3D Self-Organized Microvascular Model of the Human Blood-Brain Barrier with Endothelial Cells, Pericytes and Astrocytes

Fabrication of the microfluidic device (micro-device/macro-device)

The 3D microfluidic systems were composed of polydimethylsiloxane (PDMS; Sylgard 184; Dow Corning, MI, USA) with a single layer microchannel and two fluid channels, fabricated by soft lithography [36] (Fig. 1c, Supplementary Fig. 1a-b). Elastomer and curing agent were mixed (10:1 volume ratio), degassed and poured onto a silicon master and cured overnight at 60°C. I/O holes were created with biopsy punches, then the device was taped to remove dust and sterilized as previously described [37]. The PDMS micro and macro-devices were treated with oxygen plasma (Harrick Plasma), then bonded to a glass coverslip (Fisher Scientific) coated with poly(D-lysine hydrobromide) (PDL, Sigma-Aldrich) solution (1 mg/ml) and, finally, placed in an incubator for 3 h at 37°C, rinsed 3 times and dried overnight.

Cell culture and device seeding of BBB self-assembled vascular network model

Human iPSC-ECs (Cellular Dynamics International, CDI) were subcultured on flasks coated with human fibronectin (30 μg/ml, Millipore) in vascular medium (VascuLife VEGF Medium Complete Kit, icell media supplement, CDI). Pericytes and astrocytes isolated from human brain (ScienCell), were cultured in growth medium (ScienCell) on a poly-l-lysine (Sigma-Aldrich) coated flask, and maintained in a humidified incubator (37 °C, 5% CO2), replacing the medium every 2 days. Cells were detached using TrypLE (for iPSC-ECs) and 0.025% trypsin/EDTA for other cell types (Thermo Fisher). Experiments were performed between passages 3 and 5 for all cells.

Fibrinogen (6 mg/ml) and thrombin (100 U/ml) from bovine plasma (Sigma-Aldrich) were separately dissolved in sterile PBS. Then, thrombin was mixed with 1ml of EGM-2 MV (Lonza) and placed on ice. Cells were detached and spun down at 1200 rpm for 5 min and cell pellet was resuspended in EGM-2 MV 4 U/ml thrombin. Cell suspension was mixed with fibrinogen (final concentration 3 mg/ml) at 1:1 volume ratio. The mixture was quickly pipetted into the gel filling ports. Devices were placed in a humidified enclosure and allowed to polymerize at room temperature (RT) for 15 min before the fresh medium was introduced to fluidic channels. iPSC-ECs medium was supplemented with 50 ng/ml of vascular endothelial growth factor (VEGF, Peprotech), for the first four days of culture. Medium for the tri-culture condition was supplemented with 1% (vol/vol) Astrocyte Growth factor (AGF, astrocyte growth supplement, ScienCell).

Three different cell combinations were tested: 1) iPSC-ECs mono-culture (6×10⁶ cells/ml), 2) co-culture iPSC-ECs+PCs (add 2×10⁶ cells/ml PCs) and 3) tri-culture of iPSC-ECs+PCs+ACs (add 2×10⁶ cells/ml ACs). After fibrin polymerization, medium channels were coated for 30 min in an incubator (37°C 5% CO2) with human fibronectin (60 μg/ml) to promote endothelial cell adhesion. In each case, iPSC-ECs were subsequently seeded at 2×10⁶ cells/ml in EBM-2 (Lonza) into the fluidic channels to reduce diffusion of fluorescent dyes into the gel. Non-adherent cells were removed after 2 h. The device was kept in an incubator for 7 days (37 °C, 5% CO2), 200 μl of medium was replaced every 24 h. Devices prepared in this manner were used for both permeability measurements and immunocytochemical staining. PC conditioned medium was collected after 3 days, from a T75 flask of PCs culture, mixed 1:1 volume ratio with fresh medium and replaced every 24 h in the iPSC-ECs mono-culture in the microfluidic device.

The in vitro BBB model was developed by co-culturing human iPSC-ECs, and human brain PCs and ACs to mimic certain aspects of the organization and structure of the brain microcirculation observed in vivo (Fig. 1a,b). The BBB model formed by a vasculogenesis-like process, consisted of a well-connected and perfusable μVN in a microfluidic device (Fig. 1c, Supplementary Fig. 1a), interacting via paracrine, juxtacrine and mechanical signaling[38][39]. iPSC-ECs seeded in the side media channels reduced leakage through the side walls of the central gel region and promoted the formation of patent vessel connections to the media channels, facilitating flow into the network (Fig. 1d,e).

