Mitochondrial K_ATP channels stabilize intracellular Ca²⁺ during hypoxia in retinal horizontal cells of goldfish (Carassius auratus)

The retina is part of the central nervous system, and therefore shares the high ATP demand of the brain (Country, 2017; Wong-Riley, 2010). Normally, over 50% of ATP supply is used to pump ions across neuronal membranes, both to maintain membrane potential and to extrude Ca²⁺ from the cytosol (Ames, 1992; Nilsson and Lutz, 2004). When neurons receive insufficient O₂ (hypoxia) or blood supply (ischemia), ATP supply is reduced, neurons depolarize, and intracellular Ca²⁺ concentration ([Ca²⁺]_i) reaches toxic levels (Bickler and Buck, 1998). Several other phenomena increase [Ca²⁺]_i even further, as depolarization opens voltage-gated Ca²⁺ channels and releases excitatory neurotransmitters, such as glutamate (Szydlowska and Tymianski, 2010). High [Ca²⁺]_i is toxic to cells, and is recognized as a common final step among many pathways that trigger cell death (Bickler, 1992; Orrenius and Nicotera, 1994). This pathway from hypoxia to Ca²⁺-mediated cell death is called excitotoxicity, and is central in strokes and several types of blindness (Bickler and Kelleher, 1992; Choi, 1992; Osborne et al., 2004).

Some species have adapted strategies to survive extended periods of hypoxia. Some turtles (e.g. Trachemys spp. and Chrysemys spp.) can survive months in hypoxic, ice-covered ponds by reducing their neuronal activity to a ‘pilot light’ level (Jackson and Heisler, 1982; Lutz and Nilsson, 2004). The crucian carp (Carassius carassius) can survive months of hypoxia at ∼4°C (Holopainen and Pitkänen, 1985; Nilsson and Lutz, 2004), and the congeneric goldfish (Carassius auratus) can withstand hypoxia for days at 4°C, and for hours at room temperature (Walker and Johansen, 1977; Wilkie et al., 2008). Instead of becoming quiescent during hypoxia like turtles, Carassius spp. maintain a reduced level of neural activity and locomotion, possibly to search out waters with greater partial pressures of O₂ (P_O2) (Nilsson, 2001; Nilsson et al., 1993b; Vornanen et al., 2009; Wilkie et al., 2008). These strategies limit ATP turnover, maintain a low [Ca²⁺]_i and prevent excitotoxicity (Bickler, 1992; Bickler and Buck, 1998; Jackson, 2002; Nilsson and Lutz, 2004). The retina is also resistant to hypoxia in these species. Electroretinogram responses to light are greatly diminished during anoxia in turtle (Stensløkken et al., 2008) and carp (Johansson et al., 1997) retinas, but recover completely upon reperfusion. These studies suggest that vision withstands hypoxic insult in hypoxia-tolerant species, but the cellular mechanisms of hypoxia tolerance are completely unexplored in the retina.

Mitochondria are the major consumers of O₂ and producers of ATP, and therefore they are well positioned to sense changes in metabolite supply (Pamenter, 2014). In the turtle brain, hypoxia is sensed by mitochondrial ATP-dependent K⁺ channels (mK_ATP channels), and the resulting mitochondrial depolarization triggers a second messenger pathway which downregulates plasma membrane ion channels and limits Ca²⁺ influx (Hawrysh and Buck, 2013; Pamenter et al., 2016; Zivkovic and Buck, 2010). mK_ATP channels are also required for ischemic preconditioning (IPC) in the mammalian retina, in which brief, sublethal bouts of ischemia protect against subsequent ischemic insult (Dreixler et al., 2008; Roth et al., 2006). Mitochondrial responses to hypoxia are unknown in the ectotherm retina.

Horizontal cells (HCs) are interneurons of the inner nuclear layer of the retina, which receive glutamatergic input from photoreceptors in the dark. In return, they provide inhibitory feedback to photoreceptors to improve visual contrast and colour opponency (Thoreson and Mangel, 2012; Twig et al., 2003). The Ca²⁺-permeable ion channels, Ca²⁺ buffering mechanisms and Ca²⁺ pathways in HCs are well known, especially in teleost fish (reviewed in Country and Jonz, 2017). Teleost HCs present spontaneous Ca²⁺-based action potentials (APs) both in vitro and in situ (Country et al., 2019; Country et al., 2020; Kreitzer et al., 2012; Tachibana, 1981). The function of these APs is unknown, but they have been proposed to affect feedback to photoreceptors and assist in the transition from light to dark, by rapidly depolarizing HCs (Country et al., 2019). These APs have been observed in Ca²⁺ imaging experiments (Country et al., 2019; Country et al., 2020; Kreitzer et al., 2012). Changes in APs may provide insight into how Ca²⁺ homeostasis is being maintained or threatened.

