Predominant cone photoreceptor dysfunction in a hyperglycaemic model of non-proliferative diabetic retinopathy

Diabetic retinopathy (DR) is the leading cause of blindness for working-age individuals in developed countries (Wong et al., 2006; Morello, 2007). Diabetes afflicts ~170 million individuals worldwide, with ~78–98% progressing to DR within 15 years of diagnosis. DR is divided into initial non-proliferative DR (NPDR) and subsequent proliferative DR (PDR) stages, reflecting the inappropriate growth of new leaky blood vessels in the later stage. DR pathology stems from the hyperglycaemia associated with diabetes. Elevated blood glucose levels can modify metabolic processes owing to the overproduction of superoxide in the mitochondrial electron-transport chain (Brownlee, 2001). This can increase the production of advanced glycation end products (AGE) and the accumulation of free radicals that activate inflammatory mechanisms (Brownlee, 2001; Hernandez and Simo, 2007).

DR is a dual disorder that includes microvascular complications and neurodegeneration of the retina (Gardner et al., 2002; Simo et al., 2006). Vascular changes in NPDR include retinal capillaries with disrupted cell-cell junctions, thickened basement membranes and loss of pericytes (Garner, 1970; Simo et al., 2006; Frank, 2009). In addition, vessels become leaky and acellular, circulation is compromised and retinal oedema appears owing to plasma extravasation from the capillaries (Greenstein et al., 2000; Frank, 2004). All of these vascular changes lead to a breakdown of the blood-retinal barrier and compromise retinal function. In NPDR, damaged endothelial cells synthesise vasoconstrictors, retinal capillaries become thrombogenic, and archetypical ‘cotton wool spots’ appear, which are indicative of retinal ischemia (Frank, 2004; Simo et al., 2006).

Retinal ischemia usually progresses to severe hypoxia, triggering PDR characterized by neovascularisation (Garner, 1970; Simo et al., 2006). In response to retinopathic hypoxia, angiogenic factors including vascular endothelial growth factor (VEGF) are secreted and new blood vessels grow inappropriately into the vitreous (Frank, 2004; Hernandez and Simo, 2007). These new vessels are fragile, leading to extensive intraocular haemorrhaging. Additionally, the vessels anchor to the vitreous via fibrous tissue that is prone to contraction, producing tension on the retina and eventually causing retinal detachment and blindness (Gardner et al., 2002; Frank, 2004).

Interest in the neuropathological aspect of DR has been revived recently (Cho et al., 2000; Gardner et al., 2002; Simo et al., 2006). The functional loss of retinal neurons in diabetics was described five decades ago (Wolter, 1961), and DR patients commonly fail colour vision and contrast sensitivity tests (Daley et al., 1987). In fact, in diabetics, abnormal visual function, as assessed by electroretinography (ERG), usually precedes the clinical onset of retinal neovascularisation (Bresnick and Palta, 1987; Juen and Kieselbach, 1990; Yamamoto et al., 1996; Parisi and Uccioli, 2001; Gardner et al., 2002).

To date, the specific cell types undergoing neuropathology in the retinas of diabetics have not been clearly identified, with contradictory results in post-mortem analyses (Wolter, 1961; Barber et al., 1998; Cho et al., 2000; Carrasco et al., 2007; Carrasco et al., 2008). Recently, it has become apparent that, in DR, retinal neurodegeneration occurs prior to retinal neovascularisation. However, it is unclear whether the early vascular changes and the retinal neuropathology develop independently.

