9–28 d of exposure to elevated pCO₂ reduces avoidance of predator odour but had no effect on behavioural lateralization or swimming activity in a temperate wrasse (Ctenolabrus rupestris)

Abstract

Most studies on the impact of near-future levels of carbon dioxide on fish behaviour report behavioural alterations, wherefore abnormal behaviour has been suggested to be a potential consequence of future ocean acidification and therefore a threat to ocean ecosystems. However, an increasing number of studies show tolerance of fish to increased levels of carbon dioxide. This variation among studies in susceptibility highlights the importance of continued investigation of the possible effects of elevated pCO₂. Here, we investigated the impacts of increased levels of carbon dioxide on behaviour using the goldsinny wrasse (Ctenolabrus rupestris), which is a common species in European coastal waters and widely used as cleaner fish to control sea lice infestation in commercial fish farming in Europe. The wrasses were exposed to control water conditions (370 μatm) or elevated pCO₂ (995 μatm) for 1 month, during which time behavioural trials were performed. We investigated the possible effects of CO₂ on behavioural lateralization, swimming activity, and prey and predator olfactory preferences, all behaviours where disturbances have previously been reported in other fish species after exposure to elevated CO₂. Interestingly, we failed to detect effects of carbon dioxide for most behaviours investigated, excluding predator olfactory cue avoidance, where control fish initially avoided predator cue while the high CO₂ group was indifferent. The present study therefore shows behavioural tolerance to increased levels of carbon dioxide in the goldsinny wrasse. We also highlight that individual fish can show disturbance in specific behaviours while being apparently unaffected by elevated pCO₂ in other behavioural tests. However, using experiments with exposure times measured in weeks to predict possible effects of long-term drivers, such as ocean acidification, has limitations, and the behavioural effects from elevated pCO₂ in this experiment cannot be viewed as proof that these fish would show the same reaction after decades of evolution.

chemoreception, choice flume, climate change, ecophysiology, predator avoidance, teleost

Introduction

Animals may respond to rapid human-induced environmental change by adjusting their behaviour, which could improve an individual's chance of surviving the early stages of altered environmental conditions (Tuomainen and Candolin, 2010; Wong and Candolin, 2015). If the environment undergoes a drastic change, as can be the case with anthropogenic disturbance, behaviours may become dysfunctional under the new environmental conditions (Tuomainen and Candolin, 2010; Wong and Candolin, 2015). However, behavioural and/or phenotypic plasticity may allow animals to adjust and thereby mitigate the effects of the stressor (Endler, 1992; Wong and Candolin, 2015). As anthropogenic disturbance may alter the environment dramatically and/or rapidly, as well as in a way that the organisms have not encountered during their evolutionary history, maladaptive behavioural responses are quite likely (Ghalambor et al., 2007; Tuomainen and Candolin, 2010; Wong and Candolin, 2015).

One potentially major anthropogenic disturbance is ocean acidification associated with global climate change, caused by increased levels of carbon dioxide dissolving in the sea (Doney et al., 2009; Hönisch et al., 2012). Exposure to atmospheric CO₂ concentrations predicted by the end of the century for weeks to months may alter the behaviour of fish (reviewed in Heuer and Grosell, 2014). Among the reported behavioural effects in fish are changes in behavioural lateralization, activity level, and altered preferences towards olfactory cues (reviewed in Heuer and Grosell, 2014). Many experiments on ocean acidification and fish behaviour have used exposure times measured in days (reviewed in Heuer and Grosell, 2014). However, using such short exposure times to predict possible effects ocean acidification, which is a long-term driver, has limitations, and the behavioural effects from elevated pCO₂ in those experiments cannot be viewed as proof that these fish would show the same reaction after decades of evolution.

Behavioural lateralization refers to the preferential use of, for example, the left or the right eye and stems from the asymmetric control of cognitive functions in the brain (Vallortigara and Rogers, 2005; Bisazza and Brown, 2011). If an individual is lateralized, that is, shows a preference for using either the left or the right eye or a preference for turning left or right, this has been suggested to confer advantages in the form of enabling multiple stimuli to be processed simultaneously (Vallortigara and Rogers, 2005; Bisazza and Brown, 2011). When examining behavioural lateralization, both the preference to turn either left or right, termed relative lateralization, as well as the strength of a possible side bias, regardless of this bias being to the left or the right, termed absolute lateralization, is commonly tested (Vallortigara and Rogers, 2005). The impact of CO₂ on behavioural lateralization has been investigated in several CO₂ studies, with somewhat differing results. Domenici et al. (2012), Nilsson et al. (2012), and Jutfelt et al. (2013) report that exposure to CO₂ reduced absolute lateralization while relative lateralization was unaffected. One study, in contrast, reports a change in relative lateralization, with fish exposed to CO₂ being left lateralized while control fish were right lateralized, without any alteration in absolute lateralization, as in Domenici et al. (2014). Another study reports that CO₂ exposure reduced absolute lateralization, and in addition, there was a change in relative lateralization, this time from a left preference in control fish to no side preference in the CO₂ exposed fish (Welch et al., 2014). However, there are also examples where no effect in either aspect of lateralized behaviour was found, such as in juvenile Atlantic cod (Jutfelt and Hedgärde, 2015).

