Molecular and multimodal biomarkers in Moyamoya disease: from pathogenic mechanisms to clinical translation

Abstract

Moyamoya disease (MMD) is a chronic, progressive cerebrovascular condition marked by narrowing or blockage of the terminal segments of the internal carotid arteries, resulting in ischemic and hemorrhagic strokes. Its pathogenesis involves a multifactorial interplay between genetic susceptibility, immune–inflammatory dysregulation, endothelial dysfunction, and aberrant vascular remodeling, influenced by non-genetic and environmental factors. Despite considerable research progress, clinically useful biomarkers remain limited, lacking sufficient sensitivity and specificity for predicting disease onset, progression, or treatment response. Current management relies primarily on surgical revascularization, which restores cerebral perfusion but does not address underlying biological mechanisms, while pharmacological interventions remain largely empirical and nonspecific. This review systematically searched PubMed and Web of Science up to September 2025 using combinations of “Moyamoya disease” and “biomarker” with “genomics,” “transcriptomics,” “proteomics,” “metabolomics,” “neuroimaging,” “artificial intelligence,” and “machine learning.” We summarize recent advances in genetic and molecular biomarker discovery, including the identification of RNF213 as a major susceptibility gene in East Asian populations, alongside emerging roles of variants in MTHFR, DIAPH1, and GUCY1A3. Beyond genomics, proteomic and metabolomic profiling have revealed dysregulation of vascular repair pathways, extracellular-matrix remodeling, and lipid–amino acid metabolism, offering new insights into disease heterogeneity and progression. Noncoding RNAs and exosome-derived biomarkers—such as plasma miR-512-3p, miR-328-3p, and miR-125b-5p—have shown potential as minimally invasive tools for diagnosis and monitoring, linking posttranscriptional regulation to vascular pathophysiology. Parallel advances in neuroimaging biomarkers, enhanced by artificial intelligence (AI), are enabling the integration of morphological and hemodynamic data with molecular findings. Deep learning-based models trained on digital subtraction angiography (DSA), computed tomography angiography (CTA), and retinal imaging have achieved diagnostic accuracies exceeding 90%, while multimodal integration approaches are beginning to correlate imaging phenotypes with genetic and metabolic profiles. Future research must transcend single-omics paradigms to establish integrative, multidimensional frameworks that connect genetic variation to cellular function, vascular remodeling, and clinical phenotype. Progress will depend on international multicenter collaboration, open-access biomarker databases, and the incorporation of explainable AI to bridge discovery and clinical application. Together, these developments may usher in biomarker-driven precision diagnosis and personalized therapy for MMD.

Introduction

Moyamoya disease (MMD) is an uncommon and progressive disorder of the cerebrovascular system characterized by narrowing or blockage at the ends of the internal carotid arteries, leading to the development of delicate collateral networks that create the distinctive "puff of smoke" appearance on angiograms [1]. Clinically, MMD is a significant yet puzzling cause of both ischemic and hemorrhagic strokes throughout a person's life. Its occurrence shows considerable variation depending on ethnicity and geography, with the highest rates found in East Asia (0.5–1.5 per 100,000 person-years) and much lower rates in North America and Europe (0.1 per 100,000) [2]. Despite extensive research over the years, the cause of MMD is still only partly comprehended.

A major breakthrough came in 2011 with the identification of Ring Finger Protein 213 (RNF213) as a susceptibility gene for MMD [1, 3–7]. RNF213 encodes a large AAA + ATPase with E3 ubiquitin ligase activity, implicated in angiogenesis, vascular stability, and lipid metabolism. Research has indicated that variants linked to disease hinder the movement of endothelial cells, their ability to form tubes, and their reaction to angiogenic signals like vascular endothelial growth factor (VEGF), which may contribute to inappropriate vascular remodeling [8–14]. Additional genetic loci—such as α-Smooth muscle actin encoding gene (ACTA2), guanylate cyclase 1 soluble subunit alpha 3 (GUCY1A3), and breast cancer susceptibility gene (BRCA)1/BRCA2-containing complex subunit 3 (BRCC3)—have been reported in familial or syndromic Moyamoya phenotypes [8–10, 15–17]. Nonetheless, the practical application of genetic testing is constrained by limited penetrance, inconsistent expressivity, and significant ethnic diversity [7].

Beyond genetics, converging evidence implicates immune-inflammatory dysregulation, endothelial dysfunction, and impaired vascular remodeling in MMD pathogenesis. Histopathological studies reveal intimal hyperplasia, disruption of the internal elastic lamina, and abnormal smooth-muscle cell proliferation, all of which are features suggestive of a maladaptive vascular repair response. Circulating cytokines—including VEGF and transforming growth factor beta (TGF-β)—have been associated with disease activity, raising the possibility that aberrant angiogenic signaling drives both stenosis and fragile collateral formation [12, 18–21]. More recently, systemic modulators such as the gut microbiome have been proposed as regulators of vascular homeostasis and host immunity, although their relationship with MMD remains largely unexplored [22].

Even with these understandings, significant obstacles continue to exist. Diagnosis still heavily depends on angiography, and there is insufficient agreement on standardized clinical or molecular guidelines. There is a scarcity of biomarkers that possess both high sensitivity and specificity, and the current tools cannot accurately forecast disease progression. Treatment is dominated by surgical revascularization, which improves cerebral hemodynamics but does not address underlying biology; pharmacological therapies remain empirical and nonspecific [7]. These gaps underscore the need for integrated mechanistic models linking genetic predisposition, vascular biology, and systemic influences.

In this review, we summarize recent progress in biomarker discovery for MMD, spanning genetic markers, immune-inflammatory mediators, angiogenic and vascular remodeling factors, neuroimaging signatures, and novel candidates, such as non-coding RNAs and exosome-derived biomarkers. We also highlight how advanced technologies—including artificial intelligence, high-resolution and functional neuroimaging, and integrated proteomics/metabolomics—are reshaping biomarker research. Finally, we discuss the translational potential, current limitations, and future perspectives of biomarker application in MMD.

Biomarkers for Moyamoya disease

Genetic susceptibility factors and MMD-associated genes

In recent years, growing evidence has implicated several candidate genes in the pathogenesis of MMD, with the p.R4810K variant of RNF213 identified as the predominant genetic susceptibility factor in East Asian populations being detected in approximately 95% of familial cases and 73% of sporadic cases, but is rare in non-Asian populations [3, 4, 23]. In this section, we summarize the potential genetic susceptibility factors associated with MMD (Table 1).

Table 1.

Summary of gene biomarkers investigated in Moyamoya disease

Biomarkers	Disease models	Functions	Application of biomarkers	References
RNF213(p.R4810K and other variants)	Cohorts of MMD patients across diverse populations (East Asian and European); brain microvascular endothelial cell models with siRNA-mediated interventions; knockout mouse and immune/radiation-induced models.	RNF213 encodes an E3 ubiquitin ligase that regulates angiogenesis, immune responses, and lipid metabolism. The p.R4810K variant is a major risk allele in East Asians, associated with early-onset disease, ischemia, and posterior cerebral artery (PCA) involvement. Loss of RNF213 activates the JAK2/STAT3 pathway, promoting pathological angiogenesis	RNF213 represents the strongest known genetic susceptibility factor for Moyamoya disease, particularly in East Asian populations.. The p.R4810K variant supports early genetic screening in East Asians. Its downregulation or pathogenic mutations indicate disease severity and progression. RNF213 also represents a target for combined vascular reconstruction and immunomodulatory therapies, with implications for Moyamoya syndrome (MMS)	[1, 5–7]
ACTA2	ACTA2-mutant patients; models of multisystem smooth-muscle dysfunction syndrome (MSDMS); ACTA2 transgenic animals	ACTA2 encodes α-actin, regulating vascular smooth-muscle cell (VSMC) contraction and vascular integrity. Mutations cause intimal hyperplasia, vascular straightening, and absence of “Moyamoya-like” collaterals	ACTA2 mutations cause cerebrovascular lesions radiologically similar to MMD but clinically distinct. The ACTA2 R179C mouse model recapitulates MMD-like remodeling and supports mechanistic and therapeutic studies. Genotype–phenotype correlations assist in subtype classification and genetic counseling	[24–28]
GUCY1A3	MMD patients, mouse models, zebrafish models	GUCY1A3 encodes the α1 subunit of soluble guanylate cyclase (sGC), the NO receptor involved in vascular relaxation and remodeling. Mutations impair the NO–sGC–cGMP pathway, leading to vascular dysfunction and stenosis	Serves as a genetic marker of NO pathway abnormalities, particularly in non-Asian patients. Combined with phenotypes such as achalasia or hypertension, it aids understanding of multisystem mechanisms. Represents a candidate for targeted therapy modulating the NO–sGC–cGMP axis	[9, 16, 29–31]
Methylenetetrahydrofolate reductase (MTHFR, C677T, A1298C)	MMD patients	MTHFR mutations impair folate metabolism, causing hyperhomocysteinemia (HHcy), endothelial dysfunction, and vascular stenosis	The C677T variant is linked to increased MMD susceptibility, poor postoperative collateralization, and higher hemorrhagic risk, particularly with elevated lipoprotein(a)	[32–37]
Methionine synthase reducatse (MTRR, A66G)	MMD patients	MTRR encodes methionine synthase reductase, crucial for homocysteine remethylation. The A66G variant reduces enzyme activity, increasing HHcy levels	While associated with impaired vascular remodeling, MTRR A66G shows no link to hemorrhagic risk, suggesting utility in combined folate metabolism risk evaluation rather than as an independent predictor	[34]
Transcobalamin II (TCN2)(rs117353193)	MMD patients with HHcy and hypertension	TCN2 encodes transcobalamin II, the plasma transporter of vitamin B12 regulating homocysteine metabolism. The rs117353193 variant impairs function, elevating Hcy	The A allele of rs117353193 increases MMD risk (OR = 1.87, P = 7.89 × 10⁻1⁰) and is enriched in HHcy patients. The AA + GA genotypes are protective against hypertension, suggesting dual metabolic and vascular roles	[32, 38]
BRCC3	Syndromic MMD (Xq28 deletion); SHAM syndrome; BRCC3 knockout zebrafish; Chinese cohorts	BRCC3 encodes a deubiquitinating enzyme regulating angiogenesis via the BRCA1/BRISC complex. Loss causes vascular stenosis, gliosis, and collagen deposition	Serves as a genetic marker for syndromic MMD, especially with developmental and cardiovascular anomalies. Though absent in Chinese patients, BRCC3 remains relevant for extended screening and therapeutic target exploration	[17, 28, 39–41]
Histone deacetylase 9 (HDAC9)	MMD patients	HDAC9 encodes a histone deacetylase regulating chromatin remodeling, VSMC proliferation, and endothelial function	SNPs such as rs2107595 are strongly associated with MMD (OR = 1.64, P = 1.49 × 10⁻2⁹), identifying HDAC9 as a novel susceptibility locus and screening biomarker	[38, 42]
KRT8	MMD patients; human brain microvascular endothelial cell models	KRT8 regulates immune–metabolic signaling through STAT3, IL-6, NF-κB, and miR-21-5p pathways, modulating cytoskeletal remodeling and angiogenesis	Upregulated in MMD (log₂FC > 1.5, P < 0.001; AUC > 0.96). Enhances endothelial tube formation and may be targeted by copper chelators or STAT3 inhibitors	[43]
Human leukocyte antigen (HLA) Class II(HLA-DQA, HLA-DRB)	MMD patients	HLA class II molecules regulate antigen presentation and CD4⁺ T-cell activation, influencing inflammatory vascular remodeling	Multiple population-specific associations (e.g., HLA-DRB113:02–DQB106:09 in Koreans; HLA-DRB1*04:10 in Japanese) indicate an immune–genetic mechanism underlying familial and early-onset MMD	[44–48]
VEGF	MMD patients; chronic ischemia and burr-hole/EPO rat models	VEGF promotes angiogenesis, endothelial proliferation, and vascular remodeling via VEGF–KDR signaling	The VEGF 2634C allele predicts favorable collateral formation; KDR polymorphisms (2604C, 1192A, 1719 T) increase pediatric risk. Combined VEGF–PDGF-BB or EPO-induced VEGF expression enhances revascularization outcomes	[12–14, 49]
Platelet-derived growth factor receptor beta (PDGFRB)	MMD patients	PDGFRB regulates SMC migration and proliferation under PDGF stimulation	Variants such as rs3828610 alter PDGF-BB responses, contributing to intimal thickening and occlusion; serve as genetic risk indicators for MMD	[19–21]
TGFB1	MMD patients	TGFB1 promotes VSMC proliferation and migration, contributing to vascular remodeling	Elevated TGF-β1 levels and genetic variants (e.g., rs1800471) associate with MMD onset and progression, marking TGFB1 as a key biomarker	[19–21]
DIAPH1	MMD patients	DIAPH1 encodes mDia1, interacting with RAGE to mediate oxidative stress and actin remodeling	Mutations associate with posterior circulation involvement and thrombocytopenia, aiding classification and prognostic assessment, particularly in RNF213-negative patients	[50, 51]
Disulfidptosis-related genes (DRGs)	MMD patients	DRGs modulate actin cytoskeleton remodeling and oxidative stress, driving endothelial proliferation and migration	Four hub DRGs (WDR27, OSBPL11, MSMO1, NEIL2) are downregulated in MMD serum (P < 0.05), yielding a diagnostic model with AUC > 0.96 for noninvasive detection	[52]
EndMT-related genes (ERGs)	MMD patients	ERGs (CCL21, CEBPA, KRT18, TNFRSF11A) regulate endothelial–mesenchymal transition (EndMT) and vascular inflammation	Serve as diagnostic and prognostic markers and potential therapeutic targets for MMD	[53]
Immune-related genes (IRGs)	MMD patients	IRGs regulate immune activation, neutrophil degranulation, and T-cell differentiation	Key IRGs (PTPN11, BTK, FGR, SYK, UNC13D, AZU1, CCL11, CXCL5, CCL15) show high diagnostic accuracy (ROC-based validation), implicating immune-cell infiltration in MMD pathology	[54, 55]
Angiogenesis-related genes (ARGs)	MMD patients	ARGs (TBC1D9B, ARAP3, PITPNB, UBE2E1) regulate endothelial function, lumen formation, and immune responses via RhoA/Arf signaling and ubiquitination	Abnormal ARG expression contributes to pathological angiogenesis and immune dysregulation; potential targets for precision therapy	[56]
OXPHOS-related genes	MMD patients	OXPHOS-related genes (CSK, NARS2, PTPN6, SMAD2) participate in oxidative phosphorylation, angiogenesis, and SMC phenotype switching	Potential early genetic markers and therapeutic targets in MMD	[57]
Sphingosine-1-phosphate receptor 1 (S1PR1)	MMD patients	S1PR1 regulates endothelial proliferation and permeability via PI3K/AKT signaling, promoting SMC proliferation and vascular remodeling	Serves as a genetic susceptibility factor and potential therapeutic target through modulation of the S1PR1 pathway	[58]

