Fitness consequences of fish circadian behavioural variation in exploited marine environments

Introduction

Humans have exploited fish populations through trait-selective harvesting since the origin of our species (Allendorf & Hard, 2009). In fact, fishing is widely recognised today as a major driver of contemporaneous evolution and trait change in wild fish populations (Sullivan, Bird & Perry, 2017). There is substantial evidence that size-selective harvesting (e.g. gear selectivity) usually selects for fast life-histories and favours early maturation and high reproductive investment (Alós et al., 2014; Heino, Pauli & Dieckmann, 2015; Laugen et al., 2014; Matsumura, Arlinghaus & Dieckmann, 2011). The behavioural dimension of fisheries selection has recently gained interest among fisheries scientists and managers due to the growing evidence of consistent between-individual differences in the behaviour of exploited fish and the study of selection in real fisheries (Arlinghaus et al., 2017; Diaz Pauli & Sih, 2017, Uusi-Heikkilä et al., 2008). Currently, there is a large quantity of literature demonstrating the existence of consistent (in temporal and ecological contexts) between-individual differences of fish behavioural traits, such as boldness or aggressiveness, that define behavioural types within fish populations (Conrad et al., 2011; Mittelbach, Ballew & Kjelvik, 2014). In addition, with the recent development of aquatic telemetry, fisheries scientists have a powerful tool available to study behavioural types of free-living fishes (Hussey et al., 2015; Lennox et al., 2017a) and how fisheries may promote the selection of behavioural types in real-world fisheries (Alós et al., 2016b; Monk & Arlinghaus, 2018; Olsen et al., 2012). Together, these two developments have generated substantial empirical evidence demonstrating that bold and high-exploratory individuals (Alós, Palmer & Arlinghaus, 2012; Biro & Sampson, 2015; Härkönen et al., 2014; Klefoth, Kobler & Arlinghaus, 2011; Olsen et al., 2012) are more prone to harvest; thus, this evidence supports the idea that timidity syndrome can give rise to exploited fish populations that are composed of shy, less active and less exploratory individuals (Arlinghaus et al., 2016, 2017).

Surprisingly, behavioural traits that determine timing have been poorly considered in the context of the selective properties of fishing. Recently, Tillotson & Quinn (2017) proposed the timing of migration or breeding as candidate traits that are targeted by fisheries selection. Both the timing of migration and the timing of the breeding season have strong impacts on population dynamics (Lowerre-Barbieri et al., 2017), and selection imposed by these traits would strongly impact the long-term trajectory of the fish stocks. Similarly, an ubiquitous behaviour related to timing in fish that has been overlooked by the scientific fisheries community is the manifestation of underlying circadian rhythms. Life on earth is governed by a 24-h rotation cycle that has led the evolution of endogenous circadian clocks across taxa, including fish species (Kreitzman & Foster, 2005). Similar to behavioural types, humans and some terrestrial animals show temporally consistent between-individual variation in different circadian-related behaviours, such as awakening time or sleep onset, that are the result of the interactions between those endogenous individual circadian clocks and the environment; furthermore, these interactions define chronotypes (Roenneberg et al., 2007; Bloch et al., 2013; Rattenborg et al., 2017). Although chronotypes should be ubiquitous across animal taxa, only a few studies have demonstrated the existence of chronotypes by exploring the amount of behavioural variation explained by between-individual differences (Randler, 2014), and these studies have mainly focused on bird species (Dominoni et al., 2013; Steinmeyer et al., 2010; Stuber et al., 2014, 2015).

