Introduction

Increasing atmospheric CO2 levels are expected to cause a reduction of ocean surface water pH by 0.3–0.4 of a unit by the year 2100, a process commonly referred to as ocean acidification (OA)1. Ocean acidification projections are based on open ocean environments that are relatively stable over time1. In contrast, coastal and shallow water habitats can experience substantial natural fluctuations in pCO2 on a variety of temporal scales2, 3. These fluctuations are driven by a range of biological and physical processes4 and in some instances their magnitude can exceed mean CO2 levels projected to occur over the next century2, 3. Furthermore, natural pCO2 fluctuations are expected to increase in size throughout the century, as increased CO2 uptake by the oceans leads to reduced seawater buffering capacity5, 6. Consequently, as mean oceanic pCO2 levels rise, shallow water marine organisms will be exposed to higher pCO2 levels for longer periods of time in addition to experiencing a greater range of pCO2 levels.

Our current understanding of how natural pCO2 fluctuations will interact with rising mean oceanic pCO2 levels to affect the performance of shallow water marine organisms under future OA is limited. This is because most OA experiments have used stable pCO2 levels consistent with open ocean projections, instead of pCO2 levels naturally relevant to the study organism7, 8. While such experiments have demonstrated a range of impacts on traits across various taxa9,10,11, their ecological relevance is uncertain. Indeed, a handful of studies have shown that natural pCO2 fluctuations can significantly modify the biological responses of shallow water marine organisms to OA12,13,14,15,16,17,18,19. Consequently, there has been a call for experiments on shallow water marine organisms that include pCO2 treatments representative of their natural habitats7, 8, 20, 21. Results from such experiments will be vital for improving predictions of when the negative effects caused by elevated pCO2 will become evident in natural populations22.

Some of the most notable effects of stable, elevated pCO2 levels have been observed in coral reef fishes. Specifically, exposure to pCO2 levels between 700–1000 µatm have been shown to impair a range of sensory systems and alter ecologically important behaviours10, 23, 24. Alterations include impaired anti-predator responses25,26,27,28,29, loss of lateralization30, 31, loss of learning32, 33 and increased activity/boldness27. Such behavioural abnormalities are expected to have significant ecological consequences for fish populations. For example, as a consequence of exhibiting riskier behaviour, predation-related mortality was significantly higher when settlement stage damselfish were exposed to elevated pCO2 in the laboratory and released into their native habitat, inferring that recruitment and population sustainability will be threatened by projected future CO2 levels in the ocean27. Furthermore, the impacts that behavioural abnormalities have on predator-prey dynamics26, 34 and competitive interactions35 will likely cause shifts in community structure with unknown consequences for ecosystem functioning.

Coral reefs are highly dynamic shallow water habitats that experience diel cycles in pCO2. These daily CO2 cycles are driven by the processes of photosynthesis/respiration and calcification/dissolution over a day-night cycle, but are also influenced by physical controls such as water flow and residence time36,37,38. In shallow reef areas, diel variation in pCO2 can range anywhere from ±50 to 600 µatm around the mean38,39,40,41. Although the pCO2 of coral reef waters is not in perfect equilibrium with the atmosphere over a daily timescale, the carbonate system is still heavily influenced by flushing with offshore waters and thus the mean pCO2 of reef waters will rise in line with rising atmospheric CO2 36. To our knowledge, only three studies (all on calcifying corals) have explicitly considered diel pCO2 variation when investigating the potential impacts of OA on coral reef organisms. Importantly, they found that the negative impacts of OA on growth and calcification were buffered by the presence of a diel cycling pCO2 regime12, 13, 42. The behavioural alterations that have been observed in coral reef fishes are likely to be sensitive to the interactive effects of diel pCO2 cycles and rising mean pCO2 levels for two reasons. Firstly, previous work has shown that it takes between 24–96 h of exposure to stable elevated pCO2 levels for behavioural abnormalities to manifest, with shorter onset times at higher pCO2 levels27. Secondly, the negative effects of elevated pCO2 on behavioural responses are concentration-dependent27, 28, 31. Consequently, diel pCO2 cycles could reduce the severity of behavioural abnormalities, or prevent them from manifesting, by providing fish with a recovery period, especially if pCO2 levels drop below the onset threshold (600–700 µatm). Alternatively, experiencing higher maximum pCO2 levels daily may lead to more severe behavioural abnormalities.

