St6gal1-mediated sialylation protects retinal ganglion cells by restraining microglial phagocytosis

Abstract

The death of retinal ganglion cells (RGCs) is a pivotal pathological event that leads to irreversible vision loss following optic nerve injury. RGCs display substantial heterogeneity, with distinct subtypes exhibiting variable resilience to injury, but the mechanisms underlying the intrinsic survival advantage remain to be defined. Here, by analyzing mouse RGC single-cell RNA sequencing datasets, we found that αRGCs expressed markedly higher levels of sialyltransferases, particularly ST6Gal1. In the mouse optic nerve crush (ONC) model, we confirmed that blockade of sialylation significantly reduced αRGC survival. Conversely, overexpression of ST6Gal1 enhanced the survival of intrinsically photosensitive RGCs and other RGCs after ONC, and conferred broad neuroprotection in additional in vivo mouse retinal injury models including I/R and NMDA excitotoxicity. Notably, overexpressing ST6TGal1 promoted optic nerve axon regeneration independent of mTOR signaling activation. Mechanistically, single-cell transcriptomic profiling revealed that ST6Gal1-mediated sialylation potentially facilitated the transition of microglia toward a homeostatic state after injury, thereby attenuating phagocytic clearance of RGCs. Together, these findings uncover a previously unrecognized sialylation-mediated regulatory axis that promotes RGC survival and optic nerve regeneration, offering a neuron-directed strategy to mitigate microglia-driven neurodegeneration.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12974-026-03736-z.

Background

Retinal ganglion cell (RGC) death following optic nerve injury is a critical pathological event in neurodegenerative diseases such as glaucoma and traumatic optic neuropathy. However, RGCs represent a highly heterogeneous population, which is manifested not only in their molecular signatures and functions but also in their markedly varied responses to degeneration, with distinct survival capabilities across different subtypes [1–4]. Among the diverse RGC subtypes, αRGCs have been found to demonstrate notably high resilience and regenerative capacity following optic nerve injury [3, 5, 6]. Deletion of αRGCs-specific secreted phosphoprotein 1 (SPP1) markedly reduces αRGCs survival whereas SPP1 overexpression protects susceptible RGCs under glaucomatous, but not axotomy, conditions [7]. These findings suggest that the mechanisms underlying the survival advantage of αRGCs remain to be defined.

Microglial activation and phagocytosis play a central role in RGC degeneration by clearing dead and dying neurons [8]. However, microglia can also execute neuronal death by indiscriminately phagocytosis of stressed-but-viable neurons adjacent to damaged ones, thereby exacerbating disease progression [9–11]. Microglial phagocytosis is governed by a balance between “eat-me” signals displayed on compromised cells [12], and “don’t-eat-me” signals that suppress phagocytic recognition. Among the latter, sialic acid modification engages in inhibiting microglial engulfment and protecting viable cells [10, 13, 14]. In this context, the relatively higher survival rate of αRGCs may reflect enhanced resistance to microglial clearance, possibly through augmented “don’t-eat-me” signaling that renders these neurons less detectable to activated microglia. Modulating such mechanisms could thus preserve vulnerable yet salvageable neurons, offering therapeutic potential.

Sialylation is a key post-translational modification (PTM) mediated by a family of approximately 20 sialyltransferases, which catalyze the attachment of sialic acids to various glycan acceptors, forming specific linkages such as α2,3-, α2,6-, and α2,8- between the sialic acid residue and the underlying sugar, that regulates cell–cell interactions and immune evasion [15–17]. In the nervous system, neuronal sialylation influences synaptic plasticity and axon regeneration [15]. For instance, α2,6-sialylated immunoglobulin G (IgG) within intravenous immunoglobulin (IVIg) exerts neuroprotective effects in Guillain-Barré syndrome [18]. Additionally, Schwann cells engineered to overexpress polysialic acid (PSA) via sialyltransferase enhance spinal cord repair by improving CNS integration and remyelination [19]. In an in vitro co-culture of mouse hippocampal neurons and microglia, neurons subjected to desialylation exhibited neurite loss and reduced length, and induced an increase in microglial phagocytic uptake of neurons [20]. Another study found that hyposialylation due to deficient in essential enzymes for sialic acid biosynthesis relieves inhibition of the complement in the mouse retina, thereby triggering lysosomal microglia response in a C3-dependent manner [21]. In summary, sialylation plays an important regulatory role in protecting neurons and axons and in modulating microglial activation.

Based on these findings, we hypothesize that the remarkable resilience of αRGCs following optic nerve injury, despite concomitant microglial activation, may be attributable to enhanced sialylation, which could confer resistance to phagocytic elimination. However, whether sialylation contributes to αRGC resilience, and how optic nerve injury alters retinal sialic acid metabolism, remain largely unknown. Here, we aim to characterize the spatial and temporal dynamics of sialic acid and sialyltransferase expression post-injury, determine whether αRGC survival is associated with sialylation-dependent phagocytic inhibition, and evaluate the therapeutic potential of sialylation modulation in promoting RGC protection and axon regeneration. Elucidating these mechanisms may unveil novel strategies to mitigate microglia-mediated neuronal loss after optic nerve injury.

Materials and methods

Animals

All animal procedures followed the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research and Animals in Research: Reporting In Vivo Experiments guidelines and were approved by the Institutional Animal Care and Use Committee of the Zhongshan Ophthalmic Center (ethics approval number: ZW2022003). C57BL/6 J wild-type mice were purchased from Gempharmatech Co., Ltd. (Nanjing, China). Both male and female mice (6 to 8 weeks old; average body weight, 18 to 22 g) used in our study were randomly allocated to each experiment. All mice were housed in standard cages in a specific pathogen–free facility strictly under 12-h light/dark cycles (8:00 a.m. to 8:00 p.m.) at an ambient temperature of 24° to 26 °C and 40 to 50% humidity. Cool white fluorescent bulbs provided an ambient illumination of around 200 lx.

Optic nerve crush (ONC) model

ONC was performed in mice aged 8 weeks (average body weight, 19 to 22 g) under general anesthesia using ketamine (100 mg/kg) and xylazine (10 mg/kg). The optic nerve was exposed intraorbitally and crushed with angled forceps for 8 s at approximately 0.5 mm behind the globe. Tobramycin ointment was applied to protect the cornea after surgery. The mice were placed on a heating pad in a quiet environment for recovery.

