Personality-driven life history trade-offs differ in two subpopulations of free-ranging predators

1. Consistent individual differences in behaviour (i.e. personality) can be explained in an evolutionary context if they are favoured by life history trade-offs as conceptu alized in the pace-of-life syndrome (POLS) hypothesis. Theory predicts that faster-growing individuals suffer higher mortality and that this trade-off is mediated

Maintenance of such within-population variation can be explained when behaviour mediates trade-offs between important life-history traits (Réale et al., 2010;Wolf, van Doorn, Leimar, & Weissing, 2007). Life history trade-offs occur when the optimal response for one trait is reached at the expense of another (Stearns, 1989). Such trade-offs can maintain multiple phenotypes in a population by allowing individuals with different behavioural strategies to achieve comparable fitness (Mangel & Stamps, 2011).
For example, individuals that explore more and take more risks may discover more new food patches than their less explorative conspecifics (Spiegel, Leu, Bull, & Sih, 2017). Consequently, they would grow faster and reproduce at a younger age but experience higher mortality via predation. In turn, less explorative, riskaverse individuals would likely conduct less extensive searches for resources, leading to lower growth and delayed reproduction but higher survival due to lower predator encounter probabilities (Lapiedra, Schoener, Leal, Losos, & Kolbe, 2018). In goldfish Carassius auratus, individuals that make more outings from a refuge acquire more food than their risk-averse conspecifics but suffer higher risks from avian predation (Balaban-Feld et al., 2019). In this way, individual behaviour is expected to play a critical role in mediating trade-offs between important life-history traits such as growth and mortality (Stamps, 2007). This concept is most comprehensively conceptualized by the pace-of-life syndrome (POLS) hypothesis which integrates covariations in life history, behaviour and physiology within populations (Dammhahn, Dingemanse, Niemelä, & Réale, 2018).
Despite a large literature on life history trade-offs and on animal personality, testing the link between these two phenomena has led to equivocal evidence (Montiglio, Dammhahn, Dubuc Messier, & Réale, 2018;Royauté, Berdal, Garrison, & Dochtermann, 2018). For instance, female wild cavies Cavia aperea that explored more in a known captive environment grew faster, in accordance with theoretical predictions. But unexpectedly, exploration in a novel open field correlated negatively with growth and no behaviour correlated with the reproductive output of the animals (Guenther, 2018). In crickets Gryllus integer, bolder animals were found to have stronger cellular immune functions, challenging predictions of the POLS hypothesis (Niemelä, Dingemanse, Alioravainen, Vainikka, & Kortet, 2013). More recently, a metaanalysis revealed that personality only explained a small portion of variance in survival (Moiron, Laskowski, & Niemelä, 2020). Such conflicting results could be due to the inadequacy of the testing environment to elicit trade-offs. Indeed, the need for controlled and repeated testing inherent to studies on personality has led to a predominance of captive study systems in the literature (Archard & Braithwaite, 2010). However, local ecological conditions (e.g. predator abundance, resource abundance etc.) may be necessary to drive the emergence of life history trade-offs (Moiron et al., 2020;Montiglio et al., 2018). For instance, stronger selection on risktaking behaviours was observed in the presence of predators than in their absence in Anole sagrei lizards (Lapiedra et al., 2018). Furthermore, the relationship between personality, survival and reproduction was found to be habitat dependent in Eurasian red squirrels, Sciurus vulgaris (Santicchia et al., 2018). Therefore, the link between life history and personality might be most effectively investigated in long-term studies of free-ranging animals experiencing strong and variable ecological pressures (Jablonszky et al., 2018;Montiglio et al., 2018).
The ambiguity of evidence supporting the link between personality and life history trade-offs may also arise from the difficulty of establishing the ecological relevance of captive behavioural tests (Carter, Feeney, Marshall, Cowlishaw, & Heinsohn, 2013). Exploration of a standardized novel open field in house mice Mus musculus domesticus reflected exploratory behaviour in semi-natural environments (Krebs, Linnenbrink, & Guenther, 2019) and a similar test in great tits Parus major did not (Arvidsson, Adriaensen, van Dongen, De Stobbeleere, & Matthysen, 2017). One test might be appropriately used as a proxy for natural behaviours in some species but not in others.
Researchers in that case would benefit from working in systems selection on fast growth, we found no link between exploration personality and the growth-mortality trade-off.

