One month of hyperglycemia alters spectral responses of the zebrafish photopic electroretinogram

Diabetic retinopathy (DR) is a leading cause of blindness and visual impairment in the United States (nei.nih.gov/health/diabetic). DR occurs in individuals with long-standing diabetes mellitus, affecting 40-50% of all patients (Arden and Ramsey, 2015). Retinal complications in DR are believed to be caused by prolonged high levels of glucose (hyperglycemia) that increase inflammatory mediators (Kern, 2007; Romeo et al., 2002) and vascular permeability (Guthrie and Guthrie, 2004). Increased permeability stimulates the formation of new, fragile vessels that can subsequently rupture. The resulting hemorrhage and ischemia damages retinal nerve cells, resulting in vision loss.

In neural retina, reactive gliosis in Müller glia, characterized by an increase in glial fibrillary acidic protein (GFAP) levels, is observed in human diabetic retinas (Carrasco et al., 2008) and in animal models (Agardh et al., 2001; Barber et al., 2000; Rungger-Brandle et al., 2000; Zeng et al., 2000). Nuclear factor-κB (Nf-κB; RelA or p65) levels also increase, serving as a pro-apoptotic signal to pericytes (Kowluru and Chan, 2007; Patel and Santani, 2009). Clinically, multifocal (Xu et al., 2006) or full-field electroretinograms (ERGs) are used to determine whether retinal function is compromised in DR (Tzekov and Arden, 1999). Full-field ERGs identify photoreceptor (a-wave), ON bipolar (b-wave) and amacrine cell responses (oscillatory potentials, OPs) (Tzekov and Arden, 1999). When a long flash of light is used, the d-wave (OFF bipolar) component is also revealed (Lung et al., 2016; Sieving, 1993). Diabetic patients are reported to have altered a-wave, b-wave and OPs, with the degree of change correlated with the severity of DR. Decreased and delayed OPs are consistently observed in the early stages of DR. Decreased b-wave amplitudes also occur early (Kizawa et al., 2006), whereas changes in the a-wave component are not observed until more advanced stages (Kizawa et al., 2006; Tzekov and Arden, 1999). Decreased b-wave amplitude, smaller/slower OPs (Bui et al., 2009; Hancock and Kraft, 2004; Layton et al., 2007; Li et al., 2002) and reduced OFF responses (Samuels et al., 2015, 2012) are observed in animal models of DR, with more variable changes in the a-wave component (Bui et al., 2009; Hancock and Kraft, 2004; Kohzaki et al., 2008; Li et al., 2002; Phipps et al., 2006). Disruptions of A17 amacrine feedback onto rod bipolar axon terminals in the inner retina have been reported (Castilho et al., 2015).

Zebrafish (Danio rerio) are a new model organism in which to examine complications associated with prolonged high blood sugar levels. Zebrafish have an insulin-secreting pancreas (Argenton et al., 1999) and genes related to diabetes – such as those encoding insulin, IA-2 autoantigen and IA-2β autoantigen – have been cloned (Rubenstein, 2003). Although zebrafish retinal vessels are restricted to the inner limiting membrane (Alvarez et al., 2007; Chapman et al., 2009), they are in direct contact with Müller cell endfeet (Alvarez et al., 2007), and vascular endothelial cells are connected via junctional complexes and surrounded by pericytes (Alvarez et al., 2007). In response to hyperglycemic conditions, induced by glucose exposure in the water (Alvarez et al., 2010; Capiotti et al., 2014; Connaughton et al., 2016; Gleeson et al., 2007) or streptozotocin injections (Intine et al., 2013; Olsen et al., 2010), the inner retina thins (Gleeson et al., 2007; Olsen et al., 2010), the basement membranes of retinal vessels thicken, cone morphology is altered (Alvarez et al., 2010) and glycated proteins are formed (Capiotti et al., 2014). These are characteristics observed in other animal models as well as in humans.

Our laboratory developed an alternating immersion protocol that induces and maintains hyperglycemic conditions in zebrafish (Connaughton et al., 2016; Gleeson et al., 2007), mimicking the oscillations in blood glucose levels seen in humans with poorly controlled diabetes (Ceriello et al., 2008). Subsequent use of this technique by other investigators (Alvarez et al., 2010) identified physiological changes in glucose-treated animals. These functional changes were identified using a white-light stimulus and correlated with aberrant cone morphology (Alvarez et al., 2010). To determine whether these functional changes are caused by alterations in specific cone photoreceptor type(s), we recorded photopic ERG responses to white light and spectrally distinct light stimuli over a range of intensity levels in a superfused zebrafish eyecup preparation. In the spectral stimulation protocol, the AMPA/kainate (KA) receptor blocker 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) was added to the medium. This restricted ERG a- and b-wave signals to the outer retina, as only cones and ON bipolar cells respond in the presence of the blocker. In the white-light protocol, the medium contained no blockers, so that both the inner and outer retina circuits were functional. A major goal was to identify changes in spectral sensitivity owing to 1 month of glucose exposure and, if present, to identify the response of the cone type(s) that might be most susceptible. A further goal was to identify contributions of the inner retina to hyperglycemia-induced dysfunction.

