The Early Pathogenesis of Diabetic Retinopathy and Its Attenuation by Sodium–Glucose Transporter 2 Inhibitors

Diabetic retinopathy (DR) is a devastating complication of diabetes and remains the leading cause of visual impairment in working-age individuals worldwide (1,2). Currently, in addition to systemic treatments, such as blood glucose control and blood pressure control, there are three other treatments for DR: retinal photocoagulation, vitrectomy surgery, and intravitreal injection of vascular endothelial growth factor (VEGF) inhibitors or corticosteroids. However, these treatments target the late stage of DR, and their ability to restore already impaired vision is limited. Therefore, it is important to identify the early pathogenic mechanisms of DR. Moreover, the treatment targeting these mechanisms may prevent its aggravation.

Pericyte loss is proposed to be an initial pathological mechanism of DR. This is based on histopathological analysis of the retina of patients with DR reported in the 1960s (3,4). A recent report also showed that transient inhibition of pericyte recruitment to developing retinal vessels by injection of an anti–platelet-derived growth factor receptor b antibody in mice reproduces the characteristic features of DR (5). These findings suggest that pericyte loss may be sufficient to reproduce retinal abnormalities characteristic of DR. However, the information on the mechanism of diabetes-induced pericyte loss is limited. Moreover, there are few studies on the sequential and in-depth characterization of vascular and inflammatory alterations during the early stages of DR. Such studies may further insights into the early pathogenic mechanisms of DR and may lead to novel, effective therapeutic strategies.

Sodium–glucose cotransporter 2 inhibitors (SGLT2i) are a class of oral antidiabetes agents that block the reabsorption of glucose in the proximal renal tubules, thereby increasing urinary glucose excretion and lowering blood glucose levels (6). In recent years, an increasing number of clinical studies have indicated the beneficial effects of SGLT2i on nephropathy and cardiovascular events in patients with type 2 diabetes (7–9).

In contrast, studies that explored the effect of SGLT2i on DR are very limited, and the results are controversial (10–14). Therefore, in this study, we first investigated the mechanisms of early pathogenic alterations around retinal blood vessels in streptozotocin (STZ)-induced diabetic mice. This rodent model represents molecular and cellular processes characteristic of human nonproliferative DR (15). Next, we investigated the effect of SGLT2i on these changes leading to the potential to prevent DR.

Male C57BL/6JJcl mice (7 weeks old) were purchased from CLEA Japan (Tokyo, Japan). All the mice were bred under pathogen-free conditions at the Graduate School of Pharmaceutical Sciences Animal Center, Kyushu University. The animals had free access to tap water and standard diet (MF; Oriental Yeast Co., Ltd., Japan). Ipragliflozin and luseogliflozin were provided by Astellas Pharma Inc. (Tokyo, Japan) and Taisho Pharmaceutical Co., Ltd. (Tokyo, Japan), respectively.

Animals were intraperitoneally injected with STZ (150 mg/kg body wt in 50 mmol/L citrate buffer, pH 4.5) or citrate buffer alone (control) after overnight fasting. Mice with blood glucose levels >300 mg/dL were considered diabetic 3 days after the STZ injection.

Plasma glucose levels were determined using the glucose CII-test assay kit (Wako Pure Chemical Industries, Osaka, Japan). Plasma VEGF-A concentrations were measured using a Mouse Quantikine ELISA Kit (Proteintech Group, Tokyo, Japan).

Vascular leakage was analyzed according to the protocol described by Chikaraishi et al. (16). To examine the distribution of fluorescein-conjugated dextran in the retinas, fluorescein-conjugated dextran dissolved in 0.15 mol/L PBS was injected into the tail vein of mice at a dose of 0.25 mg/g body wt. The eyes were enucleated 30 min later for fluorescence microscopy analysis. Fluorescence images of the whole mounts were viewed using a BZ-X800 microscope (KEYENCE Corp., Osaka, Japan).

The enucleated eyeballs were fixed in 4% paraformaldehyde for 30 min at room temperature. For hematoxylin-eosin (H-E) staining of retinal sections, the cornea and lens were removed, and the samples were transferred to 30% sucrose in PBS for cryoprotection at 4°C overnight, embedded in Tissue Tek optimum cutting temperature compound (Sakura Finetek Japan Co., Ltd., Tokyo, Japan), and frozen. To estimate the changes in the hypertrophic retina, frozen tissue was cut into 8-μm-thick sections and stained with H-E (17).

