Rates of hypoxia induction alter mechanisms of O₂ uptake and the critical O₂ tension of goldfish

Environmental hypoxia is a common characteristic of many aquatic systems and is becoming increasingly prevalent, severe and long-lasting because of anthropogenic and climate change effects (Friedrich et al., 2014; IPCC, 2014; Smith et al., 2006). Many studies have examined the physiological impacts of hypoxia exposure on a diverse array of fish species, but these have focused almost exclusively on either the severity of the hypoxic exposure [i.e. partial pressure of O₂ in water (Pw_O2)] or its duration. However, a third dimension of hypoxic exposure, the rate of hypoxia induction (RHI), has received very little attention and is rarely even controlled for (or at least reported) when environmental hypoxia is experimentally induced (Rogers et al., 2016). This is unlike other abiotic variables such as temperature, which are typically altered at consistent rates across studies [e.g. 0.2–0.3°C min⁻¹ for the determination of critical thermal maxima (CT_max)] owing to the effects they have on organismal responses (e.g. temperature tolerance in fishes; Mora and Maya, 2006). Similarly, RHIs may influence the physiological responses of fishes to hypoxia, particularly, time-dependent responses related to environmental O₂ extraction.

Most fishes possess mechanisms that enhance O₂ extraction and delivery to tissues as Pw_O2 is reduced, such as increased haemoglobin (Hb) synthesis (Gracey et al., 2001) and concentration in the blood (Affonso et al., 2002), increased haematocrit (Lai et al., 2006; Turko et al., 2014), increased Hb–O₂ binding affinity (Turko et al., 2014), increased ventilation frequency and amplitude (Holeton and Randall, 1967; Itazawa and Takeda, 1978; Tzaneva et al., 2011; Vulesevic and Perry, 2006), and a redistribution of blood supply to critical tissues (Sundin et al., 1995). Some fishes, including goldfish and numerous other species, also have the ability to dramatically increase lamellar surface area in response to hypoxia exposure through apoptotic reductions to the inter-lamellar cell mass (ILCM; Anttila et al., 2015; Borowiec et al., 2015; Crispo and Chapman, 2010; Dhillon et al., 2013; Ong et al., 2007; Sollid et al., 2003, 2005; Turko et al., 2012). While these modifications to different parts of the O₂ transport cascade function to improve O₂ uptake at low Pw_O2, the time courses over which these modifications are enacted differ and may potentially impact the critical P_O2 (P_crit) of the rate of O₂ uptake (Ṁ_O2).

P_crit is defined as the Pw_O2 at which a fish's Ṁ_O2 transitions from being regulated at some stable level independent of Pw_O2 (i.e. oxyregulation) to being dependent upon Pw_O2 (i.e. oxyconformation). At P_crit, the fish's aerobic scope is theoretically zero and at Pw_O2 values below P_crit, the fish's ability to generate ATP aerobically is limited (Farrell and Richards, 2009). P_crit therefore reflects a fish's ability to acquire and use environmental O₂ as a function of Pw_O2, with a lower P_crit indicating a greater ability to extract O₂ to maintain aerobic metabolism in hypoxic environments. A low P_crit is beneficial because it allows the animal to maintain a routine level of function and activity in hypoxic environments while avoiding a reliance on anaerobic glycolysis and/or metabolic rate depression. Indeed, we have recently shown that goldfish prioritize their use of aerobic metabolism in hypoxic environments over their exceptional ability to induce metabolic rate depression, which they reserve for anoxic environments (Regan et al., 2017). Goldfish also appear to enhance their ability to extract environmental O₂ over relatively short time periods in hypoxia, which in theory should result in a lowering of their P_crit value (Regan et al., 2017). Because this ability is influenced by a suite of O₂ extraction mechanisms that are both plastic and time-dependent, we hypothesized that gradual RHIs would allow fish to induce plastic mechanisms that enhance O₂ extraction, resulting in lower P_crit values than those of fish exposed to rapid RHIs.

