Somatic variants of MAP3K3 are sufficient to cause cerebral and spinal cord cavernous malformations

Abstract

Cerebral cavernous malformations (CCMs) and spinal cord cavernous malformations (SCCMs) are common vascular abnormalities of the CNS that can lead to seizure, haemorrhage and other neurological deficits. Approximately 85% of patients present with sporadic (versus congenital) CCMs. Somatic mutations in MAP3K3 and PIK3CA were recently reported in patients with sporadic CCM, yet it remains unknown whether MAP3K3 mutation is sufficient to induce CCMs. Here we analysed whole-exome sequencing data for patients with CCM and found that ∼40% of them have a single, specific MAP3K3 mutation [c.1323C>G (p.Ile441Met)] but not any other known mutations in CCM-related genes.

We developed a mouse model of CCM with MAP3K3^I441M uniquely expressed in the endothelium of the CNS. We detected pathological phenotypes similar to those found in patients with MAP3K3^I441M. The combination of in vivo imaging and genetic labelling revealed that CCMs were initiated with endothelial expansion followed by disruption of the blood–brain barrier. Experiments with our MAP3K3^I441M mouse model demonstrated that CCM can be alleviated by treatment with rapamycin, the mTOR inhibitor. CCM pathogenesis has usually been attributed to acquisition of two or three distinct genetic mutations involving the genes CCM1/2/3 and/or PIK3CA. However, our results demonstrate that a single genetic hit is sufficient to cause CCMs.

Introduction

Cerebral cavernous malformations (CCMs) and spinal cord cavernous malformations (SCCMs) are common vascular abnormalities of the CNS, affecting ∼0.16–0.40% of the population worldwide.^1,2 The thinned, enlarged and leaky plasma membrane of endothelial cells that is characteristic of CCMs and SCCMs increases the risk of stroke, seizures and focal neurological deficits.³ Currently, no drug is available to treat SCCMs or CCMs.⁴

The propensity to develop CCMs can be inherited. Biallelic loss-of-function (LOF) mutations in three genes, namely KRIT1 (CCM1), CCM2 and PDCD10 (CCM3), have been identified as causative in familial CCMs.^5-7 Although CCM lesions were once believed to be congenital, most CCMs are sporadic and usually detected in patients between the third and fourth decades of life. The de novo formation of sporadic CCMs in patients has been frequently reported,^3,8 and nearly 85% of CCMs are sporadic.^9,10 Our team and others recently reported that somatic mutations in MAP3K3 and PIK3CA are common in sporadic CCMs and SCCMs,^11-15 but it remains unknown whether MAP3K3 mutations are the key driver of CCM and SCCM formation and pathogenesis in patients.

The aetiology of CCMs related to CCM1/2/3 mutations has been attributed to a two-hit mechanism, i.e. mutations in two of the three CCM genes.^5-7,16 Recently, however, a three-hit mechanism was proposed for severe CCMs, positing that gain-of-function (GOF) somatic mutation of the oncogene PIK3CA in addition to LOF CCM1/2/3 mutations increases CCM severity.¹⁵ Surprisingly, however, neither LOF of CCM1/2/3 nor GOF of PIK3CA alone is able to drive vascular cavernous malformations (CMs) in the brain of adult mice.^15,17-20 Here, we report that a high percentage (66%) of all MAP3K3^I441M-mutated CCM patients have a singular MAP3K3^I441M mutation without any other known mutations in CCM-related genes (e.g. CCM1/2/3, PIK3CA, etc.). Through establishing a mouse model of CCM in which MAP3K3^I441M is uniquely expressed in the endothelium of the CNS, we show that expression of the MAP3K3^I441M mutant alone caused CCMs and SCCMs in mice of different ages. The CCM pathological phenotypes in this mouse model were similar to what has been observed in human sporadic CCM lesions. In vivo time-lapse imaging of the mature brain from MAP3K3^I441M mice revealed the formation of CCMs with dilated, low-flow and leaky caverns. MAP3K3^I441M initiated CCM formation by upregulating the production of the cytoskeleton and increasing the size of individual endothelial cells. Our results also demonstrated that MAP3K3^I441M can activate the mTOR (mammalian target of rapamycin) pathway. Blocking mTOR with rapamycin slowed CCM progression.

Materials and methods

Animals

All mouse experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at the Chinese Institute for Brain Research, Beijing. NG2DsRedBAC (Cat. No. 008241) and Ai14 transgenic mouse strains (Cat. No. 007908) were purchased from Jackson Lab. Pdgfrb-Cre and Cdh5-CreER were acquired from Volkhard Lindner and Ralf H. Adams’s laboratories, respectively.