In the present study, we examined how hypoxia affects [Ca²⁺]_i in HCs. We compared responses of isolated HCs from the hypoxia-tolerant goldfish with the hypoxia-sensitive rainbow trout (Oncorhynchus mykiss). We showed that goldfish HCs maintained stable baseline [Ca²⁺]_i even after 1 h of hypoxia, whereas [Ca²⁺]_i increased in response to only 20 min of hypoxia in trout. We showed that hypoxia tolerance depends upon mK_ATP channels: blocking mK_ATP channels increased [Ca²⁺]_i in hypoxic goldfish HCs, whereas opening mK_ATP channels preserved [Ca²⁺]_i in hypoxic trout HCs. We found no evidence for K_ATP channel currents at the plasma membrane. Instead, we report that mitochondria depolarized in hypoxia, and that this depolarization was diminished with mK_ATP channel blockers. Furthermore, we present evidence that hypoxia stabilized [Ca²⁺]_i in a mK_ATP channel-dependent manner, preventing increases in [Ca²⁺]_i caused by blocking glycolysis. These results suggest that hypoxia triggers a neuroprotective pathway to maintain [Ca²⁺]_i homeostasis in goldfish HCs.

All animal care procedures were approved by the University of Ottawa Animal Care and Veterinary Services (protocol BL-1760), in accordance with regulations of the Canadian Council on Animal Care. Adult common goldfish, Carassius auratus (Linnaeus 1758) (a mix of both sexes, ranging from 7 to 60 g), were obtained from Mirdo Importations Canada (Montreal, QC, Canada), and maintained in 170 l tanks at 18°C. Rainbow trout, Oncorhynchus mykiss (Walbaum 1792) (a mix of both sexes, ranging from 112 to 142 g), were obtained from Linwood Acres Trout Farm (Campbellcroft, ON, Canada), and maintained in 170 l tanks at 13°C. Tanks received fresh, aerated and dechloraminated water from a constant flow-through system. Tanks were maintained on a 12 h light:12 h dark photoperiod. Fish were dark adapted for approximately 1 h, euthanized by rapid decapitation and pithed.

HCs were isolated according to Jonz and Barnes (2007). Unless otherwise stated, all reagents and chemicals were sourced from MilliporeSigma (Oakville, ON, Canada). Eyes were enucleated and quickly placed into ice-cold Ca²⁺-free Ringer's solution (in mmol l⁻¹: 120 NaCl, 2.6 KCl, 10 Hepes, 0.5 NaH₂PO₄ and 16 glucose, with pH adjusted to 7.8 with NaOH). The eye was punctured at the ora serrata, and the anterior chamber of the eye was removed. The retina was carefully dissected out from the eye cup and placed in hyaluronidase (100 U ml⁻¹, cat. no. H-3506) in L-15 solution (70% Leibovitz's L-15 medium with glutamine and 30% Ca²⁺-free Ringer's solution) at room temperature. Retinas were washed 3 times for 3 min in fresh L-15 solution, and moved to L-15 solution containing 7 U ml⁻¹ papain (cat. no. 3126, Worthington Biochemical Corporation, Lakewood, NJ, USA) and 2.5 mmol l⁻¹ cysteine for 40 min. Retinas were washed a further 3 times in L-15 solution, and cut into small sections (∼4 mm²). These sections were triturated 3 times to remove excess photoreceptors (Dowling et al., 1985). Then, they were moved to 1 ml of fresh L-15 solution and gently triturated to mechanically dissociate cells. The resulting cell suspension was transferred to 35 mm culture dishes (Corning Inc., Bedford, MA, USA) containing perfusion chambers (cat. no. RC-33DL, Warner Instruments Inc, Hamden, CT, USA). Dishes were pre-coated with 0.01% poly-l-lysine (cat. no. A-005-C) for 10 min, rinsed 3 times with water purified using a MilliQ system, and air-dried before use. Cells were left to settle for 15 min, and rinsed 3 times with extracellular solution (ECS; in mmol l⁻¹: 120 NaCl, 5 KCl, 2.5 CaCl₂, 2 MgCl₂, 10 Hepes and 10 glucose, with pH adjusted to 7.80 with NaOH) before dye loading.

Ca²⁺ imaging was performed according to Country et al. (2019). Briefly, to measure relative changes in free [Ca²⁺]_i, cells were loaded with a membrane-permeant form of the Ca²⁺ indicator dye Fura-2 (Fura-2-LeakRes-AM, cat. no. 1061, Ion Biosciences, San Marcos, TX, USA). Cells were incubated with 5 µmol l⁻¹ Fura-2 and 0.1% v/v of 10% w/v Pluronic F-127 (cat. no. P2443) in ECS at room temperature in the dark. After 30 min, dishes were washed 3 times with fresh ECS.

Cells were first observed with brightfield imaging using a compound microscope (FN-1, Nikon, Tokyo, Japan) and a Nikon 40× water-immersion objective (numerical aperture 0.8). HCs were identified by their characteristically large somata and thick dendrites, as in previous reports (Country et al., 2019; Dowling et al., 1985; Tachibana, 1981). For fluorescence imaging, a Lambda DG-5 wavelength changer (Sutter Instruments, Novato, CA, USA) was used to rapidly alternate between 340 and 380 nm excitation wavelengths once per second. Emission light was filtered through a 510 nm bandpass filter. Images were captured with a CCD camera (QImaging, Surrey, BC, Canada). Fluorescence ratios were recorded with NIS Elements software (Nikon), by setting circular regions of interest within HC somata. We observed no change in fluorescence ratio in cells that were not incubated in Fura-2 (n=4; not shown).

