n-3 Fatty Acid and Its Metabolite 18-HEPE Ameliorate Retinal Neuronal Cell Dysfunction by Enhancing Müller BDNF in Diabetic Retinopathy

Diabetic retinopathy (DR) is a potentially blinding complication of diabetes and a significant cause of global vision impairment (1). DR is diagnosed by onset of clinical features, which are easily determined using fundus photography; however, DR-induced functional defects often precede these signs (2,3). Electroretinography (ERG) is widely used to assess retinal function not only in humans but also in animals and is sensitive enough to diagnose the early-stage DR. In particular, ERG can reveal neuroretinal abnormalities in diabetic model rodents and patients with diabetes without gross changes by showing the reduced amplitudes of oscillatory potentials (OPs) and reduced b-wave/a-wave (b/a-wave) ratios (4–6).

The importance of detecting neuroretinal dysfunction is supported by the concept of neurovascular unit impairment, which is characterized by the neuronal, glial, and vascular cells mutually influencing each other to cause subclinical changes during the early stages of DR (7,8). Previous studies have shown that high glucose levels induce the generation of reactive oxygen species (ROS) in vascular endothelial cells, and ROS promote Müller glial and neuronal cell deaths (9,10). The corroboration of these studies introduced the concept of high cross talk of high glucose–ROS–neuronal cell damage. This concept has gradually become more widespread due to the insufficiency of the present DR treatments. Although there are several therapies in this regard, including laser photocoagulation and intravitreal injections of anti–vascular endothelial growth factor (VEGF) drugs, and although the effectiveness of vitrectomy surgery is improving, the final visual outcomes for patients with DR remain disappointingly modest (11). Therefore, early treatment strategies focusing on neuroretinal protection can have substantial advantages in the prevention of public health disasters due to DR.

Brain-derived neurotrophic factor (BDNF) has a reportedly powerful neuroprotective effect by binding to its receptor TrkB, which amacrine cells and retinal ganglion cells express, and activates extracellular signal–regulated kinase and phosphatidylinositol-3 kinase pathways (12,13). In the retina, glial cells provide structural and metabolic support for retinal neurons and blood vessels, and interestingly, Müller glia cells are widely known as a source of BDNF secretion (14). Even in the early stage of DR, Müller glia cell impairment and apoptosis reportedly occur, followed by the dysregulation of neurotrophic factors, including BDNF, which leads to neurodegeneration in DR (15,16). However, little is known about how, using the existing administration tools, BDNF expression in DR can be successfully increased. Interestingly, oral administration of n-3 fatty acids, mainly docosahexaenoic acid and eicosapentaenoic acid (EPA), was recently reported to reverse the downregulation of BDNF expression in the obese rat hypothalamus (17). In addition, the previous study, which used fat-1 transgenic mice that endogenously synthesize n-3 fatty acids, showed that they had less Purkinje cell degeneration from BDNF-mediated autophagy under streptozocin (STZ)–induced diabetic conditions (18). In the brain, n-3 fatty acid plays protective roles in neurodegenerative disorders, such as Alzheimer disease and Parkinson disease (19,20). The retina has fatty acid–rich photoreceptors and, similar to the brain, is a complex of neuronal, glial, and vascular cells. Therefore, it is possible that, by applying a similar concept of neurovascular unit impairment for brain impairment and neuroretinal dysfunction during early DR, understanding the BDNF function in DR and developing methods to increase BDNF in DR can provide strong tools for the treatment of DR neurodegeneration.

Based on these assessments, we hypothesized that neuroretinal dysfunction detected by the reduced OPs in the early stage of DR is initiated by the downregulation of BDNF expression in Müller glia cells that was triggered by the exposure of ROS from retinal vascular endothelial cells, which could induce secondary impairment of neuroretinal cells, including amacrine cells. We also examined whether dietary n-3 fatty acids, readily available for a therapeutic discipline, can attenuate cellular damage in DR. In addition, we investigated whether n-3 fatty acid or its metabolites play a pivotal role in DR using mass spectrometry.

