Animals

C57BL/6J and Alox5−/− (B6.129S2-Alox5tm1Fun/J, stock number 004155) male mice were obtained from Jackson Laboratory (Bar Harbor, ME). C57BL/6J mice were used as a congenic control for the Alox5−/− (5-LOX KO) mice. All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of California, Berkeley. Mice were housed in a controlled environment, maintained on a 12-hr light-dark cycle, and provided unrestricted food and water access throughout the study.

Silicone oil injections

OHT was induced in C57BL/6J and 5-LOX KO mice using an established silicone oil model [25, 26]. Briefly, male mice were anesthetized at 8 wks of age using intraperitoneal injection of ketamine/xylazine (100 mg/kg and 10 mg/kg, respectively). A topical anesthetic (0.5% proparacaine hydrochloride; Sandoz, Princeton, NJ) was applied to the eye. Under microscopic guidance, a sterile 31 G paracentesis needle was used to create an incision in the anterior chamber of the eye, ensuring no damage to the iris or lens. The needle was slowly withdrawn to release approximately 1–2 µL of aqueous humor. Subsequently, 1.2 or 1.8 µL of silicone oil (Alcon, Fort Worth, TX) was injected into the anterior chamber using a 33 G Hamilton syringe (Reno, NV). The syringe was slowly withdrawn after holding it for 10 s. To minimize silicone oil leakage, the eyelids were gently closed to cover the corneal incision. After the injection, an antibiotic drop (0.3% Tobramycin Ophthalmic Solution, Bausch and Lomb, Laval, Canada) was applied to the eye. Silicone oil was injected into both eyes, and the uninjected eyes from separate groups of animals were used as normotensive controls. Mice were kept on a heating pad until fully recovered from anesthesia. Intraocular pressure (IOP) was measured by Tonometer (TonoLab, Vantaa, Finland) at different time points after dilating the eyes with tropicamide solution (Akorn, Lake Forest, IL) for 10 min. At designated time points (1, 2, 4, and 6 wks) post-injection, mice were euthanized, and eyes were enucleated in sterile phosphate buffer saline (PBS) at 4 °C. Retina and optic nerve were dissected and stored in 4% paraformaldehyde (4 °C) and TRIzol (-80 °C) for immunostaining and RNA isolation.

Ocular coherence tomography

Anesthetized mice were maintained on a water-based heating pad at 37 °C. Before imaging, lubricant eye drops were instilled on the eyes (Systane Ultra, Alcon), and pupils were dilated using 0.5% tropicamide (Akorn). A lubricant gel (Tears, Alcon) was used to avoid further drying. A Bioptigen Spectral Domain OCT System (Envisu R2300, Durham, NC) was used for image acquisition. OCT imaging and analysis procedures have been described previously [27, 28]. In brief, a rectangular scan with dimensions of 1.8 × 1.8 mm was employed to capture an en-face retinal fundus image centered around the optic nerve head. Each image consisted of 100 B-scan images, with 1536 A-scans for each B-scan. Using a script written in ImageJ software (NIH, Bethesda, MD), masked observers analyzed the retinal layer B-scan images. Retinal layer thickness was quantified from both the left and right locations relative to the center of the optic nerve head. The average value obtained from these locations represented the thickness measurement.

Electroretinography (ERG)

The mice were subjected to overnight dark adaptation before the ERG measurements. Anesthesia was induced under red light using intraperitoneal injection of ketamine/xylazine (100 mg/kg and 10 mg/kg, respectively). Corneal anesthesia was achieved using topical proparacaine hydrochloride. To dilate pupils, 0.5% tropicamide (Akorn) and 2.5% phenylephrine (Paragon BioTeck, Portland, OR) were applied. ERG measurements were conducted using the Celeris system (Diagnosys LLC, Lowell, MA), employing a range of stimulus intensities from − 5.90 to 2.25 log cdm-2. The subdermal needle electrode at the tail served as the ground electrode. The positive scotopic threshold response (pSTR) was elicited using an intensity of -2.50 log cdm-2 to assess RGC function. The pSTR was recorded as the average of 20 repeats with an inter-stimulus interval of 2 s. The amplitude at ∼ 110 ms after stimulus onset was measured and used for analysis.

LXB4 treatment

For scRNA-seq experiments, mice were treated with 1 µg of LXB4 methyl ester (Cayman Chemicals, Ann Arbor, MI) by intraperitoneal injection (IP) once a day and 1 µg of LXB4 methyl ester by topical application 3 times a day for 3 days. Vehicle (ethanol) for LXB4 methyl ester was removed under a stream of nitrogen, and LXB4 methyl ester was resuspended in sterile PBS immediately prior to injection or topical treatment. For RNA-seq experiments, mice were treated with 250 ng of LXB4 methyl ester by IP and 25ng of LXB4 methyl ester by topical application every other day for 2 wks. For morphometric analysis, mice were treated with 1 µg of LXB4 methyl ester by IP and 1 µg of LXB4 methyl ester by topical application once daily for 1 wk. For the sham group, mice were treated with LXB4 equivalent volume of sterile PBS via IP and a topical route. LXB4 treatment was initiated ∼ 15 min before OHT induction.

