Ythdf2 loss in microglia aggravates ischemic retinopathy by increasing microglia activation and microvascular anomalies

Graphical abstract

Highlights

• •The expression of microglial Ythdf2 is reduced in retinal vasculopathy.
• •Microglia-specific knockout of Ythdf2 triggers microglial activation under both physiological and pathological conditions.
• •Loss of microglial Ythdf2 disrupts physiological angiogenesis and exacerbates pathological angiogenesis.
• •Ythdf2 directly regulates the mRNA stability of Ace and Bmp4 as an m⁶A reader.
• •Targeting the Ythdf2-Ace/Bmp4 network holds therapeutic potential for retinal vasculopathy.

Abstract

Introduction

Microvascular dysfunction is a key contributor to life-threatening diseases, with retinal microvascular diseases being a leading cause of blindness. Aberrant microglia activation is critical in microvasculopathies, but the underlying molecular mechanisms remain unclear. Post-transcriptional modifications, such as N⁶-methyladenosine (m⁶A) modification, are key in these processes, yet their role in retinal vasculopathy is unexplored.

Objectives

We aimed to investigate the role of the m⁶A reader, YTH domain-containing family protein 2 (Ythdf2), in microglia and its involvement in retinal vasculopathy, and to uncover its underlying regulatory mechanisms.

Methods

We assessed the expression of microglial Ythdf2 in the retinas of oxygen-induced retinopathy (OIR) mice using single-cell RNA sequencing (scRNA-seq). To investigate its function, we generated a microglia-specific Ythdf2 knockout mouse model and analyzed retinal microglial and vascular phenotypes under both physiological and pathological conditions. Additionally, RNA-seq and selective inhibitors were employed to explore the underlying signaling pathways.

Results

Down-regulation of Ythdf2 was observed in microglia from mice with microvascular diseases. Microglial Ythdf2 knockout in developing retinas caused aberrant microglial activation, disrupting capillary function, delaying sprouting, and accelerating vascular remodeling, thereby affecting physiological angiogenesis. In OIR retinas, Microglial Ythdf2 knockout intensified microglial activation and aggravated pathological angiogenesis. Mechanistically, Ythdf2 directly regulated the mRNA stability of Ace and Bmp4 as an m⁶A reader. Captopril (an Ace inhibitor) or noggin (a Bmp4 antagonist) alleviated microvascular retinopathy exacerbated by Ythdf2 insufficiency.

Conclusion

Loss of Ythdf2 in retinal microglia increased their activation and caused microvascular anomalies through Ace and Bmp4, providing insights into microvascular development and disease mechanisms. These findings suggest potential therapeutic approaches targeting the Ythdf2-Ace/Bmp4 network for microvascular diseases.

Introduction

The establishment of a mature and well-organized microvascular system is needed for organisms to maintain normal growth and function [1]. Many life-threatening human diseases are related to microvascular dysfunction, including diabetic vascular diseases, central nervous system disorders, and cancers. At the cellular level, microvascular disorders represent a destructive cycle involving vascular cells, neuron cells, and immune cells, indicating the complex and interconnected nature of the disease. The homeostasis of microenvironment, especially the immune-microenvironment, is important for maintaining regular functions of microvasculatures. However, current studies mainly emphasized on microvascular cells, especially endothelial cells, and less is known about the involvement of immune cells. Reportedly, the hyperinflammatory microenvironment, which correlates with vascular integrity disturbance and pathological neovascularization, is detected in the initial stage of microvascular disorders, indicating its unique effects on disease occurrence and progression [2]. However, current clinical applications of the anti-vascular endothelial growth factor (VEGF) therapies only target the vasoproliferative phase. Therefore, elucidating the roles of immune cells in microvascular dysfunction could provide novel insights for microvascular-related disease treatment and tissue engineering, and is urgently needed.

The retina offers an opportunity to noninvasively visualize vascular structure and function, and is thus considered as a unique window to systemic circulation [3]. Retinal microvascular diseases, such as diabetic retinopathy (DR) and retinopathy of prematurity (ROP), are leading causes of blindness worldwide [[4], [5], [6]]. Retinal microglia, the main resident immune cells in the retina, serve as immune watchdogs that constantly survey and regulate the surrounding microenvironment under both physiological and pathological conditions [2]. Located adjacent to retinal vessels, microglia extend branches to directly interact with capillary cells, including retinal microvascular endothelial cells and pericytes, thus modulating their features [7,8]. In the developing retina, microglia are essential for proper microvascular formation and remodeling through regulations of vascular branching, anastomoses, and regression [2]. In retinal microvascular diseases, pathological factors such as hypoxia and hyperglycemia directly activate and recruit microglia, which further exacerbate the proinflammatory microenvironment by secreting inflammatory cytokines [9]. Abnormal microglia function associates with vasoconstriction, pathological neovascularization and neurodegeneration in retinal microvascular diseases, and suppression of microglial activation may serve as a promising therapeutic strategy [7,10]. However, the underlying molecular events have not been fully clarified.

Shifting microglial phenotypes from pro-inflammatory to anti-inflammatory states presents a promising therapeutic strategy for retinal microvascular diseases. Recent studies have highlighted the dynamic role of epigenetic modifications in regulating microglial phenotypes, offering new insights into disease treatment. Epigenetic modifications, such as DNA methylation, histone acetylation, and non-coding RNA regulation, have been shown to play key roles in microglial phenotype alterations [11,12]. Among these, RNA methylation, particularly N⁶-methyladenosine (m⁶A), has gained considerable attention [13,14]. M⁶A modification is one of the most prevalent and reversible RNA modifications in eukaryotic cells. It is primarily catalyzed by the m⁶A methyltransferase complex (writers), removed by m⁶A demethylases (erasers), and recognized by m⁶A-binding proteins (readers) [15]. M⁶A modification regulates mRNA metabolism to govern gene expression and cellular functions [16,17]. Its dysregulation associates with various human diseases, including microvascular diseases [18,19]. The YTH domain-containing family protein 2 (Ythdf2) protein is an m⁶A reader mainly distributed in the cytoplasm of eukaryocytes. Ythdf2 modulates the fate of m⁶A-modified mRNA molecules mainly through specific recognition of and binding to those molecules to mediate their degradation. The crucial role of Ythdf2 in post-transcriptional regulation is essential for maintaining cellular homeostasis, and aberrant regulation of Ythdf2 has been linked to various diseases, including DR [20]. Although a potential role of Ythdf2 in alleviating microglia activation has been identified [21], the specific function and regulatory networks of microglia-expressed Ythdf2 in microvascular development and diseases remain unclear.

Herein, we generated the microglia-specific Ythdf2 knockout mice to study how Ythdf2 affects microglia and microvascular phenotypes under both physiological and pathological conditions. Our data demonstrated that Ythdf2 loss in the developing retina aberrantly activated microglia. Capillary cell functions were disturbed in the hyperinflammatory microenvironment, which affected vasculogenesis through delayed-sprouting and hyper-remodeling. In the oxygen-induced retinopathy (OIR) retinas, Ythdf2 ablation induced microinflammation to injure blood-retinal barrier (BRB) and aggravate pathologic angiogenesis. Mechanistically, Ythdf2 directly regulated the mRNA stability of Ace and Bmp4 as an m⁶A reader under both normoxic and hypoxic conditions. Supplementation of captopril (an inhibitor to Ace) or noggin (an antagonist to Bmp4) ameliorated vascular disorders in the OIR retinas. Our study is critical to better understanding of the molecular mechanisms and cellular events in microvascular development and diseases, and indicates the Ythdf2-Ace/Bmp4 network as potential therapeutic targets for microvascular disorders.

Methods

Mouse breeding and manipulations

C57BL/6J mice (strain ID: 219) were purchased from Charles River Laboratories (Wilmington, MA, USA), and the Rosa26 reporter mice were purchased from Shanghai Model Organisms Center (Shanghai, China). The Ythdf2^fl/fl and Cx3cr1^CreERT2 mouse strains were obtained through academic collaborations following standard material transfer agreements. All mice were bred in specific pathogen-free facilities with a 12 h light/dark cycle at 28.5 °C, and had free access to water and chow. Embryos were produced through natural mating. Our study examined male and female animals, and similar findings were reported for both sexes.

To generate OIR mice, neonatal mice at postnatal day 7 (P7) together with their nursing mother were exposed to a high oxygen environment (75 % O₂) in an oxygen chamber for 5 days from P7 to P11. The mice were returned to room air (21 % O₂) at P12. Food and water were supplied as normal.

