Wild-type C57BL/6 mice (male, 8–10 weeks) weighing 18–22 g were obtained from the College of Veterinary Medicine, Yangzhou University (Institute of Comparative Medicine) (certificate No. SCXK 2021–0003). Lrg1 knockout mice (strain designation: KOCMP-76905-Lrg1-B6N-VA) were obtained from Cyagen Biosciences Corporation on a C57BL/6N background (C57BL/6N-Lrg1^em1cyagen). Targeted disruption was achieved by deleting a ~1940 bp genomic fragment encompassing exon 1 and the TGA stop codon within exon 2 (Transcript: NSMUST00000041357). A successful knockout was confirmed via PCR-based genotyping. The Lrg1^−/− line was subsequently backcrossed for four generations to C57BL/6N mice to ensure genetic homogeneity. Experimental assays comprised 3-month-old animals (18–22 g) housed in a SPF-compliant IVC environment at Nanchang University. Environmental conditions were strictly regulated with a 12-h light-dark cycle, 22 ± 2°C, and relative humidity of 55 ± 10%. All animal procedures were performed in accordance with institutional ethical guidelines and ARRIVE criteria to ensure humane treatment and minimize distress. The primary sources of animal suffering in MCAO/R (middle cerebral artery occlusion reperfusion) models manifest as unilateral limb motor dysfunction, which, notably, does not impede normal behaviors like feeding and drinking. Additional procedures contributing to discomfort encompass tail clipping and ear tagging in mice. Throughout the experimental process, mitigating measures such as gentle petting are implemented to alleviate the pain and distress resulting from these procedures. The experimental ethics was in compliance with ARRIVE ethical guidelines, and was granted by the IACUC of The First Affiliated Hospital, Jiangxi Medical College, Nanchang University, with an approval number of CDYFY-IACUC-202306QR028.

MCAO/R was conducted on mice, followed by 1 h of occlusion and 24 h of reperfusion. The cohort included wild-type and Lrg1 knockout underwent sham-treated or MCAO/R-treated groups. Routine sequencing assay used n = 3 mice, while basic experiments typically n = 6 mice. Animals, according to the wild type and Lrg1 knockout mice, were randomly assigned a random number utilizing Excel software. The investigators were blinded to the pretreatment groups and remained so throughout the data analysis process. This method is considered an optimal choice currently for both basic and clinical research into cerebral and vascular diseases. Mice were anesthetized with 2.5% sevoflurane (Shanghai Hengrui Pharmaceutical Co., Ltd., 23122931, 24051131, Shanghai, China). This anesthetic protocol induces reversible effects on the physiological functions of mice. Animals were evaluated using laser Doppler flowmeters (LDF, FLPI2, Moor, UK), and only those with blood flow reduced to below 30% of pre-ischemia levels were selected for further experimentation. Following MCAO/R, mice exhibited unilateral motor deficits without compromising routine behaviors, such as feeding and drinking; thus, postoperative analgesics were not administered. After all procedures were completed, the mice were euthanized by cervical dislocation. All procedures strictly adhered to Standard Operating Protocols, including animal handling, weighing, and ear tagging, to minimize additional distress. Animals were promptly returned to comfortable housing post-surgery. Throughout the study, animals received gentle handling, assisted mobility support, and extended physical restraint, which was accompanied by post-procedural emotional comfort measures. In cases of unmanageable stress responses, immediate cervical dislocation euthanasia was implemented.

