Fibroblast Growth Factor 21 Protects Photoreceptor Function in Type 1 Diabetic Mice

Diabetic retinopathy (DR) is a common complication of diabetes. The incidence of DR worldwide is projected to increase to >190 million by 2030, and the number of individuals with vision-threatening proliferative DR will likely increase to >56 million (1). In DR, retinal neovascularization occurs late (proliferative DR), induced by earlier vessel loss (nonproliferative DR). Current treatments targeting retinal neovascularization have limitations. Clinical trials of anti–vascular endothelial growth factor (VEGF) drugs for DR suggests efficacy in some but not all patients. There are also safety concerns about the long-term anti-VEGF drug effects, including degeneration of normal blood vessels and choroid as well as degeneration of neural retina (2). New therapeutic approaches to prevent early retinal vessel loss to avoid the blinding end stage of DR would be of great benefit. Retinal neural electrophysiological dysfunction occurs early before vascular abnormalities in DR (3). Therefore, maintaining normal retinal function may help to prevent DR progression.

Increased oxidative stress from hyperglycemia-associated metabolic abnormalities is involved in the development of diabetic retinal microvascular changes (4). The retina is a highly metabolically demanding tissue, and photoreceptors have the highest number of mitochondria of any cell in the body (5). Of all retinal cells, photoreceptors contribute the most to diabetes-induced retinal oxidative stress and inflammation in mice (6). Superoxide production is increased in diabetic photoreceptors (7). It has been shown that photoreceptors can control retinal vascular development. In patients with diabetes who have proliferative retinopathy and later development of retinitis pigmentosa (a photoreceptor-degenerative disease), retinal neovascularization (DR) slowly regresses (8). Therefore, enhancing the antioxidant pathways in photoreceptors may prevent neurovascular damage in DR.

Fibroblast growth factor 21 (FGF21) improves metabolic homeostasis in diabetes (9). FGF21 is expressed in many tissues, although circulating FGF21 under physiological conditions is derived mainly from the liver (10). FGF21 levels in patients with type 1 diabetes are lower than in healthy control subjects (11,12). FGF21 has not been extensively studied in DR, although there is evidence in other systems to suggest that it could be important. In type 1 diabetic mice, FGF21 protects against diabetes-induced testicular apoptotic cell death and renal dysfunction (13,14) and prevents diabetic nephropathy (15) through activation of the nuclear factor erythroid 2–related factor 2 (NRF2) pathway. NRF2 upregulates antioxidant enzyme expression and is a critical defense mechanism against oxidative stress in the retina (16). We previously found that FGF21 decreases pathological retinal vessel proliferation and promotes physiological retinal vascularization in mouse oxygen-induced retinopathy (OIR), modeling late proliferative DR (17). However, there is limited knowledge of the role of FGF21 in early DR. We investigated whether FGF21 protection against early retinal neuronal dysfunction in diabetes in the following two mouse models of type 1 diabetes: 1) insulin-deficient Akita mice; and 2) streptozotocin (STZ)-induced diabetic mice. We found that the administration of a long-acting FGF21 analog, PF-05231023, improved photoreceptor dysfunction and morphology, as well as reduced retinal inflammation in vivo and in vitro.

All animal studies adhered to the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the Institutional Animal Care and Use Committee at Boston Children’s Hospital.

Ins2^Akita male mice have a spontaneous mutation in the insulin 2 gene, which leads to incorrect folding of insulin protein. Heterozygous Akita mice develop diabetes within 1 month and retinal complications at ∼6 months of age (18). The 7- to 8-month-old Akita mice were compared with littermate wild-type (WT) mice. Akita mice with retinal functional abnormalities were screened with electroretinography (ERG) and those with ERG changes were intraperitoneally injected with 10 mg/kg long-acting FGF21 analog PF-05231023 (Pfizer) or vehicle control twice a week for 4 weeks. Retinal function was again examined with ERG after 4 weeks of treatment, and body weight and blood glucose were recorded. Serum triglyceride levels were measured using the Wako L-Type TG M test.