Patch-clamp recordings were taken in the whole-cell configuration. Electrodes were pulled from capillary glass (cat. no. PG52151-4, World Precision Instruments, Sarasota, FL, USA) using a vertical pipette puller (Narishige International Inc., East Meadow, NY, USA) to yield resistances of 4–8 MΩ. Electrodes were prepared immediately before recordings, and backfilled with intracellular solution (in mmol l⁻¹): 10 NaCl, 120 KCl, 2 CaCl₂, 2 MgCl₂, 5 EGTA and 10 Hepes, and adjusted to pH 7.40 with KOH. Only cells with a stable membrane seal above 1 GΩ were used. Protocols were performed with pCLAMP 7 software, an AxoPatch 1D amplifier and a DigiData 1440A digitizer (Axon Instruments, Sunnyvale, CA, USA). Signals were filtered at 5 Hz. The liquid junction potential was calculated to be 5 mV using pCLAMP 7 software and was subtracted from the pipette potential to report membrane potential.

To measure relative changes in mitochondrial membrane potential, we used tetramethylrhodamine ethyl ester (TMRE) dye (cat. no. 87917) in quenching mode. Cells were incubated at room temperature with 50 nmol l⁻¹ TMRE for 30 min in the dark. Dishes were then rinsed 3 times with ECS, for 3 min per wash. Because TMRE uptake is partially dependent upon plasma membrane potential (Perry et al., 2011), TMRE was co-applied with 100 µmol l⁻¹ nifedipine in dimethyl sulfoxide (DMSO) to prevent spontaneous APs (Country et al., 2019) and to prevent subsequent loss of dye from cells. Nifedipine was maintained throughout dye loading, washes and perfusion throughout experiments.

Horizontal cells were identified in brightfield imaging as described above. For fluorescence imaging, cells were illuminated for 100 ms once every 10 s, and once per second for oligomycin and carbonyl cyanide m-chlorophenyl hydrazone (CCCP) controls. Excitation light was filtered through a 545 nm (±15 nm) bandpass filter; emission light was filtered through a 620 nm (±30 nm) bandpass filter. Images were captured with a CCD camera and recorded with NIS Elements software as above.

Cells were continuously superfused at ∼1 ml min⁻¹ by a gravity-driven system. Solutions were drained from the recording chamber at the same rate using a peristaltic pump (Fisher Scientific, Ottawa, ON, Canada). In all imaging experiments, cells were observed for 10 min in ECS before superfusion with treatment solution. After treatment, the bath was washed with ECS for a 10 min recovery period. To create hypoxic solutions, solution reservoirs were bubbled with 100% N₂ for at least 30 min. This procedure produced recording solution with a P_O2 of approximately 25 mmHg (Jonz et al., 2004). Control solutions were bubbled with air for the same duration. Glibenclamide (80 µmol l⁻¹; cat. no. G0639), 5-hydroxydecanoic acid (5-HD; 100 µmol l⁻¹; cat. no. H135) and diazoxide (100 µmol l⁻¹; cat. no. D9035) were dissolved in DMSO, which never surpassed a final concentration of 0.25% v/v. At this concentration, DMSO alone had no effect on Ca²⁺ APs or Ca²⁺ baseline. To block glycolysis in experiments shown in Fig. 6, glucose was replaced with equimolar 2-deoxyglucose (2-DG; cat. no. D8375). CCCP (10 µmol l⁻¹; cat. no. 215911) and oligomycin (10 µmol l⁻¹; cat. no. 495455) were dissolved in DMSO and used as positive and negative controls, respectively, in TMRE experiments.

For Fura-2 recordings, ratios were calculated by dividing the fluorescence emission at 340 nm by that at 380 nm following excitation at these wavelengths (F₃₄₀/F₃₈₀). These values are proportional to [Ca²⁺]_i. Ratios were exported to Excel (Microsoft Corp., Redmond, WA, USA) and analysed in OriginPro 2016 (OriginLab Corp., Northampton, MA, USA). OriginPro's peak analysis gadget was used to approximate baseline using a second-derivative algorithm. The peak analysis gadget was then used to detect peaks using a local maximum function compared with baseline. For analysis, baseline was calculated as the last 30 s of baseline F₃₄₀/F₃₈₀ ratio for a given time period. AP amplitude was measured as the difference between peak F₃₄₀/F₃₈₀ ratio and baseline before the AP. Values of the last two APs of a given time period were averaged. For statistics, the values at the end of a treatment or recovery period were compared with the last 5 min before the treatment, and the fold-change was reported.

For TMRE imaging, fluorescence values were averaged from the last 30 s before, during and after each time period. For CCCP controls, a local maximum function was used to report peak fluorescence.

All statistical tests were performed in Prism 8 (GraphPad Software Inc., San Diego, CA, USA). Data are presented as means±s.d. In Fig. 6E,F, a Kruskal–Wallis test was used with Dunn's post hoc test. All columns were compared with all other columns, with a family-wise significance level of 0.05. For Fig. 8A, a one-way ANOVA was used. Holm–Šidák’s post hoc test was used with a family-wise significance level of 0.05. All normoxic treatment groups were compared with normoxia alone, and all hypoxic groups were compared with hypoxia alone. One-tailed Mann–Whitney tests were used for all other analyses unless otherwise stated.

