Effective treatment of optic neuropathies by intraocular delivery of MSC-sEVs through augmenting the G-CSF-macrophage pathway

Significance

Effective therapies for the treatment of optic neuropathies are urgently needed in the clinic. In this study, we provide the proof-of-concept evidence that intraocular (i.o.) delivery of mesenchymal stem cell–derived small extracellular vesicles markedly promoted regeneration of retinal ganglion cell axons, which embroiled recruitment of macrophages by mural cell–derived colony-stimulating factor 3 (G-CSF). I.o. delivery of G-CSF reproduced beneficial effects for the treatment of optic neuropathies. Our data provide a therapeutic paradigm for the effective treatment of optic neuropathies by reprogramming immune response.

Abstract

Optic neuropathies, characterized by injury of retinal ganglion cell (RGC) axons of the optic nerve, cause incurable blindness worldwide. Mesenchymal stem cell–derived small extracellular vesicles (MSC-sEVs) represent a promising “cell-free” therapy for regenerative medicine; however, the therapeutic effect on neural restoration fluctuates, and the underlying mechanism is poorly understood. Here, we illustrated that intraocular administration of MSC-sEVs promoted both RGC survival and axon regeneration in an optic nerve crush mouse model. Mechanistically, MSC-sEVs primarily targeted retinal mural cells to release high levels of colony-stimulating factor 3 (G-CSF) that recruited a neural restorative population of Ly6C^low monocytes/monocyte-derived macrophages (Mo/MΦ). Intravitreal administration of G-CSF, a clinically proven agent for treating neutropenia, or donor Ly6C^low Mo/MΦ markedly improved neurological outcomes in vivo. Together, our data define a unique mechanism of MSC-sEV-induced G-CSF-to-Ly6C^low Mo/MΦ signaling in repairing optic nerve injury and highlight local delivery of MSC-sEVs, G-CSF, and Ly6C^low Mo/MΦ as therapeutic paradigms for the treatment of optic neuropathies.

Optic neuropathies, in which retinal ganglion cell (RGC) axons of the optic nerve are injured by various causes including trauma or neurodegeneration such as glaucoma, are the leading causes of blindness worldwide and lack a cure (1, 2). As part of the central nervous system (CNS), RGCs in adult mammals fail to regenerate their axons, leading to the loss of RGC cell bodies (3). Potential neural restoration would require both the preservation of neurons and activation of axon regrowth. One promising strategy involves modulating neuroinflammation. Stimulating an influx of hematogenous leucocytes through lens injury or intraocular (i.o.) injection of zymosan promotes RGC survival and axon regeneration in a murine optic nerve crush injury model (ONC) (4, 5). The recruited immune cells can help clear dead cells and debris or provide sustainable growth factors and chemokines such as oncomodulin, LIF, IGF, SDF1, and CCL5 to support both neuronal survival and axonal elongation (6–11). The understanding of beneficial neuroinflammation provides exciting clues for the design of immune-based therapies for nerve damage (12).

Mesenchymal stem cell–derived small extracellular vesicles (MSC-sEVs) are a promising “cell-free” therapeutic approach for tissue repair and regeneration with potential immunomodulatory effects (13). MSCs are multipotent stem cells that are known for their paracrine and immunomodulatory features in tissue regeneration and homeostasis (14). sEVs, including exosomes and microvesicles, are nanoscale membrane-enclosed particles secreted by almost all cells and carrying hundreds of bioactive molecules like nucleic acids, proteins, lipids, carbohydrates, and metabolites (15). MSC-sEVs preserve the paracrine effect of MSCs and are superior to MSCs in safety and in convenience with respect to storage and delivery (16). Various preclinical and clinical studies have demonstrated the potential of MSC-sEVs in ameliorating a wide range of tissue damage (17–19). However, MSC-sEVs are highly heterogeneous and complicated, and the key target cells and molecular mediators that trigger tissue repair remain poorly understood (17). It has been reported that MSC-sEVs relieve RGC damage in ischemic and glaucomatous retinas, and some miRNA species in MSC-sEVs may play a role (20–23). However, the therapeutic efficacy of MSC-sEVs fluctuates among reports, and the potent immunomodulatory effect and mechanism for nerve repair have not been characterized, impeding their potential clinical application.

In this study, we demonstrate a potent neuroprotective effect of MSC-sEVs via a unique immunomodulatory pathway. We found that MSC-sEVs were initially engulfed by retinal mural cells, thereby triggering the release of colony-stimulating factor 3 (G-CSF), which recruited a restorative Ly6C^low monocyte/ monocyte-derived macrophage (Mo/MΦ) population for RGC survival and axon regeneration. Our findings reveal a previously unrecognized G-CSF-to-Ly6C^low Mo/MΦ signaling axis initiated by MSC-sEVs that protects against optic nerve injury, providing viable therapeutic strategies based on MSC-sEVs or G-CSF, a proven clinical agent to treat neutropenia, for optic neuropathies.

Results

MSC-sEVs Promote Axon Regeneration and Neural Survival.

We introduced a mouse-immortalized MSC cell line, which avoided the batch effect of primary MSCs with limited passage capacity (19). The MSC-sEVs isolated by ultracentrifugation showed typical cup-shaped vesicles under transmission electron microscopy (TEM), with a size distribution and a peak diameter of approximately 135 nm by nanoparticle tracking analysis (NTA). These MSC-sEVs had enriched expression of the exosome-specific markers Alix, CD9, and CD81, with negative markers (24) barely detectable by WB (Fig. 1A).

Fig. 1.