Current treatments for DR target the vasculature, stunting the inappropriate growth of new blood vessels during late-stage PDR. Laser photocoagulation is a surgical approach that ablates unwanted vessels, but also adjacent healthy tissue. Although this intervention can halt PDR progression, it is a palliative therapy for advanced disease (Wilkinson-Berka and Miller, 2008). In addition, success is partial and associated with serious side effects including visual field loss, recurrent vitreal haemorrhage, cataracts and retinal detachment (Hernandez and Simo, 2007; Wilkinson-Berka and Miller, 2008). Molecular therapies in clinical trials block retinal neovascularisation using antibodies, small interfering RNAs (siRNAs) or small molecule inhibitors targeted to pro-angiogenic pathways (Campochiaro, 2007; Wilkinson-Berka and Miller, 2008; Frank, 2009). By contrast, no DR therapies target the earlier NPDR stage, nor do they aim to protect the endangered retinal neurons. Thus, comprehension of the neuronal and vascular abnormalities in the initial phases of NPDR may reveal therapeutic targets that more effectively block disease progression.

Numerous animal models of DR have been proposed, and although some phenocopy certain diabetic complications well, none is without limitations (see www.amdcc.org). To our knowledge, there is no animal model that recapitulates all of the neural and vascular complications that are associated with NPDR and PDR in humans. Strategies to induce DR include the destruction of beta cells in the pancreas by surgical removal or drug treatment with streptozotocin (STZ) (Rees and Alcolado, 2005). STZ, which is probably the most widely used compound to model type 1 diabetes, is a glucosamine-nitrosourea derivate and is very efficient in destroying pancreatic beta cells, but its toxicity and carcinogenicity produce unwanted side effects (Rees and Alcolado, 2005; van Eeden et al., 2006) In modelling DR, STZ treatment induces transient apoptosis in the neuroretina, glial activation, impaired electroretinograms (ERGs) and some hallmarks of early microangiopathy including open blood-retinal barriers. The phenotypes induced have a late developmental onset and do not progress to PDR (Barber et al., 1998; Zeng et al., 2000; Li et al., 2002; Park et al., 2003; Feit-Leichman et al., 2005).

Inbred rodent strains selected for hyperglycaemia include the non-obese diabetic mouse (NOD), the bio-breeding rat (BB) and the Goto-Kakisaki rat (GK) (Nakhooda et al., 1977; Makino et al., 1980). Although these models identify genetic determinants of diabetes, their application to DR research is limited owing to their reduced life span (Portha et al., 2001; Rees and Alcolado, 2005). Transgenic mice overexpressing VEGF or insulin-like growth factor (IGF) in the retina indirectly model PDR by inducing neovascularisation and retinal detachment (Ruberte et al., 2004; Shen et al., 2006; van Eeden et al., 2006); however, these models can precipitate developmental defects, may disrupt the cells expressing the transgene and, importantly, they bypass growth factor-independent pathologies arising from hyperglycaemia. In this regard, exposure to high concentrations of glucose is widely used for cell culture models of diabetes (Haselton et al., 1998; Mandal et al., 2006; Fraser et al., 2007; Ihnat et al., 2007) but fails to induce DR phenotypes in wild-type rodents (www.amdcc.org) (Rees and Alcolado, 2005).

Zebrafish have excellent potential for diabetes research (Kinkel and Prince, 2009). A simple method to induce sustained hyperglycaemia in zebrafish was described recently by Gleeson et al. (Gleeson et al., 2007). Alternating between 24 hours of glucose treatment and 24 hours of rest in system water for 30 days induces high blood glucose levels in adult zebrafish. The authors also reported a reduction in the thickness of the retinal inner plexiform layer but did not address vascular changes in the retina (Gleeson et al., 2007). The inner retina of zebrafish was also recently demonstrated to be nourished by an intricate vascular network that shares many features with the human hyaloid and retinal vasculatures (Alvarez et al., 2007). Thus, we investigated whether the retinas of hyperglycaemic zebrafish recapitulate the morphological and/or functional changes that are associated with early-stage human NPDR.

Here, we demonstrate that the retinas of hyperglycaemic zebrafish display neurodegenerative and vascular hallmarks of early-stage NPDR. Retinal capillaries have thickened basement membranes and compromised blood-retinal barrier architecture. Significantly, we observe a unique susceptibility of cone photoreceptors to hyperglycaemia, which was revealed by extensive degeneration of cone cell morphology and defective photopic ERGs. Finally, our data demonstrate that the cone dysfunction and morphological degeneration induced by hyperglycaemia develop independent of the early vascular pathologies that are induced by hyperosmolarity.

