Concede or clash? Solitary sharks competing for food assess rivals to decide

Abstract

To adapt to their environment, organisms can either directly interact with their surroundings or use social information (i.e. information provided by neighbouring individuals). Social information relates to the external features of surrounding peers, and little is known about its use by solitary species. Here, we investigated the use of social cues in a solitary marine predator by creating artificial aggregations of free-ranging sicklefin lemon sharks (Negaprion acutidens). Using a novel monitoring protocol, we analysed both dominance interactions and tolerance associations between sharks competing for food in relation with the number, the morphology and the behaviour of rivals. Sharks produced more agonistic displays and spent more time around the bait as competitors were more abundant. Moreover, the morphological attributes of competitors had very limited influence on the structure of shark social interactions. Instead, sharks appeared to establish tolerance relationships with competitors according to their individual behaviour. Furthermore, the more two sharks were observed together at a given study site, the fewer agonistic interactions they exchanged. We discuss these findings as evidence of the use of social cues in a non-gregarious predatory species and suggest directions for future research.

1. Introduction

Whether they are to find food, fight a rival or avoid predators, animals must face a great deal of challenges throughout their existence. To successfully overcome such tasks, animals can acquire information on their surrounding environment by directly interacting with it, at the risk of committing mistakes that may lead them to waste energy, be injured, or even die. Alternatively, animals may rely on social information (i.e. information provided by neighbouring individuals without necessarily interacting with them [1] and that relates to their external features, e.g. number or phenotype) to tune their adaptive response [2–4].

Using social information to assess neighbouring organisms is beneficial to all species involved, even if not permanently, in social interactions. A variety of organisms such as primates [5,6], birds [7], mammals [8,9] and fish [10–14] have been reported to exploit the social cues produced by surrounding individuals for their own interest. For instance, the outcome of dominance interactions can determine the behaviour of competitors in future interactions [15] and one may avoid confrontation with peers that overtly dominated third parties (i.e. see [16,17]). As another example, individuals looking for potential partners to engage in mutualism or cooperative interactions may avoid notorious cheaters (e.g. see [14]). Social information therefore participates in shaping both intra- and interspecific social interactions. While this phenomenon has been thoroughly investigated in group-living species, our current understanding of such processes remains scarce for solitary species.

In fact, the paucity of data on the social interactions of solitary species stems from the high technical constraints imposed by the study of elusive individuals and the rarity of interactions occurring in natural habitats. To overcome this issue, we developed a novel technical approach to investigate the social interactions of a solitary marine predator in a context of competition. Through the provision of a food stimulus, we created artificial aggregations of free-ranging sicklefin lemon sharks (Negaprion acutidens), a species that has been targeted by touristic provisioning activities for several years [18]. Sicklefin lemon sharks are commonly described as solitary predators that spend the majority of their lives without the company of conspecifics and are known to only interact occasionally during mating, foraging, defence or during the juvenile stage [19]. Solitary sharks competing for a single food source are compelled to interact with peers and may putatively use social information about rivals to tune their own social interactions, with the objective to increase their chances to feed.

To investigate this, we were able to describe the structure of artificial shark aggregations in terms of heterarchy, which is defined as ‘an organisational and/or architectural continuum that incorporates elements of both networks and hierarchies’ [20]. In other words, we investigated the combination of both vertical (dominance) and horizontal (tolerance) interactions between competing sharks. The concept of heterarchy has been used independently in a vast range of disciplines such as neuroscience [21], archaeology [22], business and organizational studies [23], ecology, computer science and politics (see [20]), offering several interesting insights into the relation between the structure and the function of a given system.

According to its original definition [24], dominance relates to the outcome of confrontations where two competing individuals produce either aggressive or submissive displays, commonly referred to as agonistic behaviours. The dominant individual consistently wins agonistic interactions over the submissive individual and gains priority of access to resources [25]. Agonistic displays therefore act as labile external cues of a competitor's internal state. Furthermore, it is considered that the outcome of agonistic encounters can often be predicted by the morphological attributes of contestants, like their body weight or the size of organs related to fighting [26]. Since they are permanently available to surrounding peers, morphological attributes act as passive signals of the fighting ability of individuals and can be exploited by rivals to avoid spending energy on active signalling.

