Microglial phagocytosis of bipolar cells triggers inner retinal degeneration in Rs1-KO mice

Abstract

Background

X-linked juvenile retinoschisis (XLRS) is a hereditary retinal disorder caused by mutations in the RS1 gene that leads to the formation of cavities in the inner nuclear layer (INL) and progressive vision loss, characterized by a disproportionate reduction of the b-wave compared to the a-wave in electroretinography (ERG). While previous research has largely focused on photoreceptor degeneration in XLRS, the specific roles of other cell populations, particularly bipolar cells and microglia, in the early stages of the disease have remained less well understood. Thus, this study aimed to elucidate the early cellular and molecular mechanisms of retinal degeneration in XLRS, with a particular focus on the role of microglia and bipolar cells.

Methods

Retinal structure and function were assessed in CRISPR/Cas9 Rs1-exon2 knockout (Rs1^−/y) mice at 8 and 24 weeks using histology, spectral-domain optical coherence tomography (SD-OCT), ERG, and optokinetic response. To analyze cell-specific changes, we performed TUNEL assay, immunofluorescence, flow cytometry, and single-cell RNA sequencing (scRNA-seq) with trajectory analysis.

Results

Rs1^−/y mice successfully recapitulated classic XLRS features, including INL schisis, reduced b/a-wave ERG ratio, and early vision loss. TUNEL assay and histological analysis revealed that cell death initiated in the INL at 8 weeks and progressed to the outer nuclear layer (ONL), while microglia displayed a progressive transition from a ramified to an ameboid morphology. scRNA-seq demonstrated a significant loss of cone bipolar cells, especially OFF-cone subtypes, which preceded photoreceptor degeneration. Importantly, microglial activation and enhanced phagocytosis of OFF-cone bipolar cells were observed prior to photoreceptor loss. This phagocytic process was found to be mediated by phosphatidylserine and complement C3b, independent of caspase-3 pathways.

Conclusions

Our findings demonstrate that bipolar cell degeneration, driven by microglial phagocytosis of stressed yet viable OFF-cone bipolar cells, is an early and critical pathological event in XLRS that precedes photoreceptor loss. This process involves "eat-me" signals and complement activation independent of classical apoptosis. These results provide a new perspective on XLRS pathogenesis and suggest that therapeutic strategies targeting bipolar cells and microglial activity could offer promising avenues for early intervention.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12964-026-02692-8.

Introduction

X-linked juvenile retinoschisis (XLRS), one of the most common forms of juvenile macular degeneration, is a hereditary retinal disorder characterized by splitting and structural disorganization within the layers of the neurosensory retina [1]. XLRS predominantly manifests during childhood or adolescence, typically between the ages of 5 and 10 years, with an estimated prevalence ranging from 1 in 5,000 to 1 in 25,000 among male individuals [2, 3]. Early symptoms often include reduced visual acuity and difficulty adapting to dark light, followed by severe visual impairment, including irreversible central vision loss and decreased ability to perform daily activities [4]. A hallmark feature of XLRS is the electronegative electroretinography (ERG) pattern, characterized by a disproportionate reduction of the b-wave compared to the a-wave [5, 6], where the a-wave is generated by photoreceptor cells, while the b-wave is believed to originate primarily from the activity of bipolar cells and Müller glial cells [7]. Another feature of XLRS is the formation of cavities in the inner nuclear layer (INL) of retina [8, 9], in which bipolar cells constitute the highest proportion, followed by Müller glia [10]. These characteristics suggest that the bipolar and Müller glial cells could play critical roles in pathophysiology of XLRS.

XLRS arises from mutations in the RS1 gene, located on the X chromosome, which encodes retinoschisin—a crucial regulator of retinal architecture and intercellular adhesion [4]. The RS1 gene comprises six exons that define the structural domains of retinoschisin [11], with its functional core centered around the discoidin domain—a conserved module essential for stabilizing retinal cell interactions [12]. Retinoschisin is synthesized by photoreceptors and bipolar cells and plays a pivotal role in maintaining retinal lamination and structural integrity by localizing on cell membranes as a disulfide-bonded functional octamer [12–14]. Mutations in RS1 can lead to impaired disulfide bond formation or misfolding of retinoschisin, disrupting its octameric assembly and consequently weakening synaptic integrity between photoreceptors and bipolar cells [15, 16]. These functional deficits contribute to aberrant photoreceptor development and signal transmission [17], dysregulated potassium ion homeostasis [18], heightened Müller cell reactivity [19], and microglial activation [20], collectively exacerbating retinal dysfunction.

Over the past two decades, investigations into XLRS have predominantly centered on photoreceptor dysfunction and degeneration, while limited attention has been directed toward elucidating the physiological roles of other retinal cell populations, including bipolar cells and microglia. Recently, the targeted expression of retinoschisin in bipolar cells using adeno-associated virus vectors encoding the RS1 gene with inner retina-specific promoters demonstrated tightly packed bipolar cells with reduced cavity formation and significant improvement in synaptic function in an XLRS mouse model [21]. These findings suggest that RS1 gene expression in bipolar cells can ameliorate the structural pathology of retinoschisis independently of photoreceptor expression, underscoring the pivotal role of bipolar cells in XLRS. In other hand, microglia, the primary immune cells of the retina, are increasingly implicated in the regulation of retinal homeostasis and pathology [22]. In XLRS, retinal architectural breakdown caused by retinoschisin deficiency may activate microglia, initiating an inflammatory response [20, 23]. While microglial activation is often neuroprotective, excessive or dysregulated activation can lead to the removal of stressed but viable cells, termed by 'phagoptosis' [24, 25]. Given the roles of bipolar cells and microglia in XLRS, their interaction is likely to play a pivotal role in the pathology of the disease, highlighting the need for further investigation into their dynamic interplay within the retinal microenvironment.

In this study, we aimed to elucidate the early cellular and molecular mechanisms underlying retinal degeneration in XLRS, with a particular focus on the role of microglia and bipolar cells. By generating an Rs1-exon2 knockout mouse model and employing a combination of histological, electrophysiological, and single-cell transcriptomic analyses, we sought to determine whether bipolar cell degeneration precedes photoreceptor loss and to explore the potential involvement of microglia-mediated phagocytosis in this process. Through this approach, we intended to uncover novel insights into the pathogenesis of XLRS and identify potential cellular targets for early therapeutic intervention.

Materials and methods

Generation of Rs1^−/y mice

Rs1-exon2-knockout (Rs1^−/y) mice were generated on a C57BL/6J background using CRISPR/Cas9 genome editing (Macrogen, Seoul, Korea). Two sgRNAs targeting exon 2 (NM_011302.3) of the Rs1 gene were used: Rs1_int1_gR1 (5′-GATTAAGGAGCTCA TTCTGGAGG-3′) and Rs1_int2_gR1 (5′-ATGAGTCTAGCACTATTGGGGGG-3′). The targeted deletion (chrX:164,114,291–164,114,487, GRCm39) removed 194 bp within exon 2, resulting in a frameshift and loss of Rs1 expression. Genomic deletion was confirmed by PCR and Sanger sequencing. Hemizygous Rs1^−/y males were obtained by mating heterozygous (Rs1^−/+) females with wild-type C57BL/6J males, and used for experiments at 8 and 24 weeks. Animals were maintained under a 12-h light/dark cycle with food and water ad libitum. All procedures were approved by the IACUC of Soonchunhyang University (protocol SCHBCA 2023–07).

Genotyping

To identify hemizygous Rs1^−/y mice, genomic DNA was extracted from tail biopsies of 4-week-old offspring using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany). PCR amplification was conducted using primers flanking exon 2 of the Rs1 gene (forward: GCTACACAGAGAAACCCTGTCTTG; reverse: CTTTCCACTCATGCCCACACC). The reaction yielded a 480 bp product for the wild-type allele and a 284 bp product for the knockout allele. PCR products were separated on 2% agarose gels and visualized using electrophoresis to confirm genotypes prior to experimental use.

Spectral-Domain Optical Coherence Tomography (SD-OCT)

Retinal structural imaging was performed using SD-OCT to assess morphological changes in WT and Rs1^−/y mice at 8 and 24 weeks of age. Mice were anesthetized by intraperitoneal injection of Zoletil 50 (Virbac, Carros, France) at 25 mg/kg and Rompun (Bayer Healthcare, Leverkusen, Germany). Pupils were dilated with 0.5% tropicamide (Hanmi Pharm, Seoul, Korea), and a thin layer of hydroxypropyl methylcellulose gel was applied to the corneal surface to prevent dehydration. OCT images were acquired using a commercial SD-OCT ophthalmic imaging system (IIScience, Yangsan, Korea). Mice were carefully positioned to align the retina with the optical axis of the device. Cross-sectional (B-scan) images were obtained across the central retina, including the optic nerve head. All procedures were performed with the animal placed on a temperature-controlled heating pad to maintain physiological body temperature during imaging.