Mice were fully anesthetized through intra-peritoneal injection of ketamine (80 mg/kg) and xylazine (4 mg/kg) before all invasive operations and sacrifice. For intra-gastrical injection of tamoxifen into the Cx3cr1^CreERT2Ythdf2^fl/fl mice, tamoxifen (Sigma-Aldrich; St. Louis, MO, USA), dissolved in corn oil at the concentration of 1 mg/ml, was intra-gastrically injected for 3 consecutive days from P1 to P3 (50 µ g per day). For intra-peritoneal injection of tamoxifen into the OIR Cx3cr1^CreERT2Ythdf2^fl/fl mice, tamoxifen was intra-peritoneally injected for 6 consecutive days from P12 to P17 (200 µg per day). For intra-peritoneal injection of captopril, captopril (Sigma-Aldrich) was dissolved in phosphate buffer saline (PBS) solution at the concentration of 1 mg/ml, and was intra-peritoneally injected into the OIR Ythdf2^cKO mice continuously for 6 days P12 to P17 (3 mg/kg/d). For intra-vitreal injection, mice pupils were dilated using 1 % cyclopentolate-HCL and 2.5 % phenylephrine. 1 µL PBS solution containing 1 µg noggin (MedChemExpress; Monmouth Junction, NJ, USA) was delivered into the vitreous chamber by an incision into the sclera (1 mm posterior of the superior limbus) using a syringe with a 33-gauge needle (Hamilton; Bonaduz, Switzerland). For adeno-associated virus (AAV) transduction, a recombinant AAV serotype 2 vector was constructed, which encapsulated the promoter region of the mouse Iba-1 gene and the coding sequence of the mouse Ythdf2 gene fused with a Flag tag (AAV-Ythdf2). A total of 1 µL (1 × 10¹²–1 × 10¹³ genome copies/mL) of either AAV-Ythdf2 or empty AAV (AAV-blank) was delivered into the vitreous chamber of mice via intravitreal injection. The transgene expression was driven by the microglia-specific Iba-1 promoter to ensure targeted delivery. Injections were made through a scleral incision 1 mm posterior to the superior limbus using a 33-gauge Hamilton syringe (Hamilton, Bonaduz, Switzerland).

Flow cytometry

Eyecups from PBS-perfused mice were dissected to isolate the retina from adjacent tissues. OIR or age-matched control mice were euthanized at P17, and retinas were carefully collected. Retinas from 8 to 10 mice per group were pooled and digested in PBS containing 15 U/mL papain (Wortington, LS003126) and 0.1 mg/mL DNase I (Biofroxx, 1121 mg010) at 37 °C for 10 min. The resulting cell suspension was passed through a 70 µm strainer and centrifuged at 300 × g for 10 min at 4 °C. Cell pellets were resuspended and incubated with anti-CD45 antibody at 1:100 dilution for 30 min at 4 °C. After washing, CD45⁺ retinal immune cells were sorted using a Moflo Astrios flow cytometer. For downstream qPCR, approximately 1 × 10⁵ CD45⁺ cells were collected per sample by pooling retinas from 8 to 10 mice. To confirm efficient induction of retinal neovascularization and avascular area in the OIR model, retinas from one mouse per litter were processed for IB4 immunofluorescence prior to qPCR analysis.

Primary microglia isolation, cell culture, and treatment

Mice at P7 were anesthetized and sacrificed for isolation of primary microglia. The mice brains were dissected from the skull with meninges and vasculatures carefully removed. The brain tissue was then cut into small pieces, enzymatically digested using Trypsin and DNase I, and passed through a 70 μm cell strainer to obtain single-cell suspensions. The cell suspension was then centrifuged to pellet the cells, and the pellet was resuspended in DMEM/High Glucose medium supplemented with 10 % fetal bovine serum (FBS; Invitrogen, Carlsbad, CA, USA), penicillin (100 U/mL; Invitrogen) and streptomycin (100 U/mL; Invitrogen). Complete medium was used short for supplemented culture medium in the following text. Cells were planted onto poly-D-lysine-coated T75 flasks (Thermo Fisher Scientific; Waltham, MA, USA) and were maintained at 37 °C with 21 % O₂ and 5 % CO₂ for adherence and grow. After 10 to 14 days of incubation, to collect microglia, the flasks were shaken at 65 revolutions per minute (rpm) for five hours with the floating cells harvested, which were recognized as purified microglia through immunofluorescence staining of Iba-1.

Primary microglia and murine microglia BV-2 cells (purchased from the Cell Bank of the Chinese Academy of Science) were cultured in complete medium at 37 °C with 21 % O₂ and 5 % CO₂. For hypoxia treatment, cells were exposed in 1 % O2 for 72 h before collection. In the actinomycin D assay, cells were maintained in complete medium supplemented with actinomycin D (5 µg/mL), and were harvested at 0, 4 and 8 h post treatment, respectively.

RNA isolation and qPCR

Total RNA was isolated from cell lysates using TRIzol reagent (Invitrogen). The concentration and purity of RNA were measured with a Nano-Drop ND-1000 spectrophotometer (Nano-Drop Technologies, Wilmington, DE, USA). cDNA was synthesized using a PrimeScript RT Kit (Takara, Otsu, Shiga, Japan). RNA expression was analyzed by qPCR using FastStart Universal SYBR Green Master (ROX; Roche, Basel, Switzerland) on a StepOne Plus Real-Time PCR System (Applied Biosystems, Darmstadt, Germany). mRNA levels of β-actin were assessed in parallel for normalization. Primer information was detailed in Supplementary Table S1.

Immunoblotting

Collected cells were initially fragmented in lysis buffer (Beyotime, Shanghai, China) supplemented with protease inhibitor cocktail (Roche, Basel, Switzerland) for protein isolation. Extracted proteins were then separated by size using sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE), and transferred to a polyvinylidene fluoride (PVDF) membrane (Millipore; Boston, MA, USA) through a wet transfer system. Membranes were further blocked by bovine serum albumin (BSA), and sequentially incubated with corresponding primary antibodies (detailed in Supplementary Table S2) and HRP-conjugated secondary antibodies (dilution: 1:10000; ICL). Blots were developed using the Tanon-5200Multi Chemiluminescent Imaging System (Tanon Science & Technology, Shanghai, China), with immunoblot images analyzed by the Image J software (https://rsb.info.nih.gov/ij/index.html).

scRNA-seq and analysis

The publicly available OIR mouse single-cell transcriptomic dataset (GSE152928) was re-analyzed for this study. Unique molecular identifier (UMI) counts of age-matched control and OIR scRNA-seq replicates were merged into one single digital gene expression (DGE) matrix and processed with R package Seurat (version 4.3.0). Cells with fewer than 100 detected genes or exceeding 20 % mitochondrial gene expression were excluded. The DGE matrix was normalized via the NormalizeData function with a scale factor of 10,000. Batch effect correction was performed using the R package Harmony (version 1.2.0). Principal component analysis (PCA) on the most variable genes in the DGE matrix identified 20 significant principal components (PCs). Dimensionality reduction and clustering were conducted using the FindNeighbors and FindClusters functions with a resolution parameter set at 0.8. The integrated dataset was visualized through uniform manifold approximation and projection (UMAP), a non-linear dimensional reduction technique. Differential gene expression analysis was performed with the FindMarkers function employing the Wilcoxon rank-sum test. Genes with |Log₂FoldChange| > 0.25 and P < 0.05 were considered statistically significant. R package Monocle2 (version 2.28.0) was used to predict a pseudotime trajectory of microglia activation. Microglia were ordered in pseudotime with DDRTree and orderCells functions.

RNA-Seq

Purity, concentration and integrity of the extracted total RNA were assessed using a NanoPhotometer spectrophotometer (Implen, Munich, Germany), a Qubit 2.0 Fluorometer (Thermo Fisher Scientific), and the Agilent 2100 Bioanalyzer system (Agilent BioTek; Winooski, VT, USA). High-quality RNA was sent for library construction, including poly(A) selection, RNA fragmentation and cDNA synthesis, using a NEBNext Ultra^TM RNA Library Prep Kit (New England Biolabs; Ipswich, MA, USA) according to the manufacturer’s instructions. Library fragments were purified with the AMPure XP Beads (Beckman Coulter; Brea, CA, USA), and library quality was assessed on the Agilent 2100 Bioanalyzer system (Agilent BioTek). Clustering was conducted on a cBot Cluster Generation System using TruSeq PE Cluster Kit v3-cBot-HS (Illumina; San Diego, CA, USA). The RNA-Seq library was further sequenced on an Illumina Novaseq 6000 platform (Illumina) to generate 150 base pair (bp) reads. Quality control and raw sequencing data filtration were achieved using the FastQC and Trimmomatic software. The STAR software was then applied to align clean reads to the mouse genome GRCm38/mm10. The HTSeq v0.6.0 was used to calculate the fragments per kilobase per million mapped reads (FPKM) of each gene, thus estimating gene expression levels. We next used the DESeq2 algorithm to identify differentially expressed genes with Log₂FC > 1 or < -1 and P < 0.05.

