Engineering of human brain organoids with a functional vascular-like system

Abstract

Human cortical organoids (hCOs), derived from human embryonic stem cells (hESCs), provide a platform to study human brain development and diseases in complex three-dimensional tissue. However, current hCOs lack microvasculature, resulting in limited oxygen and nutrient delivery to the inner-most parts of hCOs. We engineered hESCs to ectopically express human ETS variant 2 (ETV2). ETV2-expressing cells in hCOs contributed to forming a complex vascular-like network in hCOs. Importantly, the presence of vasculature-like structures resulted in enhanced functional maturation of organoids. We found that vascularized hCOs (vhCOs) acquired several blood-brain barrier characteristics, including an increase in the expression of tight junctions, nutrient transporters and trans-endothelial electrical resistance. Finally, ETV2-induced endothelium supported the formation of perfused blood vessels in vivo. These vhCOs form vasculature-like structures that resemble the vasculature in early prenatal brain, and they present a robust model to study brain disease in vitro.

Main

The human brain is composed of multiple cell types, including neurons, astrocytes, oligodendrocytes and microglia. Understanding human brain development and neurological disease mechanisms has been challenging due to the paucity of available tissues. Pluripotent stem cell-derived three-dimensional hCOs recapitulate the developing human cortex in both architecture and function. The hCOs have provided an exclusive opportunity to study initial brain development and neurologic disorders^1,2,3,4,5. Although the advent of three-dimensional brain organoids has offered innovative avenues, there are several unresolved limitations: the lack of functional blood vessels, the limited formation of microglia and of the six distinct cortical layers⁶. Notably, the absence of a vascular system is deemed detrimental to organoids, and millimeter-scale organoids under long-term culture consistently exhibit apoptotic cell death at the inner-most regions^7,8. Furthermore, differentiation of neuronal progenitors is impaired without functional vasculature⁹.

Recently, efforts have been made to implement vascular structure in human brain organoids. Human brain organoids transplanted onto the cortex of the mouse brain induced the outgrowth of murine vessels into the human tissue¹⁰, which increased cell survival and maturation. Beyond this, there is no systematic approach to produce robust vascularization in brain organoids, which limits their application in studying normal or pathogenic development in vitro. Here, we present a method to generate in vitro functional vasculature-like networks in hCOs from hESCs.

Results

Generation and characterization of cortical organoids with vascular-like networks

The expression of the transcription factor ETV2 reprograms human dermal fibroblasts into endothelial cells (ECs)¹¹. Here, we demonstrated that the expression of ETV2 induces EC differentiation in hESC under three conditions, (1) embryoid body differentiation, (2) neuron differentiation and (3) EC differentiation¹² (Supplementary Note 1). These data suggest that overexpression of ETV2 induces EC formation regardless of differentiation conditions and in the absence of growth factors (for example, VEGF) essential for differentiating and maintaining mature ECs in culture^13,14.

We hypothesized that ETV2 expression induces the formation of ECs and vascular-like structures in hCOs. We varied the ratio of cells expressing ETV2 (5, 10 and 20%) and induction time to determine the optimal condition for the formation of vascular-like structure in hCOs. We found that hCOs that were composed of 20% ETV2-infected cells and induced to express ETV2 at day 18 formed vascular-like structures most efficiently, characterized by EC markers (CDH5, CD31, KDR, TEK, vWF and CD34; Supplementary Note 2). We named the hCOs with vascular-like structures as vhCOs.

To examine whether the vasculature-like structures in ETV2-expressing organoids are functional, we evaluated the endothelial characteristics on days 30, 70 and 120. The formation of ventricular (VZ)-like zones, sub-ventricular (SVZ)-like zones and cortical layers is a distinct feature of cortical organoids^1,2. Both control hCOs and vhCOs exhibited well-organized SOX2 positive VZ and TBR1 positive SVZ around a lumen (Supplementary Fig. 1a, left panel). Moreover, the separation of early-born CTIP2⁺ and late-born SATB2⁺ neurons became distinct, leading to the formation of deep and upper cortical layers in a similar manner in control hCOs and vhCOs at day 120 (Supplementary Fig. 1a, right panel). Following neuroectoderm commitment, both control hCOs and vhCOs demonstrated a down-regulation of pluripotency genes and a concomitant up-regulation of neural genes (Supplementary Fig. 1b). Overall, these results show that both control hCOs and vhCOs differentiated to a similar extent and acquired the regional and morphological features of organized cortical structures.

We next investigated the organization of the endothelium and vascular-like networks. Whole-mount immunostaining demonstrated that CD31⁺ endothelial tubes were formed in vhCOs at day 30 while control hCOs lacked the formation of these tubes (Fig. 1a and Supplementary Videos 1 and 2). Moreover, 70 day-old vhCOs contained more complex networks of ETV2-derived CD31⁺ vascular-like structures (Fig. 1a). The quantification of vascularization via ImageJ (AngioTool) demonstrated that vhCOs had significantly more vessel area and vessel length compared to control hCOs (*P = 0.00003699 and **P = 0.00064, ***P = 0.0403; unpaired two-tailed t-test; Fig. 1a and Supplementary Fig. 1c). When we sectioned organoids, the lumen of VZ in ETV2-expressing vhCOs were stained strongly for the EC markers CD31 and CDH5, whereas the lumens of VZ in control hCOs lacked these EC markers (Fig. 1b). Quantitative PCR (qPCR) results supported our immunostaining analysis, revealing an elevated expression of CD31 and CDH5 in vhCOs compared to control hCOs at day 30 (Fig. 1b). When we cultured control hCOs up to 70 d, expression of TEK and KDR genes increased compared to day 30 control hCOs. However, vhCOs demonstrated earlier and higher induction of EC genes compared to control hCOs (Fig. 1b). Immunostaining indicated the presence of ECs, that is, CD31⁺ or CDH5⁺, in vhCOs at day 70, whereas control hCOs failed to generate these ECs (Supplementary Fig. 1d,e). Additionally, electron microscopy confirmed the presence of ECs in vhCOs (Supplementary Fig. 1f). We repeated these experiments with a different hESC line (H1) and observed a similar EC staining pattern in H1-derived vhCOs (Supplementary Fig. 2a). Overall, ETV2 induction leads to consistent generation of organoids with vascular-like architectures.

