scRNA-seq data (GSE178121), including the raw count matrix and metadata with cell type annotations and experimental conditions, were downloaded from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/) (33). The dataset included one control retinal sample and one streptozotocin (STZ)-induced DR retinal sample, each of which was generated by pooling retinas from three mice. Data preprocessing was performed using the Seurat package (v4.1.1; https://satijalab.org/seurat/) in R software (v4.5.1; R Foundation for Statistical Computing). Low-quality cells and genes were filtered out on the basis of the gene count (200-2,500) and mitochondrial content (<5%). The data were log-normalized, and 2,000 highly variable genes were selected for principal component analysis. The top 50 principal components were used for clustering and t-distributed stochastic neighbor embedding visualization. Cell clusters were visualized using DimPlot, and DotPlot was used to show the expression of selected genes across clusters (both functions were provided in the Seurat package; v4.1.1). Clusters were annotated into retinal cell types based on the expression of known marker genes, by comparison with public databases, including PanglaoDB (https://panglaodb.se/) and the Mouse Cell Atlas (https://bis.zju.edu.cn/MCA/), as well as relevant published literature (33,34). Following cell type annotation, endothelial cells were selected and further divided into UNC5B-positive and UNC5B-negative groups on the basis of UNC5B expression (raw count >0). To assess differences in pathway activity, gene set variation analysis (GSVA; v1.44.5; https://bioconductor.org/packages/release/bioc/html/GSVA.html) was performed at the single-cell level using the Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/) and Gene Ontology (GO; https://geneontology.org/) Biological Process (BP) gene sets. Pathway-level enrichment scores were compared between groups using the Wilcoxon rank-sum test with the Benjamini-Hochberg correction for multiple testing. Significantly enriched pathways (adjusted P<0.05) were visualized using bubble plots.

Immortalized human retinal microvascular endothelial cells (HRMECs; cat. no. XY-H469; Shanghai Xinyu Biotechnology Co., Ltd.) were used for high-glucose stimulation and UNC5B knockdown experiments, while immortalized human retinal microvascular pericytes (HRMVPCs; cat. no. ZQY001; Shanghai Zhongqiao Xinzhou Biotechnology Co., Ltd.) were used for UNC5B knockdown experiments and co-culture experiments with endothelial cells. HRMECs and HRMVPCs were cultured in DMEM (cat. no. 11965092; Gibco; Thermo Fisher Scientific, Inc.) containing 10% FBS (cat. no. 10099141C; Gibco; Thermo Fisher Scientific, Inc.) and 1% antibiotic-antimycotic solution (penicillin-streptomycin; cat. no. 15140122; Gibco; Thermo Fisher Scientific, Inc.) at 37°C in a humidified atmosphere containing 5% CO₂. For high-glucose treatment, HRMECs were exposed to 30 mM glucose for 24 or 48 h at 37°C to mimic hyperglycemic conditions. Cell function assays and reverse transcription-quantitative PCR (RT-qPCR) were performed using five biological replicates. Western blot experiments were conducted using four biological replicates. Transcriptome sequencing was performed in triplicates.

The lentiviral expression vectors for human UNC5B shRNA were developed and produced by Shanghai GeneChem Co., Ltd. The shRNA targeting sequence for UNC5B was 5'-CTACGAGATGTATCTACTCAT-3', while the scrambled control sequence was 5'-TTCTCCGAACGTGTCACGT-3'. These sequences were cloned into the GV112 lentiviral vector. Lentiviral particles were generated using a three-plasmid, second-generation packaging system. Briefly, 293T cells used for viral packaging were provided by the manufacturer as part of this service, and were co-transfected with the transfer vector and two helper plasmids (pHelper 1.0 and pHelper 2.0) at a mass ratio of 4:3:2 (total 45 μg DNA per 10-cm dish). Transfection was carried out at 37°C for 6 h before the medium was replaced. Viral particles were harvested 48-72 h post-transfection and concentrated via ultracentrifugation. Briefly, the supernatant was cleared of debris by centrifugation at 4°C at 4,000 × g for 10 min, filtered through a 0.45-μm filter, and ultracentrifuged at 100,000 × g for 2 h at 4°C. The resulting viral pellet was resuspended in virus preservation solution and briefly centrifuged at 10,000 × g at 4°C for 5 min before use. HRMECs or HRMVPCs were seeded in six-well tissue culture plates at a subconfluent density of 60% and transduced with lentiviral particles at an MOI of 50. After incubation for 24 h at 37°C, the medium was refreshed. Stable cell lines were selected using puromycin (1.0 μg/ml; cat. no. ST551; Beyotime Biotechnology) for 7-8 days. Following the establishment of stable colonies, the cells were maintained in medium containing a maintenance concentration of 0.5 μg/ml puromycin. Successful transduction was confirmed by the observation of green fluorescence under a fluorescence microscope. The expression levels of UNC5B in the stably transduced cells were quantified by RT-qPCR. As a control, parallel cultures of cells were infected with nonsense-scrambled shRNA lentiviral particles.

