RNF213 loss-of-function promotes pathological angiogenesis in moyamoya disease via the Hippo pathway

Abstract

Moyamoya disease is an uncommon cerebrovascular disorder characterized by steno-occlusive changes in the circle of Willis and abnormal vascular network development. Ring finger protein 213 (RNF213) has been identified as an important susceptibility gene for Asian patients, but researchers have not completely elucidated whether RNF213 mutations affect the pathogenesis of moyamoya disease.

Using donor superficial temporal artery samples, whole-genome sequencing was performed to identify RNF213 mutation types in patients with moyamoya disease, and histopathology was performed to compare morphological differences between patients with moyamoya disease and intracranial aneurysm. The vascular phenotype of RNF213-deficient mice and zebrafish was explored in vivo, and RNF213 knockdown in human brain microvascular endothelial cells was employed to analyse cell proliferation, migration and tube formation abilities in vitro. After bioinformatics analysis of both cell and bulk RNA-seq data, potential signalling pathways were measured in RNF213-knockdown or RNF213-knockout endothelial cells.

We found that patients with moyamoya disease carried pathogenic mutations of RNF213 that were positively associated with moyamoya disease histopathology. RNF213 deletion exacerbated pathological angiogenesis in the cortex and retina. Reduced RNF213 expression led to increased endothelial cell proliferation, migration and tube formation. Endothelial knockdown of RNF213 activated the Hippo pathway effector Yes-associated protein (YAP)/tafazzin (TAZ) and promoted the overexpression of the downstream effector VEGFR2. Additionally, inhibition of YAP/TAZ resulted in altered cellular VEGFR2 distribution due to defects in trafficking from the Golgi apparatus to the plasma membrane and reversed RNF213 knockdown-induced angiogenesis. All these key molecules were validated in ECs isolated from RNF213-deficient animals.

Our findings may suggest that loss-of-function of RNF213 mediates the pathogenesis of moyamoya disease via the Hippo pathway.

Moyamoya disease is a rare cerebrovascular disorder characterized by steno-occlusive changes in the circle of Willis. Ye et al. provide evidence that RNF213 loss-of-function mediates the pathogenesis of moyamoya disease via the Hippo pathway, and propose that inhibition of YAP/TAZ expression could be a therapeutic target.

Introduction

Moyamoya disease (MMD) is an uncommon cerebrovascular disorder characterized by progressive stenosis and/or occlusion of the circle of Willis with an abnormally proliferative vascular network at the base of the skull (moyamoya vessels).¹ MMD is a major cause of stroke with a relatively high risk of recurrent cerebrovascular events. This disease exhibits a diverse geographical distribution, with high incidence rates in East Asian countries, and a bimodal age of onset: 5–9 years of age and the middle 40s.^2,3 The pathological hallmarks of MMD have already been reported via a detailed scientific autopsy of intracranial large arteries, while the natural history of moyamoya vessels has been carefully described using the Suzuki staging system.^4–6 However, the aetiology and pathogenesis of MMD are not fully understood.⁷ Effective therapeutic strategies to alleviate disease deterioration and reverse disease progression are still lacking.⁸ Revascularization surgery is widely used to improve the vascular reserve capacity and prevent further cerebral vascular attacks, but postoperative hyperperfusion syndrome causes poor outcomes.⁸ Therefore, elucidation of the molecular mechanism of this disease might identify optimal treatments.

The ring finger protein 213 (RNF213) gene on chromosome 17q25.3 has been identified as an important susceptibility gene for MMD.^9,10 RNF213 is a 591 kDa protein with two functional domain structures (a Walker motif and a RING finger domain) with ATPase and ubiquitin ligase catalytic activities. Recent genome-wide association studies (GWASs) have revealed many single-nucleotide polymorphisms (SNPs) at the RNF213 locus in East Asian patients.^9–15 Of these SNPs, the mutant allele p.R4810K (c.14429G>A, rs112735431) is closely related to MMD and shows a strong genetic effect associated with earlier disease onset and worse outcome.⁷ This mutation is common and is present in 95% of patients with familial MMD, 73% of patients with sporadic MMD, and 1.4% of healthy controls.¹⁴ Although controversy persists regarding whether RNF213 mutations induce a loss-of-function or gain-of-function allele status, accumulating evidence has suggested that missense mutations might disrupt gene transcription and protein function to a certain extent. RNF213 mutation causes pathological dysregulation of substrate ubiquitination due to changes in the functional domain.¹⁶ In addition, several responsible RNF213 mutations have been identified in MMD patients through whole-exome sequencing (WES); these mutations include mainly a 21-bp in-frame insertion starting at amino acid 4951, two 3-bp deletions (p.A529del, p.K4115del) and a de novo genetic variant (c.11986_11989delCCGTinsGGGTTAG).^15,17 Some patients with MMD might present with RNF213 loss; thus, the role of these mutations is not clear. In this study, we aimed to determine the functions and underlying mechanisms of pathogenic RNF213 mutations in patients with MMD.

Materials and methods

A detailed ‘Materials and methods’ section is provided in the online Supplementary material.

Study approval

Animal experiments were conducted in accordance with the National Institutes of Health Animal Use Guidelines. Informed consent was obtained from every participant. This study was approved by the Ethics Committee for Clinical Research and Animal Trials of the First Affiliated Hospital of Sun Yat-sen University [(2022)330].

Statistical analysis

Values are presented as the means ± standard deviations (SDs). Two-tailed non-parametric tests followed by the Mann–Whitney U-test (for comparisons of two groups) or one-way ANOVA followed by the Kruskal–Wallis test (for comparisons of three or more groups) were used to calculate the statistical significance. A P-value < 0.05 was used to indicate statistically significant differences. All calculations were performed using GraphPad Prism software (v9.0) or R Project for Statistical Computing software (v4.1.2).

Results

Some MMD patients carry loss-of-function RNF213 mutations

Although only a few common RNF213 SNPs have been implicated in MMD to date, studies on the types of genetic variations have been informative and have provided clues about the pathogenesis of MMD. Donor superficial temporal arteries (STAs) from six patients with sporadic MMD undergoing direct bypass were used for whole-genome sequencing (WGS). The raw data from each sample were processed and analysed, as shown in Supplementary Fig. 1A. An average of 99.84% of the raw data aligned to the human reference genome. After 5.68% of duplicate reads were removed, the mean sequencing depth was 33.86×, and the sequencing coverage of the target region was 99.36% (Supplementary Fig. 1B and Supplementary Table 1).

First, variants with a frequency <1% in national reference databases, particularly the 1000 Genomes Project, the Single Nucleotide Polymorphism Database, and the Database of Genomic Variants (DGV), were retained. Next, variants located in the coding region or around the splicing site were retained. Ultimately, we focused principally on RNF213 variants. Except for reported missense and silent SNPs, no insertion/deletion (InDel) or structural variant (SV) of the RNF213 gene was detected in this study. However, we identified a homozygous deletion of the copy number variant (CNV) in the RNF213 coding sequence in two sporadic cases (33.33%) that might impair the transcription of exon 3 and exon 16 in the gene (Fig. 1A and Supplementary Table 2). Furthermore, the clinical significance of both CNVs was determined according to practical guidelines.¹⁸ One CNV deletion (GRch37: chr17:78287302-78291400 or GRCh38: chr17:80317600-80313502) was a benign CNV because of three similar records in the DGV. Another newly discovered CNV deletion (GRch37: chr17:78247202-78252500 or GRCh38: chr17:80278701-80273403) was compared with a known microduplication syndrome and the clinical features of patients in DECIPHER, a database of genomic variations and phenotypes in humans determined using Ensembl resources. This mutation was matched to not only CNV syndromes but also abnormalities of cardiovascular arteries and a likely pathogenic CNV deletion that was determined to be a clinically relevant CNV.

