Predation risk and the evolution of a vertebrate stress response: Parallel evolution of stress reactivity and sexual dimorphism

Predation risk is often invoked to explain variation in stress responses. Yet, the answers to several key questions remain elusive, including the following: (1) how predation risk influences the evolution of stress phenotypes, (2) the relative importance of environmental versus genetic factors in stress reactivity and (3) sexual dimorphism in stress physiology. To address these questions, we explored variation in stress reactivity (ventilation frequency) in a post‐Pleistocene radiation of live‐bearing fish, where Bahamas mosquitofish (Gambusia hubbsi) inhabit isolated blue holes that differ in predation risk. Individuals of populations coexisting with predators exhibited similar, relatively low stress reactivity as compared to low‐predation populations. We suggest that this dampened stress reactivity has evolved to reduce energy expenditure in environments with frequent and intense stressors, such as piscivorous fish. Importantly, the magnitude of stress responses exhibited by fish from high‐predation sites in the wild changed very little after two generations of laboratory rearing in the absence of predators. By comparison, low‐predation populations exhibited greater among‐population variation and larger changes subsequent to laboratory rearing. These low‐predation populations appear to have evolved more dampened stress responses in blue holes with lower food availability. Moreover, females showed a lower ventilation frequency, and this sexual dimorphism was stronger in high‐predation populations. This may reflect a greater premium placed on energy efficiency in live‐bearing females, especially under high‐predation risk where females show higher fecundities. Altogether, by demonstrating parallel adaptive divergence in stress reactivity, we highlight how energetic trade‐offs may mould the evolution of the vertebrate stress response under varying predation risk and resource availability.

The vertebrate stress response involves a complex endocrinological pathway that leads to enhanced glucocorticoid secretion, followed by increased blood glucose levels, as well as enhanced cardiovascular activity and ventilation rate, i.e. the necessities of the fight-or-flight response for predator evasion (Clinchy et al., 2013;Hawlena & Schmitz, 2010;Sapolsky, 1990;Sapolsky et al., 2000).
For instance, the hypothalamus-pituitary-adrenal/inter-renal (HPA/HPI) axis ranks among the most prevalent and evolutionarily conserved adaptations to stress (Clinchy et al., 2013;Hawlena & Schmitz, 2010;Sapolsky et al., 2000). Stress responses often show a high degree of intraspecific variability and interspecific variability (Höglund et al., 2000;Sapolsky, 1982Sapolsky, , 1990, which can result from environmental stimuli via developmental plasticity (Champagne et al., 2008;Chouinard-Thuly et al., 2018;Denver, 2009). Predation as a driver of stress response differentiation is a topic that has garnered considerable attention, but comparatively few studies have focused on the role of local variation in selection regimes for intraspecific divergence by examining both phenotypic (wild-caught) and genetic (laboratory-reared) patterns, or investigated potential sex differences in this context, even though this could prove important.
Overall, the reactivity and magnitude of the stress response should reflect its relative functional importance, manifested as a trade-off between costs and benefits associated with the specific selective environment. For instance, due to high energy requirements, stress reactivity may be dampened in environments with relatively scarce resource availability (Kitaysky et al., 1999). Physiological stress could incur substantial costs in terms of lost opportunities for foraging and reproduction, reduced growth rates, impaired immunity and inflammatory responses, as well as inhibition of sexual behaviour, and overall increased energy consumption (Gregory & Wood, 1999;McPeek et al., 2001;Oppliger et al., 1998;Sapolsky et al., 2000). For example, stress exposure resulted in a 25% reduction of the metabolic scope for activity in green sturgeon (Acipenser medirostris) (Lankford et al., 2005) and a decrease in egg size and offspring survival in brown (Salmo trutta) and rainbow (Oncorhynchus mykiss) trout (Campbell et al., 1994). Hence, whereas it may increase the chance of successful escape from acute threats, high-stress reactivity also comes with associated costs.