HRMECs or HRMVPCs with or without UNC5B knockdown were seeded in 24-well plates and cultured at 37°C overnight in DMEM containing 10% FBS to allow cell attachment. The cells were stained with a 1:1 mixture of PI and calcein-AM dye (cat. no. C1371; Beyotime Biotechnology) at 37°C for 10 min. Fluorescence microscopy (Olympus Corporation) was then employed to visualize the staining patterns, with viable cells emitting green fluorescence under a 490-nm excitation filter and necrotic cells emitting red fluorescence under a 545-nm excitation filter.

The albumin uptake assay was adapted from previously published protocols (35,36) with minor modifications. HRMECs were seeded in 24-well culture plates at a density of 5×10⁴ cells per well in 500 μl medium. Subsequently, the cells were incubated with 50 μg/ml rhodamine B isothiocyanate-labeled BSA (cat. no. SR063; Beijing Solarbio Science & Technology Co., Ltd.) at 37°C for 6 h in a medium containing 10% FBS but without antibiotics to avoid interference. Following incubation, the culture medium was removed, and the cells were gently washed with cold PBS (cat. no. BL302A; Biosharp Life Sciences) to remove any unbound or excess BSA. Subsequently, the cells were fixed at room temperature for 15 min in 4% paraformaldehyde (PFA; cat. no. BL539A; Biosharp Life Sciences) to preserve their morphological integrity and fluorescence properties for fluorescence imaging. After fixation, the nuclei were stained with DAPI (cat. no. C1002; Beyotime Biotechnology) for 5 min at room temperature, followed by three washes with PBS before fluorescence imaging.

Transwell multiplates containing cell culture inserts with a polycarbonate membrane at the lower aperture (pore size, 0.4 μm; diameter, 6.5 mm; cat. no. 353090; Corning, Inc.) were used. HRMECs with or without UNC5B knockdown were seeded at a density of 2.0×10⁵ cells/cm² on the membranes (upper chambers) in standard culture medium containing 10% FBS and 1% antibiotic-antimycotic solution. The lower chambers were filled with the same culture medium. The cells were cultured at 37°C to form a confluent monolayer within 2 days of seeding. On day 3 after seeding, 50 μg/ml FITC-conjugated dextran-40 kDa (cat. no. FD40S; MilliporeSigma) was added to the upper chamber. Dextran-40 kDa, a macromolecular tracer, was chosen because of its ability to approximate the size of macromolecules commonly encountered under physiological conditions (37). The cells were incubated with this tracer at 37°C for 6 h to allow sufficient time for permeation across the endothelial monolayer. Subsequently, the medium in the lower chambers was carefully collected, and its absorbance was measured using a microplate reader (Thermo Scientific Multiskan SkyHigh Microplate Spectrophotometer; Thermo Fisher Scientific, Inc.) as a quantitative indicator of monolayer permeability.

For the Matrigel co-culture assay, 24-well plates were primed with Matrigel (cat. no. 356234; Corning, Inc.) and subsequently incubated at 37°C for 30 min to facilitate the polymerization of Matrigel matrices. HRMECs (6×10⁴ cells per well) and HRMVPCs (4×10⁴ cells per well) were seeded onto the 24-well plates pre-coated with Matrigel and covered with DMEM supplemented with 10% FBS (38). After a culture period of 6 h at 37°C, the cells were fixed in 4% PFA at room temperature for 15 min. Subsequently, the endothelial cells were fluorescently labeled with isolectin B4 (IB4; 1:50; cat. no. I21411; Thermo Fisher Scientific, Inc.), whereas pericytes were stained with an anti-neuron-glial antigen 2 (NG2) antibody (1:200; cat. no. sc-53389; Santa Cruz Biotechnology, Inc.) at 4°C overnight. After washing using PBS, appropriate fluorescent secondary antibodies were applied at room temperature for 1 h. The antibodies used are listed in Table SI. The co-localization of these two fluorophores indicated the recruitment of pericytes to endothelial cells, and was visualized under a fluorescence microscope.

An in vitro BRB model was established using Transwell inserts with 0.4-μm polycarbonate membranes (39). HRMVPCs were seeded at a density of 3.0×10⁴ cells/cm² on the underside of the membrane in inverted inserts and incubated for 1-2 h at 37°C to allow adhesion. The inserts were then returned to the upright position, and HRMECs were seeded on the upper side at a density of 9.0×10⁴ cells/cm². Co-cultures were maintained for 48-72 h at 37°C until confluent monolayers formed on both sides. To assess permeability, the upper chamber was washed using PBS, and standard culture medium containing 50 μg/ml FITC-dextran was added, while the lower chamber received medium without dye. After 6 h of incubation at 37°C, medium (100 μl) from the lower chamber was collected, and the fluorescence intensity was measured using a microplate reader to evaluate FITC-dextran diffusion across the BRB model.

An EdU proliferation kit (cat. no. C0075; Beyotime Biotechnology) was used to assess cell proliferation. HRMVPCs were incubated with EdU solution (10 μM) at 37°C for 2 h. Subsequently, the cells were fixed with 4% PFA at room temperature for 15 min, washed using PBS, and co-stained with DAPI at room temperature for 10 min. Finally, the cells were visualized under a fluorescence microscope and analyzed using ImageJ software (v1.53; National Institutes of Health).