Immunofluorescence staining of whole-mounted retinas and retinal sections was performed according to the protocol described by Park et al. (18). Briefly, the frozen tissues were cut into 16-μm sections, flattened, and stored in methanol at −20°C until use. The samples were blocked with 5% donkey (or goat) serum in PBS with 0.3% Triton X-100 and then incubated in blocking solution with each antibody for 48 h at 4°C. After several washes, samples were incubated for 2 h at room temperature with FITC-, Cy3-, or Cy5-conjugated secondary antibodies. The antibodies used in this study are listed in Supplementary Tables 1 and 2.

Immunofluorescence images were captured separately and merged after pseudocoloring using a BZ-X800 microscope. Morphometric analyses of the retina were performed using ImageJ software (https://rsb.info.nih.gov/ij) or the BZ-X Analyzer software program (KEYENCE Corp.). Staining intensities were measured in eight representative areas of the retinal vascular regions or, if indicated, other regions in each retina, and averaged.

For analysis of hypoxia in the retina, Hypoxyprobe-1 (Hypoxyprobe Plus Kit; solid pimonidazole HCl; Hypoxyprobe, Burlington, MA) was injected intravenously at a dose of 60 mg/kg 1 h before enucleation. The retinas were then harvested and stained with FITC-conjugated anti-Hypoxyprobe antibody. Images of all samples were obtained using a BZ-X800 microscope.

Microglial cells were magnetically isolated from the cerebrum of C57BL/6JJcl mice according to the manufacturer’s instructions (Adult Brain Dissociation Kit, mouse and rat, gentleMACS Octo Dissociator with Heaters and autoMACS Pro Separator; Miltenyi Biotec, Bergisch Gladbach, Germany) as shown in Supplementary Fig. 1. The microglia pellets were gently resuspended and seeded on poly-d-lysine–coated 96-well plates (5 × 10⁴ cells per well). To induce microglial activation, lipopolysaccharide (LPS; Sigma-Aldrich, Taufkirchen, Germany) was used. After seeding, the cells were cultured in DMEM with 10% FBS and with or without SGLT2i (50 μmol/L) for 24 h. Then, the cells were stimulated with LPS. The stimulation time was 24 h for quantitative PCR and ELISA experiments.

RNA was extracted (NucleoSpin RNA XS; Macherey-Nagel, Düren, Germany) and first-strand cDNA was synthesized from 250 ng total RNA (ReverTra Ace qPCR RT Master Mix; TOYOBO, Osaka, Japan). The following genetic analysis assays for real-time PCR were performed with CFX Connect Real-Time System (Bio-Rad, Hercules, CA) using 10 ng of the first-strand cDNA reaction. The primers are shown in Supplementary Table 3.

The concentration of tumor necrosis factor-α (TNF-α) and interleukin 6 (IL-6) was measured in the cell culture supernatants using ELISA according to manufacturer’s instructions (R&D Systems, Minneapolis, MN).

To estimate Na⁺/H⁺-exchanger-1 (NHE-1) activity, we measured intracellular pH and the concentration of Na⁺ ([Na⁺]c) using BV2 cells and primary microglia, respectively (19). BV2 cells were provided by Dr. Tsuda (Kyushu University). BV2 cells were subjected to an acid load by a transient application (2 min) of a 30 mmol/L NH₄/NH₃ solution, and then the recovery of pH was measured with BCECF-AM (2′,7′-bis-[2-carboxyethyl]-5-[and-6]-carboxyfluorescein– acetoxymethyl ester; Invitrogen). The intracellular [Na⁺]c in primary microglia was detected using CoroNa Green, AM (Invitrogen).

A homology model of the protein structure of human NHE-1 was prepared using Protein Data Bank (ID: 4CZB) as a template. Molecular docking studies were performed using Webina 1.0.3 to explore possible interactions between NHE-1 and SGLT2i. Webina is a JavaScript/WebAssembly library that runs AutoDock Vina entirely in a web browser (https://durrantlab.pitt.edu/webina/). Molecular docking results were illustrated using the PyMol molecular graphic program.

All results are shown as the mean ± SEM. Statistical significance was analyzed using the Tukey-Kramer or Steel-Dwass test. The minimum level of statistical significance was set at a probability value of 0.05.

