Background
Retinal ischemia–reperfusion (I/R), a common cause of irreversible visual impairment, is involved in the pathological mechanisms of many eye diseases, including glaucoma, diabetic retinopathy, and retinal occlusion [
1‐
3]. Retinal I/R is defined as an initial restricted retinal blood supply followed by perfusion restoration [
3,
4]. This process results in retinal ganglion cells (RGC) death, morphological degeneration, function loss, and ultimately vision loss [
5] due to free radical production, mitochondrial dysfunction, exacerbated oxidative stress, inflammatory response, and activation of calpains and caspases. Current treatments for retinal I/R injury include intraocular injections, eye drops, and surgery [
6]. However, due to the complexity of the underlying mechanisms, no thoroughly effective treatments are available. The limitations of the available treatments motivated studies searching for alternatives with wide safety margins. Understanding the mechanisms controlling human retinal development is particularly important when treating vision-threatening diseases.
Previous studies have demonstrated that neuroinflammation was involved in I/R injury [
7,
8]. The inflammatory response includes rapid and transient granulocyte infiltration, slow monocytes/macrophages accumulation, and resident microglia activation, proliferation, and mobilization to the vascular barriers [
9]. Simultaneously, ischemia mediates retinal tissue damage and RGC death by triggering damage-associated molecular pattern-induced toll-like receptor 4 (TLR4), inflammasome-dependent neuroinflammation and microglial activation [
7,
8]. Furthermore, cell deaths, including apoptosis, pyroptosis, and autophagy, has been implicated in inflammatory cytokine release [
10,
11]. However, these studies focused on a limited number of cell types or mechanisms, lacking knowledge of the complex hierarchies and where the implicated genes are expressed. The full impact of I/R injury on the retina and the alternations in cell-type proportions have not been fully explored or understood. In addition, the unique interactions among the retinal cells under I/R conditions required further investigation. Therefore, a comprehensive retinal cell atlas that encompasses the influences of I/R at a single-cell resolution was urgently needed to integrate various retinal I/R interconnected components, pathways, and cell types.
Single-cell RNA sequencing (scRNA-seq) is a powerful and unbiased approach to comprehensively classify cell types and states based on their individual transcriptomes in health and disease. Its application has greatly boosted the understanding of the pathogenesis of many diseases [
12‐
14]. Herein, we applied scRNA-seq to comprehensively characterize and compared mouse retinas of an I/R model and a wild-type (blank) control, aiming to identify marked changes in the transcriptome of various cell types. Ferroptosis occurred in most retinal cell types in response to I/R, particularly in the photoreceptors, glial cells, and RGCs. We found that ferrostatin-1 (Fer-1, a ferroptosis inhibitor) administration ameliorated RGC death, reduced the immune response induced by microglia, and protected retinal structure and function. These findings offer unique insights into potential strategies for treating diseases caused by retinal I/R injury, including glaucoma.
Materials and methods
Preparation of single-cell suspension of mouse retina
3 days after ischemia–reperfusion injury, 3 retinas were taken from the model group or the blank group, respectively, as a mixed sample, and the retinas were removed after perfusion with pre-cooled 0.9% normal saline. 2 mg/ml collagenase-D and 200 U/ml DNAseI were added for digestion, and an equal volume of 10% BSA was used to terminate the digestion, then cell suspension was collected. All cells were washed twice with PBS, and 10 μl samples were stained with Trypan blue for cell count. The number and viability of cells were calculated accurately. Adjust cell terminal concentration to about 1.0 × 106 cell/ml.
ScRNA-seq data alignment, processing, and sample aggregation
Gel Bead and Multiplex Kit, Chip Kit (10X Genomics) and Chromium Single Cell 5′ Library (10X Genomics, Genomics chromium platform Illumina NovaSeq 6000) were used to convert the harvested single-cell suspensions into barcoded scRNA-seq libraries. Single-cell RNA libraries was made up of the Chromium Single Cell 5′ v2 Reagent Kit (120237; 10X Genomics) according to the manufacturer’s instructions, and FastQC software was used to inspect library quality. All sequenced data were preliminarily processed by CellRanger software (version 4.0; 10X Genomics). The count pipeline of the CellRanger Software Suite was applied to demultiplexed and barcoded sequences. Seurat package (version 3.0) was applied to filtration, normalization, dimensionality reduction, cell clustering, and differential gene expression analysis of the processed data, on the foundation of the calculated single-cell expression matrix by CellRanger. Cells with less than 200 genes detected and a mitochondrial gene ratio of more than 20% were excluded. A total of 53,701 cells (Blank, 15,617 cells; Model, 38,084 cells) were analyzed after quality control.