Here, we quantified the change in ventilation frequency over time following exposure to a threat, as a proxy of a physiological stress response (e.g. Barreto et al., 2009;Brown et al., 2005;Di Poi et al., 2016;Hawkins et al., 2004). Respiration among most teleost fishes is based on pumping water through the gills by expanding and contracting the buccal cavity in tandem with opening and closing the opercular valves (Helfman et al., 1997). A common reaction to stressors such as predation risk is increased ventilation frequency to increase oxygen uptake in preparation for behavioural responses, such as escape manoeuvres (Bell et al., 2010;Hawkins et al., 2004). Furthermore, ventilation frequency correlates with plasma cortisol levels and is proven to be a very sensitive measurement of predator-induced stress (Barreto et al., 2003;Barreto & Volpato, 2004;Queiroz & Magurran, 2005) and is both easy to quantify and non-invasive. We measured individuals' ventilation frequency in multiple populations of Bahamas mosquitofish (Gambusia hubbsi) that have evolved for thousands of years in either the presence or absence of predatory fish . To evaluate the roles of within-lifetime environmental exposure versus intrinsic, genetically based population differences, we conducted these tests using both field-collected fish and those reared for two generations in a common laboratory environment, for which differential environmental effects should be minimized.
Owing to natural variation in fish communities, inland blue holes (water-filled, vertical caves) on Andros Island, The Bahamas, present an ideal study system to test for predator-induced selection on the vertebrate stress response. Bahamas mosquitofish are small, livebearing fish (family Poeciliidae) inhabiting numerous blue holes that vary substantially in predation risk due to the presence/absence of a principal piscivorous predator (bigmouth sleeper, Gobiomorus dormitor) Langerhans et al., 2007;Martin et al., 2015). Hence, blue holes are easily dichotomized into 'high predation', where predators impose strong mortality and reduce Bahamas mosquitofish densities, and 'low predation' with no predatory highlight how energetic trade-offs may mould the evolution of the vertebrate stress response under varying predation risk and resource availability.

K E Y W O R D S
bahamas mosquitofish, poeciliidae, predation risk, predator-prey interactions, resource availability, sexual dimorphism, stress physiology, stress response, trade-offs, ventilation frequency fishes and, consequently, low-predation pressure and low mortality rates by comparison. As a consequence, several studies have demonstrated that Bahamas mosquitofish from high-and low-predation blue holes have repeatedly evolved different suites of phenotypic traits, including morphology, behaviour and life history (e.g. , 2017;Langerhans et al., 2007;Riesch et al., 2020), since the colonization of the blue holes ~15 000 years ago (Fairbanks, 1989). Importantly, gene flow among populations is low and predation regime is not associated with genetic relatedness among populations (Heinen-Kay & Langerhans, 2013;Langerhans et al., 2007;Riesch et al., 2013;Schug et al., 1998). Further, predation risk in blue holes does not systematically covary with environmental variables such as chlorophyll a density, zooplankton and phytoplankton densities, salinity, turbidity, water transparency, depth, dissolved oxygen, temperature and pH . This scenario allows us to focus explicitly on the role of predation regime in driving divergence in the physiological stress response in Bahamas mosquitofish.
In this study, we test a number of specific predictions regarding the role of predation risk in driving differentiation in the physiological stress reactivity (change in ventilation frequency) of both males and females. We expected that in environments with frequent stressful encounters, evolutionary trade-offs would fine-tune the stress reactivity to prevent unwarranted energy expenditure, while still providing adequate means for escape. Specifically, we hypothesized that populations experiencing higher predation risk would evolve reduced physiological responses to unknown threats. Therefore, we predicted (1) high-predation populations to show a smaller change in ventilation frequency, i.e., lower stress reactivity, over the course of an experimentally induced stressful scenario. This prediction was founded on earlier research and the theory on physiological stress in the wild.
Prey populations in high-predation sites often adapt to the specific stress imposed by predator exposure (Romero, 2004) and should have evolved traits that prevent excessive energy expenditure linked to glucocorticoid synthesis (e.g. Romero, 2004;Sapolsky et al., 2000) as shown in, e.g., fish prey living in high-predation localities (e.g. Archard et al., 2012;Brown et al., 2005;Fischer et al., 2014;Jenkins et al., 2021). Moreover, as the predation regimes in the blue holes have been relatively constant for thousands of years (Heinen-Kay & Langerhans, 2013;Langerhans et al., 2007;Martin et al., 2015;Riesch et al., 2013), we hypothesized that differences in stress reactivity would reflect evolutionary divergence rather than phenotypic plasticity. We therefore expected that (2) differences in stress reactivity between predation regimes would be consistent among laboratory-reared and wild-caught individuals. Finally, because sex is a recognized source of variation in responsiveness to stress, we also investigated potential sex-specific trade-offs. Here, we hypothesized that efficient energy utilization would be more important to females than to males, as female Bahamas mosquitofish bear live young and allocate more time towards foraging , and, hence, efficient energy utilization would be more important to females than to males. We predicted (3) a lower overall ventilation frequency and a lower stress reactivity in females compared with males, and, in addition, that sex differences should be larger in fish from high-predation populations with females in these environments showing the smallest stress reactivity given the increased fecundity females have evolved in highpredation populations Riesch et al., 2013Riesch et al., , 2020.