Fig. 1: Characterization of vasculature-like structures in vhCOs.

a, Left, immunostaining of whole-mount vhCOs and control hCOs at different time points (days 30 and 70) for CD31 and MAP2. Right, AngioTool analysis indicating the abundance and type of vasculature-like structures in hCOs. Data represent the mean ± s.e.m. (n = 7, from three independent batches) (*P = 0.00003699 and **P = 0.00064, ***P = 0.0403). b, Top, immunostaining for CD31 and CDH5 in sectioned hCOs and vhCOs at day 30. Bottom, expression of endothelial genes from organoids at days 30 and 70 was measured relative to HES3 hESCs. Data represent the mean ± s.e.m. (n = 5, from three independent batches). c, Illustration of FITC-dextran perfusion into organoids via bioreactor with the flow rate of 0.88 ml min^–1. d, Immunostaining of whole-mount FITC-dextran perfused vhCOs and control hCOs for CD31. Representative images were shown (n = 5, from three independent batches). e, Left, morphology and size of control hCOs and vhCOs during 120-d culture. Right, quantification of diameter (mm) from organoids at different stages (n = 20, *P = 0.000045, from four independent batches) (mean values of hCO at days 18, 30, 70 and 120 are 0.803, 2.354, 3.802 and 3.731 mm, respectively, and mean values of vhCO at days 18, 30, 70 and 120 are 0.831, 1.695, 3.697 and 3.938 mm, respectively). Error bar represents the mean ± s.e.m. f, Left, TUNEL staining of organoids after 30-, 70- and 120-d culture. Right, quantification of TUNEL⁺/DAPI⁺ cells in hCOs and vhCOs at days 30, 70 and 120. Data represent the mean ± s.e.m. (n = 8, from three independent batches) (day 30: T = 9.97, d.f. = 4 and *P = 0.000096; day 70: T = 26.02, d.f. = 4 and **P = 0.000012; day 120: T = 34.78, d.f. = 4 and ***P = 00000408). g,h, Left, voltage traces of current-clamp recordings of a cell in control hCOs and in vhCOs at days 80–90 (g) and days 50–60 (h) in response to hyperpolarizing (−10 pA) and depolarizing (+5 pA or +10 pA) current steps. Right, bar graph shows the difference in action potential (AP) incidence rate between control hCOs and vhCOs. *P = 0.0197 in g and P = 1 in h. The scale bar represents 100 μm (a,d), 50 μm (b,f) and 1, 2 and 4 mm at days 18, 30 and 70 and 120, respectively (e). The unpaired two-tailed t-test was used for all comparisons.

We further examined the perfusability of vascular structures in vhCOs by assessing in vitro capillary networks using a bioreactor (Fig. 1c and Supplementary Fig. 2b). We placed the organoids between two syringe filters under constant flow rate and perfused them with fluorescein isothiocyanate (FITC)-dextran (100 nm) (Fig. 1c). To determine the extent of perfusable vasculature in vhCOs at day 30, we performed whole-mount immunostaining for both CD31 and MAP2. We observed the presence of FITC-dextran in ETV2-induced CD31⁺ vasculature, indicating the existence of a perfusable vascular-like network in vhCOs (Fig. 1d and Supplementary Video 3). FITC-dextran filled in the lumens of VZ were co-stained with endothelial marker CDH5, and around 8% of lumens in vhCOs were FITC-dextran filled (n = 5, *P = 0.0001, unpaired two-tailed t-test; Supplementary Fig. 2c). Overall, the incorporation of FITC-dextran in CD31⁺ structures supports the notion that these vascular-like networks inside vhCOs are functional.

Next, we investigated the function of the vascular-like system in supporting the growth and viability of organoids. Both control hCOs and vhCOs were similar in size at day 18 before ETV2 induction. At day 30, when ETV2 was induced for 12 days, vhCOs were significantly smaller than control organoids (n = 20, *P = 0.000045, unpaired two-tailed t-test; Fig. 1e, right). By days 70 and 120, the growth rate of vhCOs exceeded that of control hCOs, although the size of both organoid types similarly reached 3.5–4 mm in diameter (Fig. 1e). This early lag in growth followed by a late increase appears to be temporally correlated with organoid vascularization; this phenomenon was also noted when hCOs were transplanted into mouse brains¹⁰. As the organoid size increased, control hCOs exhibited significantly higher apoptotic signals (TUNEL⁺) at day 70 (around 35%) compared to day 30 (n = 5, *P = 0.000096, **P = 0.000012, ***P = 00000408, unpaired two-tailed t-test; Fig. 1f). On the other hand, there was limited cell death in vhCOs, even at day 120 (Fig. 1f). Similarly, the analysis of hypoxia-inducible factor 1-α (HIF-1α) staining demonstrated the significantly elevated (~42%) regions of hypoxic cells in control hCOs compared to vhCOs at day 120 (~2.5%) (*P = 0.00138, **P = 0.000087 unpaired two-tailed t-test; Supplementary Fig. 2d). Collectively, these results suggest that ETV2-induced vascular-like architectures support the gradual increase in the size of hCOs by supporting the diffusion of oxygen, thus preventing cell death inside vhCOs.

