Scavenging of reactive oxygen species mimics the anoxic response in goldfish pyramidal neurons

Most vertebrates rely on a continuous oxygen supply because they cannot survive for long on the reduced ATP production of anaerobic glycolysis (Schmidt-Nielsen and Pennycuik, 1961). When the partial pressure of oxygen (P_O2) approaches zero, ATP synthesis becomes limited to anaerobic glycolysis. The amount of ATP from this pathway is only about 1/10th of the ATP normally generated through oxidative phosphorylation, making it an inefficient way to maintain normal cellular functioning (Kopp et al., 1984; Santos et al., 1996). As ATP production falls in cells, the ion-motive ATPases fail, causing a ripple effect that leads to depolarization of the membrane potential, cell swelling and cell death (Lipton, 1999; Choi, 1992; Rossi et al., 2000; Siesjö, 1989).

Numerous environments, such as oxygen minimum zones, have fluctuating ambient O₂ levels, causing local hypoxia and anoxia. Organisms in these areas adapt to these otherwise lethal conditions through a host of physiological and genetic alterations (Gorr et al., 2010; Shen and Powell-Coffman, 2003). Two closely related fish species, crucian carp (Carassiuscarassius) and goldfish (Carassius auratus), have also evolved an extraordinary anoxia tolerance by living in ice-covered ponds and lakes in northern Europe and Asia, and the core functional anoxia-tolerant pathways and mechanisms are similar in the two species (Nilsson, 2001). Similarly, two turtle species, the western painted turtle (Chrysemys picta bellii) and red-eared slider (Trachemys scripta elegans), experience winter ice cover and have also evolved strategies to cope with long-term anoxia.

Tissues vary considerably in their oxygen sensitivity, with the brain being one of the most oxygen-sensitive organs, yet also the one with the highest oxygen demand (Clarke and Sokoloff, 1999; Quastel and Wheatley, 1932; Schmidt-Nielsen, 1984). Neurons are especially sensitive to O₂ fluctuations because they have an intrinsically high metabolic rate as a result of the cost of re-establishing ion gradients following an action potential (AP). It is estimated that 50–80% of total brain ATP turnover can be used to maintain pumping at glutamatergic synapses (Howarth et al., 2012; Schmidt-Nielsen, 1984). Thus, maintaining ion homeostasis in the brain under anoxic conditions is likely the most challenging task for anoxia-tolerant organisms. Remarkably, these fish are active in the water column despite a limiting energy budget, posing a challenge to conserve energy with the brain turned on (Nilsson et al., 1993). Regardless, goldfish calorimetric studies show a 70% decrease in whole-body metabolism under prolonged anoxia at 3°C (van Waversveld et al., 1989), while the turtle species exhibit an extreme 99% decrease in whole-body metabolism when exposed to both anoxia and low temperature (Jackson, 2000).

Although the protective mechanisms of anoxia-tolerant organisms diverge, suggesting that there is not a universal pathway, carp and turtle species do exhibit three key similarities. The first strategy is metabolic depression, which is described as lowering the ATP consumption to synchronize with decreases in ATP production (Hochachka, 1986; Nilsson and Lutz, 1991). Channel arrest, the second tool for fighting anoxia, is the reduction in cell membrane conductance to match the decline in activity of the ATP-dependent ion pumps (Erecińska and Silver, 1994; Hochachka, 1986; Wilkie et al., 2008). Finally, excitatory cells implement spike and synaptic arrest, which is a reduction in brain electrical activity, through decreased AP firing (Buck and Pamenter, 2018; Sick et al., 1993). There must be a signalling mechanism that initiates these strategies and a hypoxia-mediated reduction in reactive oxygen species (ROS) concentration may serve this purpose.

ROS are known to cause cellular damage through oxidative stress (Schieber and Chandel, 2014), and neurons in particular have an increased vulnerability to oxidative stress because of their high metabolic rate and low regeneration rate (Uttara et al., 2009). Despite this sensitivity to ROS, low concentrations of ROS behave as an important signalling agent across species and tissue types (Gough and Cotter, 2011). Each type of ROS has discrete biological properties, including half-life, lipid solubility and chemical reactivity, forming a unique class of signalling molecules (D'Autréaux and Toledano, 2007). In brain tissue, ROS modify and initiate long-term synaptic potentiation while ROS produced by microglia and astrocytes mediate synaptic and non-synaptic interactions between glia and neurons. ROS signalling roles are conserved throughout evolutionary history and are found in bacteria to complex mammalian cellular networks (D'Autréaux and Toledano, 2007; Marinho, et al., 2014; Ray et al., 2012).

The main ROS species include O₂⁻, OH⁻ and H₂O₂, which all vary in their toxicity and targets. Generally, the oxyradical reactivity correlates with toxicity and thus is inversely related to signalling ability. H₂O₂ is a neutral ROS, the least reactive of the three, and typical intracellular concentrations are 10⁻⁷ mol l⁻¹ (D'Autréaux and Toledano, 2007). In addition to enzymes producing H₂O₂, O₂⁻ is converted to H₂O₂ in the cytosol. The general consensus is that H₂O₂ is the primary signalling molecule, because of a host of factors: it is the most stable of the various ROS, is the weakest oxidant and has specific targets. Additionally, it can affect the activity of tyrosine phosphatase enzymes, known to oppose protein kinase activity, create novel recognition motifs, regulate enzyme activity and impact protein–protein interactions and cellular localization of proteins (Fruehauf and Meyskens, 2007; Hamanaka and Chandel, 2010; Lee et al., 1998; Meng et al., 2002). Under physiological conditions, H₂O₂ migrates from sites of production, such as the mitochondria, into the cytoplasm, where it takes part in signal transduction pathways (Antunes and Cadenas, 2000; Rigoulet et al., 2010). Because of its neutral charge, H₂O₂ is thought to passively diffuse through membranes, although a growing body of evidence suggests that this is limited as a result of the existence of H₂O₂ gradients across the plasma membrane, which are not completely permeable to H₂O₂ extrusion (Bienert et al., 2006; Rhee, 2006). Water and H₂O₂ are structurally similar, have similar dipole moments, and can form hydrogen bonds, making aquaporins a potential route for H₂O₂ diffusion (Bienert et al., 2006).

The primary source of ROS is thought to originate from electron leakage between mitochondrial complex I, ubiquinone and complex II and III. Three possible mechanisms of ROS production have been proposed: (1) when mitochondria are damaged and NADH is reduced, reverse electron flow through complex I seems to be the primary source; (2) under conditions of high proton motive force and little ATP production, such as following ischaemia and reperfusion, again reverse electron flow through complex I seems to be the primary source; and (3) when mitochondria are actively making ATP there is ROS production from complex III (Murphy, 2009).

