Human cerebral micro-vascular endothelial cells (hCMEC/D3) were a gift from Dr. Pierre-Olivier Couraud of the Institut Cochin, INSERM, Paris, France (Weksler et al., 2013). The hCMEC/D3 used for the experiments were between passage 32 and 34. T-flasks were coated with rat-tail collagen type I solution (Thermo Fisher, United Kingdom) at a concentration of 0.1 mg/ml in DPBS (Sigma Aldrich, United Kingdom) and were incubated for 1 h at room temperature. hCMEC/D3 were sub-cultured in Endothelial Basal Medium-2 (EBM-2; Lonza Group Ltd., United Kingdom) supplemented with 0.025%v/v human epidermal growth factor (rhEGF), vascular endothelial growth factor (VEGF), and insulin-growth factor-1 (IGF), 0.1%v/v human fibroblast growth factor (rhFGF), 0.1%v/v gentamycin and ascorbic acid, 0.04%v/v hydrocortisone, and 2.5%v/v Fetal Bovine Serum (FBS) (all supplied by Lonza). Cell culture medium was changed every 2 days. Human 1321N1 astrocytoma cells were purchased from the European Collection of Authenticated Cell Cultures and were below passage 8 for all the experiments. They were grown in High glucose DMEM (Sigma-Aldrich) supplemented with 10%v/v FBS (Life Technologies–Invitrogen), 1%v/v Glutamax (Life Technologies, United Kingdom), and 1%v/v penicillin/streptomycin (Invitrogen, United Kingdom), which will be referred to as “1321N1 media.” Human SH-SY5Y neuroblastoma cells (European Collection of Cell Cultures, ECACC, Sigma-Aldrich) below passage 10 were maintained in 1:1 Minimum Essential Media (MEM; Sigma Aldrich, United Kingdom) and Nutrient Mixture F-12 Ham (Sigma-Aldrich) with 15% FBS (Life Technologies–Invitrogen, USA) and 1% Non-Essential Amino Acids (NEAA 100X, Sigma Aldrich, United Kingdom) and 1% Glutamax 1–100X (Life Technologies, United Kingdom) and 1% penicillin/streptomycin (Invitrogen, United Kingdom). All cells were cultured in a humidified incubator at 37°C and 5% CO₂. The co-culture and tri-culture media was made with EBM-2 (Lonza Group Ltd., United Kingdom) supplemented with 5% FBS (Life Technologies–Invitrogen, USA), 0.1% v/v rhFGF, 1.4 μM hydrocortisone, 5 μg/ml ascorbic acid (all supplied by Lonza), 0.1% v/v chemically modified lipid concentrate (CDLC, Thermo Fisher, United Kingdom), 2.5% v/v HEPES solution (Thermo Fisher, United Kingdom), and 1% penicillin/streptomycin (Invitrogen, United Kingdom). All media was filtered through a 0.2 μm filter before use.

For experiments, cells were cultured in the configurations shown in Figure 1A on 0.4 μm pore Greiner Bio-One ThinCert™ hanging cell culture inserts, which have a 0.33 cm² PET membrane, in 24-well plates. Once hCMEC/D3 cells were seeded, all cells were grown in EBM-2 medium (Lonza) supplemented with 10 mM HEPES, 1 ng/ml basic human FGF (rhFGF), 1.4 μM hydrocortisone, 5 μg/ml ascorbic acid, penicillin–streptomycin, chemically defined lipid concentrate, and 5% FBS to facilitate co-cultures and TJ formation. For the monoculture conditions, hCMEC/D3 cells were seeded on top of the insert membranes at 100,000 cells/cm². For the co-culture conditions, 1321N1 (astrocyte) cells were seeded on either the top or bottom side of the membrane of each insert at 40,000 cells/cm² on the 1st day and then hCMEC/D3 cells were seeded on top of each membrane at 100,000 cells/cm² after 3 days. For the tri-culture conditions, 1321N1 (astrocyte) cells were seeded on the top side of the membrane of each insert at 40,000 cells/cm² and 10,000 SH-SY5Y cells were seeded on the bottom of each corresponding well on the 1st day and then hCMEC/D3 cells were seeded on top of each membrane at 100,000 cells/cm² after 3 days. The cells were cultured for 10 days after the seeding of hCMEC/D3 cells with the media changed every other day, and then further characterization was carried out. The tri-culture model was then set up in the same way described above and validated 14 days after the seeding of hCMEC/D3 by studying the effects of the addition of LPS from Escherichia coli O111:B4 (Sigma-Aldrich) at 100 ng/ml, sodium butyrate (NaB; Sigma-Aldrich) at 100 μM, or RA (Sigma-Aldrich) at 10 μM to the cell culture media for 24 h using an EVOM Volt-Ohm meter.