Swimming activity is another behavioural measurement frequently used to investigate the effects of CO₂ exposure (reviewed in Heuer and Grosell, 2014). Many studies on fish report increased activity by CO₂ exposure (Munday et al., 2010, 2013; Cripps et al., 2011; Ferrari et al., 2011, 2012; Devine et al., 2012a; Forsgren et al., 2013). The effect sizes are generally large, with some studies reporting doubled the activity level (Cripps et al., 2011), and others up to 90 times more activity in CO₂ exposed fish compared with control fish (Munday et al., 2013). On the contrary, a number of studies found no alterations in activity (Munday et al., 2009; Nowicki et al., 2012; Bignami et al., 2013, 2014; Lönnstedt et al., 2013; Sundin et al., 2013; Jutfelt and Hedgärde, 2015; Maneja et al., 2015), and differences in activity levels in opposite directions between different species have been reported for fish occurring at natural CO₂ seeps, with some species showing an increased activity in CO₂ and others a decrease (Munday et al., 2014).

Olfaction is an important system for communication in aquatic environments as water can transfer olfactory cues over large distances, and chemical signals act as cues for a range of behaviours in fish (Magnhagen et al., 2008). The impact of elevated pCO₂ on olfactory discrimination has been extensively investigated, and most studies report large effect sizes, particularly on predatory cue avoidance behaviour. For example, Dixson et al. (2010) report a shift from 100% avoidance of the predator odour by control fish to a 100% preference of predator olfactory cues over seawater in CO₂ exposed fish. CO₂ exposed fish also showed no preference for predatory fish or non-predatory fish cue (Dixson et al., 2010). Similarly, other studies report that CO₂ exposed fish spent over 90% of the time in predator odour in a choice between predator cue and seawater (Munday et al., 2010, 2013; Nilsson et al., 2012), resulting in a 9-fold greater mortality rate in the wild (when placed on small reefs, made from live and dead coral, cleared of other fish and invertebrates) for CO₂ exposed fish compared with control (Munday et al., 2010). Additional studies report an impaired ability to learn to recognize predator odour, again resulting in lower survival in the wild (Chivers et al., 2014). Loss of antipredator responses for CO₂ exposed fish has also been reported for fish when exposed to conspecific chemical alarm cues (Lönnstedt et al., 2013). It has also been reported that there is no transgenerational acclimation of olfactory behavioural responses, since juvenile coral reef fish still spent 75–80% of their time in water containing chemical alarm cues, regardless of the parental CO₂ treatment (Munday et al., 2014). Hence, most studies found severe effects of CO₂ exposure on predator odour or conspecific chemical alarm cues avoidance behaviour, but differences between species exists, with some species, despite sharing the same ecology and life history, showing a less severe effect of CO₂ exposure on olfactory discrimination (Ferrari et al., 2011).

Despite the large number of studies reporting detrimental effects on fish behaviour by near-future CO₂ levels, a growing number of studies are failing to find effects of CO₂ on a range of behaviours, including activity and predator avoidance (Bignami et al., 2013, 2014; Jutfelt and Hedgärde, 2013, 2015; Lönnstedt et al., 2013; Maneja et al., 2013; Sundin et al., 2013; Murray et al., 2014; Näslund et al., 2015). Further, as many experiments on ocean acidification and fish behaviour use short-term high pCO₂ exposures (reviewed in Heuer and Grosell, 2014), whereas ocean acidification occurs over evolutionary time, predictions about the possible impacts of future ocean acidification are highly speculative.

As the bulk of literature is growing, great variation between species is emerging, leaving us unable to predict the phylogenetic extent of behavioural disturbances at this point. Further experiments on phylogenetically diverse species are therefore desperately needed.

In this study, we examined the possible behavioural effects of increased levels of carbon dioxide using the abundant temperate goldsinny wrasse (Ctenolabrus rupestris). Wrasses (Labridae) are widely used as cleaner fish to control sea lice infestation in commercial farming of Atlantic salmon (Salmo salar) in northern Europe (e.g. Bjordal, 1991; Treasurer, 1994). Controlling salmon lice infestations is one of the greatest challenges to the sustainability of large-scale Atlantic salmon farming (Costello, 2009). However, despite the great economical value associated with wrasses, little is known about their behavioural ecology and general population ecology (but see Sayer, 1999; Skiftesvik et al., 2014). We exposed goldsinny wrasse to elevated- and current-day levels of CO₂ for 1 month during which time we performed a series of behavioural tests, namely lateralization, activity, and prey and predator olfactory preference tests. Given the behavioural changes reported in the majority of CO₂ research to date (reviewed in Heuer and Grosell, 2014), we hypothesized that exposure to CO₂ would alter lateralization, result in increased swimming activity, lead to indifference towards the prey cue and finally to attraction to predator odour.