| Wild-caught specimens
We captured adult male (n = 91) and female (n = 85) Bahamas mosquitofish from three high-predation and three low-predation focal blue holes (populations C, S, W, E, H, and R in Figure 1; see Table S1) during 27 February-7 March 2018, using hand-held dip nets while snorkelling. Sex was easily determined visually by the detection of fully developed gonopodia in males. Blue holes were selected a priori as representative of the larger set of blue holes on Andros Island and are characterized by independent colonization events and low gene flow between mosquitofish populations (e.g. Heinen-Kay & Langerhans, 2013;Langerhans, 2017;Langerhans et al., 2007;Langerhans & Rosa-Molinar, 2021;Riesch et al., 2013). Despite varying resource availability (e.g. food and nutrients), there is a lack of covariation between known environmental parameters and the presence or absence of the predatory bigmouth sleeper Hulthén et al., 2021). After collection, each experimental subject was immediately transferred to a transparent plastic cube (5 × 5 × 5 cm; approximately 1.5-2.0 × fish body length) filled to a depth of 2 cm with water from the same blue hole as the focal fish was captured from (cubes were always rinsed and fresh blue hole water added between trials). To decrease environmental disturbance and to facilitate behavioural analysis by standardizing light conditions during trials, the cube was placed inside a white Neewer® photographic tent (80 × 80 × 80 cm) on an electronic tablet (Apple iPad, model A1822) set to full light intensity on white background. Trials were conducted near the shore of each blue hole, during daylight while shaded by white curtains, and were recorded from above using a tripod mounted DSLR camera (Canon EOS 70D; Canon Inc.) equipped with a macro lens (Canon EF 100 mm f/2.8 USM Macro 1:1). Videos were used to visually determine ventilation frequency (see Video S1 example in Appendix S1), a straightforward and established method for measuring stress in fish (see, e.g., Barreto et al., 2009;Brown et al., 2005;Di Poi et al., 2016;Hawkins et al., 2004;Queiroz & Magurran, 2005). Video recording of each trial began immediately after the cube was placed on the tablet (<40 s after the fish entered the cube), to capture immediate stress responses elicited from capture and placement into a novel, confined environment. To standardize stress exposures, all fish were also subjected to chasing as an additional stressor (e.g. Reid et al., 1994;Marentette et al., 2013, Samaras et al., 2018.
After 60 s of recording, fish were chased for 15 s by a hand-held wooden spatula. We then continued to record each fish for an additional 5 min. | 1557 VINTERSTARE ET Al.

| Laboratory-reared specimens (F2)
For the common-garden experiment, we examined the stress reactivity using the same protocol as for the wild-caught populations. All fish were raised under common laboratory conditions in a large, recirculating aquarium system comprising 72 115-L aquaria with biological, mechanical and ultraviolet filtration.
Water was maintained at approximately 25°C, a conductivity of 2850 µS/cm and a pH of 8.3, with a 14-h light/10-h dark photoperiod. Fish were fed daily with a mixture of TetraPro Tropical

Crisps, Fluval Bug Bites for Tropical Fish and Hikari freeze-dried
Daphnia, blood worms and brine shrimp. To generate the experimental animals, we collected the parental generation (F0) as newborns (to minimize maternal and environmental effects) from eight blue hole populations ( Figure 1). Wild-caught F0 specimens were transported to the laboratory facilities at North Carolina State University, United States, raised to adulthood, and each female was mated with a single male from the same population, with no fish used more than once (on average, 19 unique malefemale pairings were used from each locality, range: 10-26).
Offspring (F1) were then raised to adulthood in the same recirculating aquarium system, and females from each population were mated with multiple males from the same population (to maintain high genotypic diversity in laboratory populations). Offspring from these matings (F2) were raised separately by sex and served as experimental subjects for stress trials (male: n = 84, female: n = 87; see Table S1). All fish were reared and tested at approximately 25.0°C.