HRMVPCs (with or without UNC5B knockdown) seeded at a density of 5×10⁴ cells per chamber in serum-free DMEM were added to the apical compartment of Transwell inserts featuring a membrane with 8-μm pores (cat. no. 353097; Corning, Inc.). The basal chamber was filled with complete medium containing 10% FBS. Subsequently, the Transwell chambers were incubated in a humidified environment maintained at 37°C with 5% CO₂. After 24 h, the migrated cells were fixed with 4% PFA for 15 min at room temperature, stained with crystal violet at room temperature for 1 min and imaged under a light microscope for quantitative analysis.

Total cellular and tissue RNA was extracted using the FastPure Cell/Tissue Total RNA Isolation Kit V2 (cat. no. RC112-01; Vazyme Biotech Co., Ltd.). HiScript IV RT SuperMix for qPCR (+gDNA wiper) (cat. no. R423-01; Vazyme Biotech Co., Ltd.) was used to transcribe cDNA. Reverse transcription was performed at 50°C for 5 min, followed by 85°C for 5 sec, according to the manufacturer's instructions. The reaction mixture contained 1 μl cDNA template, 1 μl of primers, 4 μl diethyl pyrocarbonate-treated water and 4 μl 2X SYBR Green PCR Mix (cat. no. Q711-02; Vazyme Biotech Co., Ltd.). RT-qPCR was performed using a real-time PCR system (Thermo Fisher Scientific, Inc.) with the following thermocycling conditions: Initial denaturation at 95°C for 30 sec, followed by 40 cycles of 95°C for 10 sec and 60°C for 30 sec. Melting curves were generated to verify the specificity. Relative mRNA expression levels were quantified using the comparative Ct method (2^−ΔΔCq) (40), with β-actin tested as the reference gene. The primer sequences used for RT-qPCR are listed in Table SII.

Cellular and tissue lysates (20-30 μg per treatment) were prepared using RIPA lysis buffer (cat. no. P0013B; Beyotime Biotechnology) supplemented with protease inhibitor cocktail (cat. no. HY-K0011; MedChemExpress). Protein concentrations were determined using the BCA assay (cat. no. 23225; Thermo Fisher Scientific, Inc.). Equal amounts of protein (20-30 μg per lane) were separated by 6-15% SDS-PAGE and then transferred to a PVDF membrane. The membranes were blocked with 5% nonfat milk for 2 h at room temperature to prevent nonspecific binding. Primary antibodies were then applied, and the membranes were incubated overnight at 4°C to ensure optimal binding to target proteins. After thorough washing with 1X Tris-Buffered Saline with 0.5% Tween-20 (cat. no. T1085; Beijing Solarbio Science & Technology Co., Ltd.), the membranes were incubated with secondary antibodies for 1 h at room temperature. Protein bands were detected using an enhanced chemiluminescence substrate (cat. no. P0018AS; Beyotime Biotechnology). Protein bands were visualized using a Tanon 5200 series fully automated chemiluminescence imaging system (Tanon Science and Technology Co., Ltd.), which allowed a clear and quantifiable assessment of target protein expression. Band intensities were semi-quantified using ImageJ software. The antibodies used for western blotting are listed in Table SI.

Bulk RNA sequencing analysis was performed using transcriptomic data obtained from HRMECs with and without UNC5B knockdown. HRMECs transduced with UNC5B shRNA (knockdown group) or scramble shRNA (control group) were harvested at 80-90% confluence. Cells were washed with cold PBS and lysed directly in TRIzol Reagent (cat. no. 15596026; Thermo Fisher Scientific, Inc.). Lysates were aliquoted into RNase-free tubes, snap-frozen in liquid nitrogen, and stored at −80°C until shipment on dry ice to HaploX Ltd. for library preparation and sequencing. RNA quality and integrity were assessed using a NanoDrop™ One/OneC (OD260/280 and OD260/230), Qubit^® 3.0 Fluorometer (Thermo Fisher Scientific, Inc.) with the Qubit™ RNA HS Assay Kit (cat. no. Q32852; Thermo Fisher Scientific, Inc.) for precise quantification, and Agilent 4200 TapeStation (Agilent Technologies, Inc.) for RNA integrity number assessment. Libraries were prepared according to the manufacturer's instructions. Library quality was validated using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc.) to check the insert size, and quantified via RT-qPCR to ensure an effective concentration of >2 nM. Sequencing was performed on an Illumina, Inc., platform with paired-end 150-bp reads. Raw sequencing data were processed using HaploX Ltd.'s in-house open-source software fastp (v0.23.0; https://github.com/OpenGene/fastp) (41). Processing steps included adaptor removal, filtering of reads containing a high proportion of N bases (default, 5 bp), removal of reads with >40% low-quality bases (Phred score, ≤20) and sliding-window trimming (window size, 4 bp; average quality threshold, 20). Differential expression analysis was conducted using the limma package (v3.52.2; https://bioconductor.org/packages/release/bioc/html/limma.html) in R, with genes filtered on the basis of adjusted P<0.05 and |log₂ fold change|>0.5. Significantly differentially expressed genes were further subjected to GO and KEGG enrichment analysis using clusterProfiler (v4.0.0; https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html) and the enrichment results were visualized using ggplot2 (v3.3.5; https://cran.r-project.org/web/packages/ggplot2/index.html).

