Cardiac and behavioural responses to hypoxia and warming in free-swimming gilthead seabream, Sparus aurata

Cardiac performance is a core determinant of the ability of fish to survive and thrive in their environment, especially under challenging environmental conditions (Eliason and Anttila, 2017; Farrell and Smith, 2017; Stecyk, 2017). For instance, hypoxia, a reduced availability of dissolved oxygen, is a common stressor in aquatic habitats (Diaz and Rosenberg, 2008) that challenges the ability of the heart to ensure tissue oxygen supply (Randall, 1982; Taylor, 1992). Most fish are ectotherms, so increases in water temperature have direct thermodynamic effects on their metabolic rate and consequent oxygen demand, which the heart must be able to respond to (Cossins and Bowler, 1987; Rodgers, 2016; Schulte et al., 2011). Investigating how the fish heart responds to hypoxia and warming is of increasing relevance, because of the hypoxic episodes and summer heatwaves that are occurring in many aquatic ecosystems as a result of global change (Altieri and Diaz, 2019; Costa and Barletta, 2016; Eliason and Anttila, 2017; Stecyk, 2017; Stillman, 2019).

The primary cardiac response to progressive hypoxia in fish is a slowing of heart rate (f_H), known as hypoxic bradycardia (for detailed review, see Farrell, 2007; Stecyk, 2017; Taylor, 1992). Although bradycardia is a chemoreflex response, there is still debate about its actual functional significance for hypoxia tolerance (Farrell, 2007; Joyce et al., 2016; McKenzie et al., 2009; Stecyk, 2017). When progressively warmed, fish exhibit increased f_H, a tachycardia that may have multiple contributing mechanisms (Eliason and Anttila, 2017). It presumably serves to meet the increased oxygen demands caused by thermal acceleration of metabolism, such that the intrinsic capacity to raise f_H may be a determinant of a species' upper temperature tolerance (see Eliason and Anttila, 2017, for a detailed review). Although these cardiac responses to hypoxia and warming have been described in multiple species, this has almost exclusively been from acute experiments under controlled conditions with animals confined in some way and instrumented with wires connected to a measurement device (Eliason and Anttila, 2017; Stecyk, 2017). Very little is known about cardiac responses to hypoxia and temperature in free-swimming fish (Claireaux et al., 1995a,b; Lefrançois et al., 1998; Prystay et al., 2017).

Small biologging tags that record f_H from the electrocardiogram (ECG) are now available, which can be implanted into fish to measure their cardiac activity when they are recovered and swimming freely (Bjarnason et al., 2019; Brijs et al., 2018; Ekström et al., 2018; Prystay et al., 2017). It is not known whether cardiac responses to progressive hypoxia and warming will differ markedly in fish swimming freely compared with when they are confined, but responses in free-swimming animals should be a more reliable reflection of responses by wild animals in their natural environment (Claireaux et al., 1995a,b; Lefrançois et al., 1998). The tags can also log external tri-axial body acceleration (EA), such that it is possible to interpret cardiac responses of free-swimming animals against simultaneous measures of spontaneous behaviour (Clark et al., 2010).

We implanted biologgers into gilthead seabream (Sparus aurata) to compare cardiac responses to progressive hypoxia and warming when each animal was either shoaling in a tank with conspecifics or confined in an individual respirometer chamber. We expected that, although fish would exhibit hypoxic bradycardia and warming tachycardia when swimming freely in the tank (Claireaux et al., 1995a,b; Lefrançois et al., 1998), the ability to express spontaneous activity would alter responses compared with when confined in a respirometer. We expected that f_H of undisturbed animals would be higher in the tank in normoxia at acclimation temperatures, as a result of spontaneous swimming activity. Progressive hypoxia might cause progressive declines in f_H in the tank, which initially and prior to the chemoreflexive hypoxic bradycardia were due to reduced spontaneous swimming activity (Remen et al., 2015). In the respirometer, by contrast, f_H would be low and relatively stable until the onset of chemoreflexive bradycardia (Perry et al., 2009). We expected f_H to increase more rapidly in the tank with warming, because of the stimulation of activity (Claireaux et al., 1995a), with a maximum f_H reached at a lower temperature. We expected to demonstrate that swimming activity was a determinant of cardiac activity in the tank, by revealing a direct relationship between logged EA and f_H. Finally, the respirometers allowed us to investigate whether f_H was a predictor of metabolic rate in S. aurata during exposure to hypoxia and warming, as it is during aerobic exercise (Hachim et al., 2020).