| Data treatment and statistical analysis
Recordings were viewed blind by a single observer (URZ) who was not involved in the recordings. The observer used the VLC media player software (3.0.8) with optional slow-motion analysis to extract the time taken for 60 opercular beats to occur at six different time-points in each recording: start of the recording (t 0 ), immediately following the additional acute stressor (t 1 ), and then in 1-min increments until 4 min after the acute stressor (t 2 -t 5 ). If the experimental subject was swimming/moving in the arena, we paused the counting of opercular beats and continued as soon as the individual stopped moving. Paused time was subtracted from the total time in the analyses. Across all six timepoints, the average proportion of trials that required pausing were as follows, field data: 0.266; laboratory data: 0.113. F I G U R E 1 Map of the location of the eight focal blue holes used in this study. Blue: low predation, red: high predation. Study site abbreviations follow Table S1 To measure body size, we captured still frames from the videos and measured standard length (SL in mm) of each fish using the image analysis software ImageJ (version 1.52, https://imagej.nih. gov/ij/). Size of each image was calibrated from a ruler placed in the experimental cube (see Figure S1 in the Appendix S1).
Data were transformed into ventilation frequency (beats per minute, BPM) and used as the dependent variables (BPM from the six repeated time-points) in a mixed-model repeated-measures analysis of covariance (ANCOVA) to test for differences in stress reactivity between sexes, predation regimes and rearing environments.
Predation regime, sex, rearing environment and their interactions served as independent variables. To properly treat population as the unit of replication for tests of predation regime, we included population nested within predation regime as a random effect.
Also, we included log 10 -transformed (for normality) SL (mean ± SE; 24.32 ± 0.22 mm for males, and 29.94 ± 0.55 mm for females) as a covariate to statistically adjust for body size effects on ventilation frequency (we expected a negative correlation, see, e.g. Brown et al., 2005) and thus examined size-independent aspects of stress reactivity. We included an interaction term between standard length and time point to test whether size dependence varied over the course of the trial; interactions between standard length and all other effects were initially included but excluded from the final model due to unimportance (all p > 0.17).
Whereas this repeated-measures analysis provided an appropriate overall test of (1) within-subjects effects (Time), (2) whether changes in ventilation frequency over time might differ between groups (e.g. Time × Predation Regime), and (3) whether average ventilation frequencies differ between groups (e.g. Sex), we calculated two additional metrics to more explicitly evaluate variation in the stress response per se and the recovery ventilation rate. First, we measured the scope of ventilation frequency for each fish as an estimate of the relative magnitude of the stress response: average BPM during t 0 and t 1 minus the minimum BPM during the trial. This metric is designed to capture each fish's reactivity to the mild stressors of capture, confinement and chasing. Second, we calculated the recovery ventilation frequency as the average BPM during t 3 to t 5 , estimating the ventilation frequency exhibited soon after a startling experience. Using these two metrics as dependent variables, we conducted separate general linear mixed models with predation regime, sex, rearing environment and their interactions as independent variables. We again included log 10 -transformed SL as a covariate to adjust for effects of body size, as well as population nested within predation regime as a random effect. These two dependent variables (scope and recovery BPM) were not correlated with one another (r = 0.07, p = 0.17). To ensure robust results, we additionally performed all analyses excluding the two populations only examined in the laboratory (not in the wild); all results were qualitatively similar (see Tables S2 and S3). Statistical analyses were performed using the Proc Mixed procedure in SAS software (version 9.3).
Because our results indicated that an additional environmental factor (not included in our models) might play an important role in explaining among-population variation in stress response (see Results), we conducted two more analyses. Owing to the energetic costs of physiological stress responses (see Introduction), we suspected that resource availability might serve as an important selective agent, favouring reduced stress responses with scarce resources and elevated stress responses when high-quality food resources are readily available (see Discussion). We estimated resource availability in these eight blue holes using previously published estimates of zooplankton density. Specifically, previous work has documented that the availability of major prey of Bahamas mosquitofish (zooplankton, mainly copepods; Araujo et al. 2014;Gluckman and Hartney 2000;Riesch et al. unpubl. ms) consistently varies among blue holes (e.g. Hulthén et al., 2021;Sha et al., 2020) and has led to population variation in male coloration (Martin et al., 2014) and evolutionary divergence in some life-history traits, such as juvenile growth rate . Zooplankton density was estimated using a 60-m tow of a zooplankton net (20-cm diameter, 153μm mesh) at 0.5-m depth within habitat where Bahamas mosquitofish were abundant within all sites . We conducted a general linear mixed model separately for wild-caught and laboratory-raised fish that used the scope of ventilation frequency as the dependent variable; and predation regime, sex, zooplankton density (log 10 -transformed for normality) and their interactions as independent variables; and log 10 -transformed SL as a covariate. Population nested within predation regime was included as a random effect. Because the interaction between zooplankton density and sex, and the three-way interaction (zooplankton density × predation regime × sex) were uninformative (all p > 0.22), we excluded those terms from our models.

