Background
The vasculature of the brain is made up of specialized endothelial cells that form a tight blood–brain barrier (BBB) [
1]. The BBB ensures a homeostatic environment for the brain by controlling the entry of molecules from the circulation. The vasculature of the brain is surrounded by perivascular cells that support and maintain healthy BBB functioning. Among these supporting cells are astrocytes and pericytes, which strengthen the inter-endothelial adherens junctions and tight junctions and maintain proper BBB transport function [
2,
3]. The entire structure contributing to BBB function is referred to as the neurovascular unit (NVU) and includes brain endothelial cells, astrocytes, pericytes, neurons, oligodendrocytes, microglia, and the basement membrane [
4].
The NVU restricts passive diffusion of large, polar substances and potentially neurotoxic molecules into the brain. Only a selection of molecules, such as oxygen and carbon dioxide, can enter freely. Other essential molecules such as nutrients can enter the brain through specialized transporter systems, for example glucose, which enters via the highly expressed glucose transporter 1 (GLUT-1) [
5]. Conversely, more lipophilic molecules and metabolic toxins can be actively removed from the brain through efflux transporters. These efflux transporters include P-glycoprotein (P-gp), breast cancer resistance protein (BCRP), and members of the multidrug resistance protein (MRP) family [
6]. While the NVU’s barrier is essential for healthy brain functioning, it also poses a major challenge for drug delivery into the brain, as many drugs can’t freely enter the brain or are removed by efflux transporters [
7].
NVU dysfunction is observed in many neurological disorders, ranging from neurodegenerative and neuroinflammatory diseases to dysfunction caused by trauma or stroke [
8]. Stroke is the second cause of death worldwide and the leading cause of adult disability [
9,
10]. Strokes are of either hemorrhagic or ischemic nature [
10]. Hemorrhagic stroke is the result of a vessel rupturing and makes up approximately 20% of all stroke cases. The other 80% of strokes are ischemic, resulting from an occlusion of a blood vessel by a thrombus that disrupts blood flow to the brain. The brain has a very high energy demand and is responsible for approximately 20% of the body’s oxygen consumption and 25% of glucose consumption [
11,
12]. For this reason, disrupted blood flow to the brain has detrimental effects.
To date, only one therapeutic agent has been approved for ischemic stroke. Tissue plasminogen activator (tPA) can be administered to dissolve the blood clot and restore blood flow to the brain [
13]. However, tPA can only be administered during a relatively short time window (< 4.5 h), as later administration can lead to hemorrhages resulting in a poor patient outcome [
14]. Furthermore, intravenous tPA administration is often not effective in removing blood clots in the major intracranial arteries, which account for many cases of ischemic stroke [
15]. Recently, several studies have found improved clinical outcome when such cases of ischemic stroke were treated with an alternative approach. Blood clot removal via intraarterial therapy, employing mechanical thrombectomy and/or local delivery of a thrombolytic agent, resulted in improved patient outcome [
16]. However, many stroke patients are not eligible for intraarterial therapy. Moreover, the therapy only allows for a short time window, like intravenous tPA administration, and can give rise to new blood clot formation. To date, treatment of ischemic stroke remains far from optimal.
The reasons for this lack of success in treating stroke are multifactorial, but one factor may be found in the predominant use of animal models in preclinical studies [
17,
18]. While animal models of the brain’s vasculature have proven valuable, they are costly, time consuming, and allow only limited control over experimental conditions. Moreover, animal studies of neurological disease and NVU function often result in limited translational relevance due to interspecies differences, such as differential expression of important BBB transporters and immune signaling molecules [
19‐
21]. For this reason, researchers also studied stroke in vitro, using (1) traditional oxygen–glucose deprivation techniques [
22,
23], (2) chemical methods that inhibit the electron transport chain [
22,
23], such as rotenone, antimycin-A, or sodium azide, and (3) enzymatic methods employing glucose oxidase, catalase, and 2-deoxyglucose [
24,
25]. One study compared all three techniques to model renal ischemia and reported that the use of antimycin-A was most reproducible [
26]. Most in vitro stroke studies were performed using relatively simple models employing immortalized cell lines. The use of more complex in vitro models with improved physiological relevance may aid in finding new therapies for ischemic stroke and other neurological diseases.
