Embryonic hyperglycemia perturbs the development of specific retinal cell types, including photoreceptors

Diabetes is a growing epidemic, affecting 34.2 million people in the US in 2018 (DHHS, 2020; https://www.cdc.gov/diabetes/pdfs/data/statistics/national-diabetes-statistics-report.pdf). High blood sugar (hyperglycemia) is the primary symptom of diabetes and when it becomes a recurring state, various complications are likely to arise which affect tissues all over the body (Deshpande et al., 2008). Complications include neuropathies (nerve damage), nephropathy (kidney disease), stroke and retinopathy, which can cause progressive blindness (Ghaseminejad et al., 2020). Although these complications are primarily documented in adults who have experienced recurring hyperglycemia over many years, hyperglycemia during pregnancy carries its own set of complications, which can have long lasting effects on the offspring (Bianco and Josefson, 2019; Hachisuga et al., 2015; Zhao et al., 2017).

During pregnancy, glucose passes through the placental barrier from the mother to fetus; this maternally supplied glucose is necessary for fetal development but in excess it can also result in embryonic hyperglycemia. Maternal hyperglycemia may come from existing diabetes prior to pregnancy and/or increased insulin resistance developed during pregnancy to allow for increased glucose to pass through the placenta, which is essential for stimulating fetal insulin production to aid in growth (Sonagra et al., 2014). Excess insulin resistance can lead to gestational diabetes which is not diagnosed until 24 weeks into pregnancy (and occurs in up to 10% of US pregnancies; Centers for Disease Control and Prevention, 2019). The type and severity of hyperglycemia-related complication varies greatly, depending on when embryonic hyperglycemia occurs. The most prominently studied complication of embryonic hyperglycemia is heart malformation or congenital heart defect (CHD). A wide array of phenotypes are associated with CHD, and it is considered the most common birth defect associated with diabetic embryopathy (Basu and Garg, 2016). However, other developing tissues and organs are also vulnerable to the effects of embryonic hyperglycemia, and these have been less well studied. A previous study found that offspring of mothers with diabetes displayed significantly thinner inner and outer maculas as well as lower macular volume (Tariq et al., 2010). In humans, the macula contains the highest density of cone photoreceptors, suggesting a potential deleterious effect of diabetic pregnancy on retinal cone photoreceptor development. Given these data and the strong connection between diabetes and retinal degeneration leading to progressive blindness in adults, there is a critical need to study the effects of hyperglycemia on the developing retina during embryogenesis, using an animal model where the developing eye is easily accessible.

Zebrafish have recently become a favorable model for studying hyperglycemia and diabetes due to their relatively easy maintenance, high fecundity and manipulatable environment. Diabetes and hyperglycemia can be induced via ablation of pancreatic β-cells through streptozotocin injection (Intine et al., 2013) or whole-body immersion in glucose-dense fish water (Gleeson et al., 2007). Using the immersion technique with adult zebrafish, it has been demonstrated that recurring hyperglycemia results in a reduction in the number of cone photoreceptors in the retina, with remaining photoreceptors displaying an abnormal morphology, including shortened outer segments (Alvarez et al., 2010) as well as abnormal electroretinogram responses (Tanvir et al., 2018). Recent studies have utilized a genetic mutant to induce hyperglycemia (pdx1^−/−), which does not properly produce insulin. Without insulin, glucose is not metabolized as needed, leading to an increase in free glucose in the bloodstream and subsequent hyperglycemia. Characterization of the pdx1^−/− mutants showed that recurring hyperglycemia in adulthood resulted in photoreceptor degeneration, defective visual responses (Ali et al., 2020), and increased retinal angiogenesis, a hallmark of diabetic retinopathy (Wiggenhauser et al., 2020). Although these studies show the utility of zebrafish for the study of the ocular complications of hyperglycemia in adults, there has been less work on the effects of hyperglycemia during embryonic and larval retinal development. One recent study suggested that exposure to very high levels of exogenous glucose causes a decrease in retinal ganglion cells and Müller glia, as well as an increase in vasculature leakage in zebrafish larvae (Singh et al., 2019), providing evidence that embryonic hyperglycemia has deleterious effects on retinal development. However, the consequences of embryonic hyperglycemia specifically on photoreceptor development have not been closely examined. Given that there is mounting evidence that photoreceptors, which are highly metabolically demanding cells, are major contributors to the progression of diabetic retinopathy, there is a pressing need to study how hyperglycemia may impact photoreceptors during embryonic development.

In this study, we specifically explored the consequences of hyperglycemia on cell type differentiation in the developing zebrafish retina, using two complementary approaches. To model chronic hyperglycemia via lack of insulin production, similar to what is observed in type I diabetes, we utilized pdx1^−/− zebrafish, which possess a null mutation in a gene necessary for β-cell development, and therefore cannot produce insulin (Kimmel et al., 2015). A recent study showed pdx1^−/− zebrafish larvae have microvascular changes in the ocular hyaloid vasculature at 6 days post fertilization (dpf) (Wiggenhauser et al., 2020); however, an examination of photoreceptor and retinal development has not yet been reported for this mutant. To model hyperglycemia that is representative of type 2 diabetes, we developed a nutritional model in which zebrafish embryos are exposed to exogenous glucose and dexamethasone from 10 h post fertilization (hpf), just prior to optic vesicle evagination, until 5 dpf, when retinal development is largely complete. Dexamethasone is a synthetic glucocorticoid that is used in combination with glucose to elevate whole-body glucose due to its ability to stimulate gluconeogenesis and disrupt glucose transport (Tamez-Pérez, 2015). In the context of embryonic development, dexamethasone is often given to pregnant women who are at risk of preterm birth to aid in fetal lung development, and it is known that low birth weight and preterm babies are highly susceptible to hyperglycemia (Hays et al., 2006).

