2.1 Primary surgical techniques

The primary treatment modalities for CVMs include surgical resection, embolization, and SRS. Surgical resection entails the direct removal of the malformation via craniotomy, often preferred for accessible AVMs and CCMs that cause significant symptoms or pose a high risk of hemorrhage.^{1, 27-29} However, surgical intervention carries a risk of neurological damage, particularly for malformations located in eloquent brain areas.^{30, 31} Recent advancements in intraoperative imaging and neuronavigation have enhanced the precision of surgical resection and reduced the risk of neurological deficits.

Embolization is a minimally invasive procedure that injects embolic agents into abnormal blood vessels to occlude them. This technique is frequently used as a preoperative adjunct to decrease the size of AVMs or as a standalone treatment for smaller malformations. While embolization can effectively reduce AVM size, it may not eliminate the malformations, necessitating additional therapies and increasing the risk of vessel rupture, as indicated by a subgroup analysis from the TOBAS trial.^{32, 33} Furthermore, incomplete embolization may elevate the risk of AVM bleeding.^{33, 34} CCMs, which are angiographically occult, are not amenable to endovascular approaches.

SRS has emerged as a noninvasive alternative for AVM treatment, delivering focused radiation beams to induce gradual occlusion of abnormal vessels. This method is beneficial for deep-seated or surgically inaccessible malformations. Although SRS has a lower immediate risk than surgery, the effects of radiation may take 2–3 years to manifest, with complications accumulating over time. While the SRS obliteration rate of SRS is approximately 70%,^{35, 36} there is a potential risk of radiation-induced complications, including delayed cyst formation and vasculopathy issues.^37-40 The use of SRS for CCMs is controversial due to the presence of hemosiderin, a radiation sensitizer that can significantly increase the risk of complications.^41-44 Despite advancements in these techniques, the primary goal remains the physical destruction of the CVMs.

2.2 Evolution from life-saving to function-preserving treatments

While these treatments aim to save lives by preventing hemorrhage, improving quality of life has become an equally important objective. This shift has led to the development and implementation of new technologies aimed at preserving neurological function before, during, and after surgical interventions. Advances in techniques such as functional MRI (fMRI), diffusion tensor imaging (DTI), intraoperative neuronavigation, and intraoperative Doppler ultrasound enable surgeons to accurately delineate malformation boundaries and avoid critical functional areas, thus reducing the risk of postoperative neurological deficits.

fMRI allows for mapping critical brain functions, such as motor and language areas, before surgery with high temporal and spatial resolutions. This preoperative evaluation assists in planning the surgical approach to avoid critical regions and minimize functional deficits.⁴⁵ Wang et al.⁴⁶ compared the results of electrical cortical stimulation and fMRI in 43 patients with AVMs, demonstrating that fMRI exhibits high sensitivity for motor mapping.

DTI and fiber bundle tracking create spatial images by calculating the anisotropy of water molecules using MRI, which facilitates fiber tracking.⁴⁷ This technique aids in defining the lesion-to-eloquence distance (LED), the distance between the lesion and the nearest eloquent area. Jiao et al.⁴⁸ found that LED is an essential predictor of preoperative risk assessment in patients with AVM, developing the HDVL grading system that provides a detailed evaluation of vascular malformations. This system includes hemorrhagic presentation, diffuseness, deep venous drainage, and LED. It is more accurate than the Spetzler-Martin grading system in predicting surgical outcomes. This comprehensive evaluation assists in surgical planning and risk assessment, ensuring a more targeted and effective intervention.^{49, 50}

Intraoperative Doppler ultrasonography can detect residual or small CCM lesions. This technique provides immediate feedback on the hemodynamic status of the vessels involved in the malformation, enabling the surgeon to make informed decisions regarding resection. Enhanced contrast in ultrasound angiography can help identify superficial and deeper arterial blood vessels.^{51, 52} However, ultrasound may miss areas of slow blood flow or very small blood vessels (< 0.6 mm in diameter).⁵³

Intraoperative neuronavigation systems offer real-time guidance during surgery, allowing for precise localization and resection of malformations. By integrating preoperative CT or MRI data, neuronavigation enhances surgical accuracy and reduces the risk of neurological damage. Winter et al.⁵⁴ found that combining neuronavigation with fMRI significantly improved resection volume, surgical duration, and neurological outcomes in patients with supratentorial CCMs. Mathiesen et al.⁵⁵ proposed an intraoperative imaging and navigation system that utilizes intraoperatively acquired three-dimensional ultrasound data alongside preoperative MRI scans and MR angiograms, which demonstrated improved outcomes in managing AVM pathological vessels.

Postoperative rehabilitation programs tailored to individual patient needs play a vital role in restoring and optimizing neurological function. These programs may include physical therapy, acupuncture, and electrical nerve stimulation.^56-59

Despite these advancements, the fundamental principle of CVM treatment remains the physical destruction or removal of abnormal vascular structures. This underscores the necessity for alternative therapeutic approaches to address the underlying causes of these malformations.