Immunocytochemistry and confocal imaging

The 3D BBB μVNs were cultured for 7 days followed by rinsing in PBS and fixation in 4% paraformaldehyde (PFA, Electron Microscopy Sciences) for 15 min at RT. Cell membranes were permeabilized with 0.1% Triton X-100 (Sigma-Aldrich) for 5 min at RT and washed twice with PBS. Primary antibodies (1:100, volume ratio) against CD31, Glial Fibrillary Acidic Protein (GFAP), (Abcam), F-actin (Rhodamine Phalloidin, Molecular Probes), 4',6-Diamidino-2-Phenylindole (DAPI Thermo Fisher Scientific), were used to identify, respectively, iPSC-ECs, ACs, PCs, and nuclei.

F-actin is strongly expressed in all cells present in our model, whereas only iPSC-ECs highly express CD31 and only astrocytes express GFAP. We therefore used double staining of CD31/F-actin to identify iPSC-ECs and GFAP/F-actin to identify ACs, which enabled us to clearly identify the PC population as those cells that only express F-actin.

To characterize the presence of TJs and ECM proteins by immunocytochemistry, primary antibodies were used against: ZO-1 (Invitrogen), occludin, claudin-5, laminin and collagen IV (Abcam). Secondary antibodies (1:200, volume ratio) were anti-rabbit or anti-mouse IgG conjugated with Alexa Fluor (488-555, or 647) (Invitrogen). Detail on primary and secondary antibodies are listed in Supplementary Table 2. Devices were incubated with primary and secondary antibodies overnight at 4°C, placed on a shaker. After PBS washing, devices were imaged using a confocal laser scanning microscope (FMV-1000, Olympus, Japan) (aspect ratio 1024×1024) high resolution images at 10 μs/pixel scan velocity. Phase contrast imaging was used for morphological observations at different culture time-points (Axiovert 200, Zeiss, Germany). Post-processing and stitching for tiled images were performed using Imaris (Bitplane, Switzerland) and Fluoview (Olympus, Japan). Fold change average immunofluorescent (IF) intensity (relative to iPSC-ECs) was calculated by dividing total immunofluorescent intensity by cell boundary length (ZO-1, occludin, and claudin-5) or by vascularized area (laminin, collagen IV). ROIs were selected to contain only microvascular portions such that no part outside the vessels were included in the computations.

Characterization of BBB microvascular parameters

To characterize microvascular parameters, confocal images were analyzed using ImageJ software (http://rsbweb.nih.gov/ij/) and plugins (Trainable Weka Segmentation 3D, 3D geometrical measure). Briefly, raw images were prepared by enhancing contrast and removing noise. An automatic threshold was used to produce binarized images. From 2D projections, lateral vessel area (A_lateral), and total branch length (L_branch) were computed by ImageJ. Percentage of area coverage was calculated dividing A_lateral by the entire area of the region of interest. Taking advantage of the observation that most vessels are oriented in a plane parallel to the glass substrate, lateral diameters, parallel to the glass substrates of the devices, were computed as the ratio of the projected lateral vessel area to the total branch length. Transverse diameters, perpendicular to the glass substrate, were computed using the 3D vessel volume (V) and the surface area of the vessels in 3D (A_surface). Average cross-section area and circularity were computed using lateral and transverse diameters. The sequences of instructions and equations used to compute both diameters, cross-section, lateral and surface areas and circularity are shown in Supplementary methods.

Microvascular network perfusion and fluorescent dextran-based permeability assay

To assess permeability of the 3D BBB model, solutions containing 10 or 40 kDa FITC-dextran (Sigma-Aldrich) were introduced as fluorescent tracers, and time-sequential images to assess leakage through the microvascular barrier were captured. Briefly, after 7 days of culture, each device was moved to the confocal conditioning chamber (37°C, 5% CO₂), culture medium was aspirated from all reservoirs in each side channel. Then, 5 μl of dextran solution in PBS was injected in one side, simultaneously with 5 μl of medium on the other fluidic channel to maintain equal hydrostatic pressures in the device. Confocal images were acquired every 3-5 min for 6 to 8 times to create the entire 3D stack of the gel volume with microvascular formation at each time point. ROIs were selected considering vascular networks with a clear boundary between vessel wall and gel regions.

To assess perfusability, fluorescent tracers (FITC-dextran) were introduced through the microvascular networks by imposing a hydrostatic pressure drop across the gel region between two medium channels. Videos were recorded using NIS-Elements software (NIKON) on a fluorescent microscope (Nikon, TI-E ECLIPSE.) at 30 frames per second.