Here, we report that, in both genetic and nutritional models of embryonic hyperglycemia, rod and cone photoreceptor cells are significantly decreased in number, and retinal oxidative stress is increased. Notably, embryonic hyperglycemia was associated with abnormal visual behavior at 5 dpf. Additionally, the timing of retinal progenitor differentiation was altered in hyperglycemic larvae, and cone photoreceptor number remained lower than in the controls even after a return to normoglycemic conditions. These findings provide evidence that embryonic hyperglycemia impedes proper retinal development, leading to short-term, and potentially long-term, visual defects.

Prior studies of pdx1 mutant zebrafish have demonstrated that at 5 dpf, mutant larvae have elevated whole-body glucose, aberrant hyaloid vasculature, and grow to be significantly smaller compared to their wild-type and heterozygous siblings (Kimmel et al., 2015; Wiggenhauser et al., 2020). Using the pdx1^sa280 mutant line described by Kimmel et al., we found that pdx1 mutants displayed elevated whole-body glucose levels as early as 4 dpf (data not shown) and confirmed that they have significantly elevated whole-body glucose at 5 dpf (Fig. S1A). We measured whole-body glucose rather than blood glucose because the total blood volume of a 5 dpf zebrafish larvae is less than a microliter (Van Wijk et al., 2019). We measured pdx1 mutant eye size and found that it was proportional to their body size when compared with wild-type larvae, indicating that the mutation and accompanying hyperglycemia does not cause microphthalmia (Fig. S1B–D). However, further analysis at the cellular level revealed interesting abnormalities in the developing retina.

To determine whether hyperglycemia impacts photoreceptor development, we imaged and quantified red+green cone photoreceptors at 5 dpf in wild-type and pdx1^−/− retinas, using the Zpr1 antibody. Red/green cones displayed a significant decrease in number (Fig. 1B,C) in hyperglycemic compared to wild-type retinas (Fig. 1A). In the ventral region, cones possessed stunted outer segments and thinner cell bodies (Fig. 1B′) than those in wild-type larvae (Fig. 1A′). To image and quantify rods, we crossed heterozygous pdx1^sa280 adults onto the XOPS:GFP transgenic background, in which rod photoreceptors are fluorescently labeled (Fadool, 2003). We observed a significant decrease in the number of rod photoreceptors of pdx1^−/− larvae at 5 dpf compared to wild-type and heterozygous siblings (Fig. 1F). Looking closely at the rod photoreceptors in the ventral retina, the outer segments appeared to be much shorter and thinner (Fig. 1E′, arrowheads) when compared to wild-type or heterozygous retinas (Fig. 1D′). The dorsal retina also contained fewer rod photoreceptors in mutants (Fig. 1E) compared to wild-type (Fig. 1D).

Quantification revealed a significant decrease in both rod and cone photoreceptors in pdx1 mutants at 5 dpf (Fig. 1C,F). The average number of red/green double cone photoreceptors was reduced by over 20%, which was particularly striking in the ventral portion of the retina (average of 13/50 µm in wild-type or heterozygous versus 9/50 μm in pdx1^−/− mutants), and the average number of rods was decreased by 45% (average of 5.5/50 µm in wild-type or heterozygous versus 3/50 µm in pdx1^−/− mutants). Taken together, we conclude that hyperglycemic pdx1 mutants have significantly fewer photoreceptors than do normoglycemic larvae of the same age, and the photoreceptors that are present in pdx1 mutant retinas display an abnormal morphology.

Although the pdx1 mutant provides insight into hyperglycemic phenotypes resulting from a genetic defect in insulin producing cells, we also wanted to determine whether embryonic and larval hyperglycemia induced by altered nutrient uptake had similar effects on retinal development. With this approach, we can model the effects of a direct, in utero, exposure to elevated glucose, such as might occur in cases of maternal diabetes, and which involves excess glucose flow across the placenta (Desoye and Nolan, 2016). To that end, we submerged zebrafish embryos in fish water containing glucose with or without dexamethasone from 10 hpf (just prior to optic vesicle evagination from the forebrain) until 5 dpf (when retinal development is largely complete; experimental workflow shown in Fig. S2A). To avoid nonspecific effects of high glucose on embryo development, we conducted a series of dose responses for glucose concentrations and selected 50 mM, as it was the lowest concentration that resulted in whole-body glucose elevation. This is consistent with results from a previous study, which used 55 mM glucose treatments to induce hyperglycemia and which resulted in abnormal vasculature development (Jörgens et al., 2015). Other studies involving a glucose submersion technique with zebrafish to induce hyperglycemia have used higher glucose concentrations, which ranged from 110 mM to 277 mM (Singh et al., 2019; Tanvir et al., 2018).

Treatment of zebrafish embryos with glucose alone produced a significant increase in whole-body glucose in comparison to that in untreated embryos and embryos exposed to mannitol (osmolarity control; Fig. S2B); however, we found that exposure to glucose alone resulted in highly variable levels of hyperglycemia. To combat this, we added dexamethasone to the glucose treatment. Dexamethasone is a synthetic glucocorticoid that has been shown to disrupt glucose transport into cells, preventing proper breakdown of glucose and leading to increased free glucose in the blood (Garvey et al., 1989). Dexamethasone is also frequently given antenatally to facilitate fetal lung maturation (McGoldrick et al., 2020). Our results show the combination of glucose and dexamethasone treatment provides a much tighter range of significantly elevated whole-body glucose values (Fig. S2B). Additionally, dexamethasone treatment alone did not significantly affect whole-body glucose in comparison to untreated and mannitol controls (Fig. S2B).

When comparing the gross morphology of larvae from the different treatment groups, we observed some variation in body length as well as yolk and eye size. However, when eye size was normalized to body size, there was no significant difference in ocular proportions across groups, indicating that our experimental conditions do not cause microphthalmia (Fig. S2C,D). Moreover, using a fluorescent glucose analog (2-NDBG), we confirmed that exposure to exogenous glucose leads to glucose uptake in the eye (Fig. S2E). Taken together, our results show that a combination of glucose and dexamethasone exposure reliably produces hyperglycemia in zebrafish larvae. From this point forward, we will mostly present data from the mannitol and glucose plus dexamethasone (glucose+dex) treatment groups.