Figure 1.

RNF213 loss-of-function mutations exist in some MMD patients and are positively associated with intimal hyperplasia in MMD patients. (A) The integrative genomics viewer (IGV) application showed a benign copy number variant (CNV) deletion (chr17:78287302-78291400) that completely damages the genomic sequence of exon 16 and a likely pathogenic CNV deletion (chr17:78247202-78252500) that partly damages the genomic sequence of exon 3 in RNF213. The green boxes indicate a start codon, while the red boxes indicate termination codons in the three-frame translation. (B) The protein structure of the wild-type (WT) RNF213 protein (Q63HN8) was modelled with 100.0% confidence based on the single highest-scoring template of the well-known mouse Rnf213 (c7oikA). The benign CNV deletion of RNF213 showed a high identity (99.42%) with WT, while the clinically relevant CNV deletion of RNF213 showed a very low similarity (30.48%) with WT, and only 69 residues were modelled with 43.4% confidence using SWISS-Model and Phyre2. (C) Quantitative analysis of RNF213 and ubiquitin protein levels in one intracranial aneurysm (IA) patient and moyamoya disease (MMD) patient carrying the pathogenic RNF213 mutation of exon 3 (RNF213, n = 4, 1.00 ± 0.00 versus 0.84 ± 0.03, P = 0.0286; ubiquitin, n = 4, 1.00 ± 0.00 versus 0.78 ± 0.09, P = 0.0286). (D) The vascular smooth muscles were then stained with an α-SMA antibody, and the region of interest (ROI)-based quantitative analysis of α-SMA⁺ cells in the tunica media indicated a relative loss of smooth muscle in MMD patients (n = 18, 11944.81 ± 783.90 versus 9796.53 ± 962.64 cell/mm², P < 0.0001). Scale bar = 20 µm. (E) The density of Ki-67 was quantitatively analysed in the tunica intima, which indicated the presence of increased number of Ki-67⁺ cells in MMD patients with RNF213 mutation (n = 4, 7944.90 ± 706.61 versus 11 373.69 ± 1348.96 cell/mm², P = 0.0286). Scale bar = 20 µm. (F) Haematoxylin and eosin staining was used for morphometric analysis of the vascular wall structure of donor superficial temporal artery (STA) sections from MMD and IA patients. MMD patients showed a significantly thinner tunica media than IA patients (n = 18, 31.09 ± 4.35 versus 40.22 ± 4.33%, P < 0.0001). No significant difference in tunica media thickness was observed between MMD patients with RNF213 mutation and those without RNF213 mutation (n = 8 versus 10, 30.19 ± 4.86 versus 31.82 ± 4.01%, P = 0.3599). A significant difference in tunica media thickness was not observed between IA patients carrying RNF213 mutation and those without RNF213 mutation (n = 4 versus 14, 42.00 ± 5.27 versus 39.71 ± 4.11%, P = 0.4418). Scale bar = 200 µm (top); scale bar = 40 µm (bottom). (G) Weigert’s resorcin-fuchsin staining was applied for morphometric analysis of the internal elastic membranes in MMD and IA patients. MMD patients showed significantly thicker internal elastic membranes than IA patients (n = 18, 19.53 ± 11.16 versus 7.68 ± 5.12%, P = 0.0007). MMD patients carrying RNF213 mutation presented with a significantly greater tunica intima thickness than MMD patients without RNF213 mutation (n = 8 versus 10, 28.29 ± 7.87 versus 12.53 ± 8.02%, P = 0.0014). No significant difference in tunica intima thickness was observed between IA patients with RNF213 mutation and those without RNF213 mutation (n = 4 versus 14, 11.55 ± 2.54 versus 6.58 ± 5.18%, P = 0.0791). Scale bar = 200 µm (top); scale bar = 40 µm (bottom). Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the ROI. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001.

The sense DNA strand of wild-type RNF213 was downloaded from the Ensembl database, and 5207 amino acids of the RNF213 protein were obtained from the UniProt database to explore the effect of CNV deletions on RNF213 protein structure and function. The coding sequences of these CNV deletions were translated using ExPASy software. Then, coding sequence alignments between wild-type and CNV deletions showed that the benign CNV deletion only exhibited a very tiny gap (0.58%), but the clinically relevant CNV had a largely different identity from that of wild-type RNF213 (Supplementary Table 3). The 3D protein structures were automatically predicted using SWISS-MODEL and Phyre2 software. Compared with wild-type RNF213, the benign CNV deletion showed highly similar characteristics, while the clinically relevant CNV showed a completely different protein structure and functional domain (Fig. 1B).

In addition, 44 patients with MMD and 30 patients with intracranial aneurysm (IA) were enrolled in a clinical cohort at our hospital between July 2020 and July 2022. This cohort was used to measure these RNF213 mutations in peripheral blood DNA samples. The 20 STA DNA samples in MMD patients from this cohort were isolated for validation. Sanger sequencing was performed to determine the pathogenic RNF213 mutations (p.R4810K mutation, CNV deletion in exon 3) and benign CNV deletion in exon 16. These results revealed that the RNF213 p.R4810K mutation was more common in MMD patients than in patients with intracranial aneurysm (43.18% versus 33.33%). Neither CNV was observed in intracranial aneurysm patients, but a pathogenic CNV mutation in exon 3 was detected in seven MMD patients (15.91%, all heterozygous), as shown in Supplementary Table 4. Interestingly, only three STA samples (15.00%) contained identical heterozygous mutations, while one (5.00%) presented a homozygous mutation (Supplementary Table 5).

The pathogenic RNF213 p.R4810K mutation does not affect the RNF213 structure; it is a dominant mutation that deregulates E3 function and decreases overall ubiquitination levels.^16,19,20 Indeed, we isolated the STA proteins from one patient with intracranial aneurysm and one MMD patient with a heterozygous RNF213 mutation in exon 3. Protein quantification analysis showed significantly reduced RNF213 and ubiquitin levels (Fig. 1C).

In summary, this study is the first to report a novel CNV deletion in RNF213 in patients with MMD, suggesting that pathogenic RNF213 mutations might result in a loss-of-function rather than a gain-of-function in some MMD patients.

Pathogenic RNF213 mutation promotes intimal hyperplasia

MMD is a proliferative vasculopathy typically characterized by a hyperplastic intima and thin media.^4,7 Therefore, the possible relationship between a common RNF213 pathogenic mutation (p.R4810K) and MMD histopathology was explored in this study. Another set of 18 donor STA samples from MMD patients and 18 STA samples from patients with intracranial aneurysm were compared (Supplementary Table 6).

Haematoxylin and eosin staining was used for a morphometric analysis of the vascular wall structure. No significant differences in artery wall thickness were observed between MMD and intracranial aneurysm patients (251.15 ± 101.76 versus 292.11 ± 63.75 µm, P = 0.079). We analysed the ratio of tunica media to artery wall thickness in this study to exclude individual differences in artery size. A marked reduction in tunica media thickness was observed in the MMD samples compared with the intracranial aneurysm samples (Fig. 1F). The vascular smooth muscle fibres were then stained with an antibody against α-smooth muscle actin (α-SMA) using immunohistochemistry (IHC), and the average density of smooth muscle fibres per region of interest (ROI) was automatically analysed using Quantitative Pathology and Bioimage Analysis (QuPath) software (v0.3.0), which indicated a significant loss of smooth muscle in MMD patients compared with patients with intracranial aneurysm (Fig. 1D). However, subgroup analyses revealed that the RNF213 p.R4810K mutation was not associated with thinner smooth muscles in either MMD or intracranial aneurysm patients (Fig. 1F).

