Reduced older male presence linked to increased rates of aggression to non-conspecific targets in male elephants

Abstract

Males in many large mammal species spend a considerable portion of their lives in all-male groups segregated from females. In long-lived species, these all-male groups may contain individuals of vastly different ages, providing the possibility that behaviours such as aggression vary with the age demographic of the social environment, as well as an individual's own age. Here, we explore social factors affecting aggression and fear behaviours in non-musth male African elephants (Loxodonta africana) aggregating in an all-male area. Adolescent males had greater probabilities of directing aggressive and fearful behaviours to non-elephant targets when alone compared to when with other males. All males, regardless of age, were less aggressive towards non-elephant targets (e.g. vehicles and non-elephant animals) when larger numbers of males from the oldest age cohort were present. The presence of older males did not influence the probability that other males were aggressive to conspecifics or expressed fearful behaviours towards non-elephant targets. Older bulls may police aggression directed towards non-elephant targets or may lower elephants’ perception of their current threat level. Our results suggest male elephants may pose an enhanced threat to humans and livestock when adolescents are socially isolated, and when fewer older bulls are nearby.

1. Introduction

Since male fitness is mainly driven by the number of successful fertilizations [1], aggression in males is typically viewed through the lens of sexual competition, with a focus on direct mate guarding [2], defence of territory and resources to gain access to females [3], or establishment of dominance hierarchies in order to monopolize mating [4]. However, sexual segregation and bachelor groups occur in many large mammal species [5,6], providing potential for aggressive behaviours by males in the absence of females to directly contend for. Currently, we know comparatively little about the factors that influence aggressive behaviours in all-male groups. This represents an important gap in knowledge as many males spend the majority of their lives in such all-male groups. Additionally, in long-lived species with distinct life-history stages (e.g. prolonged adolescent periods with higher investment in learning and development, and lower investment in reproductive activities), the possibility arises that differences in the ages of males in all-male groups may influence the aggressive behaviours that are performed by members [7–10].

Male African savannah elephants (Loxodonta africana) dispersed from their natal herd spend most of their lives sexually segregated from females [11], with males spending 63% of their time in all-male groups, and 18% of their time alone [12]. The species is also one of the few non-predatory species whose aggressive behaviours can serve an immediate lethal threat to humans and their livelihoods [13,14], and males are disproportionately involved in human–elephant conflicts compared to females [15]. Social disruptions during development in African elephants can lead to negative behavioural outcomes, including abnormal hyper aggression [16]. Mature bulls appear to have a role in inhibiting musth (sexually active state in male elephants, characterized by high rates of aggression [17]) in younger males [7,8], suggesting both an individual's life-history stage and the social environment can influence aggression in this species. Understanding the patterns of aggression in male elephants, including the nature and targets of this aggression, and how factors such as age and social context within all-male groups can influence these behaviours, is therefore of paramount importance owing to its relevance to human safety and well-being.

Here, we quantify the agonistic behaviours of non-musth male African elephants in a male-dominated area under different social contexts. We first examined how social isolation was linked to elephants of different ages expressing ‘flight or fight’ (fear and aggression behaviours, respectively) responses towards non-elephant targets. While directing aggression to a perceived threat may be one reactive response for elephants under stress (‘fight’ response), they may also respond with more ‘flight’ type fearful anti-predator responses (i.e. running away from the perceived threat [18–20]). Male elephants form larger groups when in higher risk environments, for example, when outside of protected areas [5]. We therefore predicted, both due to their lack of previous experience in assessing and responding appropriately to real risk [11,21], as well as a greater genuine vulnerability (e.g. predation risk [22], and dispersal risks in a novel environment [23]), that adolescents would be more likely to perform fear-related behaviours when alone compared to when in the company of other males. By contrast, being alone was not expected to represent as severe a threat for adults, who are more experienced and physically larger [11]. We therefore predicted adults males that were socially isolated would express fear and aggression behaviours to non-elephant targets at equal rates to those in the company of other males.

Second, we tested if the number of males of different age classes present in the immediate environment was associated with the performance of agonistic behaviours (both to conspecifics and non-elephant targets). Specifically, we hypothesized a greater number of mature males in the immediate environment would reduce the expression of aggression and fear behaviours in male elephants.

