PEBP1 Regulates Ferroptosis in Acute Glaucoma: Targeted Therapy Using Engineered Exosomes

ABSTRACT

Glaucoma is the leading cause of irreversible blindness worldwide, primarily driven by the progressive loss of retinal ganglion cells (RGCs) under pathological high intraocular pressure (ph‐IOP). Despite the established role of ferroptosis in RGC degeneration, specific molecular targets that can be used for clinical intervention still need to be optimized, and the slow onset of conventional gene therapy vectors is incompatible with the acute clinical course of glaucoma. Here, we integrate single‐cell RNA sequencing and spatial transcriptomics to profile the dynamic transcriptomic landscape of the rat retina across acute, subacute, and chronic stages of ph‐IOP injury. Through ferroptosis‐focused screening of an early‐activated RGC gene cluster, we identify the lipid metabolism regulator phosphatidylethanolamine‐binding protein 1 (PEBP1) as a candidate mediator of RGC ferroptosis. We demonstrate that Pebp1 is specifically upregulated in injured RGCs with a trajectory mirroring ferroptosis pathway activation, and that AAV‐mediated Pebp1 knockdown suppresses ferroptosis through the GPX4/ACSL4 signaling axis, thereby preserving RGC survival, retinal structure, and visual function. To overcome the critical time‐window bottleneck—the several weeks delay required for AAV‐mediated silencing versus the rapid, irreversible RGC loss in acute glaucoma—we engineer Exosomes‐siPebp1, a mesenchymal stem cell‐derived exosome system loaded with siPebp1, which enables immediate single‐dose intervention post‐injury. This system exhibits efficient RGC uptake, prolonged intraocular retention, and robust target gene silencing, and, in a head‐to‐head comparison, significantly outperforms unloaded exosomes, liposomal formulations, and AAV vectors in RGC protection, without detectable acute systemic or local toxicity. Collectively, this study implicates Pebp1 in ph‐IOP‐associated RGC ferroptosis and supports exosome‐mediated siRNA delivery as a rapid, cell‐free intervention strategy for acute glaucomatous injury.

General schematic diagram of Spatial and Single‐Cell Transcriptomics Identifies PEBP1 as a Ferroptosis Mediator in Glaucoma: Targeted Intervention by Engineered Stem Cell Exosomes.

1. Introduction

Glaucoma is the leading cause of irreversible blindness worldwide [1]. According to the World Health Organization, approximately 76 million people were affected by glaucoma in 2020, with projections estimating 111 million cases by 2040, thereby posing a substantial burden on quality of life [2]. Glaucoma is typified by progressive visual field loss and optic nerve damage, fundamentally driven by the selective and irreversible loss of retinal ganglion cells (RGCs) [3]. Although glaucoma is classified etiologically into primary, secondary, congenital, and mixed‐mechanism types, pathological high intraocular pressure (ph‐IOP) remains the most prominent risk factor driving RGC injury and death [4]. However, the precise molecular mechanisms linking ph‐IOP to RGC loss remain poorly defined, severely limiting the development of targeted therapeutic interventions.

Under ph‐IOP conditions, RGC injury engages multiple interrelated pathological mechanisms, including: (1) mechanical compression leading to lamina cribrosa deformation and axonal transport disruption, which in turn activates apoptotic and necrotic cascades; (2) ischemic injury arising from retinal hypoperfusion; (3) immune‐mediated inflammatory cell infiltration; and (4) other contributing processes such as oxidative stress and excitotoxicity [5]. Currently, the cornerstone of clinical glaucoma management remains restricted to intraocular pressure reduction via medication or surgery [6]; yet a substantial proportion of patients continue to experience progressive visual decline despite achieving target IOP levels [7]. This clinical paradox underscores that IOP reduction alone is insufficient to halt the RGC degenerative process, revealing an urgent need to elucidate the molecular mechanisms governing RGC loss and thereby identify novel therapeutic targets amenable to clinical intervention [8].

In recent years, the rapid advancement of single‐cell RNA sequencing (scRNA‐seq) and spatial transcriptomics has provided unprecedented tools for dissecting the cellular heterogeneity and gene regulatory networks of the glaucomatous retina at single‐cell resolution [9, 10]. These technologies enable the identification of injury‐associated cell types and their differentially expressed genes within their native microenvironmental context, offering new avenues to elucidate the molecular mechanisms driving RGC loss [11, 12]. Integrating multi‐omics datasets holds the promise of systematically identifying novel potential therapeutic targets that participate in the pathological progression of glaucoma. However, target identification alone does not bridge the gap from mechanistic insight to therapeutic intervention—a tool capable of precise and efficient gene silencing is urgently required. RNA interference (RNAi) technology is ideally suited to meet this need: small interfering RNAs (siRNAs) specifically degrade target mRNAs via Watson‐Crick base pairing, thereby potently suppressing protein translation and enabling precise gene silencing [13]. To facilitate the clinical translation of RNAi, various synthetic carriers—including lipid‐based and polymeric nanoparticles—have been explored for siRNA delivery, with the goal of protecting the nucleic acid payload in vivo, promoting cellular uptake, and facilitating endosomal escape [14, 15, 16]. However, current nanocarrier‐based delivery systems continue to face critical bottlenecks—including inefficient cytosolic release and potential immunogenicity—which restrict their utility in neuroretinal disorders and underscore the pressing need for alternative delivery strategies with superior biocompatibility.

To overcome these delivery barriers, exosomes—particularly engineered exosomes derived from mesenchymal stem cells (MSCs)—have emerged as a promising natural delivery platform for siRNA. These nanoscale vesicles, approximately 30–150 nm in diameter, are actively secreted by cells and mediate intercellular communication through the transport of bioactive cargo, including nucleic acids, proteins, and lipids [17]. Compared to conventional synthetic vectors, engineered exosomes offer distinct advantages: their superior biocompatibility and low immunogenicity minimize systemic clearance and inflammatory responses; their endogenous lipid bilayer enables traversal of biological barriers and evasion of reticuloendothelial phagocytosis; and their surface adhesion proteins and integrins facilitate efficient cell recognition and internalization [18, 19, 20, 21]. These intrinsic properties enable exosomes to effectively shield siRNA from nuclease degradation and promote cytosolic release, thereby markedly enhancing delivery efficiency [22, 23]. MSC‐derived exosome‐mediated siRNA delivery may provide a rapid and biocompatible intervention strategy for glaucomatous RGC injury.

Our previous research has revealed ferroptosis as a critical pathological mechanism driving RGC loss in the ph‐IOP model [24]. However, conventional iron chelators exhibit poor target specificity and dose‐limiting toxicity, severely impeding their clinical translation and underscoring the urgent need to identify druggable molecular targets [25]. Meanwhile, we previously employed single‐cell and spatial transcriptomics to comprehensively profile the retinal transcriptome across distinct stages of ph‐IOP and identified a gene set that was robustly activated early after injury, thereby establishing a candidate pool for therapeutic target discovery [26]. Building on this foundation, we herein performed an integrative ferroptosis‐focused screen and identified the lipid metabolism regulator phosphatidylethanolamine‐binding protein 1 (PEBP1) as a gene that was specifically upregulated in injured RGCs, with expression kinetics closely mirroring the activation trajectory of the ferroptosis pathway. Subsequent in vivo AAV‐shRNA knockdown experiments demonstrated that Pebp1 suppresses ferroptosis through modulating the GPX4/ACSL4 signaling axis, thus preserving RGC viability, retinal architecture, and visual function. To address the discordance between the delayed onset of AAV‐mediated knockdown and the acute, precipitous loss of RGCs in glaucoma, we further engineered exosome‐siPebp1 (EXO‐siPebp1), a mesenchymal stem cell‐derived exosome system loaded with siPebp1, and confirmed its efficient RGC uptake, prolonged intraocular retention, and robust target gene silencing. A single intravitreal injection of exosome‐siPebp1 immediately after ph‐IOP injury afforded superior RGC protection compared to unloaded exosomes, liposomal formulations, and AAV vectors, with no appreciable acute toxicity in vivo. In summary, based on previous multi‐omics analyses, this study identifies Pebp1 as a novel pharmacological target for ferroptosis‐targeted therapy in ph‐IOP injury and establishes a strategy utilizing exosomes and siRNA for rapid delivery, thereby providing a precise, cell‐free intervention paradigm for glaucoma and other acute neurodegenerative diseases (Graphical Abstract, Scheme).

2. Materials and Methods

2.1. Animals

Female Sprague–Dawley rats (8 weeks old; Slaccas, Changsha, China) were housed and maintained under standard laboratory conditions in the animal facility of Xiangya Hospital, Central South University. All experimental procedures were approved by the Animal Ethics Committee of Xiangya School of Medicine, Central South University (Ethical Approval Code: xy2023101236), and were conducted in accordance with the principles of the Declaration of Helsinki and the statement of the Association for Research in Vision and Ophthalmology (ARVO).

2.2. Model of Ph‐IOP Injury

Anesthesia was induced in rats by intraperitoneal (i.p.) injection of 1% sodium pentobarbital (100 mg/kg; Sanshu, Beijing, China) and xylazine (10 mg/kg; Huamu, Beijing, China). The corneas were then topically administered with Oxybuprocaine hydrochloride and tropicamide phenylephrine (Santen Pharmaceuticals, Tokyo, Japan). Ph‐IOP was induced to cause injury, using a saline perfusion system as described by Yao et al. [24]. Briefly, anesthetized rats with dilated pupils and anesthetized corneas were positioned on a heated platform. A 31‐gauge needle connected to a saline infusion set was carefully inserted into the anterior chamber. The intraocular pressure (IOP) was gradually increased to 110 mmHg by elevating the saline reservoir to a height of 150 cm and maintained at this level for 60 min. Eyes that underwent needle insertion without saline pressurization served as the sham‐operated controls. Tobramycin (0.3%) ophthalmic ointment (Alcon, Houston, TX, USA) was applied during and after the procedure to prevent infection and maintain moisture. Only rats that exhibited no saline leakage or damage to the lens and iris were included in the study.