We collected vitreous samples from patients with macular hole or epiretinal membrane, as control subjects, and DR who received vitrectomy using a vitrectomy cutter before starting infusion to keep them dry and immediately stored them at −80°C. The control subjects and patients with DR had no vitreous hemorrhages. This study was conducted following the guidelines of the Declaration of Helsinki; our protocol was registered with the Clinical Trial Registry of the University Hospital Medical Information Network (registered number UMIN000024553) and was approved by the Nagoya University Hospital Ethics Review Board (Nagoya, Japan). We obtained written informed consent from all participating patients.

Eight-week-old female Sprague Dawley rats (CLEA Japan, Inc.) were used for these studies and divided into three categories: 1) control group without STZ injections and fed standard chow (CE-2; CLEA Japan, Inc.) without fishmeal but with sunflower oil (5%), 2) STZ group with intraperitoneal (i.p.) injections of STZ (65 mg/kg) (Sigma-Aldrich) and fed the same chow as the control group, and 3) EPA group with i.p. injections of STZ (65 mg/kg) (Sigma-Aldrich) and fed the standard chow without fishmeal but supplemented with 5% EPA ethyl ester (EPA-E) (Mochida Pharmaceutical Co., Ltd.) (Fig. 2A). The standard chow and the chow with EPA for one rat per day include 74 kcal and 77 kcal, respectively. All diets were stored at −80°C and thawed at room temperature before use every day. To exclude the effect of n-3 fatty acid in the food, we prepared CE-2–based chow without fishmeal and added sunflower oil 2 weeks before STZ induction (Figs. 2A and 4H). One week after STZ injection, STZ-treated rats with blood glucose levels <350 mg/dL were excluded. The concentration of β-hydroxybutyrate (β-HB) in the plasma was measured according to the manufacturer’s instructions (Cayman Chemical). All studies adhered to the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research and were conducted in accordance with the experimental protocol approved by the Nagoya University Animal Care and Use Committee.

The procedures we used in this study were similar to those described in detail in an earlier article (21). Briefly, rats were dark-adapted for at least 2 h and prepared under dim red illumination. Following anesthesia of the rats with an intramuscular injection of ketamine and xylazine, the pupils were dilated with topical 0.5% tropicamide/0.5% phenylephrine (Santen Pharmaceutical Co., Ltd.), and the animals were placed on a heating pad. Hydroxyethyl cellulose was used to hydrate the cornea and conjunctiva for better electrical conductivity of the electrodes. ERGs were recorded with a gold wire loop electrode on the cornea and gold wire reference electrode on the sclera. Responses were amplified and bandpass filtered between 0.3 and 1,000 Hz. The ERGs were averaged using a computer-assisted signal averaging system (Power Lab). A Ganzfeld bowl (LACE Elettronica SRL) with a xenon source was used for stimulation. The amplitudes of the a- and b-waves and OPs were measured.

Lipid mediators in each sample were quantified by high-performance liquid chromatography–tandem mass spectrometry, as previously described (22,23). Eyes were collected, and retinal choroidal tissue was separated, frozen immediately in liquid nitrogen, and stored at −80°C. Samples of retinal choroidal tissue were homogenized with an Automill (Tokken Inc., Chiba, Japan), and the lipids were extracted with 1 mL of methanol at −20°C. After overnight incubation, the samples were centrifuged at 5,000g for 5 min at 4°C. The supernatants were mixed with four volumes of water containing 0.1% formic acid with deuterium-labeled internal standards (Cayman Chemical). Diluted samples were loaded onto Oasis HLB cartridges (Waters, Milford, MA) and washed with 0.1% formic acid, 15% methanol containing 0.1% formic acid, and petroleum ether containing 0.1% formic acid. The samples were eluted with 200 µL of methanol containing 0.1% formic acid. The lipids were injected into a Prominence HPLC system (Shimadzu, Kyoto, Japan) and a TSQ Quantum Ultra triple-stage quadrupole mass spectrometer equipped with an H-ESI interface (Thermo Fisher Scientific, Waltham, MA). Mobile phases A and B comprised water and acetonitrile:formic acid (100:0.1 v/v), respectively. The gradient conditions were as follows: 0–7 min, 37% B; 7–19 min, 37–90% B (linear gradient); 19–21 min, 100% B; 21–22.5 min, 37% B; and flow rate, 0.12 mL/min. The separation column was a Capcell Pak C18 MGS3 (1 × 100 mm; Shiseido Co., Ltd., Tokyo, Japan) set at 45°C, and the trapping column was an OPTI-GUARD Mini C18 (1 × 15 mm; Optimize Technologies, Oregon City, OR). Lipids were analyzed in a negative-ion polarity mode under the following conditions: spray voltage, 2,500 V; capillary temperature, 225°C; vaporizer temperature, 250°C; sheath gas (N₂) pressure, 40 psi; ion sweep gas pressure, 0 psi; auxiliary gas pressure, 10 psi; collision gas (Ar) pressure, 1.5 mTorr; and the EZ Method (scheduled selected reaction monitoring method) with a cycle time of 1.0 s. For accurate quantification, calibration curves were generated for each target eicosanoid using known reference standards and the isotope-labeled internal standards. Data analysis and quantitative calculations were performed using Xcalibur 2.2 software (Thermo Fisher Scientific).