One- and 2-year-old adult zebrafish were treated with 2% glucose (~110 mM) to induce hyperglycaemia, as described previously (Gleeson et al., 2007). After 30 days, glucose-treated fish show no apparent abnormalities in their overall morphology or behaviour. Control fish that were kept in system water during the treatment period were slightly thinner. By contrast, retinal vessels from 2-year-old hyperglycaemic fish were visibly thicker than age-matched controls, especially at locations proximal to the central optic disc (Fig. 1C,D,F). Although thicker retinal vessels were apparent, no difference was observed in the overall patterning or number of retinal vessels compared with controls (Fig. 1A,B,E).

For all subsequent treatments, mannitol-treated fish were included as a hyperosmolarity control. To evaluate the effects of hyperglycaemia on the ultrastructure of the blood-retinal barrier, vessels from control, glucose-treated and mannitol-treated retinas were analysed by transmission electron microscopy (TEM) (Fig. 2). Both tight (TJ) and adherent (AJ) interendothelial junctions were significantly wider in the retinal vessels of all fish subjected to glucose or mannitol compared with water controls (Fig. 2A,C). This morphological disruption of the blood-retinal barrier architecture suggests increased vessel permeability (Vinores et al., 1992). As this early vascular phenotype was observed after glucose or mannitol treatment, it is a feature of hyperosmolarity and is not unique to hyperglycaemia. Notably, the vessel basement membrane is significantly thicker in glucose-treated fish, but not in mannitol-treated or control fish (Fig. 2B,D). Thickening of the basement membrane in retinal capillaries is the first and most widely accepted hallmark of DR (Garner, 1970; Wilkinson-Berka and Miller, 2008).

Using reverse transcription (RT)-PCR, we observed a sharp upregulation of VEGF₁₆₅ in glucose- and mannitol-treated fish (Fig. 2E). This is in agreement with the increased VEGF levels in DR (Sone et al., 1997; Frank, 2004), especially of the VEGF₁₆₅ isoform that has been linked with pathological angiogenesis (Simo and Hernandez, 2008). By contrast, no significant differences in the expression levels of tight junction protein 1 (TJP1) (ZO-1 in humans), cone transducin alpha (Tc-α) or β-actin were observed (Fig. 2E).

Retinal sections of control, glucose-treated and mannitol-treated fish were analysed to evaluate retinal cell integrity (Fig. 3). Light microscopy analyses revealed no significant difference in the thickness of the retina or any retinal sublayer (Fig. 3, and data not shown). Furthermore, the histology of control and mannitol-treated retinas were normal and indistinguishable (Fig. 3A,A’,D,D’). However, TUNEL (TdT-mediated dUTP nick-end labelling) analysis of retinal cryosections revealed high levels of apoptosis in mannitol-treated retinas, especially in the vascular and ganglion cells (supplementary material Fig. S1). Hyperglycaemia, which also results in hyperosmolarity, does not induce ganglion cell layer (GCL) apoptosis, perhaps because hyperglycaemia induces expression of the anti-apoptotic factor Bcl2 (Pampfer et al., 2001; Fraser et al., 2007).

By stark contrast, light microscopy revealed that ~60% of glucose-treated fish displayed abnormal retinal histology with prominent defects in cone photoreceptors (Fig. 3B,B’,C,C’). In approximately half of these retinas, hyperglycaemia caused severe morphological defects with pronounced signs of cone photoreceptor degeneration (Fig. 3B,B’). Some pyknotic nuclei were detected in the outer nuclear layer of the fish that were severely affected by the glucose treatment (black arrows in Fig. 3B), although the levels of apoptosis, measured by TUNEL, on retinal cryosections were not increased compared with controls (supplementary material Fig. S1). Necrotic traits observed in the severely affected retinas included cell swelling, tissue vacuolation (red arrows in Fig. 3B) and discontinuities in the plasma membrane (red arrows in Fig. 6B). Thus, necrosis, and not apoptosis, is likely to be responsible for the degenerating cell morphology that was observed in glucose-treated retinas.