To determine how diel pCO2 cycles affect the behavioural responses of coral reef fishes under OA, we reared juvenile damselfish, Acanthochromis polyacanthus (Bleeker, 1855), and clownfish, Amphiprion percula (Lacepède, 1802), under a series of stable and diel cycling pCO2 treatments in two different experiments. The aim of the first experiment was to determine if the magnitude of diel pCO2 cycles affects the behavioural performance of coral reef fishes under OA. The aim of the second experiment was to determine if the presence of diel pCO2 cycles affects the mean CO2 level at which behavioural abnormalities occur (i.e. the onset of behavioural abnormalities). Specifically, in experiment one, the behaviour of fish reared at two stable CO2 levels (480 and 1000 µatm) was compared with the behaviour of fish reared in two cycling CO2 treatments of different magnitude (1000 ± 300 and 1000 ± 500 µatm). Therefore, this experiment enabled us to test if the magnitude of diel pCO2 fluctuations affected the behaviour of fish under OA. In experiment two, the behaviour of fish reared at three stable CO2 levels (460, 750 and 1000 µatm) was compared with the behaviour of fish reared in diel cycling CO2 treatments at two different mean CO2 levels (750 ± 300 and 1000 ± 300 µatm). Therefore, this experiment enabled us to test if the effect of diel pCO2 cycles was dependent on the mean CO2 level experienced by the fish. In both experiments, we measured behavioural lateralization in A. polyacanthus and the response to a predator cue by Am. percula. These traits were chosen for each species as previous studies have demonstrated clear negative impacts of exposure to stable, elevated pCO2 conditions25, 27, 31, 43. It was predicted that diel pCO2 fluctuations could reduce the overall severity and delay the onset of behavioural abnormalities under OA conditions.

Results

Experiment one

Absolute lateralization (L A ) was significantly influenced by CO2 treatment (Fig. 1a, χ2 = 15.75, df = 3, P = 0.001). As expected, juveniles reared under stable, elevated pCO2 were less lateralized compared to those reared at control levels (P = 0.001). However, diel pCO2 cycles significantly increased how lateralized juvenile A. polyacanthus were at 1000 µatm. L A of juveniles reared under small fluctuations (±300 µatm) was intermediate, but not significantly different, to those reared at control and stable, elevated pCO2 (min. P = 0.214). L A of juveniles reared under large fluctuations (±500 µatm) was fully restored to control levels being significantly greater than those reared at stable, elevated pCO2 (P = 0.01). Mean relative lateralization (L R) in juvenile A. polyacanthus was unaffected by CO2 treatment (Fig. 1b, χ2 = 0.52, df = 3, P = 0.914). Furthermore, no group exhibited a preference for left or right turning (Fig. S1, max. χ2 = 0.84, P = 0.358). Juveniles reared under stable, elevated pCO2 tended to have a narrower L R distribution compared to the other treatments (Fig. S3), although these differences were not significant (Max. KS = 0.15, P = 0.510).

Figure 1
figure 1

Effects of stable vs diel cycling elevated pCO2 on behavioural responses in experiment one. (a) Absolute lateralization and (b) relative lateralization in juvenile Acanthochromis polyacanthus (n = 60 per treatment) were determined using a two-way T-maze. (c) Response to predator cue of juvenile Amphiprion percula (n = 16 per treatment) was determined using a two-choice flume. Different letters represent significant differences between treatments (Tukey, P < 0.05). Boxplots are sized according to the 25th and 75th quartiles, where the line identifies the median and the whiskers indicate the minimum and maximum values. +signs represent means.