| RE SULTS
Our mixed-model repeated-measures ANCOVA revealed a number of important effects of model terms on ventilation frequency (Table 1). This analysis revealed that fish from high-predation populations tended to show a dampened stress reactivity (Predation Regime × Time interaction; Table 1; Figure 2), in line with our first prediction. In low-predation populations, ventilation frequency generally dropped more precipitously over time compared with fish from high-predation populations (Figure 2). Whereas this analysis did not identify any strong evidence that the stress response differed between the wild and the laboratory within either predation regime (i.e. PR × ENV × Time interaction), consistent with our second prediction, inspection of the time-course stress reactivity curves (Figure 2) suggested that some variation may have indeed occurred: (1) in the laboratory, females appeared to show little difference between predation regimes in their average time-course curves even though they exhibited strong differences in the wild (Figure 2a) and (2) males in low-predation populations seemed to show differences between the wild and laboratory, resulting in a shift of the time-points in which the greatest differences between predation regimes occurred (i.e. during t 2 -t 5 in the wild, during t 0 -t 2 in the laboratory; Figure 2b).
Overall, fish from different predation regimes, as well as wild-caught and laboratory-raised fish, showed little consistent differences in ventilation frequency averaged across time-points (Predation Regime term, Rearing Environment term, respectively, Table 1; Figure 3a). Related to our third prediction, we uncovered strong evidence of sexual dimorphism in ventilation frequency, which varied among predation regimes, rearing environments and time-points (Table 1). Specifically, females typically exhibited a lower average ventilation frequency than males (Sex term; Table 1), a pattern that was more evident in fish from high-than from low-predation populations (Sex × Predation Regime interaction; Table 1, Figure 3a), as well as in laboratory vs. field conditions (Sex × Rearing Environment interaction; Table 1, Figure 3a). In the wild, males exhibited a ventilation frequency 7% higher, on average, than females within highpredation environments, but only showed an average difference of 1% in low-predation environments (Figure 3a). However, in the laboratory, males clearly showed higher ventilation frequencies than females within all but one population (on average, 11% higher in high-predation, 8% higher in low-predation; Figure 3a). Also, females tended to show smaller changes in ventilation frequency over the time-course of the assay, that is a lower stress reactivity as compared to males (Sex × Time interaction, Table 1, Figure 2). Average sex differences in ventilation frequency were greatest early in the assay (females ~ 19 BPM slower than males during t 0 -t 2 ) and quickly declined towards the end of the assay (females ~ 13, 11 and 8 BPM slower than males in t 3 , t 4 and t 5 , respectively). Finally, as expected there was a strong, negative relationship between ventilation frequency and body size (Table 1, Figure S2), which did not vary over the time-course of the assay (SL × Time interaction, Table 1).
Examination of the scope of ventilation frequency uncovered that several factors influenced this estimate of stress reactivity to the mild stressors imposed in our assay (Table 2). Smaller fish typically exhibited a larger scope, whereas the strongest influence on sizeindependent scope involved a dependence of the predation regime effects on the rearing environment (Predation Regime × Rearing Environment interaction): on average, low-predation populations exhibited a larger scope than high-predation populations in the wild (75% higher in females, 42% higher in males), but these differences declined in laboratory-raised F2 fish (1% higher in females, TA B L E 1 Results from mixed-model repeated-measures ANCOVA examining variation in ventilation frequency of Bahamas mosquitofish across the six time-points of the mild stressor assays  Table S1). Illustration depicts a spline fit to the least-squares means for each of the six time-points, and the shaded regions depict ±1 SE. Red: high predation, blue: low predation; light shade and dashed lines: wild-caught ( These results suggested that an unmeasured factor might help explain among-population patterns, and thus, we performed the additional analyses that tested for the influence of resource availability on the scope of ventilation frequency. Despite low statistical power to detect among-population trends, especially in the wild (three populations within each predation regime), the results depicted relatively clear patterns. In the wild, populations with greater zooplankton density tended to exhibit higher ventilatory scope (Figure 4a), although the trend was not strong and had little statistical support (Table S4). Laboratory-reared fish, however, showed an especially clear pattern where low-predation populations showed a strong positive relationship between scope of ventilation frequency and zooplankton density, but high-predation populations showed no association ( Figure 4b; Table S4).