Plasmid construction, adeno-associated virus packaging and injection

MAP3K3^I441M cDNA fragment was cloned into the pAAV-CAG-EGFP vector via adding a P2A sequence between MAP3K3 and EGFP, generating the plasmid pAAV-CAG-MAP3K3^I441M-P2A-EGFP. A pXX2-187-NRGTEWD (termed ‘BR1’) plasmid containing the adeno-associated virus (AAV) rep/cap genes with a sequence translating into a brain vascular endothelial-targeting peptide (NRGTEWD) from the laboratory of Jakob Körbelin.²¹ Three plasmids, including pAAV-CAG-MAP3K3^I441M-P2A-EGFP, pXX2-187-NRGTEWD and a helper vector (pAdDeltaF6), were transfected into HEK293T cells via Lipofectamine 2000. The AAV-BR1-MAP3K3^I441M-P2A-EGFP capsids were harvested and purified 72 h later. Mice were injected with AAV-BR1-MAP3K3^I441M-P2A-EGFP (3 × 10¹¹ genome copies per mouse) through the superficial temporal vein.²²

MRI evaluation

We performed MRI experiments on a Bruker 7.0 T MRI scanner (Bruker Pharmascan 70/16). A 23 mm surface coil and a 12-cm diameter self-shielded gradient system were equipped to acquire signals. The user interface was Paravision 5.1 software (Bruker BioSpin) and a Linux PC running Topspin 2.0. For mouse anaesthetization, 5% isoflurane and 95% O₂ mixture were used for induction in a plexiglass chamber, and 2% isoflurane/98% O₂ mixture was used for maintenance in the centre of the magnetic field. Respiratory rate was closely monitored by an animal physiological guarding system. T₂-weighted imaging was performed with a rapid acquisition method with relaxation enhancement (RARE). The detailed parameters were as follows: repetition time: 3500 ms, echo time: 33 ms, RARE factor: 4, field of view: 21 × 21 mm, acquisition matrix: 256 × 256, and slice thickness: 0.5 mm. The number of lesions was quantified from T₂-weighted images.

Cranial window preparation and live imaging

Before in vivo imaging, a chronic cranial window was made as previously reported.²³ Briefly, mice (1–2 months old) were anaesthetized with isoflurane (5% for induction, 2% for maintenance) and fixed in a homemade stereotaxic frame. Mice were kept warm with a heating pad during the operation. Dexamethasone (100 μl, 1 mg/ml in saline) was injected intraperitoneally before surgery. A craniotomy (3 mm in diameter) was made over the somatosensory cortex (1–2 mm lateral and 1–2.5 mm posterior to bregma) with a dental drill. A sterile coverslip (3 mm in diameter) with a customized titanium ring was placed on the window and sealed with cyanoacrylate glue. Dental cement was applied to the remaining exposed skull surface. Finally, a custom-made head plate was attached to the skull with dental cement. All surgical procedures were kept sterile. Ceftiofur sodium (100 μl, 0.1 g/ml, saline) was injected intraperitoneally in the following 3 days. Mice were allowed to recover for at least 2 weeks after surgery before we performed live imaging.

Observation was usually performed 2–3 weeks after AAV injection. For live imaging of the brain vasculature, head-fixed mice were placed on a customized microscope stage. To exclude the effect of anaesthesia on cerebral blood flow, mice were imaged in a state of consciousness. Mice were trained to accustom them to the platform for at least 30 min each day for 3 days before imaging. To observe blood flow, FITC-dextran dye (2000 kDa, Sigma) at a dose of 100 μl (1% w/v, saline) was injected through the retro-orbital vein. Cerebral vasculature was observed by an upright two-photon laser scanning microscope (TPLSM, FVMPE-RS, Olympus) controlled by software (F31S-SW, Olympus) and a Mai Tai HP Ti:Sapphire laser (InSight X3^TM, spectra physics) with a 25 × 1.05 NA water immersion objective lens (Olympus) at a 1024 × 1024 pixel resolution, with galvanometric scanners. The optimal excitation wavelength was set as 990 nm for both FITC and Dsred. The emitted fluorescence from FITC and Dsred was detected with a non-descanned GaAsp detector (Olympus) and emission bandpass filters were BA575-645 (red) and BA495-540 (green). Z-stack images were collected at 2-μm step size in a volume of 509 × 509 × 500 μm. For the measurement of red blood cell velocity in each capillary, X-T line scanning was conducted along a 15–50 μm range for 2000 cycles.

Histological staining

Mice were perfused with PBS followed with 4% paraformaldehyde (PFA). Brain tissues were collected and paraffin-embedded sections (6 μm) were stained with haematoxylin and eosin (H&E), Prussian blue and antibodies against Ki67. Stained sections were imaged on a VS120 Virtual Slide Microscope (Olympus).