As previously reported (Nelson and Singla, 2009), light stimulation of isolated zebrafish eyecups superfused in medium containing an AMPA/KA receptor antagonist resulted in a negative-going wave, the a1 component of the a-wave, immediately after light ON, followed by a large, positive wave, the b2 component of the b-wave. Recordings evoked in response to a white-light stimulus, without CNQX in the bath, included a-wave, b-wave and d-wave components (Fig. 1A,B). Individual ERG responses similar to these were analyzed individually and by means of the spectral model, to determine treatment-induced changes in cone sensitivities and overall retinal responses. Morphological analyses of thick sections and changes in GFAP and/or Nf-κB content were correlated with the physiological data.

Blood sugar measurements of fish alternately exposed to 2% glucose solution/0% glucose solution averaged 135±36.7 mg/dL (mean±s.e.m.; n=7 fish) after 28 days of treatment, a significant increase compared with values from mannitol-treated (n=8) and water-treated (n=6) controls (Table 1A; one-way ANOVA, P=0.011). Wet weights of these same fish were not significantly different (Table 1B; one-way ANOVA, P=0.59). These changes in blood sugar, but not wet weight, are consistent with previous reports using an alternating 2% glucose solution/0% glucose solution for 1 month, with fish <1 year of age (Connaughton et al., 2016).

Once ERG recordings began, 2 fish per day were removed from the treatment. This resulted in some fish remaining in treatment for longer than 28 days. To confirm that these fish were still hyperglycemic, blood sugar measurements were collected at the time of ERG recordings (Table 1C). In glucose-treated fish directly removed from 2% glucose, blood glucose averaged 232±10.74 mg/dL (n=3); in glucose-treated fish removed from water (0% glucose), blood glucose averaged 76±9 mg/dL (n=2). The elevated blood sugar measurements from glucose-exposed fish directly removed from glucose were significantly different (one-way ANOVA, P<0.001) from all other values, including those from water-treated controls (n=4, outlier removed), mannitol-treated controls (n=4) and untreated fish directly removed from stock tanks (n=3) (Table 1C).

To be consistent with published reports (Alvarez et al., 2010), we recorded ERGs using a white-light stimulus at saturating stimulus intensities, without AMPA/KA receptor blockade. Under these conditions, mean ERG responses with clear a-wave, b-wave and d-wave components were evoked in all treatment groups (Fig. 2A-D). These data revealed that the response of water-treated fish (Fig. 2B) was consistently reduced compared with that of untreated fish (Fig. 2A), reflecting a handling stress. Consequently, we normalized responses from glucose- and mannitol-treated tissue to those of the water-treated controls (see Materials and Methods).

Mean b-wave amplitudes (µV) modeled from 300 responses in 10 glucose-treated eyes (7.6±0.5 µV) at the brightest stimulus intensity were significantly decreased (P<0.001), compared with average b-wave amplitudes from the same number of responses in mannitol-treated eyes (10.4±0.3 µV, Fig. 2F); d-wave amplitudes were also significantly decreased in glucose-treated compared with mannitol-treated eyes (P<0.001, Fig. 2G). In contrast, there was no difference in mean peak a-wave amplitude across the groups (Fig. 2E). Similar trends were observed when the analysis was performed combining responses from all 7 light intensities (Fig. S1A-C), except for a-wave amplitude, which was significantly increased (P<0.001) in the cumulative dataset.

At the brightest stimulus intensity, mean peak implicit times (i.e. time-to-peak) for all ERG components were significantly shorter (P<0.001) in glucose-treated tissue (Fig. 2H-J; n=278-287 responses from 10 glucose eyes; n=300 responses from 10 mannitol eyes). Specifically, b-wave time averaged 97.5±1.08 ms in glucose-treated eyes, compared with an average of 105.8±19.8 ms in mannitol-treated tissue; and d-wave implicit time in glucose-treated tissue averaged 114.0±3.6 ms, versus 143.0±3.7 ms in mannitol-treated retinas. Similarly, the a-wave glucose-treated implicit time of 22.6±0.29 ms was less than the average time (23.7±0.17 ms) in the mannitol-treated group. Analysis of implicit times for all light intensities revealed the same results, and is provided in Fig. S1D-F.

To determine whether the observed changes in the native ERG (Fig. 2) after 1 month of alternating glucose immersion were caused by differential sensitivity of the different cone signals in the outer plexiform layer (OPL), we looked at responses to 4 of the 9 stimulus wavelengths – 570 nm, 490 nm, 410 nm and 370 nm – in detail. These wavelengths are close to peak opsin absorbance for adult zebrafish red (R), green (G), blue (B) and UV cones, respectively (Robinson et al., 1993). Whereas R cones might be expected to respond to all these stimuli, the shorter wavelength cones would not be expected to respond to stimuli with wavelengths significantly longer than their absorbance peaks. The ability of the wavelength filters to selectively stimulate cone types is shown in Table S1.