Correlation analyses

For the correlation analysis, Pearson correlation coefficients among the genes of interest were computed using the R package stats (version 4.3.1), and the results were visualized utilizing the Chiplot online platform (https://www.chiplot.online/).

Immunofluorescence staining

For immunofluorescence staining, collected cells were gradually fixed in 4 % PFA and permeabilized with 0.5 % Triton X-100 (Sigma-Aldrich), both at room temperature for 30 min. Cryosections and retinal flat mounts were permeabilized at room temperature for 2 h. Cells, cryosections and retinal flat mounts were then blocked in 1 % BSA at room temperature for 1 h. Samples were further sequentially incubated with primary antibodies (detailed in Supplementary Table S2) at 4 °C overnight, and corresponding fluorescence-conjugated secondary antibodies (dilution: 1:500 for tissue staining; 1:1000 for cell staining; Invitrogen) at room temperature for 2 h. Cell nuclei were counterstained with DAPI (Sigma-Aldrich). Samples were visualized and images were taken using a Stellaris STED confocal microscope (Leica) and an inverted microscope (DMi8, Leica) equipped with the THUNDER imaging system (Leica). Analyses and quantification on the fluorescent images were achieved using the Image J software (https://rsb.info.nih.gov/ij/index.html).

Mounting of mice retinas

After fully anesthetization and sacrifice, mice eyes were enucleated with the connective tissues trimmed and the anterior segments and vitreous removed. To make retinal cryosections, the remaining posterior eyecups were fixed in 4 % PFA at 4 °C overnight. The fixed eyecups were then gradually dehydrated in sucrose solution, embedded with optimal cutting temperature compound, frozen by liquid nitrogen, and sectioned at 8 μm using a Leica CM1900 cryostat (Leica; Wetzlar, Germany). To make retinal flat mounts, the posterior eyecups were fixed in 4 % PFA for 4 h. Neural retinas were then carefully separated from the fixed posterior eyecup and trimmed into a four-leaf clover-shaped retinal flat mount.

MeRIP-qPCR

MeRIP was conducted using the riboMeRIP m⁶A Transcriptome Profiling Kit (RiboBio, Guangzhou, China) according to the manufacturer’s protocol. Briefly, isolated total RNA was fragmented using the RNA fragmentation buffer with 10 % of fragmented RNA reserved as input control. To immunoprecipitate the methylated RNA fragments, the remaining RNA was then incubated with protein A/G magnetic beads coated with IgG/m⁶A antibody. To pull down the m⁶A-modified RNA fragments bound to the antibody, a magnetic separator was applied to collect the magnetic beads bound complexes. RNA extracted from the complex was then purified and sent for cDNA synthesis. qPCR was used to determine the enrichment of m⁶A-immunopurified Ace and Bmp4 mRNA with normalization to the input control.

RIP-qPCR

RIP experiment was performed using the Magna RIP Kit (Millipore) in accordance with the manufacturer’s instructions. Briefly, collected BV-2 cells were treated with the RIP lysis buffer to release the RNA-protein complex. 10 % of the RIP lysate were isolated as input control. For immunoprecipitation, protein A/G magnetic beads were sequentially incubated with IgG/Ythdf2 antibody and the cell lysates. The mixture was then placed on the magnetic separator to immobilize magnetic beads bound complexes with magnet, thus pulling down the targeted RNA-protein complex. Collected and purified RNA was further sent for cDNA synthesis and qPCR assessments with normalization to the input control.

Quantification and statistical analysis

Statistical analyses were conducted using GraphPad Prism (v 4.0; GraphPad Software, San Diego, CA). Two-tailed Student's T test was utilized for comparing between two groups, and one-way analysis of variance (ANOVA) coupled with the Bonferroni’s post hoc test was used for comparisons among four groups. All data were presented as mean ± standard error of the mean (SEM). Statistical significance was designated as P < 0.05. Detailed information regarding the replication of each experiment is provided in the figure legends. Quantifications were performed blinded by the investigators.

Results

YTHDF2 is down-regulated in diseased microglia

Herein, to determine whether aberrant m⁶A modification in microglia contributes to microvascular diseases, we detected expression of m⁶A regulators in retinal microglia of the OIR mice. Expression of m⁶A writers (Mettl3, Wtap, Mettl14, Mettl16, Zc3h13 and Rbm15), erasers (Alkbh5 and Fto), and readers (Ythdc1, Ythdc2, Igf2bp2, Igf2bp3, Ythdf1, Ythdf2 and Ythdf3) were analyzed. The expression levels of Igf2bp3 and Ythdf2 were found down-regulated in retinal microglia isolated from the OIR mice at postnatal day 17 (P17), the end phase of neovascularization and the beginning stage of neovascular regression (GEO accession number: GSE132731; Fig. 1A) [22]. We next annotated Igf2bp3 and Ythdf2 expression in microglia isolated from OIR retinas (Fig. 1B), qPCR identified Ythdf2 as the solely changed m⁶A regulator in microglia isolated from OIR retinas (Fig. 1C). To further validate this finding, primary mouse microglia were cultured and exposed to hypoxia for 72 h (Fig. 1D). The phenotype and purity of primary microglial cells were verified using Iba-1 staining (Fig. 1E). qPCR identified Ythdf2 as the solely changed m⁶A regulator in primary microglia subjected to hypoxia treatment (Fig. 1F). Immunoblotting also detected decreased Ythdf2 expression in hypoxic microglia (Fig. 1G).

Fig. 1.

YTHDF2 is down-regulated in diseased microglia. (A) Heatmap presents the expression of m⁶A regulators in retinal microglia isolated from Ctrl or OIR mice (GEO accession number: GSE132731). (B) Schematic illustration of microglia isolation from OIR and Ctrl retinas using flow cytometry sorting for CD45⁺ cells. (C) qPCR detects mRNA levels of m⁶A regulators Igf2bp3 and Ythdf2 in microglia isolated from OIR and Ctrl retinas. n = 3 for each group. (D) Schematic illustration of primary microglia isolation and treatment. (E) Immunostaining confirms positive Iba-1 expression in isolated microglia. Cell nuclei are counterstained with DAPI. Scale bar: 30 µm. (F) qPCR detects mRNA levels of m⁶A regulators Igf2bp3 and Ythdf2 in microglia treated with normoxia or hypoxia. n = 3 for each group. (G) Immunoblotting of Ythdf2 in microglia treated with normoxia or hypoxia. β-actin is used as an internal control. Representative images along with quantification of Ythdf2 protein expression are shown. n = 3 for each group. (H) Feature plot demonstrates Ythdf2 expression in different types of retinal cells in the OIR and Ctrl mice (GEO accession number: GSE152928). (I) Violin plots compare Ythdf2 expression in distinct retinal cells between the OIR and Ctrl mice (GEO accession number: GSE152928). (J) Single cell trajectory analysis indicates decreased Ythdf2 expression in microglia along the pseudotemporal trajectory based on the OIR development. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); *P < 0.05; *** P < 0.001. Two-tailed Student's T test for A, C, F, G and I.

Expression of Ythdf2 was further analyzed and compared in different types of retinal cells using single cell RNA transcriptome sequencing (RNA-Seq) data obtained from P17 OIR mice retinas (GEO accession number: GSE152928). Single cell RNA-Seq data revealed ubiquitous expression of Ythdf2 in all types of retinal cells, while differential expression of Ythdf2 between OIR and control mice was solely detected in retinal microglia (Fig. 1H,I). Single cell trajectory analysis demonstrated decreased Ythdf2 expression in microglia along the pseudotemporal trajectory based on the OIR development (Fig. 1J), further implying the involvement of Ythdf2 in OIR. Collectively, we detected down-regulated YTHDF2 expression in microglia of retinal microvascular diseases.