Adeno-associated virus (AAVs)

AAV2/2-CAG-St6gal1-IRES-EGFP was used to express ST6Gal1 in mice RGCs, and AAV2/2-CAG-GFP was used as a negative control. AAV2/2-H1-shPTEN-GFP was used to knock down Pten in RGCs, and AAV2/2-H1-shScramble-GFP was used as the negative control. AAV2/2-CAG-St6gal1-IRES-EGFP, AAV2/2-CAG-GFP, AAV2/2-H1-shPTEN-GFP and AAV2/2-H1-shScramble-GFP were obtained from Tsingke Biotech Co., Ltd. (Beijing, China). All AAV titers were diluted to at least 1 × 10¹³ vector genomes/ml.

Intravitreal injection

Mice were anesthetized by intraperitoneal injection of a mixture of ketamine (100 mg/kg) and xylazine (10 mg/kg), and topical anesthesia of the cornea was administered with tetracaine hydrochloride. For intravitreal injection of various chemical compounds, a 30-G needle (305107; BD, USA) attached to a 10 μl Hamilton syringe (80001; Hamilton, USA) was inserted into the vitreous chamber posterior to the limbus. Approximately 3 μl 3Fax-PN (1 mM; 566224; Sigma-Aldrich, USA), dissolved freshly in a solution containing 1% (v/v) dimethyl sulfoxide (196055; MP Biomedicals, USA) in PBS, were injected into the vitreous chamber immediately after ONC. 1.5 μl NMDA (20 mm) were injected into the vitreous chamber for NMDA-induced retinal neurotoxicity. For intravitreal injection of AAV, 2 μl of AAV was injected into the vitreous chamber 2 weeks before ONC to achieve stable target gene ex-pression. For anterograde tracing of regenerating axons, 2 μl of Alexa Fluor 555 (1 mg/ml)–conjugated CTB (C22843; Invitrogen, USA) were intravitreally injected 2 days before sacrifice. Eyes that developed lens injury, retinal hemorrhage, hyphema, progressive cataract formation, or endophthalmitis after injection were excluded from further analysis.

Retinal ischemia/reperfusion (I/R) model

The retinal I/R model was established as previously described [22]. After anesthesia, the pupils were dilated with 1% tropicamide. We cannulated the anterior chamber of the right eye with a 30-G needle attached to an infusion tube to allow for perfusion of sterile 0.9% saline solution for 1 h. The IOP in the eyes reached 70 mmHg. The left eyes were subjected to chamber puncture without IOP elevation as a sham procedure. After 1 h, the IOP was returned to normal by withdrawing the needle. Ischemia was confirmed by retinal whitening, and reperfusion was verified by observing the return of blood flow with an ophthalmoscope. Tobramycin ointment was applied to the eyes immediately after the operation, and the mice were placed on a heating pad for recovery.

Immunostaining, fluorescence microscopy, and image analysis

The mice were administered an overdose of anesthetic and transcardially perfused with ice-cold PBS followed by 4% paraformaldehyde (PFA, Biosharp, China). After perfusion, the eyeballs were removed and postfixed in 4% PFA at 4 °C for 2 h.

For retinal cryosection staining, eyeballs were first cryoprotected in 30% sucrose in PBS overnight and then frozen in Optimal Cutting Temperature compound (4583, SAKURA, Japan) at −80 °C. The frozen blocks containing tissue were sectioned at 14 μm (CM 1950, Leica, Germany). Sections were rinsed in PBS and blocked in 10% normal donkey serum (ASL050; Acmec biochemical, China), 5% bovine serum albumin (A1933; Millipore Sigma), and 0.2% Triton X-100 (T8200; Solarbio, China) for 1 h. Sections were then incubated in a blocking solution containing primary antibodies at 4 °C overnight. After washing three times with PBS, secondary antibodies and diluted DAPI (1:100; P0131; Beyotime, China) in blocking solution were added at room temperature for 1 h. After three additional washing steps, the sections were mounted with an antifade mounting medium (S2100; Solarbio) for microscopy. Images of retinal sections were captured by the Zeiss LSM980 confocal (Carl Zeiss, Jena, Thuringia, Germany), Leica DMi8 microscope (Leica Microsystems, Wetzlar Deutschland, Germany) and Nikon Eclipse Ni-U microscope (Nikon, Tokyo, Japan). To quantify the fluorescent intensity in RGCs, at 1.5 mm and 3 mm from the optic nerve head, four regions of identical length were selected on both sides of the optic nerve head for quantitative fluorescence analysis, and the mean immunostaining intensity of four representative fields was calculated as the intensity of each section, and the mean immunostaining intensity of three nonconsecutive sections was considered as the intensity of the retina. The number of microglia per unit area, as well as their somatic area and process length, were quantified for each retinal section using the area and line measurement tools in ImageJ. All quantifications were performed by an investigator blinded to the experimental groups.

For retinal whole-mount staining, retinas were dissected and blocked in 10% normal donkey serum, 5% bovine serum albumin, and 2% Triton X-100 for 1 h. Retinas were then incubated in blocking solution containing primary antibodies at 4 °C overnight and washed three times before another 1 h of incubation with secondary antibodies at room temperature. After three washing steps, the retinas were made four cuts to be a four-leaf clover to ensure flattening and then mounted for microscopy. Images of retinal whole-mounts were acquired using a Nikon Eclipse Ni-U microscope. For RGC counting, RGC subtype identification, four fields of 1260 × 1260 μm area were sampled from each retina.

Primary antibodies used were directed against SPP1 (1:100, AF808, R&D Systems), ST6Gal1 (1:100, 14355–1-AP, Proteintech),TUJ1 (1:500, 801202, Biolegend), RBPMS (1:500, GTX118619, GeneTex), melanopsin (1:1000, AB-N39, Advanced Targeting Systems), pCREB (1:100, ab32096, Abcam), pCaMKII (1:200, ab32678, Abcam), pS6 (1:100, 2211S, Cell Signaling Technology), pSTAT3 (1:100, 9145S, Cell Signaling Technology), pAKT (1:100, 4060S, Cell Signaling Technology), p–c-Jun (1:100, 2361S, Cell Signaling Technology), IBA1 (1:500, 019–19741, Wako). Secondary antibodies included Alexa Fluor 488 goat anti-mouse (1:500, ab150113, Abcam), Alexa Fluor 488 goat anti-rabbit (1:500, ab150077, Abcam), Alexa Fluor 488 donkey anti-goat (1:500, ab150129, Abcam), Alexa Fluor 555 goat anti-mouse (1:500, ab150114, Abcam), Alexa Fluor 555 donkey anti-rabbit (1:500, ab150062, Abcam), Alexa Fluor 594 donkey anti-goat (1:500, ab150132, Abcam), Alexa Fluor 647 goat anti-mouse (1:500, ab150115, Abcam), and Alexa Fluor 647 donkey anti-rabbit (1:500, ab150075, Abcam).

