SnRNA‐seq reveals cellular heterogeneity and proliferation mechanisms in limb venous malformations

Abstract

Background

Venous malformations (VMs) are congenital vascular anomalies characterized by abnormal vascular proliferation, with limb VMs often leading to functional impairment and physical discomfort. However, the cellular heterogeneity and underlying molecular mechanisms driving pathological proliferation in limb VMs remain incompletely elucidated.

Methods

In this study, we collected 10 tissue samples including specimens from 5 limb VM patients and 5 normal control tissues and performed single‐nucleus RNA sequencing (snRNA‐seq) to comprehensively map the cellular landscape of VMs. We first identified distinct cell subpopulations covering vascular endothelial cells, vascular smooth muscle cells, and immune cells, and compared gene expression levels between VM and normal tissues. Afterward, we conducted weighted gene co‐expression network analysis (hdWGCNA) and constructed the protein–protein interaction (PPI) network to screen critical proliferation‐related genes in VMs. We further analyzed the signaling pathways associated with these candidate genes and carried out subsequent functional validation experiments to explore the biological role of core gene TEK in proliferative vascular endothelial cells (PVECs). Besides, we also analyzed the characteristic pathways of the PVEC subpopulation to clarify its proliferation‐related molecular features.

Results

We found obvious expression differences of genes in various cell subpopulations between VM and normal tissues. Three genes, namely tyrosine protein kinase receptor (TEK), Fms‐like tyrosine kinase 1 (FLT1), and EGF‐like domain multiple 7 (EGFL7), were identified as key proliferation‐related genes with significant upregulation in VM lesions, and these three genes were closely associated with the activation of PI3K/AKT/mTOR, IL6/JAK/STAT3, and TNF‐α/NF‐κB pathways. Functional experimental results showed that TEK knockdown could significantly inhibit proliferative vascular endothelial cell (PVEC) proliferation and promote cell apoptosis, and reverse the abnormal activation of inflammatory pathways. As a vital pathogenic cell subpopulation, PVECs facilitated abnormal vascular proliferation through activating pathways including the G2/M checkpoint and E2F targets.

Conclusions

Collectively, our study systematically elucidated the cellular heterogeneity framework and proliferation mechanisms of limb VMs, identifying TEK, FLT1, and EGFL7 as key regulators of pathological proliferation. These findings provide new insights into the pathogenesis of VMs and lay a foundation for developing precise therapeutic strategies targeting proliferation‐related pathways.

To dissect the cellular heterogeneity and invasive mechanisms of limb venous malformations (VMs), this study first obtained tissue samples from four patients with VMs and four normal controls (NC). Single‐nucleus suspension was prepared, followed by transcriptome library construction and sequencing. After pretreatment, quality control, standardization, and dimension reduction clustering, marker gene analysis and cell type annotation were performed. Enrichment analysis of differentially expressed genes (DEGs) and intercellular communication analysis were then performed, along with pseudotemporal trajectory and transcription factor analysis. High‐dimensional weighted gene coexpression network analysis (hdWGCNA) was used to construct networks and screen key invasion‐related genes, identifying FLT1, EGFL7, and TEK as core genes using Cytohubba/MCODE. These genes were validated in 15 VMs and 15 NC tissue samples using reverse transcription quantitative polymerase chain reaction (RT‐qPCR), immunohistochemistry (IHC), and Western blot. A short hairpin RNA targeting TEK (shTEK) was constructed. In vitro experiments, combined with Western blot and immunofluorescence (IF), demonstrated that targeting TEK‐regulated pathways, including OXPHOS, PI3K/AKT/mTOR, RAP1, MAPK, KRAS, IL6/JAK/STAT3, and TNF‐α/NF‐κB, may provide insights into VM pathology and potential therapeutic targets.

1. INTRODUCTION

Venous malformations (VMs) are the most common type of congenital vascular malformation, arising from abnormal development of venous vessels during embryonic growth. ¹ , ² Limb VMs, in particular, exhibit high morbidity, often manifesting as localized swelling, pain, and functional limitations, significantly affecting patients' quality of life. ³ , ⁴ Current treatment strategies, including surgical resection and sclerotherapy, have limitations such as high recurrence rates and potential surgical trauma, highlighting the urgent need to explore new therapeutic targets based on pathological mechanisms. Previous studies have shown that the abnormal proliferation of vascular endothelial cells and vascular smooth muscle cells (VSMC) is the core pathological feature of VMs. ⁵ , ⁶ , ⁷ , ⁸ However, the cellular heterogeneity and key molecular regulators driving this abnormal proliferation remain unclear. With the rapid development of single‐cell omics technologies, single‐nucleus RNA sequencing (snRNA‐seq) has become a powerful tool for dissecting cellular heterogeneity and identifying key disease‐related genes in complex tissues. ⁹ Compared to traditional single‐cell RNA sequencing, snRNA‐seq more effectively captures difficult‐to‐dissociate cell types, such as VSMC and endothelial cells (EC), and preserves more complete transcriptomic information. ⁹ , ¹⁰ This study utilized snRNA‐seq to perform high‐precision profiling of limb VM tissues, aiming to reveal the heterogeneity characteristics of EC and VSMC in VMs, identify key molecular mechanisms driving disease proliferation, and provide a theoretical basis for developing precise therapeutic targets.

Previous studies have shown that the pathological process of VMs involves the abnormal activation of multiple signaling pathways that not only regulate cell proliferation and migration ¹¹ but also participate in angiogenesis and extracellular matrix (ECM) remodeling. ¹² However, the specific regulatory patterns and interaction networks of these pathways across different cellular subpopulations in VMs remain unclear. Additionally, the heterogeneous differentiation trajectories of EC and VSMC in VM tissues and their relationship with disease proliferation require in‐depth exploration. ¹³ Through integrating snRNA‐seq, pseudotemporal trajectory analysis, and transcription factor regulatory network parsing, this study systematically delineated the heterogeneity landscape of EC and VSMC in VMs and revealed the core role of key subpopulations, such as proliferative vascular endothelial cell (PVEC) in disease progression.

Findings showed that the proportions of EC and VSMC were significantly reduced in VM tissues, whereas the proportion of PVEC among EC subpopulations increased notably. PVEC highly expressed proliferation‐related genes (e.g., MKI67 and TOP2A) and promoted abnormal proliferation by activating pathways such as the G2/M checkpoint and E2F targets. Meanwhile, VSMC subpopulations exhibited distinct functional heterogeneity, with reduced proportions of contractile VSMC and increased synthetic and pericyte‐like VSMC, suggesting that phenotypic transformation of VSMC may contribute to VM pathogenesis via ECM remodeling and vascular tension dysregulation. Importantly, this study identified tyrosine‐protein kinase receptor (TEK), Fms‐like tyrosine kinase 1 (FLT1), and EGF‐like domain multiple 7 (EGFL7) as core driver genes of VM proliferation through high‐dimensional weighted gene coexpression network analysis (hdWGCNA) and protein–protein interaction (PPI) network analysis. These genes were significantly upregulated in VM tissues and closely associated with excessive activation of pathways such as phosphatidylinositol 3‐kinase (PI3K)/protein kinase B (AKT)/mammalian target of rapamycin (mTOR), interleukin‐6 (IL6)/janus kinase (JAK)/signal transducer and activator of transcription 3 (STAT3), and tumor necrosis factor‐α (TNF‐α)/nuclear factor kappa‐B (NF‐κB). In vitro experiments further confirmed that TEK knockdown significantly inhibited the proliferative capacity of PVEC and reversed the abnormal activation of inflammation‐ and apoptosis‐related pathways. This discovery not only reveals the potential value of TEK as a therapeutic target for VMs but also provides a new perspective for understanding the molecular mechanisms of VMs. Additionally, intercellular communication analysis showed markedly enhanced interactions among EC, VSMC, and fibroblasts in VM tissues, with the CD99 signaling pathway playing a critical role in communication between ECs and macrophages, potentially promoting disease progression by regulating the local immune microenvironment (Figure 1).

FIGURE 1.

Flow chart.

In this study, we employed snRNA‐seq to profile the cellular composition and gene expression patterns of limb VM tissues and normal control tissues. We aimed to (1) delineate the cellular landscape of limb VMs and identify key cell subpopulations involved in pathological proliferation; (2) screen key proliferation‐related genes and elucidate their associated signaling pathways; (3) validate the functional role of core genes in regulating vascular cell proliferation. Through integrated analysis, we identified TEK, FLT1, and EGFL7 as key genes associated with VM pathological proliferation, providing a theoretical basis for developing targeted therapies for VMs.

These achievements not only fill gaps in the understanding of the cellular and molecular mechanisms of VMs but also lay a solid foundation for developing cell subpopulation‐specific precision therapeutic strategies in the future. In summary, this study deeply elucidates the cellular heterogeneity and proliferative molecular mechanisms of VMs through multiomics integrative analysis and functional experiments, providing new targets and insights for clinical treatment. Future research will further explore the downstream effector molecules of the TEK regulatory network and their therapeutic potential in animal models, promoting the translation of VM basic research into clinical applications.