In recordings using the blue (418 nm)-adapting background, AMPA/KA receptor blocker and near-saturating chromatic stimuli, mean ERG timecourses look similar in shape across the 4 stimulating wavelengths (Fig. 3A-D), although mean peak a1 and b2 amplitudes appear reduced in glucose-treated eyes (104 responses from 13 eyes), compared with mannitol-treated eyes (100 responses from 15 eyes). These observations were confirmed statistically under 4-wavelength, saturating-irradiance stimulus conditions (Fig. 3E-H). The normalized b2 mean amplitude in glucose-treated eyes (17.4±1.4 µV) was significantly smaller (P<0.001, Fig. 3E) than the mean value from mannitol-treated eyes (31.5±2.1 µV). The mean a2 amplitude in glucose-treated eyes (7.22±0.56 µV) was also reduced compared with that of the mannitol-treated group (10.3±1.1 µV, P<0.05, Fig. 3F).

There were small, but significant, variations in implicit times. The glucose-treated b2-wave time (118.3±24.3 ms) was longer than the mannitol-treated value (111.1±1.8 ms, P<0.05, Fig. 3G), as was the implicit time for the a1 component (glucose average, 55.2±1.1 ms; mannitol average, 50.1±0.8 ms) (P<0.001, Fig. 3H). Analyzing responses averaged across all wavelengths and brightnesses, as shown in Fig. S2A-D, resulted in the same amplitude reductions.

For the red (627 nm) background and AMPA/KA receptor blocker, the mean ERG timecourses in the different treatment groups were similar for near-saturating 570 nm, 490 nm, 410 nm and 370 nm stimuli (Fig. 4A-D). The analyzed data (Fig. 4E-H) are based on 112 responses in glucose-treated eyes and 108 responses in mannitol-treated eyes under the 4-wavelength, saturating stimulus conditions illustrated (Fig. 4E-H). As in Fig. 3, peak b2 and a1 amplitudes were reduced in glucose-treated, compared with mannitol-treated, eyes. This was confirmed statistically as the mean b2 amplitude in glucose-treated eyes was 15.6±1.3 µV, a value found to be significantly smaller than the mean value in mannitol-treated controls (28.0±1.6 µV, P<0.0001). The mean a2 amplitude in glucose-treated eyes (8.4±0.6 µV) was also significantly reduced compared with that of mannitol controls (12.1±0.8 µV, P<0.001). As found with the blue-adapting background, both the mean glucose-treated b2 peak time (125.7±1.9 ms) and a1 peak time (60.1±0.9 ms) were significantly longer than the times from mannitol-treated tissue (b2, P<0.01; a1, P<0.001). The same amplitude reductions were also obtained in the responses to all brightnesses and wavelengths in the cumulative dataset (Fig. S2E-H).

Taken together, the finding that glucose decreases b2 and a1 amplitudes at all wavelengths on blue backgrounds certainly suggests the involvement of R cones, the only cones that can be stimulated by all wavelengths. The fact that this pattern of amplitude decrease also persists on red backgrounds suggests that at least one other cone could also be involved. These results further suggest that reductions in native ERG b- and d-wave amplitudes, if not implicit times, might originate in distal retina.

To analyze the physiological responses evoked by the different stimulating wavelengths in terms of cone components, the spectral properties of the cumulative spectral dataset within each treatment group were compared by fitting to Eqn 1. With this equation, saturating voltages (Vr, Vg, Vb, Vu) for signals arising from each cone can be determined. Using this approach, we compared outer retina ERG b2 signals, isolated by AMPA/KA receptor blockade, from glucose-treated and mannitol-treated control groups on the blue (418 nm) background (Fig. 5A). A significant reduction in maximal response of the glucose-treated R cone signals (V_r, Eqn 1) was found (P<0.0001), whereas no significant effect on peak amplitudes of G, B or UV cones was detected (green, P=0.20; blue, P=0.12; UV, P=0.87), even though a tendency toward smaller G and B cone signals is apparent. Because the blue background might be selectively desensitizing for B and G cones, the experiment was repeated on a red (627 nm, Fig. 5B) background, which does not desensitize them. Even though R cones are desensitized by this background, a very significant reduction in peak R cone signal amplitude was induced by glucose treatment compared with mannitol treatment (P<0.01). The ability to elicit R cone responses with a red-adapting background occurred because R cone amplitudes are not eliminated by red backgrounds. However, their half saturation (kr) does increase.

Significant reductions in G and B cone saturation amplitudes (V_g, V_b) were also observed (green, P<0.05; blue, P<0.01). In contrast, no significant signal amplitude differences were noted for UV cones (P=0.58). These findings further confirm that actions on the glucose-mannitol axis occur under AMPA/KA receptor blockade and affect R cone phototransduction and ON bipolar synaptic currents strongly; G and B cone signaling is also affected, although to a lesser extent.