Any alterations in the rate of mitochondrial ROS formation could act as a redox signalling mechanism (Murphy, 2009), and these could include respiratory rate, pH of the mitochondrial matrix, surrounding O₂ tension, amplitude of mitochondrial inner membrane potential, and damage to the respiratory chain (Hamanaka and Chandel, 2010; Starkov, 2008). How mitochondria regulate ROS production remains unclear (Hamanaka and Chandel, 2010) but it is apparent that ROS play a role in cellular signalling, suggesting that the production of these oxidants is not accidental (Cadenas et al., 1977; Han et al., 2001; Starkov and Fiskum, 2001; Turrens, 2003).

Little is known about the role of ROS signalling in anoxia-tolerant species and the majority of the work to date has involved anoxia-tolerant turtle brain and heart. We know that in turtle brain mitochondrial [ROS] decreases at the onset of anoxia (Milton et al., 2007; Pamenter et al., 2007; Hogg et al., 2015; Bundgaard et al., 2018). Furthermore, pharmacologically decreasing mitochondrial [ROS] increases NMDA (N-methyl-d-aspartate) currents by 100%, but this is countered in vivo by an increase in intracellular [Ca²⁺] and likely post-translational modification of the NMDA receptor (Buck and Bickler, 1998; Dukoff et al., 2014; Shin and Buck, 2003; Shin et al., 2005). Additionally, mitochondrial ROS also affect inhibitory neurotransmission via GABAergic pathways. Electrophysiological data (Hogg et al., 2015) collected from turtle brain sheets demonstrate that under normoxic conditions, mitochondrial ROS scavenging can augment GABAergic transmission, enhancing tonic GABA_A receptor currents, increasing mini-inhibitory postsynaptic current frequency and increasing spontaneous and giant inhibitory postsynaptic current amplitude. Additionally, pharmacologically decreasing mitochondrial ROS shifts the membrane potential to the reversal potential for the GABA_A receptor by an increase in whole-cell conductance that leads to a decrease in AP firing frequency (Hogg et al., 2015). This results in a coordinated suppression of electrical activity in pyramidal neurons, which is similar to that seen during anoxia, providing direct evidence for synaptic arrest and making mitochondrial ROS a key constituent orchestrating the anoxic defence mechanism.

Many studies on anoxia-tolerant animals focus on biochemical and systematic adaptations but fail to delve into the neurological foundations of anoxia tolerance and the signalling molecules responsible for the onset of anoxia-mediated neuroprotection. ROS are known signalling molecules in various biological systems, and a role in orchestrating the anoxic response in anoxia-tolerant painted turtles has been established (Hogg et al., 2015). However, whether this phenomenon is conserved across anoxia-tolerant organisms is unknown. Given that mitochondria are a significant ROS producer and ROS are intimately linked to respiration and the presence of oxygen, the aim of this study was to investigate whether manipulation of mitochondrial ROS production with scavengers can invoke changes in neuronal electrophysiological properties consistent with anoxia tolerance in goldfish telencephalic neurons.

This study was approved by the University of Toronto Animal Care Committee and adhered to the ethical guidelines outlined in the Guide to the Care and Use of Experimental Animals, published by the Canadian Council on Animal Care. Male and female goldfish, Carassius auratus (Linnaeus 1758), 10–13 cm total length, were obtained from AQUAlity Tropical Fish Wholesale, Inc. (Mississauga, ON, Canada) and transported to the aquatic facility housed at Ramsay Wright Zoological Laboratories (Toronto, ON, Canada). Fish were kept in plastic tanks (2×4×1.5 m) with a de-chlorinated flow-through freshwater system maintained at 20°C. Animals were subjected to a 12 h light:dark photoperiod, had environmental enrichment, and were fed goldfish pellets daily.

Rapid decapitation was followed by excision of the whole brain, which was subsequently stored in a vial containing chilled (4°C) artificial cerebral spinal fluid (aCSF) containing (in mmol l⁻¹): 20 NaHCO₃, 118 NaCl, 1.2 MgCl₂·6H₂O, 1.9 KCl, 1.2 KH₂PO₄, 10 Hepes sodium salt, 10 d-glucose, 2.4 CaCl₂; pH 7.6, osmolarity 287–290 mOsm. Both telencephalic hemispheres were sliced from the whole brain and submerged in an aCSF-filled Petri dish chilled to −18°C. A Leica VT1200S vibratome was used to axially slice hemispheres into 300 µm thick sheets, which were subsequently stored in chilled aCSF for a maximum of 36 h.

A Loligo Systems Witrox 1 O₂ meter (Viborg, Denmark) with a dipping O₂ probe was used to measure the O₂ partial pressure (P_O2) in the recording chamber (Fig. 1C). The instrument was first calibrated as per the manufacturer's instructions to 722 mmHg by placing the electrode in a glass reservoir bottle bubbled with 99% O₂/1% CO₂ and to 0 mmHg in a glass reservoir bottle bubbled with 99% N₂/1% CO₂.

To assess whether [ROS] inherently decrease with anoxia, slices were incubated with the ROS-sensitive dye CM-H2DCFDA (chloromethyl-2′,7′-dichlorofluorescein diacetate, referred to as CM-DCF hereafter; Invitrogen, Burlington, ON, Canada). In an opaque vial, slices were bathed in 5 ml of aCSF (previously bubbled with 99% O₂/1% CO₂) with CM-DCF (10 µmol l⁻¹) at room temperature for 1 h, followed by a 30 min wash in aCSF. Dye-loaded tissue was placed in a 1.2 ml Warner RC-26 open bath recording chamber with a P1 platform (Warner Instruments, Hamden, CT, USA) connected to two 1 litre reservoir bottles containing aCSF and gravity perfused from a height of 30 cm. Brain slices were held in place with a Warner 26GH/10 slice anchor (Harvard Apparatus CDN, St Laurent, QC, Canada). To ensure anoxic conditions, a cap was placed over the water immersion objective lens and gassed with 99%N₂/1%CO₂ during anoxic treatments.

CM-DCF was excited at 490 nm, and emission was detected at 520 nm via a DeltaRamX high-speed, random access monochromator and an LPS-220B light source (Photon Technology Inc., London, ON, Canada). The shutter opened for 0.85 s for each 10 s interval under 40× magnification using an Olympus BX51W1 microscope (Olympus Canada Inc., Richmond Hill, ON, Canada) and Rolera-MGi Digital EMCCD camera (Q Imaging, Burnaby, BC, Canada) to ensure maximal signal and minimal photobleaching effects.

