Heritability of behavioural tolerance to high CO 2 in a coral reef fish is masked by nonadaptive phenotypic plasticity

Abstract Previous studies have demonstrated limited potential for acclimation of adversely affected olfactory behaviours in reef fishes under elevated CO 2, indicating that genetic adaptation will be required to maintain behavioural performance in the future. Adaptation depends on the presence of heritable phenotypic variation in the trait, which may differ between populations and environments. We used parent–offspring regressions to estimate the heritability (h 2) of variation in behavioural tolerance to high CO 2 (754 μatm) in both field‐collected and laboratory‐reared families of Acanthochromis polyacanthus. Tolerance to elevated CO 2 was measured by determining the behavioural response of individuals to chemical alarm cues. Both populations exhibited high heritability of olfactory behaviour phenotype (father–mid‐offspring h 2 = 0.56 & 0.65, respectively) when offspring were acutely exposed to high CO 2 for 4 days. However, there was no heritability in the behavioural phenotype when juveniles were chronically exposed to high CO 2 for 6 weeks in the laboratory‐reared families. Parental exposure to high CO 2 during the breeding season did not alter this relationship between heritability and length of juvenile exposure to high CO 2. These results demonstrate that variation in behavioural tolerance to high CO 2 is heritable, but adaptive potential may be constrained by a loss of phenotypic variation when juveniles permanently experience a high‐CO 2 environment, as will occur with rising CO 2 levels in the ocean.