All procedures and animal care procedures were approved by the Committee on Ethics of Animal Experiments, Graduate School of Pharmaceutical Sciences, Kyushu University, and were conducted according to the Guidelines for Animal Experiments of the Graduate School of Pharmaceutical Sciences, Kyushu University.

The data and resources generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

The characteristics of experimental mice are presented in Supplementary Table 4. Plasma glucose levels were higher in STZ-treated mice than in control mice.

To investigate the changes in the hypertrophic retina of STZ mice, we stained retinal sections with H-E. Representative histological images of the whole eye and changes in the hypertrophic retina are shown in Fig. 1A and B.

The retinal thickness in 2-week-STZ mice was significantly higher at all points of the superior hemisphere than that in normal mice (Fig. 1C). However, in 4-week- and 6-month-STZ mice, the increase in retinal thickness was confined to the superior hemisphere near the optic nerve head (Fig. 1D). At 6 months after STZ administration, the retinal thickness gradually narrowed from the optic nerve head to the front of the retina. Areas of retinal thinning and detachment were also observed (Fig. 1B, right panel).

Increased fluorescein leakage, indicating increased vascular permeability, was observed in the eyes of diabetic mice (Fig. 1E). Figure 1F illustrates typical scaled-up images of the superficial and intermediate vascular layers. Vascular permeability was assessed as the sum of the leakages in any of the three vascular layers. The number of leakages significantly increased in the retina at 4 weeks and 6 months of STZ administration (Fig. 1G). In the current study, no difference was observed in the frequency of leakage among the three retinal vascular layers.

To gain insight into pericyte loss in our animal model, the whole-retinal mount was immunohistochemically stained with CD31 and NG2 as markers of endothelial cells and pericytes, respectively. Figure 2A shows NG2-positive pericyte coverage of CD31-positive vessels in the superficial, intermediate, and deep retinal vascular layers. These images show that almost all of the blood vessels were covered with pericytes. Figure 2B shows the CD31- or NG2-positive area (%) on the left-hand axis of the ordinate (colored bars), and the ratio of NG2 to CD31 on the right-hand axis of the ordinate (polygonal lines). Interestingly, not only the NG2-positive area but also the CD31-positive area gradually decreased with hyperglycemia duration, resulting in a substantially constant ratio between the NG2- and CD31-positive areas in all three layers (Fig. 2B). In addition, the vascular diameter was small in the retinas of the STZ mice (Fig. 2C). In parallel, the retinas of STZ mice exhibited extreme hypoxia (Fig. 2D and E). These results suggest that the loss of endothelial cells and pericytes occurred in parallel; thus, the ratio between the two remained constant in all three layers of retina of STZ mice, resulting in decreased retinal vessel density in the early stage of DR. In our experiment, blood vessels without pericytes were observed only in a part of the retina in two of the eight 6-month-STZ mice (Supplementary Fig. 2).

The angiopoietin-2 (Ang2)–positive region was widely distributed in the retina. In STZ mice, the Ang2 expression gradually increased in all three layers, depending on the duration of diabetes (Fig. 3A and B).

We estimated the intensity of VEGF-A and VEGF receptor 2 (VEGFR2) in the area between the surface of the retina and the deep retinal vascular plexus, as indicated by the arrows in Fig. 3C. Increased VEGF-A expression was observed along the three-layer vascular plexus of the retina in the 4-week- and 6-month-STZ mice (Fig. 3C and D). The expression of VEGFR2 in both the three-layer vascular plexus and retinal pigment epithelial cell layer also increased in the 4-week-STZ mice (Fig. 3C, E, and F). Additionally, plasma VEGF-A levels showed a significant increase in 6-month-STZ mice (Supplementary Table 4).

Systemic and local inflammation are both serious etiologies for diabetic complications. Because F4/80 is expressed in macrophages and microglia, distinguishing between them is difficult, except in the perivascular region. Therefore, we semiquantitatively analyzed the F4/80 region near the blood vessels that overlapped using CD31 immunostaining. F4/80-positive regions were scarcely observed in the retina of normal mice, but were observed mainly along the blood vessels in the 4-week- and 6-month-STZ mice (Fig. 4A and B). These results suggest that macrophages may infiltrate the retinal vessels in STZ mice.

