CD36-mediated lipid-accumulation in reactive microglia contributes to retinal degeneration via the NLRP3-IL-1β pathway in mice Authors: Tian Zhou, Ziqi Yang, Hong Zhou, Biyan Ni, Yang Zhou, Jingpeng Li, Minglu Ma, Huaicheng Wang, Peng An, Huiyi Xu, Xiaojing Lin, Shiya Lin, Shasha Wang, Shida Chen, Lixia Lin, Xialin Liu, Chang He Published: 2026 ABSTRACT Microglia are central regulators of retinal immune homeostasis, yet their pathogenic states in retinal degeneration remain less well understood. Here we identified a distinct subset of lipid-accumulated reactive microglia (aLARM), using scRNA sequencing and spatial transcriptomics in NaIO3-induced retinal degeneration mice, predominantly localized to the outer retina. aLARM were conserved across mouse models and patients and uniquely marked by high CD36 expression. Microglia-specific CD36 deletion abolished aLARM-mediated inflammation and degeneration, whereas subretinal transplantation of CD36+ aLARM exacerbated retinal structural destruction and functional impairment. Mechanistically, CD36+ aLARM activated NLRP3 inflammasome and produced IL-1β, engaging IL-1R1 on microglia/macrophages and pericytes/SMCs to amplify a feed-forward inflammatory circuit. Therapeutic CD36 blockade with the neutralizing antibody FA6-152 reduced aLARM formation and protected against neurodegeneration. Together, our findings highlight CD36+ aLARM as a targetable pathogenic microglial population linking neuroinflammation to retinal degeneration, providing a potential foundation for microglia-based precision therapies. Introduction Retinal degenerative diseases, including age-related macular degeneration (AMD), are the main cause of irreversible blindness worldwide. Despite extensive research, effective neuroprotective therapies remain lacking1. Increasing evidence points to microglia, the resident immune cells of the retina, as critical regulators of retinal health and disease2. Under homeostatic conditions, microglia help maintain neuronal integrity and immune balance; however, under stress or injury, they can adopt reactive phenotypes that contribute to neuroinflammation, photoreceptor loss, and retinal pigment epithelium (RPE) dysfunction3. Recent advances in single-cell technologies have revealed the remarkable heterogeneity of microglia4–6, yet the specific pathogenic subpopulations that drive retinal degeneration and the molecular mechanisms underlying their reprogramming remain undefined. Identifying and selectively targeting these disease-driving microglial populations in retinal degenerative disease holds promise for future therapeutic strategies to halt or reverse vision loss. A hallmark of dry AMD is the accumulation of drusen, lipid-rich extracellular deposits, beneath the RPE, indicating that dysregulated lipid metabolism plays a central role in disease pathogenesis7. The retina, one of the most lipid-rich tissues in the central nervous system, relies on tightly regulated lipid homeostasis to maintain visual function8. Within the retinal environment, microglia serve as key mediators of lipid sensing, uptake, and clearance through receptors such as Lpl, Apoe, and Trem29–12. However, in pathological conditions, microglia can become engorged with lipids, leading to neuroinflammation in diseases like multiple sclerosis (MS)13. While similar mechanisms are suspected in retinal disease, the precise link between microglial lipid overload, dysfunctional phenotypic transitions, and retinal degeneration remain poorly defined. In our study, we identified a subset of lipid-laden microglia in the diseased retina. Elucidating the mechanisms underlying their accumulation and functional shifts is essential for understanding disease progression and for the development of microglia-targeted lipid-modulating therapies. Among lipid-handling receptors, CD36 stands out as a multifunctional scavenger receptor implicated in fatty acid uptake, oxidized lipoprotein recognition, and innate immune activation14,15. CD36 has been associated with diverse neurological diseases, including stroke, Alzheimer’s disease (AD), Parkinson’s disease (PD), MS, etc, although its role appears to be highly context-dependent16–19. In atherosclerosis, CD36 facilitates foam cell formation and drives inflammation, disorders20, whereas in AD, it has been linked to both Aβ clearance and inflammatory amplification17. In the context of MS, PPARγ activation upregulates CD36 expression in microglia, thereby enhancing their phagocytic ability to clear myelin debris and reduce neuroinflammatory responses19. Despite its established role in lipid metabolism, the function of CD36 in microglial lipid handling within the degenerative retina remains unclear. Here, we identify a previously unrecognized CD36⁺ lipid-accumulated reactive microglial subset, characterized by enriched lipid-associated transcriptional programs and operationally designated as aLARM (a subset of Lipid-Accumulated Reactive Microglia), in mouse models and human retinal degeneration. These cells exhibit enriched expression of genes involved in lipid metabolism and inflammatory signaling, and spatially localize to the inferior retina, a region susceptible to degeneration. Mechanistically, we demonstrate that CD36 is required for aLARM formation and that CD36⁺ aLARM drive retinal degeneration via activation of the NLRP3–IL-1β signaling axis. Therapeutic blockade of CD36 using a neutralizing antibody attenuates microglial activation, preserves photoreceptor and RPE integrity, and mitigates retinal degeneration. Our study identifies CD36⁺ aLARM as a disease-driving microglial population that links lipid dysregulation to retinal neuroinflammation, offering CD36-targeted modulation of this subset as a promising avenue for microglia-based therapies against retinal degenerative diseases. Results Single-cell RNA-seq and spatial transcriptomics identified a lipid-accumulated reactive microglia subset (aLARM) enriched in the inferior retina during degeneration in mice To identify the repertoire of microglia participating in lipid metabolism during retinal degeneration, we performed single-cell RNA analysis of retinal microglia using the 10X Genomics platform (Fig. S1A). The CD11b-positive cells were enriched by magnetic bead selection from pooled retinae of control or NaIO3- induced retinal degeneration mice, which recapitulated the pathological process of clinical dry AMD. The retinal microglia/macrophages gave rise to six distinct clusters (Fig. 1A). In the control retinae, most microglia formed a well-defined homeostatic MG0 cluster21. In contrast, this MG0 cluster was significantly decreased in the NaIO3-degenerative retina, where a predominant MG4 cluster emerged (Fig. 1A). MG1 was enriched for Gabbr1, Dscaml1 and Ckmt, indicating roles in synaptic pruning and refinement22. MG2 exhibited a proliferative signature, characterized by Ube2c, Birc5, and Top2a23. MG3 was distinguished by elevated expression of ribosomal mRNAs including Rps24, Rps29 and Rpl1024 (Fig. S1B). Notably, MG4 cluster was characterized by transcriptional programs linked to inflammation and lipid metabolism, including lipid binding, storage, localization, and metabolic process (Fig. 1B). To refine this signature, we intersected genes enriched across lipid-related GO categories and retained those present in at least two terms, thereby identifying a core lipid-associated gene set (Fig. S1C). These genes were predominantly enriched in the MG4 cluster (Fig. 1C), supporting the existence of a distinct lipid-associated microglial subset. Furthermore, differential gene expression analysis revealed that this previously unrecognized MG4 shared substantial transcriptomic similarity with neurodegeneration-associated microglial subtypes, including DAM25 and MGnD5, yet showed limited overlap (11 shared lipid droplet-associated genes) with brain LDAM26. Importantly, MG4 exhibits distinct expression of 529 genes not observed in these known subtypes, highlighting its unique transcriptional signature and supporting its identity as a specialized and previously uncharacterized microglial subgroup (Fig. 1D).Fig. 1Single-Cell and spatial transcriptomic reveal a subset of lipid accumulated microglia in the NaIO3-induced retinal degeneration.A t-SNE plot showing the unsupervised clustering of retinal microglia/macrophages into six distinct subclusters. Bar plot depicts cluster distribution in PBS and NaIO3 groups (n = 6 eyes from 6 mice). B Dot plot of Gene Ontology (GO) enrichment for differentially expressed genes (DEGs) in MG4 versus total microglia from PBS retinae. Dot size represents gene count and color indicates statistical significance. C Dot plot of representative lipid metabolism related genes enriched in over two lipid-associated GO terms. D Venn diagram showing the overlap between MG4 DEGs and previously reported gene signatures of disease-associated microglia (DAM), microglia with neurodegenerative phenotype (MGnD), and lipid droplet-accumulating microglia (LDAM). E Spatial projection of single-cell–defined MG4 signatures onto Stereo-seq retinal sections. F Heatmaps showing spatial expression of selected lipid metabolism-related genes in retinal sections from PBS and NaIO3 groups. G Schematic of retinal sections segmented by distance from the optic nerve into center, mid-periphery, periphery zones, each subdivided into four locations (I–IV). H Dot plot showing representative lipid metabolism-related gene expression across locations I-IV in each retinal zone in both groups. I H&E staining showing subretinal cells with enlarged, foamy cytoplasm containing multiple clear vacuoles (red arrows, Scale bar: 20 μm). Transmission electron microscopy (TEM) images of NaIO₃-treated retinae showing microglial outlined in yellow and intracellular lipid droplets highlighted in red. Scale bar: 1μm, n = 6 eyes from 6 mice (J) Representative OCT image showing hyperreflective spots beneath the ONL layer in the NaIO3-induced retina (yellow arrow), and corresponding immunofluorescent images of retinal sections stained by lipid markers (Plin3 and BODIPY) in Iba1+ microglia. Quantification of fluorescence intensity is shown at right. Scale bar: 20 μm. K Immunofluorescence images of RPE-choroidal flat mounts showing