RNF213 polymorphisms and underlying molecular mechanisms in Moyamoya disease

The RNF213 gene represents the most robust susceptibility locus for MMD, displaying marked ethnic specificity among East Asian populations. Located on chromosome 17q25.3, RNF213 encodes a large (≈591 kDa) cytosolic protein with AAA + ATPase and E3 ubiquitin ligase domains that regulate stress response, lipid metabolism, and vascular homeostasis [3, 59].

The RNF213 p.R4810K variant was first identified in Japanese MMD cohorts through GWAS and has since been confirmed in 79–90% of Japanese and Korean patients and approximately 23% of Chinese patients, but remains rare in Europeans (< 0.0006) [8]. Despite its high frequency, only 1–2% of carriers develop MMD, highlighting incomplete penetrance and the likely involvement of secondary genetic or environmental triggers [8, 60]. Homozygous carriers develop MMD earlier and more severely, often with familial aggregation, indicating allele dose-dependent expressivity [28, 61, 62].

Functional studies reveal that RNF213 deficiency impairs endothelial proliferation and angiogenesis. Zebrafish knockdown models exhibit disorganized trunk vasculature, and RNF213-deficient mice display thinning of arterial walls and reduced collateral vessel formation [63]. Conversely, human endothelial cells lacking RNF213 show hyperangiogenic responses, whereas p.R4810K knock-in cells exhibit diminished angiogenic capacity and reduced ATPase activity, suggesting mutation-specific loss of function [64, 65].

Pathway analyses indicate that RNF213 regulates angiogenesis and vascular remodeling via the JAK2/STAT3 axis and caveolin-1 (Cav-1) signaling [18, 66]. RNF213 knockout increases endothelial permeability and cytokine secretion (GM-CSF, IL-6, IL-8), contributing to blood–brain-barrier (BBB) breakdown and perivascular inflammation [67, 68]. Clinically, elevated serum IL-6 and GM-CSF levels correlate with disease activity and surgical collateral formation, supporting their translational potential as inflammation-linked biomarkers [69].

Emerging evidence links RNF213 to innate immunity and lipid metabolism. The p.R4810K mutation disrupts lipid droplet turnover, enhances vascular inflammation, and alters NF-κB-dependent apoptotic signaling [70, 71]. Moreover, RNF213 acts as an E3 ligase sensor for bacterial LPS, promoting linear ubiquitin chain assembly and antibacterial autophagy [72]. Deficiency impairs Treg induction and increases susceptibility to HSV-1 and Listeria infection [73]. Collectively, these findings identify RNF213 as a pleiotropic regulator of vascular stability, angiogenesis, and immune homeostasis, implicating vascular–immune crosstalk in MMD pathogenesis.

Other candidate genetic markers

Beyond RNF213, a growing number of rare variants in vascular smooth muscle, nitric oxide signaling, and angiogenesis-related genes have been implicated in MMD or MMD-like vasculopathies.

The ACTA2 gene encodes smooth muscle α-actin, a cytoskeletal protein specifically expressed in vascular smooth-muscle cells (VSMCs) but absent in endothelial cells [26]. Early genetic studies, including linkage and genome-wide association analyses, indicated that ACTA2 mutations may predispose individuals to a spectrum of vascular disorders, including MMD [15]. However, the contribution of ACTA2 to MMD appears limited and inconsistent across populations. ACTA2 variants have been reported primarily in rare familial cases with Moyamoya-like arteriopathy and multisystem smooth-muscle dysfunction, and appear to be uncommon in unselected Moyamoya disease cohorts. Targeted next-generation sequencing in 255 Chinese MMD patients identified no pathogenic ACTA2 variants [28], and exonic sequencing in 53 Japanese and 55 Chinese patients similarly yielded no mutations [74, 75].Only a single sporadic case has been reported in Central Europe [76], suggesting that ACTA2 plays a minor role in MMD pathogenesis, although larger cross-ethnic studies are needed. Mutations at the R179 locus of ACTA2 cause smooth-muscle dysfunction syndrome (SMDS), a childhood-onset vasculopathy with MMD-like arterial occlusions. Patients with R179 mutations exhibit dilated proximal internal carotid arteries, straightened intracranial vessels, terminal ICA stenosis, and absent basal Moyamoya collaterals. In a study of 13 heterozygous R179 mutation carriers, congenital mydriasis and persistent ductus arteriosus were universal, while cerebrovascular imaging showed characteristic arteriopathy without typical MMD collaterals [77]. Murine models carrying the Acta2^R179C/+ genotype revealed immature, hypermigratory smooth-muscle cells with increased glycolytic flux. Following carotid injury, these mice developed Moyamoya-like lesions and early mortality, which were prevented by nicotinamide riboside (NR) treatment, suggesting that restoring oxidative phosphorylation can mitigate occlusive lesions [26]. In broader clinical cohorts, ACTA2 variants are rare in sporadic MMD. In a central European study of 39 non-familial MMD patients, only one heterozygous R179H variant was detected, with no significant enrichment of ACTA2 mutations overall [76]. Therefore, while ACTA2 mutations have limited relevance in typical MMD, they should be considered in differential diagnosis when MMD-like imaging features co-occur with systemic smooth-muscle vasculopathy, such as aortic disease or congenital mydriasis. Mechanistically, the murine studies highlight mitochondrial and smooth-muscle differentiation pathways, particularly oxidative phosphorylation, as potential therapeutic targets.

The GUCY1A3 gene encodes the α1 subunit of soluble guanylate cyclase (sGC), the primary receptor for nitric oxide (NO). Activation of sGC by NO generates cyclic guanosine monophosphate (cGMP), which stimulates protein kinase G signaling to mediate vascular relaxation and remodeling [16, 29, 30, 78]. Mutations in GUCY1A3 have been associated with early-onset hypertension, MMD, and achalasia [9, 16, 30, 78]. In a cohort of 96 non-familial MMD patients, predominantly of European ancestry, two individuals (~ 2.1%) carried compound heterozygous GUCY1A3 mutations linked to early onset hypertension, achalasia, and MMD [16]. Functional assays of the Cys517Tyr variant revealed markedly reduced NO-stimulated sGC activity, confirming a loss-of-function effect. In contrast, sequencing of 255 Chinese MMD patients showed no significant enrichment of GUCY1A3 variants relative to controls, highlighting potential ethnic differences in genetic susceptibility [28]. Functional studies in zebrafish, murine models, and cultured endothelial and smooth-muscle cells demonstrate that GUCY1A3 mutations impair angiogenesis, partly by disrupting the HIF-1α/VEGFA signaling pathway, a key mechanism contributing to vascular occlusion in MMD [31]. GUCY1A3 mutations have been identified in a limited number of families with Moyamoya disease, while their contribution to sporadic cases remains uncertain. Recent research in consanguineous Moyamoya angiopathy probands identified homozygous variants in both GUCY1A3 and NOS3, associated with severe posterior cerebral artery involvement, underscoring the critical role of the NO–sGC pathway in disease pathogenesis [79]. Overall, GUCY1A3 and the broader NO–sGC–cGMP pathway represent a relevant genetic marker for MMD, particularly in non-East Asian populations. Its detection may inform functional assessment and suggests potential targeted interventions, including NO-mimetic or sGC-activating therapies, to restore vascular homeostasis.

The BRCC3 gene encodes an X-linked K63-specific deubiquitinase within the BRCA1/BRCA2-containing complex, contributing to DNA damage repair, cell cycle regulation, and inflammatory signaling. Loss of BRCC3 function leads to angiogenic defects, as evidenced by impaired vascular development in zebrafish knockout models [17, 39]. Clinically, Xq28 microdeletions encompassing BRCC3 are associated with a syndromic form of MMD co-occurring with hemophilia A, termed SHAM syndrome [17, 39–41, 80]. In affected males, additional features can include short stature, hypogonadism, and cardiac anomalies. Zebrafish brcc3 morphants display defective angiogenesis that is rescued by endothelium-specific BRCC3 expression, highlighting a mechanistic link to vascular development [39]. A recent case report described a 23-year-old male with a ~ 26 kb Xq28 deletion including BRCC3, developmental delay, and reduced fibroblast expression, illustrating the clinical relevance of BRCC3 loss [17]. BRCC3 deficiency should be considered in syndromic MMD, particularly in male patients with developmental or endocrine anomalies, and family screening may reveal additional organ system risks. While therapeutic strategies are lacking, the deubiquitinase–angiogenesis connection provides a mechanistic rationale for future interventions.

