Modulation of Rac1/PAK1/connexin43-mediated ATP release from astrocytes contributes to retinal ganglion cell survival in experimental glaucoma

Connexin43 (Cx43) is a major gap junction protein in glial cells. Mutations have been found in the gap-junction alpha 1 gene encoding Cx43 in glaucomatous human retinas, suggestive of the involvement of Cx43 in the pathogenesis of glaucoma. However, how Cx43 is involved in glaucoma is still unknown. We showed that increased intraocular pressure in a glaucoma mouse model of chronic ocular hypertension (COH) downregulated Cx43, which was mainly expressed in retinal astrocytes. Astrocytes in the optic nerve head where they gather and wrap the axons (optic nerve) of retinal ganglion cells (RGCs) were activated earlier than neurons in COH retinas and the alterations in astrocytes plasticity in the optic nerve caused a reduction in Cx43 expression. A time course showed that reductions of Cx43 expression were correlated with the activation of Rac1, a member of the Rho family. Co-immunoprecipitation assays showed that active Rac1, or the downstream signaling effector PAK1, negatively regulated Cx43 expression, Cx43 hemichannel opening and astrocyte activation. Pharmacological inhibition of Rac1 stimulated Cx43 hemichannel opening and ATP release, and astrocytes were identified to be one of the main sources of ATP. Furthermore, conditional knockout of Rac1 in astrocytes enhanced Cx43 expression and ATP release, and promoted RGC survival by upregu-lating the adenosine A3 receptor in RGCs. Our study provides new insight into the relationship between Cx43 and glaucoma, and suggests that regulating the interac-tion between astrocytes and RGCs via the Rac1/PAK1/Cx43/ATP pathway may be used as part of a therapeutic strategy for managing glaucoma.


| INTRODUCTION
Connexin43 (Cx43) forms specialized intercellular connections that help maintain cell and tissue homeostasis. Mutations in the gapjunction alpha 1 gene encoding the Cx43 protein can cause oculodentodigital dysplasia, which is accompanied by ocular disorders (Laird, 2014;Wang et al., 2019). These mutations are also found in patients with open-angle glaucoma (Huang et al., 2015), suggesting that Cx43 may be involved in the pathogenesis of glaucoma.
Glaucoma, the leading cause of irreversible blindness, is characterized by optic nerve head (ONH) lesions and visual field loss. Retinal ganglion cell (RGC) apoptotic death is the fundamental pathological process in glaucoma. High intraocular pressure (IOP) is the most important risk factor for glaucoma (Frankfort et al., 2013). Because glaucoma is a neurodegenerative disease, its pathogenesis is complex and diverse. Astrocytes are the major glial cells within the ONH (Lozano et al., 2019); They surround a nonmyelinated sheath of RGC axons. Astrocytes provide energy to the RGCs via morphological remodeling, and Cx43 plays an important role in this process.
Cx43 is the main gap-junction protein in retinal glial cells including macroglia (Müller cells and astrocytes) and microglia (Chew et al., 2011;Zahs et al., 2003). A hexamer of connexins forms a connexon. The connexons in one cell are paired with the same type or different types of connexons in adjacent cells to form gap junctions.
These gap junctions allow direct transfer of small molecules (<1000 Da) between cells (Nicholson, 2003). Moreover, in the membrane, no matching connexons are found to form hemichannels. The hemichannels operate as gated pores, opening to the extracellular space to release gliotransmitters, nicotinamide adenine dinucleotide, or D-serine; they also facilitate purinergic signaling (Falk et al., 1997;Montero & Orellana, 2015). In retinal astrocytes, the metabolic network mediated by Cx43 mitigates bioenergetic stress and the impact of neurodegenerative disease processes (Cooper et al., 2020).
Although Cx43 has been implicated in the pathology of glaucoma, an exact association between Cx43 and RGC loss has not been completely elucidated. Previous studies showed that a partial transection of the optic nerve induced a biphasic upregulation of retinal Cx43 protein, and RGC loss was associated with an inflammatory response mediated by astrocytes (Chew et al., 2011). In glaucomatous retinas, IOP elevation may induce the loss of gap junction communication and the redistribution of Cx43 in astrocytes, thereby affecting homeostasis in RGC axons and stimulating glaucomatous neuropathy (Malone et al., 2007). Therefore, modulating the activity of Cx43 hemichannels may be a promising therapeutic strategy for managing eye-related diseases, including glaucoma (Chen et al., 2015).
Rac1 is an important member of the Rho GTPase small G protein family. It controls cytoskeleton rearrangement, which results in structural changes in cell junctions. Rac1-mediated signaling involves Ncadherin, MENA, and PAK1, and modulates the expression and localization of Cx43 (Adam et al., 2010;Matsuda et al., 2006;Ram et al., 2014). Previously, we showed that Rac1 was activated in retinas with chronic ocular hypertension (COH) experimental glaucoma model, and regulated RGC autophagy and apoptosis . In this study, we investigated whether and how Rac1 regulated Cx43 in retinal astrocytes and crosstalk between astrocytes and RGCs in a glaucoma mouse model.