MSC-sEV treatment recruited myeloid cells and promoted axon regeneration and RGC survival. (A) Characterization of MSC-sEVs (Left: TEM, Middle: NTA, Right: WB). (B–D) Optic nerves and retinas were harvested from ONC mice at 2 wk after i.o. treatment with vehicle (PBS) or MSC-sEVs every 3 d (starting on day 0). The sham surgery group served as a negative control. (B) Representative images of retinal whole mounts with anti-RBPMS (white) to visualize RGCs. (Scale bar, 50 μm.) The frequency of viable RBPMS⁺ RGCs was normalized to that in healthy retinas (n = 5 mice/group). (C) Representative images of optic nerves from longitudinal sections immunostained with anti-GAP43 (green). The asterisks indicate the crush site (n = 6 nerves/group). (Scale bar, 200 μm.) (D) Quantification of GAP43⁺ regenerating axons at serial distances from the crush site. Statistical significance was determined by t test or two-way ANOVA and Sidak’s post hoc comparisons (Mean ± SEM, ***P < 0.001, ****P < 0.0001). (E–H) scRNA analysis of retinal cells harvested from ONC mice at 2 wk after i.o. treatment with vehicle (PBS) or MSC-sEVs every 3 d (started on day 0) (n = 6 mice/group). (E) UMAP visualization and the percentage of the retinal cells in 10 major clusters. (F) UMAP visualization of retinal cells and (G) percentage of different clusters in each group. (H) Network centrality analysis identified the strength of cell–cell communication for each cluster in the control and MSC-sEV groups.

We explored the function of MSC-sEVs in CNS protection and axon regeneration with the classic ONC model, a mouse model of CNS injury. A dose of MSC-sEVs at 5 × 10¹⁰ particles/mL or the PBS (phosphate buffered saline) vehicle in 2 µL volume was intravitreally injected every 3 d after ONC. Two weeks after ONC injury, a standard time point when neural protection was assessed, the retina and optic nerves were harvested to measure RGC survival and axon regeneration. Compared with the sham surgery control group, an enormous loss of RBPMS (RNA Binding Protein With Multiple Splicing)⁺ RGCs was observed in ONC+vehicle group (Fig. 1B). Intravitreal treatment with MSC-sEVs significantly enhanced RGC survival (Fig. 1B). Moreover, the injured RGC axons showed significant growth with i.o. administration of MSC-sEVs (Fig. 1 C and D). The enhanced effects of RGC survival and axon growth were maintained at the end of week 4 after i.o. treatment of MSC-sEVs (SI Appendix, Fig. S1 A−C). We also examined the effects of different doses. The dose of 5 × 10¹⁰ particles/mL showed the strongest effects in promoting RGC survival and axon regeneration, while the dose of 1.5 × 10¹⁰ particles/mL exhibited intermediate effects. In contrast, the lowest dose (5 × 10⁹ particles/mL) showed minimal neuronal restoration effect (SI Appendix, Fig. S1 D and E). Thus, the dose of 5 × 10¹⁰ particles/mL was selected as the optimal dose for further study based on its superior effects. We also applied sEVs from primary human umbilical cord MSCs (UMSCs). We previously reported that UMSC-sEVs could modulate macrophages and protect against corneal damage in dry-eye disease (19). Similarly, i.o. administration of UMSC-sEVs (5 × 10¹⁰ particles/mL) also significantly promoted neural survival and axon growth (SI Appendix, Fig. S2 A−E), indicating a common effect of MSC-sEVs in RGC survival and axon regeneration regardless of their source.

Retinal Infiltration of Mo/MΦ Is Required for RGC Survival and Axon Regeneration.

To explore the potential cellular and molecular mediators of MSC-sEVs-mediated RGC survival and axon regeneration, we next performed single-cell RNA sequencing (scRNA-seq) for retinal cells from ONC mice treated with MSC-sEV vehicle (PBS) or MSC-sEVs 2 wk after ONC. Unbiased, graph-based clustering by uniform manifold approximation and projection (UMAP) identified 10 major clusters expressing classical known markers (Fig. 1E and SI Appendix, Fig. S3 A and B). Unexpectedly, we noticed that i.o. treatment with MSC-sEVs induced significant infiltration of multiple blood-borne cell types in the retina, which was barely detected in vehicle-treated ONC mice (Fig. 1 F and G). Inferring the cell−cell communication patterns (CCC) by CellChat illustrated that MSC-sEVs triggered enormous changes in interactions among the cell clusters in the retinal microenvironment (Fig. 1H), including multiple significantly upregulated signaling pathways such as complement, SPP1, CSF, CXCL, CCL, and IGF (SI Appendix, Fig. S3C). Network centrality analysis identified that microglia and Mo/MΦ became dominant mediators of CCC after MSC-sEV treatment (SI Appendix, Fig. S3D), indicating that mononuclear cells might play important roles in regulating the retinal microenvironment.