We tested this hypothesis by determining the P_crit values of goldfish exposed to progressive reductions in Pw_O2 (from normoxia to near-anoxia; referred to as P_crit trials) over different durations: ∼24 min to represent rapid RHI, ∼84 min to represent typical RHI (P_crit trials in the literature typically last 60 to 120 min; see Rogers et al., 2016) and ∼480 min to represent gradual RHI. We also ran parallel hypoxic exposures of different RHIs to investigate morphological and physiological traits of goldfish that might play causal roles in a time-related shift in P_crit, including (among other traits) gill morphometry and Hb–O₂ binding affinity. Furthermore, our use of different respirometric techniques allowed us to disentangle the effects of time and technique on the determination of P_crit, thereby addressing a longstanding concern over the use of closed-chamber respirometry and its associated metabolic end-product accumulation for the determination of P_crit (Keys, 1930; Rogers et al., 2016; Snyder et al., 2016; Steffensen, 1989). And finally, we chose goldfish as our study species because they have well-characterized responses to hypoxia exposure (Dhillon et al., 2013; Mitrovic et al., 2009), including a well-resolved P_crit as determined by closed-chamber respirometry (Dhillon et al., 2013; Fry and Hart, 1948; Fu et al., 2011; Regan et al., 2017), which could aid our analysis of how RHI might influence the underlying physiology of P_crit.

Goldfish (Carassius auratus auratus Linnaeus 1758; 2.87±0.14 g wet mass; N=84; sex unknown) were purchased from a commercial supplier (The Little Fish Company, Surrey, BC, Canada) and held under a 12 h:12 h light:dark cycle in 100-litre recirculating systems of well-aerated, dechlorinated, 17°C water (replaced weekly) at the University of British Columbia (UBC). Stocking density was <0.3 g l⁻¹. Fish were fed to satiation daily (Nutrafin Max Goldfish Flakes) except for 24 h before transfer to the experimental apparatus, when feeding ceased. UBC's Animal Care Committee approved all procedures (protocol A13-0309).

We exposed goldfish to P_crit trials of short (∼24 min), typical (∼84 min, representing a typical closed-chamber P_crit trial's duration) and long (∼480 min) durations to represent progressively reduced RHIs. These different RHIs were achieved using different respirometric techniques (details below), while the respirometer chambers, animal transfer protocol, habituation period and mean fish mass remained consistent across all trials. Each fish was used only once.

We used two 32 ml flow-through respirometer chambers made from stainless steel as described in Regan et al. (2013). For each trial, we inserted a fish into the chamber and held it under flow-through conditions for ≥16 h prior to commencing the P_crit trial. The fish chamber was supplied with flow-through water at a rate of 190 ml h⁻¹ and maintained at 17°C. Inflowing water was drawn from a well-mixed reservoir held at ∼26 kPa (manually controlled using compressed N₂ and O₂) and pumped to the respirometer chamber via a peristaltic pump (Gilson Minipuls 3, Middleton, WI, USA) through a combination of stainless steel tubing and gas-impermeable Tygon peristaltic tubing. The Pw_O2 of the inflowing water was maintained slightly hyperoxic to ensure that the outflowing water was always at or slightly above normoxic Pw_O2. Following the habituation period, we conducted our respirometry experiments.

where C_O2 is the O₂ content of the water converted to μmol l⁻¹ from Pw_O2 using the solubility factor of 14.485 μmol l⁻¹ kPa⁻¹ (Boutilier et al., 1984), T is the time period over which the change in C_O2 is calculated (5 or 2 min; see below), V is the fish chamber volume (32 ml) plus the volume of the closed-circuit water lines minus the volume displaced by the fish itself, and M is the mass of the fish. The trials were ended when Pw_O2 reached 0 kPa, at which point the short-circuit was dismantled and flow-through conditions were reestablished to return chamber Pw_O2 to habituation period conditions.