| Animals
All animal experiments were performed in accordance with the guidelines issued by the National Institutes of Health Guide for Care and Use of Laboratory Animals and were approved by the Fudan University Animal Care Committee (Shanghai, China) and the Bioethical Committee at Fudan University (approval no. 20210302-148). Every effort was made to minimize animal pain and discomfort during the experiments. C57BL/6 male mice were purchased from the SLAC Laboratory Animal Company (Shanghai, China). Rac1 flox/flox mice were obtained from the Jackson Laboratory (Bar Harbor, ME, USA). All mice were maintained under in 12-h light/12-h dark cycle with ad libitum access to food and water. Age-matched mice were randomly allocated to different groups in accordance with the requirements of each experiment.

| The chronic ocular hypertension mouse model
The COH mouse model was generated as described in our previous study . Briefly, mice were anesthetized by intraperitoneal injections of 2% sodium pentobarbital (40 mg/kg) and oxybuprocaine hydrochloride eye drops (4 mg/mL) were applied to the inner edge of the eyelids to provide topical anesthesia. Tropicamide eye drops (5 mg/mL) were used to dilate the pupils. Using an ophthalmic surgical microscope (OPMI VISU 140, Carl Zeiss, Jena, Germany), magnetic microbeads (diameter ≈ 9 μm BioMag ® Superparamagnetic Iron Oxide, Bangs Laboratories, Ins, USA) (2 μL) were injected into the anterior chamber of right eye without damaging the iris or lens. The beads were evenly distributed around the iridocorneal angle using a handheld magnet (0.45 Tesla). In the sham-operated control groups, the same procedures were performed but 2 μL of normal saline (NS) was injected. Chlortetracycline ointment was smeared onto the surface of each eyeball to prevent subsequent infections. IOP was measured using a handheld rebound tonometer (iCare Tonolab, Vantaa, Finland) at 9-10 am to circumvent any variation due to circadian rhythms. The IOP of both eyes was recorded immediately after surgery (G0d), on day 4 after surgery (G4d), and at the first, second, third and fourth week after surgery (G1w, G2w, G3w and G4w, respectively). The mean value of five consecutive measurements that deviated by less than 5% was recorded. The animals in groups with IOP values of 5-15 mm Hg higher than those in the control groups were used for further experiments; otherwise, these mice were not used in experiments.

| Western blotting analyses
Western blotting analyses were performed as described previously (Dong et al., 2015;Zhang et al., 2020). Retinas and a 5-8-mm section of optic nerve proximal to the eye globe were collected at different time points from COH mice; two optic nerves from different mice with the same treatment were pooled as one sample. The antibodies used in this study are listed in Table S1. Digital images were obtained and analyzed using FluorChem E System (ProteinSimple, Santa Clara, CA, USA) and AlphaView SA (version 3.4, ProteinSimple).

| Co-immunoprecipitation
Co-immunoprecipitation experiments were performed using a Pierce Co-immunoprecipitation kit (Thermo-Fisher Scientific, Waltham, MA, USA) in accordance with the manufacturer's instructions to test interactions between Cx43 and active Rac1 or phospho-PAK1 . The Cx43 antibody was purified using a Pierce Antibody Clean-up kit (Thermo Fisher Scientific). Retinal protein extracts were obtained from control (Ctr) and COH retinas of G1w mice, and incubated with affinity purified Cx43 antibodies coupled to resin. The immunoprecipitated complexes were detected using antibodies against phospho-PAK1 or active Rac1 following western blotting.

| Immunofluorescence staining
Immunohistochemical staining was performed as described previously (Dong et al., 2015). Briefly, the eye cup was fixed with 4% paraformaldehyde (PFA) in 0.1 M phosphate-buffered saline (PBS) (pH 7.4) for 4 h at room temperature. After dehydration, the retinas were embedded in optical cutting temperature compound (Tissue Tek, Torrance, CA), and 14-μm thick retinal sections were cut vertically using a frozen microtome (Leica, Nussloch, Germany). Flat-mounted retinas were used in some experiments. The sections or flat-mounted retinas were blocked with 5% normal donkey serum and 1% bovine serum albumin in PBS containing 0.1% Triton X-100 for 2 h at room temperature.
The samples were incubated overnight (for retinal sections) or 3 days (for flat-mounted retinas) at 4 C with primary antibodies, and then incubated with the secondary antibodies. All antibodies used in this study were listed in Table S1. Finally, samples were mounted using anti-fade mounting medium with 4 0 ,6-diamidino-2-phenylindole (DAPI, Vector Laboratories, Burlingame, CA, USA), and images were captured using an FV1000 confocal laser scanning microscope (Olympus, Tokyo, Japan) under the same excitation/detection conditions in each experiment. To avoid introducing reconstruction stack artifacts and to accurately evaluate the double markers, a single-layer optical slice was scanned at 1.0 μm intervals. To quantify immunofluorescence in the flat-mounted retina, four fields were randomly evaluated from the center (<0.8 mm from the ONH) and the periphery (0.8-1.6 mm from the ONH; Figure 9b). The mean fluorescence intensity of each micrograph was analyzed using Image-Pro Plus software by two independent investigators who were blinded to the experimental groups; data were expressed as the integrated optical density (IOD)/area.

| RNA extraction and quantitative real-time polymerase chain reaction
Total RNA was extracted from retinas using MiniBEST Universal RNA Extraction kit (TaKaRa Bio, Inc., Shiga, Japan) and reverse transcribed using PrimeScript RT reagent kit with gDNA Eraser (Takara Bio, Inc.).
Quantitative real-time polymerase chain reaction (qPCR) was performed using Green Premix Ex Taq II (Takara Bio, Inc.) and a QuantStudio 3 Real-Time PCR system (Thermo Fisher Scientific, Rockford, IL, USA), as described previously (Hu et al., 2021). The thermal cycling conditions consisted of an initial denaturation step at 95 C for 30 s, followed by 40 cycles of 95 C for 5 s, 60 C for 40 s, and 72 C for 30 s. Data were analyzed using 2 ÀΔΔct method. The primers used in this study are list in Table S2.

