|Home | About | Journals | Submit | Contact Us | Français|
The blood-brain barrier (BBB) plays an important role in brain health and is often compromised in disease. Moreover, as a result of its significant barrier properties, this endothelial interface restricts neurotherapeutic uptake. Thus, a renewable source of human BBB endothelium could prove enabling for brain research and pharmaceutical development. Herein, we demonstrate that endothelial cells generated from human pluripotent stem cells (hPSCs) can be specified to possess many BBB attributes, including well-organized tight junctions, expression of nutrient transporters, and polarized efflux transporter activity. Importantly, hPSC-derived BBB endothelial cells respond to astrocytic cues yielding impressive barrier properties as measured by transendothelial electrical resistance (1450±140 Ωxcm2) and molecular permeability that correlates well with in vivo brain uptake. In addition, specification of hPSC-derived BBB endothelial cells occurs in concert with neural cell co-differentiation via Wnt/β-catenin signaling, consistent with previous transgenic studies. This study represents the first example of organ-specific endothelial differentiation from hPSCs.
The blood-brain barrier (BBB) is composed of specialized brain microvascular endothelial cells (BMECs) that help regulate the flow of substances into and out of the brain. Complex intercellular tight junctions limit the passive diffusion of molecules into the brain and result in blood vessels exhibiting extremely high trans-endothelial electrical resistance (TEER) in vivo1. In addition, efflux transporters such as p-glycoprotein contribute to the barrier properties by returning small lipophilic molecules capable of diffusing into BMECs back to the bloodstream. As a result, BMECs are endowed with a requisite network of specific transport systems to shuttle essential nutrients and metabolites across the BBB. In addition, because of its substantial barrier properties, the BBB has significantly hampered neuropharmaceutical development by preventing uptake of the majority of small molecule pharmaceuticals and essentially all biologics2. Conversely, BBB breakdown and dysfunction is associated with a variety of neurological diseases, including Alzheimer’s disease, stroke, multiple sclerosis, and brain tumors3. These issues have collectively led researchers to develop a variety of BBB models to enable detailed mechanistic studies and drug screens in vitro.
Most in vitro BBB models have been established using brain microvessels isolated from primary animal sources such as cow, pig, rat, and mouse4. However, as a result of inevitable species differences5, 6, a robust in vitro BBB model of human origin would be of high utility for conducting high-throughput screening for brain-penetrating molecules or for study of BBB developmental, regulatory, and disease pathways in humans. Previously, human BBB models have been established by culturing primary human BMECs isolated from autopsy tissue, or more often, freshly resected brain specimens derived from tumor or epilepsy patients. As a result, issues involving BMEC availability and fidelity limit the universal use of these human BBB models7. Another proposed route toward a human BBB model has been cell immortalization8. However, immortalized BMECs suffer from poor barrier properties, including low baseline TEER8-10 and discontinuous tight junction protein expression8. Thus, to create a robust, scalable human BBB model, we sought to take advantage of the in vitro developmental potential of human pluripotent stem cells (hPSCs). hPSCs, including both human embryonic stem cells (hESCs)11 and induced pluripotent stem cells (iPSCs)12, 13, exhibit virtually unlimited self-renewal and the capacity to differentiate into somatic cell types from all three embryonic germ layers. While human endothelial cells (ECs) have been generated from hPSCs by a variety of methods, including embryoid body differentiation14-18 and OP9 stromal cell co-culture19, 20, ECs are known to develop distinct gene and protein expression profiles that depend on microenvironment cues during organogenesis21, and hPSC-derived ECs with organ-specific properties have yet to be reported. Here we report a facile hPSC differentiation method capable of reproducibly generating pure populations of endothelial cells possessing BBB properties.
In vivo, BBB specification begins as ECs forming the perineural vascular plexus invade an embryonic brain microenvironment comprised of neuroepithelial cells, radial glia, neuroblasts, and neurons, and notably much of this early BBB induction occurs in the absence of astrocytes22-25. The cells of the developing embryonic brain provide relevant BBB induction cues, such as Wnt7a and Wnt7b, that signal through the canonical Wnt/β-catenin pathway to help promote BBB specification of ECs26, 27. We therefore hypothesized that a strategy which simultaneously co-differentiates hPSCs to both neural and endothelial lineages could result in hPSC-derived ECs possessing BBB attributes. Recent studies have shown that both neural and endothelial cells can be generated from pluripotent stem cells via directed differentiation in adherent culture28-30. Thus, we developed a custom 2-dimensional hPSC differentiation strategy that promotes neural and endothelial co-differentiation, in essence providing an embryonic brain-like microenvironment in vitro. The strategy was initially implemented using the IMR90-4 iPSC line13 and protocol robustness was subsequently validated with multiple iPSC and hESC cell lines (Fig. 1a). Briefly, IMR90-4 iPSCs were expanded on Matrigel-coated plates in defined mTeSR1 medium for 2-3 days (Fig. 1a). To initiate neural and endothelial co-differentiation, the colonies were subjected to unconditioned medium (UM) for 3-4 days, at which time large numbers of cells expressing nestin (Fig. 1b [panel i], green), a marker expressed in immature neural progenitors, and βIII tubulin (Fig. 1b [panel i], red), a neuronal marker, were observed. During IMR90-4 iPSC differentiation, 86% of entire cell population could be classified as potential neural progenitors (23% nestin+/βIII tubulin−), immature neurons (58% nestin+/βIII tubulin+) or neurons (5% nestin−/βIII tubulin+), representing a largely embryonic brain-like microenvironment (Fig. 1c). A similar distribution was observed during H9 hESC differentiation (Supplementary Fig. 1a). Within this large immature neural population, very small clusters of cells exhibiting endothelial morphology and PECAM-1 expression, but lacking characteristic BBB tight junctions began to form (Fig. 1d [panel i] and Fig. 1e).