Plin3 and BODIPY accumulation in Iba1+ microglia. Corresponding fluorescence intensity quantification is shown. Scale bar: 10 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM or mean with range, ***P < 0.001, two-sided unpaired Student’s t test and Mann–Whitney test in (I). Single-Cell and spatial transcriptomic reveal a subset of lipid accumulated microglia in the NaIO3-induced retinal degeneration. A t-SNE plot showing the unsupervised clustering of retinal microglia/macrophages into six distinct subclusters. Bar plot depicts cluster distribution in PBS and NaIO3 groups (n = 6 eyes from 6 mice). B Dot plot of Gene Ontology (GO) enrichment for differentially expressed genes (DEGs) in MG4 versus total microglia from PBS retinae. Dot size represents gene count and color indicates statistical significance. C Dot plot of representative lipid metabolism related genes enriched in over two lipid-associated GO terms. D Venn diagram showing the overlap between MG4 DEGs and previously reported gene signatures of disease-associated microglia (DAM), microglia with neurodegenerative phenotype (MGnD), and lipid droplet-accumulating microglia (LDAM). E Spatial projection of single-cell–defined MG4 signatures onto Stereo-seq retinal sections. F Heatmaps showing spatial expression of selected lipid metabolism-related genes in retinal sections from PBS and NaIO3 groups. G Schematic of retinal sections segmented by distance from the optic nerve into center, mid-periphery, periphery zones, each subdivided into four locations (I–IV). H Dot plot showing representative lipid metabolism-related gene expression across locations I-IV in each retinal zone in both groups. I H&E staining showing subretinal cells with enlarged, foamy cytoplasm containing multiple clear vacuoles (red arrows, Scale bar: 20 μm). Transmission electron microscopy (TEM) images of NaIO₃-treated retinae showing microglial outlined in yellow and intracellular lipid droplets highlighted in red. Scale bar: 1μm, n = 6 eyes from 6 mice (J) Representative OCT image showing hyperreflective spots beneath the ONL layer in the NaIO3-induced retina (yellow arrow), and corresponding immunofluorescent images of retinal sections stained by lipid markers (Plin3 and BODIPY) in Iba1+ microglia. Quantification of fluorescence intensity is shown at right. Scale bar: 20 μm. K Immunofluorescence images of RPE-choroidal flat mounts showing Plin3 and BODIPY accumulation in Iba1+ microglia. Corresponding fluorescence intensity quantification is shown. Scale bar: 10 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM or mean with range, ***P < 0.001, two-sided unpaired Student’s t test and Mann–Whitney test in (I). To spatially contextualize our findings, we performed high-resolution Stereo-seq spatial transcriptomics on sagittal retinal sections (Fig. S1A). Integration with single-cell data confirmed that the MG4 were predominantly localized to the ONL and RPE regions, the areas severely affected by NaIO3-induced oxidative injury, which leads to RPE ablation and subsequent photoreceptor loss via accumulation of reactive oxygen species (ROS) (Fig. 1E)27. Moreover, Stereo-seq mapping of scRNA-seq-defined MG4 signatures were significantly enriched in the NaIO₃-treated group and spatially mapped to degenerative retinal regions, further supporting the pathological relevance of MG4 (Fig. 1F). To systematically characterize their spatial distribution, the retina was anatomically stratified into central, mid-peripheral, and peripheral zones based on the distance from the optic nerve head. Each zone was further partitioned along the dorsal-ventral and inner-outer retinal axes, generating four spatial compartments (locations I-IV) (Fig. 1G). Intersection analysis of MG4 differentially expressed genes identified from spatial and single-cell datasets was subsequently visualized across these compartments using dot-plot mapping. This analysis revealed a marked accumulation of MG4 signatures within the outer retinal regions (locations I and II) across central, mid-peripheral, and peripheral zones (Fig. 1H), indicating a preferential localization of MG4 microglia to sites of active neurodegeneration. Particularly, genes associated with lipid metabolic reprogramming, including Dgat1, Plin2, and Cd36, were prominently enriched in location I within the central retina. Notably, Cd36 exhibited the pronounced upregulation in the mid-peripheral outer retina following NaIO₃ injury (Fig. 1H). Next, the histological analysis combined with electron microscopy further exhibited the microglia with lipid deposition are present around the clustered lesions of RPE damage (Fig. 1I). These lipid-loaded microglia were identified by OCT examination and immunofluorescence staining in frozen retinal sections. As expected, the Iba1+ microglia in the NaIO3-retina was co-stained with BODIPY, a dye that specifically labels neutral lipids. Furthermore, these cells presented high expression of Plin3 and Plin2, proteins involved in the formation and maturation of lipid droplets, indicating that these lipid droplets were particularly abundant in microglia from the NaIO3-retina (Fig. 1J–K and Fig. S1D, E). These results support the idea that a specific subset of lipid-accumulated reactive microglia (aLARM), characterized by CD36 expression, lipid-associated transcriptional programs, distinct from brain LDAMs, may be involved in the degenerative retina. CD36⁺ aLARM was detected in the degenerative retina of patients with AMD Optical coherence tomography angiography (OCTA) en-face imaging of AMD patients revealed the accumulation of hyper-reflective spots in the subretinal space of macular region, a feature previously associated with activated microglia and lipid-rich deposits such as precursors of hard exudates or lipoprotein28, indicating their involvement in retinal degeneration (Fig. 2A). To investigate the role of microglia in lipid metabolism during retinal degeneration, we reanalyzed the single-cell RNA-seq data from human RPE-choroidal-complex obtained from healthy individuals and patients with macular degeneration (GSE135922)29. This analysis identified five distinct microglial clusters and a single macrophage cluster (Fig. 2B). Among these, MG4 exhibited the most prominent transcriptional reprogramming and was enriched for pathways related to cellular response to lipid and inflammatory response (Fig. 2C). Consistently, GSEA revealed preferential activation of the “hallmark of fatty acid metabolism” signature in MG4 (Fig. 2D). Furthermore, most genes defined from mouse MG4 were also significantly upregulated in the human MG4 subset (Fig. 2E). CD36 expression was markedly elevated in the macular region of AMD patients in comparison to other lipid metabolism-related genes (Fig. 2F, and S1F). Immunostaining of human retinal sections revealed CD36 co-localization with Iba1⁺ microglia beneath the ONL layer in AMD patients, while microglia are physiologically confined to the inner retinal layers in healthy donors with a resting shape (Fig. 2G). These CD36⁺ microglia also co-stained with the lipid marker BODIPY, Plin2, as well as Lrp1, indicating their activated and lipid-laden state (Fig. S1G-H). Together, these findings suggest that elevated CD36 expression is a defining feature of aLARM and present in both mouse and human AMD.Fig. 2CD36⁺ lipid-accumulated reactive microglia (aLARM) are present in the degenerative retina of AMD patients.A En-face OCTA scans of AMD patients showing hyper-reflective spots (orange arrows) in the subretinal space between the ONL and RPE, consistent with macrophage-like cells (MLCs) or lipid-rich deposits. Bar plot quantifying the number of MLCs in the macular region of healthy controls (n = 10 eyes from 10 individuals) and AMD patients (n = 10 eyes from 10 individuals). B t-SNE plot of microglia /macrophages from human RPE–choroid single-cell RNA-seq datasets (GSE135922), identifying five transcriptionally distinct microglial clusters and one macrophage cluster across healthy donors and AMD patients. The MG4 cluster exhibited the largest set of cluster-specific marker genes. n = 7 donors (4 male and 3 female). C Gene Ontology (GO) enrichment analysis of differentially expressed genes in MG4, showing enrichment of lipid-responsive and inflammatory pathways. D Gene Set Enrichment Analysis (GSEA) analysis revealing significant enrichment of the “Hallmark: Fatty Acid Metabolism” signature in MG4. E Dot plot displaying the expression of representative lipid metabolism–related genes in MG4. F Ridge plot illustrating CD36 expression in macular versus peripheral regions of AMD patients and healthy controls. G Immunofluorescence staining of retinal section from AMD patients showing CD36⁺Iba1⁺ microglia co-labeled with the lipid marker BODIPY beneath the ONL layer, while healthy control showing the resting microglia located in OPL layer. Quantification of fluorescence intensity is shown. Scale bar: 20 μm. Data are shown as mean ± SEM, ***P < 0.001, two-sided unpaired Student’s t test in (A). CD36⁺ lipid-accumulated reactive microglia (aLARM) are present in the degenerative retina of AMD patients. A En-face OCTA scans of AMD patients showing hyper-reflective spots (orange arrows) in the subretinal space between the ONL and RPE, consistent with macrophage-like cells (MLCs) or lipid-rich deposits. Bar plot quantifying the number of MLCs in the macular region of healthy controls (n = 10 eyes from 10 individuals) and AMD patients (n = 10 eyes from 10 individuals). B t-SNE plot of microglia /macrophages from human RPE–choroid single-cell RNA-seq datasets (GSE135922), identifying five transcriptionally distinct microglial clusters and one macrophage cluster across healthy donors and AMD patients. The MG4 cluster exhibited the largest set of cluster-specific marker genes. n = 7 donors (4 male and 3 female). C Gene Ontology (GO) enrichment analysis of differentially expressed genes in MG4, showing enrichment of lipid-responsive and inflammatory pathways. D Gene Set Enrichment Analysis (GSEA) analysis revealing significant enrichment of the “Hallmark: Fatty Acid Metabolism” signature in MG4. E Dot plot displaying the expression of representative lipid metabolism–related genes in MG4. F Ridge plot