Modeling ischemic stroke in a triculture neurovascular unit on-a-chip

Primary human brain microvascular endothelial cells (HBMECs, ACBRI 376, Cell Systems) were cultured in T75 flasks (Nunc™ Easy Flask, Sigma, F7552) in Promocell MV-2 medium (Bioconnect, C-22121). HBMECs were used between passage 4 and 10. iPSC-derived human astrocytes (01434, FujiFilm-CDI) were expanded in T75 flasks (734–2705, Corning) coated with GelTrex (A15696-01, Gibco) in medium comprised of DMEM (31966-021, Gibco) supplemented with 10% fetal bovine serum (FBS, F4135, Sigma), 1 × N2 supplement (17502-048, ThermoFisher), and 1% penicillin/streptomycin (P4333, Sigma). Astrocytes were expanded to P4. iPSC-derived neural stem cells (Ax0018, Axol Bioscience) were expanded in T75 flasks (734-2705, Corning) coated with growth factor reduced Matrigel (Matrigel-GFR, 356231, Corning, 80–100 µg/mL) in expansion medium (see Table 1) for up to 6 passages. Cells were subsequently differentiated on Matrigel-GFR coated 12-well plates (100,000 cells/well) in differentiation medium (see Table 1) for 3 weeks to obtain neurons. Resulting neurons were dissociated using StemPro Accutase (A11105-01, Gibco) for 20 min, after which the cells were cryopreserved in large numbers for subsequent seeding in OrganoPlate. All cells were cultured at 37 °C, atmospheric (20%) O₂, 5% CO₂ and regularly tested for mycoplasm contamination and found negative.

The OrganoPlate 3-lane platform (4004-400B, MIMETAS) was used for all experiments. Channel dimensions are 400 µm × 220 µm (w × h) and phaseguides had dimensions of 100 µm × 55 µm (w × h). Rat-tail collagen-I gel was prepared as previously described [51, 52] and dispensed in the middle lane of OrganoPlate 3-lane tissue chips by adding 2 µL to the gel inlet. After 15 min of gelation at 37 °C, a Matrigel-GFR coating (80–100 µg/mL in cold PBS) was added to the bottom channel of each chip by pipetting 40 µL into the inlet wells. The OrganoPlate was incubated at 37 °C overnight. Next, astrocytes and three-week pre-differentiated neurons were thawed, pelleted, and resuspended in differentiation medium at a density of 15,000 cells/µL, in a 1:4 ratio. 1–2 µL of astrocyte-neuron cell suspension were seeded in the bottom channel of each chip using passive pumping technique [53]. In short, Matrigel-GFR coating was aspirated from the bottom inlet of each chip and replaced with 50 µL of differentiation medium, after which the cell suspension was seeded on the connecting outlet, causing the cells to get drawn into the channel. The OrganoPlate was placed static at 37 °C for 1 h to allow cell attachment, after which medium was aspirated from the bottom inlet of the chips. Differentiation medium was then added to the top inlet and outlet wells (50 µL each) of each chip and the OrganoPlate was placed on the OrganoFlow perfusion rocker (MIMETAS, OFPR-L, 7° inclination, 8-min interval). Medium was changed twice a week by aspirating medium from the inlet and outlet wells of the top channel of each chip and replacing it with fresh differentiation medium. After 7 days, HBMECs (10,000 cells/µL) were seeded to the top channel of each chip using the passive pumping technique [53]. The OrganoPlate was incubated on the side for 3 h in the incubator to allow the HBMECs to sediment against the collagen-I gel and attach. After 3 h, all medium was aspirated from the chips and fresh medium was added. Endothelial cell medium (OrganoMedium, HBMECBM, MIMETAS) was added to the top channel (50 µL in inlet, 50 µL in outlet) and neuronal differentiation medium (see Table 1) was added to the bottom channel (50 µL in inlet, 50 µL in outlet) of each chip. The OrganoPlate was placed back on the OrganoFlow perfusion rocker and culture (37 °C, atmospheric (20%) O₂, 5% CO₂) was continued. Medium changes were performed 2–3 times a week. Assays were performed on day 14–15. A schematic representation of NVU culture in the OrganoPlate is shown in Additional file 1.

Stroke was modeled using a three-fold approach for a duration of 16 h, starting on day 14 of culture. Hypoglycemic conditions were modeled by replacing the culture media on both sides of the chips with glucose-free formulations. OrganoMedium HBMECBM-GF (MIMETAS) was added to the top channel (50 µL in inlet, 50 µL in outlet) of each chip. Neuronal differentiation was prepared as usual (see Table 1), but with Neurobasal-A medium (A2477501, ThermoFisher) instead of glucose-containing Neurobasal medium. Hypoxic conditions were modeled using 10 µM antimycin-A (A8674, Sigma), an inhibitor of complex III of the electron transport chain. Disrupted perfusion was modeled by removing the OrganoPlate from the rocker platform and placing it static in the incubator. Each approach was tested individually and in combination.

