cGAMP promotes inner blood-retinal barrier breakdown through P2RX7-mediated transportation into microglia

Abstract

Background

Impairment of the inner blood-retinal barrier (iBRB) leads to various blinding diseases including diabetic retinopathy (DR). The cGAS-STING pathway has emerged as a driving force of cardiovascular destruction, but its impact on the neurovascular system is unclear. Here, we show that cGAMP, the endogenous STING agonist, causes iBRB breakdown and retinal degeneration thorough P2RX7-mediated transport into microglia.

Methods

Extracellular cGAMP and STING pathway were determined in tissue samples from patients with proliferative DR (PDR) and db/db diabetic mice. Histological, molecular, bioinformatic and behavioral analysis accessed effects of cGAMP on iBRB. Single-cell RNA sequencing identified the primary retinal cell type responsive to cGAMP. Specific inhibitors and P2RX7-deficienct mice were used to evaluate P2RX7’ role as a cGAMP transporter. The therapeutic effects of P2RX7 inhibitor were tested in db/db mice.

Results

cGAMP was detected in the aqueous humor of patients with PDR and elevated in the vitreous humor with STING activation in db/db mouse retinas. cGAMP administration led to STING-dependent iBRB breakdown and neuron degeneration. Microglia were the primary cells responding to cGAMP, essential for cGAMP-induced iBRB breakdown and visual impairment. The ATP-gated P2RX7 transporter was required for cGAMP import and STING activation in retinal microglia. Contrary to previous thought that mouse P2RX7 nonselectively transports cGAMP only at extremely high ATP concentrations, human P2RX7 directly binds to cGAMP and activates STING under physiological conditions. Clinically, cGAMP-induced microglial signature was recapitulated in fibrovascular membranes from patients with PDR, with P2RX7 being predominantly expressed in microglia. Inhibiting P2RX7 reduced cGAMP-STING activation, protected iBRB and improved neuron survival in diabetic mouse retinas.

Conclusions

Our study reveals a mechanism for cGAMP-mediated iBRB breakdown and suggests that targeting microglia and P2RX7 may mitigate the deleterious effects of STING activation in retinal diseases linked to iBRB impairment.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12974-025-03391-w.

Introduction

The blood-retinal barrier (BRB) contains two distinct sites, the outer (o) BRB separating the choriocapillaris from the retina and the inner (i) BRB surrounding the retinal vasculature [1]. The iBRB is composed of endothelial cells, tight junctions, basement membrane and pericytes, which is analogous to the blood brain-barrier (BBB), protecting against potentially disruptive circulating substances by separating the vascular system from the neural parenchyma of the retina. Failure of the barrier function leads to vessel leakage and retinal oedema, release of inflammatory molecules and infiltration of immune cells [2]. Therefore, disruption of iBRB is a common feature in diverse retinal diseases, such as DR, age-related macular degeneration (AMD), retinal vein occlusion and uveitis [3, 4]. Currently, anti-vascular endothelial growth factor (VEGF) therapies are the first major treatment for both oBRB and iBRB breakdown. However, resistance to the treatments, recurrence of BRB disruption, and the development of retinal atrophy are common challenges in clinic, highlighting the need for more effective approaches for treating BRB disruption [5–7].

The cyclic GMP-AMP synthase (cGAS)—Stimulator of Interferon Genes (STING) is a fundamental signalling axis that senses cytosolic DNA and provokes inflammation in response to pathogens [8–10]. 2′3′-cyclic GMP-AMP (cGAMP) is the endogenous second message generated by cGAS after binding to DNA [8, 11–13]. cGAMP then binds to and activates STING, which initiates recruitment and activation of the kinase TBK1, leading to the activation of transcription factors such as IRF3/IRF7 or NFκB, and consequent production of inflammatory cytokines including interferon and interleukins [8, 14, 15]. In addition to its fundamental roles in initiating host defense against invading pathogens, chronic activation of cGAS-STING plays a crucial role in inflammatory diseases, autoimmune diseases, neurodegeneration, and aging [16–19].

Emerging evidences show cGAS-STING signalling as potential therapeutic target in major blind-causing eye diseases, including AMD, DR and glaucoma [20, 21]. We and others have recently shown the activation of cGAS-STING signalling in patients with AMD [22, 23], and inhibition of cGAS or STING alleviates the death of retinal neurons and retinal pigment epithelium (RPE) in AMD-like mouse models [22–25]. In addition, activation of cGAS-STING was detected in an oxygen-induced retinopathy (OIR) mouse model, which mimics certain aspects of diabetic retinopathy. Inhibiting STING reduced retinal vascular neovascularization and leakage in OIR [26]. However, the direct effects of STING activation on BRB and the underlying cellular and molecular mechanisms remain undefined.

Intratumoural injections of cGAMP or its analogues have demonstrated robust efficacy in murine cancer models, thus attracting considerable attention for clinical cancer treatment [27]. However, extracellular cGAMP cannot passively cross cell membranes because of its negative charge; thus, it must be imported into the cells by transmembrane transporters. Currently, five cGAMP transporters have been identified including channels (SLC19A1, SLC46A2 and LRRC8A:C/E) [28–32], transporters (ABCC1) [33] and pore (P2RX7) [34]. The ATP-gated P2X purinoceptor 7 receptor (P2RX7) is commonly expressed in immune cells and the nervous system [35, 36]. In mouse macrophages, in the presence of high concentrations of ATP, P2RX7-gated pores are opened, allowing the entry of macromolecules, such as cGAMP [34]. However, direct binding of P2RX7 to cGAMP remains unclear. Furthermore, the cGAMP transporter functions in a highly tissue- and species-dependent manner [28, 30]. However, whether P2RX7 transports cGAMP in the nervous system has not yet been determined.

DR is the most common microvascular complication of diabetes mellitus and the leading cause of blindness in working age [37, 38]. DR is generally characterized by iBRB disruption, including capillary degeneration, pericyte loss and vascular leakage. Retinal neuron inflammation and degeneration are also associated with DR pathogenesis [3]. Here, we detected extracellular cGAMP in PDR, an advanced stage of DR featured with disorganized angiogenesis and severe vitreous hemorrhage. We found a previously unknown impact of cGAMP in iBRB disruption and neuron degeneration, resembling key aspects of DR in both phenotype and gene signature. Finally, our study revealed that inhibition of cGAMMP transporter P2RX7 and thus STING activation in microglia improved iBRB integrity and neuron survival in diabetic mice.

Methods

Clinical sample collection

The usage of patient samples was approved by the Ethical Committee of the Zhongshan Ophthalimic Center, Sun Yat-sen University (Ethics Approval #IIT2023127). All involved patients were given the written informed consents prior to clinical sample collection. Aqueous humor (AH) samples were obtained from patients with PDR who required vitrectomy (n = 8). About 100 µl of AH samples were collected under a surgical microscope. The samples were cooled in ice and clarified by centrifugation (10000 rpm, 15 min, 4 ℃) before stored in liquid nitrogen.

Animals

Tmem173 (Sting) -KO mice in the C57BL/6 J background (Cat. NO. NM-KO-2105967) were obtained from Shanghai Model Organisms Center, Inc. P2rx7-KO mice in the C57BL/6 J background (Strain NO. T011510) were purchased from GemPharmatech (Nanjing, China). Twelve-week-old male BKS-Leprem2Cd479/Gpt (db/db) mice and control (m/m) mice (Strain NO. C000110) were obtained from Changzhou Cavens Model Animal Co., Ltd. (Changzhou, China). The blood glucose levels of db/db mice were ≥ 16 mmol/L. WT C57BL/6 J mice were purchased from Sun Yat-Sen University Laboratory Animal Center. Male and female mice of 6–8 week-old were randomly used for all experiments unless indicated otherwise. Mice were housed under a 12:12 light: dark cycle with ad libitum access to standard mouse chow. All procedures conformed according to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the Animal Use and Care Committee of Zhongshan Ophthalmic Center at the Sun Yat-Sen University, Guangzhou, China (Approval form #2022031).