With the establishment of a nutritional model to compare to the pdx1 mutant, we conducted similar analyses on photoreceptor development. Progeny of XOPS:GFP and TαC:eGFP (a transgenic line that fluorescently labels all cone photoreceptors; Kennedy et al., 2007) adult in-crosses were used in treatments to assess photoreceptor number and morphology. Similar to what was seen with the pdx1 mutant, we observed a significant decrease in numbers of both cone and rod photoreceptors of hyperglycemic larvae at 5 dpf (Fig. 2C,F). Cone photoreceptors were decreased by 23% in glucose-treated and 28% in glucose+dex-treated retinas compared to those treated with the mannitol control (Fig. 2A′,B′,C), whereas rods were decreased by 34% in glucose-treated and 45% in glucose+dex-treated retinas (Fig. 2F). Compared to pdx1 mutants, the decrease in cones was larger in the nutritional model, which may be because the TαC:eGFP transgene used for these experiments labels all cone photoreceptor subtypes (red, green, blue and UV), whereas only red/green cones were detected by the Zpr1 antibody used with the pdx1 larvae. In contrast, the reduction in rods was comparable in the glucose+dex-treated larvae and pdx1 mutant larvae. Confocal microscopy of whole eyes from control and hyperglycemic larvae revealed a large decrease in the number of rod photoreceptors in hyperglycemic compared to in mannitol-treated larvae (Fig. 2D,D′,E,E′). In addition, the confocal images showed that, similar to what was seen in the pdx1 mutant retinas, the outer segments of both cone and rod photoreceptors appeared stunted compared to controls (Fig. 2A,A″,C,C″), with wider inner segments (Fig. 2A′,A″,B′,B″,D′,D″,E′,E″). Taken together, our results show that embryonic hyperglycemia induced by exogenous glucose results in a reduction of photoreceptors, similar to what is noted in pdx1 mutants, further supporting a significant impact of embryonic hyperglycemia on photoreceptor development.

Our results from both models indicate that embryonic hyperglycemia causes photoreceptor defects in the larval retina, specifically in overall number and in outer segment morphology. To further assess outer segments, we used the Zpr3 antibody, which specifically labels the outer segments of rod photoreceptors and the Rh2-expressing member of the double cones (Yin et al., 2012). The results confirmed that rods and cones from hyperglycemic larvae have shorter outer segments compared to those in wild-type larvae at 5 dpf (Fig. 3A,B,D,E). Histological staining with hematoxylin and eosin (H&E) on control and hyperglycemic retinal sections also supports this finding (Fig. S3A–F). The purpose of outer segments is to capture light and, via the phototransduction cascade, convert it into an electrical signal to be sent through the retina to the brain (Baker and Kerov, 2013). Both pdx1 mutant and glucose+dex-treated larvae displayed abnormally short photoreceptor outer segments. Without full elongated outer segments, we hypothesized that hyperglycemic larvae may experience subtle visual defects (Lessieur et al., 2017). Therefore, we performed an optokinetic response (OKR) assay to measure the larval visual response. Comparison of the number of ocular saccades per minute across pdx1 genotypes revealed a significant reduction of eye movements in the pdx1 mutants compared to their wild-type and heterozygous siblings (Fig. 3C). Similarly, glucose+dex-treated larvae also showed a significant decrease in OKR performance at 5 dpf compared to untreated and mannitol-treated larvae (Fig. 3F). Glucose-treated larvae showed a wide range in response, reflective of the variability in their photoreceptor number and morphology. Together, these results show that the photoreceptor defects observed in hyperglycemic larvae are associated with reduced visual responses. Although we cannot exclude the possibility that the reduced OKR is due to abnormalities in the extraocular muscles or other sensory deficits, the photoreceptor phenotypes we documented support a connection to the visual behavior defects. Moreover, the similarity in phenotype between pdx1 mutants and glucose+dex-treated larvae, with respect to photoreceptor number, photoreceptor morphology, and visual acuity, indicates that these phenotypes are due to hyperglycemia. Therefore, we next wanted to delve into the mechanisms by which these phenotypes arise.

Hyperglycemia has been shown to induce cell death via apoptosis and autophagy as well as to increase susceptibility to necrosis (Lévigne et al., 2013; Tang et al., 2014; Volpe et al., 2018). Therefore, we wanted to quantify and compare programmed cell death in control and hyperglycemic larvae. Using the TUNEL assay we detected apoptotic cells in retinal sections at 5 dpf. We found an increase in apoptotic cells with both the genetic (Fig. S4C) and nutritional models, but only the nutritional model showed a statistically significant increase (Fig. S4F). Cell death was primarily observed in the inner nuclear layer (INL) and to a lesser extent in the outer nuclear layer (ONL) of hyperglycemic retinas (Fig. S4B,E), but not controls (Fig. S4A,D). Looking at additional timepoints leading up to 5 dpf, we found that cell death began to significantly increase at 4 dpf (Fig. S4AE), which aligns with the first time point at which we see elevation in whole-body glucose in pdx1 mutants and glucose+dex-treated larvae. Although there was a significant increase in apoptosis in hyperglycemic retinas, the total number of TUNEL+ cells per retinal section was low across all treatments and timepoints (Fig. S4G–AE).