with other environmental stressors that could limit adaptive potential (Foo, Dworjanyn, Khatkar, Poore, & Byrne, 2014). Assessing the potential for adaptation in a broader range of marine taxa and phenotypic traits affected by ocean acidification is therefore a priority.
Recent studies show that CO 2 levels predicted for the end of the century can have adverse effects on the behaviour of marine fishes (Briffa, de la Haye, & Munday, 2012;Clements & Hunt, 2014;Heuer & Grosell, 2014) with consequences for key ecological processes such as larval dispersal, habitat selection, competition and predator-prey interactions (Nagelkerken & Munday, 2016). Many reef fishes innately recognize ecologically relevant olfactory cues, such as predator odour and chemical alarm cues (CAC) from injured conspecifics, and they use these cues to avoid danger (Dixson, Pratchett, & Munday, 2012;Holmes & McCormick, 2010). Under elevated CO 2 , however, reef fish lose the ability to appropriately interpret these cues (e.g., Chivers et al., 2014;Dixson, Munday, & Jones, 2010;Ferrari, Manassa, et al., 2012;Nilsson et al., 2012). Altered responses to predation threats may have significant population-level effects due to increases in juvenile mortality, which can affect population replenishment (Chivers et al., 2014;Munday et al., 2010). Altered olfactory responses persist for weeks to months in elevated CO 2 conditions (Munday, Cheal, Dixson, Rummer, & Fabricius, 2014;Munday et al., 2013), and transgenerational experiments further demonstrate that impaired behaviours are not ameliorated when parents are held under the same elevated CO 2 levels as their offspring (Welch, Watson, Welsh, McCormick, & Munday, 2014). Nevertheless, previous studies have observed individual variation in olfactory tolerance to elevated CO 2 (Welch et al., 2014), especially at near-future CO 2 levels around 700 μatm Munday et al., 2010). Furthermore, selection for CO 2 -tolerant behavioural phenotypes has been observed in field-based experiments . This phenotypic variation and natural selection could be key for future adaptation; however, it is unknown whether variation in behavioural tolerance to high CO 2 is heritable in coral reef fishes.
A common observation from evolutionary studies is that heritability is not constant and can vary with environmental conditions (Hoffmann & Merilä, 1999). This potentially complicates attempts to assess adaptive potential to climate change and ocean acidification because adaptation could be less likely under some environmental conditions than others. Many different hypotheses have been proposed to explain why heritability can vary among environments (reviewed by Hoffmann & Merilä, 1999); however, one hypothesis relating to assessing adaptive potential to rapid environmental change is that phenotypic variation may be either increased or decreased by environmental stress (Hoffmann & Hercus, 2000). Broad-and narrow-sense heritability have been observed to decline in unfavourable conditions in a variety of animals (Charmantier & Garant, 2005;Wilson et al., 2006), suggesting that heritability may also be lower in populations exposed to chronic stress from climate change. On the other hand, heritability may also increase when a stressful environment is encountered (Hoffmann & Hercus, 2000;Hoffmann & Parsons, 1991). Heritability may be higher in a stressful (or heterogeneous) environment if increased stress causes greater expression of phenotypic variation compared with less stressful (or homogenous) environments. Therefore, estimates of heritability may differ in populations exposed to an acute environmental stress compared with chronic environmental change. The effects of acute versus chronic environmental stress are especially relevant to experiments testing adaptive potential to ocean acidification because such experiments often expose juveniles or adults to high CO 2 for just a few days or weeks (i.e., acutely), whereas ocean acidification will result in the permanent (i.e., chronic) exposure to high CO 2 throughout life.
An additional complication of assessing the adaptive potential to high CO 2 is that parental exposure to environmental stress can alter offspring phenotypes independently of genetic variation (Guillaume, Munro, & Marshall, 2016). Parents can influence the phenotype of their offspring through a range of nongenetic mechanisms that involve the transmission of nutrients, hormones, somatic factors or epigenetic marks (Bonduriansky & Day, 2009). Commonly, mothers influence the phenotype of their offspring in different environments through changes in the provisioning of eggs, embryos and juveniles (maternal effects) (Crean & Marshall, 2009;Marshall, 2008;Mousseau & Fox, 1998). There is also increasing evidence that the environmental conditions experienced in one generation can influence future generations through the inheritance of different epigenetic states (epigenetic inheritance) (Bonduriansky, Crean, & Day, 2012;Holeski, Jander, & Agrawal, 2012;Jablonka & Lamb, 1995). Consequently, the environment experienced by parents can influence the expression of phenotypic variation in their offspring through a range of nongenetic mechanisms (Bonduriansky, 2012;Day & Bonduriansky, 2011). Recent studies have shown that parental exposure to high CO 2 can ameliorate the negative effects of high CO 2 on growth and survival of juvenile fish (Miller, Watson, Donelson, McCormick, & Munday, 2012;Murray, Malvezzi, Gobler, & Baumann, 2014) and can also influence the kinematic responses of juveniles to a perceived threat (Allan, Miller, McCormick, Domenici, & Munday, 2014). Similar beneficial parental effects have not been observed in the average response of juvenile fish to olfactory cues in a high-CO 2 environment (Welch et al., 2014).
Nevertheless, parental effects could potentially influence the variation in behavioural responses to high CO 2 exhibited by offspring, and thus the heritability of CO 2 tolerance.
There are significant impediments to testing heritability of phenotypic traits in coral reef fishes because (i) the life cycle of most reef fish species cannot be completed in captivity due to difficulties rearing the small pelagic larva and (ii) those species that can be readily reared in captivity are demersal spawners with paternal egg care, making it practically impossible to cross-fertilize eggs and sperm in a diallel breeding design . The spiny damselfish, Acanthochromis polyacanthus, is a monogamous species of reef fish that broods each clutch of offspring for several months after hatching (Kavanagh, 2000). Furthermore, this species can be reared with high success in the laboratory. Thus, it is ideally suited to estimating heritability with parent-offspring regressions (Falconer & Mackay, 1996;Lynch & Walsh, 1998), which is the approach used here.
In this study, we used parent-offspring regressions to test for heritability of variation in behavioural responses to CAC under elevated CO 2 conditions in two different populations of A. polyacanthus. First, we assessed the heritability of variation in behavioural responses to CAC following acute exposure to elevated CO 2 in field-collected families of A. polyacanthus. Next, we tested for heritability of the same trait following acute exposure to elevated CO 2 in laboratory-reared families of the same species. In the laboratory-reared population, we were also able to test for heritability of variation in behavioural responses to CAC following chronic exposure of juveniles to elevated CO 2 and also when parents had been chronically exposed to high CO 2 . Consequently, the laboratory-reared population enabled us to investigate potential differences in heritability associated with acute versus chronic exposure of juvenile fish to high CO 2 and possible nongenetic parental effects from chronic exposure of adults to high CO 2 .
Eggs hatch into small, well-developed juveniles that remain with the parents for several months after hatching (Kavanagh, 2000).