5.
Our study demonstrates that the association between personality and life history is favoured in some ecological contexts but not in others. We identify predator and resource abundance as two main potential drivers of the personalitymediated trade-off and emphasize that future work on the POLS hypothesis would benefit from an approach integrating behaviour and life history across ecological conditions.

K E Y W O R D S
growth-mortality, life history, natural selection, pace-of-life syndrome, personality, sharks, trade-offs in which behaviours measured in captivity can be compared to similar behaviours in the field to ensure that the captive tests' results can be expected to regulate trade-offs in the wild (as previously suggested by Réale et al., 2007).
Here we tested whether personality measured in a novel open field mediates a growth-mortality trade-off via wild risk-taking behaviour, in juvenile lemon sharks Negaprion brevirostris. At our study site, Bimini, Bahamas (Figure 1), juvenile lemon sharks exhibit high site-fidelity and small home-ranges, which allow for regular monitoring until they die or disperse (Chapman et al., 2009;Morrissey & Gruber, 1993). Previous work suggests that life history trade-offs are important in this system, with slower-growing and smaller-bodied individuals enjoying higher apparent survival (DiBattista, Feldheim, Gruber, & Hendry, 2007) whereas individuals that exhibit riskier foraging strategies experience higher growth rate but potentially higher predation (Hussey et al., 2017). It is unclear, however, whether individual behaviour mediates the growth-mortality trade-off in this system.
Wild-caught juvenile lemon sharks from Bimini were tested yearly in a captive novel open-field assay from 2012 to 2017 (n = 286 individuals tested), with a subset of the tested sharks (n = 46) monitored in the field for up to a year using acoustic telemetry to record their distance from the shore, which is a known refuge from predators (Hussey et al., 2017). Then, using 23 years of mark-recapture data (n = 2,627 individuals), the personality measures and the telemetry data, we tested the predictions (a) that faster explorers in the novel open field were found further from shore (i.e. exhibited risk-prone behaviours) and had higher growth rates and (b) that natural selection acted against explorative, fast-growing individuals in two adjacent subpopulations of juvenile lemon sharks (North Sound and Sharkland, see Figure 1). Because these subpopulations were described to differ in their predator abundance, with Sharkland having more predators Guttridge et al., 2012), we expected selection on exploration behaviour and fast growth to be stronger there than in North Sound.

| Shark capture and handling
In Bimini, Bahamas (Figure 1), lemon sharks are born in April and May of each year (Gruber, de Marignac, & Hoenig, 2001). They exhibit high site-fidelity and can be found for the first 2-5 years of their lives in specific areas (so-called nurseries) around the islands (Morrissey & Gruber, 1993). In this study we focus on the subpopulations inhabiting two nursery areas that are adjacent but almost totally separated with respect to emigration and immigration of sharks under 2 years of age: North Sound and Sharkland (Figure 1; Gruber et al., 2001).
Sharks from both subpopulations were sampled in May and June of every year since 1995. Sampling was conducted using gillnets (180 m, length) deployed perpendicularly to the shore at three standard locations (see Figure 1) for six consecutive nights (12 hr) in each nursery.
Nurseries were sampled consecutively resulting in 12 nights of sampling per year. We checked the nets every 15 min to minimize stress of sharks and bycatch. Upon capture, sharks were checked for the presence of a uniquely coded PIT (Passive Integrated Transponder, Destron Fearing), they were tagged, sexed, measured [Pre-caudal length (PCL), nearest mm] and the condition of their umbilical wound was assessed to estimate age (see below Section 2.2). Sharks were then transferred to a nearby housing pen (details on pen construction: Guttridge et al., 2009) which prevented them from being captured on multiple occasions within the sampling period. Consequently, a rapid decrease of the capture rate was observed over the six nights of sampling, providing evidence for the very high efficiency of these methods (96% of a subpopulation captured by the fourth night of fishing as estimated by Gruber et al., 2001). Since the inception of this mark-recapture program in 1995, we recorded a total of 3,419 captures of sharks under 2 years old (0 year old, n = 2,204; 1 year old, n = 1,215) which accounted for the majority of sharks captured (n = 1,075 captures of sharks between 2 and 5 years old).

