Approximately 40% of patients with diabetic macular edema (DME) are resistant to anti–vascular endothelial growth factor (VEGF) therapy (rDME). Here, we demonstrate that significant correlations between inflammatory cytokines and VEGF, as observed in naive DME, are lost in patients with rDME. VEGF overexpression in the mouse retina caused delayed inflammatory cytokine upregulation, monocyte/macrophage infiltration (CD11b+ Ly6C+ CCR2+ cells), macrophage/microglia activation (CD11b+ CD80+ cells), and blood-retinal barrier disruption due to claudin-5 redistribution, which did not recover with VEGF blockade alone. Phosphorylated protein analysis of VEGF-overexpressed retinas revealed rho-associated coiled-coil–containing protein kinase (ROCK) activation. Administration of ripasudil, a selective ROCK inhibitor, attenuated retinal inflammation and claudin-5 redistribution. Ripasudil also contributed to the stability of claudin-5 expression by both transcriptional enhancement and degradation suppression in inflammatory cytokine–stimulated endothelium. Notably, the anti-VEGF agent and the ROCK inhibitor were synergic in suppressing cytokine upregulation, monocyte/macrophage infiltration, macrophage/microglia activation, and claudin-5 redistribution. Furthermore, in vitro analysis confirmed that claudin-5 redistribution depends on ROCK2 but not on ROCK1. This synergistic effect was also confirmed in human rDME cases. Our results suggest that ROCK-mediated claudin-5 redistribution by inflammation is a key mechanism in the anti-VEGF resistance of DME.
Introduction
The global prevalence of diabetes is increasing, and diabetic macular edema (DME) is a major cause of vision loss in patients with diabetes (1). As a direct result of progressive ischemia and inflammation, vascular endothelial growth factor (VEGF) becomes upregulated in the diabetic retina and drives DME-linked pathophysiology (2,3). Regular intravitreal injection of anti-VEGF agents (anti-VEGF therapy) can improve vision and reduce macular fluid; this therapy is now the primary treatment modality for DME (4). However, clinical trials have shown that ∼40% of patients are nonresponsive to anti-VEGFs (5,6). The molecular mechanisms underpinning anti-VEGF therapy–resistant DME (rDME) are unknown, and their clarification could drive the discovery of new drugs.
The concept of “anti-VEGF therapy resistance” was originally introduced in cancer research (7). The tumor microenvironment is rich in various cytokines in addition to VEGF (8), which could contribute significantly to anti-VEGF therapy resistance (7,9). Inflammation is closely associated with DME pathogenesis, with multiple cytokines being upregulated in the vitreous fluid of DME patients (10,11). Although tumor necrosis factor α (TNFα) is one such cytokine upregulated in DME, an inhibitor to this cytokine did not suppress edema formation (12), suggesting that multiple cytokines and related molecules are linked to rDME pathogenesis—similarly to what is observed in some neoplasms.
DME is caused by vascular leakage arising from blood-retinal barrier (BRB) disruption (13). The human central nervous system, including the retina, is separated from the bloodstream by tight junctions (TJs) between vascular endothelial cells (13,14). The integrity of TJs and of the closely associated adherens junctions (AJs) depends on complex protein-protein interactions. For example, claudin-5 is an essential TJ component protein (14–16) and its assembly is dependent on AJ–associated vascular endothelial (VE)-cadherin (17). In turn, claudin-5 becomes anchored to the actin cytoskeleton via zonula occludens protein-1 (ZO-1), resulting in its stable expression on cell membranes (14). We and others have demonstrated that claudin-5 redistribution occurs in the diabetic retina together with vasopermeability (18–21). Protection of claudin-5 integrity in the TJs may be an attractive therapeutic option to prevent BRB dysfunction.
Rho-associated coiled-coil–containing protein kinase (ROCK), in its two isoforms (ROCK1 and ROCK2), is involved in cytoskeletal reorganization, contractility, and inflammation (22). Recent studies have shown ROCK to be a candidate molecular target in cardiovascular diseases, diabetes, and eye disease (23–25). ROCK also contributes to inflammation, neovascularization, fibrovascular membrane formation, and retinal pigment epithelium (RPE) dysfunction in diabetic retinopathy (DR) (26–29). However, whether ROCK is involved in BRB integrity and diabetes-associated vasopermeability is unknown.
Here, we reveal that VEGF-independent upregulation of proinflammatory cytokines occurs in the vitreous of rDME patients. We also show that ROCK-mediated claudin-5 redistribution by these cytokines causes BRB dysfunction and contributes to anti-VEGF therapy resistance.
Research Design and Methods
Patient Characterization and Anti-VEGF Treatment
This study was conducted according to the principles of the Declaration of Helsinki. After approval from the Institutional Review Board of Kyushu University Hospital, informed consent regarding the use of vitreous fluid was obtained from each patient with macular hole (MH) or DME. Vitreous samples were collected during vitrectomy. DME was defined as center-involving macular edema secondary to type 1 or 2 diabetes with a central macular thickness (CMT) ≥300 µm.
Retinal thickness was measured by spectral domain optical coherence tomography (OCT), Biμ (Kowa Company, Ltd.), or Cirrus HD-OCT (Carl Zeiss). Untreated DME samples (naive DME [nDME]) were taken from 2008 to 2010, i.e., before anti-VEGF treatment was approved for DME. All of the DME patients after 2015 received three initial consecutive monthly intravitreal injections of 0.5 mg ranibizumab (IVR) or 2 mg aflibercept (IVA) followed by monthly examinations. They underwent pro re nata retreatment when DME recurred (i.e., when the CMT was ≥300 µm). All rDME samples we examined were taken after the VEGF suppression time to avoid the effect of anti-VEGF therapy itself (>34 days for IVR and >67 days for IVA) (30).
Intravitreal Injection of Fasudil
After approval from the Ethics Committee of the Ophthalmic Research Center, Shahid Beheshti University of Medical Sciences, we administered the combination therapy of an intravitreal injection of bevacizumab (IVB) (1.25 mg) in addition to an intravitreal injection of fasudil (IVF) (0.125 mg; Asahi Kasei Pharma) to rDME patients. We excluded patients with active proliferative DR or other macular disorders, IVB within the past 3 months, or ocular surgery within the past 6 months.
Animals
All procedures were reviewed by the Committee on Ethics in Animal Experiments of Kyushu University and were carried out according to the Guidelines for Animal Experiments issued by Kyushu University and those issued by the Japanese government. C57B6 wild-type (WT) mice and trVEGF029 (Kimba) mice (C57B6 background), 14–17 days old, were injected with 25 μg bevacizumab (Chugai Pharmaceutical) or human IgG1 (Bingo Biotech) into the vitreous cavity. An ophthalmic solution of 0.8% ripasudil (Kowa Company, Ltd.) was administered twice a day following the intravitreal injections.
Diabetic Mice
Mouse body weight and blood glucose (BS) were measured using an Accu-Chek Mobile (Roche) after a 4-h fast. Seven-week-old B57B6 mice were injected once intraperitoneally with streptozotocin (STZ) (Wako) (150 mg/kg) dissolved in 0.05 mol/L citrate buffer (pH 4.5) to induce a diabetes condition. Control mice were injected with the same volume of citrate buffer. Four days after STZ injection, body weight and BS were measured again after a 4-h fast. Mice were considered diabetic when BS exceeded 300 mg/dL. From day 5 to day 7, ripasudil was administered to control or diabetic mice three times a day. All experiments were performed on mice 7 days after STZ injection.
Real-time Quantitative PCR
WT and Kimba mice were anesthetized by intraperitoneal injection of 15 mg/kg ketamine and 7 mg/kg xylazine. Total RNA of retinal tissues or cultured cells was extracted using a NucleoSpin RNA Kit (Macherey-Nagel), and total RNA was reverse transcribed using a First Strand cDNA Synthesis Kit for RT-PCR (Roche). The real-time quantitative PCR (qPCR) was performed with SYBR Premix Ex Taq (Takara Bio) using a LightCycler 96 (Roche). Primer sequences are shown in Supplementary Table 1.