Mean percentage time that juvenile Am. percula spent in predator cue water was significantly affected by CO2 treatment (Fig. 1c, χ2 = 51.45, df = 3, P < 0.001). As expected, juveniles reared at stable, elevated pCO2 spent a greater amount of time in predator cue water compared to those reared at control levels (P < 0.001). However, diel pCO2 cycles significantly reduced the amount of time that juvenile Am. percula spent in predator cue water at 1000 µatm. Juveniles reared under both small (±300 µatm) and large (±500 µatm) fluctuations demonstrated partial restoration of antipredator behaviour spending an amount of time in predator cue water which was intermediate, and significantly different, to juveniles reared at control and stable, elevated pCO2 (max. P < 0.001).

Experiment two

As was observed in experiment one, mean L A was significantly affected by CO2 treatment (Fig. 2a, χ2 = 75.25, df = 4, P < 0.001), with juveniles reared under stable, elevated pCO2 (750 and 1000 µatm) being less lateralized compared to those reared at control levels (max. P < 0.001). Diel pCO2 cycles did not affect how lateralized juvenile A. polyacanthus were at mean pCO2 level of 1000 µatm (P = 0.986). In contrast, diel pCO2 cycles fully restored lateralization in juveniles reared at a mean CO2 of 750 µatm, being similar to those reared at control levels (P = 0.710) and significantly greater than both the stable, elevated CO2 treatments (max. P < 0.001). Also, as observed in experiment one, mean L R in juvenile A. polyacanthus was unaffected by CO2 treatment (Fig, 2b, χ2 = 4.86, df = 4, P = 0.302), and no group exhibited a preference for left or right turning (Fig. S2, max. χ2 = 3.43, P = 0.064). However, there were more individuals which were less lateralized in the 750, 1000 and 1000 ± 300 µatm CO2 treatments (Fig. S2).

Figure 2
figure 2

Effects of stable vs diel cycling elevated pCO2 on behavioural responses in experiment two. (a) Absolute lateralization and (b) Relative lateralization in juvenile Acanthochromis polyacanthus (n = 60 per treatment) were determined using a two-way T-maze. (c) Response to predator cue of juvenile Amphiprion percula (n = 24 per treatment) was determined using a two-choice flume. Different letters represent significant differences between treatments (Tukey, P < 0.05). Boxplots are sized according to the 25th and 75th quartiles, where the line identifies the median and the whiskers indicate the minimum and maximum values. +signs represent means.

Similar to experiment one, CO2 treatment significantly affected the mean percentage time that juvenile Am. percula spent in predator cue water (Fig. 2c, χ2 = 15.95, df = 4, P = 0.003). As expected, juveniles reared at 1000 µatm CO2 spent a greater amount of time in predator cue water compared to those reared at control levels (P = 0.004). The percentage time juveniles reared at 750 µatm CO2 spent in predator cue water was intermediate, but not significantly different, to those reared at control and 1000 µatm pCO2 (min. P = 0.194). Diel pCO2 cycles influenced the predator cue response of juvenile Am. percula at mean pCO2 levels of 750 and 1000 µatm. Juveniles reared at 750 ± 300 µatm CO2 spent a percentage of time in predator cue water which was more similar to those reared at control levels compared to those reared at 750 µatm CO2. Finally, juveniles reared at 1000 ± 300 µatm CO2 demonstrated partial restoration of antipredator behaviour, with juveniles spending a percentage of time in predator cue water which was intermediate to those reared at 460 and 1000 µatm CO2 (min. P = 0.309).

Discussion

This study demonstrates for the first time that diel pCO2 cycles can significantly modify the behavioural responses of fishes under OA. The negative impacts of elevated CO2 on coral reef fish behaviour have been well documented and are expected to have significant ecological consequences for reef fish populations through effects on recruitment, predator-prey interactions, competition and habitat preference10, 23, 24. However, all studies to date have exposed fish to stable levels of elevated CO2, not considering the natural diel pCO2 cycles that occur on coral reefs. Here we show that the severity of two behavioural abnormalities commonly observed under elevated CO2 are reduced when fish experience a diel cycling pCO2 regime. The extent of reduction was influenced by both the magnitude of fluctuation and mean pCO2 level experienced, as well as the behavioural trait. Overall, our results indicate that previous studies have probably over-estimated the behavioural impacts of OA on coral reef fishes once they have settled to reef habitats where diel CO2 cycles are prevalent.