2. MATERIALS AND METHODS

2.1. Sample source

Tissue specimens were collected from patients with VMs who underwent surgical resection at the Third Affiliated Hospital of Zhengzhou University between September 2021 and September 2023. Malformed venous tissues served as the experimental (VMs) group, whereas normal small veins obtained from distal resection margins were used as the normal control (NC) group. Inclusion and exclusion criteria were defined for both groups, with all VM tissues meeting the diagnostic criteria of the International Society for the Study of Vascular Anomalies (ISSVA). ¹⁴ , ¹⁵ , ¹⁶ After screening, 19 specimens were eligible (Table S1), including 4 cases each from VMs and NC groups for snRNA‐seq analysis to screen key genes and 15 cases each for validation of key genes and signaling pathways. Sample collection was approved by the Ethics Committee of the Third Affiliated Hospital of Zhengzhou University (2024‐279‐01) and obtained with written informed consent from patients or guardians.

2.2. Tissue sampling and preparation of single‐cell nuclear suspensions

All samples were collected under strict aseptic conditions. After sampling with sterile tissue scissors, impurities were removed by washing with phosphate‐buffered saline (PBS). Tissues were cut and aliquoted into enzyme‐free tubes according to experimental requirements, transported under liquid nitrogen, and stored in a refrigerator at −80°C. For tissue pretreatment, frozen samples were placed in a mixture of 1‐mL precooled lysis buffer and 1‐mL PBS, cut into ~0.1‐cm³ fragments, washed twice with PBS, and centrifuged to discard the supernatant. Then, 2‐mL precooled lysis buffer was added for prelysis on ice for 2 min. The mixture was transferred to a precooled homogenizer, gently homogenized five to six times, and incubated on ice for 6 min. The reaction was terminated by adding 2 mL of precooled 4% bovine serum albumin (BSA) solution, followed by pipetting six to eight times with a Pasteur pipette to mix. After centrifugation, the pellet was resuspended in lysis buffer containing 4% BSA and incubated on ice for 3 min. Cell debris was removed following the standard protocol of the Miltenyi Nuclei Purification Kit, and the supernatant was discarded using centrifugation. The pellet was resuspended in nuclear preservation solution, filtered through a 20‐μm cell strainer, centrifuged, and the supernatant was discarded. The nuclei were counted for subsequent use. Ten microliters of the nuclear suspension was mixed with 0.4% trypan blue staining solution, allowed to stand for 3 min, and observed using a CountStar automated cell counter. Nuclear suspensions with ≥80% intact nuclei were used for subsequent sequencing and analysis.

2.3. Single‐cell nuclear transcriptome library construction and sequencing

The process of nuclear suspension loading and generation of gel beads‐in‐emulsion (GEMs) is as follows: the quality‐controlled single‐nucleus suspension (pretreated with lysis buffer containing 0.5% NP‐40 to remove cell membranes) is mixed with 10× barcoded gel beads and enzyme mixture, and GEMs are generated using chromium microfluidic chip technology. Each GEM encapsulates a single nucleus and a gel bead. Within the droplet, the nuclear membrane is lysed by heating to release nuclear RNA. The oligonucleotide chains fixed on the gel bead surface, which contain cell barcode and unique molecular identifiers (UMIs), bind to the nuclear RNA. Then, the barcoded first‐strand complementary DNA (cDNA) is synthesized by reverse transcriptase at 42°C. Subsequently, the GEM droplet structure is disrupted by chemical demulsification to release and mix the cDNA products, which are enriched by 12‐cycle polymerase chain reaction (PCR) amplification.

During the construction of the sequencing library, the amplified products are fragmented to 200–300 bp using Covaris sonication, followed by end repair. Illumina sequencing adapters are ligated, and sequencing primer binding sites are introduced using secondary PCR to construct a standard sequencing library. After the library fragment distribution (Agilent 2100) and concentration (Qubit HS) meet the standards, the libraries are pooled for sequencing. Finally, PE150 paired‐end sequencing is performed on the Illumina NovaSeq 6000 platform. The raw data are parsed for cell barcode and UMI information using CellRanger software to generate a single‐nucleus gene expression matrix for subsequent bioinformatics analysis.

2.4. Data preprocessing and quality control

For raw data processing, Illumina's official software bcl2fastq version 5.0.1 was first used to convert BCL‐format raw sequencing data into FASTQ files. The 10× Genomics official tool Cell Ranger version 5.0.1 was then employed for gene expression quantification and preliminary filtering, including mapping data to the human reference genome GRCh38/hg38, filtering low‐quality nuclei with <500 detected genes or <1000 total UMI counts, and separating samples by sequencing indices. Subsequently, Seurat version 4.3.0 was used for rigorous quality control (QC): retaining nuclei with ≥500 detected genes, excluding nuclei with mitochondrial gene proportion ≥ 25%, filtering potential multicell droplets using DoubletFinder. Finally, the QC‐qualified data were normalized to eliminate sequencing depth differences, and 2000 highly variable genes were selected for subsequent analysis.

2.5. Data standardization and dimensionality reduction clustering analysis

The NormalizeData function in Seurat was used to standardize gene expression values and remove technical variations. The top 2000 highly variable genes were screened using FindVariableFeatures, and their expression values were further standardized using the ScaleData function to provide normalized data for dimensionality reduction analysis. Based on the standardized expression matrix of highly variable genes, principal component analysis (PCA) was performed to calculate the gene contributions of principal components (PCs). The ElbowPlot function was used to visualize the variance contribution rate curve of PCs to determine the PCs at the inflection point. The FindClusters function was applied for unsupervised clustering of all cells to generate major cell clusters. EC and VSMC were screened based on marker gene annotation, and secondary subcluster analysis was performed for each population. Finally, combined with the PCA dimensionality reduction results, RunUMAP was used for uniform manifold approximation and projection (UMAP) to display the low‐dimensional spatial distribution of cell populations.

2.6. Marker gene analysis and cell type annotation

Using Seurat's FindMarkers function, differential expression analysis was performed for each cell cluster to identify genes that are significantly highly expressed compared to all other cell populations (i.e., marker genes). Concurrently, canonical marker genes for the cell types under study were retrieved from the literature. Marker genes for each subcluster were then determined based on their known roles in cell biology and the functions of differentially expressed genes within each cluster. Cell populations were manually annotated based on the established functions of these marker genes. A DotPlot was generated to visualize the expression levels of multiple feature genes across different cell clusters, illustrating their distribution patterns.

2.7. Screening for differentially expressed genes and enrichment analysis

Differentially expressed genes (DEGs) in single‐nucleus transcriptomes were screened using the FindMarkers function combined with the Wilcoxon rank‐sum test. The criteria for significance were set as log₂ fold change (log₂FC) ≥0.26, expression in ≥10% of nuclei, and FDR ≤0.05. Results were visualized via volcano plots and used for subsequent functional enrichment analysis. Functional annotation of DEGs was performed using the compareCluster function from the clusterProfiler package for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. The top 10 significant terms were presented in bar and bubble charts. Gene set enrichment analysis (GSEA) was conducted by ranking genes based on log₂FC and using the gseGO function from the fgsea package with the MSigDB database to identify key driver pathways (false discovery rate [FDR] < 0.25). For screening key modules associated with EC proliferation, GO and KEGG enrichment analyses were performed on VM‐related modules using the DAVID database (version 6.8). This process identified critical modules closely linked to the proliferative properties of VMs.

2.8. Intercellular communication analysis

To explore the mechanisms of intercellular communication in VMs, this study comprehensively applied two methods: CellChat and CellPhoneDB. For CellChat, the createCellChat function was first used to generate an object using the counts matrix of the Seurat object as the gene expression matrix, combined with cell annotations and grouping variables. The runCellChat function was then employed to construct the communication network. Subsequently, plotCellChat was used to draw the network diagram with parameter adjustments, and exportCellChat was applied to export relevant scores and pathway information. Finally, the Benjamini–Hochberg method was used for testing, and interactions with an adjusted p < 0.05 and log₂ average expression value >0.1 were determined as significant, visualized using Circlize and igraphR. For CellPhoneDB, cells were first clustered using Seurat, with cell types annotated and gene expression matrices and cell type annotation formats converted. The cellphonedb:run_statistical_analysis function was then used to evaluate communication strength, performing 1000 permutation tests. The Benjamini–Hochberg method was applied for screening, retaining ligand‐receptor interactions with adjusted p < 0.05 and average log expression values >0.1. Finally, igraph was used to draw the interaction network diagram.

2.9. Pseudotemporal trajectory analysis and transcription factor analysis

This study employed Monocle2 (version 2.22.0) and pySCENIC (version 0.12.1) for trajectory analysis and single‐cell transcription factor regulatory network parsing, respectively. The trajectory analysis workflow was as follows: an unnormalized gene expression matrix and cell phenotype information were extracted from the Seurat object to construct a CellDataSet object, followed by library size correction and gene dispersion estimation. Highly variable genes were filtered, and dimensionality reduction was performed using DDRTree. Cells were ordered using pseudotime, and root nodes were identified. DEGs (q < 0.05) were screened based on the pseudotime model, and branch‐specific genes were identified using branch point analysis (BEAM function, p < 0.01). Finally, dynamic cell distributions and gene expression patterns were visualized using various plotting functions. The workflow for parsing the transcription factor regulatory network was as follows: The GENIE3 algorithm was used to infer regulatory relationships between transcription factors (TFs) and target genes, retaining the top 50 000 high‐confidence connections. Binding motifs were identified using RcisTarget to construct regulons. The AUCell algorithm was applied to calculate regulon activity scores and generate a binarized matrix. Activity scores were mapped to the UMAP space for visualization, and subgroup‐specific regulatory networks were identified based on the Wilcoxon rank‐sum test (p < 0.05).