| Age estimations
Lemon sharks are placental viviparous and juveniles are born with an umbilical wound which closes slowly over a few months (DiBattista was not recorded for n = 1,227 sharks. To estimate their age, we used a linear regression of age on PCL for sharks that had been precisely aged (i.e. via umbilical status) and that were under 2 years old, as we were mostly interested in 0 and 1-year old sharks for this study (Adjusted R 2 : 0.66, F 1,3,312 = 6,627, p < 0.0001). We predicted the age of the sharks based on the linear model, rounding the estimated age to the closest integer. We considered this linear estimation relevant as the relationship between body length and age (estimated via growth-band counts on vertebral centrum) shows a strong linear relationship in sexually immature lemon sharks (<180 cm PCL; mean PCL of sharks included in this study: 50.2 cm ± 3.34 SD; Gruber & Stout, 1983). Eight unaged sharks did not have a measure of PCL, we therefore used a similar model with the total length of the sharks instead. This method proved to be 90.1% accurate when used to predict the age of sharks for which umbilical opening data were available.

| Growth and apparent survival estimations
Juvenile lemon sharks are likely to disperse from their natal nursery between age 3 and age 5 (Chapman et al., 2009;DiBattista et al., 2007). In addition, most sharks captured via the mark-recapture program described above were 0 or 1 year old, therefore, we only considered in this study animals that were under 2 years old (n = 2,627).
We calculated growth as the body length (PCL) gained from one year to the next (in cm). The sharks' small home-ranges (<400 m 2 for sharks under 54 cm PCL, Morrissey & Gruber, 1993), coupled with the efficiency of the capture methods and the negligible emigration at age 0 and 1 (DiBattista et al., 2007), enabled us to estimate apparent survival based on whether sharks were recaptured (apparent survival = 1) or not (shark assumed to have died: apparent survival = 0). To avoid underestimating survival, we assumed every shark that was never recaptured but was over 2 years old to have dispersed and survived.  Figure 1). This was due to concerns regarding a dredging project on the west side of North Sound, which could have compromised visibility and made the tests impossible. In every model presented in this study, we account for the potential effect of the testing location by adding the year as a random effect.

| Personality assay
Between 12 and 18 hr prior to the assays, sharks were fed to satiation with locally caught Sphyraena barracuda filets to avoid hunger biases in behaviour (Biro & Booth, 2009). Each trial day, six sharks were observed in a sociability assay with a 25-min duration. Sharks were then individually subjected to the novel open-field assay. Data on sociability is not presented here as the sample size was significantly lower than on the novel open-field assay due to difficulties with the observations (but see Finger et al. (2018) for repeatability of Sociability). The sociability and exploration tests could not be conducted on separate TA B L E 1 Overview of sample sizes. A count of each data point is given, followed by the number of individuals for which multiple measurements are available

PCL Growth Exploration Distance
North Sound field, the start-box was closed, and 10 min of behavioural observations initiated. We recorded the shark's position in the pen using small markers on the seafloor that divided the pen into a grid of eighteen 2 × 2 m areas ( Figure 2b). The number of areas visited per second (including revisits; 'exploration score') was previously found to represent juvenile lemon sharks' reaction to novelty and to be repeatable (repeatability adjusted for body size and year: 0.37, . Further, the order at which the sharks were tested in the open field was shown to not influence the exploration score .
We included in this study only the first exploration score per year per individual, as lemon sharks are known to show short-term habituation to the novel open-field test (Finger et al., 2016). As a result, some sharks have two measures of explorations (see Table 1), but never in the same year (mean number of measurements per individual = 1.13, min = 1, max = 2).