In a prominent case study of ‘delinquent’ young male elephants in Pilanesberg National Park (South Africa), abnormal aggression and premature musth in young males was corrected once mature bulls were introduced to the population [7,8]. This observation is reminiscent to the finding that dominant individuals act as policers of subordinates’ conflicts in primates [24], and that lower adult–young ratios in horse groups leads to greater aggression in young horses due to adult regulation of young horse's aggression behaviours [10]. It is likely that aggression directed to conspecifics differs in function to the aggression directed to non-elephant targets and relates more to dominance hierarchy establishment and access to resources, as opposed to a reactive response to a perceived threat or irritant [25]. We predicted there would be increases in aggression to conspecifics with reduced mature male presence, which may indicate disruptions to the linear dominance hierarchy [7,8,26], and/or a potential policing influence of mature males on younger male's conflicts [24,27]. Additionally, mature males may also police aggression behaviours to non-elephant targets as a behaviour that is also potentially detrimental to group cohesion [24], and we also predicted elephants would direct less aggression to non-elephant targets with increased mature male presence in the environment.

Alternatively, elephants may be more likely to direct aggression to non-elephant targets with decreased mature bull presence as they may perceive themselves to be at greater risk in the absence of experienced individuals in the environment [28]. Increases in elephants performing fear behaviours to non-elephant targets with decreased mature bull presence would also support this risk perception hypothesis. In horses, informed (often older) individuals appear to play an important role in transmitting information to group mates regarding safety, for example, naïve horses have reduced fear responses when paired with informed demonstrators [29], and young foals weaned without adults express increased aggression and behavioural and physiological stress [9]. An age-structured effect on risk assessment has been in shown in female groups of African elephants, for example, where older matriarchs make better assessments about risk, which they communicate to group mates [30]. Such findings would highlight the need to investigate the social role of mature individuals in all-male groups and provide new insights to the importance of older individuals from a wildlife management perspective.

2. Methods

The study was conducted within, but at the border of Makgadikgadi Pans National Park (MPNP), Botswana, a bull area where 98% of elephant sightings are sexed as male [31]. The region adjacent to the site of data collection has the highest reported rate of human–wildlife conflict in Botswana [32], with 71% of residents in Greater Khumaga interviewed stating that elephants threatened their safety [33]. We conducted focal sampling of male African elephants aggregating at hotspots of elephant social activity along with the Boteti River, which marks the border of the MPNP (electronic supplementary material, figure S1). Data were collected between September 2015 and September 2018 at five hotspot locations. Hotspots were areas of river with easy access for elephants and were the terminal points of elephant pathways in the MPNP landscape [34]. Hotspot boundaries were defined by natural landmarks in the environment, based on the general area in which elephant aggregations remained during a visit to the river (electronic supplementary material, table S1 for locations, boundaries and approximate area covered).

(a) . Data collection

Individual subjects were filmed for the entirety of their stay within social hotspots, starting either as the subject arrived over the bank, or as he entered the hotspot having moved from another stretch of river up or downstream of the hotspot, and terminating when similar boundaries were crossed during departure. Elephants arrived at hotspots alone, or in coordinated all-male group processions [34]. However, following arrival, considerable mixing of males occurred from multiple arriving groups and original groupings became indiscriminate from the larger all-male aggregation. Males were categorized into four age classes—adolescents, 10–15 years and 16–20 years, and adults, 21–25 and 26+ years—based on body size, shoulder height [35], head size and shape, and tusk girth and splay [36]. The age class 26+ years represents an age where males are largely considered sexually and socially mature [37], begin experiencing regular annual musth periods and achieve mating success [17,37]. The age class of focal subject to be recorded was randomly preselected, and the first elephant of the assigned age class to arrive at the hotspot since the start of the session was the subject of a focal animal sample (elephants were aged in the field, if the arrival group had multiple individuals from the preselected age class, the focal was selected at random from the choice). Recordings of visits to hotspots were taken from focal individuals only once over the study period. Individuals were identified by distinguishing features such as tears, holes and notches in the ears, tusk morphology, skin wrinkles, tail length and other body abnormalities [38].

Subjects of focal animal samples were filmed using a video cam-corder (JVC quad proof AVCHD) fixed to a tripod, with the subject kept central to the frame, but zoomed out enough to allow for potential interactors to be captured. Video recordings were taken between 08.00 and 18.30 (electronic supplementary material, note S1). The research vehicle was parked at a safe distance (minimum 50 m) from points expected to receive elephants (pathway arrival points, popular drinking points, mudholes). Non-musth males in the MPNP are largely relaxed around vehicles, and if the engine was off for the entire focal session, it was common for elephants to not look in the direction of the human observer (electronic supplementary material, note S2 for methods for addressing vehicle presence).

