The calcium pump ATP2B1/PMCA1 regulates CNS vascular development by facilitating Norrin- and WNT7A/B-induced Frizzled4 signaling

SUMMARY

Frizzled4 (FZD4) is a receptor for Norrin and WNT7A/B ligands, is expressed in endothelial cells (ECs), is required for endothelial blood-central nervous system (CNS) barrier function as well as CNS angiogenesis, and transduces β-catenin-dependent signaling. Despite its fundamental importance in neurovascular biology, including as a drug target, the molecular mechanisms governing FZD4 regulation remain poorly understood. Here, we employed proximity biotinylation to identify proteins that regulate FZD4. We identified ATPase plasma membrane Ca²⁺ transporting 1 (ATP2B1, also known as PMCA1) as a FZD4 proximity interactor. Functional analyses revealed that ATP2B1 depletion increased EC Ca²⁺, activated NFAT, and significantly attenuated Norrin/Frizzled4 signaling. Endothelial-specific Atp2b1 deletion caused CNS vascular phenotypes consistent with compromised Norrin/Frizzled4 signaling. This study identifies ATP2B1 as a regulator of Norrin- and WNT7A/B-induced FZD4 signaling and suggests that in pathological contexts with elevated EC Ca²⁺-levels, EC function may be modulated by suppression of β-catenin-dependent signaling.

In brief

Jo et al. identify the plasma membrane Ca²⁺-pump ATP2B1/PMCA1 as a regulator of endothelial Norrin/Frizzled4 and Wnt signaling in the CNS vasculature. Loss of ATP2B1 elevates intracellular Ca²⁺ and activates NFAT, suppressing β-catenin signaling and linking Ca²⁺ homeostasis to angiogenesis and blood-brain barrier integrity.

Graphical Abstract

INTRODUCTION

CNS endothelial cells (ECs) participate in angiogenesis under the control of CNS-specific mechanisms and are essential for the function of the blood-brain barrier and inner blood-retina barrier (BBB and BRB). In the endothelial barriers, ECs contribute to selective transport, regulation of the neural milieu, neuroimmune modulation as well as protection from toxins and pathogens. Defects in the BBB and BRB contribute to the onset or progression of diseases, such as stroke, neurodegenerative disorders, retinopathies, and macular edema.^1–5 A key pathway for inducing and maintaining the BBB and BRB and for enabling CNS angiogenesis is β-catenin-dependent signaling transduced by Frizzled and LRP5/6 receptors in ECs.⁶ β-catenin-dependent signaling in ECs can be induced by Norrin (gene symbol NDP) or WNT7A/B. Norrin has major functions in the retina and cerebellum^7–9 and WNT7A and B carry out major functions in the developing forebrain and spinal cord.^10,11 In many mature brain regions, Norrin and WNT7A/B function in parallel.¹² Among the ten Frizzled (FZD1–10) receptors, FZD4 serves as the sole Frizzled receptor for the Norrin protein¹³ and as one of the receptors for WNT7A/B.¹⁴ FZD4 associates with additional receptor complex components: Tetraspanin12 (TSPAN12) is a Norrin co-receptor that amplifies signaling in a ligand specific manner.^15–17 The receptor complex components G protein-coupled receptor 124 (GPR124) and reversion-inducing cysteine-rich protein with Kazal motifs (RECK) amplify WNT7A/B signals.^14,18–28 The causal relationship between impaired Norrin/Frizzled4 signaling and retinal hypovascularization in Norrie disease, osteoporosis-pseudoglioma syndrome, and familial exudative vitreoretinopathy (FEVR) underscores the critical role of this pathway in retinal vascular development.²⁹ Retinal Norrin/Frizzled4 signaling is regulated by glutamatergic neuronal activity.³⁰ In the brain, loss of β-catenin-dependent signaling in ECs is implicated in several neurological conditions including ischemic stroke.³¹ Loss of Norrin/Frizzled4 signaling in cerebellar ECs creates a permissive environment for medulloblastoma.³² Together, these studies establish critical functions of FZD4 in CNS vascular biology.

Given the pivotal role of FZD4 in CNS endothelium, this receptor has emerged as a target for experimental therapies currently in clinical development.³³ The pharmacodynamic effects of Norrin-mimetics have been characterized in preclinical studies.^34–40 Although FZD4 has garnered significant attention in neurovascular biology, the mechanisms governing its regulation in ECs are not well understood.

In this study, we identify ATPase plasma membrane Ca²⁺ transporting 1 (ATP2B1 aka PMCA1) as a regulator of Norrin- and WNT7A/B-induced FZD4 signaling in CNS ECs. As a plasma membrane calcium extruder pump, ATP2B1 maintains calcium homeostasis within cells and influences the shape and duration of Ca²⁺ transients.⁴¹ The importance of these roles is underscored by the lethality resulting from its systemic deletion in mice.⁴² In addition to ATP2B1, the molecular repertoire for Ca²⁺ clearance in various cell types includes the paralogs ATP2B2–4, the Na⁺/Ca²⁺ exchanger family (NCX) and the sarcoplasmic/endoplasmic reticulum calcium ATPase (SERCA) family, which transports Ca²⁺ into the ER. It is thought that cytosolic nanodomains with low Ca²⁺ concentration in immediate proximity to ATP2B1 modulate local Ca²⁺-dependent processes, possibly supported by physical confinement, e.g., a plasma membrane invagination. The association of ATB2B1 with scaffold proteins may help to recruit receptors into low Ca²⁺ nanodomains in the vicinity of ATP2B1.⁴³ ATP2B1–4 have been studied in various excitable cell types, including sensory neurons, smooth muscle cells, and cardiomyocytes⁴⁴ as well as various non-excitable cells such as lymphocytes.^45–47 Ca²⁺ in ECs is involved in responses to signals that modulate vasodilation, angiogenesis, vascular permeability, thrombogenesis, diapedesis, or vascular inflammatory responses, including ATP, VEGF-A, bradykinin, histamine, thrombin, and shear stress.^48–53 Few studies have investigated the roles of ATP2B1–4 in ECs. ATP2B1 has been studied in relation to HUVEC migration⁵⁴ and nitric oxide production,⁵⁵ while ATP2B4 has been examined for its role in VEGF-A-induced angiogenesis.⁵⁶ Here, we identify ATP2B1 as a proximity interactor of FZD4 and elucidate a role in Norrin- and WNT7A/B-induced FZD4 signaling using cell-based and mouse genetic approaches.

RESULTS

Proximity interaction screens with a FZD4 bait

To identify novel regulatory mechanisms in FZD4 signaling, we performed proximity labeling. This approach enables detection of proximity interactions within approximately 10 nm radius in living cells using BioID, a promiscuous form of the bacterial BirA biotin ligase.⁵⁷ We used V5-FZD4-BioID as bait and GFP-BioID as a control. V5-FZD4-BioID may identify proximity interactions during intracellular trafficking through the secretory pathway, at the plasma membrane, or during endo-lysosomal trafficking (Figure 1A). The functionality of the V5-FZD4-BioID fusion protein was tested by examining V5-FZD4-BioID subcellular localization in HeLa cells. In live HeLa cells subjected to flag-AP-Norrin binding to FZD4 and endocytosis, V5-FZD4-BioID was detected both at the plasma membrane as well as in endocytic vesicles (Figure 1B), where it co-localized with FLAG-AP-Norrin (an efficiently secreted, soluble, and bioactive alkaline phosphatase fusion protein). This observation indicated that the FZD4 fusion protein underwent similar trafficking processes as FZD4 (without BioID tag).⁵⁸ TOPFlash reporter assays for Norrin-induced β-catenin-dependent FZD4 signaling in HEK 293T cells confirmed that the V5-FZD4-BioID fusion-protein transduced the Norrin signal. Signaling amplitude was increased in the presence of the Norrin co-receptor, TSPAN12, as expected,^12,15,59 indicating that V5-FZD4-BioID initiated signaling and was able to cooperate with TSPAN12 (Figure 1C). Immunoblots of 293T cell lysates transfected with V5-FZD4-BioID and treated with biotin showed substantial biotinylation of proteins of various molecular weights, whereas in lysates of V5-FZD4 expressing cells only few endogenously biotinylated proteins were detected with Streptavidin-HRP (Figure 1D), confirming biotin ligase activity in the fusion protein. Next, we performed BioID screens in three cell lines that have been previously used in cell-based studies of Norrin/Frizzled4 signaling: HEK293T cells, HeLa cells, and bEnd.3 mouse brain ECs. Collectively, these cell lines are suitable to identify mechanisms that regulate FZD4 in multiple biological contexts, including the BBB. In 293T cells, plasmids encoding LRP5 and TSPAN12 were co-transfected with the BioID constructs, whereas HeLa cells were transfected only with the BioID constructs, and bEnd.3 cells were virally transduced with vectors encoding GFP-BioID or V5-FZD4-BioID. Stable cell populations were generated by selecting blasticidin-resistant transduced cells. Staining for the V5 peptide tag showed FZD4 accumulated in the perinuclear compartment and at the plasma membrane in bEnd.3 cells (Figure S1). Biotinylated proteins were isolated from 293T, HeLa, and bEnd.3 cells and subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. The bEnd.3 cell samples were processed through a different LC-MS/MS and bioinformatic pipeline than the 293T and HeLa cells (see STAR methods). We filtered the proteins identified by (LC-MS/MS) for a combination of specificity in comparison with GFP-BioID, abundance, and protein coverage (Figures S2, S3, and S4) and displayed the shortlists of filtered proteins in a word cloud, in which the size of the letters correlates with semi-quantitative protein metrics (Figures 1E–1G). A Venn diagram was generated to highlight that multiple proteins were identified in all three experiments, including FZD4 as an auto- proximity interaction, discs large homolog 1 (DLG1) and Golgi reassembly Stacking Protein 2 (GORASP2) (Figure S5). The robust identification of these proteins validated the screens, as DLG1 is a physical and functional FZD4 interactor,^60,61 and GORASP2 is a Golgi protein that controls FZD4 trafficking.⁶² Additional proteins that were identified in all three cell lines include ATP2B1/PMCA1, a plasma membrane Ca²⁺ pump. Co-transfected TSPAN12, a Frizzled4 receptor complex component, was identified in 293T cells as expected.^15,16 The identification of Disheveled-1 (DVL-1), which is an intracellular interactor of FZD4,^58,63 further validated the screen in 293T cells. Additional proteins of interest are addressed in the discussion. For further analysis, we focused on ATP2B1 because its function in ECs remains incompletely understood (see Introduction), its proximity interaction was unexpected, and siRNA-mediated knockdown of Atp2b1 affected Norrin/Frizzled4 signaling more consistently than knockdown of other candidates (see in the following text). To validate the spatial relationship between ATP2B1 and FZD4, we performed co-immunostaining of V5-FZD4 and FLAG-ATP2B1 in both HeLa and COS-1 cells (large cells suitable for co-localization studies), revealing co-localization at the plasma membrane (Figure 1H). Furthermore, FLAG-ATP2B1-mNG2(11) reconstituted GFP fluorescence by association with co-transfected V5-FZD4-mNG2(1–10), whereas split GFP was not reconstituted with a BACE1-(aka β-secretase)-mNG2(1–10) control (Figure 1I). Together, we identified ATP2B1 as a proximity interactor of FZD4.