Previous research using stable pCO2 treatments has found that behavioural abnormalities start to manifest in coral reef fish between 600–700 µatm. Our results indicate that diel pCO2 cycles will delay the onset of behavioural abnormalities. In experiment two, we show that behavioural abnormalities present in fish reared at a stable level of 750 µatm CO2 were absent in fish reared at 750 ± 300 µatm CO2. However, in both experiments, although less severe, behavioural abnormalities were still present in the fluctuating 1000 µatm CO2 treatments. Thus, it appears that mean oceanic pCO2 levels closer to 1000 µatm will need to be reached before behavioural abnormalities could manifest in natural populations of reef fishes. Furthermore, we observed full restoration of behavioural lateralization in juvenile A. polyacanthus reared under 1000 ± 500 µatm CO2, inferring that some behavioural abnormalities may not manifest at all for populations living in habitats with large CO2 fluctuations, such as shallow reef flats and closed lagoons, even when average oceanic conditions reach 1000 µatm CO2. The observation that diel pCO2 variation can reduce and/or delay the onset of behavioural abnormalities in juvenile coral reef fish under OA is particularly important given the ecological consequences of behavioural abnormalities and past research that has shown a limited capacity for acclimation of behavioural traits to stable, elevated pCO2 31, 44. However, it is important to mention that behavioural abnormalities are still likely to occur in the pelagic larval phase of coral reef fish as they occupy a more stable CO2 environment in the open ocean. Consequently, population replenishment and sustainability of reef fish populations could still be threatened by near-future OA due to impaired behaviour in the larval phase27, 45, 46, even if behavioural effects are less severe in juveniles that have already settled to reef habitats. Finally, in experiment one, and to a lesser extent in experiment two, we observed more individual variation in predator cue responses of Am. percula at 1000 µatm CO2 if fish were reared under cycling conditions. This level of individual variation has previously been observed only at a mean stable CO2 of 700 µatm27. Thus, in addition to potentially providing more time for reef fish populations to adapt to future OA conditions, by delaying the onset of behavioural abnormalities, diel pCO2 cycles may also increase the adaptive potential of fish populations at higher CO2 levels by increasing the range of individual variation upon which selection can act.

The underlying mechanism of behavioural abnormalities in fish under OA conditions is linked to the effects of acid-base regulation on the function of type A γ-aminobutyric acid (GABAA) neurotransmitter receptors23, 43. GABAA receptors are gated ion channels with specific conductance for HCO3 and Cl. Under elevated pCO2 fish increase intracellular and extracellular HCO3 concentrations to prevent plasma and tissue acidosis47,48,49. In a recent study on A. polyacanthus this compensatory mechanism was shown to be sufficient to reverse the transmembrane gradients of HCO3 in brain tissue, which could interfere with GABAA receptor function and cause behavioural alterations50. For coral reef fish it appears that complete acid-base regulation in the brain under stable, elevated pCO2 levels may take between 24–96 h, as this is the exposure period required before behavioural abnormalities manifest27. Our results suggest that for fish reared under diel pCO2 cycles, exposure to lower CO2 levels for several hours each day is sufficient to prevent the physiological changes that would normally occur at a stable, high CO2. Extracellular and intracellular pH regulation take place at different rates, occurring more quickly in the former. For example, in gulf toadfish (Opsanus beta) exposed to 1900 μatm CO2, complete blood pH compensation was achieved after 2 h, whereas muscle intracellular pH was not adjusted until after 24h49, see also51. Thus, based on the onset times under stable, elevated pCO2 (24–96 h), it would appear that behavioural abnormalities do not manifest in coral reef fish until brain intracellular pH compensation is complete, although further testing is required. This could explain why diel pCO2 cycles alleviated the negative impacts of OA. Although no data is available, we assume coral reef fish would achieve pH compensation as fast, or faster, then toadfish in the example above, due to their higher metabolic rates and more active lifestyle. Therefore, we hypothesise that fish reared under diel pCO2 cycles were able to track disturbances in extracellular pH, but weren’t exposed to higher pCO2 levels long enough for full brain intracellular pH compensation to occur.