For each well, TEER was measured every day using an EVOM^®Volt-Ohm meter [World Precision Instruments (WPI), Sarasota, FL, USA] endowed with STX electrodes. The measured values were blanked using read-outs from hanging cell culture inserts in media, but with no cells. During measurements, the longer electrode was kept outside the Transwell insert within the well and the shorter one inside for probing the BCMECs monolayer. Readings were repeated trice to provide reproducible measurements. Cell-free collagen pre-coated inserts were also measured and used as control in the TEER baseline calculations. On day 15 of cell culture, TEER was measured upon 24 h incubation with 100 ng/ml LPS, 100 μM NaB and 10 μM RA.

Paracellular permeability was evaluated using the fluorescent marker, Lucifer yellow (LY, Lucifer yellow CH, lithium salt, Thermo Scientific). A transport buffer composed of 1%v/v HEPES solution added to Hank’s Balanced Salt Solution (HBSS) (1×, Gibco) was added to wells of a 24-well plate and placed in a humidified incubator at 37°C, 5% CO2. The hanging cell culture inserts were then rinsed with the transport buffer and then placed in the 24-well plate with the transport buffer. LY solution (300 μM) was then added to the apical side of the inserts (0.3 ml per insert) and the plate was incubated at 37°C and 5% CO₂ for 1.30 h. Samples taken from both the apical and basolateral sides of each insert were then analyzed using a Tecan Spark multimode microplate reader to record fluorescence values (excitation/emission wavelengths at 428/536 nm). A standard curve was then used to estimate the LY concentration of each sample and then apparent permeability (Papp, cm/s) was calculated using:

LY_A and LY_B are the concentrations of LY in the apical and basal compartments, respectively. Vol_A and Vol_B are the volumes in the apical and basal sides, respectively. T (seconds) is the time of incubation. C is the initial concentration of LY on the apical side (300 μM) and A is the area of the membrane.

Cells were washed with Dulbecco’s–Phosphate Buffered Saline (PBS) having Calcium and Magnesium (Sigma-Aldrich) at the beginning and in between steps. They were fixed with 4% paraformaldehyde by incubating at room temperature for 5 min. The permeabilization was done through incubation in 0.25% Triton in PBS for 10 min at room temperature and the blocking step was carried out by incubating cells in 1% BSA in PBST (0.05% Tween 20 in PBS) for 30 min at room temperature. A 1:50 rabbit polyclonal anti-ZO 1 (617300, Invitrogen, United Kingdom), 1:400 rabbit monoclonal VE-Cadherin (D87F2, Cell Signalling Technology, United States), and 1:400 rabbit polyclonal anti-glucose transporter GLUT1 (ab15309, Abcam, United Kingdom), were used as primary antibodies in 1% BSA in PBS and added to the samples, which were then incubated overnight at 4°C. Cells were then incubated for 1 h at room temperature with 1:500 of Alexa Fluor 488 Goat anti-Rabbit IgG (Invitrogen, United Kingdom) in 1% BSA in PBS. Lastly, the samples were incubated for 5 min at room temperature with 1 μg/ml Hoechst 33 342 (Sigma Aldrich, United Kingdom), mounted, and examined with a confocal microscope (ZEISS LSM 800).