Material and methods

The experiment was performed at Sven Lovén Centre for Marine Sciences, Kristineberg, at the west coast of Sweden during May to June 2014. Goldsinny wrasses were collected using fish trap cages in shallow water in the Gullmar fjord, near the marine station (58°15′N 11°28′E). The fish were caught in two bouts, 10–12 May (individuals used in lateralization tests) and 2–3 June (individuals used in activity and olfactory choice tests). The goldsinny wrasse is one of the most abundant fish species in coastal habitats in Europe (Hillden, 1981), but has been used surprisingly infrequent in experimental biology, despite the growing interest from the aquaculture industry (e.g. Bjordal, 1991; Treasurer, 1994). It readily acclimatizes to laboratory conditions, showing curiosity and feeding behaviour soon after entry into aquaria, and therefore according to Spooner (1937) “proves an excellent experimental animal”. The total mortality over the duration of the experiment was 2 of 80 fish. Although the fish were collected at time of spawning (spawning period, May–June; Hillden, 1981), we were unable to sex the individuals, probably since only the smallest individuals caught were kept and those were most likely juveniles. In the laboratory, the fish were initially housed in flow-through holding aquaria (80 × 36 × 36 cm; length, width, height), brown algae (Fucus vesiculosus, F. Serratus, and Laminaria sccarina) collected at the site of capture were provided to all fish for shelter during the 2–3-d holding periods. Fish were fed shrimps and blue mussels (Mytilus edulis) daily. Water temperature and salinity followed natural conditions in the area during the holding periods [mean ± s.d., temperature 10–12 May: 10.8 ± 0.36°C; 2–3 June: 15.4 ± 0.34°C; salinity (PSU) 10–12 May: 24.2 ± 1.19; 2–3 June: 25.4 ± 0.28, data derived from the continuous monitoring of the flow-through system at the station (http://www.weather.loven.gu.se/kristineberg/en/data.shtml)]. The light cycle was set to L16 h: D8 h to mimic natural conditions.

CO₂ exposure

The CO₂ exposure was initiated on 13 May for the first batch of fish (N = 24 control, 24 CO₂, divided over 8 tanks) and on 3 June for the second batch (16 control, 16 CO₂, divided over 8 tanks). Fish were transferred to 38 × 36 × 35 cm aquaria where they were acclimated to the final CO₂ level. We gradually increased the CO₂ by gradually decreasing the set pH value on the pH stat computers, thereby slowly increasing the amount of CO₂ added to the header tanks (see below) over 48 h until the targeted concentration of 1000 μatm was reached. Half of the aquaria were randomly assigned to the CO₂ treatment, and the other half were kept as controls. The aquaria had a constant supply of flow-through seawater (rate of 2 l min⁻¹) from four header tanks (50 l; two per treatment). Each header tank had a flow of 5 l min⁻¹ of flow-through seawater from the fjord, pumped from 5-m depth. All header tanks were aerated. The target value of 1000 μatm for the CO₂ treatment header tanks was maintained using pH stat Computers (Aqua Medic, Bissendorf, Germany) connected to solenoid valves regulating the administration of 100% CO₂ gas (AGA, Sweden). The pCO₂ of the experimental tanks was measured daily using direct pCO₂ measurements with an infrared CO₂ probe (GM 70, Vaisala, Finland) connected to a submerged gas-permeable PFTE probe (Qubit Systems, Kingston, Canada; following Hari et al., 2008; Jutfelt and Hedgärde, 2013; Green and Jutfelt, 2014). The Vaisala CO₂ meter was factory calibrated (Vaisala) before the experiment and accuracy was validated weekly throughout the experiment by testing against a known gas mixture with 1010 μatm CO₂ in air (AGA), and no drift was detected. The water carbonate chemistry was calculated using the constants of Roy et al. (1993) and Dickson (1990) in CO2calc (Hansen, USGS, USA). Temperature and pCO₂ were measured daily, and alkalinity was measured weekly, data on salinity levels were derived from the continuous monitoring of the flow-through system at the station (Table 1).

Table 1.

Water chemistry data during the exposure of the goldsinny wrasse (C. rupestris), measured daily (pCO₂, salinity, temperature) and weekly (alkalinity) for a four header tanks system (Control A and B; High CO₂ A and B).

Treatment	pCO2 (μatm)	Temperature (°C)	Salinity	Alkalinity	pHtot (calc.)
Control A	374.5 (65)	15.0 (2.5)	26.2 (4.6)	2 074.7 (158)	8.11 (0.07)
Control B	375.0 (76)	15.0 (2.5)	26.2 (4.6)	2 016.2 (168)	8.15 (0.09)
High CO2 A	1 032.2 (127)	15.0 (2.5)	26.2 (4.6)	2 009.3 (178)	7.68 (0.02)
High CO2 B	969.8 (76)	15.0 (2.5)	26.2 (4.6)	2 072.1 (148)	7.69 (0.04)
Mean Control	374.8 (70)	15.0 (2.5)	26.2 (4.6)	2 045.5 (154)	8.13 (0.08)
Mean High CO2	994.5 (122)	15.0 (2.5)	26.2 (4.6)	2 036.2 (156)	7.68 (0.03)

The pCO₂, alkalinity, salinity, and temperature were measured, whereas the total pH was calculated with CO2calc (Hansen, USGS). The data are presented as the means with s.d. in parenthesis.

Experimental design

The behavioural tests were carried out under the same environmental conditions (temperature, salinity, pCO₂) as the fish had experienced during the exposure, except for the last lateralization test where all fish were tested in control water. All tests were performed during daytime. The first batch of fish was used in the lateralization test, and the second batch of fish was used in the activity and prey and predator olfactory preference tests. Fish were carefully hand-netted from their home tank and individually introduced into the experimental arenas. The fish were weighed (g) and measured (total length, TL, in mm) after the behavioural trials (mean ± s.d., Control: length: 94.2 ± 8.46, weight: 10.1 ± 2.37; High CO₂: length: 93.7 ± 8.25, weight: 9.2 ± 2.43), where after they were returned to their home tank. The fish were released approximately at the site of capture when all experiments were completed.

The experiment was performed according to current national legislation on Animal Welfare, Swedish Board of Agriculture, and approved by the Ethical Committee on Animal Experiments, Gothenburg, under licence Dnr 151-2011 and 103-2014.

