Clinical, genomic, and histopathologic diversity in cerebral cavernous malformations

Abstract

Cerebral cavernous malformations (CCMs) are hemorrhagic vascular disorders with varied clinical and radiological presentations, occurring sporadically due to MAP3K3 or PIK3CA mutations or through inherited germline mutations of CCM genes. This study aimed to clarify the clinical, genetic, and pathological features of CCMs using a multicenter cohort across three Chinese centers. We analyzed 290 surgical specimens from symptomatic CCM patients, utilizing whole-exome sequencing, droplet digital PCR, and targeted panel sequencing, alongside immunohistology to examine genotypic and phenotypic differences. Among 290 cases, 201 had somatic MAP3K3, PIK3CA, or germline CCM mutations, each associated with distinct clinical parameters: hemorrhage risk (P < 0.001), lesion size (P = 0.019), non-hemorrhagic epilepsy (P < 0.001), Zabramski classifications (P < 0.001), developmental venous anomaly presence (P < 0.001), and MRI-detected edema (P < 0.001). PIK3CA mutations showed a higher hemorrhage risk than MAP3K3 and combined MAP3K3 & PIK3CA mutations (P < 0.001). Within PIK3CA mutations, the p.H1047R variant correlated with higher bleeding risk than p.E545K (P = 0.007). For non-hemorrhagic epilepsy, patients with single MAP3K3 mutations or combined MAP3K3 & PIK3CA mutations were at greater risk than those with PIK3CA mutations alone. Histopathologically, lesions with PIK3CA mutations displayed cyst walls, pS6-positive dilated capillaries, and fresh blood cells, while MAP3K3 and double mutation lesions exhibited classic CCM pathology with SMA-positive and KLF4-positive vessels, collagen, and calcification. PIK3CA lesions had fewer KLF4-positive cells than double mutations lesions (P < 0.001), and EndMT (SMA-positive) cells compared to double mutation lesions (P < 0.05) and MAP3K3 mutations (P < 0.001), with more pS6 compared to MAP3K3 mutations (P < 0.05). This study underscores the diverse clinical, genomic, and histopathological characteristics in CCMs, suggesting potential predictive markers based on mutation subtypes and MRI features.

Introduction

Cerebral cavernous malformations (CCMs) are vascular disorders affecting post-capillary endothelial cells of the central nervous system (CNS) and are characterized by clusters of enlarged capillaries. They rank as the second most common vascular malformation in the CNS and its prevalence ranges from approximately 0.16–0.4% [1,2,3]. CCMs can manifest incidentally or present with symptoms such as headaches, epilepsy, focal neurological deficits or hemorrhage [4, 5]. There are two different CCM types including a familial inherited and a sporadic form. Genetic analysis of CCMs has predominantly focused on the inherited form, which is associated with one of three loss-of-function genes within the CCM complex: KRIT1 (CCM1), MGC4607 (CCM2), or PDCD10 (CCM3) [6,7,8,9,10]. However, the majority of CCM cases (approximately 85%) occur sporadically [11]. Recent research has uncovered the significant role of gain-of-function mutations in MAP3K3 and PIK3CA for CCM suggesting that PIK3CA causes CCM lesion progression and hemorrhage in preclinical studies [12, 13]. PIK3CA mutations were identified as drivers of aggressive lesion burdens in CCMs, with rapamycin, which inhibits PI3K-AKT-mTOR pathways, showing promise in preclinical studies involving CCMs with PIK3CA mutations [14, 13]. However, the mutational landscape of CCMs, as well as the precise role of genomic and histopathologic diversity in shaping the heterogeneous clinical presentations and radiological appearances of CCMs remains uncertain [12, 15, 16, 13, 17].

The aim of this study is to comprehensively analyze and correlate clinical phenotypes of CCM with genotypic and histopathologic profiles obtained from surgically resected CCM lesions. Additionally, it aims to provide assistance in therapeutic stratagem and the selection of suitable patients with CCMs for targeted therapy.