All animal experiments were approved by the Institutional Animal Care and Use Committee of Nanjing Medical University (approval no. IACUC-2407092; Nanjing, China), and conducted in accordance with the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research. A total of 200 C57BL/6J male mice, aged 6-8 weeks and weighing 20-25 g, were used in the present study. Mice were purchased from Nanjing Junke Biotechnology Co., Ltd. They were bred in an air-conditioned room maintained at 22±2°C with a relative humidity of 50-60%, under a 12-h light/dark cycle, with free access to standard laboratory chow and water. Mice were anesthetized with intraperitoneal injection of ketamine (90 mg/kg) and xylazine (7.5 mg/kg) prior to any invasive procedures. For terminal experiments, including enucleation of the eyes, mice were humanely euthanized by cervical dislocation under anesthesia as aforementioned. Mice were randomly assigned to experimental or control groups according to the specific study design. Animals were assigned to the following independent groups: Wild-type (20 mice), scramble control shRNA (shC; 10 mice), UNC5B shRNA (shUNC5B; 10 mice), negative control (NC; 5 mice), UNC5B overexpression (oeUNC5B; 5 mice), DR (20 mice), RVO (5 mice), DR + 5 ng/ml Netrin-1 (5 mice), DR + 50 ng/ml Netrin-1 (5 mice), DR + 500 ng/ml Netrin-1 (5 mice), DR + 1,000 ng/ml Netrin-1 (5 mice), DR + 5,000 ng/ml Netrin-1 (5 mice), DR + shC (20 mice), DR + shUNC5B (20 mice), DR + Netrin-1 (15 mice), DR + shUNC5B + Netrin-1 (15 mice), DR + NC (15 mice) and DR + oeUNC5B (15 mice). Unless otherwise specified, all animal experiments were performed using n=5 mice per group. For western blot analysis, each group consisted of n=4 biological replicates.

C57BL/6J mice were fasted overnight and intraperitoneally injected with STZ (50 mg/kg; cat. no. EZ66D1DFD8; BioFroxx; neoFroxx) for 5 consecutive days. Starting 1 week after the final STZ injection, blood glucose levels were measured using a portable blood glucose meter once weekly using a drop of blood collected from the tail tip. Mice with blood glucose levels ≥16.7 mmol/l were considered diabetic and were included in the study.

Endothelial-tropic recombinant AAV vectors (Shanghai GeneChem Co., Ltd.) were administered by retro-orbital intravenous injection to modulate UNC5B expression in the mouse retina. For knockdown, an AAV carrying an shRNA against Unc5b driven by the intercellular adhesion molecule 2 promoter (target sequence, 5'-GGATCATGAGGTCCTTCTGCA-3') with an enhanced green fluorescent protein reporter was used. A non-targeting scramble shRNA AAV (target sequence, 5'-TTCTCCGAACGTGTCACGT-3') with the same promoter and reporter was used as the control. For overexpression, full-length mouse Unc5b cDNA under the control of the tyrosine kinase with immunoglobulin-like and EGF-like domains 2 promoter was delivered in the same manner, with an empty vector as control. AAV vectors were administered via retro-orbital injection at a total dose of ~3×10¹¹ viral genomes per mouse. Subsequent experiments were conducted ≥4 weeks post-injection. Endothelium-specific modulation of UNC5B expression was validated by western blotting and immunofluorescence staining (Fig. S1).

Mice were anesthetized via intraperitoneal injection of ketamine (90 mg/kg) and xylazine (7.5 mg/kg). The pupils were dilated with 0.5% tropicamide (Santen Pharmaceutical Co., Ltd.), and 0.5% proparacaine (Alcon, Inc.) was used for topical anesthesia. A 33G needle (cat. no. 80300; Hamilton Company) was inserted through the inferotemporal sclera under a stereomicroscope. Mice received a single intravitreal injection of 2 μl Netrin-1 (cat. no. ALX-522-124; Enzo Life Sciences, Inc.) at various concentrations (5, 50, 500, 1,000 or 5,000 ng/ml), with 1,000 ng/ml as the primary concentration used for subsequent analyses. Control eyes received equal volumes of PBS as the vehicle solution. After the injection, 0.3% ofloxacin eye ointment (Shenyang Sinqi Pharmaceutical Co., Ltd.) was used to prevent infection. The recovery, respiration and activity of mice were monitored.

Mice were anesthetized via intraperitoneal injection of ketamine (90 mg/kg) and xylazine (7.5 mg/kg), and positioned with their eyes oriented upward. A microinjector needle was inserted at a 45° angle through the medial canthus into the retro-orbital venous sinus. The aforementioned AAVs were injected slowly to avoid vascular and ocular injuries. After needle withdrawal, gentle pressure was applied with a sterile gauze to achieve hemostasis. The mice were monitored for immediate post-procedure abnormalities before being returned to their cages.