F I G U R E 2 Visualization of variation in the physiological stress response (ventilation frequency) of Bahamas mosquitofish from low-and high-predation populations for (a) females and (b) males (sample sizes in
Results for recovery ventilation rate largely paralleled findings for average ventilation rate from the mixed-model repeatedmeasures analysis. Statistically adjusting for effects of body size, we found that sex differences in recovery ventilation rate depended on the predation regime (larger in high predation) and rearing environment (larger in the laboratory) (terms involving Sex, other than the F I G U R E 3 Variation in (a) average ventilation frequency and (b) the scope of ventilation frequency (change during the mild stressor assay) for females (square) and males (circle) in all populations of Bahamas mosquitofish examined in the wild and after two generations of laboratory rearing (least-squares means ±1 SE depicted; population abbreviations follow Figure 1; samples sizes in Table S1). Light shaded symbols: population means, dark shaded symbols: predation-regime means three-way interaction; Table 2). Consistent with our third prediction, females usually showed lower recovery ventilation frequencies than males, especially in high-predation populations ( Figure S3). Similar to the average ventilation frequencies (Figure 3a), wild-caught males exhibited a recovery ventilation frequency 6% higher, on average, than females within high-predation environments, but only showed an average difference of 1% in low-predation environments ( Figure   S3). In the laboratory, males showed higher ventilation frequencies than females within all but one population (on average, 11% higher in high-predation, 6% higher in low-predation; Figure S3). Again, we found the expected strong, negative relationship between ventilation frequency and body size (Table 2).

| DISCUSS ION
In this study, we utilized the post-Pleistocene radiation of Bahamas mosquitofish to test several key hypotheses regarding the role of predation risk in moulding the vertebrate stress response. By investigating male and female stress reactivity in field-collected and laboratory-reared individuals from multiple populations, we gained several important insights into the evolution and expression of stress reactivity in a vertebrate. Interestingly, our observations did not completely align with our a priori predictions but point towards an unexpected future research direction.
Our first major result was partially in line with our prediction that populations inhabiting high-predation environments, where encounters with stressors are frequent and prolonged, should evolve a relatively lower stress response to sudden, mild stressors. We hypothesized that the frequent predator encounters experienced by Gambusia in high-predation environments would select for an adaptive fine-tuning of stress reactivity that reduces accumulating costs, consistent with the adaptive hypothesis for animals living in high-risk environments (Cooke et al., 2003).  Dobrakovova et al., 1993;Romero, 2004). Our results are partially consistent with these findings, as only wild-caught, but not second-generation laboratory-reared, Bahamas mosquitofish showed strong differences between predation regimes in physiological responses (i.e. scope of ventilation frequency) to mild TA B L E 2 Results from general linear mixed models examining variation in the scope of ventilation frequency (i.e. magnitude of the stress response) and the recovery ventilation frequency (i.e. relative baseline ventilation frequency soon after a startling experience)

F I G U R E 4
Association between resource availability (zooplankton density in the field) and the scope of ventilation frequency in Bahamas mosquitofish from low-predation and high-predation populations examined (a) the wild and (b) after two generations of laboratory rearing (least-squares means ±1 SE depicted). Regression lines drawn for each predation regime (sexes pooled) to illustrate the trends stressors. Yet, this resulted from an intriguing pattern of amongpopulation variation within predation regimes that we argue (1) highlights the strong role of predation risk in driving the evolution of stress reactivity and (2) points to another important selective agent (resource availability) in the evolution of the stress response.
Considering that these blue hole populations are genetically isolated, largely evolving independently from one another (Heinen-Kay & Langerhans, 2013;Langerhans et al., 2007;Riesch et al., 2013;Schug et al., 1998), our findings in second-generation laboratory-raised fish suggest that evolutionary trajectories have differed between predation regimes. First, high-predation populations appear to have converged on highly similar magnitudes of stress responses (see Figures 3 and 4b). This is consistent with the notion that in the presence of predators, strong and consistent selection has caused parallel evolution. Meanwhile, lowpredation populations exhibited considerable variation in their ventilatory scope and appear to have strongly responded to resource-mediated selection, with larger stress responses evident in populations that have evolved in environments with higher levels of resource availability (zooplankton density). Previous work has independently suggested important roles for predation risk and food availability/quality in stress reactivity, but these factors have not been simultaneously considered, and most prior research involved plastic (or potentially plastic) responses rather than evolutionary divergence in the stress response (e.g. Clinchy et al., 2013;Hammerschlag et al., 2017;Hawlena & Schmitz, 2010;Herring et al., 2011;Kitaysky et al., 1999Kitaysky et al., , 2007Romero & Wikelski, 2001). Together, our results suggest a scenario where three key attributes might together largely explain evolutionary patterns of the vertebrate stress response based on the relative costs and benefits of reacting to stressful events: (1) intensity of stressful encounters, (2) frequency of stressful encounters and (3) food availability. We briefly describe this hypothesis, present a simple conceptual model to illustrate its general evolutionary predictions (see Appendix S1 for details) and suggest that future studies should explore this hypothesis both theoretically and empirically.