Materials and methods

Study design and participants

We conducted a multicenter cohort study involving three reference centers in China. The study primarily encompassed patients diagnosed with symptomatic CCMs through pathological examinations, clinical symptomatology, and imaging modalities, who had treatment indications and underwent surgical procedures for the acquisition of operative specimens. The inclusion period ranged from January 1, 2013, to May 31, 2023, while specifically excluding individuals with CCMs located outside the intracranial compartment, such as those involving the spinal cord. Within this study, we conducted a comprehensive analysis, including whole-exome sequencing (WES), droplet digital PCR (ddPCR), and targeted panel sequencing, utilizing 290 surgical specimens of CCMs.

Data collection and procedures

We conducted a comprehensive review of all clinical information within our database. The diagnosis of CCMs was established based on symptoms (e.g., location-related hemorrhage, neurological dysfunction and epilepsy), with supporting evidence from imaging data, notably magnetic resonance imaging (MRI), and when available pathological examinations [18]. The imaging data predominantly relies on imagery associated with the initial symptoms and evaluation of the imaging was carried out independently by neuroradiologists. We adhered to the criteria established by the Alliance to Cure Cerebral cavernous malformation Scientific Advisory Board to define an overt hemorrhage event [19]. We employ pre-established criteria to evaluate conditions such as headaches, epilepsy, and neurological functional disorders [20]. Lesion size was defined as the maximum diameter measured on transverse T1-weighted images, while the MRI appearance was classified according to the Zabramski classification [21]. Surgical resection in our centers adhered to the standards established by the Angioma Alliance Scientific Advisory Board [4]. We utilized surgically resected fresh frozen samples or paraffin-embedded samples of CCMs to conduct genetic analyses through various methods, including WES, ddPCR and targeted panel sequencing.

Genetic analysis

The tissue samples of CCMs were sent to the sequencing facility of Nanjing Geneseeq Biotechnology Inc. (Nanjing, China) for NGS analyses. The sequencing depth for whole exome sequencing (WES) was 200X-300X. Genomic DNA isolation was performed using the commercially available DNeasy Blood and Tissue Kit (Qiagen) with established protocols. Each sample, containing 1–2 µg of genomic DNA, underwent fragmentation to 300–350 bp, followed by end-repairing, A-tailing, and adaptor ligation using the Covaris M220 sonication system and KAPA Hyper Prep Kit (KAPA Biosystems, KK8504). Subsequently, size selection and purification were conducted using Agencourt AMPure XP beads (Beckman Coulter), followed by PCR amplification and purification using Agencourt AMPure XP beads. Quality control checks were performed using NanoDropTM 2000 (Thermo Fisher Scientific) for A260/280 and A260/230 ratios, Bioanalyzer 2100 with High Sensitivity DNA kit (Agilent Technologies, 5067 − 4627) for size distribution, and Qubit 3.0 dsDNA HS Assays (Life Technology) for quantification. All procedures were conducted following the manufacturer’s recommended protocols. For hybridization capture, the xGen Exome Research Panel v1.0 (Integrated DNA Technologies) was used, following the manufacturer’s recommended protocol. After target enrichment, libraries were quantified by qPCR using the KAPA Library Quantification Kit (KAPA Biosystems). Sequencing was performed on HiSeq4000 NGS platforms (Illumina) with paired-end 150 bp sequencing chemistry according to the manufacturer’s instructions. DdPCR was performed as previously described [12].

The targeted panel used for sequencing these CCMs comprised 77 genes, including MAP3K3, PIK3CA, AKT1, KRIT1, CCM2, PDCD10, among others, with a sequencing depth target of 1000× for the tissue samples [22]. For DNA extraction, the BunnyMag FFPE DNA Isolation Kit (BunnyTeeth Technology Inc., TQ02BT) was utilized. The quantity of extracted DNA was measured using Qubit 3.0 dsDNA HS Assays (Life Technology). Library preparation required a minimum of 50 ng of genomic DNA, which was fragmented mechanically to 150–250 bp segments using the IGT™ Fast Library Prep Kit v2.0. End-repairing, A-tailing, and adaptor ligation were performed using IGT™ Adapter & UDI Primer 1–96 (for Illumina, plate). Biotin-labeled probes of the customized TargetSeq^® Target Probes T725V2 (iGeneTech Bioscience Co., Ltd.) were hybridized to the products, followed by specific capture of library fragments with targeted sequences using streptavidin-coated magnetic beads. Quality assessment involved ensuring a Qubit concentration of more than 25 ng/µl and verifying the peak of the size distribution within 220–450 bp, utilizing Qsep 100. Sequencing was carried out on the Illunima NovaSeq 6000 under PE150 mode, with a target data amount of 1G.