Sambucus nigra agglutinin (SNA) staining

The eyeballs were enucleated and fixed in FAS Fixator (Servicebio, China) at room temperature for 24 h. Subsequently, the samples were embedded in paraffin and sectioned into 10 μm-thick sections. Sections were baked at 60 °C for 2 h and then dewaxed and rehydrated by sequential incubation in xylene and a graded ethanol series and a final rinse in ddH₂O. Sections were then washed three times in PBS with gentle shaking. Heat-induced antigen retrieval was performed using citrate-EDTA antigen retrieval buffer. The slides were heated in a microwave oven on high power for 7 min followed by 14 min on medium–low power. After heating, the slides were allowed to cool naturally to room temperature and washed three times in PBS. Sections were blocked with carbon-free blocking solution (SP-5040, Vector Laboratories) at room temperature for 1 h and then incubated with SNA-FITC (1:100, L32479, Invitrogen) or SNA-Cy5 (1:100, CL-1305-1, Vector Laboratories) diluted in the blocking solution at 4 °C overnight. After washing three times with PBS, diluted DAPI were added for 10 min. After three washing steps, the sections were mounted with an antifade mounting medium for microscopy.

Pattern electroretinogram (PERG)

Pattern electroretinogram was performed using Roland-RETI scan (Electrophysiological Diagnostic Systems, Brandenburg, Germany, RETI-Scan21), according to the international standard steps described in our previous publications [6, 23]. In brief, mice were anesthetized using ketamine/xylazine and placed on the heating platform with the body temperature maintained at 37 °C under dim red light throughout the procedure. The pupils were dilated with tropicamide, and sodium hydroxypropyl methylcellulose gel was used to prevent corneal opacities. The PERG electrode was placed on the cornea and positioned to encompass the pupil without obstructing the vision. Visual stimuli were generated from a screen 18 cm in front of the eyes that displayed a black-and-white reversing checkerboard pattern. In total, 100 contrast reversals of the PERG signals were repeated twice for each eye, and 200 cycles were segmented, averaged, and recorded. The average PERG signals were analyzed to evaluate the peak-to-trough P50 to N95 amplitude.

Optic nerve clearance and evaluation of axon regeneration

The modified iDISCO method with anterograde CTB labeling was used to visualize optic nerve clearance and axon regeneration as our previously publicated [6]. In brief, after transcardial perfusion, the optic nerves were dissected and incubated in a series of 30%, 60%, 90% methanol in PBS, di-chloromethane (DCM, 270997; Millipore Sigma)/methanol (2:1) and 100% DCM for 30 min at each solution. Last, the nerves were transferred to dibenzyl ether (DBE, 33630; Millipore Sigma) and incubated over-night to achieve complete transparency. The cleared optic nerves were mounted using DBE for microscopy. CTB-prelabeled and cleared optic nerves were observed and captured by a Zeiss LSM980 confocal microscope using a Z-stack (10 μm per stack) and tile scan mode to acquire a full view of the nerves. We counted the regenerating axons that vertical to the optic nerve every 250 μm distal to the crush site until 1000 μm. Three representative Z-stacks at equal intervals for each nerve were selected to calculate the total number of axons. At each counting point in each Z-stack, the number of axons (n), the width of the nerve (w), and the thickness of the optical section (t = 10 μm) were used to calculate the axon density, d = . The mean axon density of the three Z-stacks was defined as D = . The total number of regenerating axons at each point was calculated as N = D × , where r is the radius of the optic nerve.

Single-cell RNA sequencing (scRNA-seq) and data analysis

Fresh mouse retinas were enzymatically dissociated to obtain single-cell suspensions. The experiment includes two treatment groups, with the eyes in each group undergoing simultaneous experimental processing and sample collection, and six retinas pooled together per group. The enzyme digestion mixture consists of papain (2 mg/ml), collagenase I (1 mg/ml), and DNase I (0.05 mg/ml) in DMEM/F12 medium. The dissected retina was placed in the enzyme digestion mixture and dissociated in a 37 °C water bath for 15 min, with gentle manual shaking every 5 min. The enzymatic digestion mixture was terminated by adding complete DMEM medium supplemented with 10% fetal bovine serum. The entire digestion mixture was filtered through a 70-μm cell strainer to remove undigested tissue fragments. The filtrate was then centrifuged at 350 × g for 5 min at 4 °C, and the cell pellet was resuspended. Single-cell suspensions of mouse retinas were then subjected to scRNA-seq on the Illumina NovaSeq 6000 platform (Gene Denovo Biotechnology, Guangzhou, China). The suspensions were processed using the 10 × Genomics GemCode Single Cell Instrument to generate single-cell gel beads-in-emulsion (GEMs). Library construction and sequencing were performed from the resulting complementary DNAs (cDNAs) using the Chromium Next GEM Single Cell 5′ Reagent Kit (v2.0). After the gel beads were dissolved within the GEMs, primers carrying an Illumina® R1 sequence (read 1 primer), a 16-nucleotide 10 × barcode, a 10-nucleotide Unique Molecular Identifier (UMI), and a poly(dT) sequence were released. These primers were mixed with the cell lysate and reverse transcription master mix to synthesize barcoded full-length cDNAs from polyadenylated mRNA transcripts. The resulting cDNAs were then PCR-amplified to yield sufficient quantities for subsequent library preparation and sequencing.

To analyze the gene expression of sialyltransferases in different RGC subtypes, we used publicly available murine gene sets of RGCs (GSE137400) [1]. To analyze the microglia subclasses, we used our single-cell sequencing dataset of retinal cells from mice (CRA032270). Downstream data analyses were performed using R (v4.1.3) with the R package Seurat (v4.1.1).