All of the schedules in this experiment are shown in Fig. 4H. Following an intramuscular injection of ketamine and xylazine, 18-hydroxyeicosapentaenoic acid (18-HEPE) (Cayman Chemical) (containing equal amounts of 18[S]-HEPE and 18[R]-HEPE; 10 ng/2 µL) in PBS was injected into the vitreous space of one eye, and an equal volume of PBS was injected into the other eye as a control. On the same day, a single i.p. injection of STZ (65 mg/kg) (Sigma-Aldrich) was also administered, and after 1 week, diabetes was confirmed using the same criterion for blood glucose level as mentioned above. Intravitreal injections of 18-HEPE or a control vehicle were administered on days 0 and 14. On day 28, ERG was measured, and the retinas were dissected.

After being fixed in 4% paraformaldehyde and cryoprotected in 30% sucrose, the eye cups were incubated with rabbit antibody against BDNF (EPR1292, 1:500; Abcam) and mouse antibody against glutamine synthetase (MAB302, 1:500; Millipore), followed by secondary antibodies with Alexa Fluor 488 and 594 conjugates (1:1,000 dilution; Thermo Fisher Scientific) and Hoechst 333421 (1:5,000 dilution; Invitrogen). Sections were analyzed using a fluorescence microscope (BZ-9000; Keyence Corporation of America).

Human retinal endothelial cells (HRECs) from Cell Systems (Kirkland, WA) were cultured with EGM-2 medium (Lonza) as previously described (24). HRECs were incubated with or without VEGF and with different doses of glucose (25 and 5.6 mmol/L) after the adjustment of osmotic pressure as previously described (25). Cultured MIO-M1, from a human Müller glia cell line, and human retinal pigment epithelial (hRPE) cells were purchased from E-lucid (University College London) and ATCC (Gaithersburg, MD), respectively. MIO-M1 and hRPE cells were maintained in DMEM (Invitrogen) supplemented with 100 units/mL penicillin, 100 µg/mL streptomycin, 250 ng/mL amphotericin B, and 10% FBS (Sigma-Aldrich). Neuronal cell line PC12D cells (26,27) were cultured in DMEM supplemented with 10% FBS and 10% horse serum (Medical & Biological Laboratories Co., Ltd.) and then incubated for 2 days with 50 ng/mL of nerve growth factor (Millipore) for differentiation into neurons. The high- and low-glucose DMEM contained 25 and 5.6 mmol/L glucose, respectively. We applied hydrogen peroxide (H₂O₂) (Sigma-Aldrich) as oxidative stress. Recombinant BDNF (Abcam) resuspended in DMEM was incubated for 24 h with PC12D, HRECs, and hRPE cells 1 h after H₂O₂ administration, and we evaluated the cell viability. We exposed MIO-M1 to H₂O₂ for 24 h when we evaluated cell viability and BDNF mRNA expression levels, respectively, and we exposed the cells to H₂O₂ for 72 h when we prepared the sample for ELISA. We measured the relative changes in the BDNF mRNA expression levels in MIO-M1 under oxidative stress (200 μmol/L H₂O₂) following the administration of 18-HEPE, 15-HEPE, 12-HEPE, and 5-HEPE (Cayman Chemical) for 48 h to evaluate the effects of the lipid mediators. Additionally, we evaluated the expression levels of BDNF mRNA following the administration of 15-hydroxyeicosatetraenoic acid (15-HETE) and 12-HETE (Cayman Chemical) for 48 h.