The infiltration of immune cells in MSC-sEV-treated mice reminded us of the canonical immune-driven axon regeneration scenario, in which the influx of blood-borne myeloid cells improves neural protection and axon regeneration (3–5). We wondered whether MSC-sEVs protect against nerve damage by recruiting restorative blood-borne myeloid cells. Quantification of the immune cells via flow cytometry confirmed that MSC-sEVs significantly increased the infiltration of hematogenous myeloid populations in the retina, including CD11b⁺Ly6G⁺Ly6C⁺ neutrophils and CD11b⁺Ly6G⁻Ly6C⁺ Mo/MΦ (Fig. 2A and SI Appendix, Fig. S4). Similar immune populations were also observed in the UMSC-sEV-treated group (SI Appendix, Fig. S2F). Administering anti-Gr-1, a neutralizing antibody that binds to Ly6G and weakly to Ly6C, significantly decreased the MSC-sEV-induced infiltration of both Ly6G⁺ neutrophils and Ly6C⁺ Mo/MΦ in the retina (Fig. 2B) and also Ly6C⁺ Mo/MΦ in the blood, spleen, and brain (SI Appendix, Fig. S5), indicating anti-Gr-1 efficiently reduced the recruitment of circulating myeloid cells (SI Appendix, Fig. S5). As we expected, the MSC-sEVs induced an increase in RGC survival and axon regeneration was also completely eliminated by administering anti-Gr-1, suggesting that the two infiltrated peripheral myeloid cells were necessary (Fig. 2 C−F). To further dissect which myeloid population was needed, we introduced anti-Ly6G to specifically deplete Ly6G⁺ granulocytes. In contrast to anti-Gr-1 treatment, administering anti-Ly6G still significantly promoted RGC survival to a level comparable to that of isotype control (Fig. 2 C and D). Moreover, anti-Ly6G stimulated even stronger axon regeneration than the isotype control (Fig. 2 E and F). The application of anti-Ly6G completely blocked the retinal infiltration of Ly6G⁺ granulocytes; however, the infiltration of Ly6C⁺ Mo/MΦ was instead increased (Fig. 2B). The coexistence of Ly6C⁺ Mo/MΦ but not Ly6G⁺ neutrophils with a neuroprotective effect suggested that Ly6C⁺ Mo/MΦ were essential for MSC-sEV-mediated RGC survival and axon regeneration.

Fig. 2.

Retinal infiltration of Mo/MΦ was required for RGC survival and axon regeneration. (A) Quantification of different immune cells in retinas from ONC mice at 2 wk after i.o. treatment with vehicle (PBS) or MSC-sEVs every 3 d (started on day 0) (n = 5 mice/group) using flow cytometry. (B–F) ONC mice received i.o. treatment with vehicle (PBS) or MSC-sEVs (started on day 0) after intraperitoneal (i.p.) injections of anti-Ly6g, anti-Gr-1, or isotype control (started on day −1) every 3 d. Retinal cells or optic nerves were harvested on day 14. (B) Representative flow cytometric contour map and quantification of Ly6c/Ly6g gated myeloid cells (n = 3 mice/group). (C) Representative images of retinal whole mounts immunostained with anti-RBPMS (white). (Scale bar, 50 μm.) (D) RGC survival from each group normalized to healthy retinas (n = 5 retinas/group). (E) Representative images of optic nerves from longitudinal sections immunostained with anti-GAP43 (green). (Scale bar, 200 μm.) and (F) Quantification of GAP43⁺ regenerating axons (n = 5 nerves/group). Statistical significance was determined by one-way or two-way ANOVA and Sidak’s post hoc comparisons. (**P < 0.01, ***P < 0.001, ****P < 0.0001).

Ly6C^low Mo/MΦ Subsets Display Neural Restorative Features.

To characterize the Mo/MΦ subsets with RGC survival and axon regeneration effects induced by MSC-sEVs, we next investigated their features in the MSC-sEV-induced scRNA-seq dataset. The Mo/MΦ population could be unbiasedly clustered into two subsets (Fig. 3A). Both clusters showed enriched expression of Ccr2, the marker gene for blood-borne Mo/MΦ, but barely expressed the microglia-specific marker gene Tmem119 and the granuocyte-specific marker gene Cxcr2, further confirming that they were recruited hematogenous Mo/MΦ (Fig. 3B). Both the Mo/MΦ2 and Mo/MΦ1 clusters displayed strong CCC with the other retinal cell types (Fig. 3C). In particular, Mo/MΦ2 were the prominent mediators acting on immune cells, glial cells, and endothelial cells, suggesting their role as gatekeepers in modulating the neural-restorative microenvironment (Fig. 3D).

Fig. 3.

The Ly6C^low Mo/MΦ subset displayed restorative features. (A−F) scRNA analysis of retinal cells harvested at 2 wk after ONC from mice with i.o. vehicle (PBS) or MSC-sEVs treatment every 3 d (started on day 0). (A) UMAP visualization of the Mo/MΦ subclusters. (B) Violin plots showing the normalized and scaled Ccr2, Tmem119, and Cxcr2 expression levels in microglia and the two Mo/MΦ subclusters. (C) Network centrality analysis illustrated the strength of the CCC for each subclusters in MSC-sEV-treated group. (D) Circle plot showed the inferred outgoing communication patterns among the Mo/MΦ subsets and other cell clusters. Circle sizes are proportional to the number of cells in each cell group, and the edge width represents the communication probability. (E) Dotplot showing the distribution and scaled expression level of top differential expressed genes among the two Mo/MΦ subclusters and granulocytes. (F) Barplot showing the significantly enriched pathways that were up-regulated or down-regulated in the Mo/MΦ2 subcluster compared with the Mo/MΦ1. (G) ONC mice received i.o. treatment with MSC-sEVs (started on day 0) after i.p. injections of anti-Ly6g or isotype control (started on day −1) every 3 d. (Left) Histogram of the surface Ly6c expression of Mo/MΦ cells from the retinas. (Right) the percentages of Ly6c^low and Ly6c^high Mo/MΦ were quantified (n = 3 mice/group). Statistical significance was determined by two-way ANOVA and Sidak’s post hoc comparisons (**P < 0.01).