At day 6 of UM treatment, significantly more IMR90-4-derived nestin−/βIII tubulin+ neurons were present (26%) while the nestin+/βIII tubulin+ population decreased to 29% (Fig. 1b [panel ii] and Fig. 1c). By contrast, only small clusters of cells expressing glial fibrillary acidic protein (GFAP; astrocyte marker) and α-smooth muscle actin (α-SMA; potential pericyte/smooth muscle marker) on the order of 30-50 cells per 106 could be identified by immunocytochemistry (Supplementary Fig. 2), suggesting that these cells were unlikely to play a major role in BBB specification. Coincident with the maturing neural population at 5-7 days UM treatment, the endothelial regions became larger and more prevalent, and BBB glucose transporter GLUT-1, tight junction proteins occludin and claudin-5, and p-glycoprotein were co-expressed in the endothelial population (Fig. 1f). Although peripheral endothelia are known to express one or more of these markers, the composite set of markers is quite restrictive to brain endothelium31. Therefore, the ECs that express tight junction proteins, p-glycoprotein and elevated levels of GLUT-1, will be referred to as hPSC-derived brain microvascular endothelial cells (hPSC-derived BMECs), with further vascular and BBB validation discussed in Figures 3 and and44.
The population of hPSC-derived BMECs was quantified by flow cytometry. For this analysis, BMECs were defined as cells having both PECAM-1 expression and elevated GLUT-1 expression when compared to the low basal GLUT-1 expression exhibited by cells during the first four days of differentiation (GLUT-1+/PECAM-1+, see Materials and Methods for gating details). For IMR90-4-derived BMECs, this population comprised about 30% of the cultures at day 6 of UM treatment (Fig. 1g and Table 1). Thus, the neural and endothelial populations combined represented approximately 95% of the total differentiating culture at day 6 of UM treatment. This hPSC-derived BMEC population was further expanded for 2 days in a custom EC medium that included factors known to facilitate primary BMEC growth with some selectivity (basic fibroblast growth factor (bFGF) and platelet-poor plasma derived serum32), and the percentage of GLUT-1+/PECAM1+ cells in the differentiating IMR90-4 culture increased to 66% (Fig. 1d [panel ii], Fig. 1g 6D UM 2D EC, and Table 1). The BMEC differentiation process yielded on average 11.6 PECAM-1+/GLUT-1+ cells per input hPSC after 6 days in UM and 2 days in EC medium, indicating proliferation of EC progenitors and/or ECs during differentiation. At this point, all GLUT-1+/PECAM1+ cells co-expressed the requisite BBB markers (Supplementary Fig. 3). Moreover, in EC medium, the hPSC-derived BMECs exhibited a commensurate increase in BBB properties as indicated by substantially increased expression of GLUT-1 protein (Table 1). In contrast, if instead the cultures were grown for two additional days in UM (8 days UM) rather than in EC medium, the percentage of IMR90-4-derived BMECs increased to a lesser extent although the total PECAM-1+ population increased to similar levels (Fig. 1g and Table 1), revealing the importance of the EC medium treatment for the selective BMEC expansion and conferral of BBB phenotypes on the EC population. Similar results were observed for the additional iPSC33 and hESC11 lines tested (Fig. 1a), with cell line-dependent BMEC differentiation efficiency (Supplementary Figs. 4-6 and Table 1). For the H9 hESC line, extended culture (7 days UM, 6 days EC medium) was required to improve BMEC yield from 16% to 35% of the population (Fig. 1g and Table 1). Thus, in general, 2-dimensional differentiation and expansion of hPSCs for 6-7 days in UM followed by 2-6 days in EC medium generated substantial yields of highly enriched hPSC-derived BMECs.
Murine in vivo and in vitro studies have demonstrated that canonical Wnt/β-catenin signaling is necessary for the onset of brain angiogenesis and the acquisition of BBB properties such as GLUT-126, 27 and claudin-5 expression34. Canonical Wnt ligands Wnt7a and Wnt7b have been specifically implicated in BBB development in vivo and are expressed by neural progenitors26, 27. Therefore, we assayed for WNT7A and WNT7B expression in the neural progenitor/neuron populations and monitored the temporal nuclear localization of β-catenin in the developing hPSC-derived BMEC population. Nearly all nestin+ cells and βIII tubulin+ cells (i.e. all three developing neural populations evaluated in Fig. 1c) expressed WNT7A and WNT7B transcripts at day 4 of UM treatment (Fig. 2a, Supplementary Fig. 1b-j, and Supplementary Fig. 7a). While junctional β-catenin in PECAM-1+ ECs was widespread, nuclear β-catenin was sparse at this time (7±4% of PECAM-1+ ECs; Fig. 2b [panel i]). By day 6 of UM treatment when large percentages of the culture had adopted a BMEC phenotype, the majority of the nestin+ and βIII tubulin+ cells maintained expression of WNT7A, except for some bipotent βIII tubulin+ cells (Supplementary Fig. 7b-e), while the WNT7B transcript was no longer detected. The percentage of PECAM-1+ ECs exhibiting detectable nuclear β-catenin substantially increased to 40±6% at 5 days of UM treatment (Fig. 2b [panel ii]) and nearly all PECAM-1+ ECs (90±6%) possessed nuclear β-catenin after 6 days UM and 2 days EC treatment (Fig. 2a [panel iii] and Supplementary Fig. 8 for other hPSC lines). Of note, elevated expression of GLUT-1 was nearly exclusively detected in cells that also contained nuclear β-catenin (Fig. 2b [panel ii]), which is consistent with in vivo reports of an absence of BBB GLUT-1 in endothelial-specific β-catenin knockout mutants26 and GLUT-1 downregulation in the vascular plexus of WNT7A/7B double knockout mutants27.
