Arapaima gigas maintains gas exchange separation in severe aquatic hypoxia but does not suffer branchial oxygen loss

It is broadly accepted that air breathing in fish has evolved in response to recurring hypoxia in the aquatic environment (Bayley et al., 2018). Air-breathing fish are categorized as obligate or facultative, depending on whether they rely on aerial respiration or use it as a supplementary strategy for O₂ uptake (Johansen, 1970). While there is varying dependence on aerial O₂ uptake, there is general agreement that CO₂ excretion occurs predominantly across the gills (Rahn and Howell, 1976; Singh, 1976; Brauner and Val, 1996; Graham, 1997c; Gonzalez et al., 2010; Brauner et al., 2019; Pelster et al., 2020). This represents an interesting problem, because the circulatory system of air-breathing teleosts has an undivided heart (Olson, 1994; Ishimatsu, 2012). Hence, it is conceivable that O₂-rich blood returning from the air-breathing organ (ABO) may lose O₂ to the water in hypoxia during passage through the gill basket (Smatresk and Cameron, 1982; Smatresk, 1990; Graham, 1997a; Graham and Wegner, 2010). Actual evidence for O₂ loss is limited to very few studies. Randall et al. (1981a) estimated the O₂ loss for the bowfin (Amia calva) to be 7% in hypoxia (14 mmHg) based on rather derived calculations and incomplete blood oxygen affinities across temperature, while a branchial O₂ loss of approximately 10% was found in the spotted gar (Lepisosteus oculatus) in hypoxia (12 mmHg) by direct measurement of O₂ content in blood before and after transit through the branchial basket (Smatresk and Cameron, 1982). Finally, differences in O₂ content between inspired and expired water were measured in the facultative air-breathing armoured catfish (Hypostomus pyrineusi) by Scott et al. (2017). Here, evidence for minor O₂ loss was found in only 4 of 11 animals in severe hypoxia (2 mmHg).

Furthermore, in several of the lungfishes (Johansen et al., 1968; Burggren and Johansen, 1986) and in the teleost snakehead fish Channa argus (Ishimatsu and Itazawa, 1983), separation of blood flow through the undivided heart has been documented. Such separation allows for the continued excretion of CO₂ from deoxygenated blood into the water while protecting the O₂-rich blood from the ABO from the risk of O₂ loss during branchial passage if a branchial shunt is present, such as a highly reduced gill arch or a trans-branchial vessel.

One of the most air-reliant obligate air-breathing fish is Arapaima gigas (Osteoglossiformes) (Brauner et al., 2004), which is native to tropical South America. Arapaima gigas starts life as a water breather (1–5 g) (Frommel et al., 2021) as is probably the case with almost all air-breathing fish (Bayley et al., 2020), but rapidly transitions to an air breather, which is accompanied by a radical re-modelling of the gills (Brauner et al., 2004). The inter-lamellar space fills with cell mass and ionocytes migrate to cover the now column-shaped filaments (Brauner et al., 2004). This greatly increases the blood–water diffusion distance and reduces the capacity for O₂ uptake (Brauner et al., 2004). The percentage aquatic O₂ uptake after this transition is reduced to 20–30% under normoxic conditions, with the remaining O₂ requirements being satisfied by aerial O₂ uptake (Randall et al., 1978; Brauner and Val, 1996; Gonzalez et al., 2010; Pelster et al., 2020). During progressive aquatic hypoxia, O₂ partitioning is further skewed towards the air phase where it will ultimately account for 100% of O₂ uptake (Stevens and Holeton, 1978). In normoxia, however, A. gigas continues to excrete 60–90% of its metabolically produced CO₂ to the water (Randall et al., 1978; Brauner and Val, 1996; Gonzalez et al., 2010; Pelster et al., 2020). Severe hypoxia or even anoxia frequently occurs in the tropical freshwater habitats of A. gigas (Johansen et al., 1978; Richards et al., 2007). However, no studies have yet looked at the effect of such deep aquatic hypoxia on the partitioning of CO₂ excretion or for evidence of O₂ loss to the water.

The reduced gill surface area and increased blood–water distance in A. gigas seem also to impair gill diffusion of ammonia, manifesting in an elevated blood total ammonia concentration despite mass-specific ammonia production being lower in larger fish (Gonzalez et al., 2010; Pelster et al., 2020; Wood et al., 2020). Severe aquatic hypoxia has been shown to reduce branchial ventilation in many fish (Graham et al., 1978; Mattias et al., 1998; Oliveira et al., 2004; Perry et al., 2009; Belão et al., 2011, 2015; Thomsen et al., 2017; Mandic et al., 2020). Wood et al. (2020) found a substantial reduction in ammonia excretion during aquatic hypoxia (31 mmHg) in A. gigas, suggestive of reduced branchial ventilation and/or diffusive conductance in aquatic hypoxia. If this is the case, then a shift in the partitioning of CO₂ excretion towards the air phase in severe hypoxia would be expected, which in turn would reduce the risk of branchial O₂ loss in hypoxic water. Indeed, Bayley et al. (2020) argued that the astounding growth rates observed in air-breathing fish grown in deeply hypoxic aquaculture ponds – which is certainly the case for A. gigas, growing to 10–15 kg within a year (Ramírez et al., 2018) – do not fit well with significant O₂ loss in hypoxic water.