Sections of control and hyperglycemic GFAP:GFP (a transgenic line in which Müller glia are GFP tagged) larvae revealed an increase in average number of Müller glia in glucose-treated larvae (+17%) at 5 dpf and a slight increase in the glucose+dex-treated larvae (+11%) compared to untreated and mannitol-treated controls (Fig. 4G). Looking at the distribution of Müller glia within the inner nuclear layer, glucose+dex-treated larvae displayed an irregular localization of Müller cell bodies (Fig. 4F) compared to the linear pattern of cell bodies in mannitol- and glucose-treated retinas (Fig. 4B,D). Moreover, whereas Müller glial cell bodies possessed the expected compact polygonal shape in mannitol- and glucose-treated retinal sections, the Müller glial cell bodies of glucose+dex-treated retinas appeared swollen, twisted or heart-shaped, which could represent signs of reactive gliosis (Fig. 4F). For a better understanding of Müller glia morphology, we utilized whole-mount confocal microscopy to visualize Müller glia in 3-D space. We found the increase of Müller glia in glucose-treated larvae was evident throughout the eye (Fig. 4C,C′), while the shape and size of Müller glia were comparable to mannitol-treated retinas (Fig. 4A,A′). Whole-mount imaging of glucose+dex-treated larvae revealed a variety of abnormal cell body shapes throughout the eye (Fig. 4E,E′), which were also larger than those for the glucose- and mannitol-treated groups. This finding was supported by using image analysis to calculate the ratio of regular or oval-shaped to total Müller cell bodies across treatments (Fig. 4H). Finally, a western blot revealed a significant increase in GFAP expression in glucose+dex-treated larvae (Fig. 4I,J). Together, these results indicate a gliotic response in the glucose+dex-treated retinas at 5 dpf. We also imaged retinal sections from 5 dpf pdx1 larvae that were crossed onto the GFAP:GFP background (Fig. 4J,L). Unlike glucose- and glucose+dex-treated larvae, pdx1 mutants did not exhibit an increase in the number of Müller glia (Fig. 4M). Interestingly though, we did find that Müller glia cell bodies in pdx1^−/− retinas were significantly larger than their wild-type and heterozygous counterparts (Fig. 4N) and displayed a variety of abnormal shapes similar to those in glucose+dex-treated larvae, with some disorganization in patterning (Fig. 4L,O). To determine whether there was an increase in Müller glial proliferation in hyperglycemic retinas, larvae were exposed to EdU at 5 dpf. We did not observe any colocalization of EdU-positive cells with Müller glia (Fig. S7A–E), indicating that the reactive gliosis in hyperglycemic retinas is not accompanied by significant cell proliferation. Taken together, our results suggest that embryonic hyperglycemia induces a gliotic response within the larval retina. Furthermore, exogenous exposure to glucose causes a slight increase in Müller glial number that is not observed in pdx1 mutant retinas, suggesting that this effect may be due to elevated levels of insulin signaling in our nutritional model.

To determine whether embryonic hyperglycemia resulted in altered numbers of other retinal neurons, we performed immunohistochemistry with antibodies that label various retinal cell types. Horizontal cells are located in the outermost part of the inner nuclear layer, and synapse directly to photoreceptors. We detected horizontal cells with the Prox1 antibody, which also labels progenitor cells in the INL. Horizontal cells were identified by their location and distinct oblong shape. There was a 30% decrease in the number of Prox1-positive horizontal cells per 100 µm in pdx1 mutant retinas compared with wild type (Fig. S5A–C). In the nutritional model, both glucose- and glucose+dex-treated embryos displayed a 40% and 50% decrease, respectively, in horizontal cell number (Fig. S5J–L). Bipolar cells, which also synapse directly with photoreceptors and transfer the visual signal from photoreceptors to ganglion cells (Kaneko, 1983), did not show a difference in average number per 100 µm when comparing wild-type to pdx1 mutants (labeled with PKCα antibody; Fig. S5D–F) or in glucose- and glucose+dex-treated larvae compared to controls (Fig. S5M–O). Next, we quantified the number of ganglion and amacrine cells, using the HuC/D antibody. There was no significant difference in amacrine or ganglion cell number across genotypes (Fig. S5G–I) or treatments (Fig. S5P–R). Additionally, we did not find any significant changes in number or morphology of retinal cell types with dexamethasone treatment alone (Fig. S5S–AA). These results suggest that abnormal retinal cell type differentiation in hyperglycemic larvae is limited to photoreceptors and horizontal cells, with the remaining inner retinal neurons developing in normal numbers.

Hyperglycemia has been shown to induce increased levels of reactive oxygen species (ROS), unstable oxygens that cause DNA damage and cell death (Yan, 2014). A recent study has shown that ROS production is a key component of retinal development through regulation of a switch between cellular proliferation and differentiation (Albadri et al., 2019). Furthermore, studies in adult mice and retinal explants have shown that recurring hyperglycemia leads to an increase in ROS production, specifically in photoreceptors (Bhatt et al., 2010; Du et al., 2013). To determine whether ROS production was increased in the retinas of hyperglycemic larval zebrafish we used Mitosox, an in vivo probe for superoxides. We found a trend towards increased ROS production in both glucose- and glucose+dex-treated embryos at 48 hpf (P=0.10), a key time point for the onset of photoreceptor differentiation (Fig. 5A,G) but not in pdx1 mutant larvae (Fig. 5H–L). By 5 dpf there was a significant increase in ROS in pdx1 mutants and the glucose+dex group (Fig. S6B,D,I,J). Although the ROS probe signal was observed at various places throughout the head, at 48 hpf it was especially prominent in the eyes of hyperglycemic larvae, and within the retina was detected at the border of the future ciliary marginal zone (Fig. 5C,F′). Retinal sections at 5 dpf showed the ROS in the retina to be specifically located among the outer segments of photoreceptors (Fig. S6F,H, arrows). In contrast, wild-type and untreated larvae had long straight outer segments that did not show ROS probe colocalization (Fig. S6E,G).

To get a better idea of the metabolic factors specifically impacted by embryonic hyperglycemia, we examined whether expression of enzymes involved in glucose metabolism was altered using a glucose probe array at 5 dpf. Interestingly, we found that expression of key glucose metabolism enzymes connected to ROS production, such as succinate dehydrogenase and glucose-6-phosphatase, were upregulated in the heads of glucose- and glucose+dex-treated larvae as well as pdx1 mutants compared to untreated and wild-type controls (Fig. S6K,L). The dysregulation of these enzymes can have detrimental effects on cellular metabolism, particularly in the context of hyperglycemia (Yan, 2014).