To investigate the social interactions of sicklefin lemon sharks, we analysed (i) the influence of competitor number on the agonistic behaviour and access to food of individual sharks, and (ii) the structure of hierarchical and tolerance relationships among competing sharks in relation with the morphological and behavioural attributes of rivals.

2. Material and methods

(a). Experimental setting

We conducted behavioural observations of wild sicklefin lemon sharks off Moorea, Society Islands, French Polynesia (S 17°29′ 31″; W 149°50′08″). Contrary to gregarious shark species [27,28], adult sicklefin lemon sharks have solitary lifestyles and only occasionally aggregate for foraging or mating purposes [29]. We created artificial aggregations of these predators using an appetitive stimulus and remotely recorded their behaviour while they were competing to access food. A bait cage was placed on a sandy berm on the outer slope of the reef at approximately 20 m depth at two different shark provisioning sites (i.e. Opunohu and Vaiare) that are commonly used for tourist activities [18]. The bait cage contained frozen tuna heads and flesh, and all experiments were conducted independently from touristic activities.

In order to monitor shark interactions, we developed a custom remote video recording system which comprised three wide-angle underwater cameras (GoPro, San Mateo, USA) pointing at the cage (see electronic supplementary material, figure S1). Two cameras were placed 6 m away from the cage and recorded shark behaviour from two different angles. One vertical camera was placed 15 m above the cage and recorded shark spatial positions in an approximately 180 m² area around the bait cage. We deployed the recording system for 90–120 min a day, depending on meteorological conditions.

The sicklefin lemon sharks of this study have been monitored daily for more than 10 years through photo-identification [30] and showed different degrees of attachment to the provisioning site [18]. The robustness and accuracy of the photo-identification method was demonstrated through a long-term monitoring programme [18,30] and was also cross-validated with genetic analyses [29]. In the present study, sharks were therefore individually identified according to their body marks, sex and estimated length. We determined the site fidelity of each shark as the proportion of days the individual was observed on site. For example, a shark was attributed a site fidelity score of 100% for a given provisioning site if it was identified at least once on every sampled day at that site.

We conducted behavioural observations on 52 days in two sampling seasons, from June to July 2014 and from April to July 2015, which yielded 1990 min of coded shark behaviours. For the following analyses, video recordings were divided into 10 min sampling periods as an arbitrary standard. Two consecutive sampling periods were separated by a 10 min interval to ensure independence between sampling periods. For each sampling period, the number of competitors was measured as the total number of sharks that were identified at the provisioning site on the day of the sampling period. We coded shark behaviour using behavioral observation research interactive [31], an event-logging software that permits the coding of animal behaviour from video material.

(b). Tolerance associations

We tracked the position of sharks in the approximately 180 m² monitored area and recorded their presence in a 3 m radius zone around the bait cage (see electronic supplementary material, figure S1). We consider this zone to represent the idiosphere defined by Martin [32] as ‘the volume immediately surrounding an individual’ that a shark will defend against intruders while visiting the bait cage. Here, two or more sharks co-occurring in this zone were considered to tolerate each other and form a tolerance association. A shark is considered to attack intruders that broach a perimeter equivalent to the shark's body length [32], hence we set the radius of the tolerance zone according to the mean body length of all sampled sharks (see Results).

We used social network analysis as a tool to describe tolerance relationships between sharks competing for food, with no intention to interpret the resulting network outside the boundaries dictated by the specific context of artificial aggregations. We constructed a shark tolerance network from tolerance data with the asnipe package [33] in R v. 3.2.3 [34]. Each tracked shark could be reliably identified thanks to a record book established in a previous study [30] and that differentiated individuals according to distinctive body marks, estimated size, and sex. We are confident that all sharks present at the provisioning site were recorded thanks to our multiple-angle video setting and included in the network analysis. Moreover, multiple-angle video recordings permitted to identify every shark involved in 100% of tolerance associations. Thus, we used the simple ratio index to estimate shark tolerance associations [35].