| Multiplexing of RNAscope in situ hybridization with immunostaining
The RNAscope Multiplex Fluorescent Reagent kit v2 (Advanced Cell Diagnostics, Newark, CA, USA) and RNA-Protein Co-Detection Ancillary kit (Advanced Cell Diagnostics) were used for immunostaining glial fibrillary acidic protein (GFAP) and Rac1 transcript puncta in accordance with the manufacturer's instructions. The retinal sections were fixed in fresh 4% paraformaldehyde solution and dehydrated using ethanol. Then, the sections were immersed in H 2 O 2 for 10 min at room temperature followed by co-detection target retrieval for 5 min at 98-102 C. The sections were incubated overnight at 4 C with GFAP antibodies diluted in Co-Detection Antibody Diluent (Advanced Cell Diagnostics). Next, the sections were fixed in fresh 4% PFA at room temperature and treated with RNAscope Protease Plus for 30 min at 40 C. Rac1 mRNA was detected using the RNAscope Multiplex Fluorescent Reagent kit v2 in accordance with the manufacturer's instructions. The ubiquitous Ppib gene was used as a positive control and DapB, which is not expressed in retinas, or a diluted probe dilution was used as a negative control. The secondary antibody was added immediately following RNAscope in situ hybridization. Finally, sections were mounted using anti-fade mounting medium with DAPI, and images were captured using an FV1000 confocal laser scanning microscope (Olympus, Tokyo, Japan). Six fields were randomly captured for each sample, and the numbers of Rac1 transcript puncta colocalizing with GFAP + cells in the nerve fiber layer (NFL) were recorded per unit length (for vertical samples) or per cell (for flat-mounted retinas) by two independent researchers who were blinded to the treatment groups.

| Measurement of ATP levels
Retinas were collected from eyes and ATP levels were determined by luciferin-luciferase assays using an ATP Detection kit (Beyotime Biotechnology Inc., Shanghai, China) in accordance with the manufacturer's instructions. In some experiments, the ecto-ATPase inhibitor ARL67156 (100 μM, Sigma-Aldrich, MO, USA) was added to tissue lysate to prevent ATP degradation. ATP levels were calculated using an ATP standard curve (1 nM-10 μM ATP) and adjusted for protein concentrations, which were determined using the bicinchoninic acid method (Pierce). ATP levels were expressed as ratios relative to control samples.

| Ethidium bromide uptake
Hemichannel opening-dependent ethidium bromide (EtBr) uptake assays were used to assess the function of hemichannels in astrocytes (Slavi et al., 2018). EtBr is a cell membrane impermeable compound.
However, it can pass through the opened hemichannels and shows enhanced fluorescence after binding to DNA. Retinas were dissected from the eyeballs and immediately incubated in PBS containing 4 μM EtBr (Sigma-Aldrich, MO, USA) for 10 min at room temperature. Flatmounted retinas were analyzed for GFAP by immunostaining (Slavi et al., 2018). The EtBr was excited at 530 nm and emitted fluorescence was detected at 590 nm. EtBr uptake was quantified as described in the immunofluorescence staining section.

| Intravitreal injections
Intravitreal injections were performed as described previously (Dong et al., 2015). The following drugs were dissolved in normal saline (NS):

| Fluorescence imaging of ATP sensors
AAV2-GfaABC1D-ATP1.0 (Vigene Bioscience, Maryland, USA) is a genetically encoded G protein-coupled receptor-activation-based sensor for ATP (GRABAtp) that is expressed in GFAP + cells; the quantity of extracellular ATP is indicated by the intensity of fluorescence produced by GFP (Wu et al., 2022). AAV viruses were injected into the retinas of C57BL/6 mice 3 weeks before the experiments. Normal saline or NSC23766 was injected into the vitreous bodies of eyes on the day before the operations. The retinas were collected at G1w and immediately incubated in a solution of pre-oxygenated artificial cerebral spinal fluid. The areas with fluorescence were assayed first to ensure that that retinal cells were alive inside the Nikon Live Cell Workstation (Nikon Corporation, Tokyo, Japan). Images were analyzed using Image Pro-Plus software (Image-Pro Plus, RRID: https:// scicrunch.org/resolver/SCR_007369, Media Cybernetics, Inc. USA).
The mean fluorescence intensity was expressed as the integrated optical density (IOD)/area.

| Experimental design and statistical analyses
At least three (typically n ≥ 6) animals were included in each group for every experiment. Control data were collected from animals treated with vehicles (typically normal saline). All data are presented as the mean ± SEM. Data were analyzed using GraphPad Prism software (version 6.0; GraphPad Software, Inc., San Diego, CA, USA). Data distribution was assessed using the Shapiro-Wilk test. Differences between two groups were compared using two-tailed unpaired t-tests. One-way analysis of variance (ANOVA) with Tukey's multiple comparison test was used to compare data from multiple groups.
A p value <.05 was considered statistically significant.