Intravitreal and intraperitoneal drug administration

The 2′3′-cGAMP was prepared in PBS at concentrations of 0.1–10 mM. PBS served as the vehicle control. A total volume of 1 μl of 2′3′-cGAMP or PBS was injected into the vitreous cavity of each eye of mice using a Hamilton microsyringe as previously described [39], and illustrated in Fig. 1B. A740003 was prepared in a solution consisting of 5% DMSO, 40% Polyethylene glycol 300, 5% Tween 80, and 50% H₂O, at concentrations of 5 mg/kg or 50 mg/kg. The buffer used for A740003 preparation served as the vehicle control. I.p injection of A740003 or the vehicle control was administered 1 h prior to light exposure or cGAMP injection, and was performed daily for 2 days.

Fig. 1.

cGAMP induces iBRB breakdown. A Diagrams show the mouse eye anatomy, the structure of retina, and the site for intravitreal injections or vitreous humor extraction in the vitreous cavity. GCL: ganglion cell layer, IPL: inner plexiform layer, INL: inner nuclear layer, OPL: outer plexiform layer, ONL: outer nuclear layer; RPE: retinal pigment epithelium. B ELISA analysis shows the concentration of cGAMP in the vitreous humor (VH) of control m/m and diabetic db/db mouse eyes. n = 6–8 mice. C WB analysis shows protein level in mouse retinas. Lower panel is the quantification results of WB. n = 6–7 mice. D, E RNA-seq analysis reveals the expression of cGAS-STING components. Published datasets were analyzed, including retinas from healthy controls (n = 5), retinas from patients with PDR (n = 5) (D) [44]; retinal neovascular membranes (RNV) from patients with PDR (n = 7), and internal limiting membranes (ILM) from patients with idiopathic macular holes (n = 7) (E) [45]. F Schematic diagram showing experimental procedure for G–L Mice received an intravitreal injection of 1 µl of PBS or cGAMP (10 mM). Analyses were conducted three days post-injection. G Left panels: In vivo fundus image and fundus fluorescein angiography (FFA) show retinal blood vessels. FFA was taken at the indicated time point after i.p. injection of fluorescein. Right panels: Quantification of superficial vessel diameters in the 0.3 and 0.4 mm circles centered in optic nerve head. Each dot represents value from one vessel. n = 6 mice per group. H Left panels: in vivo OCT shows retinal structure. Increased retinal thickness, detachment of retina (red arrows), abnormal retinal folding (red arrow heads) and infiltration of cells (white arrows) into the vitreous cavity were detected after cGAMP injection. Right panels: average thickness of the retinas in the 0.35 and 0.7 mm circles centered in optic nerve head. n = 6 mice per group. I Representative HE staining shows retinal morphology. Arrow heads indicate infiltrating immune cells and arrows indicate dilated or blood-filled vessels. Note retinal detachment was observed in all cGAMP but not in the PBS groups. Scale bar: 200 μm and 100 μm (enlarged boxed regions). n = 4 eyes per group. J Retinal flat mounts show hemorrhage spots. Scale bar: 1 mm, n = 8 eyes per group. K Representative three-dimensional images of IB4-labeled retinal vessels. Confocal z-stack images were captured using Zen 2.3 SP1 software. n = 3 eyes per group. L EM reveals ultrastructural of iBRB. In the cGAMP group, increased RBCs in the lumen, perivascular edema (asterisks), basement membrane degeneration (arrowheads), and abnormal tight junctions (arrows) were observed. BM, basement membrane; EC: endothelial cell; L: lumen; TJ: tight junction; P: pericyte. All results are presented as mean ± SD, *P < 0.05, **P < 0.005, ***P < 0.001. Figure 1 B,C and H: Mann–Whitney test; Fig. 1G: unpaired t-test

Optical coherence tomography (OCT) and fundus imaging

The experiment was conducted as we previously described [24, 25]. Briefly, mice were anesthetized and pupils were dilated by Tropicamide Phenylephrine Eye Drops, and the cornea was lubricated with Hypromellose GEL. OCT was performed on both eyes with a Heidelberg, Spectralis OCT device (Heidelberg Engineering). Thickness measurements were performed with a circular ring scan (circle diameter 1, 3, 6 ETDRS), centered on the optic nerve head. Central retinal thickness was calculated from four fields around the optic nerve head using the Heidelberg Eye Explorer Software. Fundus images were obtained using the Micron IV retinal imaging system (Phoenix Research Laboratories, Pleasanton, CA). Mice were i.p. injected with 2% fluorescein sodium solution (Alcon laboratories, TX, USA) (5 μl/g), and fluorescein angiographic images were recorded immediately.

Electroretinography (ERG)

ERG analysis was conducted as described before [25]. Briefly, mice were dark-adapted overnight and anesthetized with 1% pentobarbital sodium. Pupils were dilated with 1–2 drops of Tropicamide Phenylephrine Eye Drops and the cornea was lubricated with Hypromellose GEL. The ERG was recorded using a Diagnosys Celeris rodent ERG device with electrodes were placed on top of each cornea. For dark-adapted ERG, mice were stimulated by flash light varying in intensity. Analysis were performed using customized Espion ERG Data Analyzer software (version 6.63.26).

Visual behavior by optomotor test

An optomotor eye tracker system (Stria Tech) was used to detect the visual acuity of mice as we previously described [25]. The optomotor response detects animals by gazing at a moving environment/object and turning the head (eye movement) to stabilize the image on the retina. Mice were kept in a closed box during the whole process, providing a constant rotation speed (12°/s) and a constant contrast (99.72%) of the black and white stripes. The staircase method is used to determine the spatial frequency (cyc/deg), starting from 0.056 to 0.50 cycles per degree, and measuring the maximum rotation frequency that the mouse can pass. Check clockwise first, then counterclockwise. Four monitors were used to simulate the stripes, and they could be observed through the camera above the box. The instrument’s software uses a special algorithm to measure the scores of the mice under different parameters.

Western blotting (WB) analysis

The retinas were dissected in PBS and suspended in 150 μl of RIPA (per retina) containing a proteinase inhibitor cocktail and a protein phosphatase inhibitor. The total proteins were extracted by brief sonication. The tissue lysis was centrifuged at 13,000 g for 15 min at 4 ℃. The supernatant was mixed with SDS sample buffer and heated for 10 min at 100 ℃. The antibodies used are listed in the Table S2. The original blots were shown in Figure S10.

Retinal fat mounts

The procedure was conducted as described before [25]. The mouse eyeballs were enucleated immediately after death and fixed in 4% paraformaldehyde (PFA) for 5 min. A small incision was then made in the cornea, and the eyes were further fixed for an additional 5 min. After fixation, the eyeballs were dissected, and the retinas were isolated and fixed in 4% PFA for 5 more min, then washed three times with PBS. For immunostaining, the retinas were permeabilized by 0.3% PBST (0.3% Triton X-100 in PBS) for 20 min at room temperature (RT), and blocked with 2.5% BSA (prepared in PBST) for 30 min at RT. The retinas were then incubated with the primary antibody overnight at 4 ℃ and incubated with secondary antibodies for 2 h at RT. After washing with PBS, the retinas were mounted with Antifade Mounting Medium. Images were acquired using the LSM980 confocal microscope (Carl Zeiss, Germany).

Microglia depletion

Mice were fed with control chow (AIN-76) or chow containing 1200 ppm of PLX5622 for 8 days before cGAMP administration. During the period of the PLX5622-supplemented diet, no obvious behavioral or health problems were observed.

Macrophage depletion

Clodronate liposomes (around 5 mg/ml) were injected into mice at a concentration of 10 ml/kg to deplete macrophages. The control groups of mice were intravenously injected with the same amount of liposomes without clodronate. Detailed procedure is shown in Figure S8A.