Measurements of excreted CO₂ in water are notoriously difficult, because of the equilibration of this gas with the bicarbonate and carbonate pools in water (Wang et al., 2021). However, Harter et al. (2017) recently showed that by passing respirometer water through an aggregate of gas-permeable plastic, flushed on the other side with a constant gas medium, it was possible to measure water expired CO₂ using a gas analyser. Here, we expanded this approach using commercially available medical grade artificial lungs in combination with previously developed two-phase intermittent closed respirometry (Lefevre et al., 2011, 2012, 2016) to allow for steady-state simultaneous measurements of O₂ and CO₂ exchange in water and air, and nitrogenous waste (total ammonia-N plus urea-N) excretion into the water phase. This allowed us to test the hypothesis that during severe hypoxia, A. gigas will take measures to reduce the risk of O₂ loss through the gill (e.g. reduced diffusive conductance, reduced ventilation, blood shunting to non-respiratory pathways), which will then also shift partitioning of CO₂ towards the ABO. To test this, we exposed the fish to a level of aquatic hypoxia significantly below the P₅₀ of the blood, which for A. gigas at 28°C is 21 mmHg (Johansen et al., 1978). If, however, the gas exchange partitioning was not changed during hypoxia, we hypothesized a branchial O₂ loss. This was tested simultaneously with the first hypothesis as the water P_O2 levels were continuously logged. Furthermore, we hypothesized that metabolic rate would be increased in hypoxia, as increased surfacing is energetically costly (Lefevre et al., 2013). Lastly, in light of the observation of Wood et al. (2020) of reduced ammonia excretion during aquatic hypoxia, we hypothesized that this might be reflected in a shift in metabolic fuel utilization.

This study was conducted using two-phase respirometry to measure the simultaneous bimodal gas exchange of A. gigas. Intermittent closed respirometry was used to measure O₂ uptake and CO₂ excretion in the water phase and flow respirometry was used for both O₂ uptake and CO₂ excretion in the air phase. Measurement of O₂ uptake in both phases and CO₂ excretion to the air phase occurred only in the closed periods of the intermittent closed respirometry, while CO₂ excretion to the water from the closed period was measured during the open period.

Respirometry measurements were performed on 7 pirarucu, Arapaima gigas (Schinz 1822) (1055–2025 g, mean of 1583 g), at Aarhus University. The animals were obtained from the Aquaculture Association of Caqueta (ACUICA, Florencia, Colombia), transferred to holding facilities at Aarhus University, and grown to experimental size at 29.5°C in large fibreglass composite tanks (1000 l) connected to a recirculating system supplied with soft water (pH 7.7, conductance ∼160 μS cm⁻¹) at 29.5°C. The fish were fed to satiation once daily (on weekdays) using catfish feed (Sigma 811R 6 mm, BioMar A/S, Denmark; protein 44%, lipid 16%, carbohydrate 20.5%). From 24 h prior to experimentation, fish were isolated in separate tanks within the same system and fasted. Fasting was continued throughout the experimental protocol. The experiment was performed in compliance with the Danish Law for Animal Experimentation.

Aerial and water Ṁ_O2 and Ṁ_CO2 for seven A. gigas were measured in normoxic (P_O2 134.5±5.52 mmHg, mean±s.d.) and severely hypoxic (1.10±3.12 mmHg) soft non-chlorinated water (conductance ∼150 μS cm⁻¹, pH ∼7) at 28°C and a normoxic and normocapnic air phase (20.95% O₂, 0.04% CO₂). The fish were randomly divided into groups starting either in normoxic (N=4) or in severely hypoxic (N=3) water and moved into the respirometer chamber 1–2 h before the first measurement. Measurements were initiated in the late afternoon, lasting 17.5 h per treatment until the following morning to minimize disturbance. Each measurement consisted of two measurement periods, one in each treatment (i.e. normoxia or severe hypoxia) separated by a non-measurement period (∼7 h) during the daytime, totalling ∼42 h. During this non-measurement period, water oxygenation level was the same as in the first measurement period. After the second measurement period, the fish was removed from the respirometer and bacterial gas exchange was recorded. A CO₂ calibration measurement was performed by injecting a known quantity of CO₂ into the chamber.

The air and water phases were not separated, so diffusion between the two phases was possible. However, water P_O2 remained constant during background Ṁ_O2 measurements in severe hypoxia, giving no indication of diffusion from the normoxic air phase to the hypoxic water phase. The respirometer consisted of a water-filled cylinder (30.1 l) with a small air-filled box (818 ml) added at one end where the fish could air breathe, and the setup is shown in Fig. 1. The water level of the system resulted in a small intrusion of water into the air-phase box and a net air-phase volume of 704 ml. The cylinder was sealed at both ends and a closed circulation loop through the ends maintained continuous water flow for aquatic O₂ and N-waste measurements. Oxygen was therefore sampled continuously during the closed period while N-waste was sampled at the beginning and end of each closed period by taking water aliquots of 5 ml. Finally, an open-loop system for flushing the cylinder with water from a header tank (212 l) was added at each end of the respirometer. The air space of the box on top was flushed with two tubes and the outflow was measured with aerial flow-respirometry that were sealed during water flush to avoid respirometer water volume changes, using electrically controlled valves. Because of the design of the respirometer, it is possible that a small dead space was present near the water surface with reduced water circulation adding a resistance to O₂ diffusion into the water phase. If so, we judge this volume to have been very small relative to the 30 l respirometer volume.