Weigert’s resorcin-fuchsin staining was used for the morphometric analysis of the internal elastic membranes. The ratio of tunica intima (mainly internal elastic membranes) to artery wall thickness was markedly increased in the MMD samples compared to the intracranial aneurysm samples (Fig. 1G). Interestingly, subgroup analyses revealed that MMD patients carrying the RNF213 mutation (p.R4810K) exhibited significantly thicker internal elastic membrane layers (Fig. 1G). The pathogenic RNF213 mutation promoted a homogeneous distribution of thicker internal elastic membranes, which was completely different from an eccentric increase due to atherosclerosis.²¹ In addition, an ROI-based quantitative analysis of the proliferation marker Ki-67 was performed in the tunica intima and revealed an increased number of Ki-67⁺ cells in MMD patients carrying the RNF213 p.R4810K mutation (Fig. 1E). Therefore, our findings suggest that the pathogenic RNF213 mutation induces proliferative changes in the tunica intima and is strongly associated with MMD histopathology.

RNF213 deletion promotes pathological angiogenesis in vivo

As a method to observe the vascular phenotype of pathological angiogenesis in vivo, we constructed both Rnf213-deficient mice (Rnf213^−/−) and rnf213-deficient zebrafish (rnf213^−/−) to mimic the loss of RNF213 function using the CRISPR–Cas9 gene-editing system. The models were confirmed by Sanger sequencing, semiquantitative analyses of mRNA and protein levels, and agarose gel electrophoresis (Fig. 2A). The homozygous mutant mice showed a 151-bp deletion mutation in intron 6 and exon 7 of Rnf213 at chromosome 11 that caused a 46-bp loss of the coding sequence. The homozygous mutant zebrafish presented with a small deletion of 7 bp in exon 2 of rnf213 at chromosome 3.

Figure 2.

RNF213 deletion promotes pathological angiogenesis in vivo. (A) Sanger sequencing revealed a 151-bp deletion in mice and a 7-bp deletion in zebrafish. Western blots of mouse brain tissues validated an ∼90% decrease in Rnf213 expression in Rnf213^−/− mice (n = 4, 1.00 ± 0.00 versus 0.07 ± 0.02, P = 0.0286). Quantitative RT-PCR confirmed the significantly reduced gene expression levels (Rnf213, n = 4, 1.00 ± 0.11 versus 0.13 ± 0.01, P = 0.0286; rnf213, n = 4, 0.99 ± 0.06 versus 0.17 ± 0.02, P = 0.0286). Agarose gel electrophoresis further confirmed this homozygous mutation in mice. (B) No significant differences in steno-occlusive arteries in the major branches of the Willis circle were observed between Rnf213^+/+ and Rnf213^−/− mice using 7.0 T magnetic resonance angiography. (C) The complete structure of the cerebral vessels was stained with haematoxylin and eosin (H&E) and subjected to immunohistochemistry. From each section, one similar region of the hippocampus and three similar cortical regions in the unilateral cerebral hemisphere were randomly selected. The Rnf213^−/− mice showed a significantly increased vessel density in both the cerebral cortex (n = 12, 868.77 ± 133.61 versus 1167.60 ± 234.25 cell/mm², P = 0.0011) and hippocampus (n = 6, 554.19 ± 63.98 versus 802.66 ± 88.41 cell/mm², P = 0.0022). Scale bar = 1 mm (left); scale bar = 100 µm (right). (D) Vessels throughout the retinas from Rnf213^+/+ and Rnf213^−/− mice were labelled with IB-4. AngioTool software was used to automatically detect the vessel length (red) and junctions (blue). Scale bar = 500 µm. (E) No significant differences in the relative vascularized area (n = 5, 1.00 ± 0.16 versus 1.20 ± 0.21, P = 0.0952), vessel area (n = 5, 1.00 ± 0.23 versus 1.24 ± 0.23, P = 0.2222), and vessel density (n = 5, 1.00 ± 0.15 versus 1.05 ± 0.19, P > 0.9999) were observed between Rnf213^+/+ and Rnf213^−/− mice. Compared to Rnf213^+/+ mice, Rnf213^−/− mice showed significantly increased total junction numbers (n = 5, 1.00 ± 0.29 versus 1.96 ± 0.54, P = 0.0079), junction density (n = 5, 1.00 ± 0.35 versus 1.62 ± 0.27, P = 0.0159), and branch numbers (n = 5, 1.00 ± 0.25 versus 1.67 ± 0.34, P = 0.0079). Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the region of interest. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001.

Previous studies did not detect stenosis or occlusion of the intracranial arteries in Rnf213^+/+ mice or Rnf213^−/− mice at the age of 16 to 64 weeks.^22,23 Although we extended the detection time to ∼80 weeks, no significant differences in steno-occlusive changes in the major branches of the circle of Willis were observed in either mouse model using 7.0 T magnetic resonance angiography (MRA) (Fig. 2B and Supplementary Videos 1 and 2). Furthermore, we assessed whether Rnf213^−/− mice exhibited other changes in pathological angiogenesis of the cerebral small vessels. The Human Protein Atlas (HPA) database showed that RNF213 protein concentrations were higher in the cerebral cortex and hippocampus than in other human tissues (Supplementary Fig. 2A). In addition, we found that Rnf213 was expressed mainly in the cerebral small vessels but not the neurons in these areas using immunofluorescence (IF) staining, as shown in Supplementary Fig. 2B. Haematoxylin and eosin and IHC staining were then performed to visualize the complete structure of the brain and vessels in the Rnf213^+/+ mice and the Rnf213^−/− mice. Using QuPath software to automatically calculate the regional density of Cd31⁺ cells, we observed a significantly increased vessel density in the cerebral cortices and hippocampi of the Rnf213^−/− mice compared with the Rnf213^+/+ mice (Fig. 2C). The relative increases in small arteries were then confirmed by performing IF staining (Supplementary Fig. 2C and D). In addition, we applied isolectin B4 (IB-4) to mark the whole retinal vessels and used AngioTool software²⁴ to automatically analyse and compare the vascular features of the Rnf213^+/+ mice and Rnf213^−/− mice (Fig. 2D). The mature vascularized area of Rnf213^−/− retinas was similar to that of Rnf213^+/+ retinas (Fig. 2E). Although no significant difference in vessel area or density was found between Rnf213^+/+ mice and Rnf213^−/− mice, Rnf213^−/− mice showed significantly greater total junction numbers, junction density and branching than Rnf213^+/+ mice (Fig. 2E). Moreover, coronal sections of the retina exhibited more vascular lumens near the macula in the Rnf213^−/− mouse retinas (Supplementary Fig. 2E). These findings indicated the occurrence of pathological angiogenesis in the cerebral cortex, hippocampus and retinas in Rnf213^−/− mice.

Thus, loss-of-function of RNF213 may not be associated with the stenosis or occlusion of large cerebral arteries but may be positively correlated with angiogenesis of small arteries and/or arterioles.

Human brain microvascular endothelial cells are the major source of RNF213

RNF213 mutations were positively associated with pathological angiogenesis,^9,22,25–27 and the outgrowths of small vessels consisted primarily of irregular perforating small arteries and arterioles. Thus, we sought to identify the major source of RNF213 among endothelial cells and other mural cells. Using donor STA serial sections from patients with intracranial aneurysm, we briefly compared the distribution of the RNF213 protein with that of other cell markers. The IHC analysis revealed that the RNF213 protein was expressed in the walls of all vessels but that RNF213 protein expression was higher in the tunica intima and media, as shown by the similar localization of the protein with endothelial and smooth muscle markers (Fig. 3A). Then, IF staining suggested that the RNF213 protein signal was significantly stronger in the tunica intima than in the tunica media, with a location similar to that of the endothelial marker CD31 (Fig. 3B).

Figure 3.