2.10. Construction of hdWGCNA and screening of key proliferative genes

EC subpopulations were isolated from snRNA‐seq data, retaining genes expressed in ≥30% of cells. A gene similarity matrix was constructed using Pearson's correlation coefficients, and a weighted coexpression network was built using the blockwiseModules function (β = 7), converting it into an adjacency matrix. Topological overlap measure (TOM) was calculated to quantify connectivity, followed by hierarchical clustering. Modules were defined using the dynamic tree‐cutting algorithm (minimum module size: 30 genes, merge threshold: 0.25, cutting height: 0.25). Module eigengenes (MEs) were extracted using the PCA, and their correlations with disease groups (VMs vs. NC) were computed. Modules with significant correlations (p < 0.05) were visualized using dot plots. The top 30 hub genes per module were selected based on connectivity to MEs (≥0.7), and intermodule correlations were analyzed using ModuleCorrelogram. A PPI network was constructed for the top 30 genes (using the Module Eigengene connectivity value (KME) value) from four key modules (120 genes total) using the STRING database (version 12.0, confidence >0.4). After visualization in Cytoscape 3.9.1, the top 15 hub genes were identified using CytoHubba, and proliferative‐related interaction subnetworks were detected with the MCODE algorithm. Candidate genes were defined as the intersection of these two gene sets, and their differential expression between VMs and NC groups was validated using Seurat's VlnPlot.

2.11. Reverse transcription quantitative polymerase chain reaction

Fifteen patients' samples (approximately 45–60 mg) from both the VMs and NC groups were placed into sterile, enzyme‐free centrifuge tubes. Using sterile tissue scissors, the samples were cut into small pieces, followed by the addition of three steel beads and 1 mL of Trizol reagent for immersion. The samples were then ground at 65 Hz for 20 min until homogenized. Next, 200 μL of chloroform was added to each tube, vortexed for 30 s, and left to stand on ice for 10 min. After centrifugation at 12 000 rpm for 15 min at 4°C, 400 μL of the upper aqueous phase was carefully aspirated. An equal volume of isopropanol was added to the supernatant, mixed by inverting, and allowed to stand at room temperature for 10 min. The RNA was pelleted by centrifugation at 12 000 rpm for 10 min at 4°C, washed with 1 mL of 75% ethanol to remove residual salt ions by pipetting, and centrifuged again. The supernatant was removed, and the RNA pellet was air‐dried at room temperature for 10 min before being dissolved in 25 μL of Diethylpyrocarbonate (DEPC) ‐treated water.

During RNA reverse transcription and cDNA quantification, the purity and concentration of RNA were measured using a NanoDrop 2000, and only samples with an A260/A280 ratio between 1.8 and 2.0 were selected. Reverse transcription was performed by mixing the RNA solution, reverse transcription reagents, and enzyme‐free water, followed by incubation at 37°C for 15 min, 50°C for 5 min, and 98°C for 5 min. For fluorescence quantitative PCR, gene primers (Table 1), fluorescent quantitative premix reagents, and enzyme‐free water were prepared into a 10‐μL reaction system. The PCR program consisted of predenaturation at 95°C for 10 min, followed by 40 cycles of denaturation at 95°C for 15 s, annealing at 58°C for 25 s, and extension at 72°C for 35 s. The relative messenger RNA (mRNA) expression levels of TEK, FLT1, and EGFL7 were calculated using the 2^‐ΔΔCT method.

TABLE 1.

Primer sequence.

Gene	Species	Primer sequence
TEK	Human	Forward primer	5′‐CCTTGGCTCTGCTGGAATGA‐3′
		Reverse primer	5′‐GCTGGTTCTTCCCTCACGTT‐3′
FLT1	Human	Forward primer	5′‐TTCCGAAGCAAGGTGTGACT‐3′
		Reverse primer	5′‐TTATTGCCATGCGCTGAGTG‐3′
EGFL7	Human	Forward primer	5′‐GCAAGGGCTAGGGTCCATCT‐3′
		Reverse primer	5′‐CAGGGGCTGCTCTGATGCT‐3′
GAPDH	Human	Forward primer	5′‐AGAACGGGAAGCTTGTCATC‐3′
		Reverse primer	5′‐CATCGCCCCACTTGATTTTG‐3′

2.12. Immunohistochemistry

Paraffin‐embedded sections from 15 patients with VM (VMs group) and the NC group were baked at 60°C for 2 h, followed by pretreatment steps, including dewaxing, hydration, Phosphate Buffered Saline with Tween‐20 (PBST) washing, antigen retrieval, and endogenous peroxidase blocking. Sections were then incubated overnight at 4°C (8 h) with rabbit antihuman TEK, FLT1, and EGFL7 antibodies (diluted 1:400, 1:2000, and 1:2000, respectively). After rewarming, sections were rinsed with PBS, incubated with a signal enhancer at 37°C for 20 min, washed again, and probed with a goat antirabbit immunoglobulin G (IgG) secondary antibody at 37°C for 1 h. After DAB chromogenic development and reaction termination with distilled water, nuclei were counterstained with hematoxylin for 40 s, rinsed under running water for 10 min, dehydrated through gradient ethanol, absolute ethanol, and xylene, air‐dried in a fume hood, and mounted with neutral gum. Using an optical microscope, representative venous vessel fields were randomly selected. Brown‐yellow granules in EC cytoplasm indicated positive protein expression. Using ImageJ software, the H‐DAB function was applied for deconvolution, and the average optical density (AOD) was calculated as the ratio of integrated optical density to target area. Statistical graphs were generated to compare protein expression levels between groups.

2.13. Construction of shRNA knockdown of TEK

Short hairpin RNA (shRNA) sequences targeting human TEK were designed and synthesized. The fragments were ligated into the pLKO.1‐puro vector (restriction site: EcoRI; R0101V, NEB, Beijing, China). Positive clones with fully correct sequences were verified by Sanger sequencing (Shangya, China) using the primer: 5′‐TACGATACAAGGCTGTTAGAGAG‐3′. For transfection, human umbilical vein endothelial cells (HUVECs) were seeded in six‐well plates at a density of 2 × 10⁵ cells/well and cultured to 70%–80% confluence. The homo‐TEK‐shRNA‐pLKO.1‐puro plasmid (2 μg/well) was mixed with Lipofectamine 2000 reagent (4 μL/well) in serum‐free DMEM, incubated for 20 min at room temperature, and then added dropwise to the cells. After 6 h of incubation at 37°C with 5% CO₂, the medium was replaced with complete Dulbecco’s Modified Eagle Medium (DMEM) containing 10% fetal bovine serum (FBS). Cells were harvested 48 h posttransfection for subsequent experiments.

2.14. Cell culture

The HUVEC cell line was purchased from Servicebio (China, Wuhan). The cells were cultured in high‐glucose Dulbecco's modified Eagle's medium (DMEM) (Gibco Laboratories, Grand Island, NY, USA), supplemented with 10% FBS, penicillin (100 units/mL), and streptomycin. The cells were incubated at 37°C in a humidified atmosphere with 5% CO₂. The culture medium was refreshed every 2–3 days. The in vitro cellular model of VMs was established by stimulating HUVECs with IGF‐1 (50 ng/mL) combined with EGF (50 ng/mL) for 48 h. This model recapitulates key molecular features identified in human venous malformation tissues, including upregulated TEK/FLT1/EGFL7 expression and aberrant activation of proliferation‐ and inflammation‐related signaling pathways, consistent with the molecular profiles revealed by snRNA‐seq in this study. Therefore, it is suitable for the in vitro functional validation of TEK. The experimental groups included the control group, VMs model group, siTEK group, and VMs model + siTEK group.

2.15. Western blot

Tissue samples and HUVECs were homogenized using a Polytron in ice‐cold Radio‐Immunoprecipitation Assay (RIPA) buffer supplemented with Phenylmethylsulfonyl fluoride (PMSF) (G2002; G2008, Servicebio, China), sonicated, and cleared by centrifugation (12 000 × g, 10 min, at 4°C). Protein concentration in the supernatant was determined using a Bicinchoninic Acid (BCA) assay and separated on sodium dodecyl sulfate‐polyacrylamide gel electrophoresis (SDS‐PAGE) gels and transferred onto nitrocellulose membrane (Millipore, IPFL00010, Germany) by electrophoresis. Blots were blocked in 5% nonfat milk in TBST for 1 h at room temperature and probed with primary antibody, including TEK, FLT1, EGFL7, OXPHOS, IL6, JAK, p‐JAK, STAT3, p‐STAT3, TNF‐α, NF‐κB, p‐NF‐κB, Bax, Casp3, PI3K, p‐PI3K, AKT, p‐AKT, mTOR, p‐mTOR, RAP1, MAPK, p‐MAPK, KRAS, and GAPDH (Affinity, Melbourne) in TBST with 1% nonfat milk overnight at 4°C. After overnight incubation, this was followed by incubation with horseradish peroxidase (HRP)‐conjugated secondary antibody in TBST with 1% nonfat milk for 2 h at room temperature. The blots were developed using an enhanced chemiluminescence assay (BIO‐Rad). Band intensities were quantified using ImageJ software. The intensity of each target protein band was normalized to the corresponding GAPDH band intensity to calculate the relative protein expression level. All Western blot experiments were repeated at least thrice independently.