For protein collection, rat retina and cultured cells were lysed in a RIPA buffer (Sigma-Aldrich) with a protease inhibitor cocktail (Roche). The lysate was centrifuged at 15,000 rpm for 15 min at 4°C, and the supernatant was collected. Protein concentrations were determined using a Bradford Assay Kit (Bio-Rad Laboratories). For quantitative RT-PCR, total RNA was purified using a Qiagen RNeasy Mini Kit (Qiagen). RNA concentration and quality were assessed using the NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific).

MIO-M1 cells were cultured in 24-well culture plates at a concentration of 3.0 × 10⁴/well in 500 μL of DMEM exposed to different concentrations of H₂O₂ with or without 18-HEPE for 72 h. In addition, MIO-M1 cells were cultured in six-well culture plates at a concentration of 6.0 × 10⁴/well in 2 mL of DMEM exposed to different concentrations of glucose for 7 days. The culture supernatant was used according to the manufacturer’s protocol. Protein lysates were prepared from the rat retinochoroidal tissue, and BDNF protein level was measured using an ELISA kit (Human Free BDNF Quantikine ELISA Kit from R&D Systems for human BDNF; and Human BDNF ELISA Kit from Abcam for rat BDNF). Total RNA was reverse transcribed using a Transcriptor Universal cDNA Master (Roche), and RT-PCR was performed using the Thunderbird Probe qPCR Master Mix (Toyobo Co., Ltd., Osaka, Japan) and a Gene Expression Assay containing primers and a FAM dye-labeled TaqMan probe to detect human BDNF (Hs02718934_s1; Applied Biosystems, Foster City, CA), rat BDNF (Rn02531967_s1; Applied Biosystems), and eukaryotic 18S rRNA (Hs99999901_s1; Applied Biosystems). PCR cycles consisted of a predenaturation step at 95°C for 2 min, followed by 40 cycles of denaturing steps at 95°C for 15 s, and annealing and extending steps at 60°C for 60 s. The relative expressions of the target genes were determined by the 2^−ΔΔCt method.

H₂O₂ in the human vitreous samples and extracellular concentration of H₂O₂ from MIO-M1 and from HRECs cultured with high and low glucose for 72 h were measured using the OxiSelect Hydrogen Peroxide/Peroxidase Assay Kit (Cell Biolabs, Inc.) according to the manufacturer’s instructions. Oxidative stress in the retina and choroid of rats and the human vitreous sample was evaluated by quantifying the levels of 8-hydroxy-2′-deoxyguanosine (8-OHdG) as oxidatively modified DNA. Total DNA was purified from the retina and choroid using the Qiagen DNeasy Blood & Tissue Kit (Qiagen), and purified DNA was digested with an 8-OHdG Assay Preparation Reagent Set (FUJIFILM Wako Pure Chemical Corporation) according to the manufacturer’s instructions. 8-OHdG levels were measured using an ELISA kit from Oxis Research.

The cultured cells were treated with different doses of H₂O₂, recombinant BDNF (Abcam), and 18-HEPE for 24 h. MIO-M1 cells were treated with different doses of glucose for 3 and 7 days. Cell viability was determined by using the Cell Proliferation Kit I (Roche) according to the manufacturer’s instructions. Additionally, we evaluated the effect of BDNF on PC12D cells to assess the neurite outgrowth of PC12D using the IncuCyte ZOOM real-time imaging system (Essen BioScience Inc.). We administered 100 ng/mL of recombinant BDNF prior to imaging (BDNF-treated group). Phase-contrast images were taken at ×20 magnification 3 and 24 h later. After imaging, IncuCyte NeuroTrack Software (Essen BioScience Inc.) was used to determine the length of the neurite outgrowth (28).

Data are presented as mean ± SD. For comparisons among more than two groups, initial statistically significant results were further analyzed using the Scheffe test, and the Steel-Dwass test was used when significant differences were found. For comparisons between two groups, the Mann-Whitney U test or Wilcoxon signed-rank test was applied. A value of P < 0.05 was considered statistically significant.