Regardless of whether fish sleep or not (Reebs, 1992), most fish species show a circadian-related behaviour in which they change from an active state to a resting state, leading to a ‘sleep-like’ behaviour that is consistent with the sleep architecture observed in mammals (Schmidt, 2014; Siegel, 2008). This diel active/resting cycle is widely observed in free-living fish across species (Krumme, 2009; Alós, Cabanellas-Reboredo & Lowerre-Barbieri, 2012, Alós et al., 2016b; Koeck et al., 2013). Recently, Alós, Martorell-Barceló & Campos-Candela (2017) found the first evidence supporting the existence of chronotypes in fish focused on the pearly razorfish, Xyrichtys novacula. Similar to humans and birds, fish chronotypes arise from between-individual differences in circadian behavioural traits that are consistent over time and ecological contexts (Fig. 1). Far from being anecdotal, chronotypes have been frequently linked to many fitness processes in terrestrial animals, such as predation mortality or finding a reproductive mate (Roenneberg, Wirz-Justice & Merrow, 2003, and see review by Adan et al., 2012), and any directional selection pressure (i.e. either natural or human-induced) acting on chronotypes could lead to trait changes in terms of circadian behavioural rhythms (Helm et al., 2017). In fact, one recent study demonstrated how a potential environmental-induced change in a behavioural trait can influence circadian behavioural variation and impact fitness (Dominoni et al., 2013), i.e. city birds that started their activity earlier than their forest conspecifics highlighted that urban environments (i.e. those with artificial lighting) can significantly modify biologically important rhythms in wild organisms and explained the potential reproductive advantages conferred to the early-rising birds in such an artificial environments. Similarly, we assumed here that early-riser fish chronotypes would be more vulnerable to fishing simply because the number of encounters between the fish and fishers was expected to be higher.

Based on this assumption, the objective of this work was to explore the plausibility of selection acting on fish chronotypes using a spatially explicit individual-based model (IBM). Our IBM assumed relatively simple movement rules that dictated the encounters between fish and fishers, and it was based on the real properties of a general bottom coastal fishery; additionally, the IBM explicitly incorporated social-ecological factors to add realism to our model (and simulations). The selection gradient (S), as a central measure of selection in traditional quantitative genetics with heritability (Price, 1970), has been widely used to describe trait changes in commercial and recreational fisheries (Alós et al., 2016b; Monk & Arlinghaus, 2018). We aimed here to estimate mean-standardised selection gradients on circadian behavioural traits to determine whether they were different from zero, and we compared them with previously reported gradients of other traits. Although the economic consequences of fisheries selection can be addressed by proper fisheries management (Eikeset et al., 2013), it can generate undesirable consequences in terms of ecosystem functioning (Audzijonyte et al., 2013; Jørgensen et al., 2007); specifically, this selection can notably reduce the recovery of overexploited stocks (Uusi-Heikkilä et al., 2015; Walsh et al., 2006) and decrease the recreational utility of fisheries (Sutter et al., 2012). Therefore, our final objectives were to make broader predictions about our findings and to stimulate research on the topic by providing a list of empirical research needs to fully disentangle the causes and consequences of fish chronotypes in exploited environments.

Materials and Methods

To explore whether fishing selection influences circadian behavioural traits, we developed a computational IBM where a fish population spatially behaves in a 2-D landscape and is exploited by a fleet of fishing boats during a fishing session (see Fig. 2 and video in SM1). Our IBM is spatially explicit because fish and fishers move (i.e. change position every minute) across the landscape according to different types of movement models. Encounters between the fish and fishers determined the mortality of the fish. Although encounters between fish and fishers do not always predict harvest (Monk & Arlinghaus, 2018), these encounters are among the most important components related to the vulnerability of most fishes (Lennox et al., 2017b), especially in bottom coastal fisheries (Alós, Palmer & Arlinghaus, 2012; Alós et al., 2016b). Our model was built under a prototypical bottom fishery where (i) target fish performed a sedentary behaviour that lead to the establishment of a home range (HR) area, (ii) the centres of activity were patchily distributed and formed a patchy landscape (which could be the consequence of a fragmented habitat), and (iii) fish were exploited by a fleet of mobile fishing boats. Our model was parametrised using empirical data from a popular recreational baited hook-and-line fishery located in Mallorca Island (Spain) that targeted pearly razorfish (see full details in Alós et al., 2016b); however, the model is generalizable to any other system that displays these three main properties. Our computational IBM simulation was implemented and run in R (R Core Team, 2017). The R code is provided in the Supplementary Material (SM2).