Adult C57BL/6J mice were anesthetized via intraperitoneal injection of ketamine (90 mg/kg) and xylazine (7.5 mg/kg). Subsequently, Evans Blue (EB; 2%; 45 mg/kg) dye (cat. no. E104208; Shanghai Aladdin Biochemical Technology Co., Ltd.) was injected into the femoral vein and allowed to circulate for 1 h (42). This was a terminal procedure; mice were euthanized immediately after EB circulation. The eyes were then gently enucleated, and the cornea, sclera, lens and vitreous humor were carefully removed to isolate the retina. The retinas were fixed in 4% PFA for 30 min at room temperature to preserve their structural integrity and ensure optimal staining conditions. Finally, the retinas were carefully separated and examined under a fluorescence microscope. This allowed the visualization and quantification of EB leakage from the vasculature, providing a direct assessment of retinal vascular permeability.

The retinal vasculature was meticulously isolated using a standardized trypsin digestion protocol (43). Freshly excised eyes were promptly fixed in 10% buffered formalin overnight at 4°C to preserve delicate vascular structures. After fixation, the retinas were gently dissected and incubated in double-distilled water overnight at room temperature on a shaking table to ensure the thorough removal of any residual fixative or debris. Subsequently, the retinas were transferred to a 2% trypsin solution and incubated at 37°C for 1 h, allowing controlled enzymatic digestion to gently break down the surrounding tissue while preserving the vascular network. After digestion, the retinas were thoroughly washed with double-distilled water for 5 min to remove any residual trypsin. The isolated retinal vasculature was then air-dried in a well-ventilated environment to prevent unwanted morphological changes. Finally, the vasculature was stained with PAS and hematoxylin using a stain kit (cat. no. G1281; Beijing Solarbio Science & Technology Co., Ltd.) according to the manufacturer's instructions. Briefly, samples were incubated with periodic acid at room temperature for 5-10 min, followed by Schiff reagent for 15 min at room temperature, and counterstained with hematoxylin for 1-2 min at room temperature. Retinas were imaged under a light microscope, and quantitative analyses were performed using ImageJ software, revealing the intricate patterns of blood vessels within the retina.

Mouse eyes were enucleated and fixed in 4% PFA for 2 h at room temperature. Subsequently, the eyes were dissected to isolate the retinas, which were carefully cut into four-leaf clover patterns to facilitate flat mounting. The retinas were fixed in 4% PFA for 15 min at room temperature. The tissues were permeabilized and blocked simultaneously in 1% Triton X-100 (cat. no. 9002-93-1; Shanghai Hushi Laboratory Equipment Co., Ltd.) and 5% BSA (cat. no. 4240; BioFroxx; neoFroxx) at 37°C for 1 h. The retinas were incubated with primary antibodies overnight at 4°C. The following day, the retinas were incubated with the appropriate fluorophore-conjugated secondary antibodies for 1 h at room temperature. The antibodies used in the present study are listed in Table SI. Fluorescence images were acquired using a fluorescence microscope. Quantitative analysis was performed using ImageJ software.

Mouse eyes were immersed in 4% PFA and stored at 4°C for 2 h immediately after enucleation. After removal of the cornea and lens, the eyes were subjected to an additional fixation step in 4% PFA overnight at 4°C. Tissues were then dehydrated in 30% sucrose for 48 h and embedded in optimal cutting temperature compound (cat. no. 4583; Sakura Finetek USA, Inc.) for cryosectioning. Sagittal retinal sections (10 μm thick) were prepared using a cryostat and mounted onto microscope slides. The sections were permeabilized with 0.3% Triton X-100 and blocked with 5% BSA simultaneously at 37°C for 1 h. Subsequently, the sections were incubated overnight at 4°C with primary antibodies. After three washes with PBS, the sections were incubated with fluorophore-conjugated secondary antibodies appropriate for the host species of the primary antibodies for 1 h at room temperature, followed by counterstaining with DAPI for 5 min at room temperature. A list of the antibodies used is provided in Table SI. Fluorescence images were captured using a fluorescence microscope, and the fluorescence intensity and number of positive cells were quantified using ImageJ software.

C57BL/6J mice received a tail-vein injection of 1% Rose Bengal solution (cat. no. A17053; Thermo Fisher Scientific, Inc.) prepared in saline (20 mg/kg) using an insulin syringe. The mice were then monitored for adverse reactions. Within 20 min of Rose Bengal injection, the mice were anesthetized via intraperitoneal injection of ketamine (90 mg/kg) and xylazine (7.5 mg/kg). Furthermore, 0.5% tropicamide eye drops (Santen Pharmaceutical Co., Ltd.) were used for pupil dilation and 0.5% levofloxacin eye drops (Santen Pharmaceutical Co., Ltd.) were administered to prevent infection. A 514-nm laser (100 mW; 0.1 sec pulse; 50 μm spot size) was directed at the retinal veins to induce RVO (44,45).

OCT imaging (Saris; Nanjing Robotrak Technologies Co., Ltd.) was performed 1 h, 1 day and 8 days after laser treatment to assess retinal structural changes. The mice were anesthetized as described previously, and their pupils were dilated with 0.5% tropicamide (Santen Pharmaceutical Co., Ltd.). To prevent corneal dehydration, the ocular surface was lubricated throughout the procedure. The anesthetized mice were positioned on the OCT imaging platform to ensure proper eye alignment for retinal scans. Cross-sectional (B-scan) and facial images were acquired for each eye. Multiple scans were captured to ensure reproducibility, and the most focused images were selected for analysis. Retinal thickness and lesion morphology were quantified using the built-in analysis software provided by the manufacturer.