HCs were isolated from goldfish and rainbow trout, and baseline [Ca²⁺]_i was monitored over time with the ratiometric Ca²⁺ imaging dye Fura-2. HCs of both species presented spontaneous Ca²⁺-based APs for the duration of normoxic recordings as controls (Figs 1 and 2A,B). APs were similar in the two species, and matched previous reports of spontaneous, seconds- to minutes-long Ca²⁺ transients in goldfish HCs (Country et al., 2019; Kreitzer et al., 2012): APs began with a steep rise in [Ca²⁺]_i followed by a prolonged plateau and a sharp return to baseline.

To test the hypothesis that goldfish HCs (but not trout HCs) would maintain [Ca²⁺]_i baseline throughout hypoxia, we treated HCs of both species with normoxic or hypoxic solutions and monitored [Ca²⁺]_i over time. Representative brightfield images are shown in Fig. 1, along with pseudocolour representations of [Ca²⁺]_i during APs after treatment with normoxia or hypoxia. Traces of [Ca²⁺]_i over time are shown in Fig. 2A for goldfish, and Fig. 2B for trout. In goldfish HCs, [Ca²⁺]_i baseline did not change in response to 20 min normoxia (n=12) or 20 min hypoxia (n=6; Fig. 2C). Similarly, [Ca²⁺]_i did not significantly increase throughout 1 h normoxia (n=9) or 1 h hypoxia (n=5; Figs 1A–D and 2A,C).

Restoring [Ca²⁺]_i after an AP would likely require ATP for extrusion or sequestration, and APs during hypoxia may therefore strain ATP supply. We tested the hypothesis that AP amplitude would be reduced during hypoxic treatment, possibly to reduce ATP demand. Even during 1 h exposures to normoxic solution, APs persisted throughout treatment. However, in goldfish HCs, AP amplitude decreased during hypoxia (n=8; P=0.0242) when compared with normoxia (n=4; Fig. 2D). Amplitude returned to pre-treatment values upon reperfusion with normoxic solution, and there was no difference in AP amplitude between hypoxia-treated (n=7) and normoxia-treated (n=3) cells in the recovery period (Fig. 2D).

In trout HCs, [Ca²⁺]_i baseline was stable in normoxia (1.18±0.25-fold change, n=13) but increased irreversibly in response to 20 min hypoxia (1.68±0.81-fold change, n=11; Figs 1E–H and 2B,C). This represents a 50% increase in [Ca²⁺]_i compared with that in normoxia (P=0.0128). Unlike the case in goldfish HCs, AP amplitude was not significantly diminished during hypoxic treatment (normoxia n=13, hypoxia n=9; Fig. 2E). These data support the hypothesis that HCs of the hypoxia-tolerant goldfish, but not hypoxia-sensitive trout, maintain a stable [Ca²⁺]_i baseline and reduce AP amplitude during hypoxia.

Next, we tested the hypothesis that mK_ATP channels might be involved in neuroprotection against hypoxia, as has been reported in turtle cortical neurons (Hogg et al., 2014; Pamenter et al., 2008b) and in IPC in the mammalian retina (Dreixler et al., 2008; Roth et al., 2006). First, we exposed isolated goldfish HCs to the mK_ATP channel antagonists glibenclamide (Fig. 3) and 5-HD (Fig. 4), with either normoxia or hypoxia for 20 min. In normoxia, glibenclamide (80 µmol l⁻¹; Pamenter et al., 2008b) had no effect on post-treatment [Ca²⁺]_i baseline (n=11; Fig. 3A). When co-applied with hypoxia, mK_ATP channel inhibition with glibenclamide led to an increase in [Ca²⁺]_i (1.26±0.23-fold change, n=12; Fig. 3B), a 22% increase over glibenclamide alone (P=0.0344; Fig. 3C). Consistent with the effects of hypoxia in Fig. 2D, AP amplitude decreased during treatment with glibenclamide and hypoxia (n=11) when compared with glibenclamide alone (n=10; P=0.0305; Fig. 3D). AP amplitude was partially restored when glibenclamide and hypoxia were washed out, and there was no difference in AP amplitude during recovery (normoxia n=9, hypoxia n=12; Fig. 3D).

Likewise, 20 min of 5-HD (100 µmol l⁻¹; Pamenter et al., 2008b) alone had no effect on post-treatment [Ca²⁺]_i baseline (1.00±0.07-fold change, n=8; Fig. 4A). Co-application with hypoxia increased baseline 1.06±0.08-fold (n=10; Fig. 4B). This difference was subtle (6%) but significant (P=0.0273; Fig. 4C). AP amplitude decreased slightly during application of 5-HD alone (0.81±0.23-fold change, n=6) as well as during co-application of 5-HD and hypoxia (0.84±0.15-fold change, n=9; Fig. 4D). Upon recovery, amplitude was restored and there were no differences among treatments (Fig. 4D).