Gas exchange to the air phase was continuous. The air phase cycled through being open during the closed water respirometry period lasting 20 min and closed during water flushing lasting 15 min. Aerial gas exchange that occurred during the closed period was analysed upon opening of the air phase after each closed period.

Water phase gas exchange was quantified for the closed periods and hence measured for 20 out of every 35 min.

O₂ consumption from the water (Ṁ_O2,water) was measured using intermittent closed respirometry as described previously (Lefevre et al., 2011, 2012, 2016). The sides of the tank were covered in black plastic to reduce visual stimuli. Hypoxia was achieved by bubbling a header tank (212 l), covered with plastic wrap, with N₂ while P_O2 was monitored with a Hamilton optode (Visifirm DO, Hamilton Company, Bonaduz, GR, Switzerland). The normoxic water was vigorously bubbled with air to maintain normoxia in the header tank, as water returning from the respirometer was O₂ depleted as a result of both fish respiration and O₂ removal in the artificial lung used for CO₂ measurements (see below). During the flush period, the respirometer water was renewed from the header tank maintained at either normoxia or severe hypoxia in an open-loop system returning to the header tank.

To confirm the sensitivity of the system to detect O₂ loss by the fish, an analysis of detection limit was performed (see below) using the same level of hypoxia as in fish trials. Five different series were conducted with injections of O₂-saturated distilled water (0.5, 1.2, 2.5, 5 and 10 ml). Injections were made once per minute for 20 min. This corresponded to an O₂ appearance rate of 0.59, 1.48, 2.95, 5.90 and 11.66 μmol min⁻¹, respectively; for a typical 1.5 kg fish, in turn these rates corresponded to O₂ loss rates of 0.37, 0.93, 1.86, 3.73 and 7.37 μmol kg⁻¹ min⁻¹.

Upon completion of a replicate, every part of the respirometer and tank was carefully cleaned using a biocidal agent (Virkon-S), and the gas analysers were calibrated before each run using 100% N₂ gas for the zero calibration, a mixture of 95% N₂ and 5% CO₂ gas for the CO₂ span (5%), and room air for the O₂ span (20.95%).

O₂ consumption and CO₂ excretion with the normoxic and normocapnic air were measured using flow-through respirometry. Air flow was continuous except during water flush, where electronic valves sealed the air phase in the respirometer to prevent changes in water/air levels. Gas exchange occurring during the sealed period was recorded during the subsequent open air flow period. This sealing of the air phase during water flush resulted in P_O2 falling from the daily atmospheric level (ca. 155 mmHg) to a minimum of 117 mmHg and P_CO2 increasing from 0.03 to a maximum 21 mmHg. The air flow rate was measured using a mass flow meter (Sierra Instruments model 830D-L-1-V1) and water vapour was scrubbed from the air using anhydrous CaCl₂. P_O2,air and P_CO2,air were measured using a FC-10 Oxygen Analyzer and a CA-10 Carbon Dioxide Analyzer (Sable Systems Europe GmbH), respectively, and data logged with an MP100 biopac system (Biopac Systems Inc., Goleta, CA, USA) in AcqKnowledge (version 3.9.1, Biopac Systems Inc.).

Respirometer water was sampled from the closed circulation loop at the beginning and end of each closed respirometry cycle using a time-programmable autosampler (Waters Fraction Collector III, Waters Corporation, Milford, MA, USA) connected to a peristaltic pump (Ole Dich, 115LMT, Hvidovre, Denmark) run by the respirometry program described above. There was already water in the autosampler tubing deadspace (4 ml) at the onset of each sampling. To avoid collecting mixed experimental water, we sampled experimental water in duplicate (2×5 ml) and discarded the first sample. The appropriate volume of the discarded sample (i.e. 5 ml) was determined by (1) filling the autosampler tubing deadspace with non-coloured water and then (2) sampling water coloured with greed food colorant (E161b and E133) until it fully replaced the non-coloured water.

During the unsupervised respirometry measurements, the autosampler tray was filled with ice in order to maintain samples at low temperature. All samples were capped and frozen at −20°C promptly after the final sampling for later analysis. The pH of the system ranged from 6.6 to 7.0, making diffusive NH₃ loss negligible.

Total ammonia-N concentration was measured in water samples using the colorimetric method of Verdouw et al. (1978). Urea-N measurements (2 N per urea molecule) of water samples followed the colorimetric method of Rahmatullah and Boyde (1980).

Ammonia and urea excretion rates were calculated from respective colorimetric assays to calculate the nitrogen quotient [NQ=(ammonia-N+urea-N)/Ṁ_O2], and the fish-specific respiratory quotient for protein (RQ_protein; the specific RQ value for protein varies from 0.83 to 0.97 depending on the N-excretory product) according to Lauff and Wood (1996).

