Ambient CO₂, fish behaviour and altered GABAergic neurotransmission: exploring the mechanism of CO₂-altered behaviour by taking a hypercapnia dweller down to low CO₂ levels

As a result of anthropogenic CO₂ release, current atmospheric CO₂ levels have risen to approximately 0.04 kPa (∼400 µatm) – 40% higher than levels in pre-industrial times – and may reach 0.10 kPa (∼1000 µatm) by the end of the 21st century (IPCC, 2013). The world's water bodies absorb large amounts of this CO₂ and so considerable research efforts are being directed towards understanding how this elevated P_CO2 (hypercapnia) might affect aquatic organisms, with particular focus on those thought to have evolved in relatively stable CO₂ and pH conditions.

Although not immediately lethal (Doney et al., 2009; Brauner and Baker, 2009), the projected hypercapnic scenarios have been linked to striking effects on the behaviour of some marine fish species in ways that are likely to impair fitness and population survival. These effects include reversed olfactory and auditory preferences (Munday et al., 2009; Dixson et al., 2010; Simpson et al., 2011), loss of behavioural lateralization (Domenici et al., 2012; Jutfelt et al., 2013), loss of learning (Ferrari et al., 2012; Jutfelt et al., 2013), increased boldness and activity (Munday et al., 2010), reduced temporal resolution of vision (Chung et al., 2014) and increased anxiety (Hamilton et al., 2014). The physiological basis for these behavioural changes is currently thought to stem from the changes in extracellular ion concentrations that occur as a result of acid–base regulation interfering with neurotransmitter function (Nilsson et al., 2012). Typically, fishes alleviate respiratory acidosis during hypercapnia by increasing plasma HCO₃⁻ levels in exchange for Cl⁻ (Ishimatsu et al., 2008; Brauner and Baker, 2009) and the associated redistribution of ions across cell membranes seems to interfere with GABA_A receptor function. The GABA-gated ion channels conduct Cl⁻ and HCO₃⁻ fluxes (Bormann et al., 1987) when activated by the inhibitory neurotransmitter GABA, and normally reduce neural excitability by hyperpolarizing the cell through an inflow of Cl⁻. However, in hypercapnia, when new gradients of HCO₃⁻ and/or Cl⁻ are established across neural membranes, GABA_A receptor activation could result in a net outflow of these anions, depolarizing the membrane (Nilsson et al., 2012; Heuer and Grosell, 2014). Evidence for this mechanism stems from the observation that gabazine, a specific blocker of the GABA_A receptor, effectively reverses the disruptive effects of high levels of CO₂ on lateralization, olfactory preferences, learning and vision in fishes (Nilsson et al., 2012; Chivers et al., 2014; Chung et al., 2014, Lai et al., 2015). Similarly, muscimol, a GABA_A receptor agonist, causes increased anxiety in Californian rockfish exposed to high CO₂ concentrations (Hamilton et al., 2014). The widely conserved function of GABA_A receptors suggests that a wide range of aquatic animals, including invertebrates, may be susceptible to similar behavioural abnormalities (e.g. Watson et al., 2014).

Tropical freshwater systems with frequent hypoxia and organic loading can become extremely hypercapnic, with P_CO2 levels reaching as high as 8.7 kPa (Willmer, 1934; Ultsch, 1987; Furch and Junk, 1997) and inhabitants clearly adapt to these hypercapnic conditions. The Mekong River, with its high organic loading, is one such environment. Mean P_CO2 across the entire river is 0.11 kPa (Li et al., 2013), 2.7-fold higher than atmospheric levels. This value varies both spatially and temporally, from 0.023 kPa in the river's upper reaches in the flood season, to 0.61 kPa in its lower reaches in the dry season (Li et al., 2013). Overall, most regions of the Mekong are hypercapnic for most of the year (Li et al., 2013), yet the river still harbours enormous biodiversity (Valbo-Jørgensen et al., 2009), including ∼850 fish species (Hortle, 2009; Gephart et al., 2010). These fishes are probably thriving in their hypercapnic environment, so they are unlikely to display the maladaptive behavioural abnormalities observed in the hypercapnia-exposed marine fish. Moreover, in Vietnam, some fishes like the facultative air-breathing striped catfish Pangasianodon (formerly Pangasius) hypophthalmus (Sauvage 1878) are extensively cultured in ponds with no aeration and low water exchange and hence are hypoxic (Lefevre et al., 2011, 2014; Damsgaard et al., 2015), with P_CO2 of up to 4.5 kPa (Damsgaard et al., 2015). Despite the harsh conditions, P. hypophthalmus – and the entire aquaculture industry surrounding them – continue to grow at a remarkable rate (Phan et al., 2009; De Silva and Phuong, 2011).