Quantitative PCR

Total RNA was isolated from retinas using TRIzol extraction method (Invitrogen, Waltham, MA). mRNA was converted to cDNA using an iScript cDNA synthesis kit (Bio-Rad, Hercules, CA). Transcripts for C5ar1, Clec4a2, C3ar1, Ccl5, Tnf-α, Cxcl10, and Cd68 were quantified by using GoTaq PCR master mix (Promega, Madison, WI) in OneStep Plus qPCR (Applied Biosystems, Waltham, MA) system by 2− ΔΔCT method.

Immunostaining

For whole mounts of the retina and optic nerve, the tissues were dissected and fixed in 4% PFA at 4 °C overnight. The next day, tissues were blocked and permeabilized in blocking buffer (10% normal donkey serum + 2% Triton x-100) for 24 h at 4 °C. Further, tissues were incubated with primary antibodies at 1:1000 dilution (anti-Iba1, Cell Signaling (Danvers, MA); anti-CD68, BioLegend (San Diego, CA); anti-RBPMS, Phosphosolutions (Aurora, CO)) for 72 h at 4 oC. Tissues were rinsed three times with washing buffer (PBS + 0.25% Triton x-100) solution, each for 10 min, on a rocker at room temperature. Secondary antibodies were diluted in blocking buffer at 1:2000 dilution (Alexa Fluor 594 and Alexa Fluor 488, Invitrogen), and tissues were incubated with a secondary antibody cocktail overnight at 4 oC. The next day, whole mounts were rinsed three times in a washing buffer and incubated with DAPI (1:5000, Invitrogen) for 10 min at room temperature. Whole mounts were mounted using gold antifade mounting medium (Invitrogen) on a slide in a coverslip grove. For staining of optic nerve sections, the optic nerves were dissected and fixed in 4% PFA at 4 °C overnight. The next day, optic nerves were washed in PBS and dehydrated in a sucrose gradient (10%, 20%, and 30% sucrose) before embedding them in optimal cutting temperature medium (Thermo Fisher, Waltham, MA). 10 μm sections were taken in Leica CM1900 cryostat (Wetzlar, Germany). Sections were washed in PBS, blocked, and permeabilized in blocking buffer (10% normal donkey serum + 0.25% Triton x-100) for 1 h. Sections were then incubated in primary antibodies (anti-CD74, 1:100, BioLegend; anti-Iba1, 1:100; anti-p-PI3K, 1:100, Invitrogen) dissolved in blocking buffer overnight at 4 oC. The next day, sections were washed in PBS three times for 10 min each and were incubated with secondary antibodies (Alexa Fluor 488 and Alexa Fluor 594, 1:200) for 2 h at room temperature. Sections were washed with PBS, incubated with DAPI (1:2500) for 10 min, and mounted using FluorSave™ mounting media (Sigma Aldrich, St. Louis, MO).

Image analysis

RGCs were counted from whole mounts using a macro written in ImageJ, which sets the auto threshold, performs water-shedding and counts RGCs by Analyze Particles(). Two images were acquired from each flank of the retinal whole mount (Supplementary Fig. 2A), and the mean of the 8 image counts represents the count for one retina. After adjusting the threshold, Iba1 and CD74 positive microglia were counted from optic nerve sections using Analyze Particles(). For CD74 and phospho-PI3K co-localization, single cells were cropped from whole optic nerves sections and were colocalized using Coloc2 and BIOP plugins in ImageJ. Pixel intensities were measured using the Measure() function in ImageJ software. Pearson’s correlation coefficient was calculated in RStudio by cor.test().

Confocal microscopy

Images of microglia whole mounts and sections were acquired using a Zeiss LSM710, Axio Imager 2 with Plan-Apochromat 20x objective, 0.8 NA and Plan-Apochromat 63x objective, 1.4 NA. Z-stacks of the retina and optic nerve whole mounts and sections were taken at 1024 × 1024 resolution. For retina, at least 4 different images, one from each retinal whole mount flank (Supplementary Fig. 2A) were taken, and for the optic nerve, at least 3 different images (Supplementary Fig. 2B) were taken.