3.1 Genetic mutations in CVMs

3.2 Mechanisms of mutation-induced pathogenesis

3.2.1 Mechanism in AVM

Recent observations of EC-specific KRAS and BRAF mutations in AVMs suggest that the RAS/RAF/MAPK pathway significantly contributes to disease pathogenesis.⁹⁵ Nikolaev et al.⁶³ proposed that KRAS mutations could activate the MAPK/ERK pathway, leading to angiogenesis, cell migration, and proliferation.⁹⁶ Mouse and zebrafish models further confirm that KRAS mutations enhance AVM progression through the MEK/ERK pathway rather than PI3K signaling.^{97, 98} KRAS mutations can also induce endothelial-to-mesenchymal transition (EndMT), a crucial process in AVM development.⁶⁹ Xu et al.⁹⁹ suggested that the TGF-β/BMP–SMAD4 pathway mediates this process, while the knockdown of SMAD4 can reverse EndMT. Similar to KRAS, BRAF is a well-established oncogene. Studies have shown that the BRAF^V600E mutation induces typical pathological features of AVM in mice, including vascular wall instability, dilation, focal intracranial hemorrhage, and an inflammatory microenvironment.¹⁰⁰ These effects are thought to be mediated through the MAP pathway and its downstream signal transduction. However, the specific underlying mechanisms warrant further investigation. The neurovascular unit (NVU) is a complex composed of neurons, glial cells (including microglia, astrocytes, and oligodendrocytes), and vascular components (pericytes, ECs, and smooth muscle cells). Studies indicate that pericytes in the NVU are significantly reduced in AVM, and the Notch or PDGFB pathways contribute to severe AVM by disrupting pericyte homeostasis.^101-104 ECs also play a crucial role in the NVU. Studies have shown that MHC class II molecules are abnormally expressed in AVMs, potentially affecting interactions with immune cells.¹⁰¹

Although KRAS mutations in ECs of AVM are infrequent, widespread pathological changes in the endothelium indicate that low-frequency KRAS mutations may amplify signaling by disturbing cellular communication and facilitating interactions between ECs and other cell types. He et al.¹⁰⁵ found that ECs with the KRAS^G12D mutation release exosomes, which promote EndMT in adjacent normal ECs and induce extensive vascular remodeling. This discovery reveals the communication between mutated and normal endothelium in AVMs and offers new strategies for further investigating AVM progression (Figure 1).

3.2.2 Mechanism in CCM

By constructing animal models with corresponding mutations, it has been demonstrated that mutations in the CCM1/2/3, MAP3K3, and PIK3CA genes are pathogenic for CCM.^{83, 84, 86, 106-108} Several mechanistic studies have established the MEKK3–KLF2/4 signaling pathway as a central signaling axis in CCM disease pathogenesis. Additionally, various auxiliary factors contribute to CCM development.

Genomic studies of patients with CCM have identified LOF mutations in CCM1/2/3 within CCM lesions. Systemic gene mutations in animal models indicate that CCM genes are essential for normal vascular development in zebrafish and mice.^109-112 Subsequent studies using endothelial-specific CCM mutant mouse models confirmed that CCM LOF mutations are causative factors for CCM.^106-108 Protein interaction studies revealed that the CCM complex interacts with MEKK3 through CCM2, inhibiting MEKK3 under normal conditions.^{113, 114} When one of the CCM trimer subunits undergoes an inactivating mutation, its inhibitory effect on MEKK3 is reversed, activating the MEKK3 signaling pathway. In endothelial-specific CCM LOF mouse models, the knockout of MEKK3 or its downstream transcription factors, KLF2 or KLF4, is sufficient to prevent CCM formation.^115-117 Mice with KLF4 GOF mutations develop typical CCM lesions.⁸³ These studies established the MEKK3–KLF2/4 signaling pathway as the central axis in CCM pathogenesis.

The PIK3CA pathway is another crucial pathway involved in CCM formation and development. Recent genetic studies on patients with CCM found that most CCM lesions also carry PIK3CA mutations,^{83, 84} and single-nucleus sequencing indicates that PIK3CA mutations coexist with CCM mutations in the same cells.¹¹⁸ This suggests that the PI3K pathway plays a significant role in CCM pathogenesis. Additionally, studies on CCM mouse models have shown that lesions are more likely to form postnatally in areas with active angiogenesis, such as the cerebellum and retina.¹¹⁹ The earlier the CCM gene is knocked out, the more pronounced the CCM phenotype.¹²⁰ Adding the angiogenic growth factor VEGF-A or losing the antiangiogenic factor thrombospondin 1 (THBS1, TSP1) increases CCM formation.^{121, 122} These findings suggest that CCM development requires an environment conducive for EC proliferation, with PIK3CA mutations likely providing proliferative signals to ECs. Further animal model studies have shown that PIK3CA mutations exacerbate lesions in CCM and MAP3K3 mutant mouse models, and the use of the mTOR inhibitor rapamycin can inhibit CCM formation.⁸³ Due to the unique mechanism of MAP3K3–I441M, lesions with only MAP3K3 mutations cannot persist stably; however, when combined with PIK3CA mutations, lesions can remain stable for extended periods.⁸⁴ These studies demonstrate the crucial role of PIK3CA mutations and the PI3K–mTOR pathway in CCM.