The existence of preferred tolerance relationships was verified by comparing the standard deviation of association indices between the real dataset and 1000 permuted networks, while controlling for the number of observations and group size [35]. Based on the stack of group-by-individual matrices corresponding to each sampling period, this data stream permutation swaps individuals between groups within sampling periods, then recalculates the weighted summary network after each swap. Each permutation creates a new weighted network, resulting in a new stack of matrices [33]. In the resulting network, shark degrees (i.e. number of connections) provided an index of their tolerance to the presence of conspecifics around the bait.

Finally, we used the assortnet R package [36] to test the effects of individual size, sex, site fidelity, and rates of aggressive and submissive behaviours on shark tolerance assortment. We measured the significance of assortativity coefficients by comparing the observed data with 1000 randomized networks obtained from data stream permutations [35].

(c). Agonism and dominance hierarchy

Agonistic behaviours occur in competitive contexts and are expressed as aggressive or submissive displays, shaping the social dynamics and individual fitness of many taxa [25]. We recorded shark agonistic behaviours (i.e. nature of behaviour, source and target individuals) following the ethogram provided by Martin [32]. Agonistic interactions were recorded in all the monitored areas (i.e. both inside and outside the tolerance zone), so that tolerance associations and dominance interactions constituted two independent datasets.

To compare agonistic behaviour between sharks that varied in the amount of observation time, we calculated individual agonistic rates by dividing the total number of agonistic displays recorded for each shark by the total time the shark was recorded in the monitored area, while distinguishing aggressive from submissive displays (see electronic supplementary material, figure S1).

The opportunities for agonistic interactions are expected to increase proportionally to the number of competitors. We tested whether agonistic interactions increased faster than the number of competitors. For each sampling period, we divided the number of recorded agonistic interactions by the number of sharks present to obtain the number of agonistic behaviours produced for each shark.

We quantified dominance relationships from recorded agonistic interactions with the EloRating package in R and using two distinct indices: elo scores, which account for the sequencing of events and allow to calculate dominance in dynamic groups [37], and normalized David's scores (NormDS), which are robust to minor deviations in repeated interactions between individuals [38]. We tested the correlation between a shark's dominance score and its individual attributes (i.e. size, sex, site fidelity and tolerance to competitors).

As suggested by Sánchez-Tójar et al. [39], we used the aniDom R package [40] to investigate how well dominance rank predicted the probability to win an interaction. In steep hierarchies, higher ranked individuals win all dominance fights with a probability close to 1. In flat hierarchies, the outcome of dominance fights is highly unpredictable and the probability for higher ranked individuals to win is slightly superior to 0.5. Finally, we explored the transitivity of the observed interactions using the triangle transitivity Ttri, which provides an overall estimate of orderliness for the hierarchy [41]. Transitivity refers to the consistency of dominance relationships between dyads in the sense that if A wins B, and B wins C, then A should win C.

(d). Modelling resource access

Finally, we investigated the effect of aggregation parameters (i.e. size and composition of the aggregation) and individual attributes (i.e. size, sex, aggressive and submissive rates, dominance score and site fidelity) on individual shark resource access, measured as the time spent by each shark in the interaction zone for each 10 min sampling period (time at cage, TAC). We used a generalized linear mixed-effect modelling framework (GLMM), set the identity of sharks as random effects and used a binomial error structure with a logit link function. The model was built using function glme() in the lme4 R package [42] and model selection was conducted using single-term deletions with a likelihood ratio test-based (LRT) backward selection [43].

3. Results

Our custom underwater camera monitoring system (see electronic supplementary material, figure S1) allowed us to quantify (i) pairwise shark tolerance relationships from 19443 observed tolerance associations and (ii) shark dominance relationships from 368 agonistic behaviours, recorded over 52 days of experiment.

(a). Shark behaviour correlates to the number and familiarity of competitors

The number of agonistic behaviours increased faster than the number of competitors. Individual sharks displayed more agonistic behaviours for each competitor present as the number of competitors increased (Pearson product–moment correlation: n = 368, r = 0.37, p < 0.05; see electronic supplementary material, figure S2). Similarly, sharks spent more time at the cage (TAC) as the number of competitors increased (Pearson product–moment correlation: n = 592, r = 0.15, p < 0.001; see electronic supplementary material, figure S2).