For the short duration P_crit trials (24±2 min), we again used closed-circuit respirometry as described for the typical P_crit trials. To shorten the trial and hasten the Pw_O2 decline, we made initial normoxic Ṁ_O2 readings and then manually replaced the entire water volume of the respirometry chamber and its water supply lines over ∼5 min with water equilibrated to 5.3 kPa Pw_O2 using a 60 ml syringe. Pw_O2 was therefore reduced from normoxia to ∼5.3 kPa not by the fish's Ṁ_O2, but by the active replacement of the water volume. At this point, we attached the water supply lines to the peristaltic tubing, turned the pump back on to 190 ml h⁻¹, and allowed the fish to deplete the closed system's O₂ through its own respiration (typically over a ∼20 min period). We chose 5.3 kPa as our replacement Pw_O2 for two reasons: first, it allowed for reliable Ṁ_O2 measurements starting at ∼4.8 kPa, which provided enough Ṁ_O2 data points above P_crit to construct robust P_crit traces; and second, the amount of time required for the fish to reduce Pw_O2 from 5.3 kPa to anoxia put the overall duration of these P_crit trials within our targeted duration of between 20 and 30 min. Although these procedures reduced the overall duration of the P_crit trial, we must point out that the RHI below 5.3 kPa was similar to that of the typical duration trials. If mechanisms of enhanced O₂ extraction are only induced at Pw_O2 <5.3 kPa, then these two techniques could result in similar P_crit values.

Prior to actively replacing the water volume, we converted the system to closed-circuit and made a series of normoxic Ṁ_O2 readings between 25 and 19 kPa to aid in our calculation of P_crit (see below). Upon reaching 19 kPa, we converted the system back to flow-through, reestablished a normoxic Pw_O2 of ∼21 kPa, and then commenced the active water volume replacement.

For the long duration P_crit trials, we used three different respirometric techniques to ensure the mean P_crit values were the result of P_crit trial duration and not respirometric technique per se. These trials varied in average duration from 434 to 562 min depending on the technique used. We chose a time duration of ∼480 min because it was significantly longer than the typical trial durations, but likely shorter than would be required to induce gene expression acclimation responses. It is also in line with some of the longer P_crit trial durations observed in the literature (see Rogers et al., 2016).

For our first technique, we used closed-circuit respirometry where we added a 250 ml water reservoir to reduce the rate at which the fish's respiration depleted the system's O₂. This reservoir was a glass bottle placed immediately after the peristaltic pump. Water leaving the respirometer chamber was pumped into the reservoir directly over a stir bar that mixed the water volume to prevent O₂ stratification in the bottle. Water flowed out of the reservoir through a stainless steel line that punctured the bottle's rubber stopper and went directly into the stainless steel line supplying the respirometer chamber. All materials used were gas-impermeable glass or stainless steel. Attaching this reservoir to the closed-circuit system took ∼2 min, after which the peristaltic pump was turned back on and the P_crit trial was run according the closed-circuit technique described for the typical duration P_crit trials. The average duration for these closed-circuit trials was 434±56 min.

where Ci_O2 and Co_O2 are O₂ content of inflowing and outflowing water, respectively, converted from Pw_O2 as described above (we used a single P_O2 optode for these measurements) and f is water flow rate (190 ml h⁻¹). We held fish at approximately 26, 16, 5.3, 2.7, 1.3, 0.7 and 0 kPa, each Pw_O2 in series, in that order and for 1 h, and at each Pw_O2 we measured Ṁ_O2 at 10, 30 and 60 min (10 min was the minimum time required to ensure Pw_O2 had equilibrated across the respirometer and the upstream and downstream Pw_O2 measurement chambers). Because the calculated Ṁ_O2 at each Pw_O2 was nearly identical at each of the three time points, we averaged across the time points and calculated P_crit from those averaged Ṁ_O2 values for each individual. The average duration for these flow-through trials was 562±19 min, including the time required to reach target Pw_O2 values.