To test the hypothesis that a mK_ATP channel-dependent pathway could stabilize [Ca²⁺]_i in trout HCs, we applied the mK_ATP channel agonist diazoxide (100 µmol l⁻¹; Dukoff et al., 2014; Pamenter et al., 2008b) to isolated trout HCs in the presence and absence of hypoxia. [Ca²⁺]_i baseline did not change with diazoxide alone (n=6; Fig. 5A), or when diazoxide was co-applied with hypoxia (n=10; Fig. 5B,C). AP amplitude decreased during both conditions (diazoxide and normoxia, 0.66±0.19-fold change, n=5; diazoxide and hypoxia, 0.48±0.14-fold change, n=7; Fig. 5D). The decrease in hypoxia was greater but did not differ at the prescribed level of statistical significance (P=0.053). AP amplitude increased upon reperfusion, with no differences between treatment groups (normoxia n=7, hypoxia n=8).

The previous experiments suggest that a mK_ATP channel-dependent pathway may stabilize cytosolic [Ca²⁺] baseline during hypoxia. To test whether this pathway was also active in another low-energy condition, we replaced glucose with equimolar 2-DG to block glycolysis (Dvoriantchikova et al., 2012) in isolated goldfish HCs. Exposure to 2-DG for 1 h increased post-treatment [Ca²⁺]_i, compared with 1 h normoxic controls (normoxia, 1.16±0.33-fold change, n=9; 2-DG, 2.36±1.57, n=9; P=0.0106; Fig. 6A,E). However, co-application of 2-DG and hypoxia prevented this increase almost completely (1.23±0.10-fold change, n=4; Fig. 6B,E). This protection was abolished in the presence of 5-HD (2.19±1.03-fold change, n=9; P=0.0055 versus normoxic control; Fig. 6C,E). Diazoxide (100 µmol l⁻¹), like hypoxia, prevented Ca²⁺ from increasing (1.46±0.33-fold change, n=6; Fig. 6D,E). These data further suggest a role for mK_ATP channels in stabilizing [Ca²⁺]_i during hypoxia in goldfish HCs.

Next, we tested whether blocking glycolysis would reduce AP amplitude, as we observed during hypoxia in previous experiments. Amplitude decreased during all treatments, when compared with normoxic controls (Fig. 6F). This decrease was significant for 2-DG (n=6; P=0.0309) and for 2-DG plus hypoxia (n=3; P=0.0219) but was not significant for 2-DG plus hypoxia and 5-HD (n=4). A decrease in amplitude was not apparent following treatment with 2-DG plus diazoxide, but because only a small proportion of cells displayed APs under these conditions, this could not be assessed. Amplitude was partially restored in all groups upon normoxic reperfusion.

Diazoxide, glibenclamide and 5-HD are often used to activate or block mK_ATP channels, but some reports suggest they may additionally affect K_ATP channels at the plasma membrane (Baumann et al., 2002; Hanley and Daut, 2005; Hill et al., 2021). HCs express at least three plasmalemmal K⁺ channels (an inward rectifier, a delayed outward rectifier and a transient outward channel) (Jonz and Barnes, 2007; Lasater, 1991; Tachibana, 1983), but to our knowledge, plasmalemmal K_ATP channels have not been found in HCs. To test for their presence in goldfish HCs, we used whole-cell patch clamp to record membrane currents while applying the aforementioned K_ATP channel drugs. In the absence of these drugs, HCs presented a stable membrane current while voltage clamped to −65 mV. Hyperpolarizing and depolarizing voltage steps from −65 mV produced an ‘N-shaped’ profile of whole-cell currents, as in previous reports (Jonz and Barnes, 2007; Shingai and Christensen, 1986; Tachibana, 1983). Membrane currents remained stable during and after 5 min superfusion of the mK_ATP channel agonist diazoxide (100 µmol l⁻¹; Fig. 7A; n=5), and the I–V relationship did not change during or after treatment (Fig. 7B; n=5). Similarly, membrane currents were unchanged during and after 5 min superfusion of 5-HD (100 µmol l⁻¹; Fig. 7C,D; n=6) and glibenclamide (80 µmol l⁻¹; Fig. 7E,F; n=6).

Using TMRE imaging, we tested the hypothesis that mK_ATP channels open to depolarize mitochondria during hypoxia. TMRE fluorescence typically decayed gradually in normoxia (0.88±0.15-fold change within 5 min, n=23; Fig. 8A,B). Fluorescence significantly increased within 5 min of hypoxia (1.19±0.25-fold change, n=27; P<0.0001; Fig. 8A,C). Similarly, 5-HD had no significant effect alone (n=22; Fig. 8A,D), but reduced fluorescence in hypoxia by 9%. However, this effect was not significant (1.102±0.29-fold change, n=18; Fig. 8A,E). Glibenclamide alone (n=19; Fig. 8A,F) had no significant effect on fluorescence, but significantly reduced the response to hypoxia (0.99±0.15-fold change, n=14; P=0.0312; Fig. 8A,G).