The human leukocyte antigen (HLA) region has emerged as an important immune-genetic contributor to MMD susceptibility across diverse populations. Multiple associations—including HLA-DRB113:02–DQB106:09 in Koreans, HLA-DRB1*04:10 in Japanese, and HLA-DQA2 in Han Chinese—suggest that immune-mediated mechanisms may influence disease onset and progression. The major histocompatibility complex (MHC), located on chromosome 6p21 (~ 3.6 Mb), is divided into three classes: MHC-I (HLA-A, HLA-B, HLA-C), MHC-II (HLA-DR, HLA-DQ, HLA-DP), and MHC-III (which contains 55 coding genes and 5 pseudogenes) [81, 82]. These molecules regulate adaptive immune responses by mediating antigen processing and presentation, thereby influencing CD4⁺ T-cell activation and downstream inflammatory cascades [81]. Population-based studies have reinforced the immunogenetic hypothesis. In a Korean familial MMD cohort, Hong et al. (2009) identified markedly increased frequencies of HLA-DRB113:02 and DQB106:09 alleles (odds ratio > 12) [44]. Similarly, Kraemer et al. (2012) reported that HLA-DRB103 and HLA-DRB113 alleles conferred elevated MMD risk in German patients [47]. A Japanese genotyping study of 136 cases found HLA-DRB104:10 as a significant risk allele, with the haplotype HLA-DRB104:10–HLA-DQB1*04:02 further increasing susceptibility [45]. More recently, polymorphisms within the HLA region—particularly involving HLA-DQA2 and HLA-B—were confirmed as significant susceptibility factors in Han Chinese populations [46]. Collectively, these findings implicate the MHC-II-dependent immune system in MMD pathogenesis, possibly via aberrant endothelial immune activation, vascular inflammation, or maladaptive immune repair responses. Although large-scale functional studies remain limited, the consistency of these associations across ethnicities underscores a shared immunogenetic architecture that may stratify patients with familial or early onset MMD and inform precision risk assessment.

The vascular endothelial growth factor (VEGF) pathway plays a pivotal role in angiogenesis and collateral vessel formation in MMD. The VEGF gene encodes a key regulator of vascular growth that binds to its major receptor, VEGFR-2/KDR, on endothelial cells to promote proliferation, migration, extracellular-matrix remodeling, and increased vascular permeability—processes essential for neovascularization in ischemic cerebrovascular disease [11, 83–85]. Genetic studies have identified several variants in VEGF and its receptor KDR that influence MMD susceptibility and vascular outcomes. Specifically, KDR variants (2604C, 1192A, and 1719 T) are associated with increased risk of pediatric MMD, suggesting a role in developmental angiogenic dysregulation. Similarly, the VEGF -634C allele has been linked to enhanced collateral vessel formation and favorable postoperative angiogenesis, whereas the VEGF-634G allele correlates with poor collateralization in pediatric MMD, underscoring the prognostic value of VEGF genotyping [12]. At the protein level, MMD patients exhibit significantly reduced plasma concentrations of soluble VEGF receptors sVEGFR-1 and sVEGFR-2, endogenous inhibitors of VEGF signaling. Lower levels of these antagonists predict improved collateral formation after surgical revascularization, suggesting that diminished inhibitory signaling enhances compensatory angiogenesis [13]. Taken together, these findings highlight the VEGF–KDR axis as a crucial determinant of vascular remodeling capacity in MMD. Genetic and biochemical markers within this pathway may serve as prognostic indicators for collateralization outcomes and potential therapeutic targets to augment angiogenesis in patients with poor vascular reserve.

Several additional genetic variants have been identified in DIAPH1, PDGFRB, TGFB1, MTHFR, and MTRR, implicating diverse biological pathways involved in vascular contraction, endothelial and smooth-muscle cell (SMC) signaling, cytoskeletal organization, and extracellular-matrix remodeling. Although these mutations occur at low frequencies, they are increasingly recognized as modifier genes that contribute to the clinical and phenotypic heterogeneity observed in MMD [3]. For instance, rare damaging DIAPH1 variants were found in non-East Asian MMD patients via whole-exome sequencing, supporting a role for impaired actin remodeling in disease pathogenesis [51]. A study in Asian populations showed that DIAPH1 mutations might predispose to posterior circulation involvement, even if not a major risk gene overall [50]. Polymorphisms in PDGFRB and TGFB1 have been explored in association studies. In a European cohort, Roder et al. reported associations of TGFB1 and PDGFRB SNPs with MMD [9]. A Chinese study of 96 MMD patients and controls evaluated PDGFRB rs3828610 (and other loci) but did not find a statistically significant association individually, though interactions among loci were considered [21]. Variants in MTHFR and MTRR, which regulate homocysteine metabolism, have been proposed as modifiers, potentially exacerbating endothelial injury or thrombosis in vulnerable individuals. Although evidence in MMD is less direct, alterations in homocysteine metabolism are known contributors to vascular disease [86]. Although individually these modifier genes exhibit low penetrance and limited predictive power, they provide valuable mechanistic insight into convergent pathways underlying MMD. Integrating them into multi-gene or polygenic risk models may help improve predictions of disease severity, vascular phenotype, or treatment response in precision-medicine efforts.

Inflammatory and immunological biomarkers

Although genetic susceptibility forms the foundation of MMD pathogenesis, the penetrance of the RNF213 p.R4810K mutation remains extremely low, with only ~ 0.5% of heterozygous carriers developing the disease [86]. This observation supports the “second-hit” hypothesis, in which environmental or systemic immune triggers precipitate disease onset in genetically predisposed individuals [87]. A growing body of evidence indicates that immune and inflammatory mechanisms play essential roles in vascular injury, smooth-muscle proliferation, and pathological collateral formation. Consequently, a range of circulating and tissue-derived inflammatory and immunological markers (Table 2) have been investigated in MMD, providing insights into disease-associated immune-inflammatory processes.

Table 2.

Summary of inflammatory and immunological biomarkers investigated in Moyamoya disease

Biomarkers	Disease models	Functions	Application of biomarkers	References
Autoantibodies (AutoAbs)	MMD patients serum samples	Specific autoantibodies (e.g., CAMK2A, CD79A, EFNA3) indicate immune dysregulation and endothelial injury, reflecting humoral immune activation in MMD	Serve as potential diagnostic biomarkers, aiding early detection and providing insights into immune-mediated vascular pathology	[88]
Immune cells	MMD patient vascular and peripheral samples	Infiltration of macrophages, eosinophils, and T cells within cerebral vasculature indicates chronic inflammation contributing to vascular stenosis and remodeling	Patterns of immune-cell infiltration may act as immunological biomarkers for early diagnosis and for monitoring responses to immunomodulatory therapy	[54, 89, 90]
Treg cells	Peripheral blood from MMD patients	Tregs maintain immune tolerance and suppress excessive inflammatory responses. Increased counts but impaired suppressive function suggest immune imbalance in MMD	Treg frequency and function may serve as biomarkers of disease activity and indicators for evaluating immunotherapy efficacy	[91]
TGF-β	Serum of MMD patients	Key regulator of cell growth, differentiation, and extracellular-matrix remodeling; exerts anti-inflammatory and pro-angiogenic effects. Elevated levels correlate with collateral vessel formation	Acts as a biomarker for angiogenesis and vascular remodeling, useful in assessing disease progression and prognosis	[91–93]
IL-17	Serum and CSF from MMD patients	Pro-inflammatory cytokine produced by Th17 cells; promotes endothelial activation and pathological angiogenesis	Serves as a marker of inflammatory activity and aberrant angiogenesis, supporting individualized anti-inflammatory treatment strategies	[91]
Interferons(IFNs)	Serum and endothelial cell models	IFNs (particularly IFN-β and IFN-γ) regulate angiogenesis and immune responses; upregulate RNF213 expression and inhibit endothelial proliferation	Function as biomarkers for angiogenesis inhibition and immune imbalance, helping to evaluate inflammatory contribution to MMD	[65, 91]
Systemic Immune-Inflammation Index (SII)	Hematological data from MMD patients	Composite index integrating platelet, neutrophil, and lymphocyte counts to reflect systemic immune–inflammatory status	A prognostic biomarker of clinical outcomes; elevated SII values predict poorer prognosis and higher ischemic-stroke risk	[94, 95]

Immune cell-related alterations

Histopathological and immunohistochemical analyses have consistently demonstrated immune-cell infiltration in affected cerebral arteries of MMD patients. Early studies revealed proliferative smooth-muscle cell accumulation accompanied by macrophages, eosinophils, and T lymphocytes within the thickened intima and media, together with immune-complex deposits, such as IgG and S100A4 [89, 90, 96, 97]. More recent work using flow cytometry and single-cell approaches confirmed enrichment of neutrophils, monocytes, and natural killer cells in patient samples [55]. Among adaptive immune populations, regulatory T cells (Tregs) are of particular interest. Weng et al. (2017) reported increased frequencies of circulating Tregs and Th17 cells, accompanied by elevated cytokines including TGF-β, IL-10, IL-17, IL-6, and IL-23, reflecting dysregulated immune tolerance and inflammation [91]. Functionally, Treg impairment may exacerbate vascular injury, further supporting the involvement of adaptive immune dysregulation in MMD pathophysiology [91].

Autoantibody profiling further supports immune involvement. Earlier studies identified elevated anti-thyroid, anti-α-fodrin, and anti-cardiolipin antibodies [98–101]. Sigdel et al. (2013) later conducted a large-scale proteomic screening, discovering six autoantibodies—anti-CAMK2A, CD79A, GPS1, EFNA3, NKAIN4, and TMEM32—significantly upregulated in MMD serum. These autoantibodies may provide mechanistic into immune-mediated endothelial and smooth-muscle immune injury [88].

Systemic inflammatory burden can also be quantified through hematological indices. The systemic immune-inflammation index (SII), which integrates platelet, neutrophil, and lymphocyte counts, has recently been associated with clinical outcomes in MMD [94, 95]. Liu et al. (2023) and Wu, Huang et al. (2025) demonstrated that elevated SII values correlate with higher ischemic-stroke risk and poorer prognosis, suggesting that systemic inflammatory burden may be linked to disease severity. Together, these immune-related markers—cellular, humoral, and systemic—highlight the central contribution of immune dysregulation to MMD pathophysiology [94, 95].

Inflammatory factors in Moyamoya disease pathogenesis and progression

A broad spectrum of inflammatory cytokines contributes to vascular remodeling and chronic cerebrovascular inflammation in MMD. Increased serum and cerebrospinal fluid levels of pro-inflammatory cytokines, including IL-1β, IL-6, IL-23, TNF-α, and TGF-β, have been consistently reported across multiple studies [91, 102–104]. In addition to cytokines, commonly measured inflammatory markers have also been examined in Moyamoya disease. A case–control study reported elevated serum C-reactive protein (CRP) levels in patients with Moyamoya disease, with CRP correlating with circulating and cerebrospinal fluid inflammatory mediators, suggesting a nonspecific systemic inflammatory response rather than disease-specific pathology [102]. RNF213 mutations amplify endothelial susceptibility to inflammatory stress, leading to increased secretion of GM-CSF, IL-6, and IL-8, disruption of the blood–brain barrier (BBB), and recruitment of peripheral immune cells [67]. Mechanistically, pro-inflammatory cytokines, such as IL-1β and TNF-α, activate multiple signaling pathways that promote endothelial dysfunction, pathological angiogenesis, and blood–brain-barrier disruption [105, 106].

Cytokines with dual immunomodulatory and angiogenic functions, such as TGF-β and IL-17, appear particularly relevant. Cytokines with dual immunomodulatory and angiogenic functions, including TGF-β, IL-17, and interferons, have been implicated in extracellular-matrix remodeling, collateral vessel formation, and regulation of endothelial proliferation in MMD, linking immune signaling to angiogenic imbalance [65, 91–93].

Recent proteomic and immunoassay data have identified additional inflammation-associated molecules, including chemokine CXCL5 and soluble CD163, both markedly elevated in MMD plasma [107]. CXCL5, commonly linked to autoimmune disease severity, may mediate leukocyte recruitment, while CD163, a marker of M2 macrophage activation, peaks during Suzuki stage III—when Moyamoya vessels are most abundant—and contributes to angiogenesis and vessel stabilization [107, 108]. Collectively, these findings underscore the pivotal role of inflammatory and immunoregulatory pathways in MMD, offering a conceptual framework for future studies exploring immune-modulating therapeutic approaches.

Angiogenic and vascular remodeling biomarkers

MMD is characterized by progressive stenosis or occlusion of the intracranial arteries accompanied by the compensatory formation of an abnormal vascular network, known as Moyamoya vessels. This complex process is orchestrated by multiple molecular regulators that modulate angiogenesis and vascular remodeling. Among them, vascular endothelial growth factor (VEGF) and matrix metalloproteinase-9 (MMP-9) form the central regulatory axis, driving endothelial proliferation, matrix degradation, and neovessel formation. Dysregulated endothelial function is now recognized as a hallmark of MMD pathogenesis, underscoring the clinical importance of endothelial-related biomarkers for disease monitoring and therapeutic decision-making (Table 3).