Pathological analysis

The surgical specimens were obtained from the pathology database of Xuanwu Hospital, Capital Medical University. The specimens for pathological analysis were fixed in 10% phosphate-buffered formalin and embedded in paraffin. Multiple consecutive sections of 5 μm thickness were stained according to standard protocols with hematoxylin and eosin (H&E), Masson’s trichrome, and immunohistochemistry staining for SMA (Servicebio, GB111364, 1:1000), Ki67 (Servicebio, GB121141, 1:600), KLF4 (R&D, AF3640, 1:100), pS6 (CST, #4858S, 1:300), COL1A1 (Servicebio, GB11022-3, 1:1000), and CD31 (Servicebio, GB12063, 1:200). Stained sections were imaged using Pannoramic Scan(3DHISTECH Ltd). We delineated the lesions on the immunohistochemical slices and computerized image analysis was performed by Aipathwell (Servicebio) an AI-based digital pathology image analysis software. We used the positive cells to show the expression levels of SMA, pS6, and KLF4, where the positive cells = number of positive cells / total cell count.

Statistical analysis

For categorical variables, we employed statistical tests such as the Fisher exact test or Pearson chi-square test, while continuous variables were assessed using the student t-test. We systematically examined the influence of various predictive factors, symptom, maximum diameter, Zabramski classification, developmental venous anomaly (DVA), gender, edema and age of onset, among others. Subsequently, we employed Logistic analyses to scrutinize the correlation between their mutation types and clinical symptoms, evaluating the parameters for their scientific rigor in terms of relevance, accuracy, reliability, and comprehensiveness [23, 20]. Only clinical manifestations demonstrating statistical significance in univariate analysis were included in the multivariate analysis. All statistical analyses were carried out using SPSS software (version 27, IBM Corp) and Stata (version 15.0, StataCorp). Statistical significance was defined as P < 0.05 (The inter-group comparison using 2 × 3 chi-square test, α = 0.05/3 = 0.017) and all reported p-values are two-sided.