Experimental procedures were approved by the ethics committee for animal experimentation no. 036 of the French Ministère de l'Enseignement Superieur, de la Recherche et de l'Innovation, with reference number APAFIS #20294.

Experiments were performed on n=12 seabream, Sparus aurata Linnaeus 1758, with a mass of approximately 500 g and age of approximately 18 months. Sex was not verified, but fish were most probably male, as S. aurata is a protandrous hermaphrodite species. The seabream were obtained from the Ferme Marine du Douhet (La Brée les Bains, France) as post-larvae then reared at the Ifremer Aquaculture Research Station in Palavas-les-Flots, in indoor cylindrical tanks (volume 2 m³) under seasonal photoperiods, provided with a flow of biofiltered and UV-treated seawater at 21°C. Fish were fed daily with commercial pellets (B-Grower Marin, Le Gouessant, www.legouessant.com) but fasted for 24 h prior to surgery.

Fish were anaesthetized by immersion in 0.1 g l⁻¹ benzocaine (benzocaine ethyl 4-aminobenzoate, VWR, www.vwr.com) in aerated seawater, until active ventilation ceased, then weighed and placed on an operating table with their gills irrigated with aerated seawater containing 0.05 g l⁻¹ benzocaine. Heart rate loggers (DST milli HRT-ACT, 13 mm×39.5 mm, 12 g, Star-Oddi, Iceland, https://www.star-oddi.com/) were implanted in the intraperitoneal space, via an ∼2 cm longitudinal incision along the ventral midline, ∼0.5 cm posterior to the pectoral fins. Loggers were advanced as close as possible to the pericardium and fixed with sutures (silk suture and non-absorbable monofilament) such that their ECG electrodes lay against the body wall, with the incision then closed with sutures (non-absorbable monofilament). Fish were left to recover in a 1 m³ cylindrical tank provided with a flow of aerated, biofiltered and UV-treated seawater at 21°C. The tank was isolated in a room dedicated exclusively to the study, with a natural photoperiod through skylights. The tank was shielded behind an opaque black plastic curtain. During recovery, fish were checked by visual observation through small holes in the curtain (McKenzie et al., 2007a,b) in the morning (08:30 h) and evening (17:00 h). Fish were not fed during recovery or subsequent experiments.

Fish were instrumented and studied in two groups of six individuals. To allow us to compare each individual's responses to the challenges when either swimming freely in the tank or confined in a respirometer, we used a protocol where they were sequentially exposed in one holding condition and then in the other.

At 90 h after surgery (Fig. 1), three of the six fish were netted from the 1 m³ tank and placed in individual rectangular clear plastic respirometer chambers (volume 9 l), which were submerged in a small raceway in the same room, shielded behind an opaque black curtain and provided with the same water as the tank. The other three fish were left in the tank without any handling. After 24 h recovery from this disturbance (hence at 5 days after surgery for all animals; Fig. 1), experiments were conducted over 5 days. Over the first 2 days, the fish were exposed to warming and hypoxia challenges, with overnight recovery between the stressors. After this, fish were placed into their reciprocal holding condition and allowed to recover for 24 h. Then, they were once again exposed to the two challenges over two successive days. The exact order in which each fish experienced warming and hypoxia, in the tank or in a respirometer, is provided in Table S1. Fish were observed throughout all exposure protocols, at all exposure levels, through small holes in the curtains (McKenzie et al., 2007b). Care was taken to reduce all disturbance to a minimum during experiments; therefore, experimenters entered at 08:20 h to set up the trials then gave fish 30 min to recover from any disturbance before commencement.