ERG test

Mice were dark-adapted for 12 h prior to the procedure. All procedures were conducted under dim red light conditions (λ > 600 nm). Anesthesia was induced by intraperitoneal injection of a mixture of Zoletil 50 (Virbac, Carros, France) at 25 mg/kg and Rompun (Bayer Healthcare, Leverkusen, Germany). During ERG recordings, mice were maintained on a heating pad to preserve body temperature. Gold ring contact electrodes were placed on the corneas, while a ground electrode was inserted subcutaneously in the tail. Full-field scotopic flash ERGs were recorded using electrode stimulator (Celeris, Diagnosys LLC, Lowell, MA). White flash stimuli of 0.001, 0.01, 0.05, 0.1, 1.0, 5.0, and 10.0 cd·s/m² were delivered, and three responses were recorded and averaged for each intensity. During ERG recordings, mice were placed on a heated platform to maintain body temperature. ERG signals were bandpass filtered at 0.125–300 Hz and digitized at a sampling rate of 2 kHz. Scotopic a-wave and b-wave amplitudes were analyzed across the stimulus intensities, and the b/a wave ratio was calculated at 10.0 cd·s/m². Data from both eyes were acquired and averaged for analysis.

Optokinetic response (OKR) test

Mice were acclimated to the testing environment for 30 min prior to the test. The Optodrum system (Striatech, Paris, France) was used to project 360° rotating sine-wave gratings. The animals were placed in the center of the arena, and their head movements were tracked using the Optodrum system to assess the OKR. The rotation speed ranged 12°/s and contrast (99.72%) on the monitors were kept constant, and the spatial frequency started at 0.056 cycles/degree. The response was recorded as the percentage of time the mice followed the rotating stimulus.

Immunoflourescence staining

Mice were transcardially perfused with 4% paraformaldehyde (PFA), and eyecups were post-fixed and cryoprotected in 30% sucrose. Tissues were embedded in optimal cutting temperature compound and sectioned at 10 μm thickness. Sections and flatmounts were stained following permeabilization with 0.1% Triton X-100 and blocking with 5% normal donkey serum.

A polyclonal anti-RS1 antiserum was generated in Sprague–Dawley rats (DBL, Eumseong, Korea) by immunizing with a KLH-conjugated synthetic peptide (CSTEDEGEDPWYQKA) corresponding to the RS1 sequence (Peptron, Daejeon, Korea). The peptide (5 mg) was emulsified with Freund’s adjuvant and subcutaneously injected at multiple sites. Booster immunizations were given at two-week intervals. Whole blood was collected after the final boost, and antiserum was isolated by centrifugation of the coagulated blood. The serum was aliquoted and stored at − 80 °C until use.

The primary antibodies used in this study included anti-RS1 (custom-generated as described above), anti-Iba1 (Wako, Osaka, Japan; Abcam, Cambridge, UK), PKCα and PKARIIβ antibodies (BD Biosciences, Franklin Lakes, NJ), Complement C3b antibody (Bioss Antibodies, Woburn, MA), cleaved caspase-3 antibody (Cell Signaling Technology, Danvers, MA), and NRG1 antibody (Invitrogen, Waltham, MA). Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific, Waltham, MA) were used at a 1:1000 dilution. PS were stained using FITC-conjugated Annexin V (BD Biosciences, San Jose, CA). Nuclei were counterstained with Hoechst 33,354 (Thermo Fisher Scientific, Waltham, MA).

Confocal images were acquired using a Leica TCS SP8 confocal microscope (Leica Microsystems, Wetzlar, Germany), using 40 ×/1.10 water immersion and 63 ×/1.40 oil immersion objective lenses. Imaging was performed at room temperature with anti-fade fluorescence mounting medium (abcam, ab104135) as the imaging medium. Fluorescent signals were detected using photomultiplier tubes (PMTs) or hybrid detectors (HyDs), without a camera. Images were acquired using LAS X software (Version 3.3.0.16799; Leica Microsystems) and subsequently processed with gamma correction and deconvolution using the same software. All images were saved in JPEG format with 24-bit color depth.

Hematoxylin and eosin staining

Retinal cryosections were stained using a commercial hematoxylin and eosin (H&E) staining kit (Abcam, Cambridge, UK) according to the manufacturer’s instructions. Sections were equilibrated to room temperature, rehydrated in PBS, and sequentially stained with hematoxylin and eosin. After dehydration through graded ethanol and xylene, the sections were coverslipped using Organo Mounting Solution (Limonene-based; ImmunoBioScience Corp., Mukilteo, WA, USA; AR-6504–02), which is suitable for brightfield microscopy. Brightfield images were acquired using a Leica Visoria B microscope (Leica Microsystems, Wetzlar, Germany) equipped with an N Plan 20 ×/0.40 NA objective lens. Images were captured using a Leica Flexacam i5 digital color camera with a 1/2.3″ CMOS sensor and up to 12 MP resolution. Image acquisition was performed using Enersight Desktop software (Leica Microsystems), and all image files were exported in 24-bit JPEG format.

Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay

To assess retinal cell death, TUNEL staining was conducted on 10 μm-thick cryosections using the In Situ Cell Death Detection Kit (Roche, Mannheim, Germany) in accordance with the manufacturer's instructions. Tissues were permeabilized with 0.1% Triton X-100 and 0.1% sodium citrate, incubated with the TUNEL reaction mixture at 37 °C for 1 h, and counterstained with Hoechst. Images were acquired by confocal microscopy (Leica DMi8; Leica, Wetzlar, Germany).

Sholl analysis of microglia

To assess the spatial extent of microglial ramification, single-plane confocal images of Iba1-immunostained microglia were analyzed. Images were processed using ImageJ Fiji software (version 1.54), converted to 8-bit grayscale, and threshold-adjusted to refine process boundaries. Binary images were generated, and the Skeletonize function was applied to convert cellular structures into single-pixel-wide skeletons. Sholl analysis was performed using the Sholl Analysis plugin integrated in Fiji, with concentric circles drawn at 5 μm intervals from the center of the soma [26, 27]. In this study, the intersecting radius—defined as the maximum distance at which microglial processes intersected the concentric circles—was measured to quantify the spatial extent of microglial ramification.

Retinal cell dissociation and flow cytometery

Primary retinal cells were dissociated based on a previously optimized protocol [28], with slight modifications. Retinas were isolated from wild-type and Rs1^−/y mice at 8 or 24 weeks of age. For each dissociation, retinas from both eyes of a single mouse were pooled and treated as one sample.

Each pooled retina was incubated in 1 mL of digestion solution containing 20 U/mL papain (Worthington Biochemical Corporation, Lakewood, NJ) and 0.005% DNase I (Sigma-Aldrich, St. Louis, MO) at 8 °C for 40 min, followed by an additional 10-min incubation at 28 °C. After enzymatic digestion, the solution was carefully removed without disturbing the tissue. To inactivate enzymatic activity, 700 μL of pre-warmed inactivation solution was added. The tissue was then gently dissociated by pipetting 10 times. Subsequently, 700 μL of washing solution was added, and the mixture was centrifuged at 200 × g for 5 min at 4 °C. The supernatant was discarded, and the pellet was resuspended in 3 mL of Dulbecco's Phosphate-Buffered Saline (DPBS) containing 0.04% BSA (Gibco, Waltham, MA). The resulting cell suspension was filtered through a 40 μm cell strainer (Corning, Corning, NY) and immediately examined under a brightfield EVOS microscope (Thermo Fisher Scientific, Waltham, MA) to assess cell morphology and viability.

For the analysis of cell death, dissociated cells were first washed once with PBS and resuspended in 1 × binding buffer at a concentration of 1 × 10⁶ cells/mL. Then, 100 μL of the cell suspension containing 1 × 10⁵ cells was transferred into a 5 mL round-bottom FACS tube. Cells were stained with 5 μL of FITC-conjugated Annexin V and 5 μL of PE-conjugated propidium iodide (PI) using the FITC Annexin V Apoptosis Detection Kit I (BD Biosciences, San Jose, CA). After gentle mixing, the samples were incubated for 15 min at room temperature in the dark. Following incubation, 400 μL of 1 × binding buffer was added to each tube, and flow cytometric analysis was performed within 1 h.

To assess microglial activation, dissociated primary retinal cells were fixed using Fix Buffer I (BD Biosciences, San Jose, CA) and permeabilized with 1:10 diluted BD Perm/Wash Buffer (BD Biosciences, San Jose, CA). Cells were blocked in PBS containing 5% fetal bovine serum (FBS; Thermo Fisher Scientific, Waltham, MA) for 15 min at 4 °C, then stained with FITC-conjugated anti-CD11b (Elabscience, Houston, TX) and PE-conjugated anti-CD45 (BD Pharmingen, San Diego, CA) antibodies. A total of 80,000 events per sample were acquired for analysis.

All samples were acquired on a DxFLEX flow cytometer (Beckman Coulter, Brea, CA), and data were analyzed using CytExpert software (version 2.3; Beckman Coulter, Brea, CA). Compensation was performed using single-stained controls, and gating strategies were applied to exclude debris and doublets based on forward and side scatter profiles.