We next sought to understand whether hyperglycemia was affecting the cell cycle and differentiation of retinal cells during development. Previous work has suggested embryonic hyperglycemia suppresses the cell cycle via altered expression of cyclin D1 and p21 (CDKN1A), reducing cell proliferation and differentiation (Scott-Drechsel et al., 2013). As mentioned above, ROS production has also been shown to regulate cell cycle exit in the developing retina (Albadri et al., 2019). To investigate the timing of retinal progenitor differentiation in our two models of developmental hyperglycemia, we tracked the fate of cells which were proliferating from 48–50 hpf. During this window of retinal development, photoreceptor progenitors are beginning to exit the cell cycle and differentiate. Ganglion and amacrine cells have mostly differentiated before 48 hpf, leaving primarily early photoreceptors and later neurons of the INL to differentiate in the targeted time window (Stenkamp, 2007). We exposed developing control and hyperglycemic embryos to EdU from 48–50 hpf and collected larvae at 5 dpf to visualize the fate of the EdU-positive cells based on their location in the mature retina. We observed that the majority of EdU-positive cells were located in the INL across all treatments and genotypes, which aligns with known developmental timing (Fig. 6A,B,F,G,J). For untreated, mannitol-treated and wild-type larvae, there was also a large proportion of EdU-positive cells among the photoreceptor nuclei of the ONL, indicating that, as expected, many of the proliferating RPCs at 48–50 hpf ultimately differentiated into photoreceptors (Fig. 6A,B,F,G,H,J). Very few EdU-positive cells were in the ganglion cell layer at 5 dpf (Fig. 6F,J), consistent with the majority of retinal ganglion cells having already exited the cell cycle at the time of EdU exposure. In contrast, in hyperglycemic retinas, the second highest concentration of EdU-positive cells was observed in the ganglion cell layer (Fig. 6C,D,F,H,J), which suggests that, unlike in the controls, retinal ganglion cells had not yet exited the cell cycle at the time of EdU administration. To verify this phenotype was due to a delay in retinal ganglion cell differentiation (as opposed to an increase in displaced amacrine cells in the ganglion cell layer), we immunolabeled for amacrine cells using the 5E11 antibody, but did not detect any amacrine cells in the ganglion cell layer (Fig. S7F–J). Moreover, there were significantly fewer EdU-positive cells in the ONL of hyperglycemic retinas at 5 dpf compared to in controls, and the intensity of the EdU signal was also lower in cells of the ONL, suggesting that these cells were derived from retinal progenitors that had gone through more cell divisions than their counterparts in control retinas. Intriguingly, the total number of EdU-positive cells was significantly increased in pdx1 mutants compared to their wild-type siblings, but this was not observed in glucose- or glucose+dex-treated larvae (Fig. 6E versus I). This is the first phenotype to be strikingly different between the genetic and nutritional model, suggesting that there could be a differential role of insulin in RPC proliferation and differentiation. Taken together, these data indicate that embryonic hyperglycemia causes a delay in cell cycle exit and differentiation of RPCs. Since we did not observe a difference in ganglion, amacrine and bipolar cell number between control and hyperglycemic retinas at 5 dpf, these cell types must ‘catch up’ to normoglycemic levels of differentiation by 5 dpf. However, hyperglycemia appears to have a prolonged impact on the number of retinal photoreceptors and horizontal cells.

To determine whether reducing ROS could rescue photoreceptor number in hyperglycemic retinas, we first treated the nutritional model and pdx1 embryos with superoxide dismutase but had difficulty establishing a treatment concentration that did not induce harmful side effects (data not shown). Next, we treated hyperglycemic embryos with the antioxidant Methylene Blue (Rojas et al., 2012). Methylene Blue co-treatment produced a modest increase in cone photoreceptors in glucose-treated as well as glucose+dex-treated larvae compared to no co-treatment (data not shown). Finally, we utilized diphenyleneiodonium (DPI), which is an NADPH oxidase inhibitor (Niethammer et al., 2009). Treatment of hyperglycemic larvae with DPI significantly increased the number of rod photoreceptors, detected by immunolabeling with the 4C12 antibody (Fig. S8A–G). The increase in photoreceptors after antioxidant treatment supports our hypothesis that ROS production contributes to the decrease in photoreceptors in hyperglycemic retinas.

The EdU pulse-chase experiment revealed a delay in retinal cell type differentiation, which resulted in a decrease of photoreceptors at 5 dpf. To evaluate whether larvae could recover to produce photoreceptors at equivalent numbers to controls, we placed hyperglycemic larvae back into normal fish water at 5 dpf and collected them seven days later at 12 dpf. We examined retinal sections by immunohistochemistry and found that red/green cone photoreceptors were still significantly reduced in glucose+dex-treated larvae compared to controls (Fig. 7A–C,G). Similarly, the number of rod photoreceptors was also still significantly reduced in glucose+dex-treated retinas compared to in controls (Fig. 7D–F,H). The persistent reduction in both photoreceptor types is indicative of potential long-term consequences of hyperglycemia on retinal cell maintenance and vision. This is particularly important considering the susceptibility of photoreceptors to damage in adult models of hyperglycemia.

Diabetic retinopathy is a prevalent complication of diabetes and a principal cause for acquired blindness in adults (Centers for Disease Control and Prevention, 2014). A growing body of research has demonstrated that the degeneration of photoreceptor cells actually precedes the retinal vasculature defects in diabetic retinopathy (Marcovecchio et al., 2011). This underscores that the photoreceptors are a critical focal point for initiating retinal pathology as a result of chronic hyperglycemia. As the population of people with diabetes rapidly grows, research efforts must also be focused on studying the consequences of diabetes for pregnant mothers and the retinal development of their offspring.