Fuel use (%) was calculated according to Lauff and Wood (1996). In short, the method uses measurements of respiratory exchange ratio (RER) and NQ to stoichiometrically calculate fractional metabolic fuel use contributions of the major aerobic fuels: carbohydrate, protein and lipid. RQ is assumed to be equal to RER under aerobic steady-state conditions, calculated by Ṁ_CO2/Ṁ_O2.

During the initial measurements, the fish were stressed as gas exchange was elevated and RER was generally well above 1, indicating either CO₂ generation from titration of bicarbonate produced by anaerobic metabolism or non-steady-state acid–base conditions in the plasma. RERs above 1 prevent calculation of fuel use (%) because the stoichiometric approach only works within the aerobic span of RER=0.7–1.0 (Kleiber, 1961). Therefore, all initial stressed measurements were excluded, and only the second measurement period was used for fuel use calculations. Fuel use calculations could not be performed on two fish because of incomplete datasets.

Arapaima gigas appeared initially stressed after transfer to the respirometer. As the measurements started within 2 h of transfer, this stress was clearly evident in the initial Ṁ_O2 traces. The initial high O₂ uptake lasted longer in the hypoxic group than in the normoxic group. Additionally, there were occasional bouts of activity with continuous movement interspersed with more stationary behaviour with occasional surfacing glides to air breathe. Gill ventilation was observed in both normoxia and severe hypoxia and, while this was not measured, there were no apparent differences in gill ventilation between the normoxic and hypoxic water.

An exponential decline fitted to the average traces of fish started in normoxia (start normoxia) or severe hypoxia (start hypoxia) was made and compared to determine whether the decline followed the same trend in the two treatments. The two fitted curves were significantly different with a P-value of 0.0003 and could therefore not be fitted as one curve. The normoxic trendline showed a significantly faster return to baseline levels compared with the start of the hypoxic group. To exclude elevated stress values, an analysis of the exponential decline was made to obtain unbiased values of the plateau (Fig. 2). This test allowed the calculation of a time constant, τ, of 7.46 h for fish started in hypoxia and 5.84 h for fish started in normoxia: 3τ was used as a cut-off point to identify the end of stress, where 95% of the measurements did not deviate significantly from the plateau of the fitted exponential decline. This resulted in the removal of the first measurement period of fish started in severe hypoxia as their 3τ (3×7.46 h) exceeded the 17.5 h duration of the treatment. Hence, we used only the second measurement period for the statistical analysis of our hypotheses concerning CO₂ excretion and differences between normoxia and hypoxia (N=4 severe hypoxia and N=3 normoxia). For the O₂ loss measurement, all data of the 7 fish in severe hypoxia were used as stress is of interest here.

The initial stress period increased all parameters, especially in fish starting in severe hypoxia (Fig. 2). The large increases in CO₂ excretion during stress were also evident in the aerial RERs and total RERs, which were initially much higher for the fish started in severe hypoxia. Once the stress period was passed, 80% of the O₂ was obtained from the air phase in normoxia against 100% in severe hypoxia (Table 1) whereas total O₂ uptake remained almost constant between the treatments, with estimates for SMR at 33 and 36 μmol min⁻¹ kg⁻¹ in normoxia and severe hypoxia, respectively.

Surprisingly, the aerial partitioning of CO₂ excretion changed little (though almost significant, P=0.051) when comparing fish in normoxia with those in severe hypoxia. The percentage aerial excretion was 17% in normoxia against 23% in severe hypoxia, which can be compared with the corresponding O₂ values, which changed from 80% to 100% (Table 1). This large increase in aerial O₂ uptake in hypoxic versus normoxic water was also accompanied by a significant increase in aerial CO₂ excretion, which rose from 5.7 to 8.7 μmol min⁻¹ kg⁻¹. However, when viewing the total excretion of CO₂ in the two aquatic conditions, it is evident that water P_O2 of the two treatments had little impact and the water phase was the recipient of around 80% of the total metabolic excretion of CO₂ in both normoxia and severe hypoxia. The total O₂ uptake did not change significantly between treatments (P=0.42) and neither did total CO₂ excretion (P=0.34), which was reflected in a total RER of 0.94 and 0.96 for normoxia and severe hypoxia, respectively (P=0.59). Likewise, the SMR was not significantly different between fish in normoxia and those in severe hypoxia (P=0.26).

Diffusion of CO₂ from blood to water indicates a potential for O₂ loss by net diffusion in the severely hypoxic water where the blood/water P_O2 gradient would favour loss. However, no O₂ leak was detected in any of the fish (Fig. 3A), with the exception of two fish during the initial stressed period. In one of these (Fig. 3A, pink trace) there was an increase in water P_O2 in 9 of the initial 13 closed periods. The loss of oxygen in this fish corresponded to 0.5–1.36% of its total O₂ uptake in these closed respirometry loops. The second fish (Fig. 3A, orange trace) lost 0.59–0.82% of its total O₂ uptake in the corresponding loops. After around 200 min of severe hypoxia, only the small analog to digital conversion noise (which is unavoidable with the optode technology) raises the P_O2 above baseline and we concluded that O₂ loss from the fish was below our detection limit. The initial decrease in water P_O2 occurred because the initial flushes displaced the original normoxic water from the chamber to the severely hypoxic header tank, which then temporarily increased the header tank P_O2.