All procedures involving human participants were approved by The Ethics Committee of The Affiliated Eye Hospital of Nanjing Medical University (approval no. 2023003; Nanjing, China) and conducted in accordance with The Declaration of Helsinki. AH samples were collected from patients diagnosed with non-proliferative DR (NPDR), proliferative DR (PDR) or RVO, and age-related cataract at The Affiliated Eye Hospital of Nanjing Medical University (Nanjing, China) between April 2023 and April 2024. A total of 32 patients were included: Patients with NPDR (n=8; age range, 61-75 years; mean age, 67.38±5.40 years; 3 male and 5 female patients), patients with PDR (n=8; age range, 62-68 years; mean age, 65.13±2.30 years; 4 male and 4 female patients), patients with RVO (n=8; age range, 62-69 years; mean age, 65.38±2.26 years; 4 male and 4 female patients) and cataract controls (n=8; age range, 61-76 years; mean age, 70.38±4.81 years; 4 male and 4 female patients). There were no statistically significant differences in age between the cataract control group and the disease groups (NPDR, PDR and RVO) (P>0.05). The inclusion criteria were as follows: A clinical diagnosis of NPDR, PDR or RVO based on fundus examination and imaging, and availability of intraocular samples. Patients with age-related cataract without retinal or systemic vascular diseases were enrolled as controls. The exclusion criteria included the presence of other retinal diseases, ocular inflammation, prior intraocular surgery or intravitreal injection within the past 6 months, and systemic inflammatory or autoimmune diseases. Using a precision 30-gauge needle, 30-50 μl of AH was gently aspirated from each patient, with the needle insertion performed through the peripheral cornea to avoid any contact with the iris or lens tissue. This procedure was conducted under meticulous observation under a surgical microscope before cataract surgery or anti-VEGF intravitreal injection. The AH samples were transferred into sterile tubes and centrifuged at 12,000 × g for 5 min at 4°C to separate the cellular components and clarify the supernatant. The expression levels of UNC5B in the AH samples were detected using a human UNC5B enzyme-linked immunosorbent assay kit (cat. no. MM-7554761; MEIMIAN) in accordance with the manufacturer's instructions.

All data are presented as the mean ± SD to provide a comprehensive and quantitative representation of the experimental outcomes. Statistical analyses were performed using GraphPad Prism 9 software (Dotmatics). Significant differences between groups were determined using the unpaired two-tailed Student's t-test for pairwise comparisons or one-way ANOVA followed by Tukey's or Dunnett's multiple comparisons test for multiple-group comparisons, depending on the specific experimental design. P<0.05 was considered to indicate a statistically significant difference.

Initial bioinformatics analysis was performed to determine the cellular localization and differential expression of UNC5B in the diabetic retina. Following normalization, dimensionality reduction and clustering of the scRNA-seq data, 19 distinct cell clusters were identified using the first 10 principal components (Fig. 1A). These clusters were annotated into 11 major retinal cell types (Fig. 1B and C). Subsequent analysis of the expression profile revealed that UNC5B was predominantly expressed in endothelial cells, and its expression in DR retinal endothelial cells was lower than that in controls (Fig. 1D and E).

To further explore the potential functional role of UNC5B in endothelial cells, endothelial cells were stratified into UNC5B-positive and UNC5B-negative groups and subjected to GSVA using KEGG and GO BP pathways. Pathway enrichment analysis demonstrated that UNC5B expression in endothelial cells was associated with pathways such as 'cell adhesion', 'ECM-receptor interaction' and the 'Hippo signaling pathway' (Fig. 1F and G).

These findings suggested that UNC5B serves a crucial role in maintaining retinal vascular homeostasis during the pathological process of DR, particularly in preserving endothelial cell function, and potentially contributing to BRB integrity.

To validate the results of bioinformatics analysis, an STZ-induced DR mouse model was established using C57BL/6J mice. Retinal tissues were collected from different mice at 2, 4 and 8 weeks post-modeling for RT-qPCR and western blot analyses. UNC5B mRNA and protein levels in the retinas of 8-week DR mice were significantly lower than those in normal mice (mRNA, P<0.001; protein, P=0.004; Fig. 2A and B). Furthermore, in cultured endothelial cells, high-glucose stimulation significantly reduced UNC5B expression at 24 h (mRNA, P=0.039; protein, P=0.005) and 48 h (mRNA, P<0.001; protein, P<0.001) (Fig. 2C and D). Similarly, significant reductions in retinal UNC5B levels were observed in the RVO mouse model (mRNA, P=0.023; protein, P=0.003; Fig. 2E and F). Immunofluorescence staining of retinal whole mounts, using IB4 to label the retinal vasculature, further confirmed the close co-localization of UNC5B with the retinal vascular network (Fig. 2G), thereby supporting the scRNA-seq finding that UNC5B was predominantly expressed in endothelial cells.

To assess the clinical relevance, UNC5B levels were analyzed in human AH samples. The AH was collected from patients with DR showing diabetic macular edema (DME) ('NPDR + DME' and 'PDR + DME') and RVO who had not undergone intravitreal anti-VEGF injection, using samples from patients with age-related cataract as controls. ELISA results showed that the UNC5B expression in the AH of patients with DR and RVO was significantly lower than that in the control group (P<0.001 for all comparisons; Fig. 2H). These findings strongly supported the crucial role of UNC5B in the maintenance of BRB integrity.