Quantification of vessel permeability coefficient

The vascular permeability is evaluated as the flux of solute across the walls of the vascular network. Using mass conservation, the quantity of FITC-dextran crossing the vascular network equals the rate at which it accumulates outside the vessels in the tissue gel region. According to a previously described method[40], vascular network permeability, P_v, was quantified by obtaining the average intensity of vessels I_v and tissue (outside vessels) I_T at two different time points t1 and t2 and using:

Here, Δt is the time between two images, V is the tissue volume, A_surface is the surface area of all vessels in the selected ROI, computed based on the assumption that the ratio V/A_surface can be estimated as the tissue area A_lateral divided by the perimeter of the vascular region L_branch in the projected 2D images from the 3D confocal stacks. Diffusion of fluorescent dextran into the gel was minimized by introducing an iPSC-ECs monolayer in both side channels. The fluorescence intensity values, vessel surface area and tissue/gel region area were computed using ImageJ.

RNA isolation and quantitative RT-PCR

Total RNA was isolated from different conditions using TRIzol reagent (Life Science) for dissolving fibrin gel. Reverse transcription was performed using SuperScript VILO cDNA synthesis kit (Invitrogen). Quantitative Real-time RT-PCR (RT-PCR) using SYBR Premix Ex Taq (Takara) or Power SYBR Green PCR Master Mix, was performed with a 7900HT Fast Real-Time PCR System (Applied Biosystems). mRNA of endothelial cell adhesion molecule (PECAM-1) also known CD31, glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and Ribosomal Protein S18 (RPS18) were used as housekeeping genes, set to 100% as the internal standard. RT-PCR experiments were repeated at least 3 times for cDNA prepared from 6 devices. Primer sequences (Integrated DNA technology) are listed in Supplementary Table 1. RT-PCR was performed in a scaled up version a of the device (Supplementary Fig. 1b) in order to collect higher amount of total RNA.

Statistical analysis

All data are plotted as mean ± SD. One-way ANOVA with pairwise comparisons by the Tukey post hoc test was used to determine whether three or more data-sets were statistically significant. Statistical tests were performed using JMP pro (SAS Institutes, Inc.). At least four devices (≥2 regions per device) for each condition within 3 independent experiments were used for the imaging and data analysis. **** denotes p < 0.0001, *** denotes p < 0.001, ** denotes p < 0.01, * denotes p < 0.05. Non-paired student’s t-test was used for significance testing between two conditions.

Optimization of self-assembled microvasculature

Three models were established, as described in Methods, with progressively greater complexity: (i) iPSC-ECs (Fig. 2a,b (i)), (ii) iPSC-ECs + PCs (Fig. 2a,b (ii)), and (iii) iPSC-ECs + PCs + ACs (Fig. 2a,b (iii), Supplementary Figs. 4a-c). In each case, the iPSC-ECs elongated and intracellular intussusception and vacuoles appeared after 1 day followed by the formation of lumen structures after 2-3 days (Supplementary Figs. 2a and 3a,b). Further development of the μVNs resulted in a highly interconnected microvasculature by day 7 of the culture (Fig. 2b).

With iPSC-ECs alone (Fig. 2a,b (i)), vascular networks formed in 4-5 days (Supplementary Fig. 2a), however, the vessels fused, forming large, elliptical cross-section lumens, many of which contacted the bottom coverslip (Fig. 3a) and gradually degraded and regressed after 7 days (Supplementary Fig. 5a). In contrast, co-culture of iPSC-ECs with PCs formed smaller and more highly branched vessels (Figs. 2a,b (ii), 3b). No significant difference could be observed when iPSC-ECs were cultured alone or with PC conditioned medium (Supplementary Fig. 5b), suggesting that contact with PCs effectively facilitated endothelial organization, by stabilizing a mature vasculature with a morphology more similar to that found in vivo.

The addition of ACs further assisted in the development of a complex inter-connected and branched architecture found in native vasculatures (Fig. 2a,b (iii), Supplementary Fig. 3a,b). In tri-culture with ACs, the μVNs exhibited distinctive behavior during formation, with increased tortuosity and vessels extending higher up in the 3D gel (Fig. 3c). A fundamental characteristic of the BBB is the stratified organization of cells around the vessels and their direct contact interactions. In 4 replicates with 10-12 high resolution confocal images, we observed a spontaneous self-organization into multicellular BBB structures. Indeed, PCs (F-actin, red, Fig. 2c) adhered to both sides of the endothelial cell surface, surrounding the vessel (CD31, green, Fig. 2c,e, Supplementary Fig. 6a, Supplementary video 1). For example, tracing the intensity profiles of EC and PC fluorescence (Fig. 2d), F-actin expression was observed outside the vessel, clearly delineating the presence of PCs. These results showed that pericytes partially overlapped the outer surface of the EC layer exhibiting a BBB-like organization. In addition, 3D rendering of vessel bifurcations showed PCs in contact with the endothelium at multiple locations (Fig. 2e). Moreover, direct physical contacts were observed between AC endfeet (Glial Fibrillary Acidic Protein, (GFAP), violet) and the abluminal surface of the brain vessels (CD31, green, Fig. 2f; Supplementary Figs. 6b,c).