Measurement of Cytokine Concentrations in the Mouse Retina
After isolation, the neural retinas were homogenized with PBS containing 1% Triton X-100, 0.2% SDS, and Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). After centrifugation, the supernatants were collected. TNFα, IL-6, IL-8, and MCP-1 concentrations in the supernatant were determined using the Cytometric Bead Array.
Measurement of Retinal Thickness
We recorded retinal images using the Envisu R-Class (Leica) spectral domain OCT system. Retinal images 1 mm in diameter were taken around the optic nerve. We drew a vertical line from the RPE, and the retinal thickness was defined as the distance between the inner limiting membrane and the RPE. The maximum retinal thickness value in each mouse was used to compare the effect of treatments.
Visualization of BRB Integrity
PBS containing 100 µg/mL Hoechst stain H33258 (molecular weight [MW] 534 Da; Sigma-Aldrich) and 50 µg/mL fluorescence-labeled dextran (MW 4,000 Da; Thermo Fisher Scientific) was injected into the left ventricle of each mouse, while anesthetized, as previously described (18). The enucleated eyes were immediately fixed in 4% paraformaldehyde (PFA), and retinal flat mounts were observed using a BZ-X700 confocal microscope (Keyence). For quantification of vascular leakage, the number of Hoechst-positive cells in a 200-μm square was measured.
Immunohistochemistry of the Mouse Retina
Retinal tissues from the enucleated eyes were allowed to react with rabbit polyclonal antibody (pAb) against mouse claudin-5 (Thermo Fisher Scientific) or rabbit pAb against mouse CXCR1 (Abcam). Alexa Fluor 488 or 594 goat anti-rabbit IgG (Thermo Fisher Scientific) was used as the secondary antibody. Finally, cryosections and retinal flat mounts were prepared and observed using the BZ-X700 confocal microscope.
Mass Spectrometry Analysis of Phosphorylated Proteins Collected From the Mouse Retina
Phosphorylated proteins were extracted from WT and Kimba mouse neural retina using a Phosphoprotein Enrichment Kit (Clontech). Proteins were separated by SDS-PAGE and visualized with silver staining. The lanes were excised from the gel and subjected to mass spectrometry to identify the corresponding proteins.
The peptides were analyzed by liquid chromatography–tandem mass spectrometry using an Orbitrap Velos Pro system (Thermo Fisher Scientific), and the data were analyzed using Mascot software (Matrix Science). The U.S. National Center for Biotechnology Information nonredundant database was used as the reference.
Flow Cytometry of the Mouse Retina
Retinal tissues were digested in Hanks’ balanced salt solution (Sigma-Aldrich) containing collagenase D (Roche) and DNase (Roche). For the examination of monocytes, macrophages, and microglia, cells were stained with FITC-conjugated monoclonal antibody (mAb) against mouse CD80 (BioLegend), allophycocyanin (APC)-conjugated anti-mouse CCR2 mAb (BioLegend), Per-CP–conjugated anti-mouse Ly6G mAb (BioLegend), APC-Cy7–conjugated anti-mouse Ly6C mAb (BioLegend), and BV421-conjugated anti-mouse CD11b mAb (BioLegend).
For evaluation of the proportion of lymphocytes, cells were stained with FITC-conjugated anti-mouse CD3 mAb (BioLegend), phycoerythrin-Cy7–conjugated anti-mouse CD4 mAb (BioLegend), APC-conjugated anti-mouse CD19 mAb (eBioscience), and APC-Cy7–conjugated anti-mouse Ly6G mAb (BioLegend). Dead cells were stained with 7-AAD Viability Staining Solution (BioLegend). A FACSVerse flow cytometer and FlowJo software (BD Biosciences) were used for the analyses.
Cell Culture
A mouse brain microvascular endothelial cell line, bEND.3 (ATCC), and a human primary retinal microvascular endothelial cell line, HRMEC (Cell Systems), were purchased. bEND.3 cells were grown in DMEM (Sigma-Aldrich) supplemented with 10% FBS, and HRMEC cells in Complete Classic Medium (Cell Systems) with 10% FBS and CultureBoost (Cell Systems). The bEND.3 and HRMEC cells were used 7 days and 1 day after reaching confluency, respectively.
After 1 h of preincubation with ripasudil (final concentration 3 or 30 μmol/L) (Kowa Company, Ltd.), human IgG1, bevacizumab (final concentration 300 μg/mL), or dexamethasone (final concentration 1 μmol/L; Sigma-Aldrich), the cells were stimulated with mouse or human recombinant VEGF (final concentration 25 ng/mL; PeproTech), mouse or human recombinant TNFα (final concentration 25 ng/mL for bEND.3, 10 ng/mL for HRMEC; PeproTech), mouse or human recombinant IL-6 (final concentration 50 ng/mL; PeproTech), human recombinant IL-8 (final concentration 50 ng/mL; PeproTech), mouse or human recombinant MCP-1 (final concentration 25 ng/mL; PeproTech), or human vitreous fluid for 24 h.
ROCK1 or ROCK2 Knockout in bEND.3 Cells
Single-guide RNAs were designed using an online tool provided by Dr. Feng Zhang’s Laboratory at the Eli and Edythe L. Broad Institute of MIT and Harvard (https://crispr.mit.edu/). The pSpCas9(BB)-2A-Puro (PX459) V2.0 plasmid was a kind gift from Dr. Zhang (Addgene). Plasmids with single-guide RNA sequences were transfected into bEND.3 cells using Polyethylenimine HCl MAX (Polysciences). Individual clones were isolated by limiting dilution. Genotyping was confirmed by direct sequencing.
Immunocytochemistry of Cultured Cells
For the single staining of claudin-5 or the double staining of claudin-5 and ZO-1, cultured cells were fixed with 100% methanol and reacted with rabbit anti-mouse claudin-5 pAb. Alexa Fluor 488 goat anti-rabbit IgG was used alone for claudin-5 simple staining and together with rabbit anti-mouse ZO-1 mAb conjugated with Alexa Fluor 594 (Thermo Fisher Scientific) for claudin-5/ZO-1 double staining.
For ZO-1 and F-actin double staining, cells were fixed with 4% PFA and incubated with rabbit anti-mouse ZO-1 mAb conjugated with Alexa Fluor 488, Alexa Fluor 594 and Alexa Fluor 488, or Alexa Fluor 594 Phalloidin (Thermo Fisher Scientific). Cells were observed using the BZ-X700 confocal microscope.
Transendothelial Electrical Resistance
We measured the transendothelial electrical resistance (TEER) of a bEND.3 monolayer at the confluent state on collagen-coated 0.9-cm2 inserts of 0.4-μm pore size (BD Falcon) using the Millicell ERS-2 Voltohmmeter (Millipore). TEER values were calculated by subtracting the resistance of blank inserts without cells and multiplying the subtracted values by the surface area of the inserts.
Western Blotting
Cells were lysed in PBS containing 1% Triton X-100, 0.2% SDS, and Halt Protease and Phosphatase Inhibitor Cocktail. After centrifugation, the supernatants were collected. Samples were separated by SDS-PAGE and transferred to PVDF membranes (Millipore).
The membranes were reacted with rabbit anti-mouse claudin-5 pAb, rabbit anti-mouse ZO-1 pAb (Thermo Fisher Scientific), or rabbit anti-human α-tubulin mAb (Cell Signaling Technology) and then incubated with horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG (Dako). The HRP signal was visualized with an enhanced chemiluminescence system (SuperSignal West Pico or Femto Chemiluminescent Substrate; Thermo Fisher Scientific). Signal intensities were measured using the ImageJ software (National Institutes of Health). All values were normalized to α-tubulin.