Given the neural and behavioural abnormalities induced by mild hypercapnia (∼0.1 kPa) in some marine fish and recently in a temperate freshwater fish (Ou et al., 2015), it is relevant to investigate whether variation in CO₂ levels exerts similar effects in hypercapnia-tolerant fish such as P. hypophthalmus. We hypothesized that a fish species native to a regularly hypercapnic environment (such as P. hypophthalmus) would have GABA_A receptors that are functional at atypically high blood [HCO₃⁻] and low [Cl⁻]. The mechanisms that may enable this [e.g. regulation of brain intracellular pH (pH_i)] could alter ion concentrations in such a way that might even leave P. hypophthalmus vulnerable to reductions in ambient P_CO2. If this is the case, GABA_A-mediated behavioural abnormalities would be predicted when P. hypophthalmus is exposed to relatively low-CO₂ environments.

We addressed this by scoring various ecologically relevant behavioural traits of P. hypophthalmus at two P_CO2 levels. The high-CO₂ group (3.1 kPa) can be regarded as the control group as the fish are likely to have been raised under such conditions (Damsgaard et al., 2015), whereas the low-CO₂ group was acclimated to 0.04 kPa P_CO2 for 2 weeks. We used established behavioural tests: an empty tank trial to measure routine activity (Munday et al., 2010), a novel object trial to assess responsive behaviour (Jutfelt et al., 2013), a predator trial to mimic predator avoidance (Munday et al., 2010; Dixson et al., 2010) and a conspecifics trial to assess social behaviour (Munday et al., 2009). The experiments were performed before and after gabazine treatment (Nilsson et al., 2012) to determine whether GABA_A receptors are involved in any behavioural abnormalities. Our hypothesis predicted behavioural abnormalities in the normocapnia-exposed P. hypophthalmus consistent with those shown in normocapnia-acclimated fishes when exposed to relatively high-CO₂ environments: increased boldness and activity (Munday et al., 2010), an attraction to predators (Munday et al., 2010; Dixson et al., 2010) and a lack of preference for time spent near unrelated conspecifics (Munday et al., 2009). Furthermore, gabazine was predicted to attenuate these behavioural abnormalities. In addition to the behaviour trials, we measured blood pH, blood P_CO2, blood [HCO₃⁻] and brain pH_i of the fish to understand how these were affected by the different CO₂ exposures and to allow us to calculate theoretical effects of a range of CO₂ levels on neural membrane potential using the Goldman equation.

Pangasianodon hypophthalmus (∼5 months old; 13.9±5.11 g (mean±s.d.), 9.1–25.8 g; sex unknown because of juvenile life stage) were obtained from an aquaculture pond near Can Tho, Vietnam. Seven days prior to the start of the experiments, animals were transferred to 5000 litre fibreglass holding tanks (∼100 fish per tank) that were maintained at either high CO₂ (3.1 kPa P_CO2; pH 5.8±0.2) or low CO₂ (0.04 kPa P_CO2; pH 7.2±0.1) at normoxic levels. The aquarium facility was semi enclosed, allowing for natural light conditions (∼11.5 h light:12.5 h dark in December). Aquatic P_CO2 was continuously measured using a G10ps CO₂ probe and regulated by monitoring pH with a K01svpld probe, both connected to an Oxyguard Pacific System (Oxyguard International A/S, Farum, Denmark) that added CO₂ through a solenoid valve if pH increased above a threshold corresponding to the desired P_CO2 of 3.1 kPa for the high-CO₂ treatment. P_CO2 for the low-CO₂ treatment was achieved by recirculating water through a CO₂-diffusing tower where water was fully equilibrated with the air (Oxyguard International), yielding a P_CO2 of 0.04 kPa (below the limit of detection of the G10ps CO₂ probe; ∼0.1 kPa P_CO2). Previous studies have shown that behavioural effects of changes in ambient CO₂ manifest within 4 days of exposure and that longer exposure periods do not further alter behavioural responses (Munday et al., 2010), suggesting that our CO₂ acclimation period was sufficient. All animals were exposed to the same ambient temperatures of 27–29°C for the entire acclimation and experimental period of 2 weeks. Fish were fed ad libitum with commercially available catfish pellets.