Therefore, the spectral model supports the changes in b2-wave amplitudes observed for individual stimulating wavelengths and indicates that the b2-wave signals arising from R and G (double) cones, and potentially also B cones, are significantly reduced in hyperglycemic, glucose-treated fish.

Thickness measurements revealed a central retinal thickness of 136 µm (±2.91) in water-treated retinas, a value not significantly different from those of glucose- (147.5±2.54 µm) or mannitol-treated (143.9±2.98 µm) tissue. For some specific layers [inner plexiform layer (IPL), P<0.01; OPL, P<0.01], however, mean thickness values were larger in glucose and mannitol treatment groups compared with water-treated controls (Fig. 6), whereas outer nuclear layer (ONL) thickness was significantly less in these 2 treatment groups (P=0.048). Nonetheless, despite observations of physiological differences in glucose and mannitol groups, no changes in retinal layering structure were seen. The similar retinal thicknesses in both mannitol- and glucose-treated tissue is consistent with published findings using Epon-embedded sections (Alvarez et al., 2010).

Additional analysis of retinal tissue was performed using western blotting, to identify changes in protein content within retinal homogenates. We chose to probe for GFAP, a protein expressed in Müller glia, and the Nf-κB inflammatory marker, which are known to be upregulated by hyperglycemic conditions (Barber et al., 2000; Gerhardinger et al., 2005; Kowluru and Chan, 2007; Patel and Santani, 2009; Romeo et al., 2002). Both proteins were upregulated in glucose-treated zebrafish retinas (Fig. 7). GFAP content was increased 1.4× in homogenates from glucose-treated retinas, although this change was not significant (P=0.411) (Fig. 7). However, Nf-κB content (determined using the anti-RelA antibody) was increased in glucose-treated retinas by 3.5× and in mannitol-treated retinas by 2× (Fig. 7). These differences in RelA levels were significant (P=0.003).

Zebrafish were suggested as a model for type 2 diabetes when alternate immersion in a glucose solution increased blood glucose values and thinned the inner retina (Gleeson et al., 2007). Subsequent studies showed that alternating glucose immersion also thickened basement membranes in retinal vasculature (Alvarez et al., 2010) and glycated proteins in the eye (Capiotti et al., 2014), both of which are observed in diabetic patients and in other animal models. Zebrafish clearly respond to glucose in water (Alvarez et al., 2010; Capiotti et al., 2014), with uptake of a fluorescently tagged glucose probe into larvae occurring ∼30 min after immersion (Lee et al., 2013). A zebrafish type 1 diabetes model has also been developed in which repeated injection of streptozotocin ablates pancreatic β-cells, leading to elevated blood sugar levels (Intine et al., 2013; Olsen et al., 2010). These studies, along with research showing that the zebrafish pancreas has characteristics similar to those of the mammalian pancreas (Kinkel and Prince, 2009), suggest that the zebrafish is a good model in which to examine the effects of complications associated with chronic high blood sugar, as seen in diabetes.

Here, we examined changes in cone-specific physiological responses resulting from alternating exposure to 2% glucose (∼110 mM) (Alvarez et al., 2010) for 1 month. As previously reported (Connaughton et al., 2016; Gleeson et al., 2007), 1 month of exposure elevated blood sugar values to ∼3× higher than those of water- and mannitol-treated controls. To accommodate physiological testing, we extended glucose treatments to 5 weeks. Blood sugar remained elevated with continued treatment, as fish removed from glucose at the time of ERG recordings continued to display significantly elevated blood sugar levels. Although blood sugar was different across the treatment groups, wet weight and overall survival/behavior were not, consistent with published reports (Connaughton et al., 2016). This indicated that the overall health of the fish was comparable in the different treatment groups.

Whereas the overall health for the water-treated fish was like those in the other treatment groups, the ERG responses were not. b-wave amplitudes recorded from water-treated controls were reduced compared with those of untreated fish (Fig. 2A, P<0.0001), suggesting that the daily transfers are stressful to the fish, despite no overt differences in behavior and survival across groups. Handling is a well-known stressor to fish that can affect hormone levels (Acerete et al., 2004; De Lombaert et al., 2017), as well as brain chemistry (Enrico et al., 1998; Timmerman et al., 1999). Our data suggest that handling/daily transfers of the fish can also affect retinal (ERG) responses. Given that the glucose- and mannitol-treated fish were similarly handled/transferred, we normalized our ERG amplitude data to responses from the water-treated group to better visualize changes resulting from treatment conditions. Blood sugar values in water-treated fish were also reduced compared with those in untreated fish directly removed from stock tanks. However, this has been reported previously (Connaughton et al., 2016) and reflects differences in the frequency of feeding and fasting prior to blood sugar measurements.