illustrating CD36 expression in macular versus peripheral regions of AMD patients and healthy controls. G Immunofluorescence staining of retinal section from AMD patients showing CD36⁺Iba1⁺ microglia co-labeled with the lipid marker BODIPY beneath the ONL layer, while healthy control showing the resting microglia located in OPL layer. Quantification of fluorescence intensity is shown. Scale bar: 20 μm. Data are shown as mean ± SEM, ***P < 0.001, two-sided unpaired Student’s t test in (A). CD36 was essential for aLARM formation and their pathogenic role in promoting retinal degeneration in vivo To investigate the role of CD36 in aLARM formation and retinal degeneration, we first assessed its expression in mouse models, both mRNA and protein levels of CD36 were significantly upregulated following NaIO₃-induced retinal injury (Fig. 3A, B). Immunostaining revealed that CD36 was enriched in the outer layer of NaIO3-retina (Fig. S2A), and co-stained with the lipid-droplet marker BODIPY in microglia (Fig. 3C), suggesting its involvement in lipid accumulation. Single cell sequencing also showed that CD36 expression was predominantly enriched in microglia, particularly under degenerative condition (Fig. S2B, C). Although low-level expression was detected in RPE cells and photoreceptors, consistent with their lipid-rich nature, its abundance was substantially lower than in microglia (Fig. S2D, E). Minimal expression was observed in other retinal cell types, including Müller cells and RGCs (Fig. S2F, G).Fig. 3CD36 is required for aLARM formation and contributes to retinal degeneration.A Quantitative RT-PCR analysis of CD36 mRNA expression in retinas treated with NaIO₃ versus PBS controls (n = 6 retinas from 6 mice). B Western blot validation of CD36 protein upregulation in NaIO₃-treated retinas (n = 4 retinas from 4 mice). C Immunofluorescence staining of NaIO₃-treated retinas showing co-localization of CD36, BODIPY (lipid droplets), and Iba1 (microglia) in the subretinal region. Quantification of fluorescence intensity is shown. Scale bar = 10 μm. D Representative OCT images of CD36fl/fl and microglia-specific CD36 conditional knockout (CD36 cKO) mice after NaIO₃ induction. Orange arrows indicate hyper-reflective subretinal foci. Outer nuclear layer (ONL) thickness was quantified (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO). E Retinal fundus image showing scattered dot-like and linear white lesions in NaIO₃-treated CD36fl/fl mice, which were diminished in CD36 cKO mice. Quantification of retinal structural damage is shown (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO). F, G Retinal function examinations by ERG showing a-wave and b-wave amplitudes (F) and visual acuity (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO) and contrast sensitivity (n = 5 mice for CD36fl/fl; n = 3 mice for CD36 cKO) (G) in NaIO₃- treated CD36fl/fl and CD36 cKO mice. H Immunofluorescence staining of retinal sections from CD36fl/fl and CD36 cKO mice showing BODIPY⁺Iba1⁺ microglia beneath the ONL. Quantification of fluorescence intensity is shown. Scale bar: 20 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, ***P < 0.001, two-sided unpaired Student’s t test in A, B, F and Mann–Whitney test in (E, G). CD36 is required for aLARM formation and contributes to retinal degeneration. A Quantitative RT-PCR analysis of CD36 mRNA expression in retinas treated with NaIO₃ versus PBS controls (n = 6 retinas from 6 mice). B Western blot validation of CD36 protein upregulation in NaIO₃-treated retinas (n = 4 retinas from 4 mice). C Immunofluorescence staining of NaIO₃-treated retinas showing co-localization of CD36, BODIPY (lipid droplets), and Iba1 (microglia) in the subretinal region. Quantification of fluorescence intensity is shown. Scale bar = 10 μm. D Representative OCT images of CD36fl/fl and microglia-specific CD36 conditional knockout (CD36 cKO) mice after NaIO₃ induction. Orange arrows indicate hyper-reflective subretinal foci. Outer nuclear layer (ONL) thickness was quantified (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO). E Retinal fundus image showing scattered dot-like and linear white lesions in NaIO₃-treated CD36fl/fl mice, which were diminished in CD36 cKO mice. Quantification of retinal structural damage is shown (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO). F, G Retinal function examinations by ERG showing a-wave and b-wave amplitudes (F) and visual acuity (n = 10 eyes from 5 mice for CD36fl/fl; n = 6 eyes from 3 mice for CD36 cKO) and contrast sensitivity (n = 5 mice for CD36fl/fl; n = 3 mice for CD36 cKO) (G) in NaIO₃- treated CD36fl/fl and CD36 cKO mice. H Immunofluorescence staining of retinal sections from CD36fl/fl and CD36 cKO mice showing BODIPY⁺Iba1⁺ microglia beneath the ONL. Quantification of fluorescence intensity is shown. Scale bar: 20 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, ***P < 0.001, two-sided unpaired Student’s t test in A, B, F and Mann–Whitney test in (E, G). Next, we generated microglia-specific CD36 conditional knockout mice (CD36fl/fl; Cx3cr1-Cre mice, referred to as CD36 cKO mice, Fig. S3A). Under naïve conditions, age-matched CD36 cKO mice showed comparable retinal morphology and visual function to CD36fl/fl controls, as determined by ERG, OCT, and optomotor testing, suggesting that microglia-specific CD36 deletion does not alter baseline retinal homeostasis (Fig. S3B–E). Under NaIO3 stimulation, activated microglia presented CD36 expression in CD36 fl/fl controls, while CD36 expression were deficient in CD36 cKO mice (Fig. S3F). Representative OCT images showed less prominent hyperreflective foci together with better preserved ONL thickness in CD36 cKO mice compared to CD36 fl/fl controls (Fig. 3D), and funduscopy showed fewer white dots or linear lesions (Fig. 3E). The ERG examination further demonstrated significantly improved retinal function in the CD36 cKO mice, with elevated amplitudes of both a- and b-waves, improved visual acuity and contrast sensitivity (Fig. 3F, G), and decreased accumulation of lipid rich BODIPY+ microglia (Fig. 3H), indicating that CD36 deletion in microglia mitigates retinal degeneration. To further validate the pathogenic role of CD36⁺ aLARM, we performed a gain-of-function experiment by isolating CD36+CD11b+CD45int cells from the NaIO3-induced retina and transplanting them into the subretinal space of recipient mice subjected to a subthreshold dose of NaIO₃ (10 mg/kg). CD36-CD11b+CD45int cells served as controls. After transplantation, CD36+Cx3cr1-GFP+ microglia were found accumulating between the ONL and RPE layer (Fig. S4A, B). Retinal fundus images showed more prominent pale retinas and whiter dot or linear lesions in recipients of CD36⁺ aLARM compared to CD36⁻ or PBS-injected controls (Fig. 4A). Representative OCT images showed prominent hyperreflective signals (red arrows) accompanied by reduced ONL thickness in the CD36⁺ group (Fig. 4B), while H&E staining revealed significant ONL thinning accompanied by many crumby-structure patches (yellow arrows), suggestive of pathologic melanin aggregation in the RPE cells (Fig. 4C). ZO-1 immunostaining confirmed severe barrier breakdown, with disrupted hexagonal RPE tight junctions in the CD36⁺ group (Fig. 4D). Functionally, ERG analysis demonstrated significantly reduced a- and b-wave amplitudes in eyes treated with CD36⁺ aLARM, indicating profound neuroretinal dysfunction (Fig. 4E). To characterize the metabolic features associated with CD36 expression, we performed GSEA on flow-sorted retinal CD36⁺ and CD36⁻ microglia. CD36⁺ microglia were enriched for pathways related to fatty acid-responsive transcriptional programs, including lipoprotein assembly, remodeling and PPAR signaling, whereas CD36⁻ microglia preferentially expressed gene programs related to lipid clearance and other metabolic processing (Fig. 4F, and Fig. S5A). Consistently, single-cell transcriptomic analyses in both human and mouse retinal datasets and in vitro validation revealed preferential enrichment of PPAR-associated gene signatures within the aLARM subset (Fig. 4G, and Fig. S5B–E). Together, these findings indicate that CD36⁺ aLARM adopt a lipid metabolic state biased toward extracellular lipoprotein handling rather than intrinsic lipid synthesis or clearance.Fig. 4Adoptive transfer of CD36+ aLARM exacerbates retinal degenerative in vivo.A Representative retinal fundus images showing the dot-like or linear lesions in the NaIO₃-induced degeneration model following subretinal injection of PBS, CD36- (CD36-CD11b+CD45int) cells, or CD36+ (CD36+CD11b+CD45int) cells. Lesion severity was quantified (n = 8 eyes from 4 mice). B Representative OCT scanning showed retinal thinning and subretinal hyper-reflective spots (red arrow) following subretinal injection. Outer nuclear layer (ONL) thickness was quantified (n = 8 eyes from 4 mice). C H&E staining of retinal sections showing retinal structural damage, including ONL thinning, disorganization (yellow arrows), and disrupted RPE with cellular debris. Retinal folds were quantified (n = 7 eyes from 7 mice). Scale bar: 50 μm. D Whole-mount staining of RPE cells with ZO-1 (green) showing disruption of RPE morphology and loss of regular hexagonal cell structures in CD36+ group. Hexagonal RPE cells were quantified (n = 7 eyes from 7 mice). Scale bar: 50 μm. E ERG recordings showing the aptitudes of a and b waves after subretinal injection of CD36- or CD36+ cells in NaIO₃ model (n = 8 eyes from 8 mice). F Gene Set Enrichment Analysis (GSEA) of flow-sorted retinal CD36⁺ and CD36⁻ microglia showing pathway enrichment profiles, including Lipoprotein assembly, fatty acid ligand-activated PPAR signaling. G In vitro treatment of HMC3 human microglial cells with a PPAR agonist followed by assessment of PPARγ and CD36 expression and oxLDL uptake (n = 3 independent cultures). H Immunofluorescence staining of HMC3 cells after oxLDL and NaIO₃ stimulation showing CD36 distribution relative to the plasma membrane marker wheat germ agglutinin (WGA). Six images were collected randomly in each culture and the experiments were repeated at least three times. Scale bar: 10μm. Representative images and fluorescence intensity quantification are shown. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are presented as mean ± SEM. *P < 0.05, ns not significant. Kruskal–Wallis test with Dunn’s post hoc test for (A, C) one-way ANOVA with Tukey’s post-hoc test for (B) and two-sided unpaired Student’s t-test for (D, G). Adoptive transfer of CD36+ aLARM exacerbates retinal degenerative in vivo. A Representative retinal fundus images showing the dot-like or linear lesions in the NaIO₃-induced degeneration model following subretinal injection of PBS, CD36- (CD36-CD11b+CD45int) cells, or CD36+ (CD36+CD11b+CD45int) cells. Lesion severity was quantified (n = 8 eyes from 4 mice). B Representative OCT scanning showed retinal thinning and subretinal hyper-reflective spots (red arrow) following subretinal injection. Outer nuclear layer (ONL) thickness was quantified (n = 8 eyes from 4 mice). C H&E staining of retinal sections showing retinal structural damage, including ONL thinning, disorganization (yellow arrows), and disrupted RPE with cellular debris. Retinal folds were quantified (n = 7 eyes from 7 mice). Scale bar: 50 μm. D Whole-mount staining of RPE cells with ZO-1 (green) showing disruption of RPE morphology and loss of regular hexagonal cell structures in CD36+ group. Hexagonal RPE cells were quantified (n = 7 eyes from 7 mice). Scale bar: 50 μm. E ERG recordings showing the aptitudes of a and b waves after subretinal injection of CD36- or CD36+ cells in NaIO₃ model (n = 8 eyes from 8 mice). F Gene Set Enrichment Analysis (GSEA) of flow-sorted retinal CD36⁺ and CD36⁻ microglia showing pathway enrichment profiles, including Lipoprotein assembly, fatty acid ligand-activated PPAR signaling. G In vitro treatment of HMC3 human microglial cells with a PPAR agonist followed by assessment of PPARγ and CD36 expression and oxLDL uptake (n = 3 independent cultures). H Immunofluorescence staining of HMC3 cells after oxLDL and NaIO₃ stimulation showing CD36 distribution relative to the plasma membrane marker wheat germ agglutinin (WGA). Six images were collected randomly in each culture and the experiments were repeated at least three times. Scale bar: 10μm. Representative images and fluorescence intensity quantification are shown. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are presented as mean ± SEM. *P < 0.05, ns not significant. Kruskal–Wallis test with Dunn’s post hoc test for (A, C) one-way ANOVA with Tukey’s post-hoc test for (B) and two-sided unpaired Student’s t-test for (D, G). Functionally, cultured microglia robustly internalized Dil-labeled oxLDL in a dose-dependent manner, which was accompanied by a concomitant upregulation of CD36 expression (Fig. S5F). This enhanced lipid uptake was associated with progressively increased inflammatory activation at higher oxLDL concentrations (Fig. S5H). Notably, under NaIO₃ stimulation, internalized oxLDL exhibited prominent co-localization with BODIPY-positive neutral lipid droplets, indicating exacerbated intracellular lipid accumulation within activated microglia (Fig. S5G). Given the dynamic trafficking properties of CD36, we next examined its subcellular distribution. Under basal conditions, CD36 exhibited sparse and predominantly perinuclear localization. In contrast, oxLDL plus NaIO₃ treatment increased membrane-associated CD36 and enhanced early endocytic markers (Rab5a, EEA1, and Caveolin), without changes in the late endosomal marker Rab7 (Fig. 4H, S5I). This pattern indicates enhanced membrane recruitment and early endocytic trafficking of CD36 during lipid loading. Together, these findings demonstrate that CD36⁺ aLARM exhibit enhanced oxLDL uptake and membrane-associated CD36 trafficking, consistent with a lipid-laden, pro-inflammatory microglial phenotype associated with the progression of retinal degeneration. CD36+ aLARM induced neuroinflammation via activation of the NLRP3-IL-1β pathway To delineate the molecular mechanisms underlying CD36⁺ aLARM-mediated retinal degeneration, we performed Smart-seq2-based RNA sequencing on CD11b+CD45int cells isolated from the retinas of NaIO3 induced CD36fl/fl and CD36 cKO mice (Fig. 5A). The Gene Ontology analysis of upregulated genes in CD36fl/fl mice revealed significant enrichment in pathways related to cell cycle control, regulation of IL-1β production, double-strand break and DNA damage, etc (Fig. 5B). PPI network analysis further identified 11 hub genes directly connected to CD36, with NLRP3 emerging as the most frequently represented gene across GO terms (Fig. 5C). Consistent with this, immunofluorescence staining of retinal flat mounts showed markedly reduced expression of NLRP3 and IL-1β in CD36 cKO mice following NaIO₃ injury (Fig. 5D). Western blotting of the RPE-sclera complex confirmed that both proteins were significantly downregulated in CD36 cKO mice compared to CD36fl/fl controls (Fig. 5E). In a gain-of-function model, subretinal delivery of CD36⁺ microglia isolated from NaIO₃-treated WT mice resulted in higher NLRP3 and IL-1β expression in the subretinal space compared with CD36− microglia (Fig. 5F, and Fig. S6A).Fig. 5CD36+ aLARM promotes retinal degeneration via NLRP3-IL-1β pathway.A Experimental Schematic. CD11b+CD45int microglia cells were isolated by FACS from NaIO₃ induced CD36fl/fl or CD36 cKO mice and subjected to Smart-seq2 analysis (n = 3 retinae from 3 mice). Mouse and cell-sorting icons were created in BioRender: Yang, Z. (https://BioRender.com/8w5m9vq). B Dot plot showing the GO pathways enriched in NaIO₃-induced controls versus CD36 cKO microglia. C Frequency of DEGs directly connected to CD36 across enriched GO terms. D Immunofluorescence showing NLRP3, IL-1β coexpression in Iba1+ subretinal microglia after NaIO₃ induction with fluorescence quantification. E Western blotting of RPE-sclera complexes from untreated CD36fl/fl and CD36 cKO mice, NaIO₃ induced CD36fl/fl and CD36 cKO mice. n = 3 eyes from 3 mice. F Retinal sections following subretinal injection of CD36⁻ or CD36⁺ microglia into NaIO₃-injured mice stained for NLRP3, BODIPY, and Iba1 with fluorescence quantification. (G) Experimental design for subretinal transfer of CD36⁺ microglia from wild-type or NLRP3−/− mice into half-NaIO₃-injured recipients at D0 and D4. Recipients received PLX5622-containing chow for 14 days or control diet. H, I visual acuity (H) and ERG a- and b-wave amplitudes (I) after transplantation of CD36+ microglia from NaIO₃-induced WT or NLRP3−/− donors into PLX5622-treated or control recipients (n = 8 eyes from 4 mice). J, K Fundus images (J) and OCT scanns (K) showing retinal structural alterations following transplantation (n = 8 eyes from 4 mice). L H&E staining showing retinal folds and subretinal cell accumulation in mice receiving CD36+ microglia from NaIO₃-treated WT or NLRP3−/− donors (n = 4 eyes from 4 mice). M RPE-scleral flat mount stained for ZO-1, NLRP3 and Iba1 with quantification of hexagonal ZO-1–positive cells and fluorescence intensity (n = 6 eyes from 6 mice). N Western blotting of IL-1β expression (n = 3 eyes from 3 mice). Scale bar: 50 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, n.s not significant, one-way ANOVA with Tukey’s post-hoc test for E, K, Kruskal–Wallis test with Dunn’s post hoc test for H, J, Mann–Whitney test in L and two-sided unpaired Student’s t test for (M, N). CD36+ aLARM promotes retinal degeneration via NLRP3-IL-1β pathway. A Experimental Schematic. CD11b+CD45int microglia cells were isolated by FACS from NaIO₃ induced CD36fl/fl or CD36 cKO mice and subjected to Smart-seq2 analysis (n = 3 retinae from 3 mice). Mouse and cell-sorting icons were created in BioRender: Yang, Z. (https://BioRender.com/8w5m9vq). B Dot plot showing the GO pathways enriched in NaIO₃-induced controls versus CD36 cKO microglia. C Frequency of DEGs directly connected to CD36 across enriched GO terms. D Immunofluorescence showing NLRP3, IL-1β coexpression in Iba1+ subretinal microglia after NaIO₃ induction with fluorescence quantification. E Western blotting of RPE-sclera complexes from untreated CD36fl/fl and CD36 cKO mice, NaIO₃ induced CD36fl/fl and CD36 cKO mice. n = 3 eyes from 3 mice. F Retinal sections following subretinal injection of CD36⁻ or CD36⁺ microglia into NaIO₃-injured mice stained for NLRP3, BODIPY, and Iba1 with fluorescence quantification. (G) Experimental design for subretinal transfer of CD36⁺ microglia from wild-type or NLRP3−/− mice into half-NaIO₃-injured recipients at D0 and D4. Recipients received PLX5622-containing chow for 14 days or control diet. H, I visual acuity (H) and ERG a- and b-wave amplitudes (I) after transplantation of CD36+ microglia from NaIO₃-induced WT or NLRP3−/− donors into PLX5622-treated or control recipients (n = 8 eyes from 4 mice). J, K Fundus images (J) and OCT scanns (K) showing retinal structural alterations following transplantation (n = 8 eyes from 4 mice). L H&E staining showing retinal folds and subretinal cell accumulation in mice receiving CD36+ microglia from NaIO₃-treated WT or NLRP3−/− donors (n = 4 eyes from 4 mice). M RPE-scleral flat mount stained for ZO-1, NLRP3 and Iba1 with quantification of hexagonal ZO-1–positive cells and fluorescence intensity (n = 6 eyes from 6 mice). N Western blotting of IL-1β expression (n = 3 eyes from 3 mice). Scale bar: 50 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, n.s not significant, one-way ANOVA with Tukey’s post-hoc test for E, K, Kruskal–Wallis test with Dunn’s post hoc test for H, J, Mann–Whitney test in L and two-sided unpaired Student’s t test for (M, N). To determine whether the pathogenic effects of CD36⁺ microglia depend on NLRP3 signaling, we isolated CD36⁺ microglia from WT and NLRP3−/− mice and transplanted them subretinally into WT recipients subjected to a subthreshold dose of NaIO₃. Recipient mice were either pretreated with the CSF1R inhibitor PLX5622 for 14 days to deplete endogenous microglia or maintained on a control diet (Fig. 5G). Across both recipient conditions, mice receiving NLRP3−/− CD36⁺ microglia exhibited preserved retinal function compared with those receiving WT CD36⁺ microglia, evidenced by increased amplitudes of a- and b-waves and better visual acuity (Fig. 5H, I). Fundus imaging showed fewer dot-like infiltrates and white lesions in the NLRP3−/− CD36⁺ group (Fig. 5J), along with diminished