The behaviour trials used a total of 48 fish comprising six treatment groups: low-CO₂ (N=12), low-CO₂ with 4 mg l⁻¹ gabazine treatment (N=6), low-CO₂ with 8 mg l⁻¹ gabazine treatment (N=6), high-CO₂ (N=12), high-CO₂ with 4 mg l⁻¹ gabazine treatment (N=6) and high-CO₂ with 8 mg l⁻¹ gabazine treatment (N=6). The behaviour trials were conducted in a series of three identical rectangular grey fibreglass arenas (66 cm×232 cm×50 cm, 18 cm water depth) filled with water from the acclimation tanks so that fish were acclimated and tested at the same P_CO2. Four trials measuring ecologically relevant behaviours were performed on each fish in the same order: an empty tank trial (to measure routine activity), a novel object trial (to measure boldness), a predator trial (to measure predator avoidance) and a conspecific trial (to measure sociality). Each arena had a grid of squares (26 cm×26 cm) marked on the bottom, and activity was measured as the number of line crossings over time. Behaviour was recorded visually by two observers that were out of the focal fish's sight and the observers' roles were consistent across all trials and all experimental fish.

Individual fish were selected randomly from the acclimation tanks and within ∼30 s were gently transferred to the first experimental arena in a small plastic container so that the fish were never removed from water. Fish were transferred between arenas in the same way and were always released in the centre of the tank. First, routine activity (measured as line crossings) was measured for 5 min in an empty tank. After the 5 min period, a novel object (clay brick) was gently placed on the lengthwise midline of the arena, two squares (52 cm) from the experimental fish. Activity and proximity to the novel object (time spent within one grid line (26 cm) of the object) were recorded for 5 min.

Immediately following the novel object test, the fish was carefully transferred to the centre of a second arena for the predator sensitivity trial. In this arena, 31 cm of each end was fenced off with plastic mesh and a large snakehead (Channa striata; 37 cm, 740 g), a known predator of P. hypophthalmus, was placed behind the mesh in a randomly determined end of the arena. The mesh was permeable to the chemical cues of the C. striata and P. hypophthalmus, allowing both fish to confirm the other's presence visually and chemically. The arena was divided lengthwise into thirds to allow an observer to record the proportion of time the focal P. hypophthalmus spent near the predator, in the middle, or far away from the predator. Activity and location were recorded for 5 min. Only the zone closest to the predator (two grid squares, 52 cm) was used for the analysis.

Finally, the experimental fish was transferred into a third arena for the sociality test. This arena was set up in the same manner as that used for the predator-avoidance trials, except a group of five P. hypophthalmus were present behind one of the mesh screens instead of a predator (random side). Activity and location were again recorded for 5 min. As with the predator trial, only the zone closest to the conspecifics (two grid squares, 52 cm) was used in the analysis.

Gabazine targets GABA_A receptors and reduces their anion conductivity through specific binding and allosteric interaction (Ueno et al., 1997). Gabazine is known to be free of the non-GABAergic calcium-dependent potassium channel effects of bicuculline, an older and more widely used GABA_A antagonist (Heaulme et al., 1986; Hamann et al., 1988). Because gabazine is the most specific GABA_A antagonist available, it comes with the lowest likelihood of off-target effects.

Gabazine treatment followed the procedure described by Nilsson et al. (2012), placing the fish in a bucket with 3 litres of water containing gabazine (4 or 8 mg l⁻¹; Sigma-Aldrich, St Louis, MO, USA) for 30 min. The two doses of gabazine were used to ensure it remained effectual in the latter 10 min of the behaviour trials. Although data for all six trials are presented in Fig. 1, the effect of gabazine appeared to diminish over the 20 min experiment and so the subsequent figures and analyses used the 4 mg l⁻¹ treatment group to represent the gabazine-treated fish in the initial 10 min (empty tank and novel object trials) and the 8 mg l⁻¹ treatment group to represent the gabazine-treated fish in the final 10 min (predator and conspecific trials) of the behaviour trials.

Fish were euthanized immediately following the final behaviour trial with a swift blow to the head, whereupon ∼0.3 ml of blood was sampled via caudal puncture using a heparinized syringe. The brain was then excised and immediately frozen in liquid nitrogen. Blood pH, P_CO2 and HCO₃⁻ were measured using an iStat Systems portable clinical analyser. All measurements were temperature corrected to 27°C according to Harter et al. (2014). These values were further corrected according to Damsgaard et al. (2015).