Statistical analysis

Unpaired Student’s t test was used between the comparison of two groups. One-way analysis of variance (ANOVA) was used to compare multiple groups. All data were presented as means ± SEM, and P value < 0.05 was considered statistically significant. GraphPad Prism 10 was used for statistical analysis. Details of the statistical analysis for each experiment were stated in the figure legends.

Results

Optic nerve injury leads to decreased sialic acid levels in RGCs whileαRGCs maintain higher levels of sialyltransferases

To investigate the expression patterns of sialyltransferases across RGC subtypes, we analyzed raw transcriptomic data from a publicly available single-cell RNA sequencing (scRNA-seq) study, based on their established subclasses of RGC types [1]. By comparing the levels of sialyltransferases among annotated retinal RGC subtypes in wild-type (uninjured) mice, we found a preferential enrichment of several sialyltransferases in the αRGC subtype, including ST6Gal1, ST8Sia1, ST3Gal3 and ST3Gal2 (Fig. 1A).Previous studies have established that ST6Gal1-mediated α2,6-sialylation of adhesion molecules enhances cell survival and promotes immune evasion in cancer contexts [24, 25], and elevated St6gal1levels promote survival and neurite outgrowth in cerebellar granule neurons [26], both showing ST6Gal1 might spare cells and neurons from escaping death. Therefore, we prioritized ST6Gal1 for further investigation. We then analyzed the dynamic level of St6gal1 in αRGCs, ipRGCs, and other RGCs under both uninjured and post-injury conditions using the dataset. We noted that, compared with other subtypes, αRGCs consistently exhibited relatively higher St6gal1 expression levels after injury (Fig. 1B). We therefore hypothesized that αRGCs may represent a neuronal subtype concerning ST6Gal1, capable of evading degeneration following injury.

Fig. 1.

αRGCs maintain higher levels of sialyltransferases. A Dot plot showing the expression of four sialyltransferases in different RGC subtypes. The size of each circle represents the percentage of RGC subtypes expressing the gene, and the color indicates the expression level. B Line chart showing St6gal1 expression dynamics for each RGC subtype at different time points after ONC. C Representative retinal sections showing SNA expression in RGCs at the indicated time points after ONC. D Quantification of relative SNA immunostaining intensity in RGCs. n = 3 retinas per group. One-way analysis of variance (ANOVA) with Turkey’s multiple comparisons test. E Representative retinal sections showing ST6Gal1 expressions in retinas at indicated time points after ONC. F Quantification of relative ST6Gal1 immunostaining intensity in GCL. n = 6 retinas per group. One-way analysis of variance (ANOVA) with Turkey’s multiple comparisons test. G Representative retinal sections showing ST6Gal1 expressions in SPP1⁺ αRGCs and non-αRGCs at indicated time points after ONC. H-J Quantification of relative ST6Gal1 immunostaining intensity in SPP1⁺ αRGCs and non-αRGCs. n = 5 retinas per group. Unpaired t test. *P < 0.05, **P < 0.01, ns, not significant. All error bars represent means ± SEM. Scale bar, 50 μm (main views) and 10 μm (enlarged views)

To validate these findings, we assessed α2,6-sialylation dynamics of RGCs using SNA lectin staining. SNA labels α2,6-sialylated glycoproteins on the cell surface, while additional signals may be observed in plasm membrane and perinuclear membrane-associated compartments such as Golgi bodies [27, 28]. Following ONC, global retinal immunohistochemistry indicated a progressive decline in both SNA signal (Fig. 1C, D) and ST6Gal1 immunoreactivity (Fig. 1E, F) within RGCs, reflecting a loss of α2,6-sialylation upon axonal injury.

Conversely, a distinct upregulation of ST6Gal1 emerged within the αRGC subtype. Only a modest elevation of ST6Gal1 was observed in αRGCs compared to neighboring cells in uninjured retinas (Fig. 1G, H), but this disparity became pronounced following injury. After ONC, αRGCs exhibited a significant retention or upregulation of ST6Gal1, with signal intensities reaching 1.5- to 2- fold higher levels than surrounding cells (Fig. 1G, I, J). These findings suggest that unlike other RGC populations, αRGCs maintain higher ST6Gal1 levels under injury, indicating a mechanism that potentially underlies their resilience to degeneration.

Sialylation blockade exacerbates RGC death after ONC injury

3Fax-peracetyl Neu5Ac (3Fax-PN) is a kind of sialyltransferase inhibitor and can effectively inhibit sialyltransferases in a donor substrate CMP-Neu5Ac-competitive manner [29]. In uninjured retinas, intravitreal injection of 1 mM 3Fax-PN effectively reduced α2,6-sialylation expression in RGCs by SNA staining, reducing levels by over 50% compared to vehicle controls (Fig. 2A, B). Similarly, ST6Gal1 expression was also significantly decreased following 3Fax-PN treatment (Fig. S1A, B). Notably, neither 1 mM nor 5 mM 3Fax-PN affected RGC density 14 days post-injection, indicating that 3Fax-PN itself has no additional toxic effects on uninjured RGCs (Fig. S1C, D). To further investigate the protective potential of sialylation for RGCs, we performed intravitreal injection of 3Fax-PN immediately following ONC in mice and quantified surviving RGCs at 7 dpc and the survival rate of αRGCs at 14 dpc (Fig. 2C). At 7 dpc, RGC survival rate in the vehicle group was approximately 51%, whereas blocking sialylation with 3Fax-PN reduced survival to 37.24% (Fig. 2D, E). At 14 dpc, RGC survival was further diminished in the 3Fax-PN-treated group, suggesting that inhibition of sialylation exacerbated RGC loss (Fig. 2F, G). Furthermore, we also found that sialylation blockade exacerbated the loss of SPP1⁺αRGC, reducing their survival rate from 73.47% to 32.79% at 14 dpc (Fig. 2F, H), without significantly affecting non-αRGC survival (Fig. 2F, I). These results indicate that elevated sialylation could be a critical determinant underlying the superior injury resilience of αRGCs.

Fig. 2.