In addition to the critical MEKK3 and PI3K pathways, various synergistic factors contribute to CCM development. The pathological processes of CCM formation, growth, and hemorrhage are primarily related to signaling abnormalities in the NVU, including abnormal proliferative angiogenesis, blood–brain barrier hyperpermeability, inflammatory and immune-mediated processes, anticoagulant vascular domains, and gut microbiota-driven mechanisms.^123-125 Studies have identified a “gut–brain axis” in CCM mouse models, where gram-negative bacteria in the gut can secrete lipopolysaccharides that activate the MEKK3 pathway through EC TLR4 receptors, exacerbating CCM lesions.^{126, 127} Correlation studies between the gut microbiome and CCM lesion size and single nucleotide polymorphisms in TLR4 and CD14 in patients with CCM have validated this conclusion. In the CCM3 mouse model, B cell knockout significantly reduced lesion size and perilesional hemorrhage, indicating that B cells and humoral immunity also promote CCM¹²⁸ (Figure 2).

In ECH, GJA4 mutations activate the SGK-1 signaling pathway, upregulating critical genes involved in excessive cell proliferation and loss of arterial characteristics.^{93, 94} Furthermore, GNA14 mutations lead to the upregulation of the MAPK pathway and pathways related to angiogenesis.

4.1 Imaging genomics

Imaging genomics integrates high-throughput radiomics features and mathematical models to quantify CVM characteristics, linking imaging phenotypes with genetic, mutational, and expression patterns.^{129, 130} For example, integrating genomic data with MRI can reveal specific genetic mutations associated with distinct imaging phenotypes. AVMs with KRAS mutations, for instance, exhibit distinct imaging features that correlate with a higher propensity for hemorrhage. Similarly, CCMs with I441 M mutations display unique Zabramski grading patterns, informing prognosis and treatment planning.⁸⁵

A landmark study used WES to identify somatic activating mutations in KRAS in brain AVMs.⁶³ These mutations are found in many sporadic AVMs, providing a genetic basis for their formation. The study also showed that AVMs with KRAS mutations had distinct MRI characteristics, including increased flow voids and nidus size, and correlated with a higher risk of hemorrhage. This finding aligns with research on venous malformations, which suggests that PIK3CA- or TEK-mutants activate the PI3K–Akt signaling pathway, leading to “slow-flow” vascular malformations.^{131, 132} In contrast, “high flow” vascular anomalies, including sporadic brain and extracranial AVMs, are often linked with activating mutations in the RAS–MAPK pathway.^133-136 These imaging biomarkers may help identify AVMs with specific genetic profiles, improving diagnosis and personalized treatment planning. For extracranial vascular malformations, Stor et al.¹³⁷ systematically reviewed 69 studies, finding that mutations in TIE2, PIK3CA, and PIK3R1 primarily manifested as small blue, low-flow vascular malformations. Germline mutations in genes such as EPHB4 and RASA1 typically result in large pink/red lesions with high blood flow, suggesting a correlation between the imaging manifestations of vascular malformations and the associated gene mutations.¹³⁷

Similarly, genomic imaging shows promise in the diagnosis and management of CCMs. Denier et al.²⁴ and Gault et al.¹³⁸ studied the relationship between CCM1/2/3 mutations and clinical phenotypes. Denier et al.²⁴ reported that cerebral hemorrhage was the most common first symptom in patients with CCM3 mutations, and the increase in gradient-echo sequence lesions with age differed between patients with CCM1 and CCM2 mutations.²⁴ Gault et al.¹³⁸ reported that patients with CCM1 mutations had fewer cases of bleeding unrelated to lesion size or number. Shenkar et al.¹³⁹ demonstrated that CCM3 mutations lead to robust Rho-associated protein kinase (ROCK) activity in murine and human CCM vasculature and increase brain vascular permeability in humans with CCM3 mutation. The clinical phenotype of CCM3 is more aggressive compared with familial and sporadic CCM1 and CCM2, with higher lesion burden and more frequent hemorrhages at an earlier age.¹⁴⁰ Recent case series have also identified several clinical features in patients with CCM3 mutation.^{141, 142} Nikoubashman et al.¹⁴³ utilized high-resolution MRI combined with genomic sequencing to investigate the molecular underpinnings of familial CCMs. They found that specific CCM1 or CCM3 mutations were associated with distinct imaging patterns, such as multiple small lesions and a higher propensity for hemorrhage.^{143, 144} This integration of genomic and imaging data enhanced accurate disease severity assessment and progression, improving clinical management strategies for patients with familial CCMs. Interestingly, in children with CCM, the CCM2 genotype was linked to a higher risk of bleeding, and the greater number of lesions on MRI was a significant independent predictor of future symptomatic bleeding.¹⁴⁵ Zabramski classified CCM lesions into four subtypes based on MRI findings. It was previously reported that Zabramski type IV is associated with germline mutations in CCM genes.^{77, 146} Wang et al.¹⁴⁷ reported that the annual cumulative incidence of symptomatic hemorrhage in type I CCM was significantly higher than in types II and III in patients with sporadic CCM. Weng et al.⁸⁵ detected MAP3K3 c.1323C>G mutation, predominately in type II or type III patients with CCM, defining a subclass of CCM. Choquet et al.¹⁴⁸ investigated common variants in inflammatory and immune response genes that influence the severity of familial CCM1 disease. They found that IL-4, CD14, IL-6R, and MSR1 were associated with intracerebral hemorrhage and lesion counts on MRI, while TLR4, CD14, IL-6R, and IGH were linked solely with lesion counts.