Brain intracellular pH (pH_i) was measured using the methods of Pörtner et al. (1990) and Baker et al. (2009), where tissue was ground to a fine powder under liquid nitrogen using a mortar and pestle. Approximately 0.1 g of the powder was then transferred to a 1.5 ml centrifuge tube containing 0.8 ml of metabolic inhibitor cocktail comprising 150 mmol l⁻¹ potassium fluoride (KF) and 6 mmol l⁻¹ nitrilotriacetic sodium (Na₂NTA). The solution was immediately vortexed for 30 s and centrifuged at 3000 g for 45 s and the resulting supernatant represented the intracellular medium (cytosol) of the tissue sample. pH_i measurements were made on 0.2 ml volumes of this supernatant (in duplicate) using a Radiometer PHM 84 (Copenhagen, Denmark) pH meter connected to a radiometer SaS gK2401C (Radiometer Analytical, Lyons, France) pH electrode.

Activity data (line crossings) were analysed using two-way repeated-measures ANOVAs with treatment (gabazine and CO₂) and trials as the factors. One analysis was performed for the first two behaviour trials (empty tank and novel object) in which gabazine-treated fish received a 4 mg l⁻¹ dose. A second analysis was performed for the latter behaviour trials (predator and conspecific exposures) in which gabazine-treated fish received an 8 mg l⁻¹ dose. Two-way repeated-measures ANOVA was also used to analyse the proportion of time fish spent near the C. striata predator and group of conspecifics in the final two trials. All other data were analysed using one-way ANOVAs, and post hoc Holm–Šidák tests were used to test for differences between treatment groups. Non-proportional data were square root transformed when necessary to better meet the assumptions of normal distribution and equal variance. Proportional data were logit transformed when necessary (Warton and Hui, 2011). Despite the relatively small and uneven sample sizes, statistical power was generally high (>0.7) with the exception of the comparisons of time spent adjacent to predator (0.10) and blood pH (0.24). SigmaPlot 11 (Systat Software, San Jose, CA, USA) was used for all analyses (critical α=0.05). Throughout the text, values are given as means±s.e.m.

Generally, the low-CO₂ fish were more active than the high-CO₂ fish across all four behaviour trials (Fig. 1). The two-way repeated-measures ANOVA for the empty tank and novel object trials revealed a significant overall difference between high- and low-CO₂ fish untreated with gabazine (treatment P<0.05, both trial and trial by treatment interaction P>0.05; Fig. 2A,B). The same trend held for the two-way repeated-measures ANOVA run on the predator and conspecifics trials, with a significant overall difference between high- and low-CO₂ fish and a significant difference in activity between trials regardless of treatment, but no significant trial by treatment interaction (Fig. 2C,D).

Exposure to 4 mg l⁻¹ gabazine reduced activity levels of the low-CO₂ fish by approximately 50% in the first two behaviour trials (empty tank and novel object) to levels not statistically different than those of the high-CO₂ fish (P>0.05; Fig. 2A,B). During the subsequent 10 min, the effect of this low gabazine dose had apparently worn off and activity no longer differed from untreated low-CO₂ fish (t-test P=0.61 and P=0.91 for predator and conspecific trials, respectively; Fig. 1). As gabazine is rapidly taken up over the gills, it is also likely to depurate by this route and thus the low gabazine dose probably wore off after 10 min. This led us to test the effect of a double dose. When treated with the higher gabazine dose (8 mg l⁻¹), the low-CO₂ fish showed slightly higher activity during the first two behaviour trials (empty tank and novel object) relative to the low-CO₂ fish treated with 4 mg l⁻¹ gabazine, consistent with what has previously been shown for high doses of GABA receptor blockers (Turski et al., 1985). However, during the final two behaviour trials (predator and conspecifics), when gabazine levels in the 8 mg l⁻¹-treated fish were likely to have fallen closer to those of the 4 mg l⁻¹-treated fish during the first 10 min, activity of the low-CO₂ gabazine-treated fish was reduced by approximately 50% relative to the non-treated fish, to levels not statistically different than those of the high-CO₂ fish (P>0.05; Fig. 2C,D). Because the effect of gabazine appeared to diminish over the 20 min experiment, we have represented gabazine-treated fish using the 4 mg l⁻¹ dosage when averaging behavioural traits during the initial 10 min (empty tank and novel object trials) and the 8 mg l⁻¹ dosage when averaging behavioural traits during the final 10 min (predator and conspecifics trials). Neither of the gabazine treatments had any significant effect on the activity levels of the high-CO₂ fish (all P>0.05; Figs 1 and 2).