At low TMRE concentrations, as mitochondria hyperpolarize, they take up more TMRE dye and fluorescence increases. But at high concentrations of TMRE, as used in the present experiments, TMRE dimerizes and quenches itself, so that fluorescence decreases as mitochondria hyperpolarize (Perry et al., 2011; Zorova et al., 2018). To confirm that TMRE attained this ‘quenching mode’ in our preparation, we applied CCCP, a proton ionophore, to dissipate the mitochondrial H⁺ gradient and depolarize mitochondria (Brown et al., 2006; Perry et al., 2011). In quenching mode, depolarization by CCCP is expected to first increase fluorescence (as dye loss reduces quenching) and then greatly diminish it (as dye concentration drops low enough to enter ‘non-quenching mode’) (Perry et al., 2011). In line with this model, fluorescence rapidly increased with 10 µmol l⁻¹ CCCP (1.56±0.33-fold change, n=4; P<0.0001; Fig. 8A,H); then, fluorescence was nearly abolished. Oligomycin is known to block complex V of the electron transport chain, preventing the dissipation of the mitochondrial H⁺ gradient and hyperpolarizing mitochondria. In our preparation, 10 µmol l⁻¹ oligomycin decreased fluorescence within 5 min (0.68±0.28-fold change, n=44; P=0.0049; Fig. 8A,I).

The present study establishes part of a novel neuroprotective pathway, which maintains [Ca²⁺]_i throughout hypoxia in HCs of the hypoxia-tolerant goldfish retina. We showed that goldfish HCs require functional mK_ATP channels to maintain [Ca²⁺]_i. Hypoxia prevented [Ca²⁺]_i from increasing in response to glycolytic inhibition with 2-DG. This protective effect was negated by the mK_ATP channel antagonist 5-HD and mimicked by the mK_ATP channel agonist diazoxide, further suggesting that mK_ATP channels are necessary for hypoxia tolerance. We found no evidence for plasmalemmal K_ATP channel currents. Instead, our data support a role for K_ATP channels in mitochondria, as mitochondria depolarized early in hypoxia in a mK_ATP channel-dependent manner.

Previous reports have shown hypoxia tolerance in the retina of Carassius spp. In the crucian carp, electroretinogram activity decreased during anoxia and was completely restored upon reperfusion, suggesting neuronal function and vision are not lost by anoxic insult (Johansson et al., 1997). In the goldfish retina, metabotropic glutamate receptors protected against apoptosis caused by 3 h of hypoxia and reperfusion at room temperature (Beraudi et al., 2007). In the present study, we showed that neurons of the goldfish retina resisted hypoxic increases in [Ca²⁺]_i, which have been established as a precursor to excitotoxic cell death (Choi, 1992; Szydlowska and Tymianski, 2010).

The present study demonstrated that a mK_ATP channel-dependent pathway is required to stabilize [Ca²⁺]_i baseline in goldfish HCs. This work agrees with a growing body of evidence which shows a central role for mK_ATP channels and mitochondria in protection against hypoxia across a wide array of tissues and species. mK_ATP channel protection has been shown in diverse organs in mammals including the heart (O'Rourke, 2004; Obal et al., 2005; Wang et al., 2001), brain (Heurteaux et al., 1995), kidney (O'Sullivan et al., 2008; Rahgozar et al., 2003), liver (O'Sullivan et al., 2008) and retina (Dreixler et al., 2008; Ettaiche et al., 2001; Roth et al., 2006). Likewise, mK_ATP channels protect against hypoxia in hypoxia-tolerant species, including in the turtle brain (Dukoff et al., 2014; Pamenter et al., 2008b) and in cardiomyocytes of hypoxia-tolerant fish, such as the goldfish (Chen et al., 2005), the yellowtail flounder (Limanda ferruginea) (MacCormack and Driedzic, 2002) and the armoured catfish (Liposarcus pardalis) (MacCormack et al., 2003).

Among hypoxia-tolerant species, the neuronal mechanisms of hypoxia tolerance have been best studied in cortical neurons of the red-eared slider turtle (Trachemys scripta). These neurons maintained [Ca²⁺]_i for hours in hypoxia, but not when glycolysis was inhibited with iodoacetate (Bickler, 1992). Pre-treating brain slices with anoxia protected against iodoacetate insult, similar to how hypoxia prevented increases in [Ca²⁺]_i during glycolytic inhibition in the present study. Furthermore, mK_ATP channels were necessary for neuroprotection in painted turtle (Chrysemys picta) cortical neurons (Pamenter et al., 2008b; Zivkovic and Buck, 2010), and may have conferred neuroprotection in HCs in the present report.

Although the exact mechanisms of mK_ATP channel-dependent neuroprotection in turtles have yet to be established, there is evidence that hypoxia opens mK_ATP channels, which depolarizes mitochondria, triggering a subtle release of Ca²⁺ via the mitochondrial permeability transition pore (Hawrysh and Buck, 2013; Pamenter et al., 2008b; Zivkovic and Buck, 2010). This slight increase in [Ca²⁺]_i activates second messenger pathways and ultimately downregulates AMPA receptors (Pamenter et al., 2008a; Zivkovic and Buck, 2010), NMDA receptors (Pamenter et al., 2008b; Shin et al., 2005), and Ca²⁺-dependent K⁺ (K_Ca) channels (Rodgers-Garlick et al., 2013). This downregulation of membrane currents, or ‘channel arrest’ (Hochachka, 1986), would limit Ca²⁺ influx and reduce the ATP required to maintain transmembrane ion gradients, so as to conserve energy during hypoxia. Channel arrest has been demonstrated in the goldfish brain (Wilkie et al., 2008), but not yet in the goldfish retina.