If we combine the predicted influence of these key agents, we find that the predicted patterns of stress-response evolution closely resemble our findings in this study. First, selection should more strongly favour a large-magnitude stress response when encounters are more intense, where 'intensity' can range from mild to severe consequences of encounters, such as energy expenditure, lost feeding or mating opportunities, injury and death. Second, owing to the costs of mounting a stress response (e.g. Gregory & Wood, 1999;McPeek et al., 2001;Oppliger et al., 1998;Sapolsky et al., 2000), These expectations correspond well to the patterns observed here in Bahamas mosquitofish; that is, all high-predation populations showed similar, low-to-moderate ventilatory scope in response to F I G U R E 5 Predictions for the evolution of the stress response under varying intensity and frequency of stressful encounters in environments with (a) low food availability and (b) high food availability based on our simple conceptual model that assumes its evolution depends solely on selection from these three factors (see text, Appendix S1; 0.1 and 1.0 food levels depicted; full range of frequency of stressful encounters depicted; intensity of stressful encounters range from 0.25 to 1.0). Approximate regions for each predation regime for Bahamas mosquitofish denoted with LP (low predation) and HP (high predation) a mild stressor, whereas low-predation populations showed lowmagnitude stress responses when accustomed to low resource levels but high-magnitude ventilatory scope when accustomed to high resource levels. The generality of this conceptual model requires further investigation, and we suggest future studies address this topic.
The weak evidence for differences in scope of ventilation frequency between rearing environments for high-predation populations matched our prediction that evolutionary divergence, not phenotypic plasticity, explains variation in stress responses-but this only applied to high-predation populations, as low-predation populations generally showed smaller scope in the laboratory compared with the field. Blue hole populations have experienced relatively constant predation threat for thousands of generations with little gene flow across blue holes (e.g. Heinen-Kay & Langerhans, 2013;Riesch et al., 2013). This scenario should elicit little selection for plasticity in stress reactivity (Tollrian & Harvell, 1999;West-Eberhard, 2003). In line with these findings, genetic adaptation, and not developmental plasticity, has been found to underlie divergence in stress reactivity (ventilation frequency) between marine and freshwater three-spined sticklebacks (Gasterosteus aculeatus) (Di Poi et al., 2016). Yet, low-predation populations in this study showed considerable variation between the wild and the laboratory in their stress response, suggesting these fish are more sensitive to certain aspects of environmental variation in this regard. Whereas prior work has shown temporal repeatability in primary productivity and resource availability in the Bahamian blue holes (e.g. Hulthén et al., 2021), these factors do show some variation over time, and food quality and quantity surely differed between the wild and the laboratory, suggesting that resource-related factors may partially explain these findings. One potential phenomenon that could be involved is counter-gradient variation, and future studies could directly examine this hypothesis. That is, reduced food levels might induce higher stress responses (Kitaysky et al., 2007;Romero & Wikelski, 2001), but select for lower stress responses.
Considering that laboratory-raised fish from low-predation populations tended to show lower stress responses than in the wild, and food quality is probably greater in the laboratory, this pattern seems at least plausible at this point. Thus, it seems that whereas predation by piscivorous fish drives parallel evolution of stress responses in Bahamas mosquitofish, other factors such as food availability become the driving factors affecting stress responses in the absence of predators.
In contrast to patterns of ventilatory scope, fish from different predation regimes did not show differences in average or recovery ventilation frequency. In a previous study, wild-caught female fish (Brachyrhaphis episcopi) from high-predation sites in Panama similarly showed a lower ventilation frequency and a smaller scope than their low-predation counterparts when exposed to an experimental stressor, but, in contrast, a higher ventilation frequency under normal activity levels (Brown et al., 2005). Consequently, the evolution of stress responses is complex and has proven not always consistent across species. For example, wild three-spined sticklebacks facing a high risk of predation had higher opercular beat rates in response to confinement stress as compared to conspecifics from low-predation sites (Bell et al., 2010). However, these fish were juveniles and there was no effect of sex in the final model (Bell et al., 2010). Hence, we argue that it is of importance to address both biological and ecological contexts to gain a more thorough understanding of the proximate and ultimate factors affecting development and evolution of stress responses (see, e.g., Archard et al., 2012;Reeder & Kramer, 2005).