Given that all cell lines tested exhibited nuclear β-catenin localization during BMEC differentiation, linkage to Wnt-mediated processes was also assessed by evaluating transcript expression of Wnt receptors and target genes during IMR90-4 differentiation (Fig. 2c). Gene expression levels in later (7D UM) cultures containing ~30% PECAM-1+/GLUT-1+ BMECs and early (3D UM) cultures devoid of these cells were compared. As expected, transcripts indicative of BMEC differentiation, ABCB1 (p-glycoprotein) and SLC2A1 (GLUT-1), were upregulated during this time period (Fig. 2c). Wnt receptors Frizzled4 (FZD4) and Frizzled6 (FZD6) have been implicated in angiogenesis of retinal35-37 and brain ECs26, with the FZD6 transcript being highly expressed in adult brain ECs compared with lung and liver ECs26. Transcripts of Frizzled4 and Frizzled6 were upregulated in concert with the emergence of IMR90-4-derived GLUT-1+/PECAM1+ BMECs in the differentiating cultures, whereas the FZD7 gene, which encodes Wnt receptor Frizzled7 that has no known linkage to brain endothelial differentiation26, 27, was downregulated during this same time frame. The gene products for β-catenin-associated transcription factor LEF1 and Wnt-downstream gene FST (encoding for follistatin) were also upregulated during this time. In addition, the STRA6 gene, which encodes a BBB-resident vitamin A transporter38 that has been identified as a Wnt target gene39 and is enriched in adult brain ECs compared to lung and liver ECs26, was upregulated during the course of BBB differentiation. APCDD1, which has been noted to be highly enriched in adult brain ECs compared to lung and liver ECs26 and was recently shown to be an antagonist of Wnt signaling40, remained unchanged over the time course of analysis, potentially suggesting a more prominent APCDD1 role in adult BBB maintenance through Wnt pathway regulation rather than in BBB development. Treatment with the Wnt signaling inhibitors secreted Frizzled Receptor Protein 2 (SFRP2) or XAV-939 resulted in downregulation of LEF1, FST, STRA6, FZD6, SLC2A1, and ABCB1 (Fig. 2c), further supporting the involvement of Wnt signaling in regulating these transcripts. Moreover, differentiating IMR90-4 cultures treated with XAV-939 exhibited a reduction in GLUT-1+ cells from 64% to 46% (Fig. 2d and Table 1). Treatment with XAV-939 also led to a slight reduction in the overall percentage of PECAM-1+ cells (68% to 61%), thus giving a 19% net reduction in the overall percentage of PECAM-1+ ECs that became PECAM-1+/GLUT-1+ BMECs. Interestingly, the presence of XAV-939 did not alter claudin-5 or occludin localization even in GLUT-1− ECs, and western blots indicated equal amounts of claudin-5 and occludin in the differentiating cultures regardless of XAV-939 treatment (data not shown). These data compare favorably with observations in endothelial-specific β-catenin knockout mouse mutants where brain vascular malformations lacked GLUT-1 expression but still possessed tight junctions26. Thus, while certain aspects of BMEC specification appear to be influenced by the Wnt-β-catenin pathway (e.g. elevated GLUT-1 expression), this pathway is likely not exclusively responsible for BMEC specification41-43.
Although the 2-dimensional co-differentiation strategy resulted in large numbers of hPSC-derived BMECs as defined by expression of tight junction proteins, p-glycoprotein, and elevated GLUT-1, many characteristic BBB properties, including barrier formation and transporter activity, are best evaluated in purified, confluent monolayers. Furthermore, RT-PCR analysis indicated that the IMR90-4-derived BMECs, while expressing the PECAM1 transcript, lacked expression of endothelial transcripts encoding for von Willebrand factor (vWF) and VE-cadherin during the early UM treatment phase (Fig. 3a). During EC medium treatment, VE-cadherin expression was detected (Fig. 3a), consistent with sequential expression of PECAM-1 and VE-cadherin endothelial genes previously observed during EC differentiation from stem cells44. Thus, to stimulate further maturation and facilitate purification, IMR90-4-derived BMECs were subcultured from Matrigel onto plates coated with collagen/fibronectin extracellular matrix commonly used for primary BMEC culture32. In EC medium, the cells grew to confluence after 1-2 days, were contact-inhibited, exhibited characteristic EC morphology (Fig. 3b) and were capable of acetylated low-density lipoprotein uptake (Fig. 3c). Approximately 100% of the cells in the purified cultures expressed PECAM-1 and the requisite BBB markers as judged by flow cytometry and immunocytochemistry (Fig. 3d and 3e), indicating effective and facile purification using this extracellular matrix-based selective passaging method. Importantly, the cells also acquired vWF and VE-cadherin expression during subculture, indicating EC maturation (Fig. 3a and 3e), and vascular character of these BMECs was further confirmed by formation of tube-like structures in the presence of vascular endothelial growth factor (VEGF) (Fig. 3f). The extent of BBB maturation achieved during UM and EC medium treatment prior to subculture onto the fibronectin/collagen matrix was critical for obtaining this pure, confluent monolayer of BMECs. For example, when IMR90-4-derived BMECs were subcultured after just 4 days of UM treatment, cells expressing BBB markers still developed, but did not reach confluence and in many areas showed malformed and discontinuous tight junctions (Fig. 3g). This lack of BBB differentiation correlated well with the observation that at this stage of UM treatment, minimal β-catenin nuclear localization could be found (Fig. 2b [panel i]) and no PECAM-1+/claudin-5+ cells were present (Fig. 1e). Similar purification results were found for DF19-9-11T-derived BMECs when subcultured onto the collagen/fibronectin matrix (Supplementary Fig. 9 and Table 1), while DF-6-9-9T-derived BMECs (Supplementary Fig. 10) could also be purified although they exhibited some discontinuous junctions (Supplementary Fig. 10c). H9-derived BMECs could not be purified away from the large percentage of non-brain PECAM-1+/GLUT-1− ECs by the collagen/fibronectin protocol given the general affinity of ECs for the collagen/fibronectin matrix. The IMR90-4 and DF19-9-11T lines were therefore used for in-depth BBB phenotype testing as described below.