Although there is abundant research on heart development in hyperglycemic embryos, little work has focused on the retina specifically. In humans, one study showed that offspring of diabetic pregnancies had significantly thinner inner and outer macula as well as lower macular volume, suggesting the potential for harmful effects of diabetic pregnancy on retinal cone photoreceptor development (Tariq et al., 2010). These data highlight the need to visualize the effects of embryonic hyperglycemia on photoreceptor number and morphology in more detail. Our study can fill this gap by combining whole-mount imaging and retinal sections to visualize, quantify and assess the morphology of retinal cells that have developed under hyperglycemic conditions. This is the first study to characterize the development of photoreceptors specifically across two models of embryonic hyperglycemia. One of the most striking findings was the decrease in photoreceptor number in hyperglycemic larvae, with the remaining photoreceptors displaying stunted outer segments – these findings were observed in both nutritionally and genetically induced hyperglycemic models, and in both cases were associated with a reduced optokinetic response compared to controls. Interestingly, although the photoreceptors were reduced across the entire hyperglycemic retina, photoreceptor outer segments near the marginal zone (the site of post-embryonic neurogenesis in zebrafish retina) appeared somewhat more normal in length than those in the central retina (e.g. Fig. 3E). This could suggest that photoreceptors produced from the margin initially differentiate normally, but then succumb to degenerative changes as they age. Further structural analysis in different portions of the retina could help us to better understand differentiation versus maintenance issues induced by hyperglycemia.

Our results demonstrate how embryonic hyperglycemia affects the retina short term, but we were also interested in long-term effects. Looking at retinal sections of treated embryos from the nutritional model at 12 dpf, we found that both rod and cone photoreceptors continued to be reduced in number even after the larvae were returned to normoglycemic conditions. This indicates an issue in photoreceptor programming, metabolism and/or maintenance, which may result in longer-term visual defects as well as potential susceptibility to degeneration. Considering photoreceptor degeneration is an early consequence of persistent hyperglycemia in adult zebrafish, it will be important to study how photoreceptors in zebrafish that experienced embryonic hyperglycemia respond to a ‘second hit’ of hyperglycemia as adults. Furthermore, understanding how embryonic hyperglycemia affects the retina both short and long term is imperative to identifying and timing therapeutics during development to prevent lasting vision problems.

In addition to photoreceptor phenotypes, a recent study utilizing zebrafish to study retinal development showed that exposure to high levels of glucose causes a decrease in retinal ganglion cells and Müller glia as well as an increase in infiltrating macrophages (Singh et al., 2019). That study utilized a different time course of submersion and a much higher glucose concentration than our study. These differences in methodology may explain our discrepant findings in Müller glia cell number, with our model showing not only a slightly higher number of Müller glia in hyperglycemic zebrafish, but also an abnormal morphology suggestive of reactive gliosis. We also did not observe a reduction in retinal ganglion cell number in either model, which could be due to the lower glucose concentration used, or to different methods to visualize the ganglion cells.

Given the abnormal Müller glia morphology and the elevated cell death in the inner nuclear layer of hyperglycemic retinas, it is possible that reactive gliosis is a consequence of the decrease in photoreceptors or the altered metabolic environment, or a combination of both considering the role and needs of Müller glia in the retina. Interestingly, the only other class of retinal neuron that showed differing numbers or morphology in our study were the horizontal cells, which were reduced in pdx1 mutants, and glucose-treated and glucose+dex-treated larvae. The reduction in horizontal cells might be directly related to the reduction in photoreceptors, as previous studies have shown that horizontal cells play a role in cone/rod distribution and synaptogenesis (Messersmith and Redburn, 1990).

Considering the link between hyperglycemia, ROS production and cell death, we hypothesized that the decrease in photoreceptors could be due to apoptosis. Although we did observe an increase in apoptotic cells in hyperglycemic larvae, the TUNEL+ cells were not localized to the ONL where photoreceptors reside. Rather, most of the TUNEL+ cells were found in the INL. The TUNEL+ cells in the INL could represent Müller glia phagocytosis of dying photoreceptors in hyperglycemic retinas, as has been observed in other models of photoreceptor degeneration (Bailey et al., 2010; Bejarano-Escobar et al., 2017; Morris et al., 2005), which might also account for the apparent gliotic response. Alternatively, they could be due to engulfment of dying cells by microglia. In any case, because the overall number of TUNEL+ cells was relatively low at all time points, it is likely that additional factors contribute to the decrease in photoreceptor number in response to hyperglycemia, such as altered photoreceptor cell differentiation.

A recent study by Albadri et al. (2019), showed that retinal cell developmental delay can be induced by lipid peroxidation products, such as 4-hydroxynonenal (4-HNE), which downregulate HDAC1 to keep cells in a proliferative state, preventing differentiation (Testa et al., 2017). Production of ROS, such as hydrogen peroxide, is an upstream event leading to lipid peroxidation, and in human endothelial cell lines modeling hyperglycemia (Rohowetz et al., 2018). The link between photoreceptors and ROS production has yet to be elucidated in terms of mechanism. Given our observation of increased ROS in hyperglycemic retinas at 48 hpf and 5 dpf, we hypothesize that genetic or nutritionally induced embryonic hyperglycemia results in increased ROS production, which might cause retinal cell differentiation delay and abnormal photoreceptor morphology.

Indeed, utilizing EdU ‘birth-dating’, we found that hyperglycemic larvae exhibited a striking deviation from the expected timing of RPC cell cycle exit during what should have been the window of photoreceptor differentiation. The significant increase in EdU-positive cells in the retinal ganglion cell layer indicates a lag in retinal cell type differentiation in hyperglycemic larvae that is likely a major contributor to the decreased number of photoreceptors. In pdx1 mutants, we also found a significant increase in the overall number of EdU-positive cells that was not observed in our nutritional model, which indicates loss of pdx1 specifically results in an increase in RPC proliferation. Whether this is related to the lack of insulin-producing cells in pdx1 mutants warrants further investigation. Importantly, although RGCs and other retinal neurons eventually overcome this delay and differentiate in comparable numbers to control retinas at 5 dpf, photoreceptor numbers remain reduced relative to controls even after an additional period of normoglycemic conditions. This result suggests that whereas most retinal neurons exhibit adaptive plasticity in their developmental timing, photoreceptors might be particularly sensitive to early metabolic derangement. Taken together, our data now connect embryonic hyperglycemia with increased ROS production, increased RPC proliferation and extended perturbations to photoreceptor differentiation.