Western blot analysis

Retinas were dissected and homogenized in RIPA buffer (GenDepot, Barker, TX) supplemented with protease and phosphatase inhibitors (GenDepot). Protein concentration was measured using the BCA assay (Thermo Fisher Scientific, Waltham, MA).

Equal amounts of protein (10 μg) were resolved on an 8% SDS-PAGE gel and transferred to PVDF membranes (Millipore, Billerica, MA). After blocking with 5% non-fat milk in PBST, membranes were incubated overnight at 4 °C with custom-generated anti-RS1 antiserum (see Immunofluorescence Staining) and anti-α-tubulin antibody (Cell Signaling Technology, Danvers, MA). HRP-conjugated secondary antibodies (GenDepot) were applied for 2 h at room temperature. Protein bands were visualized using EzWestLumi plus (ATTO, Tokyo, Japan) and detected with the Azure C280 imaging system (Azure Biosystems, Dublin, CA).

Single-cell RNA sequencing (scRNA-seq)

The samples were processed for library preparation using Chromium GEM-X Single Cell 3p RNA library v4 (10 × Genomics, Pleasanton, CA). After cell dissociation and viability assessment, the cells were mixed with master mix and loaded onto a chromium GEM-X chip with Single Cell 3′ v4 Gel Beads and Partitioning Oil. RNA transcripts from single cells were uniquely barcoded and reverse transcribed within droplets. The cDNA molecules were pooled and processed for library construction. The libraries were then quantified using quantitative PCR and qualified using Agilent Technologies 4200 TapeStation (Santa Clara, CA). Finally, the libraries were sequenced using the HiSeq platform (Illumina, San Diego, CA).

Raw gene expression matrix was generated for each sample using CellRanger v8.0.1, and integration of four dataset, dimensionality reduction and clustering were performed using Seurat (V5.2.0). Cells that had less than 450 expressed genes or more than 8,000 expressed genes were removed. Cells in which the fraction of mitochondrial genes exceeded 10% also were removed. We identified 2,000 highly variable genes using the FindVariableFeatures function in the Seurat library for initial clustering. The unique molecular identifier (UMI) count per gene were normalized by the total UMI count in each cell and log transformed with the NormalizedData function in Seurat using 10,000 as the scale factor. The effects of the number of detected UMIs, the fraction of mitochondrial genes, and cell cycles on the gene expression values were corrected by regression using the ScaleData function in Seurat.

Before the clustering, we performed batch correction using Harmony across each single-cell sample and integrate the gene expression matrix of all samples into an integrated layer. After cell-level quality filtering, we performed dimensionality reduction with 10 principal component analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP) dimensions, followed by neighbor finding using FindNeighbor function with n_neighbors set to 10. Clustering was performed using the FindClusters function with a resolution of 0.05, and cell populations were annotated based on classical markers derived from literature [29–31] and Human Protein Atlas (https://www.proteinatlas.org/).

Trajectory analysis was performed using the Slingshot v2.14.0 package [32]. We applied the getLineages function within Slingshot to the final annotated retinal subtype UMAP to identify the global lineage structure by constructing a minimum spanning tree on the clusters. To identify genes with expression changes over pseudotime, the FindMarkers function within Seurat was used. Genes were considered pseudotime-specific and significant if their false discovery rate was less than 0.05.

Statistical analysis

All data are presented as mean ± standard error of the mean (SEM). Statistical analyses were performed using GraphPad Prism (version 10.2.3; GraphPad Software, San Diego, CA, USA) and SPSS Statistics (version 22.0; IBM Corp., Armonk, NY, USA). For comparisons involving multiple groups and two independent variables (e.g., genotype and timepoint), two-way analysis of variance (ANOVA) followed by Tukey’s multiple comparisons test was performed using GraphPad Prism (Fig. 1F–H). For non-parametric data, the Kruskal–Wallis test followed by Mann–Whitney U tests for pairwise comparisons was conducted using SPSS (Fig. 2B, E–H, I–K). A p-value < 0.05 was considered statistically significant. Significance levels are indicated as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 1.

Structural disorganization and functional deficits in the retina of Rs1^−/y mice. A Establishment of CRISPR/Cas9-mediated Rs1-Exon2-del (Rs1^−/y) mouse. Two single guide RNAs (sgRNAs) were designed to induce double-strand breaks (DSBs) within exon 2 (NM_011302.3) of the Rs1 gene resulting in the deletion of the 197 bp exon 2. This deletion introduced a frameshift mutation and premature stop codon, which produced a truncated peptide corresponding to a part of signal peptide of RS1 (MPHKIEGFFLLLLFGYEG). The exon2 deletion was confirmed using Sanger sequencing. CDS, coding sequence; UTR, untranslated region; PAM, Proto-spacer Adjacent Motif. B Western blot analysis of RS1 expression in the retinas of 8-week-old WT and Rs1^−/y mice. RS1 protein (~ 26 kDa) was absent in Rs1^−/y mice, while α-tubulin (~ 50 kDa) was used as a loading control. C Immunofluorescence staining of RS1 in retinal cryosections from WT and Rs1^−/y mice at 8 and 24 weeks of age showed strong expression in the inner segments (IS), outer plexiform layer (OPL), and inner nuclear layer (INL) in WT mice, whereas no RS1 signal was detected in Rs1^−/y retinas. Nuclei were counterstained with Hoechst. Scale bar = 50 μm. D Hematoxylin and eosin (H&E) staining of retinal sections from WT and Rs1^−/y mice at 8 and 24 weeks revealed characteristic schisis-like cavities in the INL of Rs1^−/y mice at 8 weeks and progressive retinal thinning at 24 weeks. Scale bar = 100 μm. E Representative spectral-domain optical coherence tomography (SD-OCT) images of both eyes (oculus dexter, OD; oculus sinister OS) from WT and Rs1^−/y mice at 8 and 24 weeks showed schisis-like cavities in the INL of Rs1^−/y mice at 8 weeks and marked thinning of the outer retina at 24 weeks. Scale bar = 100 μm. F Quantification of scotopic ERG responses demonstrated that a-wave and b-wave amplitudes were reduced in Rs1^−/y mice compared to WT controls at both 8 and 24 weeks across flash intensities ranging from 0.001–10.0 cd·s/m². G Scotopic b/a wave ratio measured at a flash intensity of 10 cd·s/m² was decreased in Rs1^−/y mice compared to WT controls at both 8 and 24 weeks. H Optokinetic response (OKR) test revealed that the spatial frequency threshold (visual acuity, cycles/degree) was reduced in Rs1^−/y mice compared to WT controls at both 8 and 24 weeks. Data are presented as mean ± SEM. Statistical analysis was performed using two-way ANOVA followed by Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001. GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; OPL, outer plexiform layer; ONL, outer nuclear layer; PRL, photoreceptor layer; RPE, retinal pigment epithelium

Fig. 2.

Apoptotic cell death and microglial dynamics in the Rs1^−/y retinal degeneration model. A–B TUNEL staining and quantification revealed increased apoptotic cell death in Rs1^−/y retinas, primarily localized to the INL at 8 weeks and the ONL at 24 weeks compared to WT controls. A positive control section was included. Nuclei were counterstained with Hoechst. Scale bar = 50 μm. C–D Iba1 immunostaining of retinal cryosections and flatmounts revealed increased microglial density, morphological activation, and redistribution across retinal layers in Rs1^−/y mice compared to WT controls at 8 and 24 weeks. Nuclei were counterstained with Hoechst. Scale bar = 50 μm. E–G Quantification of Iba1-positive microglia showed a significant increase in total microglial number in Rs1^−/y mice compared to WT controls at both 8 and 24 weeks, accompanied by an increase in amoeboid-shaped microglia in the INL at 8 weeks and in the ONL and PRL at 24 weeks. H Sholl analysis revealed reduced microglial branching complexity (intersecting radii) in Rs1^−/y mice. I–K Flow cytometry analysis of CD11b and CD45 expression in dissociated retinal cells showed that both CD11b⁺/CD45^high activated and CD11b⁺/CD45^low resting microglia were significantly increased in Rs1^−/y mice compared to WT controls at both 8 and 24 weeks. L Proportion of activated and resting microglia presented as 100% stacked bar graphs showed a phenotypic shift toward activated microglia in Rs1^−/y mice at both 8 and 24 weeks. Data are presented as mean ± SEM. Statistical analysis was performed using the Kruskal–Wallis test followed by Mann–Whitney U tests for pairwise comparisons. *p < 0.05, **p < 0.01, ***p < 0.001

Results

Comprehensive analysis of structural and functional retinal abnormalities in an Rs1 knockout mouse model

To establish an XLRS mouse model, we developed a CRISPR/Cas9-mediated mouse model with a targeted deletion in the Rs1 gene. We specifically excised a 194 base pair region within exon 2, which is critical for RS1 protein oligomerization [12] and leads to a frameshift mutation and premature stop codon. This genetic modification results in a truncated, non-functional RS1 peptide (MPHKIEGFFLLLLFGYEG), thereby ensuring the complete absence of functional RS1 protein (Fig. 1A). To confirm the loss of RS1 protein, we performed Western blot analysis on retinal tissues from 8-week-old wild-type and knockout (Rs1^−/y) mice. As expected, RS1 protein was clearly detected in the wild-type group, but no RS1 expression was observed in the Rs1^−/y mice (Fig. 1B and Supplementary Figure S1). Furthermore, we conducted immunofluorescence staining on retinal cryosections from both wild-type and Rs1^−/y mice at 8 and 24 weeks of age to assess the localization and expression of RS1 protein (Fig. 1C). In wild-type mice, RS1 immunoreactivity was prominent in the inner segments of photoreceptors, the outer plexiform layer (OPL), and, to a lesser extent, in the INL. This staining pattern suggests RS1 expression in both photoreceptors and inner retinal cells, such as ON bipolar or Müller glial cells. In stark contrast, no RS1 signal was detected in the retinas of Rs1^−/y mice at either age, definitively confirming the complete loss of RS1 expression due to the gene knockout.