| Risk-taking measurement in the wild
The novel open-field assay offered a quick, standardized way to obtain behavioural measurements on many individuals, but it is unclear whether it is relevant to natural behaviours. We measured individual risk-taking behaviour in the wild using acoustic telemetry on a subset of the sharks tested in the novel open field. For juvenile lemon sharks, the mangrove shore in Bimini constitutes an important protection from predators (Guttridge et al., 2012;Hussey et al., 2017). However, the main prey of lemon sharks (e.g. mojarras Gerreidae, grunts Haemulidae, Newman, Handy, & Gruber, 2009) also benefit from the refuge within the complex structure of this habitat. As a result, sharks have higher foraging success in open habitats such as seagrass patches as opposed to mangrove edges (R. Bullock, 2013, unpubl. data) and consequently have higher growth rates (Hussey et al., 2017). The decreased potential for sharks foraging away from the shore to escape predators within mangrove roots suggests the existence of a trade-off between efficient foraging and refuging (Hussey et al., 2017). In this context, we assume the distance juvenile sharks swim away from the shore to be a good proxy for their willingness to take risks, which is likely to be associated with a growth-mortality trade-off.
We randomly chose sharks within a size-matched group of individuals that had received the novel open-field assay to minimize size differences (PCL: M = 53.3 cm ± 1.8 SD). We surgically implanted acoustic transmitters (CT-82-2-I, battery life 14 months, Sonotronics) into the body cavity of sharks via a 2-3 cm incision made a few centimetres anteriorly to the pelvic fins using sterilized instruments. We closed the incision with two or three sterile monofilament resorbable sutures (75 cm × 24 mm; CP Medical ® ; see Kessel et al. (2014) for detailed methodology). Sharks were placed into a state of tonic immobility (a state of lethargy, comparable to hypnosis, Watsky & Gruber, 1990) for the entire procedure. This was preferred to the use of anaesthetic since sharks are still able to strongly ventilate their gills under tonic immobility and recovery is immediate after the surgical procedure (Kessel & Hussey, 2015). The implanted electronic transmitters weighted 9.5 g which did not exceed 2% of the sharks' body weight (as recommended by Kilfoyle & Baggeroer, 2000). After the procedure, sharks were kept in captivity for 7 days and were monitored for their healing daily. We then released them at their site of capture.
We tracked sharks in the North Sound during two consecutive  (Hansell et al., 2018).
From September to the end of April, we proceeded to the area where the tagged sharks lived using a 16 ft flat bottomed skiff powered by a 40-horsepower outboard engine. This was done twice a week, as long as the weather allowed it. To avoid observer biases in the search procedure, we followed a standardized transect to locate the sharks (see Appendix S1 for transect and exact methodology) with speed maintained below 2.5 m/s and usually completed the transect in 7 hr.
Once a shark was found, we recorded its identity, the GPS (Global Positioning System) coordinates of the boat (decimal degrees), the water depth (cm), the bearing to the shark (degrees) and estimated distance (m) from the boat to the shark. The position of the shark was then extrapolated from the position of the boat using the Haversine formula (Sinott, 1984). We then calculated the distance from the shark to the nearest shore using the 'gDistance' function provided in the rgeos package in r (version 3.5.2; Bivand & Rundel, 2017; R Core Team, 2017). Only sharks that had been encountered on more than five tracking events were included in the dataset. Therefore, if a tag was malfunctioning or a shark died or was predated upon early in the study, it was removed from the dataset. Consequently, following data filtering, the number of individuals for which distance from shore data was available in North sound was n = 19 (2014: n = 10, 2015: n = 9) and in Sharkland was n = 27 (2016: n = 12, 2017: n = 15; see Table 1). On average each shark was measured 22 times for distance from shore (min = 5, max = 65).