2.16. Immunofluorescence

The cells underwent the following series of treatments: treatment with 1 × citrate antigen retrieval solution at 98°C for 10 min, followed by washing thrice with PBS; treatment with 0.1% Triton X‐100 transparent for 10 min, followed by washing thrice with PBS; and blocking with 10% goat serum in PBS and overnight incubation with PCNA, TEK, Ki‐67, and Casp3 (1:200, Affinity) antibodies. After being washed thrice with PBS, the HUVECs were incubated with a mixture of Alexa‐fluor 555‐conjugated, Alexa‐fluor 488‐conjugated secondary antibodies (1:2000, Invitrogen) at room temperature for 2 h, and staining of nucleus with DAPI for 10 min. The sections were mounted using VectaShield medium (Vector Laboratories) and subjected to fluorescein detection using a fluorescence microscope (Leica DMI4000B, Wetzlar, Germany). The mean fluorescence intensity of the proteins was analyzed using Image J.

2.17. Statistical methods

All statistical analyses in this study were performed using R 4.4.2 and GraphPad Prism 9.5.0. For single‐nucleus data preprocessing, the Seurat package was used with QC criteria: retaining nuclei with >500 genes and mitochondrial gene proportion ≤ 25%. After normalization, batch effects were corrected using the Harmony package, followed by dimensionality reduction via PCA (top 30 PCs) and UMAP visualization.

DEGs were identified using the Wilcoxon rank‐sum test with thresholds of log₂FC ≥ 0.26 and FDR ≤ 0.05. GO/KEGG enrichment analyses were performed using Fisher's exact test in the clusterProfiler package (p < 0.05). GSEA was performed using the fgsea package to calculate normalized enrichment scores (NES), with significance defined as FDR < 0.25. For intercellular communication analysis (CellChat/CellPhoneDB), ligand‐receptor interactions were evaluated using 1000 permutation tests, requiring adjusted p < 0.05 and mean expression ≥0.1.

Pseudotime analysis (Monocle2) used a generalized linear model to screen DEGs (q < 0.05) and integrated BEAM analysis to identify branch‐specific genes (p < 0.01). Transcription factor analysis employed the SCENIC package to calculate regulon activity scores (AUCell), with differential regulons identified using Wilcoxon test (p < 0.05). For mRNA and protein expression data, GraphPad Prism was used: continuous variables were tested for normal distribution using the Shapiro–Wilk test, presented as mean ± standard deviation ($\overline{x}$ ± s), and compared between two groups using independent samples t‐tests (two‐tailed, α = 0.05). p < 0.05 was considered statistically significant.

3. RESULTS

3.1. Quantitative QC and dimensionality reduction analysis of snRNA‐seq data

After QC filtering, a total of 55 430 cells were captured. The four experimental groups were named VMs1–VMs4, and the four control groups were named NC1–NC4. The total number of nuclei in the VM experimental group was 30 391 (54.83%) and that in the control group was 25 039 (45.17%) (Table S2). Violin plots showed the number of RNA reads, gene counts, and mitochondrial gene ratios per cell before and after QC (Figure 2A). Based on high‐quality snRNA‐seq data, PCA was first used to evaluate data variation characteristics. The results showed that the standard deviations of the first 30 PCs were significantly higher than those of subsequent components, and the variance contribution rate curve stabilized after the 30th PC. Therefore, the first 30 PCs were selected as the core variation dimensions reflecting data differences (Figure 2B).

FIGURE 2.

Single‐nucleus RNA sequencing (snRNA‐seq) data quantitative quality control (QC) and cell type annotation. (A) Distribution of basic information for nuclei across samples after filtering. The X‐axis denotes sample names; left panel: Y‐axis shows the number of genes per nucleus (nFeature RNA), where abnormally low/high values may indicate cell death or contamination. Middle panel: Y‐axis shows the total gene counts (nCount RNA), with low/high values suggesting poor cell quality or abnormalities. Right panel displays the proportion of mitochondrial gene expression (percent. mito), where high values may indicate cellular stress or death. (B) Scree plot of principal component analysis (PCA). The X‐axis represents the number of principal components (PCs), and the Y‐axis shows standard deviations. The standard deviations of PCs stabilize between 30 and 50, so 30 PCs were selected for subsequent analysis. (C) Percentage distribution of eight samples (four venous malformations [VMs] and four normal controls [NC]) across 22 cell clusters. Different colors represent distinct samples; stacked bar plots show the proportional composition of each sample within each cell cluster. (D) Uniform manifold approximation and projection (UMAP) dimensionality reduction results of unsupervised clustering for single‐nucleus transcriptome data. (E) Distribution of eight cell types in UMAP space. (F) Expression heatmap of specific marker genes for each cell type. (G) Cell type proportion plot comparing the VMs and NC groups.

3.2. Cell type annotation and identification

A total of 55 430 nuclei were subjected to unsupervised clustering analysis, resulting in 22 transcriptionally heterogeneous cell clusters (clusters 0–21) (Figure 2C). UMAP was used for two‐dimensional visualization of these clusters, revealing their distribution in low‐dimensional space. All 22 disease‐related clusters were covered by eight samples from four patients (Figure 2D), annotated as eight major cell types: EC, VSMC, fibroblasts, macrophages, mast cells, T cells, skeletal muscle cells, and adipocytes (Figure 2E). EC (clusters 0, 1, 9) expressed VWF, PECAM1, etc. VSMC (clusters 4, 13, 14) showed high expression of ACTA2, MYH11, etc. Fibroblasts (clusters 2, 3, 5) showed highly expressed DCN, COL1A1, etc. Macrophages (cluster 6) expressed CD163, MS4A7, etc. Mast cells (cluster 20) expressed CPA3, TPSB, etc. T cells (cluster 17) expressed CD3E, IL7R, etc. (Figure 2F). Skeletal muscle and adipocytes were excluded from subsequent analyses due to potential contamination. Proportional analysis of cell types between the VMs and NC groups showed that EC, VSMC, and fibroblasts were the predominant types. Notably, the proportions of EC and VSMC were lower in the VMs group than in the NC group (Figure 2G).

3.3. DEGs GSEA and biological function enrichment analysis

Using differential expression analysis of single‐nucleus transcriptome data from VMs and NC samples, this study identified 772 DEGs, including 400 downregulated and 372 upregulated genes significantly associated with the pathogenesis of VMs. A volcano plot shows the top 10 significantly upregulated and downregulated genes (Figure 3A). GSEA results revealed significant enrichment of myogenesis pathways and ECM remodeling‐related pathways (epithelial–mesenchymal transition, protein secretion). Additionally, pathways involved in cell proliferation and angiogenesis regulation networks (PI3K/AKT/mTOR, KRAS, mTORC1, oxidative phosphorylation [OXPHOS], and G2/M checkpoint), transcriptional regulatory modules (MYC_v1 targets, estrogen response late), and inflammation‐related pathways (IL6/JAK/STAT3, TNF‐α/NF‐κB) were significantly enriched (Figure 3B). For biological processes (BP), enriched pathways included cell adhesion, angiogenesis, positive regulation of cell proliferation, ECM organization, and positive regulation of angiogenesis. For cellular components (CC), significant enrichment was observed in collagen‐containing ECM, focal adhesion, and cell junctions. In molecular functions (MF), enriched pathways included ribosomal structural constituent, integrin binding, and actin binding (Figure 3C). KEGG enrichment analysis showed the top 10 significantly enriched pathways (sorted by the p‐value) included PI3K/AKT signaling, focal adhesion, regulation of actin cytoskeleton, ECM‐receptor interaction, MAPK signaling, and Rap1 signaling (Figure 3D). These pathways are closely linked to angiogenesis regulation, proliferation/migration of vascular endothelial cells, and vascular structural remodeling.

FIGURE 3.

Differentially expressed genes (DEGs), gene set enrichment analysis (GSEA), and biological function enrichment analysis. (A) Volcano plot of DEGs between venous malformations (VMs) and normal control (NC) groups. X‐axis: Fold change of gene expression (log₂FoldChange). Y‐axis: Significance level (−log₁₀(p‐value)). Blue dots: Significantly downregulated genes in VMs (p < 0.05 and log₂FoldChange < −1). Red dots: Significantly upregulated genes in VMs (p < 0.05 and log₂FoldChange > 1). Gray dots: Genes with no significant difference (p ≥ 0.05 or |log₂FoldChange| ≤ 1). (B) GSEA enrichment analysis plot. Curves in different colors represent various gene set pathways. X‐axis: Position of genes in the ordered differential expression dataset (rank in ordered dataset). Y‐axis: Running enrichment score (running enrichment score), showing the enrichment trend of each pathway. The peak of each curve indicates the position of core‐enriched genes. (C) Gene Ontology (GO) enrichment analysis of DEGs. Blue bars: Biological processes (BP). Orange bars: Cellular components (CC). Green bars: Molecular functions (MF). Bar height represents the number of enriched genes. (D) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Different color modules correspond to pathway categories. X‐axis: Number of enriched genes.