2.3. Single‐Cell RNA Sequencing (scRNA‐Seq)

2.3.1. Collection of Retinal Single‐Cell Samples and Preparation of Single‐Cell Suspensions

For the retinal single‐cell suspension, eyes were collected from deeply anesthetized rats at 2 h, 24 h, or 7 days following retinal ph‐IOP injury. For each time point, retinas from three individual rats (one eye per rat) were processed independently into single‐cell suspensions, which were then pooled together prior to library construction to create one sequencing library per time point. Eyes were also collected from sham‐operated rats that did not undergo physiological saline pressure infusion, serving as controls. The cornea, iris, lens, and vitreous were excised, and the retina was carefully separated from the posterior eye globe. A 2‐mm biopsy punch (Robbins Instruments, Houston, TX, USA) was used to carefully excise the central portion of the retina to prepare the single‐cell suspension.

In brief, the retina was incubated with activated papain at 37°C for 45 min, followed by gentle trituration (using a 10 mL pipette to move up and down approximately 10–20 times). After centrifuging the cells at 300  g for 5 min, papain was removed by adsorption, and the cells were then resuspended in medium containing 10% ovomucoid, a papain inhibitor containing 5% DNase I. Complete cells were separated from the cell membrane through single‐step discontinuous density gradient centrifugation at 100  g for 6 min. To reduce acute, endogenous transcriptional changes, we applied the general transcription inhibitor actinomycin D during the dissociation process. Actinomycin D inhibits transcription mediated by all three eukaryotic RNA polymerases, offering broad‐spectrum, rapid transcriptional inhibition with minimal and slow reversibility. Actinomycin D (Sigma‐Aldrich, St. Louis, MO, USA; Cat# A1410) was added at three steps during the dissociation process: 45 mM to the papain solution, 45 mM to the papain inhibitor solution, and 3 mM during trituration.

After centrifugation, the supernatant was removed, and the cell pellet was resuspended in 100 μL of dead cell removal microbeads (Miltenyi Biotec, USA; No. 130–090‐101). The cell suspension was then passed through an MS column (Miltenyi Biotec, USA, No. 130–042‐201), placed in the magnetic field of the appropriate MACS separator (Miltenyi Biotec, USA, No. 130–042‐109). The MS column was eluted with 1× binding buffer. The cells were collected as the live cell fraction. The live cells were separated by centrifugation at 70  g for 6 min at room temperature, and the resulting pellet was resuspended in Dulbecco's phosphate‐buffered saline containing 0.4% BSA. Finally, cell viability was assessed using AO/PI staining on a cell counter (Nexcelom Bioscience LLC, Lawrence, MA, USA). The final concentration reached 100 000 cells/mL, with a viability rate exceeding 80% [27].

2.3.2. Library Construction and Sequencing

The single‐cell library was prepared according to the BD Rhapsody Single‐Cell 3′ Full Transcriptome WTA Library Construction Standardized Manual. After passing quality control, the library was sequenced using the PE150 mode on the Illumina Nova Seq6000 sequencing platform.

2.3.3. Raw Data Processing

The BD Rhapsody pipeline (https://bitbucket.org/CRSwDev/cwl/src/master/v2.0/) was used to generate a digital gene expression matrix. The UMI count matrix was converted into a Seurat object by the R package Seurat (version 4.3.0). Cells with UMI numbers < 1000 or with detected genes < 500 or with over 15% mitochondrial‐derived UMI counts were considered low‐quality cells and were removed. Genes detected in less than 5 cells were removed for downstream analyses.

After quality control, the UMI count matrix was normalized using SCTransform. Then we conducted data integration through the harmony algorithm based on gene expression profiling. To reduce the dimensionality of the scRNA‐Seq dataset, principal component analysis (PCA) was performed on an integrated data matrix. With Elbowplot function of Seurat, top 50 PCs were used to perform the downstream analysis. The main cell clusters were identified with the FindClusters function offered by Seurat, with resolution set as default (res = 0.4), and they were visualized by uniform manifold approximation and projection (UMAP) plots. To identify the cell type for each cluster, we detected gene markers for each cell cluster using the “FindMarkers” function in Seurat package, then we annotated cell types using ScType tools [28, 29, 30, 31]. Cell type annotation was performed using a set of canonical marker genes, identified from the integrated dataset across all conditions, to define stable core cellular identities (Table S1).

2.3.4. Differential Gene Expression Analysis

Differentially expressed genes (DEGs) were determined with the Find Markers/Find All Markers function from the Seurat package (one‐tailed Wilcoxon rank sum test, p‐values adjusted for multiple testing using the Bonferroni correction). For computing DEGs, all genes were probed that the expression difference on a natural log scale was at least 0.5 and the adjusted p‐value was less than 0.05.

2.3.5. Gene Set Enrichment Analysis

Based on the expression levels of key genes, samples were stratified into high‐ and low‐expression groups. Gene Set Enrichment Analysis (GSEA) was then performed to identify differentially enriched signaling pathways between these groups. The annotated gene sets from the FerrDB database served as the reference gene sets for ferroptosis‐related pathways. Differential expression analysis of pathways was subsequently conducted across the defined groups. Gene sets demonstrating significant enrichment (adjusted p‐value < 0.05) were ranked based on their normalized enrichment score (NES). GSEA is widely used to elucidate complex relationships between biological processes and disease classifications.

2.4. Spatial Transcriptome

2.4.1. Ocular Sample Collection and Preparation of Microscopic Slides

Eyes were collected from deeply anesthetized rats 2 h, 24 h, or 7 days after retinal ph‐IOP injury. For spatial transcriptomics, one eye from one rat was used per time point, constituting independent biological replicates for the ST analysis. Eyes were also collected from sham‐operated rats that did not undergo physiological saline pressure infusion, serving as controls. The eyes were then embedded in Tissue TEK OCT compound (VWR Cat No. 25608–930) and rapidly frozen on dry ice. The eyes were stored at −80°C for future analysis. The dorsal and temporal sides of the entire eye were marked with tissue labeling dye for experiments requiring anatomical orientation information.

The spatial transcriptomics library was prepared using the 10× Genomics Visium Spatial Gene Expression Kit. After tissue embedding, cryosectioning, and fixation of the rat eyeballs, optimized tissue slides were obtained. These slides were calibrated using the Aperio VERSA 200 scanning system, followed by bright‐field imaging at 5× magnification with a 200 ms exposure time. The entire tissue capture area was imaged in one shot, producing clear images of the tissue. Subsequently, the permeabilization experiment was completed with the same settings, and clear fluorescence images were obtained. It was observed that at 12 min, the fluorescence signal was strongest and signal diffusion was lowest, making this time point the optimal tissue permeabilization time (Figure S1).

2.4.2. 10× Genomics Visium Library Preparation and Sequencing

For preparing the Visium slides, the tissue block was equilibrated to −18°C, and 10 μm thick sections were placed on the active sequencing area (6.5 mm × 6.5 mm) of the 10× Genomics Visium slide. The slides were stored at −80°C in an airtight container until the time for spatial library generation. Hematoxylin and eosin (H&E) staining was performed following the 10× Genomics Visium protocol. The RIN values for all Visium‐treated samples were greater than 8.7. In brief, the sections were fixed with cold methanol, stained with H&E, and imaged on the Hamamatsu NanoZoomer S60. Following this, a template‐switching protocol was used for permeabilization, reverse transcription, and cDNA synthesis. The second‐strand cDNA was released from the slide, and a single‐index library was prepared using the 10× Genomics PCR‐based protocol. Sequencing of the library was conducted on the HiSeq 4000 (1 read per lane), with a target of 300 M raw reads per sample. The sequencing format was as follows: read 1: 28 cycles; i7 index read: 10 cycles; i5 index read: 10 cycles; and read 2S: 50 cycles. The cleaned data were subsequently processed using 10× Genomics Space Ranger (version 2.3.1) software for cell quality control, identification, and alignment with the rat reference genome (rat_NCBI_v7.2), generating the raw expression matrix. The matrix was then subjected to in‐depth downstream analysis using the Seurat package (version 4.3.0) [32].

2.4.3. Single‐Slide Processing

Filtered feature‐barcode expression matrices from SpaceRanger (v2.3.1) were used as initial input for the spatial transcriptomics analysis using Seurat (v4.3.0). The data filtering criteria were as follows: (1) Genes with less than 10 spot expressions were filtered out; (2) Spots with less than 300 measured genes and less than 500 UMIs were filtered out. Ribosomal and mitochondrial genes were excluded from this analysis. Individual count matrices were normalized, screened for hypervariable genes, and normalized using sctransform. PCA was used to reduce the dimension of the data, and the first 30 PCs were selected for clustering. Then, the spots of the 4 samples were integrated by the CCA algorithm IntegrateData function to display the clustering results of the batches with single sample tissue slices and UMAP dimension reduction, and the resolution was uniformly set to 0.5.

2.4.4. Visium Spatial Gene Expression Data Processing and Analysis Using Cell2location

To spatially resolve cell populations identified in the scRNA‐seq, we used Cell2location (v0.1.3) [33] to estimate cell type abundance for each spatial slide by taking a spatial expression matrix with mRNA counts of genes at spatial locations and a matrix of reference cell type signatures as input. All Cell2location analyses were conducted using Scanpy (v1.9.3). Before running Cell2location to map our reference cell types to each ST‐seq dataset, both the estimated reference cell type signatures (gene expression) and ST‐seq dataset from each respective biopsy were filtered to select genes shared between the two data objects. The filtered objects were both used as input to Cell2location.

To estimate reference cell type signatures, the scRNA‐seq reference signatures of cell types were loaded into Scanpy using sc.read_h5ad. First, the regression model for the single cell data was initialized with default settings, using batch as the batch_key, and the model was trained using a maximum of 250 epochs. Next, the regression model for the ST‐seq data was initialized with the scRNA‐seq reference signatures by default settings and hyperparameters selected based on cell2location's recommendations. Model accuracy was evaluated based on the evidence lower bound (ELBO) loss and reconstruction accuracy plots. We did not observe strong within‐batch variation in total RNA counts; therefore, we set detection_alpha = 200 and batch_size = 2500. By manual visual inspection, we estimated the number of cells per spot to be 16 and set N_cells_per_location = 16. The model was then trained using a maximum of 30 000 epochs. Cell2location's Non‐negative Matrix Factorization (NMF) was used to identify cellular compartments and cell types that co‐locate from the cell type abundance estimates, using n_fact = 12.

2.5. Multi‐Omics In Situ Pairwise Sequencing (MiP‐Seq)

2.5.1. Tissue Preparation

Samples served as subjects for ISS sample preparation from eyes. Animals were euthanized via isoflurane administration, followed by perfusion using a solution of 4% paraformaldehyde (PFA) in phosphate‐buffered saline (PBS). The eyes were carefully extracted and subsequently post‐fixed in 4% PFA for a duration of 12 h at 4°C. Cryoprotection was achieved using a 30% sucrose solution in PBS. Once embedded in an optimal cutting temperature (OCT) compound, the tissues were sectioned coronally to a thickness of 10 μm using a Leica cryostat. Sections encompassing retina were preserved at −80°C, pending ISS.