Immunostaining

Brains were collected and embedded with optimal cutting temperature compound (OCT compound) after perfusion as described. Frozen sections (40 μm) were incubated with primary antibodies against CD31 (1:100; BD Biosciences, 550274), Collagen IV (1:100; Merk-millipore, AB756P), p62 (1:200; Abcam, ab56416), Laminin (1:400; Sigma, L9393), S6 Kinase (1:200, Cell Signaling Technologies, 34475, or cleaved Caspase-3 (1:100, Cell Signaling Technologies, 9661) for 48 h at 4°C, and then stained with the related secondary antibodies conjugated with Alexa Fluor 546 (1:500; Thermo) or Alexa Fluor 647 (1:500; Thermo) as well as DAPI or Hoechst 33342 (0.5 μg/ml; Sigma) for 2 h at room temperature. Sections were mounted on glass slides. Images were acquired via Leica SP8 confocal microscope. For confocal imaging, the following wavelengths were used: 405 nm for DAPI or Hoechst 33342, 488 nm for GFP, 552 nm for tdTomato/DsRed or Alexa Fluor 546, and 638 nm for Alexa Fluor 647.

Clinical specimens of human samples

Frozen CM samples obtained from 104 patients were used for mutation analysis with whole-exome sequencing and/or droplet digital polymerase chain reaction (ddPCR). Paraffin-embedded sections from five patients with MAP3K3^I441M were used for histological staining and immunostaining. The control samples (n = 2 patients) were temporal lobe tissue from patients with temporal lobe epilepsy but no CCMs. One sample of a glioma was used for immunostaining for Ki67. All the patients included in this study underwent surgical resection at Xuanwu Hospital, Capital Medical University, China between May 2017 and November 2021. We should note, some patients from our previous report¹¹ were also included in this study. The study was approved by the Ethics Committee of Xuanwu Hospital (NO.2016032) and written informed consent from all patients or their guardians was obtained before surgery.

Whole-exome sequencing and droplet digital PCR of human cavernous malformations

Human tissue samples obtained after resection were immediately transferred to the laboratory, snap-frozen in liquid nitrogen, and stored at −80°C. The frozen tissue samples were sent to the sequencing facility of Nanjing Geneseeq Biotechnology Inc. for whole-exome sequencing (WES) and ddPCR analyses. For WES, xGen Exome Research Panel v1.0 (Integrated DNA Technologies) was used for hybridization capture. The libraries were quantified by qPCR using a KAPA Library Quantification Kit (KAPA Biosystems) after target enrichment. Sequencing was performed on HiSeq4000 NGS platforms (Illumina) with paired-end 150 bp sequencing chemistry. We carried out ddPCR as previously reported.²⁴ Detection of rare variants in MAP3K3 (NM_203351.1) and PIK3CA (NM_006218.2) was performed on the QX200 ddPCR system (Bio-Rad). Primers and probes for MAP3K3 c.1416C>G p.Ile441Met were customized and synthesized at Integrated DNA Technologies (IDT) with the following nucleotide sequences: forward primer 5′-TCGTGCAGTACTATGGCTGTC-3′, reverse primer 5′-TCACATGCATTCAAGGGGCA-3′, wild-type allele locked nucleic acid (LNA) probe 5′-HEX-TGACCATCTTCAT-3′, mutant allele specific LNA probe 5′-FAM-TGACCATGTTCAT-3′.

Infection of endothelial cells in vitro with lentiviruses

Human MAP3K3^I441M plasmids containing the open reading frames were amplified using primers with an added XhoI site on the 5′ end and a KpnI site on the 3′ end. PCR amplicons were then subcloned into GV141 using the XhoI/KpnI sites. MAP3K3^I441M cDNA fragment was cloned into the vector pLJM1-EGFP. These constructed plasmids were co-transfected into HEK293T cells, together with psPAX2 and pMD2.G using Lipofectamine 2000 reagent (Thermo). The lentiviruses were harvested after 24 h and then added to medium of bEnd.3 cells for 48-h culture with polybrene (10 μg/ml). The bEnd.3 cells with MAP3K3^I441M and EGFP stable co-expression were sorted via flow cytometry (BD FACSAria III) for subsequent bulk RNA sequencing.

Immunoprecipitation and western blot

HEK293T cells were cultured in Dulbecco’s modified Eagle medium (DMEM, Cytiva) with 10% foetal bovine serum (FBS, Gibco) and 1% penicillin/streptomycin (Thermo Fisher Scientific) at 37°C in a 5% CO₂ incubator. jetPRIME DNA transfection reagents (Polyplus) were used for plasmids transfection. Forty-eight hours after transfection, cells were washed twice with PBS and incubated with lysis buffer, and then gently scraped into tubes. The incubation was on an end-over-end rotator for 1 h at 4°C. The lysis buffer includes 1% NP-40, 150 mM NaCl, 1 mM NaF, 50 mM Tris (pH 7.6) and protease inhibitor mixture (Roche). After centrifugation at 4°C at 12 000g for 10 min, the supernatants were collected and incubated with the primary antibody at 4°C overnight. The immune complexes were precipitated with prewashed protein A/G agarose beads (Beyotime) for 3.5 h at 4°C. The beads were washed six times. The SDS-PAGE buffer was then added and boiled at 95°C for 10 min. The following antibodies were used in the experiments: mouse anti-Myc (ABclonal, AE010), rabbit anti-Myc (ABclonal, AE070), and rabbit anti-HA (Abcam, Ab137838).