Dimensionality reduction and clustering analysis
The ‘‘NormalizeData’’ function was used to log-normalize the counts of each cell (1+ counts per 10,000). Dimensionality reduction was achieved by “RunPCA” function. Cells were visualized by means of a two-dimensional t-SNE algorithm in the ‘‘RunTSNE’’ function. The “FindNeighbors’’ and “FindClusters” functions were used to identify significant clusters at an appropriate resolution. The function “FindAllMarkers” served as identifying marker genes of each significant cluster. The top 10 most variable genes were extracted by the “FindVariableGenes” function in Seurat with default parameters.
Differential expression analysis
Before running differential expression analysis, cell types which were missing or possessed no more than three cells in two groups were filtered out. To identify certain cell type between different groups, differential expression analysis was performed by the Wilcoxon rank-sum test implemented in the ‘‘FindMarkers’’ function of the Seurat package (version 3.0). A ferroptosis-related DEG data set was established (P value < 0.05, |Log2FC| > 0.25) after identification of DEG between groups.
Gene functional annotation
As a web-based portal, Metascape (
www.metascape.org) was used to conduct GO and pathway analysis with the input of DEG [
15]. The top 10 of 30 GO biological processes and pathways among groups and clusters were visualized by ggplot2 R package.
Determination of cell–cell interactions
Cell-level interactions among different cell types was analyzed under the help of processed scRNA-seq data. CellChat (
https://github.com/sqjin/CellChat), an R package could quantitatively calculate the intercellular communication networks and predict the main signal pathways, was used to implement the signaling pathway networks visualization. Only receptors and ligands expressed in at least 10% of specific cells were used for further analyzed, besides, communication was considered not theoretically existed if the ligand and receptor were not detected. TBtools (
www.tbtools.com) was applied to data normalization and heatmap drawing.
Scoring of biological processes
Individual cells were scored by gene signatures representing certain biological functions, then calculated with the average of normalized expression of corresponding genes. Full gene lists with biological functional signature were obtained in GO and KEGG database. For instance, the inflammatory response score was determined by calculating the average expression of genes in the GO term “inflammatory response” (GO: 0006954). Ferroptosis-related gene signatures were obtained from the KEGG Pathway data set “Ferroptosis” (ko04216).
I/R model and ferrastatin-1 treatment
Six-week-old C57BL/6J male mice were purchased from the Guangdong Medical Laboratory Animal Center. All animals were kept in a specific pathogen-free facility in Animal Laboratories of Zhongshan Ophthalmic Center and the experiments were approved by the Institutional Animal Care and Use Committee of Zhongshan Ophthalmic Center, Sun Yat-sen University. The retinal I/R model were established as previously described [
16]. In brief, a 30-gauge needle containing a balanced salt solution was cannulated into the anterior chamber to maintain the IOP at 70 mmHg. The sham operation, which acted as the control, was implemented a same operation without elevating the IOP. After 60 min, the IOP was normalized by carefully withdrawing the needle. Tobramycin ointment (Alcon, USA) was used to prevent a bacterial infection. It's worth noting that 30 μg/ml ferrostatin-1 was injected intravitreally to the operated eye immediately after I/R injury.
Histological examination
7 days after the operation, heart perfusion was applied to euthanized mice using normal saline and 4% paraformaldehyde (PFA). Whole eye balls were fixed in 4% PFA at 4 °C over night and followed by paraffin fixation. Every sample was sectioned into 10-μm-thick slices through the optic nerve plane and followed by H&E staining. Four cross-sectional measurements around the optic nerve (within 1 mm) for every retina were chosen to quantify the inner plexiform layer (IPL) thickness using a Leica DM6B device and analyzed by ImageJ software (
https://imagej.nih.gov/ij/).
Immunofluorescence staining and examination
Retinas were collected carefully then fixed by 4% paraformaldehyde, permeabilized by 0.3% Triton X-100, and blocked by 5% bovine serum albumin. For whole-mount staining, collected samples were incubated with primary antibodies at 4 °C overnight, followed by a species-compatible secondary antibody for 2 h at room temperature. The sources and dilutions of antibodies listed as Additional file
2: Table S1. Cell nuclei were stained by 1 ng/ml 40,6-diamidino-2-phenylindole (DAPI; Beyotime, China) for 10 min. Images were captured by a Zeiss LSM 880 confocal laser-scanning microscope and processed by ZEN 2 and Adobe Photoshop CS6.