Coronal brain sections, each with a thickness of 5 μm, were carefully prepared from the ischemic area. These tissue sections were first thoroughly rinsed using PBS (pH 7.4), followed by immediate fixation in ice-cold 4% paraformaldehyde for a duration of 10 min. Subsequently, they underwent a blocking and permeabilization procedure, employing a solution comprising 10% normal goat serum (Gibco, 16210064, Frederick, MA, USA), 3% bovine serum albumin (HYclone, GCC070103, Logan, UT, USA), and 0.1% Triton X-100 dissolved in PBS, with an incubation period lasting 1 h. Then, primary antibodies were applied to the sections. These included antibodies targeting anti-Fzd3 (ab217032, Abcam, Cambridge, UK, 1:200), Zmiz1 (89500, CST, Danvers, MA, 1:50), GAPDH (ab9485, Abcam, 1:50), Occludin (ab216327, Abcam, 1:100), Claudin-5 (ab131259, Abcam, 1:20), and ZO-1 (ab307799, Abcam, 1:500). The sections were then incubated at 4°C for 48 h. Following this step, the brain tissues were treated with an Alexa Fluor^®488-conjugated donkey anti-goat IgG H&L secondary antibody (Abcam, ab150129, 1:1000) and incubated overnight at 4°C. Subsequently, the cell nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI, Abcam, ab104139, 1:500) for 15 min. Finally, the immunostained brain sections were visualized using a confocal microscope (Olympus, FV300, Tokyo, Japan).

After successfully establishing the MCAO/R model, the mice were anesthetized using 2.5% sevoflurane (Shanghai Hengrui Pharmaceutical Co., Ltd., 23122931, 24051131) before being euthanized through cervical vertebra dislocation. The hemibrains were then collected and rinsed repeatedly with ice-cold saline to eliminate the remaining blood. After that, they were gently blotted dry with absorbent paper. These processed hemibrains were subsequently prepared for both immunofluorescence staining and western blot analysis. In the western blot analysis, a 40 μg protein sample extracted from the ischemic area of the brain was separated by 12.5% SDS-PAGE. After separation, the proteins were transferred onto PVDF (polyvinylidene fluoride) membranes within a Tris-glycine buffer system. The membranes were first blocked with 5% nonfat dry milk at room temperature for 2 h. Then, they were incubated with primary antibodies in an ice bath for 24 h. The primary antibodies included anti-Fzd3 (ab217032, Abcam, Cambridge, UK, 1:1000, 76 kDa), Zmiz1 (89500, CST, Danvers, MA, USA, 1:1000, 120 kDa), GAPDH (ab9485, Abcam, 1:1000, 36 kDa), Occludin (ab216327, Abcam, 1:1000, 65 kDa), Claudin-5 (ab131259, Abcam, 1:1000, 23 kDa), and ZO-1 (ab307799, Abcam, 1:1000, 225 kDa). Following the primary antibody incubation, the membranes were further incubated with an HRP-conjugated secondary antibody (ab205718, Abcam, 1:2000) at room temperature for 50 min. Enhanced chemiluminescence was employed to visualize the bands, and the protein bands were captured using a biological image system (Bio-Rad, 1658001, Hercules, CA, USA).

The mice were rendered unconscious using 2.5% sevoflurane (Shanghai Hengrui Pharmaceutical Co., Ltd., 23122931, 24051131). After being anesthetized, the mice were placed in a prone position with their skulls held steady. The fur covering the fixed head surface was carefully shaved off to expose the underlying skull. Using a skull drill, a circular region with a diameter of 5 mm was delicately thinned along the cranial suture. The exposed pia mater was then precisely positioned beneath the Lan-MC real-time vascular injury imaging system, which facilitated clear visualization of the microvasculature. Within this visual field, venules with widths between 20 and 40 μm were specifically chosen for observation [15]. Before ten minutes of the experiment, FITC-albumin was diluted in saline to reach a concentration of 1 mg/mL. This solution was then injected into the femoral vein at a dose of 50 mg/kg. This injection was timed to occur approximately ten minutes before the start of the procedure to ensure optimal circulation and labeling of the FITC-albumin. Video recording was conducted using a fluorescence microscope, enabling the capture of dynamic changes in fluorescent signals. The recorded images were subsequently analyzed using the Image-Pro Plus software package (Media Cybernetics, Version 6, Rockville, MD, USA) for quantitative evaluation. To maintain temporal precision, the video recording speed was adjusted to capture 5 frames per second. From each experimental group, nine representative frames were methodically chosen for in-depth analysis. To quantify fluorescence intensity, regions of identical size were delineated both within the vessel lumen (representing venule fluorescence intensity, Iv) and in the adjacent interstitial area (representing interstitial area fluorescence intensity, Ii). The ratio of Ii to Iv was calculated for each frame, providing an indicator of vascular permeability and the degree of FITC-albumin leakage into the interstitial space. This ratio was then compared to the baseline fluorescence level of venules observed under sham operation conditions, which served as a control for non-pathological states. After the experiment concluded, mice were anesthetized again with 2.5% sevoflurane before being euthanized by cervical vertebra dislocation.