Figure 1. Identification of a proximity interaction of FZD4 and ATP2B1.

(A) FZD4 with C-terminal BioID tag, schematic trafficking steps. Red cloud symbolizes 10 nm radius in which biotinylation is most likely to occur (not to scale). ER, endoplasmic reticulum; TGN, trans-Golgi network; RC, receptor complex; EE, early endosome; RE, recycling endosome; MVB, multivesicular body.

(B) FZD4-BioID reaches the plasma membrane (white arrows) and undergoes Norrin-induced endocytosis (intracellular puncta), as described for untagged FZD4. Representative of two experiments with similar results. Scale bars: 10 μm.

(C) TOPFlash reporter assay shows activity of fusion protein in response to Norrin, mean ± SD shown, n = 3 biological replicates.

(D) Streptavidin-HRP detects numerous biotinylated proteins in FZD4-BioID-transfected 293T cells, indicating that the BirA (R118G) component of the fusion protein is active. Representative of two experiments with similar results.

(E–G) Filtering of LC-MS/MS data by specificity (vs. GFP-BioID), abundance, coverage, and MS/MS count criteria yields prioritized lists of proximity interactors that are represented as word cloud, where the size of the letters correlates with the rank of the protein in the filtered list, see also Figures S2, S3, and S4. Proteins of interest are highlighted in color.

(H) Immuno-staining of V5-FZD4 and FLAG-ATP2B1 in HeLa and COS-1 cells. Scale bars: 10 μm.

(I) Split-GFP reconstitution in COS-1 cells co-transfected with mCherry. The membrane protein BACE1 (β-secretase 1) was used as negative control.

ATP2B1 facilitates Norrin/Frizzled4 signaling in CNS ECs

To investigate a potential functional role of ATP2B1 in Norrin/Frizzled4 signaling, we employed three distinct siRNAs targeting different regions of Atp2b1 (coding sequences and 3^′UTR), of which Atp2b1 siRNA #2 achieved only moderate knockdown efficiency in bEnd.3 cells, whereas Atp2b1 siRNA #1 and #3 knocked down ATP2B1 efficiently on the mRNA (Figure 2A) and protein level (Figure S6). We examined the expression of Axin2, an established target gene of the canonical Norrin/Frizzled4 signaling pathway in ECs.^34,64 In control siRNA-treated bEnd.3 cells, Norrin stimulation induced Axin2 mRNA expression >30-fold. As expected, this induction was abolished in FZD4-knockdown cells due to the absence of the receptor for Norrin/Frizzled4 signaling. Notably, Atp2b1 knockdown, using each of three siRNAs, significantly attenuated Norrin-induced Axin2 expression compared to controls, suggesting that ATP2B1 facilitates the Norrin/Frizzled4 pathway in ECs (Figure 2B). RNA sequencing (RNA-seq) data of bEnd.3 cells revealed that there was no major expression of other ATP2B family paralogs (Figure S7). We used P2ry1 as another Norrin signaling target gene, which we identified from our bEnd.3 cells RNAseq data. P2ry1 mRNA expression was induced by Norrin in control cells but not in FZD4-knockdown cells (Figure 2C). Atp2b1 knockdown reduced Norrin-induced P2ry1 expression, consistent with the reduced expression of Axin2. To further corroborate these findings, we used our bEnd.3 cells stably expressing a TOPFlash reporter and Renilla luciferase⁶⁵ and observed that Atp2b1 knockdown significantly reduced Norrin-induced TOPFlash activity compared to controls (Figure 2D). To determine the level at which ATP2B1 intersects with Norrin/Frizzled4 signaling, we utilized CHIR99021, a GSK3β inhibitor. GSK3β, a component of the β-catenin destruction complex, phosphorylates β-catenin, promoting its degradation.⁶⁶ GSK3β inhibition activates β-catenin-dependent signaling independently of ligand-receptor interactions. In control cells, both Norrin and CHIR99021 robustly induced Axin2 expression. Atp2b1-knockdown cells maintained their ability to upregulate Axin2 expression in response to CHIR99021, but Norrin-induced activation was significantly impaired compared to controls (Figure 2E). This differential response suggests that ATP2B1 deficiency impinges, at least in part, upstream of β-catenin, for example on the level of the receptor complex or signalosome. Using lentiviral transduction, we generated a stable bEnd.3 cell population expressing V5-FZD4 and performed a cell-surface biotinylation experiment. We found that Atp2b1 knockdown did not significantly affect FZD4 steady state levels at the cell surface, whereas Brefeldin A impaired cell-surface trafficking as expected (Figure S8).

Figure 2. ATP2B1 facilitates Norrin/Frizzled4 signaling in CNS ECs.

(A) Knockdown efficiency of ATP2B1 siRNAs in bEnd.3 cells assessed by quantitative reverse-transcription PCR (RT-qPCR) for Atp2b1 mRNA. Two technical replicates per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Dunnett’s post hoc. In this and all following figures: *; p < 0.05, **; p < 0.01, ***; p < 0.001, and ****; p < 0.0001.

(B) RT-qPCR of Axin2, a target gene of the canonical signaling, in bEnd.3 cells transfected with the indicated siRNAs and stimulated with vehicle (PBS, 0.1% BSA) or 200 ng/mL recombinant Norrin. Three independent experiments were performed, each with one biological replicate per condition. For each experiment, two technical replicates were averaged. The resulting values were subjected to paired, within-experiment normalization, in which the control condition was set to 1 and experimental conditions expressed relative to their matched control. n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Dunnett’s post hoc.

(C) RT-qPCR of P2ry1 in bEnd.3 cells transfected with the indicated siRNAs and stimulated with vehicle or Norrin. P2ry1 is an alternative target gene of canonical signaling based on our RNA sequencing in bEnd.3 cells (see Figure S7). Data were normalized and combined as described for (B). Two technical replicates per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Dunnett’s post hoc.

(D) Luciferase reporter assay using stable bEnd.3 TOPFlash cells. For each siRNA condition the data are normalized to the vehicle group to show the fold change of TOPFlash/renilla activity in response to Norrin. n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(E) RT-qPCR of Axin2 mRNA after stimulation with Norrin or 2 μM of CHIR 99021, a GSK3β inhibitor. Data were normalized and combined as described for (B). Two technical replicates per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Dunnett’s post hoc (blue bar; siCON+CHIR99021 as a comparison control, black bar; siCON+Norrin as a comparison control).

Endothelial cell-specific Atp2b1 gene ablation causes retinal angiogenesis and BRB phenotypes with variable expressivity

We re-analyzed our previously reported single-cell RNA-seq data of wild-type mouse retina (Figure S9A)³⁵ and found broad Atp2b1 mRNA expression across multiple retinal cell types, including vascular cells (Figure S9B). Additionally, the paralog Atp2b4 was expressed in a more restricted manner, including in retinal vascular cells (Figure S9C). Next, we investigated the physiological role of Atp2b1 in inducible EC-specific Atp2b1-knockout (ECKO) mice, using a Cdh5-CreERT2 driver. The murine retinal vasculature develops in three distinct layers in a central to peripheral direction: the superficial vascular plexus in the nerve fiber layer (NFL) reaches the periphery by ~ P8, the deep plexus in the outer plexiform layer (OPL) reaches the periphery around P12, and the intermediate plexus in the inner plexiform layer (IPL) develops with slight delay and reaches the periphery by ~ P15.⁶⁷ Deep vascular plexus development is particularly sensitive to loss of Norrin signaling.⁸ We administered tamoxifen daily from P2–4 and analyzed the retinal vasculature at P13 to examine deep layer angiogenesis. The genotype of Atp2b1 ECKO animals was Atp2b1^flox/−; Cre⁺. Control genotypes were Atp2b1^flox/+; Cre⁻, Atp2b1^flox/flox; Cre⁻, Atp2b1^flox/−; Cre⁻, or Atp2b1^flox/+; Cre⁺, using littermates of the Atp2b1 ECKO mice as control. All control mice exhibited normal vascular development of the superficial and intraretinal vasculature. In contrast, Atp2b1 ECKO mice displayed vascular hypovascularization, particularly at the front of the superficial vascular plexus and in the deep vascular plexus (Figure 3A). Incomplete deep vascular plexus development was occasionally concentrated to the OPL above major arteries. The expressivity of vascular phenotypes, which were present in all Atp2b1 ECKO mice, was variable, with phenotypes ranging from only partial hypovascularization of the deep vascular plexus to virtually complete absence of the deep vascular plexus, glomeruloid vascular malformations, insufficiency of superficial vascular plexus development, and invasion of hyaloid vessels into the retina. In retinas with mild phenotypes, 3D-analysis of the three retinal vascular layers revealed hypovascularization in the intraretinal capillary beds (Figure 3B). These findings identified unknown roles of ATP2B1 in retinal vascular ECs. Furthermore, the data suggested the possibility that our mouse colony on a C57BL/6J background may harbor unknown genetic modifiers that strongly influence the severity of the phenotype. Whether this modification represented an attenuation, or an exacerbation, remained unclear (see Discussion) and variable phenotypes were observed in both sexes.