subretinal cell aggregations (Fig. 5K). Consistently, ONL thickness was better preserved following transplantation of NLRP3−/− CD36⁺ microglia (Fig. 5K). Importantly, depletion of recipient-derived microglia with PLX5622 did not alter the detrimental effects induced by WT CD36⁺ cells, indicating that the observed pathology is primarily driven by the transplanted microglia rather than host microglial responses. Compared with WT CD36⁺ microglia, transplantation of NLRP3−/− CD36⁺ microglia reduced retinal folds and subretinal cell aggregation, preserved RPE barrier integrity with continuous hexagonal ZO-1 staining (Fig. 5L, M), and resulted in lower retinal NLRP3 and IL-1β levels (Fig. 5M, N, and S6B). In vitro, CD36 activation by PPARγ agonist pioglitazone increased NLRP3 and IL-1β expression (Fig. S6C). To define the downstream signaling landscape of IL-1β, we re-analyzed published human single-cell RNA-seq data29. IL-1β expression was mainly restricted to macrophage/microglia populations, whereas its receptor components, IL1R1 and IL1RAP, were detected across multiple retinal cell types and selectively upregulated in macular cells in AMD (Fig. S7A–C). Cells with high receptor expression exhibited enrichment of inflammatory or injury-related pathways (Fig. S7D–F). Immunostaining of NaIO₃-induced degenerative retinal sections confirmed increased IL-1 receptor expression in both microglial and non-microglial cells (pericytes/SMCs) (Fig. S7G–I). In vitro, recombinant IL-1β induced IL1R1 expression in endothelium and promoted their senescence (Fig. S7J). Together, these findings indicate that CD36⁺ aLARM amplify retinal neuroinflammation through NLRP3-IL-1β-dependent paracrine signaling. Therapeutic blockade of CD36 with the neutralizing antibody FA6-152 reduced aLARM formation and mitigated retinal degeneration Given the emerging role of CD36 as a therapeutic target in cancer and metabolic disorders30, we next investigated whether pharmacologic blockade of CD36 could suppress aLARM formation and alleviate retinal degeneration. We administered the CD36-neutralizing antibody FA6-152 intravitreally in mice subjected to NaIO₃-induced retinal injury and assessed its effects in vivo. Treatment with FA6-152 significantly decreased the accumulation of aLARM in the subretinal space, as shown by both cryosection imaging (Fig. 6A) and RPE flat mounts (Fig. 6B, and Fig. S8A). In addition to decreasing CD36⁺ microglia numbers, FA6-152 markedly suppressed the expression of lipid metabolism–related markers including Lrp1, P2rx7, and Abca1 in subretinal microglia (Fig. 6C, D). The NLRP3 and IL-1β expression were significantly downregulated after FA6-152 treatment at both RNA and protein levels (Fig. 6E, F), indicating attenuation of the proinflammatory signaling cascade.Fig. 6Therapeutic blockade of CD36 with FA6-152 suppresses aLARM formation and protected against retinal degeneration.A Immunofluorescence of retinal section showing reduced CD36 and BODIPY signals after FA6-152 treatment, with fluorescence quantification. Scale bar: 10 μm. B Retinal flat mount presenting the co-staining of Iba1, CD36, BODIPY and Plin3 under IgG or FA6-152 treatment after NaIO3 stimulation, with fluorescence quantification. Scale bar: 10 μm. C, D Western blotting (n = 3 eyes from 3 mice) (C) and retinal flat mount (D) showing the expression of other lipid metabolism markers including Lrp1, P2rx7 and Abca1 after FA6-152 or IgG treatment. Scale bar: 20 μm. (E, F) Western blotting (n = 4 eyes from 4 mice) (E) and qPCR (n = 7 eyes from 7 mice) (F) showing NLRP3 and IL-1β expression at the protein and mRNA level after IgG or FA6-152 treatment. (G–H) Retinal function examinations after intravitreal IgG or FA6-152 treatment by a-wave and b-wave amplitudes in ERG recordings (G) and visual acuity (H). (n = 6 eyes from 3 mice). (I–K) Representative images of structural alterations by dotlike lesions in retinal fundus examinations (n = 4 eyes from 4 mice for PBS; n = 8 eyes from 4 mice for IgG-NaIO₃; n = 8 eyes from 4 mice for FA6-152- NaIO₃ group) (I), ONL thickness and subretinal hyper-reflective spots (red arrow) in OCT scanning (n = 10 eyes from 5 mice) (J) and melanin rich aggregations (blue arrows) in H&E staining (n = 6 eyes from 6 mice) (K) in IgG or FA6-152 treated NaIO₃ mice comparing to the PBS group. L RPE-scleral flat mounts showing altered RPE morphology and loss of hexagonal structures (n = 6 eyes from 6 mice). Scale bar: 50 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, n.s not significant, one-way ANOVA test with Tukey’s post-hoc test for E, F, H, J–L two-sided unpaired Student’s t test for C, Kruskal–Wallis test with Dunn’s post hoc test for (I). Therapeutic blockade of CD36 with FA6-152 suppresses aLARM formation and protected against retinal degeneration. A Immunofluorescence of retinal section showing reduced CD36 and BODIPY signals after FA6-152 treatment, with fluorescence quantification. Scale bar: 10 μm. B Retinal flat mount presenting the co-staining of Iba1, CD36, BODIPY and Plin3 under IgG or FA6-152 treatment after NaIO3 stimulation, with fluorescence quantification. Scale bar: 10 μm. C, D Western blotting (n = 3 eyes from 3 mice) (C) and retinal flat mount (D) showing the expression of other lipid metabolism markers including Lrp1, P2rx7 and Abca1 after FA6-152 or IgG treatment. Scale bar: 20 μm. (E, F) Western blotting (n = 4 eyes from 4 mice) (E) and qPCR (n = 7 eyes from 7 mice) (F) showing NLRP3 and IL-1β expression at the protein and mRNA level after IgG or FA6-152 treatment. (G–H) Retinal function examinations after intravitreal IgG or FA6-152 treatment by a-wave and b-wave amplitudes in ERG recordings (G) and visual acuity (H). (n = 6 eyes from 3 mice). (I–K) Representative images of structural alterations by dotlike lesions in retinal fundus examinations (n = 4 eyes from 4 mice for PBS; n = 8 eyes from 4 mice for IgG-NaIO₃; n = 8 eyes from 4 mice for FA6-152- NaIO₃ group) (I), ONL thickness and subretinal hyper-reflective spots (red arrow) in OCT scanning (n = 10 eyes from 5 mice) (J) and melanin rich aggregations (blue arrows) in H&E staining (n = 6 eyes from 6 mice) (K) in IgG or FA6-152 treated NaIO₃ mice comparing to the PBS group. L RPE-scleral flat mounts showing altered RPE morphology and loss of hexagonal structures (n = 6 eyes from 6 mice). Scale bar: 50 μm. Experiments used mixed-sex mouse cohorts with an approximately balanced sex distribution. Data are shown as mean ± SEM, *P < 0.05, **P < 0.01, n.s not significant, one-way ANOVA test with Tukey’s post-hoc test for E, F, H, J–L two-sided unpaired Student’s t test for C, Kruskal–Wallis test with Dunn’s post hoc test for (I). Functionally, FA6-152 conferred robust neuroprotection. ERG revealed significantly improved amplitudes of a-wave and b-wave and enhanced contrast sensitivity in treated mice (Fig. 6G, H). Retinal fundus examination showed that FA6-152 treated group presented a more reddish-orange retina similar to PBS control, in contrast to the pale retinas with dot-like infiltrates and white lesions observed in untreated NaIO3 injected retinas (Fig. 6I). OCT scans further demonstrated restoration of a smooth, continuous, and uniform reflectiveness RPE band in the FA6-152 group, while untreated eyes showed aberrant hyperreflective foci and disruption of RPE layer (Fig. 6J). FA6-152 treatment also reversed ONL thinning and preserved photoreceptor layer integrity (Fig.6J). Histological analyses supported these findings. H&E staining revealed that FA6-152 preserved the laminar structure of the RPE and reduced melanin granule aggregation and ONL folding (Fig. 6K, and Fig. S8B). ZO-1 staining of RPE flat mounts confirmed that FA6-152 restored tight junction continuity, with more confluent hexagonal packing patterns indicative of barrier integrity, and corresponded with a decrease in subretinal microglia (Fig. 6L, and Fig. S8C). Finally, FA6-152 treatment remarkably suppressed NLRP3 and IL-1β expression in aLARM cells (Fig. S9A–C), further validating that CD36 blockade disrupts the inflammatory and degenerative axis driven by CD36⁺ microglia. Collectively, these data demonstrated that therapeutic inhibition of CD36 using FA6-152 effectively limited aLARM formation, dampened retinal inflammation, and preserved neuroretinal structure and function in a dry AMD model, highlighting CD36 as a promising target for intervention in retinal degenerative disease. Discussion In this study, we identified a previously unrecognized microglia state in the degenerative retina, defined by a lipid-loaded phenotype and a distinct transcriptional signature marked by high CD36 expression, which we term CD36+ aLARM. These cells contribute to neural damage and retinal inflammation through activation of the NLRP3-IL-1β pathway. Functionally, genetic ablation of Cd36 in microglia preserved visual function in degenerative models, while pharmacological blockade using a CD36-neutralizing antibody attenuated retinal pathology across multiple models of retinal degeneration. Together, our findings not only uncover a pathogenic microglial subset linked to lipid dysregulation and inflammation but also highlight CD36 as a promising therapeutic target for mitigating neuroinflammation and preserving retinal integrity in degenerative diseases. Our discovery of CD36⁺ aLARM reveals a previously unrecognized microglial subset in the degenerative retina, uncovering a disctinct functional phenotype of microglia in retinal neurodegeneration. Using sc-RNA sequencing, we identified MG4 subtype, corresponding to aLARM, exhibited enriched expression of genes involved in lipid uptake, trafficking, efflux, oxidation, biosynthetic across multiple retinal degeneration models. Transcriptionally, aLARM showed substantial overlap with DAM and MGnD, both of which were originally characterized in severe neurodegeneration models such as Alzheimer’s disease, amyotrophic lateral sclerosis, and multiple sclerosis4. For example, Gpnmb, a hallmark DAM gene, is a known marker for lipid-laden macrophages in Alzheimer’s disease5,31, while AnxA2, shared with MGnD, has been implicated in cholesterol-rich lipid raft microdomains critical for inflammatory signaling32–34. In contrast, the similarity between