Surface plasmon resonance (SPR) analysis

SPR was performed according to the manufacturer’s instructions. The CM5 sensor chip (Cytiva, 29149603) was first activated with 0.4 M 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride and 0.1 M N-hydroxysuccinimide at a flow rate of 30 μl/min for 400 s. Human P2RX7 (Sangon #D145129) was diluted (50 μg/ml) with 200 μl of sodium acetate (pH 4.5) and then flowed across the activated surface to couple with the chip to an 8000-target response value (RU). The remaining activated sites on the chip were blocked by 85 μl of ethanolamine (1 M, pH 8.5). Real-time detection was recorded using a Biacore 8 K instrument (Cytiva, USA) at a flow rate of 30 ml/min. Then, in order to detect the equilibrium dissociation constant (KD) between ATP and P2RX7, and between cGAMP and P2RX7, serially diluted ATP or cGAMP (in PBS-P, pH 7.4) was applied to analyze the interactions with P2RX7 coupled on the surface of the chip using the Biacore 8 K Evaluation Software.

Peripheral blood mononuclear cells (PBMC) isolation and macrophage differentiation

Whole blood from healthy, human donors was collected in 10-ml EDTA blood tubes with written informed consent. Bulk PBMCs were isolated by density gradient centrifugation using Lymphoprep as previously described [40]. The buffy coat was washed with PBS and resuspended in RPMI 1640 medium, and then used immediately for cGAMP treatment. Alternatively, PMBCs were cultured in RPMI 1640 medium supplemented with 10% (v/v) heat-inactivated FBS and 1% penicillin–streptomycin for one week to allow differentiated into macrophage for further analysis.

Cell culture and treatment

THP-1 cells were cultured in RPMI1640 containing 10% fetal bovine serum and 1% penicillin–streptomycin, respectively. For cGAMP stimulation, cells were pretreated with 100 μM of indicated inhibitors for 30 min before adding cGAMP of indicated concentrations. The treatment was performed in serum-free medium.

Immunofluorescence (IF) staining of cryosections

After fixed in 4% PFA for 1 h, the mouse eyeballs were then dehydrated gradually using 10 and 20% sucrose prepared in 0.1 M PBS and embedded in optimal cutting temperature (OCT) compound. Cryosections (thickness: 8–10 μm) through the optic disk of each eye were prepared. For IF staining, the OCT was washed by PBS, and then blocked with 2.5% BSA (prepared in PBST) for 30 min at RT. The sections were then incubated with the primary antibody overnight at 4℃ and incubated with secondary antibodies for 2 h at RT. Images were captured by the LSM980 confocal microscope (Carl Zeiss, Germany).

Transmission electron microscope imaging

Eyeballs were fixed in 2.5% glutaraldehyde for a minimum of 4 h. The samples were then washed three times with phosphate-buffered saline (PBS) for 10–15 min each. Following this, the samples were fixed in 1% osmium tetroxide for 1–2 h and washed again with PBS. Next, the samples underwent a graded ethanol series for dehydration, followed by dehydration with 100% acetone. Infiltration was performed at room temperature using acetone:embedding resin mixtures, and then with pure embedding resin at 4 °C overnight. The samples were subsequently embedded in molds.Polymerization of the embedded samples was conducted in an oven at 37 °C for 24 h, followed by 60 °C for 48 h. Ultrathin sections (~ 100 nm) were cut using an ultramicrotome (Leica UC7). The sections were stained with uranyl acetate for 20 min and then with lead citrate for 12 min.Finally, stained sections were observed and photographed using a transmission electron microscope (Tecnai G2 Spirit, Thermo FEI).

Hematoxylin and eosin (HE) staining

After fixed in FAS eye fixation solution overnight, the mouse eyeballs were then dehydrated gradually using 60%, 70%, 80%, 90%, 100% ethanol and then embedded with paraffin. The thickness of paraffin sections was 8–10 μm. After dewaxing with xylene and hydration with ethanol, paraffin sections were subjected to hematoxylin and eosin (HE) staining, and then mounts were made with neutral resin. Stained retinal sections were imaged with TissueFAXS microscopy.

cGAMP quantification in mouse vitreous humor (VH)

The mouse VH was obtained using a Hamilton needle, with a volume of 3–5 μl obtained from each eye. VH from the control and LD mice were centrifuged at 13,000 rpm for 30 min and the supernatant was collected for ELISA using a competitive 2′3′-cGAMP ELISA kit.

cGAMP pulldowns

HEK293 cells were transfected with 10 μg of either FLAG or FLAG-PXRX7 plasmids. The biotin-cGAMP-streptavidin beads were prepared by incubating biotin-cGAMP (1 μg) with streptavidin beads (10 μl) for 1 h at RT in binding buffer (50 mM Tris–HCl, pH 7.5, 1 mM EDTA, 150 mM NaCl, 10% (v/v) glycerol). Cells were harvested 24 h post-transfection and lysed in binding buffer containing proteinase inhibitor cocktail and protein phosphatase inhibitor. Proteins were extracted by passing through 26G syringe, and 0.6 mg of total proteins were incubated with biotin-cGAMP-streptavidin beads in binding buffer. After a 3 h incubation at RT, the beads were washed four times with binding buffer at RT. The bound proteins were eluted by boiling in SDS–PAGE sample loading buffer for 5 min at 95 °C. The soluble fraction was then analyzed by WB.

Retinal leukostasis assay

Mice were anesthetized and perfused with PBS containing 40 μg/ml FITC-ConA via the left ventricle. After eyeballs were collected and fixed with 4% PFA, retinas were dissected and stained with anti-CD45 antibody.

Permeability assay

Mice were injected with pre-warmed PBS (pH 7.4) containing 50 mg/ml FITC-dextran (4 kDa) via tail veins. After 2 h of circulation, mice were euthanized and perfused with PBS via the left ventricle to clear blood and unbound dye. The eyeballs were then collected and fixed with 4% PFA for 30 min before precede to IF analysis of retinal whole mounts.

Quantification of pericytes surface

The pericytes per capillary is manually determined based on their nuclei. The NG2 fluorescence intensity between two pericytes nuclei was measured by Image J, and mean intensity was obtained by dividing IF value to the selected area, which stands for the pericytes coverage.

RNA-seq analysis

The RNA isolation and sequencing data analysis was conducted as described before [25]. Briefly, two eyes from the same mouse were collected and directly lysed in 1 mL of TRIzol™ reagent. The purified RNAs were then sequenced. Bulk RNA-seq data can be accessed from NCBI Sequence Read Archive (SRA) database under accession number PRJNA1102708 and PRJNA1103420. Adapters were removed by Trim Galore v1.18. Raw sequencing data were mapped to the GRCm39 genome assembly using HISAT2 v2.2.1. The fragments per kilobase of exon per million (FPKM) was calculated by featureCounts v2.0.1. DEGs was identified using DEseq2 (fold change ≥ 2 and FDR ≤ 0.05). R package “clusterProfler” was used for function enrichment analysis on genes.

Preparation of single-cell suspension

At three days after intravitreal injection, four fresh retinas per sample were digested using the MACS Tumor Dissociation Kit (Miltenyi Biotec) for 30 min with agitation, according to the manufacturer’s instructions. The dissociated cells were subsequently passed through a 70 μm and 40 µm cell-strainer (Miltenyi Biotec) and centrifuged at 300 g for 10 min. After removing the supernatant, the pelleted cells re-suspended in washing buffer (0.04% bovine serum albumin [BSA] in Dulbecco’s phosphate-buffered saline [DPBS]) after washing two times with washing buffer. The final cell concentration was adjusted to 1000 cells/μl (viability ≥ 85%) in all samples by using Countess™ 3 Automated Cell Counter (Life).