The 6- to 8-week-old male WT and adiponectin (APN)-deficient (Apn^−/−) mice were starved 6 h prior to an intraperitoneal injection of 55 mg/kg/day STZ on days 1 and 2 followed by an injection of 60 mg/kg/day STZ from day 3 to 5. Diabetes was induced within 1 week of injection (18). The 7- to 8-month-old diabetic WT and Apn^−/− mice were screened with ERG. Mice with retinal functional abnormalities were intraperitoneally injected with 10 mg/kg PF-05231023 or vehicle control twice a week for 4 weeks. Retinal function was then re-examined with ERG.

ERG was used to assess the function of retinal neurons. Flash ERGs were obtained using an Espion e² (Diagnosys LLC, Lowell, MA) in dark-adapted, mydriatic (Cyclomydril; Alcon, Fort Worth, TX), anesthetized (ketamine/xylazine) subjects. Stimuli were “green” light-emitting diode flashes of doubling intensity from ∼0.0064 to ∼2.05 cd⋅s/m² and then “white” xenon-arc flashes from ∼8.2 to ∼1,050 cd⋅s/m² presented in an integrating sphere (Colordome; Diagnosys LLC). As shown (Fig. 2A), the saturating amplitude (Rm_P3) and sensitivity (S) of the rod photoreceptors was estimated from the a-waves elicited by the white flashes (19). The bipolar cell response amplitude and sensitivity (1/K_P2) of second-order neurons, principally bipolar cells (20), was measured by subtracting this model from the intact ERG waveform (21) and fitting the Naka-Rushton equation (22) to the response versus intensity relationship of the resulting waveform “P2.” The oscillatory potentials (OPs), which characterize activity in inner retinal cells distinct from the generators of the a- and b-waves (23), were filtered from P2 (24) and assessed in the frequency domain to determine their energy (25); the saturating energy (Em) and sensitivity (1/i_1/2) of the OPs were then assessed in a similar manner as those of P2. Finally, retinal sensitivity at threshold (Sm) was calculated by scaling the amplitude of each P2 by the intensity used to elicit it and fitting a generalized logistic growth curve, with the exponent set to negative unity, to the resulting sensitivities and determining the limit of this function as intensity approached zero (26).

Mice were anesthetized (ketamine/xylazine), and their pupils were dilated (Cyclomydril; Alcon). Spectral domain optical coherence tomography (OCT) with the guidance of a bright-field live fundus image was performed with the image-guided OCT system (Micron IV; Phoenix Research Laboratories). Photoreceptor inner and outer segment thicknesses were measured using Insight (provided by Micron IV). The thickness of photoreceptor segments was plotted at six distances (50, 100, 150, 200, 250, and 300 µm) from the optic nerve head both on the nasal and on the temporal side.

For immunohistochemistry on retinal cross sections, eyes were fixed in 4% paraformaldehyde, frozen in optimal cutting temperature compound (Tissue-Tek), and then cut into 10-μm sections and rinsed with PBS. The sections with optic nerve were treated with ice-cold methanol for 15 min and then 0.1% triton PBS for 45 min at room temperature. The sections were blocked with 3% BSA for 1 h at room temperature and stained with primary antibody against cone arrestin (1:500; catalog #AB15282; Millipore) and rhodopsin (1:500; MABN15; Millipore) overnight at 4°C. The sections were stained with corresponding secondary antibody, covered in mounting medium with DAPI (H-1200; Vector Laboratories), and visualized with a Leica SP2 Confocal Microscope or a Zeiss AxioObserver.Z1 Microscope at 200× magnification.

Fresh mouse eyes were embedded in optimal cutting temperature compound and sectioned at 12 μm in a cryostat, mounted on RNase-free polyethylene naphthalate glass slides (catalog #11505189; Leica). Frozen sections were fixed in 70% ethanol for 15 s, followed by 30 s in 80% ethanol and 30 s in absolute ethanol, and then washed with diethyl pyrocarbonate–treated water for 15 s. Laser dissection of retinal neuronal layers was performed immediately thereafter with the Leica LMD 6000 System (Leica Microsystems), and samples were collected directly into lysis buffer from the RNeasy Micro Kit (Qiagen, Chatsworth, CA).