Third, we used a variation on intermittent flow respirometry that combined flow-through and closed-circuit respirometry. We used flow-through conditions to manually reduce Pw_O2 from normoxia to ∼2.8 kPa over ∼430 min and then commenced a period of closed-circuit respirometry, which took an additional ∼15 min. We chose a target Pw_O2 of 2.8 kPa to start the closed-circuit portion of the trial based upon our earlier short-term P_crit trials (which used the same combined respirometric technique) that suggested we could reliably determine P_crit from this Pw_O2. Upon reaching 2.8 kPa, we converted to the closed-circuit setup and allowed the fish's respiration to deplete the remaining O₂ in the closed system as described previously. This combination of techniques allowed for closed-circuit Ṁ_O2 measurements with a reduced accumulation of metabolic end-products. As with the rapid RHI P_crit trials, we used closed-circuit respirometry to make a series of normoxic Ṁ_O2 readings between 25 and 19 kPa prior to the active (but in this case gradual) reduction of Pw_O2 to aid in our calculation of P_crit (see below). Upon reaching 19 kPa, we converted the system back to flow-through, reestablished a Pw_O2 of ∼21 kPa, and then commenced the active water volume replacement. The average duration of these combined flow-through and closed-circuit trials was 444±12 min.

We ran two separate but identical parallel sets of hypoxic exposures to investigate potentially causal physiological factors in a time-dependent reduction in P_crit. These parallel exposures involved manually reducing Pw_O2 of aquaria from normoxia to anoxia over 60 and 480 min periods to represent rapid and gradual RHIs, respectively. We also ran normoxic control exposures during which Pw_O2 remained normoxic for 480 min following the habituation period. Each exposure was run in two 10 litre aquaria housing four fish each, and each aquarium was fitted with a screen just below the water surface to prevent the fish from accessing the air–water interface. We mimicked the respirometric P_crit trials described above as closely as possible, with exposures run at 17°C at the same time of day (each trial commenced at ∼09:00 h) following a ≥16 h habituation period, and in complete darkness. Fish from the first set of parallel exposures were sampled to assess gill morphology and haematological parameters, and fish from the second set of parallel exposures were sampled to measure plasma [lactate].

At the end of each exposure, fish were euthanized by inconspicuously introducing anaesthetic (buffered MS-222, final concentration of 200 mg l⁻¹) to the water. Once fish ceased to respond to a tail pinch, individuals were removed, weighed, and then blood was sampled and gills were dissected. To sample blood from the fish in the first set of parallel exposures, the fish's tail was severed and blood was collected from the caudal preduncle using a 60 μl heparinized capillary tube. Ten microlitres of blood was pipetted into 1 ml Drabkins reagent for determination of [Hb], 20 μl of blood was mixed with 10 μl of heparinized Cortland's saline plus 80 μl of 3% perchloric acid for determination of the concentration of red blood cell (RBC) organic phosphates (ATP and GTP; [NTP]), and 10 μl of blood was mixed with 5 μl of heparinzed Cortland's saline for determination of Hb–O₂ binding affinity. The entire right gill basket was then removed from the fish and immediately immersed in 1 ml of Karnovsky's fixative (25% glutaraldehyde, 16% formaldehyde, 0.15 mol l⁻¹ sodium cacodylate, pH 7.4). Twenty-four hours later, the gill basket was transferred to 0.15 mol l⁻¹ sodium cacodylate and stored at 4°C until use. This procedure was repeated for all four fish in each tank, and then duplicated for the second tank of four fish, yielding N=8 for each treatment. For the second set of parallel exposures, fish were euthanized and blood was collected in the same manner as before, but the plasma was separated from the red blood cells by centrifugation and immediately assayed for plasma [lactate] (see below).

The goal of these parallel exposures was to assess the effects of RHI on physiological adjustments that may explain differences in P_crit, but there are differences between the P_crit trials and the parallel exposures that the reader should be made aware of. The main difference was vessel size (respirometer chamber was 32 ml and exposure aquaria were 10 litres), which could have affected the ability of the fish to move throughout the exposure. However, observations of the fish in the 10-litre aquaria suggest that goldfish do not increase activity during progressive hypoxia exposure. Furthermore, the parallel exposures were terminated when Pw_O2 reached 0 kPa. As the samples were taken at this point, the haematology and gill morphology measurements were not taken precisely at the point at which we observed differences in P_crit, and this could affect our ability to relate the two studies. However, the fish used for the haematology and gill morphology analyses were only exposed to an additional ∼7 to ∼15 min of progressively deepening hypoxia (for rapid and gradual RHI, respectively) beyond what they were exposed to by the time P_crit had been reached. Thus we do not believe these relatively minor differences in time would affect our ability to directly relate these components of our study.

where SA_F is the mean lamellar surface area of the five analyzed filaments, F is the number of filaments per gill arch, and 16 is the product of two hemibranchs per gill arch, four gill arches per gill basket and two gill baskets per individual fish (according to Wegner, 2011).