Endothelial cell BEND.3 (STCC2003) was purchased from Zeesan Biotech (Wuhan, China) and cultured with glucose-free medium (RPMI 1640, L270, Gbico). Cells were subjected to oxygen-glucose deprivation 3 h/reperfusion 12 h (OGD/R) conditions to establish the experimental model. The Oxygen-glucose deprivation model was simulated with a premixed gas atmosphere (1% O₂, 94% N₂, 5% CO₂) at 37°C. After OGD/R, the cells were perfused with DMEM with glucose and glutamine (10566016, Thermo Fisher Scientific, Waltham, MA, US) and transferred to a 5% CO₂, 95% O₂ air incubator for 12 h. The endothelial cells were treated for 24 h with a luciferase reporter plasmid containing the Fzd3 promoter, along with siRNA-Zmiz1 (Forward: GGAGGGCGTTGAAGAATTA and Reverse: GCATTGCCTATACCACAGG) or pcDNA3-Zmiz1 (p1017, Servicebio Technology Co., Ltd., Wuhan, China). The control plasmid employed Renilla luciferase as an internal control for the experiment. The luciferase activity of the reporter gene was then assessed using a dual-luciferase reporter assay kit (Promega, E1910, Madison, US). After treating the cells, wash them with PBS, and then add 200 μL of lysis buffer to each well. Collect the lysed cells into 1.5 mL EP tubes and centrifuge at 1000× g for 0.5 min. After collecting the supernatant, add the Stop solution. The detection was carried out using a multifunctional microplate reader (Varioskan LUX, Thermo Fisher Scientific), with the difference between the measured value and the background value representing the luciferase activity of the reporter gene.

The processing of endothelial cell samples was carried out as described in Section 2.1.6. The processed cells with the Fzd3 promoter, OGD/R, or other relevant conditions for 24 h, the cells were collected and processed using a chromatin immunoprecipitation (ChIP) Kit (Servicebio, GM1008, Wuhan, China) according to the following protocol. The cells were sonicated and centrifuged at 8000× g for 10 min at 4°C. Insoluble materials were discarded, and the supernatant was collected. Then, 50 μL of the product was mixed with 5 μL proteinase K and 2 μL 5 M NaCl, and incubated at 55°C overnight for decrosslinking. Afterwards, 100 μL of the sonicated product stored at −80°C was combined with 900 μL ChIP Dilution Buffer (1 mM PMSF, 20 μL 50 × PIC). Next, 60 μL Protein A+G Agarose/Salmon Sperm DNA was added to each sample, gently inverted and mixed at 4°C for 1 h, then incubated at 4°C for 10 min to precipitate. After centrifugation at 4000× g for 5 min, the samples were split into two 1.5 mL EP tubes. One tube got 1 μg IP antibody specific to the target protein, the other 1 μg corresponding IgG. They were gently inverted and incubated overnight at 4°C. Finally, immune complexes were precipitated, washed, and the DNA samples were recovered for RT-PCR detection. We query the binding sites of Zmiz1 on the hTFtarget database website (https://guolab.wchscu.cn/hTFtarget/#!/targets/integrate?tf=ZMIZ1, accessed on 26 November 2025). For Chromatin immunoprecipitation combined with quantitative polymerase chain reaction (ChIP-PCR), the sample volume in each well was adjusted to 2 μL of reverse primer, 2 μL of forward primer, and 25 μL of Taq Mix. The samples were then placed in a PCR machine (Eppendorf, 6336, Hamburg, Germany) for amplification. To prepare a 1% agarose gel, 1 g of agarose was dissolved in 50 mL of 1× TAE buffer by heating in a microwave until fully melted. The solution was then allowed to cool to room temperature, after which 2 μL of an ethidium bromide substitute was added and mixed thoroughly. The mixture was poured into a horizontal gel casting tray and allowed to solidify for 30 min. The prepared agarose gel was then placed in a horizontal electrophoresis tank, and an appropriate volume of 1× TAE buffer was added to cover the gel. Set the PCR reaction conditions as follows: initial denaturation at 95°C for 10 min, followed by denaturation at 95°C for 60 s, annealing at 95°C for 30 s, and extension at 72°C for 30 s, with a temperature increase rate of 0.3°C every 15 s during the ramp-up phases, then underwent 40 cycles. After the PCR reaction is completed, remove the reaction tubes and allow them to cool to room temperature. After loading the samples, electrophoresis was performed at a constant voltage of 120 V for 50 min. Upon completion, the gel was visualized for analysis. After data collection, the target genes were analyzed using Prism Demo software via the 2^−∆∆Ct method.