| Acute exposure to high CO 2
We first tested the heritability of variation in behavioural responses to CAC following acute exposure to high-CO 2 conditions in fieldcollected and laboratory-reared families of A. polyacanthus. In both populations, parents were exposed to elevated CO 2 (~750 μatm) for 5-7 days, while their offspring were exposed to ambient control (~450 μatm) or elevated CO 2 for 4 days, before the behavioural response to CAC was measured. This exposure time for the acute treatment was chosen because previous studies have demonstrated that 4-5 days is sufficient time to induce the full range of behavioural effects of high CO 2 in reef fishes (Munday et al., , 2013. The high-CO 2 treatment level of 750 μatm was chosen to match end-of-century CO 2 projections under the moderate RCP6 emissions trajectory (IPCC 2014), and because this is the CO 2 level at which the greatest individual variation in behavioural responses to high CO 2 has been observed in previous studies Munday et al., 2010Munday et al., , 2012. By testing both parents and their offspring at 750 μatm CO 2 , we were able to observe the greatest phenotypic variation in behavioural response to CAC that could be used to estimate heritability of this trait. We also measured the behavioural response to CAC of sibling juveniles that were not exposed to high CO 2 , to provide a baseline with which to compare the behavioural response of juveniles to CAC following exposure to high CO 2 . Heritability in behavioural response to CAC was not estimated in ambient control conditions because adult response to CAC was only measured following high-CO 2 exposure, and not when adults were only in control conditions.
Juveniles were approximately 5 weeks old in the field-collected families and were 6 weeks old in the laboratory-reared families, which enabled us to compare the estimates of heritability in behavioural responses to CAC following acute high-CO 2 exposure between the two populations. Further details of the experimental procedures for both the field-caught and laboratory-reared populations are provided below.

| Chronic exposure of offspring to high CO 2
To test whether chronic exposure of juveniles to high CO 2 affects the heritability of the behavioural response to CAC, we reared a subset of sibling fish from each clutch in the laboratory-reared population for 6 weeks at 750 μatm CO 2 before testing their behavioural response to CAC. Sibling juveniles in the acute and chronic high-CO 2 treatments were tested at the same age (6 weeks) to facilitate a direct comparison between the two treatments.

| Nongenetic parental effects
Previous studies have shown that parental exposure to high CO 2 can affect the phenotypic response of offspring to a high-CO 2 environment in some traits (Miller et al., 2012), but not others (Welch et al., 2014). To test whether parental exposure to high CO 2 altered the heritability of variation in behavioural responses to CAC, we repeated the acute and chronic high-CO 2 treatments described above for the laboratory-reared families, except parents were maintained in high CO 2 for 3 months until breeding, after their initial assessment of behavioural response to acute exposure to high CO 2 ( Figure 1). After the 4-to 5-day treatment period, all fish were tested for behavioural responses to CAC in a two-channel flume, as described below. Olfactory response to CAC in high CO 2 was tested for both parents in the 20 breeding pairs. The sex of each parent was determined after behavioural testing by examining the shape of the genital papillae. Behaviour was tested for 20 offspring (10 in control and 10 in high CO 2 ) from each of these breeding pairs, for a total of 400 wildcaught juveniles. Families were returned to their collection sites after testing, except for two juveniles from each family that were retained to estimate age by examination of daily increments in the otoliths (ear bones). Sagitta were extracted and prepared following standard technique (Fowler, 1990). The average age of juveniles in the field-caught families was 35 ± 6.7 (SD) days.

| Field-collected population
F I G U R E 1 Experimental design for the laboratory-reared population. All adults were first held in high CO 2 for 7 days before measuring their response to chemical alarm cues. Parent pairs were constructed based on the response of individuals to chemical alarm cues, where T = tolerant and NT = nontolerant. Pair formations, as seen in "Parent pairs" boxes, were maintained under their respective treatments, control (414 μatm) and high CO 2 (754 μatm), for breeding. Clutches of offspring from parent pairs were divided between control and high CO 2 conditions at hatching and reared in both conditions for 6 weeks. Acute high CO 2 refers to offspring reared in control conditions that were transferred to high CO 2 (754 μatm) 4 days prior to behaviour testing at the end of the 6-week rearing period