The murine central nervous system starts to vascularize at embryonic days 7.5–9.5 (ref. ^15,16), whereas human brain vascular components are identified in 6–7-week-old human embryos¹⁷. The formation of ECs is critical for the maturation of cortical neurons^9,18,19. Thus, we performed whole-cell patch-clamp recordings to characterize neuronal activity in control hCOs and vhCOs. Eight of 20 cells from vhCOs (days 80–90) produced multiple action potentials with spike frequency adaptation in response to 1-s-long depolarizing current pulses (from +5 to +20 pA with 5 pA increments; cells were held at approximately −60 mV) (Fig. 1g, left) whereas the remaining 12 cells did not produce action potentials. In sharp contrast, only one of 20 cells from control hCOs (days 80–90) produced single low-amplitude action potentials (Fig. 1g, left) while the remaining 19 cells did not produce any action potentials. The incidence rate of obtaining neurons that were able to produce action potentials was significantly higher in vhCOs compared to that in control hCOs (Fig. 1g, right, P = 0.0197, Fisher’s exact test). In 50–60 d organoids, there was no difference in the incidence rate of obtaining neurons producing action potentials. Specifically, two out of 10 cells and one out of 10 cells produce action potentials in vhCOs and control hCOs, respectively (Fig. 1h, P = 1, Fisher’s exact test). An earlier study reported spontaneous firing of neurons in 8-month old organoids while 4-month-old organoids did not exhibit this neuronal activity²⁰. In contrast, we observed neurons that generated action potentials in 3-month-old vhCOs. Moreover, vhCOs expressed higher levels of the presynaptic marker synapsin 1 (SYN1), compared to control hCOs at day 70 (Supplementary Fig. 2e). When organoids were cultured longer, SYN1 expression in control hCOs became increased at day 120, although that in vhCOs was higher (Supplementary Fig. 2e). Thus, these results suggest that functional vessel-like structures in hCOs play a critical role in neuronal maturation observed in vhCOs.

Single-cell mapping of vhCOs

To examine global differences between control and vascularized cortical organoids, we performed single-cell transcriptome analysis of a total of 9,748 and 10,278 cells derived from vhCOs and control hCOs, respectively, at day 70 (Fig. 2a). We identified a total of 26 clusters by unbiased K-means clustering from a combination of all cells from both organoids, which we systematically assigned to 15 distinct cell types, including cortical neuron, interneuron, astrocyte, radial glia cell, glial progenitor cell and neuronal progenitor cell (Fig. 2b and Supplementary Fig. 3a-c)^21,22. Our single-cell RNA-seq indicated a minimal doublet rate and a limited bias of unique molecular index (UMI) among clusters (Supplementary Fig. 3d,e). We detected an endothelial-like cell cluster (EN), which is mainly composed of both vhCO and control hCO-derived cells. This cluster is significantly enriched for EC-specific gene signatures (false discovery rate (FDR) = 0.001; Fig. 2c)^21,23, and several neuroepithelial markers (VIM and HES1)²⁴ that also act as vascular remodelers (Supplementary Fig. 3c). Analysis of cells within the EN cluster revealed that vhCOs expressed several vasculogenesis factors (FLT1, HAND1, MME and VTN), pericyte markers (PDGRFβ and THY1), collagens and cell adhesion-related genes; control hCOs lacked the expression of these markers (Fig. 2d–f and Supplementary Fig. 3f)²⁵. Although some control hCO-derived cells express endothelial-specific genes, these results suggest that ETV2 is critical for vascular EC maturation and vascular morphogenesis in vhCOs.

Fig. 2: Single-cell analysis of vhCOs.

a,b, tSNE plot of single cells distinguished by organoid type (a) and cell annotation (b). RGC, radial glia cell; GPC, glia progenitor cell; NPC, neuronal progenitor cell; CBC, Cilium-bearing cell; BRC, BMP signal-related cell; EN, endothelial-like cell; ENP, endothelial-like progenitor; PGC, proteoglycan-expressing cell; EMT, epithelial-mesenchymal transition-related cell and UPRC, unfolded protein response-related cell. Data depicts results from 20,026 cells. c, Enrichment of gene signatures for ECs, astrocytes, neurons and NPCs. Data depicts results from 20,026 cells (FDR = 0.001). d, Histogram of gene expression related to vasculogenesis. Two-sided t-test P value = 2.96 × 10^-6 (FLT1) and 4.68 × 10^–14 (MME). Statistical significance was calculated by hypergeometric test and adjusted by Benjamini–Hochberg procedure. Data are representative of 2,257 cells. e, Gene ontology enrichment for differentially expressed genes in endothelial-like clusters between vhCO and control hCO. Data are representative of 2,257 cells. dev., development; prolif., proliferation. f, tSNE plot showing gene expression related to blood vessel formation and pericyte. Data depicts results from 20,026 cells. g, Estimation of neurodevelopmental stage by enrichment of vhCO- and control hCO-specific genes. h, Monocle-based trajectory analysis of vascularized organoid, vhCO.

To further address whether organoid vascularization promotes neuronal maturation, we compared the single-cell transcriptome profiles of the neuronal clusters from vascularized organoids with those from developing human brains (GW08–23)²⁶. Gene set enrichment analysis revealed that vhCOs-derived neurons resemble in vivo neurons corresponding to gestational week 16–19 (Fig. 2g). Critically, control hCOs-derived neurons from the same timepoint resemble an earlier stage (GW10-12) than vhCOs-derived neurons (Fig. 2g). In combination with the above immunostaining and electrophysiological analysis (Fig. 1g and Supplementary Fig. 2e), the single-cell RNA-seq results further support our findings that vascularization accelerates the functional maturation of neurons. In addition, we constructed a differentiation trajectory from the single-cell transcriptome profile, and proteoglycan-expressing cells (PGC) were grouped into one branch with endothelial-like cells suggesting their potential to differentiate toward the endothelial lineage (Fig. 2h and Supplementary Note 3).