Mice were anesthetized with 2.5% sevoflurane (Shanghai Hengrui Pharmaceutical Co., Ltd., 23122931, 24051131). We designed mouse Zmiz1 gene short hairpin RNA (shRNA) via the MERCK website (https://www.sigmaaldrich.cn/CN/zh/semi-configurators/shrna?-activeLink=selectClones, accessed on 26 November 2025). The reverse complementary sequence was generated via a website (https://www.detaibio.com/sms2/rev_comp.html, accessed on 26 November 2025). The adenovirus vector encoding shRNA-Zmiz1 (sense: CGACCTTCCAAGCAATAGCAA, antisense: TTGCTATTGCTTGGAAGGTCG) and negative control-shRNA adenovirus empty vector (sense: CCGGCAACAAGATGAAGAGCA, antisense: CCAACTCGAGTTGGTGCTCTT) (Serotype: Adeno-associated virus, AAV9; Promoter: CAG) were constructed by Wuhan Servicebio Biotechnology Co., Ltd.

Mice were anesthetized with 2.5% sevoflurane (Shanghai Hengrui Pharmaceutical Co. 23122931, 24051131) and fixed in a brain solid positioner (Jiangwan I-C stereotactic device). Adjust the positioner parameters to ensure the cerebral midline is aligned symmetrically in all planes, then perform a midline incision disinfected with iodophor along the cranial sagittal suture to expose the bregma landmark. Set the stereotaxic positioning needle to the bregma landmark as the zero reference point: the injection coordinates are as follows: anterior posterior −0.9 mm, medio lateral ±1.5 mm, dorso ventral −0.9 mm, depth 3.0 mm. Fix the injection needle on the locator, carefully lower it, and insert the needle vertically into the hole of the locator. Inject 2 μL shRNA-Zmiz1 or shRNA-Zmiz1 NC (negative control) into each mouse, with the virus concentration 1 × 10¹⁰ vg/μL and the injection rate of 0.2 μL/min. At the end of the injection, hold the needle for 5 min, then slowly move the syringe up and out of the needle. The wound of operation ensures minimal discomfort and optimal recovery conditions for the animals. After all experimental results were obtained, mice were anesthetized with sevoflurane prior to euthanasia by cervical vertebra dislocation.

The t-test or Wilcoxon rank-sum test was employed to detect significant differences between two continuous groups of variables, depending on the data distribution. For the comparison of multiple groups of continuous variables, a two-way ANOVA or the Kruskal-Wallis test was utilized, with the choice guided by the data’s suitability for parametric or non-parametric analysis. To account for multiple comparisons, the Bonferroni method was applied for p-value correction, ensuring the robustness of statistical inferences. All statistical analyses were conducted using R Studio (version 4.1.3), a powerful and versatile platform for statistical computing and graphics. Statistical significance was defined as a p-value less than 0.05, unless otherwise explicitly stated, adhering to common conventions in scientific research. Prism Demo software was applied to analyze the target genes obtained from ChIP-PCR via the 2^−∆∆Ct method.