Cultures in the OrganoPlate were fixed with 3.7% formaldehyde (252549, Sigma) or 100% methanol (− 20 °C, 494437, Sigma) and immunostained as previously described [51]. In short, cells were permeabilized using a Triton X-100 solution for 10 min and blocked using a buffer containing FBS, bovine serum albumin, and Tween-20 for 45 min. Primary antibody was incubated for 1–2 h or overnight, after which secondary antibody was incubated for 1 h. For negative controls, incubation with primary antibody was omitted and only incubation with secondary antibody was performed. An overview of the antibodies used can be found in Table 2. Nuclei were stained using Hoechst (H3570, ThermoFisher) and cells were imaged with ImageXpress Micro XLS and Micro XLS-C HCI Systems (Molecular Devices).

Cells were incubated with Cal-520 (20 µM, ab171868, Abcam) and 0.4% Pluronic F-127 (P6866, Invitrogen) in serum-free medium for 60 min at 37 °C on the rocker platform, followed by 30 min at RT. Fluorescent images were acquired at 0.5 Hz, 4×, widefield setting, 20% FITC-intensity using the Micro XLS-C HCI System (Molecular Devices). Calcium recordings were corrected for photo-bleaching using a bleach correction plugin in Fiji [54]. For quantification, regions of interest were selected and fluorescent signal over time was plotted for selected individual neurons using Fiji multi measure functionality. Graphs were plotted using GraphPad Prism 8 (GraphPad Software).

Transendothelial electrical resistance (TEER) was measured at different time points using an automated multichannel impedance spectrometer designed for use with the OrganoPlate (OrganoTEER, MI-OT-1, MIMETAS). Before the baseline measurement, 50 µL HBSS was added to the middle inlets and outlets of each OrganoPlate chip. Medium solutions containing staurosporine (S4400, Sigma) or vehicle (0.001% DMSO) were dispensed into a multiwell plate. The OrganoPlate, medium solutions and TEER equipment were equilibrated in the incubator (37 °C, 5% CO₂) for at least 30 min prior to the start of the experiment. The OrganoPlate was placed in the OrganoTEER, allowing electrode pairs to be inserted into all inlet and outlet wells. Point impedance measurements were performed in the incubator (37 °C, 5% CO₂) by frequency sweep from 1000 Hz to 1 MHz (100 points; precision 0.5). A baseline measurement was performed first (1 timepoint only). Medium was then aspirated and replaced with vehicle or staurosporine-containing medium. Time-lapse measurements were then performed every 16 min for a total of 23 h and 45 min (91 timepoints). Data were analyzed using the OrganoTEER software, which automatically extracts the TEER contribution (in Ohm) from the measured spectra and normalizes it to Ohm*cm² by multiplying by the microvessel-ECM interface (estimated at 0.0057 cm²).

Barrier permeability assays were performed as previously described [52]. In short, chips were wetted with culture medium to ensure proper flow profiles and medium was aspirated from the chips. 20 µL of medium without fluorescent compound was added to the basal side of the chips (the gel inlet and outlet well and the bottom inlet and outlet well). Medium containing a fluorescent dye (10 µg/mL sodium fluorescein, F6377, Sigma; or 0.1 mg/mL 20 kDa FITC-dextran, FD20S, Sigma) was perfused through the lumen of the endothelial microvessel in the top channel (40 µL in inlet well, 30 µL in outlet well). Images were acquired from each chip every 2 min for a duration of 10 min using an ImageXpress XLS Micro HCI System (Molecular Devices). The leakage score was calculated as follows: [fluorescent signal in ECM gel channel]/[fluorescent signal in lumen of endothelial vessel]. Apparent permeability (P_app) of each HBMEC vessel was calculated as previously described [55]. Graphs were plotted using GraphPad Prism 8 (GraphPad Software).