vhCOs exhibit BBB-like characteristics

The human embryonic brain forms a blood-brain barrier (BBB) via specialized ECs that generate the perineural vascular plexus and connect cerebral capillaries by tight junctions, the major characteristics of the BBB^27,28. In addition to the expression of EC markers (Fig. 1b), we noted that most vhCOs are positive for the tight junction marker α-ZO1 in the lumens (Fig. 3a). Around 90% of lumens of vhCOs were stained for α-ZO1 compared to 18% in control hCOs (n = 4, *P = 0.0001, unpaired two-tailed t-test; Fig. 3a). At day 70, lumens of VZ in vhCOs were stained for α-ZO1 while control hCOs lacked this unique tight junction staining at the lumens (Supplementary Fig. 4a). We observed similar α-ZO1 staining at the lumens in ETV2-expressing H1-derived hCOs (Supplementary Fig. 2a). Both control hCOs and vhCOs showed a similar expression pattern of β-catenin at the lumens of VZ (Supplementary Fig. 4b). In addition to α-ZO1, occludin (OCLN) and KDR were co-stained in vhCOs at day 70 (Fig. 3b). The vhCOs showed clear but sparse OCLN and KDR co-staining at day 30 (Supplementary Fig. 4c). Control hCOs failed to express OCLN and KDR (Fig. 3b). Furthermore, glial cells marked by glial fibrillary acidic protein (GFAP) and glial-specific calcium binding protein B (S100β) were present at the lumen of VZ in vhCOs at day 70 (Fig. 3c and Supplementary Fig. 4d). Pericytes are an essential part of a neurovascular unit and contribute to vessel stability²⁹. Co-staining of pericyte marker, PDGFRβ, and astrocytic marker, GFAP, identified pericytes in vhCOs, which surrounded astrocytes but did not colocalize with them. Control hCOs did not contain any pericyte staining (Supplementary Fig. 4e). Last, vhCOs demonstrated increased expression of other tight junction-related genes, including claudin-5 (CLDN5), TJP1 and efflux transporter (ABCB1) and BBB glucose transporter (GLUT1) (Fig. 3d). Collectively, ETV2 induction gave rise to the distinct expression of tight junction markers, astrocytic and pericytic proteins and transporters, mimicking a BBB-like phenotype, albeit a structural difference from naturally formed BBB.

Fig. 3: vhCOs demonstrate BBB characteristics.

a, Left, immunostaining of α-ZO1 in hCOs and vhCOs. Right, quantification of lumens with α-ZO1. Data represent the mean ± s.e.m. (n = 5, from three independent batches). Unpaired two-tail t-test was used for comparison (T = 10 d.f. = 2 and *P = 0.0001). b, Left, co-immunostaining for OCLN and KDR in hCOs and vhCOs at day 70. Right, quantification of lumens stained with both OCLN and KDR. Data represent the mean ± s.e.m. (n = 5, from three independent batches). Unpaired two-tail t-test was used for comparison (T = 9.77, d.f. = 4 and *P = 0.0006). c, Left, co-immunostaining for GFAP and S100β in hCOs and vhCOs at day 70. Right, quantification of lumens stained with both GFAP and S100β. Data represent the mean ± s.e.m. (n = 5, from three independent batches). Unpaired two-tail t-test was used for comparison (T = 16, d.f. = 4 and *P = 0.00009). d, Expression of markers for tight junction and transporters from control hCOs and vhCOs at day 30. Data represent the mean ± s.e.m. (n = 5, from three independent batches). e, Top, depiction of TEER analysis from hCOs. Bottom, TEER was measured in hCOs and vhCOs at days 30 and 70. Data represent the mean ± s.e.m. (n = 3, from three independent batches, *P = 0.00000054). f, FITC-dextran visualization after Aβ_1–42 treatment of hCOs and vhCOs. Organoids were incubated with 10 µM Aβ_1–42 (Aβ_1–42-fibril and Aβ_1–42-oligo) for 48 h, and then FITC-dextran perfusion was performed. Data represent the mean ± s.e.m. (n = 6, from three independent batches). g, Immunostaining for tight junction marker α-ZO1 in the FITC-dextran perfused vhCOs with or without Aβ_1–42-oligo treatment. Data are representative of three independent experiments. The scale bar represents 50 μm in a–c and g.

A critical characteristic of the BBB is a tight junction barrier formed by connected ECs. To measure tight junction integrity, we used trans-endothelial electrical resistance (TEER) analysis. We measured TEER in hCOs by placing micro-electrode probes in three different regions of spheres (Fig. 3e, left panel). At 30 d, TEER in vhCOs (186 ± 4 Ω cm^–2) was significantly increased compared to control hCOs (135 ± 10 Ω cm^–2) (*P = 0.00000054, unpaired two-tailed t-test; Fig. 3e, right panel). The vhCOs grown for 70 d in culture exhibited much higher TEER (351 ± 10 Ω cm^–2) than controls (71 ± 1 Ω cm^–2) (Fig. 3e, right panel). The elevated expression of tight junction markers and the continuous cell contact via α-ZO1 stain observed in vhCOs likely resulted in higher TEER (Fig. 3a and Supplementary Fig. 4a). The TEER measurement we observed is lower than those reported before in monolayer-derived BBB models (~1,500 Ω cm^–2)³⁰. However, three-dimensionally constructed BBB models³¹ showed similar TEER values (~50 Ω cm^–2) to those measured in our system. Collectively, vhCOs not only exhibited vascular-like phenotypes and the expression of endothelial and BBB markers, but they also contain tight junctions comparable to previously described three-dimensional BBB models.