Biotinylation of Cell Surface Molecules
Cell surface molecules of bEND.3 cells were biotinylated by incubation with PBS containing EZ-Link Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific). The cell lysates were centrifuged, and the supernatants were incubated with MagnaBind Streptavidin Beads (Thermo Fisher Scientific). The beads were boiled with Laemmli sample buffer, and the supernatant was loaded on polyacrylamide gels to be processed for Western blot analysis.
Immunoprecipitation
Each sample was reacted with anti-mouse ZO-1 pAb, and immune complexes were collected using Protein G Sepharose 4 Fast Flow (GE Healthcare). The sepharose beads were boiled with Laemmli sample buffer, and the supernatant was used for Western blot analysis. HRP-conjugated rat anti-rabbit IgG TrueBlot (Rockland Immunochemicals) was used as the secondary antibody.
Isolation of Soluble and Insoluble Fractions
Cultured medium of bEND.3 monolayers was changed to ice-cold PBS and kept at 4°C for 5 min to block endocytosis. Then, the solution was replaced with 300 μL ice-cold PBS containing 1% Triton X-100–supplemented protease inhibitor cocktails. After the incubation at 4°C for 30 min under rotation at 70 rpm, the solution was collected as the soluble fraction. As the insoluble fraction, the remaining cellular components were dissolved in 100 μL of the above-mentioned lysis buffer containing cocktail inhibitors.
Chromatin Immunoprecipitation
First, bEND.3 cells were cross-linked with 1% PFA. PFA was inactivated with 150 mmol/L glycine. The cells were then washed with PBS containing 2% FBS and lysed with distilled water containing 50 mmol/L Tris-HCl (pH 8.0), 10 mmol/L EDTA, 1% SDS, and Halt Protease and Phosphatase Inhibitor Cocktail. The cells were then sonicated with a VCX130 (Sonics & Materials). After centrifugation, the supernatants were collected and diluted with distilled water containing 50 mmol/L Tris-HCl (pH 8.0), 150 mmol/L NaCl, and 1% Triton X-100.
Next, this diluted solution was incubated with Protein G Sepharose 4 Fast Flow and rabbit IgG (Dako) or rabbit anti-human β-catenin mAb. After centrifugation, the beads were collected and washed once with distilled water containing 50 mmol/L Tris-HCl (pH 8.0), 1 mmol/L EDTA, 1% Triton X-100 0.1% SDS, and 150 mmol/L NaCl and then washed twice with distilled water containing 50 mmol/L Tris-HCl (pH 8.0), 1 mmol/L EDTA, 1% Triton X-100 0.1% SDS, and 300 mmol/L NaCl.
The beads were suspended in distilled water containing 10 mmol/L Tris-HCl (pH 8.0), 5 mmol/L EDTA, 300 mmol/L NaCl, and 0.5% Triton X-100; heated at 65°C for 2 h; and then incubated at 37°C for 30 min after administration of Proteinase K (Wako). DNA fragments were purified with a FastGene Gel/PCR Extraction Kit (Nippon Genetics). Real-time qPCR was performed to compare the binding amounts of β-catenin. Primer sequences are shown in Supplementary Table 1.
Statistical Analysis
Data are expressed as the mean ± SD or SEM. Statistical analysis was performed with JMP Pro, version 13, software (SAS Institute). The significance of the differences in the values obtained from Western blot analysis, real-time qPCR, cytometric bead array, FACS, OCT, and TEER was determined by Student t tests.
The differences in patient age and sex were examined using ANOVA and χ2 tests, respectively. Significant differences in the concentrations of VEGF and cytokine expression, and in the values of HbA1c and CMT, were determined by the Wilcoxon rank sum test. The correlation between the concentrations of VEGF and cytokines in the vitreous fluid were determined by Spearman correlation coefficient. Differences were considered statistically significant at P < 0.05.
Data and Resource Availability
The data sets of mass spectrometry generated during the current study are not publicly available because we are in the middle of the pathway analysis. However, the data sets are available from the corresponding author upon reasonable request. No applicable resources were generated or analyzed during the current study.
Results
Vitreous Cytokine Profiles Implicate VEGF-Independent Inflammation in the Pathophysiology of rDME
We examined TNFα, IL-6, IL-8, and MCP-1 (representative inflammatory cytokines that have been linked to vasopermeability) and VEGF levels (13,31) in vitreous fluids from MH and DME patients with no treatment (nDME), and from rDME patients (Supplementary Table 2). Vitreous VEGF, IL-6, IL-8, and MCP-1 levels in the nDME group were significantly increased compared with those of the MH group (Fig. 1A), as shown in our previous report (32). Although vitreous TNFα, IL-6, and MCP-1 levels in rDME were also significantly higher compared with the MH group, VEGF and IL-8 expression was similar between the rDME and MH groups. IL-8 was lower in rDME patients compared with nDME patients (Fig. 1A). Furthermore, VEGF levels were significantly correlated with IL-6, IL-8, and MCP-1 levels in the nDME group, but not in the rDME group (Fig. 1B), indicating differential expression of proinflammatory cytokines between nDME and rDME.
Measurement of VEGF and cytokines in vitreous fluid reveals differences in the characteristics of retinal inflammation between nDME and rDME. A: VEGF and cytokine expression in vitreous fluid in MH (n = 13), nDME (n = 18), and rDME (n = 13). All graphs represent the mean ± SD. *P < 0.01, †P < 0.05, Wilcoxon rank sum test. B: Correlation between VEGF levels and the level of cytokines in nDME and rDME. Spearman rank correlation coefficient was used.
Measurement of VEGF and cytokines in vitreous fluid reveals differences in the characteristics of retinal inflammation between nDME and rDME. A: VEGF and cytokine expression in vitreous fluid in MH (n = 13), nDME (n = 18), and rDME (n = 13). All graphs represent the mean ± SD. *P < 0.01, †P < 0.05, Wilcoxon rank sum test. B: Correlation between VEGF levels and the level of cytokines in nDME and rDME. Spearman rank correlation coefficient was used.
A clinical trial revealed that more than one-half of DME diminished only with continuous anti-VEGF therapy (5,6), implying that VEGF-induced retinal inflammation and VEGF itself might be the main causal factors in anti-VEGF–responsive DME. Since VEGF and inflammatory cytokines in rDME patients were not correlated in the current study (Fig. 1B), we hypothesized that VEGF-independent inflammation in rDME contributed to anti-VEGF therapy resistance.
VEGF Overexpression Induces Delayed Inflammatory Cytokine Upregulation in the Mouse Retina
The differential cytokine profiles between nDME and rDME patients motivated us to investigate VEGF effects on inflammatory cytokines. To this end, we used Kimba mice, in which human VEGF is overexpressed in the retina (33), and checked the time course of TNFα, IL-6, MCP-1, and KC (the murine homolog of IL-8) mRNA levels and human VEGF in the mice. Human VEGF peaked at 10 days of age, while inflammatory cytokines peaked at 3 weeks of age (Fig. 2A), suggesting that high expression of this growth factor could induce inflammatory cytokine upregulation, with a time delay. The number of CX3CR1+ cells also markedly increased in the retina of 3-week-old Kimba mice, suggesting activation of retinal macrophages/microglia (Supplementary Fig. 1).