Photoreceptor 661 W cells were cultured at 37°C, with 5% CO₂ in a humidified atmosphere in DMEM (catalog #1196502; Gibco) supplemented with 10% FBS (catalog #S12450; Atlanta Biologicals) and 1% antibiotic/antimycotic solution. An equal number of cells per well was plated on a six-well dish. Oxidative stress was induced with 0.5 mmol/L paraquat (PQ; catalog #856177; Sigma-Aldrich) for 1 h at 37°C. The culture media were changed, and the cells were treated with vehicle (PBS) or 500 ng/mL PF-05231023 for 24 h. Cells were collected for protein and RNA.

Retinas or 661 W cells were lysed with QIAzol lysis reagent and incubated on ice for 15 min. Chloroform 20% was added and incubated for 5 min at room temperature. RNA was extracted according to the manufacturer instructions using a PureLink RNA Mini Kit (catalog #12183018A; Ambion). RNA was then reverse transcribed using iScript cDNA Synthesis Kit (catalog #1708891; Bio-Rad). Quantitative PCR (qPCR) was performed for the following: Arrestin4: 5′-GAG CAA GGG CTG CTA CTC AAG-3′, 5′-AAC CGC AGG TTC AAG TAT TCC-3′; Rhodopsin: 5′-TCA TGG TCT TCG GAG GAT TCA C-3′, 5′-TCA CCT CCA AGT GTG GCA AAG-3′; IL-1β: 5′-TTC AGG CAG GCA GTA TCA CTC-3′, 5′-GAA GGT CCA CGG GAA AGA CAC-3′; IL-6: 5′-TAG TCC TTC CTA CCC CAA TTT CC-3′, 5′-TTG GTC CTT AGC CAC TCC TTC-3′; Vegfa F: 5′-GGA GAT CCT TCG AGG AGC ACT T-3′, R: 5′-GCG ATT TAG CAG CAG ATA TAA GAA-3′; Tnfα 5′-CAT CTT CTC AAA ATT CGA GTG ACA A-3′, 5′-TGG GAG TAG ACA AGG TAC AAC CC-3′; IL-10: 5′-CTT ACT GAC TGG CAT GAG GAT CA-3′, 5′-GCA GCT CTA GG AGC ATG TGG-3′; Apn: 5′-GAA GCC GCT TAT GTG TAT CGC-3′, 5′-GAA TGG GTA CAT TGG GAA CAG T-3′; Nrf2: 5′-TAG ATG ACC ATG AGT CGC TTGC-3′, 5′-GCC AAA CTT GCT CCA TGT CC-3′; and Nfκb: forward, 5′-GGA GAG TCT GAC TCT CCC TGA GAA-3′, reverse, 5′-CGA TGG GTT CCG TCT TGG T-3′. Quantitative analysis of gene expression was generated using an Applied Biosystems 7300 Sequence Detection System with the SYBR Green Master Mix Kit, and gene expression was calculated relative to cyclophilin A (5′-CAG ACG CCA CTG TCG CTT T-3′; 5′-TGT CTT TGG AAC TTT GTC TGC AA-3′) (retinas) or β-actin (5′-CGG TTC CGA TGC CCT GAG GCT CTT-3′, 5′-CGT CAC ACT TCA TGA TGG AAT TGA-3′) (photoreceptor 661 W cells) using the ΔΔCt method. Each sample was repeated in triplicate.

Five microliters of protein lysate from photoreceptor 661 W cells were used to detect the levels of NRF2 (1:500; MAB3925; R&D Systems), phospho (p)–nuclear factor-κB (NF-κB) (1:200; catalog #3037S; Cell Signaling Technology), NF-κB (1:1,000; catalog #3034; Cell Signaling Technology), p-AKT (1:500; catalog #9271; Cell Signaling Technology), AKT (1:1,000; catalog #4691; Cell Signaling Technology) in 5% BSA overnight at 4°C. Signals were detected using 1:5,000 corresponding horseradish peroxidase–conjugated secondary antibodies and enhanced chemiluminescence (Pierce), then the digital images were visualized with a Bio-Rad ChemiDoc Touch Imaging System. β-Actin (1:5,000; catalog #A1978; Sigma-Aldrich) was used as internal control.