Microglia-specific knockout of Ythdf2 induces microglia activation in developing retina

We first explored the role of YTHDF2 in regulating microglial phenotypes in the developing retina. YTHDF2 shares a high degree of similarity in protein sequences between human and mice, suggesting that it is highly evolutionary conserved between the two species (Supplementary Fig. S1). To conditionally knock out the Ythdf2 gene in mice macrophage/microglia, we bred the Ythdf2^fl/fl mice [23] with the Cx3cr1^CreERT2 mice, a strain expressing CreERT2 recombinase driven by the Cx3cr1 promoter [24], and obtained the Cx3cr1^CreERT2Ythdf2^fl/fl mice (Supplementary Fig. S2A). Tamoxifen was intra-gastrically injected at P1, P2 and P3 of the Cx3cr1^CreERT2Ythdf2^fl/fl mice to generate the Ythdf2^cKO mice with macrophage/microglia-specific deletion of Ythdf2 (Fig. 2A). The high efficiency of Cre-mediated recombination in retinal microglia was further validated using a Rosa26 reporter mouse strain (R26-CAG-LSL-Luc-2A-EGFP mice; Supplementary Figs. S2B,C). Immunoblotting confirmed the remarkable reductions of Ythdf2 protein in microglia isolated from the Ythdf2^cKO mice at P7 (Supplementary Figs. S2D). Immunofluorescence staining further confirmed that Ythdf2 deletion in the retina was confined to microglia (Supplementary Figs. S2E).

Fig. 2.

Microglia-specific knockout of Ythdf2 induces microglia activation in developing retina. (A) Schematic illustration of tamoxifen induction. Tamoxifen is intra-gastrically injected into the Cx3cr1^CreERT2Ythdf2^fl/fl mice at P1, P2 and P3 to suppress Ythdf2 expression in microglia. (B) Volcano diagram of RNA-Seq data from Ythdf2^cKO microglia compared to Ythdf2^fl/fl microglia (Log₂FC > 1 or < -1; P < 0.05) is shown. n = 2 for Ythdf2^cKO microglia and n = 3 for Ythdf2^fl/fl microglia. (C) GO plot presents pathways enriched in Ythdf2^cKO microglia. (D) qPCR compares mRNA levels of IL6, Tnf-α, and IL-1β in Ythdf2^fl/fl and Ythdf2^cKO microglia. n = 3 for each group. (E) qPCR compares mRNA levels of Tmem119, P2ry12, Cstb, Fabp5, Trem2, Cd206 and Cd163 in Ythdf2^fl/fl and Ythdf2^cKO microglia. n = 3 for each group. (F) Immunofluorescence staining of Iba-1 in Ythdf2^fl/fl and Ythdf2^cKO microglia. Cell nuclei are counterstained with DAPI. Representative images along with quantification of the microglia area are shown. n = 7 per group. Scale bar: 30 µm. (G) The schematic illustration depicts the morphologic parameters of retinal microglia. (H-J) Immunofluorescence staining of Iba-1 in retinal flat mounts of Ythdf2^fl/fl and Ythdf2^cKO mice at P7 (H), P14 (I) and P30 (J). Representative images along with quantification results are shown. n = 7 per group (H). n = 9 per group (I). n = 8 per group (J). Scale bar: 100 µm (low-power-field), 20 µm (high-power-field). Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); * P < 0.05; ** P < 0.01; *** P < 0.001. Two-tailed Student's T test for D-F. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for H-J.

We used RNA-Seq to annotate aberrantly changed biological processes and signaling pathways induced by Ythdf2 loss in microglia isolated from the Ythdf2^cKO and the Ythdf2^fl/fl mice at P7. A total of 765 differentially expressed genes [Log₂fold change (FC) > 1 or < -1; p < 0.05], consisting of 538 up-regulated and 227 down-regulated genes, were identified in microglia isolated from the Ythdf2^cKO mice compared to the Ythdf2^fl/fl mice (Fig. 2B). Gene Ontology (GO) analyses revealed that biological processes associated with activation of macrophage/microglia were enriched upon Ythdf2 ablation (Fig. 2C). qPCR assay further revealed that Ythdf2^cKO microglia exhibited elevated expression of inflammatory cytokines (IL-6, Tnf-α and IL-1β) (Fig. 2D). In addition, the expression of resting microglia markers (Tmem119 and P2ry12) was reduced, whereas activated microglia markers (Cstb, Fabp5, Trem2, Cd206 and Cd163) were upregulated in the Ythdf2^cKO microglia (Fig. 2E).

Morphological changes of microglia were further visualized in cultured primary cells and retinal flat mounts obtained from the Ythdf2^cKO and the Ythdf2^fl/fl mice. Microglia isolated from the Ythdf2^fl/fl mice at P7 were uniform in size, resembling the resting phenotypes. However, the Ythdf2^cKO microglia expanded like a fried egg, demonstrating the activated presentations (Fig. 2F). Our data supported the regulatory effects of Ythdf2 on microglia manifestations. In agreement with the in vitro data, immunofluorescence staining revealed activation of microglia, as evidenced by morphological abnormalities (Fig. 2G) in retinal flat mounts collected from the Ythdf2^cKO mice at both P7 and P14 (Fig. 2H,I). However, such morphological change was not detected at P30 (Fig. 2J). Detailed morphological assessments identified that Ythdf2 loss in microglia led to reduction in the territory projection area and the number of interactions at P7 and P14, but not at P30 (Fig. 2H–J). An enlarged soma area was also detected at P14 (Fig. 2I). Collectively, our data indicated that loss of Ythdf2 leads to microglia activation in the developing neural retina, but such activation was vanished at P30.

Loss of Ythdf2 impairs physiological angiogenesis in developing retina

Retinal microglia play unique roles in vascular development [8,25]. We thus analyzed the effects of microglia-specific Ythdf2 knockout on retinal vasculogenesis. Three layers of vascular plexuses, termed as the superficial (P0-P7), deep (P8-P10) and intermediate (P11-P15) vascular plexuses, develop serially to form the 3D retinal vasculatures (Fig. 3A) [26,27]. Ythdf2′s function on growth of the superficial vascular layer was initially annotated using P7, P10, P14 and P30 mice. Decreased retinal vessel density within capillary networks, characterized by reduced percentage of vessel area and diameters of vessels, was noticed in retinas of the P7, P10 and P14 Ythdf2^cKO mice compared to the Ythdf2^fl/fl mice (Fig. 3B,C). Retinal flat mounts of the P7, P10 and P14 Ythdf2^cKO mice also demonstrated decreased vascular branching points and increased average vessel length, leading to reduced complexity in the remodeling zone (Fig. 3B,C). However, no obvious impairments in retinal vessel density and complexity were found in the Ythdf2^cKO mice compared to the Ythdf2^fl/fl mice at P30 (Fig. 3B,C).

Fig. 3.

Loss of Ythdf2 impairs physiological angiogenesis in developing retina. (A) Schematic illustration demonstrates the development of superficial, deep and intermediate vascular plexuses in mice retina. (B) Immunofluorescence staining of IB4 in superficial vascular plexuses of retinal flat mounts collected from WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7, P10, P14 and P30. White dots indicate branching points, and white parallel lines suggest vessel diameters. Scale bar: 30 µm. (C) Schematic illustration and quantification results of vessel area, vessel diameter, number of branching points, and average vessel length. n = 7 for each group at P7; n = 10 for each group at P10 and P30; n = 8 for each group at P14. (D-E) Immunofluorescence staining of IB4 in deep vascular plexuses of WT, Ythdf2^fl/fl and Ythdf2^cKO retinal flat mounts at P10, P14 and P30. Representative images (D) along with quantification results (E) are shown. n = 8 for each group at P10; n = 10 for each group at P14 and P30.Scale bar: 30 µm. (F-G) Immunofluorescence staining of IB4 in intermediate vascular plexuses of WT, Ythdf2^fl/fl and Ythdf2^cKO retinal flat mounts at P14 and P30. Representative images (F) along with quantification results (G) are shown. n = 8 for each group at P14; n = 10 for each group at P30. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); * P < 0.05; ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for C, E and G.

We next analyzed whether the formation of the deep and intermediate retinal vessels was affected. Consistent with findings in the superficial vascular plexuses, decreased vessel density represented by reduced vessel area and vessel diameters, as well as lowered vessel complexity shown by decreased vascular branching points and increased average vessel length, were detected in the deep layer of the retinas from the Ythdf2^cKO mice compared to the Ythdf2^fl/fl mice at P10 and P14 (Fig. 3D,E). As to the intermediate retinal vessels, decreased vessel area, vessel diameters, vascular branching points and average vessel length were found in retinas from the Ythdf2^cKO mice at P14 (Fig. 3F,G). However, characteristics of deep and intermediate retinal vascular networks at P30 were similar between the two groups (Fig. 3D–G). Thus, our data implied that microglia-specific Ythdf2 knock out affected vasculogenesis in the developing retina by impairing the density and complexity of all three layers of vascular plexuses.