Focals could stay at social hotspots for several hours (average time spent at hotspot for focal elephants seen arriving and leaving via bank = 1 h 13 min, range = 9 min–7 h 5 min, s.d. = 59 min), over which time, the males present at aggregations with focals could be highly dynamic. Since individuals arriving in all-male groups tend to arrive within 10 min of one another [34], focal follows were subdivided into 10 min follows (e.g. a focal follow of an elephant staying 40 min at the hotspot would produce four 10 min focal follows), to which a corresponding social context was assigned (see below), in order to capture the temporally dynamic nature of male aggregations at the hotspots.

In fifteen 10 min follows (from six individuals), females were also present at the hotspot. The presence of females was rare in this bull area, so it is possible this could impact on aggressive interactions between males. The presence of females did not predict the expression of any behaviours of interest by males in the study (electronic supplementary material, table S2). Nevertheless, to be conservative, the 15 focal samples where females were present were excluded from our analyses. Additionally, 52 focal animal samples (from 10 individuals) were collected on elephants in musth. Due to the established consensus that bulls act differently in musth state, with greater aggression to same-sex conspecifics [17], we excluded musth bull focals from our dataset. The electronic supplementary materials (electronic supplementary material, figure S2) provide a comparison of aggressive behaviours of musth compared to non-musth males in this study. Finally, if a subject was out of view for over 2 min within a follow, i.e. over 20% of time (n 10-min follows = 201), the 10 min focal follow was excluded from analysis. For 126 10 min focal follows, the focal elephant was out of view for 00.01–01.59 min; however, for most cases (n 10 min focal follows = 1514), the subject was in view for the full 10 min.

(b) . Scoring of behaviours

Focal follow videos were scored by one researcher (CA) to standardize scoring of behaviours, with each follow observed for behaviours three times. Behaviours of interest (aggression directed to conspecific, aggression to non-elephant target, fear to non-elephant target (table 1)) were scored as a number of events per 10 min focal follow.

Table 1.

Ethogram of behaviours recorded during focal follows and their categorization for analysis in the current study [39,40].

behavioural category	summary
conspecific aggression	Aggressive behaviours relating to dominance assertion and gaining access to resources, as well as potentially re-directed aggression including ‘advancing towards’, ‘spreading ears', ‘holding head high’, ‘ear folding’, ‘head shakes’, among other behaviours (electronic supplementary material, note S3 for full list of behaviours and detailed descriptions) directed by the focal subject towards conspecifics.
aggression directed to non-elephant target	Many of the behaviours employed during aggression to conspecifics are similarly directed at non-elephant targets that are perceived as threats or irritants, including ‘advancing towards’, ‘head high’, ‘spreading ears’, ‘Head shakes’, among others (electronic supplementary material, note S4 for full list of behaviours and detailed descriptions).
	Targets of non-elephant aggression included other animal species (e.g. ungulates, carnivores, reptiles and birds), vegetation and tourist vehicles, but in most cases the target of the aggressive behaviour was unidentifiable (electronic supplementary material, figure S3 for distribution of targets of aggression by age class).
fear directed to non-elephant target	Defensive and fearful behaviours, including ‘running away’, ‘tail raised’, ‘jaw tilted upwards’, among others (electronic supplementary material, Note S5 for full list of behaviours and detailed descriptions), employed by elephants in response to perceived threats.
	Targets of (or rather, the triggers of) these non-elephant directed fear behaviours included other species (e.g. ungulates, carnivores, reptiles and birds) and tourist vehicles, but in most cases the triggers of these behaviours were unidentifiable (electronic supplementary material, figure S3 for distribution of targets of fear behaviours by age class).

(c) . Social context

During field observations, data were collected on the number of, and ages of, all other elephants present at the hotspot with the subject elephant, such that for every 10 min focal follow, there was a corresponding recording of all ages observed as present with the focal within that time window (electronic supplementary material, figure S4). The social context at the social hotspot was unknown to researcher scoring behaviours from videos and was only matched to corresponding focals subsequent to all videos being coded for behaviours.