Sialylation blockade exacerbates RGC death after ONC injury. A Representative retinal sections showing SNA expressions in RGCs treated with 3Fax-PN. B Quantification of relative SNA immunostaining intensity in RGCs. n = 5 retinas per group. C Schematic illustration of 3Fax-PN treatment and ONC and timeline. D Representative retinal flat mounts showing TUJ1⁺ surviving RGCs in differentially treated retinas at 7 dpc. E Quantification of RGC survival rate at 7 dpc. n = 5 retinas per group. F Representative retinal flat mounts showing RBPMS⁺ RGCs and SPP1⁺/RBPMS⁺ αRGCs in differentially treated retinas at 14 dpc. G Quantification of RGC survival rate at 14 dpc. n = 5 retinas per group. H-I Quantification of percentages of surviving αRGC and non-αRGC in total surviving RGCs. n = 5 retinas per group. All statistical analyses were performed using unpaired t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns, not significant. All error bars represent means ± SEM. Scale bar, 50 μm

ST6Gal1 overexpression confers neuroprotection

Given that high sialylation is crucial for αRGC survival and that global sialylation levels are generally declined in RGCs after injury, we hypothesized that broad enhancement of sialylation may confer pan-RGC protection against ONC injury. To test this, we intravitreally delivered an adeno-associated virus AAV2-CAG-St6gal1-EGFP to overexpress ST6Gal1 in mouse retinas. This approach achieved an RGC transduction efficiency exceeding 57% (Fig. S2A, B), and upregulated expression of ST6Gal1 (Fig. S2C, D) and α2,6-sialylation (Fig. S2E, F). Wild-type mouse received either AAV2-GFP or AAV2-CAG-St6gal1-EGFP two weeks prior to ONC (Fig. 3A). At 14 dpc, ST6Gal1 overexpression significantly increased RGC survival in treated retinas (Fig. 3B, C).

Fig. 3.

ST6Gal1 overexpression confers neuroprotection. A Schematic illustration of AAV2-St6gal1 injection and ONC and timeline. B Representative retinal flat mounts showing RBPMS⁺ surviving RGCs in differentially treated retinas at 14 dpc. C Quantification of RGC survival rate. n = 7 retinas per group. D Representative retinal flat mounts showing SPP1⁺ surviving αRGCs in differentially treated retinas at 14 dpc. E–F Quantification of survival rates of αRGC and non-αRGC. n = 5 retinas per group. G Representative retinal flat mounts showing melanopsin⁺ surviving ipRGCs in differentially treated retinas at 14 dpc. H Quantification of survival rates of ipRGC. n = 7 retinas per group. I Schematic illustration of AAV2-St6gal1 injection and I/R and timeline. J Representative retinal flat mounts showing RBPMS⁺ surviving RGCs in differentially treated retinas at 7 days post injury. K-M Quantification of RGC survival rate in whole-mount, central, and peripheral retinas. n = 5 retinas per group. N Schematic illustration of AAV2-St6gal1 injection and NMDA injection and timeline of experiments. O Representative retinal flat mounts showing RBPMS⁺ surviving RGCs in differentially treated retinas at 7 days after NMDA induced injury. P Quantification of RGC survival rates. n = 5 retinas per group. Q Representative responses of PERG recordings 7 days after NMDA induced injury. R Quantification of PERG amplitudes. n = 6 eyes per group. All statistical analyses were performed using unpaired t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns, not significant. All error bars represent means ± SEM. Scale bar, 50 μm

Because αRGCs naturally maintain higher and more persistent ST6Gal1 expression both before and after injury, we next investigated whether exogenous ST6Gal1 expression could transcend subtype-specific neuroprotective preferences and confers broader neuroprotection across the RGC population. To this end, we labeled surviving RGCs at 14 dpc with subtype-specific markers SPP1 (αRGCs) and melanopsin (ipRGCs). While αRGC survival remained unchanged between groups (Fig. 3D, E), non-αRGCs in the ST6Gal1-overexpression group demonstrated prominent improved survival (Fig. 3D, F), indicating that enhanced sialylation preferentially benefits more vulnerable RGC populations. Consistent with previous single-cell RNA sequencing data showing low endogenous sialyltransferase expression in ipRGCs, overexpression of ST6Gal1 significantly promoted ipRGC survival following injury (Fig. 3G, H), underscoring its broad neuroprotective potential in the injured retina.

To further validate our findings, we assessed the neuroprotective effects of ST6Gal1 overexpression in two additional in vivo models — I/R injury and NMDA-induced excitotoxicity — which model common pathways involved in glaucomatous RGC degeneration. The results showed that overexpressing ST6Gal1 markedly increased RGC survival in both I/R (Fig. 3I) and NMDA injury model (Fig. 3N). We next evaluated whether ST6Gal1 overexpression could preserve visual function following NMDA-induced injury. RGC activity was quantified using the PERG, a noninvasive technique that records electrical response of RGCs to patterned visual stimuli. Compared with the AAV2-GFP controls, eyes overexpressing ST6Gal1 showed a significantly smaller reduction in the PERG P50-N95 amplitude at 7 days post-NMDA injury (Fig. 3Q, R), indicating that enhanced sialylation helped maintain functional RGC populations. Collectively, these data demonstrate that ST6Gal1 overexpression confers broad neuroprotective benefits across multiple models of RGC injury.