The microglia in the retina of 2-week-STZ mice showed morphological features of DR (Fig. 4C). In 2-week-STZ mice, the microglia had fewer branches and processes (amoeboid form) than those of the normal mice, indicative of their activation. This morphological change was greatest in the retinal vascular plexus of 2-week-STZ mice, after which the length of the processes gradually increased to that in control mice (Fig. 4C and D). Moreover, the number of microglia decreased in all vascular plexuses in the 2-week-STZ mice and then recovered. The proliferation of microglia depends on the duration of diabetes. In the deep vascular plexus of 6-month-STZ mice, microglial proliferation significantly increased compared with that in control mice (Fig. 4C and E). These sequential morphological changes in microglia observed in this study may indicate their diverse functions in response to retinal damage (19,20).

We confirmed the presence of SGLT2 in the retina using immunostaining together with the endothelial cell marker CD31 and the Müller cell marker glutamine synthetase (Supplementary Fig. 3). SGLT2 was slightly expressed around the retinal vascular plexus, external limiting membrane, and the retinal pigment epithelium (Supplementary Fig. 3A). In addition, we also found that SGLT was expressed in a part of Iba1-positive cells of the retinas (Supplementary Fig. 3B).

We investigated the effect of two SGLT2i, ipragliflozin and luseogliflozin, on DR at two doses (0.1 or 3 mg/kg/day) (Supplementary Table 5). SGLT2i at 3 mg/kg significantly lowered blood glucose levels, but not at 0.1 mg/kg. Oral administration of the two SGLT2i significantly suppressed the hypertrophic retina and the increased vascular permeability observed in the 4-week-STZ mice at both doses (Fig. 5A–D). Additionally, the decreased vessel density in all vascular layers in the 4-week-STZ mice was restored to the levels of that in control mice at both doses (Fig. 5E). Although Ang2-positive regions were observed in various parts of the retina, increased Ang2 expression in the 4-week-STZ mice was apparently suppressed at both doses in the inner nuclear layer area, where microglia accumulated (Fig. 6A and B). As for inflammatory agents, morphological changes and decreased numbers of microglia in the 2-week-STZ mice were also both restored to the levels of those in control mice at both doses (Fig. 6C and D). In contrast, SGLT2i had no effect on infiltration of macrophages into retinal vessels in the 4-week-STZ mice (Supplementary Fig. 4). The doses of inhibitor used in this study had no (0.1 mg/kg/day) or partial hypoglycemic effect (3 mg/kg/day) (Supplementary Table 5). Therefore, they might not affect high-glucose–induced activation of macrophages (21) and their subsequent infiltration into retinal vessels. Increased retinal VEGF-A levels in the 4-week-STZ mice appeared to be partially suppressed at both doses (Fig. 6E and G). Since macrophages are the major source of VEGF-A (22,23), VEGF-A might not be fully suppressed. In contrast, the increased expression of retinal VEGFR2 in the 4-week-STZ mice was significantly suppressed by both SGLT2i at both doses (Fig. 6F, H, and I).

Immunostaining analysis showed the coexpression of NHE-1 and SGLT2 with Iba1 in primary microglia (Fig. 7A). Cultured primary microglia exhibited a ramified form at 24 h after seeding. They underwent a transformation from a ramified to an ameboid form by LPS stimulation. This transformation was restored by cotreatment with ipragliflozin or luseogliflozin (Fig. 7B).

To clarify the protective effect of SGLT2i on microglia activation under high glucose levels, we measured the IL-6 and TNF-α protein concentrations in the culture medium using ELISA. LPS stimulation for 24 h significantly increased the protein expression of IL-6 and TNF-α. Cotreatment with luseogliflozin significantly decreased the LPS-induced protein release of both (Fig. 7C). Further, LPS stimulation resulted in a significant increase in TNF-α mRNA expression (Fig. 7D), and cotreatment with ipragliflozin or luseogliflozin significantly decreased this mRNA-expression. The level of IL-6 mRNA was slightly increased by LPS stimulation, but it was not significant (data not shown).

Recently, NHE-1 has gained increasing attention as an off-target of some SGLT2i (24,25). Therefore, we examined the binding affinity of luseogliflozin and ipragliflozin on NHE-1 using a homology model of NHE-1. Both inhibitors displayed high binding affinity to the Na⁺-binding site of NHE-1 (Fig. 7E). We also confirmed that NHE-1 was widely expressed in retinal tissues (Supplementary Fig. 5).