Hematogenous Mo/MΦ were recently classified into different subsets based on the differential expression of the surface marker protein Ly6C measured by flow cytometry (25), in which the Ly6C^hi population is the classical major Mo/MΦ with more proinflammatory properties, whereas the Ly6C^low subset is more protective and restorative with features of some alternative-activating M2-like markers and a higher capacity for phagocytosis and growth factors (26). Interestingly, compared with the Mo/MΦ1 cluster, the Mo/MΦ2 cluster displayed lower expression of the Ly6c2 gene, which encodes a component of the Ly6C complex (Fig. 3E), and genes for NF-kb signaling (Rel, Nfkb1, and Il1b), interferon responses (Ifitm1 and Isg15), antigen presentation (H2-Eb1), indicating their less proinflammatory features (Fig. 3E). The top up-regulated genes in the Mo/MΦ2 cluster included M2 markers (Chil3 and Retnla), phagocytosis-related genes (Trem2), lipid catabolic and lysosomal genes (Atp6v0d2, Apoe, and Lamp1/2), and several previously reported axon-growth-promoting factors [Igf1, Timp2, and Spp1 (27–29)] (Fig. 3E). Consistently, these features were not enriched in the granulocyte cluster. Gene set enrichment analysis (GSEA) confirmed the significantly positively enriched terms “lysosome”, “apoptotic cell clearance”, and “lipid catabolic process” in the Mo/MΦ2 cluster (Fig. 3F). Thus, the infiltrated Mo/MΦ subsets were functionally heterogenic, and the Mo/MΦ2 cluster could be referred to as the Ly6C^low Mo/MΦ population that was previously defined via flow cytometry with high restorative features (25, 26). Their partially alternatively activated status and enhanced functions of phagocytosis, lipid catabolism, and growth factors secretions could actively modulate the retinal microenvironment, help clear the cell debris, and provide trophic factors for positive neural survival and axon elongation.

Interestingly, we found that treatment with anti-Ly6G enhanced the Ly6C^low Mo/MΦ subset compared with their isotype control (Fig. 3G). The coincidence of the greater infiltration of Ly6C^low Mo/MΦ and stronger axon growth in the anti-Ly6G group further confirmed our hypothesis that Ly6C^low Mo/MΦ could be beneficial for RGC survival and axon regeneration.

G-CSF Promotes the Recruitment of Ly6C^low Mo/MΦ for Neuroprotection.

We wondered how i.o. treated MSC-sEVs induced the recruitment of Ly6C^low Mo/MΦ. We first investigated the timing of the influx of blood-borne myeloid cells. After i.o. administration of MSC-sEVs, we observed that a significant amount of CD11b⁺ cells started to accumulate in the lumen of veins around the optic disc at 6 h, and at 12 h, some CD11b⁺ cells passed through the veins and localized on the surface of the retina (Fig. 4 A–C). The dynamics of infiltrated leukocytes were also confirmed from cross-sections of the eyeballs (SI Appendix, Fig. S6).

Fig. 4.

Intraocular administration of MSC-sEVs induced significant G-CSF release. (A–C) Representative image of retina whole mount from perfused ONC mice with i.o. MSC-sEVs treatment at 1 h (A), 6 h (B), and 12 h (C) showing myeloid cells (CD11b, green) and retinal vessels (CD31, red). (Scale bar, 50 µm.) (D–G) Retinas and plasmas were harvested from ONC mice at 1 d after treatment with i.o. MSC-sEVs or vehicle (PBS). (D) Volcano plot of the differentially expressed genes and (E) GSEA of enriched pathways in retinas. (F) Quantification of Csf3 transcripts by qPCR in retinas and (G) G-CSF protein by ELISA in plasma (n = 3 mice/group). (H) Ridge plot showing the distribution of the Csf3r gene in different immune cell clusters from the scRNA-seq dataset of retinas harvested at 2 wk from ONC mice with i.o. MSC-sEVs treatment every 3 d (started on day 0). Statistical significance was determined by t test and Sidak’s post hoc comparisons. (*P < 0.05, **P < 0.01).

To explore the potential cytokines involved, we next performed bulk RNA-sequencing of retinas from ONC mice treated with or without MSC-sEVs for 1 d at the timepoint of myeloid cell infiltration. Treatment with MSC-sEVs up-regulated numerous genes (Fig. 4D). GSEA of KEGG pathways showed significantly enriched term of “cytokine-cytokine receptor interaction,” among which Csf3 (G-CSF), a colony-stimulating factor, was the most dramatically up-regulated cytokine gene (Fig. 4E). Csf3 upregulation was confirmed by qPCR (Fig. 4F). Moreover, in addition to the up-regulated Csf3 transcripts in the retina, we found significantly higher levels of G-CSF protein in the plasma from animals 1 d after administering MSC-EVs, indicating a systematic release of G-CSF at the timepoint of the peak influx of leukocytes (Fig. 4G). G-CSF is a proven clinical reagent to treat neutropenia that promotes the functions of neutrophils (30). Recently, studies have demonstrated that G-CSF is a potential tissue-reparative cytokine with multifaceted functions, including mobilizing hematopoietic stem cells and monocytes, modulating macrophages/microglia, and reducing apoptosis (31, 32). Intriguingly, the expression of Csf3r, a receptor for G-CSF, was more enriched in the Mo/MΦ2-ly6c^low subset among the observed retinal infiltrated blood-borne leukocytes from MSC-sEV-treated mice (Fig. 4H). In contrast, receptors for the other top up-regulated chemokines, such as Ccr1/Ccr2/Ccr5/Ccr4 (Ccl20/Ccl5 receptors), Cxcr1/2/3 (Cxcl1/Cxcl3/Cxcl10 receptors), showed no difference or even lower expression in the Mo/MΦ2-ly6c^low subset (SI Appendix, Fig. S7).