ST6Gal1 overexpression enhances optic nerve regeneration

Since ST6Gal1 overexpression demonstrates remarkable neuroprotective effect, we wonder whether it could promote optic nerve regeneration. Remarkably, ST6Gal1 overexpression had a pronounced pro-regenerative effect on optic nerve axons after ONC, with an efficacy rivaling that of classical Pten knockdown (Fig. 4A-C). To elucidate potential mechanisms, we examined whether ST6Gal1 overexpression modulates established neuroprotective or pro-regenerative signaling pathways in RGCs. Phosphorylation levels of adenosine 3′,5′- monophosphate (cAMP) response element-binding protein (CREB) and Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) in RGCs typically decline following optic nerve injury but can be revitalized through regenerative stimuli, eventually promoting RGC survival and axon regeneration. Notably, ST6Gal1 overexpression increased phosphorylated CREB (pCREB) immunostaining intensity in total RGCs at 3 dpc compared with AAV2-GFP controls (Fig. 4D, E). Particularly, ST6Gal1 overexpression did not further enhance pCREB levels in αRGCs (Fig. 4G), despite their intrinsically high ST6Gal1 expression. Instead, an increase in pCREB was observed in non-αRGCs (Fig. 4G). Whereas CaMKII activation remained unchanged (Fig. 4D, H). The activation of the mammalian target of rapamycin (mTOR) and related pathways stands as a well-known mediator that fosters optic nerve regeneration. However, our study did not reveal any differences in the upstream or downstream targets of the mTOR pathway, including phosphorylated protein kinase B (pAKT), phosphorylated signal transducer and activator of transcription 3 (pSTAT3), and phosphorylated ribosomal protein S6 (pS6) (Fig. 4F, I-K). The dual leucine zipper kinase (DLK)/c-Jun pathway predominantly contributes to RGC death after optic nerve injury, with the level of phosphorylated c-Jun (p-c-Jun) in uninjured retinas experiencing a marked up-regulation after ONC. However, p–c-Jun level in RGCs was not mitigated by the treatment of ST6Gal1 overexpression 3 days after ONC (Fig. 4F, L). These molecules involved in regeneration-associated pathways mentioned above, did not exhibit significant changes in either αRGCs or non-αRGCs after St6gal1 overexpressing (Fig. S3).

Fig. 4.

ST6Gal1 overexpression enhances optic nerve regeneration. A Schematic illustration of AAV2-St6gal1 injection and timeline of experiments. B Maximum projections of CTB-labeled regenerating axons. Asterisks indicate crush site. Scale bar, 250 μm. C Quantification of regenerating axons in differentially treated groups. n = 3 to 10 nerves per group. Two-way analysis of variance (ANOVA) with Turkey’s multiple comparisons test. D Representative retinal sections showing phosphorylation levels of CREB, CAMKII, S6, STAT3, AKT, c-Jun in TUJ1⁺/SPP1⁺ RGCs from differentially treated groups at 3 dpc. Scale bar, 50 μm. E–G Quantification of relative fluorescent intensities of pCREB in total RGCs (E), αRGCs (F) and non-αRGCs (G). n = 5 retinas per group. Unpaired t test. H–L Quantification of relative fluorescent intensities of pCAMKII (H), pSTAT3 (J), pAKT (K), p-c-Jun (L) in total RGCs and percentage of pS6⁺ RGCs (I). n = 5 retinas per group. Unpaired t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns, not significant. All error bars represent means ± SEM

Modulating sialylation levels affects microglial phagocytosis

Since sialic acid-receptor interactions may serve as a "don't eat me" signals between neurons and microglia, we next examined how modulating sialylation affects microglial activation after neuronal injury. To this end, we performed scRNA-seq on retinal tissue harvested at 3 dpc (Fig. 5A). Microglial clusters were subsequently extracted and subjected to subcluster analysis (Fig. 5B). Two transcriptionally distinct microglial subpopulations were identified based on established markers: a homeostatic subtype enriched for P2ry12, Tmem119, Csf1r, Cx3cr1, and Hexb [30]; and a degenerative subtype resembling microglia associated with neurodegeneration (MGnD), characterized by upregulation of Apoe, Tyrobp, Ctsb, and C1qb [31, 32], alongside downregulation of homeostatic genes (Fig. 5C, D). This MGnD-like subpopulation shares transcriptional signatures with disease-associated microglia (DAMs) commonly observed in neurodegenerative contexts and exhibits an enhanced phagocytic phenotype.

Fig. 5.

Modulating sialylation levels affects microglial phagocytosis. A Schematic illustration of AAV2-St6gal1 injection and timeline of experiments. B Uniform manifold approximation and projection (UMAP) plot visualization of retinal cells after clustering colored according to cell types. C Dot plot for the expression of homeostatic marker genes in each microglia subset. D UMAP plot visualization of microglia subpopulations. E Bar plot for the percentage of microglia subpopulations in differentially treated groups. F and J Representative retinal sections showing microglia infiltration and morphology from differentially treated groups at 3 dpc. G-I and K-M Quantifications of microglia density, soma size and process length of microglia from differentially treated groups at 3 dpc. n = 3 retinas per group (G and K), n ≥ 10 in each retina from 3 retinas per group (H-I and L-M). All statistical analyses were performed using unpaired t test. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. All error bars represent means ± SEM. Scale bar, 50 μm

Strikingly, retinas injected with AAV2-St6gal1 exhibited a marked increase in the proportion of homeostatic microglia (Fig. 5E), indicating that enhanced sialylation might shift microglial composition toward a homeostatic state. Furthermore, these homeostatic microglia displayed elevated expression of siglech compared to MGnD-like microglia (Fig. 5C), consistent with their resting phenotype.

To validate these findings, we assessed microglial abundance and morphology under manipulated sialylation conditions. Inhibition of sialyltransferases with 3Fax-PN markedly increased microglial density and induced a hypertrophic, amoeboid morphology with shortened processes — features characteristic of activated, pro-phagocytic microglia (Fig. 5F-I). Conversely, ST6Gal1 overexpression reduced microglial density and promoted a ramified morphology with fine, elongated processes, indicative of a quiescent surveillance state (Fig. 5J-M). Together, these results demonstrate that sialylation bidirectionally regulates microglial activation, with elevated sialylation and enhanced “don’t eat me” signal favoring a quiescent, homeostatic microglial phenotype that may underlie the neuroprotective and pro-regenerative effects of ST6Gal1 overexpression (Fig. 6).

Fig. 6.

Schematic of sialylation modulating the microglial phagocytosis of RGCs in nerve injury. Following injury or stress, retinal ganglion cells (RGCs) exhibit reduced expression of the sialyltransferase ST6Gal1, leading to a loss of surface sialic acids and subsequent hyposialylation. This decrease in sialylation lowers the inhibitory signaling to microglia and facilitates their phagocytic activation toward stressed-but-viable neurons. In contrast, ST6Gal1 overexpression maintains a high level of α2,6-sialylation on the RGC membrane. The abundant sialic acids engage inhibitory receptors on microglia, transmitting a “don’t-eat-me” signal that restrains phagocytic activity, promotes a homeostatic microglial phenotype, and ultimately preserves RGC survival

Discussion

Our study reveals a previously unrecognized role of sialylation in determining RGC survival and axon regeneration after injury. We found that, following ONC, αRGCs retained relatively higher α2,6-linked sialylation mediated by the sialyltransferase ST6Gal1. Inhibition of sialyltransferases exacerbated αRGC loss, while ST6Gal1 overexpression only modestly affected αRGC survival but markedly enhanced the survival of ipRGCs and other RGCs. Moreover, the neuroprotective effect of ST6Gal1 overexpression extended beyond the ONC model to multiple paradigms of retinal injury. Unexpectedly, ST6Gal1 overexpression also promoted robust optic nerve axon regeneration, with efficacy comparable to that of Pten knockdown, yet through a mechanism independent of classical mTOR activation. Furthermore, enhanced sialylation shifted microglia toward a homeostatic phenotype and reduced excessive neuronal phagocytosis. Collectively, our findings identify ST6Gal1-mediated sialylation as a previously unrecognized determinant of RGC survival and regeneration, offering new insight into glycosylation-based neuroprotection.