| Laboratory-reared population
Adult A. polyacanthus were collected from the northern Great Barrier Reef, Australia, and transported to the experimental aquarium facility at James Cook University. Adults were held under high CO 2 (754 μatm) for 7 days, after which they were tested for their response to CAC in a two-channel flume (Figure 1). Adult sensitivity to high CO 2 was categorized by their response in the flume. Individuals that spent ≤30% time in the cue were considered to be "tolerant" to high CO 2 , whereas individuals that spent ≥50% time in the cue were considered "nontol-  (Falconer & Mackay, 1996); however, our aim was to generate a maximum possible estimate of heritability (Hill, 1970;Reeve, 1961). Adult pairs were then held in 40-L aquaria at ambient control (414 μatm) or high CO 2 (754 μatm) conditions for 3 months prior to the start of the breeding season. Temperatures and laboratory photoperiod were slowly adjusted during the acclimatization period to reach summer conditions (28.5°C; 13L:11D) at the start of the breeding season. Pairs were checked daily for the presence of egg clutches once the breeding season commenced. On hatching, clutches were immediately divided and transferred to control and high-CO 2 conditions ( Figure 1). Offspring were reared under the two treatments for 6 weeks, at which point they underwent the same olfactory behavioural testing as their parents. Some offspring from control were moved to high CO 2 (754 μatm) 4 days before the 6-week testing to create an acute high-CO 2 treatment group. This acute group allowed for direct comparison with the chronic CO 2 treatment in the laboratory-reared population and with the acute high-CO 2 treatment in field-caught population. The factorial breeding design in the laboratory further allowed for examination of any nongenetic effects of parental conditioning to elevated CO 2 on the heritability of behavioural responses to CAC in offspring exposed both acutely and chronically to high-CO 2 conditions. A total of 60 breeding pairs were formed for the laboratory experiment, 30 of which were held in control conditions and 30 in high CO 2 . Twenty pairs bred in control and 18 pairs bred under high CO 2 .
Only fish from the first clutch from each breeding pair were used in the experiment. Behavioural response to CAC was tested in 60 sibling juveniles from each clutch (20 reared in control, 20 in acute high CO 2 and 20 in chronic high CO 2 , where possible), for a total of 2,258 laboratory-bred juveniles (752 in control, 755 in acute high CO 2 and 751 in chronic high CO 2 ).

| Field-collected population
Three header tanks (60 L) fed water into a total of 30 replicate 32-L aquaria where fish were held (N = 10 tanks per system). One header tank was diffused with ambient air, while the other two header tanks were dosed with 100% CO 2 to achieve the desired pH for the high-CO 2 treatment (754 μatm). Adult pairs were housed in one high-CO 2 system. Clutches of offspring were divided and maintained in both the control and the other high-CO 2 system. pH controllers (Aqua Medic, Germany) maintained the desired pH in the header tanks that supplied the tanks in each system (Table 1). Daily pH NBS (National Bureau of Standards) and temperature measurements were taken in each tank using a pH electrode (SevenGo Pro, Mettler Toledo, Switzerland) and temperature probe (Cormark C26, Norfolk, UK). A portable CO 2 equilibrator and infrared sensor (GMP343, Vaisala, Helsinki, Finland) verified seawater pCO 2 (Hari et al., 2008;Munday, Watson, Chung, Marshall, & Nilsson, 2014). Water samples were taken for the duration of the experiment and used to determine total alkalinity by Gran titration, within 1% of certified reference material (Prof. A. Dickson, Scripps Institution of Oceanography). Salinity was obtained daily from moorings around Lizard Island that are part of Australia's Integrated Marine Observing System (IMOS). Carbonate chemistry parameters were calculated using CO2SYS (Pierrot, Lewis, & Wallace, 2006) with the constants K1 from Mehrbach, Culberson, Hawley, and Pytkowicz (1973) refit by Dickson and Millero (1987), and Dickson (1990) for KHSO 4 -( Table 1).

| Laboratory-reared population
Two 10,000-L recirculating aquarium systems were set to a different pH and corresponding CO 2 level: a current-day control (414 μatm) and a mid-level end-of-century CO 2 (754 μatm). CO 2 was dosed into a 3,000-L sump using an Aqua Medic AT Control System (Aqua Medic, T A B L E 1 Mean (±SD) seawater parameters in the experimental system for adults and juveniles during the experimental seasons Temperature, pH, salinity and total alkalinity (TA) were measured directly. pCO 2 was estimated from these parameters using CO2SYS. Seawater parameters were consistent for breeding and experimental components of the study.
Germany). This allowed for maintenance of the desired pH level in each system.