We performed an experiment to determine our detection limit for O₂ leakage from the fish (Fig. 3B) by injecting known volumes of O₂-saturated water into the respirometer without fish. This revealed that the lowest detectable level of O₂ addition to the water phase was between the lowest and second lowest O₂ injection amount. For an average fish mass in this experiment (1.583 kg), this detection limit would then translate to between 1.0% and 2.6% of their SMR (i.e. 0.36–0.94 μmol min⁻¹ kg ¹). This means that O₂ loss if present in severe hypoxia would only represent a very small fraction of their O₂ uptake.

Our estimate of ammonia-N excretion rate was slightly lower (4.80±0.64 µmol kg⁻¹ min⁻¹) than found in previous studies on A. gigas, which have all been performed on animals with a lower mass than in the present study (Fig. 4B). Urea-N excretion rate was also slightly lower (0.69±0.25 µmol kg⁻¹ min⁻¹) in comparison to previous values for juvenile A. gigas (<70 g) but similar to data found in animals of intermediate size (∼0.65 kg) (Fig. 4A). Calculation of percentage urea-N [urea-N/([urea-N+ammonia-N)] revealed that urea excretion accounted for a small part of the total N-excretion (12.1±3.3%). One fish relied relatively heavily on urea excretion (25% urea-N) while the remaining fish relied much less on urea excretion (8.8±0.7% urea-N) (Fig. 4A).

RQ_protein was high (0.96±0.004) given the low percentage urea-N. NQ was substantial (0.14±0.02) considering the fish's fasted state and none of the measurements exceeded the theoretical maximum NQ of 0.27 that indicates 100% protein utilization under aerobic and steady-state conditions. Aerobic metabolism was calculated to be fuelled predominantly by protein, which accounted for more than half of fuel burned (52.95±7.1%), which was about 2-fold higher than both carbohydrate (26.76±3.3%; P=0.015) and lipid (20.29±6.3%; P=0.009) (Fig. 5A). Carbohydrate and lipid utilization were similar and did not differ significantly (P=0.40). Fuel utilization resembled prior measurements in smaller A. gigas (0.625 kg), but carbohydrate use tended to be more prominent with slightly lower protein use (Fig. 5A). Effects of severe hypoxia were not analysed because of low replication. However, on the basis of our limited data, we could find no indication that severe hypoxia affected either N-waste excretion (Fig. 4) or fuel utilization (Fig. 5B).

Our primary goal was to investigate how partitioning of gas exchange in A. gigas changed when transitioning from aquatic normoxia to hypoxia and how metabolic rate in general was affected. We hypothesized that CO₂ partitioning would shift to the air phase and if not that we would detect an O₂ leak from the gills to the water. Surprisingly, our data show that aquatic CO₂ excretion remained practically unchanged in severe aquatic hypoxia compared with normoxia and that there was no evidence of branchial O₂ loss in hypoxic water except in two fish in the stressed period immediately after introduction into the respirometer.

Although we saw a minor significant increase in aerial CO₂ excretion when A. gigas was exposed to severe aquatic hypoxia, the increase was modest in terms of the continued dominance of the water phase for CO₂ excretion. Overall, we failed to find evidence supporting our hypothesis that A. gigas changed its partitioning of CO₂ excretion when transitioning between normoxia and hypoxia. In aquatic hypoxia, 77% of the total produced CO₂ was excreted to the water phase compared with 83% in normoxia. The significant aerial elevation and small aquatic reduction are expected as the ABO is more strongly ventilated in hypoxia compared with normoxia. The aquatic CO₂ excretion in normoxic water in the present study is within the previously measured range of 63–89% (Randall et al., 1978; Brauner and Val, 1996; Gonzalez et al., 2010; Pelster et al., 2020). Despite the continued use of the water phase to excrete CO₂, the O₂ level (1.1 mmHg) of which was substantially below the blood P₅₀ of A. gigas (21 mmHg; Johansen et al., 1978), we could find absolutely no evidence for any O₂ loss from fish that had become accustomed to the respirometer. Data from two stressed individuals immediately following introduction into the respirometer confirmed that O₂ can be lost as previously suggested (Randall et al., 1981a; Smatresk and Cameron, 1982; Scott et al., 2017). Here, the O₂ loss in this initial stressed state, though detectable, presented a minor fraction of total O₂ uptake at around 1.36% in the affected closed respirometer loops. Further, as these fish calmed with respirometer familiarity, this O₂ loss rapidly decreased to below the detection limit of the system. This is in line with Scott et al. (2017), who found branchial O₂ loss in 4 out of 11 instrumented armoured catfish, at a water P_O2 approaching the P_O2 (1.7 mmHg) at which loss of equilibrium occurred (Scott et al., 2017).