All ERG data were presented as the log change from control (ΔLogNormal); by expressing the data in log values, changes in observations of fixed proportion become linear, consistent with a constant fraction for physiologically meaningful changes in parameter values (26). ΔLogNormal ERG data were plotted as the mean ± SEM and evaluated for significant effects using mixed-effects linear models (MLMs) (27). In each analysis, two MLMs were used. The first MLM, carried out on the saturating a-wave, P2, and OP parameters, had group (Akita vs. WT; or STZ vs. control), treatment (before vs. after PF-05231023 administration), parameter (amplitude vs. sensitivity), and retinal depth (photoreceptor vs. bipolar vs. inner retina). The second MLM, carried out on Sm, had factors group and treatment. Data from both eyes were included in all analyses. Differences in ERG parameters were detected by ANOVA followed by Tukey test. Two-tailed unpaired t test and ANOVA with Bonferroni multiple-comparison test were used for comparison of results as specified (Prism version 5.0; GraphPad Software, Inc., San Diego, CA). The threshold for statistical significance (α) was set at 0.05.

Circulating FGF21 levels (ELISA) were reduced in >7-month-old Akita mice compared with littermate WT control mice (Fig. 1A). Retinal expression of Fgf21 (qPCR) was not changed significantly (Fig. 1B). The physiological and pharmacological actions of FGF21 are dependent on the receptor FGFR1 and the coreceptor β-klotho (28,29). The gene expression of FGF21 receptor Fgfr1 was comparable in Akita versus WT mouse retinas and that of β-klotho was mildly increased (Fig. 1C and D) (28,29). Fgfr1 was expressed in retinal neurons isolated from retinal cross sections with laser-captured microdissection (Supplementary Fig. 1).

To examine whether FGF21 protects against retinal dysfunction in DR, we first administered PF-05231023 (10 mg/kg intraperitoneally, twice a week for a month) or vehicle control to those Akita mice that had ERG deficits at 7–8 months of age (50% of Akita mice [6 of 12 mice] had ERG deficits compared with their age-matched WT control mice). In Akita mice with ERG deficits, PF-05231023 administration did not change the body weight and serum triglyceride levels versus controls but mildly reduced blood glucose levels (Supplementary Fig. 2). Prior to PF-05231023 treatment, we examined ERG responses in Akita mice. Although the Akita mouse responses were slightly attenuated overall compared with WT controls (Fig. 2A–C), Sm, as determined by mixed-linear modeling, was significantly attenuated (F = 15.9; df = 1,27.0; P = 0.00046) relative to WT. PF-05231023 showed protective effects on Akita retinas (Fig. 2B and C). Notably, Sm in Akita mice was improved after treatment (Fig. 2D) (F = 27.9, df = 1,29.8, P = 1.1 × 10⁻⁵) to levels that were supranormal, a result of better a-wave (log S) and b-wave (log 1/K_P2) sensitivities (Fig. 2E). There was no decline in baseline retinal function (ERG signals) in any individual mouse between 7 and 8 months of age (F = 0.166, df = 1,2, P = 0.723) (Supplementary Fig. 3). In Akita mice, the change in postreceptor sensitivity (log 1/K_P2) was positively correlated with the sum of changes in photoreceptor sensitivity and saturated amplitudes (log S + log Rm_P3, a-wave) (Fig. 3A), suggesting that the changes in postreceptor cells were reflecting the deficits in photoreceptor function (30,31).