In addition to the physical features, we further evaluated the biological function of vascular-like structures in vhCOs. We developed a neurotoxicity model in cortical organoids by assessing the deposition of amyloid-β (Aβ) peptide species. The accumulation of Aβ_1–42 leads to malformation in tight junctions and the subsequent disruption of the BBB through matrix metalloproteinases^32,33,34. We examined the effects of Aβ_1–42 species, including Aβ_1–42-oligo and Aβ_1–42-fibril, on vascular structure in hCOs using the FITC-dextran perfusion assay. The Aβ_1–42-oligo treatment led to decreased FITC-dextran filling and depletion of α-ZO1 signal in lumens of vhCOs, whereas Aβ_1–42-fibril treatment did not affect FITC-dextran perfusion (Fig. 3f,g). Similar to an earlier study³², Aβ_1–42-fibril did not damage tight junctions as effectively as Aβ_1–42-oligo, and this is supported by FITC-dextran leakage in vhCOs (Fig. 3f and Supplementary Fig. 4f). Collectively, these results suggest that ETV2 induction facilitates the formation of functional endothelial tight junctions and BBB-like characteristics.

vhCOs form functional blood vessels in vivo

Last, we examined whether control hCOs and vhCOs form functional vasculature in vivo. We implanted hCOs subcutaneously into the hind limbs of immune-deficient mice, and subsequently performed magnetic resonance imaging (MRI) on the xenograft (Fig. 4a). The transplanted vhCOs relative to the surrounding muscle and skin tissues showed a high contrast, weighted by the transverse relaxation time (T₂), while the control hCO was clearly defined at neither 10 nor at 30 d postimplantation (d.p.i.) (Fig. 4b). Control organoids, lacking in vitro generated vasculature, exhibited faint contrast relative to the surrounding muscle and skin tissues due to a limited invasion of host vessels. Additionally, to assess permeability through the vasculature we used dynamic contrast enhancement with contrast weighted by the longitudinal relaxation time (T₁) shortening induced by gadolinium. The vhCOs exhibited slow but irreversible uptake, indicating a vascularized, but less permeable tissue compared to the surrounding hind limb muscle (rapid uptake with backflow; Fig. 4c). At 30 d after organoid transplantation, we perfused mice with FITC-dextran to validate whether hCOs formed functional blood vessels in vivo (Fig. 4d). Immunostaining for human-specific CD31 and FITC revealed that functional vasculature in vhCOs was formed and connected to the mouse vasculature; this was absent in control hCOs (Fig. 4e and Supplementary video 4). In addition, vhCOs not only contained significantly elevated hCD31⁺ vascular networks, but also a higher percentage of FITC-dextran filled vessels in total (*P = 0.00412, **P = 0.000183, unpaired two-tail t-test; Fig. 4e and Supplementary Videos 5–7). Control organoids exhibited limited FITC-dextran filled host vessels, which corroborated the MRI data and indicated that control hCOs lacked a functional vascular connection to the host mouse (Fig. 4b,e). Overall, an in vitro-generated vasculature-like network in vhCOs was essential for blood flow and connection with the host blood network.

Fig. 4: vhCOs possess a functional vascular system.

a, Depiction of subcutaneous implantation of control hCOs and vhCOs in the right and left leg of immune-deficient mice. b, Left, in vivo T₂ map of the implanted control hCOs and vhCOs. Right, anatomical image of hCOs and vhCOs after 10 (top) and 30 d.p.i. (bottom). Data are representative of three independent experiments. The scale bar represents 2 mm. c, Top, tissue concentration of gadolinium contrast agent as a function of time in the left leg muscle (gray trace) and implanted vhCOs (black trace). Bottom, map of the area under the curve (AUC) of the concentration curve, with regions of interest outlined for the vhCOs (green) and muscle (blue). Data are representative of three independent experiments. The scale bar represents 2 mm. d, Schematic of the method for FITC-dextran perfusion. Host blood vessels are filled with FITC-dextran, shown in green, and endogenous vessels in vhCOs are shown in magenta. e, Explanted organoids from FITC-perfused mice were stained for human-specific CD31 and hNuclei at day 30 d.p.i. The scale bars represent 50 μm. n = 3 animals and three organoids from three independent batches for MRI and n = 7 animals and seven organoids from four independent batches for FITC-perfusion (*P = 0.00412, **P = 0.000183, the unpaired two-tail t-test was used for all comparisons). Mean ± s.e.m. are shown. a.u., arbitrary units.

Discussion

In conclusion, ETV2-induced reprogramming of EC cells in organoids is a robust method to generate vhCOs with a functional vasculature-like network, thereby offering a valuable platform to investigate brain development and disease mechanisms. This vhCO model system may facilitate a more accurate physiological representation of the brain by exhibiting an interplay of neural and ECs and by reducing the apoptotic and hypoxic condition of the interior tissues that damages typical, avascular organoids.

Methods

A step-by-step protocol for the generation of vhCOs is available at Protocol Exchange³⁵ and as a Supplementary Protocol.

Animals

The Rag2^–/– GammaC^−/− mice were purchased from Jackson Laboratories. All animal experiments described in this study were approved by the Institutional Animal Care & Use Committee of Yale University and were performed under the guidelines of Yale University Institutional Animal Care and Use Committee with approved protocol (no. 2016–11347).

hESCs culture

HES3 NKX2-1^GFP/w and BC4 hESCs were cultured on Matrigel (BD Biosciences) coated dishes with mTeSR1 media (Stem Cell Technologies) and passaged every week by treatment with 0.83 U ml^–1 Dispase (Stem Cell Technologies). All experiments including hESCs were approved by Yale Embryonic Stem Cell Research Oversight (ESCRO).

Generation and differentiation of BC4 hESCs

As described previously³⁶, doxycycline-inducible (BC4) dCAS9-mCherry and rTTA were introduced into the AAVS1 locus of HES3 NKX2-1^GFP/w. Briefly, 2 million HES3 NKX2-1^GFP/w cells were electroporated with 8 µg donor plasmid, 1 µg AAVS1 TALEN-L and 1 µg AAVS1 TALEN-R by using the Amaxa Nucleofector device (AAB-1001, Lonza) and seeded in mTeSR1 plus Y27632 (10 µM). After 3 d, G-418 (Thermo Fisher Scientific) was applied for 7 d (400 µg ml^–1 for the first 3 d and 300 µg ml^–1 for the next 4 d) to obtain stable colonies. A single isogenic colony was picked and expanded. BC4 was differentiated with the induction of ETV2 in differentiation conditions: (1) embryoid body differentiation media (IMDM containing 15% FBS, 200 μg ml^–1 transferrin, 0.05 ng ml^–1 ascorbic acid, 1 mM sodium pyruvate, 100 U ml^–1 penicillin/streptomycin, 2 mM glutamine and 450 μM monothioglycerol), (2) Neuronal differentiation media (DMEM-F12, 15% (v/v) KSR, 5% (v/v) heat-inactivated FBS (Life Technologies), 1% (v/v) glutamax, 1% (v/v) MEM-NEAA, 100 µM β-Mercaptoethanol, 10 µM SB-431542, 100 nM LDN-193189 and 2 µM XAV-939) or (3) EC differentiation protocol (first cells were differentiated to mesoderm in N2B27 supplemented with 8 μM CHIR99021 and 25 ng ml^–1 BMP4, and then cells were converted into EC cells in StemPro-34, 200 ng ml^–1 VEGF-A and 2 μM forskolin) as described previously¹².