Loss of Ythdf2 disrupts vascular sprouting and remodeling during physiological angiogenesis

Microglia are critical for vascular sprouting and pruning, both of which enable the formation of stable blood vessels [2]. We next aimed to tell which step accounted for the decreased vessel density in the Ythdf2^cKO mice retina. GO analyses based on the RNA-Seq data suggested that both vascular sprouting and remodeling related biological processes were enriched in the Ythdf2^cKO microglia (Fig. 4A). We initially examined whether angiogenic sprouting, which started radially from the center at P1 and reached the periphery around P7 [28], was affected. Retinal flat mounts collected from the Ythdf2^cKO mice at P7 showed reduced vascular extension compared to the Ythdf2^fl/fl mice (Fig. 4B), indicating impairments in radial vessel outgrowth upon microglia-specific Ythdf2 deletion. Furthermore, inhibited differentiation and sprouting of endothelial tip cells, characterized by a reduced number of tip cells and fewer, shorter filopodia, was observed in P7 Ythdf2^cKO mice (Fig. 4C). IB4 staining of retinal flat mounts detected no difference in vascular extension between Ythdf2^cKO and Ythdf2^fl/fl mice at P10 (Fig. 4D) and P14 (Fig. 4E).

Fig. 4.

Loss of Ythdf2 disrupts vascular sprouting and remodeling during physiological angiogenesis. (A) GO plot presents pathways enriched in Ythdf2^cKO microglia based on the RNA-Seq data. (B) Immunofluorescence staining of IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7. Representative images, schematic diagram along with quantification of vascular extension are shown. n = 9 for each group. White circle indicates the border of the neural retina; white dot line represents border line of the retinal vasculature. Scale bar: 1 mm (left); 500 µm (right). (C) High-resolution images of IB4-stained retinas shows tip cells at the angiogenic front of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7. Representative images, schematic diagram along with quantification results are shown. n = 7 for each group. Scale bar: 30 µm. (D-E) Immunofluorescence staining of IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P10 (D) and P14 (E). Representative images, schematic diagram along with quantification of vascular extension are shown. n = 6 for each group (D); n = 7 for each group (E). White circle line represents border line of the nerual retina; white dot line represents border line of the retinal vasculature. Scale bar: 1 mm (left); 500 µm (right). (F-G) Immunofluorescence staining of Collagen IV and IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7 (F) and P30 (G). Representative images along with quantification results are shown. n = 7 for each group. Scale bar: 30 µm. (H) Schematic diagram of postnatal retinal angiogenesis and pruning in Ythdf2^fl/fl and Ythdf2^cKO mice. Loss of microglia-specific Ythdf2 disturbs angiogenic sprouting and accelerates vessel remodeling at the early developmental stage. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for B-G.

We further analyzed whether Ythdf2 in microglia also regulated vascular remodeling. Vascular remodeling is typified by empty extracellular matrix (ECM) sleeves in the capillary beds, which contain ECM proteins secreted from vascular endothelial and mural cells to stabilize nascent vessels, but lack vascular cells [29]. We thus stained retinas collected from the Ythdf2^fl/fl mice and the Ythdf2^cKO mice with the vascular marker IB4 and the ECM marker Collagen IV. Collagen IV⁺ and IB4^- areas were recognized as empty ECM sleeves. Vessel pruning starts at P5, when sprouting angiogenesis is still ongoing, and continues until P18 [30]. We found that the Ythdf2^cKO mice had more empty ECM sleeves as compared to the Ythdf2^fl/fl controls at P7 (Fig. 4F), but not at P30 (Fig. 4G). Collectively, our data suggested that loss of microglia-specific Ythdf2 disturbed angiogenic sprouting and accelerated vessel remodeling at the early developmental stages (Fig. 4H), resulting in a hypovascularized developing plexus.

Loss of Ythdf2 impairs capillary morphogenesis during physiological angiogenesis

We next verified the cellular events accounted for the suppressed vasculogenesis in the Ythdf2^cKO mice. Endothelial cells and pericytes are major cellular components of capillaries that play crucial roles throughout the entire process of vasculogenesis [1]. We thus clarified whether Ythdf2 affected vasculogenesis by regulating development of endothelial cells and pericytes. Integrity of the endothelial adherens junctions, which was essential for maintaining the blood-retinal barrier (BRB) stability, was analyzed in P7 and P30 mice. Retinal flat mounts were stained with vascular-endothelial-specific cadherin (VE-cadherin), a component of the endothelial adherens junctions. Discontinuous distribution of VE-cadherin and reduced VE-cadherin coverage to IB4 were detected in retinal flat mounts collected from the Ythdf2^cKO mice at P7, but not in the Ythdf2^fl/fl mice (Fig. 5A). Intact VE-cadherin distribution and coverage was presented in both Ythdf2^fl/fl and Ythdf2^cKO mice at P30 (Fig. 5B). We next stained mice retinas with the pericyte marker platelet derived growth factor receptor β (PDGFRβ) to visualize pericytes. Reduced pericyte coverage was noticed in retinal vessels of the Ythdf2^cKO mice compared to the Ythdf2^fl/fl mice at P7 (Fig. 5C), while such reduction was not detected at P30 (Fig. 5D).

Fig. 5.

Loss of Ythdf2 impairs capillary morphogenesis during physiological angiogenesis. (A-B) Immunofluorescence staining of VE-cadherin and IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7 (A) and P30 (B). Representative images along with quantification of VE-cadherin coverage are shown. n = 7 for each group. Scale bar: 30 µm. (C-D) Immunofluorescence staining of PDGFRβ and IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7 (C) and P30 (D). Representative images along with quantification of PDGFRβ coverage are shown. n = 7 for each group. Scale bar: 30 µm. (E-F) Immunofluorescence staining of TER-119 and IB4 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO mice at P7 (E) and P30 (F). Representative images along with quantification of TER-119⁺ cell leakage are shown. n = 7 for each group. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for A-F.

We further measured whether the suppressed capillary morphogenesis disturbed BRB construction to cause vascular leakage. We stained the retina with IB4 and TER-119, a marker of erythroid cells, to observe the positional relationship between erythroid cells and retinal vasculatures. Leakage of TER-119⁺ cells outside the retinal vessels was found in retinal flat mounts collected from the Ythdf2^cKO mice at P7, while distribution of erythroid cells along within the retinal vessels was noticed in the Ythdf2^fl/fl mice (Fig. 5E). No retinal vascular leakage was detected in the Ythdf2^fl/fl or the Ythdf2^cKO mice at P30 (Fig. 5F). Thus, our data implied that loss of Ythdf2 in microglia affected capillary morphogenesis to interrupt vasculogenesis.

Loss of Ythdf2 exacerbates microglia activation in OIR retinas

Since the loss of Ythdf2 activated microglia in the developing retina, we next analyzed whether it triggered aberrant microglia activation under pathogenic conditions. UMAP plots of the single cell RNA-Seq data from the OIR mice retinas identified three sub-clusters of microglia showing decreasing Ythdf2 expression (Fig. 6A). We termed these clusters as Ythdf2^high, Ythdf2^med, and Ythdf2^low microglia respectively (Fig. 6A). Gene Set Enrichment Analysis (GESA) plot revealed that the biological process of macrophage chemotaxis was enriched in the Ythdf2^low microglia compared to the Ythdf2^high microglia (Fig. 6B). RNA expression of markers for resting microglia (Tmem119 and P2ry12) was specifically up-regulated in the Ythdf2^high microglia, and enrichment of genes for activated microglia (Cstb and Fabp5) was detected in the Ythdf2^low microglia (Fig. 6C). RNA levels of above genes in the Ythdf2^med cluster were intermediate between the Ythdf2^high and Ythdf2^low microglia (Fig. 6C). Single cell trajectory analysis further revealed decreased expression of Tmem119 and P2ry12, as well as increased expression of Cstb and Fabp5 in microglia along the pseudotemporal trajectory (Fig. 6D). Consistently, RNA-Seq data confirmed that Ythdf2 expression was positively correlated with Tmem119 and P2ry12 expression, but was negatively associated with RNA levels of Cstb and Fabp5 in retinal microglia isolated from the OIR mice (GEO accession number: GSE132731; Fig. 6E). Herein, we also detected down-regulated Tmem119 and P2ry12 levels and up-regulated Cstb, Fabp5, Trem2, Cd206 and Cd163 expression in the hypoxic Ythdf2^cKO microglia compared to the hypoxic Ythdf2^fl/fl microglia (Fig. 6F). qPCR analysis further revealed that hypoxic Ythdf2^cKO microglia exhibited elevated expression of inflammatory cytokines (IL-6, TNF-α, and IL-1β) (Fig. 6G).

Fig. 6.