Lateralization

For the behavioural lateralization tests, we used a double T-chamber [dimensions: 50 cm long with a 9 cm wide double T-channel (PlastKapTek, Partille, Sweden)] as described in Jutfelt et al. (2013;,Supplementary Figure S1). After being introduced to the experimental arena, the fish was gently encouraged (with a plastic rod) to swim to the central channel where after they swam towards the perpendicular channel and made a left or a right choice without further interference from the observer. Ten consecutive runs in alternating order of swimming direction in the double T-chamber was performed to account for any possible asymmetry in the set up (Bisazza et al., 1998). Turning choices were recorded by direct visual observation by one observer (JS). The tests were performed on two occasions, on exposure day 9 (N = 24 per treatment group) and exposure day 19 to test for effects of exposure duration (N = 23 control, 17 CO₂, one high CO₂ tank was excluded for technical reasons). All replicates were tested within the same day. A final lateralization test was performed on exposure day 21 (N = 18 control, 17 high CO₂), where both treatment groups were tested in control water to investigate if a possible CO₂ effect in lateralization would be maintained despite an acute change in pCO₂. It has been reported previously that behavioural impairment caused by exposure to elevated CO₂ lasts for several days and is not affected by testing fish in CO₂-treated water vs. control water (Dixson et al., 2010; Munday et al., 2010). During the lateralization tests, the treatments were alternated, and the water changed for every 6th fish.

We calculated the relative lateralization index (L_R, the turning direction bias) and the absolute lateralization index (L_A, the strength of the side bias) using the formulae: L_R = [(turns to the right − turns to the left)/(turns to the right + turns to the left)] × 100, with values ranging from −100 to +100 (positive values indicating a preference for turns to the right, negative values a preference for turns to the left), and L_A= |L_R|, with values ranging from 0 to 100 where higher values indicate a stronger side bias (Bisazza et al., 1997).

Activity

For the activity tests, we used four vertical cylindrical arenas (diameter: 20 cm, water depth: 10 cm), made of white opaque Plexiglas, with flow-through water [two with control water and two with high CO₂ water (1000 μatm), in randomized order], placed on an infrared (IR) light board (Figure 1). After being carefully introduced to the experimental arena, the fish were left undisturbed, in moderate ambient light, for 60 min during which time they were monitored using an IR sensitive firewire camera (Dragonfly 2, Pointgray, Richmond BC, Canada), mounted above the arenas. We used ViewPoint Zebralab (ViewPoint, Lyon, France), which is a commercial behaviour analysis system developed for zebrafish behaviour research. It automatically quantifies total activity (total swimming distance) and swimming duration (swimming was defined as all movements above 5 mm s⁻¹) in real time (Rihel et al., 2010). Swimming duration as well as swimming distance for each animal was averaged for every minute of the experiment. We performed eight runs of the experiment using two fish from each treatment in each run, adding up to in all 16 fish from the control treatment and 15 from the high CO₂ treatment (one fish became lethargic during the exposure period and was not included in the behavioural tests). The runs were performed on exposure days 22 and 23 (i.e. all replicates were tested within 2 d). After the experiment, the fish were weighed, measured, and returned to their home tank.

Figure 1.

Experimental setup of the activity test with the four vertical cylindrical arenas made of white opaque Plexiglas, where activity was measured. The two sides of the vertical divider were independently supplied with flow-through water, one with control water and one with high CO₂ water. The cylinders were raised slightly from the bottom to ensure water replacement inside the arenas. The tracks show the movements covered during 1 min.

Prey and predator olfactory preference

The olfactory preference tests were performed using the setup described in Atema et al. (2002), modified for larger fish as described in Jutfelt and Hedgärde (2013). The choice flume consisted of a two-choice flume channel (Choice Tank, Loligo Systems, Tjele, Denmark) containing a 32 by 40 cm choice arena with a water depth of 15 cm (Figure 2). The two water masses were continuously supplied, in a flow-through manner, by two 200 l header tanks, into which the olfactory cues were introduced. To create the prey cue, we placed one recently crushed blue mussel (M. edulis) into a net in one of the header tanks and exchanged the mussel every second run. Blue mussels are a known food source for goldsinny wrasse (Hillden, 1978). To create the predator cue, we placed one Atlantic cod (Gadus morhua; wild-caught, 1 kg), in one of the header tanks ∼1 h before the start of the experiment, where it remained throughout the runs for each day. Wrasses are important prey for Atlantic cod (Salvanes and Nordeide, 1993). A main header tank upstream from the two side tanks (described further in Figure 3 in Jutfelt and Hedgärde, 2013) with a CO₂ manipulation system was used to supply the two downstream header tanks, making it possible to use the pCO₂ that the fish were acclimated to on both sides of the flume simultaneously (Jutfelt and Hedgärde, 2013). The two downstream header tanks supplied water to the two sides of the flume at a flow rate of 17 l min⁻¹ each and 1 cm s⁻¹ water speed in the arena. The flume inlets had a crossover switch, which allowed quick change of sides of the olfactory cue with no or minimal disturbance to the fish.

Figure 2.

(a) Photo of the choice flume used for the olfactory cue tests from above, with the direction of water flow from the top of the picture and downwards. The flume has four layers of honeycomb structure (top) to create laminar flow, and the olfactory choice arena is the central rectangle, upstream from the mesh. Dye tests were regularly performed to ensure laminar water flow and minimal mixing of the two water flows; here, the left side receives water with red dye. (b) The central choice arena seen through the Viewpoint analysis software, with a wrasse centrally in the bottom of the picture. The two red boxes represent the two sides of the flume, one with and one without olfactory cue. The track shows the movement covered during 1 min.