It has been suggested that behavioural abnormalities may also be influenced by alterations in gene expression related to ion regulation32, 52. Ion-regulation in blood and tissues is under circadian control in fishes53, 54. In a recent study on A. polyacanthus, variation in behavioural tolerance to stable, elevated pCO2 (754 µatm) was linked to the differential expression of genes related to circadian rhythm control55. For example, offspring of CO2 sensitive parents (i.e., those that exhibited behavioural abnormalities) upregulated the enzyme that catalyses the final reaction in the synthesis of melatonin, a key regulator of the circadian rhythm, which plays an important role in controlling ion-regulation56. This indicates that CO2 sensitive individuals might display more pronounced acid-base compensation if exposed to a sustained elevation of CO2 due to a stronger influence of circadian rhythm control, leading to larger changes of the neuronal ion gradients that determine GABAA receptor function. Our observations that diel pCO2 cycles can alleviate the negative behavioural effects of OA suggests that fish were displaying normal, or less, circadian control over acid-base regulation and thus did not respond so strongly to internal pH changes caused by elevated CO2 therefore avoiding altered brain ion gradients. Consequently, it appears that internal circadian rhythm control of acid-base regulation in coral reef fish is disrupted under stable, elevated pCO2, indicating that this process may be linked to the natural diel pCO2 cycles occurring in shallow reef habitats.

In this study we repeated the control, stable 1000 µatm CO2 and 1000 ± 300 µatm CO2 treatments in two different experiments. Although we observed similar responses to predator cue in Am. percula in both experiments, there were some differences in the effects of CO2 cycles on behavioural lateralization. In experiment one behavioural lateralization was partially restored in juvenile A. polyacanthus reared at 1000 ± 300 µatm CO2, whereas no restoration was observed in experiment two. The reason for the different results between experiments is unclear, but one possible reason is differences in the duration that the high CO2 peaks lasted. In experiment one the high peaks lasted approximately three hours, whereas in experiment two they lasted close to eight hours. Consequently, fish in experiment one had less time to adjust their acid-base status during the high peak, which may have resulted in them exhibiting less severe behavioural impairments. The reason we did not observe a similar difference between experiments for response to predator cue in juvenile Am. percula may be because the effect of elevated pCO2 on this trait was concentration dependent, as seen in experiment two. Due to logistical constraints, it was not possible to have duplicated experimental systems in experiment one. In contrast, experiment two had duplicate systems for each pCO2 treatment. As similar results were observed in each experiment we are confident that the pseudo-replication in experiment one did not affect the results. In general, the effects of stable, elevated pCO2 on lateralization and response to a predator cue observed in this study are consistent with previous work on the same species25, 27, 29, 31, with one exception. Previous studies have reported a clear attraction of Am. percula to a predator cue (>80% of time in predator cue water) at 1000 µatm CO2, whereas Am. percula in the current experiments exhibited neither attraction of avoidance of the predator cue (45–58% of time in predator cue water) at this CO2 level. The same observation was also reported in adult goldskinny wrasse, Ctenolabrus rupestris 57. Why fish in the current experiments exhibited a less dramatic change in antipredator behaviour at high CO2 compared with previous experiments is unknown, but could be related to some differences in protocol. In contrast to past studies that reset each fish to the starting position when the direction of the water sources was switched, fish were not disturbed during trials in this study. Another potential factor is the life-stage that was tested. Previous studies tested settlement stage larvae, whereas settled juveniles were used in this study. Age-specific responses to predator cues, as well as expression of odourant receptor genes, have been observed in other species of fish58, 59.