In addition to BBB marker expression and vascular phenotype, iPSC-derived BMECs should also respond to astrocyte cues, exhibit tight barrier properties, and express functional transport systems. A hallmark of the BBB is the high trans-endothelial electrical resistance (TEER) that is a consequence of tight junction protein interactions between adjacent BMECs. To measure the TEER for the BMEC monolayers, IMR90-4-derived BMECs were seeded onto Transwell filters coated with collagen/fibronectin matrix and grown to confluence in EC medium to create a standard two-compartment BBB model (Fig. 4a). Immunofluorescent detection of ZO-1, occludin, and claudin-5 demonstrated the maintenance of continuous cell-cell contacts between BMECs on the filter surface, similar to the tight junctions observed on fibronectin/collagen-coated polystyrene culture dishes (Supplementary Fig. 11). Initial TEER measurements taken at confluence were 150-175 Ωxcm2, indicative of a tightening endothelial monolayer (Fig. 4b, time 0h). TEER measurements following co-culture with either primary rat astrocytes or non-neural human embryonic kidney (HEK) 293 cells in EC medium were compared to monocultured IMR90-4-derived BMECs. After 24 hours, TEER in the astrocyte co-cultures (412±38 Ωxcm2) nearly doubled that observed in both HEK cell co-culture (236±23 Ωxcm2) and BMEC monoculture (222±51 Ωxcm2) and remained elevated up to 96 hours, indicating a specific response to astrocyte cues as expected of BMECs (Fig. 4b and Table 2). If astrocyte co-culture was instead performed in medium containing 10% FBS, TEER of the IMR90-4-derived monolayer reached 696±8 Ωxcm2 after 24 hours, while TEER in the HEK cell co-cultures was again significantly lower at 364±53 Ωxcm2 (Table 2). Optimization of seeding density to generate a more uniform monolayer yielded IMR90-4-derived BMECs with a maximum TEER of 1450±140 Ωxcm2 upon astrocyte co-culture (Table 2). DF19-9-11T-derived BMECs exhibited a maximum TEER of 777±112 Ωxcm2 during astrocyte co-culture (Table 2). The iPSC-derived BMEC monolayers typically exhibited maximum TEER between 24-48 hours of co-culture and maintained elevated TEER levels up to 8 days (Supplementary Fig. 12). In terms of robustness, iPSCs of differing passage number consistently yielded TEER in a range of 700-1450 Ωxcm2 over the course of thirty individual experiments, with an average value of 860±260 Ωxcm2. For comparison, one of the highest documented TEER values for primary human BMECs obtained from fresh biopsy is 339±107 Ωxcm2 (ref. 45) and the immortalized human BMEC cell line, hCMEC/D3, reaches a maximum TEER of 199±5 Ωxcm2 in response to hydrocortisone9, whereas peripheral endothelium exhibits a TEER in the 2-30 Ωxcm2 range.
To further validate the fidelity of the tight junctions formed by the IMR90-4-derived BMECs after co-culture with astrocytes, these cells were subjected to freeze-fracture electron microscopy (Fig. 4c). Freeze-fracture revealed networks of tight junction strands whose complexity mimicked that of high resistance BBB endothelium in vivo and the tight junction particles were primarily associated with the protoplasmic fracture face (P-face), favorably contrasting with lower resistance in vitro BBB models that oftentimes undergo a switch of tight junction particles from the P-face to the E-face46.
The BBB in vivo is not only characterized by its high TEER, but also by its impermeability to passive diffusion, expression of various transport systems used for import and export of nutrients and metabolites, and its ability to act as an active barrier to small hydrophobic molecules by employing key efflux transporters like p-glycoprotein, breast cancer resistance protein (BCRP), and members of the multidrug resistance protein (MRP) family. Purified IMR90-4-derived BMECs co-cultured with astrocytes expressed transcripts encoding a variety of receptors/transporters found at the BBB (Fig. 4d), such as LDLR (low density lipoprotein receptor), LRP1 (low density lipoprotein receptor-related protein 1), INSR (insulin receptor), LEPR (leptin receptor), BCAM (lutheran glycoprotein), TFRC (transferrin receptor), and AGER (receptor for advanced glycation endproducts, RAGE). Transcripts were also detected for members of amino acid and peptide transporter families that are highly enriched at the BBB compared to other endothelium, including STRA6 (retinol binding protein), SLC2A1 (GLUT-1), SLC7A5 (LAT1), SLC1A1 (EAAT3), SLC38A5 (SNAT5), and SLC16A1 (MCT1). Efflux transporter transcripts were also detected, including ABCB1 (MDR1/p-glycoprotein), ABCG2 (BCRP), ABCC1 (MRP1), ABCC2 (MRP2), ABCC4 (MRP4), and ABCC5 (MRP5). Of note, two assayed transcripts were not detected: PLVAP and SLC21A14. PLVAP (plasmalemma vesicle-associated protein) is initially expressed at the BBB during development and becomes downregulated with onset of barrier properties while its expression remains in peripheral vessels throughout adulthood24. Thus, the absence of PLVAP in the IMR90-4-derived BMECs is a further indicator of BBB specification and maturation. SLC21A14 (also known as SLCO1C1) encodes Oatp14, an organic anion transporter whose transcript and protein product is highly enriched at the rodent but not human BBB47, 48. Taken together, the gene expression profile of receptors and transporters is highly representative of the BBB.