| Statistical analysis
All statistical analyses were performed in the r environment (R Core However, these methods call for a high amount of repeated measures within individuals and our dataset was not sufficient to employ them (see Table 1). We therefore present below univariate methods that are more suitable to the data used for this study.
We used two sample t tests to compare growth, distance from shore and exploration score in the two subpopulations. To verify whether both subpopulations presented the same diversity of phenotypes, we compared the variance of growth, distance from shore and exploration using F tests. The difference in apparent survival between the subpopulations was tested using a chi-square test.
All generalized linear mixed models described below were performed in the MCMCglMM package (Hadfield, 2010). For all models, we set the number of iterations to 330,000 with a thinning interval of 100 and discarded the first 30,000 iterations. We obtained Monte Carlo Markov Chains with a sample size of 3,000 and with low autocorrelation. Each model was fit using a weakly informative inverse gamma prior. We used a gaussian error distribution for every model, except for the logistic regressions which were fit using the 'categorical' family. Growth, exploration and PCL were mean centred and standardized to units of one standard deviation within each year and subpopulation to set them on a similar scale and meet normality assumptions. Distance from shore was square root transformed to meet normality assumptions.
To test whether each individual's exploration score obtained in the novel open-field assay was a relevant indicator of the distance travelled from the shore (i.e. assumed to represent risk-taking) we modelled distance from shore (square root transformed) as a function of exploration score (standardized) with an interaction with subpopulation (factor). We included the interaction to allow for the detection of differences in the relationship between distance from shore and exploration between the subpopulations. Because we had on average 22 measures of distance from shore and 1.13 measures of exploration per individual, we used the value of exploration measured before acoustic tracking for each individual and their mean distance from shore throughout the study.
To test whether the exploration score predicted growth in both subpopulations we modelled growth (standardized) as a function of exploration (standardized) with an interaction with subpopulation (factor). We accounted for yearly differences and repeated measures within individuals by including year (factor) and PIT Tag number (factor) as random effects. We tested whether distance from shore predicted growth using a similar model but without random effects: the mean distance from shore of each individual that received an acoustic transmitter was used as a fixed effect, with an interaction with subpopulation.
We then determined natural selection on the traits of interest according to the method developed by Lande and Arnold (1983).
Because only sharks that had survived a year had a growth estimate, we also used PCL (which has a much larger sample size, see Table 1) in the selection analysis, as DiBattista et al., (2007) previously found by selection on growth and on body length (i.e. PCL) to act in the same direction in our system. This is likely because sharks' growth at the end of an interval is correlated to their pre-caudal length at the beginning of the interval (Pearson's ρ = 0.16, p < 0.0001). In addition, obtaining fast growth and maintaining a large body likely requires high resource acquisition (Blanckenhorn, 2000), resulting in both of those traits being selected against through risk-taking and exploration. We contend in this case that selection on PCL can be expected to be informative of a growth-mortality trade-off when sample size on growth does not allow for its detection.
To determine selection on growth, body size and exploration score we calculated standardized linear selection gradients (β′) and standard- growth, PCL and exploration. In our case, the available sample size for PCL was much higher than for growth, which was much higher than for exploration (see Table 1). Therefore, we calculated selection for year to the next, apparent survival would always be '1' for sharks that had an available growth measurement. Therefore, any model including growth was made using apparent survival to year n + 2 rather than to year n + 1. To control for potential survival and growth benefits of elongated stay in captivity due to testing in the open field we also ran selection models on subsets of the data containing only sharks that were never tested in captivity.
Because survival is binomially distributed, we used logistic regressions for significance testing (Réale & Festa-Bianchet, 2003). All selection models were computed for each subpopulation separately given difficulties of model convergence. Therefore, we checked whether the selection gradients and differentials were different between populations using post-hoc t tests on the posterior distributions of each model. We accounted for yearly differences and repeated measures within individuals by including year (factor) and PIT Tag number (factor) as random effects in all selection models.
Selection gradients and differentials could not be computed on distance from shore due to low number of individuals (see Table 1).
However, we computed a logistic regression of survival on distance from shore with an interaction with subpopulation to estimate the direction of selection on this trait and its significance. Similar to the models including distance from shore above, we used the mean distance from shore of an individual and its survival at the end of the tracking season.
The data used for this analysis are available on the Dryad Digital Repository https://doi.org/10.5061/dryad.ksn02 v71r  and an R markdown file is available as Supporting Information (Appendix S2).