To further elucidate the role of UNC5B in maintaining endothelial barrier integrity, UNC5B was knocked down in HRMECs by transfection with lentivirus-encapsulated UNC5B shRNA. Successful transfection was verified by fluorescence imaging (Fig. 3A), and the knockdown efficiency was confirmed at both the mRNA (P<0.001) and protein levels (P=0.002) (Fig. 3B and C). PI/calcein-AM staining showed that cell death was significantly increased after UNC5B knockdown (P<0.001; Fig. 3D).

The present study subsequently investigated the effects of UNC5B depletion on key regulators of the endothelial barrier. RT-qPCR analysis showed no significant changes in the expression of the paracellular junction proteins zonula occludens 1 and occludin, whereas the claudin-5 level was significantly decreased (P=0.001). Concurrently, the expression levels of caveolin-1, which is indicative of increased transcellular transport (46), were markedly upregulated (P<0.001) (Fig. 3E). Western blot analysis corroborated these findings, verifying increased caveolin-1 expression (P=0.002) and elevated levels of plasmalemma vesicle-associated protein (PLVAP), another transcellular transport marker (P=0.014) (Fig. 3F).

Functional assays demonstrated that transcellular transport, as assessed using labeled albumin, was elevated after UNC5B knockdown (P=0.002; Fig. 3G), and the paracellular leakage of monolayer endothelial cells was also increased (P=0.008; Fig. 3H). On the basis of these results, it was inferred that UNC5B serves a key role in maintaining endothelial cell barrier function.

To investigate the role of UNC5B in regulating endothelium-pericyte crosstalk, the interactions between endothelial cells and pericytes were assessed following UNC5B knockdown in HRMECs. Immunofluorescence staining of IB4 and NG2 revealed a significant reduction in pericyte recruitment to the endothelial tubes after UNC5B knockdown (P<0.001; Fig. 4A). BRB formation was then simulated by co-culturing HRMECs with HRMVPCs (Fig. 4B). The results showed that UNC5B knockdown led to increased leakage of the fluorescent tracer (P<0.001; Fig. 4C). These findings suggested that reduced UNC5B levels in endothelial cells negatively affected pericyte function.

To investigate whether UNC5B directly affects pericyte function, UNC5B was silenced in HRMVPCs using lentiviral shRNA and a stable UNC5B-knockdown HRMVPC cell line was established using puromycin selection. RT-qPCR and western blot analyses confirmed the knockdown efficiency (mRNA, P<0.001; protein, P=0.014; Fig. 4D and E). Assessment of pericyte apoptosis, proliferation and migration revealed no significant changes following UNC5B knockdown (Fig. 4F-H).

These results indicated that UNC5B served a crucial role in regulating endothelium-pericyte interactions and maintaining endothelial barrier function, with negligible direct effects on pericytes. Silencing UNC5B in endothelial cells simultaneously affected both transcellular and paracellular transport as well as interactions with pericytes, thereby disrupting BRB integrity.

The concentration-dependent effects of netrin-1 on vascular permeability were evaluated in the DR mouse model. The results of EB staining showed that compared with DR mice, high concentrations of netrin-1 (≥1,000 ng/ml) protected retinal vascular integrity (1,000 ng/ml, P=0.016; 5,000 ng/ml, P<0.001), whereas lower concentrations (<500 ng/ml) promoted vascular leakage (5 ng/ml, P=0.038; 50 ng/ml, P<0.001) (Fig. 5A).

Subsequently, the present study focused on the role of UNC5B in mediating the protective effects of high-concentration netrin-1 (1,000 ng/ml) in DR mice. Endothelial-specific silencing of UNC5B was successfully achieved in vivo by AAV-mediated gene targeting, as confirmed by reporter fluorescence (P<0.001) and protein expression (P=0.003) (Fig. S1A and B). Compared with that in DR mice treated with scrambled shRNA, intravitreal injection of netrin-1 significantly reduced retinal vascular leakage (P<0.001). Crucially, UNC5B silencing in DR mice not only increased vascular leakage (P=0.042) but also prevented netrin-1 from exerting its protective effect (Fig. 5B).

Retinal vascular morphology was assessed by PAS staining. In comparison with DR mice treated with scrambled shRNA, mice that received netrin-1 treatment showed fewer acellular capillaries (P<0.001) and increased pericyte coverage (P=0.007). By contrast, endothelial UNC5B knockdown increased the number of acellular capillaries (P=0.006) and decreased pericyte numbers (P=0.026), preventing netrin-1 from exerting its protective effects (Fig. 5C). Similar findings were obtained using IB4 and NG2 immunofluorescence staining. Pericyte coverage was significantly increased in DR mice injected with netrin-1 compared with DR mice treated with scrambled shRNA (P<0.001). By contrast, UNC5B knockdown reduced pericyte coverage (P=0.022), and netrin-1 treatment did not improve coverage in mice receiving UNC5B shRNA (Fig. 5D).