The high-CO₂ fish spent an average of 3.3±1.4% of their time in close proximity (within 26 cm) to the clay brick in the novel object trial, whereas the low-CO₂ fish spent significantly more time (21.0±2.8%) in close proximity to the novel object (ANOVA, P<0.05; Fig. 3). When treated with gabazine, the proportion of time spent by the low-CO₂ fish near the brick was reduced to a value similar to the high-CO₂ fish (12.8±4.1%; P>0.05; Fig. 3). Gabazine had no effect on the time the high-CO₂ fish spent close to the brick (3.3±1.5%; P>0.05; Fig. 3).

Overall, P. hypophthalmus spent very little time (4–23%) near the predatory C. striata (within two grid squares, or 52 cm) and this aversion was unaffected by CO₂ alone (P>0.05; Fig. 4A). However, fish acclimated to high CO₂ levels spent significantly less time near the predator (within two grid squares, or 52 cm) than low-CO₂ fish treated with gabazine (P<0.05; Fig. 4A). The same trend was observed in the sociality test (Fig. 4B).

The high-CO₂ fish had higher P_CO2 and [HCO₃⁻] in the blood than the low-CO₂ fish (P<0.05; Fig. 5A,B). Blood pH values of the high- and low-CO₂ fish were not statistically different (pH 7.50±0.05 and 7.48±0.03, respectively), whereas treatment with gabazine slightly reduced these values in both the high-CO₂ (pH 7.44±0.06) and low-CO₂ fish (pH 7.34±0.04; Fig. 6A). A similar trend was observed with brain pH_i: values of the high- and low-CO₂ fish were not statistically different (pH 6.81±0.02 and 6.79±0.04, respectively), whereas treatment with gabazine slightly reduced these values in the high-CO₂ fish (pH 6.76±0.03; P>0.05) and significantly reduced them in the low-CO₂ fish (pH 6.64±0.02; P<0.05; Fig. 6B).

We hypothesized that a fish species native to a regularly hypercapnic environment would have GABA_A receptors that remain functional at atypically high blood [HCO₃⁻] and low [Cl⁻], and thus would display GABA_A-mediated behavioural abnormalities when exposed to a relatively low-CO₂ environment. Mechanistically, these effects should be similar to those previously seen in coral reef fish that are adapted to a low-CO₂ habitat and are unable to respond appropriately to chronic hypercapnia. In both instances, neural ion gradients could change in ways that render GABA_A receptors depolarizing (excitatory) rather than hyperpolarizing (inhibitory). The underlying mechanisms could involve an inability of reef fish to regulate brain pH_i, resulting in maintained Cl⁻ levels intracellularly while these fall extracellularly (leading to a depolarizing gradient). By contrast, in the hypercapnia-adapted P. hypophthalmus, well-developed brain pH_i regulation involving atypical Cl⁻:HCO₃⁻ exchange ratios in the blood (Damsgaard et al., 2015) could lead to depolarizing GABA_A receptors; these scenarios are discussed in depth below. Consistent with some of our predictions, we observed similar behavioural alterations in normocapnia-acclimated P. hypophthalmus as found previously in normocapnia-acclimated reef fishes when exposed to relatively high-CO₂ environments: increased boldness and activity (Munday et al., 2010), and more time spent in the presence of a predator (Munday et al., 2010; Dixson et al., 2010). Furthermore, the GABA_A receptor antagonist gabazine attenuated these abnormalities in behaviour.

As predicted, the low-CO₂ P. hypophthalmus displayed significantly higher activity levels than the high-CO₂ control fish (Figs 1 and 2) and the elevated activity levels of the low-CO₂ fish were attenuated by gabazine (Figs 1 and 2). This implies an involvement of GABA_A receptors and an alteration of the neural Cl⁻ and HCO₃⁻ ion gradients gated by these receptors. Likewise, the low-CO₂ fish spent six times longer in close proximity to a novel object placed in their tank – a measure of boldness – than the high-CO₂ fish (Fig. 3). An alternative interpretation of these results might be that the low-CO₂ fish were not bold with respect to the object's presence, but rather indifferent, whereas the high-CO₂ fish were timid. This interpretation is supported by the fact that ‘close proximity’ was defined as a radius of 26 cm (one grid square) around the object, an area one quarter the size of the tank's area. If the fish was indifferent to the presence of the object, one might predict that it would be found in the quarter surrounding the object 25% of the time; this agrees well with the value for the low-CO₂ fish of 21.0±2.8%. In any case, when the low-CO₂ fish were treated with gabazine, this period of time was reduced by 40% to a level not statistically different from that of high-CO₂ fish (Fig. 3), again implying the involvement of GABA_A receptors.