Human brain microvascular endothelial cells (HBMECs) are the major sources of RNF213. (A) Immunohistochemistry staining of donor superficial temporal artery sections from patients with intracranial aneurysm (IA). The RNF213 protein was expressed in all of the vascular walls. Endothelial markers [CD31, von Willebrand factor (vWF) and E-cadherin] were obviously located inside the vascular wall. Smooth muscle markers [α-SMA,calponin-1 (CNN1), smooth muscle myosin heavy chain 11 (MYH-11)] and pericyte markers (CD146 and α-SMA) were expressed mainly in the tunica media. Fibroblast markers (fibronectin and S100A4) were expressed mainly in the tunica adventitia. Scale bar = 20 µm. (B) The RNF213 protein co-localized with CD31 in the vascular wall. Scale bar = 50 µm. (C) The RNF213 protein was expressed at higher levels in human brain microvascular endothelial cells (HBMECs) and human brain vascular pericytes (HBVPs) than in human brain vascular adventitial fibroblasts (HBVAFs) or human brain vascular smooth muscle cells (HBVSMCs). Scale bar = 100 µm. (D) Using human umbilical vein endothelial cells (HUVECs) as controls, western blots revealed significantly higher RNF213 protein expression levels in HBMECs than in other cells (n = 4, 1.28 ± 0.03 versus 0.52 ± 0.03 versus 0.36 ± 0.01 versus 0.89 ± 0.02, P < 0.0001, Kruskal–Wallis statistic =14.12), which was confirmed by performing a qRT-PCR analysis (n = 3, 1.47 ± 0.03 versus 0.67 ± 0.02 versus 0.56 ± 0.02 versus 0.84 ± 0.02, P < 0.0001, Kruskal–Wallis statistic =10.38). (E) RNF213 is mainly expressed in the cytoplasm of HBMECs, as shown by fluorescent in situ hybridization. Scale bar = 10 µm. (F) Confocal microscopy revealed stronger RNF213 protein expression in the cytoplasm than in the nucleus. Scale bar = 20 µm. Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the region of interest. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001.

Using single-cell RNA sequencing (scRNA-seq) data from 25 human tissues and blood cells in the HPA, we found that RNF213 mRNA expression in endothelial cells was higher than that in muscle and mesenchymal cells (Supplementary Fig. 3A). Although no endothelial or mural cell clusters were detected in the brain tissues using scRNA-seq, IHC revealed stronger RNF213 protein expression in endothelial cells than in glial or neuronal cells in the normal cortical sections (Supplementary Fig. 3B and C). Additionally, specific analyses of single cell types showed high expression of RNF213 mRNA in the endothelial cell cluster from heart and eye tissues (Supplementary Fig. 3D and E), indicating that RNF213 is expressed at higher levels in endothelial cells than in other vascular cells.

Furthermore, we compared RNF213 protein and mRNA expression levels in different cellular components of primary cultured cerebral vascular cells. Although the RNF213 protein was commonly expressed in all cells, an increased fluorescence intensity was observed in human brain microvascular endothelial cells (HBMECs) (Fig. 3C). Additionally, immunoblot and qRT-PCR analyses confirmed that RNF213 expression levels were higher in HBMECs than in other mural cells (Fig. 3D), indicating that HBMECs are the major sources of RNF213.

In addition, we explored the intracellular location of RNF213 in HBMECs. The HPA database revealed that the RNF213 protein was a cytosol-localized protein in A431 cells (Supplementary Fig. 3F). Although the RNF213 protein was detected in both the cytoplasm and nucleus, fluorescence in situ hybridization (FISH) with an mRNA probe and IF staining confirmed that RNF213 was expressed mainly in the cytoplasm of HBMECs (Fig. 3E and F). In addition, neither signal peptide cleavage sites nor transmembrane helices were predicted to exist in the RNF213 protein using SignalP-5.0 and TMHMM-2.0 software, suggesting that it is not a classical secreted protein (Supplementary Fig. 3G and H). SecretomeP-2.0 software was used to predict secreted proteins lacking a conventional signal peptide that travel through the non-classical pathway and confirmed that the RNF213 protein is not a non-classical secreted protein (Supplementary Table 7). Thus, our findings indicated that RNF213 is a functional non-secreted cytosolic protein that is expressed primarily in cerebral endothelial cells.

Endothelial RNF213 knockdown promotes cell proliferation, migration and angiogenesis

RNF213 is an anti-angiogenic protein that is expressed mainly in HBMECs; thus, the potential regulatory effects of loss-of-function of RNF213 on cell proliferation, migration and tube formation were explored in this study. Small interfering RNA (siRNA) and short hairpin RNA (shRNA) sequences were designed to mimic RNF213 loss-of-function conditions in a portion of MMD patients. The efficiency of specific siRNF213 sequences was validated by western blotting and qRT-PCR (Fig. 4A).

Figure 4.

Endothelial RNF213 knockdown promotes HBMEC proliferation, migration and angiogenesis. (A) Two designed siRNA/shRNA sequences were used to mimic RNF213 loss-of-function conditions, which were confirmed by western blot (siNC versus siRNF213 #1: n = 4, 1.00 ± 0.00 versus 0.43 ± 0.05, P = 0.0286; siNC versus siRNF213 #2: n = 4, 1.00 ± 0.00 versus 0.42 ± 0.03, P = 0.0286) and qRT-PCR analyses (siNC versus siRNF213 #1: n = 4, 1.00 ± 0.03 versus 0.45 ± 0.04, P = 0.0286; siNC versus siRNF213 #2: n = 4, 1.00 ± 0.03 versus 0.45 ± 0.02, P = 0.0286). (B) The siRNF213-treated human brain microvascular endothelial cells (HBMECs) showed greater cell viability than the siNC-treated HBMECs at 48 h (n = 6, 0.47 ± 0.03 versus 0.52 ± 0.05, P = 0.0931) and also showed significantly greater cell viability at 72 h (n = 6, 0.93 ± 0.07 versus 1.09 ± 0.06, P = 0.004) based on the results of CCK-8 assays. (C) A BrdU incorporation assay was conducted to measure the proliferation of siNC- and siRNF213-treated HBMECs. The siRNF213-treated HBMECs showed a significantly greater BrdU incorporation ratio than the siNC-treated HBMECs (n = 5, 2.09 ± 0.47 versus 9.84 ± 1.89%, P = 0.0079). Scale bar = 20 µm. (D) Flow cytometry was used to assess the proliferation of siNC- and siRNF213-treated HBMECs after labelling with EdU and Hoechst 33342. The siRNF213-treated HBMECs showed a significantly greater cell proliferation ratio than the siNC-treated HBMECs (n = 4, 69.75 ± 6.07 versus 93.15 ± 0.62%, P = 0.0286). (E) Transwell assays were performed to measure vertical HBMEC migration. siRNF213-treated HBMECs showed greater vertical migration than siNC-treated HBMECs (n = 6, 86.25 ± 38.92 versus 124.75 ± 57.69, P = 0.0152). Scale bar = 100 µm. (F) A wound healing assay was used to measure horizontal HBMEC migration. The siRNF213-treated HBMECs showed markedly greater horizontal migration than the siNC-treated HBMECs at 24 h (n = 5, 16.16 ± 6.44 versus 55.45 ± 9.20, P = 0.0079) and 48 h (n = 5, 49.76 ± 5.26 versus 73.36 ± 14.78, P = 0.0079). Scale bar = 100 µm. (G) A tube formation assay was performed to measure HBMEC angiogenesis. AngioTool software was used to automatically detect the vessel length (red) and junctions (green). Scale bar = 100 µm. (H) The siRNF213-treated HBMECs showed significantly greater vessel area (n = 5, 1.00 ± 0.16 versus 1.59 ± 0.36, P = 0.0079), vessel density (n = 5, 1.00 ± 0.17 versus 1.58 ± 0.36, P = 0.0079), vessel length (n = 5, 1.00 ± 0.16 versus 1.63 ± 0.42, P = 0.0079), total number of junctions (n = 5, 1.00 ± 0.22 versus 2.56 ± 1.13, P = 0.0079), junction density (n = 5, 1.00 ± 0.23 versus 2.54 ± 1.12, P = 0.0079), and branching index values (n = 5, 1.00 ± 0.18 versus 1.22 ± 0.09, P = 0.0317) than the siNC-treated HBMECs. Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the region of interest. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001.