Results

A total of 290 surgical CCM specimens were subjected to genetic sequencing, comprising 77 samples (26.6%) analyzed by WES, 213 samples (73.5%) by targeted panel sequencing, and 26 samples that underwent both WES and ddPCR. Among these, 169 lesions (58.3%) harbored somatic missense variants in the PIK3CA gene, with mutation frequencies ranging from 0.4 to 15.4%. The specific mutations identified included PIK3CA p.H1047R (NM_006218.2:21/21:c.3140 A > G:p.[His1047Arg]) in 72 lesions (24.9%), PIK3CA p.E545K (NM_006218.2:10/21:c.1633G > A:p.[Glu545Lys]) in 39 lesions (13.4%), PIK3CA p.E542K (NM_006218.2:10/21:c.1624G > A:p.[Glu542Lys]) in 31 lesions (10.7%), PIK3CA p.H1047L (NM_006218.2:21/21:c.3140 A > T:p.[His1047Leu]) in 7 lesions (2.4%), and PIK3CA p.C420R (NM_006218.2:8/21:c.1258T > C:p.[Cys420Arg]) in 6 lesions (2.1%). Additionally, 14 lesions (4.8%) contained other PIK3CA mutations, including PIK3CA p.V344G (NM_006218.2:5/21:c.1031T > G:p.[Val344Gly]), PIK3CA p.L911F (NM_006218.2:19/21:c.2733G > C:p.[Leu911Phe]), PIK3CA p.H1047Y (NM_006218.2:21/21:c.3139 C > T:p.[His1047Tyr]), PIK3CA p.G1049S (NM_006218.2:10/21:c.3146G > A:p.[Glu1049Ser]), PIK3CA p.R108H (NM_006218.2:2/21:c.323G > A:p.[Arg108His]), PIK3CA p.G1049R PIK3CA: NM_006218.2:21/21:c.3145G > C:p.[Gly1049Arg]), PIK3CA p.E707K PIK3CA: NM_006218.2:14/21:c.2119G > A:p.[Glu707Lys)], PIK3CA p.R951H PIK3CA: NM_006218.2:20/21:c.2852G > A:p.[Arg951His)], PIK3CA p.N345K PIK3CA: NM_006218.2:5/21:c.1035T > A:p.[Asn345Lys)], PIK3CA p.E453K PIK3CA: NM_006218.2:8/21:c.1357G > A:p.[Glu453Lys)], PIK3CA p.L764P PIK3CA: NM_006218.2:15/21:c.2291T > C:p.[Leu764Pro)], PIK3CA p.P449S PIK3CA: NM_006218.2:8/21:c.1345 C > T:p.[Pro449Ser)] and PIK3CA p.M1004I (NM_006218.2:21/21:c.3012G > A:p.[Met1004Ile]). Somatic missense variants in MAP3K3 p.I441M (NM_002401.3:13/16:c.1323 C > G:p.[Ile441Met]) were identified in 74 lesions (25.5%), with mutation frequencies ranging from 0.2 to 18.7%. Furthermore, 19 lesions (6.6%) exhibited loss-of-function mutations in KRIT1, CCM2, and PDCD10, including truncations, stop gained, splice donor variant and other frameshift variants, with mutation frequencies ranging from 19.73 to 66.3%. Of the 19 lesions, 10 lesions also carried PIK3CA somatic mutations (Fig. 1). These mutations were mutually exclusive with MAP3K3 mutations. In total, pathogenic mutations above were identified in 201 out of the 290 samples (69.3%), highlighting the significant genetic diversity within the cohort (Fig. 1).

Fig. 1

Mutational subgroups in 201 CCMs. Detailed information about the clinical characteristics, mutational landscape, and allele frequencies of the mutations in CCMs

Correlation with clinical phenotypes

Of the 201 CCM samples with identified pathogenic mutations, we further correlated the clinical phenotype with genotype (Fig. 1). Regarding the initial presentation patterns, 99 (49.3%) patients experienced hemorrhage as their first symptom, 50 (24.9%) patients experienced with non-hemorrhage epilepsy, 30 (14.9%) patients presented with symptoms of non-hemorrhage dizziness or headache, 16 (8.0%) patients presented with non-hemorrhagic focal neurological deficits (Table 1). Patients with single PIK3CA mutations showed a higher tendency to be hemorrhage as the initial symptom of CCM diagnosis compared to the single MAP3K3 mutations (P < 0.001) and the MAP3K3 & PIK3CA double mutations group (P < 0.001). Patients with single PIK3CA mutations tended to experience multiple hemorrhage events than MAP3K3 & PIK3CA double mutations (P < 0.001) but no statistical significance compared with single MAP3K3 mutations (P = 0.449). For onset of non-hemorrhage epilepsy, patients with single MAP3K3 mutations (P = 0.012) and MAP3K3 & PIK3CA double mutations (P < 0.001) were more than single PIK3CA mutations. The distribution of headache and local neurological functional impairment does not show statistical significance among different mutation types of CCMs (Table 2).

Table 1 Clinical characteristics of CCMs with somatic mutations and germline mutations. IQR, interquartile range; SD, standard deviation

Table 2 Clinical characteristics of the sporadic CCMs. IQR, inter quartile range; SD, standard deviation

For MRI appearance, single PIK3CA mutations lesions with Zabramski I type were more than single MAP3K3 mutations (P < 0.001) and coexistence with PIK3CA mutations (P < 0.001). CCMs’ maximum diameter of the lesion with single PIK3CA mutations were longer than MAP3K3 & PIK3CA double mutations (P = 0.007) but no statistical significance compared with single MAP3K3 mutations (P = 0.152). PIK3CA mutants were more likely to have associated DVAs on MRI compared to single MAP3K3 mutants (P < 0.001). And CCMs with single PIK3CA mutations seemed to be edema in MRI than double mutations (P < 0.001) but no statistical significance compared with single MAP3K3 mutations (P = 0.185) (Table 2).