Table 3.

Summary of angiogenic and vascular remodeling biomarkers for Moyamoya disease

Biomarkers	Disease models	Functions	Application of biomarkers	References
VEGF	MMD patients; chronic cerebral ischemia mouse model	Promotes endothelial proliferation, migration, and vascular permeability, driving pathological angiogenesis	1. Serum VEGF levels are positively correlated with collateral circulation improvement 2. Levels are higher in ischemic-type than in hemorrhagic-type MMD	[11–14]
MMP-9	MMD patients	Degrades extracellular-matrix (ECM) components to mediate vascular remodeling; overexpression compromises vessel wall stability	Serum MMP-9 > 1011 ng/mL is an independent risk factor for hemorrhagic stroke and correlates with Suzuki staging	[109–111]
EPCs (CD34 + CXCR4 +)	MMD patients	Recruited via the SDF-1α/CXCR4 axis to ischemic brain regions to facilitate vascular regeneration	In pediatric MMD patients, the number of peripheral blood CD34⁺ EPCs is higher than in adults and positively correlates with the density of Moyamoya vessels; it can serve as a predictor of EDAS treatment efficacy	[14, 112, 113]
eNOS polymorphisms	MMD patients	Reduced eNOS activity decreases NO synthesis, contributing to elevated vascular tone	Polymorphisms in the eNOS gene influence disease susceptibility	[114, 115]
​Caveolin-1	MMD patients	Ubiquitination and phosphorylation of RNF213 lead to Cav-1 dysfunction, relieving its inhibitory effect on eNOS	Low serum levels of Cav-1 are associated with adverse vascular remodeling	[116, 117]
Ang-2	MMD patients	Antagonizes Tie-2 signaling, loosens endothelial junctions	1. When plasma Ang-2 concentrations exceed 1162 pg/mL, it can identify patients in the rapid progression phase of MM 2. Elevated levels are significantly positively correlated with the extent of BBB disruption	[118]
sTie-2	MMD patients	An endogenous negative regulator of Ang/Tie-2 signaling	Lower plasma sTie-2 predicts better postoperative collateral formation; CSF–plasma gradients expand during active disease	[119]
PDGF-BB	MMD patients; chronic cerebral ischemia mouse model	Guides smooth-muscle cell (SMC) migration toward nascent vessels	1. Plasma levels are significantly positively correlated with collateral vessel maturity 2. Combined treatment with VEGF further enhances capillary maturation	[14, 104]
sCD163	MMD patients	Marker of M2 macrophage activation	1. Plasma levels are significantly positively correlated with IL-8 levels 2. The perivascular infiltration density of CD163⁺ macrophages is positively associated with the degree of vascular malformation	[107]

VEGF, a key pro-angiogenic cytokine, promotes endothelial cell proliferation, migration, and permeability, thereby facilitating both physiological and pathological angiogenesis. Sakamoto et al. detected markedly elevated VEGF expression in the dura mater and collateral vessels of MMD patients, highlighting its pivotal role in collateral formation [11]. Genetic studies revealed that the VEGF − 634G allele is associated with favorable postoperative collateral formation in pediatric ischemic-type MMD, whereas the CC genotype exerts opposite effects across age groups [12]. These findings suggest that VEGF polymorphisms may serve as predictive biomarkers for surgical outcomes and individualized therapy. Mechanistic studies further demonstrate that RNF213 silencing enhances TGF-β1 signaling, indirectly modulating VEGF-mediated angiogenesis [113]. Activation of inflammatory cytokines and the TGF-β/VEGF axis contributes to a self-reinforcing “gene–inflammation–angiogenesis” network [120]. Clinical follow-up data indicate that patients with low soluble VEGF receptor (sVEGFR-1 < 200 pg/mL; sVEGFR-2 < 1800 pg/mL) achieve an 85% improvement in collateral scores post-surgery [121]. Consistent with the dual nature of VEGF, serum levels are higher in ischemic-type than in hemorrhagic-type MMD, reflecting an imbalance between compensatory and pathological angiogenesis [14, 121].

MMP-9 plays an essential role in extracellular-matrix (ECM) degradation and vascular remodeling but contributes to vessel wall instability when overexpressed. Elevated serum MMP-9 has been linked to hemorrhagic events and disease progression in MMD. Fujimura et al. observed that MMP-9 peaks in Suzuki stage III, corresponding to the phase of most active angiogenesis [111]. Further, Lu et al. demonstrated that serum MMP-9 > 1011 ng/mL constitutes an independent risk factor for hemorrhagic stroke, particularly in adults [110]. Collectively, these studies suggest that serial monitoring of MMP-9 may help identify patients at risk of hemorrhage and serve as a dynamic indicator of vascular remodeling activity [109].

Endothelial progenitor cells (EPCs) are another crucial component of vascular repair. Elevated levels of CD34⁺CXCR4⁺ EPCs in the peripheral blood of MMD patients indicate an active compensatory angiogenic response. Ni et al. found that EPCs are recruited to ischemic regions via the SDF-1α/CXCR4 axis, promoting neovascularization [112]. Pediatric patients exhibit particularly high EPC counts, which further increase after revascularization surgery, correlating with improved collateral formation [122, 123]. These findings support the potential of EPC quantification as a biomarker for postoperative angiogenic efficacy, particularly in younger or severely ischemic individuals.

Nitric oxide (NO) is a critical mediator of vascular tone and endothelial health. Genetic studies have identified eNOS gene polymorphisms (− 786 T > C, intron 4b/a) as significant determinants of MMD susceptibility, particularly in children [114, 115]. The NO pathway is further modulated by Caveolin-1 (Cav-1), a negative regulator of eNOS. RNF213-mediated Cav-1 dysfunction—through multi-site ubiquitination and Tyr14 phosphorylation—relieves its inhibition of eNOS activity, resulting in endothelial instability [116, 117]. Clinically, reduced serum Cav-1 levels are associated with adverse vascular remodeling, suggesting that Cav-1 and eNOS variants jointly influence endothelial reactivity and may serve as biomarkers for disease severity.

Emerging angiogenic regulators, such as Angiopoietin-2 (Ang-2), soluble Tie-2 (sTie-2), platelet-derived growth factor-BB (PDGF-BB), and soluble CD163 (sCD163), have also gained attention. Ang-2 acts as a Tie-2 antagonist, disrupting endothelial junctions and pericyte attachment. Gorla et al. reported that plasma Ang-2 > 1162 pg/mL identifies patients in a rapid progression phase, while levels > 20 ng/mL predict cerebrovascular events and neurological deterioration [118]. Conversely, sTie-2 functions as a negative regulator of Ang/Tie-2 signaling; lower plasma levels predict favorable postoperative collateral formation, whereas higher cerebrospinal fluid–plasma gradients indicate active disease [119]. PDGF-BB guides smooth-muscle cell migration and vessel maturation; elevated plasma levels correlate with collateral vessel stability and act synergistically with VEGF to enhance capillary maturation [14, 104]. Finally, sCD163, a marker of M2 macrophage activation, reflects the angiogenic immune microenvironment. Plasma sCD163 levels are significantly correlated with IL-8 and perivascular macrophage infiltration, paralleling the severity of vascular malformation [107].

Collectively, these findings reveal that angiogenic and vascular remodeling biomarkers—spanning endothelial, inflammatory, and genetic axes—offer critical insights into MMD pathophysiology. They not only aid in distinguishing disease subtypes and predicting surgical outcomes but also highlight novel therapeutic targets for modulating aberrant angiogenesis and restoring vascular integrity.

Imaging biomarkers

Limitations of conventional imaging techniques and advances in novel imaging biomarkers

Although digital subtraction angiography (DSA) remains the clinical gold standard for MMD diagnosis, its invasive nature and reliance on iodinated contrast agents limit widespread use, carrying risks of allergic reactions and nephrotoxicity [124, 125]. Furthermore, DSA and conventional MRA are suboptimal for detecting subtle hemodynamic changes or evaluating postoperative collateral formation, particularly in the presence of metallic implants that introduce susceptibility artifacts [124–127]. These limitations have driven the development of noninvasive, high-resolution alternatives. High-resolution vessel wall MRI (VW-MRI) can clearly visualize concentric thickening of intracranial arteries and lumen narrowing, distinguishing MMD from atherosclerotic stenosis characterized by eccentric plaques (Table 4) [127, 128]. Quantitative criteria, such as a middle cerebral artery outer diameter ≤ 1.77 mm and an intravascular enhancement score (IVES) ≥ 2, improve diagnostic specificity and help stratify patients suitable for revascularization versus conservative management [128, 129].

Table 4.

Summary of neuroimaging biomarkers for Moyamoya disease

Biomarkers	Disease models	Functions	Application of biomarkers	References
MRI
HR-VW MRI	MMD patients	HR-VW MRI specifically differentiates MMD (concentric thickening) from Atherosclerotic Moyamoya syndrome (AS-MMS) (eccentric plaques)	1. Middle cerebral artery outer diameter ≤ 1.77 mm discriminates MMD (AUC = 0.912) 2. IVES score ≥ 2 predicts infarction risk	[124, 128, 129]
Arterial spin labeling-based dynamic magnetic resonance angiography (ASL-MRDSA)	MMD patients	Noninvasive dynamic blood flow assessment as an alternative to DSA	Short post-labeling delay (PLD) (1525 ms) quantifies CBF, while long PLD (2525 ms) quantifies arterial transit time (ATT)	[125, 130]
Pseudo-continuous ASL (PCASL)	MMD patients	Enhances signal stability by reducing labeling slab thickness	Improves visualization of distal Moyamoya vessels and evaluates collateral networks	[131]
7 T TOF-MRA	MMD patients	7 T TOF-MRA enhances the resolution of Moyamoya vessels and microaneurysms in the basal ganglia	Provides a more precise differentiation of hemorrhagic risk compared with 3 T-MRA and DSA	[132]
CVRDELAY (Blood oxygen level–dependent-magnetic resonance imaging [BOLD-MRI])	MMD patients	Assessment of cerebral perfusion reserve (using hypercapnic stimulation)	Prolonged delay serves as a warning for ischemic events; preoperative VW-CE signs predict a 67.3% risk of postoperative stroke	[43, 133]
Perfusion Parameters (FWHM/TTP)​	postoperative MMD patients	FWHM quantifies perfusion dispersion; TTP (4–6 s threshold) used as an alternative to Tmax for infarct core assessment	Postoperatively, 57% of patients showed greater improvement in FWHM compared to TTP, avoiding the need for arterial input function correction	[134]
Vessel Wall Contrast Enhancement(VW-CE)	postoperative MMD patients	VW-CE reflects vascular wall inflammation and neovascularization activity	Preoperative VW-CE regions were associated with 100% incidence of postoperative acute ischemic stroke	[133]
DSA
Orbital Grading System (OGS)	postoperative MMD patients	Quantification of collateral vessel growth	OGS is a more precise evaluation of postoperative collateral circulation than the Matsushima grading system	[135]
Ultrasound
Transcranial Doppler (TCD)​	postoperative MMD patients	Automatically tracks the superficial temporal artery	EDV > 16.62 cm/s predicts effective collateral opening	[136]
Radiomics & AI Models
Vascular Morphological Radiomics Features	MMD vs AS-MMS patients	Extraction of vascular morphological features to construct a differential diagnostic model	MMD discrimination model: OD threshold 1.77 mm (AUC = 0.912)	[128]
DSA-Based Collateral Circulation Quantification Model	MMD patients	Calculation of external carotid artery collateralization ratio (CAR) and intracranial arterial residual volume (ARV)	CAR is independently positively correlated with ARV (β = 0.385, P < 0.001), and guides revascularization strategies	[137]
Gene–Imaging Integration Model	MMD patients	WGCNA integrates molecular pathways, such as hypoxic stress and blood–brain-barrier dysfunction, with imaging phenotypes	Gene–Imaging Integration Model links microscopic molecular changes with macroscopic imaging features	[138]
3D CNN–BiConvGRU Model	MMD patients	Extraction of dynamic DSA features	Accurately predicts high hemorrhage risk	[135, 139]
GANs-based Synthetic CVR Imaging	MMD patients with contraindications to acetazolamide	Generate synthetic CVR maps from pre-ACZ ASL data	GANs-based Synthetic CVR Imaging addresses the challenge of assessing cerebrovascular reserve in patients for whom acetazolamide stimulation is contraindicated	[140]

Arterial spin labeling (ASL)-based dynamic MR angiography (ASL-MRDSA) further advances noninvasive evaluation using endogenous water as a contrast agent, eliminating nephrotoxicity risks. Multi-delay ASL allows short post-labeling delays (PLD 1525 ms) to quantify cerebral blood flow (CBF) and longer PLDs (2525 ms) to assess arterial transit time (ATT), offering a temporal resolution comparable to DSA [125, 130]. Pseudo-continuous ASL (PCASL) enhances signal stability by reducing labeling slab thickness, improving visualization of distal Moyamoya vessels and providing an effective tool for evaluating collateral networks [131]. Ultra-high-field 7 T TOF-MRA improves vascular resolution, visualizing abnormal networks and microaneurysms in the basal ganglia with greater clarity than 3 T MRA or DSA, and allowing more precise hemorrhagic risk assessment [132, 141]. Postoperative assessment is further refined by the orbit-based grading system (OGS), which quantitatively measures regional collateral growth following EDAS, providing a standardized framework for surgical outcome evaluation [135].