Moreover, the number of agonistic behaviours exchanged between two sharks was negatively correlated to the number of days the two sharks were observed together at the feeding site (Pearson product–moment correlation: n = 368, r = −0.33, p < 0.05; see electronic supplementary material, figure S3).

(b). Tolerance associations of competing sharks

The shark artificial aggregations comprised 29 adult individuals (total length = 240–330 cm, mean = 283 cm ± 19 cm s.d., lengths estimated visually and cross-validated by laser photogrammetry [18]) that appeared to form two clusters consisting of sharks that showed higher fidelity to one of the two respective study sites (figure 1a). The clusters were connected by two individuals only, demonstrating limited exchanges between the two artificial aggregations. In the following, we provide results for our network analysis that included all shark tolerance interactions sampled across both study sites. We also performed separate network analyses for each site and indicate the results in the text, where applicable.

Figure 1.

Sicklefin lemon shark tolerance network and dominance hierarchy. (a) Tolerance network. Nodes represent sharks and individuals are labelled as follows: F (females) or M (males) + ID number. Edges indicate the SRI association index. Node size and colour are respectively proportional to shark length and dominance score. (b) Sharks had more connections in the tolerance network (degree) as they showed stronger fidelity to the respective study sites. (c) Sharks produced fewer agonistic behaviours as they showed higher fidelity to the respective study sites. (d) Sharks that showed intermediate submissive rates had more connections in the tolerance network (degree) than sharks that showed high or low submissive rates.

(i). Individual variation in tolerance and agonistic behaviour

Individuals that showed high fidelity to the provisioning sites had tolerance interactions with a higher number of competitors while visiting the bait cage (figure 1b). This was also observed after separate network analysis for each site (Pearson product–moment correlation: Opunohu, n = 18, r = 0.88, p < 0.001; Vaiare, n = 14, r = 0.89, p < 0.001; electronic supplementary material, figure S4). Moreover, individuals that showed high fidelity to the provisioning sites produced less agonistic behaviours in the whole monitored area (figure 1c). Furthermore, sharks that showed intermediate rates of submissive behaviour had more connections in the tolerance network than both dominant and totally submissive individuals (figure 1d). This was also observed for each site analysed separately (electronic supplementary material, figure S4).

(ii). Tolerance assortment of competitors

Sharks visiting the bait cage did preferentially tolerate the presence of competitors showing similar submissive rates (assortativity analysis: r = 0.43 ± 0.07 s.e., p < 0.05) and site fidelity (r = 0.49 ± 0.07 s.e., p < 0.05). Assortativity analysis also revealed weak assortment of competitors according to their size (r = 0.05 ± 0.09 s.e., p < 0.05), sex (r = 0.02 ± 0.08 s.e., p < 0.05) and normalized David's score (r = 0.10 ± 0.09 s.e., p < 0.05). However, sharks did not privilege tolerance towards conspecifics of same elo score (r = −0.03 ± 0.10 s.e., p = 0.27). Sharks at both sites showed preferential tolerance towards competitors of similar submissive behaviour, but did not preferentially tolerate the presence of competitors of same size, sex or elo score (electronic supplementary material, table S5). Contrary to what was observed for the overall dataset and the Opunohu aggregation, shark assortment according to individual site fidelity was not significant for the Vaiare aggregation (electronic supplementary material, table S5).

(c). Shark dominance hierarchy

Based on recorded agonistic interactions, shark relative positions in the dominance hierarchy were estimated by calculating two distinct indexes of dominance. Elo scores and normalized David's scores showed close agreement with each other (Pearson product–moment correlation: n = 29, r = 0.86, p < 0.001), indicating that the temporal dynamics of wins and losses had limited influence on the assessment of dominance hierarchy in our shark aggregations. This also suggests that overall dominance hierarchy is stable among sampled sharks, as also indicated by the low number of recorded agonistic behaviours (n = 368) in 1990 min of behavioural observations.