In animal models, Yang et al.¹⁴⁹ employed the Dre–Cre dual recombinase system to specifically delete CCM genes in brain ECs, allowing for quantification of lesion burden in mice over time and enabling longitudinal drug testing in live animals. Similarly, Fisher et al.¹⁵⁰ developed a chronic mouse model of CCM induced by postnatal ablation of CCM1 with Pdgfb–CreERT, discovering that MRI properties of lesions correlated with cellular markers for ECs, astrocytes, and microglia. These in vivo studies provide a foundation for understanding individual lesion characteristics and offer a comprehensive preclinical platform for testing new drugs and gene therapies for CCM control.

Artificial intelligence (AI) and machine learning (ML) algorithms are increasingly applied in imaging genomics to improve diagnostic accuracy and predict disease progression. These technologies can analyze vast datasets to identify subtle patterns and correlations that may not be apparent with conventional analyses, thus facilitating early detection and personalized treatment strategies. Saggi et al.¹⁵¹ used ML, including random forest models, gradient-boosted decision trees, and AdaBoost, to predict the risk of hemorrhage in pediatric patients with AVMs who presented with or without hemorrhage and found left-sided AVM, small AVM size, and concurrent arterial aneurysm as significant risk factors for hemorrhagic presentation. In adult patients with AVM, Oermann et al.¹⁵² pioneered the use of ML to predict outcomes after SRS, showing that ML models outperformed traditional scoring systems such as the Spetzler-Martin grading scale, radiosurgery-based AVM score, and Virginia Radiosurgery AVM Scale. Jiao et al.¹⁵³ applied AI to predict postsurgical motor defects in patients with AVMs involving motor-related areas. They found that FN_10mm/50  mm can reflect the lesion-fiber spatial relationship and be a key predictor of postsurgical motor defects in patients with brain AVMs. For CCM, existing ML models have primarily focused on predicting (re)bleeding events, leading to the development of various models.^{154, 155} Although radiological features have not yet been integrated with genomic data, AI-driven genomic imaging has shown the potential to enhance diagnostic accuracy and guide personalized treatment strategies.¹⁵³

4.2 Liquid biopsy

Liquid biopsy, a noninvasive technique involving blood or other body fluids analysis, offers a valuable tool for detecting and monitoring CVMs. Circulating biomarkers such as cell-free DNA (cfDNA), noncoding RNAs (ncRNAs), proteins, and exosomes provide valuable insights into the genetic and molecular status of CVMs. cfDNA analysis can detect specific somatic mutations associated with CVMs, enabling early diagnosis and monitoring of disease progression¹⁵⁶ (Table S1).

Zenner et al.¹⁵⁷ used multiplex ddPCR to detect cfDNA in the plasma of patients with AVM, identifying cfDNA variants in two of eight cases. To circumvent the need for biopsy samples, Serio et al.¹⁵⁶ performed molecular characterization of the peripheral blood of patients with AVM using cfDNA–NGS liquid biopsy and found KRAS mutations in 60% of patients with AVM. These findings suggest that cfDNA analysis could be a noninvasive diagnostic tool for early detection and disease monitoring in AVMs.

ncRNAs, including long ncRNAs (lncRNAs), microRNAs (miRNAs), small nucleolar RNAs (snoRNAs), and circular RNAs, are involved in cellular functions and homeostasis maintenance. However, they do not code for any functional proteins. These ncRNAs are thought to accurately reflect the human transcriptome and are stable and detectable in bodily fluids.^{158, 159} Li et al.¹⁶⁰ examined lncRNA profiles in four pairs of AVM tissues and the corresponding superior temporal arteries or scalp artery fragments, discovering that certain lncRNAs are associated with seizures in patients with AVM. RNA sequencing in 10 patients with CCM and four controls revealed 1967 differentially expressed lncRNAs, with SMIM25 and LBX2-AS1, found to correlate with more protein-coding genes and implicated in critical signaling pathways, including vascular signaling and essential biological processes relevant to CCM pathophysiology.¹⁶¹ Research on miRNAs in CVMs has progressed further, with Huang et al.¹⁶² reporting decreased levels of miR-137 and miR-195* in AVM samples, both of which act as vasculogenic suppressors by altering phenotypic properties of smooth muscle cells in AVMs. Chen et al.¹⁶³ identified miR-7-5p, miR-199a-5p, and miR-200b-3p as critical miRNAs involved in VEGF signaling in unruptured AVMs. Notably, miR-135b-5p was significantly upregulated in AVM and was associated with VEGF and HIF-1α, which were further upregulated in ECs derived from patients with AVM when treated with hypoxia or VEGF.¹⁶⁴

For CCMs, Kar et al.¹⁶⁵ performed small RNA sequencing on CCM specimens and identified five key miRNAs relevant to CCM pathology (let-7b-5p, miR-361-5p, miR-370-3p, miR-181a-2-3p, and miR-95-3p). Their miRNA–mRNA network analysis revealed functional relationships between the miRNAs and critical genes MIB1, HIF1A, PDCD10, TJP1, OCLN, HES1, MAPK1, VEGFA, EGFL7, NF1, and ENG, which are thought to be critical regulators of CCM pathology.¹⁶⁵ They also preliminarily examined the potential roles of snoRNAs in three CCM specimens, identifying 271 differentially expressed snoRNAs.¹⁶⁶ Following in situ hybridization validation, SNORD115-32 and SNORD114-22 were significantly decreased.