Given the protective role of UNC5B in retinal endothelial cells, it was evaluated whether endothelial-specific overexpression preserved the integrity of the BRB in DR mice. UNC5B overexpression was confirmed by immunofluorescence staining (P<0.001) and western blot analysis (P<0.001) (Fig. S1C and D). The results showed that specific UNC5B overexpression in the endothelial cells of DR mice significantly reduced retinal vascular leakage (P<0.001; Fig. 6A), alleviated the formation of acellular capillaries (P<0.001; Fig. 6B) and preserved pericyte number (P<0.001; Fig. 6B) and coverage (P=0.004; Fig. 6C). These findings highlighted the therapeutic potential of targeting endothelial-cell UNC5B to maintain the BRB integrity in DR.

To assess the role of UNC5B in DR-associated neurodegeneration and gliosis, UNC5B was selectively silenced or overexpressed in retinal endothelial cells using AAV delivery as aforementioned (Fig. S1), and analyses were conducted 14 weeks post-DR induction. Immunofluorescence staining of retinal cryosections revealed a significant decrease in neuron-specific nuclear protein (NeuN)-positive ganglion cells and β-III tubulin (TUBB3) fluorescence intensity in DR mice in comparison with the corresponding findings in wild-type controls, confirming ganglion cell loss. Endothelial-specific silencing of UNC5B further exacerbated this neuronal loss (NeuN, P<0.001; TUBB3, P=0.031), whereas UNC5B overexpression partially mitigated ganglion cell degeneration (NeuN, P=0.020; TUBB3, P=0.022) (Fig. 7A and B). Retinal flat mounts stained with class III β-tubulin further confirmed that modulation of endothelial UNC5B expression influenced ganglion cell survival in the DR model (DR + shUNC5B, P=0.014; DR + overexpression of UNC5B, P=0.021; Fig. 7C). Vimentin staining of retinal cryosections demonstrated reactive gliosis in DR mice, with UNC5B silencing promoting glial proliferation (P=0.020), whereas UNC5B overexpression significantly attenuated this glial activation (P=0.004) (Fig. 7D). These findings suggested that endothelial UNC5B expression influenced neuronal and glial outcomes in DR by maintaining BRB integrity.

Since both DR and RVO are characterized by BRB disruption, which leads to macular edema (12), the protective role of UNC5B was further investigated in an RVO mouse model. Endothelium-specific UNC5B knockdown was achieved by AAV-mediated targeting as aforementioned (Fig. S1A and B). At least 4 weeks after AAV injection, RVO was induced by tail-vein injection of Rose Bengal followed by laser treatment (47).

Retinal edema and atrophy were dynamically monitored using OCT 1 h, 1 day and 8 days after laser induction (Fig. 8A). The OCT results showed that after endothelial UNC5B silencing, the peak retinal edema thickness was significantly increased (day 1, P=0.015), and this subsequently led to greater thinning of the retina by day 8 (P=0.037) (Fig. 8B).

To investigate the regulatory role of UNC5B in endothelial cells, transcriptome sequencing analysis of HRMECs transfected with UNC5B-shRNA lentivirus and control cells transfected with scrambled control lentivirus was performed. The volcano plot highlighted numerous significantly upregulated and downregulated genes, indicating that UNC5B silencing markedly altered the transcriptional landscape of endothelial cells (Fig. S2A). GO and KEGG enrichment analysis revealed that these differential genes were closely related to 'extracellular matrix organization', 'ECM-receptor interaction' and the 'Hippo signaling pathway' (Fig. 9A and B). Furthermore, the differential genes were enriched in cellular components such as 'collagen-containing extracellular matrix' and 'cell-substrate junction' (Fig. S2B). In terms of molecular function, these genes were primarily involved in 'extracellular matrix structural constituent' and the binding of key proteins such as integrin, fibronectin and laminin (Fig. S2C).

These ECM findings were further validated at the protein level. Western blotting confirmed that UNC5B knockdown increased the expression of several core ECM components such as collagen IV (P=0.034), fibronectin (P=0.019) and laminin (P<0.001), suggesting that UNC5B may regulate the integrity of the BRB by modulating ECM components (Fig. 9C).

The Hippo signaling pathway, which regulates cell proliferation and apoptosis, also maintains vascular stability by controlling the ECM composition and barrier protein expression (48-50). Dysregulation of this pathway can lead to excessive ECM accumulation, basement membrane thickening and subsequent BRB dysfunction, contributing to retinal diseases such as DR and RVO (51,52). Consistent with the observed increase in ECM components, UNC5B silencing inhibited the phosphorylation of the key Hippo signaling pathway kinase mammalian STE20-like kinase (MST; P=0.002), leading to elevated protein levels of the downstream effectors yes-associated protein (YAP; P=0.033) and transcriptional co-activator with PDZ-binding motif (TAZ; P=0.027). Analysis of nuclear protein extracts by western blotting demonstrated significant nuclear accumulation of YAP and TAZ in UNC5B-silenced cells compared with controls (P<0.001 for all comparisons; Fig. 9D).

Collectively, these results indicated that UNC5B influenced ECM synthesis and metabolism and inhibited the Hippo signaling pathway, thereby participating in the maintenance of endothelial cell homeostasis and vascular barrier function.