| Cx43 is mainly expressed in astrocytes in mouse retinas
Cx43 is an integral membrane protein that is expressed in glial cells (Yamamoto et al., 1990). We identified the types of glial cells that expressed Cx43 in mouse retinas. As shown in Figure 1

| Dynamic changes of Cx43 expression and phosphorylation in COH retinas
The COH mouse model was successfully generated, and the IOP of the operated eyes in these mice ranged from 14.94 ± 0.30 mm Hg at G4d (n = 41, all p < .001) to 17.70 ± 0.40 mm Hg at G4w (n = 28, all p < .001). These values were significantly higher than those observed at G0d or in corresponding unoperated eyes (9.93 ± 0.02 mm Hg, n = 141) or sham-operated eyes (9.91 ± 0.02 mm Hg, n = 150) ( Figure 2a).
(Figure 2f-i). EtBr uptake was also used to assess the function of Cx43 hemichannels. Compared with Ctr retinas, EtBr uptake in astrocytes labeled with GFAP decreased from G4d to G1w, then obviously increased at G2w, and gradually decreased again from G3w ( Figure 3a-g). These results, combined with Cx43 phosphorylation data, indicated that the frequency of hemichannels opening decreased and then increased during IOP elevation.
3.3 | Reduction in Cx43 expression was caused by alteration in astrocyte plasticity in the ONH of COH mice We investigated the mechanisms underlying the changes in Cx43 expression in astrocytes. Astrocytes form a mesh-like layer on the surface of the retina and wrap around RGC axons that converge to become the optic nerve in the ONH (Shinozaki & Koizumi, 2021). Therefore, we selected vertical retinal sections through the ONH to observe astrocyte activation. The c-Fos gene, which is a rapidly activated in stimulated cells, was used as a neural activity marker (Cruz Mendoza et al., 2022). Compared with sham-operated control retinas (Ctr), c-Fos-positive fluorescence signals progressively increased in GFAP-labeled astrocytes found in COH retinas from G4d to G4w (indicated with arrow heads, Figure 4b3-f3). These signals were present in the ganglion cell layer (GCL) (mainly RGCs) and in both the inner nuclear layer (INL) and outer nuclear layer (ONL) from G1w ( Figure 4c3-f3, indicated with arrows). To assess the temporal activation of astrocytes and RGCs, we examined vertical retinal sections that did not pass through the ONH ( Figure S1); we found that c-fos-positive signal appeared first in the GFAP-positive NFL and then in the GCL. We also labeled RGCs with Brn3a antibodies to show the relative positions of astrocytes and RGCs in the retina ( Figure S1h).
These results suggest that astrocytes are activated earlier than neurons in COH retinas. The increase in GFAP protein levels in COH retinas confirmed that macroglial cells were activated after IOP elevation (Figure 4g,h). Activated astrocytes may produce inflammatory factors in response to injury. In our COH retinas, mRNA levels of proinflammatory factors [e.g., tumor necrosis factor (TNF)-α, interleukin (IL)-6, and anti-inflammatory factor transforming growth factor (TGF)-β] increased immediately, and mRNA level of inducible nitric oxide synthase (iNOS) was upregulated at G2w (Figure 4i-l). Therefore, activated astrocytes may stimulate the expression of inflammatory factors.
Next, we identified the glial cells that express growth-associated protein 43 (Gap43) in the coronal sections of the optic nerves; Gap43 is implicated in axon and astrocyte plasticity and regeneration (Hung et al., 2016;Tedeschi et al., 2009). Müller cells are distributed throughout the retina and Müller cell bodies are found in the INL. It is noted that GS-positive cells of the optic nerve were fibrous astrocytes (Domercq et al., 1999). Gap43 was mainly expressed in astrocytes labeled with GFAP. There was little Gap43 in astrocytes labeled with GS, less in oligodendrocytes labeled with O4, and none in microglia labeled with Iba1 ( Figure S2).
F I G U R E 3 Changes in Cx43 hemichannel activity in COH retinas. (a-g) Representative images showing hemichannel activity as measured by ethidium bromide (EtBr) uptake (red) in astrocytes taken from flat-mounted COH retinas at different times (a-f) and corresponding quantification (g). a2-f2 are enlarged images of the white rectangles in a1-f1, respectively. Scale bar: 50 μm. n = 5-6. **p < .01 versus Ctr. One-way ANOVA with Tukey's multiple comparison test.
Although the number of Gap43-labeled cells in the optic nerve remained unchanged throughout IOP elevation (G4d-G4w) ( Figure 5a,b), the fluorescence intensity of Gap43 increased (Figure 5a,c). The number of Gap43 and Cx43 double-positive cells in COH optic nerves significantly decreased from G4d to G2w (Figure 5a,d). The changes in Cx43 protein level and F I G U R E 4 Activation of astrocytes in the COH retinas. (a-f) Double-immunofluorescence staining showing expression of c-Fos (red) and GFAP (green) in the optic nerve heads of sham-operated retinas (Ctr) in retinal vertical slices (a) and corresponding images at different times after the operations (b-f). Representative images from at least three mice. Scale bar: 50 μm for all images. Arrow heads indicate c-Fos expression in astrocytes. Arrows indicate c-Fos expression in RGCs. (g,h) Representative immunoblots (g) and quantification (h) of GFAP expression in Ctr and COH retinas (G4d to G4w). All data are normalized to their corresponding β-actin levels and to Ctr. n = 6 for each group. *p = .0199; **p < .01; and ***p < .001 versus Ctr. (i-l) Bar charts showing the changes in mRNA levels of pro-inflammatory factors (iNOS, TNF-α, and IL-6) (i-k) and anti-inflammatory factor (TGF-β) (l) in COH retinas under different conditions as shown in panel g. All data are normalized to Ctr. n = 5. *p < .05; **p = .0013; and ***p < .001 versus Ctr. One-way ANOVA with Tukey's multiple comparison test.