ERG b2 amplitudes isolated from the distal retina by CNQX treatment were significantly decreased in glucose-treated tissue at all stimulus wavelengths. As only cones and ON bipolar cells respond to light in these conditions, these ERG deficits originate from cones and their synapses onto ON bipolar cells. Our results further suggest that R cone signals are robustly susceptible to hyperglycemic insult, regardless of adaptation state. Susceptibility of G and B cones was also revealed using red adaption, where these types are essentially dark adapted. These results concur with reported glucose-induced anatomical changes to zebrafish cones (Alvarez et al., 2010) and suggest that the anatomical changes are correlated with functional deficits.

Native ERG b-wave amplitudes from eyecups not perfused with CNQX were decreased in glucose-treated fish (Alvarez et al., 2010). These white-light responses reflect signals from all cone and neuronal types. ERG d-waves were previously reported to decrease in glucose-treated tissue (Alvarez et al., 2010), a finding we replicate. The current, more extensive, data provided in Table S1 and Fig. S3 show that a-wave amplitudes increased in response to glucose treatment, and a-, b- and d-wave implicit times became quicker.

A surprising finding was the large amplitude response of mannitol-treated tissue versus water controls. Previous studies (Alvarez et al., 2010) used untreated fish as a water control, and thus did not encounter this effect. In other studies by our group (Connaughton et al., 2016; Gleeson et al., 2007), in most parameters assessed here (blood sugar, wet weight, overall health) and in cell culture studies that use mannitol as an osmotic control for glucose treatment (Chen et al., 2013; Costa et al., 2013), the mannitol group reflects control (Chen et al., 2013; Gillies et al., 1997; Santiago et al., 2007) and/or normoglycemic (Dennis et al., 2015; Schrufer et al., 2010) responses. However, our ERG recordings revealed differences between control groups. The reasons for this are unknown but they could be due to either a hyperosmotic response or a protective effect of mannitol. Our mannitol and glucose treatment groups used the same concentrations (2%) and the same exposure regime. However, it is likely that glucose was metabolized by the fish as they have functioning insulin and insulin receptors. Not all ingested glucose is immediately stored, however, as evidenced by high blood sugar levels in the treatment phase (Table 1C). Mannitol is not taken up by cells nor does it affect blood sugar levels, suggesting hyperosmotic effects, potentially localized to the cellular level. Exposure to hyperosmotic solutions increases spontaneous quantal release of neurotransmitter at frog neuromuscular junctions (Kashani et al., 2001) and enhances transmitter release onto ganglion cells in retinal slices (Kashani et al., 2001). In mice, however, mannitol can be metabolized by gut bacteria (Xiao et al., 2013), indirectly providing nutritional benefit, not manifesting as deleterious high serum glucose.

Alternatively, mannitol could be protective to retinal tissue. Clinically, mannitol is used in the treatment of glaucoma (Jin et al., 2007; Lam et al., 2002; Nagaraju et al., 2007; Rhee et al., 2006), as an adjuvant to facilitate stem cell delivery to the brain (Gonzales-Portillo et al., 2014) and to reduce ischemic damage in cardiac muscle cells (Powell et al., 1976). Similar effects might be occurring in the retina. However, perfusion with a mannitol solution disrupts the blood-brain barrier in rodents (Kaya et al., 2004). Intravitreal injection of mannitol causes retinal detachment (Marmor, 1979; Marmor et al., 1980), alters retinal pigment epithelium (RPE) c-wave activity (Marmor, 1979; Shirao and Steinberg, 1987) and, over time, decreases ERG a- and b-wave components (Marmor, 1979). Whatever the physiological mechanism underlying mannitol's potential protective effect on retinal tissue, our data suggest that it is long-term, as there was no mannitol in the eyecup perfusate during recordings.

Altered ERG implicit times and response amplitudes (Klemp et al., 2004), visual-evoked potentials and pattern ERG responses (Parisi et al., 1998) can be observed prior to any clinical signs of retinopathy (Tzekov and Arden, 1999). These physiological changes increase in severity with DR progression (Tzekov and Arden, 1999; Zakareia et al., 2010). In rodents, changes in ERG a- and/or b-wave components do not begin immediately after hyperglycemic induction. Rather, most are reported to occur after 10-12 weeks (Bui et al., 2009; Hancock and Kraft, 2004; Li et al., 2002; Phipps et al., 2006; Ramsey et al., 2006), when retinal thinning is observed (Martin et al., 2004), although some reports indicate that changes can occur earlier, at ∼6 weeks (Layton et al., 2007; Li et al., 2002). In diabetic patients, disease progression occurs over many years. Thus, in both rodents and humans, there is a time course associated with hyperglycemia-induced changes in retinal function. This time course appears to be accelerated in the zebrafish model.

No changes in retinal thickness were observed in glucose- versus mannitol-treated tissue, similar to previously reported results in Epon-embedded sections (Alvarez et al., 2010), but different from measurements in cryostat sections (Gleeson et al., 2007). Glucose-induced differences in cone morphology have also been reported, with the greatest defects observed in double (R/G) cones (Alvarez et al., 2010). Our physiological data support photoreceptor-specific deficits, even though we did not detect any morphological differences in the photoreceptor layer in our thick sections.