Library construction for scRNA-seq

DNBelab C Series High-throughput Single-Cell RNA Library (940-000519-00) was utilized for scRNA-seq library preparation. In brief, Briefly, the cells were diluted to a concentration of 1000 cells/mL and loaded into the cell reservoir of the microfluidic chip. Barcoded beads and droplet-generation oil were successively added to the beads and oil reservoirs. Encapsulated droplets were generated and collected using a DNBelab C4 system. Beads that captured the mRNA were recovered for reverse transcription (RT). Following polymerase chain reaction (PCR) amplification by polymerase chain reaction, complementary DNA (cDNA) was purified and quan-tified using a Qubit dsDNA kit (Thermo Fisher Scientific). Libraries of 3′-end transcripts were subsequently constructed through cDNA fragmentation, size selection, end repair and A-tailing, adapter ligation, PCR for indexing libraries, and cyclisation of sequencing libraries, according to the manufacturer’s instructions. Sequencing libraries were purified and quantified using the Qubit ssDNA kit (Thermo Fisher Scientific) and Qsep100 (Bioptic).

Single-cell RNA sequencing using DNBelab C4

The DNBelab C4 Series Single-Cell Library Prep Set (MGI) was used for sequencing. DNBs were loaded into the patterned nano arrays and sequenced on the DNBSEQ-T7 sequencer with pair-end sequencing. The sequencing reads contained 30 bp read 1 (including the 10 bp cell barcode 1, 10 bp cell barcode 2 and 10 bp unique molecular identifiers (UMI)), 100 bp read 2 for gene sequences and 10 bp barcodes read for sample index. The raw data can be accessed with accession number PRJNA1103951.

scRNA-seq data visualization and analysis

For mouse retinal dataset, count matrices of the unique molecular identifiers were processed using Seurat R package [41]. Cells with more than 15% mitochondrial reads and < 300 genes were excluded. The NormalizeData function in the Seurat package was used to generate log-normalized expression values for the raw counts. SCTransform function was used to normalize and scale the expression matrices and identify highly variable genes in each sample. Following canonical correlation analysis-based integration for all datasets using FindIntegrationAnchors and IntegrateData functions, we performed principal component analysis (PCA) dimensionality reduction and graph-based cell clustering using the FindNeighbors (dims = 30) and FindClusters (resolution = 1.0) functions. The RunUMAP function was used to generate UMAP maps using the top 30 principal components. Signature genes for each cell population were identified by FindAllMarkers function (min.pct = 0.5, logfc.threshold = 0.5 and test.use = “MAST”) and retained with p-value < 0.05. Cell–cell communication was conducted using normalized scRNA-seq data by CellChat (https://github.com/sqjin/CellChat) with default parameter. Receptors and ligands expressed in at least 50 cells of specific cluster were further analyzed.

Human scRNA-seq was obtained from GEO (#GSE137537) [42]. The analysis was performed as described previously [25]. Briefly, low-quality cells (< 200 genes, > 6000 genes or mitochondrial genes > 10%) were removed. R package Seurat was used to perform clustering analysis of single cells. Clusters were visualized using t-distributed stochastic neighbor embedding (tSNE). Each cluster of cells was annotated by known marker genes from cellmarker 2.0 [43]. Microglia cells were selected by the special markers (P2ry12, Hexb, C1qa, Trem2, Fcrls, Tgfbr1, Olfml3 and Tmem119).

For scRNA-seq data analysis from patient with PDR, data were downloaded from GEO (#GSE245561) [48]. To score individual cells for pathway activities, we used the R package AUCell. First, for each cell we used an expression matrix to compute gene expression rankings in each cell with the AUCell_buildRankings function, with default parameters.For each gene set and cell, area-under-the-curve (AUC) values were computed (AUCell_calcAUC function) based on gene expression rankings, where AUC values represent the fraction of genes within the top-ranking genes for each cell that are defined as part of the pathway gene set.

Quantification and statistical analysis

Statistical analysis results are expressed as mean ± SD. GraphPad Prism 8.0 software (GraphPad software, Inc., La Jolla, CA) was used for statistical analysis. Differences between two groups were analyzed using an unpaired t-test. Multiple-group comparisons were performed by one-way ANOVA.

Results

Extracellular cGAMP is present in the eye of patients with PDR and leads to iBRB disruption

To measure extracellular cGAMP in the eye, ELISA was performed on both retina and vitreous humor (VH), a transparent fluid in the posterior cavity of the eye (Fig. 1A). Unexpectedly, the cGAMP level in VH was much higher than that in the retina (Figure S1), suggesting cGAMP are mainly present in secreted form within the eye. Elevated cGAMP levels were found in the VH of diabetic db/db mice (Fig. 1B), comparable to those in the aqueous humor of patients with PDR (Figure S2). Western blot (WB) analysis showed that cGAS-STING signaling was activated in the retinas of db/db mice (Fig. 1C), and an increase in cGAS-STING components was observed in the retinas of patients with PDR [44](Fig. 1D). Additionally, cGAS-STING signaling was activated in neovascular membranes from patients with PDR, compared to internal limiting membrane (ILM) from patients with macular hole (MH), where ILM serves as a scaffold for cell proliferation [45] (Fig. 1E). Next, intravitreal cGAMP injection was performed to directly investigate the effects of cGAMP on the retina (Fig. 1F). Unexpectedly, cGAMP treatment led to evident retina vascular leakage and dilation in vivo (Fig. 1G). Optical Coherence Tomography (OCT) showed that increased retinal thickness, possibly due to retinal oedema, and infiltration of cells into the vitreous cavity (Fig. 1H). Retinal vasodilation in the superficial vascular plexuses and the presence of blood cell-filled vessels within the superficial retina were evident following cGAMP injection (Fig. 1I). Retinal haemorrhages were also observed on retinal flat mounts after cGAMP injection (Fig. 1J). Three-dimensional confocal images further showed disruption of blood vessels in both the superficial and deep vascular plexuses following cGAMP treatment (Fig. 1K). A lower cGAMP dose induced leakage of red blood cells (RBCs) from capillaries, whereas leakage occurred in both the major vessel and capillaries upon treatment with higher cGAMP (Figure S3). Pericytes maintain iBRB integrity by regulating tight junction protein expression, and cGAMP injection decreased pericyte coverage (Figure S4). Finally, we used electron microscopy (EM) to examine the ultrastructure of retinal vessels (Fig. 1L). In larger vessels, cGAMP treatment resulted in an increased presence of RBCs within the vessel lumen, basement membrane (BM) collapse (arrowheads), and mitochondrial swelling in endothelial cells (EC) (Fig. 1L). In small vessels, cGAMP induced perivascular parenchymal oedema (stars), abnormal pericyte morphology and tight junction appearance (arrows) (Fig. 1L). Taken together, we found that extracellular cGAMP is elevated in diabetic mouse eye and causes iBRB breakdown.

STING is essential for cGAMP-induced iBRB disruption and retinal degeneration

Next, we confirmed that cGAMP-induced iBRB breakdown is structure specific, as 2′5′-GpAp, the hydrolyzed linear dinucleotide analog of cGAMP, has no impact of iBRB integrity (Figure S5). cGAMP binds to and activates STING to elicit an inflammatory response. Thus, we assessed the retinal structure and vessels in STING knockout (STING KO) mice. STING depletion suppressed cGAMP-induced retinal vasodilation and iBRB leakage (Fig. 2A–C). Next, we determined the effects of cGAMP on retinal neurones, non-neuronal Müller glial cells, and astrocytes (Fig. 2D). In control retinas, Müller cells and astrocytes (GFAP +) were confined to the inner margin of the retina, whereas cGAMP induced an increase in the GFAP signal in the inner plexiform layer (IPL), and depletion of STING suppressed cGAMP-induced GFAP immunoreactivity (Fig. 2E). In retinal neurones, cGAMP injection led to a reduction of retinal ganglion cells (RGCs, Brn3a +) and cell death in the ganglion cell layer (GCL), whereas STING-depleted retinas were protected from the cGAMP-induced RGC loss and cell death (Fig. 2F, G). Further, we accessed retinal neuron connectivity by labelling the ON-rod bipolar cells (PKCα +) and horizontal cells (calbindin +) (Fig. 2H). In control retinas, the horizontal cells exhibited a punctate staining pattern along the dendrites and axons that extended from the cell bodies and projected into the outter plexiform layer (OPL) (Fig. 2I, arrowheads). The retraction and loss of horizontal cell neurites were detected in cGAMP-injected retinas, which were reversed in STING KO mice (Fig. 2I, arrowheads). For rod bipolar cells, the control retinas showed abundant dendrites projecting from the cell body through the large dendritic arbours, while cGAMP injection led to the retraction of dendrites and reduction of dendritic branches compared to control retinas, and this phenotype was rescued by STING KO (Fig. 2J, arrowheads). In addition, contacts between the horizontal and bipolar processes were scarce in cGAMP-injected retinas, and were restored by STING depletion (Fig. 2K, arrows). Finally, we determined retinal response by ERG analysis. Injection of cGAMP resulted in a reduction of both photoreceptor phototransduction (a-wave) and rod bipolar cell depolarization (b-wave), while depletion of STING prevented the cGAMP-induced impairment of retinal function (Fig. 2L). Together, these results revealed that cGAMP leads to retinal vascular inflammation, neuronal degeneration in a STING-dependent manner.