Exosomes are nanoscale vesicles secreted by cells containing diverse components, including DNA, ncRNA, proteins, metabolites, and lipids, and can stably exist in various body fluids. Li et al.¹⁶⁷ collected blood samples from 14 patients with AVM and 14 healthy volunteers, identifying 117 dysregulated exosomal lncRNAs and 1159 dysregulated exosomal mRNAs. They further conducted an exosomal lncRNA–miRNA–mRNA-related ceRNA regulatory network, identifying three significant modules involving 31 dysregulated exosomal lncRNAs, 114 dysregulated exosomal mRNAs, and KEGG pathways associated with Rap1, Ras, MAPK, and platelet activation. Another study, utilizing miRNA microarray analysis, found that miR-3131, secreted by KRAS-mutant ECs, can induce EndMT, with an increasing trend also observed in the peripheral blood of some patients with AVM.¹⁰⁵ Given that EndMT is associated with the bleeding phenotype, this study suggests that exosomes may help predict the bleeding risk of CVMs in a noninvasive manner.

Renedo et al.¹⁶⁸ analyzed blood samples from patients with AVM in the UK Biobank. They found that APOE ε4 status may identify patients with AVM at particularly high risk of intracerebral hemorrhage. Previous studies on the transcriptome, gut microbiota, and plasma proteome of CCM lesions have suggested that disease severity is linked to the brain–gut axis and proinflammatory responses. Srinath et al.¹⁶⁹ employed liquid chromatography–mass spectrometry to evaluate the plasma metabolomics of patients with CCM and those with symptomatic bleeding. They found that plasma metabolites, including cholic acid and hypoxanthine, could distinguish patients with CCM, while arachidonic and linoleic acids differentiated those with symptomatic hemorrhage. These molecules are known to contribute to maintaining blood–brain barrier integrity, downregulating apoptosis, and alleviating inflammatory damage following hypoxia.¹⁷⁰ These studies suggest a potential role for metabolites or lipids as biomarkers in CVMs, warranting further investigation.

4.3 Endovascular biopsy

An innovative approach to molecular profiling of CVMs without the need for open surgery was introduced by Winkler et al.¹⁷¹ through the use of endoluminal biopsy for the molecular profiling of human brain AVMs. They performed endoluminal biopsy and computational fluid dynamics (CFD) modeling in adults with unruptured AVMs during cerebral angiography. The endoluminal biopsies were successfully conducted in four patients without complications, revealing 106 differentially expressed genes that enriched pathogenic cascades, including RAS–MAPK signaling. The findings strongly correlated with genome-wide expression data from surgically excised tissues. CFD analysis correlated wall shear stress with inflammatory pathway upregulation, and comparisons of pre- and postembolization samples confirmed flow-mediated gene expression changes. This study illustrated that endoluminal biopsy enables molecular profiling of AVMs in living patients, identifying potentially targetable RAS–MAPK signaling abnormalities and facilitating precision medicine approaches.

These advances in biopsy techniques, encompassing cfDNA, ncRNAs, exosomes, and metabolites, highlight the potential of these technologies to revolutionize the diagnosis, monitoring, and treatment of CVMs. By providing noninvasive or minimally invasive means to profile genetic and molecular changes, these approaches facilitate early detection, dynamic monitoring, and the identification of targeted therapies. Although liquid biopsy offers a noninvasive method for identifying genetic mutations, it has inherent limitations, including the risk of misdiagnosis due to biomolecules from non-CVM sources, leading to potential false positives. Similarly, endovascular biopsy poses procedural risks such as vascular injury and inflammation, which can complicate treatment outcomes.

5.1 Targeted therapies

Targeted therapies have emerged as a promising treatment approach, given the identification of specific genetic mutations that drive CVMs. These therapies aim to inhibit molecular pathways activated by these mutations, thereby preventing disease progression and reducing the risk of complications. The drugs used to treat CVMs may target different aspects of CVM progression and rupture to provide therapeutic benefits. Current studies on AVMs are mainly based on in vitro proof-of-concept studies, with drug targets primarily focused on the KRAS/BRAF–MAPK, VEGF, and mural cell pathways.

Although KRAS mutations exist in many sporadic AVMs, the lack of direct KRAS inhibitors has led to the use of small-molecule MEK inhibitors such as U0126 and SL327. These inhibitors, originally used to treat cancer, have shown efficacy in alleviating the phenotype in vitro and KRAS^G12V zebrafish models.^{63, 97} In another in vitro study, researchers used human umbilical vein ECs engineered to overexpress the MAP2K1^K57N mutation rather than the KRAS mutation to study pathway effects. They found that trametinib reduced ERK pathway activation and reversed vascular network formation, revealing its potential therapeutic effects in AVMs.¹⁷² Furthermore, Park et al.⁹⁸ constructed KRAS^G12V/bEC mice and verified that trametinib effectively inhibited AVM growth and improved behavioral defects in this model.

In a BRAF^fl/V600E animal model, Tu et al.¹⁰⁰ initiated dabrafenib treatment on the first-day postinduction and continued the treatment for 6 weeks. They found that targeted inhibition of BRAF^V600E prevented AVM formation but did not rescue existing AVMs. Given that AVM lesions often develop fully before clinical treatment, dabrafenib may not be clinically effective. A prospective phase II open-label pilot clinical trial is underway in Canada (NCT06098872), involving participants preparing for AVM surgical resection without a control group. Enrolled patients will receive trametinib orally once daily for 60 days before surgery, with the primary outcome being AVM imaging volume after the last dose.