Figure 3. Vascular phenotypes in Atp2b1 ECKO retinas.

(A) Atp2b1 ECKOs were induced with tamoxifen at P2, P3, and P4 and the retinas were harvested at P13. All Atp2b1 ECKO retinas show a similar vascular phenotype, but the expressivity of the phenotype was variable and is illustrated in three examples. Blue arrows indicate a delay of the progression of the vascular front in the superficial vascular plexus. White dashed lines demarcate a boundary between areas with deep vascular plexus development vs. areas lacking the deep vascular plexus. Green arrowheads show glomeruloid vascular malformations. Yellow arrowheads show areas with anastomosis of retinal and hyaloid vasculature. Scale bars: 500 μm.

(B) Left: whole mount retinas of the indicated genotype are represented by stitched images. The white box indicates an area that is shown enlarged at 3 depth levels in the right. Scale bars: 500 μm left, 100 μm right.

(C) Quantification of the area with intraretinal hypovascularization in an allele series of the indicated genotypes. n = 3–14 retinas from 3 to 14 mice, graphs represent median, one-way ANOVA with Tukey’s post hoc.

Quantification of intraretinal vessel deficiency confirmed the hypovascularization phenotype in the comparison of Atp2b1 ECKO vs. control (Figure 3C). To corroborate that ATP2B1 functions in the Norrin/Frizzled4 pathway, we attempted to test for genetic interactions of Atp2b1 and the Norrin co-receptor Tspan12.^16,17 In the analysis of the allele series (Figure 3C), we observed a non-significant trend toward an exacerbated hypovascularization in Atp2b1 ECKO; Tspan12^+/− compound mutant retinas compared to Atp2b1 ECKO retinas. This analysis was complicated by the high variability of the Atp2b1 ECKO phenotype as well as the statistical power required for a comparison of 5 groups.

Because Norrin/Frizzled4 signaling is a key pathway for the induction and maintenance of the BRB,⁹ we investigated if BRB function was compromised. We assessed plasmalemma vesicle associated protein (PLVAP) expression, an EC fenestration/diaphragm component that is normally absent from retinal ECs, but is expressed in ECs of Ndp, Lrp5, and Tspan12 gene ablated mice.¹⁵ While control retinas showed no substantial PLVAP expression (Figure 4A), several Atp2b1 ECKO mice exhibited localized PLVAP expression corresponding to regions of defective deep layer angiogenesis (Figure 4A). The expressivity of the phenotype was highly variable, but the mean PLVAP⁺ area was significantly different between Atp2b1 ECKO retinas and control retinas (Figure 4B). Several Atp2b1 ECKO; Tspan12^+/− mice displayed widespread PLVAP expression across the retinal vasculature; however, mean PLVAP expression in Atp2b1 ECKO; Tspan12^+/− mice was similar to those in Atp2b1 ECKO mice. We further used sulfo-NHS-LC-biotin, a small molecule biotinylation reagent that penetrates disrupted barriers⁶⁸ to evaluate BRB function. The increase in biotin⁺ area in Atp2b1 ECKO retinas did not reach significance compared to controls, whereas leakage was significantly different in Atp2b1 ECKO; Tspan12^+/− mice compared to controls (Figure 4C).

Figure 4. Analysis of BRB function in Atp2b1 ECKOs.

(A) Example images of retinal petals stained for IB4, PLVAP (a fenestration marker), and biotin (using streptavidin-Alexa 488) from an allele series with the genotypes indicated in the figure. Scale bars: 200 μm.

(B) Quantification of PLVAP⁺ area. n = 3–9 retinas from 3 to 9 mice, graphs represent median, one-way ANOVA with Tukey’s post hoc.

(C) Quantification of biotin⁺ area. n = 3–15 retinas from 3 to 15 mice, graphs represent median, one-way ANOVA with Tukey’s post hoc.

In addition to the Atp2b1 ECKOs of the Atp2b1^f/−; Cre⁺ genotype, we also generated Atp2b1 ECKOs with the genotype Atp2b1^f/f; Cre⁺. The outcomes for hypovascuarization, PLVAP expression, and biotin leakiness were similar as in Atp2b1^f/−; Cre⁺ retinas, albeit milder (Figures S10A–S10C). We also analyzed aggregated data from Atp2b1^f/−; Cre⁺ and Atp2b1 ^f/f; Cre⁺ ECKO retinas combined. In this analysis all phenotypic parameters (hypovascularization, PLVAP expression, biotin leakage) trended higher in the Atp2b1 ECKO; Cre⁺; Tspan12^+/− group compared to Atp2b1 ECKO, but the difference did not reach significance (Figures S10D–S10F).

Together, these results demonstrate that EC-specific deletion of Atp2b1 causes vascular defects and that the expressivity of the vascular phenotypes is variable (see Discussion).

Elevated intracellular Ca²⁺ inhibits Norrin/Frizzled4 signaling

Having established a role for ATP2B1 in Norrin/Frizzled4 signaling, retinal angiogenesis, and BRB maintenance, we investigated the mechanistic basis of this regulation. Given the function of ATP2B1 as a plasma membrane calcium pump, we hypothesized that its deletion would elevate intracellular calcium levels in bEnd.3 cells, especially since ATP2B1 is the major paralog in bEnd.3 cells (Figure S7). To test this hypothesis, we knocked down Atp2b1 in bEnd.3 cells, which resulted in a significant increase of Fluo-4 signal (a membrane-permeable Ca²⁺ indicator dye whose emission was normalized to Hoechst fluorescence) compared to bEnd.3 cells transfected with non-targeting control siRNA (Figures 5A and 5B). Next, we employed ionomycin, a calcium ionophore, to increase intracellular calcium concentrations in 293T cells co-transfected with FZD4, LRP5, TSPAN12, and the reporter constructs. In control cells, Norrin treatment induced a robust increase in luciferase activity. However, this response was dose-dependently and significantly attenuated by ionomycin (Figure 5C). To further validate the calcium-dependent regulation of Norrin/Frizzled4 signaling, we utilized SEA0400, a selective inhibitor of the Na⁺-Ca²⁺ exchanger (NCX), which extrudes Ca²⁺ in its forward mode. 1 μM SEA0400 treatment of bEnd.3 cells suppressed Norrin-induced TOPFlash activity compared to controls (Figure 5D). Consistent with these findings, ionomycin treatment significantly decreased the mRNA expression of Norrin target genes Axin2 and P2ry1 in bEnd.3 cells (Figures 5E and 5F). Collectively, these data support that elevated intracellular calcium, resulting from ATP2B1 deficiency, negatively regulates Norrin/Frizzled4 signaling in CNS ECs.

Figure 5. Elevated intracellular Ca²⁺ inhibits Norrin/Frizzled4 signaling.

(A) Images of two wells of a 48-well plate containing bEnd.3 cells loaded with Fluo4 Ca²⁺ indicator. Cells were transfected with siRNAs as indicated in the figure.

(B) Quantification of Fluo-4 intensity normalized by Hoechst intensity. n = 3 biological replicates, graphs represent mean ± SEM, unpaired Student’s t test.

(C) 293T cells transfected with the indicated constructs and TOPFlash and CMV-Renilla plasmids were stimulated with vehicle or Norrin in the presence or absence of ionomycin. n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(D) Stable bEnd.3 TOPFlash cells stimulated with vehicle or Norrin in the presence of ionomycin or SEA0400, a Na⁺-Ca²⁺ exchanger inhibitor. n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(E) Total RNA from bEnd.3 cells subjected to RT-qPCR for Axin2 mRNA after stimulation with vehicle or Norrin in the presence of DMSO or ionomycin. Two technical replicates per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(F) Total RNA from bEnd.3 cells subjected to RT-qPCR for P2ry1 mRNA after stimulation with vehicle or Norrin in the presence of DMSO or ionomycin. Two technical replicates per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

Calcineurin-mediated NFAT activation negatively regulates Norrin/Frizzled4 signaling

Because several nuclear factor of activated T cells (NFAT) family transcription factors (NFATc1–4) are controlled by the Ca²⁺-dependent phosphatase calcineurin, we investigated whether NFAT-signaling modulates Norrin/Frizzled4 signaling. Intracellular calcium promotes calmodulin-dependent activation of calcineurin, which dephosphorylates multiple NFAT paralogs, facilitating their nuclear translocation and subsequent target gene transcription.⁶⁹ Our RNA sequencing of bEnd.3 cells showed that NFAT2 (aka NFATc1), NFAT4 (aka NFATc3) and Ca²⁺-independent NFAT5 are the predominant family members in bEnd.3 cells (Figure S7). We assessed the expression of known NFAT2 target genes, Egr3 and Nr4a2⁷⁰, in bEnd.3 cells treated with Atp2b1 or non-targeting control siRNA (Figures 6A and 6B). ATP2B1 knockdown significantly induced the expression of both NFAT2 target genes compared to controls. Treatment with the calcineurin inhibitor cyclosporin A (CsA), which suppresses NFAT activity, significantly reduced Egr3 and Nr4a2 expression in Atp2b1-knockdown cells, confirming that these target genes are under control of calcineurin/NFAT signaling in ECs. Furthermore, we transfected EGFP-hNFAT2 with C-terminal EE tag into bEnd.3 cells treated with non-targeting siRNA or Atp2b1 siRNA. While NFAT2 localized predominantly outside the nucleus in control cells, we observed significant translocation to the nucleus after Atp2b1 knockdown (Figures 6C and 6D). To investigate the relationship of NFAT signaling and Norrin/Frizzled4 signaling, we inhibited NFAT signaling in bEnd.3 TOPFlash cells. CsA treatment rescued the reduction of canonical Norrin/Frizzled4 signaling after Atp2b1 knockdown (Figure 6E). This finding indicates that calcineurin/NFAT activation contributes to inhibiting the Norrin/Frizzled4 pathway. Furthermore, we used lentivirus to transfer constitutively active NFAT2 (aka NFATc1) into bEnd.3 cells and selected a stable population using blasticidin. This population exhibited significantly elevated expression of Egr3 and Nr4a2 compared to parental cells (Figures 6F and 6G). While Norrin treatment successfully induced the expression of its target genes Axin2 and P2ry1 in control bEnd.3 cells, this response was virtually abolished in CA-NFAT2-overexpressing cells (Figures 6H and 6I). Together, these results support that ATP2B1 deficiency activates NFAT signaling, which in turn contributes to suppressing Norrin/Frizzled4 signaling.