Among the 182 patients, epilepsy was observed in 67 cases (36.8%). To identify factors associated with epilepsy events, we conducted binary logistic regression analyses. In univariable analyses, we found that Zabramski classification (unadjusted odds ratio [OR] = 5.5, 95% CI: 2.8–10.7, P < 0.001), lesion location (unadjusted OR = 39.4, 95% CI: 9.2-168.6, P < 0.001), age of onset (adjusted OR = 1.0, 95% CI: 0.9-1.0, P < 0.001), maximum diameter (unadjusted OR = 0.5, 95% CI: 0.3–0.8, P = 0.002), and mutational subgroups with MAP3K3 & PIK3CA double mutations (unadjusted OR = 3.8, 95% CI: 1.9–7.6, P < 0.001) were associated with epilepsy. In Multivariable analyses, age of onset (adjusted OR = 1.0, 95% CI: 0.9-1.0, P < 0.001), maximum diameter of lesions (adjusted OR = 0.5, 95% CI: 0.3–0.9, P = 0.014) and locations (adjusted OR = 43.7, 95% CI: 9.9-210.1, P < 0.001) were independent risk factors for epilepsy (Table 3).

Table 3 Factors impacting epilepsy events of sporadic CCMs. IQR, interquartile range, CI, confidence interval; OR, odds ratio; SD, standard deviation

To shed further light on the impact of mutational subgroups in the clinical characteristics of symptomatic CCMs, we conducted a comparative analysis of clinical characteristics between patients (n = 182) in a sporadic form with somatic mutations in MAP3K3 and PIK3CA, and those (n = 19) in a familial form with germline mutations in KRIT1, CCM2, PDCD10. Notably, patients with germline mutations displayed an earlier onset of symptoms (P = 0.004) and larger size of lesion (P = 0.016) compared to those in a sporadic form with somatic mutations (Table 1).

Histopathologic findings between different mutations

We performed paraffin embedding, sectioning, and staining (including HE, Masson’s trichrome, and immunohistochemistry) on 14 surgical specimens. Histopathological analysis revealed diverse lesions with varying genotypes, correlating with their hemorrhagic characteristics and MRI appearance (Fig. 2). Lesions with single PIK3CA p.H1047R and p.C420R mutations were characterized by cyst walls filled with cystic fluid. Fresh blood cells were present within the cyst walls, indicating recurrent hemorrhages and continuous high intensity on T1-weighted MRI. These walls were primarily composed of sinusoidal collagen arrangements with interstitial spaces largely lacking endothelial-lined lumina. Within these sinusoidal collagen structures, numerous newly formed microvessels are observed, some displaying ruptured luminal shapes, with red blood cells present both inside and outside the vessels. These ruptured microvessels may serve as the source of hematoma fluid and contribute to the progressive enlargement of CCMs (Fig. 2). Some lumina, particularly dilated capillaries within the cyst wall, showed overexpression of the downstream pS6 protein from the PI3K-AKT-mTOR pathway within the endothelial cytoplasm. This overexpression of pS6 in neovascular tissue reflects the dynamic aggressiveness of lesions with single PIK3CA p.H1047R and p.C420R mutations (Fig. 3. A). Conversely, lesions with single MAP3K3 mutations and MAP3K3 & PIK3CA double mutations exhibited typical CCM pathology, characterized by numerous irregularly dilated vascular lumina, abundant collagenous tissue, and calcification (Fig. 2). Some lumina, particularly those in vascular tissues, were encased by dense collagenous tissue, which appeared to confine the hemorrhage (Fig. 2), and the endothelial layers expressed SMA, potentially indicating endothelial-mesenchymal transition (EMT) cells associated with collagen secretion. KLF4 was detected in the nuclei of endothelial cells or EMT cells within abnormally dilated lumina (Fig. 3. B & C). Lesions with PIK3CA p.H1047R and p.C420R mutations showed a notable reduction in KLF4-positive cells than double mutations (P < 0.001) and EndMT (SMA positive) cells compared to lesions with double mutations (P < 0.05) and MAP3K3 mutations (P < 0.001), although all types exhibited Type I collagen. The expression of pS6 in PIK3CA p.H1047R and p.C420R mutations lesions were more than MAP3K3 mutations (Fig. 3. D).