The first attempts at in vitro NVU modeling started with cultures of primary brain endothelial cells in traditional two-dimensional (2D) culture systems [
27,
28]. Aiming to improve physiological relevance and complexity, the first models in Transwell were developed [
29,
30]. In this system, brain endothelial cells were cultured on one side of a semi-permeable membrane and supporting cells on the other. Although the Transwell presented a step forward in physiological NVU modeling, the presence of a membrane and the lack of flow and direct cell–cell contact posed limitations.
In response to those limitations, microfluidic platforms made their appearance in the field of NVU modeling [
31,
32]. These platforms make use of tissue culture chips comprising small channels that allow the development of layered three-dimensional (3D) cell cultures under flow [
31]. After early work with NVU models based on hollow fiber apparatuses [
33‐
35], Booth and colleagues developed the first NVU model in a chip using vertically stacked planar structures made from polydimethylsiloxane (PDMS) [
36]. These planar chips held much thinner membranes than the hollow fiber apparatuses, allowing for improved cell–cell contact in co-culture setups. Many others followed similar approaches in subsequent years, using primary cells and cell lines from various species [
37‐
44]. The most recent microfluidic NVU models still show resemblance to the chip reported by Booth et al., but special focus has been placed on all-human models, using primary material [
45], or induced pluripotent stem cell (iPSC)-derived cells [
46,
47], allowing for potential use in personalized therapies. Lyu et al. recently applied such an NVU model to study ischemic stroke, using a complex co-culture of endothelial cells, astrocytes, pericytes, microglia, and neurons [
48].
While many have developed microfluidic platforms for complex NVU modeling, most of these models are very low in throughput and cumbersome to use. There is a large unmet need for higher throughput, more user-friendly platforms that unite microfluidic NVU models with routine experimentation and the possibility of drug candidate evaluation [
49]. We previously reported a BBB model in a microfluidic platform that allows culture of 40 chips in parallel, while being compatible with standard laboratory equipment and automation [
50,
51]. The model comprised immortalized human brain endothelial cells grown against an extracellular matrix (ECM) gel in co-culture with immortalized human astrocytes and pericytes.
We here report a microfluidic human NVU model that incorporates primary brain endothelial cells, in co-culture with iPSC-derived astrocytes and neurons grown under bidirectional, gravity-driven perfusion. To showcase the model’s use in studying NVU dysfunction, we developed a protocol to mimic stroke. Under stroke mimicking conditions, the NVU on-a-chip cultures showed reduced BBB integrity, mitochondrial membrane potential, and adenosine triphosphate (ATP), which are common features of ischemic stroke. In contrast to many other microfluidic approaches, the high-throughput and pump-free nature of the platform used renders this method suitable for routine experimentation. The NVU on-a-chip model can be used for fundamental studies of the NVU in health and disease as well as for evaluation of drug candidates under disease mimicking conditions.
Discussion
We established a triculture NVU on-a-chip model that accounts for many key features of the NVU without the need for cumbersome procedures or long culture times. The NVU on-a-chip model consists of a vessel of primary brain endothelial cells in co-culture with iPSC-derived astrocytes and neurons. The model presents with tight in vivo-like barrier function as observed by the retention of the small molecule sodium fluorescein. Exposure of the NVU on-a-chip model to a known disruptive compound decreased the TEER of the barrier and increased its permeability to sodium fluorescein. This finding indicates the model’s use in assessing BBB-disrupting compounds and potential restorative therapies. In addition to tight barrier function, the NVU on-a-chip cultures also demonstrated spontaneously active neurons and expression of relevant endothelial transporters.
The NVU on-a-chip model presented with an average TEER of 12.6 Ω × cm
2 at baseline. Although TEER measurements can in theory be compared between different culture setups and systems, practice shows otherwise [
67‐
69]. Highly conflicting TEER values have been reported by different studies even when the same cells and culture setups were used. When different culture systems are employed, such as Transwell and microfluidic systems, the discrepancies are even larger. The discrepancies result from a combination of different factors that influence TEER values, including thickness and pore size of membranes, electrode type and position, electrical and mathematical approaches, temperature, surface area and shape, and current line distribution [
70]. For this reason, we recommend that TEER values are compared only within a study rather than between studies. Additional permeability studies using fluorescent molecules and imaging-based readouts can aid in further characterization of a model’s barrier function. In this study, we observed a concentration-dependent loss of barrier function over time using the BBB-disrupting anticancer drug staurosporine [
59,
60]. While exposure to 0.033 µM staurosporine resulted in a 66% reduction in TEER compared to vehicle control, exposure to 0.1 µM staurosporine resulted in a near complete loss of TEER. These findings were supported by a concentration-dependent increase in leakage of fluorescent molecule sodium fluorescein. These findings indicate that TEER measurements in our NVU on-a-chip system can be used to assess disruption of the NVU’s barrier.
