Characterizing the hypoxic performance of a fish using a new metric: P_AAS-50

How an animal's metabolic rate interacts with its ambient environment has long fascinated biologists. For example, characterizing how ambient oxygen constrains whole-organism oxygen uptake (Ṁ_O2) has a long history (Warburg, 1914; Hyman, 1929; Tang, 1933), whether the interest lies with birds flying above the Tibetan Plateau (Black and Tenney, 1980; Lague et al., 2017), with mammals living at alpine regions (Terrados et al., 1988; Ivy and Scott, 2017), or with fishes exploiting oceanic oxygen minimum zones (Douglas et al., 1976) and oxygen-depleted lakes (Lefevre et al., 2014). While severely depleted levels of ambient oxygen ultimately constrain an animal's basal or standard metabolic rate (SMR), a less severe level of hypoxia constrains its maximum Ṁ_O2 (Ṁ_O2,max). For clarity, we define Ṁ_O2,max as the highest attainable Ṁ_O2 measured under normoxic conditions and use the term peak Ṁ_O2 (Ṁ_O2,peak) as the highest attainable Ṁ_O2 measured under all hypoxic conditions. Knowing the exact nature of hypoxia constraint has important physiological and ecological implications because Ṁ_O2,peak sets the ceiling for all aerobic activity above SMR. However, measuring Ṁ_O2,peak during progressive hypoxia to assess how hypoxia constrains peak aerobic performance is far more technically challenging than measuring routine Ṁ_O2 (Ṁ_O2,routine). As such, despite its obvious importance, relatively few studies have characterized how hypoxia constrains peak aerobic performance in fish.

In fish, the limiting effect of declining ambient oxygen levels on the maximum exercise-induced Ṁ_O2 (by critical swim or exhaustive chase protocols), which we collectively term hypoxic performance models, can take three general forms (Fig. 1A; Fry and Hart, 1948; Neill et al., 1994; Claireaux and Lagardère, 1999; Chabot and Claireaux, 2008; Zhang et al., 2021). For all these models, one anchor point is Ṁ_O2,max and the other is the critical ambient partial pressure of oxygen (P_crit, mmHg). SMR, Ṁ_O2,routine and Ṁ_O2,peak of a fish should all converge at P_crit (i.e. zero absolute aerobic scope, AAS; Fry and Hart, 1948). In between these two anchor points are three general models. The simplest is a linear model where Ṁ_O2,peak conforms with the partial pressure of oxygen (P_O2) (model I). A partial oxyregulation model (model II) is where Ṁ_O2,peak under modest hypoxic conditions is independent of ambient oxygen and identical to Ṁ_O2,max (Fig. 1A; Fig. S1), followed by a linear phase of oxyconforming. Model III has two linear phases of oxyconforming. Variations on models II and III occur when the Ṁ_O2,peak dependence on P_O2 is curvilinear (models IIa and IIIa) rather than linear. A curvilinear model requires more data points to adequately model the hypoxic performance curve relative to a linear model, where a curvilinear model sometimes uses a third value as a reference point. Indeed, curvilinear models could use the midway point between SMR and Ṁ_O2,max, the minimum P_O2 that would support 50% of AAS (P_AAS-50), as a reference point. Regardless of the model, Ṁ_O2,peak needs to be repeatedly measured over a range of P_O2 values. It is this challenge that we address here.

Improvements in the techniques and protocols used in aquatic respirometry have done much to pave the way for repeatedly exercising individual fish at multiple levels of hypoxia, which is a requirement for generating a hypoxic performance model. In particular, chasing a fish inside rather than outside a static respirometer allows Ṁ_O2 to be closely monitored during and immediately following exercise, ensuring that Ṁ_O2,peak is not missed (Zhang and Gilbert, 2017; Zhang et al., 2020). Analysis of the decline in P_O2 due to fish respiration has also improved. For example, Ṁ_O2,peak is estimated more accurately by minimizing the duration of the sampling window for each Ṁ_O2 determination and using an iterative algorithm to specifically identify Ṁ_O2,peak (Zhang et al., 2019; Prinzing et al., 2021).