Cell culture of primary retinal microglia
Retinal microglia were isolated from the retina of 3–5 day C57BL/6J neonatal wild-type mice as previously described with a slight modification [
17]. In brief, mixed glial cultures were separate from the retinas, followed by a chemical dissociation (trypsin–EDTA 0.05%, Gibco) conducted at 37 °C with a gentle blow for 1 min until the tissue masses disappeared and digestion was terminated with the equivalent amount of DMEM/F12 (Gibco, USA) supplemented with 10% (vol/vol) FBS. Retinal mix cells were then resuspended in DMEM/F12 (Gibco, USA) supplemented with 10% (vol/vol) FBS, and continued to culture until reaching confluency. Detached microglia were collected by a horizontal shaker at 180 rpm, 37 °C for 2 h. Transferred the collected cells into a 6-well plate and continued to culture for another 1–2 weeks before an identification by flow cytometry.
Establishment of the OGD/R model
As an in vitro model of retinal ischemia/reperfusion, we used oxygen–glucose deprivation/reperfusion (OGD/R) in primary microglia cells. In brief, cells were cultured in glucose-free DMEM(Gibco) after washing twice by PBS(Gibco), then subjected to an anoxic chamber (Billups-Rothenberg, Inc., USA) containing 5% (vol/vol) CO2 and 95% (vol/vol) N2 at 37 °C for 3 h. For the same duration, cells incubated in serum-free medium supplemented with 4.5 g/l d-glucose under normoxic conditions acted as control. At the end of the exposure period, cells were then reoxygenated [5% (vol/vol) CO2 and 95% (vol/vol) air] with normal medium for 24 h.
Western blotting
Total protein was isolated from the retinas with radioimmunoprecipitation assay (RIPA) lysis solution (Beyotime, China) and run on 12% (wt/vol) gradient polyacrylamide gels following a standard protocol. The expression of target proteins was normalized to β-actin obtained from the same sample (taken as 1.0) and then quantified using ImageJ Software (
https://imagej.nih.gov/ij/). The primary antibodies and dilutions were listed in Additional file
2: Table S1.
Quantitative real-time PCR and qPCR
Total RNA was extracted from the retina and cultured cells with TRIzol Reagent (Invitrogen) according to the manufacturer’s instructions. cDNA was synthesized with the PrimeScript RT Master Mix (TaKaRa) according to a standard protocol. Quantitative analysis was conducted by the Light Cycler 480 Real-Time PCR System (Roche Molecular Systems, Inc., SUR). The expression of target mRNAs was measured and normalized to that of GAPDH. Additional file
2: Table S2 lists the primers.
Flow cytometry and detection of reactive oxygen species (ROS)
Cells were isolated from eye balls and prepared to single-cell suspensions. The cell activity was detected with Zombie NIR™ Fixable Viability Kit (APC-Cy7, catalog 423105, Biolegend, San Diego, CA, USA). Then, cells were stained with fluorochrome-conjugated mAbs surface markers CD11b (Bv605, catalog 101237, Biolegend) for 15 min. For detection of reactive oxygen species (ROS), cells next were incubated with H2DCFDA (FITC, catalog HY-D0940, MedChemExpress, Monmouth Junction, NJ, USA) to detect the ROS at 37 °C for 20–30 min. Finally, washing cells with cold PBS for three times before analyzing with flow cytometry (BD LSR Fortessa, BD Bioscience, San Jose, CA, USA). All data were analyzed using FlowJo (TreeStar, Ashland, OR, USA).
Statistics
Data quantification were performed blindly and presented as mean ± SE of measurement (SEM). Data were analyzed statistically using one-way ANOVA with Bonferroni’s post hoc test for comparisons of three and more groups or two-tailed Student’s t test for two group comparisons. To assess significance, a value of P < 0.05 was considered statistically significant (*P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001). The sample sizes and P values are indicated in the figure legends.
Data availability
The scRNA-seq data are deposited in the Genome Sequence Archive in BIG Data Center, Beijing Institute of Genomics (BIG,
https://bigd.big.ac.cn/gsa/), Chinese Academy of Sciences, under the Project Accession No. PRJCA008174 and GSA Accession No. CRA006042.
Discussion
We applied scRNA-seq analysis and, for the first time, provided single-cell transcriptomics that examined the gene expression in mice with retinal I/R injury. We comprehensively understood the I/R impact on various retinal cell types using cell-type composition, subset-specific gene expression, enriched pathways, and cell–cell communication. The major findings of this study were: (i) the scRNA-seq technology revealed unique gene signatures of retinal cells in mice with retinal I/R injury; (ii) ferroptosis genes’ expression was upregulated in photoreceptor cells, RGC, and glial cells; (iii) ferroptosis occurred in RGC after retinal I/R, alleviated, at least in part, by Fer-1 inhibition of ferroptosis; (iv) the myeloid participated in ferroptosis development after retinal I/R, while Fer-1 attenuated I/R-induced IBA-1 positive cells activation and neuroinflammation.