Targeting αRGC resilience to bridge the survival disparity among RGC subtypes

A major challenge in neuroprotection lies in the heterogeneous response among RGC subtypes to injury. Whereas αRGCs exhibit remarkable resilience, the majority of other RGCs rapidly undergo apoptosis [33, 34]. Our research proposes sustained high sialylation as an important intrinsic factor that contributes to αRGC resistance, complementing other well-established mechanisms. Specifically, αRGCs maintain high α2,6-linked sialylation after ONC, correlating with their superior survival; in contrast, more other vulnerable subtypes exhibit significant sialylation loss. This causal link is further supported by a compensatory intervention: ST6Gal1 overexpression preferentially enhanced the survival of injury-desialylated ipRGCs, without further benefiting αRGCs. The differential sialylation profiles thus explain the divergent injury resilience across RGC subtypes and reveal a promising therapeutic strategy: artificially elevating or preserving sialylation in vulnerable subtypes could mimic the innate resilience of αRGCs.

Therapeutically, enhancing sialylation — through precursors, sialyltransferase induction, or sialidase inhibition — may confer broad protection by acting through dual mechanisms. Intrinsically, it fortifies vulnerable RGCs against cytotoxic and apoptotic signals. Extrinsically, it modulates the retinal immune milieu, where a sialylated neuronal surface can inhibit microglial phagocytosis and promote a homeostatic state. By concurrently enhancing neuronal resilience and fostering a quiescent neuroimmune environment, sialylation-based interventions offers a strategic path to bridge the survival disparity among RGC subtypes.

Mechanistic implications of sialylation in preserving RGCs

Sialylation, a widespread glycosylation modification in the nervous system, plays a pivotal role in neurodevelopment, synaptic plasticity, immune homeostasis, neuroprotection and the maintenance of myelin integrity [35]. Its protective actions likely derive from sialic acid's function as a "self" molecular marker that engages with specific immunoregulatory receptors [15]. As terminal residues on cell-surface glycoconjugates, sialic acids can bind complement factor H or be recognized by sialic acid-binding immunoglobulin-type lectins (siglecs) expressed on microglia, thereby suppressing complement activation, phagocytosis, and oxidative burst [13, 35]. Conversely, the loss or enzymatic removal of sialic acids exposes underlying glycan structures, facilitating the binding of opsonins such as C1q, calreticulin, or galectin-3, which in turn promote phagocytic clearance [36].

In the immune system, tumor cells often exploit this pathway by upregulating sialylation to engage with siglec receptors (e.g., Siglec-7, -9, -E) on immune cells, transmitting a "don't eat me" signal that suppresses cytotoxic T-cell and NK-cell activity, thereby facilitating immune evasion [37–41]. Analogously, our finding that αRGCs retain enriched α2,6-linked sialylation and superior survival after ONC suggests a similar mechanism in the injured retina: highly sialylated αRGCs may evade microglial phagocytosis by maintaining a robust “self” signature through siglecs-mediated interactions. This provides an interesting explanation for their innate resistance and suggests sialylation may act as a central regulator of neuroimmune crosstalk following injury.

Building on evidence that microglia-derived neuraminidase (Neu1) exacerbates glutamate toxicity via neuronal desialylation [42], we asked whether intrinsic sialylation levels govern RGC resilience. Pharmacological inhibition of sialyltransferases with 3Fax-PN markedly reduced overall RGC survival and accelerated αRGC loss at 14 dpc, coinciding with an expansion of phagocytic microglia. These results indicate that loss of neuronal sialylation not only sensitizes RGCs to injury but also drives a permissive, pro-phagocytic microenvironment.

Although therapeutic modulation of hypersialylation for CNS repair remains underexplored, early studies established the regenerative potential of PSA. Lentiviral expression of polysialyltransferase in glial scars or Schwann cells markedly enhanced Purkinje axon regeneration, which is typically absent in adult mammals, and similar strategies also promoted corticospinal tract regrowth after spinal cord injury [43, 44]. These findings defined sialylation as a primarily extracellular modulator of the inhibitory environment.

In contrast, our work unveiled a distinct, cell-autonomous role for sialylation in neuroprotection and regeneration. Elevating sialylation intrinsically within RGCs enhanced their regenerative capacity through augmentation of pCREB signaling, and modulated microglial phagocytic activity, which possibly curbing excessive clearance of stressed-but-viable neurons and thereby promoting RGC survival after injury. Although it remains unclear whether the observed reduction in microglial phagocytic activation is a cause of decreased RGC death or a consequence of attenuated RGC degeneration. Direct manipulation of RGC-microglia interactions, such as microglial depletion by PLX5622, would be required to resolve this causality in the future.

Sialylation as a novel modulator of optic nerve regeneration

Optic nerve regeneration represents one of the foremost challenges in CNS repair, largely due to the intrinsically limited regenerative capacity of mature neurons [45, 46]. Therapeutic efforts for optic nerve regeneration converge on boosting intrinsic growth potential and overcoming extrinsic inhibition [47]. Within this framework, PTMs stand out as pivotal integrators of the axon regenerative process. Previous research has focused on PTMs of cytoskeletal proteins, especially tubulin, showing the roles of phosphorylation, acetylation, tyrosination, and polyglutamylation in regulating microtubule stability, axonal transport, and growth cone motility [48–51]. These modifications critically shape the intrinsic regenerative state of injured neurons and influence their capacity to respond to extrinsic inhibitory cues.