In the present study, we had three objectives. The first was to determine the best statistical model to apply to Ṁ_O2,peak data. The second was to define the minimum number of P_O2 values needed for accurate modelling and prediction of P_AAS-50. While minimizing the number of P_O2 values reduces the number of separate exercise bouts per fish, the potential confounding effect of cumulative stress, which could reduce Ṁ_O2,peak, is decreased but not completely eliminated. Therefore, our third objective was to use a control group of fish to independently measure Ṁ_O2,max and Ṁ_O2,peak at only one level of hypoxia, near the P_AAS-50, to determine the degree to which cumulative stress may affect P_AAS-50. By satisfying these objectives, we arrived at a recommendation for a testing protocol that reliably quantified the hypoxic performance of individual rainbow trout (Oncorhynchus mykiss). This testing protocol advances our ability to capture inter-individual variation relative to an earlier methodology that characterized hypoxic performance in a group of European sea bass (Dicentrarchus labrax) by using different individuals at multiple levels of hypoxia (Claireaux and Lagardère, 1999). Moreover, we generated a reliable estimate of P_AAS-50, which has the potential to be a standardized metric to compare hypoxic performance within and across fish species.

The experiments were performed on a hatchery-reared, Fraser Valley strain of rainbow trout, Oncorhynchus mykiss (Walbaum 1792) (n=23 fish, body mass: 126.4±4.8 g) (Fraser Valley Trout Hatchery, Abbotsford, BC, Canada; Freshwater Fisheries Society of British Columbia). They were held in 200 l circular tanks containing dechlorinated Vancouver tap water in the Department of Zoology, University of British Columbia (UBC) Aquatics Facility for over 1.5 years prior to experimentation. Monitoring of water temperature (11°C for at least 3 months before experimentation started) and fish feeding (a maintenance ration, 1% body mass, of commercial trout pellets; Skretting Canada Inc., Vancouver, BC, Canada) were performed daily. Animal holding and experimental procedures were approved by The UBC Animal Care Committee (A18-0340).

The hypoxic performance trials were conducted on individual fish (fasted for 48 h) placed into one of four replicate 4.1 l Loligo-type, static respirometers (water volume to fish mass ratio ∼33:1; Loligo Systems, Tjele, Denmark). Fish were habituated to normoxic conditions with the respirometer in a water flush mode for at least 30 min before the first chase (see below) to ensure that fish at least repaid the high-energy phosphate and oxygen stores that might be partially consumed during handling (Zhang et al., 2018). The respirometers were submerged in a water table (2.4×0.8×0.4 m) filled with fully aerated freshwater, which accommodated all four respirometers and allowed simultaneous measurements on four individuals. A water temperature regulator (Ecoplus ¼ HP chiller, Vancouver, BC, USA) maintained the temperature of the water table at 11°C and circulated the contents of the water table. The entire respirometry system was located inside a thermally regulated (11°C) environmental chamber. The water volume of the entire system was replaced in between trials. Every 5 days, the entire respirometry system was disinfected with Virkon Aquatic (10 g l⁻¹; Syndel, Nanaimo, BC, Canada), rinsed thoroughly, and the water in the respirometers and water table replaced.

The water from the water table was intermittently circulated through each respirometer using individual, computer-controlled (AquaResp v.3, Aquaresp.com) flush pumps (Universal 3400, EHEIM, Deizisau, Germany). Flush pumps were stopped at the beginning of each 600 s Ṁ_O2 monitoring period (AquaResp v.3, Aquaresp.com), which was followed by 30 s flush and 40 s equilibration periods. Each respirometer also had a recirculation loop which contained a dedicated external pump (Universal 600, EHEIM) and an optical oxygen probe (Robust Oxygen Probe OXROB10) to continuously record the P_O2 (mmHg) of ambient water inside the respirometer (recording frequency ∼1 Hz, response time <15 s, PyroScience GmbH, Aachen, Germany). Oxygen probes were calibrated to 0 mmHg (0 kPa, water saturated with sodium sulphite and bubbled with nitrogen gas) and fully aerated water (157 mmHg=20.9 kPa). The background Ṁ_O2 in each respirometer was measured for 25 min after each trial had been completed. On the few occasions where background exceeded 1% of Ṁ_O2,routine, a correction was applied to the Ṁ_O2 measurement.