| Rac1 regulates the astrocyte response in COH retinas
Cytoskeletal changes affect the plasticity of astrocytes and Rac1 is involved in regulation of the actin cytoskeleton (Machesky & Hall, 1996. As in our previous study, we found that the ratio of active to total Rac1 in COH retinas significantly increased at G4d and G3w (Figure 6a Representative immunoblots (e) and densitometric quantification (f-h) showing changes in Cx43, p-Cx43 ser373 , and p-Cx43 ser368 protein levels in the optic nerves of control (Ctr) mice and COH mice at different times (G4d-G4w). Red dashed lines indicate the ratios of p-Cx43 Ser373 /Cx43 in (g) and p-Cx43 Ser368 /Cx43 in (h). All data are normalized to corresponding β-actin and to Ctr. n = 5-6. *p < .05, **p < .01; ***p < .001 versus Ctr. One-way ANOVA with Tukey's multiple comparison test.
in COH retinas treated with normal saline (NS G1w), and this increase was reversed by administering NSC23766 (NSC), an inhibitor of Rac1 (Figure 6c,d). The morphological changes in astrocytes labeled with GFAP (i.e., smaller cell bodies with slender processes to the larger cell bodies with hypertrophic protrusions after IOP elevation) were partially reversed by the Rac1 inhibitor NSC23766 (Figure 6e,f). At G1w, NSC23766 treatment significantly increased EtBr uptake through open hemichannels (shown in red) in astrocytes labeled with GFAP (shown in green) and uptake was blocked by the Cx43 inhibitor Gap26 (Figure 6g,h). These results suggest that Rac1 regulates hemichannels opening via Cx43 in astrocytes.
The opposing expression patterns of Rac1 and Cx43 in COH retinas suggest that Rac1 may regulate the expression of Cx43. Indeed, intravitreal injections of the Rac1 inhibitor NSC23766 significantly increased Cx43 protein levels in COH retinas (Figure 7a,b). Additionally, p-Cx43 Ser373 levels increased and p-Cx43 Ser368 decreased ( Figure 7a,c,d). Furthermore, Cx43 protein levels in membrane components significantly increased (Figure 7e,f). The effectiveness of NSC23766 and IPA-3 (a PAK1 inhibitor) were confirmed by the reduced activity of Rac1 and its downstream effector PAK1, respectively ( Figure S3). The effect of IPA-3 treatment on Cx43 phosphorylation was similar to that of NSC23766, although Cx43 protein levels were unchanged (Figure 7g-j). The interactions between Cx43 and active Rac1 or p-PAK1 were further revealed by Co-immunoprecipitation experiments (Figure 7k,l). These results suggest that the Rac1/PAK1 signaling pathway directly regulates the phosphorylation of Cx43.

| Inhibition of Rac1 mediated ATP release from astrocytes
In COH retinas, the extracellular ATP concentration increased gradually from G2w to G4w (Figure 8a). However, when the ecto-ATPase inhibitor ARL67156 was added to the tissue lysate to prevent ATP degradation, a significant increase in ATP was also observed at G4d and G1w (Figure 8b). This suggests that ATP is rapidly degraded soon after IOP elevation. Furthermore, treatment with NSC23766 promoted to ATP release at G1w and G2w, respectively, which was reversed by the Cx43 inhibitors Gap26 or Gap19 (Figure 8c,d). The levels of ATP in the Gap26 and Gap19 groups were slightly reduced, but did not show significant difference from those in the NS group at G1w and G2w (Figure 8c,d).
Cx43 is expressed in astrocytes, Müller cells and microglia. To determine the source of the Rac1 inhibition-mediated ATP release, microglia were removed by a single dose intraperitoneal injection of clodronate liposomes (Clo-lip). In total, 66.1% ± 2.9% of the microglia were removed from the retinas after Clo-lip treatment ( Figure S4). However, removing retinal microglia did not affect the increased ATP level caused by Rac1 inhibition in COH retinas (Figure 8e), indicating that microglia were not the source of the Rac1 inhibition-mediated ATP release. To determine whether ATP may be released from astrocytes, we injected the AAV2-GfaABC1D-ATP1.0 into the retinas of C57BL/6 mice ( Figure S5). Then, 3 weeks after the AAV injections, the mice were treated with NSC23766 before the COH operations. GFP from the ATP probe was detected in retinas at G1w. Additionally, NSC23766 significantly increased the fluorescence signal, compared with controls (NS) (Figure 8f,g). These results indicate that astrocytes were the source of the increased extracellular ATP.
To further clarify the role of Rac1 in ATP release from astrocytes, we generated a Rac1 conditional knockout in astrocytes by injecting of GFAP-Cre-AAV in Rac1 fl/fl mice. The specificity of the knockout was confirmed in vertical sections and flat-mounted retinas using the RNAscope method (Figure 8h,i, and Figure S6a

| ATP release mediated by Rac1 inhibition promoted the survival of RGCs in COH retinas
Next, we investigated how ATP release from astrocytes, which was  (Figure 9i-k). These results suggest that it was via adenosine receptors that Rac1 inhibition-mediated ATP release promotes RGC survival.
F I G U R E 7 Rac1 regulates Cx43 expression through PAk1. (a-d) Representative immunoblots (a) and the densitometric quantification (b-d) showing the changes in Cx43, p-Cx43 Ser373 , p-Cx43 Ser368 levels in COH retinas at G1w with normal saline (NS) and NSC23766 (NSC) injections.