3.4. Intercellular communication analysis

In the VMs group, interaction frequencies among fibroblasts, VSMC, EC, macrophages, and other cell types significantly increased. Fibroblasts showed the most pronounced changes in interaction frequencies with EC and macrophages (Figure 4A). Communication intensity between EC, fibroblasts, VSMC, macrophages, and other cell types was notably enhanced in the VMs group. EC exhibited the most frequent interactions with fibroblasts and macrophages, whereas interaction intensity between EC and VSMC/fibroblasts increased most significantly (Figure 4B), suggesting that these cell populations form a key interaction network in the pathogenesis of VMs. In the NC group, 113 significant ligand‐receptor pairs were identified across eight cell types, involving 47 signaling pathways, including classic pathways (TGF‐β, VEGF, PDGF, IGF, NOTCH) and vascular development‐related pathways (ANGPT, EPHA/EPHB, ADGRL) (Figure 4C). The VMs group showed an increase in 129 significant ligand‐receptor pairs across 31 signaling pathways, including TGF‐β, VEGF, PDGF, NOTCH, EGF, BMP, SEMA3, ANGPT, laminin, and collagen (Figure 4D). EC‐macrophage interactions: in the NC group, EC primarily signaled via PECAM1‐PECAM1 and APP‐CD74. In the VMs group, CD99‐CD99 homophilic binding and APP‐TNFRSF21 became dominant pathways. This shift may remodel the local immune environment via CD99‐mediated inflammation. Compared to the NC group, the VMs group showed deeper color intensity in the CD99 signaling pathway interaction region (Figure 4E,F), indicating significantly enhanced interaction importance. EC acted as both signal sender and receiver in the CD99 pathway, serving as a core hub for signal transduction and network coordination via its mediator role.

FIGURE 4.

Intercellular communication analysis between normal control (NC) and venous malformations (VMs) groups. (A) Difference in interaction frequency between cell types. Color depth represents the relative change in interaction frequency. (B) Difference in interaction intensity between cell types. Red indicates enhanced interaction in the VMs group, whereas blue indicates weakening. (C, D) Ligand‐receptor pairs between endothelial cells (ECs) and macrophages/vascular smooth muscle cell (VSMC)/fibroblasts in NC and VMs groups. Y‐axis: Ligand‐receptor pair names. X‐axis: Cell types. Color gradient: Communication probability. Black dots: Statistical significance markers, showing communication intensity and significant differences of ligand‐receptor pairs between groups. (E, F) Interaction relationship patterns of CD99 signaling pathway network, with different cell types (EC, macrophages, fibroblasts, etc.) acting as sender, receiver, mediator, and influencer. Color depth corresponds to “importance” values (range: 0–1). Values closer to 1 indicate stronger importance of intercellular interactions, reflecting key differences in cell interaction significance within the CD99 signaling network.

3.5. Heterogeneity analysis of VSMC

Unsupervised clustering of 6225 VSMCs identified 13 transcriptomic subpopulations (clusters 0–12), categorized into three major subtypes based on marker genes and functional characteristics: (1) contractile VSMC1 (clusters 0, 1–4, 11): highly expresses contractile marker genes (ACTA2, MYH11), maintaining vascular tension and contraction; (2) synthetic VSMC2 (clusters 5, 8–10): downregulates contractile genes while highly expressing ECM synthesis genes (e.g., COL1A1, COL3A1), exhibiting a matrix‐secreting phenotype; (3) pericyte‐like VSMC3 (clusters 6, 7, 12): specifically expresses CD36, EBF2, and STEAP4, involved in microvascular stabilization (Figure 5A–C). Statistical analysis (Figure 5D) showed that compared to the NC group, the proportions of pericyte‐like VSMC3 (30.98% vs. 6.57%) and synthetic VSMC2 (26.48% vs. 14.17%) significantly increased in the VMs group, whereas the proportion of contractile VSMC1 significantly decreased (42.54% vs. 79.26%). Pseudotime analysis revealed the differentiation trajectory of VSMC subpopulations (Figure 5E), and dynamic gene expression patterns were summarized into three modules (Figure 5F). Among key genes, TPM2 expression decreased with pseudotime in Module 2 (pink), whereas TMSB4X, DES, EEF1A1, and S100A6 showed significant expression decline in Module 3 (green) (Figure 5G).

FIGURE 5.

Heterogeneity analysis of vascular smooth muscle cell (VSMC). (A) Uniform manifold approximation and projection (UMAP) dimensionality reduction in VSMC subpopulations in the normal control (NC) group. (B) UMAP dimensionality reduction in VSMC subpopulations in the venous malformations (VMs) group. (C) Expression levels of specific marker genes for each VSMC subpopulation. (D) Proportional distribution of VSMC subpopulations in NC and VMs groups. (E) Developmental trajectory and subpopulation distribution of VSMC along the pseudotime dimension. (F) Expression heatmap of pseudotime‐related differentially expressed genes (DEGs). Color intensity represents expression levels (blue = low, red = high). (G) Expression trends of the top six significantly changed genes during pseudotime progression.

3.6. Heterogeneity analysis of EC

3.6.1. Dimensionality reduction, clustering, and subpopulation annotation of EC

This study identified 11 subpopulations from 19 418 ECs (Figure 6A), all expressing common markers such as PECAM1 and VWF. After QC excluding cluster 2 with high mitochondrial gene expression, ECs in VMs were classified into five types: venous EC (VEC, clusters 0, 1, 4, 6, 9): marked by NR2F2. Arterial EC (AEC, cluster 5): marked by NOTCH4. Capillary EC (CEC, cluster 3): marked by CD31. Lymphatic EC (LEC, clusters 7, 8): marked by PROX1. PVEC (cluster 10): marked by MKI67/TOP2A (Figure 6B,C). Statistical analysis showed that compared to the NC group, the VMs group exhibited a significant increase in the proportion of PVEC (2.61% vs. 0.85%) and VEC (77.43% vs. 65.03%), whereas the proportions of LEC (1.87% vs. 7.7%), CEC (13.64% vs. 18.53%), and AEC (4.45% vs. 7.89%) significantly decreased (Figure 6D,E; Table 2).

FIGURE 6.

Endothelial cell (EC) subpopulation clustering uniform manifold approximation and projection (UMAP) plots and proportion analysis. (A) UMAP plot of EC dimensionality reduction, showing spatial distribution characteristics of cell subpopulations. (B) Visualized UMAP distribution of five EC subpopulations. (C) Dot plot of subpopulation marker gene expression, with dot size and color indicating expression frequency and level. (D) UMAP plots of EC subpopulations in venous malformations (VMs) (red) and normal control (NC) (dark blue) groups, highlighting intergroup differences in cell distribution. (E) Bar chart of EC subpopulation proportions in VMs and NC groups.

TABLE 2.

Percentage of different cell types.

Cell type (%)	PVEC	VEC	LEC	CEC	AEC
NC1	0.92	74.04	3.66	15.83	5.55
NC2	0.12	28.58	42.48	9.24	19.58
NC3	0.99	43.95	18.19	24.96	11.91
NC4	1.21	81.22	4.46	10.68	2.43
VM1	1.17	86.48	2.86	6.1	3.39
VM2	2.76	53.21	8.23	24.16	11.63
VM3	1.34	71.58	3.39	18.77	4.92
VM4	3.87	85.59	4.66	4.57	1.31

Note: NC1–NC4: four control groups; VM1–VM4: four venous malformations groups.

Abbreviations: AEC, arterial endothelial cell; CEC, capillary endothelial cell; LEC, lymphatic endothelial cell; PVEC, proliferative vascular endothelial cell; VEC, venous endothelial cell.

3.6.2. DEGs analysis of EC populations

A total of 534 upregulated genes and 426 downregulated genes were identified (Figure 7A). After GO and KEGG enrichment analyses of all DEGs: GO biological processes revealed significant enrichment in pathways related to angiogenesis, positive regulation of cell population proliferation, cell differentiation, actin cytoskeleton organization, positive regulation of transcription by RNA polymerase II, cell–cell adhesion, and positive regulation of cell migration (Figure S1A). KEGG pathways showed significant enrichment in metabolic pathways, pathways in cancer, focal adhesion, calcium signaling pathway, PI3K‐AKT signaling pathway, MAPK signaling pathway, Ras signaling pathway, and Rap1 signaling pathway (Figure S1B).

FIGURE 7.

Differentially expressed genes (DEGs) analysis and functional heterogeneity of endothelial cell (EC) populations. (A) Scatter plot of DEGs between EC groups. Red dots: Genes with significant intergroup differences between venous malformations (VMs) and normal control (NC) groups. Gray dots: Genes without significant differences. (B) Intergroup expression analysis of MKI67 and TOP2A in proliferative vascular endothelial cell (PVEC). Violin plot showing expression differences of MKI67 and TOP2A in PVEC between VMs and NC groups. X‐axis: Cell subpopulations; Y‐axis: Gene expression levels. (C,D) Gene set enrichment analysis (GSEA) of PVEC subpopulation. (C) X‐axis: Rank of genes by expression difference (rank in ordered dataset). Y‐axis: Running enrichment score. Different color curves represent various gene set pathways. (D) Top 30 enriched pathways ranked by p‐value. X‐axis: Normalized enrichment score (NES). Y‐axis: Gene set pathways. Color gradient (blue to red) indicates p‐value from high to low.