VEGF-induced retinal inflammation promotes retinal edema resistant to anti-VEGF therapy. A: Real-time PCR results for the mRNA expression of VEGF (human), TNFα, IL-6, MCP-1, and KC in the retina of WT and Kimba mice. 10d, 10 days old; 3w, 3 weeks old; 8w, 8 weeks old. n = 5. B: OCT images of mouse retinas (3-week-old WT and Kimba mouse, top) and human retinas (healthy control subject and DME patient, bottom). C and D: Comparison of the therapeutic effects of IVB and Rip-OS on retinal edema. OCT images (C) and their corresponding quantitative analyses for retinal thickness (D). n = 6. E and F: Comparison of the therapeutic effects of IVB and Rip-OS on BRB breakdown. E: BRB disruption was visualized with Hoechst leakage (blue), and the number of Hoechst-positive cells was measured for the quantification of leakage (F). n = 4. Scale bar = 50 μm. G: Therapeutic effects of IVB and Rip-OS on claudin-5 expression in retinal vessels. n = 3. Scale bar = 50 μm. H: Silver staining of phosphorylated proteins expressed in the mouse retina (upper left). Right: expression levels of phosphorylated proteins in the retina of WT and Kimba mice. Lower left: confirmation of the mass spectrometry results by immunoblotting of phosphorylated moesin. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Students t test. emPAI, exponentially modified protein abundance index.
VEGF-induced retinal inflammation promotes retinal edema resistant to anti-VEGF therapy. A: Real-time PCR results for the mRNA expression of VEGF (human), TNFα, IL-6, MCP-1, and KC in the retina of WT and Kimba mice. 10d, 10 days old; 3w, 3 weeks old; 8w, 8 weeks old. n = 5. B: OCT images of mouse retinas (3-week-old WT and Kimba mouse, top) and human retinas (healthy control subject and DME patient, bottom). C and D: Comparison of the therapeutic effects of IVB and Rip-OS on retinal edema. OCT images (C) and their corresponding quantitative analyses for retinal thickness (D). n = 6. E and F: Comparison of the therapeutic effects of IVB and Rip-OS on BRB breakdown. E: BRB disruption was visualized with Hoechst leakage (blue), and the number of Hoechst-positive cells was measured for the quantification of leakage (F). n = 4. Scale bar = 50 μm. G: Therapeutic effects of IVB and Rip-OS on claudin-5 expression in retinal vessels. n = 3. Scale bar = 50 μm. H: Silver staining of phosphorylated proteins expressed in the mouse retina (upper left). Right: expression levels of phosphorylated proteins in the retina of WT and Kimba mice. Lower left: confirmation of the mass spectrometry results by immunoblotting of phosphorylated moesin. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Students t test. emPAI, exponentially modified protein abundance index.
Moreover, OCT imaging of Kimba mice retinas showed significantly increased retinal thickness (Fig. 2B–D). Our vascular perfusion assay with FITC-dextran (MW 4,000 Da) and Hoechst 33258 (MW 534 Da) confirmed Hoechst leakage (i.e., vascular hyperpermeability) in Kimba but not WT mice (Fig. 2E). We also observed the disappearance of claudin-5 in the retinal endothelial cells of Kimba mice (Fig. 2F).
We investigated the therapeutic effects of anti-VEGF therapy on the retinal thickness of the mice after IVB (Supplementary Fig. 2). The retinal thickness of the Kimba mice was significantly reduced with either IVB × 1 or IVB × 2, but there was no difference between the two (Fig. 2C and D). Moreover, vascular barrier dysfunction and claudin-5 disappearance only partially recovered, even with two IVB administrations (Fig. 2E–G), suggesting that anti-VEGF therapy alone was insufficient.
We then sought to identify a molecule activated downstream of various inflammatory cytokine signals by analyzing phosphorylated protein expression in the retinas of WT and Kimba mice by electrophoresis and mass spectrometry (Fig. 2H). Various proteins involved in cytoskeleton reorganization and endocytosis were phosphorylated (Fig. 2H, right), although there was a marked increase in ROCK substrates: moesin, radixin, and vimentin (22) (Fig. 2H, right and lower left).
We therefore examined the impact of a selective and potent ROCK inhibitor, ripasudil, on hyperpermeability and retinal edema in the VEGF-overexpressing retina. Although retinal thickness in the Kimba mice was significantly decreased even with 0.8% ripasudil ophthalmic solution (Rip-OS) alone (34), vascular hyperpermeability and claudin-5 disappearance only partially recovered, as with IVB (Fig. 2C–G). Notably, IVB and Rip-OS had a synergistic effect, and retinal thickness decreased the most with IVB × 2 and Rip-OS in combination. In addition, vascular hyperpermeability and claudin-5 expression recovered to WT levels (Fig. 2C–G). IVB × 1 and Rip-OS in combination produced effects similar to those of IVB or Rip-OS monotherapy, suggesting that continuous VEGF suppression is necessary for ripasudil’s therapeutic effect (Fig. 2C–G).
Ripasudil Regulates Retinal Inflammation in Kimba Mice by Suppressing Granulocyte Infiltration and Activation
We then investigated whether ripasudil and bevacizumab had different mechanisms of action. The effects of IVB and Rip-OS on retinal cytokine expression levels were confirmed (Supplementary Fig. 3), and all four cytokines examined (TNFα, IL-6, MCP-1, and KC) were upregulated in Kimba mice compared with WT mice, where they were undetectable. IVB × 1 treatment did not suppress cytokine expression, but IVB × 2 treatment suppressed TNFα and IL-6 expression. Unlike IVB, Rip-OS suppressed TNFα, MCP-1, and KC expression. IVB × 1 + Rip-OS and IVB × 2 + Rip-OS were similar to IVB or Rip-OS monotherapy.
We next investigated IVB and Rip-OS effects on the retina leukocyte profile using flow cytometry (Fig. 3). CD11b+ immune cells in the retina were significantly increased in Kimba mice compared with WT mice (Fig. 2A and B). IVB × 2 + Rip-OS combination therapy, but not IVB or Rip-OS monotherapy, decreased their proportion. IVB × 2 + Rip-OS combination therapy, but not IVB × 2 or Rip-OS monotherapy (Fig. 3C and D), decreased the proportion of inflammatory monocytes (CD11b+ Ly6C+ CCR2+ cells).
Ripasudil suppresses retinal inflammation by inhibiting leukocyte infiltration and activation. A and B: FACS plots of CD11b+ cells in the retina (A) and the corresponding quantitative results (B). C and D: FACS plots of CD11b+ Ly6C+ CCR2+ cells in the retina (C) and the corresponding quantitative results (D). E and F: FACS plots of CD11b+ CD80+ cells in the retina (E) and the corresponding quantitative results (F). G and H: FACS plots of Ly6G+ cells in the retina (G) and the corresponding quantitative results (H). I: Quantitative analysis of CD19+ cells in the retina. J: Quantitative analysis of CD3+ cells in the retina. n = 4 WT mice; n = 8 Kimba mice. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
Ripasudil suppresses retinal inflammation by inhibiting leukocyte infiltration and activation. A and B: FACS plots of CD11b+ cells in the retina (A) and the corresponding quantitative results (B). C and D: FACS plots of CD11b+ Ly6C+ CCR2+ cells in the retina (C) and the corresponding quantitative results (D). E and F: FACS plots of CD11b+ CD80+ cells in the retina (E) and the corresponding quantitative results (F). G and H: FACS plots of Ly6G+ cells in the retina (G) and the corresponding quantitative results (H). I: Quantitative analysis of CD19+ cells in the retina. J: Quantitative analysis of CD3+ cells in the retina. n = 4 WT mice; n = 8 Kimba mice. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
We also investigated whether ripasudil inhibits macrophage activation in the Kimba mouse retina. The proportion of CD11b+ CD80+ cells was increased in the Kimba compared with WT mice (Fig. 3E and F). Unlike IVB, Rip-OS monotherapy could inhibit macrophage activation. However, combined therapy with IVB × 2 + Rip-OS enhanced the reduction of CD11b+ CD80+ cells. Macrophage activation was also significantly suppressed by intravitreal IgG injection compared with noninjected controls, suggesting that IVB did not enhance the therapeutic effect of ripasudil. Neutrophil proportion (Ly6G+ cells) was also significantly increased in the Kimba mice and significantly decreased by IVB monotherapy. Rip-OS did not influence neutrophil proportion, and no synergistic effect was observed with combination therapy (Fig. 3G and H). B-cell (CD19+) proportion was also significantly increased in Kimba mice and unaffected by IVB or Rip-OS (Fig. 3I). T-cell (CD3+) proportion was unchanged in Kimba mice and unaffected by IVB or Rip-OS (Fig. 3J). These results indicate that ripasudil inhibits retinal inflammation, with ripasudil and bevacizumab having synergistic effects.