(d) . Statistical analyses

For our analyses, we ran generalized logistic mixed-effects models (GLMMs) in R. Within each 10 min focal follow, each of the three behaviours of interest (table 1) were transformed to a binary 1/0 (present/absent) term due to a considerable right skew in the dataset (e.g. for aggression directed at non-elephant targets, 1047 10 min focal follows had 0 events, 312 10 min follows had 1 event, and 168 10 min follows had greater than 1 events of aggression (range 2–12 events)). Due to a small sample size for 10- to 15-year-old focals sighted alone (eight 10 min focal follows), we merged age classes of focal elephants into the categories ‘adult’ (21+ years; n = 846 10 min focal follows from 147 individuals) and ‘adolescent’ (10–20 years; n = 681 10 min focal follows from 134 individuals) to test the effect of social context on the behaviours of subjects.

First, we explored if social isolation was related to elephants’ (i) expression of aggressive behaviours to non-elephant targets and (ii) expression of fear behaviours to non-elephant targets. For these GLMM's, each behaviour (dependent variables) was modelled in relation to season, hotspot location, age category (adult or adolescent), social isolation condition (where 1 represented a subject being alone at a hotspot, and 0 represented other elephants being present with the subject), and the interaction between age category and social isolation condition (whereby reference class of age category was switched to explore the influence of social isolation on the aggression and fear behaviours for adolescent and adult bulls separately). Elephant ID was included as a random effect in both models.

Second, we investigated if the number of mature bulls (26+ years) at the hotspot was related to the probability that a subject directed aggressive behaviours at (i) conspecific targets and (ii) non-elephant targets, and (iii) fear behaviours at non-elephant targets. For these models, only males observed with other elephants at the hotspot were included (lone subjects were excluded). We fit GLMMs predicting each behaviour (dependent variable) by focal age category (adult or adolescent), season, hotspot location and number of each age class present during the 10 min focal follow (i.e. number of each age class 10–15, 16–20, 21–25 and 26+ years were included as separate predicting variables). This allowed us to compare whether the number of other age classes present also influenced behaviours. In cases where the expression of a behaviour was only predicted by number of mature bulls and not the presence of individuals from other age classes, we re-ran this analysis to include interaction terms between focal age category and number of mature bulls, to test if the number of mature bulls in the environment had a different effect on adolescents compared to adults. All non-significant fixed effects from the initial model were excluded in this second interaction model. Elephant ID was again included as a random effect in all models.

In all the above analyses, we also included a fixed effect of whether this type of behaviour had also been performed in the preceding 10 min follow to control for the potential influence of temporal autocorrelation (electronic supplementary material, note S6). We also included season in all our GLMMs because availability of resources, and potentially body condition, are linked to season [41] which may influence elephants' tolerance in sharing limited resources, or influence linear dominance hierarchies [26] (electronic supplementary material, note S7 for season determination methods). Furthermore, focal observations conducted in the wet season had higher numbers of other elephants present at the hotspot compared to the dry season (electronic supplementary material, figure S4), and we wanted to account for this seasonal difference in aggregation sizes. Lastly, season also represented the best indicator of numbers of other species (potential targets of behaviours) sharing the hotspot resource with elephants, with some 20 000 zebra and wildebeest frequenting the Boteti River over the dry season, but absent in the wet season [42]. As a control, hotspot location was also included as a fixed effect in all models, since the five hotspot locations differed in factors such as proximity to human-dominated landscapes and tourist presence, which may influence behaviours.

3. Results

Social isolation significantly predicted the likelihood of adolescents, but not adults, performing both aggressive and fear-based behaviours to non-elephant targets, with adolescent males more likely to perform both these behaviours when alone compared to when observed with other elephants (figure 1; adjusted odds ratio (aOR) for directing fear behaviours to non-elephant targets when alone compared to with other elephants: adolescents = 2.775, p = 0.013; adults = 1.206, p = 0.736; aOR for directing aggression behaviours to non-elephant targets when alone compared to with other elephants: adolescents = 2.624, p = 0.021; adults = 1.387, p = 0.400; electronic supplementary material, tables S3 and S4 for full outputs of GLMMs including 95% confidence intervals).

Figure 1.