Baseline fluorescence readings were maintained under normoxic conditions from t=0 min to t=5 min. At t=5 min, an anoxic solution was bulk perfused for 10 min and at t=15 min, normoxic solutions were reintroduced. H₂O₂ was added at the end of some experiments to confirm the responsiveness of the dye. For control experiments, ROS scavengers were bulk perfused (from a 1 litre reservoir bottle) at t=5 min in normoxic saline and were washed out at t=15 min. For each experimental recording, regions of interest (ROI) were drawn around the centre of a minimum of 8 neurons, which were then selected for analysis. Neurons that exhibited low to moderate fluorescence were used for analysis. Fluorescence experiments were limited to under 20 min, because of issues pertaining to drift and photobleaching.

Data are shown as percentage change relative to the baseline regression at the end of the normoxic period. The percentage change was calculated by using the median point of each treatment divided by the extrapolated baseline regression. Percentage change in CM-DCF fluorescence of all neurons in all tissue slices from one animal was averaged and used as one n-value for statistical analysis.

Slices were anchored in the RC-26 recording chamber as above. To achieve normoxic conditions, aCSF was gassed with 99% O₂/1% CO₂ in 1 litre bottles. Attached to the bottle was an intravenous (i.v.) dripper, which continuously gravity-perfused aCSF through the chamber during all normoxia recordings. A second 1 litre bottle with an i.v. dripper containing aCSF was gassed with 99% N₂/1%CO₂ to achieve anoxic conditions. A 3-way valve was used to switch between bottles during recordings when necessary. Both bottles contained black Viton tubing (Cole-Parmer, Montreal, QC, Canada) to prevent oxygen diffusion into the saline. Experiments were conducted at room temperature (22°C).

Fire-polished borosilicate glass capillary tubes were pulled using a P-97 micropipette puller (Sutter Instruments, Novato, CA, USA) to create 5–9 MΩ micropipette electrodes. Electrodes were filled with a solution composed of (in mmol l⁻¹): 110 potassium d-gluconate, 20 KCl, 1 MgCl₂, 0.3 GTP, 2 NaATP, 10 Hepes sodium salt, 0.0001 CaCl₂, 8 NaCl, 1 EGTA; pH 7.4, osmolarity 297 mOsm. Ag/AgCl wires were used for both the recording and reference electrodes.

A detailed protocol for blind whole-cell patch clamping is described elsewhere (Blanton et al., 1989; Wilkie et al., 2008). Briefly, the whole-cell patch configuration was established by slowly advancing the electrode towards the tissue slice with a manual Marzhauser MM33 micromanipulator (Wetzlar, Hausse, Germany). The final approach was made remotely with a stepper motor (Newport 860A stepper motor, Irvine, CA, USA.) while maintaining a slight positive pressure and applying a −2 mV square wave pulse to the electrode. Once the −2 mV square wave was compressed by approaching a cell, a 1–10 GΩ seal was formed by applying light suction while the holding potential was set to −70 mV, before applying a brief, rapid suction to break open the patch. Once the whole-cell configuration was established, patches with series resistance values between 120 and 200 mΩ were kept, and patches were discarded if series resistance changed more than 20% over the course of the recording. Data were filtered at 5 kHz using an Axopatch 1D amplifier, a CV-4 head stage and a Digidata 1200 A/D interface; Clampex 10 software was used for controlling the amplifier and analysing the data (Molecular Devices, Sunnyvale, CA, USA).

Neurons were characterized as either stellate or pyramidal based on electrophysiological properties. Prior to experimental recordings, a current step protocol was used to inject current to elicit AP firing. A 150 pA, 450 ms current injection resulted in a 30 mV depolarization of membrane potential (V_m) and initiation of AP firing. Pyramidal neurons exhibited accommodation, with decreasing AP amplitude over time, and AP cessation. Stellate neurons produced a train of APs that did not change in amplitude over time and did not exhibit accommodation. This classification of neurons was based on amplitude, frequency and refractory periods of APs as described elsewhere (Hossein-Javaheri et al., 2017). Experiments were carried out only in quiescent pyramidal cells and a subset of spontaneously active pyramidal cells.

Measurements of V_m and action potential firing frequency (AP_ff) were made following a 3–5 min patch stabilization period, then neurons were passively recorded for 25 min. In some experiments, the period from t=1 min to t=10 min was a normoxic baseline period to ensure stable patches and for comparison with subsequent treatment conditions. Patches were discarded if V_m changed more than 4 mV. At t=5 min, N-acetyl cysteine (NAC), (2-(2,2,6,6-tetramethylpiperidin-1-oxyl-4-ylamino)-2-oxoethyl)triphenylphosphonium chloride (MitoTEMPO) or H₂O₂ was perfused onto the slice using a secondary drug perfusion system (VC-6 perfusion valve controller, Warner Instruments) for 10 min, and at t=15 min, normoxic conditions were re-established and neurons were permitted to recover. This drug-perfusion system consisted of six different saline reservoirs attached to six electrically operated pinch valves that led into a common manifold with a single output. The output was a length of PE 10 tubing and its terminus was placed 1–2 mm upstream of where the recording electrode entered the tissue. A subset of pyramidal neurons that spontaneously fire was also examined. For neurons, AP_ff was also recorded as the number of APs per minute before, during and after treatment to investigate whether ROS contribute to the spike arrest phenomena documented in anoxic telencephalic neurons. For whole-cell conductance (G_w) recordings, neurons were clamped at −100 mV and stepped in 10 mV increments for 250 ms each to −40 mV. The amount of current that traversed the membrane was measured at each step, and G_w was determined from the slope conductance of the resulting current–voltage (I–V) relationship (Ghai and Buck, 1999). Using the same time line as in the passive recording protocol, action potential threshold (AP_th) was measured in current-clamped neurons by ramping voltage in 2 mV increments for 250 ms until an AP was generated. AP_th was the threshold value recorded for the first AP elicited.

Measurement of V_m, G_w, AP_th and AP_ff was similar to that described above. To ensure consistency with previous findings, bulk-perfusion with anoxic saline began at t=5 min; at t=25 min, normoxic saline was reintroduced. V_m, AP_th, G_w and AP_ff were measured during each of these treatments.