In this study, we show that a diel pCO2 cycle can substantially reduce the severity of behavioural abnormalities caused by elevated CO2 in coral reef fishes. In contrast, behavioural impairments were still present in a temperate shark species reared under elevated CO2 in a mesocosm that experienced diel CO2 variation, although there was no stable, elevated CO2 treatment to compare against60. A handful of other studies have also shown that daily pCO2 fluctuations can significantly modify the biological responses of shallow water marine organisms to OA12,13,14,15, 17, 19. This highlights the importance of considering natural pCO2 variability when trying to determine the response of shallow water marine organisms to OA. While our understanding of the magnitude and frequency of pCO2 fluctuations in situ is growing, many shallow water habitats remain under- or un-sampled8. Consequently, there is a need for more high resolution in situ studies that characterise natural CO2 variability both spatially and temporally. Such data will establish ecologically relevant pCO2 treatments to be used in laboratory experiments and allow us to better interpret results from past OA studies that have employed stable pCO2 levels7, 61. This will be critical for accurately assessing the likely effects of OA on shallow water marine organisms and which species and ecosystems may be at greatest risk.

Materials and Methods

Study species

Acanthochromis polyacanthus and Amphiprion percula are common throughout the Indo-Pacific region. Both species are demersal spawners, laying their eggs within small caves and crevices in the reef matrix. In A. polyacanthus, eggs hatch into small juveniles, with both parents providing care to the eggs and offspring for up to 45 d post-hatching62. In contrast, Am. percula has a relatively short larval phase of approximately 11 d before settling on the reef 63. Both species can be bred and reared in captivity with high success, which has led to their establishment as models for investigating the potential impacts of OA on coral reef fishes25, 27, 31, 50, 55.

Brood-stock and general rearing protocol

Adult A. polyacanthus were collected using hand nets from the Bramble Reef area (site 1: 18°22′S, 146°40′E; site 2: 18°25′S, 146°40′E) of the Great Barrier Reef in July 2015. Fish were transported to an environmentally controlled aquarium research facility at James Cook University (JCU) (Townsville, Australia) where they were housed as breeding pairs in 60 L aquaria at temperature conditions matching the collection location. An existing brood-stock of Am. percula at JCU was used. These pairs had been collected from the Cairns Region of the Great Barrier Reef and housed at JCU for four years. Adult A. polyacanthus and Am. percula pairs were maintained under stable, ambient pCO2 (~490 µatm). Temperatures were increased at a rate of 0.5 °C per week until the summer breeding temperature of 29 °C was reached in the first week of November 2015. Adult pairs were provided with half a terracotta pot to act as a shelter and spawning site. Aquaria were checked each morning for the presence of newly laid clutches. Pairs were fed ad libitum on commercial fish feed pellets (INVE Aquaculture Nutrition NRD 12/20) once daily outside the breeding season and twice daily during the breeding season (November–May).

Acanthochromis polyacanthus juveniles were fed a combination of freshly hatched Artemia naupli and weaning fish feed (INVE Aquaculture Nutrition Wean-S 200–400 µm) daily for the first four days post hatch (dph). 5–21 dph they were fed daily on the weaning feed and then switched to a small pellet fish feed (INVE Aquaculture Nutrition NRD 5/8) at 22 dph. Rearing of larval Am. percula was performed using methods described by Munday et al.28. Settled juveniles were fed daily on the weaning fish feed.

Experimental design

Experiment one was carried out at the aquarium research facility at JCU. For details on the experimental system refer to Supporting Information. Fish were reared at two stable (480 and 1000 µatm) and two cycling (1000 ± 300 and 1000 ± 500 µatm) CO2 treatments (Table 1 and Figure S3). The stable 1000 µatm pCO2 treatment represented the open ocean projection for the end of this century, typically used in many OA experiments11. The cycling pCO2 treatments matched levels that have been observed in some tidal lagoons38. Diel pCO2 fluctuations of between ±50–150 µatm are more typical in other reef areas40, 41, however, the magnitude of fluctuations seen in tidal lagoons today may become more common in other reef areas by the year 2100, as a amplification in diel pCO2 fluctuations is predicted to occur over this time period5. Mean values for seawater parameters in experiment one are presented in Table 1.