For progressive hypoxia, oxygen partial pressure in the tank and raceway was decreased simultaneously by bubbling with 100% nitrogen, from 100% (normoxia) to 80%, then in steps of 10% from 80% to 20%, then finally to 15%. Each step had a duration of 30 min, water oxygen levels were recorded using optical oxygen probes (Firesting sturdy dipping probes, Pyroscience, www.pyroscience.com) and oxygen meter (FireSting FSO2-4), with data displayed and recorded in Pyro Oxygen Logger software, with nitrogen flow and setpoints controlled manually. For acute warming, temperature was raised simultaneously in the tank and respirometers, in steps of 1°C every 30 min, from 21 to 31°C, using automated temperature control systems (AquaMedic T controller twin, www.aqua-medic.de) that reached incremental setpoints precisely by controlling activity of a submerged pump, in the tank or the raceway, that generated a flow of water through heat exchange coils immersed in a reservoir (1 m³) of tapwater held at 40°C. All fish were exposed to these levels of hypoxia and warming. The limit of 15% saturation in hypoxia was chosen because seabream of a similar size have been reported to be able to tolerate exposure to below 10% at 21°C (Remen et al., 2015). A limit of 31°C in warming is some degrees below the critical thermal maximum (CT_max) for S. aurata, which is at least 34°C (Kir, 2020; Madeira et al., 2014; Madeira et al., 2016). Thus, these exposures were sublethal but should nonetheless have engendered significant cardiac and behavioural responses (Claireaux et al., 1995a,b; Remen et al., 2015).

Rates of oxygen uptake (Ṁ_O2, mmol kg⁻¹ h⁻¹) were measured on the fish in the respirometers, using automated intermittent stopped-flow respirometry (Steffensen, 1989), over a 15 min cycle with a 6 min closure and 9 min flush period, providing two measures of Ṁ_O2 for each 30 min step of hypoxia or warming. Water oxygen concentrations were recorded continuously in each respirometer using Firesting sturdy dipping probes and meter, with data displayed and recorded in Pyro Oxygen Logger. Each fish's Ṁ_O2 was then calculated considering the rate of decline in oxygen concentration in the chamber, chamber volume and the mass of the fish (McKenzie et al., 1995, 2007a). Background measurements, on empty chambers, were made prior to placing the fish and at the end of each series. These were always negligible and so no corrections were applied.

The loggers were programmed with Mercury software (Star-Oddi) according to the manufacturer's instructions. Fish were all operated on in the morning, with tags activated at 07:00 h the following day. For f_H, the ECG data were sampled at 200 Hz for 4 s. During recovery from surgery and at night during experiments, f_H was measured once every 2 h. It was measured once every 5 min from 07:00 h to 18:00 h during the exposure trials. An ECG trace was saved with each measure of f_H, for visual confirmation of data quality. For EA, data were sampled at 10 Hz for 1 min, only during experiments: once every 5 min from 07:00 h to 16:00 h for the first series, and from 09:00 h to 17:00 h for the second series. Water temperature, date and time were also recorded with each f_H and EA measurement.

f_H was measured in beats min⁻¹, calculated by PatternFinder v.1.16.0 software (Star-Oddi) from R–R intervals in the QRST wave of the ECG (Altimiras et al., 1997). Each measure was confirmed by visual inspection of the ECG trace and manual calculation of R–R interval within PatternFinder. The variance of EA (VAR_m) was used to identify periods where variation in acceleration indicated bouts of activity or agitation. It was calculated as the variance of the 600 EA measurements per minute, which indicated when the sensor was measuring acceleration above 1g, in units of mg (1000 mg=1g), where EA=0 is equal to 1g and EA=1000 is equal to 2g. Each measure of f_H or VAR_m was associated with temperature, date and time recorded on the logger. Date and time information on the logger were used to establish the associated oxygen levels for the hypoxia trials, based on the oxygen probe recordings.