Shark hierarchy appeared to be transitive (Ttri = 0.8, p = 0.018) and of moderate steepness (figure 2a). Our ratio of interactions to individuals of 14.7 was within the interval of 10–20 recommended by Sánchez-Tójar et al. [39], suggesting that although our sampling effort could be improved in the future, the resolution of our data is suitably informative. Interestingly, the upper part of our shark hierarchy did not seem to be very well defined, suggesting that a few dominant individuals were not willing to submit to one another. In the following analyses, we retain elo scores as our dominance index.

Figure 2.

Shark dominance hierarchy. (a) Relationship between the difference in rank and the probability for the higher-ranked individual to win the interaction. Value and significance of the hierarchy transitivity (Ttri) and the ratio of interactions to individuals are also indicated. (b) Dominance interaction matrix. Cell colour captures how often individual sharks (rows) dominate rivals (columns), while controlling for the co-occurrence of individuals.

(i). Dominance status and resource access

The most dominant sharks did not have the highest degrees in the artificial aggregations (Pearson product–moment correlation: n = 29, r = 0.08, p = 0.68). However, dominant individuals at the Opunohu site gained privileged resource access since sharks spent more time at the cage as they had higher dominance scores (see electronic supplementary material, table S6). In order to gain further perspective on the structure of shark dominance hierarchy, we combined agonistic and tolerance data to investigate putative patterns of dominance within our shark artificial aggregation.

(ii). Dominance interactions within the tolerance network

We described the directionality of dominance outcomes in pairwise interactions within the Opunohu artificial aggregation while ordering sharks according to their weighted degree. Figure 2b sums up all dominance interactions that were recorded during the whole monitoring campaign at the Opunohu site. While most hierarchical relationships were dominated by central sharks over peripheral competitors, two interaction patterns were the exception and retained our attention. In the first pattern, one peripheral shark dominates several central competitors. In the second pattern, several peripheral sharks dominate the same central competitor. These patterns were observed several times (figure 2b). This result prompts thought on which individual attributes determine a shark's tendency to dominate conspecifics [44]. Remarkably, while morphological and phenotypic attributes are assumed to predict the outcome of agonistic interactions (e.g. [26,45–48]), dominance in our shark artificial aggregation was determined neither by shark size (Pearson product–moment correlation: n = 25, r = 0.37, p = 0.06) nor by sex (Wilcoxon rank sum test: n = 25, p = 0.21).

(iii). Shifts in the direction of agonistic behaviour

By looking at the temporal sequence of dominance interactions recorded within the artificial shark aggregation, we identified a shift in the direction of submissive displays pictured in three steps (figure 3). Most sharks initially submitted to the most dominant and central individual (step 1). Then, one peripheral shark overtly assaulted and dominated the central dominant shark (step 2). Following this, central individuals submitted to the victorious peripheral shark (step 3). A similar pattern could be observed as peripheral sharks regularly assaulted the second most central individual that was overtly submissive to the central dominant shark.

Figure 3.

Behavioural shift within the shark aggregation. Green arrows show the direction of submissive displays pictured in three steps.

4. Discussion

We analysed the social behaviour of solitary sharks that were competing to access food to investigate the putative use of social cues among rivals. Sharks might directly interact with surrounding individuals to tune their adaptive response. In this case, sharks are expected to interact with all potential competitors through dominance fights and establish random tolerance associations. Alternatively, sharks might rely on social information about their rivals and establish tolerance and dominance relationships accordingly to increase their chances to access food, for example by avoiding or submitting to larger individuals.

The morphological attributes of sharks could not predict the outcome of dominance interactions as previously thought, nor did they correlate to patterns of segregation among competitors. However, sharks preferentially tolerated rivals with similar agonistic behaviour. The agonistic behaviours produced by individual sharks and the time they spent near the bait cage increased with the number of competitors. Moreover, the agonistic interactions exchanged between sharks were all the fewer as sharks were often observed together at the study site. Finally, sharks produced fewer agonistic interactions as they showed increased fidelity to the study site. This, we argue, indicates that a major proportion of our sharks might resort to social information to interact with competitors in a context of artificial aggregations.

Although solitary sharks might occasionally gather around a common food source (e.g. whale carcasses [49]), we caution that the artificial aggregations described in this study do not reflect situations that commonly occur in the wild. Still, we accept as a core idea that the ability to assess peers through social information is beneficial in all contexts where individuals gather and interact with one another (see [50] for an example). Following this idea, if sharks are to use social information in artificial aggregations, they might do so in other contexts such as reproduction, foraging or defence.