Fig. 2.

cGAMP-induced retinal vascular inflammation and neurodegeneration depends on STING. Intravitreal injection of PBS or cGAMP (10 mM) was performed in wild type (WT) or STING knockout (STING KO) mice, followed by analysis three days post-injection. A FFA showing retinal blood vessels. n = 5 mice per group. B Retinal flat mounts showing hemorrhage spots. Scale bars: 1 mm, n = 5 eyes per group. C Left: Three-dimensional images of IB4-labeled retinal vessels and TER119-labeled RBCs. Right: Qualification results of IB4 area per field. n = 4 eyes per group. Mann–Whitney test. D Schematic image illustrating the localization of the Müller cells and neurons investigated. E–K IF analysis of the indicated proteins or dead cells (TUNEL) in retinal cryosections. The nuclei were counterstained by DAPI, n = 3–4 eyes per group. Scale bars: 50 μm. Results presented as mean ± SD, each dot represents a value from one capture filed. *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant, one-way ANOVA, Tukey's multiple comparisons test. E Left: GFAP immunostaining to label Müller cell and astrocytes. Right: quantification of GFAP intensity per captured field. F Left: Brn3a immunostaining to label retinal ganglion cells (RGCs). Right: Qualification of RGC cell number per captured field. G TUNEL labeling shows cell death after cGAMP injection in WT mouse retinas. Right: quantification of TUNEL- positive cells per captured field. H–K Immunostaining of PKCα (rod bipolar cells), calbindin (horizontal cells). I–K Enlarged images from boxed regions showing retracted dendrites (arrowheads) and deteriorated contacts (arrows) after cGAMP injection, rescued in STING KO retinas. L Representative dark-adapted ERG results. Right panels: Luminance-response results for the a- and b-waves. n = 6 mice per group. Ns: not significant, # or *p < 0.05; ## or **p < 0.01; ***p < 0.0005, (#: WT-cGAMP versus PBS, *: WT-cGAMP versus KO-cGAMP), two-way ANOVA, Tukey’s test

Retinal microglia are reprogrammed to neurodegenerative states by cGAMP

To comprehensively determine the effects of cGAMP in retina, scRNA-seq was performed on retinas of PBS- and cGAMP-injected mice (Fig. 3A). A total of 36,805 cells passed quality control were subject to unsupervised clustering analysis and 14 distinct clusters were identified (Fig. 3B). Differential expression gene (DEG) analysis revealed that microglia show the most pronounced gene alterations following cGAMP treatment (Fig. 3C). As microglia play a critical role in initiating retinal vascular degeneration and immune cell infiltration under pathogenic conditions [46, 47], we determined transcriptome alteration in microglia. The cGAMP-upregulated genes were enriched in interferon signalling, TYROBP causal network in microglia, pathogen phagocytosis, p38, IL-24, chemokine and VEGFA/VEGFR2 signallings (Fig. 3D). Microglia from cGAMP treatment increased genes associated with VEGFA/VEGFR2 signature (Gpx1, Stat1, Fn1, Cybb, Iqgap1, Sdcbp, Ldha and Mmp14), retinal degeneration signature (Lgals3, Lgals1, Lpl, Mmp12 and Spp1) and disease associated microglia signature (H2-Q7, B2m, Cd74, Apoe) [48, 49] (Fig. 3E). On the other side, cGAMP downregulated homeostasis genes in microglia (Fig. 3E). Next, the localization of microglia was investigated. In the control retina, the microglia displayed a typical ramified phenotype, with long and thin processes distributed along the retinal vessels (Fig. 3F). cGAMP injection led to an increased number of microglia, the appearance of an active amoeboid phenotype, and enhanced the STING signal in these reactive microglia (Fig. 3F). In retinal cryosections, cGAMP-induced contact between microglia/macrophages (IBA1 +) and vessel endothelial cells (CD31 +) was evident (Fig. 3G, b-c), and the presence of microglia/macrophages in the vessel lumen was detected (Fig. 3G, a). STING depletion inhibited cGAMP-induced microglia/macrophage activation (Fig. 3G). Bulk RNA-seq revealed that cGAMP treatment activated genes involved in antigen presentation, chemokine production and T cell activation (Fig. 3H). Consistent with the upregulation of leukocyte adhesion signature, retinal leukostasis showed significant increases in both leukocytes adhering to the retinal vessels and the infiltration of CD45-positive leukocytes into the retina after cGAMP administration (Fig. 3I, arrows). IF analysis showed increased major histocompatibility complex (MHC) II I-A/E alloantigens signal after cGAMP treatment, which was enriched in microglia (Fig. 3J). In addition, close contact between CD45-positive cells and MHC II-positive microglia were detected following cGAMP administration (Fig. 3K). These results suggest that microglia may become antigen-presenting cells (APCs), direct the recruitment of leukocytes into retina in response to cGAMP. Together, we found that cGAMP reprogrammes microglia from a homeostatic state to neurodegenerative, angiogenic, and disease-associated activation states. Additionally, cGAMP directs these reactive microglia towards retinal vessels, where they may act as APCs to trigger leukocyte infiltration.

Fig. 3.

cGAMP reprograms homeostatic microglia to neurodegenerative-associated states in mouse retina. Intravitreal injection of PBS or cGAMP (10 mM) was performed, followed by analysis three days post-injection. A–E scRNA-seq analyses of retinas with either PBS (CTRL) or cGAMP injection. Each group consists of two replicates, with each replicate containing four retinas. A Sequenced cells were visualized by Uniform Manifold Approximation and Projection (UMAP). B Clustering strategy of major retinal cell populations based on scaled expression heatmap of canonical markers for each cluster. Color scheme is based on z-score distribution. C The up- and down-regulated deferentially expressed genes (DEGs) from each cluster. D Enriched pathways of up-regulated genes within the microglia cluster following cGAMP treatment. E Representative neurodegenerative-associated genes up-regulated and homeostatic genes down-regulated in the microglia cluster following cGAMP treatment. F Representative images of retinal flat mounts for microglia (IBA1 +), blood vessel (IB4 +) and STING in PBS- or cGAMP-injected mice. Scale bars: 50 μm. n = 4 retinas per group. G Representative images from retinal cryosections showing microglia (IBA1 +) and endothelial cells (CD31 +). Nuclei counterstained by DAPI. Scale bar: 50 μm, n = 3 retinas per group. H Bulk RNA-seq analysis of retinas with either PBS (CTRL) or cGAMP injection, n = 4–5 retinas per group. Upper panel: gene set enrichment analysis (GSEA) profiles demonstrate significant enrichment of gene sets in the cGAMP group. Lower panel: Heatmap shows increased cell adhesion and antigen processing/presentation genes in the cGAMP group compared to the PBS. p < 0.05. I, J Representative three-dimensional confocal images of retinal whole mounts. n = 3 retinas. Vasculature and adherent leukocytes (arrows) were labeled with ConA-lectin (CoA), while infiltrated leukocytes were labeled with an anti-CD45 antibody (I). MHC II proteins were labeled by I-A,I-E antibody (J). K Representative confocal images showing retinal whole mounts in cGAMP treated-mice. n = 3 retinas

Depletion of microglia inhibits cGAMP-induced iBRB breakdown

Next, we determined the effect of microglia on cGAMP-induced iBRB disruption. The mice were supplemented with PLX5622 (PLX), a commonly used Csf1r antagonist previously shown to induce retinal microglial death with minimal effects on circulating systemic immune cells [46] (Fig. 4A). Retinal microglia were depleted within seven days of PLX treatment (Figure S6). Depletion of microglia suppressed cGAMP-induced STING activation in mouse retina (Fig. 4B, C). Reduced retinal vasodilation, oedema, haemorrhage, and increased pericyte IF signals were observed in PLX-treated mice after cGAMP injection (Fig. 4D–H). EM further showed that the depletion of microglia protects the iBRB ultrastructure against cGAMP treatment, as indicated by normal pericyte morphology, reduced perivascular edema, and intact basement membranes (Fig. 4I). Therefore, cGAMP-induced iBRB breakdown requires retinal microglia.