Table 1 Seawater parameters for experiment one. Values are means ± 1 SD for daily average, minimum, maximum and range of pHT and pCO2.

Experiment two was carried at the National Sea Simulator (SeaSim) facility at the Australian Institute of Marine Science (AIMS) (Cape Cleveland, Australia). Fish were reared at three stable (460, 750 and 1000 µatm) and two cycling (750 ± 300 and 1000 ± 300 µatm) CO2 treatments (Table 2 and Figure S4). For details on the experimental system refer to Supplementary Information. Previous experiments indicate that behavioural abnormalities are first evident at around 700 µatm CO2, although the magnitude of effect is often not as large as observed at higher CO2 levels27, 28, 31. Therefore, the inclusion of the 750 and 750 ± 300 µatm CO2 treatments enabled us to determine how diel pCO2 cycles may affect the onset threshold of behavioural abnormalities. Mean values for seawater parameters in experiment two are presented in Table 2.

Table 2 Seawater parameters for experiment two. Values are means ± 1 SD for daily average, minimum, maximum and range of pCO2.

A similar protocol was employed in both experiments. Three offspring clutches were used per species, each from a different parental pair. In experiment one, A. polyacanthus and Am. percula clutches were transferred to the experimental system and split between pCO2 treatments in duplicate tanks (12–15 A. polyacanthus per tank and 10 Am. percula per tank) at 1 and 12 dph respectively. In experiment two, offspring clutches were transferred to the experimental system and split between pCO2 treatments in duplicate tanks (one tank per line; 15 A. polyacanthus per tank and 13–15 Am. percula per tank) at 14 and 12 dph respectively. A. polyacanthus clutches were transferred at 14 dph in experiment two, compared with 1 dph in experiment one, due to logistical reasons.

Behavioural lateralization trials on A. polyacanthus were performed 40–42 dph, which equated to approximately six and four weeks of exposure to pCO2 treatments in experiments one and two respectively. Predator cue trials on Am. percula were performed 18–20 dph, which equated to approximately 1 week of exposure to pCO2 treatments in both experiments. All behavioural trials were performed between 09:00 and 17:00. Fish were gently transferred to the behavioural arenas using a glass beaker to minimise handling stress. Fish from each pCO2 treatment were tested at random times throughout the day to account for any possible time of day effects in the fluctuating treatments. Each fish was tested once, being placed in an isolation chamber within their experimental tank after a trial for the rest of the day. Research was carried out under approval of the James Cook University animal ethics committee (permit: A2210) and according to the University’s animal ethics guidelines.

Behavioural assays

Behavioural lateralization trials

Behavioural lateralization (i.e., favoring the left or right side during behavioural activities) is an expression of brain functional asymmetry and a strong determinant of fish behaviour. Lateralized individuals show higher performance in cognitive tasks64, schooling behaviour65 and escape reactivity66. Lateralization in juvenile A. polyacanthus was determined using a detour test in a two-way T-maze using methods similar to those described by31. The two-way T-maze consisted of an experimental arena (60 cm × 30 cm × 20 cm), with a runway in the middle (25 cm × 2 cm, length × width), and at both ends of the runway (2 cm ahead of the runway) an opaque barrier (12 cm × 12 cm × 1 cm) was positioned perpendicular to the runway. The maze was filled to a depth of 4 cm with the respective treatment water of the fish being tested, being changed after each trial. A single fish was placed at one end of the T-maze and given a 3 min habituation period, during which time it could explore the apparatus. At the end of the habituation period the fish was gently guided into the runway using a plastic rod with the observer standing directly behind the fish (the plastic rod was never placed closer than approximately twice the body length of the fish). At this point to minimise human interference affecting direction turned the observer slowly stepped back from the maze and the fish was allowed to swim to the end of the runway. In instances when a fish did not swim to the end, encouragement was provided by gently moving the plastic rod around at the beginning of the runway. Direction choice was recorded as the first direction turned when the fish exited the runway. Ten consecutive runs were recorded per fish. Twenty fish from each clutch (ten per tank) were tested per CO2 treatment. To account for any possible asymmetry in the maze, turns were recorded alternately on the two ends of the runway. Turning preference (i.e. bias in left or right turns) at the population level was assessed using the relative lateralization index (L R, from −100 to +100, indicating complete preference for left and right turning, respectively) according to the following formula: L R = [(Turn to the right − Turn to the left)/(Turn to the right + Turn to the left)] * 100. The strength of lateralization (irrespective of its direction) was also assessed at the individual-level using the absolute lateralization index L A (ranging from 0 (an individual that turned in equal proportion to the right and to the left) to 100 (an individual that turned right or left on all 10 trials)). Lateralization trials in experiment two were performed with the observer blinded to the experimental treatments.