(a) Being alone significantly predicted the likelihood of adolescents performing fear behaviours to non-elephant targets, but not adult elephants (electronic supplementary material, table S3 for full output of GLMM). (b) Being alone significantly predicted the likelihood of adolescents performing aggression behaviours to non-elephant targets, but not adult elephants (electronic supplementary material, table S4 for full output of GLMM). Significant regression coefficients indicated with asterisk (*), 95% confidence intervals indicated. (Online version in colour.)

Excluding subjects alone at hotspots, 10 min focal follows had on average 2.85 (s.d. = 3.98, max = 22) 10–15 year olds, 4.22 (s.d. = 4.88, max = 28) 16–20 year olds, 2.15 (s.d. = 2.44, max = 21) 21–25 year olds and 1.04 (s.d. = 1.48, max = 10) 26+ year olds present with the focal subject. However, there were differences between adolescent and adult subjects concerning the mean number of other age classes present with them. Adolescent subjects had more 10–15 year olds present with them at hotspots than adult subjects did, and adult subjects had more elephants aged 16–20, 21–25 and 26+ years present with them at hotspots than adolescent subjects did (electronic supplementary material, table S5).

Adults were more likely to direct aggression to conspecifics compared to adolescents (aOR adult compared to adolescent = 1.686, p = 0.014). The number of elephants of each age class present at a hotspot did not predict the likelihood of subjects directing aggression to conspecifics (electronic supplementary material, table S6 for output of GLMM).

Adults were less likely to direct fear behaviours to non-elephant targets compared to adolescents (aOR adult compared to adolescent = 0.556, p = 0.016). Only the number of 10–15 year olds present at a hotspot predicted the likelihood of subjects directing fear behaviours to non-elephant targets, with elephants directing more fear to non-elephant targets when a greater number of 10–15 year olds were present (Regression coefficient: 0.113, p = 0.015; electronic supplementary material, figure S5 and table S7 for output of GLMM).

The number of 26+ year olds present at a hotspot did predict the probability of a subject directing aggression to non-elephant targets. As the numbers of mature bulls present increased, the likelihood of subjects directing aggression to non-elephant targets decreased (Regression coefficient: −0.242, p = 0.001; figure 2). No relationship was found between the likelihood of a subject directing aggression to non-elephant targets and the number of elephants present of all the other age classes (electronic supplementary material, table S8). Adults were less likely to direct aggression to non-elephant targets than adolescents (aOR adult compared to adolescent = 0.378, p < 0.001; electronic supplementary material, table S8), but there was no significant interaction between age category of the subject and the number of 26+ year olds present at a hotspot in predicting the likelihood of the subject directing aggression to non-elephant targets (electronic supplementary material, table S9). That is, when greater numbers of mature bulls were present, the probability of males of any age acting aggressively to non-elephant targets decreased.

Figure 2.

Elephants were less likely to direct aggression to non-elephant targets with greater numbers of 26+ year olds present at social hotspots. Grey area represents 95% confidence intervals based on standard errors (electronic supplementary material, table S8 for output of GLMM).

Season had no influence on probability of an elephant directing aggression to either conspecific targets (electronic supplementary material, table S6) or non-elephant targets (electronic supplementary material, tables S4 and S8), nor on probability of directing fear behaviours to non-elephant targets (electronic supplementary material, tables S3 and S7). Hotspot location did not predict likelihood of behaviours being performed in any of our models, apart from in the main effects model predicting aggression directed to non-elephant targets by numbers of each age class present, whereby aggression was more likely to be performed at hotspot 1 compared to hotspot 4 (electronic supplementary material, tables S4–S9). In all models, the performance of behaviours in a 10 min follow were also predicted by whether that type of behaviour had also been performed in the 10 min follow immediately previous, apart from the model predicting fear directed to non-elephant targets by numbers of each age class present (electronic supplementary material, tables S4–S9).