| Dynamic changes in the Cx43 hemichannels of astrocytes
Cx43 is expressed in astrocytes and is the most widely studied gapjunction forming protein in the central nervous system. Previous studies have shown that Cx43 is expressed in retinal Müller cells (Zahs et al., 2003) and microglia (Danesh Meyer et al., 2016). In this study, we demonstrated that Cx43 is found in NFL and GCL, which colocalized with astrocytes ( Figure 1). In COH retinas, expression of Cx43 exhibited biphasic downregulation (from G4d to G1w and then to G4w) (Figure 2). In fact, Cx43 expression pattern differ in various diseases. For example, increased expression of Cx43 was observed in cerebral hypoxic preconditioning, hippocampal seizures, transient retinal ischemia/perfusion, epilepsy, and in the drug-resistant cerebral cortex (Danesh-Meyer et al., 2012;Garbelli et al., 2011;Lin et al., 2008;Mylvaganam et al., 2010). Downregulation of Cx43 was also reported in heart failure (Ai & Pogwizd, 2005) and cerebral ischemia/reperfusion . The different expression patterns of Cx43 may be due to tissue specific expression of the protein or its involvement in various pathological processes. Therefore, the strategy of modulating Cx43 should be carefully selected according to the different time window during the pathological process of disease.
The mechanisms underlying the downregulation of Cx43 in glaucomatous retinas remains unclear. Changes of Cx43 expression may be associated with remodeling of astrocytes during IOP elevation.
First, in response to IOP elevation, astrocytes were activated and underwent morphological and functional changes (Figure 4). Gap43 is a marker of activity-dependent plasticity in astrocytes (Hung et al., 2016). In Gap43-positive astrocytes, expression of Cx43 was reduced (Figure 5a,d). Second, during the plastic changes that occurred in response to IOP elevation, astrocytes underwent structural remodeling and Cx43 was redistributed, which may have involved transient breakdown of Cx43 gap junctions (Yoshioka et al., 2005). Third, the degradation of Cx43 in glaucoma may be accelerated. In response to elevated pressure, epidermal growth factor (EGF) receptor was activated, leading to a decrease in gap junction intercellular communication via tyrosine phosphorylation of Cx43 in ONH astrocytes (Malone et al., 2007); this could increase the internalization and degradation of Cx43 in a proteasome-dependent manner (Leithe & Rivedal, 2004).
Phosphorylation of Cx43 is crucial for regulating the function of gap junctions, trafficking, and connexin assembly (Ribeiro Rodrigues et al., 2017). Protein phosphorylation is a rapid regulatory mechanism, and phosphorylation at a single site of connexin is sufficient to alter the function of gap junction channels . Phosphorylation of Cx43 at serine 368 induces internalization and a conformational change in Cx43, decreasing the relative frequency of the full open state and the hemichannel permeability (Bao et al., 2004;Leithe & Rivedal, 2004). By contrast, phosphorylation of Cx43 at serine 373 induces an acute increase in gap junction size (Dunn & Lampe, 2014), an enhancement of channel formation and Cx43 hemichannel opening (Batra et al., 2014;Yogo et al., 2006). In this study, we found that the p-Cx43 S368 /Cx43 ratio increased at an early stage, whereas the p-Cx43 S373 /Cx43 ratio increased at later stage both in total and membrane component of retinal proteins or in optic nerves in COH mice, suggesting that in response to IOP elevation, Cx43 hemichannels in astrocytes may be modulated, thus affecting its function. Opening of Cx43 hemichannels may exacerbate inflammation (Acosta et al., 2021). In our COH retinas, mRNA levels of pro-inflammatory factors had increased only G4d after IOP elevation (Figure 4k). The Rac1/PAK1-mediated changes in Cx43 phosphorylation may stimulate ATP release from astrocytes soon after IOP elevation. Interestingly, relative level of p-Cx43 Ser368 increased in COH retinas at G4d and G1w (Figure 2e), but extracellular ATP concentrations also increased ( Figure 8b). Perhaps Cx43 undergoes structural rearrangement and the relationship between Cx43 expression and ATP release from astrocytes is not linear. Additionally, treatment with Gap26 or Gap19 blocked Cx43 F I G U R E 8 Inhibiting or conditional knockout of Rac1 in astrocytes increases ATP release. (a,b) Changes in ATP levels determined by luciferinluciferase assays in retinas of Ctr and COH mice at different times after the operations without (a) or with (b) the ecto-ATPase inhibitor ARL67156. All data are normalized to Ctr. n = 5-7. *p = .0105; **p < .01; and ***p < .001 versus Ctr. (c,d) Changes in ATP levels in COH retinas at G1w (c) or G2w (d) with normal saline (NS), NSC23766 (NSC), NSC23766 + Gap26 (NSC + Gap26), NSC23766 + Gap19 (NSC + Gap19), Gap26 or Gap19 injections. All data are normalized to NS groups. n = 5-10. ***p < .001 versus NS; #p < .05 and ###p < .001 versus NSC. Oneway ANOVA with Tukey's multiple comparison test. (e) Changes in ATP levels in COH retinas with or without microglia deletion by clodronate liposomes (Clo-Lip) at G1w. n = 6. (f,g) Rac1 inhibition (NSC23766, NSC) increased ATP1.0 immunofluorescence in astrocytes. Representative images (f) and relative mean intensity of ATP 1.0 fluorescence (g) in COH retinas at G1w with NS or NSC injections; flat-mounted retinas were injected with AAV2-GfaABC1D-ATP1.0 in C57BL/6 mice. Scale bar: 20 μm. n = 6. ***p = .0002 versus NS. (h) Representative images showing immunostaining of GFAP and Rac1 transcript puncta by RNAscope in retinal vertical slices; subretinal injections of GFP-AAV (h1-h3) as a control or GFAP-Cre-AAV (h4-h6) in Rac1 flox/flox mice, highlight decreased expression of Rac1 transcript puncta in astrocytes in the GFAP-Cre-AAV group (h6 and h6'). The yellow squares show no changes in Rac1 expression in other cell types (including RGCs, and cells of the inner and outer nuclear layers). Scale bars: 20 μm. (i) Bar charts showing the number of Rac1 mRNA puncta colocalizing with GFAP in the NFL per length unit. n = 3 for each group. **p = .0053 versus GFP-AAV G1w. (j,k) Representative immunoblots (j) and densitometric quantification (k) showing changes in Cx43 in COH retinas with subretinal injections of GFP-AAV or GFAP-Cre-AAV in Rac1 flox/flox mice. All data are normalized to GFP-AAV G1w. n = 6 for each group. **p = .0025 versus GFP-AAV G1w. (l) Bar charts showing mean ATP levels in GFP-AAV G1w, and GFAP-Cre-AAV G1w retinas of Rac1 flox/flox mice. All data are normalized to the GFP-AAV group. n = 5 for each group. *p = .0297 versus GFP-AAV G1w. Two-tailed unpaired t test. channels but not significantly inhibit ATP release from COH retinas at G1w (Figure 8c,d). Cx43 hemichannels may be principally responsible for mediating ATP release, leading to inflammatory response. However, oxidative stress, excitatory neurotransmitters, and pannexin channels may also be involved (Acosta et al., 2021;Bennett et al., 2012). Inhibition of Rac1 may remodel astrocytes and maintain the syncytial body of astrocytes in a state of coordinated response, which provides mechanical stability and enables cells to resist pressure .