RNA sequencing and data analysis

Total RNA of endothelial cells was extracted with RNeasy Micro Kit (QIAGEN, Cat. No. 74004) as suggested by the manufacturer’s protocol. RNA quality was measured with an Aglient RNA 6000 Pico kit. We used the RNA samples with RIN higher >8 for library preparation. The cDNA library was constructed using SMARTer® Stranded Total RNA-Seq Kit v2 (Takara).

Preprocessing and analysis of bulk RNA-seq

RNA-seq data were first subjected to ‘FastQC’ for data quality evaluation. Clean reads were mapped to the reference genome (Version: GRCm38) downloaded from Gencode by HISAT with the parameter ‘−5 30 −3 30’. Uniquely mapped reads were utilized to calculate the counts of each gene by feature counts with the parameter based on the GTF annotation downloaded from Gencode (Version: vm23). Read count matrix was normalized to library size by DESeq2 with default parameters. Hierarchical clustering was performed by factoextra package in R. Volcano plots were generated using the ggplot2 package in R, with cut-off of adjusted P-value < 0.01 and log₂(fold change) absolute value >1.5, measured by DESeq2. The heat map was generated by pheatmap with the z-score of each gene. Gene set enrichment analysis (GSEA) was performed using the cluster Profiler package in R, with adjusted P-value < 0.05.

Drug administration

Rapamycin (APExBIO, A8167) was dissolved in DMSO (10 mg/ml, stock solution) and stored below −20°C. It was diluted with sterile saline before use (working concentration, 0.4 mg/ml). Mice were administered rapamycin (4 mg/kg body weight per day) via intraperitoneal injection (i.p.) starting on the second day or 2 weeks after AAV injection and continuing daily for 4 weeks. Brains of all animals underwent MRI scanning on the second and fourth weeks after drug injection. The brains of these mice were harvested during the fourth week after MRI scanning.

EdU staining

5-Ethynyl-2’-deoxyuridine (EdU) was dissolved with saline (2 mg/ml, working solution) and stored at −20°C. Mice were administered EdU (6 mg/kg body weight) intraperitoneally once every 2 days after they received the AAV injection.

Tamoxifen preparation and treatment

Tamoxifen was dissolved with corn oil (10 mg/ml). Mice were administered tamoxifen (100 mg/kg body weight per day) for five consecutive days after they received the AAV injection.

Data analysis and statistics

Animals were grouped with randomization during the experiments. NIH ImageJ software was used to quantify the data of immunostaining analysis from confocal microscopy. All data were analysed with GraphPad Prism software and are presented as mean ± standard error of the mean (SEM). P-values were calculated using an unpaired two-sided Student’s t-test, or one-way ANOVA. Two-sided P < 0.05 was considered statistically significant and are denoted as follows: *P < 0.05; **P < 0.01; ***P < 0.001.

Data availability

The authors confirm that the data supporting the findings of this study are available within the article and its Supplementary material. Derived data supporting the findings of this study are available from the corresponding author on request.

Results

MAP3K3^I441M drives cerebral cavernous malformation formation in mice of different ages