As a positive control, tissue slices were exposed to H₂O₂ under anoxia, slices were both bulk and drug perfused with normoxic saline from t=0 min to t=5 min, and then both bulk and drug perfusion were switched to anoxic saline. At t=20, H₂O₂ bubbled with 99% N₂/1% CO₂ was drug perfused until t=25 min; bulk anoxic perfusion was maintained. At t=25 min, drug perfusion of H₂O₂ was switched back to anoxic drug perfusion without altering the bulk perfusate. At t=35 min, both normoxic bulk perfusate and drug perfusate were reintroduced.

Pharmacological agents were both bulk and drug perfused into the chamber; a difference between the two methods was not observed. Slices were perfused with either the general ROS scavenger NAC (125 µmol l⁻¹) or the mitochondrial ROS scavenger MitoTEMPO (30 µmol l⁻¹), to decrease endogenous ROS under normoxic conditions. H₂O₂ (25 or 750 µmol l⁻¹) was exogenously added to increase ROS concentration. Drug concentrations were determined using a dose–response curve (data not shown). CM-DCF was purchased from Invitrogen Life Technologies (Burlington, ON, Canada). All other chemicals were obtained from Sigma-Aldrich Inc. (Burlington, ON, Canada).

Fluorescence and electrophysiological data were analysed using SigmaPlot v.11.0 (Systat Software, Inc., San Jose, CA, USA). A one-way ANOVA (Tukey test) was used to compare the means of normoxic controls and treatment groups for all fluorescence recordings. For electrophysiological recordings, a two-way ANOVA was used to compare the means of normoxic controls and treatment groups, followed by either a one-way ANOVA (Tukey test) or a paired t-test as post hoc tests. AP_ff was stable with a single neuron but varied between neurons; thus, data were normalized. Treatment within treatment groups was analysed using a one-way repeated measures (RM) ANOVA followed by a Tukey post hoc test. All results are expressed as means and s.e.m.

To determine how rapidly the recording chamber became severely hypoxic and how much O₂ remained, P_O2 measurements were made using an O₂ probe placed in the recording chamber. When the recording chamber was perfused with saline bubbled with 99% O₂/1% CO₂, a P_O2 of 717 mmHg O₂ was measured. In our open chamber, there is some loss of O₂ to the environment. The closed vessel calibration with 99% O₂/1% CO₂ bubbled saline could achieve a P_O2 of approximately 722 mmHg. When subsequently switched to saline bubbled with 99% N₂/1% CO₂, a P_O2 of 4.3 mmHg was recorded after 10 min. Anoxia was likely never reached (0 mmHg); therefore, our anoxic treatments are considered to be severely hypoxic (Fig. 1C).

To ensure adequate dye uptake, an excitation scan (excitation 450–550 nm) was conducted prior to each experiment. Emission values peaked around 500 nm, in agreement with the dye specifications. In dye-free tissue, there were no changes in fluorescence during a scan. Auto-fluorescence from a variety of cellular constituents can interfere with the fluorescence signal. Both reduced and oxidized forms of flavin adenine dinucleotide (FADH₂/FAD excitation/emission: 410 and 440 nm/510–530 nm) and nicotinamide adenine dinucleotide (NADH/NAD⁺ excitation/emission: 340±30 nm/460±50 nm) auto-fluoresce. Because the electron transport chain arrests under anoxia, the accumulation of reduced NADH and FADH₂ in the cell may artificially augment the signal. To investigate this possibility, fluorescence experiments were run in non-dye-loaded tissue under anoxic conditions to measure any native fluorescence. There were no detectable changes in fluorescence in the absence of the dye. However, background fluorescence in non-dye-loaded tissues did account for ∼300 a.f.u. (arbitrary fluorescence units). Dye-loaded tissue had fluorescence >600 a.f.u., and those cells that emitted fluorescence >900 a.f.u. were selected for analysis. The background fluorescence was subtracted from the raw fluorescence readings obtained from the cells.

Normoxic baseline recordings steadily increased over 20 min but the change was not significant (−0.35±0.18%, n=4; Fig. 2Ai). To assess the sensitivity of the dye to ROS, H₂O₂ (25 µmol l⁻¹) was added exogenously after each experiment and resulted in an immediate large increase in CM-DCF fluorescence (Fig. 2Aii). H₂O₂ can saturate the response and can be converted into reactive oxygen singlets, contributing to higher photobleaching; therefore, the addition of H₂O₂ was restricted to the end of the recovery period. The lowest dependable concentration of H₂O₂ that could elicit a signal was 25 µmol l⁻¹, is similar to previously reported values (Hogg et al., 2015). This value was used for subsequent experiments.

In western painted turtles, [ROS] decreases by −8.0±1.2% under anoxia (Hogg et al., 2015). To investigate whether this change is seen in goldfish neurons, we bulk perfused anoxic saline and measured CM-DCF fluorescence. Fluorescence significantly decreased compared with that in normoxia (−10.7±0.8%, n=6, P<0.001; Fig. 2iii).

Addition of the general scavenger NAC significantly decreased fluorescence (−10.96±0.72%, n=5, P<0.001; Fig. 2iv). To determine whether mitochondria were the source of the ROS, we used the mitochondria-specific scavenger MitoTEMPO. Fluorescence significantly decreased in relation to that in normoxia (−10.85±0.30%, n=4, P<0.001; Fig. 2v) and this was not significantly different from the NAC results. The decrease in fluorescence by either scavenger was also not significantly different from that in anoxia.

About 90% of the neurons in the goldfish telencephalon are pyramidal neurons and about 10% are stellate neurons; therefore, given this disproportional abundance of pyramidal neurons, we chose to only study pyramidal neurons. To electrophysiologically identify neuron type before an experiment, a neuron identification protocol was run (see Materials and Methods). In short, after a brief current injection, traces with accommodation of AP firing were categorized as pyramidal neurons, whereas traces with sustained AP firing were identified as stellate neurons (Fig. 3A,B). Based on the firing pattern, only pyramidal neurons were selected. There were spontaneously active neurons exhibiting a firing pattern similar to that of pyramidal neurons, except that AP_ff increased with the same current injection (Fig. 3C). These neurons were selected to study the effects of pharmacological ROS modulators on AP_ff.

V_m, G_w, AP_ff and AP_th did not significantly change after 30 min of normoxic perfusion (V_m=−72.33±3.35 mV to −72.11±3.62 mV, n=5, raw trace in Fig. 4i,ii; G_w=3.95±0.36 nS to 4.16 nS±0.30 nS, n=4; AP_th=−35.80±1.19 mV to −35.5±1.32 mV, n=3; and AP_ff=0.21±0.04 Hz to 0.22±0.01 Hz, n=3, raw traces/data not shown).