Each respirometer was equipped with a chasing device (a soft piece of 25 cm cable tie attached to a 20 cm metal stem bent at a right angle and inserted through a 1.5 cm diameter sealed port located mid-way along the top of the respirometer; Zhang et al., 2020), which when repeatedly turned from the outside would agitate the fish to perform C-start turns until it fatigued.

Modelling hypoxic performance above the P_crit (17.6 mmHg or 2.3 kPa; see below) required a preliminary trial (n=3 fish) to strategically select the ambient P_O2 test range to bracket the anticipated P_AAS-50, much like acute toxicity testing brackets median lethal concentrations. Based on these preliminary data, three testing protocols (four-point, five-point and seven-point P_O2 levels) were designed (n=7–8 fish for each protocol): the four-point model used P_O2 levels of 142, 96, 70 and 41 mmHg (18.9, 12.8, 9.3 and 5.5 kPa), the five-point model used P_O2 levels of 141, 103, 86, 72 and 40 mmHg (18.8, 13.7, 11.5, 9.6 and 5.3 kPa) and the 7-point model used P_O2 levels of 142, 110, 97, 84, 69, 57 and 41 mmHg (18.9, 14.7, 12.9, 11.2, 9.2, 7.6 and 5.5 kPa) (all values are the mean value for the mid-point of the ambient P_O2 while the respirometer was closed). Each testing protocol involved a stepwise deoxygenation of the respirometers with an Ṁ_O2,peak determination (described below) at each P_O2 level. Thus, the three testing protocols all started with an Ṁ_O2,peak at normoxia for each fish and then examined 3, 4 or 6 levels of hypoxia for that fish.

Ṁ_O2,peak was measured at each of these P_O2 levels by turning off the flush plumps and sealing the respirometer. Each fish was then individually exercised to fatigue by rapidly turning the chasing devise for 5 min, after which it was left undisturbed for 5 min to capture the initial portion of the excess post-exercise oxygen consumption (EPOC; Fig. 2), i.e. a closed monitoring cycle of 10 min. Subsequent off-line analysis captured the highest Ṁ_O2 values during this 10 min Ṁ_O2 monitoring period, and Ṁ_O2,peak was assigned to the highest of these values (see below and representative individual traces in Fig. S1), and the associated P_O2 was defined as peak P_O2 (P_O2,peak­; see calculation in ‘Data analysis’, ‘Ṁ_O2, Ṁ_O2,peak and SMR analysis’, below; and values in Figs 1 and 3). At the end of each 10 min Ṁ_O2 monitoring period, the respirometer flush pumps were turned on to introduce hypoxic water into the respirometer and gradually achieve one of the predetermined P_O2 levels outlined above using a 4 min cycle of a 30 s flush period, a 40 s equilibration period and a 170 s Ṁ_O2 monitoring period. Hypoxic water was generated using a custom-built, 10 l gas equilibration column situated upstream of the water table (it received the normoxic water at the top and nitrogen gas was injected at the bottom) and reduced the water P_O2 of the entire respirometry system. This exercise protocol was repeated at each P_O2 test level. Thus, each 10 min period when a fish was exercised and began recovery was followed by a 20 min recovery period during which Ṁ_O2 was monitored on-line (data not presented). This same protocol was followed for each P_O2 test level except the final one (39–42 mmHg or 5.2–5.6 kPa), which required a slightly longer degassing period (∼30 min) to reach the desired ambient P_O2. At the end of the trial, the fish were removed from their respirometers and returned to three holding tanks where the four-, five- and seven-point test groups were kept separated. After a 10-day recovery, the same fish from the five-point or seven-point test groups were re-tested to check the repeatability of the five-point and seven-point models. The four-point protocol was not repeated after it produced an inferior hypoxic performance curve (see Results). Additional details concerning the respirometry system and the testing protocols are available from figshare (https://doi.org/10.6084/m9.figshare.19658568.v3, table S1).