Contrary to our expectations, low-CO₂ fish were not attracted to the predatory C. striata. Indeed, the low- and high-CO₂ fish spent statistically similar amounts of time near the C. striata (Fig. 4A). These trends were unaffected by gabazine treatment and are therefore contrary to previous results that have demonstrated a clear role of GABA_A receptors in the olfactory sensing of predators (e.g. Nilsson et al., 2012). These inconsistencies may be caused by the use of different species, different CO₂ exposures (e.g. high-to-low CO₂ in our study versus low-to-high CO₂ in other studies) or different methodologies. In our study, the fish could use at least three cues (vision, olfaction, hearing) to detect the predator, whereas previous studies (e.g. Munday et al., 2010; Nilsson et al., 2012) tested predator avoidance based only on chemical cues. It may be that any sensory impairment caused by the altered CO₂ environment is compensated for by the use of more than one sensory modality (Rauschecker and Kniepert, 1994; Chapman et al., 2010).

P. hypophthalmus regulates brain pH_i when faced with drastic changes in environmental CO₂ levels (Fig. 6B). This is not surprising given that extracellular pH (pH_e) was also regulated over the duration of CO₂ exposure (Damsgaard et al., 2015) and is consistent with other CO₂-tolerant fishes (Brauner and Baker, 2009) exposed to short-term hypercapnia [white sturgeon (Baker et al., 2009); Amazonian catfish (Brauner et al., 2004)] and long-term hypercapnia [gilthead seabream (Michaelidis et al., 2007)].

What benefits come with brain pH_i regulation, and how might GABA_A-mediated behaviour be affected by it? Defending brain pH_i maintains the routine excitability of brain cells that may otherwise be increased or decreased depending on the magnitude of hypercapnia (Siesjö et al., 1972). The primary mechanism of active pH_i regulation is believed to involve the equimolar exchange of intracellular Cl⁻ for extracelluar HCO₃⁻ (Brauner and Baker, 2009), where intracellular [Cl⁻] and [HCO₃⁻] decrease and increase, respectively. In turn, this increases extracellular [Cl⁻], and since GABA_A channels are approximately five times more conductive to Cl⁻ than to HCO₃⁻, at least in mammals (Farrant and Kaila, 2007), the net result is an increased likelihood of a neuroinhibiting hyperpolarization of the brain cell upon GABA_A receptor activation. Therefore, the relatively low activity levels observed in our high-CO₂ fish may be partially explained by putative increases in extracellular [Cl⁻] resulting from the active regulation of brain pH_i, which itself also helps to preserve routine neural excitability.

The observation that a moderate dose of gabazine reduced the activity levels of the low-CO₂ fish relative to those of the high-CO₂ fish suggests that GABA_A receptors are involved in this behavioural response. GABA_A channels are conductive to Cl⁻ and HCO₃⁻, and environmental CO₂ levels influence these ion concentrations in both the blood and the cytosol (Ishimatsu et al., 2008; Brauner and Baker, 2009; Damsgaard et al., 2015). This is related to acid–base regulatory mechanisms, and during this regulation in P. hypophthalmus exposed to 3 kPa CO₂, [Cl⁻] decreases by 13 mmol l⁻¹ with an increase in [HCO₃⁻] of 20 mmol l⁻¹ (Damsgaard et al., 2015). We did not measure [Cl⁻] in either compartment of our fish, but our measurements of blood [HCO₃⁻] during hypercapnia are in line with this (Fig. 5B). Combining our P_CO2 measurements (Fig. 5A) with our brain pH_i measurements (Fig. 6B), Henderson–Hasselbach calculations estimate intracellular [HCO₃⁻] of the brain to increase from approximately 1.53±0.23 mmol l⁻¹ at 0.04 kPa P_CO2 to 4.32±0.77 mmol l⁻¹ at 3.1 kPa P_CO2. And as P. hypophthalmus brain cells appear to actively regulate pH_i under high CO₂ (Fig. 6B), the increase in intracellular [HCO₃⁻] is probably matched by an equimolar decrease in intracellular [Cl⁻]. It is possible that GABA_A receptors of hypercapnia-adapted fishes have evolved to function under shifted neural anionic set points and that deviating from these set points could result in a reversal of GABA_A receptor function.