We (Hossein-Javaheri et al., 2017) previously found that with the onset of anoxic perfusion, V_m depolarized from −72.18±2.30 mV to −57.70±1.60 mV, G_w increased from 1.75±0.34 nS to 2.83±0.17 nS and AP_th remained unaltered. To confirm these findings and establish a baseline for comparison following the pharmacological manipulations, we replicated these experiments. In agreement with previous findings, V_m significantly depolarized from normoxia to anoxia (−73.57±1.91 mV to −61.43±2.30 mV, n=7, P<0.001; Fig. 4Aiii,B), G_w significantly increased (5.69±0.63 nS to 8.02±0.46 nS, n=4, P<0.05; Fig. 5) and AP_th remained unchanged (−37.80±0.93 mV to −37.90±0.90 mV, n=10; Fig. 6). In a subset of spontaneously firing neurons, anoxic treatment significantly decreased AP_ff (0.177±0.06 Hz to 0.033±0.01 Hz, n=5, P<0.001; Fig. 7C). The effects were reversed after 20 min of normoxic perfusion. We defined spontaneously firing neurons as those that fired APs in the range of 0.2 to 0.3 Hz in the absence of any external stimulus and as compared with many neurons that did not fire APs unless depolarizing current was injected.

To test the hypothesis that anoxia-induced decreases in [ROS]_i depolarize V_m, NAC was administered during normoxia. Following 10 min of NAC perfusion, V_m significantly depolarized from normoxic values (−76.63±1.23 mV to −66.23±1.21 mV, n=7, P<0.001; Fig. 4Aiv,B). To source the ROS decrease, MitoTEMPO was perfused into the chamber for 10 min under normoxia. Compared with that in normoxia, V_m depolarized significantly (−71.50±1.08 mV to −62.50±1.82 mV, n=6, P<0.001; Fig. 4Av,B). V_m values in the presence of ROS scavengers were not significantly different from each other from those in anoxia (Fig. 4B). Washout of both NAC and MitoTEMPO failed to reverse the drug effects and led to hyperactivity or neuronal death in most neurons; the data were not included in any analysis.

In order to evaluate whether increases in [ROS] played a role in the anoxic response, the predominant mitochondrial ROS product, H₂O₂, was perfused through the chamber. After 10 min of 25 µmol l⁻¹ hydrogen peroxide bath perfusion, V_m did not significantly change (−74.33±2.10 mV to −72.63±2.10 mV, n=5; Fig. 4Avi,B). At a higher concentration of 750 µmol l⁻¹, H₂O₂ did not significantly affect V_m after 10 min (−73.17±1.58 mV to −72.00±1.10 mV, n=5; Fig. 4Avii,B). V_m values in the two H₂O₂ conditions were not significantly different from each other but were significantly different from those in anoxia (anoxia versus 25 µmol l⁻¹ H₂O₂, P<0.001; anoxia versus 750 µmol l⁻¹ H₂O₂, P<0.001; Fig. 4B). Recovery remained stable for neurons treated with 25 µmol l⁻¹ H₂O₂, but after <5 min of recovery, some neurons treated with 750 µmol l⁻¹ H₂O₂ depolarized and fired excessively.

During anoxia, G_w increased from 5.69±0.63 nS to 8.02±0.46 nS. To determine whether decreased [ROS]_i influenced this, NAC and MitoTEMPO were separately applied under normoxic conditions. NAC caused G_w to significantly increase (5.97±0.51 nS to 7.54±0.28 nS, n=7, P<0.05; Fig. 5). Treatment of patches with MitoTEMPO caused G_w to significantly increase (6.02±0.49 nS to 7.58±0.55 nS, n=3, P<0.05; Fig. 5). Effects of the two scavengers were not significantly different from each other from those of anoxia (Fig. 5).

To test whether ROS addition increases G_w similar to anoxia, H₂O₂ was applied to tissue slices. No significant changes in G_w were measured after 10 min with 25 µmol l⁻¹ H₂O₂ (5.49±0.84 nS to 4.81±0.30 nS, n=4) or 750 µmol l⁻¹ H₂O₂ (5.04±0.31 nS to 5.12±0.57 nS, n=4; Fig. 5). However, the anoxia response was significantly different from that of the two H₂O₂ responses (P<0.01 for 25 μmol l⁻¹ H₂O₂ and 750 μmol l⁻¹ H₂O₂; Fig. 5).

During anoxia, AP_th remained unchanged, unlike in turtle cortical neurons (Hogg et al., 2015). To test whether low levels of ROS affect AP_th, NAC or MitoTEMPO was added exogenously to patches. AP_th was unchanged following application of NAC (−38.14±0.94 mV to −38.29±0.47 mV, n=7; Fig. 6Aiii,B) or MitoTEMPO (−37.71±1.04 mV to −37.86±0.90 mV, n=7; Fig. 6Aiv,B). Addition of either 25 or 750 µmol l⁻¹ H₂O₂ did not elicit a change in AP_th (−39.08±1.19 mV to −39.42±1.39 mV, n=6; and −40.33±1.06 mV to −40.00±1.30 mV, n=6, respectively; Fig. 6Av,vi,B). No statistical difference was detected with any of the treatments compared to anoxia (Fig. 6B).

Under anoxia, AP_ff decreased from 0.177±0.06 Hz to 0.033±0.01 Hz in goldfish. To determine whether decreases in ROS mimic spike and synaptic arrest, spontaneously active pyramidal neurons were treated with NAC for 10 min. AP_ff decreased significantly during treatment compared with that in normoxia (0.267±0.05 Hz to 0.063±0.010 Hz, n=4, P<0.001; Fig. 7Aii,Bii,C). To determine whether mitochondrially-derived ROS are the source for the decrease, neurons were treated with MitoTEMPO for 10 min. Again, AP_ff decreased significantly (0.347±1.58 Hz to 0.083±0.0004 Hz; n=7; P<0.001, Fig. 7Aiii,Biii,C). AP_ff values of patches treated with either scavenger were not significantly different from each other from those in anoxia (Fig. 7C).

Drug perfusion of H₂O₂ was used to determine whether ROS addition affects the anoxic response. Addition of 25 or 750 µmol l⁻¹ H₂O₂ to spontaneously firing pyramidal neurons did not significantly affect AP_ff under normoxic conditions (0.113±0.055 Hz to 0.104±0.047 Hz, n=4; and 0.10±0.044 Hz to 0.11±0.049 Hz, n=3, respectively; Fig. 7Aiv,v,Biv,v,C). AP_ff of patches treated with either oxidant was significantly different from AP_ff during anoxia (P<0.001 for both; Fig. 7).