GFAP and Nf-κB content increased in glucose-treated retinal homogenates, with a large, significant increase in Nf-κB. Early stages of DR are associated with glucose-induced upregulation of pro-inflammatory and pro-apoptotic markers (Adamis and Berman, 2008; Joussen et al., 2001; Kern, 2007), which stimulate leukostasis in capillary endothelial cells (Adamis and Berman, 2008; Kern, 2007; Tang and Kern, 2011), resulting in vessel leakage, capillary dropout and other changes characterizing the clinical onset of DR. Experiments have identified a common set of inflammatory markers, including Nf-κB, that are activated by high glucose (Kern, 2007; Romeo et al., 2002; Tang and Kern, 2011). The 3.5× increase in Nf-κB in glucose-treated retinas observed here is consistent with a hyperglycemia-induced inflammatory response. GFAP levels were also elevated in glucose-treated retinas, but not significantly. Previous reports indicate an increase in GFAP levels only in mannitol-treated tissue, suggesting an osmotic effect (Alvarez et al., 2010). Upregulation of GFAP occurs in response to hyperglycemia (Barber et al., 2000; Gerhardinger et al., 2005; Lieth et al., 1998; Mizutani et al., 1998) as well as other retinal insults, including light damage (Grosche et al., 1995), photoreceptor degeneration (DiLorento et al., 1995; Lewis et al., 1995) and retinal detachment (Lewis et al., 1995).

Zebrafish exposed to 1 month of alternating 2% glucose/0% glucose display elevated blood sugar levels. Retinal homogenates identified upregulation of Nf-κB, suggesting an inflammatory response. Using spectrally distinct stimuli and AMPA/KA receptor blockade in the inner retina, glucose exposure decreased a1 and b2 response amplitudes, owing to an impairment primarily between R cones and their ON bipolar synapses. In response to bright-white light, native ERGs from glucose-treated eyecups displayed reduced b- and d-wave amplitudes and decreased implicit times for a-, b- and d-waves; effects on a-wave amplitude depended on stimulus brightness. These findings indicate that prolonged hyperglycemia adversely affects zebrafish retinal electrophysiology through separate actions in both distal and proximal retina, with distal effects on b-wave amplitudes and proximal effects on response component implicit times.

Wild-type adult zebrafish (Danio rerio) (4 months to 1 year) of mixed sex were purchased from a vendor or bred from the in-house zebrafish colony and housed within the Fish Facility at American University. All procedures were approved by the American University Institutional Animal Care and Use Committee. The fish were acclimated for at least 1 week to standard laboratory conditions (28°C; 14 h light/10 h dark) prior to experiments. The fish were fed a diet of Tetramin flakes+zebrafish crumble (Aquatic Habitats) daily until the beginning of the experiments.

Hyperglycemia was induced in zebrafish by exposing the fish to alternating conditions of 2% glucose/0% glucose every 24 h (i.e. a change in solution composition was made every 24 h) (Gleeson et al., 2007). Osmotic control animals were exposed to 2% mannitol/0% mannitol every 24 h and the water-treated control animals (for handling stress) were exposed to 0% glucose/0% glucose (water/water solution change) every 24 h. Deer Park Spring Water was used to make the experimental solutions. Fish were fed every other day while in the experimental containers. Temperature, pH, water quality and general fish behavior were recorded daily. Fish were maintained on the alternating exposure protocol for at least 28 days (4 weeks) before ERG analysis. Starting on day 28, 2 fish were removed and tested each day. Alternations between the 2 solutions continued for all remaining fish until they were analyzed for ERG responsiveness.

At the end of the 28-day exposure period, a subset of fish from each of the 3 treatment groups was removed from treatment and anesthetized in 0.02% tricaine solution. For some experiments, untreated fish taken directly from stock tanks were also used. Once anesthetized, the fish were weighed (wet weight, in mg; Sartorius microbalance) and decapitated so that a blood sugar measurement could be obtained (Connaughton et al., 2016; Gleeson et al., 2007). Blood sugar was measured from cardiac blood using a Freestyle-Lite Blood Glucose Meter (Connaughton et al., 2016). Differences in wet weight and blood sugar were assessed for outliers using Dixon's outlier test and then compared across treatments using a one-way analysis of variance (SPSS software) at an α-level of 0.05.

Fish were sacrificed by decapitation. Heads were hemisected to remove the eyes and the anterior segment was removed, forming an eyecup. Each eyecup was placed in the recording chamber and superfused with an oxygenated (95% O₂/5% CO₂, 20°C) minimal essential medium (MEM) solution (Millipore) for at least 30 min. A 28-gauge nonmetalic syringe needle (WPI, Sarasota, FL, USA) was placed inside the eyecup to perfuse the retina at a rate of 0.3 ml min⁻¹. For responses using spectrally distinct stimuli, the MEM contained 50 µM CNQX, an AMPA/KA receptor antagonist (Tocris), to isolate a- and b-wave responses. Under these conditions, the major b-wave component is the b2 response (Nelson and Singla, 2009), which is the non-AMPA/KA receptor response of the ON bipolar cells postsynaptic to stimulated cone photoreceptors. The major a-wave component is the a1 response (Nelson and Singla, 2009). For ERG responses elicited by white-light stimuli, CNQX was not used, allowing native a-wave, b-wave and d-wave components to be recorded. Glucose, mannitol and water treatments were not continued during the recording sessions, although all recordings were made in 0.1% glucose, the amount standard to MEM.