Fish behaviour was video recorded by a Dragonfly II camera (Pointgrey, Richmond, Canada) positioned above the flume. After 10 min, the flows were switched (the olfactory cue switched side of the flume within 1 min as confirmed by dye test before the experiment). After the switch, the flow was maintained for another 10 min (i.e. total trial time 20 min). We performed the prey cue preference experiment on exposure day 22–24 (i.e. all replicates were tested within 3 d), alternating the control and high CO₂ treatments, N = 16 control and 15 high CO₂ (excluding one lethargic CO₂ fish). The predator cue avoidance experiment was performed on exposure day 27–28 (i.e. all replicates were tested within 2 d).

We used ViewPoint Zebralab to analyse the videos, analysing total duration for the left and the right side of the flume. For each run, we analysed 8 + 8 min, that is, excluding the first 2 min after trial start (regarded as acclimatization time) and the first 2 min after water switch.

Data analysis

We first tested for possible tank effects within each treatment for each behavioural test using nested ANOVAs. For the lateralization test, no such effects were found (all three runs, all tanks p > 0.1, Supplementary Table S1). However, for the activity tests (swimming duration and total distance), we found tank effects with decreased activity in one CO₂ tank (tank T3A, swimming duration: t = −2.12, p = 0.045; total distance: t = −2.27, p = 0.033, all remaining tanks p > 0.1, Supplementary Table S2), which was therefore excluded from further analysis (nested ANOVA excluding tank T3A, all remaining tanks p > 0.1, Supplementary Table S2). For the olfactory preference test, no tank effects were found (prey cue attraction and predator avoidance, all tanks p > 0.1, Supplementary Table S3). Statistical methods are described in brief, see Supplementary material for error distributions and link-functions.

For the lateralization experiment, we analysed test run 1 and 2 separately from test run 3, as the experimental setup differed between these runs, with all fish tested in control water in test run 3. For relative and absolute lateralization (used to allow for easier comparison to previous studies), test run 1 and 2, we performed Generalized Linear Models (GLMs) with appropriate error distribution and link-functions (Quinn and Keough, 2002), with the relative and absolute indexses as response variables and treatment, test run, and the interaction between them as fixed effects. For relative and absolute lateralization, test run 3, we used GLMs with treatment as fixed effect. Given the binomial error distribution, rather than Gaussian, of the detour test (choice of turning left or right), we also conducted two GLMs (with binomial error distribution), using the number of turns to the left as the response variable (corresponding to the relative index), and the maximum number of turns to the preferred side as the response variable (corresponding to the absolute index), both GLMs with the total number of left and right turns as the denominator, and treatment as the fixed factor.

For the activity test, we analysed proportion swimming duration (swimming duration out of total duration) and mean total distance using GLMs with treatment as fixed effect and fish length as covariate. To test whether there was any temporal change in activity across the testing period, we used a Generalized Linear Mixed Model, GLMM, with proportion inactive duration as the response variable, time, treatment, and the interaction between them as fixed effects, and the ID of each fish as random effect. All analyses on activity were performed excluding tank T3A, for which we found tank effects, including tank T3A, however, yielded similar results (Supplementary Table S4).

For the olfactory choice tests, we analysed the proportion of time spent in the odour cue using GLMs with treatment as the fixed effect and fish length as covariate. To test whether there was any temporal change in cue preference across the testing period, we used GLMMs with time spent in odour cue as the response variable, time, treatment, and the interaction between them as fixed effects, and the ID of each fish as random effect.

We used JMP 11 (SAS Institute Inc., Cary, NC, USA) for all statistical analyses, except for the GLMMs where GenStat 8 (VSN International Ltd, Hemel Hempstead, UK) was used.

Results

Lateralization

For the first two trials, performed day 9 and day 19, we found no effect of treatment in the relative lateralization index (non-significant effect of test run and non-significant interaction removed, Table 2, Figures 3 and 4). In the absolute lateralization index, the interaction between test run and treatment was significant (Table 2, Figures 3 and 4), showing that absolute lateralization was reversed between the treatments when tested the second time (with CO₂ exposed fish showing the strongest absolute lateralization on the first test run and Control fish showing the strongest absolute lateralization on the second test run, Figures 3 and 4).

Table 2.

The effect of treatment (high CO₂ and control) in the goldsinny wrasse (C. rupestris), for the behavioural tests lateralization, activity, and olfactory choice.

Test	Response variable	Explanatory variable	χ2	p-value
Lateralization	Relative lat, run 1 and 2	Treatment (CO2)	0.13	0.722
		Test run (1)	2.79	0.095
		Run (1) × Treatment (CO2)	0.15	0.701
	Relative lat, run 3	Treatment (CO2)	4.33	0.037*
	Absolute lat, run 1 and 2	Treatment (CO2)	0.63	0.428
		Test run (1)	5.17	0.023*
		Run (1) × Treatment (CO2)	4.81	0.028*
	Absolute lat, run 3	Treatment (CO2)	0.12	0.695
Activity	Swimming dur.	Treatment (CO2)	0.17	0.684
		Length	0.68	0.408
	Accl. time removed	Treatment (CO2)	0.17	0.680
		Length	0.97	0.324
	Total distance	Treatment (CO2)	2.05	0.152
		Length	0.20	0.652
	Accl. time removed	Treatment (CO2)	2.46	0.117
		Length	0.48	0.490
			Wald χ2	p
	Temporal change	Treatment (CO2)	0.01	0.940
	Swimming dur.	Time	37.23	<0.001***
		Time × Treatment (CO2)	0.07	0.796
			χ2	P
Olfactory choice
	% time in prey cue	Treatment (CO2)	0.00	0.989
		Length	0.69	0.407
	% time in predator cue	Treatment (CO2)	3.18	0.075
		Length	0.02	0.893
			Wald χ2	P
	Temporal change	Treatment (CO2)	0.00	0.947
	% time in prey cue	Time	0.19	0.659
		Time × Treatment	0.62	0.430
	Temporal change	Treatment (CO2)	2.70	0.100
	% time in predator cue	Time	6.09	0.009**
		Time × Treatment (CO2)	21.57	<0.001***