Another crucial drug target for AVMs is the VEGF signaling pathway. Elevated levels of VEGF have been validated in various studies from plasma samples to specimens, particularly in high-flow AVMs.¹⁷³ Using an ALK1^fl/fl mouse AVM model, Cheng et al.¹⁷⁴ found that elevated VEGF levels increased bleeding and mortality in mice. They further observed that the administration of bevacizumab, a humanized monoclonal anti-VEGF antibody, reversed abnormal vascular density and dysplasia in the mouse mode.¹⁷⁵ Seebauer et al.¹⁷³ treated three adult patients with sporadic extracranial AVMs off-label with bevacizumab, resulting in improved lesion deformity, although some side effects were noted.

Abnormalities in mural cells of AVMs are primarily related to hemodynamics, specifically manifested by decreased pericyte number and density. Pericyte coverage is reduced in ruptured AVMs, and a decreased pericyte number correlates with the severity of microbleeds and blood flow in the lesions.¹⁰⁴ In the ALK1^fl/fl AVM mouse model, blood vessels exhibited increased diameter and loss of surrounding pericytes.¹⁷⁶ Zhu et al.¹⁷⁷ found that thalidomide, an immunomodulatory drug, reduced hemorrhage in this model.

For CCM, drug targets primarily include MEKK3–KLF2/4, RhoA/ROCK, PI3K/AKT/mTOR, and superoxide dismutase mimetics (Table 4). Defective ECs lacking any of the three CCM proteins exhibit overactivation of MEKK3–KLF2/4, which plays a crucial role in disease pathology. Ponatinib inhibits MEKK3 activity and normalizes downstream KLF expression in animal models, thereby inhibiting the formation of new CCM lesions and reducing the growth of established lesions.¹⁷⁸ TSP1 and TGFβ/BMP, significant downstream targets of KLF2/4, are involved in CCM pathogenesis. Direct intervention with these targets can reverse the CCM phenotype both in vitro and in vivo.^{117, 121, 179}

The Rho/ROCK signaling pathway is abnormally activated when CCM proteins are inactivated, intersecting with the MEKK3–KLF2/4 signaling pathway, but appears to be downstream of MEKK3.¹¹⁶ Studies on mouse and human CCM lesions have shown that RhoA and its downstream effector ROCK1/2 are pathologically activated, leading to phosphorylation of myosin light chain (MLC) and MLC phosphatase and inhibition of MLC phosphatase catalytic activity.^180-182 Initially, researchers observed that Rho/ROCK inhibitors Y-27632 and C3 transferase reduced JNK activation, blocked the stress fiber response, and rescued endothelial barrier function in vitro.^{180, 183} Mice treated with BA-1049, a novel ROCK 2 selective inhibitor, exhibited a significant dose-dependent reduction in lesion volume and bleeding around the lesions.¹⁸⁴ In a CCM2 mice model pretreated with simvastatin, an HMG-CoA reductase inhibitor, the permeability response to VEGF was significantly reduced.¹⁸⁰ Shenkar et al.^185-187 further demonstrated that fasudil, high-dose atorvastatin, and simvastatin significantly reduced chronic bleeding in CCM lesions, with fasudil and high-dose atorvastatin being more effective than simvastatin in improving survival and slowing lesion development in vivo. PI3KCA overactivation is commonly observed in CCM lesions, leading to abnormal cell proliferation and symptomatic bleeding. Marchi et al.¹⁸⁸ utilized rapamycin and other mTORC1 inhibitors, such as torin1, and found that they could slow the progression of CCM lesions in vivo and in vitro.⁸³ Several studies have focused on the biological process of CCM-related angiogenesis, proposing promising therapeutic targets, including ERK, EphB4, and VEGFR2 inhibitors.^{120, 122, 189-191}

For noninvasive CCM treatments, some researchers have innovatively used data-driven, unbiased, small-molecule drug screening and ML platforms.^{192, 193} Gibson et al.¹⁹² conducted automated immunofluorescence and ML-based preliminary screening of the structural phenotype of CCM2-deficient human ECs and further screening based on changes in endothelial function and animal model outcomes. They ultimately screened 2100 known drugs and bioactive compounds, identifying vitamin D3 and tempol as candidate drugs.¹⁹² Otten et al.¹⁹³ applied 5268 substances to CCM-mutant worms, zebrafish, mice, and human ECs. Their findings were integrated with a whole transcriptome profile of zebrafish CCM2 mutants using a systems biology-based target prediction tool to reveal disease-related signaling pathways and potential targets for small-molecule-based therapeutics.