Figure 6. ATP2B1 deficiency regulates Norrin/Frizzled4 signaling through the calcineurin/NFAT pathway.

(A and B) bEnd.3 cells were transfected with Atp2b1 siRNA#3 or control siRNA and analyzed with quantitative reverse-transcription PCR (RT-qPCR) for Erg3 or Nr4a2 after stimulation with vehicle 1 μg/mL cyclosporin A, a calcineurin inhibitor. Two technical replicates per biological replicate were averaged, n = 2 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(C and D) bEnd.3 cells transfected with EGFP-hNFAT2-EE and the indicated siRNAs were scored for nuclear NFAT2 localization to determine NFAT2 activation. The EE tag is glutamate-rich tag derived from the polyoma virus medium T antigen. 3 images per biological replicate were averaged, n = 3 biological replicates, graphs represent mean ± SD, homoscedastic t test.

(E) Stable bEnd.3 TOPFlash cells were transfected with the indicated siRNAs and stimulated with vehicle, Norrin or Norrin and 1 μg/mL cyclosporin A. n = 9 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(F and G) bEnd.3 cells were infected with CA-NFAT2 lentivirus and selected with blasticidin. mRNA expression of Erg3 and Nr4a2 was assessed by RT-qPCR in parental bEnd.3 cells vs. CA-NFAT2 expressing cells. Two technical replicates per biological replicate were averaged, n = 2 biological replicates, graphs represent mean ± SEM, unpaired t test.

(H and I) Parental or CA-NFAT2 expressing bEnd.3 cells were stimulated with vehicle or Norrin and Axin2 and P2ry1 mRNA expression was assessed by RT-qPCR. Two technical replicates per biological replicate were averaged, n = 2 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

ATP2B1 facilitates WNT7A/B signaling and brain angiogenesis

Similar to Norrin/Frizzled4 signaling in retinal vascular development, neuroepithelial-derived WNT7A and WNT7B orchestrate critical aspects of neurovascular development in the developing brain and spinal cord. These ligands activate β-catenin-dependent signaling in ECs, thereby promoting both angiogenesis and BBB specialization. Next to Frizzled receptors, GPR124 functions as an essential co-activator in the WNT7A/B signaling axis (see Introduction). To investigate the role of ATP2B1 in Wnt/β-catenin signaling, we used 293T cells transfected with plasmids encoding WNT7A or WNT7B and co-cultured these cells with siRNA-treated bEnd.3 cells carrying the TOPFlash reporter. While control siRNA-treated cells activated canonical signaling in response to WNT7A and WNT7B, signaling amplitude was significantly reduced in Atp2b1-knockdown cells (Figure 7A). To further elucidate the requirement of ATP2B1 in Wnt/β-catenin signaling in vivo, we generated Atp2b1 ECKO; Gpr124^−/− compound mutant mice (Cre⁻ littermates were used as control, this included littermates that were heterozygous for the Gpr124 null or the Atp2b1 floxed allele). After maternal intraperitoneal injection of tamoxifen at E10.5, E11.5, and E12.5, brain sections were collected from embryonic day 13.5 (E13.5) embryos (Figure 7B). Previous studies have established that Gpr124-knockout embryos display pronounced defects in vascularization of the cerebral cortex and medial ganglionic eminence (MGE), accompanied by EC de-specialization and reduction of glucose transporter GLUT1 expression.¹⁴ Immunohistochemical analysis revealed that both wild-type and Atp2b1 ECKO mice exhibited largely normal vascularization patterns in the cerebral cortex and MGE. In contrast, both Gpr124 KO and Atp2b1 ECKO; GPR124^−/− compound mutant mice displayed glomeruloid vascular structures in the cerebral cortex and aberrant vascular aggregation in the MGE (Figure 7C). Notably, the severity of these phenotypes was significantly more pronounced in the Atp2b1 ECKO; Gpr124^−/− mice compared to Gpr124 single KO mice, indicating that ATP2B1 deficiency exacerbates the angiogenesis defects observed in Gpr124^−/− mice (Figures 7D and 7E). Collectively, these data support that ATP2B1 genetically interacts with components of WNT7A/B signaling during brain vascular development.

Figure 7. ATP2B1 regulates WNT7A/B-induced Frizzled signaling in the developing brain.

(A) Stable bEnd.3 TOPFlash cells were co-seeded with 293T cells, the latter were transiently transfected with plasmids encoding WNT7A or WNT7B before co-seeding. n = 3 biological replicates, graphs represent mean ± SEM, one-way ANOVA with Tukey’s post hoc.

(B) Orientation of sections analyzed in (C).

(C) E13.5 forebrain 20 μm sections of the indicated genotype. White boxes show areas in the two developing hemispheres and are shown enlarged in the right images. White boxes labeled a show the developing forebrain cortex, white boxes labeled b show the medial ganglionic eminence (MGE). White arrows point to glomeruloid vascular malformations. Scale bars: 200 μm.

(D) Quantification of a total number of glomeruloid vascular malformations in the cerebral cortex. n = 3–6 embryos, graphs represent median, one-way ANOVA with Tukey’s post hoc.

(E) Maximum length of vascular malformation in MGE, n = 3–6 biological replicates, graphs represent median, one-way ANOVA with Tukey’s post hoc.

DISCUSSION

A major conclusion from this study is that ATP2B1 facilitates Norrin- and WNT7A/B- signaling in CNS ECs. This conclusion is supported by the effects of Atp2b1 reduction in multiple independent cell-based signaling assays and the phenotypes of genetic mouse models. Mechanistically, increased Ca²⁺/NFAT signaling associated with Atp2b1 knockdown is involved in suppressing Norrin/Frizzled4 signaling. The mechanisms downstream of NFAT remain to be determined, but it is noted that NFAT can exert its functions via both transcriptional⁷⁰ and non-transcriptional mechanisms.⁷¹ The evidence that ionomycin or CA-NFAT2 overexpression each reduce Norrin/Frizzled4 signaling does not rule out that ATP2B1 deficiency could affect Norrin/Frizzled4 signaling via additional mechanisms besides NFAT signaling. Atp2b1 knockdown in ECs did not inhibit canonical signaling induced by the GSK3β inhibitor CHIR99021 to the same extent as Norrin/Frizzled4 signaling, suggesting functions of ATP2B1 or Ca²⁺/NFAT signaling upstream of β-catenin.

A second major conclusion is that ATP2B1 functions in the regulation of retinal angiogenesis and the BRB. There is overlap between phenotypes resulting from loss of Norrin/Frizzled4 signaling and Atp2b1 LOF in the retinal vasculature. In both situations deep vascular plexus formation is impaired, the fenestration marker PLVAP is upregulated, and BRB dysfunction is evident.⁷² Yet, given the broad roles of Ca²⁺ in EC signal transduction,^48,52,53 Atp2b1 ECKO phenotypes may not be exclusively due to impaired Norrin/Frizzled4 signaling. Norrin and WNT7A/B signaling are pathways that enable angiogenesis in the CNS but are not required for systemic angiogenesis. We observed angiogenesis phenotypes in Atp2b1 ECKO mice in the retina, and exacerbated angiogenesis phenotypes in Gpr124;Atp2b1 ECKO compound mutant mice in the embryonic brain. Further studies are needed to determine if ATP2B1 functions in ECs are CNS specific.