Next, radiolabeled small molecules with varying sizes, lipophilicity, and efflux transporter recognition were screened for relative permeability using the IMR90-4-derived BMEC/astrocyte co-culture model (Fig. 4e and Supplementary Table 1). The measured permeability (Pe) values showed excellent correlation (R2=0.98) to in vivo rodent brain uptake measured by in situ brain perfusion49. Importantly, sucrose permeability (Pe=3.4×10−5 cm/min), which is often used to benchmark the passive barrier of in vitro models, was on the order of high-fidelity animal BBB models (10−4-10−5 cm/min)49, 50 and significantly lower than the hCMEC/D3 line (Pe=1.65×10−3 cm/min)8. Glucose, a small (180 Da), hydrophilic molecule that is actively influxed across the BBB by the GLUT-1 transporter, accordingly demonstrated a 7-fold higher permeation compared to sucrose (Pe=2.2×10−4 cm/min). Inulin, a large, hydrophilic, polysaccharide polymer, showed similar permeation to sucrose (Pe=2.9×10−5 cm/min). Diazepam, a very lipophilic small molecule that demonstrates high BBB penetration in vivo because it is not an efflux transporter substrate, showed the highest in vitro permeation (1.1×10−3 cm/min). Colchicine and vincristine, which are also lipophilic small molecules but have limited brain uptake in vivo since they are substrates of p-glycoprotein and MRP family members, were accordingly found to have lower Pe values in vitro (9.2×10−5 and 6.2×10−5 cm/min, respectively). Prazosin, which is effluxed by BCRP but also thought to be influxed by an organic cation transporter, showed slightly higher Pe (2.9×10−4 cm/min) than colchicine and vincristine, correlating to its in vivo ranking. Importantly, the 40-fold dynamic range of Pe values (diazepam/sucrose) indicates the potential utility of the iPSC-derived BBB model as a drug screening tool, although future testing with a broader spectrum of compounds will be needed to fully validate the predictive power of the model.
While the relative exclusion of the aforementioned lipophilic small molecules like colchicine, vincristine, and prazosin implies functionality of the relevant BBB efflux transporters described in Figures 3e and and4d,4d, we also assessed their functionality and polarity using selective inhibitors (full compound details found in Supplementary Table 2). First, p-glycoprotein function in monocultured IMR90-4-derived BMECs was probed using rhodamine 123, a cell-permeable, fluorescent p-glycoprotein substrate. There was a 1.4-fold increase in cellular accumulation of rhodamine 123 (decrease in efflux) in the presence of cyclosporin A, a known p-glycoprotein inhibitor, but not when Ko143 or MK 571, inhibitors of BCRP and the MRP family respectively, were used, indicating the presence of functional p-glycoprotein in IMR90-4-derived BMECs (Fig. 4f [panel i]). When doxorubicin, a substrate for p-glycoprotein, BCRP, and the MRPs, was added to monocultured IMR90-4-derived BMECs, its accumulation was increased in the presence of cyclosporin A, Ko143, or MK 571, demonstrating the functional efflux activity of each class of transporter (Fig. 4f [panel i]). Next, directional transport assays using the IMR90-4-derived BMEC/astrocyte co-culture model were used to demonstrate preferential efflux function in the brain-to-blood direction (basolateral to apical). For p-glycoprotein, transport of rhodamine 123 from the apical to basolateral chamber was 2.3-fold higher in the presence of cyclosporin A, but not Ko143 or MK 571, as a result of inhibited efflux in the basolateral to apical direction (Fig. 4f [panel ii]). Conversely, transport of rhodamine 123 in the basolateral to apical direction slightly decreased upon treatment with cyclosporin A as a result of blocking p-glycoprotein extrusion of rhodamine 123 at the apical interface (Supplementary Fig. 13). Transport of doxorubicin from the apical to basolateral chamber was increased 1.3-fold, 1.3-fold, and 1.2-fold by cyclosporin A, Ko143, and MK 571, respectively, indicating net functional polarization of p-glycoprotein, BCRP and MRP family efflux transporters (Fig. 4f [panel ii]). Together, these data demonstrate both function and polarization of efflux transporters. Importantly, the degree of polarization, as measured by increased flux in the presence of inhibitors, is consistent with that found in primary cultured in vitro rodent BBB systems49. Overall, the composite of BBB gene and protein expression, tight junction fidelity, correlative permeability for small molecules, and polarized efflux transporter activity benchmarks the iPSC-derived BMECs as possessing significant BBB character.
Traditional methods of primary animal and human BMEC culture are employed by a limited number of laboratories because of their complexity. Moreover, primary BMEC culture is hampered by yield constraints. Immortalized cell lines can alleviate yield concerns but are instead limited by their diminished BBB phenotype. In contrast, the approaches described here for generating hPSC-derived BMECs capable of forming robust BBB models are of comparative ease and readily scalable given the efficiency of endothelial differentiation (>60%) and the capability to perform facile extracellular matrix-based purification. By comparison, other hPSC differentiation strategies utilizing embryoid bodies, OP9 co-culture, and 2D differentiation with endothelial factors have lower reported EC differentiation efficiencies (1-43%), and must be coupled with antibody-assisted purification methods to yield pure populations of ECs14-20, 29, 30. Also, the differentiation process reported here resulted in a high yield of 11.6 PECAM-1+/GLUT-1+ BMECs per input hPSC after 8 days of differentiation whereas recent publications using OP9 co-culture and embryoid body/directed differentiation methods have reported yields of 0.6 and 7.4 ECs, respectively15, 19. Thus, the stem cell basis for this model presents a highly scalable source of human BMECs that, for example, can easily be scaled to generate sufficient cells for thousands of filters for drug screens. In addition, none of these previous approaches has reported an EC population possessing organ-specific properties. In this study, BBB specification occurred in the presence of co-differentiating neural cells, which likely supplied many of the necessary cues normally provided by the embryonic brain microenvironment in vivo24, 26, 27, 34, 41-43. It is possible that co-differentiation strategies could also specify ECs of other lineages by providing organ-specific microenvironmental cues during development in vitro.