Fig. 2

MRI appearance, pathological features, and hemorrhage risk across different mutational subgroups of CCMs. The typical MRI manifestations and gross pathological appearance of hematoxylin and eosin (H&E), Masson’s trichrome, and immunohistochemistry staining for COL1A1 and CD31 of among different mutational subgroups. PIK3CA mutations typically manifest as circular hematomas on MRI, classified as Zabramski type (I). Pathologically, these lesions are characterized by sinusoidal collagen structures enclosing the hematoma fluid. High-magnification images reveal newly formed microvessels within the collagen, filled with red blood cells both inside and outside the vessels, with some exhibiting rupture, potentially serving as the source of the hematoma fluid. In contrast, patients with MAP3K3-PIK3CA mutations exhibit a characteristic popcorn-like appearance on MRI, corresponding to Zabramski type (II). Pathologically, these lesions display typical CCM changes, including more extensive and irregularly dilated vessels surrounded by a thin collagen layer, along with erythrocyte accumulation within the vessels. Scale Bars 1000 μm. The dashed box (Scale Bar 50 μm) indicates the original image magnified twenty times.

Fig. 3

Representative immunohistochemical staining of cases from different mutational subgroups of CCMs. (A-C) The SMA, pS6, and KLF4 immunohistochemical staining on paraffin-embedded samples of cavernous malformations with different mutations. All three mutation types exhibited expression of SMA in the innermost layer of the vascular lumen. In PIK3CA and MAP3K3-PIK3CA mutations, pS6 was expressed in the cytoplasm of endothelial cells of newly formed microtubules, whereas KLF4 was expressed in the nuclei of endothelial cells lining the mature, larger lumens in cases of MAP3K3 mutations or PIK3CA-MAP3K3 mutations. (D) Quantification of KLF4, pS6 and SMA expression. Data are shown as mean ± SEM; *P < 0.05 with paired t test; **P < 0.001 with t test. The dashed box (Scale bar 20 μm) indicates the original image (Scale bar 100 μm) magnified four times

PIK3CA p.H1047R mutations tend to be hemorrhage

PIK3CA p.H1047R (72/169) and PIK3CA p.E545K (39/169) were identified as the most frequent mutations in surgically resected CCMs with PIK3CA mutations, located at distinct mutation sites (Fig. 4). Notably, patients with the PIK3CA p.H1047R mutation, which resides in the kinase domain of PIK3CA, exhibited a significantly higher risk of bleeding compared to those with the PIK3CA p.E545K mutation, located in the helical domain of PIK3CA (P = 0.007). The distribution of somatic mutation types in PIK3CA across core functional domains of the p110α protein and their mutation frequencies in CCMs are also presented, highlighting the prevalence and significance of these mutations within the context of the protein’s functional structure (Fig. 4). Additionally, a significant difference in bleeding risk was observed when comparing carriers of the PIK3CA p.H1047R mutation to those with other PIK3CA mutations (excluding p.H1047R) (P = 0.013). These findings indicate that the PIK3CA p.H1047R mutation is associated with a higher risk of bleeding compared to other mutation types within the PIK3CA gene. Notably, among the 72 patients with PIK3CA p.H1047R mutations, a lower mutation frequency was associated with the presence of bleeding (unadjusted OR = 0.6, 95% CI: 0.5–0.8, P < 0.001).