Predator cue trials

The ability to detect and elicit appropriate antipredator behaviour is critical for survival, especially in early life-stages that experience a greater predation threat67. The response of juvenile Am. percula to a predator cue was tested in a two-channel choice flume using methods similar to those described by29. The flume combination was predator cue water versus untreated water. Water at the same pCO2 level from two different sources (9 L buckets) was gravity fed into the choice flume, which was divided down half of its length. A constant flow rate of 100 ml min−1 was maintained and monitored using a flow meter and dye test after every water change. Water was changed after each trial. Fish were tested under the mean pCO2 level of their respective treatments (i.e. fish reared under both 1000 and 1000 ± 300 µatm were tested at 1000 µatm), due to the logistical difficulties involved in manipulating predator cue water pH across a daily cycle. While this resulted in fish from cycling treatments experiencing a change in pCO2 between experimental and test water, recent work has shown this has no effect on the response of Am. percula to a predator cue at far greater changes than experienced in this study29. For each trial, a single test fish was placed in the centre of the downstream end of the choice flume and given a 2 min acclimation period. The position of the fish was then recorded every five seconds for a total of 2 min. A rest period of 4 min followed, during which time the water sources were switched to eliminate potential side preferences. The position of the fish was then once again recorded every five seconds for a total of 2 min. Fish were not disturbed during the trial. Temperatures during the trials were kept within 1 °C of the temperature in the rearing tanks. Eight fish from each clutch were tested per pCO2 treatment (4 per tank). Predator cues were obtained from three common coral-cod, Cephalopholis miniatus, as described by ref. 29. Response to predator cue was assessed as the percentage of time spent in the cue water. In experiment one, the control fish from one clutch exhibited no response to the predator cue (i.e. did not avoid the predator cue) and so this clutch was excluded from data analysis.

Statistical analyses

The effects of pCO2 treatment on absolute lateralization (L A), relative lateralization (L R) and percentage time spent in cue water were tested using mixed-effects logistic regressions68. Models for L A data from experiment one and predator cue data from experiments one and two were over dispersed and so were re-run using a penalised quasi-likelihood. In all models, parental pair and tank were included as random factors, with tank nested within parental pair. Pairwise comparisons were performed using Tukey’s post hoc tests. To determine if a treatment group demonstrated a turning direction preference Pearson’s Chi-square tests were used, where we expected a 50:50 ratio for left/right turning preference. Finally, differences in the relative frequency distribution of L R between treatments were tested using Kolmogorov-smirnov tests. Mixed-effects logistic regressions with and without penalised quasi-likelihood were conducted in R version 3.3.269 using the lme470 and MASS71 packages respectively. Pairwise comparisons were conducted using the multcomp72 package. Pearson’s Chi-square tests were performed using Minitab 17.

Data availability

The datasets generated during and analysed during the current study are available from the corresponding author on request or via the Tropical Research Data Hub (doi:10.4225/28/5923bfed71f8d).