In addition to TEER, we assessed barrier function of NVU on-a-chip cultures using sodium fluorescein, a commonly used small molecule dye for studying BBB permeability [
71‐
74]. Hawkins et al. reported that sodium fluorescein is subject to transport by organic anion transporter 3 (Oat3) and MRP2 in rats and therefore may result in an overestimation of a culture’s barrier properties [
75]. Although sodium fluorescein presents with this disadvantage, it presents with several other highly favorable characteristics [
74], including its inability to accumulate inside cells [
76]. We have assessed both sodium fluorescein and lucifer yellow, a dye that has not been reported to be substrate to transport, in HBMEC monocultures and found that the cellular barriers retained both dyes (data not shown). Moreover, the permeability for sodium fluorescein in our NVU on-a-chip cultures (P
app of 3.08 × 10
–6 cm/s) falls within the range of the molecule’s in vivo permeability reported for rat brain microvessels (P
app of 0.11 × 10
–6 [
75] to 2.71 ± 0.76 × 10
–6 cm/s) [
77]), indicating tight barrier formation. We do recommend that like TEER, P
app values are also compared within a study rather than between studies, as they are also subject to large discrepancies in reported values due to biological as well as technical and analytical parameters [
78].
Expression of BBB transporters was confirmed at the RNA level for glucose influx transporter GLUT-1, efflux transporters P-gp, BCRP, and MRP1, as well as for TfR, an important transporter for receptor mediated transcytosis. In the presence of P-gp inhibitors cyclosporin-A [
65,
66] or zosuquidar [
66], accumulation of fluorescent P-gp substrate calcein was observed, indicative of functional P-gp activity [
62‐
64]. We have not performed functional assessment of other relevant transporters. Future work may include functional assessment of BCRP1, which was shown to be upregulated in the NVU on-a-chip model compared to 2D, and GLUT-1, which is reported to become upregulated following ischemic stroke [
79]. Lastly, investigation of the sodium-dependent glucose transporter (SGLT) may be of interest, as some have reported a combined role for GLUT-1 and SGLT in ischemic stroke [
80].
Following previous work with cell lines [
51], the NVU on-a-chip presented here employs primary brain endothelial cells. Potential concerns with the use of primary brain endothelial cells include dedifferentiation and loss of certain characteristic features after removal from their in vivo environment [
81,
82]. In addition, the use of primary cells is subject to donor variation. For this reason, the use of iPSC-derived brain endothelial cells (iBECs) has gained much attention over the last decade. However, recent studies acknowledge that the current protocols for iBEC generation often result in suboptimal cellular phenotypes [
46,
83,
84], showing a predominantly epithelial phenotype and a lack of active transport across the cells. Our experience with iBECs is in line with these reports (data not shown). For this reason, we employed primary human brain endothelial cells in our NVU on-a-chip model. The resulting endothelial vessel shows a relevant phenotype, including expression of relevant BBB transporters and tight barrier formation. For consistency, endothelial cells from only one donor were used in this study. However, we have worked with three different donors without finding obvious donor to donor differences (data not shown), indicating that donor variation does not necessarily pose insurmountable issues. Lastly, the use of primary brain endothelial cells rather than iBECs also allows for strongly reduced culture times. As the field continues to improve iBEC differentiation protocols, the replacement of primary brain endothelial cells by iPSC-derived ones in our NVU on-a-chip model may be possible in the near feature when for example donor matched models are desired for personalized medicine applications.
The OrganoPlate platform allows flexible tissue model design [
85]. In addition to the cell types present in the model described here, one could easily add pericytes, which play a major role in healthy NVU functioning [
3,
51]. The model can also be expanded to include microglia, the resident macrophages of the brain [
86,
87]. Additionally, the role of circulating immune cells may be investigated. Impaired BBB function and inflammation is observed in many acute and chronic neurological diseases and results in the entrance of immune cells from the systemic circulation into the brain [
8,
88,
89]. After entering the brain these immune cells further exacerbate BBB disruption, either directly—via release of inflammatory factors such as cytokines, free radicals and matrix metalloproteinases—or indirectly, via activation of other constituents of the NVU, such as astrocytes or microglia [
90‐
92]. The expression of endothelial cell adhesion molecules (CAMs) such as intercellular CAM-1 (ICAM-1) can be studied in our NVU on-a-chip model at baseline and after mimicking stroke or other neurological disorders. Immune cells can be perfused through the lumen of the endothelial vessel in the OrganoPlate and immune cell adhesion to the vascular wall can be quantified, as reported by Poussin et al. [
93]. Subsequent extravasation and migration towards the brain side of the chip may be studied using an approach similar to the one reported by Gjorevski et al. [
94] or De Haan et al. [
95]. Furthermore, samples can be taken from apical and basal compartments and cytokine contents can be analyzed, as shown in a study by Gijzen et al. [
85].