A Molecular Probes™ Neurite Outgrowth Staining Kit (Life Technologies, United Kingdom) was used according to the manufacturer’s protocol for measuring the neuron viability and neurite outgrowth of SH-SY5Y cells in both co-culture and tri-culture models. Briefly, 10 μl of Cell Viability Indicator and 10 μl of Cell Membrane Stain were added in 10 ml of PBS containing calcium and magnesium (1× working Stain Solution). After incubation for 20 min at 37°C and 5% CO_2, cells were washed once with PBS and 1× working Background Suppression Dye (10 μl per 1 ml PBS) was added to the wells for confocal imaging (ZEISS LSM 800) and fluorescence quantification using a Tecan Spark microplate reader. For confocal microscopy, the Cell Viability Indicator dye settings were: excitation 495 nm and emission 515 nm; for Cell Membrane Stain: excitation 555 nm and emission 565 nm. Plate reader settings: Cell Viability Indicator: excitation 483 nm and 525 nm emission, bandwidth 10 nm; Cell Membrane Stain: excitation 554 nm and 567 nm emission, bandwidth 5 nm. Quantitative readouts were obtained from three replicates for each condition except for the BMECs-neurons dataset where only 2 datapoints were generated out of 3 (“over” readout). Neurite outgrowth analysis was performed using the semi-automated ImageJ plugin, NeuronJ. For each condition (i.e., BMECs-neurons, astrocytes-neurons, and tri-culture), three confocal images were used for analysis and 90 neurites were manually traced for quantification.

Analysis of statistical differences was performed, specifically: One-way ANOVA analysis was used for TEER analysis and comparison. Dunnett’s multiple comparison test was used for the TEER values comparison between the control tri-culture model (untreated) vs. the treated samples; Tukey’s multiple comparison test was used to compare % TEER change between each condition. Statistical analysis was performed with GraphPad Prism9.0.0 (121). Data were represented as mean of three replicates ± standard deviation.

To generate a biomimetic model of the human NVU, the following human immortalized cell lines were cultured: hCMEC/D3 cells as representative of brain microvascular endothelial cell (BMECs), 1321N1 cells for astrocytes and SH-SY5Y cells for neurons. The schematics of the NVU in vivo architecture and in vitro models developed in our study are shown in Figures 1A,B. Using Transwell systems, we developed 3 different model setups with increasing level of complexity and biomimicry, namely: (1) a monoculture model (i.e., BBB), the most reductionist configuration where BMECs were located in the apical side of the insert (Figure 1B, left); (2) a co-culture setup in which BMECs were seeded on the apical side along with 1321N1 seeded either on the underside of the insert (i.e., “indirect” contact co-culture, Figure 1B, center-left) or BMECs seeded on top of 1321N1 cells on the apical side of the insert (i.e., contact co-culture, Figure 1B, center-right). The establishment of these two co-culture conditions was aimed to test changes in barrier formation and its properties depending on the type of physical contact between BMECs and astrocytes (i.e., intimate and direct cell-cell interaction in the contact co-culture); lastly, (3) a tri-culture setup with both BMECs and 1321N1 growing in contact on the inside of the insert and SH-SY5Y cells seeded on the bottom of the well plate (Figure 1B, right). Although the tri-culture model is lacking other important NVU cell types (e.g., pericytes, microglia, oligodendrocytes), it comprises the main components of a functional NVU, such as the endothelial barrier representative of the brain vasculature, neural cells for emulating the brain tissue and glial cells acting as mediators between the two modules. Thus, the tri-culture best recapitulates the physiological and structural composition of the human NVU compared to the monoculture and co-culture layouts. Indeed, it has been previously reported that the crosstalk and exchange of growth factors between each cell type are enhanced and promoted in such multicellular microenvironment than in more simplified and reductionist single cell type models (Schreiner et al., 2022). Following optimization of the cell culture media composition, cell density, and cell seeding strategies we established a 10-day cell culture protocol where each cell type of the model was seeded in the respective compartment at a specific time point of the study (Figure 1C).