Microglia feature analysis

The features of microglia morphology were analyzed by the method published by Heindl et al. [29] using custom scripts in MATLAB (R2022a, MathWorks, Natick, MA), which relies on the Image Processing Toolbox and Statistics and Machine Learning Toolbox for its functionalities. The fully automated analysis of morphological features extracted from confocal image stacks of Iba1 stained microglia involved four primary steps. First, the image quality and preprocessing were controlled to ensure reliable results. Second, microglial cells were segmented from the background, and within each cell, further segmentation was performed to distinguish the nucleus, soma, and branches. Third, a skeleton representation is constructed to capture the spatial structure of the cell bodies and branches. Finally, morphological features were extracted using the properties derived from the cell surface area, volume, and skeleton. The output feature file was imported into RStudio v.4.2.0, and the mean values for individual retina and optic nerves were calculated for each replicate. Mean values were plotted using GraphPad Prism 9 software (La Jolla, CA).

MorphOMICs analysisReconstruction of microglia morphology

Microglia morphology was reconstructed using Imaris 9.2.v (Oxford Instruments, Abingdon, UK) software. Briefly, raw z-stack confocal image files were imported into Imaris, and the surface module was used to construct the surfaces on the Iba1 channel by setting smoothing = 1. The new masked channel was created on the surface. Further, microglial processes were analyzed in three dimensions using a filament-tracing plugin. Starting points (soma) for tracing were identified by setting a maximum diameter of 12 μm, and the seeding points (dendrites) were identified by setting a diameter of 1 μm. Following the tracing process, we manually excluded cells located at the image border that were only partially traced, thereby ensuring that these cells were not included in the analysis. However, it is highly challenging to manually remove minor artifact filament generated by Imaris. The .ims files were then converted to .swc files for individual microglia using the ImarisReader toolbox (https://github.com/rcubero/Matlab_Imaris_converter). To count microglia, soma statistics were used after using filament tracer plugin.

MorphOMICs pipeline

Microglia morphology was mapped using MorphOMICs (https://github.com/siegert-lab/MorphOMICs). MorphOMICs uses microglia’s topological morphology descriptor (TMD) combined with bootstrapping and dimension reduction techniques, Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP), to visualize unsupervised clustering of microglia from different treatment conditions. The bootstrap sample size was 100, and the number of bootstraps collected was 400. The first ten principal components were used as input to UMAP with n_neighbors = 50, min_dist = 1.0, and spread = 3.0. Importantly, bootstrapping and PCA reduce the impact of minor artifact filaments introduced by Imaris in the final clustering outcome.

Monocle trajectory analysis

Each bootstrapped sample is represented by an array of 10,000 pixels from its persistence image, therefore a pixel from a sample can be seen as analogous to a gene from a cell within a single-cell transcriptomic setting. Pixels in proximity covary with one another in a manner similar to how genes may covary.

A pseudo-temporal trajectory-inference algorithm called Monocle [30,31,32] was used. Monocle uses a partitioned approximate graph abstraction-like algorithm for Louvain clustering, which learns a principal graph using reversed graph embedding to generate lineages and pseudotimes (https://github.com/cole-trapnell-lab/monocle3/).

The bootstrapped samples from MorphOMICs were used as input for Monocle. The principal components were obtained using preprocess_cds with num_dim = 10, after which UMAP dimension reduction was performed using reduce_dimension with umap.metric = ‘manhattan’, umap.min_dist = 1.0, and umap.n_neighbors = 50. Clusters were determined using cluster_cells with cluster_method = ‘leiden’, and the pseudo-temporal trajectory was obtained using learn_graph with use_partition = FALSE and close_loop = FALSE.

Single-cell transcriptomicsSingle-cell dissociation

Retinas were dissociated into single-cell suspensions using a papain dissociation kit (Worthington, Columbus, OH). Briefly, dissected retinas were placed in papain solution for 30 min at 37 o C with exposure to 5% CO2 in the incubator. Retinas were gently tapped and incubated for 15 more mins for complete dissociation. The reaction was quenched by adding ovomucoid inhibitor solution, and cells were collected after centrifugation and dissolved in resuspension buffer containing PBS + 5%BSA + DNase to form a single-cell solution of cells. Cells were passed through a 40µM cell strainer (Sigma) to remove debris and clumps. Before processing the cells for single-cell transcriptomics, rod cells were depleted to enrich the rest of the population. Cells were incubated with Biotin-CD133 antibody for rod-specific labeling and were depleted in the magnetic column after anti-biotin-magnetic bead labeling (Miltenyi Biotech, Bergisch Gladbach, Germany).

Barcoding and library preparation

Barcoding and library preparation was achieved using a 10X Chromium Single Cell 3’ reagent (v3.1 chemistry, 10x Genomics, Pleasanton, CA), according to the manufacturer’s instructions. Briefly, cells were loaded on the chromium chip to create gel beads containing unique oligo barcodes with single cells, creating cell-gel droplets. RNA from each cell was captured within the droplets, and cell-specific barcodes were linked to their respective transcripts. Reverse transcription and amplification of captured RNA were performed to generate cDNA libraries. Quality of cDNA preparation was checked using Bioanalyzer (Applied Bioscience).