We further investigated whether SGLT2i exhibited effects on NHE-1 by studying intracellular [Na⁺]c and pH. Figure 7F shows a reduced recovery of intracellular pH after NH4⁺ pulse with ipragliflozin or luseogliflozin. Furthermore, [Na⁺]c was reduced with these SGLT2i (Fig. 7G).

In this study, we demonstrated sequential alterations in vascular structural and inflammatory responses in the early stages of DR in STZ-induced diabetic mice. Thus, this study provides new insights into the pathogenic mechanisms of DR. To the best of our knowledge, this is the first study to show that SGLT2i attenuate these early pathogenic alterations, including retinal vascular hyperpermeability and edema.

Microglia/macrophages are major immune cells involved in various types of damage. In this study, morphological changes in microglia were observed around deep retinal capillaries 2 weeks after the onset of diabetes. This preceded the various alterations in vascular structure observed at 4 weeks. This morphological change in microglia has been reported in the retina of diabetic rodent models (26,27) and high-glucose culture (28). Although it is difficult to determine whether microglia are directly activated by high glucose levels or activated in response to the minimal retinal damage, our findings showed that microglia activation may play an important role in the initial stage of DR. Activated microglia can induce vascular or neuronal damage via the excessive release of inflammatory cytokines and reactive oxygen species (29). Simultaneously, the inflammatory reaction is considered to cause Ang2 release (30) and expression of VEGFR2 (31). Ang2 then destabilizes quiescent endothelial cell-pericyte interactions (30). Under vascular destabilization, pericytes withdraw from the blood vessels, resulting in endothelial cell death and vascular regression (32,33). Our data showed that these processes can also occur in the early stages of DR. In fact, the retinal blood vessels of STZ mice had a similar coverage of pericytes compared with normal mice but showed a lower density. Furthermore, the retina of STZ mice had extreme hypoxia compared with that of normal mice. Hypoxic conditions further increase the expression of Ang2 (34), VEGF-A (35), and VEGFR2 (31). In diabetic mice, angiogenesis may be activated by VEGF-A signaling via VEGFR2 (36,37). The destabilizing effect of Ang2 also promotes the angiogenic effect of VEGF-A–VEGFR2 signaling (38). Thus, these factors are mutually induced by local inflammation caused by hyperglycemia. In our diabetic mice, although the blood vessels were covered with pericytes in the retina, the inflammation, Ang2 release, vascular instability, hypoxia, and VEGF-A–VEGFR2 signaling were induced. This demonstrates that simultaneous loss of endothelial cells and pericytes may be an important pathogenic event in the early stage of DR. Recent studies using optical coherence tomography angiography have demonstrated decreased vascular density and retinal structural changes in the retinas of patients with diabetes with no or mild nonproliferative retinopathy (39). These results are very similar to the findings observed in STZ mice in our study.

In DR, even in the absence of overt retinal vascular damage, destabilization of the vascular structure and preparation for angiogenesis may be in progress. Therefore, initiating treatment in the early stage of DR is very important. As illustrated in Fig. 8, our study showed that the protective effect of SGLT2i may be mainly due to the attenuation of inflammatory responses by microglia, the downregulation of Ang2, the attenuation of decreased vessel density, and the suppression of increased VEGFR expression in the VEGF-A–VEGFR system. Thus, it is plausible that SGLT2i protected the retina from vascular permeability and edema formation through these mechanisms. The findings in this study clearly suggested that SGLT2i may prevent early pathogenic mechanisms and, thereby, may have a potential role in preventing DR.

Several recent several animal studies have also demonstrated beneficial effects of SGLT2i on retinopathy in rodent models of diabetes (40,41). Tofogliflozin improved impaired retinal neurovascular coupling in db/db mice (42). Another report showed that empagliflozin may decrease the damaging effects of hyperglycemia by directly acting on SGLT2 because they confirmed the presence of SGLT2 in the retina using immunostaining (41). However, it is difficult to exclude the possibility that the beneficial effects were mediated by the glucose-lowering effect. In our study, SGLT2i (both luseogliflozin and ipragliflozin) had protective effects even at a very low dose (0.1 mg/kg/day) that did not affect blood glucose levels. Therefore, these effects were unlikely to be mediated by glycemic control. Recent reports have also reported the beneficial effects of low-dose SGLT2i with no glucose-lowering effect (43–45).