These photopic ERG signals were amplified (DAM80, World Precision Instruments) using a bandpass from 0.1 Hz to 1 kHz and a Digidata 1440A (Axon Instruments) with a sampling rate of 2 kHz. Traces averaged 4× for each stimulus condition were boxcar filtered over a period corresponding to one cycle of 60 Hz line frequency. Mean response component amplitudes, peak times and spectral properties were obtained from the different treatment groups. Data were collected with pCLAMP 10 software (Axon Instruments) and analyzed using Origin 8.5, 9.0 or 9.1 (2015, 2016).

A white-light stimulus was tested on 10 eyes per treatment group. White-light ERG responses had a-wave, b-wave and d-wave components (Fig. 1A). In the white-light protocol, the background was infrared (IR, RG780 filter). The white-light protocol generated a response dataset with 70, 4-trace-average ERGs. These ERGs were in response to 10 replicates of a 7-step irradiance response series covering a 3-log unit range. The white-light source was a 150 W Xenon arc lamp, as imaged through UV compliant optics, providing a flat energy spectrum across the visual sensitivity range of zebrafish from 650 nm to 370 nm. This range includes zebrafish UV cones, as well as R, G and B cones. The maximum delivered irradiance (−3.0 log units attenuation) was 0.15 µW cm⁻². The white-light protocol was run 3× on each eyecup, covering a recording period of ∼75 min, to generate a total of 30 ERGs for each brightness level from each eye, or a total 300 ERGs at each of 7 brightness levels from 10 eyes.

For statistical comparisons between treatments, all white-light datasets were combined into a single, cumulative dataset for each treatment group. This data analysis was consistent with the modeling analysis of spectral responses (see below). Amplitudes of a-, b- and d-waves were measured as means within fixed, characteristic, time intervals after onset or cessation of the stimulus (pre-stimulus baselines set to ‘0’). The time intervals were selected to include a-, b- or d-wave peaks. This method allowed consistent measurements of response amplitude as the time-to-peak for each component varied with treatment. The native ERG a-wave amplitude (μV) was the mean amplitude within 15-35 ms after stimulus onset. The a-wave trough time (ms) was the time of maximal excursion during a 10-100 ms poststimulus interval. Both b-wave amplitude (μV) and peak times (ms) were measured in the interval 51-200 ms after stimulus onset. For the d-wave, amplitude and peak times were measured in the interval 25-250 ms after stimulus offset. Means during characteristic time segments were always used as amplitude indexes for a, b or d-waves, but in the absence of a peak within an interval, no implicit time was scored.

In a separate set of experiments, 7-15 eyes per treatment were tested with spectral stimuli. The spectral protocol used 9 wavelengths ranging from 330 nm to 650 nm (40 nm increments, 20 nm half bandwidth) over 7 intensities (0.5 log unit increments covering 3 log units). The spectral protocol generated a response dataset with 70, 4-trace-average ERGs. These ERGs were in response to 64 unique combinations of wavelength and intensity, with the remaining 6 ERGs being replicates. Mean amplitudes and peak times of ERG components were measured as described above, except the a1-wave amplitude time segment was 10-65 ms, because the a1-wave, as revealed in CNQX, is slower than the native a-wave.

Two chromatic backgrounds were used to better isolate individual cone mechanisms and to prevent rod dark adaptation: red (627 nm) and blue (418 nm). The former background helped to isolate G, B and UV cone signals from the R cone signal, while the latter reduced the B cone signal and, to a lesser extent, the G cone signal in comparison to the R and UV cone signals. The irradiance of the red background was 4.79 log(quanta µm⁻² s⁻¹) and the irradiance of the blue background was 3.84 log(quanta µm⁻² s⁻¹). The spectral protocol was run on each eyecup 4 times, twice with each background. Recordings were accepted if amplitudes and waveforms remained stable during the 17 min, 70-stimulus protocol, as stability is critical to spectral analysis. The resulting ERG recordings had characteristic a (a1)- and b (b2)-waves (Fig. 1B).

Treatment-level datasets included all the responses from all eyes in each treatment group, allowing the model to run properly and to provide a detailed analysis of changes in cone inputs across the different stimulus intensities and wavelengths. The model also reports the number of responses used for each analysis; we provide this value as the N in the results. We chose to use the spectral model to examine hyperglycemia-induced effects on cone photoreceptors when analysis using average responses across eyes revealed no significant differences across treatment, owing to the higher variability, and lability, in the responses from eyes of glucose-treated fish (Fig. S3). However, differences in ERG traces due to treatment are clearly apparent (Figs 2-4). By using all the data points in each dataset, the spectral model identified and quantified these differences. This spectral model has been used previously to quantify the magnitude of inputs from the different cone photoreceptors in zebrafish horizontal (Connaughton and Nelson, 2010) and amacrine (Torvund et al., 2017) cells.