Electric cell-substrate impedance sensing (ECIS) analysis

The ECIS assay was used to evaluate two-dimensionally endothelial differentiated cells barrier function, as described previously³⁷. The wells of ECIS chamber slides were functionalized with 1 mM cysteine and coated with Matrigel at 37 °C for 1.5 h, respectively. Chamber slides were inserted into the ECIS device (8W10E+, Applied Biophysics) that was situated within a cell culture incubator. BC4 hESCs were inoculated into the chambers, and impedance was recorded in real time during the three different differentiation protocols described above.

Generation of human cortical organoids (hCOs) with ETV2 induction

For the generation of EC-containing hCOs, BC4 hESCs were infected with the lentivirus containing the FUW-tetO-ETV2 as described previously¹¹. The lentivirus containing inducible ETV2 was transduced for 7 d in the absence of dox. As we described earlier²¹, hCOs were generated by mixing ETV2-infected BC4 and noninfected parental HES3 hESCs. Briefly, hESCs were dissociated into single cells via Accutase and 9,000 cells, containing 5, 10 or 20% ETV2 infected BC4 hESCs, were plated into a well of U-bottom ultra-low-attachment 96-well plate in neural induction media (DMEM-F12, 15% (v/v) KSR, 5% (v/v) heat-inactivated FBS (Life Technologies), 1% (v/v) Glutamax, 1% (v/v) MEM-NEAA, 100 µM β-mercaptoethanol, 10 µM SB-431542, 100 nM LDN-193189, 2 µM XAV-939 and 50 µM Y27632). Basal activation of ETV2 was started on day 2 by adding 0.5 µM dox, and FBS and Y27632 were removed from days 2 and 4, respectively. At day 10, organoids were transferred to ultra-low-attachment six-well plate in hCO media with minus vitamin A (1:1 mixture of DMEM-F12 and Neurobasal media, 0.5% (v/v) N2 supplement, 1% (v/v) B27 supplement without vitamin A, 0.5% (v/v) MEM-NEAA, 1% (v/v) Glutamax, 50 µM β-mercaptoethanol, 1% (v/v) penicillin/streptomycin and 0.025% insulin) for spinning culture and the media was changed every other day. After day 18, the media was changed to hCO media with vitamin A (the same composition as described above except B27 with vitamin A was used and 20 ng ml^–1 brain-derived neurotrophic factor and 200 µM ascorbic acid were supplemented). The media was changed every 4 d after day 18. Activation of ETV2 was performed beginning on either day 10 or 18 by adding 2 µM dox continuously in the media.

Live imaging of hCOs

Live images were captured in hCOs at days 30 and 70 to visualize the mCherry-expressing cells and structures. A Leica TCS SP5 confocal microscope, equipped with a controlled cell chamber possessing 37 °C temperature and 5% CO₂, was used to generate z-stack images. Three-dimensional reconstruction of images was attained by using Leica Las-X software (v.1.1.0.12420).

FITC-dextran perfusion

The in vitro capillary network was developed by using two Cole Palmer silicone tubing (size L/S160) connected to a Master-flex peristaltic pump. Organoids were located in between two EZFlow Syringe filters (MedLab) that were attached to the silicone tube. Perfusion of FITC-dextran (Sigma, average molecular weight) was performed for 1 h under constant flow rate (0.88 ml min^–1). All experiments were performed in the dark and organoids were fixed as described below for characterizing the FITC localization.

Electrophysiological recordings

The control hCOs and vhCOs (D80-D90) were cut into two halves in chilled oxygenated slicing medium containing (in mM) 85 NaCl, 75 sucrose, 2.5 KCl, 25 glucose, 1.25 NaH₂PO₄, 4 MgCl₂, 0.5 CaCl₂ and 24 NaHCO₃. They were kept in the same solution at room temperature (RT) until recording. We performed whole-cell patch-clamp recordings using artificial cerebrospinal fluid containing (in mM) 126 NaCl, 2.5 KCl, 26 NaHCO₃, 2 CaCl₂, 2 MgCl₂, 1.25 NaH₂PO₄ and 10 glucose) gasses with 95% O₂/5% CO₂. Cortical cells in hCOs were visualized using an upright microscope (Eclipse FN1; Nikon Instruments Inc.) with infrared differential interference contrast optics and water immersion ×60 objective. Borosilicate glass pipettes (5–7 MΩ) were filled with intracellular solution containing (in mM) 126 K-gluconate, 4 KCl, 10 HEPES, 4 ATP-Mg, 0.3 GTP-Na and 10 phosphocreatine (pH 7.2 and osmolality of 290 mOsm). They were used to record activities from cortical neurons. MultiClamp700B amplifier and Digidata 1440A digitizer (Molecular Devices) were used for recordings. Voltage signals were filtered at 3 kHz using a Bessel filter and digitized at 10 kHz. We used the Clampfit 10 software (Molecular Devices) to analyze the data.

Electron microscopy

The cultured organoids were fixed in 2% glutaraldehyde and 0.1 M Na-cacodylate (pH 7.4), at different time points, for 1 h at RT. The organoids were postfixed via incubation in 1% osmium tetroxide for 1 h, staining en bloc in 2% aqueous uranyl acetate, dehydration in ethanol and propylene oxide and infiltration with Embed 812. After the hardening of the blocks, 60 nm sections were generated, poststained with uranyl acetate and lead citrate and analyzed by electron microscopy.