We examined the retinal histopathology in our mouse model to understand the structural consequences of functional RS1 loss. We performed hematoxylin and eosin (H&E) staining on retinal sections from wild-type and Rs1^−/y mice at 8 and 24 weeks of age (Fig. 1D). Wild-type mice maintained well-organized retinal architecture with clearly defined layers, including the outer nuclear layer (ONL), INL, and ganglion cell layer, at both time points. In contrast, Rs1^−/y retinas at 8 weeks displayed schisis cavities, primarily within the INL, which is an early and consistent structural hallmark of XLRS. By 24 weeks, these mice showed pronounced thinning of the ONL and overall retinal atrophy, indicating progressive neurodegeneration. Detailed quantification of ONL thickness revealed that while Rs1^−/y retinas showed a trend toward thinning at 8 weeks, the difference was not statistically significant (Supplementary Figure S2A). However, by 24 weeks, a significant reduction in ONL thickness was evident, confirming progressive photoreceptor degeneration (Supplementary Figure S2B). We confirmed these observations using SD-OCT, performed on both eyes of wild-type and Rs1^−/y mice at 8 and 24 weeks of age. In 8-week-old Rs1^−/y mice, we observed distinct hyporeflective schisis cavities within the INL (Fig. 1E), which were absent in wild-type retinas. By 24 weeks, Rs1^−/y mice exhibited severe retinal thinning, indicative of progressive retinal degeneration, while wild-type mice maintained normal retinal architecture and thickness. At this later stage, the distinct laminar structure in Rs1^−/y mice was lost, with a significant reduction in the OPL and shortening of the ONL, INL, outer segments, and inner segments. Quantitative analysis of the schisis cavities confirmed that the cavity area within the INL was significantly larger in 8-week-old Rs1^−/y retinas compared to 24-week-old mice, indicating that structural disruption peaks at the earlier stage (Supplementary Figure S2C).

To compare retinal functions between wild-type and Rs1^−/y mice during XLRS progression, we assessed scotopic ERG responses at 8 and 24 weeks of age. As shown in Fig. 1F, scotopic a-wave amplitudes increased with stimulus intensity in all groups, with 8-week-old wild-type exhibiting the highest responses. In contrast, 24-week-old wild-type, 8-week-old Rs1^−/y, and 24-week-old Rs1^−/y mice all showed significantly reduced a-wave amplitudes compared to the 8-week-old wild-type group (p < 0.01), indicating progressive rod photoreceptor dysfunction in both RS1-deficient and aged retinas. Similarly, scotopic b-wave amplitudes were significantly decreased in 24-week-old wild-type, 8-week-old Rs1^−/y, and 24-week-old Rs1^−/y mice compared to the 8-week-old wild-type group (p < 0.001), with the most pronounced reductions observed in the Rs1^−/y groups. Notably, in 8-week-old Rs1^−/y mice, the b-wave was more severely reduced than the a-wave, suggesting early dysfunction of the INL, particularly ON bipolar cells, prior to photoreceptor degeneration. To further evaluate the balance between inner and outer retinal dysfunction, we calculated the b/a wave ratio at 10 cd·s/m² (Fig. 1G). The b/a wave ratio was significantly reduced in both 8-week-old and 24-week-old Rs1^−/y mice compared to wild-type controls (p < 0.01, Fig. 1G), indicating disproportionately greater inner retinal impairment. In contrast, no significant difference in the b/a ratio was observed between 8-week-old and 24-week-old wild-type mice, suggesting that aging alone does not substantially impair inner retinal function in the presence of functional RS1. Representative ERG traces are shown in Supplementary Figure S2D.

To evaluate functional vision in our Rs1^−/y knockout mice, we performed an OKR test to measure visual acuity (cycles/degree, c/d). As shown in Fig. 1H, visual acuity was significantly reduced in Rs1^−/y mice compared to age-matched wild-type controls at both 8 and 24 weeks of age (p < 0.01). We observed no significant difference between 8-week and 24-week wild-type mice, indicating that aging alone did not substantially affect visual acuity within this timeframe. However, the visual acuity in 8-week-old Rs1^−/y mice was already as low as that in 24-week-old wild-type mice, suggesting that RS1 deficiency leads to early-onset visual impairment. These findings are consistent with our ERG results and confirm that RS1 loss causes early and progressive visual dysfunction.

Retinal cell death and spatiotemporal dynamics of microglial activation in the Rs1 knockout retina

To understand the temporal pattern of retinal cell death in Rs1^−/y mice, we performed TUNEL staining on retinal cryosections from wild-type and Rs1^−/y mice at 8 and 24 weeks of age (Fig. 2A). At 8 weeks, we observed a significant increase in TUNEL-positive cells primarily within the INL of Rs1^−/y retinas, with some also present in the ONL. No TUNEL signal was detected in the wild-type controls. By 24 weeks, the TUNEL-positive cells in Rs1^−/y retinas were mainly localized to the ONL and appeared less frequent compared to the 8-week time point. Quantitative analysis revealed a significant increase in the percentage of TUNEL-positive cells in both the INL and ONL of Rs1^−/y retinas compared to wild-type at 8 weeks (p < 0.05). However, no significant difference was observed at 24 weeks (Fig. 2B). These findings suggest that RS1 deficiency induces early retinal cell death, beginning in the INL around 8 weeks of age and subsequently spreading to the ONL.

To investigate microglial responses in Rs1^−/y retinas, we performed Iba1 immunostaining on both retinal cryosections and whole-mounts (Fig. 2C-L). In wild-type mice, Iba1-positive microglia were consistently restricted to the nerve fiber layer, inner plexiform layer, and OPL at both 8 and 24 weeks (Fig. 2C). These microglia exhibited a ramified morphology, indicative of a resting, immune-surveillant state. In contrast, Rs1^−/y cryosections at 8 weeks showed a noticeable increase in ameboid-shaped microglia within the INL and at the OPL–ONL boundary, alongside prominent INL-centered schisis (Fig. 2C). By 24 weeks, microglia were more broadly distributed across the INL, OPL, ONL, and photoreceptor layer (PRL), displaying a mix of both ramified and ameboid forms. Whole-mount images (Fig. 2D) further supported these observations, revealing ameboid microglia in the INL and ONL at 8 weeks. By 24 weeks, these activated microglia had extended into the PRL, with cavity formation in the INL being more prominent at the 8-week time point.

Quantification of Iba1-positive microglia revealed a significant increase in total microglial numbers in Rs1^−/y mice at both 8 and 24 weeks, with a more pronounced difference at 8 weeks (Fig. 2E). We observed an increase in ameboid microglia in the INL at 8 weeks, and in the ONL and PRL at 24 weeks, indicating temporally regulated microglial activation across different retinal layers (Fig. 2F–G). Sholl analysis further showed a reduced number of intersecting radii in Rs1^−/y microglia, particularly at 8 weeks, reflecting an early loss of their characteristic ramified, resting morphology (Fig. 2H).

To assess functional activation of microglia, we analyzed CD11b⁺/CD45^high (activated) and CD11b⁺/CD45^low (resting) microglia using flow cytometry (Fig. 2I). Both activated and resting microglial populations were significantly elevated in Rs1^−/y mice at both 8 and 24 weeks (Fig. 2J-K). Importantly, the proportion of activated cells was markedly higher in the RS1-deficient retina (Fig. 2L), confirming a significant shift toward an activated phenotype in these mice.

Preferential phagocytosis of bipolar cells by activated microglia in RS1-deficient retina

To investigate retinal cell-type-specific changes in our Rs1^−/y mouse model, we performed scRNA-seq on retinas from both wild-type and Rs1^−/y mice (Fig. 3). We successfully identified all major retinal cell populations based on canonical marker gene expression (Fig. 3A and Supplementary Figure S3). Comparative analysis of cell type proportions revealed a significant reduction in cone bipolar cells in Rs1^−/y retinas (3.1%) compared to wild-type controls (13.3%, p = 0.008) at 8 weeks (Fig. 3B). Proportions of other retinal cells, including photoreceptors, remained unchanged (p > 0.05). While microglia represented a relatively low proportion of total retinal cells and thus did not reach statistical significance, their proportion in Rs1^−/y mice increased six-fold (1.3%) compared to wild-type (0.2%), indicating a notable expansion of the microglial population (Fig. 3B).