The morphological attributes of animals can determine their distribution in the habitat (e.g. [51]) and may limit potential conflicts between size or sex classes. They can also predict the outcome of dominance interactions by signalling the fighting ability of contestants [26]. Here, shark segregation by body size, albeit statistically significant, was of very limited magnitude. Contrary to previous findings [49,52,53], shark assortment by sex was also insignificant. Moreover, sharks established relatively stable dominance relationships that were related neither to size nor to sex. These results may stem from our limited ability to estimate shark size accurately enough. Otherwise, given the limited variation of body size among the sampled sharks (see Results), we suggest it did not constitute a reliable phenotypic cue to be exploited by competitors. As a consequence, sharks might have relied on other social information (e.g. number of competitors, behavioural or olfactory cues) to tune their adaptive response.

As the number of competitors increased, sharks spent more time around the bait and produced more agonistic behaviours towards each competitor present. The number of neighbouring individuals conveys information on the surrounding habitat (e.g. the quality of shelter or resources [45]) and influences the behaviour of a wide range of species (e.g. [3,4,17]). In this study, sharks may be drawn to the bait cage by copying the behaviour of conspecifics or by olfactory or acoustic cues that are released as rivals engage with the alimentary resource. A more complex alternative could be that sharks interpret increased rival numbers as reduced chances to feed, and therefore try to secure their access to the alimentary resource by tuning both their spatial position and agonistic behaviour. Future investigations will have to establish whether positive feedback (i.e. the ‘amplification of events through recruitment or reinforcement’ [3]) applies to aggregations of sharks and determine the respective contribution of visual, olfactory and acoustic cues.

When visiting the bait cage, sharks did segregate according to their individual levels of submissive behaviour, reflecting behaviour-dependent associations reported in other taxa such as birds [54] or fish [55,56]. In gregarious species, individuals sometimes synchronize to the behaviour of immediate neighbours [57] because it provides them advantages such as reduced predation risk [57]. Two opportunistic sharks simultaneously visiting the bait cage might synchronize their submissive behaviour if it provides them with the opportunity to approach the alimentary resource. However, the individual levels of submissive behaviour that we used to test for preferential shark pairing were estimated by including agonistic interactions that occurred outside the tolerance zone, and that were recorded over the entire study period. Consequently, this measure does not appear to us as fully appropriate to interpret brief shark interactions around the bait cage. It does, however, rather depict general tendencies of individual shark behaviour.

A more plausible explanation of shark behavioural segregation is that sharks preferentially tolerated competitors that showed similar levels of submissive behaviour. In other words, in the same way as juvenile sharks reduce predation risk by avoiding larger individuals [58], adult sharks competing for food might avoid competitors that produce few submissive displays and against which they stand fewer chances to prevail. This does not necessarily translate into shark assortment by dominance score, since the latter is a general estimation of the overall hierarchy and does not describe the dominance relationship between two individuals [25]. Still, the hypothesis that sharks assess their rivals' agonistic behaviour and avoid dominant individuals is consistent with dominant sharks being more isolated in the aggregation than sharks showing intermediate submissive rates (figure 1).

As discussed above, the observed structure of shark social interactions can be explained by solitary sharks assessing their rivals' behaviour through social information. Still, the conditions promoting the use of social information are yet to be clarified. Further insight can be gained by highlighting additional correlations between shark social behaviour and site fidelity. As outlined by Sims and colleagues [59], aggregations may be an important prerequisite for the development of social behaviour. Although adult sicklefin lemon sharks mainly aggregate for reproduction, their ability to use social information might evolve from early life stages where juveniles gather in shallow-water habitats. Most sharks in the sampled population are genetically related [29] and probably mixed in the same nurseries. Hence, dominance and tolerance relationships [60] could have emerged through social learning [61] as juvenile sharks engaged in repeated interactions [62].