TEER analysis

TEER in three-dimensional organoids was measured by placing micro-electrode probes in three different regions of the spheres. TEER measurements were performed by using an electrochemical impedance analyzer (Z100, eDAQ). For each control hCO and vhCO (days 30 and 70, n = 3), three different measurements were performed and Z navigator was used to produce electrochemical impedance spectroscopy data from the organoids. ZMAN software (v.2.32), both metal and battery supercapacitor circuits, was used to analyze the data. Finally, resistivity is calculated as follows:

$$\begin{array}{rcl}{\rm{TEER}}\ {\rm{value}}\ (\Omega\,{\rm{cm}}^{-2}) & = & {\rm{TEER}}_{{\rm{average}}}\ (\Omega, {\rm{ZMAN}}\ {\rm{software}})\\ &&\times\ {\rm{hCO}}\ {\rm{surface}}\ {\rm{area}}\ ({\rm{cm}}^2,\ 4{\mathrm{\pi}} {r}_{{\mathrm{organoid}}}^2)\end{array}$$

Where r represents the radius of the organoid.

Library preparation for scRNA-seq

hCOs at 70 d old were randomly obtained from three different culture dishes (10 hCOs pooled together). As described previously²¹, organoids were first dissociated via the papain dissociation system according to the manufacturer’s instructions. After washing with HBSS, hCOs were dissected into small pieces in papain solution and oxygenated with 95% O₂:5% CO₂ for 5 min. Then, the tissue was oxygenated for 5 min and incubated at 37 °C for 1 h. After generating a single-cell suspension via trituration, single cells were suspended in 1% bovine serum albumin (BSA)/PBS supplemented with 10 μM Y27632 and stained for propidium iodide. FACS-sorted propidium iodide cells were re-suspended in 0.04% BSA/PBS (128 cells per μl) and used to generate cDNA libraries by using the Single Cell 3′ Reagent Kits. In short, cells were partitioned into nanoliter-scale Gel Bead-In-Emulsions (GEMs), and microfluidic cells were streamed at limiting dilution into a stream of Single Cell 3′ Gel Beads and then a stream of oil. After cell lysis, primers, an Illumina P7 and R2 sequence, a 14 base pair 10x Barcode, a 10 bp randomer and a poly-dT primer were released and mixed with the cell lysate and a bead-derived Master mix. In each individual bead, full-length cDNA from poly-adenylated messenger RNA was generated. After individual droplets were broken down, homogenized and cleaned from the remaining noncDNA, the libraries were size-selected followed addition of R2, P5, P7 sequences to each selected cDNA. Finally, each library was sequenced using the Illumina HiSeq4000 2 × 150 bp in Rapid Run Mode.

Data processing of scRNA-seq

Alignment to human (hg19) genome, initial quality control and UMI counting per Ensembl genes were processed by CellRanger with the count function by default parameters (v.2.1.0). Then, all libraries were normalized to the same sequencing depth and analyzed for dimensional reduction and K-means clustering with ‘aggr’ function of CellRanger by default parameters. Batch effect was then minimized by Seurat (v.2.2.1)³⁸. Each organoid type’s raw UMI count was normalized to total UMI with log-normalization and scaled by a factor of 10,000. Highly variable genes were identified with more than one dispersion and an average expression of 0.1–8. Canonical correlation analysis (CCA) was then performed with common variable genes from both types of the organoid. Single-cell transcriptome profiles from vascularized and nonvascularized organoids were merged into CCA subspace using the first to the twentieth CCs. For dimensional reduction, t-distributed stochastic neighbor embedding (tSNE) was implemented with the first to the twentieth CCs. Finally, cell clusters were identified by a shared nearest neighbor method by FindClusters function with ‘k.param = 30, resolution = 1.5, reduction.type = ‘cca’’ options. Differentially expressed genes in each cluster were identified by 1.5-fold change and P < 0.05 with two-sided t-test. Gene ontology analysis was implemented to the differentially expressed genes by GOstats (v.2.36.0) in the Bioconductor package. Multiple statistical tests were adjusted by the Benjamini–Hochberg method using the p.adjust function in R. An FDR < 0.05 was used as statistical significance.

Cell-type labels were assigned to each cluster by cell-type specific gene signatures, significant gene ontology terms, genes involved in specific cellular events, neurotransmitter transporter and synthetic enzymes as described previously (Supplementary Fig. 3b)^21,22. Further scRNA-seq analyses are described detail in Supplementary Note 4.

Aβ_1–42 peptide generation and treatment

Amyloid-β protein fragment 1–42 (Aβ_1–42, Sigma) was used to prepare fibrils (Aβ_1–42-fibril) and oligomers (Aβ_1–42-oligo) as described previously³². Briefly, 2 mM stock was prepared by dissolving Aβ_1–42 in dimethylsulfoxide (DMSO, Sigma). For generating fibrils, 2 mM Aβ_1–42 was diluted to 100 µM by adding 10 mM HCl and vortexed for 30 s followed by 24 h incubation at 37 °C with continuous shaking. For oligomer conditions, 2 mM Aβ_1–42 was diluted in neural differentiation media to 100 µM vortexed for 30 s followed by 24 h incubation at 4 °C. After generating fibrils and oligomers, 10 µM Aβ_1–42 peptide variants were incubated with control hCOs and vhCOs for 2 d. Then, the effects of Aβ1_–42 on FITC-dextran perfusion and tight junction marker expression were analyzed.

Subcutaneous implantation of hCOs

Control and ETV2-infected hCOs grown for 40–50 d were embedded with Matrigel and used for implantation. Mice were placed into a chamber by providing 2% isoflurane for anesthetization. Then, mice were fixed in a laminar hood to cut a small incision at each back leg. Matrigel-embedded control hCOs and vhCOs were subcutaneously placed into the incision of the right and left back legs, respectively. Once organoids were inserted, the wound was closed with sutures and buprenorphine (0.1 mg kg^–1) was administered for pain relief. Subsequently, implanted organoids were analyzed via MRI analysis and immunofluorescence staining.