Our study expands this paradigm by demonstrating that sialylation, a key PTM on neurons, strengthens neuronal resilience and dynamically regulates neuron-glia communication to promote optic nerve regeneration. Notably, the regenerative effect of ST6Gal1 overexpression occur independently of canonical mTOR activation classically engaged by Ptendeletion [52]. This distinction underscores the therapeutic potential of sialylation as an alternative strategy that circumvents the off-target effects and clinical safety concerns associated with chronic PI3K/Akt/mTOR activation [53, 54]. Sialylation-based modulation introduces an orthogonal and complementary dimension to the current repertoire of regenerative interventions.

Because sialylation operates through a distinct neuron-glia axis, it is well suited for combination with existing approaches, for example, transcription-factor reprogramming, epigenetic modifications, enhancement of mitochondrial energy metabolism, zinc chelation, or dopamine supplementation, to simultaneously reinforce intrinsic growth programs and cultivate a permissive extrinsic environment [47, 55–57]. In this way, sialylation upregulation could act both as a standalone therapy and as a versatile adjuvant that synergizes with other interventions.

Going forward, priorities include long-term evaluation of visual functional recovery, clarification of how sialylation influences central projection reformation and target reinnervation, and rigorous behavioral assessment. Addressing these questions will be critical to determine whether sialylation modulation can be translated into a clinically viable strategy for optic-nerve repair.

Neuron-directed fine-tuning of microglial function

The role of microglia in neuronal injury is notoriously complex and double-edged. Overactivation drives secondary neurodegeneration, through the release of pro-inflammatory cytokines and excessive phagocytosis of stressed-but-viable neurons [11]. Whereas complete ablation of microglia deprives neural tissue of neuroprotective cytokines and trophic factors, ultimately impairing recovery [58].

Our findings on sialylation offer a refined solution to this dilemma. Rather than broadly suppressing or eliminating microglia, which disrupts their essential functions, enhanced neuronal sialylation enables selective modulation of microglial activity from the neuronal side. Highly sialylated RGCs present a "don't eat me" signal that engages inhibitory microglial receptors such as siglecs, thereby raising the threshold for phagocytic activation. This mechanism allows microglia to remain in situ, continuing their surveillance and homeostatic functions, while their inappropriate engulfment of stressed neurons is restrained.

Notably, activated microglia can release neuraminidases that strip sialic acids from neuronal glycoconjugates, marking neurons for clearance and aggravating injury [42]. Consistent with this, desialylation of microglia by 3Fax-PN stimulated phagocytosis of beads in vitro, and co-culture with desialylated microglia induces neuronal loss even in the absence of neuronal desialylation [29]. By preventing such desialylation, enhanced sialylation may correct microglial hyper-responsiveness. Accordingly, ST6Gal1 overexpression appeared to foster a more homeostatic microglial milieu after injury, which may partly explain how enhanced sialylation protects RGCs from degeneration. This approach elegantly targets the interaction between the neuron and the immune cell, rather than the immune cell itself, offering a more precise and potentially safer therapeutic avenue for neuroprotection.

Several limitations of the present study should be noted. First, although our findings suggest that enhanced ST6Gal1-mediated sialylation modulates the phagocytic state of microglia, the underlying molecular mechanisms remain unclear, including the specific sialylated glycoproteins on RGCs and their interactions with inhibitory receptors on microglia. In addition, the mechanisms by which ST6Gal1 overexpression promotes axonal regeneration may extend beyond enhanced RGC survival and warrant further investigation. Second, our analyses were limited to early post-injury stage, and neither long-term survival nor functional outcomes were assessed. While our data demonstrated elevated survival and regeneration associated with enhanced sialylation and microglial homeostasis, whether these changes translate into sustained functional recovery requires further investigation. Finally, ST6Gal1 overexpression was induced prior to injury, which does not recapitulate a post-injury therapeutic paradigm. Future studies incorporating long-term follow-up, functional assessments, and clinically relevant post-injury interventions will be essential to fully establish the translational potential of targeting ST6Gal1.

Conclusion

In conclusion, our study elevates sialylation from a correlational marker of neuronal resilience to a potential mechanistic regulator of both neuroprotection and regeneration. By harnessing this pathway, future therapies could simultaneously empower vulnerable neurons and recalibrate immune–neuronal dialogue, offering a precise and sustainable route toward functional repair of the injured visual system.

Abbreviations

RGC

Retinal ganglion cell

ONC

Optic nerve crush

SPP1

Secreted phosphoprotein 1

PTM

Post-translational modification

PSA

Poly-sialic acid

CNS

Central nervous system

AAV

Adeno-associated virus

I/R

Ischemia/reperfusion

SNA

Sambucus nigra agglutinin

PERG

Pattern electroretinogram

dpc

Days post crush

3Fax-PN

3Fax-peracetyl Neu5Ac

ipRGC

Intrinsically photosensitive RGCs

MGnD

Microglia associated with neurodegeneration

DAM

Disease-associated microglia

Siglec

Sialic acid-binding immunoglobulin-type lectins

Authors’ contributions

LYL conducted the experiments, analyzed the data, prepared the figures and wrote the manuscript. JHT and QZ performed some experiments, analyzed the data, and edited the manuscript. ZL, BZ, and YXQ performed some experiments. QZ and YDL helped the initial analysis of the scRNA-seq dataset. LYL and YQL designed the project. YHZ and YQL supervised all the experiments, and acquired funding.

Funding

This study was supported by the National Natural Science Foundation of China (82471067 to YQL, 82471074 to YHZ), the Guangdong Basic and Applied Basic Research Foundation (2024A1515013296 to YQL), the Guangdong Basic Research Center of Excellence for Major Blinding Eye Diseases Prevention and Treatment (2024-PIZC-022 to YHZ), the Research Funds of the State Key Laboratory of Ophthalmology, the Open Research Funds of the State Key Laboratory of Ophthalmology (SZ2025KF02 to YQL) and the Guangdong Basic and Applied Basic Research Foundation (2023A1515110922 to ZL)

Data availability

The data supporting the conclusions of this article are included within the article and its additional files. The data are available from the corresponding authors on reasonable request.

Declarations

Ethics approval and consent to participate

The animals experiments were approved by the Institutional Animal Care and Use Committee of the Zhongshan Ophthalmic Center (ethics approval number: ZW2022003).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.