Transendothelial electrical resistance (TEER) measurement is an extensively used parameter to quantitatively and non-invasively evaluate barrier formation and stability in in vitro barrier model systems (i.e., endothelium or epithelium monolayers) (Srinivasan et al., 2015). In our study, TEER measurements were performed daily in all NVU model configurations 24 h post-endothelial cell seeding (i.e., Day 1, Figure 1C). As reported in Figure 2A and Supplementary Figure 1, each NVU model configuration is characterized by different initial TEER values which undergo fluctuations over the cell culture timeline. However, from day 8 of cell culture a significant increase in TEER was observed in the tri-culture setup, reaching plateau at day 10 (∼ 40 Ω cm², ^****p < 0.001). Lower TEER values were found in the other model configurations (i.e., monoculture ∼20 Ω cm², “indirect” contact co-culture >20 Ω cm², “contact” co-culture >10 Ω cm²), Figure 2B. These findings are in line with previous studies showing similar, increased TEER values when endothelial cells are co-cultured with other cell types of the NVU (Ito et al., 2019; Stone et al., 2019; Gericke et al., 2020). It is worth mentioning that the BBB in vivo has been estimated to reach higher TEER values which are difficult to reproduce in vitro (>1000 Ω⋅cm²) (Sivandzade and Cucullo, 2018; Bhalerao et al., 2020; Erickson et al., 2020); in addition, most of the in vitro BBB model developed to date exhibit different ranges of TEER values depending on the cell source in use [i.e., iPSCs, primary cells or rodent-derived cells, (Srinivasan et al., 2015)], the cell culture conditions and timelines. Following previous studies (Elbakary and Badhan, 2020), we also tested the effects of shear stress on the monoculture barrier tightness with respect to the static condition (Supplementary Figure 2), noting that induced shear stress has beneficial influence on barrier formation and tightness in hCMEC/D3 derived BBB. As a simplified means of applying shear stress, we used the orbital rotation approach by setting the rotator to 150 rpm (i.e., ∼7–8 dyn/cm² of shear) as described in previous studies (Barichello, 2019). However, the remaining cell types of the model (i.e., 1321N1 and SHSY5Y cells) did not exhibit a healthy phenotype when subjected to orbital shaking, which is likely explained by the fact that within the NVU microenvironment the bloodstream is only interfacing with endothelial cells of the BBB, whereas astrocytes and neurons are not exposed to its physical cues. Thus, the application of flow induced shear stress was not carried forward to the optimized model.

Taking into consideration the reductionist approach of a tri-culture model of the NVU, we demonstrated that the co-presence of astrocytes and neurons within the biological model promotes barrier formation and enhances its integrity with respect to the barrier monolayer (Figures 2A,B).

Next, permeability assays were carried out for each biological model at day 10 of the protocol, as an endpoint following the final TEER recordings. In the case of the tri-culture setup, this timepoint was considered optimal for testing the barrier permeability given the higher TEER value (∼40 Ω cm²) should have corresponding low permeability values. It is well-known that BBB endothelial cells exhibit very limited paracellular permeability to hydrophilic molecules, due to the formation of TJs. BBB permeability was assessed by measuring the flux of a small fluorescent marker across the barrier generated, known as LY (MW = 457 Da). Both the LY flux (%) and the apparent permeability (Papp) were calculated in each condition for comparison. Although there is no statistical significant difference in barrier permeability between the monoculture and the other model configurations, we observed lower permeability values (<1 × 10^–6 cm/s) in all the setups with respect to the cell-free control and similar to those reported in most of the studies using hCMEC/D3 as barrier model (Figures 2D,E; Poller et al., 2008; Eigenmann et al., 2013; Zhao et al., 2019). Interestingly, a decrease in Papp was also found when both the monoculture (i.e., ≤5 × 10^–7 cm/s dynamic vs. >6 × 10^–7 cm/s static) and tri-culture (i.e., ∼7.1 × 10^–7 cm/s dynamic vs. ∼8.4 × 10^–7 cm/s static) were exposed to shear stress (Supplementary Figure 3), supporting the theory of physical cues like shear stress as enhancers of barrier strength and tightness.

It is worth noting that the use of different cell lines, cell density, Transwell membrane coating, membrane porosity and cell culture media composition are all factors contributing to different TEER and paracellular permeability values across NVU in vitro models generated in similar studies (Ito et al., 2019; Stone et al., 2019; Bhalerao et al., 2020; Gericke et al., 2020; Simöes Da Gama and Morin-Brureau, 2022). From day 5 of cell culture we observed an overgrowth of astrocytes in both the co-culture and tri-culture setups, meaning that the endothelial cells sitting on top of the astrocytes could be prevented from forming an even and flat barrier monolayer, resulting in a wavy and groovy cellular monolayer. Although the TEER values seem to not be affected by such a phenomenon, this could explain the slightly higher apparent permeability found in these multicellular model configurations with respect to the monoculture where cells adhere directly on the insert porous membrane (Figures 2C–E).