Considering the previously reported function of G-CSF (31, 32) and the interesting expression pattern of G-CSF and Csf3r, we proposed that G-CSF played an important role in recruiting restorative ly6c^low Mo/MΦ cells and neural repair. To validate this hypothesis, we next used an anti-G-CSF neutralizing antibody. Intraperitoneal administration of the anti-G-CSF antibody significantly reduced the infiltration of the Ly6G⁺ granulocytes, the Ly6C^low, and Ly6C^high Mo/MΦ induced by MSC-sEVs in the retina (Fig. 5A) and also reduced the percentage of circulating Ly6G⁺ granulocytes and Ly6C^low Mo/MΦ in the blood (SI Appendix, Fig. S8A). Moreover, the axon regeneration effect was also eliminated (Fig. 5 B and C), suggesting that G-CSF was required for the MSC-sEV-induced recruitment of myeloid cells and axon regeneration. Next, we explored the capacity of G-CSF to recruit the Ly6C^low Mo/MΦ and repair optic nerve injury. Similar to MSC-sEV treatment, i.o. injection of G-CSF induced significant infiltration of both Ly6G⁺ granulocytes and Ly6C⁺ Mo/MΦ cells in the retina (Fig. 5D). Moreover, compared with MSC-sEVs, G-CSF dramatically increased the percentage of the Ly6C^low Mo/MΦ population, confirming that G-CSF alone preferentially enhanced the recruitment of Ly6C^low Mo/MΦ (Fig. 5D). RT-qPCR of retina samples showed G-CSF treatment increased expression of Arg1, Mrc1, and Igf1 transcripts (Fig. 5E), the genes that were enriched in the Ly6C^low Mo/MΦ2 subset in scRNA (Fig. 3E). The preferential increase in the restorative Ly6C^low Mo/MΦ might lead to a stronger axon-regenerating effect. Indeed, we observed that i.o. administration of G-CSF significantly activated axon regeneration compared with the vehicle (PBS) control and even more strongly than administration of MSC-sEVs (Fig. 5 F and G).

Fig. 5.

Intraocular G-CSF promoted the recruitment of restorative Ly6C^low Mo/MΦ for axon regeneration. (A–C) ONC mice received treatment with i.o. MSC-sEVs (started on day 0) and i.p. injections of anti-G-CSF or isotype control (started on day −1) every 3 d. Cells or tissues were harvested on day 14. (A) Quantification of Ly6c/Ly6g gated myeloid cells in retinas. (B) Representative images and (C) quantification of GAP43⁺ (green) regenerating optic nerves (n = 4 nerves/group). (Scale bar, 200 μm.) (D and E) ONC mice received treatment with i.o. MSC-sEVs every 3 d or i.o. G-CSF every day for a total of 5 d (started on day 0) and retinal samples were harvested at 1 wk. (D) Representative flow cytometric contour map and quantification of Ly6c/Ly6g gated myeloid cells (n = 3 mice/group). (E) Arg1, Mrc1, and Igf1 transcripts were quantified by qPCR (n = 3). (F and G) ONC mice received treatment with i.o. MSC-sEVs every 3 d or i.o. G-CSF every day for a total of 5 d (started on day 0) and nerve samples were harvested at 2 wk. (F) Representative images and (G) quantification of GAP43⁺ (green) regenerating optic nerves. (Scale bar, 200 μm.) Statistical significance was determined by two-way ANOVA and Sidak’s post hoc comparisons. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

To validate whether the G-CSF recruited Ly6C^low Mo/MΦ directed stronger RGC survival and axon regeneration than other infiltrated myeloid subsets, we next introduced an adoptive transfer approach. Sorting the retinal infiltrates from the G-CSF-treated mice only yielded a low number of immune cells. We noticed that i.o. administration of G-CSF also induced more Ly6C^low Mo in the peripheral blood (SI Appendix, Fig. S8 B and C), suggesting that G-CSF not only increased retinal infiltrated Ly6C^low Mo/MΦ but also systemically expanded the Ly6C^low monocyte population. We instead harvested the peripheral myeloid cell subsets from the blood of G-CSF-treated donor mice and administered i.o. injection to the recipient mice with ONC injury (Fig. 6A). Three myeloid populations, namely, Ly6C^hi and Ly6C^low Mo and Ly6G⁺ neutrophils, were separately sorted and compared with the vehicle (PBS) control. Donor Ly6C^low Mo induced significantly more RGC survival and strongly activated axonal regeneration to a level similar to that with i.o. G-CSF (Fig. 6 B−E). Ly6C^hi Mo also induced significant but much less neural protection, whereas Ly6G⁺ neutrophils stimulated almost no effect compared with the vehicle (PBS) control (Fig. 6 B−E). These results illustrated that G-CSF-induced Ly6C^low Mo were capable of neural protection and axon regeneration and exerted the strongest effect among the myeloid populations.

Fig. 6.