In clinical studies, research on effective drugs for CCM began with a series of case reports on propranolol.^{194, 195} Based on this, preclinical studies conducted by Oldenburg et al.¹⁹⁶ suggested that propranolol may help reduce and stabilize vascular lesions, positioning it as a potential medication for CCM. Mabray et al.¹⁹⁷ conducted a small prospective randomized controlled trial pilot study (NCT01764451) to evaluate the safety of simvastatin in patients with CCM and the feasibility of using imaging measures for efficacy assessment. However, the study was unable to draw reliable conclusions regarding its efficacy. A phase I/IIa placebo-controlled, double-blind, single-center clinical trial is currently underway (NCT0176445) to investigate the effectiveness of atorvastatin for treating symptomatic bleeding in CCM.¹⁹⁸ The treatment group will receive 80 mg of atorvastatin orally daily as the starting dose, while the control group will receive a placebo. CCMs showing symptomatic bleeding within the past year will be evaluated by MRI quantitative magnetic susceptibility mapping to determine whether atorvastatin produces a significant difference in lesional iron deposition compared with the placebo. However, the results of this study have not yet been published.

5.2 Immunotherapy

Identifying immune components in CVM pathogenesis has spurred interest in immunotherapy as a potential treatment modality. Studies have shown that immune cells, particularly macrophages, are crucial in the development and progression of CVMs.^{101, 199} Furthermore, immune cell infiltration correlates with lesion progression, suggesting that targeting these immune cells or modulating their activity may provide a novel approach to disease management.

Chen et al.²⁰⁰ found a significant increase in neutrophils and macrophages in AVM tissues, while T lymphocytes and B lymphocytes were infrequently observed. The study identified myeloperoxidase (MPO) levels as a potential indicator to evaluate neutrophil count; a higher neutrophil count in AVMs corresponded to elevated MPO and matrix metalloproteinase-9 (MMP-9) levels. Weinsheimer et al.²⁰¹ characterized the blood transcriptional profiles of patients with AVM, and found that 29 genes were differentially expressed between patients with unruptured AVM and controls, with 13 potentially predictive of AVM. Another study found that IL-6 levels in the blood and activated MMP-9 in the tissues of patients with ruptured AVM were significantly higher than those in the nonruptured and control groups, suggesting a potential role of inflammation in AVM pathogenesis.²⁰² Noshiro et al.²⁰³ examined 18 sporadic AVM specimens and found that the mRNA expression of IL-6 was also significantly higher than that in control tissues. Notably, macrophages and other inflammatory cells are frequently observed, even in AVMs without a history of rupture, indicating their intrinsic role in AVM pathogenesis rather than merely a response to hemorrhage.^{204, 205} Winkler et al.¹⁷¹ classified cerebrovascular-related immune cells at the single-cell level, identifying 17 cell subsets, including nine myeloid cell subsets, dendritic cells, glial cells, three perivascular macrophage subsets (pvMφ), and three monocyte (Mo) subsets. Myeloid immune cells constitute a significant proportion of AVMs, indicating immune activation during the disease process.¹⁷¹ In contrast, a single-cell resolution molecular atlas of human cerebral vasculature recently published in Nature showed that the proportion of myeloid cells in AVMs was lower than that in control tissues.¹⁰¹ These findings indicate that the components of the immune microenvironment in AVMs warrant further investigation at the single-cell level. Given that both ruptured and unruptured AVMs are associated with inflammatory processes, researchers have rationally targeted this process by using anti-inflammatory drugs in preclinical models. Treatment with the MMP-9 inhibitors minocycline and pyrrolidine dithiocarbamate reduced brain MMP-9 activity and attenuated VEGF-induced stroke in a mouse model.²⁰⁶ Similarly, doxycycline lowered MMP-9 levels in the same mouse model.²⁰⁷ The effects of doxycycline were also explored in patients with AVMs before surgical resection. However, due to the short duration of medication and the small patient sample, only a trend toward reduced MMP-9 levels was observed, which did not reach statistical significance.²⁰⁸ Antiangiogenic drugs, such as bevacizumab, have been shown to reduce vascular density and dysplasia in a mouse model of brain AVM.¹⁷⁵ Furthermore, thalidomide reduced the burden of CD68+ cells and the expression of inflammatory cytokines in AVM lesions, although potential side effects cannot be ignored.¹⁷⁷

Regarding the immune microenvironment of CCM, the Kahn laboratory at the University of Pennsylvania reported a novel and intriguing role of the gut–brain axis in CCM pathogenesis. In a CCM mouse model, lipopolysaccharides activated the TLR4 receptor on ECs and its downstream MEK5–ERK5–KLF2/4 pathway, promoting CCM development while knocking out the TLR4 receptor significantly reduced CCM lesion formation.¹²⁶ Given the critical role of TLR4 in CCM pathophysiology, TLR4 antagonists (LPS-RS) may serve as potential therapeutic agents for CCM. In a subsequent mouse model, the team increased CCM formation after chemically disrupting the intestinal barrier with sodium dextran sulfate, suggesting that the intestinal barrier is the primary determinant of CCM progression rather than the microbiome.¹²⁷ They further revealed the combined effects of dexamethasone on both brain ECs and intestinal epithelial cells, suggesting that dexamethasone could effectively inhibit CCM formation in mice. Previous studies have also reported that TLR4 and Fc receptors (FcR) can synergistically activate common downstream molecules, producing cumulative effects.²⁰⁹ TLR4 and FcR expression in CCM ECs is significantly higher than that in normal cerebral vascular ECs (Cao et al., unpublished data).