The vertebrate retina is one of the most well-characterized regions of the central nervous system [
30,
31]. It serves as a good model system for studying the neurological impairment mechanisms following I/R. While extensive investigations have well-characterized I/R-related diseases using bioinformatic methods [
32‐
34], the transcriptional features of retinal I/R have not been explored using high-throughput approaches, and a detailed retinal I/R cell atlas was missing before this study. It was very important to reveal the retinal cells’ transcription characteristics to understand the damage caused by I/R. To our knowledge, this was the first study to delineate the retinal cell landscapes of both I/R and blank mice. We identified 12 cell types and described their proportional variations during I/R.
So far, our study has first identified the influx of three immune cell types, myeloid, Neutrophil, and T&DC, dominated the acute inflammatory phase at a single-cell level, accounting for the known inflammatory cascade accumulated by the damage of the blood-eye barrier during I/R [
35‐
37]. These findings suggest that targeted regulation of hyperexpressed genes in the blood-derived immune cells, or inhibition of their recruitment, may be a new approach for protective treatment of retinal ischemia–reperfusion injury, including glaucoma. Although the ratio of retinal Mag decreased, the Mag subsets increase following I/R might be associated with considerable immune cell infiltration and a subsequent decrease in Mag proportion. Furthermore, it has been suggested that RGC, CBC, and the photoreceptors were highly sensitive to I/R [
38,
39]. Almost all neuronal cell types were damaged in our study, with ROD seemingly more than CONE. The mouse I/R retinal atlas illustrated aspects of cell-type diversity and function, and revealed transcriptomic and cell-type compositional differences.
Cell death plays a critical role in the response to various stresses and the activation of the immune system. Studies have emphasized the role of deleterious inflammation and oxidative stress in various cell death modes of I/R, including apoptosis, necroptosis, and pyroptosis [
40‐
43]. However, blocking one of these signaling pathways cannot completely ameliorate inflammation, suggesting the mutual involvement of several cell death pathways, as indicated by GO analysis of various cell types in the I/R retinas. Apoptosis, the most well-characterized PCD to date, is immunologically silent (“clean”) and its activation does not promote a significant inflammatory or autoimmune response [
44,
45]. The non-apoptotic PCDs, including necroptosis, pyroptosis, and ferroptosis, are a much “dirtier”. The most well-characterized inflammation associated cell death models defined primarily by acute injury, including those induced by I/R, pathogenic infection, and neurodegenerative and neuroinflammatory diseases. These pathologies have a significant combination of cell death components, resulting in a severe inflammatory state. Furthermore, considering the role of microenvironmental molecular changes in RGC damage, we showed through ligand–receptor interaction map that the SEMA3 pathway was downregulated in RGC after I/R. The SEMA3 pathway is known for its role in axon guidance, facilitating retinal axon crossing in the chiasm [
26], possibly accounting for the difficulty in nerve regeneration after I/R damage. In summary, retinal I/R-induced nerve injury involves various cell death pathways.
Of importance, ferroptosis overexpression in almost all cell types can enhance our understanding of retinal I/R. Unlike the other classical non-apoptotic PCD, ferroptosis is a unique PCD characterized by iron-dependent accumulation of lipid hydroperoxides to lethal levels [
46‐
48]. Iron homeostasis is tightly regulated as its deficiency and overload could lead to various pathological conditions [
49,
50]. Excessive iron release can induce intraocular peroxidation of unsaturated phospholipids and retinal inflammation [
51,
52]. Iron intake might increase the risk of glaucoma [
53]; however, very little research has been done in this area. Indeed, iron uptake and transport were found in our GO pathway analysis, suggesting iron imbalance in I/R retinas. Emerging evidence indicated that ferritin light chain 1(Ftl1) and ferritin heavy chain 1 (Fth1) were involved in iron storage, entry, and homeostasis [
54‐
58]. Our DEG analysis found that these were upregulated in most of cell subsets. Ferroptosis and iron homeostasis imbalance have been implicated in several neuroinflammatory diseases, including ischemic stroke, subarachnoid hemorrhage, Alzheimer’s disease, and Parkinson disease [
59,
60]. Our cell atlas and bioinformatic methods revealed considerable enrichment of ferroptosis response gene networks (e.g., Slc7a11, Gpx4 and Ptgs2) in the I/R retinas in vivo and in vitro. Ferroptosis onset involves an intracellular iron overload and upregulation lipid peroxides. Ptgs2, Slc7a11 and Gpx4 are widely accepted ferroptosis biomarkers (57, 63). Networks revolving around glutathione peroxidase 4 (GPX4) regulate ferroptosis, among which the system Xc