Next, the monoculture, co-culture and tri-culture setups underwent optical characterization using confocal microscopy to visualize the expression of key biomarkers of the BBB. Despite the fact that the “indirect” contact co-culture exhibits higher (yet not statistically significant) TEER compared to the monoculture, there is almost no difference in permeability between the two setups. It is worth mentioning that within the human NVU a close physical interaction exists between the BBB and the surrounding pericytes/astrocytes (Iadecola, 2017; Bhalerao et al., 2020; Kadry et al., 2020) (i.e., end-foot in proximity of the endothelial cells); this scenario is not fully recapitulated by the “indirect” contact co-culture condition, where the Transwell membrane separates the apical endothelial monolayer from the astrocytes underneath. Thus, the “contact” co-culture setup (i.e., endothelial cells are seeded on top of astrocytes in inserts) was the only co-culture approach continued for further experiments and constituted the rationale for the development of the tri-culture model configuration. As shown in Figure 3, all setups showed a strong expression of relevant BBB markers involved in regulating barrier functional properties, specifically: TJs like ZO-1, which contribute to the formation of a highly selective and tight separation between the vasculature and brain compartments; adherens junctions like VE-cadherin, whose role is to ensure adhesion between endothelial cells and maintenance of the brain microenvironment; and finally influx/efflux transporters such as the glucose transporter GLUT-1, for glucose uptake. It is worth nothing that ZO-1 was expressed on the border of the cells, however, the staining across all the conditions also showed a cytoplasmic expression of this biomarker (Figures 3A–C), as observed elsewhere (Adriani et al., 2017; Biemans et al., 2017). In addition, VE-Cad expression pattern seemed to be uneven in the co-culture and tri-culture setups, yet this biomarker was found consistently expressed along the perimeter of the cells (Figures 3D–F).

The selective and restrictive nature of the NVU toward foreign substances and pathogens is mostly dictated by the BMECs, which work in concert with other ancillary cells (e.g., pericytes, astrocytes, and neurons) in maintaining the brain homeostasis. Previous studies on advanced multicellular in vitro NVU models (Ito et al., 2019; Stone et al., 2019; Gericke et al., 2020) together with our findings reported above on TEER have demonstrated the effects of the cellular compartment interplay (i.e., vascular and neuronal) on barrier integrity. Compared to simplistic monoculture BBB models, sophisticated in vitro setups incorporating multiple NVU cell types seem to encourage or improve barrier properties, such as barrier integrity and resistance. However, very limited investigation has been carried out on the influence of a healthy BBB on neuronal behavior and phenotype. To this end, we developed a co-culture model of BMECs and neurons (i.e., without glia cells) and a co-culture model of astrocytes and neurons (i.e., without BBB) to be compared with the complete tri-culture NVU model (Figure 4A). Our aim was to evaluate any discrepancies in neural health and morphology among these conditions that can be inferred as BBB driven. By means of specific membrane and intracellular fluorescent dyes, which allow the simultaneous detection of viable neurons and their neurites in the sample (Kaur et al., 2012), we firstly assessed the direct influence of the BBB in the BMEC-neuron model on neuronal cell viability and neurite outgrowth. As reported in Figure 4B and Supplementary Figure 4, there is higher neural cell viability (level of green indicator dye) compared to the tri-culture setup, whereas the measured axonal lengths (relative fluorescence units, RFU) are comparable between samples. Similar results were obtained with the astrocyte-neuron model, in which the cell viability of the neurons was detected to be slightly lower than the BMEC-neuron setup yet closer to the tri-culture (Figure 4C). Interestingly, fluorescent images acquired for each cell model condition showed that neural cells of the tri-culture NVU model seem to possess a high number of neuronal processes and branched neurites (orange indicator dye) forming dense interconnected neural networks (Figure 4A and Supplementary Figure 4). All together, these findings may suggest that the presence of an endothelial barrier within the in vitro cell model strongly enhances neural health and viability, thus having beneficial influence on inducing neural networks formation and maturation. The decrease in the neural cell viability fluorescent signal found in both astrocytes-neurons and tri-culture may be due to the very high proliferation rate of the immortalized astrocytoma cell line used in our models, which likely took over the neuronal culture and partially affected the formation of a flat barrier monolayer in the tri-culture setup (see Section “Conclusion”). Finally, we further traced and estimated neurite projections in all the three cell model configurations to convert RFU in μm (see Section “Materials and methods” and Supplementary Figure 5). As reported in Figure 4D, we measured neurite length means of 27.53 ± 8.5 μm (range 9.248–54.642 μm) in the BMEC-neuron model, 30.14 ± 8.9 μm (12.4–52.4 μm) in the astrocyte-neuron co-culture and 31.17 ±6.9 μm (17.2–53.4 μm) in the tri-culture.