Hb–O₂ binding affinity was determined within 60 min of blood sampling by constructing an oxygen equilibrium curve (OEC) using the thin film spectrophotometric technique (Lilly et al., 2013) and a 96-well microplate spectrophotometer (Molecular Devices, Sunnyvale, CA, USA). A Wostoff gas mixing pump (H. Wösthoff Messtechnik GmbH, Bochum, Germany) mixed compressed O₂ and N₂ to each of nine P_O2 values between 0 and 21 kPa P_O2, always starting with 0 kPa and working toward 21 kPa, and each P_O2 was maintained for 20 min, during which Hb–O₂ saturation was determined spectrophotometrically. A sigmoidal OEC was fit through the % Hb–O₂ saturation versus P_O2 data for each fish, and Hb P₅₀ (the P_O2 at which Hb is 50% saturated with O₂) was determined using SigmaStat 11.0.

We measured whole blood [Hb] spectrophotometrically at 17°C and 540 nm after conversion to cyanomethemoglobin using Drabkin's reagent (Sigma-Aldrich). The measurements were made using a Shimadzu UV-160 spectrophotometer and a millimolar extinction coefficient of 11.

We measured RBC [NTP] spectrophotometrically at 17°C using the GAPDH- and PGK-catalyzed reactions converting glycerate 3-phosphate to glyceraldehyde 3-phosphate, where the oxidation of NADH to NAD⁺ was measured at 340 nm (Bergmeyer et al., 1983). Finally, we measured plasma [lactate] spectrophotometrically at 17°C using the LDH-catalyzed reaction converting lactate to pyruvate, where the reduction of NAD⁺ to NADH was measured at 340 nm (Bergmeyer et al., 1983). [NTP] and [lactate] were measured using a 96-well microplate spectrophotometer (Molecular Devices).

The accumulation of metabolic end-products during closed-chamber/circuit respirometry is regarded as a major shortcoming of the technique (e.g. Keys, 1930; Rogers et al., 2016; Snyder et al., 2016; Steffensen, 1989), but measurements of metabolic end-products in the respirometer chamber are rarely, if ever, reported. To address this knowledge gap, we ran a separate set of closed-circuit P_crit trials (91±10 min) to measure accumulated levels of CO₂ and nitrogenous end-products (NH₃+NH₄⁺). For each of four fish, we took water samples from the respirometer chamber at three time points: start of the habituation period; end of a 16 h habituation period immediately prior to starting the P_crit trial; and end of the P_crit trial as soon as the respirometer's Pw_O2 reached 0 kPa. Pw_CO2 was determined using the Henderson–Hasselbalch equation and measurements of total CO₂ content in the water (CO₂+HCO₃⁻; Corning 965 Carbon Dioxide Analyzer, Corning, NY, USA) and pH (probe: SaS gK2401C, Radiometer Analytical, France; meter: VWR Symphony SB70P, VWR, Radnor, PA, USA). Total ammonia (NH₃+NH₄⁺) was measured using an API ammonia test kit, and unionized ammonia (NH₃) was calculated from this value in combination with the particular trial's water pH and temperature (17°C).

P_crit is defined as the Pw_O2 at which an organism's stable Ṁ_O2 transitions from being independent of to being dependent upon Pw_O2. There are different methods to calculate P_crit, but analyses performed by Mueller and Seymour (2011) suggest that most of the methods used yield comparable values. We therefore decided to use a variation on a two-segment linear regression model (details below) to identify P_crit as the Pw_O2 at which the two linear trend lines (one representing the Pw_O2 range of oxyregulation, the other of oxyconformation) intersect on a graph plotting Ṁ_O2 as a function of Pw_O2 (BASIC program of Yeager and Ultsch, 1989). This method is employed widely throughout the literature (see Rogers et al., 2016) and has been used on goldfish (Fu et al., 2011; Dhillon et al., 2013; Regan et al., 2017).