The data sets generated and/or analyzed during the current study are available in figshare (submitted by H.K., “n-3 Fatty Acid and Its Metabolite 18-HEPE Ameliorate Retinal Neuronal Cell Dysfunction by Enhancing Müller BDNF in Diabetic Retinopathy,” posted 10 November 2019, at https://doi.org/10.6084/m9.figshare.9973616.v1). No applicable resources were generated or analyzed during the current study.

The levels of 8-OHdG and H₂O₂ significantly increased in human vitreous samples of the eyes of patients with DR compared with those from the eyes of patients with macular holes and epiretinal membranes (Fig. 1A and B). After confirming the presence of ROS in DR, we verified the influence of high glucose on HRECs and MIO-M1. The extracellular concentration of H₂O₂ levels from HRECs in 25 mmol/L glucose without VEGF was not significantly higher than that in 5.6 mmol/L glucose without VEGF (Supplementary Fig. 1). In contrast, the extracellular concentration of H₂O₂ levels from HRECs in 25 mmol/L glucose containing VEGF significantly increased compared with that in 5.6 mmol/L glucose with VEGF (Fig. 1C). In contrast, 25 mmol/L glucose did not increase the extracellular concentration of H₂O₂ from MIO-M1 but significantly elevated the cell viability of MIO-M1 after exposure for 72 h compared with 5.6 mmol/L glucose (Fig. 1D and E). Based on these findings and previous studies that have shown that hyperglycemia itself directly induced neuroretinal damage (29,30), we assumed that ROS produced in HRECs under high-glucose conditions was another trigger for diabetic neuroretinal damage. Therefore, in this study, we focused on oxidative stress and provided medium containing 25 mmol/L glucose within 72 h for all of the experiments, and oxidative stress was induced to mimic the condition of early DR. In MIO-M1 cells, H₂O₂ significantly reduced cell viability and BDNF mRNA levels compared with that in the control as well as BDNF protein level in the culture medium compared with that in the control (Fig. 1F–H). Similarly, H₂O₂ significantly reduced the PC12D cell viability in a dose-dependent manner (Fig. 1I). Moreover, the administration of recombinant BDNF statistically ameliorated the reduction of cell viability in H₂O₂-treated PC12D cells dose-dependently and the decrease rate of the relative change in the neurite length per cell body cluster from the beginning in PC12D cells time-dependently (Fig. 1J and K).

Based on previous results indicating that n-3 fatty acid increased BDNF in other tissues (17,31), we investigated the effect of EPA-E on BDNF expression in the eye and on ERG in STZ-induced diabetic rats. STZ group showed significant differences in body weight, blood glucose, and β-HB compared with the control group at 2 and 8 weeks after STZ injection, but the STZ and EPA groups did not show significant differences in all of the three parameters, including β-HB, which is widely used as a diabetic ketoacidosis marker (32) (Table 1 and Supplementary Table 1). These findings indicated that treatment with EPA-E did not affect the metabolic variables or ketosis. To evaluate the neuroprotective effects of EPA on DR, full-field ERG was applied. We first examined when a significant change occurred in the OPs of ERG after STZ injection and confirmed that it occurred 4 weeks after compared with the control (Fig. 2B). Thereafter, we evaluated the effect of oral administration of EPA-E for 8 weeks on ERG. The b/a-wave ratio of the ERG in the STZ group was significantly lower than that of the control group, and there was no significant difference between the b/a-wave ratio in the STZ and the EPA group (Fig. 2C). The OPs of ERG in the STZ group were significantly lower than those of the control group, but those in the EPA group were significantly higher than those in the STZ group (Fig. 2D). In summary, the decrease in b/a-wave ratio and OP amplitudes in the STZ group suggested a decreased neuroretinal function in DR (27), and significant improvement of the OPs in the EPA group indicated that oral EPA-E administration ameliorated the decreased neuroretinal function in DR. Immunohistochemistry showed that glutamine synthetase and BDNF were colocalized in the inner retina (Fig. 2F), suggesting that BDNF is produced from Müller cells, ganglion cells, and astrocytes in the rat retina. Furthermore, 8-OHdG showed a significant increase in the STZ group compared with that in the control group and a significant decrease in the EPA group, indicating that oxidative stress was statistically suppressed by EPA-E (Fig. 2G). In addition, the Bdnf mRNA level and the relative BDNF protein level in the retina of the STZ group were significantly lower than those of the control and the EPA group (Fig. 2H and I). These results suggested that oral EPA-E administration reduced ROS and ameliorated BDNF reduction in the retina, resulting in the recovery of neuroretinal cell damage after ROS exposure in the early stage of DR.