Given our ROS data in the nutritional model, we tested whether an antioxidant (DPI) could rescue photoreceptor number in hyperglycemic larvae. We found that addition of DPI in our nutritional model increased the number of rod photoreceptors, with outer segment morphology that better resembled controls (Fig. S8F). Interestingly, the density of rods in normoglycemic retinas exposed to DPI was reduced compared to controls (Fig. S8D), indicating that normal photoreceptor development might require a baseline level of ROS. We hypothesize that a combination of antioxidants and ROS inhibitors with a more targeted delivery method could be necessary to fully rescue the decrease in photoreceptors specifically. Furthermore, analysis of the particular forms of ROS as well as other reactive compounds that are produced in response to hyperglycemia is needed to identify better targets for pharmacologic intervention.

In a hyperglycemic state, there is an influx of glucose undergoing glycolysis, which can alter the expression of various enzymes in the glycolytic pathway and have further downstream effects (Lund et al., 2019; Kanow et al., 2017; Yan, 2014). Using a glucose probe array, we found a significant increase in enzymes that play important roles in NADH production that feeds into the electron transport chain (ETC) where it is oxidized at complex I. When excess NADH is fed into the ETC, superoxide production is increased, which we noted in our hyperglycemic larvae. Superoxide can induce DNA damage (Keyer and Imlay, 1996) and delays in cell differentiation. Given our results, we created a model that connects the metabolic effects of hyperglycemia to the photoreceptor phenotypes we observed (Fig. 8). We propose that hyperglycemia induces delayed photoreceptor differentiation early in retinal development, while a gliotic response is induced later via ROS production and cell death. It is also possible that hyperglycemia contributes to the photoreceptor and Müller glia phenotypes via indirect effects, such as hypoxia; these possibilities must be further explored to fully understand the mechanism of action of embryonic hyperglycemia in the retina.

Embryonic development requires a complex coordination of events at the cellular level, which can be altered through metabolic stressors. As hyperglycemia becomes increasingly common, it is necessary to understand how it affects development of all tissues. Given the known deleterious effects of chronic hyperglycemia, examining retinal development in a hyperglycemic state is critical. Our findings clearly show that hyperglycemia negatively impacts retinal development, and suggest a potential mechanism of action, which may inform the search for therapeutic targets. Our future studies will include elaborating on our embryonic hyperglycemia models through metabolic assays and alternative ‘rescue’ experiments. We also look forward to studying how the hyperglycemic larval retina continues to develop, maintain itself and function long term.

Zebrafish were bred, raised and maintained in accordance with established protocols for zebrafish husbandry. All zebrafish lines were bred and raised at 28.5°C on a 14-h light–10-h dark cycle as previously described (Coomer and Morris, 2018). The Tg(3.2TαC:EGFP) transgenic line (TαC:EGFP), as previously described (Kennedy et al., 2007), was generously provided by Susan Brockerhoff (University of Washington, Seattle, WA). The Tg(XlRho:EGFP) transgenic line (XOPS:GFP), has been previously described, and was obtained from James Fadool (Florida State University, Tallahassee, FL, USA; Fadool, 2003). The Tg(gfap: GFP)mi2001 or GFAP:GFP transgenic line (Bernardos and Raymond, 2006), was obtained from the Zebrafish International Resource Center (ZIRC, Eugene, OR, USA). Heterozygous pdx1^sa280 adults were generously provided by Jeffrey Mumm (Johns Hopkins University, Baltimore, MD, USA). The Pdx1 genotype was determined by PCR using the primers 5′-TGGCTCATGTGCTCGTGTA-3′ and 5′-GTGCGTGTGAGATTTGGTTG-3′ followed by restriction fragment length polymorphism analysis with DraI. Embryos were anesthetized with ethyl 3-aminobenzoate methanesulfonate salt (MS-222, Tricaine; Sigma-Aldrich, St Louis, MO, USA). All animal procedures were carried out in accordance with guidelines established by the University of Kentucky Institutional Animal Care and Use Committee and the ARVO statement on the use of animals in research.

GFAP:GFP, XOPS:GFP and TαC:eGFP adult fish were in-crossed to generate embryos that were euthanized at the following time points: 48, 72, and 96 hpf, and 5 dpf. To generate hyperglycemic embryos/larvae, 10 hpf embryos were dechorionated with pronase (Sigma), and randomly sorted into groups of 25 and placed in the following treatments: untreated fish water, or 50 mM glucose, 50 mM mannitol, 10 μM dexamethasone (Sigma), or 50 mM glucose plus 10 μM dexamethasone (glucose+dex) in 1× phenylthiourea (Sigma) dissolved in fish water. At each time point of interest, heads were removed for cryosectioning, mRNA or protein extraction, while the body was used to quantify whole-body glucose concentration. For the pdx1 line, larvae were trisected; heads were used for retinal sectioning, the mid-body area was used to quantify glucose levels using the glucose assay, and genomic DNA was extracted from the tail for genotyping.

Larvae were submerged in fish water containing a fluorescent analog of glucose [2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NDBG); Invitrogen, Grand Island, NY)] at 5 dpf. 2-NDBG was dissolved in DMSO and diluted in fish to a final concentration of 500 µM. Larvae were submerged for 1 h in a dark incubator at 28.5°C then imaged with a Nikon inverted fluorescent microscope (Eclipse Ti-U, Nikon Instruments).

A glucose colorimetric assay kit from Biovision (Biovision, Milpita, CA) was used to quantify whole-body glucose concentration, following the manufacturer's instructions. In short, following euthanization, animals were homogenized in glucose assay buffer from the kit, mixed with a glucose oxidation enzyme and a colorimetric probe, and incubated for 20 min at 37°C. A spectrophotometer was used to quantify glucose concentration. Reads were translated into pmol/larvae through generation of a standard curve from a set of samples of defined glucose concentrations.