In order to test whether the anoxia-mediated depolarization of V_m could be reversed by ROS reintroduction, 25 µmol l⁻¹ H₂O₂ was added to pyramidal neurons during anoxic exposure. Anoxia caused a significant depolarization in V_m (−76.75±2.70 mV to −69.9±3.20 mV, n=3, P<0.001; Fig. 8A,B), consistent with previous findings (Fig. 8A). Perfusion of 25 µmol l⁻¹ H₂O₂ repolarized the membrane close to normoxic V_m values (−74.58±2.70 mV, n=3; Fig. 8A,B), which were significantly different from anoxic values (P<0.01). Cessation of H₂O₂ perfusion significantly depolarized V_m back to anoxic values (−70.83±2.90 mV, n=3, P<0.001 compared with normoxia, not significant compared with anoxia; Fig. 8Ai,Bi). Switching the perfusate to normoxic saline repolarized V_m close to baseline values (V_m=−75.46±1.40, n=2). The use of 750 µmol l⁻¹ H₂O₂ did not impact the anoxia response (Fig. 8Aii,Bii).

We show a decrease in [ROS] during the anoxic transition, which is comparable to the decrease following either NAC or MitoTEMPO administration. This result indicates that ROS concentrations decrease proportionally with O₂ tension and that the main source of ROS production is mitochondria, in agreement with other literature (see review by Turrens, 2003).

During the transition from normoxia to anoxia, [ROS] steadily decreases. Interestingly, studies in anoxia-intolerant organisms show a burst of mitochondrial ROS production under hypoxia (Hamanaka and Chandel, 2010; Hernansanz-Agustín et al., 2014). The reason for the burst is that, under hypoxia, electrons leak to oxygen, forming O₂⁻ at complex III, and increasing [ROS] in the intermembrane space. Genetic targeting of the Rieske iron–sulfur protein in complex III abolished the hypoxic ROS increase in rats, indicative of the role of mitochondrial complexes in ROS generation in hypoxic mammalian tissue (Hamanaka and Chandel, 2010). However, this idea is often contested. In isolated mitochondria, ROS concentration was unaffected by changes in O₂ from 250 µmol l⁻¹ to 5–7 µmol l⁻¹, and no increase in ROS concentration was observed at any level of O₂ tension, concluding that if there is a hypoxic increase, it is not sourced from the mitochondria (Starkov, 2008).

The differences in [ROS] during hypoxia compared with that in anoxia may be due to the usage of two divergent pathways for combating hypoxic and anoxic stress. In mammals, long-term hypoxia is stressful and the change in redox state, postulated to orchestrate a hypoxic defence, initiates several defence cascades including the activation and binding of the transcription factor HIF-1a to DNA (Duranteau et al., 1998; Huang et al., 1996; Salceda and Caro, 1997). HIF-1a is one of the primary hypoxia signalling proteins utilized to return the body back to homeostasis at the cellular and organismal level (see review by Wenger, 2000). Hypoxia-tolerant fish, such as the longjaw mudsucker, also upregulate several genes that are known HIF-1 targets under hypoxia (Chandel et al., 2000; Gracey et al., 2001). At O₂ tensions of ∼0 mmHg, organisms do not actively upregulate HIF-1 and do not experience an increase in [ROS], irrespective of tolerance (Schroedl et al., 2002). HIF-1-induced changes are counterintuitive to long-term anaerobiosis and, thus, HIF-1-mediated responses are inhibited in anoxic painted turtles (Storey, 2007). The idea of [O₂]-dependent differential pathway activation is supported in nematodes. Nematodes can survive under hypoxia and enter a depressed metabolic state during anoxia. hif-1 knockouts show that HIF-1 is essential to hypoxic but not anoxic survival (Jiang et al., 2001; Padilla et al., 2002). Further, anoxic stabilization of HIF-1a does not require mitochondrial ROS, possibly explaining why there is no increase in [ROS].

Painted turtles do not show a burst in ROS under hypoxia or anoxia (Pamenter et al., 2007). Perhaps the lack of a ROS burst is because the transition to anoxia is within the homeostatic range for anoxia-tolerant species. In these turtles, heat shock protein concentrations remain stable for 12 h of anoxia, indicative that a short-term decrease in O₂ tension may not cause the same level of molecular or cellular stress as in other organisms (Ramaglia and Buck, 2004). However, the response of the red-eared slider turtle and crucian carp appears to be different in that both heat shock protein and RNA increase within hours of anoxic exposure (Kesaraju et al., 2009; Prentice et al., 2004; Stecyk et al. 2012; Stenslokken et al., 2010). In fact, hypoxia-intolerant organisms, such as rat or rainbow trout, that are not naturally exposed to fluctuating ambient O₂ tension, show increases in hsp70 gene expression during periods of low O₂, indicative of cellular stress (Airaksinen and Råbergh, 1998). In these cases, ROS bursts are likely utilized to orchestrate a hypoxic defence cascade.

In our study, severely hypoxic goldfish pyramidal neurons showed a depolarized V_m, an increased G_w, a decreased AP_ff and AP_th maintenance, which corroborates a previous study from our lab (Hossein-Javaheri et al., 2017). The depolarization in goldfish pyramidal neurons is attributed to the anoxia-mediated release of GABA (Hossein-Javaheri et al., 2017). Extracellular [GABA] doubles during anoxia in the crucian carp telencephalon (Hylland and Nilsson, 1999); and although this is modest compared with that in the painted turtle (∼80-fold increase; Nilsson and Lutz, 1991), GABA plays an important role in spike or synaptic arrest in fish. Also, the expression of GABA transporter proteins decreases by 80% during anoxia, which likely increases the synaptic [GABA] (Ellefsen et al., 2008) and increases substrate availability for post-synaptic receptor binding. In goldfish telencephalon, GABA is inhibitory and decreases the likelihood of excitatory activity within pyramidal neurons (Martyniuk et al., 2005). Because of the anoxia-mediated surge of GABA in the synapse, the amount of excitatory input required to depolarize V_m to AP_th increases, resulting in a lower AP_ff. Goldfish telencephalon, like painted turtle cortical neurons, experiences shunting inhibition as an inhibitory control. This shunting can be reversed by blocking GABA_A receptors in both species (Hossein-Javaheri et al., 2017; Pamenter et al., 2011). The possible mechanisms underlying this response are discussed elsewhere (Hossein-Javaheri et al., 2017).