Donor Ly6G^low Mo/MΦ cells promoted RGC survival and axon regeneration. (A) Schematic graph of the adoptive transfer in (B−E). A 2.5-μL cell (1 × 10⁵ cells/μL) suspension was i.o. injected to recipient mice on day 0, day 3, and day 6. Optic nerves and retinas were harvested at 2 wk. (B) Representative images of retinal whole mounts immunostained with anti-RBPMS (white) from each group of recipient mice. (Scale bar, 50 μm.) (C) RGC survival normalized to healthy retinas was quantified (n = 8 retinas/group). (D) Representative images of optic nerves from each group showing GAP43⁺ (green) regenerating axons. (Scale bar, 200 μm.) (E) Quantification of GAP43⁺ regenerating axons. (n = 4 nerves/group). Statistical significance was determined by one-way or two-way ANOVA and Sidak’s post hoc comparisons. (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

MSC-sEVs Primarily Target Retinal Mural Cells for G-CSF Release.

We next wondered how MSC-sEVs induced G-CSF release. We first explored MSC-sEV-targeting cells in vivo. sEVs were isolated from MSCs that stably expressed an exosome-reporter, that is, a mutated pHluorinM153R fluorescent protein fused within the second loop of CD63 (33). We identified a high and stable fluorescent signal in the MSC-pHluorin-sEVs sample from the Flow NanoAnalyzer (SI Appendix, Fig. S9). After i.o. administration of the MSC-pHlurin-sEVs, we observed that a significant pHluorin fluorescent signal started to accumulate around the optic disc and distribute along with the vessels on the inner surface of the retina at 6 h, the timepoint before myeloid cell infiltration (SI Appendix, Fig. S10A). The retinal capillaries are mainly composed of endothelial cells, with mural cells, including pericytes and vascular smooth muscle cells covering the outside and branches of astrocytes and microglia contacting them (34). Immunofluorescence staining showed that most of the pHluorin signal was localized in mural cells stained with NG2 or PDGFRB (Fig. 7A and SI Appendix, Fig. S10B), whereas it was barely observed in endothelial cells stained with CD31 or branches of astrocytes or microglia marked by GFAP or IBA1 (Fig. 7 B and C). We also observed significant pHluorin signal in the infiltrated myeloid cells and some in microglia in the inner retina at 1 d (SI Appendix, Fig. S11 A and B). In summary, i.o. MSC-sEVs were primarily taken up by mural cells before myeloid cell infiltration and were later partially engulfed by infiltrated immune cells and resident microglia.

Fig. 7.

MSC-sEVs were primarily taken up by retinal mural cells. (A–C) Representative image of retina whole mount from perfused ONC mice 6 h after i.o. MSC-pHluorin-sEVs (white). Vessels were visualized by anti-CD31 immunostaining (red). (A) The Upper row shows the landscape of the retinal mural cells (NG2, green) on the vessels around the optic nerve head at low magnification. The white arrows indicated the distribution of the pHluorin signal along the blood vessels. (Scale bar, 100 µm.) The Middle and Bottom rows show the overlapping pHluorin signal (white) in retinal mural cells (NG2, green) surrounding the endothelial cells (CD31, red) of the vessels at high magnification zoomed in on the areas labeled by the cyan boxes. (Scale bar, 5 µm.) (B and C) Nonoverlapping (yellow arrows) pHluorin signal (white) in (B) astrocyte branches (GFAP, green) or (C) microglia (IBA1, green). (Scale bar, 10 µm.)

An important function of mural cells in the CNS system is regulating the permeability of vessels for the blood−brain/blood−retina barrier and the recruitment of immune cells (34). We proposed that mural cells stimulated by MSC-sEVs secreted G-CSF and initiated the influx of myeloid cells. To validate this hypothesis, we utilized primary CNS mural cell cultures in vitro. Significant uptake of fluorescently labeled MSC-sEVs by primary mural cells was observed as early as 1 h (Fig. 8A). MSC-sEV treatment for 12 h significantly changed the transcriptome of mural cells, as determined by bulk RNA-sequencing analysis (Fig. 8B). GSEA showed positively enriched pathways related to leukocyte chemotaxis, cell−cell adhesion, and cytokine production (Fig. 8C). Transwell chemotaxis experiments showed that conditioned medium (CM) from mural cells treated with MSC-sEVs but not vehicle significantly promoted the migration of bone marrow (BM)–derived leukocytes, whereas medium or MSC-sEVs alone could not directly increase their mobilization (Fig. 8D and SI Appendix, Fig. S12). The up-regulated cytokine genes largely overlapped with those in retinas from mice i.o. treated with MSC-sEVs for 1 d, confirming that mural cells are an important source of cytokines in response to i.o. MSC-sEVs in vivo (Fig. 8E). Csf3 was significantly up-regulated (Fig. 8F), and G-CSF protein levels in CM from MSC-sEV-treated mural cells also strongly increased (Fig. 8G). Cotreatment with MSC-sEVs and dynasore, an inhibitor of clathrin-dependent endocytosis (35), largely blocked the accumulation of MSC-sEVs in the mural cells in vitro (Fig. 8H) and the upregulation of Csf3, validating that increased Csf3 release relied on the engulfment of MSC-sEVs by mural cells (Fig. 8I). Coincidently, immunofluorescence staining showed that a significant G-CSF signal was localized in mural cells in retinas from ONC mice treated with MSC-sEVs for 6 h (SI Appendix, Fig. S13). We further explored the requirement of mural cell-derived G-CSF for mobilizing the myeloid cells by the transwell migration assay. The CM from mural cells treated with MSC-sEVs significantly increased migration of BM-derived Ly6C^low Mo. However, their migrations were significantly reduced by anti-G-CSF neutralizing antibody (Fig. 8J). These findings indicate that MSC-sEVs-inducing mural cell-derived G-CSF signal is required for mobilizing the myeloid cells in vitro. In summary, mural cells primarily engulfed MSC-sEVs for G-CSF release.

Fig. 8.