Prior to ISS analysis, eye slices underwent the following preparatory procedure: Initially, slices were sealed within Secure‐Seal hybridization chambers. Subsequent to this, they were rinsed twice with ice‐chilled DEPC‐PBS, followed by a fixation period of 10 min in 4% PFA. Post‐fixation, the slices were subjected to three washes using DEPC‐OBSTR (PBS supplemented with 0.1% Tween‐20 and 0.1 U/μL RRI). Thereafter, the slices were immersed in pre‐cooled methanol and maintained at −80°C for a span of 15 min. Subsequently, the slices were exposed to a solution of 2 mg/mL pepsin (Sigma, P0525000) in 0.1 M HCl and incubated at 37°C for 90 s. After a final rinse with DEPC‐PBSTR, the tissue slices were aptly conditioned for ISS analysis [34].

2.5.2. Probe Design

For ISS analysis, a set of specific genes from retina were chosen. The probe assembly employed in ISS is comprised of a duo of padlock probes paired with an RCA initiator primer probe, specifically designed for target‐dependent RCA. Additionally, detection probes for RCA and in situ sequencing probes were incorporated. The design methodology for these probes has been reported in prior publications. Briefly, the ISS probes were generated according to the following criteria: (1) Exons located near the 5′ end were preferred, with two target sites selected for each gene. (2) The padlock probe had complementary ends (12–16 bp) that matched the target sequence, while the middle section consisted of a complementary anchor primer sequence and a barcode sequence, (3) The RCA initiator primer had a 5′ end that matched the 3′ end of the target sequence (12–16 bp), and its 3′ end complemented the padlock probe. (4) The efficiency of RCA was assessed using the detection probe (cy3‐p4). (5) The probes for in situ sequencing included three anchor primers (anchor‐1, anchor‐2, and anchor‐3) and four query probes (seg‐1, seg‐2, seg‐3.and seg‐4), each labeled with different Alexa Fluor dyes.

2.5.3. In Situ Sequencing

The ISS procedures we used were described previously. Briefly, padlock probes were first phosphorylated at their 5′ termini using T4 polynucleotide kinase (Vazyme, N102‐472 01). Subsequently, the phosphorylated padlock probes, at a concentration of 30 nM per probe, were combined with the RCA initiator (also 30 nM per probe) in a hybridization buffer comprising 2× SSC (SangonB548109), 10% formamide (Sangon, A100606), and 20 mM RVC (Bevotime, R0108). This mixture was added to a hybridization chamber housing pretreated eye slices and incubated overnight at 37°C. Following this, the samples underwent two wash cycles with DEPC‐ PBSRT, succeeded by a 4X SSC wash. The chamber then received the ligase reaction mixture for a 2 h incubation at 37°C. After an additional two wash cycles with DEPC‐PBSTR, the samples were treated with a rolling circle amplification reaction solution, comprised of 1 U/μl Phi29 (Vazyme, N106‐01), 1X RCA buffer, 0.25 μM dNTP, 0.2 μg/μl BSA, and 5% glycerol, for 2 h at 30°C. Post‐wash, the samples were exposed to a sequencing mixture, containing 1X BSA, 0.2 U/μl T4 DNA ligase (Thermo Fisher Scientific, EL0012), 1X T4 DNA ligase buffer, 1 μM anchor primer, and 1 μM query primer for a 2 h incubation at 25°C. The samples were then thrice washed using a 10% formamide in 2X SSC solution, followed by DAPI staining. Imaging was conducted on a Leica THUNDER Imaging Systems, 20 × (NA = 0.80). After each imaging round, signals were eradicated with a stripping buffer (60% formamide in 2X SSC) twice for 10 min at ambient temperature. This was followed by the addition of the subsequent round's sequencing mixture to facilitate multiple sequencing rounds.

2.6. Adeno‐Associated Virus (AAV) Intravitreal Injection

The experimental viral vectors, pAAV‐U6‐shPebp1‐CMV‐EGFP‐WPRE and the mock control pAAV‐U6‐spgRNA‐CMV‐EGFP‐WPRE, were packaged and purchased from OBiO Technology (Shanghai, China). For intravitreal injection, anesthetized rats with dilated pupils were positioned under a stereomicroscope. A volume of 2 μL of AAV solution (1.5 × 10¹² viral genomes [vg]/mL) was injected into the vitreous cavity of each eye using a glass micropipette connected to a microinjection pump (R‐480, RWD, China). The injections were performed on 5‐week‐old rats, and the analyses were conducted at 8 weeks of age.

2.7. Cell Culture

Human umbilical cord (hUC) samples were obtained from full‐term healthy women undergoing elective cesarean section at Changsha Hospital for Material and Child Health Care. Written informed consent was obtained from each donor prior to tissue collection, and donors were informed that the tissue would otherwise be discarded as medical waste and that they could withdraw at any time without affecting their medical care. The following exclusion criteria were applied: (1) maternal age < 18 or > 40 years; (2) positive serological tests for infectious diseases (HBV, HCV, HIV, or syphilis); (3) pregnancy complications (gestational diabetes, pre‐eclampsia, intrauterine infection, or premature rupture of membranes); (4) multiple gestation; (5) known fetal genetic disorders or congenital anomalies; and (6) maternal malignancy or chronic immunosuppressive therapy. Only samples meeting all inclusion criteria (full‐term uncomplicated singleton pregnancy, negative infectious disease screening) were used. The study protocol was approved by the Institutional Ethics Committee of Changsha Hospital for Material and Child Health Care (Ethical Approval Code: EC‐2021106‐07). hUC‐derived mesenchymal stem cells (hUC‐MSCs) were then isolated and cultured in Shandong Yinfeng Life Science Institute (Jinan, China) and Changsha Longchuang Biotechnology Co. Ltd. (Changsha, China). The cells were initially cultured in a humidified incubator at 37°C with 5% CO₂, using DMEM/F12 medium (Thermo Fisher Scientific, USA) supplemented with 10% fetal bovine serum (FBS; Thermo Fisher Scientific) and 1% penicillin/streptomycin (Thermo Fisher Scientific). For subsequent experiments, hUC‐MSCs were maintained in an animal component‐free (ACF) culture medium (Catalog #05448, STEMCELL Technologies Inc., Canada) specifically formulated for human hUC‐MSCs, according to the manufacturer's instructions. Cell passaging was performed using an Animal Component‐Free Cell Dissociation Kit (Catalog #05426, STEMCELL Technologies Inc., Canada). When hUC‐MSCs at passages 2–3 reached 70%–80% confluency, the culture medium was replaced with ultra culture serum‐free medium. The cells were then cultured for an additional 24–48 h, after which the conditioned supernatant was collected. hUC‐MSCs identification was performed by immunofluorescence staining to detect the expression of surface markers CD105 (GB150113, Service bio, 1:1000), CD73 (GB115174, Service bio, 1:600), and CD90 (GB113753; Service bio, 1:400) in passage 4 hUC‐MSCs.

2.8. Isolation and Exosome Purification

Exosomes were isolated from the serum‐free conditioned medium of human umbilical cord mesenchymal stem cells (hUC‐MSCs) by ultracentrifugation using a Beckman Coulter ultracentrifuge (Brea, CA, USA), following the manufacturer's guidelines and established protocols. Briefly, the medium was sequentially centrifuged at 300  g for 10 min to remove cells, followed by 2000  g for 10 min to pellet apoptotic bodies and cell debris. The resulting supernatant was then subjected to centrifugation at 10000  g for 30 min to eliminate smaller cellular debris and subsequently filtered through a 0.2 μm pore‐size filter (Millex, Germany). The filtrate was ultracentrifuged at 100000  g for 70 min to pellet the exosomes. The exosomal pellet was resuspended in phosphate‐buffered saline (PBS) and subjected to a second ultracentrifugation wash at 100000  g for 70 min. All centrifugation steps were performed at 4°C. The protein concentration of the isolated exosomes was quantified using a BCA kit (23 225, Pierce, USA). Finally, the purified exosomes were stored at −80°C for subsequent experiments.

2.9. Preparation of siRNA‐Loaded Exosomes and Encapsulation Efficiency

siRNA targeting Pebp1 (siPebp1) was loaded into hUC‐MSC‐derived exosomes using an Exosome RNA Loading Kit (Cat# ELSR‐06, ExoLoad, Echo‐biotech, China) according to the manufacturer's instructions. Briefly, a reaction mixture containing 1000 pmol of siRNA, exosomes (3 × 10¹¹ particles/mL), Exosome Transporter Peptide (ETP), and reaction buffer was incubated in the dark at 37°C for 2 h with constant shaking at 150 rpm. Following incubation, the sample was transferred to an ultrafiltration device to remove unencapsulated (free) oligonucleotides. The sample was then washed twice with wash buffer to obtain the purified siPebp1‐loaded exosomes. The prepared siPebp1‐loaded exosomes were aliquoted and stored at −80°C. siRNAs were purchased from geneharma. Sequences of siRNAs used were Pebp1 sense GCAUUUCAUGGGAUGGCCUTT, antisense AGGCCAUCCCAUGAAAUGCTT.

To calculate the siRNA encapsulation efficiency, a standard curve was generated. A series of twofold serial dilutions of the siRNA in PBS were prepared, with concentrations ranging from 1 μM to 0 nM, each in a 100 μL volume. The loaded exosome preparation was mixed thoroughly, and a 3 μL aliquot was diluted 40‐fold with PBS to a final volume of 120 μL. Then, 80 μL of each diluted standard and the diluted loaded‐exosome sample were transferred in duplicate to a black 96‐well plate. The fluorescence intensity was measured using a microplate reader. The standard curve of fluorescence versus concentration was plotted from the serial dilutions, and the total amount of encapsulated siRNA in the sample was calculated based on its fluorescence intensity. The total exosome quantity was determined in parallel, either by measuring total protein content using a BCA assay (23 225, Pierce, USA). The siRNA loading capacity was calculated using the following formula: X = (A × B × C)/D. Where: X is the amount of siRNA loaded per unit of exosomes (e.g., pmol/μg protein or pmol/particle); A is the siRNA concentration derived from the standard curve (in nM); B is the dilution factor for fluorescence measurement; C is the total volume of the loaded exosome sample (in μL); and D is the total quantity of loaded exosomes (in μg of protein or total particles).