This study modeled ischemic stroke using a threefold approach that combines hypoglycemia, chemical hypoxia, and halted perfusion. When perfused was continued, omission of glucose alone showed limited effects on NVU barrier function, endothelial mitochondrial membrane potential, or ATP levels on blood- and brain side. This may be explained by a compensatory mechanism, such as a switch to cellular respiration mechanisms that don’t rely on the presence of glucose. Upon low glucose levels, endothelial cells have shown to increase fatty acid oxidation, also known as β-oxidation, for energy production [
96,
97]. When perfusion is halted, delivery of new fatty acids is hampered, possibly explaining why the combination of no glucose and halted perfusion does result in reduced NVU barrier function and mitochondrial membrane potential.
Similarly, when perfusion was continued, chemical hypoxia alone did not strongly affect NVU function in the tested assays. This may be explained by several factors. The concentration of antimycin used in this study likely does not result in a full inhibition of the electron transport chain [
98]. In addition, endothelial cells rely primarily on glycolysis for their energy production rather than mitochondrial respiration [
99] and are therefore less likely to be affected by inhibition of the electron transport chain. While endothelial cells rely primarily on glycolysis, energy metabolism in astrocytes and neurons remains subject to debate [
100]. While neurons contain many mitochondria and likely rely heavily on mitochondrial respiration [
101], it is hypothesized that upon hypoxic conditions, they can switch to glycolysis as a primary source of ATP production: a less efficient, but faster alternative [
100,
102,
103]. This may underly our finding showing increased ATP on the brain side of the NVU on a chip model upon exposure to antimycin-A only. Overall, we found that the combination of low glucose, chemical hypoxia, and halted perfusion resulted in impaired NVU barrier function, reduced mitochondrial potential, and lower ATP levels.
Further studies may investigate other components of the complex cascade of events that follows ischemic stroke, such as excitotoxicity, production of reactive oxygen species (ROS), and reperfusion injury [
104,
105]. Presence of excess glutamate causing excitotoxicity [
106,
107] may be investigated in the NVU on-a-chip model using a fluorescent calcium indicator or by determining glutamate levels in medium samples taken from the brain side of the chip. ROS production [
108] can be studied in the NVU on-a-chip model using fluorescent or luminescent assays, or by measuring the ROS contents of apical and basal medium samples. Reperfusion injury [
109,
110] may be studied by removal of the stroke conditions and addition of glucose-containing medium without antimycin-A and placing the OrganoPlate back on the rocker platform to reintroduce flow.
Many traditional in vitro NVU models do not incorporate fluid flow. However, numerous publications have reported that incorporation of fluid flow in in vitro NVU models is beneficial, showing reduced BBB permeability, decreased cell division, and increased expression of drug and nutrient transporters [
36‐
38,
42,
43,
46,
111,
112]. Although direct in vivo measurements of shear stress in brain vasculature are lacking and vary dependent on local vessel diameter and curvature, it is estimated that capillaries experience shear stress ≥ 6 dyne/cm
2. Endothelial cells in our NVU on-a-chip model experience shear stress of ~ 1.2 dyne/cm
2, falling within the range reported for post-capillary venules (1–6 dyne/cm
2) [
113‐
115]. Furthermore, the flow in the NVU on-a-chip model reported here is of bidirectional nature, unlike cerebral blood flow in vivo, and flow disturbances are associated with diminished vascular health [
115]. By using systems employing pumps and syringes, higher shear stress and unidirectional flow can be achieved in the NVU on-a-chip model presented here. However, this will come at the cost of ease of use and throughput, which is undesirable. While the nature of flow in our system may make the model suboptimal for those research questions requiring full control of all aspects of flow, it has been shown to improve cellular differentiation, polarization, junctional organization, and barrier function compared to static culture [
51,
52], indicating that bidirectional flow still holds advantages over static culture in in vitro modeling. More importantly, the lack of a complex setup employing pumps and syringes makes this model amenable to routine experimentation and automation.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.