Loss of Ythdf2 exacerbates microglia activation in OIR retinas. (A) A UMAP plot shows 3 subpopulations in retinal microglia isolated from the OIR mice based on Ythdf2 expression (Ythdf2^high, Ythdf2^med, Ythdf2^low; GEO accession number: GSE152928). (B) GSEA plot demonstrates that macrophage chemotaxis pathway is enriched in the Ythdf2^low microglia. (C) Dot plot compares the expression of Ythdf2, Tmem119, P2ry12, Cstb and Fabp5 in the 3 microglia subpopulations. (D) Expression patterns of indicated genes along the pseudotemporal trajectory. (E) Pearson correlation coefficients of indicated genes in retinal microglia isolated from the OIR mice (GEO accession number: GSE132731). (F) qPCR compares mRNA levels of Tmem119, P2ry12, Cstb, Fabp5, Trem2, Cd206 and Cd163 in hypoxic Ythdf2^fl/fl and Ythdf2^cKO microglia. n = 3 for each group. (G) qPCR compares mRNA levels of IL6, Tnf-α, and IL-1β in hypoxic Ythdf2^fl/fl and Ythdf2^cKO microglia. n = 3 for each group. (H) Schematic illustration of the mouse OIR model, tamoxifen induction and microglia phenotypes. (I-K) Immunofluorescence staining of Iba-1 in retinal flat mounts of WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice at P17 (I), P21 (J) and P25 (K). Representative images along with quantification of microglia percentage are shown. n = 7 for each group. Scale bar: 30 µm. (L) Schematic diagram demonstrating the microglia phenotypes in retinas of Ythdf2^cKO and Ythdf2^fl/fl OIR mice. Loss of microglia-specific Ythdf2 triggers aberrant microglia activation in the OIR retinas. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. * P < 0.05; ** P < 0.01; *** P < 0.001. Two-tailed Student's T test for F-G. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for I-K.

For morphological assessments, the OIR mice model was generated as described in Fig. 6H, with microglia features annotated at P17 (end of neovascularization and beginning of neovasculature regression), P21 (in the middle of neovasculature regression), and P25 (end of neovasculature regression). Immunofluorescence staining confirmed the successful establishment of the OIR model (Supplementary Fig. S3). Microglia were classified into activated (ameboid) and resting (dendritic) status based on their morphology (Fig. 6H). Accumulation of ameboid microglia around the retinal neovasculatures was detected in the Ythdf2^cKO OIR mice compared to the Ythdf2^fl/fl OIR mice at all three time points (Fig. 6I-K). Collectively, Ythdf2 depletion intensifies extensive activation of microglia in OIR retinas, represented by accumulation of ameboid microglia (Fig. 6L).

Loss of Ythdf2 aggravates neovascularization and BRB breakdown in OIR retinas

We then analyzed whether microglia-specific Ythdf2 loss associates with abnormal vascular features in OIR retinas. Consistent with above assessments on microglia phenotypes, retinal vascular structures were investigated at P17, P21 and P25 respectively. More extensive areas of neovascular tufts (NVTs), formed in the superficial vascular plexuses, were detected in the Ythdf2^cKO OIR mice compared to the Ythdf2^fl/fl OIR mice at P17 (Fig. 7A,B), suggesting that loss of microglia-specific Ythdf2 contributes to pathogenic retinal vascularization. Consistently, more aggressive NVTs were also found at P21 and P25 Ythdf2^cKO OIR mice (Fig. 7A,B), implying that the regression of NVTs was inhibited upon Ythdf2 loss. No difference in the avascular area in the central retina was identified between the Ythdf2^cKO and the Ythdf2^fl/fl OIR mice at P17, but the enlarged avascular areas were detected in the Ythdf2^cKO OIR mice at P21 and P25 (Fig. 7A–C), suggesting that the growth of physiological retinal vessels into the avascular area was interrupted upon microglia-specific Ythdf2 deletion.

Fig. 7.

Loss of Ythdf2 aggravates neovascularization and BRB breakdown in OIR retinas. (A) Immunofluorescence staining of IB4 in whole-mount retinas from WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice at P17, P21 and P25. Representative images and schematic diagrams are shown. Red dots represent NVT area, and grey area represents avascular area. Scale bar: 1 mm. (B-C) Quantification of retina NVTs (B) and avascular areas (C) in WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice. n = 10 for each group at P17; n = 8 for each group at P21; n = 8 for each group at P25. (D) Immunofluorescence staining of VE-cadherin and IB4 in NVT and non-NVT areas of retinal flat mounts from WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice at P21. Representative images along with quantification of VE-cadherin coverage are shown. n = 8 for each group. Scale bar: 30 µm. (E) Immunofluorescence staining of PDGFRβ and IB4 in NVT and non-NVT areas of retinal flat mounts from WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice at P21. Representative images along with quantification of PDGFRβ coverage are shown. n = 8 for each group. Scale bar: 30 µm. (F) Immunofluorescence staining of TER-119 and IB4 in NVT and non-NVT areas of retinal flat mounts from WT, Ythdf2^fl/fl and Ythdf2^cKO OIR mice at P21. Representative images along with quantification of TER-119⁺ cell leakage are shown. n = 8 for each group. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); * P < 0.05; ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for B-F.

Since above data indicated that Ythdf2 loss suppressed capillary morphogenesis to cause vascular leakage, we next measured whether BRB integrity was disturbed in the Ythdf2^cKO OIR mice. Retinal flat mounts were stained with VE-cadherin to visualize the endothelial adherens junctions. Discontinuous junctional distribution and reduced VE-cadherin coverage to IB4 was revealed in both NVTs and non-NVTs vessels of retinal flat mounts isolated from Ythdf2^cKO OIR mice at P21 compared to the control group (Fig. 7D). Immunofluorescence staining also detected reduced coverage of the pericyte marker PDGFRβ in retinal vasculatures of Ythdf2^cKO OIR mice at P21 (Fig. 7E). Vascular leakage was further measured by immunofluorescence staining of the erythroid cells marker TER-119. Leakage of TER-119⁺ erythroid cells outside the retinal vessels, both NVTs and non-NVTs, was observed in Ythdf2^cKO OIR mice at P21, while erythroid cells were mainly distributed along within the retinal vessels in Ythdf2^fl/fl OIR mice (Fig. 7F). These data further indicated that microglia-derived Ythdf2 deletion plays an indispensable role in facilitating pathological NVTs development and increasing BRB permeability in OIR retinas.

Ythdf2 supplementation mitigates microvascular dysfunction in OIR retinas

We next examined whether Ythdf2 supplementation could mitigate microvascular dysfunction in OIR retinas. The AAV-Ythdf2 was designed for microglia-specific overexpression of Ythdf2. It was packaged into an AAV serotype 2 vector and consists of the mouse Iba-1 promoter driving the expression of mouse Ythdf2 coding sequence fused to a C-terminal Flag tag. And AAV-blank, an empty vector, was used as the control (Supplementary Fig. S4A). We then intravitreally injected AAV-Ythdf2 or AAV-blank into mice at P12, and isolated retinas at P21 to identify the overexpression effect (Fig. 8A). The Ythdf2-Flag fusion protein was detected in nearly all microglia of mice receiving AAV-Ythdf2 (Supplementary Fig. S4B). The efficiency of Flag tag expression (Supplementary Fig. S4C) and the overexpression of Ythdf2 protein (Supplementary Figs. S4D,E) in the neural retina were both verified.

Fig. 8.

Ythdf2 supplementation mitigates microvascular dysfunction in OIR retinas. (A) AAV-mediated transduction of microglia in OIR mice. Neonatal mice received intravitreal injection of AAV at P12, and retinas were collected for analysis at P21. (B) Immunofluorescence staining of IB4 in whole-mount retinas from mice injected with AAV-Ythdf2 or AAV-blank at P21. Representative images, schematic diagrams, and quantification results are shown. Red dots represent the NVT area, and grey area represents the avascular area. n = 6 for each group. Scale bar: 1 mm. (C) Immunofluorescence staining of TER-119 and IB4 in retinal flat mounts collected from mice injected with AAV-Ythdf2 or AAV-blank at P21. Representative images along with quantification of TER-119⁺ cell leakage are shown. n = 6 for each group. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); * P < 0.05; ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for B-C.

We further assessed the therapeutic effect of AAV-Ythdf2. Intravitreal injection of AAV-Ythdf2 reduced the expansion of both the avascular and neovascularization territories in the OIR retinas (Fig. 8B). Consistently, immunofluorescence staining showed a reduced number of TER-119⁺ erythroid cells outside retinal vessels in OIR mice treated with AAV-Ythdf2 (Fig. 8C). Collectively, these results demonstrate that intravitreal delivery of AAV-Ythdf2 exerts a therapeutic effect on microvascular dysfunction in the OIR retina.