Sequencing

The cDNA libraries were subjected to high-throughput sequencing on the Illumina Novaseq S1 100SR platform (San Diego, CA). Sequencing reads containing cell-specific barcodes and transcript information were obtained for each cell. Raw files were demultiplexed using Illumina bcl2fastq2 software for downstream analysis. Library preparation and scRNA-seq were performed at the QB3 Genomics Core Facility, UC Berkeley, Berkeley, CA, RRID: SCR_022170.

Downstream data analysis

The sequencing reads were processed using Cellranger software (10x Genomics). Cellranger ‘count’ pipeline aligned the sequencing reads to the existing mouse genome (mm10, University of California Santa Cruz). Each read was assigned to its respective cell barcode, enabling cell identification and quantification. The output matrix was used for further downstream analysis and clustering of the cell populations in RStudio v.4.2.0 using Seurat V4.0 [33]. Briefly, quality control and filtering were performed after determining the percentage of mitochondrial transcripts. The data were normalized using the NormalizeData() function, ensuring that gene expression values were comparable across cells. Variable features were selected using the FindVariableFeatures() function to identify the genes that exhibited significant variation and contributed to heterogeneity within the dataset.

The FindIntegrationAnchors() function was executed on both objects to enable comparisons across different datasets. This process identified cell-to-cell pairings between the two datasets based on the first 30 principal components (PCs). The objects were then integrated using the IntegrateData() function on the anchorset and the first 30 PCs. This integration step combined the gene expression profiles from different datasets while preserving the underlying biological variability.

After integration, counts were scaled using the ScaleData() function to normalize the expression values across cells. Principal components (PCs) were computed using the RunPCA() function, capturing the major sources of variation within the integrated dataset. Furthermore, clustering analysis was performed by executing the FindNeighbors() function on the first 30 PCs, identifying the nearest neighbors for each cell based on their gene expression profiles. Subsequently, the FindClusters() function (resolution- 0.8) was used to assign cells to distinct clusters based on their similarity in gene expression. A UMAP was constructed using the RunUMAP() function on the first 30 PCs, allowing for the visualization of the cells in a lower-dimensional space while preserving the global structure of the data.

To assign cell types to the identified clusters, markers specific to each cluster were identified using the FindAllMarkers() function and compared to known cell-type markers [34,35,36], facilitating the annotation of the clusters with specific cell types. Differential gene expression in microglial cells was determined using the FindMarkers() function. Data were visualized using VlnPlot() and DotPlot() functions.

Bulk RNA sequencing

Total RNA was isolated from optic nerves by TRIzol method (Invitrogen) and RNA quality was measured on Bioanalyzer (Applied Bioscience). mRNA was converted to cDNA using SMARTer v4 Ultra Low Input RNA Kit (Clontech, Mountain View, CA). A Diagenode Bioruptor Pico was used to fragment the cDNA, and libraries were generated using the KAPA Hyper Prep Kit for DNA (Roche, Basel, Switzerland) for sequencing on a NOVAseq S4 flow cell (Illumina). Library preparation and RNA-seq were performed at QB3 Genomics, UC Berkeley, Berkeley, CA, RRID: SCR_022170. The raw sequencing reads were demultiplexed by Illumina bcl2fastq2 software, and read quality was assessed using FastQC v.0.11.9. Adapters were subsequently trimmed from the reads using Trim Galore v.0.6.6. The processed reads were then aligned to the mouse genome (mm39, University of California Santa Cruz) using the STAR alignment tool v.2.7.1a and read counts for each gene were obtained using FeatureCount v.1.5.3. The resulting feature count matrix was imported into RStudio v.4.2.0, and normalization and differential gene expression analyses were conducted using the DESeq2 package. Heatmap and venn diagram was generated using pheatmap() and ggvenn() packages in RStudio v.4.2.0.

Additionally, pathway enrichment analysis was performed using the clusterProfiler package, allowing for the identification of biological pathways enriched with differentially expressed genes. The interaction network of genes was created by String DB, and the Highly interacting network of genes was analyzed by MCODE v. 2.0.2 in Cytoscape v.3.9.1 software. Pathway enrichment of the MCODE-generated networks was performed using ClueGO v.2.5.9 in Cytoscape v.3.9.1, using a significance cutoff of p < 0.05.

Statistical analysis

Student’s t-test was used to determine the significance of differences (p < 0.05) between the two groups. One-way analysis of variance with post-hoc Tukey’s multiple comparison tests was used to compare multiple groups. Values are presented as mean ± SEM (standard error of the mean).