In addition to neuronal function, we examined whether cone and rod photoreceptor structure was influenced by PF-05231023 administration. PF-05231023 administration increased cone-specific arrestin4 expression (32) and did not change rhodopsin expression at the mRNA levels in Akita mice (Fig. 3B). Color vision is initiated by a G-protein–coupled receptor–mediated phototransduction cascade in cones. Arrestins bind to the phosphorylated G-protein–coupled receptors and rapidly desensitize the activated receptors (32), contributing to the maintenance of photoreceptor function and health. Cone-specific arrestin 4 deficiency leads to diminished visual acuity and contrast sensitivity, as well as to significantly reduced neuronal responses detected by ERG in mice (33). Cone photoreceptor outer and inner segments were oriented in parallel in WT mice but were disorganized in Akita mice. PF-05231023 administration normalized the photoreceptor segment arrangement (Fig. 3C). Rhodopsin staining was comparable between WT and PF-05231023-treated Akita mouse retinas, whereas there appeared to be a reduction in the thickness of rod photoreceptor segments in Akita mice (Fig. 3D). With OCT measurements, there was a significant reduction in the total thickness of photoreceptor inner and outer segment in Akita mice (blue line) versus WT mice (black line), but the reduction in the inner and outer segment thickness was no longer observed with PF-05231023 administration (orange line) (Fig. 3E). These observations suggest that PF-05231023 protection against DR takes place through the restoration of photoreceptor function and structure. We do not exclude the possibility of the contribution from other retinal cells as Fgfr1 is also expressed in the inner nuclear layer and retinal ganglion cell (Supplementary Fig. 1).

Retinal inflammation induces retinal neurovascular abnormalities in diabetes (6,7,34,35). Significantly increased retinal IL-1β and decreased Vegfa mRNA expression was observed in Akita versus WT mice (Fig. 4A). PF-05231023 administration reduced retinal IL-1β mRNA expression in Akita mice (Fig. 4B). Interleukin (IL)-1β inhibits energy production in retinal neurons and induces retinal microvascular changes in rats (36–38). PF-05231023 administration did not change the expression levels of Vegfa, Tnfα, IL-6, IL-10, and Apn in Akita mouse retinas (Fig. 4B).

Hyperglycemia induces oxidative stress, a crucial contributor to DR (39). Photoreceptors are the most metabolically active cell in the body, and are very susceptible to metabolic derangement and resulting oxidative stress (40). Modulating photoreceptor oxidative stress protects against retinal neurodegeneration (41). In PF-05231023–treated versus vehicle-treated Akita mouse retinas, there was a significant increase in total AKT levels, although there was no significant change in the ratio of p-AKT to AKT (Fig. 5A), suggesting that the absolute level of p-AKT was higher in PF-05231023–treated mouse retinas. Activation of the AKT pathway regulates NRF2 activity in retinal pigment epithelium in vitro (42). There was a large variation of NRF2 protein levels in Akita versus WT mice. PF-05231023 administration decreased the variability of retinal NRF2 levels (Fig. 5B, blot 1) and showed a trend (nonsignificant) toward increased NRF2 levels (Fig. 5B, blot 2) in Akita mice. Taken together, the results suggested that PF-05231023 modulates retinal NRF2 levels by activating the AKT pathway. We further validated the observation in 661 W cells in vitro. We first evaluated the effect of hyperglycemia on 661 W cells. The 661 W cells tolerate very high levels of glucose (up to 100 mmol/L, in which 40% of total medium volume is glucose [stock is 555 mmol/L]) (Supplementary Fig. 4). The higher the glucose level, the more osmotic stress is induced. Hyperglycemia induces retinal oxidative stress, a crucial contributor leading to DR (39). Oxidative stress, resulting from highly metabolic photoreceptors, induces inflammation, which induces DR (6,7,34,35,40). Modulating oxidative stress prevents the progression of DR (43). The NRF2 pathway has antioxidant capability (41) and regulates IL-1β transcription (44,45). Therefore, we induced oxidative stress alone instead of inducing hyperglycemia with osmotic stress in 661 W cells for further confirmation of the mechanism. To test whether PF-05231023 protects photoreceptors against oxidative stress, we induced oxidative stress with the use of PQ, a nonselective herbicide, to induce the production of reactive oxygen species in mitochondria (46). In 661 W cells (the only photoreceptor cell line available currently) in vitro, oxidative stress induced with PQ increased IL-1β expression; PF-05231023 treatment prevented IL-1β induction (Fig. 5C). Both the activation of antioxidant transcriptional factor NRF2 and the phosphorylation of NF-κB modulate IL-1β transcription, which can be modulated by FGF21 (44,45,47). PF-05231023 treatment increased gene expression of Nrf2 but not Nfκb in 661 W (Fig. 5D). In PQ-treated 661 W cells (to induce oxidative stress), the induction of NRF2 expression by PF-05231023 was dose-dependently inhibited by perifosine, an AKT inhibitor (48) (Fig. 5E). PQ induced NRF2 protein levels and PF-05231023 treatment further increased NRF2 production, but did not change NF-κB phosphorylation in photoreceptors with PQ-induced oxidative stress (Fig. 5F). Cell antioxidant systems control reactive oxygen species for cell survival (49). PQ induces an increase in cellular reactive oxygen species (46). As a response, cells may induce their antioxidant levels as a defense, such as inducing NRF2 expression (Fig. 5F). PF-05231023 further increased NRF2 protein levels, suggesting further protection against oxidative stress.

APN is a key mediator of FGF21 modulation of glucose and lipid metabolism in mice (50,51). Changes in the APN pathway may contribute to the development of neovascular eye diseases (52). To test whether APN mediated the protective effects of PF-05231023, diabetes was induced with an injection of STZ in 6- to 8-week-old WT and Apn^−/− mice. Retinal function was then examined by ERG at 7–8 months of age. Again, PF-05231023 administration did not change body weight, blood glucose levels, or serum triglyceride levels (Supplementary Fig. 5). Furthermore, neither the amplitude nor sensitivity of the a-wave, b-wave, or the OPs differed significantly between STZ-treated mice (Fig. 6A–C), but Sm was significantly attenuated after STZ treatment (F = 12.2, df = 1,6.0, P = 0.013) (Fig. 6D). PF-05231023, administered as described above, again improved log Sm values (F = 45.2, df = 1,5.9, P = 0.001) in the STZ-treated mice to levels that were supranormal (Fig. 6D and E). The protective effects of PF-05231023 on retinal sensitivity (log Sm) in STZ-induced WT diabetic mice were again found in Apn^−/− diabetic mice (F = 23.8, df = 1,2, P = 0.040) (Fig. 6F–H), suggesting that the rescue was partly independent of APN. In the STZ-induced diabetic mice, PF-05231023 decreased IL-1β expression in diabetic WT and Apn^−/− retinas (Fig. 6I), suggesting that PF-05231023–induced reduction in IL-1β was independent of APN, in line with our phenotypic observation cited above (Fig. 6F–H).

New therapeutic approaches to prevent and manage DR are needed. Dysfunction in photoreceptors and postreceptor neurons are among the early retinal changes seen in patients with diabetes, anteceding ophthalmoscopic signs of retinopathy (3). In diabetes, hyperglycemia causes oxidative stress, a crucial inducer of DR (39). Activation of the antioxidant protein NRF2 protects against retinal neuronal degeneration (41), particularly in photoreceptors, as photoreceptors are the most metabolically active cells in the body (53,54) and are more vulnerable to oxidative stress damage. Our study demonstrated that, in insulin-deficient diabetic mice, the administration of the long-acting FGF21 analog PF-05231023 reversed diabetes-induced retinal neuronal deficits with improved photoreceptor function and morphology, and decreased photoreceptor-derived inflammation (Fig. 7). PF-05231023 administration suppressed proinflammatory IL-1β expression. IL-1β causes neurovascular damage in the retina (36–38). Therefore, we propose that FGF21 (PF-05231023) regulates retinal NRF2 levels and reduces IL-1β production and photoreceptor dysfunction in DR.