2.10. Characterization of Exosomes

Total protein was extracted from exosomes using RIPA Lysis Buffer (Biomiga, Santiago, USA). Subsequently, the following proteins were detected using a capillary‐based automated Western blot analysis system (ProteinSimple, California, USA): ALIX (ab275377, Abcam, 1:3000), CD63 (ab68418, Abcam, 1:3000), and GM130 (ab52649, Abcam, 1:3000).

Images of siRNA‐loaded and non‐loaded exosomes were acquired using transmission electron microscopy (TEM, JEM‐1200, JEOL Ltd., Japan). Simultaneously, their diameter and concentration were determined by nanoparticle tracking analysis (NTA, Nano Sight NS300, Malvern, UK).

Fluorescent co‐localization of siRNA and exosomes was observed under a confocal microscope (LSM980, Carl Zeiss, Germany). The siRNA was labeled with FAM (Cell Signaling Technology, Danvers, MA, USA), and the exosomes were labeled with PKH26 (Sigma‐Aldrich). Following labeling, the samples were washed via ultracentrifugation in PBS at 100000  g for 2 h. The resulting pellet was resuspended in 200 μL of PBS. The zeta potential of exosomes, both before and after siRNA loading, was measured using nanoparticle tracking analysis.

2.11. Labeling of EXO‐siPebp1 and Uptake by R28 Retinal Precursor Cells In Vitro

The R28 cell line is an immortalized, adherent retinal precursor cell line derived from the retinas of neonatal Sprague–Dawley rats. This cell line is commonly used for in vitro studies on the neuroprotection and physiological functions of retinal ganglion cells (RGCs). The R28 cells were kindly provided by the Department of Anatomy and Neurobiology at Central South University (Changsha, China). The R28 cells were cultured in low‐glucose DMEM (11 885 084, Gibco, USA) supplemented with 10% fetal bovine serum (0500, ScienCell, USA), 1% non‐essential amino acids (M7145, Sigma, USA), 1% penicillin–streptomycin (60162ES76, Yeasen, China), and 1% L‐glutamine (G3126, Sigma, USA).

To examine the uptake of exosomes by retinal ganglion cells (RGCs), EXO‐siPebp1 was labeled with PKH26 (Sigma‐Aldrich) according to the manufacturer's instructions. RGCs were incubated with 50 μg/mL of the labeled EXO‐siPebp1 in serum‐free medium for 1, 2, 4, or 6 h, respectively. Subsequently, the cells were washed with PBS, and the intracellular distribution of the exosomes was observed using a fluorescence microscope (Nikon, Japan).

2.12. Ocular and Systemic Distribution of EXO‐siPebp1

To investigate the ocular distribution and systemic metabolism of exosomes following intravitreal injection, SD rats were administered a 2 μL (1 mg/mL) injection of Cy5.5‐labeled EXO‐siPebp1 into the vitreous body. The rats were euthanized at 1, 8, 24, and 72 h post injection. Following cardiac perfusion, their eyeballs and major organs (including the heart, liver, spleen, lungs, and kidneys) were collected. The Cy5.5 fluorescence signal in each tissue was then detected using an IVIS spectrum imaging system (PerkinElmer, USA). The fluorescence intensity was quantified using the region of interest (ROI) tool in the Living Image software (v4.2).

2.13. Toxicity Assessment

At 72 h post‐intravitreal injection of 2 μL (1 mg/mL) EXO‐siPebp1, the rats were anesthetized and euthanized. Their eyeballs, serum, and major organs (including the heart, liver, spleen, lungs, and kidneys) were then collected. Apoptosis in retinal tissue was assessed using the terminal deoxynucleotidyl transferase‐mediated dUTP nick end labeling (TUNEL) assay. Serum biochemical parameters related to hepatic and renal function were measured using an automated biochemical analyzer (Chemray 800/240, Rayto, China). The histomorphological structure of each organ was examined by H&E staining.

2.14. Animal Treatment

To evaluate the therapeutic advantage of EXO‐siPebp1, the animals were allocated into the following six experimental groups: (1) Normal control group; (2) pathologically high intraocular pressure (ph‐IOP) model + PBS group (2 μL/eye); (3) ph‐IOP model + EXO‐siPebp1 group (2 μL, 100 μg/mL); (4) ph‐IOP model + EXO group (2 μL, 100 μg/mL); (5) ph‐IOP model + liposome‐siPebp1 group (2 μL, containing 80 nM siRNA); (6) ph‐IOP model + AAV‐shPebp1 group (2 μL, 1.5 × 10¹² vg/mL). All interventions were administered via intravitreal injection immediately after model induction.

2.15. Electrophysiological Examination

After the rats were anesthetized, they were placed on a table, and three recording electrodes were inserted separately under the skin of the anterior bregma (cathode), occipital bone (anode), and ear (ground electrode). Unilateral flash VEP data were obtained from the right and left eye in a dark room. After the recordings were completed, the rats were placed in a warmed cage for recovery. A professional technician performed the data recording process in a blinded manner. VEP was performed using the Roland ophthalmic electrophysiological diagnostic system (Germany).

2.16. Retrograde Labeling RGCs by Superior Colliculus Injection

Three days prior to sacrifice, rats were anesthetized and secured in a stereotaxic apparatus (RWD Life Science, Shenzhen, China) using blunt ear bars. A volume of 0.5 μL of 4% Fluoro‐Gold (FG; Fluorochrome, Denver, CO, USA) solution was then slowly injected into the superior colliculi bilaterally at the following coordinates relative to bregma: 6.0 mm posterior, 1.8 mm lateral to the midline, and 4.0 mm deep from the skull surface, as previously described36. Three days post‐injection, the retinas containing retrogradely labeled FG(+) cells were collected and flat‐mounted on microscope slides. FG‐positive retinal ganglion cells (RGCs) were visualized using a fluorescence microscope (DM5000 B; Leica, Wetzlar, Germany). For each retina, twelve images were captured at specified radial distances from the optic disc (0.625 mm, 1.25 mm, and 1.875 mm, representing central to peripheral retina) within the superior, inferior, nasal, and temporal quadrants. FG‐labeled RGCs in each micrograph were counted using Image‐Pro Plus 6.0 software.

2.17. Retina Morphological Analysis

Anesthetized rats were subjected to cardiac perfusion with pre‐chilled saline. The entire eyeballs were enucleated and fixed overnight at 4°C in FAS eyeball fixative (G1109; Servicebio, Wuhan, China). Subsequently, the fixed eyeballs were dehydrated, embedded in paraffin, and sectioned at a thickness of 5 μm near the optic nerve. Retinal sections were stained with H&E using a commercial kit (G1005; Servicebio). To better characterize retinal morphological changes, we measured the total retinal thickness and ganglion cell complex (GCC) thickness in each retinal section using Image‐Pro Plus 6.0 software (Media Cybernetics, Rockville, MD, USA). The GCC, which comprises the retinal nerve fiber layer, ganglion cell layer, and inner plexiform layer, corresponds to the anatomical distribution of RGCs in the retina [35]. Measurements were taken perpendicular to the retinal pigment epithelium at positions ±360, ±720, ±1080, ±1440, ±1800, and ±2160 μm from the center of the optic nerve. GCC thickness was quantified by performing four cross‐sectional measurements within a 1‐mm area surrounding the optic nerve in each retina using Case Viewer 2.3 software (3DHISTECH).

Retinal immunofluorescence (IF) staining was performed as previously described [24]. The following primary antibodies were used: anti‐ACSL4 (ab155282; Abcam; 1:150), anti‐GPX4 (ab125066; Abcam; 1:100), and anti‐RBPMS (1832‐RBPMS; PhosphoSolutions, Aurora, CO, USA; 1:500).

2.18. Reverse Transcription‐Quantitative Polymerase Chain Reaction (RT‐qPCR)

The total RNA was extracted using TRIzol, and reverse transcription of total RNA to cDNA was performed using a QuantiTect reverse transcription kit (Qiagen). RT‐qPCR was performed using SYBR Green Master Mix (Qiagen, Hilden, Germany). For tissues, results were normalized to the internal control GAPDH, and the primer sequences were: GAPDH (forward: 5′‐CTGGAGAAACCTGCCAAGTATG‐3′; reverse: 5′‐GGTGGAAGAATGGGAGTTGCT‐3′) and Pebp1 (forward: 5′‐TTTAACTGCACTCGGGACTCG‐3′; reverse: 5′‐CCATGAAATGCTGCTTGGTCTAT‐3′).

2.19. Western Blot Analysis

Total protein was extracted from retinal tissues using RIPA Lysis Buffer (Biomiga, Santiago, USA) [36]. Subsequently, the following proteins were detected using a capillary‐based automated Western blot analysis system (ProteinSimple, California, USA): GPX4 (ab125066; Abcam; 1:3000), PEBP1 (GB113118; Servicebio, Wuhan, China), ACSL4 (ab155282; Abcam; 1:3000), and β‐actin (GB15003‐100; Servicebio, Wuhan, China).

2.20. Measurement of Retinal Oxidative Stress

Retinal MDA content was measured using a Lipid Peroxidation MDA Assay Kit (S0131S; Beyotime, Shanghai, China). Retinal NADPH levels were quantified with an NADP/NADPH Assay Kit (ab65349; Abcam, Cambridge, UK). Retinal GSH content was assessed using a Micro Reduced Glutathione (GSH) Assay Kit (BC1175; Solarbio, Beijing, China). All measurements were normalized to the protein concentration of the samples. All experimental procedures were performed according to the protocols provided in the respective kits [37, 38, 39].

2.21. Statistical Analysis

Sample sizes were based on standard protocols in the field, and at least three biological independent replicates were performed for each experiment. Experiments were blinded to the person performing data extraction and analysis. SPSS version 22.0 (IBM, Armonk, NY, USA) and GraphPad Prism 9.0 (GraphPad Software, La Jolla, CA, USA) were used for statistical analyses. Data are expressed as the mean ± standard deviation (SD). Prior to applying parametric tests, normality was assessed using the Shapiro–Wilk test for each group, and homogeneity of variance was assessed using Levene's test (for multi‐group comparisons) or F‐test (for two‐group comparisons). For datasets meeting both assumptions (Shapiro–Wilk p > 0.05 and variance homogeneity p > 0.05), comparisons between two groups were performed using two‐tailed Student's t‐test, and comparisons involving more than two groups were performed using one‐way ANOVA followed by Tukey's honestly significant difference (HSD) post hoc test. For datasets violating either assumption, non‐parametric alternatives were applied: the Mann–Whitney U test for two‐group comparisons and the Kruskal–Wallis test followed by Dunn's post hoc test for multi‐group comparisons. Statistical significance was set at p < 0.05.