To determine whether hyperglycaemia affects the expression of cell type-specific markers in the retina, cryosections were stained by immunohistochemistry (Figs 4, 5 and supplementary material Fig. S2). Staining of ganglion cells with zn5 (Fig. 4A) and amacrine cells for parvalbumin (Fig. 4F) resulted in no significant differences between control, glucose-treated or mannitol-treated fish. Müller glia span almost the entire zebrafish retina and directly contact the retinal vessels at the GCL. In mannitol-treated retinas, glial fibrillary acidic protein (GFAP) labelling of the Müller glia revealed more intense staining, particularly in the end-feet, suggesting glial activation (Fig. 4D,E). In agreement with the light microscopy defects (Fig. 3), cone photoreceptor immunolabelling was aberrant in the glucose-treated retinas (Fig. 4B; Fig. 5). Comprehensive analysis of the individual cone subtypes [ultraviolet (UV), blue and double cones] revealed that all subtypes had less intense labelling following hyperglycaemia (Fig. 5 and supplementary material Fig. S2). However, double cones (stained with zpr1) were the most obviously affected, exhibiting discontinuous labelling and an irregular stubby-shaped cone morphology (Fig. 4B; Fig. 5A and supplementary material Fig. S2C). Staining of the rod outer segments revealed no significant difference between glucose-treated, mannitol-treated or control retinas, although milder defects cannot be excluded, as the zpr3 antibody does not depict individual cell morphology.

To determine whether the morphological abnormalities induced by hyperglycaemia correlate with defects in visual function, ex vivo ERGs were conducted on one eye, and retinal histology examined in the contra-lateral eye (Fig. 6). The ERG records electrical activity from the outer retina in response to flashes of light. Under photopic (light-adapted) settings, the b- and d-wave amplitudes of glucose-treated fish were reduced compared with control or mannitol-treated fish (Fig. 6C-E). The reductions in the b- and d-waves were consistent with diminished signal initiation in ON and OFF bipolar cells, respectively; most likely reflecting perturbed upstream photoreceptors. In agreement, severe morphological defects were observed in the cone photoreceptors of hyperglycaemic fish (Fig. 3). Indeed, the ultrastructure of the retinas from fish with the most diminished ERGs had acute signs of cone neurodegeneration, including abnormal squat cell morphology, deteriorated mitochondria and disrupted outer segments (Fig. 6B). Vacuolations and plasma membrane discontinuities in these hyperglycaemic cones support neurodegeneration by necrosis (Fig. 6B). In agreement with the immunohistochemistry, cone but not rod photoreceptors are particularly sensitive to hyperglycaemia (Fig. 6B).

DR is a prevalent disease that results in progressive blindness. Here, we describe novel pathological features, in particular dysfunctional cones, in a zebrafish model of adult-onset DR that was confirmed by histology, immunohistochemistry, ultrastructure and electroretinography.

Hyperglycaemic treatment results in a thickening of the vessels on the surface of the zebrafish retina, a breakdown of the tightness of interendothelial cell-cell junctions, and a thickening of the vessel basement membrane. No evidence of retinal neovascularisation or significant pericyte loss was observed following the short 30-day treatment. Overall, this data is consistent with archetypal features of early-stage NPDR (Garner, 1970; Gardner et al., 2002; Hernandez and Simo, 2007; Wilkinson-Berka and Miller, 2008) and validates the hyperglycaemic zebrafish as an effective model of DR.