| Do fast explorers in captivity swim further from shore in the wild?
Sharks from both subpopulations did not score differently in the novel open-field test (mean North Sound = 0.26 ± 0.05 SD, mean Sharkland = 0.26 ± 0.04 SD, t 321.6 = 1.15, p > 0.05), but there was a difference in variance between the subpopulations, with North Sound having a higher variance than Sharkland (F 26,174 = 1.48, p = 0.01, Figure 3a). Sharks from North Sound were found significantly further away from shore (mean square root distance = 8.74 ± 4.48 SD) than sharks from Sharkland (mean square root distance = 5.12 ± 3.13 SD), t 1,027.5 = 15.26, p < 0.001 (Figure 3b). Additionally, the variance in distance from shore of the two subpopulations was different with North Sound presenting the most variance (F 163,596 = 2.05, p < 0.0001, Figure 3b).  Figure 4).

| Do fast explorers grow faster?
Growth rate was significantly different between subpopulations, sharks from Sharkland grew faster than North Sound sharks

| Are fast explorers, fast growers and larger sharks selected against?
Apparent survival was significantly higher in North Sound (mean proportion of surviving sharks per year = 0.43 ± 0.09

SD) than in Sharkland (mean proportion of surviving sharks per
year = 0.41 ± 0.11 SD, Chi-squared = 5.23, p = 0.02). We found significant negative selection differentials and gradients on PCL and growth in North Sound for all sub-samples of the data apart from those containing only sharks tested for personality (Table 2; Figure 6a-c). In Sharkland, we also found significant selection on growth and PCL in every sub-sample apart from the ones containing only the sharks tested for personality (Table 2). Additionally, selection on PCL was also significant in Sharkland when only sharks tested for personality that had a growth measurement were included in the model. This means that smaller sharks and slow growers survived better in both subpopulations (Figure 6a,b). The gradients and differentials were always significantly different in the subpopulations with selection differentials on PCL being stronger in North Sound than Sharkland and with selection gradients being generally stronger in Sharkland than North Sound (see Table 2). We found no selection on exploration in either of the subpopulations (Table 2; Figure 6c).
Since DiBattista et al. (2007) made their selection analysis on relative growth (proportion of the body length gained from one year to the next) rather than absolute growth, we provide in Appendix S3,  Figure 6d).

