**4. Discussion**

Diabetic retinopathy causes both neural and vascular defects, with neural deficits preceding vascular changes [6,39–42]. Even before the onset of clinically detectable retinopathy, diabetic patients have a reduced ERG implicit time [43] and high-frequency flicker amplitude [44]. Later, they experience decreased vascular density [45]. In this study, we have shown that HFD feeding results in a suitable model of prediabetes, with the HFD cohort exhibiting insulin resistance and hypercholesterolemia without hyperglycemia. The retinopathy that is exhibited occurs over a slower time course than in T2D models, where both hyperglycemia and hyperinsulinemia exist.

The HFD mouse has previously been described as a model for T2D [7,46], as C57BL/6J mice fed HFD develop obesity and insulin resistance [47,48], but as we show in this study, this model has a distinct timeline and different characteristics than those seen in T2D. We show that HFD mice have hypercholesterolemia and insulin resistance but the absence of hyperglycemia, which is typical of T2D models.

In agreemen<sup>t</sup> with the literature, our study shows that mice fed a HFD have a sustained increase in body weight [6,49,50]. As confirmed by EchoMRI, the increase in body weight is primarily due to elevated body fat mass. After 12 weeks of feeding, HFD mice showed a two-fold increase in body fat mass over control LFD mice. Despite the marked increase in fat mass, HFD mice did not develop overt hyperglycemia. Glycated hemoglobin levels measured at 6 months and 12 months showed that both groups had normal glycated hemoglobin, thus indicating a key difference between the HFD model and other T2D rodent models, many of which are genetic. However, HFD mice develop hyperinsulinemia (Figure 1E,F), and their insulin production is sufficient to maintain euglycemia, as indicated by their glycated hemoglobin levels. The marked hyperinsulinemia we observed is supported by the literature [6,51–54]. In contrast, T2D in humans is characterized by not only insulin resistance but also the presence of sustained hyperglycemia and elevated HbA1c levels. When only insulin resistance is present, individuals are described as prediabetics [55,56].

Insulin resistance is believed to play a key role in diabetic neuropathy by increasing oxidative stress and mitochondrial dysfunction [57,58], and may also drive the early neural retinal dysfunction that we observed in our HFD mice. Thus, the HFD mice secrete sufficiently elevated insulin to maintain a normal glucose level, and as such the HFD model may be better characterized as a prediabetes model. Importantly, the incidence of prediabetes is often higher than that of diabetes [59]. The prevalence of prediabetes is also increasing; it is estimated that more than 470 million people worldwide will be suffering from prediabetes by 2030 [60]. Most importantly, the three classical microvascular complications, retinopathy, neuropathy, and nephropathy, have all been documented in individuals with prediabetes [61].

While classifications of diabetes remain "glucose-centric", our study draws attention to the importance of earlier events, when glucose levels are still normal. Thus, in our model, hyperinsulinemia with hypercholesterolemia will likely lead to the retinal pathology observed. Not surprisingly, these pathologies take a longer time to develop than those typically seen when hyperglycemia is also present.

Systemic and retinal lipid abnormalities have been shown to promote retinal damage [16,62,63]. Previously, we demonstrated that diabetes-induced disruption of the LXR axis results in abnormal lipid metabolism, inadequate vascular repair, and localized and systemic inflammation [16,64]. The LXRs (LXR α and LXRβ) play important roles in cholesterol homeostasis [65]. They regulate the expression of reverse cholesterol transporters [12]. Activation of LXRs using pharmacological agents repress inflammatory genes such as TNFα and IL-1β [66], inhibit the expression of pro-apoptotic factors [67], and prevent the development of DR [12]. We showed that use of GW3965, an LXR agonist, resulted in normalization of cholesterol homeostasis and repression of inflammatory genes, such as iNOS, IL-1β, ICAM-1, and CCL2 in the retina [16]. We found that inadequate cholesterol removal due to deficiency in LXR and reduced oxysterol production in the retina due to loss of cytochromes p450 27A1 and 46A1 resulted in widespread retinal pathology [68]. In the current study, we showed that concentrations of 40% fat in the diet were su fficient to reduce expression of LXR in the inner and outer nuclear layers.

Our study showed that HFD mice develop neural retinal deficits after 6 months of feeding, as both a-waves and b- waves were reduced under scotopic conditions. Unexpectedly, the a- and b- wave responses for LFD mice was significantly less after 12 months compared to the response after 6 months of feeding (*p* < 0.01 for both scotopic and photopic conditions), which suggests that the LFD may have detrimental e ffects on the neural retina. Because the composition of the diets must be isocaloric, when the amount of fat is reduced, some other dietary component needs to be increased to compensate. Inn the LFD, the amount of sucrose increases from 72 g to 354 g and 315 g of corn starch is also added so that the LFD can be isocaloric with the HFD. However, this largely occurs at the expense of making the diet high in carbohydrates. The literature supports that LFD may be detrimental [69–71]. While we were unable to find literature supporting the impact of LFD specifically on ERGs, the systemic consequences of LFD may indirectly a ffect the retina, for example by reduced availability of fat-soluble vitamins or changing retinal cholesterol metabolism. Moreover, the increased sucrose and cornstarch in the LFD may have direct deleterious e ffects [72,73]. LFDs promote insulin resistance, and while most of the research has been performed in humans, these findings may have relevance to murine studies. LFD, typically considered a high carbohydrate diet, is known to promote inflammation [74–76]. A recent study compared ERGs in HFD fed rats, Streptozotocin (STZ) rats and type 2 diabetes (T2D) rats at 6 months to controls. Kowluru found di fferences between the diabetic ERGs and controls, but no di fferences between the ERGs of the HFD rats compared to controls; however, Kowluru did not look at 12 month tests and the study was performed in rats, not in mice [77]. Thus, it is di fficult to compare these findings with our results.

While neural damage was detected at 6 months, the vascular damage was not observed until much later. This is in agreemen<sup>t</sup> with Rajagopal et al. [6], who demonstrated that vascular damage was not observed at 6 months of HFD feeding. However, despite the absence of vascular damage after 6 months of HFD, we observed the presence of "lipid-laden like" lesions, and also neural infarcts similar to what is described in humans as "cotton-wool" spots. These lesions, which appeared to increase as the retinopathy progressed in the HFD mice, could become a useful measure of retinal damage and may be sensitive enough to use as a novel endpoint for the preclinical investigation of therapeutic agents.

GFAP is normally expressed in retinal astrocytes in rodents; however, during stress and inflammation, Muller cells [37] respond by increasing GFAP expression. In this study, we show that WD induces GFAP expression in selective Muller cells, supporting the presence of increased stress and inflammation in the retina of these mice. Kim et al. have reported increased inflammation in other tissues such as adipose tissue and intestines [78]. Lee et al. showed increased numbers of activated macrophages in the retina of HFD mice [79]. In both humans and rodents, obesity-induced diabetes is associated with hypoxia in tissues such adipose tissue, and suppression of HIF-1 α mitigates tissue-specific pathological changes associated with HFD [80]. The liver, brain, kidney, and heart display tissue-specific regulation of HIF-1 α under systemic hypoxia [81]. After 6 months, but not after 3 months, we observed that HIF-1 α expression is increased in the WD retinas compared to LFD controls. Similar to our observation in the retina of 3-month-old mice on WD, Prasad et al. showed the absence of pimonidazole staining in the kidneys of 10–11-week old db/db mice [82], also indicating the absence of hypoxia response in the kidneys at this time point.