aLARM and LDAM found in aged brain tissue was limited, likely due to differences in disease severity, microenvironmental context, and lipid composition26. The retina is uniquely enriched in long-chain polyunsaturated fatty acids, particularly within photoreceptor outer segments35, whereas brain lipids primarily support membrane integrity and myelination36,37. These observations suggest that although aLARM, DAM, and MGnD may activate conserved lipid-responsive programs, their effector functions are shaped by tissue-specific cues. aLARM thus represents a retina-specific microglial phenotype with distinct lipid-driven, pro-inflammatory features that may require tailored therapeutic strategies. The functional role of lipid-laden microglia has been a topic of debate. Lipid droplets in myeloid cells have been associated with both protective and pathological roles depending on the context38. In peripheral diseases, foamy macrophages in atherosclerosis and lipid-rich eosinophils in allergic inflammation are linked to pathology39. Conversely, lipid-enriched dendritic cells have been shown to enhance antigen presentation and host defense40. In the CNS, Drosophila glia utilize lipid droplets to buffer toxic peroxidized lipids, suggesting a neuroprotective function41, while in aging and neurodegenerative mammalian brains, LDAM exhibit impaired phagocytosis, heightened oxidative stress, and chronic cytokine secretion26. In the retina, subretinal lipid-laden microglia have been observed at the origin of drusen-like deposits in mouse models, but their identity and pathogenic role remained unclear. In this study, our data supports a load-dependent impact of lipid accumulation in microglia. While moderate lipid uptake may support clearance and metabolic adaptation, excessive lipid load drives metabolic imbalance and inflammatory reprogramming. The aLARM likely represents this maladaptive state arising from overwhelmed lipid-buffering capacity, providing a unifying framework for the dual roles of lipid-laden microglia in retinal degeneration. These insights carry important implications for AMD, a disease marked by lipid-rich drusen accumulation, chronic inflammation, and progressive photoreceptor loss42. The retina’s reliance on precise lipid turnover makes it particularly vulnerable to disruptions in lipid homeostasis43,44. When microglia are overwhelmed by lipid burden or fail to appropriately process lipid debris, they may undergo pathological reprogramming toward an aLARM-like state. Our results show that aLARM not only accumulate lipids but also promote neuroinflammation via the CD36–NLRP3–IL-1β axis. Upregulation of IL-1β has been reported to be associated with an increased risk of incident AMD45. This highlights a critical immunometabolic axis in retinal degeneration and supports the notion that targeting CD36-mediated lipid sensing in microglia could restore homeostasis and mitigate disease progression. More broadly, our study underscores the double-edged nature of innate immune responses in lipid-rich environments and advocates for tissue-specific strategies to modulate microglial metabolism in neurodegenerative diseases. CD36 functions as a key lipid scavenger receptor, mediating the uptake of oxidized low-density lipoproteins (oxLDL), long-chain fatty acids, and other modified lipids20. Beyond CD36-mediated lipid uptake, lipid-sensing nuclear receptors such as PPARs may also modulate inflammatory reprogramming, given their role in coordinating lipid metabolism and regulating CD36 expression. Dysregulated lipid accumulation in degenerative retinas may therefore perturb PPAR-dependent transcriptional balance, reinforcing metabolic stress and inflammasome activation. This possibility merits future investigation. Under physiological conditions, such lipid internalization may support debris clearance and metabolic adaptation, thereby contributing to tissue homeostasis. However, in degenerative retinas characterized by sustained lipid stress, persistent uptake can exceed the buffering capacity of microglia. Excessive lipid uptake through CD36 leads to intracellular lipid accumulation, triggering metabolic stress and organelle dysfunction, including lysosomal overload and mitochondrial impairment46,47. In addition, lipid-laden microglia exhibit reduced phagocytic capacity, heightened oxidative stress, and impaired resolution of inflammation, all features that compromise their neuroprotective roles and exacerbate neurodegeneration. These findings suggest that CD36-mediated lipid dysregulation acts as a critical upstream event that reprograms microglial function toward a pathogenic state in retinal degenerative disease. Given its dual role in lipid sensing and immune activation, CD36 has emerged as a promising therapeutic target in diverse diseases, including cancer and metabolic disorders. Several groups have explored pharmacological strategies to inhibit CD36 function using monoclonal antibodies, small molecules, or mRNA-based therapies. In preclinical cancer models, CD36 blockade significantly reduced metastatic spread, prompting the launch of early clinical development programs48,49. In our retinal degeneration models, treatment with FA6-152, a murine IgG1 that blocks fatty acid transport and thrombospondin-1 interaction, effectively suppressed retinal damage, whereas JC63.1, another anti-CD36 antibody with distinct binding properties, failed to achieve similar benefits. These findings underscore the importance of epitope specificity and antibody isotype in modulating CD36 function. Although CD36 is detectable on retinal microglia under physiological conditions, its baseline expression remains relatively low. In contrast, CD36 is markedly upregulated in response to stress, lipid accumulation, or inflammatory stimuli, as observed across models of retinal degeneration. Our cross-species single-cell and histological analyses indicate that CD36 enrichment is most pronounced in microglia, particularly within disease-associated state. While low-level expression can be detected in other retinal cell types, both the magnitude and inducibility of CD36 are substantially greater in microglia, supporting their central role in CD36-driven pathology. This inducible pattern underscores the importance of temporal context in therapeutic targeting. Intervening during the window of pathological CD36 upregulation, rather than during basal expression, may allow for selective suppression of disease-associated microglial states while minimizing disruption of physiological lipid handling. Thus, timing and precision of anti-CD36 therapy will be critical to maximizing therapeutic benefit and avoiding unintended off-target effects. Further exploration of this area is clearly justified. Together, we identified a lipid-loaded microglial subtype, termed aLARM, in retinal degeneration using single-cell RNA sequencing and validated its spatial distribution through transcriptomic mapping. We demonstrated that CD36 is a defining feature of aLARM and mediates their neurotoxic effects via activation of the NLRP3-IL-1β inflammatory pathway. Both genetic ablation and antibody-based blockade of CD36 effectively attenuated retinal degeneration and preserved visual function, highlighting its therapeutic potential. Future studies are needed to determine whether aLARM represents a common feature across diverse forms of retinal neuroinflammation or participate in broader neurodegenerative processes beyond the eye. Nevertheless, our findings position CD36⁺ aLARM as a mechanistically defined and druggable target for modulating microglial-driven neuroinflammation and restoring tissue homeostasis in retinal degenerative diseases. Methods Animals All mice were maintained in a specific pathogen-free facility with a 12-h light/12-h dark cycle in Animal Laboratories of Zhongshan Ophthalmic Center, and all the experiments were carried out in accordance with the ARVO Statement and followed the guidelines of Animal Care and Use Committee of Zhongshan Ophthalmic Center (Ethics ID: Z2024001). Eight- to twelve-week-old C57BL/6 J (GemPharmatech, Nanjing, China), NLRP3-/- (Strain: #021302) and CD36 cKO mice were used in this study. Male and female mice were included in approximately equal numbers across experimental cohorts, and sex was not analyzed as an experimental variable. The CD36 cKO mice were constructed by crossing CD36fl/fl mice (Strain: #032276) and Cx3cr1cre mice (Strain: #025524). Cx3cr1-GFP mice (Strain: #005582) were obtained from the Jackson Laboratory. To establish the retinal degeneration model, the mice were injected intravenously with 20 mg/kg of NaIO3 (Sigma-Aldrich Corp., St. Louis, MO, USA), which was dissolved in phosphate-buffered saline (PBS). The mice receiving the same volume of PBS served as controls. For gain of function, isolating CD36+CD11b+CD45int cells from the NaIO3-induced retina and transplanting them into the subretinal space of recipient mice subjected to a subthreshold dose of NaIO₃ (10 mg/kg). For microglial depletion experiments, mice were administered a chow diet containing the CSF1R inhibitor PLX5622 (1200 ppm; Changzhou SYSE Bio-tec Co., Ltd) for 14 days prior to NaIO₃ induction or adoptive cell transfer, as previously described50. Control animals received standard chow. This treatment efficiently depleted endogenous retinal microglia before transplantation experiments. For therapeutic intervention, NaIO₃-induced C57BL/6 J mice received intravitreal injections of the CD36-blocking antibody FA6-152 (Abcam, ab17044) or mouse IgG1 isotype control (Selleck, A2106). A dose gradient of 40, 200, and 1000 ng/mL was tested to determine the optimal dosage, and 200 ng/mL was selected as the lowest effective concentration for subsequent experiments. All mice were euthanized by carbon dioxide inhalation seven days after NaIO₃ treatment unless otherwise specified, and eyes were collected for subsequent analyses. Human samples Postmortem human eyes were obtained from the Eye Bank of Guangdong Province, China. Donors were 55–75 years of age, with equal numbers of males and females. The eyes were processed within 3.5 h of procurement. Before tissue collection, each globe was examined by an ophthalmologist for retinal pathology. Eyes