Functional imaging assessment

Functional imaging in MMD focuses on quantifying the brain’s compensatory capacity under hemodynamic stress. Cerebrovascular reactivity (CVR), typically evaluated via hypercapnic stimulation with BOLD-MRI, reflects vascular reserve capacity and predicts ischemic events [43, 142]. Prolonged CVR delays indicate exhaustion of perfusion reserve due to impaired smooth-muscle responsiveness and inadequate collateral flow, providing critical guidance for surgical timing. Multi-delay ASL improves perfusion asymmetry assessment by capturing both early flow (short PLD) and delayed transit (long PLD), with strong correlations to PET-derived CBF and mean transit time (MTT) (r = 0.63–0.71) [125, 143]. Post-revascularization studies demonstrate CBF increases in affected regions and reductions in ATT and Tmax, validating surgical efficacy. Novel perfusion parameters, including full width at half maximum (FWHM), quantify perfusion dispersion, capturing hemodynamic changes that TTP or Tmax may miss, and obviate complex arterial input function corrections [134]. Phase-contrast PET (PC-PET) also enables noninvasive whole-brain CBF measurement, revealing dose–response relationships between stenosis severity and cerebrovascular reserve [144]. Vascular wall contrast enhancement (VW-CE) correlates with inflammatory activity and neovascularization, with preoperative VW-CE regions predicting 100% of subsequent acute ischemic strokes, supporting its role as a perioperative risk biomarker [133]. Transcranial Doppler (TCD) monitoring provides objective postoperative measures, where superficial temporal artery end-diastolic velocity > 16.62 cm/s indicates effective collateral circulation [136].

Radiomics and artificial intelligence-assisted diagnosis

Radiomics and artificial intelligence (AI) technologies, through high-throughput feature extraction and deep learning algorithms, are driving MMD diagnostics toward automation and precision, although challenges remain regarding standardization and clinical translation (Table 4).

Radiomics offers objective, quantitative approaches that gradually replace traditional subjective assessments. For instance, by extracting vascular morphological parameters such as the outer diameter of the middle cerebral artery and the remodeling index, researchers developed a differential diagnostic model to distinguish MMD from atherosclerotic Moyamoya syndrome (AS-MMS), achieving an OD threshold of 1.77 mm and an AUC of 0.912 [128]. In surgical planning, quantitative evaluation of the correlation between the CAR and the ARV provides objective indicators to assess collateral capacity and optimize revascularization strategies [145]. Moreover, the integration of structural MRI (sMRI)-derived brain atrophy features with functional PET metabolic parameters, such as cerebral glucose metabolism rates, has enabled the construction of predictive models for cognitive impairment in MMD patients, highlighting the predictive value of striatal hypometabolism as an early warning sign of cognitive decline [146].

In intelligent diagnostic model development, the integration of multi-omics data with deep learning techniques has shown significant potential across clinical scenarios. For differential diagnosis, a logistic regression model combining morphological parameters from high-resolution vessel wall imaging (HR-VWI), such as vascular outer diameter and wall thickness, with perfusion indices derived from ASL techniques, including CBF and ATT, significantly improved diagnostic accuracy for distinguishing MMD from AS-MMS [124]. Similarly, gene–imaging fusion models constructed using weighted gene co-expression network analysis (WGCNA) identified and validated key molecular pathways implicated in MMD pathophysiology, including hypoxic stress responses regulated by AKT1/ISG20, blood–brain-barrier dysfunction mediated by CLDN3/TGFB2, and immune microenvironment dysregulation involving PIK3CG/CXCL10. These molecular features were mapped onto high-resolution imaging phenotypes, establishing an initial framework that bridges microscopic molecular alterations with macroscopic radiological manifestations [138].

In risk assessment, dynamic DSA-based models employing advanced hybrid deep learning architectures such as 3D CNN-BiConvGRU successfully extracted hemodynamic parameters including blood flow propagation velocity and microvascular stasis indices, enabling accurate prediction of hemorrhagic risk in MMD patients [139]. Clinical validation demonstrated that such AI-based risk warning systems can identify high-risk cases in advance and provide opportunities for timely intervention [135, 139], marking a transition toward precision prevention in MMD management. For patients contraindicated for acetazolamide challenge testing, Liu et al. proposed the generation of CVR maps using generative adversarial networks (GANs-CVR). By leveraging adversarial training between a generator and discriminator, this method synthesized post-challenge-like CVR maps from pre-administration ASL images, offering a novel solution for evaluating cerebrovascular reserve function in otherwise ineligible patients [140].

Other emerging biomarkers

Noncoding RNAs and exosome-based biomarkers

In recent years, noncoding RNAs and exosomes have attracted increasing attention as emerging biomarkers in MMD (Table 5). Among microRNA biomarkers, plasma exosomal miR-512-3p has been found to be significantly upregulated in children with MMD, where it disrupts RHOA-mediated angiogenic signaling by targeting and inhibiting ARHGEF3. Functional assays demonstrated that suppression of miR-512-3p expression enhances the angiogenic capacity of endothelial colony-forming cells (ECFCs), highlighting its potential as a therapeutic target [147]. Similarly, exosome-derived miR-151a-3p and miR-125b-5p are enriched in the plasma of MMD patients, where they promote aberrant vascular remodeling by inducing endothelial-to-mesenchymal transition (EndMT), characterized by the downregulation of endothelial markers (CD31/VE-cadherin) and the upregulation of mesenchymal markers (α-SMA/vimentin) [148]. In addition, the combined upregulation of hsa-miR-328-3p and hsa-miR-200c-3p, which target cytoskeleton-regulatory genes such as DIAPH1 and RND3, has been proposed as a highly sensitive and noninvasive diagnostic tool for MMD. Their abnormal elevation in peripheral blood also contributes to cytoskeletal remodeling defects and functional impairment of VSMCs [149].

Table 5.

Summary of emerging molecular biomarkers for Moyamoya disease

Biomarkers	Disease models	Functions	Application of biomarkers	References
Noncoding RNAs & exosomes
miR-512-3p	MMD patients	Targeted inhibition of ARHGEF3 to disrupt the RHOA-mediated angiogenic pathway	Inhibition restores the angiogenic capacity of ECFCs; potential therapeutic target	[147]
miR-151a-3p /miR-125b-5p	MMD patients	Induction of EndMT	Plasma enrichment levels reflect abnormal vascular remodeling activity	[148]
hsa-miR-328-3p /200c-3p	MMD patients	Targeted inhibition of DIAPH1 and RND3	Circulating biomarkers indicate vascular smooth-muscle dysfunction	[149]
IPO11_circRNA, PRMT1_circRNA	MMD patients	Promotion of pathological angiogenesis	Expression fluctuations correlate with the severity of arterial stenosis	[150]
circZXDC	MMD patients	circZXDC acts as a sponge for miR-125a-3p, relieving ABCC6 inhibition and activating the ERS pathway	circZXDC acts as a potential biomarker driving vascular stenosis	[151]
CACNA1F_circRNA	MMD patients	Driver of vascular stenosis/occlusion	A potential peripheral blood diagnostic biomarker	[50, 150]
Neutrophil-related circRNA	Asymptomatic MMD patients	Upregulates VEGF via HIF-1α; regulates neutrophil recruitment and cytokine release	Neutrophil-derived circRNA indicates involvement in immune response, metabolism, and angiogenesis	[152]
lncRNA	Intracranial arteries of MMD patients	lncRNAs are involved in immune-related pathways, including inflammatory responses, Toll-like receptor signaling, and MAPK signaling	lncRNAs indicate that immune activation and impaired vascular remodeling are central pathological processes	[153, 154]
Proteomics & Metabolomics
CDH18	MMD patients	Inhibition of abnormal proliferation and migration of VSMCs	Postoperative upregulation predicts favorable neurological recovery	[124, 155]
Haptoglobin/α-1B-glycoprotein (A1BG)	MMD patients	Association with RNF213 loss-of-function–driven pathological processes	Characteristic upregulation; potentially involved in inflammatory regulation	[150]
CFL1(Cofilin-1) / actin-related protein 2/3 (ACTR2/3)	Serum-derived exosomes (SDEs) from MMD patients and cerebrovascular ECs in mice	Abnormal actin dynamics and mitochondrial dysfunction	In ischemic/hemorrhagic MMD, CFL1/ACTR2/3 are downregulated in SDEs, and SDEs induce mitochondrial dysfunction	[156]
FLNA	MMD patients	FLNA is involved in ECM remodeling and the regulation of angiogenesis	Proposed targeting FLNA to regulate ECM remodeling	[157]
DG/triglycerides (TG) ratio	MMD patients	Activate vascular abnormal proliferation pathways	DG/TG ratio distinguishes MMD from intracranial atherosclerotic disease	[158]
L-methionine, L-glutamate, glutamine	MMD patients	Reflects accelerated amino acid catabolism and disordered energy metabolism	Multiple differential amino acids (such as β-alanine and O-phospho-L-serine) exhibit high diagnostic sensitivity	[150]
Betaine/Hcy	MMD patients	Betaine activates MetO reductases b1 (Msrb1)/Msrb2 to alleviate oxidative stress, while elevated Hcy promotes endothelial injury	Methionine cycle metabolite risk scoring model	[159]
Other emerging biomarkers
Desmosine	MMD patients	Reflects the intensity of elastin catabolism	Predicts the risk of progression of intracranial arterial stenosis–occlusion	[160]
Lysosomal genes (EPDR1, DENND3, NCSTN)	Vascular tissue in MMD Patients	Regulation of impaired cholesterol reverse transport and foam cell formation	Machine learning models demonstrate diagnostic potential	[137]
Vasoactive intestinal peptide (VIP)/CCK/SST	MMD Patients	Interaction between impaired blood–brain barrier and inflammatory cytokines	Decreased levels of VIP, CCK, and SST are independent predictors of MMD occurrence	[103]
Bacteroidetes/R. gnavusGut	Gut microbiota in MMD Patients	Gut microbiota dysbiosis accompanied by reduced concentrations of brain–gut peptides	Potential therapeutic target for gut–brain-axis interventions	[22, 150]

Significant advances have also been made in the study of circular RNAs (circRNAs) in MMD. Functional studies have demonstrated that IPO11_circRNA and PRMT1_circRNA promote pathological angiogenesis, whereas CACNA1F_circRNA primarily contributes to vascular stenosis and occlusion, highlighting their potential utility as diagnostic biomarkers detectable in peripheral blood [161]. Liu et al. further revealed that circZXDC acts as a competing endogenous RNA (ceRNA) by sponging miR-125a-3p, thereby releasing the inhibition of ABCC6. This process activates the endoplasmic reticulum stress (ERS) pathway, drives VSMCs toward a synthetic phenotype, and accelerates neointimal hyperplasia, ultimately resulting in vascular narrowing in MMD [151]. These findings underscore circZXDC as a promising biomarker for both diagnosis and prognosis. Moreover, Corey et al. identified 123 differentially expressed circular RNAs [circRNAs (54 upregulated, 69 downregulated)] in neutrophils from asymptomatic MMD patients via microarray analysis. These circRNAs were enriched in immune response, metabolic, and angiogenic pathways, potentially promoting endothelial proliferation and abnormal vascular network formation through HIF-1α-mediated VEGF upregulation, while simultaneously regulating neutrophil recruitment and cytokine release (e.g., IL-8, CXCL12), thereby exacerbating BBB disruption [152]. Collectively, the fluctuating expression profiles of these specific circRNAs in serum/plasma provide preliminary evidence supporting their potential as novel noninvasive biomarkers for assessing disease severity in MMD.