where R is the ideal gas constant (8.315 J K⁻¹ mol⁻¹), T is temperature (300.15 K), F is Faraday's constant (96,485 Coulombs mol⁻¹) and P is the relative HCO₃⁻/Cl⁻ permeability of the GABA_A channel. This relative permeability varies from 0.2 to 0.6 in mammals (Farrant and Kaila, 2007) and values of 0.3 and 0.4 have previously been applied to fish (e.g. Heuer and Grosell, 2014). Blood [HCO₃⁻] ([HCO₃⁻]_o) in our high- and low-CO₂ fish (Fig. 5B), and the corresponding P_CO2 values (Fig. 5A) and brain pH_i (Fig. 6B), yields an intracellular [HCO₃⁻] ([HCO₃⁻]_i) of 4.32 and 1.53 mmol l⁻¹ for the high and low-CO₂ fish, respectively. Blood [Cl⁻] ([Cl⁻]_o) of the low-CO₂ fish was assumed to be 110 mmol l⁻¹ (Damsgaard et al., 2015) and we considered two scenarios for the high-CO₂ fish: one with the typical equimolar (1:1) exchange of blood Cl⁻ for HCO₃⁻ (e.g. Brauner and Baker, 2009) and another with a 1:2 exchange of blood Cl⁻ for HCO₃⁻, as observed by Damsgaard et al. (2015) in P. hypophthalmus. Such a ratio might be related to hypercapnia tolerance or could even be unique to P. hypophthalmus. Intracellular [Cl⁻] ([Cl⁻]_i) in the low-CO₂ fish was estimated to be 8.0 mmol l⁻¹ based on Delpire and Staley (2014) and 5.2 mmol l⁻¹ in the high-CO₂ fish based on an equimolar exchange of intracellular Cl⁻ for HCO₃⁻ resulting from active pH_i regulation of the brain. Finally, we assumed the commonly used resting neural membrane potential of −70 mV (Moyes and Schulte, 2006). All values are summarized in Table 1.

We plotted a linear trend line between the high- and low-CO₂ values for each ion concentration and used its equation to determine ion concentrations under 20 different P_CO2 conditions, ranging from 0 to 4 kPa. These ion concentrations were then used together in the Goldman equation to calculate E_GABAA reversal potentials as a function of P_CO2 at each of the 20 P_CO2 conditions. We did this for P_{HCO3⁻/Cl⁻} ratios of 0.3 and 0.4, and for hypercapnic blood Cl⁻:HCO₃⁻ exchange ratios of 1:1 and 1:2 (Fig. 7).

The results of the Goldman calculations (Fig. 7) agree with both our hypothesis and our behavioural observations. The neural anion gradients of high-CO₂ fish result in a hyperpolarizing GABA_A reversal potential when assuming either 0.3 or 0.4 P_{HCO3⁻/Cl⁻} permeability ratios, while the neural anion gradients of low-CO₂ fish result in a depolarizing GABA_A reversal potential assuming either 0.3 or 0.4 P_{HCO3⁻/Cl⁻} permeability ratios. The P_CO2 at which the membrane potential changes from hyperpolarizing to depolarizing varies with both P_{HCO3⁻/Cl⁻} permeability ratio and blood Cl⁻:HCO₃⁻ exchange ratios (Fig. 7). A hyperpolarizing membrane potential is retained over a larger P_CO2 range when the P_{HCO3⁻/Cl⁻} permeability ratio of GABA_A is lower and when the exchange ratio of blood Cl⁻:HCO₃⁻ is at 1:2, with the latter ratio seemingly playing a greater role. Therefore, a possible effect of the unusual 1:2 anion exchange ratio in P. hypophthalmus may be an avoidance of hyperactivity in the high CO₂ levels that are typical of its natural environment. In any case, our calculations of E_GABAA reversal potentials agree with our behavioural findings, where high-CO₂ fish exhibit inhibitory membrane potentials and relatively low activity, and low-CO₂ fish exhibit excitatory membrane potentials and relatively high activity. By reducing the activity of GABA_A channels, gabazine effectively counteracts this GABA_A-mediated excitation and subsequently reduces the activity levels of the low-CO₂ fish.