3.6.3. Functional heterogeneity of EC subpopulations

Through clustering analysis of 11 EC subclusters (clusters 0–10, 19 418 nuclei total), this study defined functional characteristics of two key subpopulations: CEC and PVEC. CEC highly expressed a series of genes, including fibroblast marker genes (e.g., COL15A1), ECM‐related genes (e.g., COL3A1), extracellular region‐related genes, and angiogenesis/developmental genes (e.g., ROBO2). GO enrichment analysis showed significant enrichment in cell migration and extracellular space (Figure S1C), indicating potential roles in cell migration and proliferation during VMs pathogenesis. PVEC significantly overexpressed proliferation‐related genes (e.g., MKI67, TOP2A) (Figure 7B), DNA repair genes (e.g., FANCA), and cell division‐related genes (e.g., RRM2, CENPK). These genes promote abnormal cell proliferation by regulating the cell cycle, maintaining genomic stability, and driving chromosome segregation. GO analysis further revealed enrichment in cell cycle, cell division, and DNA repair (Figure S1D). GSEA identified core pathways, including the G2/M checkpoint, mitotic spindle, and E2F targets, involving mTORC1 and MYC_V1 signaling pathways (Figure 7C,D). Collectively, abnormal expression of proliferation and cell division‐related genes in PVECs disrupts critical biological processes such as cell cycle regulation and DNA repair. This not only activates proliferation‐related pathways (e.g., G2/M checkpoint, E2F targets) but also enhances cell invasiveness (e.g., migration, matrix degradation) via mTORC1 and MYC_V1 signaling, indicating that this subpopulation plays a key role in VM initiation and proliferative progression.

3.6.4. Pseudotime analysis of EC subpopulation

Using the Monocle 2 algorithm, a pseudotime axis for EC differentiation was constructed (Figure 8A), identifying five cell states (State) (Figure 8B). Analysis showed that EC in VMs exhibited two major differentiation directions originating from State 1. VMs samples had EC primarily distributed in States 1, 4, and 5, with differentiation paths differing from normal venous EC: PVEC were enriched in States 4–5, whereas CEC concentrated in State 1 (Figure 8C–E). Based on dynamic gene expression patterns, significantly changed genes were classified into three modules (Figure S2A): Module 1 (green) genes (e.g., CD34, NOS1AP) were downregulated along pseudotime. Module 2 (pink) genes (e.g., VAV3, CHRM3‐AS2) showed upregulated expression. Key genes exhibited expression trends consistent with subpopulation differentiation: downregulation of CD34 aligned with CEC differentiation, whereas the upregulation of VAV3 and CHRM3‐AS2 coincided with PVEC differentiation (Figure S2B).

FIGURE 8.

Pseudotime analysis trajectory plots of endothelial cell (EC). (A) Cell differentiation trajectory based on pseudotime, visualizing developmental progression from initial to terminal states. (B) State‐based trajectory plot, categorizing cells into discrete transitional states during differentiation. (C) Trajectory plot based on annotated cell types, color‐coded by EC subpopulations (e.g., venous endothelial cell [VEC], proliferative vascular endothelial cell [PVEC]; capillary endothelial cell [CEC]). (D) Group‐based trajectory plot, comparing differentiation paths between venous malformations (VMs) and normal control (NC) groups. (E) Differentiation trajectory distribution by EC type, showing lineage‐specific developmental patterns.

3.6.5. Intercellular communication between PVEC, CEC, and VSMC subpopulations/fibroblasts

The EC subsets exhibit distinct characteristics: PVEC shows proliferative traits, whereas CEC is associated with ECM remodeling and migration. Using CellPhoneDB, this study analyzed interactions between PVEC/CEC and other cell types in malformed veins of VMs (Figure 9A,B). Results showed VSMC primarily interacted with PVEC via ANGPT2‐TEK (Figure 9C), with no significant difference in interaction intensity among three VSMC subpopulations. PVEC engaged with VSMC through ligand‐receptor pairs such as JAG1‐NOTCH3 and DLL1‐NOTCH3. In contrast, CEC exhibited stronger interactions with fibroblasts: CEC interacted with fibroblasts via TGFB1‐TGFBR3. Fibroblasts reciprocally interacted with CEC through complex ligand‐receptor pairs, including FN1‐integrin α5β1 and FBN1‐integrin α5β1 (Figure 9D).

FIGURE 9.

Intercellular communication and transcription factor analysis. (A) Heatmap of interaction intensity between endothelial cell (EC) subsets (proliferative vascular EC [PVEC], capillary EC [CEC]) and vascular smooth muscle cell (VSMC)/fibroblasts. (B) Network diagram of interaction frequencies. (C, D) Interaction profiles of PVEC and CEC with other cell types, respectively. (E) Distribution of transcription factor AUC values in uniform manifold approximation and projection (UMAP) space. (F) Density analysis of transcription factor AUC distribution.

3.6.6. Transcriptional regulatory analysis of the PVEC subpopulation in EC

To identify the key regulatory factors governing the PVEC cell population—a distinct EC subset—this study employed the SCENIC algorithm to analyze transcription factors (TFs) potentially driving EC heterogeneity in VMs tissues. By calculating regulatory activity scores (RAS), we successfully distinguished regulon activity differences among five EC subpopulations (Figure S3A). Ranking regulon specificity scores (RSS) in VMs revealed elevated regulatory activity of HMGB2, BRCA1, RUNX1, PBX3, and EZH2 in PVECs, with RSS values of 0.28, 0.24, 0.23, 0.21, and 0.21, respectively. UMAP dimensionality reduction analysis based on area under the curve (AUC) values of 24 regulons, stratified by cell type, showed that EZH2 exhibited high specificity among EC‐associated TFs in UMAP visualizations integrating AUC values, gene set activity, and expression (Figures S3B and 9E,F).

3.7. Identification of key proliferation‐related genes via hdWGCNA

As shown in Figure S4A, when the scale‐free topology fitting index R ² reached 0.85 with a soft threshold (β) of 7, the connectivity curves (average, median, and maximum) plateaued, indicating optimal network connectivity. A total of 18 gene modules were identified (Figure S4B), and their correlation matrix is presented in Figure S4C. Six modules exhibited elevated expression in the VMs experimental group (Figure S4D). GO and KEGG enrichment analyses revealed that four modules (red, salmon, cyan, turquoise) (Figure S5A) were significantly enriched in angiogenesis, integrin‐mediated signaling pathways, vascular development, focal adhesion (Figure S5B), as well as PI3K‐AKT, focal adhesion, and RAP1 signaling pathways (Figure S5C), suggesting their involvement in VM pathogenesis and proliferation. The midnight blue and green‐yellow modules were primarily associated with macrophage‐derived foam cell differentiation, lysosome function, and other inflammation‐related pathways (Figure S5D). The gray module, containing unclustered genes, was excluded from further analysis.

3.8. Validation of key proliferation‐related gene expression

Based on four modules (Figure S5A), CytoHubba analysis identified 15 key genes associated with the occurrence and proliferation of VMs: CTNNB1, CRK, YWHAE, FLT1, CBL, MAP3K1, FOXO1, TEK, MCL1, ASAP1, VAV3, MACF1, ATG7, GRB10, and EGFL7. Concurrently, MCODE analysis highlighted TEK, FLT1, and EGFL7 as critical genes. The intersection identified TEK, FLT1, and EGFL7 as key candidate genes significantly upregulated in VMs. Validation using single‐cell RNA sequencing data revealed that TEK, FLT1, and EGFL7 were specifically expressed in EC, with significantly higher expression levels in the VMs group compared to the NC group (Figure 10A,B). RT‐qPCR further confirmed elevated mRNA expression of TEK, FLT1, and EGFL7 in the VMs group (Figure 10C–E). Immunohistochemistry demonstrated positive cytoplasmic staining (brown or dark brown) for TEK, FLT1, and EGFL7 in EC, with notably higher protein expression in the VMs group (Figure 10F–I). Western blot analysis corroborated these findings, showing increased protein levels of TEK, FLT1, and EGFL7 in the VMs group (Figure 10J,K).

FIGURE 10.

Validation of key proliferation‐related genes. (A, B) Expression of TEK, FLT1, and EGFL7 across different cell types. A: Normal control (NC) group; B: Venous malformations (VMs) group. The X‐axis denotes distinct cell types, whereas the Y‐axis represents gene expression levels. (C–E) Reverse transcription quantitative polymerase chain reaction (RT‐qPCR) results comparing the VMs and NC groups reveal significant differences in the relative messenger RNA (mRNA) expression levels of TEK, FLT1, and EGFL7. n = 15. (F) Immunohistochemical staining for TEK, FLT1, and EGFL7 in the VMs and NC groups. n = 15. Scale bar = 20 μm. (G–I) Average optical density (AOD) values for TEK, FLT1, and EGFL7. (J) Expression of TEK, FLT1, and EGFL7 proteins. n = 4. (K) Statistical plot of expression of TEK, FLT1, and EGFL7 proteins. Compared to the NC group, **p < 0.01, ***p < 0.001.

