Association between Diabetes and Keratoconus: A Retrospective Analysis

Keratoconus (KC) and chronic diabetes mellitus (DM) are both associated with significant defects in the human corneal structure. Studies have long suggested that DM is linked to KC, mainly via the crosslinking mechanism, but scientific evidences are lacking. The role of altered systemic metabolism is well-established in both DM and KC with studies suggesting localized altered cellular metabolism leading to the development of corneal pathologies. We have previously characterized the metabolic defects associated with both conditions using targeted metabolomics. To compare metabolic differences between KC and DM-derived corneal fibroblasts, we performed a respective study of two cohorts of the KC and DM populations using a retrospective analysis of targeted metabolomics data. The goal of this study was to identify the group of differentially regulated metabolites, in KC versus DM, so that we may unravel the link between the two devastating corneal pathologies.

increased Collagen III production 22 , though heavily favoring alternative pathways including tryptophan metabolism depending on Type 1 DM (T1DM) or Type 2 DM (T2DM) 23 . This work has led us to hypothesize that differential metabolic regulation may be a key factor in promoting the pathological changes in corneal structure, e.g. thinning and thickening, in the context of KC development and DM progression, respectively.
The aim of the current study was to draw parallel and contrasting features of the metabolic network of both KC-and DM-derived corneal stromal cells as a means to determine key characteristics in bioenergetics that contribute to pathological changes in corneal structure associated with each disease. In order to do that, we performed a retrospective analysis of multiple targeted metabolomics datasets with a cross comparison of the metabolome of corneal fibroblasts isolated from healthy controls, KC-, T1DM-, and T2DM-patients. These studies utilized our established 3D in vitro model where primary cells are stimulated by stable ascorbic acid (Vitamin C) to assemble their own ECM. Using targeted mass spectrometry, we retrospectively performed a higher-order analysis of regulated metabolic pathways by comparing the relative abundance of 134 metabolites in order to determine similarities and differences between the DM and KC samples.
Metabolomics has been used, by us and others 20,21,[23][24][25][26] , for the investigation of ocular diseases. However, it has not been utilized to its full potential in terms of identifying therapeutic targets for diseases. To our knowledge, this is the first study attempting to shine a light on the corneal metabolic associations between KC and DM.

Materials and Methods
Ethical approval and informed consent. The study was performed with the Institutional Review Board (IRB) approval from the University of Oklahoma Health Sciences Center (IRB protocol #3450). Written, informed consent was obtained prior to tissue collection. All methods were performed in accordance with federal and institutional guidelines. All human samples were de-identified prior to analysis. This research adhered to the tenets of the Declaration of Helsinki.
Study rationale and objectives. A retrospective analysis of previously reported metabolomics studies of Healthy, KC, T1DM, and T2DM corneal stromal cells, in our 3D in vitro model, was performed using a cross-comparison analysis of multiple experiments 22,27 . These metabolic experiments have previously highlighted dysregulated metabolic pathways or drug-regulated pathways (e.g. prolactin) related to each disease compared to healthy controls. The objectives of the current study included the following: i) identify differential regulation of metabolic pathways that may be shared (up-or down-regulated) between KC and DM; and ii) directly compare similarities and differences in metabolism, in KC and DM, that may contribute to corneal crosslinking. Data compilation. Targeted metabolomics data was collected and processed at the Mass Spectrometry Core at Beth Israel Deaconess Medical Center. All samples were isolated and analyzed using identical protocols, as previously described 16,[21][22][23] . Two cohorts of each population (healthy, KC, and DM) 22,27 were chosen with no added chemicals or treatments, as used in other metabolomics studies previously reported by our lab 21 . inclusion and exclusion criteria. Only metabolites identified in all biological replicates in each group were included in the analysis. Two independent datasets 22,27 were evaluated using HCFs obtained from healthy controls, HKCs isolated from KC patients, and T1DM and T2DM isolated from DM patients. Any metabolites with no expression detected in a biological replicate were excluded from further analysis. cell isolation and expansion. Human corneal stromal cells were isolated from healthy, KC, T1DM, and T2DM patients. All healthy samples, with no ocular or systemic diseases, were obtained from NDRI (National Disease Research Interchange; Philadelphia, PA). All KC samples were obtained from Aarhus University Hospital (Aarhus, Denmark) and our collaborator Dr. Jesper Hjortdal. The average age range for HCF and HKC donors reported for one independent data set 27 was 46 ± 22 and 47 ± 14 years old. T1DM and T2DM samples were obtained from the Oklahoma Lions Eye Bank (Oklahoma City, OK), as well as the NDRI. The average age range for HCF and T1DM and T2DM donors for the second data set 22 were 58 ± 6, 55 ± 7, and 59 ± 5 years old, respectively 22 . An average duration of DM was reported as 15.71 ± 4.17 years with a range of 3-30 years 22 . Tissue was processed, as previously described 28 . Briefly, the corneal epithelium and the corneal endothelium were removed from the stroma by scraping with a razor blade. The stromal tissue was then cut (~2 × 2 mm pieces) and positioned into T25 culture flasks. Explants were allowed to adhere and cultured with Eagle's Minimum Essential Medium (EMEM: ATCC; Manassas, VA), 10% fetal bovine serum (FBS: Atlantic Biologicals; Miami, FL) and 1X antibiotic-antimycotic (Gibco, Life Technologies; Grand Island, NY). All cells used for our experiments were between passages 3 and 7. 3D in vitro constructs. As previously described [20][21][22] , HCFs, HKCs, T1DMs, and T2DMs were plated on 6-well transwell polycarbonate membrane inserts with 0.4 μm pores (Corning Costar; Charlotte, NC) at a density of 10 6 cells/well. All cells were cultured in EMEM with 10% FBS and 1X antibiotic-antimycotic and were further stimulated with 0.5 mM 2-O-α-D glucopyranosyl-L-ascorbic acid (American Custom Chemicals Corporation, San Diego, CA, USA). The cultures were allowed to grow for 4 weeks before processing.