Our finding that Ca²⁺/NFAT signaling inhibits Norrin/Frizzled4 signaling implies that factors that regulate EC Ca²⁺, including VEGF,⁵³ ROS,⁷³ thrombin,^48,74 bradykinin,^75,76 or mechanosensory stress⁵¹ could reduce the levels of Norrin/Frizzled4 signaling in ECs and contribute to increased vascular permeability, for example in diabetic retinopathy. The data suggest that other genes that are involved in Ca²⁺ homeostasis, including the ATP2B1 paralog ATP2B4, may functionally interact with β-catenin-dependent signaling in ECs. Furthermore, Ca²⁺-signaling and Wnt/β-catenin signaling potentially intersect in additional biological contexts.⁷⁷

Our study revealed additional proximity interactions of FZD4 beyond ATP2B1 (Figure S5). These interactions include DLG1, GORASP2, CTNND1/p120, and ITGB1. DLG-1 is a PDZ domain scaffold protein that was identified as a Frizzled4 interactor in a yeast-two-hybrid screen.⁶¹ Importantly, EC-specific gene ablation of Dlg1 results in a reduction of retinal deep vascular plexus density and increased PLVAP expression. Furthermore, Dlg1 genetically interacts with Fzd4 and Tspan12 with respect to retinal vascular phenotypes, and with Tspan12 and Ndp regarding BBB phenotypes. These phenotypes are rescued by stabilized β-catenin. DLG1 interacts with FZD4 via PDZ domains 1 and 2.⁶⁰ This study provides compelling evidence for complex formation of DLG1 and FZD4 and our identification of DLG1 among the proximity interactors in three cell lines validates the proximity biotinylation screens on a technical level. GORASP2 promotes the transport of cargo with PDZ recognition motifs, including FZD4.⁶² Thus, our screen identified at least two proteins that can interact with the PDZ recognition motif of FZD4. Additional proteins related to trafficking that were identified in all three cell lines include ACBD3 (a Golgi-resident protein) and SNAP23 (a T-SNARE). Another protein consistently identified in all three cell lines is ITGB1. The significance of this finding is unclear, but it is noted that the ITGB1 interactor integrin-linked kinase (ILK) is implicated in regulating retinal angiogenesis through β-catenin-dependent signaling and is genetically linked to FEVR.⁷⁸ Furthermore, we identified catenin delta 1 (aka p120-catenin, CTNND1) in 293T and HeLa cells. This is of interest, as a role for CTNND1 in β-catenin signaling has been attributed to binding LRP5/6 and the regulation of internalization of the signalosome into multi-vesicular bodies,⁷⁹ which is thought to lead to β-catenin stabilization.^58,80 Mutations in human CTNND1 or mouse Ctnnd1 cause retinal hypovascularization reminiscent of FEVR, due to roles in adherens junctions, or β-catenin-dependent signaling, or both.⁸¹ The finding of a proximity interaction between CTNND1 and FZD4 suggests that additional roles of CTNND1 in FZD4-signaling remain to be uncovered.

FZD4 proximity interactors may have functions beyond β-catenin-dependent signaling, for example, the identification of VANGL1 in all three cell lines stands out. Furthermore, there may be functionally significant interactors that we detected only in one of three cell lines, e.g., some EC-specific interactors may not be expressed in 293T or HeLa cells. UNC5B was only detected in bEnd.3 ECs, and this protein is implicated in BBB function.⁸² TSPAN12, which is physically associated with FZD4,¹⁶ was co-transfected into 293T cells and was only identified in those cells. Finally, proximity interactors may not necessarily modulate FZD4 function, instead, FZD4 may modulate the function of the nearby proteins. Together, the proximity biotinylation screens yielded numerous interactors of interest and led to the identification of ATP2B1 as a facilitator of Norrin and WNT7A/B signaling. Our mechanistic studies identify Ca²⁺/NFAT as an inhibitor of β-catenin-dependent signaling in ECs and imply that pathways and pathological conditions that increase EC Ca²⁺ have the potential to modulate β-catenin-dependent signaling and regulate angiogenesis and barrier function. In addition, this mechanism may be relevant in the context of pharmacological approaches that transiently disrupt endothelial barriers for drug delivery.

Limitations of the study

Our genetic experiments conducted on a C57BL/6J background revealed significant variability in the retinal vascular phenotypes of Atp2b1 ECKO mice and the reason(s) for this variability remain undefined. Possible explanations include variable efficiency of recombination, variable expression of redundant factors in ECs (e.g., ATP2B4), or variable compensation in Ca²⁺-handling. The fact that the machinery regulating intracellular Ca²⁺ is highly complex may make Atp2b1 ECKO mice susceptible to the effect of genetic modifiers acquired through genetic drift.

Whether Atp2b1 ECKO mice compounded with homozygous loss of Tspan12 display a genetic interaction remains to be determined. The complete lack of deep vascular plexus formation and very severe BRB dysfunction in Tspan12 KO mice may obscure the detection of further reduction of deep vascular plexus formation and BRB function when ATP2B1 is reduced in addition to loss of TSPAN12.

Our identification of proximity interaction, co-localization, and split-GFP reconstitution is consistent with proximity of FZD4 and ATP2B1, but experimental support for complex formation or the identification of a direct interaction interface is currently lacking. The mechanisms by which NFAT signaling intersects with Norrin/Frizzled4 signaling in ECs remain to be investigated.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact Harald J. Junge (junge@umn.edu).

Materials availability

Materials generated in the Junge laboratory will be available from the lead contact but may require a completed materials transfer agreement (MTA).

Data and code availability

• RNA-seq data have been deposited at NCBI GEO under accession number GSE299490.

• Proteomics data are available in the PRIDE database under PXD065058 and PXD077093.

• This study does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

Atp2b1 floxed mice (Atp2b1^{tm1c(KOMP)Wtsi}) were generated from Atp2b1^{tm1a(KOMP)Wtsi} sperm by Dr. Klaus-Dieter Fischer’s lab⁴⁷ and imported to Minnesota with permission from MMRRC UC Davis. Atp2b1 floxed mice were imported on a C57BL/6J background and subsequently backcrossed at the Minnesota site for 7 additional generations using C57BL/6J mice obtained from Jackson laboratories. Experimental mice were up to 13 days old (P13). The Atp2b1 null allele was generated by recombining the floxed Atp2b1 allele in the germline using Sox2-Cre (Jackson lab stock # 008454). Tg(Cdh5-cre/ERT2)1Rha⁸³ mice were used for EC-specific recombination and provided by Dr. Ralf Adams under MTA with CANCERTOOLS.ORG. The Tem5 aka Gpr124 null allele¹⁸ was kindly provided by Dr. Brad St. Croix under MTA with NCI. Studies were conducted using animals of both sexes and data were not analyzed disaggregated for sex as the study investigating multiple compound mutant genotypes was not powered for this analysis. Mice were housed in a specific pathogen-free animal facility. The Animal Care and Use Committee of the University of Minnesota, Twin Cities, approved all animal protocols.

Cell lines

bEnd.3 cells, COS-1 cells, HEK293T cells, and HeLa cells (all female-derived) were cultured in high-glucose DMEM supplemented with 10% FBS at 37°C in a humidified atmosphere containing 5% CO₂. Cell lines were routinely tested for mycoplasma contamination. Cell lines were received from ATCC and not further authenticated other than by morphological criteria.

METHOD DETAILS

Plasmids and virus

pcDNA 3.3 V5-hFZD4-BioID was generated by PCR-cloning the BioID coding sequence into the BamHI and AgeI sites of pcDNA3.3 V5-hFZD4-DmrAA, which we described previously.¹⁶ This construct was used for transient transfections into 293T and HeLa cells. pLenti-CMV-V5-hFZD4-BioID-IRES-BlastR and pLenti-CMV-V5-hFZD4 were generated by PCR amplification and subcloning into the AgeI and XhoI sites of a pLenti-CMV-MCS-IRES-BlastR vector. This vector was obtained by excising the AgeI/XhoI turboRFP insert from Addgene plasmid #102343 (deposited by Ghassan Mouneimne) and replacing it with a primer dimer harboring an MCS. pLenti-CMV-flag-hATP2B1-IRES-BlastR was generated by subcloning a gene synthesis-derived flag-hATP2B1 fragment into the BamHI and NheI sites of pLenti-CMV-MCS-IRES-BlastR – this fragment harbors silent mutations that transfer resistance to specific siRNAs. pLenti-CMV-3xHA-CA-NFAT2-IRES-BlastR was generated by PCR cloning 3xHA-CA-NFAT2 into the AgeI and XhoI sites of pLenti-CMV-MCS-IRES-BlastR. The 3xHA-CA-NFAT2 sequence was obtained from Addgene plasmid #11102, deposited by Anjana Rao. pcDNA3.3 hWnt7b was described previously.¹⁶ pcDNA Wnt7a was obtained from Addgene (#35914, deposited by Marian Waterman). Virus was generated by the UMM Viral Vector and Cloning Core.

TOPFlash luciferase assay

293T were transfected with a plasmid mix that we described previously,¹⁶ in brief: 4 ng pcDNA.3.3 hFZD4, 8 ng pcDNA.3.3 hLRP5, 8 ng pcDNA3.3 hTSPAN12 and 140 ng reporter mixture were transfected with TransIT LTI according to manufacturer’s instruction per 48 well. 6–8 h after transfection, cells were stimulated with 200 ng/mL Norrin (R + D Systems, 3014-NR-025) for 12–16 h and processed with the Promega Dual-Glo lysis buffer and substrates according to manufacturer’s instructions. For raising intracellular Ca²⁺, ionomycin was present from the time cells had adhered and maintained throughout the experiment. For TOPFlash experiments in bEnd.3 cells, we used our previously described⁶⁵ stable bEnd.3 population infected with two lentivirus encoding TOPFlash (puromycin resistance) and CMV-Renilla (blasticidin resistance), cultured under selection pressure in high glucose DMEM, 10% FBS, 1% penicillin/streptomycin, 1 μg/mL puromycin and 10 μg/mL blasticidin. 30K cells were seeded per 48-well in full DMEM without puromycin and blasticidin and incubated overnight. Medium was replaced (200 μL per well) and cells were stimulated for 18 h with 200 ng/mL Norrin (R + D Systems) before harvest with Dual Glo lysis buffer and substrates (Promega, E1910). For raising intracellular Ca²⁺, ionomycin or SEA0400 (or DMSO as vehicle) was present 16 h before adding norrin and maintained throughout the experiment. For bEnd.3 TOPFlash experiments in the presence of Cyclosporin A, cells were transfected with siRNAs. 48 h after transfection, cells were stimulated for 24 h with vehicle, Norrin, or Norrin in the presence of 1 μg/mL CsA. TOPFlash data are normalized to the respective condition without stimulus (-Norrin) within one experimental group, where each siRNA transfection is a separate experimental group. To stimulate bEnd.3 TF cells with WNT7A or WNT7b, bEnd.3 TF cells were seeded into a 6 well plate (30K cells per well) and transfected the following day with either 20 nM final concentration of control siRNA or ATP2B1 siRNA #3 and incubated for 48 h 293T cells were seeded into a 6 well plate (1 million cells per well) and transfected 6 h later with 2 μg pcDNA3.3 EGFP, pcDNA-WNT7a, or pcDNA3.3 hWnt7b) expression vectors. Transfected 293T cells were incubated for 42 h post-transfection. For co-culture, siRNA-transfected bEnd.3 TOPFlash cells (30K cells per well) were combined with GFP- or Wnt7-transfected 293T cells (60K cells per well) in a 48 well plate. After 24 h of co-culture, cells were subjected to the Dual Glo luciferase assay.