The most striking and most novel phenotype observed was the prevalent cone photoreceptor dysfunction and morphological degeneration in 2-year-old fish. Abnormal cone histology and severe degeneration of cone morphology were apparent in ~60% and ~30%, respectively, of the glucose-treated retinas. This variability in penetrance and severity is consistent with the unequal susceptibility of diabetics to develop retinopathy (Brownlee, 2001; Wong et al., 2006). Preliminary data also suggest that the neurodegeneration is age-dependent. Younger fish (1-year old) exposed to hyperglycaemia display more open interendothelial junctions but no evidence of neurodegeneration (supplementary material Fig. S3). This is in agreement with epidemiological studies wherein the severity of DR in humans is related to a longer duration of diabetes and a younger age of diagnosis (Klein et al., 1984; Wong et al., 2006). Rod but not cone photoreceptor perturbations have been reported in some rodent models of DR (Li et al., 2002; Park et al., 2003; van Eeden et al., 2006). In transgenic mice overexpressing VEGF₁₆₅, significant thinning of the outer nuclear and inner plexiform layers was observed at postnatal day (P)28 (van Eeden et al., 2006). Conflicting results have been reported for STZ-treated retinas: in rats, dominant photoreceptor degeneration occurs, whereas in mice, GCL apoptosis was observed (Park et al., 2003; Martin et al., 2004). In our zebrafish DR model, we observe morphological perturbations of rods, but these changes are significantly milder than the prevalent cone photoreceptor degeneration. However, studies in rodents would not easily identify cone defects owing to the rod-dominant retinas of these nocturnal animals.

Is disrupted cone physiology likely to occur in diabetics? ERG analysis of non-retinopathic diabetics reveals significant reductions in the b-wave amplitude of S-cones (Greenstein et al., 2000; Gardner et al., 2002; Mortlock et al., 2005; Simo et al., 2006). Indeed, an abnormal S-cone ERG is proposed to predict the development of PDR in asymptomatic patients (Yamamoto et al., 1996). Our data suggest that, in diabetics, attenuated ERG responses result from perturbed cone photoreceptor morphology followed by cone degeneration. It is intriguing that, compared with rods, zebrafish cones have an enhanced susceptibility to hyperglycaemia-induced neurodegeneration. We can only speculate on the mechanism, but this finding is recapitulated in humans. Kurtenbach et al. report that, in humans exposed to hyperglycaemia, cone metabolism is significantly enhanced compared with rod metabolism (Kurtenbach et al., 2006).

There is debate about whether the development of neuropathology and microangiopathy in DR are independent events (Simo et al., 2006; Hernandez and Simo, 2007). Our data demonstrate that the cone dysfunction induced by hyperglycaemia develops independent of the early vascular pathologies and GCL apoptosis induced by hyperosmolarity. As mannitol permeabilises the brain/retinal-blood barriers, it has been administered clinically to release intracranial pressure after head trauma, to relieve intraocular pressure in glaucoma, and to facilitate the delivery of drugs to the central nervous system (Lam et al., 2002; Jin et al., 2007; Engelhard et al., 2008). Our results demonstrating neuronal apoptosis in the GCL suggest cautious long-term, clinical use of mannitol. By contrast, hyperglycaemia disrupts the blood-retinal barrier, thickens the retinal vessel basement membrane, and induces cone degeneration (Table 1). Thus, blood-retinal barrier perturbation and elevated VEGF expression are not sufficient to induce cone neurodegeneration.

The pro-angiogenic VEGF is significantly elevated in NPDR where it is thought to induce vascular leakage (Sone et al., 1996; Sone et al., 1997; Brownlee, 2001; Simo et al., 2006). Anti-VEGF treatments targeting vascular pathology are in clinical trials for DR (Campochiaro, 2007). However, VEGF also controls neuronal function. In one report, diminishment of VEGF expression in the retina resulted in photoreceptor degeneration (Saint-Geniez et al., 2008). In another report, overexpression of VEGF in the retina resulted in photoreceptor degeneration (van Eeden et al., 2006). Thus, new therapies for DR may have to move away from VEGF with the goal of inhibiting vascular pathologies and providing neuroprotection for the photoreceptors.