| Response to CAC
Response to CAC was tested in a two-channel choice flume. The size of the chamber differed between adults (30 cm × 13 cm) (Heuer, Welch, Rummer, Munday, & Grosell, 2016) and juveniles (13 cm × 4 cm) (Gerlach, Atema, Kingsford, Black, & Miller-Sims, 2007). Individuals were given the choice between two water streams in the flume: seawater containing conspecific CAC versus untreated seawater. Water chemistry in the flume matched the respective treatment for each fish. Water from the two different sources was gravity fed into the choice flume, which is divided down half of its length. A constant flow rate of 450 and 100 ml/min was maintained for the adults and juveniles, respectively. Flow rates were monitored using a flow meter and dye test after every water change.
To produce CAC, control donor fish were euthanized with a quick blow to the head. Donor fish were the same size and approximate age as the fish being flumed in each trial. Superficial cuts were made along the sides of the donor fish. Adult donors were then rinsed with 60 ml of control water, while juvenile donors were rinsed with 15 ml of control water . The rinse water was collected and immediately mixed with 10 L of treatment water in the tank used to supply CAC to the flume. CAC was replenished after every second fish to ensure a consistent concentration of fresh CAC for the duration of each trial. A ratio of one donor fish to one test fish was used.
For each trial, a single test fish was placed in the centre of a downstream end of the choice flume and given a 2-min acclimation period.
The position of the fish was then recorded every 5 s for a total of 2 min. A rest period of 1 min followed, during which time the water sources were switched and the fish was re-centred in the downstream end of the flume. The entire acclimation and trial process was then repeated to eliminate potential side preference.

| Behaviour repeatability
A total of 20 adults were tested for repeatability of olfactory behaviour: 10 individuals from the field-caught population and 10 individuals from the laboratory-reared population. An equal number of males and females were tested. Individual olfactory behaviour was measured in the two-channel choice flume, following the protocol above.
Individuals were placed into separate holding tanks for 24 hr after the first behavioural trial. The same olfactory behaviour test was then repeated. Per cent time spent in the chemical alarm cue was correlated for the first and second trials to estimate trait repeatability.

| Data analysis
The average time in CAC for each group of offspring (mid-offspring value) was used in the parent-offspring regressions (Åkesson, Bensch, Hasselquist, Tarka, & Hansson, 2008). We focused our analysis on the father-mid-offspring regression as this should provide the least biased estimate of narrow-sense heritability (Falconer & Mackay, 1996). Heritability (h 2 ) was calculated as two times the slope of the least-squares regression between the father and mid-offspring values (Lynch & Walsh, 1998

| Field-collected population
Acute exposure to elevated CO 2 altered juvenile olfactory response to CAC, with two to four times more time spent in CAC for CO 2 -treated fish compared to sibling fish kept in current-day control seawater ( Figure 2a; F 1,36 = 7.97, p < .01).
Heritability in the behavioural response to CAC was high (0.56) in wild-caught fish acutely exposed to high CO 2 (Table 2) indicating a large component of additive genetic variation. This estimate was strongly influenced by offspring from three of the 20 fathers.
Offspring from these three fathers exhibited a response to CAC that was more similar to control fish (i.e., low time in CAC) compared with other offspring acutely exposed to high CO 2 (Figure 2a). The average time in CAC for these high-CO 2 offspring was only 2%-5% more than their siblings that had been reared and tested in control conditions.
Heritability declines to 0.26 (±0.11 SE) if these three families with tolerant offspring responses are removed from the analysis.
Individual repeatability in responses to CAC was very high for fieldcaught fish with a correlation of r = .992 (p < .001) between the first and second trial for time spent in CAC by adults exposed to high CO 2 .

