shear stress of 0.28 to 8.91 dynes/cm², Yeon et al²⁰⁸ successfully demonstrated that the permeability of this system to fluorescein isothiocyanate-dextran inversely correlated with its size. Additionally, by comparing the permeability of 5 well-known drugs in this system with their in vivo data, this group further validated the reliability of this model, suggesting its application in predicting the CNS permeability of new compounds.²⁰⁸ The TEER measurement in this system, however, has not been reported. Two major concerns of this device are (1) it lacks cell–cell contact, a key feature of the in vivo BBB and (2) it does not replicate the dimensions of microvasculature in vivo. Recently, this microfluidic device has been modified and a new microfluidic-based synthetic microvasculature model of the BBB (SyM-BBB) has been developed.²⁰⁹ This SyM-BBB model contains 2 microchannels separated by microfabricated pillars with 3-μm gaps (Figure 3D).²⁰⁹ The pillars separated by 3-μm gaps mimic the porous membrane in the μBBB model. Endothelial cells will be infused to the blood chamber via ports 1 and 2, whereas astrocyte-conditioned medium or astrocytes can be infused to the brain chamber from port 3. The flow speed of medium in these chambers determines the shear stress. This design better mimics the in vivo microcirculatory system by including the diverging and converging bifurcations.²⁰⁹ It has been reported that astrocyte-conditioned medium significantly increases the expression of TJP in RBE4 cells and decreases their permeability to fluorescein isothiocyanate-dextran in this model.²⁰⁹ TEER values were not available because of the lack of electrodes in the system at this stage. Additionally, efflux activity assay revealed a functional P-glycoprotein efflux system,²⁰⁹ suggesting that the Sym-BBB model can be used in drug discovery and BBB permeability studies. Further work should focus on including other cell types (astrocytes or pericytes) and integrating electrodes for TEER measurement.

Compared with the dynamic in vitro BBB model, these microfluidic in vitro BBB models have many advantages. First, because the thickness of the membrane or pillars is <10 μm,^205,206 they allow transmigration/trafficking studies in a condition that better replicates the in vivo BBB structure. Second, nondestructive microscopy is possible because of the transparency of the materials.^205,206 Third, it takes less time (3–4 days) to reach steady-state TEER.^205,206 Fourth, they only require a small amount of cells and are less demanding in terms of technical skills. These microfluidic models, on the other hand, also have some shortcomings. For example, TEER value is not high enough (250–300 Ω·cm²) in these models. This may be because of the short time in culture (3–4 days). Future work should examine whether the TEER continues to increase at later time points. Additionally, such as the dynamic in vitro BBB model, these microfluidic-based models can only incorporate 2 cell types, given that the membrane and pillars (hollow fibers in the dynamic in vitro BBB model) have only 2 sides. Although there are only limited publications about these microfluidic-based BBB models to date (probably because they are relatively new), they show promising results as dynamic in vitro BBB models.^205,206 With the accumulation of research data, these microfluidic models may be used to aid neurovascular research and new-drug R&D in the future because of its small size, short time to reach steady-state TEER, low-to-moderate technical skill requirement, and low cost. The advantages and disadvantages of these in vitro BBB models are summarized in the Table.