Fig. 3.

Single-cell profiling of retinal cell populations and microglial dynamics in the Rs1^−/y mouse. A UMAP visualization of retinal cell populations in wild-type (WT) and Rs1^−/y mouse at 8 weeks. Integrated single-cell RNA sequencing analysis was performed using Seurat package (version 5.2.0) to compare retinal cell compositions between genotypes. Cell clusters were identified and annotated based on the expression of well-established cell type-specific marker genes. B Comparison of cell type proportions between WT and Rs1^−/y retinas. The bar graph displays the relative abundance of each identified cell type, calculated as a percentage of the number of total cells in each sample. Statistical significance of the differences in cellular composition between the two groups was assessed using a Chi-square test. C Gene Ontology (GO) analysis of Biological Processes. Enrichment analysis was performed to identify biological processes associated with differentially expressed genes. The bar graphs display the most significant terms for genes down-regulated in Rs1^−/y cone-bipolar cells (top) and genes up-regulated in Rs1^−/y microglia (bottom) relative to WT. D Temporal analysis of rod photoreceptor and cone-bipolar cell-derived transcripts retained within microglia. The scatter plot compares the presence of non-microglial transcripts within the microglia population between 8 and 24 weeks. The x-axis represents the temporal dimension (left: 8 weeks, right: 24 weeks), and the y-axis discriminates the cellular origin, with photoreceptor-derived transcripts (upper) and cone-bipolar cell-derived transcripts (lower). The number of differentially expressed genes with an absolute log₂FoldChange > 1 and the corresponding percentile among 281 DEGs (FDR < 0.05) are indicated in each quadrant. E UMAP visualization of microglial sub-clusters. The microglia population was subsetted from the global dataset and re-clustered to visualize distinct subpopulations within WT and Rs1^−/y samples at 8 weeks and 24 weeks. F Trajectory analysis of microglial sub-clusters. Lineage inference was performed using Slingshot package (version 2.18.0) to construct a minimum spanning tree on the clusters. Arrows indicate the global lineage structure used to identify pseudotime-dependent gene expression changes (FDR < 0.05). G Pseudotime expression dynamics of bipolar cell- and rod photoreceptor-derived transcripts. The graph illustrates the changes in transcript abundance (log counts) along the microglial trajectory. P_bipolar and P_rod denote the pseudotime points corresponding to the peak expression of bipolar- and rod-derived transcripts, respectively

Differential gene expression (DEG) analysis revealed significant changes in gene activity within Rs1^−/y retinas. At 8 weeks, pathways related to the regulation of synaptic signaling and synaptic assembly were notably enriched among the genes downregulated in Rs1^−/y cone bipolar cells compared to wild-type (Fig. 3C and supplementary Figure S4). Conversely, microglia in Rs1^−/y retinas at 8 weeks showed an upregulation of genes associated with phagocytosis, response to cytokine and cell migration compared to wild-type (Fig. 3C and supplementary Figure S5). This indicates an activated and migratory state of microglia, consistent with their role in responding to retinal pathology. To determine the temporal sequence of neuronal loss, we analyzed microglial transcriptomes to identify cell-derived transcripts. Microglia from 8-week Rs1^−/y retinas were enriched with bipolar cell–derived transcripts, suggesting that bipolar cells are a primary target of early microglial activity. In contrast, 24-week microglia predominantly contained rod photoreceptor–derived transcripts (Fig. 3D), indicating a later involvement of rod photoreceptor degeneration and subsequent microglial engagement.

To further investigate these temporal changes in microglial behavior, we performed microglial subclustering and trajectory analysis (Fig. 3E and F). As anticipated from earlier findings, the number of microglia increased in Rs1^−/y retinas compared to wild-type at both 8 and 24 weeks (Fig. 3E and Supplementary Figure S6A). In wild-type mice, Cluster 4 represented the predominant microglial population, consistent with a resting state (Fig. 3E and Supplementary Figure S6B). However, in Rs1^−/y mice, we observed an increase in various activated microglial subtypes. Specifically, Cluster 0 was present in Rs1^−/y microglia at 8 weeks and maintained by 24 weeks. Conversely, Cluster 2 increased in Rs1^−/y microglia at 24 weeks compared to 8 weeks (Fig. 3E and Supplementary Figure S6). Pseudotime trajectory analysis further elucidated this temporal progression, revealing that Cluster 0 emerged earlier than the cluster 2 (Fig. 3F). This temporal pattern was reinforced by the expression levels of bipolar cell markers, such as Isl1 and Grm6, which showed a peak in Cluster 0 (P_bipolar in Fig. 3G). This peak appeared earlier than Cluster 2, which contained rod photoreceptor-specific transcripts (P_rod), Rcvrn and Rho, further solidifying the sequence of cellular events.

OFF-cone bipolar cells were more largely reduced than ON- bipolar cells in Rs1^−/y retinas by microglial phagocytosis

To investigate if microglia specifically target certain bipolar cell subtypes in the RS1-deficient retina, we conducted a detailed cluster analysis of bipolar cells, identifying 10 distinct clusters. These clusters were categorized into rod bipolar, ON cone bipolar, and OFF cone bipolar cells using established gene markers (Supplementary Figure S7A and B). Our analysis focused on cone bipolar cells (Fig. 4A), as the number of rod bipolar cells was consistent between wild-type and Rs1^−/y mice (Fig. 3B). We observed a reduction in cone bipolar cells in the Rs1^−/y retina compared to wild-type at both 8 and 24 weeks (Fig. 4B and Supplementary Figure S7C). Notably, OFF cone bipolar cells showed a significantly greater reduction in Rs1^−/y mice at 8 weeks (dropping from 36.4% in wild-type to 5.0%) compared to ON cone bipolar cells (which decreased from 63.6% to 28.2%). Similar trends were observed at 24 weeks. To validate these molecular findings, we performed immunofluorescence staining on retinal sections from wild-type and Rs1^−/y mice, using markers for ON bipolar cells (Goα) and OFF bipolar cells (PKARIIβ). Iba1⁺ microglia were consistently found in close association with PKARIIβ-positive cells (Fig. 4C). Crucially, we observed microglia visibly engulfing PKARIIβ-positive elements, while remaining spatially separated from Goα-positive ON bipolar cells (Fig. 4D).

Fig. 4.

Single-cell RNA sequencing and immunofluorescence reveal cone bipolar cell dysregulation. A Identification of cone bipolar cell clusters. UMAP projection showing distinct clusters of OFF cone bipolar cells (cyan) and ON cone bipolar cells (orange) from Rs1^−/y mouse retinas. B Relative proportion of cone bipolar cell populations. Quantification of the relative proportion of OFF cone bipolar cells (cyan) and ON cone bipolar cells (orange) in WT and Rs1^−/y retinas at 8 and 24 weeks, demonstrating a more significant reduction in OFF cone bipolar cells in Rs1^−/y mice. C-D Microglial engulfment of PKARIIβ- and PKCα-positive bipolar cells. Both PKARIIβ- (C) and Goα-positive cells (D) were engulfed by Iba1-positive microglia (red asterisks). Scale bar = 20 μm

Phagocytosis of OFF-cone bipolar cells by microglia via phosphatidylserine recognition but independent to cleaved caspase-3 in Rs1^−/y retinas

To investigate the molecular mechanisms underlying cell removal, we first examined caspase expression profiles across bipolar cell subtypes. As shown in Fig. 5A, Casp3 was predominantly expressed in OFF-cone bipolar cells. However, despite this enrichment, we found that only a subset (~ 35%) of ON- and OFF-cone bipolar cells co-expressed the pro-apoptotic gene Trp53 and Casp3 (Supplementary Figure S8A and S8B). This aligns with our earlier observation in Fig. 2, which showed less than 10% TUNEL⁺ cells even in the INL of the Rs1^−/y retina. These findings suggest that classical apoptosis may not be the sole or primary mechanism driving the substantial loss of bipolar cells.

Fig. 5.