| D ISCUSS I ON
Here Note: Sample sizes are given in parenthesis. Significant results are shown in bold (.p = 0.05, *p < 0.05, **p < 0.01, ***p < 0.001). a Sample only includes sharks that were never tested for personality. b Selection measured using survival at year n + 2 due to the inclusion of the growth variable.
TA B L E 2 Linear (directional) selection differentials (S′i) and gradients (β′) on precaudal length, growth and exploration of sharks from North Sound and Sharkland explorative sharks would be selected against as part of a growthmortality trade-off. In both subpopulations, we found that selection acted against large body size and faster growth, confirming the presence of a trade-off between growth and mortality. But we found that sharks that were more active in the novel open-field assay, took higher risks in the wild and had higher subsequent growth rates only in North Sound (which has lower predator abundance). Neither subpopulation showed selection directly on personality. Therefore, our results provide strong support for trade-offs between growth and mortality but suggest that behaviour's role in mediating this tradeoff is less clear.
In North Sound, our results are in accordance with predictions we made from the POLS hypothesis: faster explorers take more risks, grow faster and are more likely to die. The maintenance of interindividual differences in risk-taking behaviour and growth rates in this subpopulation can be explained via the presence of a growth-mortality trade-off. In goldfish C. auratus risk-prone individuals, grew faster and in cavies C. aperea exploration of a known environment predicted growth (Balaban-Feld et al., 2019;Guenther, 2018). Similarly, juvenile lemon sharks who explore more, take more risks and profit from higher growth rates. This has the potential to make them reach a size at which they are less susceptible to predation faster (sizeselective mortality, Sogard, 1997) and to make them reach sexual maturity earlier than slower-growing conspecifics. These advantages are mitigated by higher exposition to predation before they reach a less dangerous size class. In our case, the trade-off is likely favoured because open habitat foraging areas such as seagrass patches are more productive but also more dangerous (DiBattista et al., 2007;Hussey et al., 2017) which is supported by our finding that sharks that swim further from shore have lower apparent survival. Bigger risk-takers in the wild were not found to grow faster, but we contend that this could be due to a lot of risk-takers dying and therefore not having a growth measurement at the end of the tracking year. It is worth noting however that we observed a loss of significance of the selection on growth and PCL when only sharks tested for exploration were included in the model and that exploration was never selected against in our models. Previous work on selection has shown high probability of making type II errors (i.e. failure to reject the null hypothesis) when selection is weak and sample size is low. For instance, 700 data points were necessary to have a 0.8 probability of detecting a selection of 0.12 (Hersch & Phillips, 2004). In this case, we contend that the absence of significance for selection in smaller sample sizes could be due to a lack of power. This seems especially reasonable for growth and PCL because inspection of survival rates ( Figure 6a,b) reveals that the sharks tested in the novel open field were not driving the significance found in the larger samples.
In Sharkland, a growth-mortality trade-off was also detected but exploration in captivity did not predict growth or risk-taking and survival was not predicted by distance from shore. Failure to support the POLS hypothesis can be for several reasons. First, several different behaviours could covary with life history trade-offs (Montiglio et al., 2018). The constraints placed on predator avoidance and resource acquisition affect which behaviours covary with life history and which do not. In our study, we measured animals' exploration in F I G U R E 6 (a) Pre-caudal length, (b) growth rate, (c) exploration and (d) mean distance from shore as a function of apparent survival in North Sound and Sharkland . The original data points are shown in light grey. For pre-caudal length and growth, we highlight in colour the data points of sharks tested for personality. Significance was determined by logistic regressions at the maximum sample size available (.p = 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, see Table 2) semi-captivity and their distance from a refuge in the wild. Exploration and risk-taking behaviours can be expected to covary with life history when an animal's vulnerability to predators depends on its ability to avoid predator encounters and/or when resources are clumped or ephemeral (Montiglio et al., 2018). In North Sound, sharks that had higher captive exploration, were found further from shore and were less likely to survive, which is likely due to increased predator encounters. Fast explorers also grew faster, suggesting increased resource encounters, which points to potentially low resource abundance or clumped resources. But in Sharkland, distance from shore did not predict survival, and sharks were mostly found close to the shore, where they are less vulnerable to predators. Furthermore, their growth was not predicted by exploration, suggesting that exploration does not impact resource acquisition. It seems possible in this case that in Sharkland, the life history trade-off is mediated through different behaviours than the one we investigated.
The association (or not) of traits could also be driven by feedbacks between behaviour and state (Dingemanse & Wolf, 2013).
State-dependent feedbacks on behaviour predict that animals behave according to their state (e.g. energy reserves), and such feedback mechanisms also depend on environmental conditions such as predation risks and resource availability (see Luttbeg & Sih, 2010).