Increased diffusive resistance to O₂ seen in air-breathing fish gills compared with that in closely related water breathers is often seen as a strategy to reduce O₂ loss in hypoxia (Randall et al., 1981a,b; Graham, 1997b; Brauner et al., 2004; Phuong et al., 2017; Frommel et al., 2021). This includes both reductions in respiratory surface area and increases in epithelial thickness of the gills (Phuong et al., 2017, 2018). As A. gigas develops from the larval stage, its interlamellar spaces are filled with cells, resulting in a very small diffusive surface in animals of more than 5 g (Brauner et al., 2004; Frommel et al., 2021). Interestingly, these interlamellar cells do not appear to have been evaluated in terms of any potential plasticity with respect to water oxygen and animal metabolic requirements (e.g. Tzaneva et al., 2014; Phuong et al., 2018), which might be a promising line for future research.

Irrespective of any gill modifications that might reduce O₂ loss in hypoxia, our data show that most of the produced CO₂ is still excreted to the water, meaning that blood and water are not fully isolated. This continued CO₂ excretion to the water phase in deep hypoxia has been indirectly estimated in several other species of air-breathing fish based on very low aerial RER values (Hughes and Singh, 1970, 1971; Smatresk and Cameron, 1982; Graham, 1983; Yu and Woo, 1985). Even though the Krogh's diffusion constant for CO₂ is much larger than that for O₂, the excretion of CO₂ in the teleost gill is still limited by diffusion (Malte and Weber, 1985; Perry and Gilmour, 2002). The reduced diffusion conductance accompanying reduced gills decreases the ability of the gills to excrete CO₂, resulting in an elevated arterial P_CO2 (Brauner et al., 2004) to a new level where production is again matched by diffusive loss. In A. gigas, arterial P_CO2 has been measured to lie between 26 and 31 mmHg, which is approximately 10 times greater than for a typical water-breathing fish (2–3 mmHg) (Randall et al., 1978; Brauner et al., 2004). Hence, the reduced gills fit well with the measured arterial P_CO2. This would also help explain the lack of concomitant O₂ loss in hypoxia. The high barrier for diffusion would especially limit O₂ compared with CO₂ as Krogh's diffusion constant (K) is 20–25 times higher for CO₂ (de Moraes et al., 2005). Whether these adaptations are sufficient to prevent O₂ loss to the degree documented here requires further investigation. Johansen et al. (1978) measured P_O2 in pre-branchial aortic blood in A. gigas as between 45 and 50 mmHg, leading these authors to predict a modest branchial O₂ loss.

An adaptation that might also facilitate this pattern of aquatic CO₂ excretion while completely preventing O₂ loss involves a trans-branchial shunt vessel, allowing blood to pass the gills without coming into contact with water. This would further require a continuous separation of blood flow from the ABO and systemic venous vessels rather than a convergence, through the heart and into either the arterial vessels leading to the ABO or the systemic circulation (Olson, 1994), because even with a gill shunt, there is still the problem of blood mixing during cardiac transit. This optimal organization of blood flow past the gills is known from several lungfish species but has also been functionally proven in the air-breathing teleost snakehead Channa argus. The lungfishes (Lepidosiren, Neoceratodus, Protopterus) have developed the most extensive modifications with various degrees of intracardiac septation and shunts approaching a parallel circulatory system not seen in other fish (Olson, 1994). The snakehead has extensive anatomical modifications inside the heart chambers and multiple ventral aortas that allow preferential diversion of oxygenated blood through degenerated gill arches (Ishimatsu and Itazawa, 1983). In another teleost, Monopterus albus, oxygenated and deoxygenated blood is shunted through branchial arteries that bypass the gills to the dorsal aorta (Iversen et al., 2013), although whether this represents an inefficiency in vascular layout or a functionally useful trait remains to be shown. Arapaima gigas does possess a possible shunt vessel given it has a vastly reduced 5th gill arch, which is devoid of lamellae; but again, any role this might play in avoiding O₂ loss remains an interesting avenue for future research.

The arrangement we therefore envisage that best fits our data is as follows. The CO₂-rich and O₂-poor blood of the venous return passes through the heart without fully mixing with the oxygenated but low-CO₂ blood from the ABO. This venous blood then loses CO₂ during branchial passage. In hypoxia, this can then occur with minimal oxygen loss because it is already de-oxygenated. After branchial passage, this blood is then transferred to the ABO where it is oxygenated, and where there is some minor further CO₂ excretion. In normoxia, this branchial passage will also allow for some oxygen uptake, but the greatly reduced A. gigas gills only then allow for 20% of the total oxygen uptake. The subsequent passage to the ABO allows for the remaining 80% and again here CO₂ is lost at the gills and therefore is not seen in expired gas from the ABO to any great degree. Conversely, the O₂-rich and decarbonated blood leaving the ABO passes the heart without effective mixing with the venous flow and must therefore then be shunted past the branchial basket for distribution to the body.