Embryos were fixed overnight in 4% paraformaldehyde, then incubated overnight in 10% followed by 30% sucrose at 4°C. Transverse 10 µm sections were taken beginning in the anterior tip of the head, moving posteriorly through the eye. For imaging and cell quantification, only sections containing an optic nerve were used for consistency. Photoreceptors in the dorsal, central and ventral portions of the retina were quantified and normalized to the curvilinear length of the outer nuclear layer. For the HuC/D, PKCα, and Prox1 quantification, counts were conducted on 50 µm wide regions of interest, 50 µm dorsal to the optic nerve for consistency. For the Müller glia shape analysis, images of retinal sections from both nutritional and genetic models on the GFAP:GFP background were uploaded into the Nikon Imaging Analysis Program. Within the program, the ratio of normally shaped to total Müller glial cell bodies was quantified using a standardized wide oval shape fitted to the wild-type GFP-positive Müller glia, followed by software auto-detection of similar shapes within the images. Images were taken using a 20× objective on a Leica SP8 Confocal microscope or a Nikon Eclipse inverted fluorescent microscope (Eclipse Ti-U, Nikon Instruments). At least five embryos were analyzed per treatment/genotype, and at least 2 separate biological replicates were performed for each experiment.

Protein was extracted from the heads of 25 pooled 5 dpf larvae from each nutritional model treatment by homogenizing in RIPA buffer and Complete EDTA mini tablets (Roche), which contains protease inhibitors. Following extraction, protein was quantified with a Bradford assay. For the western blot, 30 µg of protein from each sample was loaded onto a pre-cast gel (Bio-Rad) and run to separate proteins by size, then transferred to a nitrocellulose membrane. The membrane was blocked for 30 min at room temperature then incubated overnight at 4°C with the following antibodies: β-actin (1:500; cat. no. sc-47778, Santa Cruz Biotechnology) then GFAP (1:1000, cat. no. GTX128741, GENETEX). Membranes were then incubated in secondary antibody (1:500, Santa Cruz Biotechnology) at room temperature for 1 h, developed (Bio-Rad) and imaged. Bands were quantified using ImageJ (Rasband, n.d.), normalized to β-actin bands, and a paired, two-tailed t-test was used to determine significant differences in band intensity.

Sectioning and immunohistochemistry were conducted as previously described (Wen et al., 2015) and immunolabeled sections were imaged on either a Nikon inverted (Nikon Ti-U) or confocal microscope (Leica SP8, Leica). The following antibodies were used: anti-Zpr1 (red and green cones, mouse, 1:20, ZIRC); anti-Zpr3 (photoreceptor outer segments, mouse, 1:100; ZIRC), anti-HuC/D (ganglion and amacrine cells, mouse, 1:40; ZIRC), anti-PKCα (bipolar cells, mouse, 1:100; cat. no. sc-17769, Santa Cruz Biotechnology), and anti-Prox1 (horizontal cells, rabbit, 1:1000; cat. no. AB5475, Millipore, Burlington, MA) and 5E11 (amacrine cells; 1:1000) and 4C12 (mature and immature rods; 1:100), generously provided by Jim Fadool (Florida State University, Tallahassee, FL). Alexa Fluor-conjugated secondary antibodies (Invitrogen) and Cy-conjugated secondary antibodies (Jackson ImmunoResearch, West Grove, PA) were used at 1:200 dilution. Slides were incubated in DAPI to label nuclei (1:10,000 dilution; Sigma). The TUNEL assay was conducted with ApopTag Fluorescein Direct In Situ Apoptosis Detection Kit (Millipore, Billerica, MA, USA) on retinal cryosections according to manufacturer's instructions. The EdU assay was conducted with the EdU Cell Proliferation Assay Kit (EdU-555, Millipore). Embryos were soaked in 0.75 mM EdU for 2 h at room temperature from 48–50 hpf, then washed in fresh fish water, and raised to 5 dpf. Heads were cryosectioned and the Clik-It assay was conducted on retinal cryosections according to the manufacturer's instructions.

A Zebrafish Glucose Metabolism RT2 Profiler PCR Array (Qiagen, location) was used to quantify 86 different enzymes involved in pathways related to glucose metabolism (glycolysis, TCA cycle, electron transport chain, etc.) according to the manufacturer's instructions. Briefly, mRNA was extracted, and cDNA generated as previously described (Coomer and Morris, 2018) from larval heads at 5 dpf for all treatment groups and pdx1 genotypes. cDNA, SYBR Green and ultra-purified water were mixed and pipetted into the plate and a Roche Light Cycler (Roche) was used to run a program consisting of 45 cycles, 95°C for 15 s, 60°C anneal for 60 s. Gene expression was quantified with Roche Light Cycler Analysis program and normalized to the housekeeping gene encoding β-actin.

Treated and control 48 hpf and 5 dpf larvae were submerged in 5 mM MitoSOX ROS probe (Invitrogen, Carlsbad, CA) for 20 min at 28°C in the dark. Larvae were washed three times in fish water, mounted in 4% low melting agarose with 1% tricaine, and imaged with a Leica SP8 confocal microscope. Following imaging, eyes were microdissected for retinal sections and imaged on the Leica SP8 Confocal microscope. ROS production was quantified by measuring the fluorescent pixel density using ImageJ. Diphenyleneiodonium (DPI, D2926, Sigma) was included with nutritional model treatments at 25 µM and larvae were analyzed as described above.

The optokinetic response test was conducted as described previously (Brockerhoff, 2006). Larvae were assessed at 5 dpf, one at a time for 1 min at 6 rotations per minute. Each larva was tested four times, twice in each direction of the rotating drum, and averaged. Saccadic eye movements were quantified and compared across treatments and genotypes with an one-way ANOVA with unpaired two-tailed t-test.

Data were checked for normal distribution via a skewness test in Excel; statistical analysis consisted of one-way ANOVA followed by two-factor, unpaired t-test, using GraphPad software. P<0.05 was considered significant and is indicated as *P<0.05, **P<0.01 and ***P<0.001. Boxplots were generated using R (version 3.6.2)/R studio (version 1.2.5033) ggplot2 package (Wickham and Sievert, 2016). Fig. 8 and Fig. S2A were created using Biorender (biorender.com). All figures were constructed using Photoshop (Adobe version 21.0.2).

The authors would like to thank Evelyn Turnbaugh and Lucas Vieira Francisco for exceptional zebrafish care. We also thank Dr Hannah Henson for conceptual input and Dr Cagney Coomer for editorial input.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259187.