To further investigate the protective mechanisms of EPA in DR, we examined the lipid mediators in rat retinal tissue. The possible metabolites made from EPA, 18-HEPE, 15-HEPE, 12-HEPE, and 5-HEPE were significantly detected in the retinal tissue of EPA-E–fed rats (Fig. 3). In contrast, 12-HETE was significantly lower in the EPA group than in the control and STZ groups. The raw data for each subject are shown in Supplementary Table 2. Furthermore, we performed single-cell culture of MIO-M1 to confirm that Müller cells is one of the sources of BDNF. Relative BDNF mRNA level and BDNF protein level in MIO-M1 cells treated with H₂O₂ were increased by 2 μmol/L 18-HEPE but not by 0.5 μmol/L 18-HEPE compared with the control (Fig. 4A and B). In contrast, relative BDNF mRNA was not significantly increased by 2 μmol/L 15-HEPE, 2 μmol/L 12-HEPE, or 2 μmol/L 5-HEPE compared with the controls (Fig. 4C). Meanwhile, the relative BDNF mRNA level in MIO-M1 cells exposed to 2 μmol/L 15-HETE and that exposed to 2 μmol/L 12-HETE were significantly lower than those of the controls (Fig. 4D). Additionally, H₂O₂ significantly reduced the cell viabilities of all PC12D, HRECs, and hRPE cells, but 18-HEPE did not inhibit these reductions (Fig. 4E–G). These observations confirmed that 18-HEPE played the most prominent role in BDNF increase in Müller glia cells but did not affect the other ocular cell types. Moreover, OPs and the Bdnf mRNA level in the retina of STZ rat eyes with 18-HEPE intravitreal injection were significantly higher than those with PBS (Sham) (Fig. 4J and K). These corroborating data suggested that 18-HEPE specifically induced BDNF expression in the STZ rat retina and that resulted in OP recovery in ERG, as was seen after EPA-E administration in vivo.

It is already revealed that the oral administration of n-3 fatty acids, including EPA, has anti-inflammation, antiangiogenesis, and antioxidation effects in DR (33). However, only a few studies focused on the neuroprotective effect of EPA (34). BDNF is one of the key factors in neuroprotection, and we showed the protective effect of BDNF on PC12D. Our study is the first to demonstrate that EPA has significant neuroprotective effects; specifically, it rescued amacrine cell function, which can be evaluated by OPs in ERG in DR by enhancing BDNF expression in Müller glia cells. We further found that some EPA metabolites, including 18-HEPE, were highly produced in the retina of rats administered EPA-E, and 18-HEPE specifically induced BDNF expression in Müller glia cells. The novel findings in our study were as follows: 1) Although previous studies showed a BDNF increase in other tissues following the administration of n-3 fatty acid (17,31), the concurrent increase in BDNF in the retina has not been reported to date. 2) Although EPA metabolites in the eyes have been analyzed already (35), we detected the increase in 18-HEPE in eyes with EPA treatment for the first time. 3) Although the other HEPEs also increased in the retina after EPA administration, only 18-HEPE increased BDNF mRNA in Müller glia cells.