Isolation, extraction of metabolites and targeted mass spectrometry. Sample preparation and
processing. All constructs consisting of HCFs, HKCs, T1DMs, and T2DMs were processed and metabolites were isolated as previously described [20][21][22] . Briefly, samples were washed with 1xPBS and lysed with ice-cold 80% methanol, incubated on dry ice for 15 minutes, and homogenized briefly to ensure complete cell lysis. Samples were then centrifuged at 13,500 rpm overnight at 4 °C and stored at −80 °C until further analysis. Pellets were re-suspended in 20 μL HPLC-graded water for targeted tandem mass spectrometry.
Mass spectrometry processing. Briefly, 5 μL of each sample was injected and analyzed using a hybrid 5500 QTRAP triple quadrupole mass spectrometer (AB/SCIEX) coupled to a Prominence UFLC system (Shimadzu) using an Amide HILIC column (Waters) and analyzed with selected reaction monitoring (SRM) with positive/ negative polarity switching. Peak areas from the total ion current for each of 134 metabolite SRM transition were integrated using MultiQuant v2.1 software (AB/SCIEX). Relative metabolite abundance was provided as integrated peak intensities.
Data analysis. Retrospective data, from previous studies 22,27 , was analyzed and plotted. Each of these experiments was performed from two independent studies using at least 3 biological replicates per group (HCFs, HKCs, T1DMs, and T2DMs). The data for each test group (HKC, T1DM, and T2DM) were first normalized relative to their respective experimental control (HCF) allowing for comparability between groups and experiments based on the up-/down-regulation relative to control. The original raw data from the MultiQuant software was uploaded to MetaboAnalyst (http://www.metaboanalyst.ca) for subsequent data processing and analyses 29,30 . Over Representation Analysis (ORA) was performed using only the metabolites that are regulated by 2:1 ratio, as previously described 16 , in order to ensure that only the vastly abundant metabolites were included. pathway enrichment. We performed Pathway Enrichment Analysis using a Metaboanalyst (www. Metaboanalyst.ca), which is intended for the analysis of metabolomics data 29 . As previously reported, only the metabolites that were up or down regulated by 2-fold were included in the analysis. The 2-fold cutoff ensures that only the vastly abundant metabolites were included. Furthermore, only metabolites that were detected in all biological samples were included. The metabolites fulfilling our criteria were input into the software and the pathway enrichment analysis was executed. The output of this algorithm highlights metabolic pathway or pathways affected. Statistical analysis. Data was generated from our previous metabolic studies on diabetes in the cornea and keratoconus separately 22,27 and was compiled for this retrospective analysis. All data was plotted with at least an n ≥3 with statistical significance determined using a one-way ANOVA with Tukey's multiple comparison test. A p-value of less than 0.05 was considered statistically significant.

Results
Altered Metabolic pathways. A retrospective analysis of the metabolomics data indicated that out of 134 targeted metabolites, 11, 11, and 18 metabolites were downregulated at least 2-fold in HKCs, T1DM, and T2DM cells, respectively, compared to HCFs (Fig. 1A). Quinolinate, an agonist of the glutamate receptor, was the only metabolite downregulated in both KC and T1DM (p ≤ 0.05) and not in the T2DM. Both HKCs and T1DM cells exhibited significant upregulation of 24 and 29 metabolites relative to the HCFs, respectively, compared to only 5 metabolites upregulated in T2DM (Fig. 1B). Interestingly, dimethyl-L-arginine, an inhibitor of nitric oxide synthase, was the only metabolite shared in a 2-fold upregulation in HKCs (p ≤ 0.05) and T1DM (p ≤ 0.01) compared to HCFs, with no significant variation found in T2DM.