Statistics were performed with R version 3.5.3 (R Core Team 2019, http://www.R-project.org/) with P=0.05 taken as the fiducial level for statistical significance. To evaluate f_H across recovery days following surgery, data for each individual were averaged into 6 h bins (3 two-hourly measures), starting at tag activation until 84 h post-surgery, when fish were handled for experimentation. A one-way ANOVA with repeated measures was used to compare the sequential 6 h values, using the aov function from the stats package (http://www.R-project.org/). Holm–Bonferroni post hoc tests were used to identify where significant differences lay. For the 5 day experimental protocol, ‘undisturbed’ f_H at 21°C in normoxia was evaluated from a single measure at 07:00 h for each fish each morning. A two-way ANOVA was used to compare the undisturbed values with one factor being holding condition and the second being the sequential days of the protocol. The overall mean for each individual was compared between the tank and the respirometers by paired t-test. For these tests, normality, homoscedasticity and independence of residuals were verified visually.

Effects of progressive hypoxia or warming on f_H were evaluated and compared between holding conditions by two-way ANOVA with repeated measures using the aov_car function from the afex package (https://CRAN.R-project.org/package=afex) where one factor was holding condition and the other was either oxygen level or temperature. Because measures were repeated, both factors were treated as within-subject factors. Undisturbed values at 07:00 h prior to each trial were included in the ANOVA. As a two-way ANOVA with repeated measures does not tolerate missing data, 1.7% of the f_H measures were inputted using either the nearest neighbours or linear regression method, using the kNN function from the DMwR package (Torgo, 2010). Linear regression models were always significant and calculated f_H was always plausible. Normality of the data was verified with a Shapiro–Wilk test. Sphericity of the data was not met; therefore, a Greenhouse–Geisser correction was applied. Homoscedasticity of the residuals was verified visually by plotting them as a function of their fitted values. Holm–Bonferroni post hoc tests were used to identify where significant differences occurred. A paired t-test was also used to compare maximum and minimum f_H during trials and the oxygen partial pressure or temperatures at which these occurred, with normality, homoscedasticity and independence of the residuals verified visually.

As VAR_m data were not normal, effects of the stressors and comparison between holding conditions were evaluated with a generalized linear mixed model, with fixed factors being holding condition and either oxygen partial pressure or temperature, and individual as a random effect. Tukey post hoc tests were used to identify where any significant differences lay. Models were calculated with the function glmer from the lme4 package (https://CRAN.R-project.org/package=lme4) and post hoc analysis was performed with the emmeans package (https://CRAN.R-project.org/package=emmeans).

A linear relationship between VAR_m and f_H was established for all individuals using a generalized linear mixed model, and between f_H and Ṁ_O2 using a linear mixed model, with individuals as a random effect. Regression slopes between temperature and hypoxia trials were compared using the function lstrends from the lsmeans R package (https://CRAN.R-project.org/package=lsmeans). A linear relationship was also established between f_H and Ṁ_O2 for each individual, using a linear model. Homoscedasticity and independence of the residuals were verified visually.

A complete dataset was collected for n=10 seabream (four from the first series and six from the second), with a mean (±s.d.) mass of 534±86 g, ranging from 363 to 801 g.

Over 72 h of recovery from surgery with absolutely minimal disturbance (Fig. 1), all individuals showed a progressive decline in f_H (Fig. 1). The mean (±s.e.m.) f_H declined significantly until 60 h, after which it showed no further significant change (Fig. 1). At 90 h of recovery, corresponding to 15:00 h in the afternoon, fish were disturbed by netting and transfer of individuals to chambers, to start experiments. Over the ensuing 5 day protocol, the mean undisturbed f_H (i.e. the mean f_H at 07:00 h each day) was significantly lower in the tank than in the respirometers. This was irrespective of the number of days elapsed or individual exposure history during the protocol (Table 1, Figs 2 and 3; Fig. S1).