Mural cells engulfed MSC-sEVs to release G-CSF and recruit leukocytes. (A) Uptake of fluorescently labeled MSC-sEVs (PKH dye, green) by primary mural cells in vitro at 1 h. (Scale bar, 10 μm.) (B and C) Bulk-RNA-seq dataset of mural cells treated with MSC-sEVs or vehicle (PBS) for 12 h. (B) Volcano plot of differentially expressed genes and (C) GSEA of enriched KEGG pathways visualized by dotplot. (D) Quantification of migrated cells in the chemotaxis experiment (n = 3). (E−G) Mural cells were seeded on 6-well plates and treated with MSC-sEVs (4 × 10⁹/mL) or vehicle (PBS) for 12 h. (E) Retinas were harvested from ONC mice at 1 d after treatment with i.o. MSC-sEVs or vehicle (PBS). Heatmap of bulk RNA-seq dataset showing the significantly upregulated cytokine genes in MSC-sEV-stimulated mural cells or retinas. (F) Csf3 transcripts in primary mural cells were quantified by qPCR (n = 3). (G) The G-CSF protein level in mural cell-CM was quantified by ELISA (n = 3 mice/group). (H) Uptake of fluorescently labeled MSC-sEVs (green) by primary mural cells was largely eliminated by cotreatment with dynasore (20 μM, started at −1 h). (I) MSC-sEVs induced Csf3 upregulation was significantly blocked by dynasore (20 μM, started at −1 h) in primary mural cells at 12 h (n = 3). (J) Quantification of migrated BM-derived Ly6g^-Ly6c^low Mo/MΦ cells by flow cytometry in the chemotaxis experiment (n = 4). Statistical significance was determined by t test or one-way ANOVA and Sidak’s post hoc comparisons. (**P < 0.01, ***P < 0.001, ****P < 0.0001).

Given the applied exogenous MSC-sEVs promoted RGC survival and axon regeneration, we were interested in the role of potent endogenous MSCs in the retina. Gli1 is a marker for resident MSCs in multiple tissues (36, 37). Utilizing the Gli1-CreERT; LSL-tdTomato reporter mice, we examined the Gli1⁺ cells. We demonstrated that Gli1⁺ cells exist in the retina. Interestingly, these Gli1⁺ cells remain closely associated with the retinal vasculature (SI Appendix, Fig. S14A). After ONC, the Gli1⁺ cells showed a similar distribution in the retina as observed in the physiological control condition (SI Appendix, Fig. S14 A and B). In the optic nerve, we observed that Gli1⁺ cells mainly localized in the optic nerve head, and the positive signals were indistinguishable between the ONC and control groups (SI Appendix, Fig. S14 C and D). Consistent with the Gli1 reporter mice, when analyzing a public scRNA database (38), we found that low abundant Gli1 expression in vascular cell cluster in the retina and in general Gli1⁺ cells represent a small population (SI Appendix, Fig. S15A). Furthermore, ONC did not significantly alter Gli1 expression in the retina (SI Appendix, Fig. S15 A and B). We also performed flow cytometry analysis to quantitatively define MSC-like cells using independent markers other than Gli1. The CD105⁺CD29⁺Sca1⁺CD45⁻CD11b⁻ MSC-like cell population (14) was identified and their percentage was not significantly changed after ONC (SI Appendix, Fig. S15 C and D). Together, our findings demonstrate that MSC-like cells may exist in the retina and optic nerve, but did not alter much after ONC.

Discussion

We report a potent neuroprotective effect of MSC-sEVs via an immunomodulatory pathway. Intraocularly administered MSC-sEVs were primarily engulfed by retinal mural cells, triggering the release of G-CSF that facilitates the recruitment of a neural-restorative Ly6C^low Mo/MΦ population into the injured site to repair optic nerve injury (Fig. 9). Our findings provide a rationale to optimize MSC-sEV-related therapies. Moreover, the identified key roles of G-CSF and the restorative Ly6C^low Mo/MΦ infiltrates in optic nerve regeneration could lead to the development of therapies for optic neuropathies.

Fig. 9.

Schematic graph of the therapeutic mechanism of MSC-sEVs for repairing optic nerve injury. MSC-sEVs delivered by i.o. injection are primarily taken up by retinal mural cells, inducing the release of the colony-stimulating factor G-CSF. This leads to the recruitment of a Ly6C^low Mo/MΦ population that significantly promotes RGC survival and axon regeneration.

Although MSCs are known for their immunomodulatory properties, the immunoregulatory function of MSC-sEVs in repairing retinal injuries has not been reported. It was hypothesized that some cargoes such as miRNAs and growth factors in MSC-sEVs directly regulate the neurons and preserve their functions (20–23, 39). In our study, by dissecting the target cells and molecular responses of MSC-sEVs in vivo and in vitro depleting myeloid mediators, we identified that MSC-sEVs stimulated beneficial neuroinflammation to achieve positive neural outcomes. Elie Metchnikoff, the father of natural immunity, believed that there could be no cure without inflammation (40). After optic nerve or spinal injury, acute inflammatory stimulation improves neural survival and provides fuel for axon regeneration (41). Our findings revealed that i.o. administered MSC-sEVs acted as a unique neural immunomodulating reagent, stimulating the recruitment of hematogenous immune cells and reprogramming of the retinal immune microenvironment for RGC survival and axon regeneration. Of note, the mesenchymal stem cells used in our study were not treated with a specific stimulus. In fact, MSC-sEVs preparation from this type of “non-stimulated” mesenchymal stem cells are widely used. The non-stimulated mesenchymal stem cell–derived EVs exhibit a wide range of tissue-repairing effects and the therapeutic efficacy could be further enhanced through some specific relevant stimulation of MSC (17).