| Laboratory-reared population
Similar to the field-caught population, juveniles that were acutely exposed to high CO 2 spent two to seven times more time in CAC than sibling fish kept in current-day control seawater (Figure 2b; F 1,36 = 3.23, p > .05).
Heritability in the behavioural response to CAC was high (0.65) in the laboratory-reared fish that were acutely exposed to high CO 2 (Table 2) indicating a large additive genetic component. Heritability estimated in the laboratory-reared fish that were acutely exposed to high CO 2 (±0.17 SE) overlapped with the estimate of heritability in the wild-caught fish acutely exposed to high CO 2 (0.56 ± 0.23 SE).
In direct contrast to the acutely exposed fish, heritability was very low (0.02) for siblings that had been chronically exposed to high CO 2 from hatching (Table 2). With a SE of ±0.11, the estimate of heritability for fish that were chronically exposed to high CO 2 could not be distinguished from zero.
The comparison of regression slopes (Figure 3) further demonstrates the effects of acute versus chronic high-CO 2 exposure on heritability of behavioural response to CAC. For offspring from control fathers, there was a significant interaction between the slopes for acute high-CO 2 offspring and chronic high-CO 2 offspring ( Figure 3a; F 1,36 = 9.55, p < .01). This interaction was due to the positive relationship between fathers and their offspring that were acutely treated with high CO 2 , compared with no relationship for the same fathers and sibling offspring that were chronically exposed to high CO 2 .
F I G U R E 2 Father-mid-offspring regressions for olfactory preference of fish acutely exposed to high CO 2 for (a) the fieldcollected population and (b) the laboratory-reared population. Per cent time spent in the chemical alarm cue was measured. Father responses are from acute CO 2 exposure and are plotted along the xaxes. Mid-offspring responses are plotted along the y-axes for direct relationships with their fathers. Regressions are grouped by offspring treatments: control and acute high CO 2 . Laboratory-reared juveniles are from parents held in ambient control during the breeding season. h 2 = 2b and is depicted on the graphs Heritability in the behavioural response to CAC for fish from parents exposed to high CO 2 for 3 months before breeding was moderate (0.28) in offspring acutely exposed to high CO 2 and could not be distinguished from zero for offspring chronically exposed to high CO 2 (Table 2). For fish that were acutely exposed to high CO 2 , heritability was lower in the group from parents maintained continuously in high CO 2 (0.28) compared with the group from parents that were maintained in control conditions after initial behavioural testing (0.65), indicating a negative effect of parental CO 2 exposure on heritability in the behavioural response to CAC.
Comparison of regression slopes showed a significant interaction between the acute high-CO 2 offspring and chronic high-CO 2 offspring from high-CO 2 fathers (holding condition) (Figure 3b; F 1,32 = 7.29, p < .05). This interaction was similar to that seen between offspring from control fathers, resulting from the positive relationship between fathers and their offspring that were acutely treated with high CO 2 , but lack of relationship for the same fathers and sibling offspring that were chronically exposed to high CO 2 . Indeed, when compared with the same offspring treatments (Figure 3a,b) from parents held in control conditions, there was an overwhelming effect of acute versus chronic high-CO 2 exposure on heritability and a relatively minimal effect of parental CO 2 treatment on heritability.
As observed in field-collected fish, repeatability was very high for laboratory-reared fish with a correlation of r = .997, p < .001, between the first and second trial for adults exposed to high CO 2 .