Characterization of microvascular parameters

To determine the geometrical changes in the μVNs (Fig. 3a-c, Supplementary Fig.4a-c), lateral and transverse vessel diameter distributions, percentage of image area containing vascular networks, and total branch length were each quantified (Fig. 3d-i). As expected, in the iPSC-ECs+PCs co-culture, the lateral vessel diameters (30 to 100 μm, Fig. 3e (i)) were significantly lower than in mono-culture conditions (50 to 150 μm with a few outliers to 200 μm, Fig. 3d (i)). Lateral diameters were further reduced by adding ACs (most values between 25 and 50 μm (Fig. 3f (i))). The overall transverse diameter distributions were similar for all three conditions, ranging between 10 and 40 μm, and centered around 30 μm (Fig. 3d-f (ii)).

Hence, lumens with nearly circular cross-section and consequently smaller cross-section area and higher circularity (Supplementary Fig. 6d-f) formed in the tri-culture condition (average lateral diameter: 42 ±13 μm, average transverse diameter: 25 ± 6 μm, Fig. 3g), while lumens were flattened and had elliptical cross-section in mono-cultures (average lateral diameter: 108 ± 14 μm, average transverse diameter: 29 ± 10 μm, Fig. 3g), and in co-cultures (average lateral diameter: 64 ± 13 μm, average transverse diameter: 27 ± 7 μm, Fig. 3g).

Moreover, the cumulative average μVN branch length decreased from mono-culture (226 ± 40 μm), to co-culture (179 ± 31 μm), and tri-culture (136 ± 24 μm) conditions, respectively (Fig. 3h), demonstrating a highly complex and intertwined vascular network. Accordingly, the networks with iPSC-ECs, iPSC-ECs+PCs, and iPSC-ECs+PCs+ACs conditions covered progressively less area in the projected image (62%, 42%, and 28%, respectively (Fig. 3i). Indeed, in tri-culture conditions, the μVNs showed improved morphology provided by the co-culture with ACs and PCs, with reduced vessel diameters and average branch length. These results mirror similar observations that have been attributed to the secretion of angiogenic growth factors by PCs and ACs [19][20].

In summary, these results indicate that the networks formed with all three cell types -- iPSC-ECs+PCs+ACs -- contained more stable and shorter vessel branches, with more circular cros-ssections and smaller vessel diameters compared to the other conditions. These networks also exhibited more random interconnections and improved 3D structural orientation into the gel region: such structure is more similar to in vivo vessel morphology [30].

Protein synthesis and gene expression related to blood-brain barrier (BBB)

To analyze whether the engineered 3D brain microvascular model constitutes a functional barrier and exhibits physiological characteristics typical of the BBB present in vivo, we validated and compared protein expression measured by immunocytochemistry assays and quantitative real-time RT-PCR, performed after 7 days. Firstly, immunocytochemistry images of vascular networks obtained under different culture conditions were compared from multiple regions of interest (ROIs) within the vessels. The expression of endothelial-specific junction proteins ZO-1, occludin, and claudin-5 (Fig. 4a-c), and ECM constituents such as laminin (Fig. 4d) and collagen IV (Fig. 4e) was analyzed by confocal microscopy (See also Supplementary fig. 7b-f). Interestingly, the increase of TJ protein expression in μVNs was observed by introducing PCs and ACs (Fig. 4a-c). Therefore, the BBB μVNs obtained by iPSC-ECs+PCs+ACs tri-culture (Fig. 4a (iii)) relatively expressed much higher level of ZO-1, occludin and claudin-5 compared to mono-culture of iPSC-ECs and iPSC-ECs+PCs (Fig. 4a-c). Quantitative analysis of fold change average immunofluorescent (IF) intensity (relative to iPSC-ECs) confirmed qualitative observations (Fig. 4f). Average immunofluorescent (IF) intensity was calculated by dividing the total immunofluorescent intensity by the cell boundary length in each ROI in the case of tight-junction proteins (ZO-1, occludin, and claudin-5). In the case of basement membrane protein deposition (laminin, collagen IV), average IF intensity was calculated by dividing the total IF intensity by the vascularized area in each ROI. ROIs were selected to contain only microvascular portions (Fig. 4f). Continuous cell-cell junctions lining the rhomboidal boundaries of endothelial cells along lumens were observed in co-culture and tri-culture conditions, as demonstrated by the clear delineation of ZO-1 along the cell-cell border (Supplementary Fig. 7a).