SGLT2 is specifically located in kidney proximal tubular epithelial cells, but in recent years, it has also been reported to be expressed in mesangial cells (46), podocytes (47), brain pericytes (44), and other cells. Additionally, the expression of SGLT2 may be upregulated under pathological conditions such as diabetes (48), high glucose levels (43), intracellular nuclear factor-κB activation (47), and cancer (49). The role of SGLT2 may not be the same under pathological and physiological conditions. Thus, these results suggest that SGLT2i may directly affect the pathogenic mechanisms of DR. One of the factors complicating the development of therapeutic drugs for DR is the presence of a blood-retinal barrier between circulating blood and the retina with tight junctions and limited transfer of molecules. SGLT2i may pass through the blood-retinal barrier. In this study, we observed SGLT2 around the retinal vascular plexus, external limiting membrane, and retinal pigment epithelium using immunostaining. We also confirmed the presence of SGLT2 in a part of Iba1-positive cells. These data suggested the possibilities that microglia and vascular plexus may be directly affected through SGLT2i, resulting in the protective effect for DR. In contrast, the SGLT2i empagliflozin was recently reported to exert anti-inflammatory effects in primary microglia via NHE-1 (25). Microglial activation was reported to depend on NHE-1 mediated H⁺ homeostasis (50). Of interest, it was also reported that some SGLT2i directly binds to the Na⁺-binding site of NHE-1 and reduces intracellular Na⁺ and intracellular Ca²⁺ concentrations in cardiomyocytes (24).

In our study, we also indicated that luseogliflozin and ipragliflozin both inhibited the LPS-induced expression of proinflammatory mediators in primary microglia along with morphological changes from an amoeboid form to a ramified form. Additionally, both SGLT2i displayed high binding affinity to the Na⁺-binding site of NHE-1 (Fig. 7E), suggesting that both drugs might directly attenuate microglia activation in the diabetic retina through inhibition of NHE-1. Thus, SGLT2i may have the protective effects on DR through the direct effects on SGLT2 (on-target) and/or on NHE-1 (off-target). In the future, the pathophysiological role of SGLT2 and NHE-1 in the retina and the more detailed mechanism underlying the effect of SGLT2i should be clarified.

At present, there is no specific treatment for early-stage retinopathy in which there are preclinical latent lesions and mild nonproliferative lesions with preserved visual acuity. Our study suggested that SGLT2i may be effective in the prevention and early treatment of DR, based on their mechanism of action. In addition, since SGLT2i can be administered orally, SGLT2i may have significantly fewer adverse effects compared with the antibodies that must be injected directly into the eye. In particular, if the beneficial effects of low-dose SGLT2i that do not affect blood glucose levels are translatable to humans, SGLT2i can be used more safely as drugs specific to DR. Additionally, the combination of antibodies and SGLT2i could be more effective and useful in reducing the frequency of administration of antibodies, thus resulting in the reduction of adverse effects. Well-designed randomized prospective trials are needed in the near future to establish the beneficial effect of SGLT2i against DR.

Acknowledgments. Luseogliflozin was kindly provided by Taisho Pharmaceutical Co., Ltd., and ipragliflozin was kindly provided by Astellas Pharma Inc. The authors thank Yuki Honda for assistance with mRNA analyses. The authors thank the technical support provided by the Graduate School of Medical Sciences Research Support Center at Kyushu University and thank Editage (www.editage.com) for English language editing.

Funding. This work was supported in part by an AMED Core Research for Evolutionary Medical Science and Technology (AMED-CREST) grant (JP22gm0910013 to K.Y.), Japan Society for the Promotion of Science Grants-in-Aid for Scientific Research (JSPS KAKENHI) (23H05481, 22H05572, and 20H00493 to K.Y.), and Takeda Science Foundation to K.Y. This work was also supported by the Platform Project for Supporting Drug Discovery and Life Science Research from the AMED.

The study funders were not involved in the design of the study, the collection, analysis, and interpretation of data, writing the report, and did not impose any restrictions regarding the publication of the report.

Duality of Interest. This work was partly supported by research funds from Taisho Pharmaceutical Co., Ltd., and Astellas Pharma Inc. T.I. is an inventor of a patent related to this work (patent no. US10821126B2, JP6563376B2). No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.Y. and N.K. performed the experiments. M.Y. and N.K. analyzed the data. M.Y., K.Y., and T.I. revised the manuscript. M.Y. and T.I. designed the study. M.Y. and T.I. wrote the manuscript. K.Y. provided the critical reagents. All of the authors approved the manuscript. M.Y. 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.