In CCM, mutant ECs can reduce intercellular junction protein expression and destroy the CCM blood–brain barrier. Mutant ECs can also recruit inflammatory and immune cells by overexpressing immune adhesion chemokines, promoting the CCM immune microenvironment. Mutated ECs lining blood vessels are in close proximity to immune cells in circulation, leading to an accumulation of myeloid cells, lymphocytes, and other immune cells around CCM lesions.^{123, 210-213} Shi et al.²¹⁴ suggested that plasma cells, differentiated from activated mature B cells, exist in CCMs, with IgG and complement membrane attack complexes co-localizing in CCM lesions. The IgG in CCM lesions amplifies in an oligoclonal pattern, suggesting that this amplification was antigen-specific.²¹⁵ Furthermore, Shi et al.²¹⁴ sequenced B cells obtained via laser microdissection of tissue specimens, revealing the clonal expansion of in situ B cells, which suggests specific antigen stimulation occurs in CCM. Zhang et al.²¹⁶ analyzed laser-captured microdissected plasma cells. They found that antibodies produced by plasma cells in CCM lesions typically target cytoplasmic and cytoskeletal autoantigens, including NMMHC IIA, vimentin, and tubulin.²¹⁶ Given that plasma cell-mediated immunity may play a role in CCM lesions, studies have examined the effects of B cell knockout on imaging findings on CCM lesions. Research has shown that nonheme iron deposition and ROCK activity are reduced in the lesions of B cell-depleted mice, suggesting the potential application of B cell inhibitors in CCM treatment.²¹⁷ Mechanisms such as neutrophil extracellular trapping and macrophage inflammatory infiltration have also been validated in clinical specimens, opening new avenues for research in this field.^{148, 218-221}

6.1 Future in imaging genomics

Imaging genomics is still in its early stages, with most studies focusing on identifying novel imaging markers. The future lies in developing big-data analytics and AI-driven methods. A key objective is establishing a link between genetic mutations and imaging features to better predict patient prognosis. Future research should also prioritize developing standardized protocols for imaging genomics to ensure consistency and reproducibility across research and clinical settings. Additionally, integrating AI and ML algorithms will enhance the diagnostic accuracy and predictive capabilities of imaging genomics, facilitating early detection and personalized treatment strategies.

6.2 Role of liquid or endovascular biopsy

The future of CVM treatment lies in precision medicine, which tailors therapeutic interventions to a patient's genetic and molecular profile. Advances in liquid biopsy will play a crucial role in realizing this goal. By identifying specific genetic mutations and molecular pathways involved in individual patient, clinicians can select targeted therapies that are effective, minimizing unnecessary treatments and optimizing outcomes.

In 2023, Fan et al.²²² conducted a nontargeted metabolite detection analysis of patient plasma, revealing significant changes in the metabolic profiles of patients with extracranial AVMs. This highlights metabolomics as a valuable diagnostic tool for CVMs. Similarly, lipidomics offers a new dimension to liquid biopsy approaches, providing insights into lipid alterations underlying CVMs. Future studies should explore integrating metabolomics, lipidomics, and other liquid biopsy approaches, such as circulating ncRNA and exosome analysis, to develop comprehensive biomarker panels for the early diagnosis and monitoring of CVMs.

6.3 Advancing targeted therapies

Recent advancements in targeted therapies have shown promise in treating CVMs. In patients with tumors, research on upstream inhibitors such as direct KRAS^G12C or KRAS^G12D inhibitors is emerging. Fraissenon et al.²²³ demonstrated the effectiveness of sotorasib, a specific KRAS^G12C inhibitor, in reducing vascular malformation volume and improving survival in two mouse models with mosaic KRAS^G12C. Additionally, two adult patients with KRAS^G12C-associated AVMs showed rapid symptom and lesion size reduction after sotorasib treatment. However, similar drugs targeting KRAS^G12D have yet to be developed and validated for brain CVMs. Future research should focus on developing and testing these inhibitors in clinical trials to evaluate their efficacy and safety.

6.4 Exploring the immune microenvironment

Understanding the role of the immune microenvironment in CVM pathogenesis has spurred interest in immunotherapy as a potential treatment modality. Tertiary lymphoid structures (TLSs), known for their role in rapid and efficient T cell-dependent B cell activation, have gained attention and have been reported to be ectopic around chronic inflammatory sites in diseases, such as those involving tumors, infections, autoimmune diseases, and organ transplantation.^{224, 225} TLSs promote B cell maturation, antibody subtype conversion, and somatic hypermutation. After clonal type screening and affinity maturation, B cell clones differentiate into plasma cells to produce specific IgGs targeting specific antigens.^{226, 227} However, this phenomenon has not been reported in CVMs.

Future research should investigate the presence and potential role of TLSs in CVMs and investigate targeting these structures for therapeutic purposes.

6.5 Addressing challenges and improving outcomes

Despite significant progress in understanding and treating CVMs, several challenges remain unresolved. The genetic and clinical heterogeneity of CVMs complicates the development of universally effective therapies. Additionally, the long-term safety and efficacy of novel treatments must be rigorously evaluated, particularly those targeting genetic and immune pathways. Ensuring new treatments are safe and effective across diverse patient populations is crucial for their successful implementation in clinical practice.

Collaboration among researchers, clinicians, and pharmaceutical companies is essential for translating these advancements into clinical practice. Large-scale multicenter clinical trials are necessary to validate the efficacy and safety of emerging therapies. Continued investment in innovative diagnostic and therapeutic approaches is critical to improving outcomes for patients with CVMs.