DR is a complex disorder. Our understanding of DR has increased with established in vivo and in vitro models. However, alternative animal models that reveal novel aspects of DR pathophysiology and/or that facilitate discovery of improved therapeutic intervention are welcome. An advantage of the zebrafish DR model is the ease of assessing the well-characterised microvascular and neuronal networks in the retina (Alvarez et al., 2007; Cao et al., 2008; Kitambi et al., 2009). In addition, zebrafish are suited to targeted and random pharmacological screens which may uncover novel drugs that prevent vascular pathology or that protect from neuropathology (Chan et al., 2002; Bayliss et al., 2006; Hong, 2009; Kitambi et al., 2009; Mandrekar and Thakur, 2009).

Tg(fli1:EGFP) zebrafish were maintained according to standard procedures on a 14-hour light/10-hour dark cycle at 28°C (Westerfield, 2000). Three independent glucose treatments were performed on a total of 33 1-year-old zebrafish (control, n=5; glucose, n=28) and 125 2-year-old zebrafish (control, n=29; mannitol, n=32; glucose, n=64), essentially following the protocol of Gleeson et al. (Gleeson et al., 2007). Briefly, groups of five fish per 2-litre tank were subjected to 24-hour treatments with 2% glucose or 2% mannitol (~110 mM), followed by 24 hours in freshly prepared system water, for 30 days. Control fish were kept in the system water for the entire 30-day treatment. To avoid bacterial contamination, system water was prepared using distilled water and adjusted to approximately pH 7 and a conductivity of ~1000 μS. After every 24 hours of glucose/mannitol treatment, fish were rinsed in distilled water for 5–10 seconds before transfer to a clean tank of system water. Fish were fed once a week and monitored twice a day. Distressed animals (~20%) were euthanised immediately. All experiments were carried out under ethical approval granted by the UCD animal research ethics committee and subject to a licence from the Irish Department of Health and Children.

Adult zebrafish were euthanised with benzocaine and rinsed in PBS. Eyes were enucleated and fixed in 4% PFA at 4°C overnight. Retinas were dissected and prepared for imaging of the vasculature, as described previously (Alvarez et al., 2007). The retinas of 1-year-old fish were stored in methanol before analysis, which reduced vessel thickness by ~25%. Vessels were visualised using bright-field, epifluorescence, and confocal microscopy (Zeiss Lumar V12 stereo microscope; Olympus SZX16 stereo zoom microscope; Zeiss UV510 META LSM confocal microscopy system).

Total RNA was isolated from pools of at least 2–4 control, glucose and mannitol retinas using the RNeasy kit (Qiagen). For each condition, RNA extraction was conducted on two separate pools of retinas (from two independent treatments) to have duplicates of the data. cDNA was synthesised by reverse transcription using a Superscript III kit (Invitrogen). Standard PCR was carried out using 1 U of Taq-DNA polymerase (Invitrogen) per reaction. PCR products were visualized on 1% agarose/ethidium bromide gels. The primer sequences and product sizes for each amplicon analysed are listed in supplementary Table S1.

Eyes were fixed in 4% PFA overnight and cryoprotected with 15% sucrose in PBS. Eyes were subsequently embedded in Tissue-Tek (Sakura Finetek Europe) and kept at −80°C before 10-μm cryosections were cut. Cryosections were dried at room temperature for 30 minutes, gradually dehydrated in methanol and then re-hydrated in PBS. Sections were permeabilised with 0.1% Tween in PBS for 10 minutes, treated with blocking solution (PBS, 0.1% Tween, 1% BSA, 5% goat serum) at room temperature, followed by incubation of the primary antibody (diluted 1:50 to 1:100 in blocking solution) at 4°C overnight.

The primary antibody was washed three times, for 5 minutes each, in 5% goat serum-PBS and incubated in the same solution for 1 hour at room temperature with Cy2- or Cy3-labeled anti-mouse or anti-rabbit antibodies, diluted 1:200 in blocking solution (Jackson Labs). Sections were photo-protected with Vectashield (Vector), coverslipped and sealed prior to confocal microscopy. The following primary antibodies used were: (1) zn5, for retinal ganglion cells; (2) zpr1, for double cone photoreceptors; (3) anti-UV opsin, for UV cones; (4) anti-blue opsin, for blue cones; (5) zpr3, for rod outer segments; (6) GFAP, for Müller cells; and (7) parvalbumin, for amacrine cells.