Interestingly, instead of directly regulating immune cells in a wide range of MSC and MSC secretome-mediated tissue-repair conditions (14), we reported that MSC-sEV signals were primarily sensed and transduced by retinal mural cells to initiate immunomodulation. Before the influx of myeloid cells, we found predominant engulfment of MSC-sEVs by retinal mural cells instead of other retinal resident cells, such as microglia or astrocytes in vivo (Fig. 7). MSC-sEVs alone were not sufficient to recruit immune cells in vitro, while the secretome from MSC-sEV-stimulated mural cells could mobilize immune cells (Fig. 8). These observations demonstrated that retinal mural cells are primary MSC-sEV-targeting cells, which might be correlated with several features. Mural cells, especially pericytes, show specific features in the CNS. Vessels in the CNS have the highest pericyte coverage, with an endothelial: pericyte ratio estimated between 1:1 and 3:1, in contrast to a ratio of 100:1 in other tissues such as the muscle (42), which makes the CNS pericytes unique in that they contact the nervous parenchymal cells and the blood circulation system. The retinal mural cells on the vessels around the optic nerve head could better sense the intravitreal MSC-sEVs, as they are cells on the inner surface of the retina facing the vitreous body (Fig. 9). In addition to their advantages in physical location, CNS pericytes have been recently recognized as active secretors of leukocyte effectors for leukocyte trafficking (43). Brain pericytes but not endothelial cells secreted cytokines such as G-CSF and Ccl2 in vitro when stimulated by IL-17 and in vivo upon systemic inflammation (44, 45). Thus, it is reasonable that retinal mural cells stimulated by MSC-sEVs represent the initial source of key cytokines coordinating the following immune responses. In addition to mural cells, microglia also engulf MSC-sEVs at later stages, and it would be interesting to explore the role of microglia in regulating immune cascades.

Our data described a population of Ly6C^low Mo/MΦ as a key immune population for RGC survival and axon regeneration, and administering G-CSF could enhance its recruitment. A recent study reported that a subset of Ly6G^low neutrophils promoted CNS restoration (8). We also observed that MSC-sEVs induced infiltration of Ly6G⁺ neutrophils in addition to the Ly6C^low monocytic cells (Fig. 2B). However, when the infiltration of all Ly6G⁺ neutrophils was depleted via an anti-Ly6G antibody, MSC-sEV-mediated neural protection remained (Fig. 2), and direct i.o. administration of Ly6G⁺ neutrophils barely protected against optic nerve injury. Instead, when the infiltration of the Ly6C^low Mo/MΦ subset was depleted, the neuroprotective effect disappeared, and when the Ly6C^low Mo/MΦ subset was enriched by administering anti-Ly6G, i.o. G-CSF stimulation, or adoptive transfer, the axon regeneration effect was more dramatic (Figs. 2, 5, and 6). The Ly6C^low Mo/MΦ subset displayed high levels of growth factors such as Igf1 and phagocytosis and lipid catabolism signaling (Fig. 3). Such features could help replenish deficient growth signaling, remove debris, and prevent gliosis that hinders the growth of CNS axons (46, 47). It is plausible that the MSC-sEVs and their induced Ly6C^low Mo/MΦ cells could have a similar restorative effects on other retinal neuroretinal cell types, including photoreceptors in other rapid or chronic retinal neurodegeneration models, whose pathogenic environment including debris accumulation, lack of survival signals, and harmful inflammation contributes to the death of the retinal neurons (48). Thus, our findings proposed that the Ly6C^low Mo/MΦ population represents a significant immune subset for neural survival and axon regeneration. Moreover, our data demonstrated that i.o. G-CSF displayed a unique function in modulating the infiltration of restorative Ly6C^low Mo/MΦ in the injured retina, not only providing a local cytokine gradient for attracting Ly6C^low Mo/MΦ but also stimulating their production and release from the BM. G-CSF is a medicine that has long been applied in the clinic for neutropenia (30). Our findings reveal a previously unrecognized G-CSF-to-Ly6C^low Mo/MΦ signaling axis initiated by MSC-sEVs that protects against optic nerve injury, representing a viable immune therapeutic strategy based on G-CSF and MSC-sEVs for optic neuropathies.

Materials and Methods

Animals.

C57BL/6J mice aged approximately 8 wk were provided by Gempharmatech Biotechnology Co., Ltd. All mice were raised in a specific pathogen-free room in the Animal Care Center of Zhongshan Ophthalmic Center, Sun Yat Sen University.

ONC Surgery and i.o. Injection.

The optic nerve was exposed through an incision in the conjunctiva. Immediately after ONC, the posterior chamber of the eye was injected (i.o.) with 2 μL of sEV vehicle (PBS) or sEV (1 × 10⁸ particles in total in PBS) every 3 d. Detailed descriptions can be found in SI Appendix, Materials and Methods.

Study Approval.

The present study with animals was reviewed and approved (protocol No. Z2022018) by the ethics committee of Zhongshan Ophthalmic Center, Sun Yat Sen University.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Our sequencing data are available at Gene Expression Omnibus under the accession number GSE224068. A public scRNA dataset [https://singlecell.broadinstitute.org/single_cell/study/SCP1785/mouse-adult-retina-and-eyecup-atlas-and-optic-nerve-crush-time-series (38)] was analyzed to identify the Gli1⁺ cells in the retina. All data needed to evaluate the conclusions in the paper are present in the paper and/or the SI Appendix.