We calculated Ṁ_O2 values by measuring the change in Pw_O2 over sequential time intervals: 5 min between 25 and 5.3 kPa and 2 min between 5.3 and 0 kPa. To standardize our estimates of a stable, oxyregulated Ṁ_O2, we used the mean of each fish's calculated Ṁ_O2 values between 21 and 18.7 kPa Pw_O2. This represented a normoxic routine Ṁ_O2 that was likely close to standard Ṁ_O2 as a result of it being made following a habituation period that was ≥16 h. We then determined P_crit as the intersection of this horizontal line with a linear regression through the Ṁ_O2 values that were >15% below the mean routine Ṁ_O2 value. This technique was carried out according to McBryan et al. (2016).

We compared all average values of P_crit, normoxic Ṁ_O2, blood properties, gill morphometrics and accumulated Pw_CO2 and nitrogenous end-products using one-way, two-tailed ANOVAs with a critical α=0.05 (repeated measures for the water pH, Pw_CO2 and nitrogenous end-products comparisons). Post hoc Tukey tests were used to test for differences between treatment groups. Any data set that did not meet the assumptions of normality or equal variance were log-transformed prior to analysis. All analyses were performed using SigmaStat 11.0. Values reported in the text are presented as means±s.e.m.

Long duration P_crit trials resulted in P_crit values that were approximately half those of short and typical duration P_crit trials (ANOVA, P<0.001; Fig. 1A, Fig. S1). P_crit values determined by short and typical trial durations did not differ from one another, nor did the P_crit values determined by the three respirometric techniques used for the long duration trials (Fig. 1A). Each of the five respirometric techniques yielded statistically similar normoxic Ṁ_O2 values (ANOVA, P=0.276; Fig. 1B).

RHI significantly affected the mass-specific lamellar surface areas of goldfish (ANOVA, P=0.004; Fig. 2), whereby fish exposed to gradual RHIs had ∼60% larger lamellar surface areas than fish exposed to rapid RHIs and normoxic controls, which did not differ.

RHI significantly affected Hb–O₂ binding affinity (ANOVA, P=0.007; Fig. 3). Goldfish exposed to rapid RHIs had ∼60% higher Hb P₅₀ values than goldfish exposed to gradual RHIs and normoxic controls, which did not differ.

RHI did not affect whole-blood [Hb] (ANOVA, P=0.334; Fig. 4A), but it did affect RBC [NTP] (ANOVA, P=0.001; Fig. 4B), whereby gradual RHI fish had RBC [NTP] values that were approximately half those of the rapid RHI and normoxic control fish, which did not differ.

Goldfish exposed to rapid and gradual RHIs both accumulated similar concentrations of plasma lactate to a level significantly higher than that observed in normoxic control fish (ANOVA, P=0.001; Fig. 4C).

Respirometer chamber Pw_CO2 doubled over the course of a 16 h habituation period under flow-through conditions, and increased 6.5-fold over the course of a typical duration closed-circuit P_crit trial (91±10 min; ANOVA, P<0.001; Fig. 5A). Water pH was concomitantly reduced from 7.61 to 6.93 over the course of the P_crit trial (ANOVA, P<0.001; Fig. 5B). The concentration of total ammonia (NH₃+NH₄⁺) in the chamber also increased (ANOVA, P<0.001; Fig. 5C). Unionized ammonia (NH₃) accumulated in a different way because of pH changes of the water, with [NH₃] tripling over the 16 h habituation period, then falling to an intermediate value by the end of the P_crit trial (ANOVA, P<0.001; Fig. 5D).

We hypothesized that gradual RHIs would allow goldfish to induce time-dependent plastic phenotypes that enhance O₂ uptake. This hypothesis predicted that the P_crit of goldfish exposed to long duration P_crit trials would be lower than those of goldfish exposed to short or typical duration P_crit trials, and our results agree with these predictions regardless of the respirometric technique used. Furthermore, fish exposed to gradual RHIs developed greater lamellar surface areas and Hb–O₂ affinities over their exposures than fish exposed to rapid RHIs, both characteristics that would enhance O₂ extraction under hypoxia and therefore likely explain the lower P_crit associated with the long duration trials. Taken together, our results suggest that time (more precisely, RHI) is a significant determinant of P_crit in goldfish.