Another sign of vessel maturation was the deposition of basement membrane proteins, exhibiting a similar trend to TJ expression. Laminin and collagen IV immunofluorescence intensity (Fig. 4d-f) approximately doubled in the case of the microvascular networks obtained by iPSC-ECs+PCs+ACs tri-culture (Fig. 4d,e (iii)) compared to iPSC-ECs mono-culture (Fig. 4d,e (i)) and was significantly higher than for iPSC-ECs+PCs co-culture (Fig. 4d,e (ii)).

To confirm immunocytochemistry results, total RNA was extracted from the total cell population in the microfluidic device (Fig. 5a) and purified from different conditions at several time points (day 0, 4, and 7). RT-PCR analysis was conducted considering gene markers of TJ proteins, ECM production and several endothelial membrane transporters such as efflux-pumps, passive transports, solute carriers, and receptor-mediated mechanisms. Vessel maturation was investigated in terms of the expression of several markers and proteins, in the case of co-culture and tri-culture conditions, and was compared to the control condition (iPSC-ECs). The mRNA expression of each gene was measured relative to the expression of CD31 and GAPDH (fold change). TJ proteins such as ZO-1, occludin, and claudin-5 were highly up-regulated in the tri-culture condition at day 7 compared to mono-culture and co-culture conditions. Interestingly, the expression of TJ markers in the tri-culture case increased as a function of culture time (Fig. 5b, Supplementary Fig. 10a-e). As expected, GFAP was regulated exclusively in the presence of ACs. PDGFR gene expression was slightly higher in the tri-culture condition while alpha-smooth muscle actin (αSMA) expression was reduced, possibly due to the increased proliferation of iPSC-ECs and PCs stimulated by ACs. Furthermore, basement membrane proteins (collagen IV, laminin) were highly expressed over time in the tri-culture condition compared to the mono- and co-culture cases. In addition, gene expression of several BBB-specific membrane transporters and receptors which exploit several transport mechanisms (passive diffusion, ATP-binding efflux transporter, solute carriers and receptor-mediated transcytosis), such as P-GP, MRP1, MRP4, TF-R, LRP1, LAT-1, GLUT-1, CAT1, MCT1, ABCA1, and BCRP widely increased over time in the tri-culture BBB model (iPSC-ECs+PCs+ACs), compared to iPSC-ECs+PCs and iPSC-ECs microvascular network conditions. Overall after 7 days, the tri-culture condition displayed a constantly increased maturation and upregulation of all examined genes (Fig. 5b, Supplementary Fig. 10a-e, Primer sequences in Supplementary Table 1).

Distinct cell contributions to BBB permeability

The permeability of the microvessels in our BBB μVN models was computed to assess the practical potential to use our system to mimic solute transport in vivo. In all culture conditions, vessels comprising the entire vascular network were well formed and completely perfusable at day 7 (Supplementary Fig. 8d, Supplementary video 2, 3). Permeability coefficients were measured by introducing solutions containing FITC-dextran tracers in the vasculature (10 & 40 kDa), and capturing confocal images at 5 min intervals and computing themed as explained in Methods (Fig. 6a-d, Supplementary Fig. a-c). With side-channels seeded with iPSC-ECs, permeability to 40 kDa FITC-dextran of the μVN obtained under mono-, co-, and tri-culture conditions progressively decreased: 6.6, 2.5, and 0.89 ×10⁻⁷ cm/s, respectively. A similar trend was observed for the 10 kDa FITC-dextran: 12, 4.8 and 2.2 ×10⁻⁷ cm/s, respectively (Fig. 6e, f). When iPSC-ECs were not added to the side channels, leakage of tracer across the side-walls of the gel region gave rise to higher permeability values, roughly a two-fold increase, due to the artifact associated with the additional source of tracer influx. Side channel seeding helped in several ways. It improved coverage of the exposed side gel surface with an endothelial monolayer, filled gaps that sometimes formed at the gel-post borders, and increased the number and patency of connections between the networks and the main channel (Fig. 7a and Supplementary Fig. 9a-e).