RNA was collected from proliferating 2D HBMEC cultures (1 × T75 flask, cultured in MV2 medium) or from the HBMEC fraction of NVU on-a-chip cultures by lysation with RLT buffer (79216, Qiagen). For NVU on-a-chip cultures, lysates from 5 chips were pooled into one sample before proceeding to RNA isolation. RNA isolation was performed using an RNeasy Micro kit (74004, Qiagen) according to manufacturer’s instructions. RNA yield and purity were measured using the NanoDrop OneC Microvolume UV–Vis Spectrophotometer (ND-ONE-W, ThermoFisher). Complementary DNA was synthesized using M-MLV reverse transcriptase (28025013, ThermoFisher) according to manufacturer’s instructions. The following primers were used in this study: Glut-1 (forward: 5′-AACTCTTCAGCCAGGGTCCAC-3′, reverse: 5′-CACAGTGAAGATGATGAAGAC-3′), P-gp (forward: 5′-GGCACCAACATGGACAACC-3′, reverse: 5′-GTATTCCTGGGACACGCAGT-3′), BCRP1 (forward: 5′-AGATGGGTTTCCAAGCGTTCAT, reverse: 5′-CCAGTCCCAGTACGACTGTGACA-3′), MRP1 (forward: 5′-GCCGAAGGAGAGATCATC-3′, reverse: 5′-AACCCGAAAACAAAACAGG), TfR (forward: 5′-CTGCTATGGGACTATTGCTGTG-3′, reverse: CGACAACTTTCTCTTCAGGTC), and ACTB (forward: 5′-CTCTTCCAGCCTTCCTTCCT-3′, reverse: 5′-AGCACTGTGTTGGCGTACAG-3′). Quantitative PCR was performed using FastStart Essential DNA Green Master (06402712001, Roche) on the LightCycler® 96 Instrument (05815916001, Roche), performing each measurement in triplicate (technical replicates). The data was analyzed using the corresponding software according to the ΔΔCq method [56]. In brief, (1) ΔCq, (2) ΔΔCq, and (3) fold change were calculated in Microsoft Excel using the following equations: 1) ΔCq = Cq gene of interest – Cq endogenous control; 2) ΔΔCq = (Cq gene of interest – Cq endogenous control)_{sample A} – (Cq gene of interest – Cq endogenous control)_{sample B}; and 3) Fold change = 2-ΔΔCq. Target genes were normalized to the endogenous control beta-actin. Fold changes for NVU on-a-chip cultures were calculated relative to control cultures (2D HBMEC). Graphs were plotted using GraphPad 8 (GraphPad Prism Software).

P-glycoprotein (P-gp) assays were performed as previously described [55]. In short, calcein-AM (C3099, ThermoFisher), a substrate of P-gp, was perfused through the lumen of the model in presence or absence of P-gp inhibitor cyclosporin-A (30042, Sigma) or zosuquidar (SML1044, Sigma) for 60 min. Next, the chips were washed with cold Opti-HBSS buffer (1:3 mix of Opti-MEM, 31985062, Thermo Fisher and HBSS, H6648, Sigma) and perfused with Hoechst (H3570, ThermoFisher) to stain the nuclei. Z-stacks were acquired using the ImageXpress® Micro Confocal High Content Imaging System (Molecular Devices) and green-fluorescent calcein signal was normalized to Hoechst cell count to quantify the intracellular calcein level in conditions with and without P-gp inhibitor. Graphs were plotted using GraphPad Prism 8 (GraphPad Software).

Tetramethylrhodamine, methyl ester (TMRM, T668, ThermoFisher) was used to determine the mitochondrial membrane potential according to manufacturer’s instructions. Medium was aspirated from all inlets and outlets and 50 µL of 250 nM TMRM staining solution was added to the top and bottom inlet and outlet wells of the chips. The plate was incubated at 37 °C on the rocker platform for 30 min followed by washing with HBSS (55037C, Sigma). Nuclei were stained using Hoechst (H3570, ThermoFisher) and z-stacks were acquired using the ImageXpress® Micro Confocal HCI System (Molecular Devices). SUM projections were loaded into Fiji and a region of interest was selected to extract a mean intensity value from each chip. Background signal was subtracted by deducting the mean intensity from cell-free control chips from each chip. Graphs were plotted using using GraphPad Prism 8 (GraphPad Software).

A CellTiter-Glo® 3D cell viability assay (G9681, Promega) was used according to manufacturer’s instructions. ATP standard curve solutions (0.016-0.08-0.4-2-10-20 µM) were prepared using ATP disodium salt (A7699, Sigma) in HBSS (55037C, Sigma). CellTiter-Glo® 3D ready-to-use reagent was mixed in a 1:1 ratio with HBSS and added to the OrganoPlate chips (50 µL in top and bottom inlets and outlets) for 15 min incubation at 37 °C on the rocker. The ATP standard curve solutions and the lysates from the OrganoPlate cultures were transferred to white, flat bottom 384-well plates (262360, ThermoFisher). Luminescence was measured in duplo for each sample using a multiwell plate reader (Fluoroskan Ascent®, ThermoFisher). ATP concentrations of the OrganoPlate samples were interpolated from the luminescent values following calibration curve fitting in GraphPad Prism 8 (GraphPad Software). Graphs were plotted using the same software.