The majority of P_crit measurements are made using closed-chamber respirometry over the course of 60 to 120 min (Rogers et al., 2016). Here, our representative closed-circuit P_crit trials lasted ∼84 min and resulted in a P_crit of 2.5±0.2 kPa (Fig. 1A). Our values are in general agreement with values previously reported for goldfish [∼3.6 kPa (Fry and Hart, 1948); 3.0 kPa (Fu et al., 2011); 3.3 kPa (Dhillon et al., 2013); and 3.0 kPa (Regan et al., 2017)], though slightly lower owing to a possible combination of experimental temperature differences and the fact that our study used closed-circuit respirometry as opposed to static closed-chamber respirometry. Reducing the trial duration to ∼24 min did not affect P_crit (Fig. 1A), which may not be surprising considering the RHI below a Pw_O2 of 5.3 kPa was similar between the short and typical duration P_crit trials (see Materials and methods for details). However, our results clearly indicate that increasing the trial duration from ∼84 min (i.e. reducing its RHI) to ∼480 min resulted in significantly lower P_crit values. The reasons for this variation could be related to time, technique or some combination of the two, and we will explore these possibilities below.

Goldfish exposed to gradual RHIs developed significantly larger lamellar surface areas than those of normoxic controls and goldfish exposed to rapid RHIs, which did not differ. Hypoxia-induced gill remodeling was first observed in goldfish and the closely related crucian carp (Carassius carassius) 13 years ago (Sollid et al., 2003, 2005) and in numerous fish species since [e.g. mangrove killifish, Kryptolebias marmoratus (Ong et al., 2007; Turko et al., 2012); African cichlids (Crispo and Chapman, 2010); various carp species (Dhillon et al., 2013); Atlantic salmon Salmo salar (Anttila et al., 2015); and Atlantic killifish Fundulus heteroclitus (Borowiec et al., 2015)]. Dhillon et al. (2013) observed a near-doubling in the lamellar surface area of goldfish following 8 h acclimation to a constant Pw_O2 of 0.7 kPa, but to our knowledge, the present study is the first time gills have been shown to remodel over such short time scales under progressively decreasing Pw_O2. Increases to lamellar surface area are typically the result of apoptotic reductions to the ILCM (Sollid et al., 2003). ILCM reductions per se also enhance the gill's diffusion capacity (Bindon et al., 1994; Greco et al., 1995) and contribute to a reduced P_crit in crucian carp (Sollid et al., 2003) and Atlantic killifish (Borowiec et al., 2015). However, a study that examined (among other things) O₂ diffusion across the gills of two groups of goldfish with temperature-induced differences in gill surface area found that the differences in gill surface area had no effect on arterial P_O2 when acutely exposed to hypoxia (Tzaneva et al., 2011). Although this seems to run counter to what Fick's first diffusion law would predict, the authors speculated that the goldfish that started hypoxia exposure with a smaller gill surface area may have been rapidly remodeling their gills to increase lamellar surface area over the course of the acute hypoxia exposure. Our gill morphometric results lend empirical support to this speculation.