TUNEL was carried out on 10-μm cryosections following the manufacturer’s instructions (In Situ Cell Death Detection kit, Roche). Briefly, sections were treated with proteinase K (2 μg/ml) and dehydrated with a mixture of 70% ethanol/30% acetic acid. Sections were re-fixed in 4% PFA for 5 minutes at 4°C, and 50 μl of labelling mix and 50 μl of TUNEL mix were added to each section. Sections were covered with coverslips and incubated in the dark at 37°C overnight. After washing in PBS, sections were photo-protected with Vectashield (Vector), coverslipped and sealed prior to confocal microscopy.

Retinas for transmission electron microscopy (TEM) were prepared as described previously (Alvarez et al., 2007). Briefly, eyes were fixed in Sorensen phosphate buffer (pH 7.3) containing 4% PFA and 2.5% glutaraldehyde, then embedded in Epon resin. Semi-thin (1 μm) sections were examined under light microscopy for histological analysis, and ultra-thin sections (100 nm) were examined using a Tecnai 12 BioTwin transmission electron microscope (FEI Electron Optics) for ultrastructural investigation.

ERGs were recorded from adult zebrafish eyecups. Briefly, zebrafish were euthanised in 4% 3-aminobenzoic acid methylester and the eyes enucleated. The cornea and lens were removed by microdissection. The eyecup was immediately placed onto filter paper in the recording chamber and oxygenated (95% O₂/5% CO₂). Minimal essential medium (MEM) was perfused directly into the eyecup at ~0.4 ml/minute. The recording electrode, a thick-walled glass capillary (tip diameter: 80–100 μm) containing 0.9% saline, was placed within the eyecup, not touching the retina. The reference electrode, a bleached silver/silver chloride wire, was placed underneath the filter paper. The voltage signal was amplified using a differential P55 pre-amplifier (Grass Instruments) with a bandpass of 0.1 and 100 Hz. A 50-Hz line filter was switched on in order to remove cyclical noise. The eyecup was left to equilibrate in the dark for 30 minutes to adapt to the new conditions. A 300-W tungsten light source was used for stimulus and an LED light source for background light.

The unattenuated irradiances for the background and the stimulus were 50 μW/cm² and 2.8×10³ μW/cm², respectively. Stimulus flash duration was controlled by a mechanical shutter (Newport Corporation) and the rate was controlled by an S48 stimulator (Grass Instruments). Data were acquired using a PC with a NiDAQ 6024E board, running Windows Whole Cell Program (WinWCP, University of Strathclyde). Full-field white flashes (300 mseconds) were presented every 30 seconds to the dark-adapted eye. Following 20 minutes of light adaptation, full-field white flash (300 mseconds) presentation was carried out every 10 seconds. The data were signal averaged (5–10 signals). For all recordings, flashes attenuated by 2.0, 1.0 and 0 (unattenuated) log units were used. The amplitude of the various waves was measured as described previously (Wong et al., 2004).

P values were calculated based on a t-test with two samples and assuming unequal variances.

We would like to thank Professor David Hyde for providing us with anti-UV and anti-blue opsin antibodies; Beata Sapetto-Rebow for technical assistance; Dr David C. Cottell for expert technical advice on TEM; and Professor Alan Stitt, Professor Cormac Taylor, Professor Tom Gardiner, Dr Ralph Nelson and Dr Derek Brazil for valuable comments on the manuscript. This work was supported by a Health Research Board (Ireland) Postdoctoral Fellowship (to Y.A.; PD2006-32), a Science Foundation Ireland Investigator Grant (to B.N.K.; 04/IN3/B559), a Travelling Fellowship from the Company of Biologists (to A.L.R.) and a Science Foundation Ireland UREKA Site Award (to K.C.; 06/UR/B913).