3. Results

3.1. Multi‐Omics Analysis Identifies Pebp1 as a Potential Ferroptosis Target in Ph‐IOP‐Induced Retinal Ganglion Cell Injury

To identify the molecular mechanisms of retinal ganglion cell (RGC) death induced by pathological high intraocular pressure (ph‐IOP), we conducted single‐cell RNA sequencing (scRNA‐seq) on adult rat retinal tissues at three time points after ph‐IOP induction (2 h, 24 h, and 7 days), capturing transcriptional dynamics across acute, subacute, and chronic stages. Following quality control, data processing, and annotation (Figure S2), Uniform Manifold Approximation and Projection (UMAP) analysis revealed 32 distinct cell clusters. Cell Marker annotation further identified 14 cell types, with a particular focus on RGCs, highlighting the substantial cellular heterogeneity induced by ph‐IOP injury (Figure 1A). Similarly, spatial transcriptomics on ocular sections from corresponding time points, analyzed with the cell2location algorithm, confirmed the predominant localization of RGCs to the inner retinal layer and optic nerve head, consistent with established histopathology (Figure 1B). To screen for potential therapeutic targets, we analyzed RGC transcriptomes post‐ph‐IOP (Figures S3–S5), revealing a specific gene cluster that was significantly activated during early injury, providing candidate targets for early intervention against RGC injury (Figure 1C). Given the established role of ferroptosis in RGC death, yet with specific regulatory targets remaining elusive, we prioritized the analysis of differentially expressed genes (DEGs) within this early‐injury RGC cluster. A dot plot screening identified ferroptosis‐associated genes, among which Pebp1 expression was notably upregulated (Figure 1D,E). Phosphatidylethanolamine‐binding protein 1 (PEBP1), a known regulator of lipid metabolism, has been shown to drive ferroptosis in pathologies such as asthma, renal failure, and brain injury [40]. To investigate its role in glaucoma, we performed Gene Set Enrichment Analysis (GSEA) using the FerrDB database [41](a dedicated resource for ferroptosis regulators and disease associations), which demonstrated significant enrichment of Pebp1 in ferroptosis driver signaling pathways (Figure 1F). Furthermore, multi‐omics in situ spatial sequencing confirmed Pebp1 expression within RGCs, revealing a dynamic pattern characterized by an initial increase followed by a decline (Figure 1G). Finally, RT‐qPCR validated that Pebp1 mRNA levels were significantly elevated in the retina at 2 and 24 h post‐ph‐IOP compared to controls (Figure 1H).

FIGURE 1.

Multi‐omics Analysis Identifies Pebp1 as a Potential Ferroptosis Target in ph‐IOP‐Induced Retinal Ganglion Cell Injury. (A) UMAP clustering plot generated after quality control, normalization, and dimensionality reduction of single‐cell data using Seurat. Cell populations were annotated into 14 types using the scType automated annotation tool and published marker genes, including: Amacrine cells (AC), Astrocytes, Bipolar cells (BC), Cones, Endothelial cells (EC), Erythroid cells, Epithelial stem cells (ESC), Fibroblasts, Macrophages, Microglia, Müller glial cells, Retinal ganglion cells (RGC), Rods, and Retinal pigment epithelial cells (RPE). (B) Spatial distribution and abundance of RGCs across four spatial transcriptomics (ST‐seq) samples. Color intensity represents relative cellular abundance. (C) Gene expression trend analysis in RGCs identifies an early‐upregulated gene cluster following injury. (D) A dot plot displaying differentially expressed genes (DEGs) in RGCs during the early injury phase (thresholds: p‐value < 0.05 and |log₂FC| > 0.5). Red and blue dots represent significantly up‐ and down‐regulated genes, respectively. (E) Violin plot shows significant upregulation of Pebp1 expression in RGCs at 2 and 24 h post ph‐IOP injury. (F) Gene Set Enrichment Analysis (GSEA) based on the FerrDB database. The x‐axis shows the rank of genes in the expression dataset; the y‐axis shows the running enrichment score. Enrichment in the ferroptosis driver pathway suggests the biological process regulated by the target gene. Line color indicates expression direction (red for upregulation, blue for downregulation), and line width represents correlation strength. (G) Multi‐omics in situ spatial sequencing (MIP‐seq) validates the spatiotemporal expression pattern of Pebp1 in retinal sections across different time points. Nuclei are stained with DAPI (blue), RGCs are labeled with Tubb3 (green), and Pebp1 is shown in red. Scale bar: 50 μm. (H) qRT‐PCR analysis of Pebp1 expression levels in retinal tissues at different time points (n = 3 in each group). Data are presented as the mean ± SEM. ns, not significant; ***p ≤ 0.001 (compared with the control group using one‐way ANOVA followed by Tukey's HSD post hoc test).

In summary, these findings identify Pebp1 as a pivotal driver of RGC ferroptosis in the ph‐IOP model, suggesting it as a potential and specific therapeutic target for glaucoma.

3.2. AAV‐Mediated Pebp1 Knockdown Protects RGCs via Ferroptosis Inhibition

Given the critical role of Pebp1 as a lipid metabolism regulator in ferroptosis, we first verified its expression dynamics in the ph‐IOP model. At 24 h post‐injury, both immunofluorescence and western blotting revealed markedly elevated PEBP1 protein levels in the retina relative to controls (Figure 2A–C), implicating PEBP1 in the acute injury response. To perform in vivo functional validation, we screened three distinct short hairpin RNAs (shRNAs) targeting Pebp1. Three weeks after intravitreal injection, RT‐qPCR confirmed that AAV3‐shPebp1 achieved the highest silencing efficiency (87.4%; Figure 2D) and was thus selected for all subsequent experiments. To determine whether AAV‐shPebp1‐mediated neuroprotection operates through ferroptosis inhibition, rats were assigned to four groups—control, control+AAV‐shPebp1, ph‐IOP, and ph‐IOP + AAV‐shPebp1—and retina tissues were collected 1 day post‐injury for analysis. Measurement of the lipid peroxidation product malondialdehyde (MDA) revealed a significant increase in retinal MDA content in the ph‐IOP group compared to controls, which was markedly suppressed by AAV‐shPebp1 treatment (Figure 2E). Concurrently, retinal levels of the nicotinamide adenine dinucleotide phosphate (NADPH) and antioxidants glutathione (GSH) were significantly diminished in the ph‐IOP group, and AAV‐shPebp1 treatment effectively restored them (Figure 2F,G), suggesting that Pebp1 knockdown rescues oxidative stress‐induced antioxidant stress. Western blot analysis of key ferroptosis regulators further demonstrated that, relative to the ph‐IOP group, AAV‐shPebp1 treatment significantly downregulated the proteins PEBP1 and ACSL4 while upregulating the ferroptosis suppressor GPX4 (Figure 2H–K). Immunofluorescence staining confirmed these findings: in the retinal ganglion cell layer, ph‐IOP injury enhanced ACSL4 signal intensity and attenuated GPX4 signal, both of which were reversed by AAV‐shPebp1 intervention (Figure 2L,M). Furthermore, these protein alterations were largely confined to the inner retina, indicating that ferroptosis predominantly occurs in RGCs.

FIGURE 2.

AAV‐mediated Pebp1 Knockdown Protects RGCs via Ferroptosis Inhibition. (A) Representative photomicrographs of immunofluorescence staining for PEBP1 (green) in retinal ganglion cells (counterstained with RBPMS; red). (B) Representative micrographs of Simon Western blot analysis for PEBP1 in control and ph‐IOP‐injured rats. (C) Western blotting detection of retinal levels of PEBP1 (normalized to that of actin; n = 3 in each group). (D) Retinal Pebp1 mRNA levels in rats 3 weeks after intravitreal injection of pAAV‐spgRNA‐EGFP (control) or pAAV‐shPebp1‐EGFP, assessed by RT‐qPCR (n = 3 in each group). Comparison of MDA (E), NADPH (F), and GSH (G) contents between control and ph‐IOP injured groups, with or without AAV‐sh Pebp1 treatment (n = 5 in each group). (H) Representative micrographs of Simon Western Blot Analysis for PEBP1, ACSL4, and GPX4. (I–K) Western blotting detection of retinal levels of PEBP1, ACSL4, and GPX4 (normalized to that of actin; n = 3 in each group). Representative photomicrographs of immunofluorescence staining for ACSL4 (L; green) and GPX4 (M; green) in retinal ganglion cells (counterstained with RBPMS; red). Data are shown as the mean ± SD; **p < 0.01, ***p ≤ 0.001 (compared with the control group using one‐way ANOVA followed by Tukey's HSD post hoc test); ^# p < 0.05, ^## p < 0.01, ^### p < 0.001 (compared with the ph‐IOP group using one‐way ANOVA followed by Tukey's HSD post hoc test).

In summary, these results demonstrate that ph‐IOP injury activates the ferroptosis pathway in RGCs, and that AAV‐mediated Pebp1 knockdown suppresses ferroptosis—via attenuation of lipid peroxidation, restoration of antioxidant defenses, and regulation of the GPX4/ACSL4 axis—thereby conferring neuroprotection.

3.3. Pebp1 Knockdown Alleviates Ph‐IOP‐Induced Impairments in RGC Survival, Retinal Structure and Visual Function

To evaluate whether Pebp1 knockdown alleviates ph‐IOP‐induced retinal structural and functional impairment, we performed histological analyses, retrograde RGC labeling, and electrophysiological recordings, as outlined in the experimental schematic (Figure 3A). Hematoxylin and eosin (H&E) staining revealed that, compared to the ph‐IOP group, Pebp1 knockdown (ph‐IOP + AAV‐shPebp1) significantly preserved ganglion cell complex (GCC) thickness—comprising the retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), and inner plexiform layer (IPL) (p < 0.001; Figure 3B,C). To quantify RGC survival, we performed retrograde labeling with fluorogold. This analysis demonstrated that Pebp1 knockdown significantly increased RGC survival relative to the ph‐IOP group (p < 0.001; Figure 3D,E). Beyond morphological preservation, the maintenance of visual function is equally critical. We therefore employed flash visual evoked potential (FVEP) recordings to objectively evaluate the functional integrity of the entire visual pathway from the retina to the occipital cortex. Consistent with the histological findings, Pebp1 knockdown significantly attenuated ph‐IOP‐induced visual dysfunction (p < 0.001; Figure 3F,G).