MAP3K3 encodes MEKK3, a CCM2 binding partner.²⁵ LOF mutations in any of the three CCM1/2/3 genes activate MEKK3–KLF2/4 signalling and contribute to the development and progression of CCMs.^18,26,27 We recently detected somatic mutations of MAP3K3 and PIK3CA in CCMs of human patients. Mutations in both genes were found mainly in endothelial cells.¹¹ To investigate the role of MAP3K3^I441M in CCM formation in vivo, we intravenously injected vascular endothelial-targeting adeno-associated virus BR1 (AAV-BR1) carrying MAP3K3^I441M (AAV-BR1-CAG-MAP3K3^I441M-P2A-EGFP, i.e. AAV-MAP3K3^I441M) into mice on postnatal Day 35 (P35) (Fig. 1A). AAV-BR1-CAG-P2A-EGFP (i.e. AAV-EGFP) was used as the control vector (Supplementary Fig. 1). These mice were subjected to T₂ MRI once per week. MRI revealed no obvious microlesions at 1 week after AAV-MAP3K3^I441M injection. At 2 weeks post-injection, however, many intracranial lesions with low MRI signal were identified that resembled human CCMs. The number of the lesions increased significantly between the second and fourth week after AAV-MAP3K3^I441M administration, with the number plateauing during the fifth and sixth weeks (Fig. 1B–D). We further performed MRI in mice at 4 months after injection of the AAV-MAP3K3^I441M. We found that the number of the lesions increased between the second and fourth months, while no larger lesions were identified in the MRI, which indicated the MAP3K3 GOF mutations cause quiescent CCMs after a prolonged exposure to MAP3K3^I441M (Supplementary Fig. 2). Notably, single mutations of CCM1/2/3 or PIK3CA have not been reported to cause CCM lesions in the adult mouse brain.^15,17 Therefore, to further determine whether MAP3K3^I441M could drive lesion formation in both the developing and mature brain, we injected AAV-MAP3K3^I441M into mice at P5, P35, P150 and P330 (Fig. 1E) and then performed T₂ MRI 2 weeks after injection. Surprisingly, the lesions were distributed throughout the entire brain even in the P330 group (Fig. 1E). MRI showed that all mice developed lesions similar to human CCMs, regardless of the postnatal age at which they were injected (Fig. 1E). Dissection of each mouse brain and imaging with a stereoscope revealed numerous small vascular lesions located in both the brain and spinal cord (Fig. 1F and G and Supplementary Fig. 3). Staining of brain sections with H&E and Prussian blue revealed dilated vascular sinusoids lined by endothelial cells. Haemorrhages and staining for haemosiderin were observed around these lesions (Fig. 2A and B). Lesions were absent in the brain of control mice injected with AAV-EGFP. The MAP3K3^I441M-induced CCMs in mouse brain resembled MAP3K3^I441M-mutated CCMs in human patients (Fig. 2B). Together, these findings suggested that a single hit, i.e. MAP3K3^I441M mutation, can cause CMs in the brain and spinal cord of developing and adult mice. To determine what percentage of CCM patients having MAP3K3^I441M mutations alone or together with other mutations that have been reported in CCM patients, we systematically analysed all known CCM-related mutations in the WES results acquired for surgically resected CCM lesions from 104 patients (Supplementary Fig. 4A–D). These mutations included MAP3K3 (p.I441M), PIK3CA (p.C378Y), PIK3CA (p.C420R), PIK3CA (p.E542K), PIK3CA (p.E545K), PIK3CA (p.H1047R), AKT1 (p.E17K), CCM1, CCM2 and CCM3, all of which have previously been found in CCM lesions.^11-13,15 The same patient-tissue samples were also analysed with ddPCR. The landscape of pathological mutations revealed that MAP3K3^I441M mutations were present in 45.2% of all CM lesions, including 57.3% of sporadic CMs, but were not seen in familial CMs. Of all MAP3K3^I441M-mutated CMs, 66.0% had no other known pathological mutations (Supplementary Fig. 4F–H). MAP3K3^I441M-mutated CMs occurred at comparable rates in male and female patients (57.4% and 42.6%, respectively; n = 47 in total). Interestingly, of all MAP3K3^I441M patients with CMs, 76.6% had SCCMs (Supplementary Fig. 4K). The patients with MAP3K3^I441M were distributed across a wide range of ages (mean ± SEM: 35.8 ± 2.2 years, range, 10–80 years old, n = 47 patients) (Supplementary Fig. 5L), but most were adults (87.2%, ≥18 years old). These results suggested that MAP3K3^I441M mutations are widely and exclusively present in sporadic CMs, mostly affecting adults. Together, these results illustrated that MAP3K3^I441M mutation alone—without any other known mutation—is a common genotype in sporadic CCMs and especially SCCMs. This strongly indicates that MAP3K3^I441M alone can induce CCMs in human patients, which is consistent with and supported by the results acquired with our CCM mouse model after administering AAV-MAP3K3^I441M (Fig. 1). Unlike other CCM-related mutations (i.e. to CCM1, CCM3 and PIK3CA) reported previously,^15,17 our results for humans and mice strongly suggest that MAP3K3^I441M (i.e. one hit) can drive CCMs and SCCMs in both young and adult mice.