Blood-brain barrier (BBB) disruption or dysfunction generally correlates with lower TEER and greater permeability as ions can leak through the barrier tissue (Williams-Medina et al., 2021). Numerous substances, such as microbiota derived metabolites, soluble factors in the media or extracellular matrix (ECM), have been known to affect BBB integrity and thus permeability. For instance, 10 μM RA increased the TEER of iPSC-derived BBB models by more than 1,000 Ω cm² (Butt et al., 1990; Sivandzade and Cucullo, 2018; Bhalerao et al., 2020). Conversely, BBB disruption occurs in response to pro-inflammatory stimuli, such as LPS and tumor necrosis factor–α (Lazear et al., 2015). Moreover, in recent years, the emerging area of gut-brain-microbiota research has revealed that gut bacteria can also influence BBB permeability by producing metabolites, such as Short Chain Fatty Acids (SCFAs) like NaB, which decreases BBB permeability and prevents BBB disruption after cerebral ischemia and LPS exposure (Braniste et al., 2014; Hoyles et al., 2018). Therefore, RA, LPS, and NaB were firstly chosen as soluble factors for validating the human NVU models developed herein by studying the effects of these soluble factors on the BBB integrity (Figures 5A,B).

In this experiment the tri-culture setups were used for barrier integrity assessment at day 14 of cell culture, following barrier steady-state which occurred from day 10 onward (i.e., TEER values reaching plateau). As illustrated in Figures 5A,B after 24 h of treatment with 10 μM RA, TEER increased by over 60% from 31.75 Ω cm² (Control) to 51 Ω cm² (^****p < 0.0001). This is expected, as all-trans-retinoic acid (RA) is a metabolite of vitamin A and a well-known antioxidant, known to enhance the barrier properties of brain endothelial cells by regulating junctional proteins like occludin (Tóth et al., 2014). It was decided to treat the cells with 100 μM of NaB because that is close to the SCFAs potency range and human plasma concentration (Cleophas et al., 2019; Villabona-Rueda et al., 2019). After 24 h, this led to a 35% rise in TEER to 43 Ω cm² (^****p < 0.0001). This is not very different from a previous, similar study that also found that 24 h of treatment with 200 μM of NaB reduced the BBB permeability of brain endothelial cells in hanging inserts (Kim et al., 2021). It has been reported that butyrate restores the BBB, affects BBB permeability and has neuroprotective effects in murine models (Kim et al., 2021). On the contrary, LPS reduced TEER by almost 50% to just 16 Ω cm² (^****p < 0.0001). In another model, 24 h of LPS (100 ng/ml) treatment increased 7 kD FITC-Dextran permeability and reduced TEER by 15.85% (Ni et al., 2017), whereas this TEER reduction was between 30 and 50% in another study (Kim et al., 2021). As LPS did not cause morphologic changes in a previous study, its mechanism for increasing BBB permeability is not well understood. LPS, which can be found on gram-negative bacteria cell walls, is an endotoxin known to disrupt and penetrate the gut barrier in addition to impairing BBB integrity and being associated with brain diseases. Whilst there was no significant difference in LY permeability of the control compared to the cells treated with LPS, LPS seemed to cause a slight variability in LY Papp as seen in Figure 5C, whilst both RA and NaB treated samples exhibit a reduced Papp with respect to the control and the LPS condition.