Metabolic Differences between HKC and T2DM. Six metabolites were significantly different among the
HKCs and T2DM. Two of the metabolites, 1,3-disphosphoglycerate and phosphoenolpyruvate are components of the glycolytic pathway and are contiguous in nature, as seen in Fig. 3. When compared to HKCs, T2DMs exhibit increased expression by 5-fold (p ≤ 0.001) and 1.7-fold (p ≤ 0.05) in both metabolites, respectively (Fig. 3D,E).
Interestingly, lactate and citrate were both significantly downregulated in T2DMs, when compared to HKCs. www.nature.com/scientificreports www.nature.com/scientificreports/ The prolonged hyperglycemic states that the human cornea experiences during DM also lead to endogenous CXL within the corneal stroma. In the case of DM, however, the process happens naturally with age through a cascade of intramolecular and intermolecular mechanisms primarily mediated by crosslinking of advanced glycation  www.nature.com/scientificreports www.nature.com/scientificreports/ end products (AGEs) that are produced during chronic hyperglycemia 3 . Furthermore, studies have suggested that riboflavin-mediated CXL may also proceed via in situ production of AGEs from endogenous glycosaminoglycans (GAGs) present on core proteoglycan proteins 32 . A few case studies have suggested that there may be a protective effect of diabetes on the development of KC. With both KC and DM shown to possess metabolic pathology, we have looked further into finding a relationship in the metabolic network of the two diseases in order to determine if this could be important into the hypothesized protective effect of DM on KC.

Discussion
The actual number of studies reported are few in number and have been conflicting. The original postulate was derived from the biomechanical differences that the two diseases possessed, where the authors concluded that DM stiffens while KC weakens the human cornea 7 . From a retrospective case-control study, Seiler et al. determined that DM had a statistically significant protective effect against KC. Another study reported a lower prevalence of KC diagnosis in DM patients when compared to a control population 6 . In contrast, Kuo and co-authors observed no differences in prevalence among diseased and control populations but did find a negative association between DM presence and KC severity 33 . A conflicting report came from Kosker et al. in 2014 where the authors noted a positive association in both prevalence and severity between KC and DM 34 . These studies highlight the complexity of the potential interplay between KC and DM in the context of human corneal microenvironment.
Numerous studies have associated KC with increased oxidative stress in vitro 16,35 and ex vivo 14,36 . Recent work has further suggested that alterations in levels of reactive oxygen species (ROS)-scavenging enzymes, such as superoxide dismutase, may contribute to defects in responding to endogenous and exogenous oxidative species by HKCs 37 . Moreover, the metabolic changes observed in our study, including an upregulation in aerobic glycolysis in HKCs characterized by increased lactate production, may correlate to altered ROS levels that in turn affects glucose metabolism 20,21 , though this trend appears absent or inversely regulated in T1DM and T2DM constructs. Though oxidative stress has been posited to play a fundamental role in both KC and DM [38][39][40] , we hypothesize that the metabolic phenotype found in HKCs may correspond to an inherent defect in metabolic regulation compared www.nature.com/scientificreports www.nature.com/scientificreports/ to the combined phenotypic and epigenotypic changes that occur following prolonged exposure to elevated glucose in the case of DM. Given that the abundance of glycolytic metabolites in T2DMs remains similar to control levels, our results suggest that the effects of DM on promoting CXL in situ may be more related to exogenous glucose levels than cytosolic flux. Thus, the fundamental differences in disease causation between KC and DM may partially explain the inverse regulation in basal cellular metabolism observed in vitro with HKCs exhibiting increased glycolytic metabolite levels compared to T1DMs. Variations between T1DM and T2DM were also evident in this study with higher expression of select metabolites in T2DM, including dimethyl-L-arginine and guanosine monophosphate; however, glycolytic metabolism appeared relatively consistent (excluding a single metabolite, fructose-6-phosphate, which was significantly higher in T1DM than T2DM). Our lab 22,23 has previously identified significant variations between ECM protein expression by corneal fibroblasts and corneal tissue in T1DM and T2DM. While our current study did not assess relative HbA1c levels in DM patients prior to tissue isolation, it likely that chronic glycemic levels and the degree of blood glucose control impact epigenetic markers of each cell type, thus contributing to differential cell phenotype detected in vitro.
Of importance, the metabolomics studies evaluated in this retrospective analysis were performed in euglycemic media, thus, further investigation of the differential responses of T1DMs and T2DMs challenged with a high glucose environment is justified. Furthermore, though a report suggesting that acute exposure to elevated glucose alone may not promote altered keratocyte marker expression, e.g. keratocan and lumican, by healthy corneal stromal stem cells in vitro 41 , stimulation with chronic hyperglycemia may promote a differential response in corneal fibroblasts isolated from DM patients due to permanent mitochondrial damage and increased ECM thickness 22 . We posit that the epigenetic changes that occur during prolonged DM 42,43 may give rise to the characteristic defects in corneal structure, including differences in ECM composition 22,23 . Further studies are required to elucidate the connection between altered collagen isoform secretion favoring a fibrotic phenotype characterized by increased Collagen III and changes in glucose metabolism in HKCs 16,21 and DM 22,23 , i.e. aerobic glycolysis versus TCA and the pentose phosphate pathway. These pathways may also play a significant role in the response to hyperglycemia by T1DMs and T2DMs. The mechanism by which this pathological change occurs during DM and more importantly, if this process can be reversed with pharmacological intervention, remains a significant question. Our study is the first to our knowledge to shed light on the metabolic similarities and differences of KC and DM to better define the observed biomechanical alterations that occur in the cornea.

Data Availability
The datasets generated during the current study are available from the corresponding author on reasonable request.