In the field of long non-coding RNA (lncRNA) research, Mamiya et al. demonstrated aberrant lncRNA expression in the intracranial arteries of MMD patients, suggesting that immune activation and impaired vascular remodeling represent core pathological processes [153]. Differentially expressed lncRNAs were found to be involved in antibacterial humoral responses, T-cell receptor signaling, cytokine production, and vascular morphogenesis. Co-expression network analyses of lncRNA–mRNA further revealed close associations with immune-related pathways, including inflammatory signaling, Toll-like receptor cascades, and MAPK signaling [154].

It is particularly noteworthy that dysregulated functional microRNAs (e.g., miR-512-3p, miR-151a-3p) in the pathological milieu of MMD mediate long-distance communication between lesional brain regions and the peripheral circulation. This process critically depends on the active transport capacity of exosomes—nanoscale vesicles capable of traversing the blood–brain barrier. Acting as efficient carriers, exosomes can precisely deliver pathogenic RNA molecules to target cells, thereby triggering pathological cascades such as EndMT and pro-fibrotic responses [147–149]. Owing to these fundamental properties, exosomes have emerged as the most promising translational medium for liquid biopsy in MMD, providing a molecular foundation for the development of noninvasive diagnostic tools, therapeutic monitoring, and targeted intervention strategies.

Integrating proteomics and metabolomics for precision diagnosis and treatment

Proteomic studies have identified cadherin-18 (CDH18) as a valuable prognostic biomarker. Its significant upregulation in CSF has been observed primarily in subgroups of MMD patients who experience favorable postoperative neurological recovery and long-term outcomes [155, 162]. Mechanistic investigations revealed that CDH18 directly suppresses abnormal proliferation and pathological migration of VSMCs, and its in vivo overexpression effectively inhibits pathological vascular remodeling, thereby improving clinical outcomes [155]. In addition, haptoglobin and A1BG also show characteristic upregulation in the CSF of MMD patients [150]. Wang et al. reported that differentially expressed proteins are implicated in cell growth, actin dynamics, and immune processes. For instance, in serum-derived exosomes (SDEs) from ischemic and hemorrhagic MMD patients, expression of CFL1 and ACTR2/3 was downregulated, while SDEs from hemorrhagic MMD patients induced mitochondrial dysfunction in cerebrovascular ECs [156]. Furthermore, Carrozzini et al. established a proteomic atlas of the dura mater (DM) in MMD patients, revealing direct molecular interactions between the dura and the nervous system. Their analysis identified aberrant expression of filamin A (FLNA), which participates in ECM remodeling and angiogenesis regulation, potentially driving MMD progression. FLNA was thus proposed as a diagnostic marker of vascular pathology and a potential therapeutic target for ECM-remodeling interventions [157].

Metabolomics research has revealed marked dysregulation of lipid metabolism in MMD, characterized by the accumulation of circulating diacylglycerol (DG) accompanied by reduced triacylglycerol (TG) levels, ultimately leading to aberrant vascular signaling pathways [158]. Notably, the DG/TG ratio has demonstrated excellent diagnostic performance as a composite biomarker for distinguishing MMD from intracranial atherosclerotic disease (ICAD), suggesting that lipid metabolic disturbances may represent a distinctive pathological hallmark of MMD. Moreover, differential lipid composition in intracranial arteries of MMD patients has shown a significant negative correlation with plasma iron levels [158, 163], implicating iron dysregulation in disease pathogenesis and highlighting the potential of therapeutic strategies targeting arterial lipid composition or plasma iron homeostasis.

Significant advances have also been achieved in amino acid metabolism. Liu et al. identified 12 amino acids that differed significantly between MMD patients and healthy controls using serum amino acid profiling, with L-methionine, L-glutamate, β-alanine, and O-phosphoserine showing high sensitivity and specificity [164]. Building on this, Li et al. developed an MMD risk score model based on methionine cycle-related metabolites, demonstrating that betaine attenuates oxidative stress and neuroinflammation by activating methionine sulfoxide reductases (Msrb1/Msrb2), thereby potentially suppressing disease progression. In contrast, elevated Hcy levels were associated with endothelial injury and both ischemic and hemorrhagic pathology in MMD [159], supporting the utility of noninvasive metabolic biomarkers for early disease screening. Coordinated upregulation of key plasma amino acids, such as L-methionine, L-glutamate, and glutamine, further reflected a systemic disruption of energy metabolic networks in MMD, characterized by accelerated amino acid catabolism and aberrant biosynthetic pathways [150, 165].

Importantly, integrated models combining proteomic and metabolomic biomarkers hold promise for substantially enhancing diagnostic specificity in MMD, while also paving the way for the development of personalized therapeutic strategies targeting the metabolic–immune interface.

Emerging biomarkers and research directions

Plasma desmosine, a specific biomarker reflecting the intensity of elastin degradation, is elevated in patients with ongoing elastic fiber breakdown in the vascular wall. Studies have demonstrated that when plasma desmosine concentrations consistently exceed the lower limit of quantification (LOQ) of the assay, the risk of progressive deterioration of intracranial arterial stenosis–occlusion is significantly increased [160]. This finding suggests that dynamic monitoring of plasma desmosine levels could serve as an early warning indicator of vascular elastin degradation and may hold predictive value for intracranial arterial stenosis progression.

Single-cell omics studies have further identified multiple lysosome-related genes, including markedly downregulated EPDR1 and aberrantly upregulated DENND3 and NCSTN. These genes may drive lipid metabolic abnormalities—by impairing cholesterol reverse transport and promoting foam cell formation—thereby facilitating abnormal immune-cell infiltration into the vascular wall and jointly accelerating the progression of MMD [137]. Although machine learning models based on these gene signatures remain in the exploratory stage, their discriminatory potential suggests that incorporating lysosome-associated pathways into diagnostic frameworks could represent a promising direction.

In recent years, gut microbiota–brain-axis research has opened new avenues for MMD diagnosis and treatment. Analyses of neuropeptides have revealed that vasoactive intestinal peptide (VIP), cholecystokinin (CCK), and somatostatin (SST) may interact with inflammatory cytokines through a compromised BBB, contributing to MMD pathogenesis [103]. Emerging evidence further indicates that MMD patients exhibit gut microbiota dysbiosis characterized by abnormal abundance of Lachnoclostridium and Fusobacterium, Bifidobacterium and Enterobacter, accompanied by significantly reduced plasma levels of several neuroactive gut peptides [22, 150]. Future animal model and cohort studies are required to clarify whether gut–microbiota imbalance acts as a driver of MMD or arises secondarily from cerebrovascular pathology—an issue of critical importance for developing gut–brain axis-targeted interventions.

Translational applications of biomarkers and limitations

Biomarkers are increasingly central to precision medicine in MMD, offering potential for early diagnosis, risk stratification, therapeutic guidance, and dynamic disease monitoring. Integration of diagnostic, predictive, and monitoring biomarkers enhances sensitivity and specificity, enabling real-time assessment of disease progression and treatment efficacy. Here, we summarize the advantages and disadvantages of clinical applications of MMD-associated genes, non-coding RNAs, proteomics, and metabolomics (Table 6).

Table 6.

Clinical applications, advantages, and limitations of representative biomarkers in Moyamoya disease

Biomarker catagery	Specific biomarkers	Clinical application	Advantages	Disadvantages
Genetic	RNF213 p.R4810K	Early diagnosis, risk stratification	High specificity in East Asians; aids in diagnosis	Low penetrance; low positive predictive value
Noncoding RNAs	miR-512-3p, miR-151a-3p, miR-125b-5p, CircRNAs	Diagnostic markers, therapeutic targets	Involved in disease mechanisms; detectable in blood	Requires further validation; variability in expression
Proteomics	CDH18, FLNA	Prognostic indicators, therapeutic targets	Associated with disease outcomes; potential therapeutic targets	Invasive sampling required; need for broader validation
Metabolomics	DG/TG Ratio, Amino Acids (L-methionine, L-glutamate)	Diagnostic biomarkers, metabolic profiling	Distinctive metabolic profiles; potential diagnostic markers	Influenced by various factors; need for standardized protocols

The RNF213 gene, particularly the p.R4810K variant, remains the most established susceptibility marker in East Asian populations, supporting high-risk screening, early diagnosis, and postoperative prognostication [166–168]. In cases with atypical or ambiguous imaging, RNF213 genotyping may provide supportive contextual information, but it cannot be used as a stand-alone discriminator between Moyamoya disease and atherosclerosis. Phenotypic severity correlates with zygosity: homozygous carriers often exhibit more aggressive disease requiring earlier intervention, whereas heterozygous carriers frequently demonstrate robust collateral formation post-revascularization [62, 169]. Despite these advantages, single-gene models have limited predictive value. The p.R4810K variant occurs in 1–2% of healthy East Asians, yet only ~ 0.5–2% of carriers develop MMD, yielding a low positive predictive value (~ 8.3%) [10]. Polygenic risk score (PRS) models integrating RNF213, HLA loci, and epigenetic markers show promise for improved risk stratification; a preliminary 12-SNP PRS increased sensitivity for identifying high-risk individuals to 67% [78]. However, PRS approaches remain limited by population specificity and incomplete incorporation of environmental modifiers, such as infection, hypoxia, or autoimmune disorders, which can significantly modulate penetrance. Retrospective data suggest, for example, that thyroiditis combined with RNF213 variants may elevate MMD risk fivefold, though prospective validation is lacking. Moreover, the p.R4810K mutation is largely restricted to East Asians, with minimal prevalence in other ethnicities (< 5%), and substantial clinical heterogeneity is observed even among carriers of the same genotype [3, 8, 9, 170].

Inflammatory and immune biomarkers hold promise for assessing disease activity, predicting recurrence risk, and monitoring therapeutic efficacy. Inflammatory cytokines such as TNF-α and interleukin-6 (IL-6) show stronger prognostic value in patients without RNF213 variants, highlighting their potential as complementary biomarkers in clinical practice [171]. The construction of integrated models that combine multi-omics information (including genomics, proteomics, and imaging) enables personalized risk stratification and tailored treatment planning, supporting individualized surgical decision-making and follow-up strategies. However, several challenges limit their clinical translation, including poor specificity, lack of standardized detection protocols, and substantial inter-individual variability. Currently, no single inflammatory or immune biomarker has been universally accepted as an independent diagnostic indicator. Future research should focus on developing combinatorial biomarker panels, integrating dynamic monitoring, tissue-specific validation, and machine learning-based prediction models to improve clinical utility. Although a wide array of potential biomarkers—encompassing genetic, inflammatory, immune, and imaging markers—have been proposed with potential applications in high-risk screening, early diagnosis, recurrence prediction, and therapeutic monitoring, the limitations of single biomarkers in terms of sensitivity and specificity remain prominent, hindering the development of robust diagnostic and prognostic models. For example, GUCY1A3 and BRCC3 were initially suggested to be associated with MMD [16, 29, 39, 40], yet subsequent multicenter case–control studies failed to confirm their significant associations [28], reflecting the lack of cross-population replication for some genetic markers. Similarly, inflammatory indices, such as the neutrophil-to-lymphocyte ratio (NLR), the SII, and the neutrophil-to-albumin ratio (NAR), have been linked to disease activity in several studies [94, 95] (Wang, Wang et al. 2025), but as systemic response markers, they are prone to external interference, and their stability and specificity remain insufficient for clinical use.