FITC-dextran perfusion into mice containing subcutaneous implanted hCOs

After 30 d of organoid implantation, mice were injected with 15 mg ml^–1 FITC-dextran via a perfusion device. Briefly, the perfusion device was inserted in the left ventricle, and then we punctured the right ventricle. FITC-dextran was immediately injected into mice with a 5 ml min^–1 flow rate via a peristaltic pump. After monitoring the color change in the blood vessels of mice (around 3–5 min), the mice brains, vhCO and control hCOs were explanted and imaged for FITC via confocal microscopy. Then, explanted tissues were fixed and sliced for further immunofluorescence staining.

Immunofluorescence staining

For explanted organoids, residual Matrigel was removed by washing with PBS. As described earlier²¹, all organoids were fixed in 4% paraformaldehyde (PFA) at 4 °C overnight following three washes with PBS at RT. Then, organoids were transferred to a 30% sucrose solution for 2 d at 4 °C. For FITC-dextran perfused hCOs, organoids were fixed in 4% PFA at RT for 15 min and incubated in 30% sucrose at 4 °C for 1 d in the dark. Organoids were equilibrated with optical cutting temperature compound at RT for 15 min, transferred to base molds, and embedded in optical cutting temperature compound on dry ice. Then, 40-μm cryosections were generated and organoid blocks were stored at −80 °C. Slides were dried for 2 h at RT and incubated with 0.1% Triton-100 for 15 min at RT in a humidified chamber. 3% BSA was used for blocking at RT for 2 h, and then organoids were incubated with a primary antibody diluted in 3% BSA overnight at 4 °C. After two washing steps, organoids were incubated with Alexa Fluor Dyes (1:1,000) for 1 h following nuclei staining with DAPI (1:1,000). Finally, slides were mounted with ProLong Gold Antifade Reagent and images were taken with a Leica TCS SP5 confocal microscope. The tunnel assay (C100247, Invitrogen) was performed to detect apoptotic cells following the manufacturer’s protocol. A list of antibodies is presented in Supplementary Table 1.

Whole-mount Immunostaining of organoids

We performed whole-mount immunostaining followed by confocal microscopy to examine the localization and organization of EC networks within the cortical organoids. Organoids were washed with PBS and fixed overnight in 4% paraformaldehyde (PFA) at 4 °C. After washing the organoids with PBS for 5 h, the organoids were blocked overnight at RT in 0.5% BSA and 0.125% Triton-100 in PBS. Organoids were incubated in primary antibodies (anti-Cd31 1:100, anti-MAP2 1:200) and diluted in 0.5% BSA and 0.125% Triton-100 in PBS for 2 d at 4 °C. Unbound antibodies were removed via washing with PBS for 1 d at RT. Then, organoids were incubated with Alexa Fluor Dyes (1:500) for 4 h following nuclei staining with DAPI (1:1,000) for 2 h. After washing with PBS for several hours, clearing of the organoids was performed as described previously³⁹. The organoids were applied to a sequential dehydration series of 30, 50, 70 and 99% 1-Propanol (diluted in PBS pH adjusted to 9.5 via triethylamine) for 4 h at 4 °C. After dehydration, the organoids were incubated with ethyl cinnamate for 1 h at RT in light protected and air-sealed tubes.

Flow cytometry

For flow cytometry analysis, EC cells generated via induction of ETV2 were dissociated with Accutase. Single cells were incubated with CD31 (Abcam, Ab119339, 0.5 μg per 5 × 10⁵ cells) for 30 min on ice. The cells were washed and subsequently incubated with secondary FITC-labeled antibody for 30 min on ice. Finally, the cells were washed, filtered and analyzed on a BD LSRII instrument.

MRI data acquisition and analysis

Images were acquired using an 11.74T horizontal-bore spectrometer with Bruker console, using a dual-coil setup (volume and surface coils in transmit-receive configuration, RAPID Biomedical GmbH). The animal was anesthetized using 1.5% isoflurane in oxygen, and body temperature was maintained at 35–37 °C throughout the procedure using a circulating warm water pad. First, anatomical images were acquired using a T₂-weighted turboRARE sequence with four averages and a RARE factor of four (effective echo time = 15 ms). Then, an intrinsic T₁ map was fitted from a multi-recycle time acquisition (300, 700, 1,000, 2,000, 4,000 and 8,000 ms). Finally, a spoiled-gradient sequence was used for the dynamic contrast enhancement (flip angle = 15°, recycle time = 41.6 ms, dual echo time = 2.5 per 5 ms), acquired every 4 s for 22 min, with a bolus injection of Gadobutrol in the tail vein (0.1 mmol kg^–1) after 2 min. The dynamic signal change was converted to contrast agent concentration calibrated from voxel-level intrinsic T₁, and an additional apparent T₂ correction (T₂*) from the second echo due to magnetic susceptibility of Gadobutrol, where the relaxivity of Gadobutrol (r₁) was previously measured in vitro.

Real-time qPCR

The whole organoids were used for total RNA isolation via RNeasy Mini Kit (Qiagen). Then, 1 µg of RNA was converted to complementary DNA using iScript Select cDNA Synthesis Kit. For the quantification of gene expression, qPCR was carried out on a CFX96 Real-Time PCR system (Biorad) using SsoFast EvaGreen Supermix (Biorad). The PCR conditions were: 95 °C for 15 min, followed by 40 two-step cycles at 94 °C for 10 s and 60 °C for 45 s. A list of primers used in this study is presented in Supplementary Table 2.

Statistics

Data are presented as mean ± s.e.m. The paired or unpaired two-tail t-test (GraphPad Prism software v.7.0), hypergeometric test adjusted by Benjamini–Hochberg procedure and two-sided t-test (R v.3.5.0 software) were used to determine the statistical significance. Statistical tests and biological replicates for each experiment are presented in the figure legends.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.