Caspase-3 expression and non-apoptotic phagocytosis of OFF-cone bipolar cells by microglia. A Dot plot showing the expression of various caspase genes across bipolar cell subtypes (Rod-BC, ON-CBC, OFF-CBC) from single-cell RNA-seq data. The size of the dot represents the percentage of cells expressing the gene, and the color intensity indicates the average expression level. B Quantification of microglial phagocytosis preference. The box plot displays the percentage of Rod-BCs, ON-CBCs, and OFF-CBCs found within Iba1 + microglia, stratified by cleaved caspase-3 (CC3) positivity. *p < 0.05, **p < 0.01, ***p < 0.001. C Engulfment of PKARIIβ-positive OFF bipolar cells by microglia in retinal cryosections of Rs1^−/y Mice. In retina of Rs1^−/y mice, PKARIIβ-positive OFF bipolar cells were engulfed by Iba1-positive microglia, regardless of cleaved caspase-3 (CC3) expression. Scale bar = 50 μm. D Microglial phagocytosis of CC3-positive and -negative PKARIIβ-positive cells in retinal flatmounts. This panel confirms microglial phagocytosis of both CC3-positive (yellow arrowheads) and CC3-negative (white asterisks) PKARIIβ-positive cells. Nuclei are counterstained with Hoechst. Scale bar = 20 μm. Rod-BC, rod bipolar cell; ON-CBC, ON-cone bipolar cell; OFF-CBC, OFF-cone bipolar cell

To directly visualize the cell removal process, we performed immunofluorescence staining for cleaved caspase-3 (CC3) alongside the microglial marker Iba1 in 8-week-old Rs1^−/y retinas (Fig. 5B - D). Interestingly, we observed two distinct patterns of microglial engagement. While some Iba1 + microglia were associated with CC3⁺ OFF-bipolar cells (indicating the clearance of apoptotic cells), we also frequently observed Iba1 + microglia engulfing CC3-negative OFF-bipolar cells (Fig. 5B, C and Supplementary Figure S8C). This finding suggests that microglia may actively phagocytose compromised bipolar cells via non-apoptotic mechanisms during the early stages of retinal degeneration, in addition to clearing apoptotic debris. CC3⁺ cells were not detected in wild-type mice. These observations were further confirmed by whole-mount triple staining (Fig. 5D), where both CC3⁺ and CC3⁻ OFF bipolar cells were found engulfed within Iba1⁺ microglia, providing compelling evidence for the phagocytic removal of non-apoptotic OFF bipolar cells by microglia in the Rs1^−/y retina.

To investigate the mechanisms driving this non-apoptotic removal of OFF bipolar cells, we focused on phosphatidylserine (PS). Our single-cell RNA-seq analysis confirmed that cell viability metrics, such as the number of features (genes) and Malat1 expression, were comparable between rod-, ON-cone and OFF-cone bipolar cells (Fig. 6A). However, we found a substantially higher expression of Xkr4 in OFF-cone bipolar cells compared to rod- or ON-cone bipolar cells (Fig. 6B). As Xkr4 is involved in presenting PS on the cell surface as an 'eat-me' signal in the central nervous system [33], this suggests that OFF cone bipolar cells might be actively eliminated from the INL primarily through phagocytosis rather than apoptosis.

Fig. 6.

Phosphatidylserine exposure-mediated microglial clearance of live OFF-bipolar cells in Rs1^−/y retinas. A Comparison of cell viability metrics between Rod bipolar cells (Rod-BC), ON-cone bipolar cells (ON-CBC), and OFF-cone bipolar cells (OFF-CBC). Violin plots display quality control metrics from the single-cell RNA-seq dataset. Two key parameters indicative of cell health were assessed: the number of detected features per cell (left) and the expression levels of the nuclear-retained transcript Malat1 (right). B Violin plots showing the expression level of Xkr4 across bipolar cell subtypes. Note the specific upregulation of Xkr4 in OFF-CBCs. C Phosphatidylserine (PS) exposure on PKARIIβ-positive OFF-CBC in Rs1^−/y retinas. In Rs1^−/y retinas, PS signal co-localized with PKARIIβ, a marker of OFF-CBC (upper panel, white asterisks), but not with PKCα, a marker of rod-BC (lower panel). Scale bar = 20 μm. D Confirmation of microglial engulfment of PS-exposing cone OFF-CBC. Retinal flatmounts demonstrate co-localization of PKARIIβ, PS, and Iba1, indicating microglial engulfment of PS-exposing cone OFF-CBC (upper panels, red asterisks). In contrast, PS does not co-localize with PKCα + rod-BC (lower panels, white asterisks). Nuclei were counterstained with Hoechst. Scale bar = 20 μm

To confirm this molecular finding, we performed fluorescence staining for PS using FITC-conjugated Annexin V on retinal sections and whole-mount preparations from 8-week-old wild-type and Rs1^−/y mice. In the Rs1^−/y retinas, PKARIIβ-positive OFF bipolar cells frequently co-localized with PS (Fig. 6C, upper panel). This strongly indicates that these cells were externalizing PS, an "eat-me" signal, even in the early stages of degeneration. In stark contrast, PKCα-positive rod bipolar cells showed no PS signal (Fig. 6C, lower panel). These observations were further corroborated by whole-mount triple staining (Fig. 6D), where PS⁺/PKARIIβ⁺ cells were clearly located within Iba1⁺ microglial profiles, providing direct evidence for the phagocytic removal of OFF bipolar cells exposing PS. As expected, PKCα⁺ rod bipolar cells lacked PS exposure and showed no spatial association with Iba1⁺ microglia (Fig. 6D, lower panel).

Selective activation of complement C3b in OFF-bipolar cells in the RS1-deficient retina

It's known that microglia can remove stressed-but-viable neurons through PS recognition and complement C3b activation [34]. To investigate if this mechanism applies in our model, we examined the expression of Itgam and Itgb2, genes encoding the complement receptor alpha and beta subunits, respectively, within the trajectory of the retinal microglia transcriptome. As shown in Fig. 7A, both genes were highly expressed in cluster 0. This cluster corresponds to a peak pseudotime that was significantly enriched with bipolar cell-derived transcripts (P_bipolar), indicating a specific microglial state associated with bipolar cell interactions.

Fig. 7.

Complement system involvement in microglial phagocytosis of OFF-bipolar cells. A Microglial mRNA expression of complement receptor subunits in pseudotime trajectory. Microglial mRNA expression levels of Complement 3 receptor subunit alpha (Itgam) and beta (Itgb2) are shown in a pseudotime trajectory derived from single-cell RNA sequencing data. Expression of both genes is relatively high at pseudotime points enriched by microglia containing bipolar cell-derived transcripts (P_bipolar). B Selective co-localization of complement C3b with cone OFF-Bipolar Cells in Rs1^−/y Mice. Immunofluorescence staining of retinal cryosections from 8-week-old mice shows that complement C3b selectively co-localizes with PKARIIβ-positive cone OFF-bipolar cells in Rs1^−/y retinas (upper panel), but not in wild-type (WT) controls. In contrast, complement C3b does not co-localize with PKCα-positive rod ON-bipolar cells in either Rs1^−/y or WT retinas (lower panel). Nuclei were counterstained with Hoechst. Scale bar = 20 μm

To confirm this molecular finding, we evaluated complement C3b activation around ON and OFF bipolar cells in the Rs1^−/y retina. Prominent C3b activation was observed around OFF bipolar cells in Rs1^−/y retinas, but not in wild-type controls (Fig. 7B, upper panel). In contrast, C3b expression was barely detectable around ON bipolar cells in Rs1^−/y retinas and showed little difference compared to wild-type controls (Fig. 7B, lower panel). These results strongly suggest local complement activation specifically targets OFF cone bipolar cells.

Impaired synaptic assembly and reduced Neuregulin-1 signaling in OFF-cone bipolar cells of Rs1^−/y retina

Gene ontology (GO) analysis of differentially expressed genes in OFF-cone bipolar cells of 8-week-old Rs1^−/y mice compared to wild-type revealed distinct molecular alterations (Fig. 8). Upregulated genes were associated with inflammatory and immune responses, such as regulation of monocyte differentiation, negative regulation of response to wounding, chemotaxis, and cytokine signaling (Fig. 8A). Conversely, downregulated genes were significantly enriched in pathways related to positive regulation of TGF-β signaling, detection of light stimulus, and notably, synapse assembly and maintenance (Fig. 8B). Specifically, among the genes related to synapse assembly, Nrg1 which encodes Neuregulin-1 (NRG1), a growth factor vital for synaptic maintenance and neurotrophic signaling [35], were significantly downregulated and fewer cells expressed these genes compared to wild-type mice (Fig. 8C and Supplementary Figure S9).

Fig. 8.