During progressive hypoxia, S. aurata displayed bradycardia in both the tank and respirometers (Fig. 2). Within each condition, f_H did not vary significantly from undisturbed normoxia down to 40%, but then f_H decreased significantly at 30%, 20% and 15%. There was, however, a significant difference in f_H of fish in the tank and the respirometers during hypoxia trials (P<0.01) and a significant interaction between holding condition and oxygen level (P<0.01). Thus, mean f_H was significantly lower in the tank than in the respirometers at all oxygen steps between 80% and 40%. Although undisturbed normoxic f_H differed significantly (Table 1, Fig. 2), this was not true of the measures taken in normoxia at the first step of the exposure trial, probably because of increased individual variation in f_H in the respirometers (Fig. 2). This was presumably because the fish had been slightly disturbed by the presence of the experimenters. Once bradycardia occurred, namely at 30%, 20% and 15%, there were no significant differences in f_H between the tank and the respirometers (Fig. 2). These different patterns of f_H during hypoxia, between the two holding conditions, were reflected in the fact that mean maximum f_H, whenever this might have occurred during hypoxia trials, was significantly lower (P<0.01) in the tank than in the respirometers (Table 1). In contrast, mean minimum f_H was similar and occurred at a similar very low oxygen saturation (Table 1).

The VAR_m of all individuals was generally low in hypoxia, with no significant differences between the tank and the respirometers at any level of hypoxia (Fig. S2). Visual inspection of the tank revealed that the seabream were moving slowly around the perimeter in hypoxia and tended to stop swimming entirely and rest on the bottom of the tank at hypoxic levels that caused bradycardia. Inspection of the respirometers showed no signs of agitation at any of the hypoxic levels.

Oxygen uptake in hypoxia showed a typical teleost response, where it was regulated at levels similar to normoxia until 40%, from where there was a significant and progressive decline in mean Ṁ_O2 (Fig. S3).

During progressive warming, S. aurata displayed tachycardia in both the respirometers and the tank (Fig. 3). Mean f_H was statistically similar between holding conditions at all temperatures, despite having been different when undisturbed at 21°C (Table 1, Fig. 3). Once again, f_H at the initial 21°C step of the exposure protocol was variable among individuals, presumably as a result of the mild disturbance. There was, however, a significant interaction between holding condition and temperature (P<0.01). Thus, f_H was similar between the undisturbed and 21°C steps, but then increased significantly from 21°C up until 27°C in the tank, but only increased from 21°C up until 23°C in respirometers (Fig. 3). Furthermore, the mean temperature at which maximum f_H occurred was significantly higher in the tank (30.3±0.34°C) than in the respirometers (28.6±0.68°C), being closer to the maximum temperature tested (31°C) in the tank (Table 1). One fish in a respirometer (fish no. 4; Table S1) showed an aberrant cardiac response to temperature, with a drastic drop in f_H at 28°C, which then remained low until 31°C (Fig. 3), which probably contributed to the interaction between holding condition and temperature. Visual inspection revealed that the fish was very dark coloured and striped at 28°C, and showed signs of losing equilibrium at the end of the trial. It was immediately removed and transferred to the tank. It then performed normally in subsequent trials, so we had no reason to exclude this animal from analyses.

During the warming trials, VAR_m was highly variable in the tank but generally showed a progressive increase (Fig. 4). It increased significantly from 21 to 23°C (P<0.01), the first warming step, and then showed various further significant changes, being highest overall at 31°C, the highest temperature (P<0.001) compared with 21°C (Fig. 4). As would be expected in confined fish, VAR_m was usually low in the respirometers and did not vary significantly with temperature (Fig. 4). As a result, mean VAR_m in the tank was significantly higher than in the respirometer at most temperatures (Fig. 4). Visual inspection of the tank showed that the fish were swimming actively around the perimeter at high temperatures, with occasional bursts of speed, especially at 31°C. Inspection of the respirometers only revealed signs of activity in one individual, starting at 26°C, which was reflected in a high VAR_m at 26 and 28°C (Fig. 4). Unfortunately, VAR_m data were not collected from this animal at higher temperatures because of an error in our programming of the Star-Oddi tag.

During warming, animals in respirometers generally showed an increase in Ṁ_O2 with warming that was significant at temperatures above 23°C (Fig. S4). The individual that showed the aberrant cardiac response also showed an aberrant metabolic response, with a decline in Ṁ_O2 from 28°C (Fig. S4).