Data was analyzed using GraphPad Prism, version 8. Gaussian distribution was assessed using the Shapiro–Wilk normality test. In case the assumptions were not violated, one-way ANOVAs were performed. In case of normally distributed data in which the assumption of equality of variances was violated (Figs. 4d, 5a–d), the Brown-Forsythe and Welch test was performed with a Dunnett T3 multiple comparisons test. When normal distribution could not be confirmed (Fig. 4b, left panel graph), the nonparametric Kruskall-Willis test with Dunn’s multiple comparisons test was performed. A 2-way ANOVA with Sidak’s multiple comparison test was used to compare vehicle control data of the two graphs shown in Fig. 4b (with medium composition as variable 1 and presence or lack of perfusion as variable 2). For multiple comparisons tests, all groups were compared to the vehicle control group only. Statistical significance was indicated by one or more asterisks. *P < 0.05, **P < 0.01, ***P < 0.001, or ****P < 0.0001.

The OrganoPlate 3-lane allows parallel culture of 40 miniaturized tissues in microfluidic chips (Fig. 1a, b) [50‐52]. In each chip, a 3D human NVU triculture model was grown under medium perfusion (Fig. 1c, Additional file 1). The NVU cultures were characterized by immunostaining. Figure 1d shows a 3D reconstruction of a representative NVU on-a-chip culture. The culture comprised a vessel of brain endothelial cells, grown against a collagen-I gel, in co-culture with neurons and astrocytes. Cultures remained viable for a minimum of two weeks (Additional file 2).

The vessel of primary brain endothelial cells showed expression of adherens junction proteins vascular endothelial cadherin (VE-cadherin) and platelet endothelial cell adhesion molecule 1 (PECAM-1) as well as tight junction proteins claudin-5 and zona occludens 1 (ZO-1) (Fig. 1e). The presence of these markers and their localization at the cell–cell contacts is indicative of barrier formation [1, 57]. Located on the opposite side of the collagen-I gel were networks of iPSC-derived neurons and astrocytes, positive for neuronal marker beta-III-tubulin (TUBB3) and astrocytic marker astrocytic marker S100 calcium-binding protein B (s100β), respectively (Fig. 1f).

To ensure neuronal functionality, electrophysiological activity was detected by means of calcium imaging [58], confirming spontaneous neuronal firing in the NVU cultures (Additional file 3).

After observing expression of adherens- and tight junction proteins in the NVU on-a-chip cultures, we investigated barrier formation of the endothelial vessel at a functional level, at baseline and in response to staurosporine, an anticancer drug that disrupts BBB integrity and induces apoptosis [59, 60]. NVU on-a-chip cultures were exposed to two concentrations of staurosporine for a duration of 24 h. To assess barrier integrity, TEER was measured over time during the exposure.

The measurements showed a concentration-dependent decrease in TEER in response to staurosporine (Fig. 2a, b). The lower concentration of 0.033 µM resulted in a near-linear decrease in TEER during the first 4 h of exposure before reaching a plateau that showed a 66% reduction in TEER compared to vehicle control. The higher concentration of 0.1 µM staurosporine results in near-zero TEER values after 4 h of exposure, which remained unchanged for the remainder of the experiment.

Following the 24 h exposure to staurosporine, the barrier integrity of each chip was also assessed using a fluorescent assay. Small molecule sodium fluorescein (0.45 nm radius [61]) was added to the lumen of each chip and its leakage into the adjacent gel lane was monitored over time by timelapse imaging and quantification. In line with the TEER measurements, a concentration-dependent decrease in barrier function was observed (Fig. 2c, d), with untreated chips proving leak-tight for sodium fluorescein (P_app of 3.1 × 10^–6 cm/s) and treated chips showing leakage (P_app of 4.2 × 10^–5 cm/s and 1.6 × 10^–5 cm/s for 0.033 µM and 1 µM staurosporine, respectively).

After confirming the barrier function of the NVU on-a-chip model, we assessed expression of relevant transporters. Expression of influx transporter GLUT-1 and efflux transporters P-gp, breast cancer resistance protein 1 (BCRP1), and multidrug resistance protein 1 (MRP1) were confirmed at the RNA level. In addition, we confirmed expression of the transferrin receptor (TfR), which is of interest for drug delivery of biologicals into the brain (Fig. 3a). P-gp expression was similar in NVU on-a-chip cultures compared to HBMECs cultured in 2D. A small upregulation was observed for GLUT-1 and MRP1, while a small downregulation was observed for TfR. Notably, a 24-fold increase was found for BCRP1 (Fig. 3b).