Downregulation of synaptic genes and Neuregulin-1 in OFF-cone bipolar cells of Rs1^−/y mice. A Gene ontology of upregulated genes in Rs1^−/y OFF bipolar cells compared to WT OFF bipolar cells. B Gene ontology of downregulated genes in Rs1^−/y OFF Bipolar Cells. C Dot plot of synapse assembly-related genes. Expression levels and percentage of cells expressing selected synapse assembly-related genes (e.g., Cntn5, Bdnf, Lrtm2, Gpm6a, Cstn2, Pcdh17, Adgrg3, Nrg1, Gria1) in WT and Rs1^−/y retinas. D Immunofluorescence staining of NRG1 in ON-and OFF-bipolar cells. Retinal cryosections showing co-localization of PKARIIβ (green), a marker for OFF bipolar cells, with NRG1 (red) in WT mice, and reduced NRG1 expression in Rs1^−/y mice (upper panels). In contrast, PKCα (green), a marker for ON bipolar cells, does not co-localize with NRG1 (red) in either WT or Rs1^−/y retinas (lower panels). Nuclei are counterstained with Hoechst (blue). Scale bar = 10 μm

To validate these molecular findings, we performed immunofluorescence staining on retinal sections from wild-type and Rs1^−/y mice, using markers for rod bipolar cells (PKCα) and OFF bipolar cells (PKARIIβ), alongside NRG1. Our analysis revealed that PKARIIβ-positive OFF-cone bipolar cells in the Rs1^−/y retina exhibited severe structural disorganization and a marked reduction in NRG1 signal intensity (Fig. 8D, upper panel). In contrast, PKCα-positive rod bipolar cells in the Rs1^−/y retina showed less disrupted cellular alignment and morphology, with no obvious reduction in NRG1 expression compared to wild-type (Fig. 8D, lower panel). The selective reduction of NRG1 in PKARIIβ-positive regions suggests that RS1 deficiency specifically compromises OFF-cone bipolar cells by disrupting their local neurotrophic support mechanisms, potentially contributing to their vulnerability and synaptic dysfunction.

Discussion

In this study, we aimed to elucidate the pathogenesis of XLRS using a Rs1-exon2 knockout mouse model, in comparison with wild-type controls. We observed characteristic schisis in the INL of the Rs1^−/y mouse retina, accompanied by a more pronounced reduction in the ERG b-wave relative to the a-wave, as well as significantly impaired visual acuity. Notably, Rs1^−/y mice exhibited a marked increase in microglial infiltration and activation within the INL at 8 weeks of age, which subsequently shifted to the ONL and PRL by 24 weeks. Single cell RNA-seq analysis revealed that, although the overall number of photoreceptors in Rs1^−/y mice was comparable to that in wild-type mice, cone bipolar cells were significantly reduced at both 8 and 24 weeks of age. Importantly, our data indicated that cone bipolar cells were phagocytosed by microglia prior to photoreceptor degeneration. Among the cone bipolar cell subtypes, OFF-cone bipolar cells exhibited a more substantial reduction than ON-cone bipolar cells in Rs1^−/y mice. Interestingly, these OFF-cone bipolar cells showed low expression of Trp53 and Casp3, but high expression of Xkr4. Immunohistochemistry further confirmed that OFF-cone bipolar cells were preferentially targeted by microglial phagocytosis over ON-bipolar cells in the INL in Rs1^−/y mice at 8 weeks. This process appeared to be independent of cleaved caspase-3 but dependent on complement component C3b. Collectively, our findings, for the first time, suggest that microglia-mediated removal of OFF-cone bipolar cells is an early pathological event that precedes photoreceptor degeneration in XLRS.

We developed a mouse model with exon 2 deletion, which results in the complete loss of functional RS1 protein. Several RS1-deficient animal models have been previously established to explore the pathogenesis of XLRS. One well-characterized model is the Rs1h^−/y mouse, which carries a complete Rs1 deletion and displays retinal layer splitting, synaptic dysfunction, and early, robust upregulation of immune response genes indicative of microglia-driven proinflammatory activation [20]. In a separate model, the Rs1-exon1 knockout rat exhibited a very early onset and rapidly progressive photoreceptor degeneration, with outer limiting membrane disruption by postnatal day 15 (P15), mislocalization of photoreceptor nuclei into the subretinal space, significantly reduced dark-adapted ERG a- and b-wave amplitudes, and activation of microglia and Müller glia as early as P7 [36]. Similarly, the Rs1-exon3 knockout rat developed schisis cavities by P15, photoreceptor displacement into the subretinal space and OPL, photoreceptor loss by P21, and profound reductions in both a- and b-wave amplitudes by P28, accompanied by impaired synaptic transmission to bipolar cells [37]. In addition, point mutation models of Rs1, including C59S and R141C, have been reported, all consistently reproducing key features of human XLRS, such as intraretinal schisis and selective b-wave reduction [38]. Most recently, a mouse model carrying the patient-derived RS1 missense mutation R213W was shown to faithfully recapitulate the human XLRS phenotype [39]. Our Rs1-exon2 knockout mouse model not only confirms the core pathological features shared with previous RS1-deficiency models but also provides novel insights into the cellular and molecular mechanisms, particularly highlighting the roles of microglia and bipolar cells in disease progression.

Microglia are the primary resident immune cells and macrophages of the central nervous system, including the retina [40]. In the healthy retina, microglia are typically localized within the plexiform layers, where they actively monitor the retinal microenvironment for pathogens, dead cells, and protein aggregates to maintain tissue homeostasis and ensure the functional integrity of the retina [41, 42]. Importantly, they are also capable of phagocytosing stressed but still viable cells within the neural retina [43]. However, upon activation—particularly when adopting a pro-inflammatory (M1) phenotype—microglia secrete a range of pro-inflammatory cytokines and chemokines, including TNF-α, IL-1β, IL-6, and IFN-γ [44]. These inflammatory mediators can induce chronic inflammation and exacerbate neurodegeneration, thereby contributing to the progression of retinal damage [44]. In models of retinitis pigmentosa (RP), a distinct retinal degenerative disease, microglia have been shown to promote photoreceptor loss by actively phagocytosing stressed, yet still viable, rod cells [45]. This process is initiated by the exposure of 'eat-me' signals on dysfunctional photoreceptors, even prior to the onset of apoptosis [45]. Although XLRS presents a different primary pathology compared to RP, the progressive photoreceptor degeneration and early microglial activation observed in XLRS models [20, 36] suggest that a similar mechanism—microglial-mediated removal of stressed but not yet fully degenerated retinal cells—may contribute to disease progression and vision loss in XLRS.

Our results demonstrated that microglial infiltration and activation predominantly occurred in the INL of Rs1^−/y mice at 8 weeks, subsequently shifting to the ONL and PRL by 24 weeks. This spatial transition suggests that microglia primarily target INL cells before engaging photoreceptors in the ONL. Supporting this, transcriptomic analyses of microglia revealed higher levels of bipolar cell-specific transcripts at 8 weeks, which declined by 24 weeks, while rod cell-specific transcripts became more prominent at the later time point. These findings indicate that microglial phagocytosis of bipolar cells precedes photoreceptor degeneration. Coupled with the significantly reduced b/a-wave ratio observed in Rs1^−/y mice at 8 weeks, our data strongly suggest that RS1 deficiency leads to early inner retinal dysfunction, with bipolar cell degeneration emerging prior to photoreceptor loss. Although bipolar cell degeneration in XLRS has often been overshadowed by the focus on photoreceptor degeneration, accumulating evidence supports the involvement of bipolar cells in XLRS pathology. In a RP model, the development and degeneration of bipolar cells were reported to occur independently of photoreceptors [46]. Additionally, schisis cavities in XLRS are most prominent in the foveomacular region, typically affecting the inner retina, including the OPL and INL [47]. This structural splitting disrupts the highly organized neuronal architecture, directly impacting bipolar cells, which reside in the INL and extend dendrites into the OPL to form synapses with photoreceptors [48]. Recent studies demonstrated that targeted Rs1 expression in bipolar cells of Rs1-KO mice resulted in more densely packed bipolar cells, fewer schisis cavities, and notable improvements in inner retinal structure and synaptic function [21]. Clinically, XLRS patients typically exhibit a marked reduction in b-wave amplitude, which reflects bipolar cell function, while the a-wave—primarily generated by photoreceptors—remains relatively preserved [49]. Moreover, a longitudinal study of ERG changes in Rs1h-KO mice from 1 to 16 months showed that b-wave amplitudes were significantly reduced as early as 1 month, with the greatest decline occurring between 1 and 4 months, while a-wave reduction progressed more gradually with age [50]. Notably, the b/a-wave ratio was lowest at 4 months and increased rapidly between 4 and 8 months [50]. These observations align with our findings and further support the notion that bipolar cell degeneration is more rapid than photoreceptor degeneration in RS1-deficient mice within 4 months postnatally.

Caspases serve as central effectors of apoptosis and have been implicated in retinal degeneration; however, the specific contributions of different caspase subtypes in RS1-deficient retinas remain complex. Previously, Gehring et al. reported elevated mRNA and protein levels of caspase-1, but not caspase-3, in Rs1h^−/Y retinas using whole retinal extracts [51]. However, relying on whole retinal lysates may mask cell-type-specific signals, such as those in bipolar cells, and their study did not distinguish between the inactive pro-enzymes and the cleaved active forms. Furthermore, a recent study by Gehrke et al. demonstrated that genetic ablation of caspase-1/11 in Rs1-KO mice failed to rescue the phenotype, showing no improvement in cyst severity, ONL thickness, or ERG amplitudes [52]. This suggests that caspase-1 may not be the primary driver of degeneration in this context. In contrast to previous findings focused on photoreceptors, our study specifically identified cleaved caspase-3 activation within bipolar cells. Although caspase-3 has been traditionally associated with photoreceptor degeneration in models like rd1 mice [53], recent evidence supports its activation in the inner retinal layers, such as bipolar cells, in traumatic brain injury models [54]. Our observation of specific caspase-3 activation in bipolar cells (Fig. 5) highlights a distinct degenerative mechanism in the inner nuclear layer that may have been overlooked in bulk tissue analyses. Interestingly, we also observed signs of degeneration that appeared independent of caspase-3. This aligns with the concept proposed by Zeiss et al., who suggested that retinal cell death can proceed via caspase-independent mechanisms even in the absence of caspase-3 [55]. Collectively, our findings imply that bipolar cell degeneration in Rs1^−/y retinas is likely multifaceted, involving both caspase-3-dependent apoptosis and caspase-independent non-apoptotic pathways. Future studies are warranted to dissect the precise interplay between these cell death mechanisms and their contribution to functional loss.