For the lateralization test, we included test run (1 and 2) and the interaction between test run and treatment as factors. Test run 3 was analysed separately due to differing experimental procedure. For the activity test, length was included as a covariate for all tests on swimming duration and total distance moved (with and without the removal of acclimation time). For swimming duration, we further analysed temporal change over time, with time and the interaction between time and treatment as factors. For the olfactory choice test, length was included as a covariate, analysing proportion of time spent in cue, as well as the temporal change over time of time spent in cue, with time and the interaction between time and treatment as factors. *p < 0.05, **p < 0.01, ***p < 0.001.

Figure 3.

Frequency distribution of relative lateralization [blue/black bars: control (370 μatm), orange/grey bars: high CO₂ (995 μatm)] in the goldsinny wrasse (C. rupestris), after (a) 9 d of exposure, (b) 19 d of exposure (all fish tested in their respective acclimation water), and (c) 21 d of exposure (all fish tested in control water). Individuals with a negative score were behaviourally left biased, and individuals with a positive score were right biased (relative lateralization index: L_R = [(turns to the right − turns to the left)/(turns to the right + turns to the left)] × 100).

Figure 4.

(a) Relative lateralization in the goldsinny wrasse (C. rupestris), showing the mean turning bias, positive scores indicate a right-turning bias, and negative scores indicate a left-turning bias. (b) Absolute lateralization, showing the strength of the bias, irrespective of side preference. The fish tested after exposed to control water (370 μatm; blue/black diamonds) or high CO₂ water (995 μatm; orange/grey squares) for 9 and 19 d. At 21 d, all fish were tested in control water (means ± s.e.m.).

For the last trial, where both treatments were tested in control water (experiment performed day 21), there was a significant effect of treatment on relative, but not on absolute, lateralization (Table 2). In this trial, the control fish had a mean L_R index [8.9 ± 10.51 (±s.e.)] that did not deviate from 0 (one-sample t-test against an expected mean of 0: t₁₈ = 0.85, p = 0.409), whereas the CO₂-exposed fish had the mean L_R index [−20.0 ± 8.91(±s.e.)] that did deviated from 0 (one-sample t-test against an expected mean of 0: t₁₇ = −2.24, p = 0.039) indicating a turning preference towards the left (Figures 3 and 4). The binomial GLMs showed similar results (Supplementary Table S5).

Activity

We found no effect of CO₂ on either swimming duration or total distance moved (Table 2).

For the analysis of temporal change in activity (inactive duration) across the testing period, there was again no effect of treatment while time had a significant effect, showing that the fish became less active during the testing period (Table 2, Figure 5). However, with a mean proportion of time spent swimming for the first minutes of 0.95, compared with a mean for the last 10 min of 0.91, the effect is small and likely due to habituation to the novel arena. Excluding the first 10 min of each run, to allow for habituation, yielded similar results in swimming duration and total distance moved (Table 2).

Figure 5.

(a) The proportion of time spent moving in the goldsinny wrasse (C. rupestris), after 22–23 d of exposure to control water (blue/black) or elevated CO₂ levels (orange/grey). The value for each minute is the mean from continuous measurements of activity (n_control = 16; $n_{\mathrm{C}{\mathrm{O}}_2}=11$ ) and the error bars show s.e.m. The two treatments are slightly offset on the x-axis to show the error bars. (b) Total distance moved each minute in the same experiment (mean ± s.e.m.).

Prey and predator olfactory preference

When testing the mean time spent in the cue for the total testing period, we found no effect of CO₂ in the choice of olfactory cue, neither on prey nor on predator cue (Table 2). For the analysis of temporal change in cue preference across the testing period, we found no effect of CO₂ treatment, time, or the interaction between treatment and time on time spent in the prey cue (Table 2). The wrasses actively avoided the prey cue, regardless of treatment and of what side the cue was presented on in the flume (Figure 6). For the predator cue, there was an interaction between time and treatment, showing that the control fish actively avoided the predator odour at the beginning of the experiment, whereas the CO₂ exposed fish showed indifference, and both control and CO₂ showed indifference towards the predator odour towards the end of the test run (Table 2, Figure 6).

Figure 6.

The mean % time the goldsinny wrasse (C. rupestris) spent in (a) prey odour and (b) in predator odour after 22–24 d of exposure to control water (blue/black) or elevated CO₂ levels (orange/grey). The value for each minute is the mean from continuous measurements of time spent on the side of the flume containing the odour (n_control = 16; $n_{\mathrm{C}{\mathrm{O}}_2}=15$⁠), with the first 2 min set as acclimation time (i.e. no data presented) and the first 2 min after the odour side was switched in the flume removed to ensure complete side switch of the odour (i.e. minute 11 and 12). The shaded area represents the time with mixed water after side switch, and the dashed line is the 50% mark representing no active side choice. Error bars show the s.e.m. The two treatments are slightly offset on the x-axis to show the error bars.