Time-lapse imaging of CCM formation in MAP3K3^I441M mice

To better understand the development of CCMs caused by MAP3K3^I441Min vivo, we used transcranial two-photon laser scanning microscopy (TPLSM) to analyse NG2DsredBACtg and Pdgfrb-Cre::Ai14 mice after injection of AAV-MAP3K3^I441M at P35 (Fig. 3A). NG2DsredBACtg is a transgenic mouse line that is frequently used for labelling pericytes and smooth-muscle cells in the brain vasculature.^28,29 Daily imaging under a stereomicroscope revealed lesion formation in the superficial vasculature of the brain. The de novo vascular abnormalities appeared 2–3 weeks after AAV-MAP3K3^I441M injection (Fig. 3B), which was similar to the phenotype observed in patients (Supplementary Fig. 4B and C). To determine the distribution and alteration of blood flow in the cerebral vasculature in the CCM lesions, FITC-dextran dye was injected into the retro-orbital vein with subsequent analysis with time-lapse imaging via TPLSM. Massively bulged vascular caverns were observed in small vessels (Fig. 3C and D). Blood leakage was observed, reflecting a typical characteristic of CCMs in humans with CCM1/2/3 mutations,^30-32 and dye leakage was also seen in the caverns of the MAP3K3^I441M-induced CCMs (Fig. 3F), indicating the disruption of the blood–brain barrier in the vasculature of CCM lesions. Interestingly, not all caverns leaked (8.9%, or n = 16 of 180 caverns) (Fig. 3H). Similar results were obtained with brain sections using endothelial cell-specific transgenic Cdh5-CreER::Ai14 mice³³ after injection of AAV-MAP3K3^I441M (Fig. 2C and D). Pericytes have been reported to play a critical role in stabilizing newly formed capillaries³⁴ and in the maintenance of the blood–brain barrier.^35,36 The use of NG2DsredBACtg mice allowed us to measure the coverage of pericytes in normal and cavernous vessels. A low proportion of pericyte-free caverns was observed among the newly formed caverns (5.6%, n = 10 of 180 caverns; Fig. 3G). These results were confirmed in brain sections stained with anti-collagen IV in Pdgfrb-Cre::Ai14 mice (Fig. 2E and F), another transgenic line in which pericyte labelling can be readily quantified.³⁷ To further investigate whether blood flow was altered in the vascular caverns induced by MAP3K3^I441M, we measured red blood cell velocity with line scanning under TPLSM. This analysis revealed both turbulent and stagnant blood flow in the caverns, which resembled the reduced perfusion of MAP3K3^I441M-mutated CCMs observed in patients (Fig. 3M–O and Supplementary Fig. 4C). Thus, long-term in vivo imaging of individual CCMs in the MAP3K3^I441M mouse brain revealed no significant alteration of pericyte coverage during the early stage of formation of vascular caverns. Also, leakage and alterations of blood flow were detected in vascular caverns.

MAP3K3^I441M initiates cerebral cavernous malformation formation without an obvious increase in endothelial cell proliferation

Previous studies have revealed that GOF of oncogenic PIK3CA somatic mutations drives excessive cell proliferation in CCMs.¹⁵ To determine whether the GOF MAP3K3^I441M mutation initiates CCM formation via a similar mechanism, we used P35 Cdh5-CreER::Ai14tg mice as an experimental model. The mice were first infected with AAV-MAP3K3^I441M; after 4 days, they received a daily injection of EdU for 10 days to label dividing cells. After two additional days, an abundance of dividing cells (i.e. EdU⁺) were detected in the subventricular zone of the brain, a site of pronounced neurogenesis. Interestingly, no dividing endothelial cells were found in the 50 MAP3K3^I441M-mutated CCMs that we imaged (0.0%, n = 0 of 50 lesions, 0.0%, n = 0/100 normal vessels) (Fig. 4A and B). We then carried out immunohistochemistry for Ki67, a cell proliferation marker, in human tissues of CCMs expressing MAP3K3^I441M. Ki67⁺ endothelial cells were rarely detected in CCM tissue (Fig. 4C and D, n = 2 patients). These results suggested that endothelial cell proliferation is likely not the main driving force for initiation of vascular caverns, which is consistent with our results obtained for human CCMs.

To further determine how MAP3K3^I441M alters endothelial cells and causes vascular CMs, we compared the size of endothelial cells of MAP3K3^I441M-induced vascular caverns with the size of those in normal capillaries. Interestingly, the area of individual endothelial cells expressing MAP3K3^I441M was significantly larger than that of individual cells lacking MAP3K3^I441M expression (Fig. 4F–H). Autophagic degradation is crucial for cell survival.³⁸ We stained blood vessels with p62, a marker for autophagy and observed many p62 bodies surrounding caverns of blood vessels in mice with AAV-MAP3K3^I441M infection (Supplementary Fig. 5). We also detected there were higher cleaved caspase-3 signals in blood vessels of MAP3K3^I441M-induced vascular caverns (Supplementary Fig. 6).