Ripasudil Blocks the Inflammation-Induced Breakdown of the Retinal-Vascular Barrier by Stabilizing Claudin-5 Expression
While ripasudil suppressed some key retinal inflammation parameters (Fig. 3 and Supplementary Fig. 3), it could not account for the synergistic effect of IVB and Rip-OS in recovering retinal edema, vascular hyperpermeability, or claudin-5 dislocation (Fig. 2C–F), as the cytokine expression levels promoting the breakdown of the vascular barrier were similar between IVB and Rip-OS monotherapy and their combination therapy (Fig. 2). We therefore investigated the potential of ripasudil to prevent “secondary” vascular barrier dysfunction.
In microvascular endothelial cells, ripasudil and bevacizumab prevented TEER reduction induced by TNFα, IL-6, or MCP-1, and ripasudil was more effective than bevacizumab (Fig. 4A). Claudin-5 is anchored to the actin cytoskeleton via ZO-1, and this structure is crucial for stable claudin-5 expression in the cell membrane (Fig. 4B). Immunostaining confirmed that ripasudil treatment both restored claudin-5 expression and maintained ZO-1 expression on the endothelial cell membranes (Fig. 4C). In addition, ripasudil, but not bevacizumab, decreased cytokine-induced F-actin polymerization (Fig. 4C and D).
Ripasudil blocks cytokine-induced barrier dysfunction by inhibiting claudin-5 and ZO-1 redistribution and actin cytoskeleton reorganization. A: The effects of ripasudil and bevacizumab on TEER values under cytokine-stimulated conditions. n = 9. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. B: Schematic representation of the junctional complex of TJs and AJs in retinal endothelial cells. C: The effects of ripasudil and bevacizumab on claudin-5 (green) and ZO-1 (red) distribution under cytokine-stimulated conditions. n = 3. Scale bar = 10 μm. D: The effects of ripasudil and bevacizumab on the actin filament network (green) and ZO-1 distribution (red) under cytokine-stimulated conditions. n = 3. Scale bar = 10 μm. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
Ripasudil blocks cytokine-induced barrier dysfunction by inhibiting claudin-5 and ZO-1 redistribution and actin cytoskeleton reorganization. A: The effects of ripasudil and bevacizumab on TEER values under cytokine-stimulated conditions. n = 9. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. B: Schematic representation of the junctional complex of TJs and AJs in retinal endothelial cells. C: The effects of ripasudil and bevacizumab on claudin-5 (green) and ZO-1 (red) distribution under cytokine-stimulated conditions. n = 3. Scale bar = 10 μm. D: The effects of ripasudil and bevacizumab on the actin filament network (green) and ZO-1 distribution (red) under cytokine-stimulated conditions. n = 3. Scale bar = 10 μm. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
Western blot analysis after biotinylation of cell membrane–localized molecules demonstrated that ripasudil maintained claudin-5 expression after TNFα, IL-6, or MCP-1 stimulation (Fig. 5A and B). As TJs have a lipid-rich structure and are not soluble in Triton X-100 at 4°C, we examined ZO-1 expression in the soluble (S) and insoluble fraction (I) by Western blotting, and calculated the I/S ratio (Fig. 5C and D). Ripasudil did not change the ZO-1 distribution in the control condition but significantly increased the ZO-1 I/S ratio under TNFα, IL-6, and MCP-1 stimulation (Fig. 5C and D). These results indicate that ripasudil maintains ZO-1 expression in the insoluble fraction of Triton X-100, where TJs should localize.
Ripasudil inhibits claudin-5 and ZO-1 redistribution and restores the binding ratio between claudin-5 and ZO-1 decreased by cytokine stimulation. A and B: Results of the Western blot analysis (A) and the corresponding quantitative analysis (B) for cell membrane–localized claudin-5. n = 3. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. C and D: Western blot results (C) and the corresponding quantitative analysis (D) for ZO-1 in the 1% Triton X-100 S and I fractions. n = 3. E and F: Results of immunoprecipitation (IP) analysis to examine the interaction of ZO-1 with claudin-5 (E) and quantitative analysis of the claudin-5–to–ZO-1 ratio (F). n = 3. G: Ripasudil mechanism of action in claudin-5 redistribution under inflammatory conditions. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001. Student t test.
Ripasudil inhibits claudin-5 and ZO-1 redistribution and restores the binding ratio between claudin-5 and ZO-1 decreased by cytokine stimulation. A and B: Results of the Western blot analysis (A) and the corresponding quantitative analysis (B) for cell membrane–localized claudin-5. n = 3. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. C and D: Western blot results (C) and the corresponding quantitative analysis (D) for ZO-1 in the 1% Triton X-100 S and I fractions. n = 3. E and F: Results of immunoprecipitation (IP) analysis to examine the interaction of ZO-1 with claudin-5 (E) and quantitative analysis of the claudin-5–to–ZO-1 ratio (F). n = 3. G: Ripasudil mechanism of action in claudin-5 redistribution under inflammatory conditions. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001. Student t test.
We next examined whether ripasudil influences the bond between ZO-1 and claudin-5 (Fig. 5E and F). There was no significant difference in the claudin-5–to–ZO-1 ratio among the treatment groups under the control condition. However, the binding of claudin-5 and ZO-1 was reduced by TNFα, IL-6, and MCP-1, while ripasudil treatment restored it to the control level. In contrast, dexamethasone did not increase TEER values under TNFα, IL-6, or MCP-1 stimulation (Supplementary Fig. 4A). Claudin-5 and ZO-1 expression did not recover, and F-actin polymerization remained after dexamethasone treatment (Supplementary Fig. 4B).
These results suggest that 1) ripasudil treatment maintains both ZO-1 expression and the claudin-5/ZO-1 bond under cytokine stimulation by inhibiting actin polymerization and 2) cytoskeletal stabilization contributes to stable claudin-5 expression (Fig. 5G).
Ripasudil Enhances the Transcription of Claudin-5 by Inhibiting the Intranuclear Transport of β-Catenin
Claudin-5 expression is regulated by its degradation and synthesis (17). We thus investigated the effect of ripasudil on claudin-5 transcription in endothelial cells using real-time PCR. In both control and inflammatory conditions, ripasudil significantly increased claudin-5 mRNA (Fig. 6A, left), but not that of claudin-3, which is another TJ protein required for epithelial barrier formation (Fig. 6A, right). The effect of ripasudil on the transcription of ZO-1 and TJ proteins (other than claudin-5) such as occludin and claudin-12, which are expressed in the retinal vascular endothelium (13), was also examined (Supplementary Fig. 5). Transcription of ZO-1, but not occludin or claudin-12, was enhanced by ripasudil. These results suggest that ripasudil’s protective barrier function is specific to the claudin-5–ZO-1 complex in vascular endothelial cells. We also confirmed the ripasudil-induced increase of claudin-5 transcription in HRMECs (Fig. 6B). In Kimba mice, claudin-5 transcription levels were elevated not only by ripasudil but also by IVB. Notably, ripasudil and IVB combination therapy significantly increased claudin-5 transcription compared with monotherapy with either agent (Fig. 6C).