Moreover, size-specific differences in ventilation frequency between predation regimes on the scale observed here (~14-20 BPM in the wild) could certainly translate to fitness-relevant differences in energy usage and is similar to or exceeds findings in previous work of about 10-BPM differences between predatory environments in similarly sized fish (Bell et al., 2010;Brown et al., 2005).
The sex-specific effects on ventilation frequency indicate that stress reactivity can evolve differentially among males and females, and independently of sex differences in overall body size. Sexspecific stress responses have been demonstrated in earlier studies (e.g. Donelan & Trussell, 2020), where female responses to stress have been suggested to build on processes related to attachment and caregiving that ultimately would downregulate the HPA-axis (Sapolsky et al., 2000;Taylor et al., 2000). The fact that females showed lower overall ventilation frequencies than males, especially in laboratory-raised fish and in high-predation populations, suggests that the sexual dimorphism has a strong genetic basis and that predation threat has influenced its evolution. We suggest a high premium is placed on energy efficiency in females compared with males.
Bahamas mosquitofish females are viviparous, breed year-round, invest heavily into reproduction (embryos make up ~12%-13% of the body weight), are almost constantly pregnant and even provide nutritional provisioning to embryos during pregnancy in some populations (Riesch et al., 2013). These great energetic demands should result in stronger selection for energy efficiency in the stress response of females compared with males. A possible explanation for why sex differences were greater in the laboratory derives from sex-specific housing in the rearing environment; laboratory-reared females did not contend with male harassment, whereas laboratoryreared males may have experienced elevated male-male interactions.
Avoiding wasteful energy expenditure may be even more important in high-predation localities, where fecundity is higher Riesch et al., 2013) and energy-demanding rapid locomotor performance is critical for surviving predatory threats (Langerhans, 2009). Similar patterns were recently described in the foraging behaviour of mosquitofish, with stronger divergence between predation regimes in females than males . Overall, female energy efficiency is therefore a potential explanation to why the sexes differ more strongly in high-predation compared with lowpredation populations.
We however only found inconclusive support for reduced stress reactivity (i.e. scope of ventilation frequency) in females compared with males, indicating that whereas females have clearly evolved lower ventilation rates, the relative sex differences in the physiological stress response are more variable and minor. In contrast, a recent study demonstrated that in another live-bearing fish species (Trinidadian guppies, Poecilia reticulata), females have a lower cortisol release than males when exposed to a stressor (Chouinard-Thuly et al., 2018). Overall, it seems likely that lifehistory differences between the sexes may largely explain differential stress responses, as these can be tightly linked to individual variation in physiological stress responsiveness in vertebrates (Furtbauer et al., 2015;Reeder & Kramer, 2005;Sapolsky, 1990).
In Bahamas mosquitofish, the sexes differ in a large range of morphological and behavioural traits, and genetic correlations among these traits may prove important in sex-specific patterns of stress physiology and reactivity.

| CON CLUS IONS
Natural populations of Bahamas mosquitofish inhabiting blue holes provide an excellent model system for evolutionary studies of local selection and adaptation. In the current study, we utilized both wild-caught and laboratory-reared mosquitofish originating from multiple, independent high-predation and low-predation sites to provide evidence that predation risk is indeed a major factor moulding stress responses in vertebrates. High-predation pressure appears to have driven very similar stress responses in high-predation populations, whereas low-predation populations show more variation among populations, apparently responding to site-specific resource availabilities. We argue that these patterns are likely explained by evolutionary trade-offs to reduce stress-related energy expenditure among environments that differ in frequency and intensity of stressful encounters, as well as in food availability. Furthermore, we demonstrate significant sexual dimorphism in ventilation frequency and propose that this is strongly influenced by evolution under varying predation risk. In a broader context, our results suggest that animals exposed to frequent, high-intensity stressful encounters benefit from a dampened acute stress response, especially for individuals that require high energy intake (e.g. reproductively active, livebearing females). Meanwhile, a dampened stress response should also evolve in environments with low resource availability, but this effect may be strongest when encounters are less frequent and potentially less intense. The link between the vertebrate stress response and overall energy allocation, as well as potential effects thereon from predation risk and sex, remains important issues for future research.