−–glutathione–GPX4 axis is most well-known one [
55,
61]. Slc7a11, a key component of the cystine/glutamate antiporter system Xc
−, transfers cystine to glutathione biosynthesis and regulates cellular lipid peroxidation [
54‐
58,
62]. The expression or activity of Gpx4 restrain lipid peroxides at the expense of GSH [
63]. In models of CNS injury, microglial activation has been shown to induce iron overload and subsequently trigger neuronal ferroptosis [
64]. We showed that ferroptosis inhibitors reduced inflammatory cytokines, partially prevented RGC death, and mitigated the activation of IBA-1 positive cells, the executor of PCD and mediator of immune storm. These findings provide compelling evidence that ferroptosis is a promising therapeutic target for immune programming. In a newly published study, Qin and colleagues investigated the interactions among different modes of cell death (apoptosis, necroptosis, and ferroptosis) in RGC loss during retinal I/R injury [
65], which is consistent with our findings. In addition to RGC, we discovered that almost all retinal cells, such as CONE, ROD, Mag and myeloid underwent ferroptosis with retinal I/R injury as well. Importantly, ferroptosis in myeloid took a non-negligible part in retinal I/R injury by recruiting inflammatory factors and amplifying inflammatory cascades. Given the strong pro-inflammatory effects of ferroptosis [
62], its regulation could help control the inflammatory and immune responses, crucial knowledge when designing optimized therapies for retinal I/R-induced neuroinflammation. Targeted inhibitors could alleviate ferroptosis-mediated neuronal damage and reduce neuroinflammation by IBA-1 positive cells activation.
Post-I/R inflammation is a complex process involving diverse signaling pathways and various cell types. Cell death has been reported to trigger neuroinflammation following plasma membrane bursting and subsequent promotion of pro-inflammatory responses from the immune system [
66‐
68]. Microglia, the immunocompetent cells of the CNS and the first responders to neuronal injury and death, become reactive upon retinal I/R and produce various cytokines, including TNF-α, IL-6, and IL-1β [
69‐
71]. We and others have investigated the mechanisms underlying myeloid cell-mediated initiation and resolution [
11,
72] of inflammation. Our findings demonstrate that myeloid are the main effector cells to induce cell mediator release and inflammatory infiltration. Upregulated inflammatory cytokines (IL-1, IL-6, and TNF-α) and pathways (CCL, TNF, and myeloid leukocyte activation) are involved in inflammatory spreading. In turn, inhibited ferroptosis in myeloid, combined with a reduction of cell death, resulted in the inhibition of myeloid activation. Therefore, targeting retinal I/R injury induced ferroptosis can “kill two birds with one stone” by inhibiting cell death-mediated damage and reducing inflammatory activation. Since ferroptosis was the only PCD mode overexpressed in almost all retinal cells, we showed here for the first time that Fer-1 reduced inflammatory cytokines and IBA-1 positive cells activation, and enhanced RGC survival. Furthermore, similar to the synchronized responses of macroglia and IBA-1 positive cells to neuron injury, their bi-directional interactions are critical for building and amplifying neuroinflammation and dictating neurologically compromised outcomes [
73]. Müller cells, the principal retinal macroglia, undergo reactive gliosis after acute injury or chronic neuronal stress, which exerts dual functions [
74]. GO analysis of Müller cells indicated the induction of a pro-inflammatory response through the upregulation of the IFNG and MAPK pathways. In addition, we discovered an interesting phenomenon, in which inflammatory pathways were upregulated in non-immune cell subsets, including RGC and photoreceptor cells. Traditionally, photoreceptor cells were assumed highly specialized, terminally differentiated neurons that detect photons and transmit light information to bipolar cells in the retina [
75]. However, their structural and metabolic requirements render them susceptible to many acquired and genetic injury sources. GO analysis of photoreceptor subsets showed that CONE-SC3 and ROD-SC2 were specifically involved in inflammatory responses (MAPK and IFNG signaling pathways). These novel evidences suggest that ferroptosis may be a new target for retinal I/R injury diseases, including retinal vessel occlusion, diabetic retinopathy and glaucoma. However, the exact mechanism of ferroptosis has not been elucidated yet. In addition, most of the existing ferroptosis inhibitors are non-specific, which could also lead to the limitation of clinical drug research. Therefore, targeting I/R-induced neuroinflammation by blocking ferroptosis might be an attractive strategy to treat retinal I/R injury.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.