Ythdf2 regulates retinal microglial and vascular features by modulating the mRNA stability of Ace and Bmp4 in vitro

We next aimed to identify the down-stream regulators of Ythdf2 in microglia. As an m⁶A reader, Ythdf2 directly binds to the m⁶A sites on mRNA, thus affecting mRNA stability to down-regulate gene expression [31]. Therefore, we initially focused on the 538 up-regulated genes in Ythdf2^cKO microglia. A total of 251 genes were sorted out as targeted genes of Ythdf2 in mice tissues/cell lines by RM2Target database (https://rm2target.canceromics.org/#/home) with ≥ 2 evidences [32] (Fig. 9A). Among the 251 genes, 7 were reported to be associated with the pathogenesis of retinal microvascular diseases. We next selected the bone morphogenetic protein 4 (Bmp4) and angiotensin-converting enzyme (Ace) genes, top two up-regulated genes in Ythdf2^cKO microglia among the 7 genes, as potential down-stream regulators of Ythdf2 in microglia for further assessments. ACE is a key enzyme in the renin-angiotensin system (RAS), which converts the inactive angiotensin I into angiotensin II, a powerful vasoconstrictor that maintains body fluid homeostasis via blood pressure regulation [33]. BMP4, a secreted growth factor of the transforming growth factor β (TGF-β) superfamily, participates in multiple cellular processes, including regulations of glia activity, vascular development and angiogenesis [34,35].

Fig. 9.

Ythdf2 regulates retinal microglial and vascular features by modulating the mRNA stability of Ace and Bmp4 in vitro. (A) Flow chart showing the identification of Ace and Bmp4 as downstream genes regulated byYthdf2 in microglia. (B) qPCR shows the mRNA expression of Ace and Bmp4 in microglia isolated from Ctrl and OIR retina. (C) qPCR shows the mRNA expression of Ace and Bmp4 in primary isolated microglia under normoxia or hypoxia. (D) Immunoblotting shows the protein expression of Ace and Bmp4 in primary isolated microglia under normoxia or hypoxia. β-actin is used as an internal control for Ace, α-actinin is used as an internal control for Bmp4. Representative images along with quantification results are shown. n = 3 per group. (E) Abundance of m⁶A peak in the Bmp4 and Ace transcripts detected by MeRIP-Seq is presented. (F) MeRIP-qPCR analysis of m⁶A enrichment on the Ace and Bmp4 transcripts in BV-2 cells. n = 3 per group. (G) Ythdf2-RIP-qPCR validates the binding between the Ythdf2 protein and the Ace/Bmp4 mRNA in BV-2 cells. n = 3 per group. (H) The Ace and Bmp4 mRNA levels detected by qPCR in the Ythdf2^fl/fl and Ythdf2^cKO microglia at 0, 4 and 8 h post actinomycin D treatment. n = 3 per group. (I-J) The mRNA expression of Ace and Bmp4 in the Ythdf2^fl/fl and Ythdf2^cKO microglia under normoxic (I) or hypoxic (J) condition. n = 3 per group. (K-L) The protein expression of Ace and Bmp4 in the Ythdf2^fl/fl and Ythdf2^cKO microglia under normoxia (K) or hypoxia (L). β-actin is used as an internal control for Ace, α-actinin is used as an internal control for Bmp4. Representative images along with quantification results are shown. n = 3 per group. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. * P < 0.05; ** P < 0.01; *** P < 0.001. Two-tailed Student's T test for B-D and F-L.

We first checked if Ace and Bmp4 levels changed in microglia from OIR retinas or treated with hypoxia. qPCR detected higher expression of Ace and Bmp4 in microglia from OIR retinas (Fig. 9B). Both qPCR and immunoblotting confirmed that, after 72 h of hypoxia, microglia had increased mRNA and protein levels of Ace and Bmp4 compared to normal conditions (Fig. 9C,D). We next analyzed whether Ace and Bmp4 were direct down-stream targets of Ythdf2 in mouse microglia. M⁶A sites in mice Ace and Bmp4 genes revealed by RMBase v3.0 database (https://rna.sysu.edu.cn/rmbase3/) (Fig. 9E). Direct bindings between the m⁶A antibody and the m⁶A sites in the Ace and Bmp4 transcripts were validated using the methylated RNA immunoprecipitation (MeRIP)-qPCR experiments in microglia (Fig. 9F). We also confirmed bindings between the Ythdf2 protein and the Ace and Bmp4 transcripts in microglia using the RNA immunoprecipitation (RIP)-qPCR assay (Fig. 9G). To further tell whether Ythdf2 directly regulates Ace and Bmp4, we annotated their mRNA stability and levels in primary microglia isolated from the Ythdf2^cKO and Ythdf2^fl/fl mice. Prolonged half-life of the Ace and Bmp4 transcripts was detected in Ythdf2^cKO microglia treated with the transcription inhibitor actinomycin D compared to Ythdf2^fl/fl microglia (Fig. 9H), indicating that Ythdf2 disturbs mRNA stability of Ace and Bmp4. Consistently, both qPCR (Fig. 9I,J) and immunoblotting (Fig. 9K,L) identified promoted Ace and Bmp4 expression in microglia with Ythdf2 knocked out compared to the control group under both normoxic and hypoxic conditions. Collectively, these data revealed that Ythdf2 regulates mRNA stability of Ace and Bmp4 in microglia as an m⁶A reader in both biological and pathological status.

Supplementation of captopril or noggin ameliorates microvascular dysfunction induced by loss of Ythdf2 in OIR retinas

Since Ythdf2 in microglia mediates retinal microvascular phenotypes by regulating the mRNA stability of Ace and Bmp4, we therefore tested whether application of antagonists to Ace or Bmp4 could partly alleviate the microvascular dysfunction induced by microglia-specific Ythdf2 loss in the OIR retinas. We showed that intra-peritoneal injection of captopril, an angiotensin-converting enzyme inhibitor (ACEI) that blocks the enzymatic activity of Ace (Fig. 10A), partly ameliorated the enlargement in both the avascular and the NVT areas caused by microglia-specific Ythdf2 loss in the OIR retinas (Supplementary Fig. S5A). We further tested whether captopril could also suppress vascular leakage in the Ythdf2^cKO OIR mice. Consistently, immunofluorescence staining identified decreased amount of TER-119⁺ erythroid cells outside the retinal vessels in the Ythdf2^cKO OIR mice supplemented with captopril (Supplementary Fig. S5B). We next checked whether noggin, an antagonist to Bmp4 (Fig. 10B), could alleviate the pathogenic retinal features in the Ythdf2^cKO OIR retinas. Our data demonstrated that, similar to intra-peritoneal injection of captopril, the enlarged avascular areas and accelerated NVTs developments in the Ythdf2^cKO OIR mice was also partly suppressed by intra-vitreal injection of noggin (Supplementary Fig. S5C). Consistently, supplementation of noggin also restrained retinal vascular leakage in the Ythdf2^cKO OIR mice (Supplementary Fig. S5D).

Fig. 10.

Supplementation of captopril or noggin ameliorates microvascular dysfunction induced by loss of Ythdf2 in OIR retinas. (A) Schematic illustration of captopril administrations and biological functions of captopril. Captopril is intra-peritoneally injected into the Ythdf2^cKO OIR mice for 6 continuous days from P12 to P17. (B) Schematic illustration of noggin administrations and biological functions of noggin. Noggin is intravitreally injected into the Ythdf2^cKO OIR mice at P12. (C) Immunofluorescence staining of IB4 in whole-mount retinas collected from mice treated with combined captopril and noggin at P21. Representative images, schematic diagrams, and quantification results are shown. Red dots represent the NVT area, and grey area represents the avascular area. n = 6 for each group. Scale bar: 1 mm. (D) Immunofluorescence staining of TER-119 and IB4 in retinal flat mounts collected from mice treated with combined captopril and noggin treatments at P21. Representative images along with quantification of TER-119⁺ cell leakage are shown. n = 6 for each group. Scale bar: 30 µm. Data represent different numbers (n) of biological replicates. Error bars represent mean ± SEM. NS: not significant (P > 0.05); * P < 0.05; ** P < 0.01; *** P < 0.001. One-way ANOVA coupled with the Bonferroni’s post hoc test was used for C-D.