with dry AMD but no other retinal disease were assigned to the AMD group, whereas eyes without retinal disease served as controls. Written informed consent for eye donation was obtained from the donors or their legal next of kin by the eye bank. Following the clinical use of the corneas, the remaining retinal tissues were provided for research in accordance with the eye bank’s established procedures. The use of these residual donor tissues was approved by the Ethics Committee of Zhongshan Ophthalmic Center, Sun Yat-sen University (No. 2024KYPJ082), and the study was registered at www.medicalresearch.org.cn (MR-44-25-043663). All procedures involving human donor tissues were conducted in accordance with the relevant ethical regulations. En-face OCTA imaging of subretinal macrophage-like cells in patients Retinal images were captured using the Topcon DRI OCT Triton SS OCT Angio™ system (Topcon Corp., Tokyo, Japan). High-resolution 3D volume scans centered on the fovea (3 × 3 mm) were acquired using the Angio mode with motion correction and automated layer segmentation enabled. To visualize macrophage-like cells (MLCs) potentially residing beneath the neurosensory retina, en face structural OCTA slabs were reconstructed at subretinal levels with slab thickness set to 10 μm. Hyperreflective foci (HRF) appearing as discrete, high-intensity punctate signals within these subretinal slabs were considered candidate subretinal MLCs. Each detected HRF was confirmed in cross-sectional B-scan views to exclude projection artifacts, vessels, or drusen-related reflectivity. The study was approved by the Clinical Research Ethics Committee of Zhongshan Ophthalmic Center and was registered on ClinicalTrials.gov (Identifier: NCT06803082). Participants were recruited consecutively from the observational ophthalmic study at Zhongshan Ophthalmic Center and enrolled according to predefined inclusion and exclusion criteria. Both AMD patients and healthy controls were recruited through the same process. Single-cell RNA sequencing Sample collection Retinae from six C57BL/6 J mice per group, treated with either PBS or NaIO3 were harvested and pooled. Retinae were minced into small pieces and incubated in Ames’ Medium containing 15 U/mL papain (Worthington) and 0.1 mg/mL DNase I (Sigma) at 34 °C for 15 min. Subsequently, the cell suspensions were filtered through a 40 µm filter as previously described21, and centrifuged at 300 g for 8 min at 4 °C. Cell pellets from NaIO3 group were resuspended, incubated with CD11b MicroBeads (STEMCELL Technologies), and magnetically separated using an EasySep™ Magnet (STEMCELL Technologies) according to the manufacture instruction. After three washes with Selection Cocktail buffer, CD11b⁺ cells were isolated and resuspended in 1 mL Ames’ Medium for downstream processing. Samples with a viability >85% were used for single-cell RNA sequencing library preparation. Library preparation and data analysis Single-cell RNA-seq libraries were constructed using the Chromium Single Cell 3′ Reagent Kits v3 (10x Genomics, Pleasanton, CA, USA) following the manufacturer’s protocol. Libraries were sequenced on an Illumina NovaSeq 6000 platform at a depth of 50,000–100,000 reads per cell. Raw sequencing data were demultiplexed and aligned to the mouse reference genome (mm10) using Cell Ranger (10x Genomics, version 3.0.2). Subsequent analysis was performed in R (version 4.1.2) using the Seurat package (version 4.1.0). Cells expressing fewer than 200 genes or exhibiting >15% of total UMIs derived from mitochondrial genes were excluded. Gene expression data were log-normalized and scaled. Datasets from PBS and NaIO₃-treated mice were integrated using Seurat v3 and further scaled for downstream analysis. Candidate genes were subjected to gene set enrichment analysis using the ClusterProfiler package, and results were visualized using the cnetplot function in R. Non-linear dimensionality reduction was performed using t-SNE based on top principal components (PCs). Clustering was achieved through graph-based methods utilizing the FindNeighbors and FindClusters functions in Seurat (clustering resolution = 0.2). Microglia/macrophage populations were identified based on the expression of canonical markers including Itgam, Cx3cr1, P2ry12, and Csf1r. A second round of principal component analysis was conducted on the microglia/macrophage subset to refine clustering. Differential gene expression analysis was performed using FindAllMarkers or FindMarkers with the Wilcoxon rank-sum test (adjusted p < 0.05). Top marker genes for each cluster were identified by pairwise comparisons against all other clusters. Clusters were annotated based on enrichment of canonical microglial subtype markers and genes involved in defined biological processes. Functional enrichment analysis was performed using G:Profiler. Enriched Gene Ontology (GO) Biological Process terms were visualized based on fold enrichment and -log₁₀ (FDR). For cross-species validation, human single-cell RNA-seq data were obtained from the NCBI Gene Expression Omnibus (GEO) under accession number GSE135922, as published previously29,51. Spatial transcriptomics sample preparation, library construction and data preprocessing Spatial transcriptomics sample preparation and imaging Mouse eyes were enucleated at day 7 following NaIO₃ or PBS treatment and immediately embedded in optimal cutting temperature (OCT) compound. Eyeballs were rapidly frozen in liquid nitrogen–cooled isopentane and stored at −80 °C until sectioning. Sagittal cryosections (10 μm thickness) crossing the optic nerve head were obtained using a Leica CM1950 cryostat and immediately mounted onto Stereo-seq chips (STOmics, BGI). Tissue sections were fixed in pre-cooled methanol at −20 °C for 30 min. After fixation, sections were stained with ssDNA fluorescent dye (Thermo Fisher Scientific, Q10212) to visualize tissue morphology and imaged using a Nikon Eclipse Ti-7 fluorescence microscope. High-resolution images were acquired for spatial alignment and downstream analysis. Permeabilization, reverse transcription, and library construction Permeabilization was performed according to the manufacturer’s protocol to release mRNA molecules onto DNA nanoball-patterned arrays on the Stereo-seq chip. Captured transcripts were reverse transcribed in situ to generate spatially barcoded cDNA. Following reverse transcription, cDNA products were released, purified, and amplified for library preparation. Indexed sequencing libraries were constructed using the Stereo-seq Library Preparation Kit following standard STOmics protocols. Final libraries were sequenced on an MGI DNBSEQ-T7 platform using paired-end sequencing. Primary data processing and spatial matrix generation Raw sequencing data were processed using the Stereo-seq Analysis Workflow (SAW, BGI Research). Low-quality reads and invalid coordinate identity (CID) sequences were filtered with default parameters. Clean reads were aligned to the mouse reference genome (GRCm38/mm10) using STAR aligner to generate spatial gene expression matrices. The original bin1 expression matrix was aggregated into bin50 resolution for downstream analyses unless otherwise specified. Spatial coordinates were assigned to each bin based on chip layout information. Downstream bioinformatic analysis Subsequent analyses were performed using the StereoMap pipeline and Seurat (v4.0). Spatial bins with fewer than 100 detected genes were excluded. Gene expression matrices were normalized using SCTransform, followed by principal component analysis (PCA). Clustering analysis was performed using shared nearest neighbor (SNN) graph-based clustering with the top principal components. Differentially expressed genes were identified using the Wilcoxon rank-sum test, with adjusted P < 0.05 considered statistically significant. Spatial gene expression patterns were visualized by overlaying transcriptomic data onto corresponding ssDNA fluorescence images to reconstruct retinal layer-resolved transcriptional architecture. Smart-seq2 analysis Cell sorting Retinae were dissected, dissociated and stained with CD45-APC, CD11b-BV421, CD36-PE antibodies and Live/Dead dye. CD11b⁺CD45int cells from NaIO₃-treated CD36fl/fl or CD36 cKO mice, CD36+CD11b⁺CD45int and CD36−CD11b⁺CD45int cells from NaIO₃-treated wild-type mice were sorted using a FACSAria Fusion cell sorter (BD Biosciences) for downstream analysis. RNA Extraction, library construction and sequencing Cells were lysed in 10× Lysis Buffer (Vazyme, #N711). For Smart-seq2 library generation, a priming buffer mix containing dNTPs and oligo-dT primers was added to the lysate and denatured at 72 °C. cDNA synthesis and pre-amplification were carried out using the Discover-sc WTA Kit V2 (Vazyme) according to the manufacturer’s instructions. Amplified cDNA was then purified using VAHTS DNA Clean Beads. Sequencing libraries were prepared using the TruePrep™ DNA Library Prep Kit V2 (Vazyme, #TD503). Briefly, 1 ng of pre-amplified cDNA was incubated with the TruePrep Tagment Enzyme Mix, which includes Tn5 transposase and adapters, at 55 °C for 10 min to fragment the cDNA. Library amplification was performed using N5/N7 and P5/P7 primer pairs. Library quality was assessed with an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) and sequencing was performed on a NovaSeq X Plus platform by Guangzhou OmicsMaster Biotechnology Co., Ltd. (Guangzhou, China). Bioinformatics analysis Raw sequencing reads were processed using fastp (v0.18.0) for adapter trimming and quality control, followed by alignment to the mouse reference genome (mm10) using HISAT2 (v2.2.4). Gene expression levels were quantified as transcripts per million (TPM) using StringTie (v1.3.1), and genes with TPM > 0 were considered detected. Differential expression analysis between PBS and NaIO₃ groups was conducted using DESeq2. Genes or transcripts with a false discovery rate (FDR) < 0.05 and absolute fold change ≥1 were considered differentially expressed (DEGs/DETs). Functional enrichment analysis of DEGs was performed using G:Profiler. Key genes related to CD36 were mapped to protein-protein interaction networks using the STRING database (v11.0, http://string-db.org/), and the top 11 genes were selected