Ripasudil promotes claudin-5 transcription under inflammatory conditions by blocking β-catenin nuclear translocation. A: Effects of ripasudil and bevacizumab on claudin-5 (left) and claudin-3 (right) transcription levels in bEND.3 cells under cytokine-stimulated conditions. n = 3. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. B: Effect of ripasudil on claudin-5 transcription levels in HRMECs under cytokine-stimulated conditions. n = 3. C: Effect of Rip-OS on claudin-5 transcription levels in the retina of Kimba mice. n = 4. D: Effects of ripasudil and bevacizumab on β-catenin (green) and VE-cadherin (red) distribution under cytokine-stimulated conditions. Dot-like stains of β-catenin in the cytoplasm are indicated by arrowheads. n = 3. Scale bar = 10 μm. E: Effects of ripasudil and bevacizumab on the binding of β-catenin to the claudin-5 promoter region (top) and its binding ratio (bottom). n = 3. F: The action mechanism of ripasudil on claudin-5 transcription under inflammatory conditions. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
Ripasudil promotes claudin-5 transcription under inflammatory conditions by blocking β-catenin nuclear translocation. A: Effects of ripasudil and bevacizumab on claudin-5 (left) and claudin-3 (right) transcription levels in bEND.3 cells under cytokine-stimulated conditions. n = 3. Bev, 300 μg/mL bevacizumab; Rip, 30 μmol/L ripasudil. B: Effect of ripasudil on claudin-5 transcription levels in HRMECs under cytokine-stimulated conditions. n = 3. C: Effect of Rip-OS on claudin-5 transcription levels in the retina of Kimba mice. n = 4. D: Effects of ripasudil and bevacizumab on β-catenin (green) and VE-cadherin (red) distribution under cytokine-stimulated conditions. Dot-like stains of β-catenin in the cytoplasm are indicated by arrowheads. n = 3. Scale bar = 10 μm. E: Effects of ripasudil and bevacizumab on the binding of β-catenin to the claudin-5 promoter region (top) and its binding ratio (bottom). n = 3. F: The action mechanism of ripasudil on claudin-5 transcription under inflammatory conditions. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test.
VE-cadherin, a component of the AJs between endothelial cells (Fig. 4B), controls claudin-5 transcription by regulating β-catenin localization (11). We used immunostaining to examine the effect of ripasudil on β-catenin and VE-cadherin localization in cytokine-stimulated conditions (Fig. 6D). In the control group, β-catenin was expressed in a linear pattern just under the cell membrane. In the TNFα and MCP-1 groups, a change in β-catenin localization from the cell membrane to the cytoplasm was observed (Fig. 6D, arrowheads), which was suppressed by ripasudil. In the IL-6 group, the expression of β-catenin under the cell membrane was decreased, but no change in cytoplasmic localization was observed.
TNFα, IL-6, or MCP-1 addition resulted in intermittent VE-cadherin expression on the cell membrane, and many cytoplasmic dot-like stains were observed. Ripasudil attenuated this effect under stimulation with TNFα or MCP-1 but not IL-6 (Fig. 6D). We then used chromatin immunoprecipitation (ChIP) to examine the amount of β-catenin bound to the claudin-5 promoter region (17) (Fig. 6E, top) and its binding ratio (Fig. 6E, bottom). Both TNFα and MCP-1 enhanced β-catenin binding, and both the binding amount and binding ratio were significantly weakened by ripasudil (Fig. 6E).
Based on these results, we propose a mechanism (Fig. 6F) in which in the inflammatory state (particularly with high TNFα or MCP-1 expression), actin polymerization triggers β-catenin nuclear translocation and its binding to the claudin-5 promoter, inhibiting the transcription of this essential junction component. Ripasudil appears to prevent claudin-5 transcription suppression by inhibiting actin polymerization.
Ripasudil Inhibits Claudin-5 Redistribution Under Stimulation With the Vitreous Fluid of Human nDME
We next examined whether the vascular barrier function and the expression of claudin-5 in endothelial cells changed upon stimulation by human nDME vitreous fluid. Stimulation with vitreous fluid significantly reduced TEER values (Supplementary Fig. 6A). Claudin-5 redistribution was also confirmed by immunostaining and Western blot analysis after biotinylation of cell membrane–localized molecules (Supplementary Fig. 6B–D). Notably, only ripasudil recovered the TEER values and claudin-5 expression to normal levels, suggesting that ROCK also acts as the downstream molecule of the VEGF signaling pathway.
ROCK Acts as the Downstream Molecule of the VEGF Signaling Pathway and Contributes to VEGF-Induced Vascular Barrier Dysfunction
We investigated whether ROCK acts downstream of the VEGF signaling pathway (Supplementary Fig. 7). The drop in TEER values caused by VEGF recovered with ripasudil in a concentration-dependent manner (Supplementary Fig. 7A). Immunostaining also confirmed that ripasudil inhibited claudin-5 redistribution and actin polymerization in a concentration-dependent manner (Supplementary Fig. 7B and C). Western blot analysis of cell surface proteins indicated that ripasudil maintained claudin-5 expression in cell membranes (Supplementary Fig. 7D).
In contrast, ZO-1 expression on cell membranes (Supplementary Fig. 7B and C) and I/S ratio (Supplementary Fig. 7F and G) were unchanged by VEGF stimulation and unaffected by ripasudil (Supplementary Fig. 7B, C, F, and G). However, ZO-1 and claudin-5 binding was attenuated by VEGF stimulation, and this effect was inhibited by ripasudil (Supplementary Fig. 7H and I). Ripasudil also restored the VEGF-induced decrease in claudin-5 mRNA levels (Supplementary Fig. 7J). Immunostaining showed β-catenin relocalization from the cell membrane to the cytoplasm, while VE-cadherin disappeared from cell membranes (Supplementary Fig. 7K). These phenomena were attenuated by ripasudil administration (Supplementary Fig. 7K).
ChIP showed that VEGF-enhanced β-catenin nuclear translocation resulted in increased β-catenin binding to the claudin-5 promoter (Supplementary Fig. 7L and M). Furthermore, ripasudil inhibited claudin-5 transcriptional repression by retaining β-catenin localization on cell membranes (Supplementary Fig. 7L and M). These results demonstrate that ROCK acts as the downstream molecule for VEGF signaling pathway.
ROCK2, but Not ROCK1, Is Involved in Endothelial Claudin-5 Redistribution Under VEGF and Cytokine Stimulation
We examined which ROCK isoform (ROCK1 and/or ROCK2) (22) could affect the vascular barrier using ROCK1 or ROCK2 knockout (KO) endothelial cells (Fig. 7A). Neither ROCK1 nor ROCK2 affected claudin-5, ZO-1, or F-actin expression under normal conditions, but ROCK2 promoted claudin-5 and ZO-1 redistribution and actin polarization under VEGF- or inflammatory cytokine–stimulated conditions (Fig. 7B and C). We also confirmed that VEGF and inflammatory cytokine stimulation failed to decrease TEER values in ROCK2 KO endothelial cells (Fig. 7D). The expression level of claudin-5 mRNA in ROCK1 KO cells was similar to that of MOCK cells. In ROCK1 KO cells, VEGF or TNFα decreased claudin-5 mRNA expression, while IL-6 increased it (Fig. 7E). Interestingly, in ROCK2 KO cells, the claudin-5 mRNA expression level was significantly lower compared with MOCK cells, even without stimulation (Fig. 7E). VEGF, TNFα, and IL-6 further decreased claudin-5 mRNA expression in ROCK2 KO cells, although claudin-5 protein expression was maintained in cell membranes (Fig. 7B and E). ChIP demonstrated that ROCK2 KO and ripasudil inhibited β-catenin nuclear migration under VEGF, TNFα, or MCP-1 stimulation (Fig. 7F), implying the possible degradation of claudin-5 mRNA in ROCK2 KO cells due to positive feedback. These results suggest that ROCK2, but not ROCK1, is closely involved in endothelial cell junction integrity.