A critical question arising from our findings is whether microglial activation acts as the primary driver of bipolar cell degeneration or occurs secondarily to intrinsic neuronal damage. While our data demonstrate that microglia actively engulf non-apoptotic OFF-cone bipolar cells via the PS-dependent pathway, the precise causal sequence warrants careful interpretation. One plausible scenario is that microglial activation is a secondary response to Rs1 deficiency-induced synaptic instability. RS1 is an extracellular adhesion protein crucial for maintaining the structural integrity of the synaptic cleft in the outer plexiform layer [56]. Its absence may lead to subtle synaptic disorganization or stress in bipolar cell dendrites, which could trigger "eat-me" signals like PS exposure even before frank apoptosis occurs. In this context, microglia would be sensing and eliminating these "stressed but viable" neurons—a phenomenon known as "phagoptosis" or primary phagocytosis [24, 34]. This implies that while the initial trigger is intrinsic to the bipolar cell (loss of RS1), microglia exacerbate the loss by prematurely removing potentially salvageable cells. Conversely, it is also possible that microglia are activated directly by the retinoschisis microenvironment. The physical separation of retinal layers (schisis) creates mechanical stress and may release damage-associated molecular patterns (DAMPs), priming microglia into a reactive state [57]. These activated microglia could then become neurotoxic, actively targeting vulnerable neuronal populations such as OFF-cone bipolar cells. These possible mechanisms under especially interaction between bipolar cell and microglia in INL degeneration of XLRS should be evaluated in further study. Regardless of the initiation mechanism, our identification of non-apoptotic phagocytosis suggests that targeting this immune interaction could offer a novel window for preserving retinal function in early-stage XLRS.

Interestingly, our findings revealed that OFF-cone bipolar cells were more susceptible to microglial phagocytosis than ON-cone bipolar cells. scRNA-seq analysis showed a greater reduction in the number of OFF-cone bipolar cells compared to ON-cone bipolar cells, and immunohistochemical analysis demonstrated that OFF-cone bipolar cells exhibited more frequent colocalization with microglia within the INL. In some OFF-cone bipolar cells, microglial phagocytosis appeared to be mediated by PS translocation to the cell surface and complement C3b activation, yet this process was independent of cleaved caspase-3, suggesting that microglia may be actively engulfing stressed but viable cells—a process known as ‘phagoptosis’—in parallel with efferocytosis. Unlike classical apoptosis or necrosis, where cell death precedes clearance, phagoptosis refers to the targeted removal of living cells that have aberrantly exposed "eat-me" signals including PS and C3b or have lost "don’t-eat-me" signals due to cellular stress, damage, or aging [34, 43]. This inappropriate removal of viable neurons is increasingly recognized as a significant contributor to neuronal loss in various neurological diseases, including those occurring during brain development, as well as in conditions involving inflammation, ischemia, and neurodegeneration [34]. In pathological states, abnormal microglial activation and excessive phagocytosis can directly drive neurodegeneration [43]. Moreover, exposure to low levels of inflammatory stimuli, such as TNF-α, glutamate, superoxide, nitric oxide, or peroxynitrite, can induce neuronal stress sufficient to trigger the presentation of "eat-me" signals, rendering cells vulnerable to phagoptosis by microglia [43]. This mechanism has been directly implicated in retinal degeneration, as demonstrated in RP models, including rd10 mice, where microglial phagocytosis of stressed yet living rod photoreceptors significantly contributes to their degeneration [45]. Live-cell imaging studies have shown that infiltrating microglia can dynamically interact with photoreceptors and rapidly engulf rods that have not yet entered apoptosis [45]. Similarly, our findings suggest that in the context of functional RS1 deficiency in XLRS, synaptic disruption may directly predispose stressed bipolar cells—particularly OFF-cone bipolar cells—to active phagocytic removal by microglia, representing a key mechanism of early neuronal loss in this disease.

This study has several limitations. First, we selected 8- and 24-week-old mice based on prior longitudinal ERG studies in Rs1h-KO mice, which reported a more rapid decline in the b-wave compared to the a-wave within the first four months after birth [50]. However, other RS1-deficient mouse models have demonstrated key pathological features, such as retinal schisis and photoreceptor degeneration, as early as the first four weeks postnatally [20, 36, 37]. As a result, our study may not fully capture the earliest pathophysiological events of XLRS. Future studies should investigate microglial activation and bipolar cell changes at earlier stages in this model to better understand the initial disease mechanisms. Second, our single-cell RNA-seq analysis did not include biological replicates for each group, which raises the possibility that our findings could have been influenced by batch effects or individual variability. To strengthen the robustness of our conclusions, additional independent experiments with proper replication are needed. Finally, although we propose that live bipolar cells may be actively targeted by microglial phagocytosis mediated by PS translocation and complement C3b activation, the precise cellular and molecular pathways underlying this process remain to be elucidated. It is particularly important to determine whether bipolar cell vulnerability is directly caused by the absence of functional RS1, whether alterations in ‘don’t-eat-me’ or other ‘eat-me’ signals occur on the bipolar cell surface, and whether these cells are truly viable at the time of microglial engulfment. Addressing these questions will be essential for fully characterizing the early degenerative processes in XLRS.

In conclusion, our study provides novel insights into the early pathogenesis of XLRS. We demonstrate that bipolar cell degeneration, particularly OFF-cone bipolar cells, precedes photoreceptor degeneration in Rs1-exon2 knockout mice. This degeneration appears to involve active microglial phagocytosis of viable bipolar cells, potentially through mechanisms such as phosphatidylserine exposure and complement activation, independent of classical apoptotic pathways. These findings highlight the previously underappreciated role of bipolar cell vulnerability and microglial activity in the progression of XLRS. Our results not only advance the understanding of XLRS pathophysiology but also suggest that therapeutic strategies aimed at preserving bipolar cells or modulating microglial phagocytic activity may offer promising avenues for early intervention in XLRS.

Abbreviations

CC3

Cleaved caspase-3

DEG

Differential gene expression

ERG

Electroretinography

GCL

Ganglion cell layer

H&E staining

Hematoxylin and eosin staining

INL

Inner nuclear layer

IPL

Inner plexiform layer

IS.

Inner segments

OKR

Optokinetic Response

OMR

Optomotor response

ONL

Outer nuclear layer

OPL

Outer plexiform layer

PCA

Principal Component Analysis

PFA

Paraformaldehyde

PRL

Photoreceptor layer

PS

Phosphatidylserine

RP

Retinitis pigmentosa

RPE

Retinal pigment epithelium

scRNA-seq

Single-cell RNA sequencing

SD-OCT

Spectral-domain optical coherence tomography

TUNEL assay

Terminal Deoxynucleotidyl Transferase dUTP Nick End Labeling Assay

UMAP

Uniform Manifold Approximation and Projection

XLRS

X-linked juvenile retinoschisis

Authors’ contributions

J.Y. Yang: conceptualization, data curation, formal analysis, investigation, methodology, resources, validation, visualization, and writing (original draft and editing). H.S. Chang: conceptualization, data curation, single cell transcriptome analysis, investigation, methodology, visualization, and writing (original draft, review and editing). Y.J. Kim: data curation, formal analysis, investigation, methodology, software, validation, and visualization. S. An: investigation and methodology. H.S. Park: investigation and methodology. J.H. Kim: investigation and methodology. J.W. Han: formal analysis, investigation, and methodology. S.S. Paik: investigation and data curation. J. Lyu: data curation. I.B. Kim: investigation. T.K. Park: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, and writing (original draft, review and editing).

Funding

This work was supported by the Bio&Medical Technology Development Program of the National Research Foundation (NRF) funded by the Korean Government (MSIT) (grant number RS-2025–02217948); the National Research Council of Science and Technology (NST) (grant number GTL24022-000), Republic of Korea; and supported by the Soonchunhyang.

Data availability

The single cell RNA sequence dataset supporting the conclusions of this article is available in the NCBI's Gene Expression Omnibus repository, GEO Series accession number GSE 303947 ( https:/www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE303947).

Declarations

Ethics approval and consent to participate

All procedures were approved by the IACUC of Soonchunhyang University (protocol SCHBCA 2023–07).

Competing interests

The authors declare no competing interests.