Given that Ace and Bmp4 each partially mitigated the pathological phenotypes, we explored whether their combined use would yield a more robust therapeutic effect. Our data indicate that combined captopril and noggin treatment more effectively suppressed the enlarged avascular areas and the accelerated NVTs development than either intervention alone (Fig. 10C,D). Collectively, both captopril and noggin showed inhibitory roles in the Ythdf2 depletion induced pathogenic NVTs development, interrupted physiological vessels growth, and vascular leakage in the OIR retinas. Thus, our data further supported Ace and Bmp4 as down-stream regulators of Ythdf2 in retinal microglia, and implied potential therapeutic options for microvascular retinopathies caused by Ythdf2 insufficiency.

Discussion

While endothelial-intrinsic pathways are undeniably critical for vascular development, immune-derived signals from microglia are equally important modulators of the vascular microenvironment, particularly under ischemic or pathological conditions [36,37]. Microglia influence vascular cells through both paracrine signaling and direct contact, maintaining retinal function via physical interactions with blood vessels and contributing to angiogenesis through the secretion of soluble factors [10,38]. They also regulate vessel diameter and blood flow velocity through neurovascular coupling [7,39]. Critically, aberrant microglial activation accelerates microvascular anomalies, and a bidirectional causal relationship exists between microglial activation and disrupted angiogenesis [40]. Activated microglia release inflammatory mediators that compromise vascular barrier function [41,42], while hypoxia arising from vascular insufficiency further exacerbates microglial dysfunction, establishing a pathological feedback loop [43]. However, the molecular mechanisms driving this interplay remain elusive. Increasing evidence highlights the potential roles of m⁶A modification in regulating microglia phenotypes, yet its specific involvement in microglial dysfunction during retinal vasculopathy is poorly understood [44,45]. This study identifies microglia-expressed YTHDF2 as a crucial regulator of microglial activation, retinal vascular sprouting, and remodeling—processes essential for both normal vascular development and the resolution of pathological features.

Although YTHDF2′s role in post-transcriptional regulation is established in various cell types, its function in microglia during retinal diseases remained elusive. Prior studies implicated Ythdf2 in mitigating microglia activation during central nervous system inflammation via circHIPK2 degradation [21] and in betaine-mediated suppression of microglia pyroptosis in dementia [46]. However, its pathological involvement in ocular diseases had not been investigated. To address this gap, we generated the microglia-specific Ythdf2 knockout mice. These mice exhibited pronounced microglia activation in both developing retinas and in the OIR model. During normal development, Ythdf2 knockout retinas displayed transient vascular abnormalities, including delayed sprouting, accelerated pruning, and BRB disruption, peaking around P7-P14, the period of maximal microglial involvement in vascular remodeling. By P30, coinciding with the transition of microglia to a surveillance state and vascular maturation [47], these defects resolved. This resolution may also involve compensatory mechanisms, potentially through other m⁶A readers compensating for YTHDF2 loss post-development [48]. Crucially, however, such transient developmental defects can establish long-term vulnerability, mirroring the lifelong visual impairments seen in ROP patients stemming from early vascular delays [49].

Under pathological conditions like OIR, sustained hypoxia exacerbates microglial dysfunction. Our findings demonstrate that deficiency in microglia-expressed YTHDF2 consistently worsens microglial activation and drives abnormal neovascularization in this setting. This establishes a clear causal link between microglial Ythdf2 deficiency and the progression of retinal microvasculopathy. The role of Ythdf2 in retinal vasculopathy has been documented in previous studies. Specifically, Ythdf2 exert a DR-inhibitory effect in endothelial cells by regulating the mRNA instability of targets including CDK2, MAP4K4, ITGB1 and PIEZO1 mRNA instability [43,[50], [51], [52]]. Based on current evidence, Ythdf2 appears to have similar regulatory effects across different retinal vascular diseases, though the cell type–specific effects may differ. Collectively, Ythdf2 is a key regulator of both developmental and pathological angiogenesis, and its activity within specific time windows underscores the dynamic nature of microglial-vascular interactions.

Mechanistically, as an m⁶A reader, Ythdf2 directly regulated mRNA stability of Ace and Bmp4 in microglia under both normoxic and hypoxic conditions. A complete RAS exists in both developing and adult retinas, with components in vasculatures, neurons and glia cells [53,54]. As key elements of the RAS, ACE and angiotensin II influence the developing retinal vasculatures and play pathogenic roles in retinal microvascular diseases like DR and ROP [53,55]. Angiotensin II activates retinal microglia in ischemic retinopathies [56], and retinal RAS activation stimulates pro-angiogenic and pro-inflammatory factors, driving angiogenesis, BRB destruction, and vascular leakage [55,57]. Consequently, ACEI, which blocks the RAS, protect against retinal microglial activation and vascular abnormalities in vasculopathies [55,56,58]. Similarly, BMP4 also demonstrates pro-angiogenic and pro-inflammatory roles. It induces microglial polarization, and antagonizing BMP4 signaling mitigates microglial activation and reverses M1/M2 imbalance [35,59]. In endothelial cells, BMP signaling regulates sprouting angiogenesis and vascular remodeling [60], and BMP4 contributes to endothelial dysfunction in DR models, impairing barriers and exacerbating fibrosis [34,61,62]. While Ace and Bmp4 are key downstream targets, other YTHDF2 effectors in microglia likely exist, warranting further investigation into the full regulatory network of microglial YTHDF2 in ischemic retinopathy.

Therapeutically, while anti-VEGF agents are the first-line treatment for many retinal vascular diseases due to their efficacy in halting abnormal vessel growth, limitations exist. These include the need for repeated injections, inability to address the early immune-inflammatory stage, and variable or diminishing patient responses. Our findings suggest that targeting the microglial YTHDF2-ACE/BMP4 axis offers a promising complementary or alternative strategy. YTHDF2 supplementation, Captopril (an ACEI), and Noggin (a BMP4 antagonist) modulate microglial activation and the associated immune response, thereby improving vascular abnormalities. Key advantages include targeting the source of pro-inflammatory factors (microglia), potentially covering a broader disease course, and acting upstream of VEGF. This upstream targeting could reduce dosing frequency compared to anti-VEGF therapy, which primarily suppresses existing VEGF. Captopril, an established oral antihypertensive with decades of safety data [63], offers the added benefits of improved patient adherence and reduced injection-related complications.

Small exploratory studies suggest oral ACE inhibitor, including Captopril, Lisinopril and Enalapril, may reduce retinal vascular leakage in DR [[64], [65], [66]], though clinical evidence for YTHDF2 supplementation and Noggin in ophthalmology remains lacking. Future clinical research is essential to confirm the safety and efficacy of targeting this axis and to explore optimal formulations and administration routes.

Conclusions

Our study highlights the crucial role of microglia-expressed Ythdf2 in regulating retinal microglia and microvascular functions. We have also annotated the downstream regulatory network of Ythdf2 in microglia. Our study offers deep insights into the molecular mechanisms and cellular events in microvascular development and disorders. Our findings imply therapeutic potentials for microvascular diseases with pharmacological modulation of the Ythdf2-Ace/Bmp4 network.

CRediT authorship contribution statement

Hong-Jing Zhu: Conceptualization, Methodology, Investigation, Visualization, Writing – original draft, Writing – review & editing. Yi-Chen Zhang: Methodology, Investigation, Visualization, Writing – original draft, Writing – review & editing. Ye-Ran Zhang: Methodology, Investigation, Visualization, Writing – original draft, Writing – review & editing. Bing Qin: Methodology, Investigation. Xin Cao: Methodology, Investigation. Jia-Nan Wang: Methodology, Investigation. Ying Wang: Methodology, Investigation. Zi-Qin Ding: Methodology, Investigation. Meidong Zhu: Investigation. Jiang-Dong Ji: Investigation. Qing-Huai Liu: Investigation. Biao Yan: Investigation, Visualization, Supervision, Writing – review & editing. Song-Tao Yuan: Investigation, Supervision, Writing – review & editing. Xue Chen: Conceptualization, Supervision, Writing – original draft, Writing – review & editing.

Funding

This study was supported by National Natural Science Foundation of China (82471105 to XC); Natural Science Foundation of Jiangsu Province (BK20231371 to XC); Jiangsu “333” Advanced Talent-training Project to XC; Social Development Program of Jiangsu Province (BE2022805 to QHL); Jiangsu Province Hospital High-level Talent Cultivation Program (Phase I; CZ0121002010037 to XC). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Compliance with Ethics Requirements

All animal experiments were approved and consistently reviewed by the Ethical Review Committee of Nanjing Medical University (approval number: IACUC-2307023 for mice) in accordance with the guidelines for the care and use of laboratory animals (published by NIH Publication No. 86-23, revised 1996).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: