Oral Peroxisome Proliferator–Activated Receptor-α Agonist Enhances Corneal Nerve Regeneration in Patients With Type 2 Diabetes

Diabetic neuropathy is one of the most common microvascular complications of uncontrolled diabetes (1). Chronic hyperglycemia results in the accumulation of advanced glycation end products and the production of reactive oxygen species, both of which increases oxidative stress and inflammation to pericytes and endothelium of capillaries, thereby reducing microvascular supply to Schwan cells or neurons (2). Prolonged hyperglycemia also inhibits nerve growth factor (NGF) for neuronal health and myelin formation, leading to reduced nerve velocity (3). Collectively, these pathological processes damage neuronal tissues, leading to corneal denervation. The human cornea is densely innervated, and corneal nerves play an important neurotrophic role in maintaining ocular surface homeostasis, integrity, and functionality (4). Any degeneration or injury to the corneal nerves leads to a neuropathic ocular surface, resulting in clinical presentations ranging from decreased corneal sensitivity to severe corneal perforation (5).

Diabetic corneal neuropathy (DCN) and its resultant diabetic keratopathy are underreported yet estimated to affect ∼47–64% of patients with diabetes (6). Despite the risk for corneal perforation and vision loss, the current mainstay management for diabetic keratopathy only focuses on symptomatic treatment, such as lubricants and prophylactic antibiotics. These treatments do not address the underlying pathophysiology (i.e., corneal neuropathy). Currently, recombinant NGF eye drops are the only U.S. Food and Drug Administration–approved drug for treating patients with neurotrophic keratopathy. However, this treatment course is costly and requires frequent topical application (7). In view of the significant disease incidence and limited treatment options, novel, effective, and affordable treatment is crucially needed.

Fenofibrate is a peroxisome proliferator–activated receptor-α (PPAR-α) agonist (8) known to regulate lipid and glucose metabolism (9). Through the activation of PPAR-α, this antilipidemic drug increases lipoprotein lipase activation, resulting in the rapid degradation of LDL and triglycerides (TGs) while increasing the synthesis of HDL cholesterol and stimulating the fatty acid oxidation pathway in tissues (10). We believe that these effects are helpful in protecting against nerve damage, as hypertriglyceridemia, hyperlipidemia, and decreased HDL cholesterol are known significant risk factors for the development of peripheral neuropathy (11).

Recently, two mouse studies (12,13) have shown that oral fenofibrate reduces neuroinflammation and improves axon regeneration of peripheral nerves via the activation of PPAR-α receptors. A separate study on diabetic mice demonstrated that fenofibrate treatment significantly prevented the development of diabetic peripheral neuropathy (DPN) by ameliorating neural, Schwann cell, and endothelial damage through the activation of the PPAR-α-AMPK-peroxisome proliferator–activated receptor γ coactivator 1α pathway (14). Additionally, the Fenofibrate Intervention and Event Lowering in Diabetes (FIELD) study reported that patients on fenofibrate had a significantly greater reversal of DPN than those on placebo, with an associated risk reduction in amputations of 37% (15). Fenofibrate may also be effective against DCN. PPAR-α knockout mice phenotypically have decreased corneal nerve density, impaired corneal sensitivity, and increased incidence of corneal lesions (16). The same study found that treatment with systemic fenofibrate significantly ameliorated the loss of corneal nerve fibers and restored glial cell–derived neurotrophic factor levels in the corneal epithelial and subbasal nerve plexus (16). These results suggest that PPAR-α agonists may have therapeutic potential for DCN. However, to our knowledge, the therapeutic effects of PPAR-α agonists against DCN have not been investigated and confirmed in a clinical trial. In the current study, we evaluated the potential effects of oral fenofibrate on the prevention of corneal nerve degeneration and the stimulation of corneal nerve regeneration in patients with type 2 diabetes. We also investigated the underlying mechanisms of the therapeutic effects with comprehensive tear neuromediator and proteomic analysis.

This was a single-arm, open-label, interventional study conducted at the Singapore National Eye Centre and Singapore General Hospital. Approval for the study was granted by the institutional review board of SingHealth, Singapore (ref. no. 2020/2050 and R1678/1/2020), and the study was conducted in accordance with the Declaration of Helsinki.

A total of 30 individuals with type 2 diabetes and 20 age-matched healthy control subjects were recruited from October 2020 to October 2021. Patients with diabetic nephropathy were recruited from the Singapore General Hospital’s Diabetes and Metabolism Centre. Adults (≥21 years) with type 2 diabetes of at least 3 months duration and mild to moderate diabetic nephropathy, defined as a diminished creatinine clearance time (CCT) between 30 and 60 mL/min and/or albuminuria, were invited to participate. Patients had to have an HbA_1c of <9% (<75 mmol/mol) with no significant changes in the dosages of their lipid- and glucose-lowering agents in the past 3 months. Patients were excluded if they had severe renal impairment (CCT <30 mL/min) or received renal replacement therapy. They were also excluded if they had any non–diabetes-related renal glomerular disease, chronic liver disease, or significant alcohol intake. Patients with concurrent use of substances that would interfere with the intervention drug (e.g., tacrolimus, bile acid sequestrants, systemic steroids) were excluded as well. The full list of the inclusion and exclusion criteria for the diabetes cohort is detailed in Table 1. Control subjects comprised volunteers who were free of any health problems, were not on any long-term medications, and had no corneal pathology and history of ocular or corneal surgery.

Eligible subjects were treated with oral fenofibrate for 30 days. Fenofibrate is contraindicated in individuals with severe renal impairment (CCT <30 mL/min), and dose adjustment is required in the presence of renal impairment. The dosages of fenofibrate in this trial were based on the local drug regulatory authority’s renal dosing recommendation. Subjects with a CCT >60 mL/min were treated with fenofibrate at 300 mg/day, and subjects with a CCT of 30–59 mL/min were treated with fenofibrate at 100 mg/day.

Patient age, diabetes duration, HbA_1c, and BMI were recorded. Additionally, blood tests were done to assess their pre- and postlevels of HDL, TGs, LDL, cholesterol, and fasting glucose. Neuropathic ocular surface assessment, in vivo confocal microscopy (IVCM) scans for corneal nerve plexuses and epithelial cells, tear neuromediator, and proteomic analysis were performed before and after treatment.

Laser scanning IVCM (HRT3 with Rostock Corneal Module; Heidelberg Engineering, Heidelberg, Germany) was performed by an experienced technician masked to the study group assignment as previously described (17). Briefly, scans were obtained at the center of the patient’s cornea first. The patient was then instructed to fixate on a light source in various directions with the contralateral eye to stabilize the scanning view, and the superior, inferior, nasal, and temporal region (each ∼3 mm away from the corneal apex) were also scanned. The scanned depth was from the superficial corneal epithelium to midstroma, at ∼350 μm. The time of image acquisition was 10 frames/s, and the field of view was 400 × 400 μm.

To avoid observer bias, the images were deidentified prior to image analysis. For the corneal nerve analysis, five best-focused and most representative images of the subbasal nerves were selected for each scanned area. Each main nerve trunk and nerve branch was selected only once. These 25 micrographs selected for each eye were analyzed by an automated software program (ACCMetrics; University of Manchester, Manchester, U.K.) (18) for the following seven nerve parameters: 1) corneal nerve fiber density (CNFD) (the number of fibers/mm², each frame area = 0.16033 mm²), 2) corneal nerve branch density (CNBD) (the number of branch points on the main fibers/mm²), 3) corneal nerve fiber length (CNFL) (total length of fiber mm/mm²), 4) corneal nerve fiber total branch density (CTBD) (total number of branch points/mm²), 5) corneal nerve fiber area (CNFA) (total nerve fiber area mm²/mm²), 6) corneal nerve fiber width (CNFW) (average nerve fiber width mm/mm²), and 7) nerve fiber fractal dimension (CFracDim) (metric of corneal nerve morphology to measure the spatial loss of nerves) (19).

For the corneal epithelial cells analysis, the five best-focused micrographs selected for each eye were analyzed using AIConfocal Rapid Image Evaluation System software (ARIES; ADCIS, Saint-Contest, Basse-Normandie, France), which automatically calculates the cell density (μm⁻²), average cell size (μm²), and cell circularity (20).

A handheld Cochet-Bonnet esthesiometer (Luneau Ophthalmologie, Paris, France) containing a thin retractor nylon monofilament was used to assess the patients’ corneal sensitivity. The 6.0-cm adjustable nylon filament was applied perpendicularly to the central and peripheral four quadrants of the subject’s cornea, starting from 6.0 cm and progressively reducing in 5-mm steps until the first response occurred (total scales 0–30 cm for five areas). Schirmer I test was performed by placing the Schirmer strips (Clement Clarke, Essex, U.K.) over the inferior temporal one-half of the lower lid margin in both eyes, without prior anesthesia, and the wet length (mm) after 5 min was recorded (21). Tear breakup time (TBUT) was evaluated by placing a single sterile fluorescein strip over the inferior fornix. The time interval between eye opening and the appearance of the first dry spot in the tear film was subsequently recorded (22). Ocular surface integrity was evaluated with fluorescence dye under the slit lamp using the cobalt blue illumination. Ocular surface staining was graded with the Oxford grading scale, ranging from 0 (absence of staining) to 5 (severe staining) (23). Corneal punctate staining was examined using the National Eye Institute (NEI) scale by dividing the cornea into five sections and assigning a value from 0 (absent) to 3 (severe) to each section, based on the amount, size, and confluence of the punctate lesions, for a maximum of 15 points.

The strips collected from the Schirmer I test were stored at −80°C until the day for analysis. For tear neuromediator analysis, ELISA was performed. The wetted part of each Schirmer strip was first cut into small pieces and submerged in 200 μL of ice-cold tear elution buffer containing 0.55 mol/L NaCl, 0.33% Tween 20, 0.55% BSA, and 1× protease inhibitor. Tear samples were then homogenized by sonication for 20 s, followed by overnight incubation at 4°C with gentle shaking at 450 rpm. The homogenization step was repeated once before centrifugation at 11,000 rpm at 4°C for 20 min (24). The tear proteins containing clear supernatants were collected and stored at −80°C until ELISA analysis.

For plasma neuromediator analysis, blood from each patient at each visit was collected in EDTA-anticoagulated tubes, followed by centrifugation at 3,000 rpm at 4°C for 10 min. The plasma was collected and stored at −80°C until ELISA analysis.

The level of tear and plasma neuromediators, including substance P (SP), calcitonin gene-related peptide (CGRP), neuropeptide Y (NPY), and NGF, was analyzed separately using ELISA kits according to the manufacturer’s instructions. The details of the dilution factors and kit information are provided in Supplementary Table 1. The optical density measurement was read at 450 nm using an Infinite M200 plate reader (Tecan, Männedorf, Switzerland), and optical density reading at 540 nm was set as the reference. The final concentrations of each analyte were interpolated from the standard curve using GraphPad Prism 8 software (GraphPad Software, San Diego, CA).

Quantitative tear proteomic analysis was performed as previously described (25). The Schirmer strips were cut into small pieces, and 100 μL of lysis buffer was added and mixed in a ThermoMixer for 1.5 h at 20°C. The total protein concentration was measured using Bio-Rad DC Protein assay. One hundred micrograms of eluted tear proteins were reduced, alkylated, tryptic digested, and desalted using the reagents from EasyPep Mini Sample Prep Kits (Thermo Fisher Scientific, San Jose, CA). The total peptide amount was quantified with a fluorometric peptide quantification kit. The peptide sample was dried with SpeedVac and resuspended in 2% acetonitrile with 0.1% formic acid containing iRT (1:10 ratio; Biognosys). All peptide samples (1 μg of total peptides per sample) were analyzed using an EASY-nLC 1200 system coupled to an Orbitrap Exploris 480 mass spectrometer via an EASY-Spray source. For liquid chromatography separation, an Acclaim PepMap 100 C18 was used as the precolumn and a PepMap RSLC C18 as the analytical column. Peptides were separated by a 60-min gradient using 0.1% formic acid as buffer A and 80% acetonitrile in 0.1% formic acid as buffer B. The reagent, kits, and columns were purchased from Thermo Fisher Scientific. Quantitative proteomic analysis was performed using data independent acquisition (DIA) experiments. In total, 19 windows of 45.7 Da were used with an overlap of 3 Da. The normalized automatic gain control target value for fragment spectra was set at 1,000%, and maximum injection time mode was set to auto. DIA data were processed using library-free directDIA workflow in Spectronaut 15 (Biognosys).

Raw protein abundance values were derived after fragment ions were selected for quantification based on default quality control criteria as implemented in mapDIA (26). The downstream data analysis, data visualization, and Gene Ontology term enrichment were performed using custom scripts in R (64-bit, version 4.1.1). The raw abundance data were median normalized and log transformed for all statistical analyses. Based on the proteomics data set, pathway enrichment analysis was performed with gene set enrichment analysis (GSEA) (27) using the Kyoto Encyclopedia of Genes and Genomes database. The enrichment score (ES) corresponds to the (weighted) Kolmogorov-Smirnov statistic and reflects the degree to which the gene (protein) in a gene (protein) set is upregulated (positive ES) or downregulated (negative ES).

The required sample size was calculated based on the pilot data of the primary outcome, which was CNFD, from five patients. The mean CNFD was 9.5 ± 5.1 fibers/mm² and 12.3 ± 5.3 fibers/mm² before and after treatment, respectively. Hence, a sample size of 22 patients, with a power of ≥80% and at a 5% level of significance, was sufficient to detect the difference before and after treatment. Considering a 15% rate of loss to follow-up, we recruited 30 patients with diabetes. Linear mixed models were used to analyze the data before and after treatment from both eyes, as well as the data of control subjects and patients before treatment, to take into account the correlation of the paired time points and both eyes. Linear mixed models were also used to evaluate the association between the neuromediator and nerve profiles. All data are expressed as mean ± SD. The statistical analysis was performed using Stata 17 (StataCorp, College Station, TX), and P < 0.05 was considered significant.

All data generated or analyzed during this study are included in the published article and its online supplementary files. Resources will be made available upon request.

Of the 30 patients with diabetes, 21 (70%) and 9 (30%) were prescribed oral fenofibrate 100 mg/day and 300 mg/day, respectively. The mean age was 60.8 ± 9.3 years, and the majority (80.8%) were men and Chinese. Table 2 shows the clinical and demographic characteristics of the patients with diabetes and control subjects. Patients with diabetes had significantly higher HbA_1c and LDL levels (both P < 0.001) compared with the control subjects. The serum TGs decreased from 1.8 ± 1.1 to 1.4 ± 0.8 mmol/L (P = 0.002) after fenofibrate treatment, suggesting good adherence to the intervention drug. The other baseline characteristics were comparable to the control group (all P > 0.05).

Compared with the control subjects, patients with diabetes had significantly lower CNFD (P = 0.04) and CNFL (P = 0.03) and higher CNFW (P = 0.03). Their corneal nerves were more tortuous and had a greater number of nerve beadings (Fig. 1A). After fenofibrate treatment, CNFD significantly increased from 9.5 ± 6.2 to 12.6 ± 4.4 fibers/mm² (P = 0.01) and was restored to the level comparable to the control group (12.3 ± 6.0 fibers/mm²). Following treatment, CNFW significantly decreased from 0.024 ± 0.001 to 0.022 ± 0.001 mm/mm² (P = 0.01), suggesting less swelling of the nerve fibers. There was also a trend of improvement in CNBD, CTBD, CNFA, and CFracDim, though it did not reach statistical significance (Table 3 and Fig. 2).

Morphological improvement with a more regular cell border was observed in the corneal epithelial layers after treatment, as evidenced by a significant decrease in the circularity of the cells from 0.73 ± 0.02 to 0.72 ± 0.02 (P = 0.04) (Table 3). We also observed that the epithelial cell density became more compact along with a reduction in cell size (Fig. 1B) from 131.0 ± 22.1 to 127.1 ± 9.50 μm² (Table 3). After treatment, the epithelial cell parameters improved to a level close to those of control subjects (Table 3 and Fig. 2H–J).

Corneal sensitivity (24.6 ± 1.8 vs. 27.5 ± 1.4 mm; P < 0.001), TBUT (5.7 ± 1.5 vs. 8.1 ± 1.7 s; P < 0.001), ocular surface Oxford score (0.8 ± 0.8 vs. 0.3 ± 0.6; P = 0.004), and corneal staining NEI score (1.2 ± 1.1 vs. 0.2 ± 0.5; P = 0.001) were significantly worse in the patients with diabetes than control subjects (Table 4). After fenofibrate treatment, ocular surface superficial punctate keratopathy, which stained green with fluorescein dye under cobalt blue light (Fig. 1C), was significantly reduced, as also evidenced by the significant reduction in Oxford and NEI scores. The mean Oxford score significantly decreased from 0.8 ± 0.8 to 0.6 ± 0.6 (P = 0.01). The mean NEI scores also significantly improved from 1.2 ± 1.1 to 0.7 ± 0.6 (P = 0.008). The tear film became significantly more stable (Fig. 1C), as demonstrated by the significant improvement in TBUT from 5.7 ± 1.5 to 6.2 ± 1.8 s (P = 0.006) (Fig. 2).

The patients with diabetes had a significantly lower tear SP (P = 0.05) and a significantly higher NGF level (P = 0.04) than the control subjects. After treatment, the tear SP levels significantly increased from 1,239.4 ± 719.2 to 1,669.0 ± 948.4 pg/mL (P = 0.03) (Table 4). Of note, the changes in tear SP level were significantly associated with an increase in CNFD (β = 0.003; P = 0.013). There was no significant alteration in the CGRP, NPY, and NGF levels in tears after treatment (Table 4).

The SP level in plasma was 228.3 ± 87.8 and 203.9 ± 76.6 pg/mL before and after treatment, respectively (P = 0.08). The CGRP levels in plasma were 77.0 ± 15.4 and 77.8 ± 12.0 pg/mL before and after treatment (P = 0.86). Fenofibrate did not alter blood neuromediator profiles.

In the diabetes group, apolipoprotein A-IV (APOA4) (log₂ fold change [log₂FC] = 2.28), and serum amyloid A-1 protein (SAA1) (log₂FC = −3.82), both related to inflammation, were significantly increased (Fig. 3A). Elevated APOC1 (log₂FC = 2.68) and complement C1q subcomponent subunit A (C1QA) proteins, which are essential for several complement-mediated activities, were also observed (log₂FC = 3.08) (Fig. 3A). The top 25 tear proteins that were significantly up- and downregulated in the eyes of patients with diabetes versus control subjects are listed in Table 5 and Supplementary Fig. 1A. These significant dysregulated protein expressions were linked to significantly downregulated cholesterol (ES −0.75; P = 0.001), carbohydrate (ES −0.93; P = 0.002), fat (ES −0.80; P = 0.008), and protein (ES −0.94; P = 0.01) metabolism. In particular, neuroactive ligand-receptor interaction (ES −0.89; P = 0.04) and PPAR signaling pathway (ES −0.67; P = 0.01) were significantly downregulated in the patients with diabetes (Fig. 3A and Table 6).

When comparing tear proteomic profiles before and after fenofibrate treatment, β-galactoside α-2,6-sialytransferase 1 (ST6GAL1) (log₂FC = 1.42), human tetratricopeptide repeat protein 9A (TCC9A) (log₂FC = 0.75), and ras-related protein (RAB5A) (log₂FC = 0.63), all involved in the regulation of nervous system function, were significantly increased after treatment (Fig. 3B). Significant upregulation of suppressor of mothers against decapentaplegic homolog 1 (SMAD1) (log₂FC = 1.72), which is important for the maintenance of neural cells (28), was also observed after treatment. On the other hand, cytochrome c oxidase assembly factor 6 homolog (COA6) (log₂FC = −2.66), which is involved in the transcriptional network in neurons (29), was significantly downregulated. Lamin-B1 (LMNB1) (log₂FC = −1.44), a scaffolding component of nuclear envelope, was also significantly downregulated (Fig. 3B). Table 5 and Supplementary Fig. 1B list the top 25 tear proteins that are significantly up- and downregulated after treatment. Upon GSEA analysis, we found that three neuronal regulation–related pathways were significantly upregulated after treatment: 1) MAPK signaling pathway (ES 0.53; P = 0.009), 2) neurotrophin signaling pathway (ES 0.59; P = 0.01), and 3) linoleic acid (LA) metabolism (ES 1.00; P = 0.01) (Table 6). Ribosome family was significantly downregulated (ES −0.67; P = 0.002). Fenofibrate treatment also significantly modulated cholesterol and fat metabolism (ES 0.77 [P = 0.004] and 0.90 [P = 0.006], respectively). Ocular surface complement and coagulation cascades (ES −0.78; P = 0.001), neutrophil extracellular trap formation (ES −0.69; P = 0.001), and platelet activation (ES −0.66; P = 0.01) were significantly downregulated after treatment (Fig. 3B and Table 6).

In this clinical trial, we demonstrated for the first time the neurotrophic effect of oral fenofibrate on diabetic corneas. The beneficial effects on neuropathic ocular surface are evidenced by the significant improvement in CNFD, CNFW, epithelial cell morphology, and tear SP level. These further contributed to an objective improvement in the clinical ocular surface function through an improvement in tear stability and the reduction of corneal and conjunctival punctate keratopathy. We also found that treatment with fenofibrate was associated with alterations in three pathways: 1) neuronal pathway, 2), lipid modulation, and 3) anti-inflammation and anticoagulation.

Treatment with oral fenofibrate improves corneal nerve metrics in diabetic corneas. At baseline, we observed a significantly lower CNFD and CNFL, as well as significantly higher CNFW, in the diabetes cohort compared with the age-matched control cohort. This finding is supported by previous literature showing that CNFD and CNFL were significantly reduced in patients with diabetes (30,31). Thickening of subbasal nerves found in diabetic corneas (32) may be indicative of the progressive deposition of advanced glycation end products and an increase in DPN severity (6,33). After fenofibrate treatment, CNFD and CNFW improved significantly, with CNFD restored toward the level of control subjects. These findings indicate that fenofibrate leads to the regeneration of nerve fiber and a reduction in nerve edema. Moreover, we observed a trend of improvement in CFracDim posttreatment, indicating a decrease in the spatial loss of nerve fibers. Fenofibrate treatment also improved the morphology of the epithelial cells in terms of cell circularity. The corneal subbasal plexus is thought to contribute to a healthy epithelial surface by providing trophic influences and maintaining epithelial metabolism (34). Several studies have shown that hyperglycemia compromises epithelial barrier function, resulting in morphological changes to the corneal epithelial cells (35). Neurotrophic factors, such as epithelial growth factor and ciliary neurotrophic factor, have been shown to promote corneal epithelial wound healing (36). In line with these observations, the upregulation of neurotrophin signaling and neurotrophic factors could explain the posttreatment improvement in corneal epithelial morphology.

Clinically, the diabetes cohort had significantly worse corneal sensitivity, TBUT, and corneal and ocular surface punctate keratopathy compared with the control cohort. Diabetes-induced denervation reduces blink reflex, trophic support, and the viability of the corneal epithelial cells, putting patients at an increased risk of ocular surface damage (36). Studies have also suggested that ocular surface dysfunction is correlated with the duration and severity of diabetes (37,38). We found that fenofibrate promoted ocular surface homeostasis, as evidenced by the significant improvement in TBUT and Oxford and NEI scores. The stimulation of corneal nerves also promotes lacrimal gland secretions and maintains tear volume and composition (39). Hence, the improvement in corneal nerve metrics would also provide favorable effects on patients’ TBUT and ocular surface integrity. We did not observe significant changes in Schirmer values after treatment, and this may be due to the high variability and low repeatability of Schirmer I tests (40). While fenofibrate improved several nerve parameters, corneal sensitivity remained unchanged, highlighting the issue that the current clinical evaluation tool (i.e., corneal esthesiometer) is too crude to detect subtle nerve changes. A greater degree of neuronal restoration may also be needed before any clinically appreciable improvement in corneal sensitivity could be observed.

Neuromediators, which are released in response to nerve damage or stimulation, are crucial in maintaining ocular surface homeostasis. SP stimulates the proliferation of epithelial cells and promotes diabetic corneal epithelial wound healing via neurokinin-1 receptors (41). The reduction in tear SP levels is associated with an increase in DPN severity (42) and predisposes peripheral nerves to early neuropathic changes (42). In our study, we found that tear SP concentrations were significantly lower in patients with diabetes, and its level significantly increased after fenofibrate treatment. Notably, this increment in tear SP was significantly associated with the increase in CNFD, demonstrating that tear SP may serve as a surrogate biomarker for corneal nerve regeneration in concordance with previous studies illustrating a significant correlation between tear SP and CNFD (42,43). Tear NGF concentrations were also found to be significantly higher in the patients with diabetes than in control subjects, which could be due to 1) a consequence of neuroinflammation (44), 2) a compensatory reaction to decreased corneal nerves (44), 3) an impaired uptake and transportation system of NGF in patients with diabetes (45,46), or 4) a protective response to degenerative nerves (47).

Neuroactive ligands are important for the maintenance of protective corneal reflexes, support for epithelial regeneration, and the recovery of nervous function (48). We found a significantly downregulated neuroactive ligand-receptor interaction and PPAR signaling pathway in the patients with diabetes cohort, which could explain their clinical manifestations, such as decreased corneal sensitivity and superficial punctate keratopathy. More recently, a study demonstrated that PPAR-α levels were reduced in the corneas of diabetic mice, and the knockout of PPAR-α alone was sufficient to induce corneal nerve degeneration, impaired corneal sensitivity, and increased corneal lesions (16). The downregulation of PPAR-α thus plays a pathogenic role in diabetic keratopathy, which may be amendable to treatment using PPAR-α agonists such as fenofibrate.

The neuroprotective effects of fenofibrate are not fully understood but can occur via several mechanisms (Table 7). Based on our GSEA, oral fenofibrate stimulated the neurotrophin and MAPK signaling pathways as well as LA metabolism. Neurotrophins are essential for the modulation of corneal nerve branching, maintenance of corneal nerve density, and the success of nerve regeneration (49). The activation of MAPK mediates neurite outgrowth-promoting effects in vitro (50). Moreover, overexpression of LA metabolism generates γ-LA, which serves as an important constituent of neuronal membrane phospholipids and helps to preserve nervous blood flow to facilitate nerve regeneration (51). We also found that fenofibrate suppressed the expression of ribosome. Axonal ribosomes serve as a marker for diseased axons and indicate involvement in neurodegenerative disorders (52). Taken together, these data suggest that fenofibrate potentially attenuates neurodegeneration and enhances neuroprotection.

Elevated triglycerides are associated with corneal nerve damage in patients with small fiber neuropathy (53). We found that the metabolism of cholesterol and fatty acid was significantly downregulated in the diabetes cohort. As fenofibrate acts by lowering triglycerides and increasing HDL levels, this could explain the posttreatment improvement in DCN. A recent review demonstrated the mechanistic link between altered lipid metabolism and peripheral nerve dysfunction (54), suggesting the importance of lipid modulation in DCN. From the angle of drug discovery, fenofibrate stands in a more favorable position compared with other investigational drugs, as it is a drug prescribed routinely for the management of dyslipidemia.

We also found that fenofibrate significantly suppressed neutrophil extracellular trap formation, complement and coagulation cascades, and platelet activation pathways. Similarly, it was reported that topical PPAR-α treatment inhibits proinflammatory markers, suppresses inflammatory cells, and promotes corneal wound healing in the cornea of a rat alkali-burned model (55). The same study found that topical PPAR-α suppresses nuclear factor κB expression (55). The inhibition of nuclear factor κB has been reported to reduce neuroinflammation as well as stimulate neurogenesis (56), neuritogenesis (57), and axoneogenesis (58). In our study, we found that ST6GAL1, which regulates neutrophilic inflammatory response (59), increased following treatment. Furthermore, low platelet time and plateletcrit levels were found to have a significant association with poorer nerve conduction function and the presence of neuropathy in patients with type 2 diabetes (60). Collectively, these suggest that the suppression of inflammation and platelet activation are possible mechanisms of fenofibrate.

On the protein level, APOA4 was significantly upregulated in the diabetes cohort. Increased APOA4 was found in the early stages of nerve degeneration (61). This increase was postulated to be in response to denervation and chronic neuroinflammation due to hyperglycemia (62). SAA1, on the other hand, was significantly decreased in patients with diabetes, which is in agreement with a previous study (63). APOC1 and C1QA, which are both associated with inflammation and diabetic nephropathy (64,65), were significantly upregulated in the tears of the patients with diabetes. After treatment, there was an increase in the expression of neurotrophic receptors by the upregulation of TCC9A and RAB5A. TCC9A and RAB5A regulate neuronal plasticity (66) and signaling of neurotrophic receptors (67), respectively. Both proteins are involved in the neurotrophin and MAPK signaling pathways, which were found to be upregulated in our GSEA. SMAD1, which is critical for axon generation (28), was found to be significantly increased after treatment. On the other hand, we found LMNB1 levels to be significantly downregulated after treatment. Increased expression of LMNB1 has been reported to cause central myelin breakdown and demyelinating neuropathy (68).

The current study has several limitations. First, the changes in the age-matched control cohort were not evaluated, but one would expect few changes in the parameters of interest in healthy individuals. Second, a longer end point duration could be considered and will be included in our future studies. Furthermore, we did not see significant differences in other nerve parameters, despite a trend of improvement. This could be explained by CNFD being chosen as the primary outcome for the sample size calculation, but it should be noted that CNFD is the most reliable parameter measured and is the most important clinical indicator for nerve regeneration (69). Moreover, the nerve parameters could have been underestimated when analyzed by an automated software, especially in low-denervated corneas, such as diabetic corneas (69). The newly regenerated corneal nerve fibers are relatively short and fine, making nerve parameters like CNFL less sensitive to be detected. Finally, although we reported alterations in several tear proteins and pathways following fenofibrate treatment, additional mechanistic studies are required to confirm the causal relationship between these pathways and the improved clinical phenotype.

In conclusion, with comprehensive investigation comprising clinical ocular surface assessment and corneal nerve and epithelial cell imaging analysis, we demonstrated that fenofibrate improves ocular surface integrity in diabetic corneas. In addition, detailed ocular surface proteomics suggest that neurotrophic reaction, lipid modulation, and anti-inflammatory and anticoagulation response may contribute to the neuroprotective effects of fenofibrate. Our data support the further evaluation of fenofibrate as a potential novel treatment for DCN.

Acknowledgments. The authors thank colleagues Chen Qi Fan and Vieon Wu Ai Ni from the Department of Endocrinology, Singapore General Hospital, who coordinated and helped with the clinical trial.

Funding. This study was supported by Singapore National Medical Research Council grants MOH-CSAINV21jun-0001 and NMRC/OFLCG/001/2017.

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

Author Contributions. C.H.Y.T. contributed to the manuscript writing, formal analysis, data curation, and visualization. M.T.-Y.L. and I.X.Y.L. performed the laboratory experiments and data curation. S.-K.K. performed the laboratory experiments. L.Z., D.S.G., H.C., and H.W.L.K. contributed to the formal analysis and visualization. A.Y.R.L. contributed to the acquisition of data. P.S.L. contributed to the methodology. J.S.M. and J.-P.K. reviewed the manuscript. T.M.C. contributed to the funding acquisition. H.C.T. contributed to the conceptualization, methodology, formal analysis, manuscript review, and supervision. Y.-C.L. contributed to the conceptualization, methodology, investigation, formal analysis, manuscript review, supervision, and funding acquisition. H.C.T. and Y.-C.L. 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 orally at the Association for Research in Vision and Ophthalmology Annual Meeting, Denver, CO, 1–4 May 2022; in poster form at the 32nd International Congress of Medical Women International Association, Taipei, Taiwan, 24–26 June 2022; and orally at Wong Hock Boon Society Research Day, National University of Singapore, Singapore, 29 January 2022.