This proposed arrangement would seem utterly implausible were it not for the fact that such trans-cardiac flow separations and branchial shunts are already known in lungfish and the snakehead C. argus (Olson, 1994). Lungfish separate blood flow through the heart (Johansen et al., 1968; Graham, 1997d) with the aid of both ventrolateral and spiral folds in the bulbus arteriosus lumen (Scadeng et al., 2020) as well as incomplete septa and valves within the heart (Johansen et al., 1968; Graham, 1997d). Channa argus achieves intra-cardiac flow separation through a simpler arrangement with many ventricular pockets and longitudinal ridges in the bulbus (Ishimatsu and Itazawa, 1983), which resemble the amphibian heart. Based on the high O₂ saturation and P_O2 in the ventral aorta (80–90% and 45–50 mmHg, respectively; Johansen et al., 1978), Greenwood and Liem (1984) argued that mixing of systemic venous blood with the return from the ABO was probably small in A. gigas. The heart of A. gigas also possesses interesting morphological features that may promote flow separation within the heart, including a completely trabeculated ventricle (Buzete Gardinal et al., 2019; Scadeng et al., 2020) and a bulbus arteriosus lumen filled with elaborate tissue folds (Scadeng et al., 2020). The exact balance between diffusion and perfusion of oxygen and CO₂ and the degree of necessary flow separation is rather complex and a current focus through modelling in our laboratories. Nevertheless, the actual solution to produce the phenomenon measured in the present study would clearly not require complete separation of the ABO and systemic venous return. Thus, in addition to theoretical models of gas exchange, it would be of interest to measure blood gases at several sites and to describe the anatomy of the inflow to the heart as well as the arterial supply to the branchial basket.

We were able to fully automate the intermittent respirometry system allowing for unsupervised measurements of O₂ and CO₂ in both phases and sampling of chamber water for excreted nitrogen in the water. The long duration (17 h per treatment) of multiple uninterrupted measurements allowed for calculation of partitioning of gas exchange and fuel utilization.

As measurements were started 1–2 h after the fish had been moved to the respirometer, the overall shapes of the O₂ and CO₂ traces (Fig. 2) follow the expected trend of an exponential decline until plateau (Keys, 1930), usually seen with an initial phase with stress and the likely involvement of anaerobic metabolism. Handling has been shown to increase the O₂ consumption of juvenile salmon by 39–98%, while also increasing whole-body lactate concentration by 500% compared with controls (Davis and Schreck, 1997). In the present study, the initial Ṁ_O2,total was elevated by 89% above the unstressed plateau (Fig. 2). The elevation in CO₂ excretion was even more pronounced and disappeared more slowly, particularly in fish started in hypoxia. A high cytosolic ATP breakdown would have resulted in reduced pH titrating HCO₃⁻ and elevating blood P_CO2 and hence CO₂ excretion (Robergs et al., 2004), while a high production of lactate would aid in the elevated post-exercise gas exchange requirement, as it is known in teleosts that the lactate ion has a strong stimulatory effect on ventilation (Thomsen et al., 2017; Thomsen et al., 2019). This non-metabolically produced CO₂ can then elevate RER above 1, which was clearly evident in the entire first measurement period (Fig. 2). Robust evidence for elevated post-stress RER of this magnitude does not exist in the fish literature but is well documented in terrestrial tetrapods. For example, RER can reach 1.4 after exercise in cane toads (Rhinella marinus) and may reach 2.5 in the American alligator (Alligator mississippiensis) (Andersen and Wang, 2003; Hartzler et al., 2006).

In contrast to our initial hypothesis, we found no elevation of total Ṁ_O2 or SMR in severe hypoxia compared with normoxia. The cost of surfacing was probably very low in this study as surfacing behaviour is similar in normoxia and hypoxia, and the distance to the water surface is small.

O₂ uptake from water accounted for 20% of total O₂ uptake in normoxia. This O₂ partitioning agrees well with other studies of similarly sized A. gigas that have estimated this fraction to range from 21.9% to 25% (Table 2). Aerial O₂ uptake increased to 100% in severe hypoxia, which is expected based on earlier observations (Stevens and Holeton, 1978) and given that, in our experiment, ventral aortic blood P_O2 was almost certainly well in excess of water P_O2 in the water hypoxia treatment. The mass-specific rates of O₂ uptake and CO₂ excretion found in this study were considerably lower than most previous findings in A. gigas (Table 2), probably because of differences in fish mass between studies, and the use of intermittent flow respirometry instead of closed respirometry. Using the allometric scaling coefficient for Ṁ_O2 (0.776) and Ṁ_CO2 (0.843) from Pelster et al. (2020) to upscale their 0.687 kg fish to the mean mass (1.583 kg) of the current fish, our Ṁ_O2 and Ṁ_CO2 values are still, respectively, 27% and 32% lower than theirs. Both studies use a temperature of ∼28°C and the difference must therefore be attributed to the different methodology.

The SMR found in this study is at the low end of the range measured in other air-breathing fish species at similar temperature (Lefevre et al., 2014). The SMR of A. gigas is even low when considering its carnivorous sit and wait predation lifestyle and comparing it with that of other tropical, benthic and sluggish fish (Lefevre et al., 2014). Arapaima gigas shows a near-endothermic growth rate of around 10–15 kg year⁻¹ (de Oliveira et al., 2012), and it may well be that its low householding costs contribute to this.