The mean Hb P₅₀ value of gradual RHI goldfish (1.6 kPa, similar to long duration P_crit of ∼1.2 kPa) was significantly lower than that of rapid RHI goldfish (2.4 kPa, similar to typical duration P_crit of 2.5 kPa), but no different than normoxic controls (1.4 kPa). This implies that rapid hypoxia induction reduces Hb–O₂ binding affinity, but gradual induction does not. The underlying mechanism(s) might involve RBC [NTP] and/or pH. Nucleoside triphosphates (ATP and GTP, collectively NTPs) reduce Hb–O₂ binding affinity by binding to sites on the Hb tetramer that stabilize its deoxygenated conformation and consequently increase the P₅₀ (Jensen et al., 1998; Wood and Johansen, 1972). The significantly lower [NTP] values of our gradual RHI fish (Fig. 4B) at least partly explain their lower Hb P₅₀ values compared with those of the rapid RHI fish (Fig. 3), but the similar [NTP] values in the rapid RHI and normoxic control fish exclude this mechanism as the cause of the rapid RHI fish's elevated Hb P₅₀ values. Another possible mechanism for the rapid RHI fish's reduced Hb–O₂ binding affinity is RBC pH. We did not measure RBC pH, but if protons accumulated to higher concentrations in the RBCs of rapid RHI goldfish than gradual RHI goldfish, then this would likely have reduced Hb–O₂ binding affinity via goldfish's Bohr/Root effect (Rodewald and Braunitzer, 1984). Regardless of the causal mechanism(s), the different Hb–O₂ binding affinities of the rapid and gradual RHI fish are likely to at least partly explain their different P_crit values.

Respirometric techniques can be broadly categorized as closed (closed-circuit or static closed-chamber), flow-through or intermittent flow. Though none of these techniques are ideal for all experimental scenarios, intermittent flow respirometry is generally regarded as superior because it avoids the potential accumulation of metabolic end-products presumed to occur in closed respirometry and it has greater temporal resolution than flow-through respirometry (reviewed by Clark et al., 2013; Steffensen, 1989; Svendsen et al., 2016). It has been suggested that the choice of respirometric technique used to determine P_crit may influence the results, and indeed P_crit in shiner perch (Cymatogaster aggregata) shifted from ∼9.9 kPa to ∼6.1 kPa when using closed-chamber versus intermittent flow respirometry, respectively (Snyder et al., 2016). The authors attribute this to technique, but also discuss the possibility that duration of the P_crit trials (∼1 h for closed-chamber, ∼5 h for intermittent flow) may play a role (Snyder et al., 2016). In the present study, we used modified versions of all three respirometric techniques for our long duration P_crit trials, and, despite technique specific-differences and challenges (e.g. flow-through trials demanded a step-wise reduction in Pw_O2; closed-circuit trials resulted in higher Ṁ_O2 values in the mid-Pw_O2 range; Fig. S1), each technique yielded nearly identical P_crit values, which were all lower than the typical or short duration P_crit trials. This suggests that the different P_crit values observed between our short and typical duration P_crit trials and those of the long duration trials are the result of RHI rather than technique, and this may also be the case with the results of Snyder et al. (2016).

The fact remains that closed respirometry leads to end-product accumulation (Fig. 5), and this could theoretically influence P_crit (Keys, 1930; Snyder et al., 2016; Steffensen, 1989). However, within our long duration P_crit trials, closed-circuit trials (where CO₂ accumulated) and combined flow-through/closed-circuit trials (where CO₂ did not accumulate) resulted in nearly identical P_crit values (Fig. 1A). This suggests that the levels of metabolic end-products that accumulate with closed-circuit respirometry are not high enough per se to have a significant effect on P_crit.

Our results demonstrate that RHI significantly alters the P_crit of goldfish, whereby long duration P_crit trials (i.e. gradual RHIs) yield lower P_crit values than short duration P_crit trials (i.e. rapid RHIs). These reduced P_crit values are caused by time-dependent effects on mechanisms that enhance environmental O₂ extraction, including gill morphology and Hb–O₂ binding affinity. Fishes generally possess numerous time-dependent mechanisms that enhance O₂ extraction in response to hypoxia, so RHI is likely an important factor to consider in all fish species when carrying out P_crit trials and experimental hypoxia exposures. Because longer duration P_crit trials allow for some degree of acclimation that may consequently reduce P_crit, shorter duration P_crit trials are likely to best represent the innate abilities of a hypoxia-exposed fish to extract and use O₂ at the time of analysis. Thus, similar to the standardized rate of temperature change used when determining a fish's CT_max, an RHI should be chosen that is fast enough to avoid acclimation during the trial. Our data suggest that P_crit trials of <90 min are probably sufficient to achieve this.

We thank Tara McBryan for her assistance with gill morphometrics, and Phillip Morrison for his assistance with OEC measurement and construction. We also thank the two reviewers for their helpful, insightful comments.