Photoreceptor high-energy consumption makes it susceptible to neurovascular disease. Blood vessels supply nutrients and oxygen to neurons and evacuate waste. Disturbances in neuronal activity trigger vascular remodeling (55). In diabetic animal models, photoreceptor responses to hyperglycemia induce retinal blood vessel loss (6,7,34). In addition, low rod sensitivity is associated with abnormal retinal vasculature. Rod photoreceptor demands contribute to the vascular recovery in hypoxia-induced retinal neovascularization (56). Photoreceptor metabolic dysfunction dictates pathological retinal angiogenesis (57). Therefore, maintaining photoreceptor function may prevent vascular abnormalities in DR. In insulin-deficient Akita mice, we found reduced sensitivity in the postreceptor retina, which is in line with clinical observations (3). Meanwhile, the changes in postreceptor cells were actually reflecting the deficits in photoreceptor function (30,31). We found that the administration of an FGF21 analog, PF-05231023, reversed the diabetes-induced morphological changes in photoreceptors and restored Sm; it also reduced disorganization of the photoreceptor segments. This suggests that the effect of PF-05231023 on retinal function ERG may be due to improved photoreceptor structure and function.

There is a strong correlation between hyperglycemia and the development of DR. Hyperglycemia leads to many cellular metabolic alterations that could serve as therapeutic targets. However, although pharmacological interventions disrupt putative biochemical signaling pathways between hyperglycemia and DR, an effective and safe drug is not yet available. Patients with type 1 diabetes have low circulating FGF21 levels (11,12), and FGF21 administration reduces hyperglycemia and lessens renal dysfunction in type 1 diabetic mice (13). FGF21 also improves the lipid profile (i.e., decreased triglycerides) of obese monkeys and patients with type 2 diabetes (58), indicating that FGF21 may have positive effects on diabetes and diabetic complications.

We found no significant impact of PF-05231023 administration on serum triglyceride levels in either Akita or STZ-induced diabetic mice. Although PF-05231023 slightly reduced hyperglycemia in Akita mice, this finding was not replicated in STZ-induced mice. The protective effects of PF-05231023 on retinal neurons is, therefore, likely to be independent of circulating glucose and lipid modulation. We also found that PF-05231023 protection against retinal neuronal deficits was preserved with APN deficiency. In the OIR mouse model of late vasoproliferative retinopathy, FGF21 inhibits pathological neovessel growth mediated by APN (17). Our current findings indicate that FGF21 regulates retinal neuron and neovessel growth through other mechanisms.

Oxidative stress resulting from highly metabolic photoreceptors induces inflammation, which induces DR (6,7,34,35,40). Modulating oxidative stress prevents the progression of DR (43). PF-05231023 administration attenuated the diabetes-induced IL-1β expression in Akita mice. FGF21 reduces oxidative stress and inhibits the NF-κB pathway in mice (44). In photoreceptors in vitro with PQ-induced oxidative stress, we observed that PF-05231023 treatment decreased IL-1β expression and activated the NRF2 pathway, which is known for its antioxidant capability (41) and for the regulation of IL-1β transcription (44,45). Additionally, PF-05231023–induced effect on IL-1β was independent of APN in diabetic retinas, which is in line with the neuronal observation. APN inhibits retinal neovessel growth via tumor necrosis factor-α (TNF-α) (59), and FGF21 also reduces TNF-α in neovascular mouse retinas (17). However, in Akita mice, there was no significant change in retinal Tnfα expression between PF-05231023– and vehicle-treated groups. Taken together, we concluded that, in diabetic retinas, PF-05231023 protected neuronal activity through the regulation of retinal levels of NRF2 and IL-1β, which was at least to some degree independent of the APN-TNF-α pathway that we showed to be involved in retinal neovascularization in OIR (60). IL-1β causes a decline in mitochondrial membrane potential and ATP production in retinal neurons (36). A reduction of retinal IL-1β may prevent the induction of early vessel loss in DR because IL-1β induces retinal microvascular abnormalities in rats (37,38).