Given that BMEC populations were obtained from both hESC and iPSC lines derived via distinct reprogramming strategies, this hPSC-derived BMEC model could be readily adopted by the broader research community for studies of brain development, disease mechanisms, and drug delivery. The in vitro BBB specification process appears to involve Wnt/β-catenin signaling which is consistent with in vivo mechanisms of brain development demonstrated in mice26, 27, 34. This model could therefore enable the elucidation of other molecular mechanisms leading to human BBB differentiation, studies which is intractable in humans. Information regarding human BBB development could be clinically relevant for promoting BBB repair after stroke or inhibiting recruitment of blood vessels by brain tumors. Having shown the hPSC-derived BMECs possess excellent barrier characteristics with appropriate molecular exclusion and functional transport systems, it is anticipated this cellular platform will also be useful in drug permeability screens to either develop pharmaceuticals that can reach the brain or to limit the brain uptake of drugs that may have neurotoxic side effects.
Human embryonic stem cells (H9)11 and induced pluripotent stem cells (iPS(IMR90)-4 (ref. 13), iPS-DF19-9-11T33, and iPS-DF6-9-9T33) were maintained on irradiated mouse embryonic fibroblasts in standard unconditioned medium: Dulbecco’s Modified Eagle’s Medium/Ham’s F12 containing 20% Knockout Serum Replacer (Invitrogen), 1× MEM nonessential amino acids (Invitrogen), 1 mM L-glutamine (Sigma), 0.1 mM β-mercaptoethanol (Sigma), and human basic fibroblast growth factor (bFGF, 4 ng/mL for hESCs and 100 ng/mL for iPSCs; Waisman Clinical Biomanufacturing Facility, University of Wisconsin-Madison). Prior to differentiation, cells were passaged onto Matrigel (BD Biosciences) in mTeSR1 medium (STEMCELL Technologies). After 2-3 days in mTeSR1 (ref. 51), medium was switched to unconditioned medium lacking bFGF (referred to as UM throughout the manuscript) to initiate differentiation. Major morphological changes were observed by day 5-7 of UM treatment, at which point the medium was switched to endothelial cell (EC) medium: human Endothelial Serum-Free Medium (Invitrogen) supplemented with 20 ng/mL bFGF and 1% platelet-poor plasma derived bovine serum32 (PDS; Biomedical Technologies, Inc.). Following 1-2 days of EC medium treatment, cells were dissociated with dispase (2 mg/mL; Invitrogen) and plated onto 12-well tissue culture polystyrene plates or 1.12 cm2 Transwell-Clear® permeable inserts (0.4 μm pore size) coated with a mixture of collagen IV (400 μg/mL; Sigma) and fibronectin (100 μg/mL; Sigma). Culture plates were incubated with the coating for at least 30 min at 37 °C, while the inserts were incubated for a minimum of 4 h at 37 °C. One well of differentiated hPSCs from a standard 6-well tissue culture plate (9.6 cm2) could be used to seed either three wells of a collagen/fibronectin-coated 12-well plate (11.4 cm2) or four collagen/fibronectin-coated inserts (4.48 cm2). Cells were then cultured in EC medium until they reached confluence (typically 1-2 days). Over the course of dozens of differentiation and purification experiments, multiple lots of PDS (5 lots), Knockout Serum Replacer (at least 3 lots), Matrigel (at least 3 lots), collagen IV (new batches every 2-3 months), and fibronectin (new batches every 2-3 months) were used with no observable effects on differentiation efficiency or BMEC barrier fidelity. In addition, it is important to note that none of the aforementioned materials were qualified or prescreened for their capacity to promote efficient differentiation.
For co-culture experiments, primary astrocytes were isolated as previously described52. Briefly, cortices were isolated from P6 neonatal Sprague Dawley rats (Harlan) and minced in Hank’s Balanced Salt Solution (HBSS; Sigma). This tissue was digested in HBSS containing 0.5 mg/mL trypsin (Mediatech, Inc.) in a 37 °C shaker bath for 25 min, followed by digestion in HBSS containing 114 U/mL DNase I (Worthington Biochemical) in a 37 °C shaker bath for 5 min. After trituration and filtration, cells were cultured on collagen-I-coated flasks (100 μg/mL; Sigma) in DMEM containing 10% qualified heat-inactivated fetal bovine serum (FBS; Invitrogen), 10% heat-inactivated horse serum (Sigma), 2 mM L-glutamine, and 1% antibiotic-antimycotic (Invitrogen). Human embryonic kidney 293 cells (HEK cells; ATCC) were cultured in DMEM supplemented with 10% FBS, 1 mM sodium pyruvate (Sigma), 2 g/L sodium bicarbonate (Fisher Scientific), 30 mM HEPES (Sigma), and 1% antibiotic-antimycotic, and used as a non-neural cell control. Co-culture of hPSC-derived BMECs was initiated with primary rat astrocytes or HEK cells in either EC medium (called 1% PDS medium in Table 2) or 70:30 (v/v) DMEM/F12 (Sigma/Invitrogen) supplemented with 1% antibiotic-antimycotic, 2% B27 (Invitrogen), and 10% FBS52 (called 10% FBS medium in Table 2). Trans-endothelial electrical resistance (TEER) measurements were performed using an EVOM voltohmmeter (World Precision Instruments) at the start of co-culture and every 24 h thereafter. The resistance value (Ωxcm2) of an empty filter coated with collagen/fibronectin was subtracted from each measurement. To determine the Pe of radiolabeled ligands, compounds were diluted to 0.4 μCi in transport buffer (distilled water with 0.12 M NaCl, 25 mM NaHCO3, 3 mM KCl, 2 mM MgSO4, 2 mM CaCl2, 0.4 mM K2HPO4, 1 mM HEPES, and 0.1% bovine serum albumin [BSA; Sigma]) and 0.5 mL were added to the upper chamber. 200 μL aliquots were extracted from the basolateral chamber every 15 min and replaced by fresh transport buffer. The rate of accumulation of radioactive ligand in the basolateral chamber over the course of 1 h was used to calculate Pe values for [14C]-sucrose, [3H]-inulin, [3H]-colchicine, [3H]-diazepam, [3H]-prazosin, [14C]-glucose, and [3H]-vincristine. [3H]-vincristine was purchased from American Radiolabeled Chemicals, while all other radiolabeled compounds were acquired from PerkinElmer. All compound incubations were conducted at 37 °C, and the radioactive permeability experiments were carried out on a rotator. Triplicate filters were used for all permeability studies.