Biomarkers also demonstrate clear advantages in preoperative evaluation and postoperative management. Currently, surgical revascularization (direct or indirect) remains the mainstay of treatment for MMD. Genotypic information can be used to estimate expected revascularization efficacy and complication risk, guiding the optimal timing of surgical intervention. Meanwhile, peripheral blood inflammatory indices, such as the SII and the NAR, serve as easily accessible, cost-effective monitoring tools that provide evidence for predicting poor outcomes after stroke and informing early intervention strategies [94]. Moreover, inflammatory and immune-related biomarkers may serve as predictors of hemorrhagic subtypes, opening new avenues for immunomodulatory therapies [91, 95].

Despite the growing repertoire of genetic, inflammatory, immune, and imaging biomarkers, their clinical translation is constrained by fundamental challenges related to the quality and generalizability of the underlying evidence. As Moyamoya disease is a rare disorder, the majority of findings discussed in this review, including many of the biomarker candidates highlighted, derive from small-scale cohorts, retrospective case series, or single-center studies. This reliance on limited and often heterogeneous samples reduces statistical power, increases the risk of selection bias, and limits the external validity of the conclusions [172].

Substantial heterogeneity across studies further complicates biomarker validation. Notably, markers with strong associations in East Asian populations, such as the RNF213 p.R4810K variant, frequently demonstrate low prevalence or absent association in non-Asian cohorts, highlighting distinct genetic and possibly pathophysiologic backgrounds. At the molecular level, numerous reported candidates, such as specific miRNA signatures, metabolomic profiles, and proteomic features, remain preliminary. This is due to inconsistencies in assay methodologies, sample processing, and analytical pipelines, coupled with a lack of standardized validation protocols, which collectively result in sporadic findings that await replication in independent, multi-ethnic cohorts. Many initially reported genetic associations, including GUCY1A3 and BRCC3, have not been consistently replicated across populations [16, 28, 29, 39, 40]. Similarly, inflammatory indices exhibit limited specificity and are prone to external interference.

Therefore, while the expanding biomarker landscape in MMD offers valuable mechanistic insights and generates important hypotheses, the current body of evidence should be interpreted as largely exploratory rather than definitive. Advancing these discoveries into reliable clinical tools will necessitate rigorously designed, prospective, multi-center international studies that employ standardized protocols to adequately address challenges related to sample size, phenotypic and ethnic heterogeneity, and technical reproducibility.

Future perspectives and directions

Biomarker research in MMD has progressed from the exploration of single molecular candidates to a new stage driven by large-scale data and empowered by multidisciplinary integration. Looking ahead, the central task is no longer merely the expansion of biomarker catalogs, but rather the integration of multidimensional information to construct dynamic, multilayered, and predictive disease models that capture the complete causal chain—from upstream genetic susceptibility to downstream clinical phenotypes. Achieving this vision will require three key pillars: high-quality multi-omics datasets as the foundation, powerful analytic engines that combine AI with bioinformatics, and a global collaborative network as the cornerstone for translation and implementation.

Emerging biomarkers and research directions

To date, single-omics studies have provided important molecular insights into MMD. Genomic analyses have identified RNF213 variants as major risk factors in East Asian populations, which disrupt vascular development [162]. GWAS have identified synergistic effects among MTHFR C677T, DIAPH1, and other variants, forming a polygenic susceptibility network [173]. Epigenetic studies have shown that hypomethylation of CpG sites in the SORT1 promoter increases its transcription, suppresses endothelial tube formation, and modulates MMP-9 expression, thereby contributing to vascular remodeling. Transcriptomic studies highlight the impact of non-coding RNAs. For example, miR-328-3p and miR-512-3p regulate the RHOA pathway by targeting ARHGEF3 and DIAPH1, influencing cytoskeletal remodeling and VSMC function. Meanwhile, miR-151a-3p and miR-125b-5p, which are enriched in plasma, promote EndMT, characterized by the downregulation of endothelial markers (CD31, VE-cadherin) and the upregulation of mesenchymal markers (α-SMA, vimentin), which drive pathological vascular remodeling and underscore the multilayered mechanisms of MMD [147, 149].

Proteomic studies indicate that upregulating CDH18 in ECs suppresses VSMC proliferation and migration, whereas overexpression may enhance vascular remodeling [155]. Metabolomic profiling reveals lipidomic alterations: intracranial DG and TG levels are positively correlated, indicating distinctive MMD-associated signatures and an inverse relationship with plasma iron [158]. Methionine cycle disruptions—characterized by elevated Hcy and reduced betaine—have been shown to support risk score models that predict disease progression [159]. Diagnostic models using plasma amino acid signatures, such as β-methionine and β-alanine, achieved an AUC greater than 0.90 [164]. Overall, these findings outline a molecular map of MMD that remains fragmented across genomics, transcriptomics, proteomics, and metabolomics.

While single-omics studies reveal disease-associated changes at individual biological levels (such as DNA, RNA, protein, or metabolites), they fall short of reconstructing the full causal chain from genetic variation to clinical phenotype. This limitation highlights the need for integrative multi-omics approaches (combining multiple "-omics" datasets) to advance biomarker research in MMD. Multiomics integration aims to transform fragmented data into biologically meaningful networks. Initial progress in horizontal integration has been achieved. For instance, Guo et al. combined proteomic and metabolomic datasets using WGCNA and KEGG pathway analysis. They linked aberrant ferroptosis activation with upregulated transferrin receptor protein-1 (TFRC), elevated L-glutamate, and reduced uracil in MMD [174]. These findings bridge different omics layers. Vertical integration, in contrast, seeks to uncover causal chains across biological hierarchies. For example, Zhou et al. applied machine learning to identify differentially expressed genes and integrated immune infiltration analyses. This highlighted key genes associated with reduced Treg cell abundance and validated the regulatory role of ARAP3 in endothelial function in vitro. This established a mechanistic “gene–immune–cell function” pathway [56], mapping gene networks onto cellular pathophysiology. These integrative strategies provide the foundation for initiatives such as MOYAOMICS, aiming to shift from unidimensional analyses to system-level approaches [150].

Despite these advances, significant challenges remain. Most current studies have constructed only MMD-associated networks and have not demonstrated causality. Future research will need to integrate longitudinal cohort data (data collected from participants over time) and use spatially resolved omics technologies (methods that analyze genetic, protein, and metabolic information with spatial context). The application of causal inference algorithms (computational approaches that attempt to identify cause-and-effect relationships) is also required. Such efforts will let the field move beyond correlative associations toward causal validation and dynamic prediction. This progress will allow for precision interventions.

Integration of artificial intelligence and bioinformatics as a new paradigm for biomarker discovery

Multiomics studies generate vast datasets. Bioinformatics transforms these into biologically interpretable knowledge that can inform clinical decision-making. Moving from static networks to dynamic predictive models requires powerful analytical tools. In this context, integrating AI, especially machine learning, with bioinformatics is emerging as a transformative paradigm for biomarker discovery and disease modeling in MMD. The strength of this paradigm comes from combining the predictive capacity of machine learning with the biological interpretability of bioinformatics. Researchers typically begin by applying algorithms such as support vector machine–recursive feature elimination (SVM-RFE), Boruta, or least absolute shrinkage and selection operator (LASSO). These methods extract features with the highest diagnostic or prognostic value from high-dimensional data. Next, pathway enrichment tools, such as Gene Ontology (GO) and KEGG, validate the biological relevance of candidate biomarkers. For example, Han et al. identified keratin 8 (KRT8) as a candidate genetic marker using SVM-RFE, Boruta, and LASSO, achieving excellent model performance (training set AUC = 0.945; validation set AUC = 0.958). Functional analyses confirmed its involvement in angiogenesis pathways through protein–protein interaction (PPI) networks [137, 175]. In a related study, the same group integrated RNA-sequencing with machine learning. They identified oxidative phosphorylation-related genes, including CSK and SMAD2. These genes yielded a diagnostic nomogram with an AUC greater than 0.8 and mechanistic validation through gene set enrichment analysis (GSEA), highlighting their role in immune infiltration [57].

Integration of AI and bioinformatics also facilitates the discovery of more complex biological mechanisms. Tang et al. applied WGCNA to identify cell death-associated gene modules linked to MMD [176]. Weng et al. combined metabolomics with deep learning to distinguish patients with MMD based on serum metabolic fingerprints (AUC = 0.958), and used K-means clustering to stratify patients into risk subgroups associated with metabolic–immune dysregulation [177]. Wang et al. showed that apolipoproteins (APOA1, APOA4) involved in cholesterol metabolism were significantly downregulated, with plasma APOA4 correlating positively with HDL-C (r = 0.52, p < 0.01) [178]. In parallel, TFRC upregulation and glutamate metabolism abnormalities have been implicated in ferroptosis, contributing to endothelial injury [9]. Neuroendocrine peptide analyses further suggested that VIP, CCK, and SST may cross the disrupted blood–brain barrier and interact with inflammatory cytokines to promote MMD pathogenesis [103]. Collectively, these findings move beyond single-biomarker exploration and begin to map the molecular pathology of MMD.

A major frontier of AI and bioinformatics integration is its application to neuroimaging, which holds promise for decoding the heterogeneous clinical and radiological phenotypes of MMD and accelerating clinical translation. Disease course and imaging manifestations in MMD are highly variable, and traditional methods often fail to capture subtle features. Deep learning models, such as DenseNet and ResNet, can extract microscopic features from DSA and CTA images that are invisible to the human eye, thereby improving diagnostic accuracy, disease staging, and prognostic prediction. Lu et al. developed a DenseNet-121 model that achieved outstanding diagnostic performance (internal test set AUC = 0.977 [95% CI: 0.928–0.995]; external validation AUC = 0.880 [95% CI: 0.786–0.937]), comparable to radiologists [179]. Ma et al. demonstrated that deep learning-based reconstruction (DLR-brain) enhances the visualization of small vessels on CTA at low radiation doses. This results in improved image quality and spatial resolution, accompanied by reduced noise [180]. Lei et al. applied ResNet-152 to DSA sequences, achieving 97.64% accuracy (AUC = 0.99) in the diagnosis of unilateral lesions. Their MV-CNN-C model, which combines demographic and clinical features, predicted hemorrhagic risk with 94.12% sensitivity [181]. Beyond brain imaging, Hong et al. demonstrated that a retinal deep learning model could accurately screen for MMD (AUC = 94.6%, sensitivity = 89.8%, and specificity = 90.4%). Stage prediction AUCs exceeded 78% across disease phases [182]. These advances demonstrate AI’s potential for diagnostic improvement and the integration of macroscopic imaging with molecular pathology.

The next breakthrough lies in achieving system-level integration of clinical, imaging, and omics data. Future AI models must move beyond recognizing imaging features to mechanistically linking them with molecular processes and clinical outcomes. Tan et al. exemplified this approach by constructing a diagnostic nomogram based on hypoxia–immune-related genes (AKT1, CLDN3, ISG20, TGFB2) (AUC = 0.924), and validating its correlation with vascular stenosis severity through radiomic features [138]. A meta-analysis by Santana et al. of seven studies confirmed the high diagnostic performance of AI models for MMD (sensitivity 0.89–0.92; specificity 0.89–0.91; pooled AUC 0.94), providing robust evidence for clinical translation [183].

However, key challenges are still unresolved. First, the “black box” nature of deep learning reduces clinical trust; explainable AI will be crucial for transparency [184]. Second, generalizability remains limited due to the reliance on small, single-center datasets, which lack sufficient external validation [183]. Third, multimodal data integration lacks standardized frameworks, and the combination of model complexity and the need for rigorous data harmonization presents obstacles to clinical implementation [139, 146]. To overcome these barriers, establishing large-scale, standardized, multi-center databases is a critical priority. Taken together, the integration of AI and bioinformatics is transforming biomarker discovery, risk stratification, and mechanistic research in MMD, with particular promise in imaging-based precision medicine [176]. Realizing its full clinical potential will depend on explainability, robust validation, and standardized multimodal integration.