| DISCUSSION
We found that offspring exposed to elevated CO 2 exhibited impaired antipredator behaviour, spending an increased amount of time in a water stream containing CAC compared with control fish, consistent with previous findings (e.g., Welch et al., 2014). More importantly, our father-offspring regressions revealed strong heritability of variation in behavioural response to CAC in offspring that are acutely exposed to high CO 2 , but there was no heritability in this trait when offspring were chronically exposed to high CO 2 . Our results show that parents that were tolerant to acute elevated CO 2 conditions produced offspring that also exhibited behavioural tolerance under acute high-CO 2 treatments, but this effect was not evident in sibling offspring that were reared in high CO 2 from hatching. Parental exposure to high CO 2 also diminished the heritability of behavioural tolerance to high CO 2 , but not to the same extent as chronic exposure to high CO 2 in juveniles. These results have important ramifications for understanding the likelihood that fish will be able to adapt to projected future CO 2 levels in the ocean.
Our results show that the duration of exposure to high CO 2 can substantially influence the variation in behavioural responses exhibited in juvenile fish, and thus the heritability of behavioural responses to high CO 2 . In the laboratory-reared population, there was high heritability of behavioural response to CAC in offspring that had been acutely exposed to high CO 2 , but not in offspring that had been chronically exposed to high CO 2 . Importantly, fish in these two treatments were siblings and there was almost no mortality during the 6-week rearing period that could have led to the selection of less tolerant genotypes; therefore, the difference must be due to plasticity, not genetic differences. The reduction in heritability for offspring chronically exposed to high CO 2 occurred due to previously tolerant offspring (low percentage time in CAC) in acute high CO 2 losing this tolerance in the chronically exposed siblings (high percentage time in CAC). When chronically exposed to high CO 2 , all juveniles exhibited a high percentage time in CAC. This suggests that the behavioural phenotype of CO 2 -tolerant offspring is modified by nonadaptive plasticity when they are chronically exposed to high CO 2 . Innate responses in animals, such as predator and CAC avoidance, tend to occur rapidly and be both favourable to the individual and heritable (Agrawal, Laforsch, & Tollrian, 1999). However, individual variation in responses may diminish in the longer term due to the convergence of cellular and physiological processes in a common environment (Day & Bonduriansky, 2011). Here, we observed a reduction in phenotypic variation expressed in juveniles that were chronically exposed to high CO 2 compared with juveniles that were acutely exposed to high CO 2 , causing a reduction in the heritability of behavioural response to CAC in a high-CO 2 environment. While parent behaviours were only recorded after acute high-CO 2 exposure for this experiment, it is likely that heritability would remain low for chronically exposed juveniles if parent behaviours were measured after chronic high-CO 2 exposure due to the reduction in phenotypic variation seen in our results. Our results are consistent with reduced heritability in unfavourable conditions in other animals (Charmantier & Garant, 2005;Wilson et al., 2006), but they also demonstrate that the mechanism may be nonadaptive plasticity in stressful environments. Critically, our results suggest that the T A B L E 2 Heritability (h 2 ) of olfactory behaviour for offspring in elevated CO 2 conditions estimated from father-midoffspring regressions collapse in phenotypic variation under chronically high-CO 2 conditions may reduce the potential for adaptation of fish populations to consistently high-CO 2 levels in the future.
High heritabilities for behavioural response to CAC in the acute high-CO 2 treatment in both the field-caught and laboratory-reared populations were strongly driven by a few families that exhibited a high behavioural tolerance to high CO 2 . Offspring from these families had similar phenotypes in both high CO 2 and control conditions, spending minimal time in CAC, indicating that their behavioural response to CAC was tolerant to the effects of high CO 2 . This is consistent with previous observations that some individuals appear to be completely unaffected by this level of CO 2 (700 μatm) and exhibit no change in behaviour compared with controls . Furthermore, these highly tolerant individuals are selectively favoured in their natural habitat because they suffer lower rates of predation . These earlier observations suggested that there could be rapid adaption of behavioural tolerance to high CO 2 in fish populations.
However, those studies involved juveniles that had been acutely exposed to high CO 2 for the same duration as our acute treatments. Our results indicate that these individuals may lack this behavioural tolerance when chronically exposed to high CO 2 from hatching, and this will constrain the adaptive potential of fish populations to high CO 2 .
There was a negative effect of parent holding conditions on the heritability of behavioural response to CAC in fish that were acutely exposed to high CO 2 , with a reduction in h 2 from 0.65 to 0.28 for fathers maintained in control versus high-CO 2 conditions, respectively.
This suggests that nongenetic parental effects might reduce variation in the behavioural response to high CO 2 and consequently reduce the pace of adaptation. Heritable variation was present in offspring treated with acute high CO 2 , indicating that adaptation would be expected to occur in instances where individuals experience short-term exposure to high-CO 2 levels, as might occur in natural upwelling zones and coastal habitats (Hofmann et al., 2011). Nevertheless, higher anthropogenic CO 2 levels will be permanent in the future, exposing successive generations of individuals to high CO 2 for extensive periods of time, which might reduce adaptive potential. Our chronic CO 2 treatment sought to mirror this scenario, and there was no additive genetic variation observed, regardless of parental treatment. Similar to Welch et al. (2014), our results also showed no evidence for transgenerational acclimation of behavioural response to high CO 2 because there was an increased attraction to CAC compared with controls for all offspring that were chronically exposed to high CO 2 , regardless of the parental CO 2 holding condition.