Brain tissue samples destined for analysis were carefully extracted, and a precise amount was weighed into a mortar previously chilled with liquid nitrogen. 1 mL of TRIzol reagent was meticulously added per 50–100 mg of tissue sample to isolate RNA from the brain samples. Subsequent to the quantification of RNA concentration, an amount of 1 μg of total RNA from each group was selected for the construction of cDNA libraries. The transcriptome sequencing was expertly handled by the Nova6000 high-throughput sequencer (Illumina Co., Ltd., San Diego, CA, USA). The resulting bulk RNA-seq data, a testament to the rigor of the process, have been publicly accessible in NCBI’s Sequence Read Archive (assigned the unique accession number PRJNA1147915). Genes exhibiting a |log₂ fold change (FC)| ≥ 1 and an adjusted p-value (P_adj) < 0.05 were rigorously scrutinized and identified as differentially expressed genes (DEGs). To provide a clear and intuitive visualization of their expression patterns, hierarchical clustering analysis was meticulously carried out. Additionally, enrichment analyses were conducted to delve deeper into the biological functions and pathways associated with these DEGs.

Brain tissue from the right hemisphere was obtained from each mice group and subjected to two washes with precooled RPMI 1640 medium (12633020, Thermo Fisher Scientific) supplemented with 0.04% BSA. The cleaned tissue was then processed into a single-cell suspension. This suspension was then resuspended in a red blood cell lysis buffer (10× Genomics, MACS, Catalog #130-094-183, Catalog#000268, Miltenyi Biotec, Bergisch Gladbach, Germany) to eliminate erythrocytes. After erythrocyte lysis, the lysis was incubated at 4°C for 10 min in a refrigerator, followed by centrifugation at 500× g for 5 min. Following centrifugation, the cell pellet was rinsed once more with RPMI 1640 medium (12633020, Thermo Fisher Scientific) and subjected to a second centrifugation step. The freshly prepared single-cell suspension was adjusted to the appropriate concentration as detailed in the protocol. The library preparation procedure was meticulously executed, adhering to the manufacturer’s protocols for the 10× Genomics Chromium Next GEM Single Cell 3’ Reagent Kit v3.1 (10× Genomics, Catalog#000268). Subsequently, the constructed libraries underwent high-throughput sequencing on the Illumina NovaSeq 6000 PE150 platform, guaranteeing comprehensive and precise transcriptome profiling of each individual cell.

The data was analyzed utilizing the Seurat pipeline (version 4.1.1, available at https://github.com/satijalab/seurat/, accessed on 26 November 2025) to conduct quality control measures, incorporating dimensionality reduction techniques and unsupervised clustering approaches [16]. The quality control criteria in this project are as follows: high-quality cells are retained if their gene count and UMI count fall within the range of the median ±2 times the median absolute deviation, and their mitochondrial UMI proportion is below 10%. Doublets are subsequently removed. To mitigate potential batch effects across different experimental groups, the BBKNN tool (Version 1.5.1, accessible via https://github.com/immunogenomics/harmony, accessed on 26 November 2025) was employed [17], with parameter settings as batch = 20, n_neighbors = 40, trim = 50. This correction step ensured the integrity of the data for downstream analyses. Subsequently, we identified individual cells based on universal markers (Supplementary file S1) specific to various cell types, leveraging established knowledge from textbooks and published literature. The identified cell clusters were then classified according to their expression of generic markers, providing a comprehensive characterization of the cellular composition within the analyzed samples.

We calculated the ROGUE index using ROGUE (version 1.0, https://github.com/PaulingLiu/ROGUE, accessed on 26 November 2025) [18]. The permutation test, a widely employed statistical method for analyzing small–sample data, was employed to assess differences in ROGUE values among distinct groups with 1000 permutations [19]. To conduct a rigorous assessment of the variations in ROGUE values (The input expression matrix: formatted with genes as rows and cells as columns. The matr.filter() function is applied to filter out: set as low-abundance genes (min.genes = 10) and low-quality cells (min.cells = 10). Default parameter: 0.6; Permutation test settings: number of permutations: 1000, UMI: k = 45, full-length transcript data: k = 500. ROGUE >0.9 means high consistency, <0.9 needs more clustering or data check) among different groups.