| Rac1 inhibition led to ATP release from astrocytes via Cx43 in COH retinas
Increasing evidence indicates that Rho family proteins are importance in gap junction formation (Adam et al., 2010;Matsuda et al., 2006;Ram et al., 2014). We show that Rac1 activation reduced Cx43 expression in the astrocytes of COH retinas. Inhibition or deletion of Rac1 in astrocytes stimulated the expression and function of Cx43 (Figures 6 and 7). Co-immunoprecipitation experiments demonstrated interactions between Cx43 and Rac1 or its downstream effector PAK1 (Figure 7). Interactions between PAK1 and Cx43 have been observed in cardiac myocytes, and activated PAK1 may dephosphorylate Rac1 as part of a negative feedback loop (Ai et al., 2011). Constitutively activated Rac1 may redistribute Cx43 (Ram et al., 2014). As a signal transduction hub, Rac1 may also mediate signals from N-cadherin that regulate the localization of Cx43 in aligned cardiac myocytes (Matsuda et al., 2006). The interaction between Cx43 and Rac1 is regulated by Mena, a member of the Ena/VASP family (Ram et al., 2014). Rac1 can be activated by cAMP/protein kinase A signaling (Kobayashi et al., 2013;Zhao et al., 2014), and elevated intracellular cAMP increased Cx43 phosphorylation (Darrow et al., 1995). cAMP immunoreactivity was increased in retinal astrocytes of DBA/2J spontaneous glaucoma mice, making these astrocytes more vulnerable to oxidative stress (Shim et al., 2018). Therefore, Rac1 may regulate Cx43 function both directly and indirectly in glaucoma.
Our results showed that extracellular ATP levels were increased in glaucomatous retinas ( Figure 8). This extracellular ATP may originate from retinal neurons or glial cells, because mechanical pressure may trigger ATP release from these cells (Beckel et al., 2014). In this study, we showed that Rac1 inhibition led to ATP release from astrocytes in COH retinas, as demonstrated by the location of ATP probe in astrocytes (Figure 8f,g). Partial depletion of microglia had no significant effect on ATP release, which excludes microglia-mediated ATP release in our model. The enhancement of ATP concentrations by inhibiting Rac1 in COH retinas was mainly from astrocytes, but the possibility that it also comes from Müller cells cannot be completely ruled out because of the location of Müller cells' end-feet in the NFL.
4.3 | Rac1/Cx43/ATP/A3R signaling mediated the crosstalk between astrocytes and RGCs in COH retinas ATP exerts its functions via two mechanisms: direct stimulation of purinergic receptors and indirect stimulation of adenosine receptors after being converted to adenosine by extracellular ectonucleotidases (ecto-ATPase) (Abbracchio et al., 2009). ATP can also induce retinal neuronal death by activating the P2X7 receptor (Zhang et al., 2005).
By contrast, adenosine may elicit neuroprotective effects by stimulating adenosine A1 and A3 receptors (Boia et al., 2020;Newman, 2003). Therefore, the net effect of increased extracellular ATP on RGC survival in COH retinas may depend on the relative availability of ATP and adenosine. In the present study, we showed that in the presence of the ecto-ATPase inhibitor ARL67156, IOP elevation induced an increase in extracellular ATP levels. Rac1 deletion in astrocytes further increased ATP release. However, ATP concentrations only increased immediately after IOP elevation (G4d and G1w) in the absence of ARL67156, suggesting ATP may be rapidly degraded to F I G U R E 9 Conditional knockout of Rac1 in astrocytes promotes RGC survival via adenosine receptors in COH retinas. (a) Representative micrographs of Brn3a-positive RGCs from normal (Ctr) and COH flat-mounted retinas at G1w with injections of GFP-AAV, GFAP-Cre-AAV, and GFAP-Cre-AAV + Gap26 in Rac1 flox/flox mice, respectively. Images were taken from the central (a1-a4) and peripheral (a5-a8) regions. Scale bar: 50 μm for all images. (b) Diagram showing eight chosen fields for evaluating RGC survival per retina. (c) Quantification of Brn3a-positive RGCs under different conditions as shown in (a). n = 6 retinas for each group. **p < .01 and ***p < .001 versus Ctr; #p = .0149 and ###p = .0007 versus