mTOR pathway is activated in endothelial cells expressing MAP3K3^I441M

To further explore the underlying molecular pathways altered by MAP3K3^I441M in endothelial cells, the mouse endothelial cell line bEnd.3 was first infected with lentivirus-MAP3K3^I441M (LV- MAP3K3^I441M-P2A-EGFP; LV-EGFP was used as the control). Total RNA from these cells was then subjected to RNA sequencing. The transcriptome of the MAP3K3^I441M-expressing cells differed significantly from that of the control cells (Fig. 5A–E and Supplementary Fig. 7). Considering that MAP3K3^I441M mutations increase the size of the plasma membrane of individual endothelial cells, GSEA was carried out, revealing upregulation of the structural constituent of cytoskeleton pathway in MAP3K3^I441M-expressing endothelial cells (Fig. 5Cand Supplementary Fig. 7). Among the top eight most enriched actin-encoding genes in endothelial cells, five ACT genes were upregulated in cells expressing MAP3K3^I441M. Previous results reported that MAP3K3 regulates the phosphorylation of p62 and promotes mTOR activation.³⁹ Pull-down experiments were performed in cells with overexpression of three different constructs, Myc-GFP, HA-mTOR, and Myc-MAP3K3. We also performed co-immunoprecipitation experiments in two groups of cells co-transfected with HA-mTOR plus Myc-GFP, and HA-mTOR plus Myc-MAP3K3, repectively. Proteins were pulled down with the Myc antibody. We found that Myc-MAP3K3 interacts with mTOR directly. These results indicated that MAP3K3 activation increases the activity of the mTOR pathway, which provides direct evidence for how blocking mTOR with rapamycin alleviates symptoms in mice with MAP3K3^I441M-induced CCMs (Fig. 5F). The results of the transcriptomic analysis implied that MAP3K3^I441M can activate the mTOR pathway, leading to alterations of the cytoskeleton before cavern formation in the brain.

Blocking mTOR activity with rapamycin prevents progression of MAP3K3^I441M-mediated formation of cerebral cavernous malformations

Rapamycin is an mTOR inhibitor that has been approved by the US Food and Drug Administration. As such, rapamycin is a promising agent for targeted pharmacological management of vascular malformations.^15,40,41 To test whether blocking the mTOR pathway could slow the formation of CCMs, rapamycin was administered to mice with MAP3K3^I441M-induced CCMs. The fact that the mTOR pathway was activated in endothelial cells expressing MAP3K3^I441M (Fig. 5) suggested that rapamycin might mitigate CCM formation in MAP3K3^I441M mice. Mice injected with AAV-MAP3K3^I441M were treated with rapamycin (4 mg/kg, i.p. injection) or vehicle control daily for 4 weeks. All animals underwent MRI at 2 and 4 weeks after AAV injection. At the 4-week time point, each brain was imaged and harvested (Fig. 6A). The results of MRI and H&E staining of the brain sections revealed that 4 weeks of rapamycin treatment significantly decreased the number of CCM lesions (Fig. 6B–H), suggesting that blocking mTOR with rapamycin might be an effective therapeutic option for MAP3K3^I441M-driven sporadic CCMs. To further verify our results, we also stained the section with p62, a marker for autophagy. The number of p62 bodies significantly decreased in rapamycin-treated mouse group (Fig. 6I and J). As sporadic CMs with MAP3K3^I441M mutations are usually found in adult patients, to further test whether rapamycin treatment works on mice after caverns are formed, we infected the aged mice (age, 18–24 months old) with AAV-MAP3K3^I441M for 2 weeks before we treated these mice with rapamycin (Fig. 6K). We observed that rapamycin could also substantially inhibit the progression of CCM formation after a prolonged exposure with MAP3K3^I441M in these mice (Fig. 6L). We performed pathological analyses (staining for autophagy, microlesion/bleeding and apoptosis) between the vehicle and rapamycin-treated mice. We found that rapamycin significantly alleviated the symptoms in mice with MAP3K3^I441M-induced CCMs. The density and size of microlesions decreased in the brain sections of rapamycin-treated groups. Furthermore, we also detected that the number of autophagy bodies decreased dramatically in rapamycin-treated mice. We treated the MAP3K3^I441M CCM mice with ponatinib (AP24534), a blocker for MAP3K3. However, we detected severe side effects in survival, body weight and neurological behaviours in both CCM and wild-type mice. Thus, we did not continue the treatments.

Discussion

Both CCMs and SCCMs are common vascular abnormalities of the CNS. They are major causes of stroke, seizures and neurological deficits. Activating, single somatic mutations in the RAS–MAPK pathway have been reported to drive arteriovenous malformations.^42,43 Although multiple genes may contribute to CCM formation, no reported single gene variant has been demonstrated to be sufficient for CCM induction in an adult mouse model.^15,17 Thus, previous accounts mainly proposed two-hit hypotheses involving biallelic LOF via CCM1/2/3 mutations or a three-hit hypothesis including both CCM1/2/3 LOF and PIK3CA GOF.^5-7,15 In mice, CCMs have only been observed when LOF of CCM1/2/3 or GOF of PIK3CA occurs in early postnatal stages. After P21, CCM phenotypes are rare in the mouse brain, typically occurring only when CCM1/2/3 LOF and PIK3CA GOF co-exist in the same endothelial cells.¹⁵ Unlike CCM1/2/3- and PIK3CA-induced CCMs, MAP3K3^I441M mutations alone were widely identified in our patient cohort with sporadic CCMs and SCCMs (Fig. 3). This strongly suggests that a one-hit mechanism (i.e. MAP3K3 →MAP3K3^I441M) may underlie the formation of CCMs, and this is supported by results from our mouse model of CCM. The introduction of MAP3K3^I441M mutations into mice at age P5, P35, P150 and P330 caused CMs in the brain or spinal cord. These results demonstrate that expression of MAP3K3^I441M specifically in endothelial cells of the brain or spinal cord is sufficient to drive CCM formation with dilated, low-flow leaky caverns (i.e. CCM lesions) in adult mice. Interestingly, our EdU pulse-chase experiment with mice did not reveal an obvious increase in endothelial-cell proliferation in CCM lesions. We conclude that the mechanism of MAP3K3^I441M-mediated CCM formation differs from that expected based on the simple activation of MEKK3 downstream signalling pathways,^11,12 but this conclusion must be confirmed in a future study.