Oxidative stress plays an important role in DR (10). Indeed, we confirmed that H₂O₂, for which production was elevated in DR and under high-glucose conditions in an in vitro study, reduced BDNF secretions in Müller glia cells and cell viability in both neuroretinal and Müller glia cells. However, EPA-E intake reduced oxidative stress in DR, which contributed to BDNF recovery, resulting in the amelioration of neuroretinal function. To the best of our knowledge, this finding has not been reported to date. STZ-induced diabetic models in rodents reportedly exhibited differences between the control and STZ groups with regard to their general condition (e.g., body weight) (36). In our study, the β-HB levels in the STZ and EPA groups were significantly higher than those in the control group, suggesting that STZ injection must be a trigger of ketoacidosis. In contrast, it is already known that the quantitative measurement of β-HB could be useful to diagnose diabetic ketoacidosis, and serum β-HB levels of ≥3.0 mmol/L were one of the criteria for the diagnosis in previous studies (37–39). In addition, a study defined serum β-HB levels of ≥1.0 mmol/L as a ketosis model (40). Therefore, in our study, we considered that the animals developed ketosis instead of ketoacidosis. Nevertheless, we did not find significant systemic improvement in STZ-induced rats by EPA-E, suggesting that EPA-induced retinal recovery was not only due to an improvement in the general condition. Nevertheless, it was unclear whether the EPA-induced retinal recovery was due to other specific factors in addition to the reduction in oxidative stress in DR. To answer this question, we further investigated which EPA metabolites were increased in DR and specifically induced BDNF expression in the retina. For the first time in our study, we showed that an increase in 18-HEPE particularly contributed to an increase in the Müller glia cell–derived BDNF, both in vivo and in vitro. Recent studies have demonstrated that EPA is metabolized by vascular endothelial cells, and macrophages introduced from bone marrow may locally produce 18-HEPE (41,42). Consistent with the previous study, which demonstrated that vascular endothelial cell injury in diabetic mice was not improved by oral administration of n-3 fatty acids (43), we confirmed that 18-HEPE did not improve a decrease in HREC viability by oxidative stress as well as PC12D and hRPE cells. However, 18-HEPE induced a BDNF increase in Müller glia cells in vitro. In contrast, the potential roles of 12/15-lipoxygenase (12/15-LOX) and LOX-derived eicosanoids in DR have been documented in several articles, and it has been reported that the treatment of HRECs with 12-HETE and 15-HETE increased ROS generation (44,45). In contrast to the current study, other studies showed a significant increase in 12/15-LOX and its metabolites in patients and mice with DR. Indeed, 8-OHdG was improved by EPA-E administration in the current study. It is possible that the oral administration of EPA-E did not only increase 18-HEPE and Müller glia BDNF but also promoted a change in the activity of 12/15-LOX from metabolizing arachidonic acid to EPA (46,47) and further reduction in 12-HETE, resulting in the protection of Müller glia BDNF.

The other questions and the limitations of the study are as follows: 1) The precise mechanisms of n-3 fatty acids, including their metabolites, remain unclear, and specific receptors for 18-HEPE have not been identified. Resolvin E1, an 18-HEPE metabolite, reportedly binds to the receptor ChemR23 and plays a pivotal role in suppressing neovascularization and inflammation (41). However, we did not detect resolvin E1 in our retinal samples. Nevertheless, it is unclear whether the metabolite was present in undetectable quantities due to the limited sample volume of the rat retina. Further analysis is necessary to determine the precise mechanism involving lipid mediators in DR. 2) Although OP reduction was reportedly observed 6 weeks after STZ administration (48), we noted OP reduction 4 weeks after STZ administration. This discrepancy may be due to the following reasons: i) The previous study was performed with 60 mg/kg STZ, whereas we used 65 mg/kg STZ to induce the diabetic condition, and the higher STZ dose resulted in faster ERG change. ii) We created a special chow by removing fishmeal and adding sunflower oil and started using chow 2 weeks before STZ induction. The effects of EPA reportedly began 2 weeks after administration (49). Sunflower oil contains large amounts of n-6 fatty acid, whereas CE-2, the basal chow that we modified for this study, contains fishmeal, which is rich in n-3 fatty acid. Therefore, our STZ rats had a large amount of n-6 fatty acid but no n-3 fatty acid for 6 weeks before ERG measurement. The difference in the amount of n-6 and n-3 fatty acids in the food could affect the time of OP change. iii) In our study, the food was not available ad libitum, and the total calories were controlled. The relationship between total calories and diabetes prevalence has been reported (50,51), and the difference in calorie control might affect the difference in the time of OP reduction. 3) Although BDNF was reportedly detected in human vitreous fluid (52), we did not detect BDNF in the vitreous of patients with DR or that of the control subjects (data not shown), as previous studies using ELISA presented (53,54). It is not ethically possible to collect human retina samples during surgery. Therefore, to confirm BDNF expression in the diabetic eye, we measured the increase in BDNF in the eyes of rats in our study. 4) Various studies have reported that high glucose concentrations do not cause oxidative stress in HRECs when they are pure and properly cultured (55). However, we detected a slight but significant increase in ROS production from the HRECs in a high glucose concentration compared with those in a lower concentration. This was consistent with the findings of several previous studies (56–58). Two reasons might explain this divergence in findings among studies. First, the previous studies used primary HRECs obtained from donors, whereas we used HRECs from a manufacturer. Supplementation with VEGF may also have affected the ROS production; there was no increase in the extracellular concentration of H₂O₂ from HRECs in the medium without VEGF, even with 25 mmol/L glucose. VEGF can promote the production of ROS by HRECs (59). Nevertheless, we considered it reasonable to use supplementation with VEGF to mimic the intraocular condition of retinal damage from diabetes. A second potential reason for the divergence among studies may have been the difference in the method used to detect ROS. We used OxiSelect Hydrogen Peroxide/Peroxidase Assay Kits (Cell Biolabs, Inc.) to detect extracellular ROS, whereas most of the previous studies used 2′7′-dichlorodihydrofluorescin diacetate to detect intracellular ROS. 5) Many studies have reported high glucose–induced apoptosis in Müller cells, but a further distinction among types of cell death has been proposed. Many studies have reported that Müller cells die in a hyperglycemic environment via pyroptotic cell death (60). In the current study, MTT assay showed that high glucose exposure for 3 days increased MIO-M1 cell viability, whereas exposure for 7 days significantly reduced cell viability, consistent with the reduced secretion of BDNF in MIO-M1 (Supplementary Fig. 2). Thus, the decrease in BDNF production by Müller cells in DR may be due to oxidative stress caused by hyperglycemia and also due to hyperglycemia. Further studies are needed to clarify the mechanism underlying the damage to Müller cells in a diabetic environment.