Fig. 4.

Microglia are essential for cGAMP-induced iBRB breakdown. A Schematic diagram depicting the experimental procedure. B Left: WB analysis of protein levels in mouse retinas. Right: quantification results showing relative STING levels normalized to GAPDH. n = 3 independent experiments. C Representative confocal images of retinal wholemounts showing blood vessels (IB4), microglia (IBA1), and STING. Scale bars: 50 μm. n = 4 retinas per group. D FFA showing retinal blood vessels. n = 5 mice per group. E In vivo OCT images showing retinal structure. Arrows indicate cells infiltrating the vitreous cavity. n = 5 mice per group. F Retinal wholemounts showing hemorrhage spots. n = 5 retinas per group G Representative three-dimensional images of IB4-labeled retinal vessels. n = 3 retinas per group. H Left: confocal images of pericytes (NG2 +), with artery (A) and vein (V) vessels distinguished by vascular diameter. Scale bars: 50 μm. Right: quantification of NG2 coverage. Twenty pericytes per group were randomly quantified, n = 3 retinas per group. I Electron microscopy reveals ultrastructural of iBRB. Perivascular edema (asterisks), basement membrane degeneration (arrowheads), and mitochondria destruction (arrows) were observed in cGAMP treatment. BM, basement membrane; EC, endothelial cell; L: lumen; TJ: tight junction. All results presented as mean ± SD, *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant. F: unpaired t-test; B and H: one-way ANOVA, Tukey's multiple comparisons test

Depletion of microglia ameliorates cGAMP-induced retinal inflammation and neuron degeneration

Cell communication inferred from scRNA-seq showed that microglia were the primary source of galectin, SPP1 and CXCL signalling that may send signal to T cells, macrophage, Müller glia and bipolar cells following cGAMP treatment (Fig. 5A and Figure S7). Bulk RNA-seq analysis revealed that depleting microglia reversed these cGAMP-induced signallings in retina (Fig. 5B). Further, genes upregulated by cGAMP but reversed by PLX treatment were enriched in pattern recognition, antigen binding and MHC protein binding pathways (Fig. 5C). Consistent with transcriptomic changes, microglia depletion suppressed leukocyte adhesion and infiltration and Müller cell gliosis in mice injected with cGAMP (Fig. 5D, E). Upon cGAMP treatment, galectin signalling was a major pathway through which microglia may influence other retinal cells (Fig. 5F). IF analysis showed that cGAMP-induced β-galactose binding protein LGALS3 (galectin-3) signal in mouse retina was repressed by depletion of microglia (Fig. 5G). Notably, administration of PLX restored the expression of visual perception genes in cGAMP-treated retinas (Fig. 5B, Pde6h, Arr3, Rdh5), suggesting that microglia contribute to cGAMP-induced neuron impairment. Indeed, depletion of microglia suppressed cGAMP-induced photoreceptor degeneration, RGC loss and cell death in GCL (Fig. 5H–J). Finally, visual acuity was determined using an optomotor response test (Fig. 5K). cGAMP injection led to lower visual acuity; in contrast, microglial depletion improved the visual acuity (Fig. 5K). Together, these findings indicated that microglial depletion inhibited retinal inflammation, degeneration and visual impairment following cGAMP treatment.

Fig. 5.

Microglia are essential for cGAMP-induced vascular inflammation and neuron degeneration. A Inferred cell communication network from scRNA-seq of CTRL and cGAMP groups, the ligand–receptor pairs are indicated. B, C Bulk RNA-seq analysis of mouse retinas. The treatment was depicted in Fig. 4A. n = 4–5 retinas per group. B Volcano plot showing the fold change of genes (log2 scale) and significance (log10 scale) between PLX + cGAMP and cGAMP groups. C Upper: Overlap of genes up-regulated by cGAMP and down-regulated in PLX + cGAMP-treated retinas. Bottom: Gene ontology (GO) analysis of the overlapping genes. D Confocal images showing retinal leukostasis. The vasculature and adherent leukocytes were labeled by CoA, the infiltrated leukocytes were labeled by anti-CD45 antibody; E IF analysis of cryosections shows Müller cell gliosis (GFAP). F Circle plot illustrating the inferred Galectin signaling sending from microglia in cGAMP group based on scRNA-seq analysis. G–J IF analysis of retinal cryosections. The confocal images show IBA1 + microglia and LGALS3 (G), PDE6H + photoreceptors (H), Brn3a + RGC (I) and TUNEL + dead cells (J). n = 3–5 retinas per group. Scale bars: 50 μm. K Visual acuity measured as the spatial frequency threshold in mice. Left: A snapshot of the recording video; Right: Qualification results, with each dot representing an individual animal. *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant, one-way ANOVA, Tukey's multiple comparisons test

Depletion of macrophage did not inhibit cGAMP-induced iBRB breakdown, retinal inflammation and degeneration

The scRNA-seq analysis showed that macrophages were also increased following cGAMP injection (Fig. 3B). Furthermore, CD68, MHCII, and STING are also highly expressed in macrophages. To assess the role of macrophages, we performed macrophage depletion experiments. Liver macrophages (F4/80 +) were successfully depleted by injecting clodronate liposomes (CLO), a widely used reagent that induces macrophage death [50] (Figure S8A-B). Depletion of macrophages partially reduced cGAMP-induced STING activation, but it was less effective compared to microglia depletion (Figure S8C). However, macrophage depletion did not reverse the cGAMP-induced galectin activation (Figure S8D), Müller cell gliosis (Figure S8E), RGC loss (Figure S8F), or iBRB breakdown (Figure S8G-I). These results indicate that microglia, rather than macrophages, play a dominant role in cGAMP-induced iBRB breakdown, retinal inflammation, and degeneration.

P2RX7 is required for cGAMP-induced STING activation in retinal microglia

Because cGAMP is a charged molecule that cannot passively enter cells, we next determined how cGAMP is transported into retinal microglia. Analysis of a scRNA-seq database showed that among the five known cGAMP transporters, only P2RX7 was mainly expressed in human retinal microglia (Fig. 6A). Intriguingly, bulk RNA-seq revealed that only the P2RX7 transcript was induced by cGAMP injection (Fig. 6B), and scRNA-seq showed that cGAMP increased P2rx7 as well as STING downstream targets Isg15, Irf7 in microglia (Fig. 6C). Next, we assessed the effect of P2RX7 on cGAMP-induced STING activation. In control retinas, P2RX7 was barely detectable in the vascular layers (Fig. 6D). After cGAMP treatment, P2RX7 expression was evident in IB4-positive reactive microglia/macrophages, which also exhibited a strong STING signal (Fig. 6D). The administration of the specific P2RX7 inhibitor A740003 (A74) inhibited the upregulation of STING in mouse retinas (Fig. 6E, F). Additionally, A74 prevented STING expression in microglia and reduced reactive microglia (CD86 +) as well as MHC II-positive microglia following cGAMP treatment (Fig. 6G, H). Finally, knockout of P2RX7 suppressed cGAMP-induced STING activation in retina as well as STING immune signals in microglia (Fig. 6I, K). Reduction of MHC II-positive microglia and protection of retinal neuron RGC were observed in P2RX7 KO mice after cGAMP injection (Fig. 6L, M). Together, our results indicate that P2RX7 is essential for cGMAP-induced STING activation in retinal microglia.