Because previous studies on olfactory choice in general have been conducted by noting the position of the fish in the flume in a certain time interval (every fifth or every tenth second, e.g. Dixson et al., 2010; Munday et al., 2010; Devine et al., 2012b; Munday et al., 2012; Nilsson et al., 2012; Jutfelt and Hedgärde, 2013) we, in addition to our high-precision automated flume analysis, used the same method here for the prey cue choice test, which allows cross validation of these two methods. We thus reanalysed the prey videos by noting what side the fish was on every 10th second and then analysed the proportion of times observed in the prey cue using similar statistics as described above. With this method, we obtained similar results to what was found when analysing the videos through ViewPoint (proportion of times observed in prey cue: effect of treatment: ${\chi }_{1,26}^2=0.01$⁠, p = 0.924, effect of length: ${\chi }_{1,26}^2=0.15$⁠, p = 0.699), suggesting that these two fish detection methods are comparable.

Discussion

In this study, on the goldsinny wrasse, we conducted a range of standard behavioural assays for behaviours in which adverse effects of CO₂ exposure have previously been reported in marine fish. We failed to detect any major effects of carbon dioxide in most behaviours investigated, except predator avoidance. This study thus adds to the growing literature where the effects of elevated CO₂ levels on behavioural responses were either minor or not observed (reviewed in Heuer and Grosell, 2014).

Prey and predator olfactory preference

The impact of CO₂ exposure on olfactory discrimination is one of the most well studied areas in the ocean acidification literature, with most studies reporting that CO₂ exposed fish show a preference for predator cue (reviewed in Heuer and Grosell, 2014). We found reduced predator avoidance in CO₂ exposed fish, but the avoidance did not turn into attraction as reported for many coral reef fish (reviewed in Heuer and Grosell, 2014). Instead the CO₂ exposed fish showed indifference to the cue, staying in predator odour 50% of the time. If the effect of CO₂ reducing predator olfactory cue detection is repeatable and still present after longer exposures, it could be detrimental to fish in the wild under high CO₂ conditions. Transgenerational acclimation of CO₂-induced behavioural disturbance has been suggested to be limited in some species (Welch et al., 2014), but not in others (Miller et al., 2012; Allan et al., 2014; Murray et al., 2014), hence it should be investigated if the effect could persist in fish during multiple generations and therefore may cause maladaptive behaviour in a future ocean.

The control fish avoided the predator odour for the first part of the test run but showed indifference during the last part, a behaviour that could be explained by initial predator avoidance followed by sensory or neural habituation to the cue. However, habituation to cues has been suggested to be slow or non-existent in fish (Vilhunen, 2006; Jutfelt and Hedgärde, 2013). To answer why the control fish only showed predator avoidance during the first part of the run, detailed follow-up experiments are required.

Although the effect of CO₂ exposure on predator avoidance behaviour has been investigated in many studies, the possible effect of CO₂ on prey cue attraction is a largely neglected area with a few exceptions showing reduced prey detection (Cripps et al., 2011) or no effect (Bignami et al., 2013, 2014). In the present experiment, we failed to detect any effect of CO₂ on prey cue attraction and both treatments avoiding crushed mussel cues equally, spending 36% of the time in the mussel cue. In general, food odours elicit a strong stimulus action on the feeding behaviour and search behaviour in fish (Kasumyan and Döving, 2003), and our hypothesis was that crushed mussels would attract the fish and that CO₂ would alter this attraction. The fish readily fed on mussels in their exposure aquaria. It is therefore possible that the behavioural response to crushed mussel cues is highly context-dependent and that the transfer from the school in the familiar exposure aquaria to being alone in the flume arena caused a shift from considering the cue food to considering the cue a disturbance. It is also possible that olfactory cues from food items are not enough to elicit a feed search response in the goldsinny wrasse and that visual cues are needed in addition. As the fish were food deprived for only 24 h before olfactory choice experiments, it is possible that their feeding motivation was low. Similar results have been reported previously, where the introduction of food odour did not cause a noticeable search for food in the common carp (Cyprinus carpio; Kasumyan et al., 2009). The prey cue avoidance behaviour, while surprising, was nonetheless strong and consistent between treatments, showing that the wrasses responded identically regardless of their pCO₂ exposure history as well as the pCO₂ in the flume during olfactory detection. This finding is in contrast to the strong effects of CO₂ exposure on many cues observed in several tropical species (reviewed in Briffa et al., 2012).

The different effects of the CO₂ exposure on prey cue and predator cue avoidance behaviour is interesting, possibly suggesting that CO₂ can affect the olfactory detection at the olfactory-receptor level rather than at the central nervous system level via the GABA_A receptor function as suggested previously (Nilsson et al., 2012). Elevated CO₂ has been suggested to chemically reduce the olfactory receptor response in European sea bass (Dicentrarchus labrax; Porteus et al., 2014), leading to higher detection thresholds for chemical cues and a reduction in maximum distance to the olfactory source wherefrom detection could occur by 48%. It is therefore possible that the observed discrepancy between control and CO₂ exposed fish in predator cue avoidance may be due to a higher detection threshold when CO₂ is present in the water. The potential for mitigation of this effect by acclimation or adaptation is unknown.