Norrin-induced FZD4 endocytosis

HeLa cells were transfected with FZD4 and subjected to endocytosis as described previously.⁵⁸ In brief: 36 h after transfection, cells were incubated with ice-cold conditioned medium containing flag-alkaline phosphatase-Norrin for one hour on ice, washed with cold medium, and transferred to 37°C for 1 h to induce endocytosis. Non-internalized flag-AP-Norrin was removed by incubation with 150 mM NaCl, 50 mM glycine, pH 2.4 for 3 min at RT before cells were fixed, permeabilized, and stained using antibodies directed against the V5-tag and FLAG tag.

BioID proximity labeling

293T, HeLa, or bEnd.3 cells were cultured as described above. For the experiment using 293T cells, four 10 cm diameter dishes per condition were each seeded with 6 million cells. 3 h later, each dish was transfected with 400 ng pcDNA.3.3 V5-hFZD4-BioID or GFP-BioID, 800 ng pcDNA.3.3 hLRP5, 800 ng pcDNA 3.3 hTSPAN12, 6000 ng empty vector, 2000 ng pcDNA.3.3 EGFP using 15 μL TransIT-LTI. After 24 h, medium was replaced and 50 μM biotin was added for 24 h 293T cells were harvested by scraping, washed in PBS twice, and frozen. Sample IDs were S479 and S482 for duplicate GFP-BioID samples, and S480 and S483 for duplicate FZD4- BioID samples. For the experiment using HeLa cells, 2 million cells were seeded into each of four 10 cm diameter dishes and transfected with 250 μL linear PEI per 10 μg total plasmid DNA (9 μg pcDNA3.3 EGFP, which we used as filler vector for vector balancing, plus 1 μg of either V5-FZD4-BioID or GFP-BioID). After 24 h, cells were incubated for an additional 24 h with 50 μM biotin. Cells were washed in PBS, harvested using enzyme free cell dissociation solution (Gibco #13151–014), and frozen. GFP-BioID was sample S455, V5-FZD4-BioID was sample S456. For the experiment using bEnd.3 cells, cells were infected with lentivirus for gene transfer of V5-FZD4-BioID or GFP- BioID. After selection with blasticidin, four 150 cm² dishes per sample were incubated for 24 h with 50 μM biotin. Duplicate samples were generated. Cells were washed in PBS and directly lysed. Cells from all experiments were processed according to a detailed method paper.⁸⁴ In brief, cells were lysed in 8 M urea 50 mM Tris, pH 7.4, buffer with protease inhibitor (87785; Thermo Fisher Scientific) and DTT under sonication, precleared with gelatin-sepharose beads (17095601; Cytiva), incubated with streptavidin-sepharose beads (17511301; Cytiva/GE Healthcare), washed in 8 M urea, 50 mM Tris, pH 7.4, and resuspended in 50 mM ammonium bicarbonate saturated with 1 mM biotin.

Protein digestion and mass spectrometry

HeLa and 293T cell samples were reduce-alkylated in 8 M urea, 50 mM ammonium bicarbonate buffer with 10 mM tris(2-carboxyethyl)phosphine (TCEP) at 30°C for 60 min and 30 mM iodoacetamide (IAA) in the dark at room temperature for 30 min. Samples were diluted to 1M urea and digested with mass spec grade Trypsin/LysC mix overnight (V5071; Promega). The peptide eluates were acidified with formic acid and purified via a C18 matrix (5190–6532; Agilent) using Agilent AssayMap BRAVO liquid handling system. Solvent was removed in a SpeedVac and subjected to LC-MS/MS. The 293T and HeLa samples were analyzed at the SBP Discovery Core San Diego using Proxeon EASY nanoLC system coupled to a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). MS spectra were analyzed with MaxQuant 1.5.5.1 software and searched against the Homo sapiens Uniprot protein sequences database (version January 2018) and GPM cRAP sequences (common contaminants). Precursor mass tolerance was set to 20 ppm for the initial mass recalibration and then 4.5 ppm for the main search. Product ion mass tolerance was set at 0.5 Da and maximum precursor ion charge state was 7. Carbamidomethylation of cysteines was searched as a fixed modification, while oxidation of methionines and acetylation of protein N-terminal were searched as variable modifications. Enzyme was set to trypsin with max 2 missed cleavages. The target-decoy-based false discovery rate (FDR) filter for spectrum and protein identification was set to 1%.

bEnd.3 cell samples were processed at the Center for Metabolomics and Proteomics at the University of Minnesota. We reconstituted the dried peptide mixtures in 98:2:0.1,H2O:acetonitrile (ACN):formic acid (FA) and analyzed ~200 nanogram of each sample by capillary LC-MS on an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific, Inc., Waltham, MA) online with a Thermo UltiMate 3000 RSLCnano LC system. Peptides were separated on a 40 cm self-packed C18 capillary column with 100 μm inner diameter, with Dr. Maisch GmbH ReproSil-PUR 12 ÅC18-AQ, 1.9 μm particle size; the column was maintained at 55°C with a column heater from Sonation (Biberach, Germany). Peptides were loaded directly on column at 325 nL/min with 98:2:0.1,H2O:ACN:FA. We performed elution of the peptides with the following gradient: 5–32% solvent B from 0 to 50 min, 32–45% solvent B from 50 to 55 min, 45–90% solvent B from 55 to 60 min 325 nL/min, where solvent A was 0.1% formic acid in water and solvent B was 0.1% formic acid in ACN. We operated the mass spectrometer with the following parameters: ESI voltage 2.1kV, ion transfer tube 275°C; Orbitrap MS1 scan 120k resolution in profile from 380 to 1580 m/z with 50 msec injection time and 120% normalized automatic gain control (AGC); MS2 triggered on precursors charge states 2–5 above 20000 counts; precursor isolation window 1.6 Da; MIPS (monoisotopic peak determination) set to Peptide; higher energy collisional dissociation (HCD) at 35% with MS2 Orbitrap detection at 50k resolution (at 200 m/z), 86 msec injection time, 100% (1000) AGC, dynamic exclusion duration 15 s with ±10 ppm mass tolerance.

We processed peptide tandem MS using SEQUEST (Thermo Scientific) in Proteome Discoverer 4.0. The mouse Universal Proteome database (UP000000589) was downloaded from UniProt on 11/19/2021 and merged with a common lab contaminant protein database (https://www.thegpm.org/crap/) (55,214 total protein sequences). We applied the precursor mass recalibration node with precursor mass tolerance 20 ppm, product ion tolerance 0.2 Da with fixed carbamidomethyl (CAM) modification of cysteine 57.0215 m/z. The SEQUEST⁸⁵ database search parameters included enzyme trypsin full specificity, 2 missed cleave sites; precursor tolerance 15 ppm, fragment ion tolerance 0.05 Da and maximum 4 dynamic modifications per peptide. We specified CAM cysteine (+57.021 Da) as a fixed modification and the dynamic modifications were acetylation of protein N terminus (+42.011 Da), oxidation of methionine (+15.995 Da), conversion of glutamine to pyroglutamic acid (−17.027 Da), methionine loss at the protein N terminus (−131.040 Da), methionine loss + acetylation at the protein N terminus (−89.030 Da), biotinylation of lysine (226.078 Da) and deamidation of asparagine or glutamine (+0.984 Da).

We applied 1% protein and peptide False Discovery Rate (FDR) filters using the Percolator algorithm (https://doi.org/10.1038/nmeth1113) in PD.

We used the label free quantification workflow in PD 2.4 that included steps for feature extraction, chromatographic alignment, peptide mapping to features, protein abundance calculation, normalization, protein relative abundance ratio calculation and hypothesis testing for significance of relative fold chance. We applied the Minora Feature Detector algorithm in PD 2.3 for the assignment of chromatographic features for isotopically related peaks within a 0.2 min retention time range. After features were assigned, chromatographic alignment was performed; the file with the largest number of features was assigned as the reference to which features from each file were aligned within a 4 min (RT) window tolerance and 12 ppm mass tolerance. Retention times were adjusted from a non-linear regression-based fit of the differences in retention time between reference and sample. Peptide were mapped to retention time-aligned consensus features across samples with the requirement that at least one sample contains a peptide spectral match. We applied total peptide summed normalization across samples. Protein abundance calculations were made from summed peptide abundances. We used the background t test for hypothesis testing with the null hypothesis of equal protein abundance. Protein relative abundances between groups were reported with p-values adjusted with the Benjamini-Hochberg method for multiple testing corrections.

Filtering of priority candidates

Candidate interactors were prioritized based on multiple criteria: an FZD4/GFP label-free quantification (LFQ) abundance ratio >10 for HeLa and HEK293T cells and >5 for bEnd.3 cells; sequence coverage thresholds of >20% (HeLa), >15% (HEK293T), and >14% (bEnd.3); inclusion among the top 50 genes by absolute LFQ abundance in FZD4-BioID samples; MS/MS spectral count filters (excluding proteins with <5 counts in HeLa and HEK293T or <3 counts in bEnd.3); and finally sorted by the absolute MS/MS count of FZD4-BioID. Selected candidate interactors from each dataset were visualized using WordCloud.com and are also displayed in the supplement in tabular form.