GFP-AAV G1w; &p = .0193 and &&&p = .0007 versus GFAP-Cre-AAV G1w. (d) Double immunostaining showing the changes in A1R (d1-d3) and A3R (d4-d6) expression in Brn3a-positive RGCs in retinal vertical slices taken from Ctr and COH mice at G1w with normal saline (NS) and NSC23766 injections. Inhibition of Rac1 increased the expression of A3R in RGCs, but did not change the expression of A1R. Scale bars: 10 μm. (e,f) Relative mean intensity of A1R (e) and A3R (f) fluorescence signals in different groups as shown in (d). n = 5 retinas for each group. ***p < 0.001 versus Ctr; ###p < .001 versus NS G1w. One-way ANOVA with Tukey's multiple comparison test. (g) Triple immunofluorescence staining showing changes in A3R expression in Brn3a-positive RGCs in retinal vertical slices taken from COH retinas at G1w with injections of GFP-AAV (g1-g4) and GFAP-Cre-AAV (g5-g8) in Rac1 flox/flox mice. Rac1 conditional knockout in astrocytes increased expression of A3R in RGCs. Scale bars: 10 μm. (h) Relative mean intensity of A3R fluorescence signals in different groups as shown in (g). n = 4 retinas for each group. *p = .0208 versus GFP-AAV G1w. Two-tailed unpaired t test. (i) Typical micrographs of Brn3a-positive RGCs captured from COH flat-mounted retinas at G1w with injections of GFAP-Cre-AAV, GFAP-Cre-AAV + ARL67156 (an ecto-ATPase inhibitor) and GFAP-Cre-AAV + CGS15943 (an adenosine receptor antagonist) in Rac1 flox/flox mice. Scale bar: 50 μm. (j,k) Quantification of Brn3a-positive RGCs under different conditions as shown in (i). n = 5-6 retinas for each group. **p < .01 and ***p < .001 versus GFAP-Cre-AAV G1w. One-way ANOVA with Tukey's multiple comparison test.
adenosine. At the same time, A3R expression in RGCs increased.
Together, these responses may promote RGC survival in COH retinas, at least during the early stages of glaucoma. Inhibition or deletion of Rac1 in astrocytes increased the expression of Cx43 and ATP release; this may further increase the level of adenosine, providing a neuroprotective effect (Figure 9). Interestingly, progressive IOP elevation resulted in more ATP accumulation (Figure 8a), which may damage RGCs by stimulating the P2X7 receptor (Zhang et al., 2005).
Importantly, the effects of astrocytes on retinal neurons may also depend on the functional states of astrocytes. Acute stress changes the structure and function of astrocytes, inducing cell hypertrophy and reducing gap-junction coupling between cells. This may lead to detachment of astrocyte from the network, reducing their ability to nourish and support neurons (Boal et al., 2021). Deletion of Rac1 increased the expression of Cx43, remodeled astrocytes, and reduced astrocyte reactivity. Furthermore, the Cx43-mediated astrocyte metabolic network as an endogenous mechanism can alleviate the bioenergy stress in the stress response (Cooper et al., 2020). Because the Cx43 expression was suddenly downregulated at G4d and G1w, which was matched to the time course of astrocyte activation, we targeted the G1w time point to inhibit Rac1 and restore Cx43 expression. The neuroprotective effects of Rac1 inhibition persisted for at least 2 weeks, as evidenced by our supplementary results in Figure S8.
In this study, we provide robust evidence that Cx43 is primarily expressed in retinal astrocytes. IOP elevation in a glaucoma mouse model activated astrocytes. The activated astrocytes decreased Cx43 expression and altered phosphorylation of Cx43. We also showed, for the first time, that Rac1 deletion in astrocytes stimulated ATP release.
This ATP release was mediated by Cx43 hemichannels and promotes RGC survival via adenosine receptors during the early stages of glaucoma ( Figure 10). These results, together with our previous study , suggest that modulating Rac1/Cx43 signaling may have neuroprotective effects in patients with glaucoma.

CONFLICT OF INTEREST STATEMENT
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The data supporting the findings of this study are available within this article and its supplemental information files. All other relevant data are available from the corresponding authors upon reasonable request.