For comparison, we can attempt to estimate the E_GABAA reversal potential for a ‘typical’ hypercapnia-intolerant reef fish when exposed to climate change-relevant high-CO₂ (0.1 kPa). Here, we assume that the reef fish does not display P. hypophthalmus’ 1:2 exchange of blood Cl⁻ for HCO₃⁻, which, when combined with the fish’s seawater environment, influences intracellular and extracellular ion concentrations (detailed in Table 1) in a way that yields depolarizing (excitatory) E_GABAA reversal potentials (−68.9 or −67.8 mV for 0.3 and 0.4 P_{HCO3⁻/Cl⁻} ratios, respectively) when used in the Goldman equation. In contrast, a typical reef fish under normocapnic CO₂ conditions (0.04 kPa) produces hyperpolarizing (inhibitory) E_GABAA reversal potentials (−71.8 and −71.4 mV for 0.3 and 0.4 P_{HCO3⁻/Cl⁻} ratios, respectively; Table 1; [HCO₃⁻]_o from Esbaugh et al., 2012; [HCO₃⁻]_i from Henderson–Hasselbach calculations; [Cl⁻]_o from Marshall and Grosell, 2006; [Cl⁻]_i from Delpire and Staley, 2014). These calculations not only support the results of previous studies on reef fishes that have shown excitatory behaviour under high-CO₂ conditions but also suggest that some of P. hypophthalmus’ mechanisms of hypercapnia tolerance (i.e. brain pH_i regulation, 1:2 exchange ratio of blood Cl⁻ for HCO₃⁻) help preserve the neuroinhibiting effect of GABA under hypercapnia.

An alternative explanation for our activity results involves the anaesthetic effect of CO₂ (Fish, 1943; Bell, 1964). That is, the high activity of the low-CO₂ fish may only be ‘high’ relative to the activity of the high-CO₂ group, whose behaviour may have been influenced by the anaesthetic effect of CO₂. Although it is possible for CO₂ to have an anaesthetic effect on fishes at a P_CO2 as low as 2.4 kPa [e.g. rainbow trout (Iwama et al., 1989)], hypercapnia-tolerant species tend to require a much higher P_CO2 [e.g. 0.89–4.44 kPa pCO₂ for common carp (Yoshikawa et al., 1988, 1991)]. P_CO2 levels higher than 3.1 kPa would probably be required to elicit an anaesthetic effect in P. hypophthalmus. Furthermore, other aspects of water chemistry, such as P_O2, have been shown to strongly influence anaesthetic effect of CO₂ on fish (Bernier and Randall, 1998). For example, an anaesthetizing P_CO2 in one study (Iwama et al., 1989) may have no anaesthetic effect on the same species (rainbow trout) when it is not coupled with hyperoxia (Bernier and Randall, 1998). We did not hyperoxygenate our water, which may have mitigated any anaesthetic effect the CO₂ may have otherwise had. Finally, our qualitative observations of the fish provided no evidence of anaesthesia. We therefore find it unlikely that our high-CO₂ fish were under CO₂-induced anaesthesia.

As a species native and presumably adapted to typically hypercapnic environments, P. hypophthalmus allowed for a novel test of the involvement of GABA_A channels in the behavioural response of fishes to high levels of environmental CO₂ (assuming the hatchery-derived individuals used in our study were representative of wild P. hypophthalmus). We hypothesized that P. hypophthalmus would have GABA receptors that are functional at atypically high blood [HCO₃⁻] and low [Cl⁻], and thus would be predicted to display GABA­_A-mediated behaviour abnormalities when exposed to a relatively low-CO₂ environment. The increased activity levels of normocapnia-exposed fish, together with the reversal of this hyperactivity when treated with the GABA_A receptor blocker gabazine, support this hypothesis. Further support is provided by our theoretical calculations of E_GABAA reversal potentials, suggesting the hyperpolarizing GABA_A reversal potentials of hypercapnic P. hypophthalmus are indeed reversed to depolarizing potentials when the fish are in normocapnic environments.

In an increasingly hypercapnic world, we can look to native species of today's hypercapnic environments for some insight into how other species might cope with increases in environmental CO₂ and acidity. Our results suggest that P. hypophthalmus retains functional GABA_A receptor activity in high-CO₂ environments and consequently avoids the CO₂-induced behavioural alterations observed in numerous normocapnia-native reef fishes (Munday et al., 2012). Assuming that P. hypophthalmus or its ancestors were at one point native to normocapnic environments, this would suggest that fish are indeed capable of adapting to increases in environmental CO₂ in a way that mitigates the adverse behaviours observed in recent studies (Welch et al., 2014; but see Munday et al., 2014). However, the ability of natural selection to adapt species to hypercapnia will of course depend on the presence of relevant gene variants in the gene pool and on the rate at which environmental CO₂ levels increase, and there is a considerable risk that the present rate is too high to allow adaptation through natural selection.

The authors wish to thank Dr Sigal Balshine and the two anonymous reviewers for their valuable input on the manuscript.