Immunofluorescence colocalization assay

30K cells HeLa cells were seeded into a 8-well chamber slide and transfected with 80 ng pcDNA3.3 V5-FZD4 and 80 ng pLenti-CMV-flag-hATP2B1-IRES-BlastR using 0.3 μL TransIT LTI (Mirus). Twenty-four hours post-transfection, cells were fixed for 10 min with ice-cold methanol, washed three times with PBS, and then blocked and permeabilized for 30 min in PBS containing 0.1% Triton X-100 and 5% goat serum. Cells were incubated overnight at 4°C with the following primary antibodies: mouse α-V5 (1:500, Bio-Rad, MCA1360) and rabbit α-Flag (1:500, Cell Signaling clone D6W5B #14793). The next day, cells were washed three times with blocking buffer and incubated with secondary antibodies: goat α-mouse Alexa 488 (1:2000, Invitrogen, A11001), goat α-rabbit Alexa 555 (1:2000, Invitrogen, A21428), and DAPI (1:1000) for nuclear staining. After final washes, cells were mounted using Fluoromount-G (Invitrogen, 00–4958-02). Images were acquired using a Leica DMIL epifluorescent microscope or a Keyence BZ-X810 digital microscope.

Split-GFP reconstitution

40K COS-1 cells were seeded into a gelatin-coated 8-well chamber-slide (LABTEK) and transfected with pLenti-CMV-flag-hATP2B1-mNG2(11) and pcDNA3.3 V5-hFZD4-mNG2(1–10) or BACE1-mNG2(1–10) and mCherry using a total of 160 ng plasmid DNA and 0.3 μL TransIT-LTI (Mirus). After 48 h, cells were fixed with 4% PFA for 10 min at RT and imaged using a Keyence BZ-X810 digital microscope.

siRNA knockdown in bEnd.3 cells

bEnd.3 cells were cultured as described above. 300K cells were seeded in each well of a 6-well plate and allowed to adhere overnight. The following day, cells were transfected with siRNA at a final concentration of 20 nM using Lipofectamine RNAiMAX (Invitrogen, 56532) according to the manufacturer’s instructions. For Western blot analysis of knockdown efficiency after 48 h, bEnd.3 cell lysates were probed with anti-ATP2B1(Abcam ab190355) and anti-β-actin (Novus NB600–501H). For TOPFlash reporter experiments in bEnd.3 cells, 48 h after siRNA transfection, cells were treated with either recombinant Norrin (R + D Systems, 200 ng/mL) or the GSK3β inhibitor CHIR99021 (2 μM) in fresh DMEM. Cells were incubated for an additional 24 h before RNA extraction.

Mouse Atp2b1 and Fzd4 siRNA was from Sigma. Nucleotides are ribonucleotides unless otherwise denoted with d for deoxynucleotide. siAtp2b1 #1 (5^′- GUCAUGGGCCAGUGGUCAA-dTdT-3^′ and 5^′- UUGACCACUGGCCCAUGAC-dTdT-3^′), siAtp2b1 #2 (5^′-GGCUAAACACGAUCUCUGU-dTdT -3^′ and 5^′- ACAGAGAUCGUGUUUAGCC-dTdT-3^′), siAtp2b1 #3 (5^′-CUUUAUACCUCCUA AGAAG-dTdT-3^′ and 5^′-CUUCUUAGGAGGUAUAAAG-dTdT-3^′) and siFzd4 (5^′-CCUGUUAUUUCUAUGAAAU-dTdT-3^′ and 5^′- AUUUCAUAGAAAUAACAGG-dTdT-3^′). AllStars Neg. Control siRNA (Qiagen, 1027281) was used as non-targeting control.

RNA isolation and quantitative PCR

Total RNA was extracted from cultured cells using TRIzol reagent (APB Bioscience, FP312) according to the manufacturer’s instructions. Equal amounts of total RNA were reverse-transcribed into complementary DNA (cDNA) using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR (Thermo Fisher Scientific, K-1642). Quantitative PCR (qPCR) was carried out using SYBR Green master mix, and relative gene expression levels were calculated using the comparative Ct (ΔΔCt) method. Primers were designed to span exon–exon junctions to avoid genomic DNA amplification. Primer sequences are provided in Table S1.

Bulk RNA-sequencing

bEnd.3 cells were stimulated with vehicle (PBS, 0.1% BSA) or 200 ng/mL Norrin (R + D systems, 3014-NR-025) for 24 h. Total RNA was extracted from quadruplicate samples using Trizol (Invitrogen, Cat# 15596026) according to manufacturer’s instructions. Genomic DNA contamination was removed with TURBO DNA-free DNase treatment (Invitrogen, AM1907). Libraries were made using the TakaraBio SMARTer Pico Mammalian kit. Novaseq 150PE was carried out, generating approximately 45 million reads per sample. Sequence alignment to the mouse reference genome was conducted using the RNA-Seq module of SeqMan NGen (DNASTAR Lasergene). Differential expression analysis was carried out using ArrayStar (DNASTAR Lasergene). Data are available at NCBI GEO under GSE299490.

FZD4 cell-surface biotinylation

A stable population of V5-FZD4 expressing bEnd.3 cells was generated by lentiviral transduction and was maintained under blasticidin selection (10 μg/mL). 300K cells per 6-well were seeded into 2 mL of full DMEM without blasticidin. The next day, cells were transfected with 20 μM final siRNAs as described above. 48 h later, cell-surface biotinylation was performed. Brefeldin A treatment (3 μg/mL) was for 24 h before cell surface biotinylation. Cultures were transferred to ice to prevent endocytosis, the medium was removed, and cells were briefly rinsed with ice-cold PBS (pH 8.0). Surface proteins were labeled by incubating intact cells with 1.5 mL PBS (pH 8.0) containing 0.25 mg/mL EZ-Link Sulfo-NHS-SS-biotin (Thermo Fisher) for 45 min on ice with gentle rocking. Excess reagent was removed and then biotinylation was stopped by adding 400 μL PBS supplemented with 200 mM glycine (pH 8.0) for 5 min at 4°C. Cells were washed with 2 mL cold TBS. Cells were lysed in 0.8 mL cold RIPA buffer containing 0.5 mM CaCl₂, 2 mM MgCl₂ (to improve DNAseI activity), 0.2 mg/mL DNase I, and EDTA-free protease inhibitor cocktail. Lysis proceeded for 20 min on ice with intermittent vortexing, followed by addition of EDTA to 20 mM final. Insoluble material was removed by centrifugation (10 min, 20,000 × g, 4°C). For input samples, 104 μL of cleared lysate was mixed with 16 μL 1M DTT and 40 μL 4× LDS sample buffer and incubated at 45°C for 10 min. The remaining lysate was incubated with 50 μL pre-equilibrated NeutrAvidin agarose (Thermo Fisher) for 1 h at 4°C with rotation to capture biotinylated proteins. Beads were transferred to spin filter columns connected to a vacuum manifold and washed 10 × 0.8 mL with 0.1x RIPA diluted with 150 mM NaCl solution at room temperature. Bound proteins were eluted by incubation with 105 μL of 1x LDS sample buffer containing 100 mM DTT for 20 min at 45°C with agitation, followed by collection of eluates by centrifugation. Samples were analyzed by immunoblotting using HRP-conjugated antibodies against V5 (Bio-Rad MCA1360P).

Tamoxifen preparation and administration

Tamoxifen (Sigma, T5648) was dissolved in sterile corn oil (Sigma, C8267) by rotating overnight at RT while protected from light using aluminum foil. The resulting solution was sterile filtered, aliquoted, and stored at −80°C for up to one month. For Cre induction, postnatal pups received 50 μg tamoxifen s.c. once daily on postnatal days P2, P3, and P4. Embryos received Tamoxifen through a pregnant dam by intraperitoneal injections of 100 μL tamoxifen solution (20 mg/mL; 2 mg per injection) once daily on embryonic days E10.5, E11.5 and E12.5.

Vascular analysis of retinal whole mounts

Mice were anesthetized using an isoflurane drop jar and euthanized via cervical dislocation. Eyes were enucleated and fixed immediately in 4% PFA for 15 min at RT. Retinas were dissected and blocked in 5% goat serum with 0.5% Triton X-100 in PBS for 1 h at RT. Immunostaining was performed overnight at 4°C in blocking buffer with primary antibodies: mouse α-PLVAP (1:100, BD Bioscience, 550563) and secondary antibodies: Goat α-rat cross-adsorbed Alexa 555 (1:500, Invitrogen, A21434), Isolectin B4 Alexa 647 (1:100, Invitrogen, I32450), and α-streptavidin Alexa 488 (1:400, Invitrogen, S11223). Between staining steps, retinas were washed five times for 30 min each in PBS containing 0.1% Triton X-100 at RT with shaking. Retinas were post-fixed and mounted for imaging.

Images were acquired using a Keyence BZ-X810 digital microscope. Whole retina images were captured using 10× objective with tile stitching. Areas lacking intermediate or deep vascular plexuses were manually measured using ImageJ. Individual vascular layers were captured at 20× magnification. PLVAP⁺ area and biotin⁺ area were quantified from 10× stitched images using a threshold established from control samples, which was consistently applied across all genotypes using ImageJ. PLVAP⁺ area and biotin-⁺ area was normalized by total retina area.

Administration of biotin

To assess vascular permeability, sulfo-NHS-LC-biotin (Thermo Fisher, 21335) was administered intraperitoneally at a dose of 10 μL/g body weight (20 mg/mL solution) to postnatal day 13 (P13) pups. The biotin tracer was allowed to circulate for 60 min prior to sacrifice. Eyes were enucleated, and retinas were immersion-fixed in 4% PFA for 15 min at RT. Biotinylated proteins were visualized using 1:400 streptavidin conjugated to Alexa Fluor 488 (Invitrogen, S11223).