FIGURE 3.

Pebp1 Knockdown Alleviates ph‐IOP‐Induced Impairments in RGC Survival, Retinal Structure, and Visual Function. (A) This study employed a pathologically high intraocular pressure (ph‐IOP) model to evaluate the neuroprotective effects of Pebp1 knockdown on the retina. Sprague–Dawley rats first received an intravitreal injection of AAV‐shPebp1 to generate Pebp1‐knockdown rats 21 days later. The ph‐IOP retinal injury was then established by anterior chamber perfusion (110 mmHg for 60 min). The experimental timeline is summarized as follows: Fluorogold (FC) retrograde tracing was performed on Day 4. On Day 7, retinal function and structure were assessed using flash visual evoked potential (FVEP) detection, hematoxylin and eosin (H&E) staining, and retinal ganglion cell (RGC) counting. (B) Representative H&E‐stained images of retinas from control, ph‐IOP‐injured, and ph‐IOP‐injured rats 3 weeks after AAV‐shPebp1 injection. (C) Quantification of ganglion cell complex (GCC) thickness. The GCC includes the retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), and inner plexiform layer (IPL), indicated by double‐headed arrows. (D) Representative images of fluorogold‐labeled surviving RGCs in retinas from control, ph‐IOP injured, and ph‐IOP‐injured rats 3 weeks after AAV‐shPebp1 injection. (E) Quantitative analysis of RGC counts. (F) Representative flash visual evoked potential (FVEP) waveforms from control, ph‐IOP injured, and ph‐IOP injured rats 3 weeks after AAV‐shPebp1 injection. (G) Quantitative analysis of N1‐P1 amplitudes for each group. Data are shown as the mean ± SD; ***p ≤ 0.001 (compared with the control group using one‐way ANOVA followed by Tukey's HSD post hoc test); ^### p < 0.001 (compared with the ph‐IOP group using one‐way ANOVA followed by Tukey's HSD post hoc test). Scale bars: 20 μm (B), 100 μm (D).

In summary, AAV‐mediated Pebp1 knockdown confers multi‐level protection against ph‐IOP injury, preserving retinal structure, RGC survival, and visual function. These findings further validate Pebp1 as a promising therapeutic target for glaucoma.

3.4. Preparation and Characterization of Exosome‐siPebp1

Although the preceding results established Pebp1 as a key driver of RGC ferroptosis in ph‐IOP injury, the 3‐week timeframe required for AAV‐shRNA to reach peak silencing conflicts sharply with the clinical reality of acute glaucoma, in which RGCs are irretrievably lost within hours to days. To overcome this therapeutic time‐window constraint, we therefore explored an exosome‐mediated siRNA delivery strategy capable of immediate post‐injury intervention. We hypothesized that intravitreal delivery of Exosome‐siPebp1 would enable rapid, targeted intervention in RGCs following ph‐IOP injury, thereby intercepting ferroptosis‐driven neuronal loss during the acute phase [42]. To this end, we isolated exosomes from mesenchymal stem cell (MSC) culture supernatants by ultracentrifugation and loaded siPebp1 using exosome transit peptide (ETP) technology, thereby constructing the engineered delivery system Exosome‐siPebp1 (experimental schematic, Figure 4A). Immunofluorescence staining confirmed that the MSCs were positive for CD73, CD90, and CD105, verifying their identity as reliable exosome‐producing cells (Figure 4B). Fluorometric quantification using a microplate reader determined that the siRNA encapsulation efficiency was 15.8% per 10⁹ exosomes (Figure 4C). To comprehensively validate siPebp1 loading at the nanoscale, we characterized both naive exosomes (EXO) and Exosomes‐siPebp1 (EXO‐siPebp1) using nanoparticle tracking analysis (NTA), transmission electron microscopy (TEM), zeta potential measurement, confocal microscopy, and western blot [43, 44, 45]. NTA revealed mean hydrodynamic diameters of 108 ± 37 nm for EXO and 118 ± 42 nm for EXO‐siPebp1 (Figure 4D). TEM imaging showed that both displayed the characteristic cup‐shaped bilayer morphology (Figure 4E). A significant shift in zeta potential following siPebp1 loading further confirmed successful incorporation of the nucleic acid cargo into the exosomes (Figure 4F). Confocal microscopy demonstrated pronounced co‐localization of FAM‐labeled siPebp1 with PKH26‐labeled exosomal membranes, directly supporting efficient siRNA loading (Figure 4G). Western blot analysis confirmed that the exosomal markers CD63 and ALIX were highly enriched in both EXO and EXO‐siPebp1, whereas the Golgi marker GM130 was nearly undetectable, indicating high purity and an absence of organelle contamination (Figure 4H).

FIGURE 4.

Preparation and Characterization of Exosome‐siPebp1. (A) Schematic diagram illustrating the preparation protocol for EXO‐siPebp1, including collection of conditioned medium from cell culture, isolation of exosomes via ultracentrifugation, and loading of siPebp1 using the Exosomal Transport Peptide (ETP) technique. (B) MSC membrane markers (CD73, CD90, and CD105) were identified via immunofluorescence. (C) Encapsulation efficiency of siPebp1 per 10⁹ exosomes. (D) Nanoparticle tracking analysis (NTA) showing the size distribution of naive exosomes (EXO) and EXO‐siPebp1. (E) Representative transmission electron microscopy (TEM) images of EXO and EXO‐siPebp1. Scale bar, 100 nm. (F) Comparison of zeta potential between EXO and EXO‐siPebp1. (G) Confocal microscopy image of EXO‐siPebp1 after ETP‐mediated loading. Colocalization of exosomes (red, PKH26) and siPebp1 (green, FAM) is highlighted. Scale bar, 1 μm. (H) WB analysis exosomal marker proteins (ALIX, CD63) and a negative marker (GM130) Data are presented as mean ± SD; Student's t‐test was used for comparison between groups. **p < 0.01, ***p ≤ 0.001.

In summary, these data demonstrate the successful construction of the EXO‐siPebp1 engineered exosome platform, providing a foundation for subsequent functional validation and in vivo delivery.

3.5. Cellular Uptake and Biodistribution of Exosome‐siPebp1

To evaluate the feasibility of EXO‐siPebp1 as an siRNA delivery vehicle, we first assessed its uptake by RGCs. PKH26‐labeled EXO‐siPebp1 was incubated with RGCs, and its intracellular localization was tracked by confocal microscopy at 1, 2, 4, and 6 h post‐incubation. These experiments revealed that PKH26‐labeled EXO‐siPebp1 was efficiently internalized by RGCs, with intracellular fluorescence intensity progressively increasing over time (Figure 5A,B). We next evaluated the in vivo intraocular delivery efficiency of this system. Rats received a single intravitreal injection of Cy5.5‐labeled EXO‐siPebp1 (2 μL, 1 mg/mL); in vivo imaging system (IVIS) detected a robust intraocular fluorescence signal as early as 1 h post‐injection, confirming efficient ocular delivery (Figure 5C). This signal remained detectable for up to 72 h, indicating prolonged intraocular retention of EXO‐siPebp1 (Figure 5D). Biodistribution analysis in major organs was performed by IVIS imaging at 1, 8, 24, and 72 h post‐injection. Cy5.5 fluorescence accumulated predominantly in the liver and kidney, peaking at 24 h, while only minimal signals were detected in the lung, heart, and spleen (Figure 5E), indicating that circulating exosomes are primarily cleared via hepatobiliary and renal routes. To confirm the in vivo gene‐silencing efficacy of EXO‐siPebp1, retinal tissues were collected 24 h after ph‐IOP injury and analyzed by immunofluorescence, western blotting, and RT‐qPCR. These results demonstrated that intravitreal EXO‐siPebp1 significantly reduced both PEBP1 protein and mRNA levels in the retina relative to controls (Figure S6), confirming effective in vivo target silencing.

FIGURE 5.

Cellular Uptake and Biodistribution of Exosome‐siPebp1. (A) Representative confocal microscopy images showing time‐dependent cellular uptake of PKH26‐labeled EXO‐siPebp1 (red) by R28 retinal precursor cells over 1, 2, 4, and 6 h. Scale bars, 50 μm (overview), 100 μm (magnified view). (B) Quantitative analysis of fluorescence intensity of R28 cells at 1, 2, 4, and 6 h after incubation with PKH26‐labeled EXO‐siPebp1 (n = 5 per group). (C) Representative ex vivo fluorescence images showing the biodistribution of Cy5.5‐labeled EXO‐siPebp1 within the rat eye and major organs at indicated time points. (D) Quantitative analysis of real‐time in vivo fluorescence intensity in rat eyes at different time points (1, 8, 24, and 72 h) following intravitreal injection of Cy5.5‐labeled EXO‐siPebp1 (n = 6 per group). (E) Temporal changes in fluorescence intensity in major organs following intravitreal injection of Cy5.5‐labeled EXO‐siPebp1 (n = 3 per group). The data represent as mean ± SEM. *p < 0.05, **p < 0.01, ***p < 0.001, Mann–Whitney test in (B), student's t test (two‐tailed) in (D).

Collectively, EXO‐siPebp1 exhibits efficient RGC uptake, rapid intraocular entry with prolonged retention following intravitreal injection, and robust in vivo target gene silencing—thus establishing a solid feasibility foundation for subsequent therapeutic efficacy studies.