Complex cell–cell communication plays a crucial role in the interactions between cells. The CellphoneDB package (version 5.0.0) in Python (version 3.9.0) was used to infer and analyze cell–cell communication by identifying potential mechanisms of interaction through ligand-receptor pairs. Single-cell transcriptome data in Hdf5 format was set as the input file; the parameters should be set to the default values, requiring that at least 10% of the cells express genes related to a specific interaction; otherwise, that interaction will be excluded. Additionally, the CSOmap tool was employed to simulate direct cell-to-cell communication in spatial contexts based on single-cell sequencing data. In this study, the CSOmap R package (version 1.0) was utilized to identify cell communication networks within the brain tissues, providing deeper insights into the nature of intercellular interactions. Use the gene-cell expression matrix generated from scRNA-seq as the input file for CSOmap, with the parameters set as follows: Enriched interactions arise when one cell type is significantly overrepresented near another (right-tailed p < 0.05, q < 0.05), while depleted interactions occur when one type is significantly underrepresented (left-tailed p < 0.05, q < 0.05). Cell-cell interaction significance is assessed via 1000 random permutations of cluster labels to generate expected interaction distributions. Right/left-tailed tests compute p-values for enrichment (observed > expected) and depletion (observed < expected). Benjamini-Hochberg correction estimates q-values; pairs with enrichment/depletion q < 0.05 are deemed significant (interacting/dispersing), while those with both q > 0.05 are labeled “other”.

To evaluate the functional differences among different groups of cells, we utilized Seurat’s AddModule Score function at the single–cell level to calculate the scores of functional modules in cell clusters. Convert CSV and other files into list files for import into R software. We referenced the macrophage functional gene set summarized by Cheng et al. [20] to assess the function of microglial cells and macrophages. The calculated functional modules included the M1 score and the M2 score. In the AddModuleScore function, setting ctrl = 100 means randomly selecting 100 control genes from the expression data for normalization. Higher scores of various functional assessments indicate stronger corresponding functions of the cell. To unravel the functional differences among diverse cell groups, we leveraged Seurat’s AddModuleScore function at the single-cell level, enabling us to quantify the scores of functional modules within the identified cell clusters. For this purpose, we referenced the endothelial-relevant functional gene sets curated in the Gene Set Enrichment Analysis (GSEA) database (Supplementary file S2), encompassing critical processes like hypoxia response, oxidative phosphorylation, adhesion, and chemotaxis. These functional assessments aimed to evaluate the specialized roles of endothelial cells. Notably, higher scores across various functional evaluations indicate a more pronounced manifestation of the respective cellular functions.

To gain insights into the intricate lineage differentiation dynamics among cells, we adopted a comprehensive analytical approach that uses Monocle2 version 2.26.0 (a widely available tool, but generally available at https://github.com/cole-trapnell-lab/monocle-release, accessed on 26 November 2025, for trajectory analysis) [21]. These tools facilitated the examination of cellular differentiation trajectories, providing insights into the progression of cells from one state to another. We have prepared the metadata file for input. Additionally, we utilized Ro/e as a quantitative metric to assess the degree of preference exhibited by cells towards individual subpopulations. We utilize the calculate Ro/e function, with the calculation formula defined as: Ro/e = Observed/Expected. The calculation of the Ro/e ratio is based on the following approach: Both the observed and expected frequencies were derived using the chisq.test function (which performs a chi-square test). An Ro/e value exceeding 1 indicates that the specific cell cluster is overrepresented (enriched) within the group, whereas an Ro/e value below 1 suggests that the cell cluster is underrepresented (not enriched) in the group.

To detect differences between two continuous variable groups, we employed either the t-test or the Wilcoxon test. For comparing differences across multiple groups of continuous variables, ANOVA (when data met parametric assumptions) or the Kruskal-Wallis test (for non-parametric data) was utilized. To address the issue of multiple comparisons, the Bonferroni method was applied for p-value correction. All statistical analyses were conducted using R Studio (version 4.1.3) or Python (version 3.9.12), ensuring the rigor and reproducibility of our findings. Statistical significance was defined as a p-value less than 0.05.