based on their frequency across enriched GO terms. Electroretinogram (ERG) recordings Following overnight dark adaptation, mice were anesthetized and their pupils were dilated with 0.5% tropicamide. ERG recordings were performed using gold-plated wire loop electrodes placed on the corneal surface as active electrodes, and a stainless steel needle electrode inserted into the tail served as the ground. Full-field scotopic flash stimuli were delivered at intensities of 0.03, 0.1, 0.3, 1.0, 3.0, and 10.0 log cd·s/m². The amplitudes of a- and b-waves were recorded using the Celeris-Diagnosys system52. Visual acuity Visual acuity test began when the mouse on the platform ceased voluntary movement. A 100% contrast grating with a low spatial frequency (0.042 cycles/degree) was presented, and the mouse’s tracking behavior was observed. The grating was then rotated counterclockwise to test the contralateral eye. The acuity threshold was determined using a staircase procedure: if no response was observed, a lower frequency was presented; if tracking occurred, the frequency was increased until the highest non-responsive spatial frequency was found. The threshold was defined as the highest frequency eliciting a behavioral response53. Optical coherence tomography (OCT) Murine OCT imaging was performed using the SPECTRALIS system (Heidelberg Engineering, Germany) equipped with a mouse-specific objective. The optic disc was centered during imaging, and scans were acquired in automatic real-time (ART) mode to enhance signal-to-noise ratio. Scans with a quality index above 28 were included. A total of 12 circular scans were collected per eye using a 30° field-of-view, with 60–100 frames per scan. Retinal outer nuclear layer (ONL) thickness was quantified in ImageJ (NIH), defined as the distance from the outer plexiform layer to the outer limiting membrane. Measurements were averaged from 10 positions spaced every 200 μm across five locations on each side of the optic disc. Blinded investigators performed all measurements. Retinal fundus imaging and grading After anesthesia and pupil dilation, retinal fundus images were captured using a Micron IV camera (Phoenix Research Labs, Pleasanton, CA, USA) with a mouse-specific objective and a 50° field of view (1.8-mm diameter) centered on the optic nerve head53. Hematoxylin and eosin staining and thickness analysis Eyeballs were fixed in 4% formalin overnight, embedded in paraffin, and sectioned vertically at 3 μm thickness. Sections were stained with H&E and examined under a light microscope (Leica DM4000, Germany) as previously described21. Transmission electron microscopy Eyes were enucleated and dissected to remove the anterior segment and vitreous. Small retinal-RPE-choroid tissue blocks (~1 mm³) were fixed in EM fixative (Servicebio, G1102) at 4 °C for 2–4 h. Samples were rinsed three times in 0.1 M phosphate buffer (PB, pH 7.4) for 15 min each. Post-fixation was performed in 1% osmium tetroxide in 0.1 M PB for 2 h at room temperature, followed by three more rinses. Tissues were dehydrated through graded ethanol (50%–100%) and acetone, then infiltrated with acetone:resin (1:1 for 2–4 h; 1:2 overnight) and pure Poly/Bed 812 resin for 5–8 h. Samples were embedded in resin and polymerized at 60 °C for 48 h. Ultrathin sections (60–80 nm) were cut with a Leica EM UC7 ultramicrotome and stained with 2% uranyl acetate and lead citrate for 15 min each. Sections were imaged using a Hitachi HT7700 transmission electron microscope (Hitachi Co., Ltd., Tokyo, Japan). Quantitative PCR analysis and Western blotting For RNA extraction, the RPE-sclera complex was dissected by removing the anterior segment, lens, vitreous, and neuroretina. Tissues were lysed in Trizol (Thermo Fisher Scientific), and qPCR was performed as previously described54. Each 20 μL qPCR reaction contained 2 μL cDNA, 10 μL 2× SYBR Premix Ex Taq, 7 μL ddH₂O, and 10 μmol/L primer pairs. Primer sequences are listed in Table S1, and reactions were run on a Roche LightCycler480 II system. Western blotting was performed as described previously21, and primary antibodies are detailed in Table S2. Immunofluorescence Immunofluorescence staining was performed on human retinal cryosections, murine retinal whole-mounts, and cryosections. The primary antibodies are listed in Table S2. Confocal microscope images were captured using LSM980 (Carl Zeiss)55. For visualization of microglial morphology and spatial distribution across retinal layers, retinal whole-mounts and cryosections were stained with Iba1 and counterstained with DAPI to identify microglia located within the retinal nerve fiber layer (RNFL), inner plexiform layer (IPL), and outer plexiform layer (OPL), as shown in Fig. S10A. Flow cytometry Retinal single-cell suspensions were prepared as previously described with minor modifications56. Briefly, mouse eyeballs were enucleated, and retinae were carefully dissected. Retinal tissues were transferred into RPMI-1640 medium supplemented with HEPES buffer and mechanically dissociated by gentle trituration using wide-bore pipette tips. The resulting cell suspension was passed through a 70 μm cell strainer (BD Biosciences) to remove debris and cell aggregates. Cells were washed with phosphate-buffered saline (PBS) containing 1% fetal bovine serum (FBS) and incubated with fluorochrome-conjugated antibodies against CD45, CD11b, and CD36 and Live/Dead dye for 30 min at room temperature in the dark. Flow cytometric acquisition was performed using a BD flow cytometer (BD Biosciences), and data were analyzed with FlowJo software. Retinal microglia were identified as CD45int CD11b+live cells, and the proportion of CD36⁺CD11b⁺ cells was subsequently quantified (Fig. S10B). Cell culture and treatments Human microglial HMC3 cells (Procell, CL-0620) were maintained in MEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin, bEnd.3 (Procell, CL-0598) cells were cultured in DMEM containing 10% FBS and 1% penicillin–streptomycin. All cells were maintained at 37 °C in a humidified incubator with 5% CO₂. To mimic oxidative stress in a lipid-loaded microenvironment, HMC3 cells were treated with sodium iodate (NaIO₃, 0.5 mM; Sigma) together with human oxidized low-density lipoprotein (oxLDL, 50 μg/mL; Yiyuan Biotechnologies, China, YB-002) for 24 h. To determine whether PPARγ regulates activation of the CD36–NLRP3–IL-1β axis in microglia, HMC3 cells were treated with the PPARγ agonist pioglitazone (MCE, HY-14601) at 30 μM for 24 h. Dil-oxLDL uptake assay To evaluate oxLDL internalization by microglia, HMC3 cells were incubated with Dil-labeled oxLDL (Yiyuan Biotechnologies, China, YB-0010) at 0, 1, 2.5, 10, 50, or 100 μg/mL under NaIO₃ (0.5 mM) stimulation for 24 h, followed by fluorescence imaging. oxLDL dose-response stimulation and RNA sequencing To assess the effects of oxLDL across low-dose to high-dose ranges, HMC3 cells were treated with oxLDL at 0, 1, 2.5, 10, 50, or 100 μg/mL for 24 h. Total RNA was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. RNA quality and integrity were assessed by agarose gel electrophoresis and an Agilent 2100 Bioanalyzer system prior to library preparation. RNA sequencing libraries were constructed using the NEBNext® Ultra™ RNA Library Prep Kit for Illumina following the manufacturer’s protocol. Briefly, mRNA was enriched using poly-T oligo-attached magnetic beads, fragmented, and reverse-transcribed to synthesize first- and second-strand cDNA. After end repair, adapter ligation, and PCR amplification, libraries with insert sizes of approximately 250–300 bp were generated and sequenced on an Illumina NovaSeq platform to obtain paired-end 150-bp reads. Raw sequencing reads were processed to remove adapter sequences and low-quality reads. Clean reads were aligned to the human reference genome using HISAT2 (v2.0.5), and gene expression levels were quantified as fragments per kilobase per million mapped reads (FPKM) using featureCounts. Differential gene expression analysis was performed using DESeq2, with adjusted P < 0.05 considered statistically significant. For functional interpretation, Gene Ontology (GO) and pathway enrichment analyses were conducted using the clusterProfiler R package. Genes belonging to selected biological processes, including intracellular lipid transport, phagocytosis, TNFα signaling via NF-κB, and NLRP3 inflammasome pathways, were extracted to generate heatmaps illustrating transcriptional responses across oxLDL dose gradients. IL-1β stimulation assays To evaluate the effects of IL-1β on IL-1R-expressing cells, bEnd.3 cells were treated with recombinant IL-1β (MCE, HY-P7073) at 10 ng/mL for 24 h. Statistics and reproducibility For immunofluorescence, Hematoxylin & eosin staining and TEM on retinal section, at least three sections of each retina and three retinae of each group were used. In every section, three images were captured in the center, mid-periphery, and periphery area, respectively. For immunofluorescence on whole mounts, at least six images were captured in the center, mid-periphery, and periphery areas of each retina respectively. six retinae or RPE-sclera complex from six mice were used in each group for analysis. Representative images were shown in the according Figures. In experiments including qPCR and western blot, samples were collected from three retinae or RPE-sclera complex of three individual mice. All in vitro experiments were performed in triplicate and repeated independently for at least three times. Data was presented as mean ± standard error of mean (SEM) with the number of samples provided in the figure legends. Normality was assessed using the Shapiro-Wilk test. For normally distributed data, two-sided unpaired Student’s t test for two groups or one-way ANOVA with Tukey’s post hoc test were used for over three groups. For non-normal data, Mann-Whitney U test or Kruskal–Wallis test with Dunn’s post hoc test were applied. A p value < 0.05 was considered statistically significant. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information supplementary Information file Reporting summary Transparent Peer Review file supplementary Information file Reporting summary Transparent Peer Review file Source data Source Data Files Source Data Files