In mice expressing MAP3K3^I441M, the number of brain lesions is higher with respect to spinal cord CCMs while the opposite has been observed in CMs collected by patients.^11,44 Based on our clinical experience and observation of >100 patients with MAP3K3 mutation at Xuanwu Hospital, the patients with MAP3K3^I441M mutation have a lower likelihood of bleeding in the CNS than patients without this specific mutation. Since MAP3K3^I441M patients do not have severe symptoms in the brain, they tend not to see doctors. As a result, these patients with micro-lesions in the brain have a lower probability of being diagnosed in hospitals than those with CCM lesions in the spinal cord. Since the spinal cord is directly involved in various sensory functions, small lesions in the spinal cord caused by MAP3K3^I441M mutation have a higher potential to cause symptoms that are bothersome to patients.

Understanding the mechanism underlying the initiation of MAP3K3^I441M-induced sporadic CCMs is crucial for developing new therapies for the ∼40% of CCM patients with this mutation. Previous reports have posited various explanations for CCMs and proposed that they are likely congenital lesions.^45,46 The lifetime risk of haemorrhage has been underestimated. Our animal model, which takes into account postnatal age at CCM onset, will prompt a rethinking of how sporadic CCMs form. This animal model will also provide the CCM research community with a means for studying lesion development, haemorrhage onset, and the occurrence of other symptoms, such as seizures, in patients with sporadic CCMs.

Surgical resection remains the only treatment for most CCMs, as no pharmacological option currently exists.⁴⁷ However, resection carries the risk of surgical complication and is extremely challenging for lesions located in deep-brain regions; consequently, morbidity and mortality rates are non-negligible.^3,47 Recent studies have reported that somatic MAP3K3 mutations activate MAP3K3 downstream signalling pathways.^11,12 The consequent increase in CCM–MAP3K3–KLF2/4 signalling upregulates PI3K–mTORC1 signalling, and hence rapamycin can prevent CCM lesion formation stemming from CCM1/2/3 LOF and PIK3CA GOF.¹⁵ Somatic mutation MAP3K3^I441M is common in sporadic CCMs and SCCMs and it has been shown as a GOF mutation. This GOF mutation in MAP3K3 has been proposed to have similar functional consequences of LOF mutations in CCM genes.⁴⁸ It was also reported that the MAP3K3 GOF mutation causes quiescent CCM.⁴⁹ Our results reveal that MAP3K3^I441M mutations drive the formation of CCMs and SCCMs via activation of MAPK and mTOR signalling in endothelial cells and altering the cytoskeleton. Furthermore, our RNA sequencing results demonstrate that MAP3K3 can interact with mTOR directly, although LV-MAP3K3^WT-P2A-EGFP is a more reasonable control for the mutant protein. Since we could not treat CCM mice with MAP3K3 inhibitor ponatinib (AP24534) for a longer time, we could not evaluate its effect.

Our results also demonstrate that MAP3K3^I441M-induced CCMs can be prevented by treatment with the mTOR inhibitor rapamycin. These results provide insights into the pathogenesis of sporadic CCM and suggest a therapeutic alternative.

Acknowledgements

We thank Xinwei Gao, Mingyue Jia, and Dr Qingchun Guo from CIBR Imaging facility, Yingzi Ju from CIBR Vector Core, Jingmei Chen and Dr Li Zhang from Genomics Center, Shufang Huang and Dr Wenlong Li from CIBR Laboratory Animal Resource Center for their support. We thank the lab members at Sun and Ge lab and colleagues from Chinese Institute for Brain Research, Beijing for discussion. Dr Bridget Samuels for critical reading.

Funding

This work is supported by National Natural Science Foundation of China with grant (No. 32170964) and CIBR start-up funds to W-P.G., grant (No.81971104) to T.H., grant (No. 82201440) to J.R., and grant (No. 81971113) to H.Z.; Beijing Municipal Science and Technology Commission with grant Z201100005520024 to T.H. Beijing Municipal Administration of Hospitals with grant DFL20180801 to H.Z. and QML20190802 to T.H. Beijing Municipal Education Commission with grant CIT&TCD201904095 to T.H.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

References

Author notes