The mK_ATP channel protection we observed likely involves a different pathway from that of the turtle brain. In contrast with the turtle model, we did not observe an immediate increase in [Ca²⁺]_i after applying hypoxia to goldfish HCs, suggesting any mK_ATP channel pathway in goldfish HCs is Ca²⁺ independent. Instead, we report that mK_ATP channel blockade increased [Ca²⁺]_i in hypoxia, whereas in the turtle, mK_ATP channel activation increased [Ca²⁺]_i (Hawrysh and Buck, 2013; Pamenter et al., 2008b). Mitochondrial Ca²⁺ release may not be a feasible strategy in HCs because they appear to have few mitochondria. Mitochondria are scarce in electron micrographs (Dowling and Boycott, 1966; Yamada and Ishikawa, 1965) and cytochrome c oxidase labelling is weak in HC somata (Kageyama and Meyer, 1988). Furthermore, Ca²⁺ is elevated in darkness, so a Ca²⁺-dependent pathway might not be feasible (Country and Jonz, 2017; Hayashida and Yagi, 2002).

Our results may align more closely with IPC in the mammalian retina, which is also Ca²⁺ independent. In the rat retina, mK_ATP channels are both necessary and sufficient for IPC (Dreixler et al., 2008; Roth et al., 2006; Stokfisz et al., 2017). Once open, mK_ATP activates two Ca²⁺-independent isoforms of protein kinase C (PKC): PKC-δ and PKC-ε (Dreixler et al., 2008). HCs likely benefit because IPC prevents thinning of the inner nuclear and outer plexiform layers (Roth et al., 1998; Toprak et al., 2002; Zhu et al., 2007), and the δ isoform localizes to mammalian HCs (Dreixler et al., 2008). It is possible that neuroprotection is Ca²⁺ independent in both mammalian and goldfish HCs, because [Ca²⁺]_i is already elevated throughout darkness as a result of constant glutamatergic input from photoreceptors. Future experiments could compare this HC pathway with neuroprotection in the inner retina (e.g. ganglion cells), as inner retinal neurons are not expected to undergo prolonged increases in [Ca²⁺]_i during the dark, unlike HCs.

These models may offer insight into how mK_ATP channels open in the goldfish retina. Local decreases in [ATP] have been proposed to increase mK_ATP channel open probability in the turtle (Hogg et al., 2014), yet this is unlikely in goldfish HCs because blocking glycolysis dysregulated [Ca²⁺]_i in the current study. Adenosine and A1/A2a receptors are necessary for neuroprotection in the turtle (Buck and Bickler, 1995; Pérez-Pinzón et al., 1993) and they open mK_ATP channels during IPC in the mammalian retina (Li and Roth, 1999; Li et al., 2000). Adenosine may be a likely candidate as adenylates are released by goldfish HCs, where they may buffer the synaptic cleft pH and modulate feedback to cones (Vroman et al., 2014). Other candidate mK_ATP channel openers include nitric oxide and mitochondrion-specific isoforms of PKC (Korge et al., 2002; Sasaki et al., 2000). PKC is present in HCs of the catfish (Pfeiffer-Linn and Lasater, 1998) and nitric oxide synthase is present in goldfish H1 cells (Daniels and Baldridge, 2011).

The specificity of mK_ATP channel agonists and antagonists has been called into question, as has the existence of mK_ATP channels (reviewed in Garlid and Halestrap, 2012; Roth et al., 2006). Glibenclamide may inhibit plasmalemmal K_ATP channels, and 5-HD is a fatty acid which can interfere with lipid oxidation in mitochondria (Hanley and Daut, 2005). Although diazoxide is over 1000 times more specific for mK_ATP channels than for plasmalemmal K_ATP channels (Garlid et al., 1996), it has been shown to inhibit complex II of the electron transport chain and to lead to mitochondrial uncoupling (Kowaltowski et al., 2001). However, these off-target effects may require higher concentrations than those used in mK_ATP channel experiments (Garlid and Halestrap, 2012; Kowaltowski et al., 2001), and they are not themselves sufficient to explain the drugs’ neuroprotective impacts in hypoxia (reviewed in O'Rourke, 2004; Roth et al., 2006). The present report provides further evidence for mK_ATP channels, as we show mitochondrial depolarization was dependent on K_ATP channels yet we found no evidence for K_ATP channel currents at the plasma membrane. Perhaps most convincingly, mK_ATP channel proteins were recently reconstituted and shown to confer protection via IPC and activation by diazoxide (Paggio et al., 2019), supporting the existence and protective role of mK_ATP channels. They are expressed and mediate hypoxic protection in a wide range of species, including Caenorhabditis elegans (Wojtovich et al., 2012), turtles (Dukoff et al., 2014; Pamenter et al., 2008b), rats (Piriou et al., 2000; Roth et al., 2006) and humans (Jiang et al., 2006). In fish, mK_ATP channels have been found in the goldfish heart (Chen et al., 2005) and trout liver (Onukwufor et al., 2016).