Results

The first simulation scenario that considered real wild circadian behavioural variation (i.e. observed fish chronotypes) adequately reproduced the existence of chronotypes (Fig. 1). The R scores in this scenario were 0.43 [0.37–0.6] for awakening time and 0.45 [0.39–0.6] for rest onset, which were similar to the scores obtained from the real data by Alós, Martorell-Barceló & Campos-Candela (2017). Fish started their activity as late as 400 min after sunrise, and among-individual differences in awakening time were clearly recognisable, enabling the identification of an early-riser chronotype (Fig. 1). In contrast, fish finished their activity within a shorter period (up to 20 min after sunset), but some individuals extended their activity by an average of a few minutes according to the real data (Fig. 1).

The mean and s.d. of the four behavioural traits in the initial and exploited populations are shown in Table 1. In total, 650 individuals survived (exploitation rate = 67.5%) in the simulation scenario, and in general, the exploited population was composed of individuals with later awakening times, similar rest onset times, smaller HRs and slower exploration (Table 1). These results generated significant S_μ that differed from zero in terms of awakening time (mean S_μ = 0.85), HR size (mean S_μ = −0.52), and exploration rate (mean S_μ = −0.22) (Table 1). These results were consistent along the simulated gradient of fishing effort, and the strength of significant S_μ values increased as fishing effort increased (Fig. 3).

In the second simulation scenario, i.e. where no fish chronotypes were simulated, the number of surviving individuals was 315 (exploitation rate = 84.2%). In this case, the exploited population was composed of individuals with similar awakening times and rest onset times and smaller HRs and exploration rates (Table 1). These results generated significant S_μ different form zero values for only the HR size (mean S_μ = −0.49) and exploration (mean S_μ = −0.36, Table 1), and we discarded significant S_μ that differed from zero for the circadian behavioural traits (Table 1). These results were also consistent across the simulated gradient of fishing effort (Fig. 3). Therefore, we assumed that the results from the observed fish chronotypes simulation scenario were caused by factors other than circadian behavioural variation.

Discussion

Circadian behavioural variation have important implications for individual fitness, and many eco-evolutionary trends are dependent on the realised expression of circadian rhythms (Roenneberg, Wirz-Justice & Merrow, 2003; Wicht et al., 2014); however, very little information is known about the consequences of fish chronotypes. Here, we found that fishing selection may influence the variation in circadian behaviours by differentially harvesting early-riser chronotypes, and the strength of this selective process is linked to the fishing pressure. We demonstrated these potential consequences of fish chronotypes in exploited environments using a novel social-ecological IBM. IBMs are especially appropriate for formulating and testing emergent population properties from individual processes in predator–prey systems (Barbier & Watson, 2016; Watkins & Rose, 2017), including fisheries (Alós, Palmer & Arlinghaus, 2012). The R score or the within-population behavioural variation are classic examples of an emergent population property from individuals, and this value makes IBMs particularly suited to test our hypotheses (Bell, Hankison & Laskowski, 2009). In addition, our IBM allowed us to test our working hypotheses using two different ecological simulation scenarios using real data and a wide range of fishing pressure scenarios. Therefore, we feel our approach, although theoretical, properly reproduce some of the potential fitness consequences of circadian behavioural variation in exploited marine environments and provide novel insights in the selective properties of fishing.