ROCK2 is essential for claudin-5 redistribution under VEGF or cytokine stimulation. A: Sanger sequencing results in ROCK1 or ROCK2 KO cells. Deletions are marked by dashes. PAM, protospacer adjacent motif. B: Claudin-5 distribution in MOCK, ROCK1 KO, or ROCK2 KO endothelial cells under VEGF or cytokine stimulation. (-) indicates no stimulation. n = 4. Scale bar = 10 μm. ICC, immunocytochemistry. C: ZO-1 distribution (green) and actin filament network (red) in MOCK, ROCK1 KO, or ROCK2 KO cells under cytokine stimulation. n = 4. Scale bar = 10 μm. D: TEER values of MOCK and ROCK2 KO bEND.3 cells under cytokine-stimulated conditions. n = 9. E: Claudin-5 mRNA expression in MOCK, ROCK1 KO, or ROCK2 KO cells under VEGF or cytokine stimulation. n = 4. F: Binding of β-catenin to the claudin-5 promoter region (left) and its binding ratio (right) in MOCK or ROCK2 KO cells under VEGF or cytokine stimulation. n = 4. All graphs represent the mean ± SEM. ‡P < 0.05, §P < 0.01, †P < 0.001, *P < 0.0001, Student t test.
ROCK2 is essential for claudin-5 redistribution under VEGF or cytokine stimulation. A: Sanger sequencing results in ROCK1 or ROCK2 KO cells. Deletions are marked by dashes. PAM, protospacer adjacent motif. B: Claudin-5 distribution in MOCK, ROCK1 KO, or ROCK2 KO endothelial cells under VEGF or cytokine stimulation. (-) indicates no stimulation. n = 4. Scale bar = 10 μm. ICC, immunocytochemistry. C: ZO-1 distribution (green) and actin filament network (red) in MOCK, ROCK1 KO, or ROCK2 KO cells under cytokine stimulation. n = 4. Scale bar = 10 μm. D: TEER values of MOCK and ROCK2 KO bEND.3 cells under cytokine-stimulated conditions. n = 9. E: Claudin-5 mRNA expression in MOCK, ROCK1 KO, or ROCK2 KO cells under VEGF or cytokine stimulation. n = 4. F: Binding of β-catenin to the claudin-5 promoter region (left) and its binding ratio (right) in MOCK or ROCK2 KO cells under VEGF or cytokine stimulation. n = 4. All graphs represent the mean ± SEM. ‡P < 0.05, §P < 0.01, †P < 0.001, *P < 0.0001, Student t test.
ROCK Inhibition Prevents Retinal Hyperpermeability in rDME
Lastly, we investigated the ability of ripasudil to protect vascular barrier function under stimulation with rDME vitreous fluid. TEER was restored even with the administration of bevacizumab alone, but the effect of ripasudil was significantly stronger (Fig. 8A).
Combination of anti-VEGF therapy and ROCK inhibition achieves rDME disappearance. A: Effect of ripasudil on TEER values under rDME stimulation. Rip and Bev indicate 30 μmol/L ripasudil and 300 μg/mL bevacizumab, respectively. n for control and - (rDME stimulation only) = 4. n for Rip and Bev = 7. B: Therapeutic effects of IVB + IVF on rDME patients. These two patients were extracted from those who were enrolled in our previous clinical trial (ref. 47). At least 3 months have passed between the final IVB and IVB + IVF. C: Schematic representation of “VEGF-dependent” and “VEGF-independent” inflammation occurring in the DME retina. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test. BCVA, best corrected visual acuity.
Combination of anti-VEGF therapy and ROCK inhibition achieves rDME disappearance. A: Effect of ripasudil on TEER values under rDME stimulation. Rip and Bev indicate 30 μmol/L ripasudil and 300 μg/mL bevacizumab, respectively. n for control and - (rDME stimulation only) = 4. n for Rip and Bev = 7. B: Therapeutic effects of IVB + IVF on rDME patients. These two patients were extracted from those who were enrolled in our previous clinical trial (ref. 47). At least 3 months have passed between the final IVB and IVB + IVF. C: Schematic representation of “VEGF-dependent” and “VEGF-independent” inflammation occurring in the DME retina. All graphs represent the mean ± SEM. †P < 0.05, *P < 0.01, §P < 0.001, ‡P < 0.0001, Student t test. BCVA, best corrected visual acuity.
Clinical observations of the effects of combination therapy (anti-VEGF antibody and intravitreal injection of fasudil [IVF], another ROCK inhibitor) on two rDME patients are illustrated in Fig. 8B. Fluorescein angiography and OCT imaging showed that combination therapy reduced fluorescein leakage from retinal vessels and the central foveal macular edema. Retinal thickness was decreased, and best-corrected visual acuity improved in both patients (Fig. 8B).
The proposed mechanisms of ROCK inhibition in rDME are shown in Fig. 8C. ROCK inhibition has a marked therapeutic effect by preventing BRB dysfunction in a manner distinct from that of anti-VEGF therapy, specifically, through the inhibition of “secondary” monocyte/macrophage infiltration and activation (VEGF-independent inflammation) by suppression of macrophage/microglia activation and the blockage of secondary vascular barrier dysfunction due to inflammatory cytokines.
Discussion
Retinal inflammation is involved in DR pathogenesis, including vasopermeability (11). While the precise mechanism underpinning rDME remains unknown (35), the current study suggests that 1) VEGF-dependent and VEGF-independent inflammation are present in the DME retina, 2) VEGF causes VEGF-independent retinal inflammation, and 3) VEGF-independent inflammation is involved in rDME pathology.
A clinical trial showed that only approximately one-half of DME cases could be treated with anti-VEGF therapy alone (5,6). In this study, both VEGF and inflammatory cytokine concentrations in the vitreous fluid were increased and positively correlated in nDME patients (32). Furthermore, good responses were reported in DME patients with increased aqueous humor cytokine concentrations that correlated with VEGF (36), suggesting that “VEGF-dependent retinal inflammation” occurs in DME. However, in the vitreous of our rDME patients, VEGF concentration was decreased, but TNFα, IL-6, and MCP-1 were unchanged, suggesting that VEGF-independent retinal inflammation can also occur in rDME. If responsiveness to anti-VEGF therapy depends on the quality of inflammation, the key to rDME resolution is to identify the factor that can suppress VEGF-independent inflammation.
In this study, we used Kimba mice and showed that prolonged VEGF overexpression causes VEGF-independent inflammation. This mouse model also shows retinal edema, which is not normally seen in the retinas of diabetic mice. As expected, we observed that VEGF overexpression alone in Kimba mice could cause retinal inflammation. Similar to the DR retina (37), macrophages/microglia were activated and secreted several inflammatory cytokines. The significant reduction of TNFα and IL-6 expression in the retinas of Kimba mice treated with IVB × 2 also confirms that VEGF-dependent inflammation occurs. IVB treatment did not cure retinal edema in Kimba mice or suppress MCP-1 or KC expression. Further, the proportions of infiltrated monocyte/macrophage (CD11b+ Ly6C+ CCR2+ cells) as well as activated macrophage/microglia (CD11b+ CD80+ cells) were unchanged. These results imply that VEGF-independent inflammation can occur in the VEGF-overexpressed retina.
Analysis of phosphorylated proteins in the Kimba mouse retina showed increased ROCK-related signaling activity, which is related to various disease pathophysiologies, including DR and age-related macular degeneration (26–29,38). Indeed, Rip-OS and IVB showed different inhibitory effects on retinal inflammation in Kimba mice.
Actin cytoskeleton reorganization is necessary for leukocyte infiltration. We previously demonstrated that retinal macrophage infiltration and polarization can be suppressed by ROCK inhibition (38). The increase in the number of infiltrating leukocytes in Kimba mice is thought to be driven not only by an active mechanism (leukocyte skeletal changes) but also by a passive mechanism (leakage due to disruption of the vascular barrier). The differentiation and activation of macrophages are affected by actin filament structure (39). In the current study, Rip-OS significantly suppressed TNFα, MCP-1, and KC expression levels and inhibited CD11b+ CD80+ macrophage/microglia increases in the retinas of Kimba mice, indicating that ROCK inhibition can suppress monocyte/macrophage infiltration and change the characteristics of infiltrated macrophages and microglia in VEGF-overexpressed retinas.