Fig. 6.

Inhibition of P2RX7 suppresses cGAMP-induced STING activation in mouse retinal microglia. A Dot plot of known cGAMP transporter genes in published human retina cells by sc-RNA data analysis. B Bulk RNA-seq analysis showing cGAMP transporters expression in PBS- or cGAMP-injected mouse retinas, n = 5 retinas per group. C Gene expression in microglia cluster in PBS (control) or cGAMP-injected mouse retinas. D IF analysis of mouse retinal flat mounts. n = 3 retinas per group. Scale bars: 20 μm. E Schematic diagram showing experimental procedure. F Left: WB analysis showing protein level. Right: quantification results showing relative STING levels normalized to tubulin. n = 3 independent experiments. G, H IF analysis of retinal flat mounts. Scale bars: 20 μm (G) and 50 μm (H). n = 3 retinas per group. Reactive microglia were labeled by IBA1 and CD86 (G). MHC II proteins were labeled by I-A,I-E antibody (H). I Schematic diagram showing experimental procedure. J Left: WB analysis showing protein level. Right: quantification results showing relative STING levels normalized to tubulin. n = 4 independent experiments. K, M IF analysis of retinal cryosections. Scale bars: 50 μm. n = 3 eyes per group. All results are presented as mean ± SD, *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant, one-way ANOVA, Tukey's multiple comparisons test

Human P2RX7 binds to and imports cGAMP under physiological conditions

P2RX7 was thought to non-selectively import cGAMP in ATP (high concentration, 0.125–0.5 mM)-mediated pore open mechanism [34]. The function of cGAMP transport varies among species and cell types [28, 30], and the role of human P2RX7 as a cGAMP transporter is unknown. Therefore, we determined the effect of P2RX7 on the cGAMP transporter in primary human macrophages. Extracellular cGAMP activated STING and downstream TBK1 in macrophages from all tested healthy donors without the addition of extra ATP, which were repressed by the P2RX7 inhibitor (Fig. 7A, B). Similar results were obtained using the human monocyte cell line, THP-1(Fig. 7C). Another P2RX7 inhibitor, JNJ-55308942 (JNJ), also repressed cGAMP-induced TBK1 phosphorylation (Fig. 7C). SLC19A1 has been identified as a major transporter of cGAMP in human immune cells [16, 28, 29]. The inhibitory effect of A74 and JNJ was comparable to that of methotrexate (MTX), a SLC19A1 inhibitor that suppresses cGAMP-mediated STING in PBMC (Fig. 7D). Next, we determined the effect of P2RX7 on cGAMP-induced STING signalling activation in HEK293T cells lacking endogenous P2RX7 and STING. Co-transfection of P2RX7 and STING resulted in enhanced STING activation following cGAMP treatment, compared to co-transfection of P2RX7 with the control vector (Fig. 7E). These results demonstrated that human P2RX7 imports cGAMP without excessive ATP and suggested a direct interaction between P2RX7 and cGAMP. Indeed, a pull-down assays using streptavidin beads showed that P2RX7 was precipitated by biotin-labelled cGAMP (Fig. 7F). Finally, surface plasmon resonance (SPR) analysis revealed a direct interaction between cGAMP and P2RX7 (Fig. 7G). Compared to ATP, cGAMP exhibited a higher binding affinity to P2RX7, with dissociation constants (KD) of 224 μM for ATP-P2RX7 and 106 μM for cGAMP-P2RX7, respectively (Fig. 7G).

Fig. 7.

P2RX7 binds to cGAMP and activates STING in physiological conditions. A WB analysis of protein levels in primary human macrophages. Cells were treated with cGAMP combined with DMSO or A740003 (100 μM) for 24 h before analysis. Repeated experiment from three donors were shown. B Quantification results of A. P-STING was normalized to total STING. C WB analysis of protein levels in THP-1. Cells were treated with cGAMP (100 μM) combined with DMSO or indicated drugs (100 μM) for 24 h before analysis. Right panel: quantification results of three independent experiments. D PBMC were pre-treated with or without 100 μM of indicated drugs for 30 min before adding cGAMP (5 μM) for 2 h (B–D). Right panel: quantification result of two independent experiments. E WB analysis shows P2RX7 enhances extracellular cGAMP-induced STING activation. HEK293T cells transfected with indicated plasmid were treated with or without cGAMP (100 μM) for 2 h before WB analysis. Right panel: quantification results of three independent experiments. F WB analysis of pulldowns by biotin-cGAMP-streptavidin beads of human P2RX7. HEK293T cells transfected with either FLAG or FLAG-P2RX7 were subject to pull-down assay. Two independent repeats were shown. G SPR analysis showing the binding profiles of P2RX7 to increasing concentrations of cGAMP or ATP. Experiments were repeated two times independently. *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant, one-way ANOVA, Tukey's multiple comparisons test

Inhibition of P2RX7 alleviates STING activation and iBRB breakdown in diabetic mouse

To assess the clinical relevance, we analyzed a scRNA-seq dataset from fibrovascular membranes of patients with PDR [51] and identified seven distinct clusters (Fig. 8A, B). The gene set upregulated in mouse retinal microglia following cGAMP treatment was utilized to score each cell in fibrovascular membranes from patients with PDR and area-under-the-curve (AUC) values were then calculated based on gene expression rankings (Fig. 8C). The highest AUC scores were observed in microglia, with lower scores in stromal and T cells (Fig. 8C). These findings suggest that the cGAMP-induced gene signature in mouse retinal microglia can be recapitulated in the microglia of fibrovascular membranes. Additionally, P2RX7 was predominantly expressed in microglia and macrophages (Fig. 8D). Other cGAMP transports, such as ABCC1, were primarily expressed in B and T cells, LRRC8A in endothelial cells, and SLC46A1 in microglia (Fig. 8D). Next, we sought to evaluated the therapeutic effects of P2RX7 inhibitor in db/db diabetic mice retina (Fig. 8E). Administration of A74 reduced cGAMP level and inhibited STING level in db/db mouse retinas (Fig. 8F, G). Additionally, A74 treatment inhibited STING immune signaling in microglia (Fig. 8H) and reverted microglia to more ramified, resting morphology (Figure S9). The iBRB is also protected by A74 treatment in db/db mice, as indicated by increased NG2 coverage (Fig. 8I). RGC death are prominent features of DR [38]. In db/db mouse retina, administration of A74 promoted RGC survival (Fig. 8J) and reduced TUNEL-positive cells in the GCL (Fig. 8K). Together, these results indicate that cGAMP-induced microglia signature was present in fibrovascular membranes cells from patients with PDR. Inhibition of cGAMP transporter P2RX7 inhibited STING activation, protects iBRB and neuron in diabetic db/db mouse model.

Fig. 8.