SMR, Ṁ_O2,peak in normoxia and Ṁ_O2,peak at P_AAS-50 were independently measured on a group of 8 naive fish using the same test apparatus, measurement protocols and chasing protocols as described above. However, once Ṁ_O2,peak was measured in normoxia, the fish were left undisturbed in the respirometer, which was continuously flushed with normoxic water for a 2 day quiescent period during which Ṁ_O2,routine was continuously measured and SMR was estimated from these measurements (see below). After this quiescent period, the water was gradually deoxygenated at a similar rate (total time ∼50 min) to near the P_AAS-50 estimated from the hypoxic performance curve (around 75 mmHg=9.9 kPa,), at which point another Ṁ_O2,peak determination was made as described above. These estimates of Ṁ_O2,peak in normoxia and at P_AAS-50 were statistically compared with those obtained with the hypoxic performance testing protocol to examine whether cumulative exercise in hypoxia affected these estimates. SMR was statistically compared with an independent measurement of SMR from another group of eight naive fish that were never chased. These additional fish were placed into the respirometers and left undisturbed under normoxic conditions for a similar 2 day quiescent period during which Ṁ_O2,routine was continuously measured to determine SMR. After this quiescent period, a well-established hypoxia challenge test (Claireaux et al., 2013) was performed to provide an estimate of P_crit (Claireaux and Chabot, 2016).

SMR was estimated from ∼288 Ṁ_O2 calculations using Eqn 2 collected during a 2 day quiescent period in the respirometer and applying a 20th quantile algorithm (Chabot et al., 2016, 2021). P_crit was estimated by fitting a linear regression function of ambient water P_O2 to the declining Ṁ_O2 (calculated using Eqn 2) during the final stages of the hypoxia challenge test. P_crit was assigned to the P_O2 by solving this regression for SMR (Claireaux and Chabot, 2016), i.e. the intersection of the regression line with SMR (Fig. 1).

The responses of Ṁ_O2,peak (y) as a function of ambient water P_O2 (x) were statistically modelled with both a linear regression equation (Eqn 3) and an asymptotic equation (Eqn 4) (Mueller and Seymour, 2011). The quality of these statistical fits for the four-point, five-point and seven-point datasets was assessed by least-squares regressions following Akaike information criterion (AIC) (for model parameters, see table S2 in figshare, https://doi.org/10.6084/m9.figshare.19658568.v3).

The repeatability of the five-point and seven-point hypoxic performance models was statistically tested by comparing the 1st and 2nd determinations using a non-linear mixed-effects model (Eqn 5) (lmerTest package in R v.4.1.1).

P_AAS-50 was estimated by normalizing individual Ṁ_O2,peak values to a percentage of their individual AAS (AAS=Ṁ_O2,max−SMR) and interpolating P_AAS-50 as the minimum P_O2 supporting the 50% of AAS from the best-fitted hypoxic performance model. The extrapolated SMR values were calculated based on P_crit using the best-fitted hypoxic performance model.

Measurements points were presented as mean±s.e.m. for Ṁ_O2,peak and P_O2,peak­, and 95% confidence interval (CI) was provided for the hypoxic performance equations. Logarithm transformations were applied on the metrics that failed normality tests to meet the assumptions of normality of residuals, homoscedasticity of the residuals and no trend in the explanatory variables. Statistical comparisons among different individuals, and for measured and extrapolated Ṁ_O2 values used one-way ANOVA with Holm–Šídák post hoc tests. The comparisons of the same individuals used paired t-tests. The statistical analyses were conducted in SPSS v.26 (SPSS Inc. Chicago, IL, USA). The best-fitting regression analyses were conducted using Prism v.9 (GraphPad Software, San Diego, CA, USA). Significance was assigned when α<0.05. Analysis of the respirometry data was performed in R v.4.1.1 software and R studio (https://www.rstudio.com/products/team/) using either the fishMO2 package (Claireaux and Chabot, 2016) or LabChart v.8.0 (ADInstruments, Colorado Springs, CO, USA).

Ṁ_O2,peak was typically reached during the chase (sometimes more than once), but rarely during the EPOC period (Fig. 2; Figs S1–S3). While Ṁ_O2 typically declined exponentially when P_O2 was above ∼100 mmHg, Ṁ_O2 remained quite constant during EPOC below ∼70 mmHg (Fig. 2). Thus, we confirm that chasing fish inside a respirometer provides a more reliable measurement of Ṁ_O2,peak than quickly transferring an exhausted fish into a respirometer and measuring Ṁ_O2 during the initial phase of EPOC (Zhang et al., 2020). Thus, a key criterion of measuring Ṁ_O2,peak is that a sustained workload on an individual is a prerequisite of a sustained peak performance (Midgley et al., 2007; Copp et al., 2009). In addition, we confirmed the normoxic anchor points by showing that an independently measured Ṁ_O2,peak at normoxia was indistinguishable from that determined with the hypoxic performance protocol (F=0.574, P=0.637, power=0.662; Fig. 1; Fig. S3).