3.9. Knockout and validation of key genes

Figure S6A displays the map of the homo‐TEK‐shRNA‐pLKO.1‐puro plasmid. The constructed plasmid was verified by sequencing, and the results showed that the sequence was correct (Figure S6B), and the silent effect was validated using Western blot analysis (Figure S6C,D). The results indicated the successful establishment of the shTEK15 cell line. Compared to the control group, the expression of PCNA and TEK proteins was significantly upregulated in the VMs model group. In contrast, after intervention, the expression of PCNA and TEK proteins was decreased in the VMs model + shTEK group compared to the VMs model group (Figure 11A,B).

FIGURE 11.

Effects of shTEK on venous malformations (VMs) models. (A) Expression of PCNA and TEK proteins. (B) Statistical plot of expression of PCNA and TEK proteins. (C) PCNA and TEK immunofluorescence detection. Scale bar = 50 μm. (D) Statistics plot of mean fluorescence intensity of PCNA and TEK. (E) Effects of shTEK on OXPHOS complex (CI–CV) proteins. n = 3. Compared to the control group, **p < 0.01, ***p < 0.001, ns = no significance; compared to the VMs model group, ^## p < 0.01; compared to the shTEK group, ^△△ p < 0.01, ^△△△ p < 0.001.

Meanwhile, immunofluorescence detection of PCNA and TEK protein colocalization showed that the average fluorescence intensity of PCNA and TEK proteins was enhanced in the VMs model group compared to the control group, whereas it was weakened in the VMs model + shTEK group compared to the VMs model group (Figure 11C,D). Colocalization of Ki‐67 and Casp3 revealed that the average fluorescence intensity of Ki‐67 was enhanced and that of Casp3 was weakened in the VMs model group compared to the control group. Conversely, compared to the VMs model group, the average fluorescence intensity of Ki‐67 was weakened and that of Casp3 was enhanced in the VMs model + shTEK group (Figure S6E,F).

In addition, the expression of OXPHOS complex (CI–CV) proteins decreased in the VMs model group compared to the control group, whereas it increased in the VMs model + shTEK group compared to the VMs model group (Figure 11E). In conclusion, shTEK affects the expression of proliferation‐related proteins such as PCNA and Ki‐67, cell apoptosis, and OXPHOS proteins in the VMs model, suggesting its potential regulatory role in the pathological process of VMs.

3.10. Signal pathway analysis

This study conducted an in‐depth analysis of signal pathways enriched based on key genes, exploring the effects of shTEK on the PI3K/AKT/mTOR, RAP1, MAPK, KRAS, IL6/JAK/STAT3, and TNF‐α/NF‐κB signaling pathways. Compared to the control group, the VMs model group showed increased expression of p‐PI3K, p‐AKT, p‐mTOR, RAP1, p‐MAPK, KRAS, IL6, p‐JAK, p‐STAT3, TNF‐α, and p‐NF‐κB, while inhibiting cell apoptosis (reduced expression of Casp3 and Bax). When compared to the VMs model group, the VMs model + shTEK group exhibited decreased expression of the aforementioned pathway proteins and increased expression of apoptosis‐related proteins (Figure 12). In summary, shTEK can regulate multiple signaling pathways in VMs models: it inhibits the overactivation of inflammatory pathways (PI3K/AKT/mTOR, IL6/JAK/STAT3, TNF‐α/NF‐κB), modulates apoptosis‐related protein expression (Casp3, Bax), and affects cell proliferation (PCNA, Ki‐67). These findings suggest that targeting the TEK gene may be a potential strategy to interfere with the pathological progression of VMs by regulating key signaling pathways.

FIGURE 12.

Effects of shTEK on signaling pathways in venous malformation (VMs) models. (A) Expression of PI3K, p‐PI3K, AKT, p‐AKT, mTOR, p‐mTOR, RAP1, MAPK, p‐MAPK, and KRAS proteins. (B) Statistical plot of expression of p‐PI3K, p‐AKT, p‐mTOR, RAP1, p‐MAPK, and KRAS proteins. (C) Expression of IL6, JAK, p‐JAK, STAT3, p‐STAT3, TNF‐α, NF‐κB, p‐NF‐κB, Bax, and Casp3 proteins. (D) Statistical plot of expression of IL6, p‐JAK, p‐STAT3, TNF‐α, p‐NF‐κB, Bax, and Casp3 proteins. n = 3. Compared to the control group, ***p < 0.001, ns = no significance; compared to the VMs model group, ^# p < 0.05, ^## p < 0.01, ^### p < 0.001; compared to the shTEK group, ^△ p < 0.05, ^△△ p < 0.01, ^△△△ p < 0.001.

4. DISCUSSION

This study utilized snRNA‐seq technology to systematically reveal for the first time the cellular heterogeneity landscape of limb VMs and the molecular mechanisms underlying their proliferation. The main findings are as follows: (1) pro‐angiogenic PVECs were identified as the core pathogenic cell type in VMs; (2) it was confirmed that VSMC undergo a phenotypic transition from a contractile to a synthetic/pericyte‐like state, participating in the pathological process of VMs through ECM remodeling; (3) TEK, FLT1, and EGFL7 were identified as key genes upregulated in VMs and linked to proliferation‐related pathway activation, including PI3K/AKT/mTOR, IL6/JAK/STAT3, and TNF‐α/NF‐κB; (4) The critical role of CD99‐mediated interactions between EC and macrophages in microenvironmental reprogramming was revealed; (5) experimental validation showed that targeting the TEK gene effectively reversed the pathological phenotype in VMs models. These findings provide a new perspective for deepening the understanding of VMs pathogenesis and developing precise therapeutic strategies.

In this study, a PVEC subpopulation was identified in VMs, with a significantly higher proportion (2.61%) in VM tissues compared to normal tissues (0.85%) and high expression of proliferation‐related genes (such as MKI67, TOP2A, and FANCA). PVECs promote abnormal cell proliferation by activating pathways like the G2/M checkpoint and E2F targets, a feature similar to proangiogenic endothelial cells in the tumor microenvironment. Additionally, PVECs are mainly enriched in the terminal states (State 4–5) of the EC differentiation trajectory, suggesting that they may originate from abnormal differentiation of VECs. In contrast, CECs participate in ECM remodeling and cell migration through high expression of genes such as COL15A1 and ROBO2, consistent with the phenotype of disordered vascular structures in VMs.

The study found significant changes in the proportion of VSMC subpopulations in VM tissues: the proportion of contractile VSMCs decreased from 79.26% to 42.54%, whereas the proportions of synthetic (26.48% vs. 14.17%) and pericyte‐like VSMCs (30.98% vs. 6.57%) significantly increased. Synthetic VSMCs highly express ECM synthesis‐related genes (such as COL1A1 and COL3A1), disrupting vascular tension homeostasis through the secretion of abnormal matrix; the expansion of pericyte‐like VSMCs (CD36, EBF2) may affect microvascular stability. Pseudotime analysis further confirmed downregulated expression of contractile markers (such as TPM2 and DES) in the VSMC differentiation trajectory, indicating that the “contractile‐synthetic phenotypic transition” plays a key role in the pathological process of VMs, similar to the pathogenic mechanism of VSMC phenotypic transition in atherosclerosis.

Through hdWGCNA and PPI network analysis, this study screened TEK (TIE2), FLT1 (VEGFR1), and EGFL7 as core genes driving VM proliferation. Single‐cell data showed that these three genes are specifically highly expressed in endothelial cells, with significantly upregulated mRNA and protein levels in VM tissues. As a tyrosine kinase receptor, TEK mediates ANGPT2 signaling. This study confirmed that knocking down TEK (shTEK) inhibits PVEC proliferation, promotes cell apoptosis, and reverses the expression of OXPHOS complexes, suggesting that TEK influences VM progression by regulating cellular energy metabolism.

TEK, FLT1, and EGFL7 are individually associated with the activation of three key pathways: the PI3K/AKT/mTOR pathway, where shTEK treatment significantly inhibits the phosphorylation of p‐PI3K, p‐AKT, and p‐mTOR, and this pathway is known to be involved in endothelial cell survival and vascular permeability regulation in vascular malformations ¹⁷ , ¹⁸ ; inflammatory pathways, including IL6/JAK/STAT3 and TNF‐α/NF‐κB, whose excessive activation is associated with pain and swelling phenotypes in VMs ¹⁹ , ²⁰ ; and the RAP1/MAPK pathway, with increased expression of RAP1, p‐MAPK, and KRAS, possibly mediating ECM‐cell adhesion disorders through integrin signaling. ²⁰ , ²¹ Upon activation, TEK directly recruits and phosphorylates downstream adaptor proteins to trigger the PI3K/AKT/mTOR signaling cascade, thereby regulating cell proliferation, survival, and metabolism. The present study demonstrates that TEK knockdown markedly reduces the phosphorylation levels of PI3K, AKT, and mTOR, indicating a direct signaling linkage between TEK and these pathways rather than a simple correlation. Furthermore, TEK knockdown suppresses the excessive activation of inflammatory pathways, including IL6/JAK/STAT3 and TNF‐α/NF‐κB, which may be indirectly mediated by attenuating the pro‐inflammatory microenvironment remodeled by dysfunctional endothelial cells. These results further establish the core driver role of TEK in the pathogenic network of VMs.