A functional assay was employed to confirm activity of one of these transporters, namely P-gp, an important member of the ATP-binding cassette family and involved in drug efflux from the brain. The lumen of NVU on-a-chip cultures was perfused with calcein-AM, a fluorescent substrate of P-gp [62‐64]. Inhibition of P-gp with first-generation inhibitor cyclosporin-A [65, 66] (10 µM, 60 min) or the more selective third-generation inhibitor zosuquidar [66] (0.4 µM, 60 min) resulted in a significant increase in intracellular fluorescence (65% and 32%, respectively), indicative of inhibited efflux of the fluorescent substrate out of the cells (Fig. 3c).

Next, the NVU on-a-chip cultures were used to investigate the effects of ischemic stroke on NVU function. Stroke was mimicked using a threefold approach: (1) glucose depletion was modeled using glucose-free medium, (2) hypoxia was mimicked using antimycin-A [22, 26], an inhibitor of complex III of the electron transport chain, and (3) halted perfusion was mimicked by removing the OrganoPlate from the rocker platform and placing it static. Each approach was assessed separately and in combination with the other approaches and compared to control chips. Results showed that when perfusion is continued, no-glucose alone or antimycin-A alone do not alter NVU barrier integrity, but the combination of no-glucose and antimycin-A induces a marked disruption of the endothelial barrier, causing leakage of 20 kDa FITC-dextran out of the models’ lumen (Fig. 4a, b, left panels; P_app of 4.9 × 10^–5 cm/s versus 9.3 × 10^–7 cm/s for vehicle control). In contrast, in absence of perfusion, no-glucose alone or antimycin-A alone both did cause a decrease in barrier integrity compared to control. A combination of no-glucose and antimycin-A in absence of perfusion again showed a further disruption of barrier integrity (Fig. 4a, b, right panels; P_app of 3.3 × 10^–5 cm/s versus 2.0 × 10^–6 cm/s for vehicle control).

Similar effects were observed when studying mitochondrial function using a fluorescent mitochondrial membrane potential assay. Under perfusion, no-glucose alone or antimycin-A alone did not significantly alter mitochondrial membrane potential, but the combination of no-glucose and antimycin-A caused a strong decrease (Fig. 4c, d, left panels; 55.6% reduction in fluorescent intensity compared to vehicle control). In absences of perfusion, mitochondrial membrane potential was drastically reduced in all conditions compared to perfused chips, with also control chips (with glucose, without antimycin-A) showing significantly lowered mitochondrial membrane potential (Fig. 4c, d right panels; 29.3% reduction compared to perfused vehicle control, P = 0.0079). In line with results from the barrier integrity assay, mitochondrial membrane potential was also affected in no-glucose only and antimycin-A only conditions if perfusion was stopped. The strongest reduction was again observed when both approaches were combined (80.5% reduction compared to static vehicle control; 86.2% compared to perfused vehicle control).

In stroke, the disrupted blood flow to the brain leads to hypoxic and hypoglycemic conditions and hence a reduction in ATP, the molecule that supplies energy for processes in living cells. Using a CellTiter-GLO assay, we assessed ATP levels in the apical (blood) and basal (brain) side of the NVU on-a-chip cultures after applying stroke conditions.

Under perfused conditions, no-glucose alone or antimycin-A alone resulted in a reduction in ATP levels on the blood side of the chips compared to vehicle control (19.4% and 31.3% reduction, respectively) (Fig. 5a). The combination of glucose removal and presence of antimycin-A resulted in a strong decrease in ATP levels (83% reduction). Samples taken from the brain side of the perfused chips showed higher baseline ATP levels than on the blood side, explained by the larger number of cells present in that compartment. On the brain side of the perfused chips, no-glucose alone did not affect ATP levels (Fig. 5b). Addition of antimycin-A showed a trend of increased ATP, although not significant. Like the blood side, the brain side of the perfused chips showed a strong reduction in ATP when no-glucose and antimycin-A were combined (89.7%).

When perfusion was halted, the blood side of the chip showed similar findings as when perfusion was continued, with combination of no-glucose and antimycin-A showing a strong decrease in ATP concentrations (77.8% reduction) (Fig. 5c). The brain side of the chips again showed higher baseline ATP levels due to a larger number of cells present. In samples taken from the brain side, no-glucose alone or antimycin-A alone did not significantly alter ATP concentrations (Fig. 5d). A combination of the two, however, resulted in strongly decreased ATP (91.1%).