The necessity and implementation pathways of international multicenter collaboration and biomarker database construction

Establishing a unified, multimodal international database for MMD is the cornerstone for achieving data and technology standardization, as well as for building a global collaborative network. As a rare disease, MMD research has long been constrained by the limitations of single-center, small-sample studies, resulting in insufficient statistical power and limited generalizability. A lack of high-quality randomized-controlled trials and observational studies—especially the limited availability of Western datasets—continues to impede the identification of reliable, generalizable biomarkers and the development of AI models [185]. Thus, fostering international multicenter collaboration and establishing open, shared, and sustainable research infrastructures are crucial for validating the generalizability of biomarkers and enabling their clinical translation, representing a necessary path for advancing the field.

Major initiatives like the Human Connectome Project (HCP), UK Biobank, and the Chinese Human Connectome Project (CHCP) serve as important examples of international collaborative databases [186]. Such cooperation enables the aggregation of diverse population datasets, facilitating the validation of core genetic markers, such as RNF213, and molecular subtypes across different ethnic groups [121], while also allowing for the identification of region-specific factors. A dedicated MMD database should integrate multi-ethnic genomic, transcriptomic, proteomic, and metabolomic data alongside standardized clinical phenotypes and imaging records. Initiatives like MOYAOMICS have already offered a preliminary blueprint for this endeavor [150].

However, the technical framework must advance in parallel with ethical and governance frameworks. Cross-border data sharing is challenged by ethical, legal, and cultural complexities [187]. The UK’s broad consent model, Uganda’s sample custodianship system, and the cautious collaborative approaches adopted in the Arab region collectively underscore the critical importance of establishing internationally agreed-upon standards for data governance and ethics [188, 189]. Beyond ethical consensus, technological innovations are also required. Federated learning (FL) provides a promising approach by allowing multiple centers to collaboratively train models without sharing raw patient data; only model parameters are securely exchanged [190]. This approach has demonstrated potential within collaborative networks such as PCORnet in the United States and the international OHDSI consortium, providing a secure and feasible pathway for global collaborative analyses while safeguarding data sovereignty and patient privacy.

To guarantee data quality and comparability, it is essential to create and implement internationally standardized operating procedures (SOPs) that encompass every stage from sample collection and processing to storage and analysis. Detailed specifications should include variables such as blood collection timing and MRI scanning parameters, alongside the establishment of biomarker reference standards that can serve as cross-cohort quality controls [191]. Building an open, shared, and sustainable global MMD research ecosystem is an imperative for the future of precision medicine. Such a system should encompass standardized analytic pipelines, open-access biobanks, and capacity-building initiatives in data governance, such as those pioneered by BBMRI-ERIC [192]. Only through the co-construction of an open, standardized, and dynamically updated international resource platform, can the discovery, validation, and clinical translation of MMD biomarkers be accelerated, ultimately achieving the global implementation of precision medicine.

Conclusion

The search for biomarkers in Moyamoya disease is undergoing significant advancements, linking genetic research, vascular science, immune response studies, and modern imaging techniques, as summarized in Fig. 1. The RNF213 gene, especially prevalent in East Asian populations, remains the strongest genetic risk factor. Other rare gene variants—such as ACTA2, GUCY1A3, and BRCC3—broaden the range of genetic contributors seen in MMD and related conditions. Inflammation, immune mechanisms, and factors involved in blood vessel formation and repair (including VEGF, MMP-9, EPCs, eNOS gene variations, and PDGF-BB) highlight the importance of ongoing monitoring to guide diagnosis and therapy. Advances in imaging are improving diagnostic capabilities: while DSA is still the reference method, it has limitations for early detection and follow-up. New MRI techniques, enhanced by artificial intelligence, offer more accurate and automated evaluations but need further standardization and testing in clinical settings. At the same time, non-coding RNAs, exosome-based indicators, and comprehensive omics approaches may provide less invasive and more targeted tools for diagnosis and personalized treatment. It is important to note that many of the biomarker candidates discussed herein require validation in larger, prospective, and multi-ethnic cohorts to confirm their clinical utility, given the inherent challenges of studying a rare disease. Looking ahead, combining detailed omics data, advanced vascular imaging, and AI-powered analyses could lead to predictive and individualized care models for MMD. Realizing this goal will require global collaboration, validation across multiple centers, and the creation of shared biomarker resources to speed up translation from research to patient care.

Fig. 1.

Schematic illustration of ongoing biomarker research and future perspectives in Moyamoya disease

Materials and methods

A systematic and comprehensive literature search was conducted using the PubMed and Web of Science databases to identify relevant publications up to September 2025 (Fig. 2). The search strategy combined the terms “Moyamoya disease”, “moyamoya”, and “moyamoya syndrome” paired with “biomarker”, in addition to domain-specific keywords including “genomics”, “immunomics”, “pathology”, “single-cell omics”, “metabolomics”, “functional imaging”, “proteomics”, “artificial intelligence”, and “machine learning”. Titles and abstracts were screened for eligibility. Studies were included if they investigated biomarkers related to the diagnosis, prognosis, or therapeutic response of MMD. Articles not primarily focused on biomarkers were excluded. No restrictions were applied regarding study design or methodological approach.

Fig. 2.

Schematic overview of the research strategy

Abbreviations

A1BG

α-1B-glycoprotein

ACTA2

α-Smooth muscle actin encoding gene

ACTR2/3

Actin-related protein 2/3

AI

Artificial intelligence

AIF

Arterial input function

Ang-2

Angiopoietin-2

ARGs

Angiogenesis-related genes

ARV

Arterial residual volume

ASL

Arterial spin labeling

ASL-MRDSA

Arterial spin labeling-based dynamic magnetic resonance angiography

AS-MMS

Atherosclerotic Moyamoya syndrome

ATT

Arterial transit time

AUC

Area under the curve

AutoAbs

Autoantibodies

BBB

Blood–brain barrier

BMECs

Brain microvascular endothelial cells

BOLD

Blood oxygen level-dependent

BRCC3

BRCA1/BRCA2-containing complex subunit 3

CAR

Carotid artery collateralization ratio

Cav-1

Caveolin-1

CBF

Cerebral blood flow

CCK

Cholecystokinin

CDH18

Cadherin-18

cEPCs

Circulating endothelial progenitor cells

ceRNA

Competing endogenous RNA

CFL1

Cofilin-1

cGMP

Cyclic guanosine monophosphate

CHCP

Chinese Human Connectome Project

circRNAs

Circular RNAs

ClpP

Caseinolytic mitochondrial matrix peptidase proteolytic subunit

CVB3

Coxsackievirus B3

CVR

Cerebrovascular reactivity

CXCL12

C–X–C motif chemokine ligand 12

CXCR4

C–X–C motif chemokine receptor 4

DG

Diacylglycerol

DM

Dura mater

DRGs

Disulfidptosis-related genes

DSA

Digital subtraction angiography

ECFCs

Endothelial colony-forming cells

ECM

Extracellular matrix

ECs

Endothelial cells

EDV

End-diastolic velocity

EndMT

Endothelial-to-mesenchymal transition

EPC

Endothelial progenitor cell

EPDR1

Ependymin related 1

EPO

Erythropoietin

ERGs

EndMT-related genes

ERS

Endoplasmic reticulum stress

FL

Federated learning

FLNA

Filamin A

FWHM

Full width at half maximum

GANs

Generative adversarial networks

GO

Gene Ontology

GSEA

Gene set enrichment analysis

GWAS

Genome-wide association study

hCMECs

Human cerebral microvascular endothelial cells

HCP

Human Connectome Project

Hcy

Homocysteine

HDAC9

Histone deacetylase 9

HHcy

Hyperhomocysteinemia

HLA

Human leukocyte antigen

HR-VWI

High-resolution vessel wall imaging

HSV-1

Herpes simplex virus type 1

ICAD

Intracranial atherosclerotic disease

IL-6

Interleukin-6

IRGs

Immune-related genes

IVES

Intravascular enhancement sign

JAK2/STAT3

Janus kinase 2/signal transducer and activator of transcription 3

KDR

Kinase insert domain receptor (also known as VEGFR2)

KRT8

Keratin 8

LASSO

Least absolute shrinkage and selection operator

LDs

Lipid droplets

lncRNA

Long non-coding RNA

LOQ

Limit of quantification

LPS

Lipopolysaccharide

LUBAC

Linear ubiquitin chain assembly complex

MHC

Major histocompatibility complex

MMD

Moyamoya disease

MMP-9

Matrix metalloproteinase-9

MMS

Moyamoya syndrome

MRA

Magnetic resonance angiography

mRS

Modified Rankin Scale

MSDMS

Multisystem smooth-muscle dysfunction syndrome

Msrb1

MetO reductases b1

MTHFR

Methylenetetrahydrofolate reductase

MTRR

Methionine synthase reductase

NAR

Neutrophil-to-albumin ratio

NCSTN

Nicastrin

NF-κB

Nuclear factor kappa-light-chain-enhancer of activated B cell

NGS

Next-generation sequencing

NO

Nitric oxide

NR

Nicotinamide riboside

OGS

Orbit-based grading system

OXPHOS

Oxidative phosphorylation

PCA

Posterior cerebral artery

PCASL

Pseudo-continuous arterial spin labeling

PC-PET

Phase-contrast positron emission tomography

PDGF

Platelet-derived growth factor

PDGF-BB

Platelet-derived growth factor-BB (homodimer of PDGF B chain)

PDGFRB

Platelet-derived growth factor receptor beta

PI3K/AKT

Phosphoinositide 3-kinase/protein kinase B signaling pathway

PKG

Protein kinase G

PLD

Postlabeling delay

PPARγ

Peroxisome proliferator-activated receptor gamma

PPV

Positive predictive value

PTP1B

Protein tyrosine phosphatase 1B

RNF213

Ring finger protein 213

RSV

Respiratory syncytial virus

S1PR1

Sphingosine-1-phosphate receptor 1

SDEs

Serum-derived exosomes

sGC

Soluble guanylate cyclase

SII

Systemic immune-inflammation index

SMC

Smooth muscle cell

SMDS

Smooth muscle dysfunction syndrome

sMRI

Structural magnetic resonance imaging

SNPs

Single nucleotide polymorphisms

SOPs

Standardized operating procedures

SST

Somatostatin

sTie-2

Soluble Tie-2

SVM-RFE

Support vector machine–recursive feature elimination

TCD

Transcranial Doppler ultrasound

TCN2

Transcobalamin II

TFAM

Mitochondrial transcription factor A

TFRC

Transferrin receptor protein-1

TG

Triglycerides

TGF-β

Transforming growth factor beta

TNF-α

Tumor necrosis factor-alpha

TTP

Time-to-peak

VEGF

Vascular endothelial growth factor

VEGFR

Vascular endothelial growth factor receptor

VIP

Vasoactive intestinal peptide

VSMC

Vascular smooth-muscle cell

VW-CE

Vascular wall contrast enhancement

VW-MRI

Vessel wall magnetic resonance imaging

WGCNA

Weighted gene co-expression network analysis

Author contributions

Conceptualization, J.L., C.D., and X.H.; methodology, C.D. and J.L.; software, C.D. and J.L; validation, C.D. and X.H.; formal analysis, J.L., L.Z. and C.D.; investigation, J.L., L.Z. Y.Z., and C.D.; resources, J.L., Y.Z., and L.Z.; data curation, C.D. and X.H.; writing—original draft preparation, J.L., L.Z. and C.D.; writing—review and editing, C.D. and X.H.; visualization, C.D. and X.H.; supervision, C.D. and X.H.; project administration, C.D. and X.H.; funding acquisition, C.D..All authors have read and agreed to the published version of the manuscript. Acknowledgments: We are deeply grateful to my colleagues for their insightful guidance, which has greatly improved the quality of my manuscript.

Funding

This study was supported by the National Natural Science Foundation of China under Grant No. 82302108.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethical approval and consent to participate

Not available.

Competing interests

The authors declare no competing interests.