P-glycoprotein, BCRP, and MRP functionality were assessed using rhodamine 123 (Sigma), a preferred substrate for p-glycoprotein, and [14C]-doxorubicin (PerkinElmer), a substrate for all aforementioned efflux transporters. To assess activity, hPSC-derived BMEC monolayers cultured on polystyrene (absent astrocyte co-culture) were pre-incubated for 30 min on a rotator at 37 °C with or without 5 μM cyclosporin A (Sigma), 1 μM Ko143 (Sigma), or 10 μM MK 571 (Sigma), which are inhibitors of p-glycoprotein, BCRP, or various MRPs, respectively. BMECs were then incubated with rhodamine 123 (10 μM) or doxorubicin (0.25 μCi) for 1 h on a rotator at 37 °C with or without inhibitors. Cells were then washed three times with ice-cold PBS and lysed with 5% Triton X-100 (TX-100; Fisher). Fluorescence (485 nm excitation and 530 nm emission) was measured using a plate reader and normalized to cell counts obtained using a hemacytometer, while radioactivity was measured using a liquid scintillation counter. To quantify apical-to-basolateral transport, hPSC-derived BMEC monolayers on Transwell filters were co-cultured with astrocytes for 24 h and then pre-incubated with or without inhibitors for 60 min, followed by addition of rhodamine 123 or doxorubicin to the upper chamber. After another 60 min, aliquots were extracted from the bottom chamber and transport was quantified on a plate reader or scintillation counter. To quantify basolateral-to-apical transport, hPSC-derived BMEC monolayers on Transwell filters were pre-incubated with or without cyclosporin A for 60 min, followed by addition of rhodamine 123 to the lower chamber. After 3 h, aliquots were extracted from the upper chamber and fluorescence quantified on a plate reader. All measurements of accumulation and transport were normalized to accumulation and transport in the absence of inhibitor. Rhodamine accumulation and transport studies were carried out in the 10% FBS co-culture medium, while doxorubicin studies were conducted in transport buffer described above. Sucrose permeability and TEER measurements were used to confirm monolayer integrity in the presence of inhibitors.
24-well tissue culture plates were coated with 500 μL of Matrigel for 1 h at 37 °C. Collagen/fibronectin-purified hPSC-derived BMECs were dissociated using trypsin, and 100,000 cells were plated into each Matrigel-coated well in EC medium supplemented with 40 ng/mL VEGF (R&D Systems) and imaged after 12 h. A control sample of cells lacking VEGF was also employed. For acetylated LDL uptake, purified hPSC-derived BMECs were incubated with 10 μg/mL acetylated LDL conjugated to Alexa Fluor 488 (Invitrogen) for 4 h at 37 °C, washed twice with PBS, and visualized immediately with an Olympus epifluorescence microscope. Images were taken using a Diagnostic Instruments camera run by MetaVue software.
Cells were washed once with PBS and fixed in either 2% paraformaldehyde or 100% ice-cold methanol for 15 min. The cells were then blocked with 40% goat serum in PBS (40% PBSG). If probing for an intracellular antigen, 0.1% TX-100 was present in the 40% PBSG. The cells were then incubated in 40% PBSG containing primary antibodies (see Supplementary Table 3 for list) for 1 h at room temperature or overnight at 4 °C. After three washes in PBS, cells were incubated in 40% PBSG containing goat anti-rabbit Texas Red (1:500; Invitrogen) or goat anti-mouse Alexa Fluor 488 (1:500; Invitrogen) for 1 h at room temperature. Cell nuclei were labeled with 300 nM 4′,6-Diamidino-2-pheny-lindoldihydrochloride (DAPI) for 10 min. Cells were washed three times in PBS and visualized. Vascular and BBB markers were tested at the time points indicated in the Results text. Markers characteristic of basal epithelial and epithelial progenitor cells, cytokeratin K14 and the p63 transcription factor, were tested in IMR90-4 derived cells after 6 days of UM and 2 days of EC medium treatment or in purified IMR90-4-derived BMECs and were not detected. The simple epithelial cytokeratin K18 was expressed in undifferentiated IMR90-4 iPSCs and its expression was still detected in all cells of the mixed differentiating population after 6 days of UM and 2 days of EC treatment. In situ hybridization for detection of WNT7A and WNT7B transcripts was conducted similar to the method described by Planell-Saguer et al53. Briefly, cells were washed once with PBS and fixed in 2% paraformaldehyde for 10 min, followed by permeabilization in PBS containing 0.1% TX-100 for 5 min. Pre-hybridization was performed with a water-based solution containing 3% BSA and 4× saline-sodium citrate buffer (SSC; Fisher), followed by a 1 h hybridization in 4× SSC and 10% dextran sulfate (Fisher) at room temperature. Digoxigenin (DIG)-labeled locked nucleic acid probes were purchased from Exiqon and DIG-labeled locked sense probes were used as negative controls (sequences found in Supplementary Table 4). Following three washes in 4× SSC/0.1% Tween-20, one wash in 2× SSC, one wash in 1× SSC, and one wash in PBS (all washes conducted at 50 °C), cells were blocked in PBS containing 4% BSA for 20 min and labeled overnight at 4 °C with a monoclonal anti-digoxigenin antibody (Sigma). Secondary antibody and DAPI labeling were carried out as described above.