3.6. Therapeutic Efficacy of Exosome‐siPebp1 In Vivo

The preceding results demonstrated that Pebp1 knockdown confers neuroprotection against ph‐IOP‐induced RGC injury and established the successful construction of the EXO‐siPebp1 delivery platform. On this basis, we hypothesized that EXO‐siPebp1 would offer superior therapeutic efficacy over conventional delivery strategies in a clinically relevant scenario of immediate post‐injury intervention. To test this hypothesis, rats with IOP‐induced retinal injury were randomized into five groups receiving a single intravitreal injection of PBS, EXO‐siPebp1, unloaded EXO, Liposome‐siPebp1, or AAV‐shPebp1, with untreated normal rats as controls. Seven days post‐injection, H&E staining, fluorogold retrograde labeling, and FVEP recordings were performed (experimental schematic, Figure 6A). H&E staining revealed that ph‐IOP injury resulted in marked thinning of the GCC, disorganization of the nerve fiber layer, and RGC pyknosis with nuclear condensation at day 7. Among all treatment groups, EXO‐siPebp1 produced the most pronounced restoration of retinal architecture (Figure 6B,E). Quantification of fluorogold retrograde labeling demonstrated a significant reduction in RGC numbers in the ph‐IOP group relative to normal controls. The EXO‐siPebp1 group exhibited the highest density of RGC‐positive cells per unit area among all treatment groups—2.28‐fold higher than the ph‐IOP group (Figure 6C,F). Unloaded Exosome, Liposome‐siPebp1, and AAV‐shPebp1 all exhibited partial RGC protection, but none approached the efficacy of EXO‐siPebp1. For functional assessment, FVEP recordings revealed that ph‐IOP injury markedly attenuated the N1‐P1 amplitude. EXO‐siPebp1, unloaded EXO, and Liposome‐siPebp1 each ameliorated the FVEP amplitude deficit to varying degrees, with EXO‐siPebp1 achieving the greatest recovery (Figure 6D,G).

FIGURE 6.

Therapeutic Advantages of EXO‐siPebp1 In Vivo. (A) This study employed a Pathologically high intraocular pressure (ph‐IOP) model to evaluate the neuroprotective effects of different treatments on the retina. The experimental timeline is summarized as follows: On Day 0, a rat model of ph‐IOP retinal injury was established by anterior chamber perfusion (110 mmHg for 60 min). Immediately after modeling, intravitreal injections were administered, with the treatment regimens being PBS, EXO‐siPebp1, EXO, liposome‐siPebp1, and AAV‐shPebp1. Subsequently, fluorogold retrograde tracing was performed on Day 4, and retinal function and structure were assessed on Day 7. These assessments included electroretinography, hematoxylin and eosin (H&E) staining following eyeball dissection, and quantification of fluorogold‐labeled retinal ganglion cells (RGCs). (B) Representative H&E‐stained images of retinas from the different treatment groups. (E) Quantitative analysis of the ganglion cell complex (GCC) thickness. The GCC, comprising the retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), and inner plexiform layer (IPL), is indicated by double‐headed arrows. (C) Representative images of surviving fluorogold‐positive RGCs in the retinas across treatment groups. (F) Quantitative analysis of RGC counts. (D) Representative flash visual evoked potential (FVEP) waveforms from each treatment group. (G) Quantitative analysis of the N1‐P1 amplitudes for each group. Data are shown as the mean ± SD; n = 8/group. *p < 0.05, **p < 0.01 (compared with the control group using one‐way ANOVA followed by Tukey's HSD post hoc test); ^## p < 0.01 (compared with the shPebp1 group using one‐way ANOVA followed by Tukey's HSD post hoc test). Scale bars: 20 μm (B), 100 μm (C).

In summary, in ph‐IOP‐induced acute RGC injury, EXO‐siPebp1 exhibited significantly superior therapeutic efficacy compared to unloaded exosomes and conventional delivery systems including liposomes and AAV vectors. Notably, under this immediate post‐injury dosing design and the 7‐day assessment window, EXO‐siPebp1 provided greater early neuroprotection than AAV‐shPebp1. This comparison primarily reflects the more rapid onset of siRNA‐mediated silencing and confers significant neuroprotection, underscoring its translational potential.

3.7. Safety Assessment of Exosome‐siPebp1 Treatment in Rats

To assess the in vivo safety of EXO‐siPebp1, we conducted toxicity evaluations in the retina and major organs including the heart, liver, spleen, lung, and kidney. TUNEL staining revealed no significant difference in the number of retinal apoptotic cells between EXO‐siPebp1‐treated and saline‐treated control groups (Figure 7A), indicating that intravitreal administration of EXO‐siPebp1 did not induce appreciable retinal apoptosis. Serum biochemistry analysis demonstrated that hepatic function markers alanine aminotransferase (ALT) and aspartate aminotransferase (AST), as well as renal function markers blood urea nitrogen (BUN) and creatinine (CREA), did not differ significantly from control levels, and all values remained within normal reference ranges (Figure 7B). H&E staining of major organs revealed intact tissue architecture in all EXO‐siPebp1‐treated rats, with no evidence of inflammatory infiltration, necrosis, or fibrosis (Figure 7C). The FVEP recordings were performed at 7 days post‐injection. Quantitative analysis revealed no significant difference in N1‐P1 amplitudes between the EXO‐siPebp1‐treated group and the PBS control group (Figure S7).

FIGURE 7.

Safety Assessment of Exosome‐siPebp1 Treatment in Rats. (A) Representative TUNEL assay images of eye sections from saline‐treated controls (CON) or EXO‐siPebp1 treated rat. (B) Serum biochemical parameters of rats treated with EXO‐siPebp1 or PBS (n = 5 per group). ALT, alanine aminotransferase; AST, aspartate aminotransferase; BUN, blood urea nitrogen; CRE, creatinine. (C) Representative hematoxylin and eosin (H&E)‐stained sections of the heart, liver, spleen, lungs, and kidneys from rats treated with either EXO‐siPebp1 or PBS (control). Scale bar = 100 μm.

In summary, EXO‐siPebp1 was well tolerated following intravitreal injection, with no observable acute systemic toxicity or local retinal injury within the monitoring window, thereby providing an initial safety foundation for further preclinical development.

4. Discussion

This study aimed to identify specific therapeutic targets of ferroptosis in ph‐IOP‐induced retinal ganglion cell (RGC) injury and to explore engineered exosome–siRNA delivery strategies capable of achieving intervention during the acute phase of glaucoma. Integrating single‐cell and spatial transcriptomics with in vivo functional validation, we identify the lipid metabolism regulator Pebp1 as a key driver of RGC ferroptosis in acute glaucoma: Pebp1 is robustly and selectively upregulated in injured RGCs, where it promotes ferroptosis via the GPX4/ACSL4 signaling axis, and its targeted silencing effectively preserves RGC survival, retinal architecture, and visual function. To resolve the inherent mismatch between the delayed onset of AAV‐mediated silencing and the rapid, irreversible RGC loss characteristic of acute glaucoma, we engineered exosome‐siPebp1—MSC‐derived exosomes loaded with siPebp1—which enabled single‐dose intervention immediately after injury, significantly outperforming unloaded exosomes, liposomal formulations, and AAV vectors in RGC protection, without detectable acute toxicity. In summary, these findings implicate PEBP1 in ph‐IOP‐associated RGC ferroptosis and support exosome‐mediated siRNA delivery as a rapid therapeutic strategy for acute glaucomatous injury.

Identifying the specific molecular drivers of RGC ferroptosis is central to overcoming the current therapeutic impasse in glaucoma. By integrating single‐cell and spatial transcriptomics, we uncovered a dynamic expression trajectory of Pebp1 in injured RGCs—characterized by rapid acute‐phase upregulation, subacute‐phase peak, and chronic‐phase return to baseline—with spatial enrichment in the retinal ganglion cell layer and functional convergence on ferroptosis driver pathways. This spatiotemporally restricted expression profile positions Pebp1 as a candidate molecular hub that couples ph‐IOP‐derived mechanical and ischemic insults to the ferroptotic execution program [46, 47]. Consistent with our observations, prior evidence indirectly implicates Pebp1 in ferroptosis regulation: as a phosphatidylethanolamine‐binding protein, PEBP1 can bind lipid hydroperoxides and propagate membrane lipid peroxidation chain reactions, and has been reported to drive ferroptosis‐dependent cell death in models of lung injury, asthma, and hepatocellular carcinoma [48, 49, 50]. However, none of these prior studies explored the functional role of Pebp1 in the context of glaucoma or retinal disease, nor evaluated it as an actionable therapeutic target in vivo. Thus, our multi‐omics discovery provides the glaucoma field with a novel potential target for ferroptosis‐directed therapy.

PEBP1 belongs to the phosphatidylethanolamine‐binding protein family and is a highly conserved, multifunctional protein implicated in cell development, proliferation, apoptosis, and ferroptosis regulation [51, 52, 53]. To test the hypothesis that Pebp1 drives RGC ferroptosis through lipid peroxidation, we performed AAV‐shRNA‐mediated silencing in vivo. Pebp1 knockdown markedly suppressed the accumulation of the lipid peroxidation product malondialdehyde (MDA) and restored glutathione (GSH) and NADPH levels, demonstrating that Pebp1 deficiency mitigates the redox imbalance triggered by ph‐IOP. Concurrently, the ferroptosis suppressor GPX4 was upregulated and the pro‐ferroptotic protein ACSL4 was downregulated, indicating that Pebp1 governs RGC ferroptosis through the GPX4/ACSL4 signaling axis. In vivo functional experiments further demonstrated that targeted silencing of Pebp1 significantly enhanced RGC survival in the ph‐IOP model. These findings align with the established ferroptosis‐regulatory function of Pebp1 in disease models including renal injury, asthma, and cerebral ischemia [40, 54, 55]. However, Wei et al. reported that Raf‐1 Kinase Inhibitory Protein (RKIP, also known as PEBP1) expression decreased after optic nerve crush and that RKIP overexpression promoted RGC survival and axonal regeneration through ERK/AKT signaling [56]. This apparent discrepancy may reflect differences in injury paradigm, observation period, cellular context, and downstream signaling pathways. Prior work has demonstrated that PEBP1 complexes with 15‐lipoxygenase isoforms (15LO1/15LO2) to catalyze the production of phospholipid hydroperoxides (PUFA‐PE‐OOH), which competitively inhibit GPX4‐mediated substrate reduction and thereby promote ferroptosis execution [57, 58]. This mechanism offers a plausible molecular basis for the GPX4 suppression and heightened lipid peroxidation observed in our study. However, whether PEBP1 drives RGC lipid peroxidation in the ph‐IOP model through the canonical 15‐LOX/PEBP1 complex pathway, or whether it engages a retina‐specific regulatory mechanism, remains to be determined through further investigation.