There was a significant linear relationship between VAR_m and f_H in the tank during the warming trials (P<0.001), which was the only condition where animals showed significant activity (Fig. 5). There was no relationship of f_H to VAR_m under any other condition.

There was a significant positive linear relationship between f_H and Ṁ_O2 in the respirometers during both hypoxia [Ṁ_O2=f_H(0.034)+0.4; marginal R²=0.31; conditional R²=0.65; P<0.001] and warming trials [Ṁ_O2=f_H(0.02)+3.37; marginal R²=0.085; conditional R²=0.72; P<0.001]. There was no significant difference between these two slopes, so a single linear relationship between f_H and Ṁ_O2 was fitted for all hypoxia and warming values plotted together, which was highly significant (P<0.001; Fig. 6). Heart rate was also a predictor of metabolic rate for each individual fish (Table S2).

This study provides the first explicit demonstration of cardiac responses to progressive hypoxia and progressive warming in a free-swimming fish, and comparison with responses by the same individuals confined in a respirometer. The cardiac loggers have, however, already been used on several species (Arvén Norling, 2017; Brijs et al., 2018, 2019a,b; Davidsen et al., 2020; Prystay et al., 2017; Skeeles et al., 2020), including in sockeye salmon (Oncorhynchus nerka) migrating in the wild to demonstrate that f_H is significantly affected by water temperature (Prystay et al., 2017). A major unexpected result was that the f_H of undisturbed seabream was higher when they were confined than when they were swimming freely. As a consequence, cardiac responses to hypoxia and warming in the free-swimming animals did not follow our predictions, although they did exhibit some interesting differences when compared with those of the fish when confined.

The relatively high f_H after surgery presumably indicates an acute stress response, which may have included a release of circulating catecholamines (Reid et al., 1998; Gallo and Civinini, 2003) and/or removal of inhibitory cholinergic neural control (Randall, 1982; reviewed in Farrell, 1984). The progressive decline in f_H during the ensuing recovery presumably indicates an associated decline in stress and recovery of autonomic control (Campbell et al., 2004; McKenzie et al., 2007b; Sandblom and Axelsson, 2011; Taylor et al., 2010). Recovery of mean f_H by 60 h is somewhat faster than studies on salmonids. Brijs et al. (2019b) reported that a minimum of 3 days was necessary for free-swimming rainbow trout (Oncorhynchus mykiss) to recover a stable f_H after implantation of these loggers, while Føre et al., (2021) found that an average of 4 days was required in free-swimming Atlantic salmon (Salmo salar). The undisturbed mean f_H of the seabream following 60 h recovery, at around 60 beats min⁻¹ at 21°C, is amongst the lowest resting values reported for this species (Aissaoui et al., 2000, 2005, 1997; Hachim et al., 2020).

Throughout the 5 day exposure protocol, the seabream were necessarily subjected to daily disturbance from the presence of the experimenters, plus the exposures to hypoxia and warming. Nonetheless, when undisturbed in the morning, seabream consistently had higher f_H when in the respirometers, irrespective of whether they had been handled or exposed to stress on the previous day. The most obvious explanation would be a stress response to confinement, as stress is known to increase f_H in fish (Farrell, 1991; Lefrançois et al., 1998; Rabben and Furevik, 1993; Sopinka et al., 2016; Svendsen et al., 2021). The proximate mechanism for the high f_H in seabream confined in respirometer chambers requires further investigation. This implies, nonetheless, that allowing the seabream to shoal with conspecifics was less stressful than being confined alone. This finding indicates that confinement may introduce bias into studies of physiological responses by fish to environmental stressors, in a manner that may differ among species. Notably, it could bias measures of metabolic rate by static respirometry, given that f_H can be a predictor of Ṁ_O2, which is the case for S. aurata (Hachim et al., 2020). The experiments also revealed how sensitive seabream were to disturbance as, despite taking great care, our simple presence was enough to obscure differences in f_H between fish in the tank and the respirometers at trial commencement in normoxia at 21°C.