Activity

Activity is one of the behavioural traits where CO₂ exposure seems to elicit a broad range of responses. For fish occurring in naturally high CO₂, at CO₂ seeps, some species show an increase in activity while some species show a decrease, compared with control fish from otherwise comparable areas (Munday et al., 2014). In this study, we found no effect of CO₂ on either swimming duration or total distance moved, matching a number of studies failing to detect effects of CO₂ on activity (Munday et al., 2009; Nowicki et al., 2012; Bignami et al., 2013, 2014; Lönnstedt et al., 2013; Sundin et al., 2013; Jutfelt and Hedgärde, 2015; Maneja et al., 2015). On the other hand, many studies report increased activity when exposing fish to CO₂ (Munday et al., 2010, 2013; Cripps et al., 2011; Ferrari et al., 2011, 2012; Devine et al., 2012a; Forsgren et al., 2013). Thus, although some studies report great effects of CO₂ on activity, the results are not congruent, possibly suggesting that the response of increased levels of CO₂ varies between species and ecological settings.

Lateralization

When testing behavioural lateralization under the same environmental condition as the fish had experienced during the exposure (i.e. testing control fish in control water and CO₂ exposed fish in CO₂ water; test run 1 and 2), we found no difference between CO₂ and control fish in relative lateralization. However, for absolute lateralization, the significant interaction between test run and treatment indicated that the CO₂ exposed fish displayed the strongest side bias on the first test run, whereas the control fish showed the strongest bias on the second test run. It appears the control fish showed a similar strength in lateralization throughout the test runs (also in the third test run), displaying a mean absolute lateralization around 30 for each run, whereas the CO₂ exposed fish had a mean absolute lateralization of 40 in the first run and 16 in the second run. This could be interpreted as the control fish persistently showing the same strength of the side bias while the CO₂ exposed fish exhibited alterations in the strength of side bias as the CO₂ exposure progressed. This inconsistency of behavioural lateralization in CO₂ exposed fish suggests that it may be necessary to monitor lateralization repeatedly during a treatment to describe the full effect on lateralization and that a snapshot at one time point may not tell the whole story. Hence, forthcoming studies should focus on assessing the impact CO₂ may have on lateralization over time, investigate the repeatability of the lateralization pattern, and further determine the adaptive significance of the observed effect (if any) in context of the specific species ecology. Since the relative and/or absolute lateralization can change within individuals depending on a range of factors, such as time in captivity (Bisazza et al., 1997), sexual motivation (Bisazza et al., 1998), and parasite prevalence (Roche et al., 2013), this further suggests that laterality most likely is a context-dependent trait rather than fixed, which may be altered thorough individual experience (Bisazza et al., 2000; Brown, 2005; Bisazza and Brown, 2011).

When the lateralization experiment was performed in control water for both treatment groups (i.e. testing control fish and CO₂ exposed fish in control water; test run 3), there was an effect of treatment on relative lateralization, suggesting that an acute change in pCO₂ could elicit a treatment effect not present when testing the fish under the same CO₂ conditions as what they were exposed to. Previous studies on other species report that behavioural impairment caused by exposure to elevated CO₂ lasts for several days and is not affected by testing fish in CO₂ exposed fish in control water (Dixson et al., 2010; Munday et al., 2010). This seemed not to be true for the goldsinny wrasse. Therefore, to avoid possible artefacts from acute water changes, it could be advisable that behavioural assays take place in water with the same pCO₂ as the fish were exposed to. As fish (Atlantic cod and the Japanese sea catfish) can detect the acute water pCO₂ (Jutfelt and Hedgärde, 2013; Caprio et al., 2014), perhaps the water the fish were tested in could have immediate effects on behaviour through sensory pathways, which in the present experiment could explain why effects of CO₂ exposure on lateralization only appeared under an acute change in pCO₂.

Conclusions

We show that behavioural lateralization, activity, and olfactory preferences, all behaviours, where disadvantageous responses of fish exposed to near-future CO₂ levels have previously been reported, were largely unaffected by CO₂ in the goldsinny wrasse. The exception was predator smell avoidance behaviour, which was decreased by the CO₂ exposure. This study adds valuable balance in the reporting of the effects of CO₂ on fish, as the vast majority of studies on the impact of elevated pCO₂ report major detrimental behavioural responses (reviewed in Heuer and Grosell, 2014). In addition to the results found here, a growing number of studies report minor or no effects of CO₂ particularly studies using temperate species (Jutfelt and Hedgärde, 2013; Maneja et al., 2013, 2015; Sundin et al., 2013; Näslund et al., 2015), but also in some coral reef fish experiments (Ferrari et al., 2011; Nowicki et al., 2012; Lönnstedt et al., 2013). The interspecific variation in susceptibility demonstrates that we are far from able to predict possible future impacts of elevated pCO₂ on marine fish and ecosystems.

Authors’ contributions

JS and FJ designed the experiments, JS and FJ performed the experiments, JS analysed the data and wrote the manuscript draft. Both authors contributed to and approved the final manuscript.

Conflict of interest

The authors declare no conflicts of interest.

Acknowledgements

We thank Johan Rudin, Laura Vossen, and Arianna Cocco for field assistance. The Sven Lovén Centre for Marine Sciences, Kristineberg, provided excellent laboratory facilities. Thanks to Bengt Lundve for technical assistance. This work was funded by the Swedish Research Council Formas (2013-947 to JS and 2009-596 to FJ), the Swedish Research Council VR (621-2012-4679 to FJ), and by the Royal Swedish Academy of Sciences (FOA14SLC027 to JS).

Author notes

Handling editor: Howard Browman