The proliferation of HBMECs was measured by using a CCK-8 assay, a BrdU incorporation assay, and flow cytometry using EdU and Hoechst 33342 staining. Using the CCK-8 assay, we observed greater viability of siRNF213-treated HBMECs than siNC-treated HBMECs at 48 h and a significant increase in cell viability at 72 h (Fig. 4B). Cell proliferation was measured by competitive binding with BrdU/EdU or thymine during DNA replication. We randomly selected five locations to calculate the average numbers of DAPI⁺ cells and DAPI⁺/BrdU⁺ cells. The ratio of DAPI⁺/BrdU⁺ cells was significantly higher in siRNF213-treated HBMECs than in siNC-treated HBMECs (Fig. 4C). EdU and Hoechst 33342 were used to label proliferating cells and live cells, respectively. Flow cytometry analysis was applied to identify the ratio of EdU⁺/Hoechst 33342⁺ cells in each group, and increased HBMEC proliferation was observed in the siRNF213-treated group (Fig. 4D).

Cell migration was evaluated using Transwell and wound healing assays. A membrane with a pore size of 8 µm was used to measure vertical HBMEC migration upon treatment with siNC and siRNF213. The number of HBMECs that migrated to the lower surface of the membrane was significantly greater in the siRNF213-treated group than in the siNC-treated group (Fig. 4E). Then, we constructed stable RNF213-knockdown HBMECs and inserted the GFP gene using lentiviral transfection. A two-well silicone insert was used to form a defined cell-free gap of 500 µm at 0 h, and HBMECs were continuously cultured in serum-free medium for another 48 h. The results showed a significantly decreased gap area and reduced gap depth at 24 h compared to those in control HBMECs, and the difference was more pronounced at 48 h (Fig. 4F), suggesting that horizontal HBMEC migration was increased.

We also explored the effect of RNF213 loss on angiogenesis using tube formation assays, and we detected increases in the vascular area, density, length, number of vascular junctions and number of branches in siRNF213-treated HBMECs compared with siNC-treated HBMECs, as shown in Fig. 4G and H. In summary, our findings suggest that RNF213 knockdown promotes HBMEC proliferation, migration and angiogenesis in vitro and thus that RNF213 plays an important role in MMD pathogenesis.

The Hippo pathway is associated with RNF213-related MMD pathogenesis

In this study, RNA-seq was performed to determine the potential mechanism of RNF213 mutations in MMD by using HBMECs and donor STA samples. Data were analysed and processed using the protocol outlined in Supplementary Fig. 4A. After gene annotation and quality control (Supplementary Fig. 4B and Supplementary Table 8), genes with a count < 2% and/or a degree of variation < 5% were filtered, and a differential expression analysis was performed with the edgeR package.

First, a total of 2633 differentially expressed genes (DEGs) between siNC-treated and siRNF213-treated HBMECs were identified using the Network Analyst platform (v3.0),²⁸ as shown in Fig. 5A. The top 100 significantly upregulated and downregulated DEGs were selected based on adjusted P-values (Supplementary Fig. 4C and Supplementary Table 9). The Database for Annotation, Visualization and Integrated Discovery (DAVID) bioinformatics resource (v6.8)²⁹ was then used to discover the biological functions in which each of the genes was enriched. We obtained 136 significantly enriched Gene Ontology (GO)³⁰ terms from the top 100 upregulated DEGs, among which the biological processes were positively associated with the development of the vasculature (Fig. 5B and Supplementary Table 10). We identified 45 significantly enriched GO terms for the top 100 downregulated DEGs, among which the biological processes were negatively related to cell adhesion and endothelial cell proliferation (Fig. 5B and Supplementary Table 10). Then, 78 important signalling pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database were screened for all DEGs³¹; these pathways were associated primarily with transcription and translation, biological metabolism, the cell cycle and proliferation (Fig. 5C and Supplementary Tables 11 and 12). In addition, we identified 584 DEGs between STAs from intracranial aneurysm patients and STAs from MMD patients carrying the RNF213 mutation (p.R4810K) using the same approach (Fig. 5D), and the top 100 upregulated and downregulated DEGs were used for functional annotations (Supplementary Fig. 4C and Supplementary Table 13). Bioinformatics analyses revealed 185 positively enriched GO terms, 60 negatively enriched GO terms and 56 KEGG pathways associated with MMD pathogenesis, as shown in Fig. 5E and F (Supplementary Tables 14–16). Among these pathways, 29 signalling pathways enriched in the siRNF213-treated HBMECs and MMD arteries were screened (Supplementary Fig. 4D), mainly cell proliferation and junction-associated pathways.

Figure 5.

The Hippo signalling pathway might be involved in RNF213-related MMD pathogenesis, as shown through bioinformatics analyses. (A) Volcano plot showing 2633 differentially expressed genes (DEGs) between siNC- and siRNF213-treated human brain microvascular endothelial cell (HBMECs), including 1431 upregulated (red) and 1202 downregulated (blue) DEGs. (B) GO analysis indicated functions that were positively and negatively regulated by the top 100 RNF213-related DEGs. (C) Functional enrichment analysis of pathways significantly enriched in the RNF213-related DEGs. (D) Volcano plot showing 584 DEGs between intracranial aneurysm (IA) donor superficial temporal arteries and moyamoya disease (MMD) donor superficial temporal arteries, including 301 upregulated (red) and 283 downregulated (blue) DEGs. (E) GO analysis of functions that were positively and negatively regulated by the top 100 MMD-related DEGs. (F) Functional enrichment analysis of pathways significantly enriched in MMD-related DEGs.

According to previous research, the Hippo signalling pathway might play key roles in vascular remodelling and barrier formation.³² The Hippo signalling pathway was suppressed in our study via reduced gene expression, and its suppression was positively associated with the proliferation of vascular disease. We similarly observed impaired adherens junctions and tight junctions in MMD patients with the RNF213 p.R4810K mutation (Supplementary Fig. 4E); this impairment might inhibit Hippo signalling. Although researchers have not clearly determined whether Hippo signalling is the major pathway involved in MMD pathogenesis, our findings indicate that the pathogenic RNF213 mutation may participate in regulating the Hippo signalling pathway.

Endothelial RNF213 knockdown activates the Hippo effector YAP/TAZ

We measured the levels of key molecules in RNF213-knockdown HBMECs to explore the effect of RNF213 mutations on the Hippo pathway. The transcriptional co-activators Yes-associated protein 1 (YAP) and tafazzin (TAZ) are master regulators of the Hippo signalling pathway.³³ In brief, MST1/2 phosphorylate and activate LATS1/2, which then phosphorylate YAP/TAZ, leading to the cytoplasmic sequestration, degradation and inactivation of YAP/TAZ. However, YAP/TAZ shuttle from the cytosol to the nucleus, where they mainly interact with TEAD transcription factors to regulate the expression of a large group of target genes primarily involved in cell growth, survival and migration.

The key upstream molecules of MST1/2 and LATS1/2 were explored in siRNF213-treated HBMECs compared with siNC-treated HBMECs using western blot analyses (Supplementary Fig. 5A). Quantitative RT-PCR analyses revealed increased TAZ and YAP mRNA expression levels between the groups (Supplementary Fig. 5B). The RNA polymerase inhibitor actinomycin D (ActD) was then applied to determine whether this finding was due to transcriptional activation or increased mRNA stability. After incubation with ActD (500 µg/ml) for 30 min, qRT-PCR analyses confirmed elevated transcription of TAZ and YAP in siRNF213-treated HBMECs (Supplementary Fig. 5B). In addition, TAZ mRNA expression levels were significantly higher in siRNF213-treated HBMECs incubated with ActD for an extended period of 0–6 h than in control HBMECs (Supplementary Fig. 5C).