We rejected a two-segmented linear regression model for the rainbow trout and compared a linear regression model with an asymptotic model for the best statistical fit. The former is often used to evaluate the dependence of Ṁ_O2,routine on P_O2 (Ultsch et al., 1980), while the latter models are used for hypoxic performance modelling (Mueller and Seymour, 2011). An asymptotic equation (curvilinear) modelled the hypoxic performance of rainbow trout no better than a linear regression equation based on AIC (AIC four-point: 287.8 versus 286.7, five-point: 378.0 versus 381.9, seven-point: 513.8 versus 517.7) and only marginally better based on R² (R² four-point: 0.900 versus 0.897, five-point: 0.936 versus 0.926; seven-point: 0.954 versus 0.949) (Fig. 1; see table S2 in figshare https://doi.org/10.6084/m9.figshare.19658568.v3). However, the curvilinear model was superior to the linear regression in terms of predicting the second anchor point of hypoxic performance, SMR; solving and extrapolating the linear models for the independently measured P_crit (17.6±0.9 mmHg, Fig. 1B,C) consistently overestimated SMR (76.0–90.9 mg O₂ h⁻¹ kg⁻¹) by a very large amount (74–143%) compared with the independently measured SMR value (SMR=43.9 mg O₂ h⁻¹ kg⁻¹; F=46.85, P<0.0001; Fig. 1). In contrast, extrapolation of the five- and seven-point asymptotic equations (35.1±3.8 and 43.4±3.8 mg O₂ h⁻¹ kg⁻¹, F=2.3, P=0.12, power=0.394; Fig. 1) reliably predicted the measured SMR. Thus, using either five or seven strategically selected P_O2 values with an asymptotic equation generated a reliable characterization of the hypoxic performance curve for rainbow trout, one that accurately predicted SMR using a known P_crit. However, we rejected a four-point curvilinear model because it also overestimated SMR by 50% (SMR=65.6 mg O₂ h⁻¹ kg⁻¹, F=38.11, P<0.001) despite a good statistical fit (R²=0.900; AIC=287.8).

We proposed P_AAS-50 as a potential third anchor point of a hypoxic performance curve. P_AAS-50 interpolations were statistically indistinguishable (t=0.10, P=0.92, power=0.051) for the five-point (first test 71.6±2.3 mmHg, second test 68.8±2.2 mmHg) and seven-point (first test 72.1±4.2 mmHg, second test 68.2±1.5 mmHg) curvilinear models. Also, the interpolated Ṁ_O2,peak values at P_AAS-50 were indistinguishable from the independent measurements (F=2.047, P=0.131, power=0.466; Fig. 1; Fig. S3). Consequently, five strategically selected P_O2 levels are likely the minimum number of data points needed to reliably model a hypoxic performance curve and predict P_AAS-50. Nevertheless, further studies are needed to understand whether a curvilinear or a linear model applies to the hypoxic performance of other fish species.

The reproducibility of the five-point and seven-point hypoxic performance curve protocols was confirmed by retesting the same individuals after a 10 day recovery period. The re-tested five-point hypoxic performance curve only manifested the slower rate of decline than the first test (t=2.49, P=0.015; see table S3 in figshare https://doi.org/10.6084/m9.figshare.19658568.v3; Fig. 3) whereas the re-tested seven-point hypoxic performance curve manifested a slower rate of decline (i.e. the lower tangent; t=2.49, P=0.015) and a lower asymptote (t=2.07, P=0.04) than the first test. Consequently, Ṁ_O2,peak at normoxia for the re-tested five-point hypoxic performance curve was only 9% lower than that for the first test (t=2.86, P=0.024; Fig. 3) and 12% lower with the re-tested seven-point hypoxic performance curve. Ṁ_O2,peak for the re-tested five-point hypoxic performance had an 8% lower Ṁ_O2,peak at 85 mmHg (t≥4.05, P≤0.007) than the first test. Despite the inconsequential difference in the Ṁ_O2,peak values, P_AAS-50 values for the re-tests were indistinguishable from those of the first tests (t=0.22, P=0.82, power=0.055) (Fig. 3).