In the current study, PF-05231023 was administrated intraperitoneally, and circulating FGF21 levels were measured. PF-05231023 administration did not alter retinal Fgf21, Fgfr1, and Klb expression in Akita mice (Supplementary Fig. 6). These data implicate circulating/peripheral FGF21 as a primary driver of retinal protection rather than autocrine/paracrine effects of FGF21 in the retina. Although FGF21 is expressed in liver, white adipose tissue, and brown adipose tissue, it is a hepatokine, and liver is the primary source of circulating FGF21 under fasting and refeeding conditions (61). In humans, liver is also the primary source of circulating FGF21 in a pattern that is consistent with a hormonal response (62). Although liver-derived FGF21 is critical for the adaptive response to fasting or starvation in rodents, in humans FGF21 plays an important role in fructose metabolism (63). Circulating FGF21 has been shown to cross the blood-brain barrier in humans in a nonsaturable, unidirectional manner (64). FGF21 regulates metabolism and circadian behavior, and sweet and alcohol preferences by directly acting on the nervous system (65,66). The blood-retinal barrier, which is essential for normal visual function (67), is broken down in DR (68). The leaky blood-retinal barrier potentiates the transport of FGF21 from blood into retina. Because local expression of FGF21 receptors has been detected in total retina and in retinal neurons, we speculate that circulating FGF21 could directly act on retinal neurons to exert protective effects in DR. However, we do not exclude the potential contribution of local FGF21.

In summary, there is an unmet need for the prevention and treatment of DR. Maintaining retinal structure and function, particularly photoreceptor activity, will improve retinal vascular stability, which can be achieved in the following two ways: 1) modulating photoreceptor metabolism to match the energy supply and 2) slowing down the visual cycle to reduce the energy demand (69). We determined that FGF21 (independent of VEGFA) may be a novel therapeutic target for the prevention and treatment of DR, possibly through restoring photoreceptor structure and function, and reducing photoreceptor oxidative stress and inflammation. Further exploration of FGF21 effects on photoreceptor metabolism and blood vascular changes in DR is desirable.

Acknowledgments. The authors thank Dr. Jing Chen, Boston Children’s Hospital, for her expert advice. The authors also thank Pfizer Cardiovascular and Metabolic Diseases for providing native FGF21 and PF-05231023.

Funding. L.E.H.S. is supported by National Institutes of Health National Eye Institute grants 1R24EY024868 and EY017017, Boston Children’s Hospital Intellectual And Developmental Disabilities Research Center grant 1U54-HD-090255, the Lowy Medical Research Institute, and European Commission FP7 project 305485 PREVENT-ROP. A.H. is supported by the Swedish Research Council (DNR# 2011-2432), Gothenburg County Council (grant ALFGBG-426531), long-term support by De Blindas Vänner and Kronprinsessan Margaretas Arbetsnämnd för synskadade, and European Commission FP7 project 305485. Z.F. is supported by the Knights Templar Eye Foundation (grant #76293) and Blind Children’s Center (seed grant #89282). C.-H.L. is supported by the Knights Templar Eye Foundation. B.C. is supported by the German Research Foundation (DFG). R.L. is supported by German Research Foundation (DFG) grant Li2650/1-1. Y.S. is supported by a Boston Children’s Hospital OFD/BTREC/CTREC Faculty Career Development Grant.

Duality of Interest. S.T. was formerly an employee and stockholder of Pfizer. He is a current employee of Merck and was at the time the study was conducted. No other potential conflicts of interest relevant to this article were reported.

Merck has had no role in any aspects of the work.

Author Contributions. Z.F. contributed to study conception and design; the acquisition, analysis, and interpretation of the data; and the drafting or revision of the article. Z.W., C.-H.L., Y.G., B.C., R.L., Y.S., S.S.M., S.B.B., I.A., E.M., R.D., A.P., and S.S.C. contributed to the acquisition of data. S.T. and A.H. contributed to the drafting or revision of the article. J.D.A. contributed to the analysis and interpretation of the data and the drafting or revision of the article. L.E.H.S. contributed to study conception and design, analysis and interpretation of the data, and the drafting or revision of the article. L.E.H.S. 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.

Prior Presentation. Parts of this study were presented in abstract form at Cell Symposia: Metabolic Disease Therapies, San Diego, CA, 15–17 October 2017.