Although AAV‐shRNA‐mediated Pebp1 silencing effectively protects RGCs through ferroptosis inhibition, the 3‐week timeframe required for peak transgene expression is fundamentally incompatible with the clinical trajectory of acute glaucoma, in which RGCs are irretrievably lost within hours to days—patients cannot realistically receive prophylactic gene therapy 3 weeks before an acute attack. This clinical reality provided the central impetus for transitioning from a slow‐onset gene therapy approach to a rapidly acting interventional strategy [59, 60]. Naked siRNA is inherently susceptible to nuclease degradation, lacks tissue tropism, and is unable to independently traverse biological barriers [61, 62]. Moreover, the retina is a highly specialized neurosensory tissue whose blood‐retinal barrier—a network of complex tight junctions—while indispensable for maintaining the photoreceptor microenvironment, poses a formidable obstacle to intraocular drug delivery [63, 64]. An ideal siRNA delivery vehicle must therefore reconcile multiple demands: protection of the nucleic acid cargo, penetration of biological barriers, efficient cytosolic release, and favorable biocompatibility. The EXO‐siPebp1 system developed in this study—MSC‐derived exosomes engineered to carry siPebp1—satisfies these criteria. In contrast to prior studies employing unmodified exosomes alone, EXO‐siPebp1 exerts its neuroprotection through a dual‐synergy mechanism: MSC‐derived exosomes inherently supply neurotrophic factors and anti‐inflammatory molecules that support neural repair, while siPebp1 simultaneously silences the pro‐ferroptotic target gene, thereby interrupting lipid peroxidation‐driven RGC loss. This “carrier‐cargo” functional integration achieves concurrent ferroptosis inhibition and neurorepair, rather than the mere additive effects of its individual components [65, 66, 67, 68]. More importantly, head‐to‐head comparisons demonstrated that a single intravitreal injection of EXO‐siPebp1 administered immediately after ph‐IOP injury significantly outperformed unloaded exosomes, Liposome‐siPebp1, and AAV‐shPebp1 in preserving RGC numbers, GCC thickness, and FVEP N1‐P1 amplitude, with no detectable acute systemic or local toxicity. These findings convincingly establish that the therapeutic superiority of EXO‐siPebp1 arises not from the exosomal carrier alone, but from the synergistic action of the siRNA payload and the delivery vector, while its favorable safety profile provides a foundation for further preclinical advancement. Collectively, our findings support EXO‐siPebp1 as a rapid post‐injury delivery strategy with therapeutic potential in acute glaucoma.

Several limitations should be acknowledged. First, the relatively small sample sizes limit the statistical power of some analyses, and the discovery‐stage transcriptomic datasets lacked independently replicated sequencing libraries. For scRNA‐seq, retinal cell suspensions from three animals were pooled to generate one library per time point, whereas ST‐seq used one eye from one animal per time point. Inter‐animal variability therefore cannot be distinguished from temporal injury‐associated effects, and the DEG, GSEA, and spatial findings should be considered hypothesis‐generating rather than population‐level estimates. Although RT‐qPCR and quantitative PEBP1/RBPMS co‐immunofluorescence in larger independent cohorts supported the temporal Pebp1 expression pattern, these targeted validations do not establish the reproducibility of the broader transcriptomic programs. Moreover, Visium provides spot‐level rather than single‐cell resolution, making cell‐type localization dependent on probabilistic integration with scRNA‐seq. Independently replicated sequencing libraries and higher‐resolution spatial methods are therefore required. Second, the multi‐omics associations do not by themselves establish causal mechanisms, and the absence of an in vivo EXO‐siNC group precludes formal exclusion of nonspecific or off‐target siRNA effects. Although EXO‐siPebp1 outperformed unloaded EXO across multiple functional outcomes, this comparison supports but does not prove a Pebp1‐dependent mechanism. Future studies should incorporate non‐targeting siRNA controls and cell‐type‐specific genetic or ligand‐receptor perturbation experiments in adequately powered cohorts. Third, efficacy was evaluated only in the ph‐IOP rat model, and safety assessment was limited to retinal TUNEL staining, serum hepatic and renal biochemistry, and major‐organ histology 72 h after a single intravitreal injection. These findings therefore indicate preliminary acute tolerability, but not long‐term ocular, systemic, or immunological safety. This limitation is particularly relevant because human umbilical cord MSC‐derived exosomes were administered xenogenetically to rats; delayed inflammation, complement activation, anti‐exosome antibody formation, altered pharmacokinetics, and the consequences of repeated dosing remain unknown. Future studies should include longitudinal single‐ and repeated‐dose assessments at 7, 14, and 28 days and at least 8–12 weeks; evaluate ocular and systemic inflammation, complement activation, anti‐exosome IgM/IgG and neutralizing activity, biodistribution, and clearance; compare xenogeneic with species‐matched EV preparations; and confirm efficacy and safety in large‐animal glaucoma models. Clinical translation will also require scalable, standardized manufacturing with consistent critical quality attributes and improved RGC‐targeted delivery.

In summary, by integrating single‐cell and spatial transcriptomics, this study identifies PEBP1 as a key molecular target that bridges ph‐IOP injury and RGC ferroptosis, and delineates its regulatory mechanism through the GPX4/ACSL4 signaling axis. To overcome the narrow therapeutic time window inherent to acute glaucoma, we engineered EXO‐siPebp1, an exosome‐siRNA delivery system that enables immediate single‐dose intervention post‐injury, and demonstrated that it outperforms conventional delivery strategies in RGC protection with a favorable safety profile in vivo. This work supports a cell‐free strategy for acute glaucomatous injury and provides mechanistic evidence relevant to targeted intervention in other acute neurodegenerative conditions.

Author Contributions

Yijia Huang conceptualized the study, curated and investigated the data, performed formal analysis and validation, developed the methodology and software, acquired funding, created visualizations, and wrote the original draft. Di Gong participated in data curation, formal analysis, methodology, and visualization. Junhong Guo contributed to data curation, formal analysis, methodology development, and software. Wei Huang and Shengbin Tang provided critical resources. Tingyu Hu contributed to methodology and visualization. Li Li provided resources. Bingkai Feng and Kuanrong Dang were involved in data curation and methodology. Simin Deng and Yong Liu contributed to data curation. Chunyan Tan contributed to the study conceptualization, supervised the research, and reviewed and edited the manuscript. Fei Yao conceptualized and supervised the study, curated data, performed formal analysis and investigation, acquired funding, developed and validated the methodology, administered the project, and wrote and reviewed/edited the manuscript. Jiantao Wang contributed to the study conceptualization and manuscript review and editing. All authors approved the final version of the manuscript for submission.

Funding

This work was funded by the Shenzhen Fund for Guangdong Provincial High‐level Clinical Key Specialties (SZGSP014), the National Natural Science Foundation of China (82201182), the Shenzhen Key Medical Discipline Construction Fund (SZXK037), the Shenzhen Science and Technology Program (JCYJ20220818103207015), the Shenzhen Science and Technology Program (KCXFZ20230731093359004), the Sanming Project of Medicine in Shenzhen (SZSM202311012, SZSM202411009), the Natural Science Foundation of Hunan Province, China (2024JJ6670), and the Basic Research Project of Shenzhen (JCYJ20240813152617023).

Ethics Statement

All animal experiments were approved by Xiangya School of Medicine, Central South University (Ethical Approval Code: xy2023101236). The use of human umbilical cord tissue for mesenchymal stem cell isolation was approved by the Institutional Ethics Committee of Changsha Hospital for Material and Child Health Care (Ethical Approval Code: EC‐2021106‐07), with written informed consent obtained from all donors in accordance with the Declaration of Helsinki and ISSCR guidelines.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Visium Spatial Tissue Optimization Experiment Results Image (A) Bright‐field imaging of the rat eyeball tissue optimization slide. (B) Corresponding fluorescence imaging of the permeabilization experiment (showing the negative control, permeabilization times of 3 min, 6 min, 12 min, 18 min, 24 min, 30 min, and positive control, magnification: 5×; exposure time: 200 ms).

Figure S2: Bubble plots illustrate the expression patterns and characteristics of marker genes used for annotating each cell cluster (Circle sizes indicate the percentage of cells expressing the gene in a particular cluster; colors represent average gene expression).

Figure S3: Bar charts compare the proportional abundance of each cell cluster across different experimental groups. The compositional proportions of distinct cell types were analyzed, including amacrine cells (AC), astrocytes, bipolar cells (BC), cones, endothelial cells (EC), erythroid cells, epithelial stem cells (ESC), fibroblasts, macrophages, microglia, Müller glial cells, retinal ganglion cells (RGCs), rods, and retinal pigment epithelial (RPE) cells. The results indicate that rod photoreceptors constitute the predominant cell type across all groups. Compared to the control group, the proportions of ACs, BCs, and microglia were significantly increased at 2 h post‐model induction, whereas the proportions of cones and ESCs were markedly reduced. Several cell types exhibited dynamic, time‐dependent alterations in their relative abundances. Notably, by day 7 post injury (ph‐IOP 7d), the proportions of erythroid cells, fibroblasts, macrophages, and RGCs were ultimately lower than those in the sham group. Müller glial cell proportions initially increased at the 2 h (ph‐IOP 2 h) and 24 h (ph‐IOP 24 h) time points, followed by a subsequent decline by ph‐IOP 7d. These findings demonstrate that injury systematically alters the cellular composition of the retina, suggesting that these cell types may contribute to the pathological process through modulation of their functional states.

Figure S4: Temporal Gene Expression Patterns in Retinal Ganglion Cells (RGCs) During ph‐IOP Injury: Analysis of Expression Trends for Six RGC Subclusters (C1‐C6).

Figure S5: Functional Pathways Enriched in Retinal Ganglion Cells (RGCs) During the ph‐IOP Injury Time Course. Heatmap displaying the significantly enriched Gene Ontology Biological Process (GO‐BP) terms for genes in Clusters 1, 2, and 4 (red indicates upregulation, blue indicates downregulation).

Figure S6: The immunofluorescence (A), western blotting (B) and RT‐qPCR (C) analysis of Pebp1 expression in the retinal tissues of control and ph‐IOP injured rats, treated with or without EXO‐siPebp1 (n = 3 in each group). Data are presented as the mean ± SD; ***p ≤ 0.001 (compared with the control group using one‐way ANOVA followed by Tukey's HSD post hoc test); ^## p < 0.01 (compared with the ph‐IOP group using one‐way ANOVA followed by Tukey's HSD post hoc test). Scale bars: 50 μm.

Figure S7: (A) Representative flash visual evoked potential (FVEP) waveforms between the PBS control group and the EXO‐siPebp1‐treated group. (B) Quantitative analysis of N1‐P1 amplitudes for each group.

Table S1: Cell marker.

Table S2: Changes in retinal Pebp1 levels in normal rat receiving AAV‐shRNA injection.

Data Availability Statement

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.