The results of the first simulation scenario, i.e. that which used real wild circadian variability, revealed a significant selection gradient in terms of the awakening time. Fish that survived the simulated fishing season were clearly not a random sample of the initial population, and early-riser chronotypes were more prone to capture by the fleet of boats. This finding adds a new variable to the complex concept of the vulnerability of fish to fishing (Lennox et al., 2017b). This result was consistent across all fishing pressure levels, suggesting that even in low fishing pressure scenarios (i.e. two boats per km²), fishing selection may influence circadian behavioural traits. In fact, the strength of selection was expected to increase as fishing pressure (i.e. mortality) increased. In contrast, no evidence was found for any selective properties regarding the time of rest onset, which was likely related to the fact that simulated fishing activity mainly occurred during the daytime. In the second scenario, where no wild circadian behavioural variation was simulated, the selection gradient of the awakening time was not significantly different from zero, confirming that chronotypes were major drivers of selection force; this result agrees the results of the first simulation scenario that was based on real-world data.

The potential for eco-evolutionary changes in chronotypes under human pressure has been recently proposed (Helm et al., 2017). In fact, Dominoni et al. (2013) demonstrated that city European blackbirds, Turdus merula began their activity earlier and had faster circadian oscillation than did their forest con-specifics. The results by Dominoni et al. (2013) suggested that humans (through artificial lighting) may have selected for individuals by favouring those with large circadian period lengths. In this example, the selective force imposed by artificial lighting acts in the opposite direction than that of our working hypothesis. In our work, the selective force is imposed by the timing of the fishing pressure (Fig. 2); thus, the selective force should favour small foraging periods. What is relevant in this context is that both artificial lighting in the city and fishing pressure in the sea may impose selection gradients in circadian behavioural traits and may act as eco-evolutionary drivers in wild populations, and this information should be further studied (Helm et al., 2017). In addition, our work provides the first evidence that suggests fishing may play a role in the circadian rhythms found in oceans.

Our theoretical selection gradients were mean-standardised, which allowed them to be compared with other traits. First, we found significant selection gradients in the two spatial behavioural traits considered here, indicating selection against large HRs and fast exploration rates. Although both were smaller than the values obtained by the circadian behavioural traits, significant selection gradients were consistent between the two simulation scenarios and across all fishing pressures. Interestingly, the direction of selection was consistent with the empirical selection gradients of hook-and-line recreational fisheries on these spatial behavioural traits, indicating that our IBM was robust (Alós et al., 2016b). Moreover, the strength of the obtained selection gradient on awakening time was also stronger when compared with other life history (S_μ = 0.66) and morphological (S_μ = 0.29) traits that have previously been reported (Hereford, Hansen & Houle, 2004), although in function of the fishing scenario simulated. In fact, the strength of selection may vary according to the morality pressure, as revealed by the different fishing effort scenarios in our simulations. This fact highlights the relevance of estimating selection gradients in real populations that are exposed to mortality pressure and the importance of using realistic scenarios (i.e. those based on data from the wild). However, our results demonstrated that the potential of selection on circadian behavioural traits certainly exists, and the selection strength could be similar or even stronger than that of previously considered traits.

Although our work is mainly theoretical, we can derive some ecological implications about the selective properties of fishing acting against early-riser chronotypes. Chronotypes are important determinants of reproductive success in birds; for instance, females choose males with early awakening times (Helm & Visser, 2010). Assuming this also occurs in fish, one could predict a reduction in the overall reproductive output of a population due to the absence of highly reproductive early-rising males. In addition, fish such as the pearly razorfish play a key role in the food-web by preying on other taxa (Castriota et al., 2005) and serving as prey for larger animals, such as dolphins. Thus, a change in the daily timing in a population of pearly razorfish could induce foraging behavioural changes with impact in the lower and upper levels of the food-web. We can also speculate that fishing-induced selection against early risers is currently occurring, and the results observed by Alós, Martorell-Barceló & Campos-Candela (2017) are the result of such selective processes. Therefore, we suggest that the ecological consequences of the selective properties acting on circadian behavioural traits are plausible but may already be occurring. In all cases, there is a need to delve into the causes and consequences of fish chronotypes selection, and more empirical work is needed in clarifying the ecological consequences (Bloch et al., 2013; Helm & Visser, 2010).