Cells were harvested via Accutase (Invitrogen) incubation for 2-3 min, fixed in 100% ice-cold methanol for 20 min, and blocked with 40% PBSG for 20 min at room temperature. Primary antibody labeling (Supplementary Table 3) was performed in 10% PBSG for 1 h at room temperature. IgG controls were used at the same concentration. After a wash with 5% FBS in PBS, secondary antibodies (goat anti-rabbit Alexa Fluor 488 and goat anti-mouse Alexa Fluor 647; 1:200) in 10% PBSG were added to each sample for 30 min at room temperature. After two washes with 5% FBS in PBS, the samples were analyzed on a FACScaliber flow cytometer. For quantification purposes, PECAM-1+ events were quantified using a PECAM-1/forward scatter dot plot referenced to a rabbit IgG control. Events demonstrating elevated GLUT-1 expression (GLUT-1+) were quantified using a GLUT-1/forward scatter dot plot referenced to a 4D UM culture that lacked BMECs to account for low basal GLUT-1 expression in the differentiating culture. Events that were found in both of these positive gates were classified as PECAM-1+/GLUT-1+ cells. PECAM-1+ events having basal GLUT-1 expression were referred to as PECAM-1+/GLUT-1−. For inhibition of BMEC differentiation, 10 μM of XAV-939 (Sigma)54 or equivalent DMSO vehicle control was added to IMR90-4 iPSCs starting at day 2 of UM treatment, and PECAM-1/GLUT-1 expression were evaluated as described above. Data are presented as two-dimensional dot plots with color codes for ease of viewing. βIII tubulin and nestin expression were quantified against rabbit and mouse IgG controls, respectively, using two-color flow cytometry.
Cells were differentiated as previously described. For inhibition of Wnt signaling, cells were treated with 250 ng/mL of mouse secreted frizzled-related protein 2 (SFRP2; R&D Systems)55 in UM for 4 days, followed by 750 ng/mL of SFRP2 in UM for an additional 3 days, or with 10 μM XAV or equivalent volume of DMSO vehicle control starting at day 2 of UM treatment. For RNA collection, cells were washed once with PBS and dissociated with trypsin or accutase. Total RNA was extracted using an RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. cDNA was generated from 1 μg of RNA using Omniscript reverse transcriptase (Qiagen) and oligo-dT primers (Invitrogen). Quantitative PCR (qPCR) was then performed using 1 μL of cDNA and iQ SYBR Green Supermix (Bio-Rad) on an iCycler (Bio-Rad). RT-PCR was also performed using GoTaq Green Master Mix (Promega). Primer sequences are supplied in Supplementary Table 4. Relative expression was quantified using the comparative cycle threshold (CT) method, normalizing to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) expression. Fold difference was calculated as x−ΔΔCT, where x refers to primer efficiency calculated according to LinRegPCR version 12.3 (ref. 56). Transcript amplification was analyzed by 2% agarose gel electrophoresis of the qPCR or RT-PCR products.
After 24 h of co-culture with rat astrocytes, IMR90-4-derived BMECs were washed once with PBS, fixed in 1.5% glutaraldehyde (Sigma) for 60 min, washed several times with PBS, cryoprotected with glycerol (30%), scraped from the filters, loaded into gold specimen carriers (Bal-Tec part # LZ 02125 VN), and plunged in liquid ethane. Frozen specimens were transferred to a Balzers 301 Freeze Fracture Apparatus for fracturing and 1 min of sublimation at −110 °C. Etched fracture surfaces were replicated with platinum at 45° supported by carbon deposited from 90°. Replicas were cleaned for 1 h using commercial household bleach followed by rinsing with double distilled water. Bare 400 mesh copper grids were used to pick up the cleaned replicas. Images were obtained using a JEOL 1230 TEM equipped with a Gatan Model 894 2kX2k CCD camera.
This work was funded in part by National Institutes of Health grants NS056249 (E.V.S.), AA020476 (E.V.S), EB007534 (S.P.P), and National Science Foundation grant EFRI-0735903 (S.P.P). E.S.L is the recipient of a National Institutes of Health Chemistry Biology Interface Traineeship (T32 GM008505) and S.M.A. is the recipient of a National Science Foundation Graduate Research Fellowship. The authors would like to thank the WiCell Research Institute for providing research support and Dr. William M. Pardridge for the kind gift of GLUT-1 antiserum.
E.S.L., S.M.A., S.P.P., and E.V.S. conceived the hPSC-derived BMEC strategy, designed all experiments, analyzed all data, and wrote the paper. E.S.L. and S.M.A. performed the majority of the experiments. J.E.K. performed RT-PCR experiments, R.A.N. performed the freeze-fracture microscopy and contributed to its interpretation, H.K.W. performed fluorescence in situ hybridization experiments, and A.A. contributed to the characterization of the BMECs.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.