Dominance relationships are thought to spare individuals from permanent competition through the establishment of stable rules, such as the priority to access mates or food (see [25]). The relatively stable dominance relationships occurring among our sharks could possibly be enhanced by conditioning as the same sharks are regularly drawn together at a static food source by provisioning tourism [18]. This hypothesis could explain that sharks showing increasing fidelity to the study sites were involved in a greater amount of tolerance interactions and displayed less agonistic behaviour (figure 1). In this connection, shark segregation patterns were highly similar between the overall dataset and the Opunohu site alone, where shark provisioning activities have been undertaken for decades [18]. Conversely, shark segregation patterns at the Vaiare site alone, where provisioning tourism has only been introduced recently, did not appear to be clearly established (electronic supplementary material, table S5). We suggest two possible, non-mutually exclusive interpretations for this result. First, sharks in the Vaiare aggregation did not differ enough in their individual parameters for our analysis to highlight marked segregation patterns. Second, sharks sampled at the Vaiare site were not subject to the putative long-term conditioning that might have been occurring at the Opunohu site. Also, the eventuality that sharks become familiar to each other is further supported by our observations that sharks established preferential tolerance interactions with rivals showing equivalent site fidelity, and that the more often two sharks were observed together at the study site, the fewer agonistic behaviours they exchanged. This does not exclude that ‘naive’ sharks (i.e. individuals that do not regularly frequent provisioning sites) may directly confront competitors when attracted to the provisioning site (figure 3).

This being said, causal relationships cannot be clearly established at this stage and alternative hypotheses are still to be tested. For example, some sharks might be more likely to come near provisioning sites due to their ‘socially tolerant’ nature. Moreover, shark dominance interactions might be established through cues different from the agonistic behaviours that were sampled in this study (e.g. olfactory signals). Even so, previous studies report that high ratios of brain mass to body mass in sharks [63] and patterns of shark brain allometry [64] indicate their potential for remarkable learning abilities [65], including the ability to recognize familiar conspecifics [62]. This, added to the fact that sharks appeared to copy the agonistic behaviour of neighbouring individuals (figure 3), sets the stage for future research on shark social conditioning.

Although only a few studies provide quantitative information on shark social interactions [28,60,66], this study aligns with a recent surge of work [67–70] that suggests the importance of inter-individual behavioural differences on critical aspects of shark ecology. Moreover, a recent study [71] highlights that simple inter-individual differences in social behaviour can explain the apparently complex structure of social networks. Albeit ethically questionable and methodologically challenging, captive or semi-captive manipulative studies that test how the individual parameters of sharks transcend at the collective scale could lead to novel insights into the use of social information in solitary predators. Furthermore, we emphasize that describing animal aggregations through the heterarchy framework and across a gradient of group parameters, such as group size or composition, provides a promising avenue to uncover organizational patterns in animal groups.

5. Conclusion

We provide experimental evidence of the use of social cues among artificial aggregations of solitary predators. Using an original monitoring protocol, the analysis of hierarchical and tolerance interactions of sharks in a context of competition suggested that sharks exploited information related to the number and the individual attributes of rivals. This ability may provide individual benefits by minimizing negative interactions with dominant counterparts and maximizing foraging opportunities. Future studies on other social contexts in sharks as well as similar methodological approaches on other solitary species would be particularly valuable to improve our understanding of (i) how social information participates in the organization of animal groups and (ii) which are the conditions promoting the use of social information.

Ethics

This work was approved by the Direction à l'Environnement (DIREN) of French Polynesia and Ministère de la promotion des langues, de la culture, de la communication et de l'environnement de Polynésie française under Arrêté 9324 of 30 October 2015.

Data accessibility

Data available at http://dx.doi.org/10.5061/dryad.4dp8c [72].

Authors' contributions

All authors contributed to the study and first draft; P.F.B. and S.P. designed the experiment; P.F.B. collected and analysed the data with the help of J.M. All authors agreed to be held accountable for the content therein and approved the final version of the manuscript.

Competing interests

We have no competing interests.

Funding

This study benefited from the financial support of the Ministère de l'Ecologie, du Développement Durable et de l'Energie (MEDDE) of France, the Fonds Français pour l'Environnment Mondial (FFEM) and the Délégation Régionale à la Recherche et à la Technologie (DRRT) of French Polynesia.