Our experiments utilized two control groups: an osmotic control (mannitol) and a handling stress/transfer control (water). The mannitol treatment controls for the increased osmolarity due to the presence of glucose in tank water and the water treatment controlled for daily transferring of the fish. Interestingly, in all ERG recordings, the response amplitudes from the water-treated control group were consistently the lowest recorded. We believe that this reflects an effect of handling stress/daily transfers, even though blood sugar levels were not elevated in these fish (Table 1A) and they displayed no overt stressful behavior(s). Responses from the water-treated control group were also reduced compared with responses from untreated fish directly removed from stock tanks (Fig. 2A), again suggesting handling stress. Given that fish in the glucose and mannitol treatment groups were also transferred daily and would, therefore, experience similar handling stress(es), we normalized the ERG response amplitudes of the glucose- and mannitol-treated fish to those of the water treatment group to more accurately index the effects of hyperglycemia.

Statistical analyses were performed using SPSS ver. 22 software (IBM Corporation), Prism ver 6 (GraphPad Software) or Origin (OriginLab) and evaluated at α=0.05; graphs were generated in SigmaPlot (ver. 12) or Origin. Statistical differences within each parameter at each wavelength were assessed using either a one-way analysis of variance (ANOVA) or a Student's t-test with α=0.05. If the results of the ANOVA were significant, a Tukey post hoc multiple comparison test was used.

To determine whether there were any changes in retinal layer morphology due to glucose exposure, 5 fish per treatment group were euthanized in 0.02% tricaine solution. Fish were then decapitated, and the eyes were removed and fixed in 4% paraformaldehyde. Each eye was washed 3× in 0.9% saline solution and dehydrated through an ethanol series (30%, 50%, 70%, 95%, 100%). Tissue was then washed 2× with either propylene oxide or acetone and infiltrated in a 1:1 propylene oxide: resin or acetone: resin solution for 1.5-2 h, followed by overnight infiltration in 1:2 solutions (propylene oxide: resin). The next day, tissue was embedded in fresh resin and allowed to polymerize for 3 days at 60°C.

Thick (3-5 µm) sections were cut with a diamond knife on a Leica UL-7 ultramicrotome. Sections were mounted on gelatin-coated glass slides and stained with a mixture of 1% Methylene Blue, 1% Azure A, and 1% sodium borate (Richardson et al., 1960; Tarboush et al., 2014). Once dried, sections were viewed at 5×, 10× and 40× using a Zeiss inverted microscope, photographed using QCapture Plus software (ver. 3.1.3.10) and measured using ImageJ (National Institutes of Health).

Measurements of retinal layer thickness were made from sections of central retina, behind the lens. Thickness of the whole retina, IPL, INL, OPL, ONL and photoreceptor layer (PL; from the distal side of the OPL to the outer edge of the retinal pigment epithelium) were measured in the vitread-sclerad direction. Each measurement was made 3× and then averaged to reduce error. For consistency, only sections with an identified cornea and lens were measured. A one-way ANOVA was used to determine significant differences between mean thickness measurements across treatment, followed by a Tukey post hoc test (α=0.05).

Following measurement of blood sugar, eyes were enucleated, and the cornea and lens removed. Retinal tissue was then separated from the sclera and placed into RIPA buffer (3-5 retinas per replicate per treatment) on ice. Total protein lysates were prepared from pooled retinas using M-PER reagent (Thermo Fisher Scientific, Rockford, IL, USA) containing Halt protease/phosphatase inhibitor (Thermo Fisher Scientific) in accordance with the manufacturer's protocol. Proteins were quantified using a BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA, USA) and separated using 4-12% Bis-Tris protein gels (Life Technologies, Grand Island, NY, USA). Gels were electrophoresed and transferred to polyvinylidene difluoride membrane using the iBlot 2 Dry Blotting System (Life Technologies). Membranes were blocked in 5% bovine serum albumin (BSA). Primary GFAP, RelA (V-Rel Avian Reticulodneotheliosis Viral Oncogene Homolog A) and β-actin antibodies were applied in a 1:1000 dilution in 5% BSA. Anti-rabbit (RelA, β-actin) or anti-mouse (GFAP) horseradish peroxidase-conjugated secondary antibodies were applied at a dilution of 1:2000 (Cell Signaling Technology, Danvers, MA, USA). West Dura Chemiluminescent substrate (Thermo Fisher Scientific) was used for signal detection. Membranes were visualized using a ChemiDoc-It imaging system (UVP, Upland, CA, USA) and densitometry was quantified using ImageJ. Results were normalized to β-actin, which served as a loading control.