The selection gradient is, however, only one component that addresses trait change and derives eco-evolutionary trajectories (Price, 1970). The heritability, or the degree of variation in a phenotypic trait in a population caused by genetic variation between individuals, is a key component that can be used to forecast the population-level consequences of any mortality pressure (including fishing). There is no information on the heritability of chronotypes in marine fish. However, Helm & Visser (2010) quantified the heritability of the chronotypes in the great tit, Parus major to be 0.86, which is certainly high. In addition, our study is a computational simulation, and it is possible that our results are overestimations because we did not consider other sources of mortality or connectivity; furthermore, we did neither consider other traits that may experience fisheries selection (e.g. size, personality-related behavioural traits, age), nor quantified the fitness in terms of expected reproductive lifetime (i.e. cumulated offspring). The early-life stages of the pearly razorfish are pelagic, and the connectivity of the surrounding non-exploited populations should be integrated to estimate the selection gradients (Alós et al., 2014). Therefore, there is a need to provide empirical data to support our predictions and to develop more complex meta-population dynamics that provide a more accurate view of the strength of the selection gradients on the circadian behavioural variation. Next-generation individual-based ecological models that aim to make predictions in a changing world would help in this task by accounting for spatial and temporal resources that merge individual fish and fisher behaviour and bioenergetics with potential micro-evolutionary adaptations (Ayllón et al., 2016).

Conclusion

Our work demonstrates that the timing associated with fleet activity may generate significant selection on fish circadian behavioural traits. In fact, the direct selection acting on chronotypes can indirectly be a mechanism of fishing selection on migration or breeding behaviours (Graham et al., 2017). Therefore, our work proposes a novel view for understanding the selection properties of fishing acting behavioural traits and generates a list of research needs.

First, we should explore how widespread chronotypes are across fish taxa. Adequate technology and approaches used to measure chronotypes in the wild is certainly available (Alós, Martorell-Barceló & Campos-Candela, 2017; Helm et al., 2017; Rattenborg et al., 2017), and further work should also consider nocturnal species or species that focus their activity during the crepuscular hours to evaluate the generality of our findings.

Second, there is a need to validate our theoretical predictions by performing selection experiments in the wild, where fish are monitored while they are being exploited by real fishers (Alós et al., 2016b); additionally, this should include different fleet timing dynamics and fish-fishers behavioural interactions. This future work should also help disentangle the synergistic effects of predation risk and fishing from the potential eco-evolutionary dynamics generated by the existence of circadian behavioural variation.

Third, we should identify the mechanisms behind the expression of wild circadian behavioural variation. Chronotypes are the emergent pattern of the interaction between circadian clocks and the environment, which includes potential light entrainment and responses to predation risk, and their study requires a combination of field and laboratory experiments (Helm et al., 2017). In addition, we should explore the plasticity and additive genetic variation (including its heritability) of fish chronotypes to evaluate the potential for evolution in circadian behavioural traits. In a quantitative genetic way, one potential route would be the exploration of candidate genes and polymorphisms linked to chronotypes, such as the CLOCK or the NPAS2 genes (Stuber et al., 2016), and how they are translated across generations (Helm & Visser, 2010; Zhang et al., 2017).

Fourth, in our previous study, we found chronotypes as an independent axis of activity as fish personality trait (Alós, Martorell-Barceló & Campos-Candela, 2017). However, there is a need to extend our research to other fish personality traits, such as boldness, exploration, aggressiveness or sociability (Conrad et al., 2011), and their feasible interactions. This would help us understand the role of circadian rhythms in the architecture of behavioural variation in fish. In addition, it would be helpful to explore how chronotypes are correlated with individual growth (productivity) or reproductive success, as it has been done with other behavioural traits (Biro & Stamps, 2008).

Once this information and population-based approaches become available, we will be able to forecast the relevance of eco-evolutionary consequences of the wild circadian behavioural variation and how human is affecting it. We hope our work stimulates research and debates on this topic.

Supplemental Information