We demonstrated that ROCK inhibition blocked claudin-5 redistribution and recovered vascular barrier dysfunction in cytokine-stimulated conditions, suggesting that ROCK 1) weakens claudin-5 and ZO-1 binding by promoting actin stress fiber formation and 2) promotes claudin-5 degradation. Surprisingly, ROCK inhibition both prevented claudin-5 degradation and promoted its transcription—a novel finding. Our present results showed that cytokine (other than IL-6) stimulation promotes β-catenin nuclear migration and suppresses claudin-5 transcription. Preservation of VE-cadherin on cell membranes is needed to inhibit β-catenin migration (17), and stress fiber formation and decreased ZO-1 expression are known to promote VE-cadherin redistribution (40), suggesting that claudin-5 transcription enhancement by a ROCK inhibitor took place via actin cytoskeleton stabilization.
The claudins are a 27-member multigene family and contribute to TJ function with a tissue-specific expression pattern (41,42). In central nervous vessels, including the retina, a strong barrier is formed by claudin-5 (15), but in general, more than two claudins are expressed in each cell, and the permeability of water and ions is regulated by their bonding pattern (homophilic or heterophilic interactions) (43). The regulatory mechanisms of claudin expression are diverse, and the proteins controlling transcription are different for each claudin (44,45). A ROCK inhibitor is already used for the treatment of glaucoma, and ROCK inhibition decreases intraocular pressure by disrupting the Schlemm canal endothelial barrier (25). We speculate that this contradictory reaction occurs due to differences in TJ molecular components between the endothelial cells of the retina and Schlemm canal.
Unlike vascular endothelial cells, claudin-11 is closely involved in TJ formation between Schlemm endothelial cells (46). As ROCK specifically affected claudin-5 transcription levels (claudin-3 transcription was unchanged) in different vascular endothelial cells, the ROCK–β-catenin pathway may be the universal regulatory mechanism of claudin-5 transcription in endothelial cells. A ROCK inhibitor was able to upregulate claudin-5 transcription in vivo and in vitro and may therefore be effective for retinal diseases with vascular barrier dysfunction other than DR.
The breakdown of the vascular barrier function with nDME vitreous stimulation completely recovered with the ROCK inhibitor alone, indicating that ROCK acts downstream of the VEGF signaling pathway. VEGF-induced claudin-5 degradation and transcription suppression were inhibited by ROCK inhibition, with claudin-5 maintained at the cell membrane.
We hypothesized that VEGF-dependent and VEGF-independent inflammation occurs simultaneously in rDME. Large-scale trials have demonstrated that VEGF is the most crucial molecule for DME formation (4–6). For suppression of VEGF, the continuation of anti-VEGF therapy is essential. Our observation that the therapeutic effect of IVB × 1 + Rip-OS was inferior to that of IVB × 2 + Rip-OS in Kimba mice confirms the importance of continuing anti-VEGF therapy.
We previously reported the possible inhibitory effect of combined anti-ROCK/anti-VEGF therapy for refractory macular edema (47). Our current results in rDME and in mice with VEGF-overexpressed retinas show that prevention of retinal inflammation and vasopermeability could be achieved by combination therapy. In addition, the integrity of claudin-5 within junctional complexes in the retina of Kimba mice was significantly increased by combination therapy compared with each monotherapy. Moreover, our in vitro experiments showed that ROCK2, but not ROCK1, inhibition could block vascular hyperpermeability. Taken together, our data suggest that a ROCK2-specific inhibitor may be more effective than unspecific ROCK inhibitors against anti-VEGF resistance.
A phase 2 trial (Protocol U) using anti-VEGF therapy and vitreous injection of a sustained dexamethasone implant for rDME patients showed that the combination group had a significantly greater reduction in retinal thickness compared with the anti-VEGF monotherapy group (48). Our current study reveals that ROCK inhibition attenuates the upregulation of several cytokines, which can increase vascular permeability. The pertinent difference between ripasudil and steroids is that ripasudil works directly on vascular endothelial cells and protects barrier function by blocking claudin-5 redistribution. Our clinical trial suggests that unlike steroids, ripasudil has a lower risk of increasing intraocular pressure and cataracts (47). If ripasudil eye drop therapy is shown to be efficacious in human rDME, ripasudil may be accepted as a safer drug. The therapeutic effect of ripasudil on rDME also needs to be compared with steroids as well as targeted drugs against specific cytokines, such as TNFα, IL-6, and MCP-1.
The route of administration of ripasudil also needs to be carefully considered. Eye drop therapy is less invasive than intravitreal injections, but whether a concentration sufficient to eliminate macular edema in the human retina can be maintained with eye drops remains unclear. However, using animal models, we previously showed that ripasudil is able to reach the retina even with eye drops and that the sufficient retinal concentration of ripasudil was maintained for at least 8 h (34). In addition, Chihara et al. (49) reported that ripasudil instillation twice a day enhanced the peripapillary retinal vascular density using OCT angiography. Thus, it will be worthwhile to investigate the effects of anti-VEGF therapy and ripasudil eye drops on rDME in the future.
An important limitation of our approach was the use of the Kimba mouse model, in which diabetes is not present and VEGF is overexpressed in photoreceptor cells. Using conventional mice with STZ-induced diabetes, we also examined the efficacy of ripasudil (Supplementary Fig. 8). In this model, vascular hyperpermeability occurs but retinal edema does not. As shown in Supplementary Fig. 8, administration of ripasudil suppressed the increase in vascular permeability and the decrease in the expression of claudin-5. These results suggest that ROCK is involved in the disruption of vascular barrier function in DR, supporting our proposal that ROCK may be a novel therapeutic target.
Another limitation was the inability to collect vitreous from patients whose edema disappeared with anti-VEGF therapy alone. If the cytokines in the vitreous of these patients are also reduced along with VEGF, the presence of VEGF-dependent inflammation can be confirmed. Since several cytokines and VEGF in the aqueous humor declined in parallel in the group with a good response to anti-VEGF therapy (50), it is expected that a similar phenomenon occurs in the vitreous.
In conclusion, we demonstrated that VEGF-independent inflammation causes rDME, which can be mediated by ROCK—especially ROCK2.
Article Information
Acknowledgments. The authors thank Dr. Elizabeth P. Rakoczy (University of Western Australia) for the gift of Kimba mice. The authors also thank Mitsuhiro Kurata, Takako Iwasaki, Masayo Eto, and Mizuho Oda (Kyushu University) for technical assistance.
Funding. This study was partly supported by Japan Society for the Promotion of Science (JSPS) KAKENHI (17K16997 [M.A.] and 17K11456 [S.N.]), Charitable Trust Fund for Ophthalmic Research in Commemoration of Santen Pharmaceutical’s Founder (S.N.), and the Takeda Science Foundation (S.N.). The powder and ophthalmic solution of ripasudil used in this study were provided free of charge from Kowa Company, Ltd., to Kyushu University.
Duality of Interest. This study was also supported by the Bayer Retina Award (S.N.), Novartis Pharmaceuticals research grants (S.N.), Alcon research grants (S.N.), and grants from Kowa Company, Ltd. This research was conducted in collaboration with Kyushu University and Kowa Company, Ltd. S.N. and K.-H.S. received research support fees from Kowa Company, Ltd. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. S.N. conceived the project and designed the experiments with M.A. With critical assistance from A.W.S., T.I., and K.-H.S., M.A., M.Y., Y.F., and K.S. performed the in vivo experiments and M.A., H.F., Y.K., and I.W. conducted the in vitro experiments. H.A. performed the clinical study of intravitreal fasudil injection with DME patients. M.A. and S.N. were involved in data analysis. M.A. and S.N. wrote the manuscript. S.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.