Elucidating the potential involvement of ROS in low-oxygen sensing and arrest of pyramidal neuron firing during severe hypoxia was the goal of the present study. From the fluorescence data, we showed that during anoxia, [ROS] decreased and then hypothesized that a decline in ROS production by mitochondria could orchestrate the anoxic response via spike or synaptic arrest. The result showed that V_m depolarizes with the addition of either scavenger (i.e. NAC or MitoTEMPO) during normoxia but does not elicit APs in quiescent pyramidal neurons. Next, G_w was assessed and the addition of either scavenger resulted in an increase in G_w. Thus, these changes are linked to spike or synaptic arrest.

We previously demonstrated that an increase in G_w results in the change of V_m to the GABA_A current equilibrium potential in goldfish (Hossein-Javaheri et al., 2017). Addition of NAC and its congeners in human epithelial cells increased conductance to Cl⁻, whereas addition of MitoTEMPO in rat smooth muscle cells increased whole-cell voltage-gated K⁺ currents, suggesting that both NAC and MitoTEMPO can change the electrical properties of cells, specifically ion conductance (Köttgen et al., 1996). In the anoxia-tolerant painted turtle, pyramidal cortical neuron G_w increased when supplemented with MPG or MitoTEMPO, similar to our results in goldfish pyramidal neurons reported here (Hogg et al., 2015).

Furthermore, a subset of goldfish pyramidal neurons were spontaneously active and fired at regular intervals just as was found in turtle pyramidal neuronal, which demonstrated a greater than 70% decrease in AP_ff with the onset of anoxic perfusion or addition of either ROS scavenger (Pamenter et al., 2011; Buck and Pamenter, 2018; Hogg et al., 2015). Goldfish pyramidal neurons also displayed decreased AP_ff with the addition of either ROS. The decrease in AP firing could be attributed to several mechanisms, including: a decrease in glutamate release, an increase in GABA release, changes in voltage-gated channel properties, flux in ion concentrations and changes in neurotransmitter receptor sensitivity. All of the proteins involved have ROS modulation sites (Reczek and Chandel, 2015). Of these, GABAergic transmission is of particular interest, as it is the primary constituent required for spike/synaptic arrest, and its release is thought to be redox sensitive in turtle dorsal cortex (Hogg et al., 2015). Application of GABA to turtle pyramidal neurons during normoxia decreases AP_ff while blocking GABA_A receptors during anoxia increases AP_ff, indicating that the GABA_A receptor is an important part of the anoxia tolerance mechanism. These results were mirrored by application of ROS scavengers and oxidants (H₂O₂), indicating that ROS may be another important part of the mechanism (Hogg et al., 2015). However, the effect on pyramidal neurons seems to result from increased stellate neuron activity and GABA release that causes V_m depolarization in the pyramidal neuron, which appears to be mediated by a decrease in stellate neuron mitochondrial [ROS] (Hawrysh and Buck, 2019).

The relationship between GABA and mitochondrial ROS has been previously established in other organisms. GABA release is inhibited by increased ROS generation while ROS scavenging restores GABA_A receptor-mediated transmission in mouse substantia gelatinosa of the spinal dorsal horn (Yowtak et al., 2011). In the painted turtle model, all three types of GABA receptor currents (synaptic, peri-synaptic and tonic) were amplified with the addition of MitoTEMPO (Hogg et al., 2015). This amplification of GABA currents occurs during anoxia, indicative of GABAergic redox modulation (Hogg et al., 2015). Based on the turtle model, a likely assumption would be that decreases in mitochondrial ROS are also required to induce GABAergic spike/synaptic arrest in goldfish pyramidal neurons.

Interestingly, AP_th in goldfish pyramidal neurons did not change during treatment with ROS scavengers or during anoxia; however, anoxia depolarized AP_th of pyramidal cells in the turtle model (Pamenter et al., 2011). An increase in AP_th is beneficial because the amount of excitatory input needed to elicit an AP increases, resulting in a reduction in AP generation, and thus less ATP expenditure – a strategy utilized by the western painted turtle (Pamenter et al., 2011). However, in goldfish, the benefit of maintaining AP_th may outweigh the benefit of a depolarized AP_th. During anoxia, goldfish remain active in the water column, with only a ∼70% decrease in whole-body metabolic rate whereas turtles reduce their whole-body metabolic rate by >90%, remaining quiescent until O₂ availability increases (Johansson and Nilsson, 1995; van Waversveld et al., 1989; Jackson, 2000). The telencephalon in fish is analogous to the hippocampus and amygdala, which are responsible for functions such as fear and spatial memory (Broglio et al., 2003; Portavella et al., 2004). These functions are integral to fish survival, as they aid in predator avoidance and navigation. Although a reduction in many energy-consuming pathways is vital for surviving anoxic periods, perhaps it is beneficial to be able to excite neurons in this brain region if necessary, as goldfish remain motile during anoxia.

Addition of H₂O₂ did not alter V_m or G_w, corroborating the previously published results on the turtle model (Hogg et al., 2015). In rat hippocampal CA1 neurons, addition of 330 µmol l⁻¹ H₂O₂ did not alter V_m, whereas 1 mmol l⁻¹ H₂O₂ caused inconsistent hyperpolarizing V_m shifts. Only addition of a higher concentration (3.3 mmol l⁻¹ H₂O₂) resulted in a repeatable hyperpolarization of V_m and increased a K⁺ conductance (Seutin et al., 1995). This increase in conductance, however, may be species specific as H₂O₂ administered to the same area in guinea pig brain caused no change in K⁺ conductance (Pellmar, 1987). Concentrations used for our experiments were below 1 mmol l⁻¹ and had no measurable effects on any of the parameters measured during normoxia. Additionally, goldfish have a remarkable antioxidant system. Compared with other organs, the brain has constitutively higher levels of superoxide dismutase (SOD) and glutathione S-transferase activity to prevent oxyradical overproduction (Lushchak et al., 2001). Because H₂O₂ plays a concentration-dependent dual role, the 750 µmol l⁻¹ dose we used may have caused neuronal oxidative stress, resulting in the measured changes in neurophysiological parameters being a product of cellular defences attempting to re-establish homeostasis (Halliwell et al., 2000). Concentrations greater than 50 µmol l⁻¹ H₂O₂ produce cytotoxic effects in mammalian tissues, although LD₅₀ values are dependent on a host of factors, including length of H₂O₂ exposure, physiological state and cell type (Halliwell et al., 2000).