Fig. 4

Distribution of somatic mutation types in PIK3CA across core functional domains of the p110α protein and their mutation frequencies in CCMs. Common mutations are highlighted in pink, while mutations shown in gray have not been reported in the literature to date, and their impact on the function of p110α remains uncertain. The domains include the p85B regulatory subunit binding domain, C2 (calcium-dependent phospholipid-binding domain), helical domain (Pl3K helical domain), and kinase domain (Pl3/4-kinase domain)

Discussion

Our study provides a comprehensive analysis of the clinical, genomic, and histopathologic characteristics of cerebral cavernous malformations (CCMs), offering valuable insights into how genetic mutations shape their clinical presentations and imaging findings. By analyzing 290 surgical specimens from symptomatic CCM patients, we were able to correlate distinct mutational subgroups with specific clinical and radiological features, significantly contributing to the understanding of this heterogeneous disease. One of the key findings of this study is the strong association between somatic mutations in MAP3K3 and PIK3CA with different clinical outcomes. Specifically, PIK3CA mutations were significantly associated with an increased risk of hemorrhage, especially in patients harboring the PIK3CA p.H1047R mutation. This finding suggests that PIK3CA-driven CCMs may follow a more aggressive clinical course, which could inform clinical decision-making and risk stratification. In contrast, patients with MAP3K3 mutations and MAP3K3 & PIK3CA double mutations exhibited a higher prevalence of non-hemorrhagic epilepsy, highlighting the varying impact of different genetic alterations on clinical outcomes. Radiologically, the study revealed correlations between mutation types and Zabramski classifications, lesion size, and associated findings like DVA and edema on MRI. These associations underscore the importance of considering both clinical and imaging characteristics when assessing CCM patients, as they may reflect underlying genetic differences that influence disease progression. Histopathologically, we observed distinct differences in the morphology of CCM lesions across different mutational subgroups. PIK3CA mutations were associated with cystic lesions characterized by pS6-positive dilated capillaries and fresh hemorrhage within cyst walls, while MAP3K3 and double mutation groups displayed classic CCM pathology with dilated vascular lumina, abundant collagen, and calcification. This phenotypic diversity further strengthens the idea that underlying genetic differences can influence not only clinical behavior but also the structural properties of the lesions. The reduction in KLF4-positive and SMA-positive cells in PIK3CA-mutated lesions points to a differential regulation of endothelial-to-mesenchymal transition (EndMT), which may be a key driver of lesion morphology and clinical presentation. Interestingly, pS6 expression in PIK3CA were more than MAP3K3 mutations, while the variability in KLF4 and SMA expression among lesions with different mutations suggests that these factors may serve as potential biomarkers for distinguishing CCM subtypes. The observed reduction in EndMT in PIK3CA mutations, compared to MAP3K3 & PIK3CA double and MAP3K3 mutations, raises the possibility of targeted therapies that address these molecular differences.

Our previous scRNA-seq data identified a diverse array of cell types within CCM lesions, including endothelial cells, mural cells, mast cells, astrocytes, B cells, natural killer (NK) cells, dendritic (DC) cells, oligodendrocytes, microglia, macrophages/monocytes, epithelial cells, and T cells. By sorting CD31 + and CD31- cells, we detected somatic mutations in MAP3K3 and PIK3CA, with these mutations predominantly localized in endothelial cells. Furthermore, we observed comparable mutation frequencies of MAP3K3 and PIK3CA in patients harboring double mutations, suggesting that these mutations may originate from the same cellular population [12, 15]. This hypothesis is strongly supported by findings from a recent study [24]. Using single-nucleus DNA sequencing, the authors provided direct evidence that MAP3K3 and PIK3CA mutations co-exist within the same cell in patients with double mutations. These findings substantiate the notion that a single-cell origin may underlie the pathogenesis of CCMs in patients with co-occurring mutations, offering critical insights into the molecular mechanisms driving lesion development.

Recent evidence on CCM and gain-of-function (GOF) mutation in PIK3CA confirmed the growth of CCM lesion and hemorrhagic conversion similar to cancer driving cell proliferation [13]. Our previous preclinical study revealed that endothelial cells in samples with PIK3CA mutations displayed a reduced rate of apoptosis [12]. This observation suggests that CCM lesions with PIK3CA mutations may be more susceptible to vascular obstruction. However, there is currently no certain evidence explaining why a single PIK3CA mutation increases the risk of bleeding. Notably, CCMs manifest in early-born mice following the loss of CCM genes in endothelial cells, but CCM formation in adult mice requires the activation of PI3K signaling [25, 13]. This underscores the crucial role played by PIK3CA in both the development and hemorrhagic events associated with CCMs.