The data demonstrate that hypoxic bradycardia is observed in free-swimming fish but the cardiac response did not comprise a progressive decline linked to a progressive drop in spontaneous activity. Instead, it was very similar to ‘typical’ responses reported for many species under confined experimental conditions (Perry et al., 2009). Hypoxic bradycardia is a reflex response in teleosts, the sensory arm being chemoreceptor nerve cells in and around the gills that sense oxygen levels in ventilatory water and bloodstream and transmit this information to the brainstem. The reflex response occurs via cholinergic fibres in the cardiac branch of the vagus nerve, which slow the heart (reviewed by Farrell and Smith, 2017; Stecyk, 2017; Taylor, 1992). The functional significance of hypoxic bradycardia is still debated but it may protect the function of the cardiac pump, a purely aerobic organ, by conserving contractility and reducing myocardial energy requirements when oxygen supply in the blood is below a critical level (Farrell, 2007; Iversen et al., 2010; Joyce et al., 2016; McKenzie et al., 2009).

It is noteworthy that, although fish f_H was significantly lower in the tank than in the respirometers at oxygen levels above the threshold for hypoxic bradycardia, this threshold did not differ, being between 40% and 30% in both conditions. The higher f_H in seabream confined in a respirometer should, presumably, have been accompanied by a higher Ṁ_O2 than when swimming freely in the tank, given the direct relationship between these two variables. It might be expected, therefore, that the threshold for bradycardia would be higher in fish in the respirometer than in the tank. The fact that the threshold was the same and that, once bradycardia did occur, f_H was similar between fish in the tank and respirometer, requires further investigation.

The low VAR_m during progressive hypoxia, and absence of differences between fish in the tank and respirometers, presumably indicates that movements in the tank did not involve changes in speed, which are necessary to engender variation in acceleration (Hinch et al., 2002; Kawabe et al., 2003; Palstra et al., 2021; Tanaka et al., 2001). Thus, the gentle movements observed during hypoxia in the tank were clearly below the sensitivity of the accelerometer in the tag.

Although these tags have been used to study cardiac responses to acute warming in an anaesthetized fish (Skeeles et al., 2020), this is the first report of responses to progressive warming in a fully recovered free-swimming animal. As for hypoxia, cardiac responses were generally similar between the tank and respirometer, with a pronounced tachycardia in both cases. Warming tachycardia in fish presumably represents a response to increased oxygen demand when metabolism is accelerated by warming, as demonstrated by the linear relationship of f_H and Ṁ_O2 in the fish during warming in the respirometer. In terms of the heart itself, this response may reflect both direct effects of temperature on pacemaker function and modulation of autonomic control (see Eliason and Anttila, 2017, for a detailed review). In the seabream, the maximum f_H observed during warming, 205 beats min⁻¹ at a temperature of 31°C, was about double the maximum achieved during forced exercise in a swim-tunnel at 16°C in this species (Hachim et al., 2020). These high f_H in the seabream were all confirmed by visual inspection of the traces, with clear ECG waveforms.

It is interesting that, unlike in hypoxia, the accelerometer detected activity in the tank during warming, especially at the higher temperatures. The consistently higher VAR_m in the tank relative to the respirometer, for most warming steps, would explain why f_H did not differ significantly between the holding conditions. Thus, the VAR_m data confirm that, in free-swimming individuals, the tachycardia was not only due to the warming itself but also to behavioural responses to increasing temperature (Claireaux et al., 1995a). When free to express spontaneous behaviour, the seabream were extremely sensitive: the significant increase in VAR_m upon initial exposure to 23°C may have been because the fish were reared their whole post-larval life at 21°C. The fact that the mean temperature for maximum f_H was reached at a lower temperature in the respirometer than in the tank was contrary to our expectation, but entirely consistent with the fact that the confined fish already exhibited tachycardia when undisturbed at 21°C. The activity observed in the tank, especially the bursts of speed at high temperatures, may have reflected attempts to escape the conditions, although fish did not become agitated at the same temperature when confined in the respirometer. Overall, the behavioural responses to warming may provide useful indicators of tolerance that are much more sensitive than, for example, loss of equilibrium at CT_max (McDonnell and Chapman, 2015). In S. aurata, CT_max ranges from about 34.3°C to 36.6°C, depending on the acclimation temperature (Kir, 2020; Madeira et al., 2014, 2016).