The addition of H₂O₂ was postulated to reverse the anoxic response, similar to what was observed in painted turtle neurons (Hogg et al., 2015). Administration of a low concentration (25 µmol l⁻¹) of H₂O₂ repolarized V_m similar to pre-anoxic values. An explanation for this could be the redox state of the neuron. Under normal oxygen conditions, the cell is predominantly oxidized. Under anoxia, the redox balance shifts as a result of the absence of oxygen and ROS. Addition of a surplus of H₂O₂ may have indiscriminately oxidized cellular constituents and inhibited the anoxic GABA response. The redox-modulatory role of oxidants on GABA is documented within several species and cell types. In lobster muscle, H₂O₂ addition resulted in a decrease of presynaptic GABA release at the neuromuscular junction, whereas superoxide decreases GABA_A receptor activity in rat cortical synaptosomes (Colton et al., 1986; Schwartz et al., 1980). In turtles, GABA-mediated changes in V_m and PSCs, which were initiated by low mitochondrial [ROS] and by anoxia, were reversed with the addition of exogenous H₂O₂, suggesting a redox-dependent GABAergic system. This ability to reverse the anoxic response further provides evidence that the transition to anoxia is redox modulated.

Although many cellular constituents and organelles can detect changes in O₂ tension, the positioning of the mitochondria as the central hub for oxygen sensing is a plausible scenario (Wenger, 2000). Being the largest consumers of oxygen in the cell, mitochondria are unable to function without sufficient oxygen. This reliance on oxygen, in addition to the signalling role of ROS, renders mitochondria a potential primary O₂ sensor. In tissue specialized to sense minute changes in O₂ tension, such as pulmonary arterial smooth muscle cells and glomus cells, inhibition of the electron transfer system (ETC) affects O₂-sensing ability, illustrating the role of mitochondria in O₂-sensitive cells (Ganfornina and Lopez-Barneo, 1991; Wyatt and Buckler, 2004). Mitochondrial oxygen-sensing mechanisms that rely on redox state and mitochondrial ROS formation are an inexpensive and natural method of initiating signal transduction pathways.

During aerobic respiration, a by-product of electron movement through the ETC is the partial reduction of O₂ metabolites (Adam-Vizi and Starkov, 2010). Under anoxia, mitochondrial ROS generation is arrested as a result of the lack of oxygen molecules available to accept electrons; the ETC becomes reduced and the ratio of reduced glutathione (GSH) to oxidized glutathione disulfide (GSSG) increases, resulting in a reduced intercellular redox state. The cessation of ROS production and the change in redox state may serve as a cytoprotective mechanism in this low-energy state. The shift in redox state during anoxia is likely a signal as several ion channels, pumps and co-transporters have redox sites that are sensitive to low P_O2, including proteins associated with spike/synaptic arrest (Hogg et al., 2015; Pitlik et al., 2009). This positions ROS as a unique, intercellular signalling molecule.

Unexpectedly, the NAC concentration curve showed a bimodal effect. Increasing [NAC] above 250 µmol l⁻¹ counterintuitively depressed the response. Low molecular weight thiols, such as NAC, can act as both anti-oxidants and pro-oxidants (Munday, 1989; Nath and Salahudeen, 1993). Increasing NAC concentration would thereby augment H₂O₂ production. As both NAC and H₂O₂ affect V_m, this may have an additive effect, and decrease the net change in V_m (Hogg et al., 2015).

Although the targets of ROS have not been explored in this study, the hallmark of a target is the possession of an amendable cysteine and methionine residue. Transmembrane ion channels have these groups and are targets for ROS, resulting in altered conducting and gating properties (Ramírez et al., 2016). ROS targets are innumerable, with some of the primary cellular protein targets being ion co-transporters, pumps, exchangers and channels (Kourie, 1998). Additionally, it is likely that multiple targets exist within the same system. In guinea pig ventricular myocytes, addition of H₂O₂ inhibited L-type Ca²⁺ currents but enhanced the delayed-rectifying K⁺ currents, alluding to multiple targets within a cell type (Satoh and Matsui, 1997). ROS are generally thought to be superimposed signals that intersect various cellular activities through timing, tuning and recruiting of a cellular signalling pathway in response to the metabolic state of the cell (D'Autréaux and Toledano, 2007).

As with most signalling molecules, there is likely an interplay between different cellular signals and ROS, mediating its effects. Like ROS, tight regulation of calcium is imperative for cellular homeostasis. Calcium and ROS have a mutual interplay that is required for signal fine tuning (Brookes et al., 2004; Görlach et al., 2015). Pathways that require calcium, such as long-term potentiation and consolidation of memory, are enhanced in rats when 1 µmol l⁻¹ H₂O₂ is applied, and blocked after subsequent application of L-type voltage-dependent calcium channel inhibitors (Kamsler and Segal, 2003). In aortic human endothelial cells, 10 µmol l⁻¹ H₂O₂ enhanced inositol triphosphate-mediated calcium release, but caused no change in [Ca²⁺]_i in the absence of inositol triphosphate (Hu et al., 2000). The relationship between ROS and calcium is bidirectional, as calcium decreases ROS generation from complex I and III under normoxic conditions (Hoffman and Brookes, 2009). In anoxia-tolerant turtles, there is a modest rise of [Ca²⁺]_i, sourced from the mitochondria during anoxia, that is responsible for interfering with an otherwise ROS-induced NMDA current amplification (Hawrysh and Buck, 2013). Preliminary results in goldfish tissue show a modest increase in [Ca²⁺]_i and depolarization of mitochondrial V_m under anoxia, alluding to the possibility that the goldfish and turtle models possess a similar biochemical response under anoxia. The intracellular ROS–Ca²⁺ interplay is a topic that should be investigated as it may be an evolutionarily conserved anoxia tolerance mechanism.

ROS are evolutionarily conserved as a signalling molecule from unicellular to complex multicellular organisms (Marinho et al., 2014) and play an important role in areas such as host defence and oxidative biosynthetic reactions. At lower levels, ROS function as a signalling molecule, maintained at a relatively similar concentration across taxa (Marinho et al., 2014). Given the affiliation of ROS with oxygen, this multifaceted molecule has a probable role as an oxygen sensor. In anoxia-intolerant organisms, the decrease in oxygen results in a detrimental molecular cascade, eventually leading to cell death (Abele et al., 1990; Choi, 1992; Rossi et al., 2000). However, anoxia-tolerant organisms have evolved a mechanism by utilizing the innate decrease in ROS to facilitate a neuroprotective cascade during environmental anoxia. To our knowledge, this is the first study to investigate the role of ROS in prompting the anoxic response in fish and supports the idea of a homeostatic redox state that can be altered with varying [ROS].