BDNF is reported to be upregulated by antioxidants (61). According to the recent study that indicated that BDNF is expressed not only in Müller glia cells but also in retinal ganglion cells, consistent with our immunohistochemistry result, BDNF expression from both the retinal ganglion and the retinal neuronal cells may be reduced under diabetic conditions (14). Therefore, further studies are required to elucidate EPA involvement with these other cells in DR. Although previous studies have shown the neuroprotective role of BDNF in the eyes, they used subretinal transplantation of the transfected cells that can express BDNF or BDNF intravitreal injection, and, thus, many problems need to be overcome for clinical applications (14,62). However, our study successfully suggests a very safe and effective drug delivery method: straightforward oral administration of EPA-E is effective for neuroprotection at early-stage DR.

Acknowledgments. The authors thank Drs. Shu Kachi (Miyake Eye Hospital) and Tadasu Sugita (Sugita Eye Hospital) for important scientific suggestions and Reona Kimoto, Makiko Kato, and Shiang-Jyi Hwang (Nagoya Univeristy) for technical assistance. The authors also thank Enago (www.enago.jp) for the English language review.

Funding. This work was partially supported by the following Japan Society for the Promotion of Science KAKENHI grants: Grants-in-Aid for Young Scientists (B) (17K16963 and 19K09988 to H.K. and 18K16922 to N.N.), Grants-in-Aid for Scientific Research (C) (19K07357 to T.O.), Grants-in-Aid for Scientific Research on Innovative Areas (15H05904 to T.Y.), and Grants-in-Aid for Scientific Research (B) (18H02627 to T.Y.). This work was also partially supported by the Takeda Science Foundation and the Eye Research Foundation for the Aged.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.S., H.K., and F.Y. were responsible for conceptualization. A.S., H.K., T.O., and T.H. were responsible for methodology. A.S., H.S., and K.Y. were responsible for software. K.T., N.N., and K.K. were responsible for validation. A.S., H.K., T.O., Y.O., and T.Y. conducted formal analysis. A.S., H.K., Y.F., M.N., S.I., T.O., T.H., and R.N. were responsible for investigation. A.S., S.I., and F.Y. obtained resources. A.S., H.K., T.O., and Y.O. were responsible for data curation. A.S. and H.K. were responsible for writing (original draft preparation). A.S., H.K., T.O., Y.O., and T.Y. were responsible for writing (review and editing). H.K. was responsible for visualization. T.Y. and H.T. conducted supervision. H.K. handled project administration. H.K., T.O., N.N., and T.Y. were responsible for funding acquisition. H.K. 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.