Furthermore, we measured the levels of active (unphosphorylated) and inactive (phosphorylated) YAP/TAZ proteins. Significantly reduced p-TAZ/TAZ ratios and p-YAP/YAP ratios were detected at the total protein level in siRNF213-treated HBMECs (Fig. 6A). The intracellular distribution of active YAP/TAZ was then explored in this study. Although cytoplasmic YAP/TAZ protein levels were slightly reduced, the nuclear YAP/TAZ level was significantly increased (Fig. 6B and C), and its nuclear translocation ratio was elevated in siRNF213-treated HBMECs (Fig. 6D and E). Moreover, a stronger YAP/TAZ fluorescence intensity was detected in donor STA sections from MMD patients with RNF213 mutations than in those from patients with intracranial aneurysm (Supplementary Fig. 5D). Co-immunoprecipitation (Co-IP) experiments confirmed a direct protein–protein interaction between RNF213 and YAP/TAZ (Supplementary Fig. 5A). Thus, RNF213 knockdown promotes YAP/TAZ activation.

Figure 6.

Endothelial RNF213 knockdown activates the Hippo effectors TAZ and YAP. (A) Quantitative analysis of p-TAZ/TAZ and p-YAP/YAP protein levels in siNC- and siRNF213-treated HBMECs (p-TAZ/TAZ, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.61 ± 0.12, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.56 ± 0.09, P = 0.0286; p-YAP/YAP, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.55 ± 0.03, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.43 ± 0.05, P = 0.0286). (B) Quantitative analysis of nuclear (N) and cytoplasmic (C) TAZ protein levels in siNC- and siRNF213-treated HBMECs at the nuclear and cytoplasmic levels (nuclear TAZ, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 1.39 ± 0.09, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 1.49 ± 0.14, P = 0.0286; cytoplasmic TAZ, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.67 ± 0.06, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.66 ± 0.04, P = 0.0286). (C) Quantitative analysis of nuclear and cytoplasmic YAP protein levels in siNC- and siRNF213-treated HBMECs at the nuclear and cytoplasmic levels (nuclear YAP, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 1.31 ± 0.16, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 1.45 ± 0.13, P = 0.0286; cytoplasmic YAP, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.65 ± 0.04, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.82 ± 0.07, P = 0.0286). (D) Analysis of the difference in the distribution of TAZ protein expression between siNC- and siRNF213-treated HBMECs (N > C, n = 6, 28.30 ± 6.42 versus 39.16 ± 4.33%, P = 0.0152; N = C, n = 6, 42.83 ± 10.61 versus 44.55 ± 9.51%, P > 0.9999; N < C, n = 6, 28.87 ± 12.25 versus 16.29 ± 8.89%, P = 0.0931). Scale bar = 20 µm. (E) Analysis of the difference in the distribution of YAP protein expression between siNC- and siRNF213-treated HBMECs (N > C, n = 6, 55.42 ± 8.72 versus 73.75 ± 11.09%, P = 0.0152; N = C, n = 6, 35.34 ± 8.67 versus 17.87 ± 8.22%, P = 0.0087; N < C, n = 6, 9.25 ± 6.78 versus 8.39 ± 4.71%, P = 0.7879). Scale bar = 20 µm. Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the region of interest. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Endothelial RNF213 knockdown promotes VEGFR2 expression via YAP/TAZ-TEAD

We measured the intranuclear levels of transcription factors in the Hippo pathway, and qRT-PCR analyses revealed prominently elevated TEAD1-3 mRNA expression levels and reduced TEAD4 mRNA expression levels in siRNF213-treated HBMECs (Supplementary Fig. 5E), which indicated that Hippo signalling was switched to an ‘off’ state through the upregulation of YAP/TAZ and TEADs. Interactive YAP/TAZ-TEAD binding activates classic regulators of cell growth and organogenesis, including Axin2, cycD, Sox2, Nkd1, myc, Slug, Birc2/5, CTGF and Gh2.³³ Among these well studied molecules, VEGFR2, an important gene regulating vascular development via the VEGF-VEGFR2 axis, was recently identified as a downstream target molecule of YAP/TAZ-TEAD.³⁴ Therefore, we tested whether RNF213 knockdown activated VEGFR2 expression to promote pathological angiogenesis through interactive YAP/TAZ-TEAD binding.

Although the ratio of VEGFR2 phosphorylated at different sites (p-VEGFR2) to VEGFR2 varied in siRNF213-treated HBMECs (Fig. 7A), flow cytometry analysis showed a marked increase in VEGFR2 signals in siRNF213-treated HBMECs compared with siNC-treated HBMECs (Supplementary Fig. 5F). Significantly increased membrane protein expression and decreased intracellular protein expression of VEGFR2 were detected in siRNF213-treated HBMECs (Fig. 7B), and qRT-PCR confirmed the increase in VEGFR2 mRNA expression (Supplementary Fig. 5B). In addition, post-translational modifications (PTMs) of the VEGFR2 protein depend on cytoplasmic organelles, and mature proteins are transferred from the Golgi apparatus to the cell surface. We detected the co-localization of VEGFR2 and the Golgi apparatus marker TGN46 and found that RNF213 knockdown increased the distribution of VEGFR2 in the Golgi apparatus, which may suggest an increase in PTMs (Fig. 7C). Our data indicate that loss of endothelial RNF213 promotes overexpression of VEGFR2 through ‘off-state’ Hippo signalling.

Figure 7.