Our research indicates that the H1047R mutation is associated with a higher risk of bleeding compared to other mutation types, including more common variants such as E545K, E542K, and C420R. Different mutation sites affect specific functional domains of PIK3CA, which encodes the p110α protein (Fig. 4). Studies in tumors have demonstrated that the H1047R mutation in the kinase domain enhances membrane binding by disrupting the self-inhibitory effect of the C-terminal tail and altering the conformation of the critical H917 residue [29,30,31,32,33]. This conformational shift significantly increases PI3K activation, leading to robust activation of downstream pathways. However, similar findings have not yet been reported in CCMs. Regarding mutation frequency and the presence of bleeding, we hypothesize that patients with hemorrhage experience more severe endothelial cell damage, which may contribute to the observed lower mutation frequency.

The emerging understanding of pathological mutations in CCMs holds promise for targeted therapies, particularly in CCMs with PIK3CA mutations, which represent an aggressive phenotype in patients [12, 16, 13, 24, 17]. Inhibitors targeting PI3K-AKT-mTOR pathways, such as the PI3Kα inhibitor Alpelisib and the mTOR inhibitor rapamycin, have been widely used in PIK3CA-related overgrowth syndromes and lymphatic malformations [26,27,28]. Rapamycin treatment has shown efficacy in reducing lesions in preclinical models with CCM1 and PIK3CA mutations [14, 13]. However, since PIK3CA mutations in CCMs are somatic mutations, they cannot be detected by blood samples, and detecting these somatic mutations still relies on genetic testing of surgically removed lesions. This limitation diminishes the potential benefit of targeted therapy for patients. The main strength of our study lies in identifying preoperative clinical characteristics of CCMs correlated with mutational subgroups. Specifically, MRI appearances characterized by a longer maximum diameter and grouped as Zabramski I category were more likely to be associated with PIK3CA mutations. These findings suggest that the radiological appearance of CCM lesions can serve as a reliable predictor for both CCM mutational subgroups, providing valuable insights for clinical management and facilitating the selection of suitable patients for targeted therapy.

This study has several limitations. Selection bias may have influenced the results, as the genetic data used in this study were derived from surgical specimens. This source of genetic data may have introduced bias, as it predominantly represents patients with symptoms or surgical indications. Consequently, there may be a significantly lower proportion of mutational subgroups associated with milder phenotypes, such as MAP3K3 mutations. Patients with these milder phenotypes may not experience severe symptoms and, as a result, may be less likely to seek medical attention compared to those with PIK3CA mutations or double mutations. Additionally, patients with MAP3K3 mutations have a relatively lower risk of bleeding, while PIK3CA mutations are associated with a higher risk of bleeding. This leads to MAP3K3 mutants being more likely to be hospitalized and surgically treated for epilepsy, whereas PIK3CA mutants are more likely to seek treatment due to bleeding, particularly in the brainstem. As a result, more MAP3K3 mutants tend to have lesions in the frontal and temporal lobes, while PIK3CA mutants typically present with lesions in deeper regions or the brainstem. Furthermore, due to the high surgical risk and recurrent episodes associated with multiple lesions carrying CCM germline mutations, these patients are often managed conservatively. Consequently, there were few surgical specimens available for analysis from patients with germline mutations. Furthermore, the referral bias to a tertiary medical center may have resulted in the overrepresentation of patients with more severe clinical presentations. Additionally, it should be noted that our cohort primarily consisted of Asian patients, which could limit the generalizability of our findings to other populations. Therefore, there is a need for an international multicenter cohort study to further investigate the correlation between mutational subgroups and clinical classifications in CCMs in the future.

Conclusions

Our study underscores the diverse clinical, genomic, and histopathologic features of CCMs, and demonstrates the potential to predict mutational subgroups based on clinical and MRI characteristics. The association between mutational subgroups and clinical presentations may have significant implications for identifying lesions amenable to targeted therapies.