Fluo-4 assay

bEnd.3 cells were seeded into a 48 well plate at ~50% confluency. The day after plating, the cells were transfected with a control siRNA or ATP2B1 siRNA #3. After 72 h, intracellular calcium was measured using the Fluo-4 Direct Calcium Assay Kit (Thermo Fisher Scientific, F10472). A 2X Fluo-4 Direct calcium reagent loading solution was prepared following the manufacturer’s protocol and was supplemented with 5 mM probenecid and 2 μg/mL Hoechst 33342. An equal volume of the 2x loading solution was added to the cells (e.g., 250 μL/well for a 48 well plate, the final concentration for probenecid is 2.5 mM and for Hoechst 33342 is 1 μg/mL). After incubation at 37°C for 30 min to allow Fluo-4 and Hoechst 33342 dyes to load into the cells, the fluorescent signals from Fluo-4 and Hoechst 33342 were captured and quantified using a Celigo Image Cytometer (Revvity). For each well, the integrated Fluo-4 fluorescence intensity per well was normalized to that from Hoechst 33342.

NFATC2 nuclear localization

35K bEnd.3 cells in full high-glucose DMEM were seeded in a gelatin-coated 8-well chamberslide (LABTEK) and transfected with 20 nM final siRNA 2 h later. After additional 14 h, the medium was replaced. Additional 2 h later, cells were transfected with 50 ng EGFP-hNFAT2-EE-WT (Addgene #24219, deposited by Jerry Crabtree) plus 110 ng empty vector using 0.3 μL Lipofectamine 3000 and 0.3 μL P3000 condenser. After additional 36 h, cells were fixed with 4% PFA for 10 min at RT and imaged using a Keyence BZ-X810 digital microscope. Cells were counted in two categories: nuclear GFP signal weaker than cytoplasm, and nuclear GFP signal stronger than cytoplasm.

Immunostaining of embryonic brain sections

Embryos were collected at E13.5 and fixed in 4% PFA overnight. Subsequently, embryos were washed with PBS and immersed in 30% sucrose for 24 h. Embryos were then rinsed in PBS and frozen in OCT. Embryonic brains were sectioned in 20 μm thickness and blocked in 5% goat serum with 0.5% Triton X-100 in PBS for 1 h at RT. Brains were incubated overnight at 4°C with the following primary antibodies: rabbit α-mouse Glut-1 (1:200, Cell signaling, 73015T). After washing, the following secondary antibody or lectin were applied: Goat α-rabbit Alexa 647 (1:1000, Invitrogen, A21245) and Isolectin B4 Alexa 488 (1:100, Invitrogen, I21411). Images were acquired using a Keyence BZ-X810 digital microscope.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were conducted using either the Statistics Kingdom web application or GraphPad Prism software. The specific statistical tests used are indicated in the corresponding figure legends. The sample size (n) is provided and defined in the figure legends. Data were assessed for normality and variance and are presented as mean ± standard error of the mean (SEM) or standard deviation (SD), as indicated in the figure legends. For comparisons among multiple groups, one-way ANOVA followed by Tukey’s post hoc test or Dunnett’s post hoc were used for normally distributed data, while the Kruskal–Wallis test was applied for non-parametric data. Two-group comparisons were performed using an unpaired t test, Welch’s t test, or the Mann–Whitney U test, as determined by the Statistics Kingdom app. A p-value <0.05 was considered statistically significant, with significance denoted as follows: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2026.117605.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
V5	BioRad, Cat#MCA1360	AB_322378
Flag	Cell Signaling, Cat#14793	AB_2572291
ATP2B1 (PMCA1)	Abcam, Cat#190355	AB_2893200
β-actin	Novus, Cat#NB600-501H	AB_1216153
PLVAP	BD Bioscience, Cat#550563	AB_393754
GLUT-1	Cell Signaling, Cat#73015T	AB_3064908
Bacterial and virus strains
pLenti-CMV-V5-hFZD4-BioID-IRES-BlastR	This paper	N/A
pLenti-CMV-V5-hFZD4	This paper	N/A
pLenti-CMV-flag-hATP2B1-IRES-BlastR	This paper	N/A
pLenti-CMV-3xHA-CA-NFAT2-IRES-BlastR	This paper	N/A
Critical commercial assay
Dual-Luciferase® Reporter Assay System	Promega	Cat#E1910
Maxima First Strand cDNA Synthesis Kit	Thermo Fisher Scientific	Cat#K-1642
Fluo-4 Direct Calcium Assay Kit	Thermo Fisher Scientific	Cat#F10472
Chemicals, peptides, and recombinant proteins
Ionomycin	Sigma	Cat#407950
SEA0400	Selleck Chemicals	Cat#S6920
Cyclosporin A	Sigma	Cat#C1832
CHIR99021	Sigma	Cat#SML1046
Recombinant human Norrin	R&D systems	Cat#3014-NR
EZ-Link Sulfo-NHS-SS-biotin	Thermo Fisher	Cat#21331
EZ-Link Sulfo-NHS-LC-biotin	Thermo Fisher	Cat#21335
Tamoxifen	Sigma	Cat#T5648
Experimental Models: Cell lines
Human: 293T	ATCC	CRL-3216
Human: Hela	ATCC	CCL-2
Monkey: Cos-1	ATCC	CRL-1650
Mouse: bEnd.3	ATCC	CRL-2299
Mouse: bEnd.3-TOPFlash + Renilla stable population	This paper	N/A
Mouse: bEnd.3 V5-FZD4 stable population	This paper	N/A
Mouse: bEnd.3-CA-NFAT2 population	This paper	N/A
Experimental Models: Organisms/Strains
Mouse: Atp2b1 floxed mice		N/A
Mouse: Atp2b1 +/- mice	This paper	N/A
Mouse: B6.Cg-Edil3Tg(Sox2-cre)1Amc/J	https://doi.org/10.1016/s0925-4773(03)00099-6	JAX: 008454
Mouse: Cdh5creER+:Tg(Cdh5-cre/ERT2)1Rha	https://doi.org/10.1038/nature09002	N/A
Mouse: Gpr124 null allele	https://doi.org/10.1016/s0925-4773(03)00099-6	N/A
Oligonucleotides
Mouse Atp2b1 siRNA #1 sequence: 5'- GUCAUGGGCCAGUGGUCAA-dTdT-3'	This paper	N/A
Mouse Atp2b1 siRNA #2 sequence: 5'-GGCUAAACACGAUCUCUGU-dTdT -3'	This paper	N/A
Mouse Atp2b1 siRNA #3 sequence: 5'-CUUUAUACCUCCUAAGAAG-dTdT-3'	This paper	N/A
Mouse Fzd4 siRNA sequence: 5'- CCUGUUAUUUCUAUGAAAU-dTdT-3'	This paper	N/A
qPCR primers for Atp2b1, Axin2, P2ry1, Egr3 and Nr4a2, see Table S1 in supplemental methods	This paper	N/A
Recombinant DNA
pcDNA3.3 V5-hFZD4-BioID	This paper	N/A
EGFP-hNFAT2-EE-WT	Beals CR et al. Nuclear localization of NF-ATc by a calcineurin-dependent, cyclosporin-sensitive intramolecular interaction. Genes Dev. 1997	Addgene plasmid #24219
pcDNA Wnt7a	Najdi R et al. A uniform human Wnt expression library reveals a shared secretory pathway and unique signaling activities. Differentiation. 2012	Addgene plasmid #35914
pcDNA3.3 hWnt7b	Lai M et al. TSPAN12 Is a Norrin Co-receptor that Amplifies Frizzled4 Ligand Selectivity and Signaling. Cell Reports. 2017	N/A
pcDNA3.3 V5-hFZD4-mNG2(1-10)	This paper	N/A
pLenti-CMV-FLAG-hATP2B1-mNG2(11)	This paper	N/A
pCAGGS BACE1-mNG2(1-10)	This paper	N/A
Software and Algorithms
MaxQuant software	Max Planck Institute, Version 1.5.5.1	https://maxquant.org/
SEQUEST	Thermo Scientific, Proteome Discoverer 4.0	N/A
SeqMan Ngen genome assembly software	DNASTAR	https://www.dnastar.com/software/lasergene/seqman-ngen/
GraphPad Prism software	Dotmatics	https://www.graphpad.com/
Deposited data
Proximity biotinylation data (bEnd.3 cells)	PRIDE	[PRIDE]: [PXD077093]
Proximity biotinylation data (HeLa and 293T cells)	PRIDE	[PRIDE]: [PXD065058]
Bulk RNA-Seq data	NCBI, GEO	GSE299490
Other
TransIT-LT1 Transfection Reagent	Mirusbio	Cat#MIR2304
Lipofectamine RNAiMAX	Invitrogen	Cat#56532
Enzyme free cell dissociation solution	Gibco	Cat#13151-014
Protease inhibitor	Thermo Fisher Scientific	Cat#87785
Gelatin-sepharose beads	Cytiva	Cat#17095601
Streptavidin-sepharose beads	Cytiva/GE Healthcare	Cat#17511301
Trypsin/LysC	Promega	Cat#V5071
C18 matrix	Agilent,	Cat#5190-6532
TRIzol reagent	APB Bioscience	Cat#FP312
TRIzol reagent	Invitrogen	Cat#15596026
TURBO DNA-free DNase treatment	Invitrogen	Cat#M1907
SMARTer Stranded Total RNA-Seq kit v3 Pico Input Mammalian	Takara Bio	Cat#634485
NeutrAvidin agarose	Thermo Scientific	Cat#29202
DNase I	Worthington Biochemical Corporation	Cat#LK003170
Griffonia Simplicifolia Isolectin B4-Alexa 647	Invitrogen, Cat#I32450	SCR_014365
Streptavidin, Alexa Fluor™ 488 Conjugate	Invitrogen, Cat#S11223	N/A

Highlights.

• BioID approach identifies aproximity interaction of FZD4 and ATP2B1

• Endothelial ATP2B1 deficiency causes angiogenesis and barrier defects in the CNS

• ATP2B1 deficiency reduces canonical Wnt signaling by activating NFAT signaling