The consistent values for Ṁ_O2,peak at P_AAS-50 by independent means suggest that the cumulative stress associated with repeat chasing to fatigue did not impact this variable. This finding implies that rainbow trout exercised to fatigue for 5 min and allowed to recover for 25 min can repeat their peak performance even though the fish only partially repaid their EPOC. Indeed, exhausted salmonids take about 30 min to replenish the majority of their oxygen stores and high-energy phosphate (Wood, 1991; Scarabello et al., 1991, 1992; Zhang et al., 2018) and salmonids can repeat a critical swimming test after just a 20–40 min recovery from the fatigue experienced in the previous test while carrying some unpaid EPOC (Randall et al., 1987; Jain et al., 1997, 1998; Farrell et al., 1998, 2001; Wagner et al., 2006; Steinhausen et al., 2008). Likewise, European sea bass (D. labrax) repeated four constant acceleration tests with only a 5 min rest period between tests (Marras et al., 2010). Whether or not other fish species can repeatedly perform as quickly remains to be determined.

The low coefficient of variation for the P_AAS-50 (<10% for both five-point determinations) suggests that it may be a robust and potentially valuable metric to compare hypoxic performance curves across fish species or within a species across different ambient environments. Also, it is tempting to suggest that the curvilinear nature of the rainbow trout hypoxic performance curve reflects the sigmoidal shape of the blood oxygen equilibrium curve (Weber et al., 1987). However, the P_AAS-50 was around 70 mmHg, which is approximately twice the P₅₀ of exercised rainbow trout haemoglobin (∼35 mmHg; Rummer and Brauner, 2015). Therefore, other physiological factors must affect the positioning of the curvilinear hypoxic performance of rainbow trout.

What is clear from our data, however, is that a linear model is inaccurate over a wide range of P_O2 values. A linear model with the anchor points of P_crit and the zero intercept for Ṁ_O2 and P_O2 has been used in a different context to describe hypoxic performance (Seibel et al., 2021). However, all our hypoxic performance curve models had non-zero intercepts (see table S2 in figshare https://doi.org/10.6084/m9.figshare.19658568.v3; Fig. 1B,C), suggesting that forcing the intercept through the origin lacks biological relevance, at least in trout. Moreover, such a linear model was suggested to be able to predict Ṁ_O2,max (Seibel et al., 2021), presumably at normoxia. However, this extrapolation for our measured P_crit in rainbow trout predicted a Ṁ_O2,max that was 13% lower (t=2.73, P=0.009; 357 mg O₂ h⁻¹ kg⁻¹) than our measured value. While such a difference can be considered small, larger differences might be expected when such extrapolations are made for fishes with a lower P_crit than rainbow trout. In particular, the linear extrapolation through the origin for the measured P_crit value in European sea bass (D. labrax) (Zhang, 2021) predicted a Ṁ_O2,max that was 47% higher than that measured (t=12.7, P<0.0001). Thus, any such linear models and extrapolations should be conducted with caution.

In conclusion, the hypoxic performance curve protocol provides a respiratory phenotyping platform to compare a fish's ability to exercise under hypoxia. We recommend the five-point hypoxic performance curve as a reliable, time-efficient and reproducible methodology to functionally quantify the hypoxic performance of individual fish and for interpolation of the P_AAS-50, which may prove to be a valuable comparative tool for hypoxic performance both within and across species. Zoologists can now measure the limiting effects of hypoxia in fish that was envisaged by Fry (1947) over 75 years ago.

The corresponding author particularly appreciates the numerous insightful discussions with Dr Guy Claireaux and Dr Denis Chabot over the years. We appreciate the logistical assistance from Dr Phillip Morrison and staff in the aquatic facility and workshop at the Department of Zoology, University of British Columbia. We appreciate the constructive feedback provided by two anonymous reviewers.