Inhibition of P2RX7 reduced STING activation, iBRB disruption and neuron degeneration in db/db diabetic mouse retinas. A Analysis of published scRNA-seq data [51] of fibrovascular membrane samples derived from four patients with PDR. The data were integrated in a single data set for downstream analyses. UMAP plot shows seven different cell clusters. B Clustering strategy of cell populations based on scaled expression heatmap of canonical markers for each cluster. C Area under the curve (AUC) score of individual cells based for the gene set upregulated in mouse retinal microglia following cGAMP treatment. D Dot plot of known cGAMP transporter genes. Note P2RX7 were mainly expressed in microglia and macrophages. E Schematic diagram depicting the experimental procedure. F ELISA analysis shows the cGAMP level in mouse retinas. n = 8 retinas per group. G Left: WB analysis showing protein level in mouse retinas. Right: quantification results showing relative STING levels normalized to Actinin. n = 7–10 retinas per group. H IF analysis of retinal cryosections showing STING and microglia (IBA1 +). n = 4 eyes per group. Scale bars: 50 μm. I IF analysis of retinal whole mounts. About 20 pericytes were randomly selected in each group for area measurement. n = 4 eyes per group. Scale bars: 50 μm. J, K IF analysis of retinal cryosections showing RGC (Brn3a +) (J) and dead retina cells (TUNEL +) (K). n = 4 eyes per group. Scale bars: 50 μm. *P < 0.05, **P < 0.005, ***P < 0.001, ns: not significant, one-way ANOVA, Tukey’s multiple comparisons test

Discussion

The breakdown of iBRB in conditions such as DR, AMD, retinal vein occlusions and other retinal diseases plays a critical role in retinal edema and neuron damage, ultimately leading to vision loss [3]. Change of iBRB permeability can be caused by increased growth factors, cytokines and advanced glycation end products [2, 3]. Our study revealed for the first time that besides those known molecules, cGAMP leads to breakdown of iBRB in STING dependent manner. We further identified P2RX7 as the primary cGAMP transporter that imports cGAMP into retinal microglia, which is essential for cGAMP-induced iBRB disruption and neuron degeneration.

The most prominent phenotype observed after cGAMP administration were retinal vessel dilatation and leakage. During the preparation of this manuscript, Su et al. showed that cGAMP reduced vasoconstriction in the isolated aorta which depended on the activation of PKGI on the endothelium [52]. The retinal vasculature can be investigated in live animals because of its unique localisation in the eye. To our knowledge, this is the first in vivo evidence that cGAMP promotes vascular dilation. Aberrant activation of STING has been observed in patients with vascular and pulmonary syndromes, and STING inhibition alleviates aortic degeneration or cardiotoxicity [53–55]. In the cardiovascular system, STING has been investigated in endothelial and smooth muscle cells [52, 55]. Our results showed that STING was expressed in both retinal vascular endothelial cells and microglia, whereas depletion of microglia prevented cGAMP-induced iBRB breakdown and neuronal degeneration. These results indicate that in the retina, and possibly in the brain given the highly similarity between iBRB and BBB, microglia are the primary cell population that initiates vascular disruption upon STING activation. To date, more than ten clinical trials of Csf1r inhibitors have been successfully completed, and PLX5622 has been investigated in Phase I for the treatment of rheumatoid arthritis (NCT01329991). Our study sheds light on the therapeutic potential of Csf1r inhibitors against STING-induced iBRB impairment.

A limitation of our study is the lack of data on cGAMP levels in the aqueous humor under normal conditions, due to the difficulty of obtaining fresh samples from healthy individuals post-mortem. Future studies using aqueous humor from patients with macular hole may help address this gap. Furthermore, the source of cGAMP in the VH of diabetic mice remains unclear. Recent studies have demonstrated that diabetic stress causes mitochondrial damage and mtDNA leakage in retinal endothelial cells and RGCs [56, 57]. Therefore, we speculate that the extracellular cGAMP detected in the eyes of diabetic mice may be secreted by the injured vascular endothelial cells and RGCs.

We found that microglia were activated and became closely associated with pathogenic vasculature after cGAMP treatment. A close association between microglia and pathogenic retinal neovascularization has been observed in patients with AMD and macular telangiectasia, and in mouse models of vascular inflammation [46, 47]. However, the signals responsible for microglial recruitment remain unclear. We postulate that cGAMP may contribute to the mobilisation of microglia to the retinal vessels under these disease conditions. Our scRNA-seq uncovered that cGAMP promotes a microglial gene expression program associated with both neurodegenerative and angiogenic states. Analysis of cell communication indicates that these cGAMP-reprogrammed microglia may attract T cells by Galectin, CXCL and SPP1 signalling. Microglial depletion experiments supported that reactive microglia facilitate leukocyte infiltration by secreting cell adhesion and antigen presentation molecules, which, in turn, ultimately contributes to cGAMP-induced retinal degeneration and vision impairment.

Increased galectin-3 has been found in reactive microglia in distinct neurodegenerative environments, with its neurotoxic or neuroprotective effects appearing to depends on the subcellular localizaiton, the type of cell involved, and the disease context [58]. A recent report showed that galectin-3 was upregulated in subretinal microglia by Trem2, which is necessary for the phagocytosis of dead photoreceptors and protecting the retina during retinal degeneration [59]. However, other studies showed that inhibition of galectin-3 prevented RGC cell death in glaucoma or delayed retinal damage during light injury [60, 61]. Here, our sc-RNA seq data showed downregulation of Trem2 but upregulation of galectin-3 in microglia, which does not align with the protective Trem2-galectin-3 signaling observed in subretinal microglia population. However, the exact role of galectin-3 induced in microglia following cGAMP treatment remains unclear and will require further investigation through genetic or pharmacological inhibition studies.

We identified P2RX7 as the major cGAMP transporter in mouse retinal microglia. The cGAMP transporter functions in a cell type- and species-dependent manner. For example, SLC19A1 imports cGAMP in the human macrophage THP-1, but not in the epithelial cell lines HEK293 or HeLa, possibly owing to its differential expression [29]. In the mouse retina, P2RX7 was the sole transporter upregulated by cGAMP treatment, suggesting that P2RX7 acts as a major transporter in microglia that takes up cGAMP and amplifies neuroinflammation. Importantly, in human microglial cell lines and primary human macrophages, P2RX7 imports cGAMP and induces STING activation under physiological conditions. P2RX7 directly binds to cGAMP, suggesting selective transport of cGAMP by human P2RX7. As an ATP receptor, P2RX7 has been implicated in the development of glaucoma, AMD, and DR [62, 63]. Here, our results revealed a novel function of P2RX7 as a cGAMP transporter, promoting iBRB destruction and inflammation, which may be related to the pathogenic effects of P2RX7 in these retinal diseases.

Currently, the prevalent treatment for retinal diseases associated with iBRB impairment is anti-VEGF therapy. However, resistance to anti-VEGF and off-target effects which promote photoreceptor degeneration due to the neurotrophic effects of VEGF have been often reported in clinical practice [64, 65]. Emerging evidence suggest that targeting STING may be an effective therapy for treating AMD and DR [21]. During preparation of this manuscript, Liu et al. reported increased cGAS and STING expression in retinas from patients with DR, and strong STING immunostaining in microglia/macrophages surrounding occluded capillary or a hyalinized capillary microaneurysm [66]. Depletion of STING alleviated retinal capillary degeneration in diabetic mice [66], and reduced neovascularization, vascular leakage and myeloid cell activation in ischemia-induced retinopathy [26]. Our study revealed that cGAMP-mediated iBRB degeneration and retinal inflammation share phenotypic characteristics with DR, such as retinal haemorrhage, leakage, and leukocyte adhesion and RGC cell loss, a molecular signature that includes increased expression of inflammatory, angiogenic, and adhesion factors. These results provide the first proof of concept evidence that activation of STING led to vascular inflammation and vision impairment, and demonstrated that targeting cGAMP transportation into microglia may be a new alternative treatment strategy for DR and other retinal diseases associated with iBRB breakdown.

Author contributions

Conceptualization: LLG, XYD, XCT; Discussion: RJ, ZTZ, MD, LW, XBH; Writing: LLG; Methodology: XYG, XFZ, WL, MSL, MZ, HFC, ZQC.

Funding

This study is supported by National Natural Science Foundation of China (Grants 82070969), Guangzhou Municipal University Joint Funding Project (SL2023A03J00488), Guangdong Basic and Applied Basic Research Foundation (2024A1515010518, 2021A1515010689) and the 2024 Graduate Education Innovation Program Project.

Data availability

Bulk RNA-seq data can be accessed from NCBI Sequence Read Archive (SRA) database with the primary accession code PRJNA1102708 and PRJNA1103420. Raw data of scRNA-seq can be accessed with accession number PRJNA1103951.

Declarations

Competing interests

The authors declare no competing interests.