Inhibition of YAP/TAZ expression reverses RNF213-induced angiogenesis. (A) Quantitative analysis of the p-VEGFR2/VEGFR2 protein ratio in siNC- and siRNF213-treated HBMECs (Tyr.1175, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.39 ± 0.03, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.57 ± 0.02, P = 0.0286; Tyr.1212, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 1.69 ± 0.26, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 1.70 ± 0.37, P = 0.0286). (B) Quantitative analysis of surface and intracellular VEGFR2 protein levels between siNC- and siRNF213-treated HBMECs (surface VEGFR2, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 1.28 ± 0.29, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 1.49 ± 0.50, P = 0.0286; intracellular VEGFR2, n = 4, siNC versus siRNF213 #1 = 1.00 ± 0.00 versus 0.59 ± 0.06, P = 0.0286; siNC versus siRNF213 #2 = 1.00 ± 0.00 versus 0.44 ± 0.04, P = 0.0286). (C) Analysis of the VEGFR2 protein distribution between siNC- and siRNF213-treated HBMECs. Scale bar = 20 µm. (D) Quantitative analysis of nuclear TAZ, nuclear YAP, and surface VEGFR2 protein levels in siNC-, siRNF213−, siNC+YAP/TAZ inhibitor-1−, and siRNF213+YAP/TAZ inhibitor-1-treated HBMECs (nuclear TAZ, n = 4, siNC versus siRNF213 = 1.00 ± 0.00 versus 1.11 ± 0.11, P = 0.0286; siNC+YAP/TAZ inhibitor-1 versus siRNF213+YAP/TAZ inhibitor-1 = 0.43 ± 0.05 versus 0.56 ± 0.06, P = 0.0286; nuclear YAP, n = 4, siNC versus siRNF213 = 1.00 ± 0.00 versus 1.46 ± 0.26, P = 0.0286; siNC+YAP/TAZ inhibitor-1 versus siRNF213+YAP/TAZ inhibitor-1 = 0.36 ± 0.03 versus 0.50 ± 0.04, P = 0.0286; surface VEGFR2, n = 4, siNC versus siRNF213 = 1.00 ± 0.00 versus 1.28 ± 0.17, P = 0.0286; siNC+YAP/TAZ inhibitor-1 versus siRNF213+YAP/TAZ inhibitor-1 = 0.36 ± 0.10 versus 0.56 ± 0.07, P = 0.0286). (E) Quantitative analysis of TAZ mRNA levels in normal HBMECs incubated with VEGF₁₆₅ (5 ng/ml) at 0, 1, 2, 3, 4.5 and 6 h (n = 3, 1.00 ± 0.04 versus 1.24 ± 0.10 versus 1.42 ± 0.07 versus 0.92 ± 0.07 versus 0.68 ± 0.02 versus 0.72 ± 0.14, P = 0.0086, Kruskal–Wallis statistic =15.46). (F) A tube formation assay was conducted to measure angiogenesis in siRNF213-treated HBMECs after incubation with a YAP/TAZ inhibitor and/or VEGF₁₆₅. Scale bar = 100 µm. (G) AngioTool software was used to automatically detect the vessel length and junctions. The siRNF213-transfected HBMECs treated with a YAP/TAZ inhibitor showed significant reductions in the vessel area (n = 6, 1.00 ± 0.07 versus 1.69 ± 0.21 versus 0.59 ± 0.08 versus 0.89 ± 0.03, P = 0.0005, Kruskal–Wallis statistic =17.58), vessel density (n = 6, 1.00 ± 0.02 versus 1.66 ± 0.17 versus 0.60 ± 0.07 versus 0.86 ± 0.03, P = 0.0005, Kruskal–Wallis statistic =17.86), vessel length (n = 6, 1.00 ± 0.41 versus 2.57 ± 0.76 versus 0.36 ± 0.20 versus 0.59 ± 0.34, P = 0.0044, Kruskal–Wallis statistic =13.09), total number of junctions (n = 6, 1.00 ± 0.15 versus 2.11 ± 0.40 versus 0.47 ± 0.11 versus 0.71 ± 0.15, P = 0.0007, Kruskal–Wallis statistic =16.91), junction density (n = 6, 1.00 ± 0.12 versus 2.08 ± 0.36 versus 0.48 ± 0.13 versus 0.69 ± 0.14, P = 0.0009, Kruskal–Wallis statistic =16.51), and branching index (n = 6, 1.00 ± 0.10 versus 1.73 ± 0.24 versus 0.54 ± 0.07 versus 0.77 ± 0.08, P = 0.0005, Kruskal–Wallis statistic =17.86) compared to those of siRNF213-transfected HBMECs treated with VEGF₁₆₅. Each dot represents one sample. The means ± SDs are shown. The dashed line indicates the region of interest. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Inhibition of YAP/TAZ expression reverses RNF213 knockdown-induced angiogenesis

YAP/TAZ expression was inhibited via RNAi or incubation with YAP/TAZ inhibitor-1 (50 nmol/ml) for 24 h to further explore the major role of YAP/TAZ in RNF213 knockdown-induced angiogenesis. The decrease in YAP/TAZ protein expression was consistent with an obvious reduction in VEGFR2 protein expression, and the inhibition of YAP/TAZ expression in inhibitor-treated HBMECs alleviated VEGFR2 overexpression at the protein level (Fig. 7D). This finding was subsequently confirmed by qRT-PCR (Supplementary Fig. 5G). A dual-luciferase reporter assay previously showed that YAP/TAZ induces the transcription of VEGFR2.³⁴ In addition, we found that inhibition of YAP/TAZ reduced the YAP/TAZ distribution in the nucleus and increased the VEGFR2 distribution in the Golgi apparatus (Supplementary Figs 5H and 7C).

VEGF is generally accepted as a proangiogenic factor that regulates vascular remodelling by binding cell-surface VEGFR2. Thus, a small dose of VEGF₁₆₅ (5 ng/ml) was applied for 0–6 h to explore the time-dependent effect of VEGF on TAZ expression in HBMECs. Quantitative RT-PCR analysis revealed that TAZ mRNA expression peaked at 2 h (Fig. 7E). Next, a tube formation assay was employed using siRNF213-treated HBMECs incubated with a YAP/TAZ inhibitor and/or VEGF₁₆₅ (Fig. 7F). Loss of endothelial TAZ restricted tube formation, while stimulation of TAZ expression enhanced vasculogenesis (Fig. 7G). These findings suggest that endothelial YAP/TAZ expression is positively associated with tube formation and that YAP/TAZ inhibition reverses RNF213 knockdown-induced angiogenesis.

External validation of Hippo pathway dysregulation using RNF213-deficient animals

Vascular endothelial cells were isolated from RNF213-deficient animals and used to measure key molecules in the Hippo pathway for further verification. Since the vascular endothelial cells of the Tg(flk:EGFR; gata1:DsRed) transgenic line exhibited systemic green autofluorescence but those of normal zebrafish (AB line) did not, fluorescence-activated cell sorting (FACS) was used to separate the endothelial cells from single-cell suspensions derived from rnf213^+/+ and rnf213^−/− zebrafish at 120 h post-fertilization (hpf) (Supplementary Fig. 6A). An insufficient number of antibodies is an acknowledged obstacle in zebrafish research; thus, we measured the mRNA expression levels of taz, yap and vegfr2. Endothelial cells from the rnf213^−/− zebrafish exhibited significantly higher taz, yap and vegfr2 levels than those from the rnf213^+/+ zebrafish (Supplementary Fig. 6B).

Furthermore, we extracted vascular endothelial cells from the brains of Rnf213^+/+ and Rnf213^−/− mice at 64–80 weeks of age (Supplementary Fig. 6C). Endothelial deletion of Rnf213 not only promoted Taz, Yap and Vegfr2 mRNA expression (Supplementary Fig. 6B) but also significantly increased nuclear Taz and Yap expression and surface Vegfr2 expression (Supplementary Fig. 6D). In addition, endothelial deletion of Rnf213 increased the level of nuclear YAP/TAZ expression, and a marked increase in cell-surface Vegfr2 expression was observed in endothelial cells from Rnf213^−/− mice (Supplementary Fig. 6E). These findings support the hypothesis that loss-of-function of RNF213 increases VEGFR2 expression by activating the Hippo effector YAP/TAZ.

Discussion

Although missense SNPs in the RNF213 gene have been studied for more than 10 years using peripheral blood or saliva samples from MMD patients, our data indicated that STA samples from some patients with sporadic MMD contained a novel CNV deletion that markedly damaged RNF213 protein structure and function, as shown through WGS analysis. Homozygosity of MMD-specific pathogenic mutations was present in the arteries, and heterozygosity was present in the blood samples. These differences suggest that environmental factors might exacerbate RNF213 mutation in MMD vascular lesions. However, MMD lesions should be located primarily in steno-occlusive areas and moyamoya vessels. Due to ethical and legal restrictions, we could not obtain these samples during surgical revascularization. Donor STAs were the best alternative tissue samples, but the results from such samples may not completely reflect the precise pathogenesis of MMD. Nevertheless, loss-of-function RNF213 mutations in MMD were detectable.