In the present study, AP amplitude was reduced as ATP supply was challenged with hypoxia or with inhibition of glycolysis. Reductions in AP amplitude may be a neuroprotective response to hypoxia, so as to limit Ca²⁺ influx and reduce the ATP demand for extrusion via Ca²⁺ ATPases. In support of this theory, mK_ATP channel activation in trout reduced AP amplitude and stabilized [Ca²⁺]_i baseline. However, reductions in AP amplitude were not sufficient to maintain [Ca²⁺]_i, as [Ca²⁺]_i was dysregulated in multiple conditions despite amplitude reductions (i.e. glibenclamide and hypoxia; 2-DG; 2-DG, hypoxia and 5-HD). This may be an example of channel arrest, either of L-type Ca²⁺ channels or of ryanodine receptors (Country et al., 2019), to limit increases in [Ca²⁺]_i during APs. Channel arrest is known to protect neurons of the goldfish telencephalon (Wilkie et al., 2008) and the turtle cortex (Pamenter et al., 2008a,b; Rodgers-Garlick et al., 2013); however, to our knowledge it has yet to be shown in the retina.

Rainbow trout have been shown to be sensitive to hypoxia in previous reports (e.g. Hylland et al., 1995; Iftikar et al., 2010; Nilsson et al., 1993a), and a-, b-, and c-waves of the trout electroretinogram were all negatively affected by hypoxia (Ubels, 1979). Our results confirmed the hypoxia sensitivity of trout retinal neurons. Furthermore, activating mK_ATP channels in trout HCs prevented hypoxia from increasing [Ca²⁺]_i baseline, suggesting the mechanisms of hypoxia tolerance downstream of mK_ATP channel activation are conserved among fish. An additional take-home point from these studies relates to the hypoxia sensitivity of HCs, more generally. In mammals, HCs were resistant to excitotoxic cell death (Kim et al., 2010), possibly because they have adapted to long periods of high [Ca²⁺]_i in the dark. By showing that [Ca²⁺]_i is dysregulated in trout but not goldfish, our results suggest that neuroprotection is specific to goldfish, and not to HCs more generally. Furthermore, we showed that trout HCs are susceptible to hypoxic cell death even in the absence of glutamate. But in the dark, glutamate from photoreceptors increases HC Ca²⁺ to a steady concentration of ∼600 nmol l⁻¹ (Hayashida and Yagi, 2002), so that more ATP turnover would be required to maintain Ca²⁺ homeostasis. The mK_ATP channel pathway and other forms of neuroprotection are likely even more crucial for cell survival in darkness.

Turtles are often considered the champions of hypoxia tolerance, partly because they greatly reduce electrical activity in their brains during anoxia, compared with modest reductions in goldfish and crucian carp (Nilsson and Lutz, 2004). But in the retina, this paradigm may be reversed. Electroretinogram recordings suggest electrical activity in the turtle retina is reduced modestly (50%) in hypoxia (Stensløkken et al., 2008), and turtles respond to visual stimuli during their hibernation (Madsen et al., 2013). Carp shows upwards of a 90% reduction in electroretinogram activity in hypoxia (Johansson et al., 1997). Given this difference, the Carassius retina merits consideration as a novel model of neuroprotection against hypoxia. This model may be important because strokes, and the majority of retinal diseases of the eye, involve ischaemic cell death (Hayreh and Zimmerman, 2005; Osborne et al., 2004). Blood flow is negatively affected in the eye in branch and central retinal artery occlusions, retinal vein occlusions and glaucoma (Hayreh and Zimmerman, 2005; Osborne et al., 2004, 1999). Other diseases, such as age-related macular degeneration and diabetic retinopathy, can leave affected areas of the retina hypoxic, leading to cell death and vision loss (Blasiak et al., 2014; Sim et al., 2014). By clarifying the cellular mechanisms by which goldfish retinal neurons survive hypoxia, we may learn of new strategies to mitigate or prevent retinal disease and protect vision.

The present study revealed a novel mK_ATP channel-dependent mechanism which may protect the vertebrate retina from hypoxia. We showed that [Ca²⁺]_i was dysregulated in hypoxia in HCs of the hypoxia-sensitive trout; in contrast, HCs of the hypoxia-tolerant goldfish maintained [Ca²⁺]_i even after 1 h of hypoxia. mK_ATP channels were necessary to stabilize [Ca²⁺]_i in goldfish HCs and sufficient for Ca²⁺ homeostasis in trout HCs. We ruled out the involvement of plasmalemmal K_ATP channels, while showing that mK_ATP channels depolarize mitochondria early in hypoxia. Finally, we showed that hypoxia stabilized [Ca²⁺]_i even as glycolysis was inhibited. This novel in vitro model is the first to explore intrinsic hypoxia tolerance in retinal neurons, and may lead to novel insights to protect against ischaemic diseases of the eye.

The authors thank Benjamin F. N. Campbell for advice, and Dr Kathleen M. Gilmour for generously providing rainbow trout.