The significant dependence of f_H on VAR_m during warming in the tank is further proof that spontaneous activity was responsible for raising the f_H of free-swimming fish. The relationship was, nonetheless, rather noisy with low predictive power. This may be because increases in VAR_m, especially at high temperatures, reflect agitation and burst swimming movements powered by fast-twitch glycolytic muscle (Webb, 1978). The metabolic costs of such movements are paid during recovery, rather than during the activity itself (Webb, 1978; Kieffer, 2000), so changes in f_H may have been out of phase with changes in VAR_m. Palstra et al. (2021) concluded that acceleration was most reliable as an index of unsteady burst swimming activity in the seabream; for example, when fish are feeding. Although measures of body acceleration have also been used to predict metabolic rate in fish (Bouyoucos et al., 2017; Gleiss et al., 2010; Metcalfe et al., 2016; Wilson et al., 2013; Wright et al., 2014), it seems unlikely this will ever have the same predictive power as f_H, not least because movement is only one component of metabolic activity in fish.

The data demonstrate that f_H is a predictor of metabolic rate in S. aurata during hypoxia and warming, as it is during aerobic swimming (Hachim et al., 2020). The fact that a single linear relationship between mean f_H and mean Ṁ_O2 could be described, irrespective of whether data derived from exposure to hypoxia or warming, demonstrates a tight coupling of cardiac pumping activity to metabolic oxygen demand under diverse environmental conditions in this species. The relationships for individual animals were highly significant, which Thorarensen et al. (1996) cite as a necessary condition to use f_H as a predictor for metabolic rate. However, their predictive power differed markedly among fish, with variation in f_H explaining less than 70% of variation in Ṁ_O2 in 6 of the 10 seabream. Also, the relationship may break down under multiple stressors and elevated workloads, or at high temperatures where f_H can become thermally limited (Thorarensen et al., 1996, Brijs et al., 2019a,b). For this reason, we did not perform the exercise of predicting individual Ṁ_O2 from their f_H when in the tank. Further research is required to establish the extent to which this variation among individuals is methodological, for example because f_H and Ṁ_O2 were measured over different time scales, or is physiological. Nonetheless, the results are promising in terms of calibrating the relationship of f_H to Ṁ_O2 using respirometry and then using logged f_H data to estimate patterns of energy use by free-swimming seabream (Clark et al., 2010; Cooke et al., 2016; Lucas, 1994; Treberg et al., 2016). The need to retrieve the tag is still a limitation on performing such studies on fish released into their natural environment (Prystay et al., 2017, 2019).

Our results demonstrate that hypoxic bradycardia and warming tachycardia are observed in fish whether they are free to shoal in a tank or confined in a respirometer. The fact, however, that confining S. aurata in a respirometer raised their f_H, presumably as a result of stress, and that f_H is a predictor of metabolic rate, has clear implications for estimating metabolic traits by static respirometry in some fish species. Tachycardia in free-swimming fish during warming was due, to some degree, to increased spontaneous activity. Thus, the combined measures of f_H and VAR_m in free-swimming fish provided novel insight into drivers of cardiac responses to temperature, and revealed highly sensitive behavioural responses to warming. Overall, the results demonstrate that biologging of physiological and behavioural responses to hypoxia and warming in free-swimming fish can provide more valid and reliable data than responses from confined fish, and has the potential to reveal sensitive sub-lethal thresholds for the impact of these stressors.

The authors are grateful to Marc Vandeputte and Ferme du Douhet for donating the fish. The authors are also grateful to Ásgeir Bjarnason of Star-Oddi Ltd for technical advice and assistance, and to Germain Salou and Aurélien Leddo of Ifremer Palavas-les-Flots for help in setting up the experiments. The authors also thank Tobias Wang for stimulating discussions about the experiments.