Here, the subpopulations did not differ in mean exploration scores in captivity, where resource availability and predation risk are the same for every individual. But exploration in the wild was different between the populations and we know them to differ in their predator abundance (with Sharkland having more predators) and in their growth rates (with sharks from Sharkland growing faster). Both high predator pressure and high resource availability can be hypothesized to promote cautious behaviour via negative feedbacks (Luttbeg & Sih, 2010). If resources are abundant and accessible, sharks may have higher energy reserves and be less likely to take risks while foraging (Biro & Booth, 2009). If predators are abundant and foraging is dangerous, sharks may take risks while foraging only when their energy reserves are low (Biro, Abrahams, Post, & Parkinson, 2004). Even if sharks in Sharkland were rarely found far away from the mangrove edge, and thus displayed more risk-averse behaviours, their growth rates were still on average higher than in North Sound. This suggests that these individuals did not need to exhibit typically risky behaviours to gain the resources needed to maintain their faster growth rates. This seems especially likely as we occasionally observed juvenile lemon sharks foraging close the mangrove edge while tracking in Sharkland, but never made such an observation in North Sound.
A recent description of the faunal community in Sharkland found that a larger abundance of fish is found within 50 m of the mangrove edge than offshore, further supporting this (Grimmel et al., 2020).
In this case, in Sharkland, low distance from shore could be maintained via a negative feedback due to their higher state (illustrated here by faster growth). In other words, sharks from Sharkland might not display their full range of behaviour because the environment they live in negatively impacts risk-taking behaviours. This issue is especially interesting because Sharkland sharks did not exhibit lower exploration in captivity, despite being found closer to shore, which suggests that the diversity in exploration and in distance from shore in Sharkland is highly plastic (i.e. adjusted on the short term to the current conditions) or is promoted through different mechanisms.
The fact that our exploration measure in captivity did not predict risk-taking in the wild in one population is of further interest.
Previous studies have stressed that novel open-field assays are often hard to interpret, as an animal's behaviour in a novel environment might represent its willingness to explore and/or its reaction to a stress and/or its general activity (Carter et al., 2013). Here, we expected it to predict risk-taking in the field, and our prediction was confirmed in one subpopulation only. Our study emphasizes the need for researchers to carefully validate their personality assays, as the ecological relevance of the assay here was different between two geographically adjacent habitats.
It is important to note that despite providing important new insights into the understanding of a topic that has recently received much research interest , our study has several limitations. Our dataset was only collected in two subpopulations, making firm conclusions on differences between populations impossible. We studied the exploration personality of 28 juvenile lemon sharks in a third Bimini subpopulation with high predator abundance at the beginning of the project. Unfortunately, the main refuge for this nursery was damaged in 2014, leading to a depletion of the lemon shark population and the impossibility to conduct further testing. Continued work on lemon sharks could benefit from including more subpopulations from neighbouring areas. In addition, the lack of repeated measures within individuals in our dataset prevented the use of multivariate analysis and structural equation modelling which are highly suitable for studies on trait association and POLS hypothesis (see Moiron et al., 2019;Polverino et al., 2018;Santostefano et al., 2017). Therefore, we could not further investigate the causal relationship between traits. Finally, the POLS hypothesis usually encompasses phenotypic traits that are involved in a trade-off between current and future reproduction . Lemon sharks reach sexual maturity at approximately 12 years of age (Chapman et al., 2009), and here, logistical constraints compelled us to use sharks of 0 and 1 year. It is important to keep in mind that our study only investigates a part of lemon sharks' life history and thus offers a partial test of the POLS hypothesis.
In conclusion we tested a part of the POLS hypothesis using a novel open-field assay, measuring risk-taking in the field, and testing a growth-mortality trade-off in wild juvenile sharks. We found that the prediction that personality-mediated growth-mortality trade-off via wild risk-taking behaviour was present in one subpopulation but not the other, suggesting that the emergence of personality driven life history trade-offs is context dependent.
Importantly, the subpopulations of juvenile sharks are known to differ in their predator abundance. We discuss mechanisms driving the conditional association of behaviour and life history and suggest that they are driven by predator abundance but also likely by resource availability. Understanding the processes that promote personality driven life history will require scrutinizing behaviour and life history across ecological gradients. Long-term, multi-population datasets from the wild and controlled captive experiments where conditions can be manipulated will be crucial to our future understanding of how behaviour can mediate life history.

ACK N OWLED G EM ENTS
We are thankful to all the staff and volunteers at the Bimini Biological Field Station who helped in collecting the data presented in this study over the years. In particular, we would like to thank