Using CellChat and CellPhoneDB to analyze the VM cell interaction network, this study discovered the EC‐macrophage CD99 signaling axis: CD99‐CD99 homophilic interactions in VMs replace the PECAM1‐PECAM1 interactions in normal tissues. As an immunomodulatory molecule, CD99 promotes leukocyte extravasation and inflammatory responses, and enhanced CD99 signaling may exacerbate the local inflammatory microenvironment by recruiting macrophages. ²² The EC‐VSMC‐fibroblast triangular network: PVEC interact with VSMC through ANGPT2‐TEK and JAG1‐NOTCH3, activating the Notch pathway to participate in angiogenesis; CECs interact with fibroblasts through TGFB1‐TGFBR3 and FN1‐integrin α5β1, driving ECM deposition and vascular fibrosis.

The theoretical innovations include the first application of snRNA‐seq technology to analyze the VM cell landscape, overcoming the limitation of low capture efficiency for VSMC and ECs in traditional single‐cell RNA sequencing; identification of PVEC as a new therapeutic target cell population, whose high expression of TEK provides a molecular basis for precise intervention; and clarification that TEK regulates VM proliferation via multiple signaling pathways, whereas FLT1 and EGFL7 are also significantly upregulated in the disease. TEK inhibitors (such as Rebastinib) have been used in clinical trials for the treatment of TEK mutation‐related VMs. ²³ , ²⁴ This study further confirms that knocking down TEK can simultaneously inhibit cell proliferation, promote apoptosis, and reverse the excessive activation of inflammatory pathways, providing experimental support for TEK‐targeted drug therapy. Future research could explore PVEC‐specific delivery of TEK inhibitors, combined blockade of the PI3K/AKT and IL6/JAK pathways, and the use of CD99 antagonists to regulate the immune microenvironment.

This study also has limitations: the snRNA‐seq analysis in this study was performed on four patients with VMs and four control subjects, representing a relatively small snRNA‐seq cohort. Although we employed stringent quality control, batch correction, and robust bioinformatic methods, the limited sample size may still reduce the statistical power for detecting rare cell populations and subtle expression changes. All core conclusions of this study have been repeatedly validated in an independent cohort of 15 pairs of clinical samples using RT‐qPCR, immunohistochemistry, and Western blot analysis, which significantly enhances the reliability of our findings. In addition, it is further discussed that as extremity VM is a rare disease, conducting large‐scale single‐nucleus sequencing is challenging. Future multicenter studies with larger cohorts are warranted to further verify the universality of the PVEC subpopulation, VSMC phenotypic transition, and the core TEK regulatory network. This study only verified the function and regulatory mechanism of TEK in in vitro cell models, without establishing in vivo functional verification using animal models of VMs (such as TEK‐mutant mice, organoids, or patient‐derived xenograft models). Therefore, the therapeutic potential of TEK as an intervention target still needs to be further confirmed at the in vivo level. We established an in vitro cell model of VMs by treating HUVECs with a combination of IGF‐1 and EGF. This model is rational and reliable for several reasons: IGF‐1 and EGF signaling pathways are significantly activated in human VM tissues; this model stably recapitulates the core pathological features of VMs, including abnormal endothelial proliferation, enhanced migration, ECM remodeling, and upregulation of TEK, FLT1, and EGFL7, which are highly consistent with the snRNA‐seq results in this study. Nevertheless, this model has certain limitations, as it cannot fully mimic the complex multicellular microenvironment, abnormal vascular structure, and three‐dimensional structural disorders of human VM tissues in vivo. Future studies will utilize patient‐derived VM ECs, organoid models, and TEK‐mutant animal models to further validate the pathogenic mechanisms and regulatory network of TEK, as well as assess the efficacy and translational value of TEK‐targeted interventions, thereby providing more robust experimental evidence for clinical translation.

5. CONCLUSION

This study systematically elucidated the cellular heterogeneity framework and proliferative driving mechanisms of VMs through multiomics integrative analysis: The PVEC subpopulation drives abnormal vascular proliferation; TEK, FLT1, and EGFL7 are significantly upregulated and associated with this process, which synergistically activates PI3K/AKT/mTOR, inflammatory signaling, and ECM remodeling pathways. This process is further facilitated by VSMC phenotypic transformation and CD99‐mediated reprogramming of the immune microenvironment. TEK may serve as a promising candidate target for precision intervention of VMs, pending further in vivo validation. Future research should further promote the clinical translation of TEK‐targeted therapies and explore cell subpopulation‐specific intervention approaches.

AUTHOR CONTRIBUTIONS

Junjie Lin: Writing – original draft. Tingting Liu: Writing – original draft. Bin Fang: Formal analysis. Xiaojuan Feng: Methodology. Wenting Jiao: Software; validation. Changkuan Chen: Resources. Yaqing Ding: Software. Gaozan Zhu: Investigation. Wenqiu Wang: Resources; software. Wenbo Liu: Formal analysis. Yuanqi Li: Software. Shoufu Hou: Data curation. Jianshe Wei: Funding acquisition; supervision. Junbo Qiao: Funding acquisition; supervision.

FUNDING INFORMATION

This work was supported partly by National Natural Science Foundation of China (32161143021, 81271410), Henan University graduate “Talent Program” of Henan Province of China (SYLYC2023092), Henan Natural Science Foundation of China (182300410313), and Key Research and Development Project of Henan Province (231111311400).

CONFLICT OF INTEREST STATEMENT

The authors have declared that no competing interests exist.

ETHICS STATEMENT

The study was approved by the Ethics Committee of the Third Affiliated Hospital of Zhengzhou University (approval no.: 2024‐279‐01) and written informed consent was obtained from patients or guardians.

Supporting information

Figure S1. Endothelial cell (EC) subpopulation differentially expressed genes (DEGs) enrichment analysis. (A) Gene Ontology (GO) enrichment analysis of EC DEGs. The bar chart illustrates functional enrichment in three GO categories: biological processes (BP), cellular components (CC), and molecular functions (MF). (B) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of EC DEGs. (C) GO enrichment bubble plot of capillary EC (CEC) subpopulation. (D) GO enrichment analysis of proliferative vascular EC (PVEC) subpopulation. The horizontal axis represents the rich factor, whereas the vertical axis denotes significantly enriched GO terms. Bubble size indicates gene count, and color gradient reflects p‐values (darker red signifies smaller p‐values).

Figure S2. Pseudotemporal gene expression heatmap and dynamic gene plots of endothelial cells (EC). (A) The heatmap displays the expression profiles of the top 50 genes in ECs along the pseudotime axis. (B) Dynamic expression plots of the top six pseudotemporal genes. Each point represents an individual cell, color‐coded by EC subpopulation type. The Y‐axis indicates gene expression levels, and the X‐axis denotes pseudotime.

Figure S3. Transcriptional regulatory analysis of proliferative vascular endothelial cell (PVEC) subsets. (A) Heatmap of regulon activity in EC subsets. NC is the control group, and VM is the experimental group. The figure shows the regulon activity regulated by transcription factors (TFs) in five EC subsets analyzed by SCENIC. Each row corresponds to a regulon, each column represents a cell, and the color depth indicates the size of the area under the curve (AUC) value. (B) Analysis results of the top five transcription factors ranked by regulon specificity scores (RSS).

Figure S4. Enrichment analysis results of key modules. (A) The top 25 key genes in four modules of red, salmon, cyan, and turquoise. Nodes represent genes, edges represent coexpression relationships, and the top 10 key genes with KME values are located in the center of the network. (B) Gene Ontology (GO) enrichment analysis of the four key modules. The vertical axis is biological processes (BP), cellular components (CC), and molecular functions (MF), with −log10(p‐value) indicating the significance level, and the horizontal axis is the number of related genes for each GO term. (C) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of the four key modules. (D) Enrichment analysis of midnight blue and green‐yellow modules.

Figure S5. Construction of shTEK in human umbilical vein endothelial cells (HUVECs) and signaling pathway analysis. (A) homo‐TEK‐shRNA‐pLKO.1‐puro map. (B) shTEK plasmid sequencing results. (C) Western blot was used to detect the expression of TEK protein, ensuring the successful construction of shTEK HUVECs cell line. n = 4. (D) Statistical plot of TEK protein expression. Compared to the control group, **p < 0.01, ns = no significance; compared to the vector group, ^## p < 0.01. (E) Ki‐67 and Casp3 immunofluorescence detection. Scale bar = 50 μm. (F) Statistics plot of mean fluorescence intensity of Ki‐67 and Casp3. n = 3. Compared to the control group, ***p < 0.001, ns = no significance; compared to the venous malformation (VMs) model group, ^## p < 0.01, ^### p < 0.001; compared to the shTEK group, ^△△ p < 0.01, ^△△△ p < 0.001.

Supplementary Table S1. Clinical information of patients with venous malformations (VMs).

Supplementary Table S2. Overview of single‐nucleus RNA sequencing (snRNA‐seq) data before and after quality control.

Lin J, Liu T, Fang B, et al. SnRNA‐seq reveals cellular heterogeneity and proliferation mechanisms in limb venous malformations. Anim Models Exp Med. 2026;9:1567‐1586. doi: 10.1002/ame2.70256

DATA AVAILABILITY STATEMENT

All relevant data are within the manuscript and its Supporting Information files.