Our findings from donor artery sections confirmed that thickening of the tunica intima and thinning of the tunica media are two main pathological characteristics of MMD in patients. However, the pathogenic RNF213 p.R4810K mutation was positively associated with concentric intimal hyperplasia only, and changes in the number of smooth muscle fibres were not explained by these mutations. Although quantitative analyses indicated that HBMECs were the major source of RNF213, higher expression was detected in smooth muscle cells (SMCs) within arterial and heart tissues than in fibroblasts. The limited research on the effect of RNF213 mutation on smooth muscle remains controversial. In vitro studies using induced pluripotent stem cell-derived SMCs have demonstrated that RNF213 mutation does not damage the cellular proliferation, migration, and contraction abilities of SMCs in MMD patients.³⁵ A co-culture study has confirmed that smooth muscle progenitor cells are recruited and regulated by endothelial colony-forming cells in MMD patients.³⁶ In contrast, morpholino-based rnf213 knockdown of zebrafish results in marked decreases in myofibrils and motor neurons.³⁷ Since mutations in the smooth muscle marker ACTA2 have been reported in some family numbers of MMD patients and as phenotypic conversion of SMCs is regulated by circZXDC, which is involved in MMD pathogenesis,^38,39 we did not explore the effect of RNF213 on smooth muscles. Because MMD is a chronic and progressive disease, a minimum change in SMCs from RNF213 mutation could be obtained in a long-term observation.

In addition, RNF213 is expressed at high levels in immune cells (the highest levels were detected in peripheral NK and T cells) and plays a key role in resisting bacterial infection via the ubiquitylation of lipopolysaccharide.⁴⁰ Recent studies have revealed that dysregulated RNF213 function is associated with autoimmune diseases.^41,42 Nevertheless, our study did not explore the potential immunomodulation of RNF213 in angiogenesis. Such research might provide new insights into MMD pathogenesis in future studies.

Some of the most important features of MMD are steno-occlusive changes in the distal portions of the internal carotid arteries and/or the proximal portion of the circle of Willis.¹ Similar to the case in previous studies,^22,23 MRA did not reveal significant differences in the major arteries of the head and neck in Rnf213^−/− mice. However, another feature of MMD is abnormally proliferative vasculopathy consisting of artery and arteriole proliferation.¹ We observed increased vessel size, number and density in the cerebral cortex, hippocampus and retinas of Rnf213^−/− mice. Using rnf213^−/− zebrafish, we have previously observed similar blood vessel sprouting in the intersegmental arteries at 72 hpf and detected a possible mulberry-like cluster of disordered-flow vascular channels at 120 hpf.^25,26 Our study showed that loss-of-function of RNF213 may not lead to stenosis and/or occlusion of the intracranial large arteries but may promote pathological angiogenesis in the small arteries. Previous studies have shown that these abnormally proliferative small arteries contain an incomplete vascular structure and that the development of moyamoya vessels occurs ahead of regional stenosis and/or occlusion.^43,44 A clinical follow-up study with MRA has also revealed that the reappearance of moyamoya vessels is independent of previously steno-occlusive arteries in MMD patients who are undergoing surgical revascularization.⁴⁵ Although moyamoya vessels are viewed as evidence of a physiological compensatory mechanism to supplement regional blood flow, our findings suggest that the proliferative vasculopathy caused by RNF213 mutations in MMD is a typical feature.

Endothelial knockout of the RNF213 gene has been found to damage cell–cell junctions and promote endothelial cell migration by activating inflammatory pathways in a cell culture model in vitro.⁴⁶ Overexpression of RNF213 significantly inhibits tube formation by suppressing the PI3K-Akt signalling pathway.^47,48 Furthermore, our study confirmed that HBMECs were the major sources of non-secreted RNF213 protein, and knockdown of RNF213 expression significantly increased endothelial cell proliferation, migration and tube formation in vitro. However, these abilities were markedly enhanced early in the experimental treatment period, and the effect of RNF213 knockdown gradually weakened over time. Indeed, the long-term cumulative effects of RNF213 mutations should be explored in the future.

We next sought to obtain a more comprehensive mechanistic understanding of the relationship between loss-of-function of RNF213 and moyamoya vessels. Our data showed that endothelial RNF213 knockdown activated the Hippo pathway effector YAP/TAZ and induced the expression of the downstream effector VEGFR2 (Fig. 8). Inhibition of YAP/TAZ resulted in an altered cellular distribution of VEGFR2 due to defects in trafficking from the Golgi apparatus to the plasma membrane and reversed RNF213 knockdown-induced angiogenesis.

Figure 8.

Schematic showing the mechanism by which RNF213 loss-of-function mutations regulate the Hippo signalling pathway in endothelial cells. Green = IB-4; red = α-SMA.

Furthermore, the VEGF-VEGFR2 axis induced vascular remodelling by regulating the SCR kinase family and cytoskeleton, but its effects were constrained by YAP/TAZ protein activity.^32,34 Upregulated YAP/TAZ expression promoted VEGFR2 protein expression, and the VEGF-VEGFR2 axis suppressed MST1/2 and LAST1/2 through a feedback loop.⁴⁹ Our data also revealed that YAP/TAZ levels were positively correlated with angiogenesis. Compared with YAP/TAZ inhibitor treatment, YAP/TAZ knockdown in HBMECs only slightly reduced VEGFR2 expression. In this study, we tried to rectify the reduction in RNF213 expression through plasmid DNA transfection. Supplementation with a small amount of RNF213 led to a decrease in YAP/TAZ mRNA expression, but overexpression of RNF213 by more than 12-fold led to a significant increase in YAP/TAZ expression, indicating that strongly elevated RNF213 expression led to dysfunction of this negative feedback loop. Moreover, Hippo signalling reciprocally promoted Notch signalling while antagonizing NF-κB and Wnt/β-catenin signalling.^50–52 Thus, RNF213 knockdown-induced MMD pathogenesis involves pathways in addition to the Hippo signalling pathway, and more molecular networks should be explored in the future.

This study had a few limitations. First, we identified a novel RNF213 mutation in several patients with sporadic MMD, but we did not test their parents. The heterozygous mutation cannot be concluded to be inherited. Because the evidence supporting it as a loss-of-function mutation is not sufficiently strong (lack of validation in a large sample), all experiments using human samples were conducted on STA samples from MMD patients carrying the pathogenic p.R4810K mutation. In addition, we observed an increased number of endothelial cells from RNF213-deficient animals by labelling vessel markers, but the effects of veins could not be excluded. Moreover, p-VEGFR2 is the active component and starting factor initiating downstream signalling pathways in angiogenesis. However, the level of the VEGFR2 protein phosphorylated at different sites varied substantially in siRNF213-treated HBMECs. Owing to a cross-reaction with VEGFR1 and VEGFR3 (which have protein weights similar to that of the p-VEGFR2 antibody), changes in these confounding proteins might have influenced the actual p-VEGFR2 concentration. Because the targeted tyrosine sites of VEGFR2 were not fully discovered, variations in the phosphorylation of several sites were not sufficiently accurate to confirm VEGFR2 activation. Although our study identified a significant increase in membrane VEGFR2 levels in endothelial cells, this finding must be validated in the future. Finally, the increased mRNA expression of YAP/TAZ in siRNF213-treated HBMECs could not be explained by low ubiquitination activity, but YAP/TAZ protein stability is controlled by an appropriate balance between E3 ubiquitin ligases and deubiquitinases. Since loss-of-function of RNF213 disrupted the K63-linked polyubiquitin chains,⁵³ we hypothesized that endothelial RNF213 knockdown could stabilize and activate YAP/TAZ via hypo-ubiquitination to a certain extent; this hypothesis should be tested in further studies.

Taken together, our results reveal that RNF213 knockdown mediates VEGFR2 overexpression by activating the Hippo pathway effector YAP/TAZ during pathological angiogenesis. Inhibition of YAP/TAZ expression might represent a therapeutic target for MMD patients with RNF213 loss-of-function.

Data availability

The authors declare that all supporting data are available within the article, the Supplementary material, and the National Genomics Data Center (https://ngdc.cncb.ac.cn/, accession numbers: HRA002920 and HRA002922).

Funding

This work was funded by the National Natural Science Foundation of China (82071286, 81671132, and 81471180) and the National Natural Science Foundation of Chongqing, China (CSTB2023NSCQ-BHX0204).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.