Table 3 shows N-waste and fuel use parameters in A. gigas from the literature. Using the scaling coefficient from Pelster et al. (2020) for ammonia-N (0.724), urea-N (0.730) and total-N (0.718) to upscale their data for 0.687 kg A. gigas to 1.583 kg, our ammonia-N and total-N measurements, respectively, are 56% and 47% lower than expected while our urea-N is 7.3% higher. We judge these differences to be minor when considering the lower overall metabolic rate in this study as discussed above. Contrary to our expectations based on a previous report where A. gigas were exposed to moderate hypoxia (Wood et al., 2020), severe water hypoxia did not seem to affect total ammonia-N excretion rates (Fig. 4). Total urea-N excretion rates were also unaffected, which is in agreement with Wood et al. (2020). Similarly, no trends in metabolic fuel use were apparent as a result of changes in water oxygenation (Fig. 5B), in contrast to our initial hypothesis. As in previous studies on fasted individuals of this species, we found that nitrogen was predominantly excreted as ammonia-N. Urea-N contributed 12.1±3.3% to the total N-excretion, which is slightly more than in the previously studied smaller fish (0.02–0.7 kg) where the urea-N contribution ranged from 5% to 10%. However, this discrepancy was in large part due to a single individual with 25% urea-N, while the remaining fish averaged just 8.8±0.7% in line with previous data (Gonzalez et al., 2010; Pelster et al., 2020; Wood et al., 2020).

NQ, which correlates with the fractional contribution of protein oxidation to the entire aerobic metabolism, was 0.14, and well below the theoretical maximum of 0.27 (Lauff and Wood, 1996). Our values are close to earlier measurements on A. gigas that ranged from 0.16 to 0.22 (Gonzalez et al., 2010; Pelster et al., 2020). Additionally, an NQ of 0.16 and protein use of 58% can be estimated for Wood et al. (2020) if assuming an Ṁ_O2 as measured by Pelster et al. (2020) for A. gigas of similar age (3–4 months) and size (675 and 625 g, respectively), which is similar to the present study. For much smaller A. gigas (5 and 67 g), protein use was found to be higher at ∼80% (Gonzalez et al., 2010; Pelster et al., 2020).

This great reliance on protein to fuel aerobic metabolism in A. gigas during fasting is in line with values in satiation-fed juvenile rainbow trout (Oncorhynchus mykiss) (50–70% protein use) that rely much less on protein burning during fasting (15% protein use) (Alsop and Wood, 1997). However, this high protein use may simply reflect the diet. How fuel use changes after feeding has been studied in zebra fish (Danio rerio), where there was a post-prandial increased reliance on protein oxidation as digestion proceeded. The 48 h fasted zebra fish had a 26.8% protein fuel use while in fed zebra fish, protein fuel use progressively increased to about 52% at 7–10 h after feeding (Ferreira et al., 2019). This post-prandial increased reliance on protein burning is expected as dietary free amino acids in excess of anabolic needs must be oxidized if not converted to other fuel classes (Cowey, 1980). Therefore, it is possible that the high protein fuel use measured in A. gigas is a consequence of measurements being made relatively soon after feeding (≤3 days) (Gonzalez et al., 2010; Pelster et al., 2020). The nutritional status of our 24 h fasted A. gigas was probably still affected by the last meal consumed, as we sometimes detected tiny amounts of faeces in the respirometer. Feeding experiments followed by longer fasting periods might elucidate whether the current literature reflects truly fasted fish oxidizing their energy stores.

In conclusion, our two-phase intermittent closed respirometry in combination with nitrogenous waste sampling allows for complete respirometric measurements to assess fuel utilization by the stoichiometric approach. This method allowing multiple uninterrupted measurements makes it possible to separate activity from basal metabolism. We found no functional change in the partitioning of CO₂ excretion between water and air during exposure to severe hypoxia, with the aquatic fraction remaining around 80%. There was a small but significant increase in aerial CO₂ excretion, which we believe results from the increased ventilation of the ABO in aquatic hypoxia, where 100% of the O₂ uptake is aerial. We did not detect any sign of branchial O₂ loss in severe hypoxia, except in two very stressed fish soon after the fish was introduced to the respirometer. We argue that the absence of oxygen loss is probably a result of both the reduced gills in this species, which would especially limit O₂ compared with CO₂ as Krogh's diffusion constant (K) is 20–25 times higher for CO₂ (de Moraes et al., 2005), and also an unknown anatomical arrangement promoting diversion of oxygenated blood away from the branchial respiratory surfaces. We found no effect of hypoxia on total ammonia-N and urea-N excretion. This is in accordance with the finding of no change in aquatic CO₂ excretion. We found that metabolic fuel use was dominated by protein (53%), with near-equal contributions from carbohydrate (27%) and lipid (20%).

We thank Heidi M. Jensen, Erik Paima and Claus Wandborg for assistance in the animal facility. We also thank John S. Jensen and Jesper R. Voetmann for building our respirometer chambers and Kristian Beedholm for hardware construction and programming. Finally, we thank Per G. Henriksen for laboratory guidance.