There was heritable phenotypic variation in behavioural responses of juvenile fish exposed to high CO 2 for 4 days, but this variation was absent in fish reared for 6 weeks in high CO 2 . However, the rate of decay in phenotypic variation through time is unknown. The rate of decay could be important because mortality rates of reef fish from predation are highest in small size classes and diminish rapidly with increasing size and age (Almany & Webster, 2006;Jones & McCormick, 2002). Therefore, the strongest selection for appropriate behavioural responses to the threat of predation is likely to occur at an early age when juveniles will have had a shorter period of time exposed to high CO 2 . It is possible that selection of CO 2 -tolerant phenotypes could occur during this critical window of early life, leading to the evolution of more CO 2 -tolerant populations, even if that tolerance is obscured by phenotypic plasticity later in life. Further studies assessing the rate of decay in phenotypic variation through time are required to determine the window of opportunity for selection to act on genetic variation in the behavioural response to predation threat in a high CO 2 environment.
We used parent-offspring regressions to estimate heritability of variation in the behavioural response of fish to CAC in a high-CO 2 environment. Assortative mating can influence estimates of heritability in parent-offspring regressions; in particular, it may generate higher estimates of heritability than random mating (Lynch & Walsh, 1998). We assumed that natural breeding pairs in the field were mated randomly, and thus, our estimates of heritability from the field population are unlikely to be biased by assortative mating. However, our laboratory breeding pairs were deliberately constructed to include pairs of similar high or low behavioural tolerance to high CO 2 . This assortative mating design had the potential to inflate our estimate of heritability. Nevertheless, there was no significant difference in heritability estimated between the field-caught and laboratory-reared populations for the acute high-CO 2 treatments, with both populations demonstrating high heritabilities of behavioural response to CAC in the acute high-CO 2 treatment. Our estimate of heritability in acutely exposed fish from the laboratory population (0.65) was higher than from the field-caught population (0.56), but the values were not significantly different due to the relatively large standard errors (±0.23). It is possible that with a larger sample size, these values might have been significantly different. More critical to our findings, however, is that the assortative mating design would have produced the highest estimate of heritability in the fish that were chronically exposed to high CO 2 . Despite a possible positive bias, our estimate of heritability in the chronic CO 2 treatment was extremely low with an error range that encompassed zero. Consequently, we have increased confidence that there is negligible heritability of behavioural response to CAC when juvenile fish have been permanently exposed to high CO 2 from hatching.
While father-mid-offspring regressions provide the best possible estimate of narrow-sense heritability, they may still include multiple sources of genetic variation (Lynch & Walsh, 1998). There may also be nongenetic effects, such as from parental egg care (Robertson, 1973), or other environmental factors that could affect phenotypic variation in parents and their offspring. If our estimates of heritability from father-mid-offspring regressions contained a large environmental component, we would expect to see very different heritability values between the field-caught and laboratory-reared populations due to substantial differences in the general environments they experienced and because the offspring were separated from their parents at hatching in the laboratory-reared population, but had ongoing parental care in the wild-caught population. Despite the many environmental differences that adults and their offspring would have experienced in the two populations, the estimated heritability of behavioural response to CAC in the field-caught and laboratory-reared populations was remarkably similar. This similarity in heritability estimates, despite the substantial environmental differences between populations, supports the presence of a large genetic component to the phenotypic variation in behavioural response to high CO 2 in acutely exposed fish. In another recent study, Schunter et al. (2016) found that patterns of gene expression in the brain of juvenile A. polyacanthus exhibited a strong connection to the assignment of their parents as either tolerant or nontolerant to high CO 2 , which is also consistent with the presence of genetic variation in CO 2 tolerance for this species.
Our results show that heritability in behavioural responses to high CO 2 is obscured by nonadaptive phenotypic plasticity when juveniles are exposed to high CO 2 for prolonged periods of time. Reef fishes initially display heritable phenotypic variation in behavioural response to high CO 2 that could favour selection of tolerant genotypes during a short ontogenetic window, but this phenotypic variation is lost with longer exposure to high CO 2 . This underscores the importance of investigating genetic variation in phenotypic traits over time scales relevant to environmental change in natural habitats. More generally, our results demonstrate the potential difficulty in estimating adaptive potential to a rapidly changing environment because estimates of heritability of phenotypic variation can differ markedly depending on the duration of the environmental stress applied in experiments. Our results also highlight the potential importance of nonadaptive plasticity to estimating evolutionary potential. Theory and empirical research has generally concentrated on whether adaptive plasticity constrains or facilitates adaptive evolution (Hendry, 2015;Merilä, 2015), with little consideration of nonadaptive plasticity. However, in a recent study, Ghalambor et al. (2015) showed that nonadaptive plasticity might promote adaptive evolution by increasing the strength of natural selection. In contrast, our results suggest that nonadaptive plasticity could constrain adaptive evolution by reducing the phenotypic variation, and thus heritability, of traits expressed in some environments. Either way, considering the role of plasticity in facilitating or constraining adaptive potential will be critical to making reliable predictions about the future of animal populations under rapid climate change.