4. Discussion

When alone, adolescents were more likely to perform aggression and fear behaviours to non-elephant targets compared to when with other males at hotspots, and overall, adolescent male elephants were more likely to direct aggression and fear behaviours to non-elephant targets than adult males. These ‘fight or flight’ type responses to non-elephant targets may be a reflection of the physiological and psychological state of elephants, driven by their perception (both real or perceived) of their current risk and threat level [25,28]. Aside from human threats, adult bulls have no other natural predators [43]. Adult elephants may be less fearful in the exposed habitat of the riverbed hotspot environment that they may have frequented multiple times over their lifetime and thus have a greater level of familiarity with [11]. Adolescents, on the other hand, are still vulnerable to a real threat of predation from lions [22]. Adolescents are also more likely to be recently dispersed from their natal herd and may be more sensitive to perceive the potentially novel, unknown environment as risky [11,23,44,45]. Less experienced adolescents may also perceive the social hotspots as dangerous due to close proximity to human settlements, to which they are not yet habituated (the hotspots mark the boundary of a protected area and a human-dominated landscape) [31,46]. Indeed, elephants are very sensitive to human scent [18], and adolescents may additionally be less habituated to tourist presence, hence more likely to perform self-defence type aggression and fear behaviours in the national park [25,47]. Animals adjust vigilance rates in response to group size and respond with flexible heightened anti-predator and flight behaviour when they perceive human or predatory threats [48,49]. When socially isolated, the real and perceived risks described are likely exacerbated (e.g. individual risk of predation is greater [22]) and younger males may experience a further lowered threshold of risk perception [25,44,49], demonstrated by their increases in fear and aggression behaviours to non-elephant targets. By contrast, the behaviour of adult males did not appear to be influenced by social isolation, suggesting that physically larger and more socially experienced adults do not experience a change to their real or perceived threat level when alone [45].

In many species that experience an adolescent life-history stage, where individuals are not fully socially mature, hormones in the adolescent's physiology can drive exploratory tendencies, novelty seeking and motivation for risk-taking behaviours that could be more likely to put the individual in dangerous situations [50,51]. This highlights a potential dilemma of cause and effect in our findings. It may not be possible to discern whether adolescents are more prone to social context influencing their behaviour compared to adults (i.e. their increased sensitivity in performing more agonistic behaviours to non-elephant targets when alone), or alternatively whether adolescents with temporary hormonal and aggressive ‘surges' separate themselves and choose to be alone, or are excluded from groups owing to their disruptive hyper-aggressive and fearful behaviours. Furthermore, the observed lack of variation in adult agonistic behaviours to non-elephant targets depending on grouping condition may be due to selective disappearance of the individuals that are overly fearful and aggressive when alone [52] (i.e. individuals that express heightened fear and aggression behaviours when alone don't reach adulthood). While a longer-term study would be needed to address the potential of selective disappearance of individuals with a low threshold to coping with risk in adulthood, we believe it is unlikely that the sample of lone elephants represented individuals that were actively excluded from groups, or choosing to be alone. Hotspots were routinely visited by large numbers of elephants, and our method of scoring social context quantified the presence of all elephants at the hotspot, not necessarily reflecting the individuals preferred choice of social companions. While it is possible that individuals excluded from groups or choosing to be alone can fissure from groups out in the larger landscape of the MPNP, the hotspots are a large, shared and popular resource, and elephants have no control over the arrival of conspecifics.

For both adult and adolescent elephants, the probability of performing aggressive behaviours to non-elephant targets was greater when there were fewer older male elephants in the immediate environment. One interpretation of this result could be that elephants perceived themselves to be at higher risk in these cases. Male elephants of all ages prefer to have the oldest males in a population as their nearest neighbours, potentially to reap benefits from their heightened ecological knowledge, which could include knowledge regarding environmental risk assessment [53]. Some researchers suggest that due to their heightened experience with age, older males hold a similar role as matriarchs do in female family groups in their importance to the wider bull society [12,30,34,53]. In elephant family groups, older matriarchs are better at assessing risks in the environment, which provides survival benefits to their group mates [30]. We suggest that, for males too, with fewer older mature males present in environment, males may perceive themselves to be at higher risk, and experience lower levels of certainty about their safety [28], which is expressed though the observed increases in aggression to non-elephant targets. In other words, older males may act as particularly effective partners in social buffering [54], relieving stress and anxiety in group mates. In addition, we also found elephants were more likely to direct fear behaviours to non-elephant targets when greater numbers of 10–15 year olds were present; this may reflect a social contagion and spread of fear behaviours triggered by greater numbers of more skittish, fearful young adolescents being present.

While the increased probability of performing aggressive behaviours to non-elephant targets when in higher-risk social contexts may represent responses to targets actually perceived as threatening by elephants with a heightened sensitivity, this aggression may alternatively or additionally be a form of re-directed or displaced aggression linked to an acute stress response induced by a perceived threatful social condition [39,55]. Indeed, aggression to non-elephant targets often appeared not to be a true anti-predator defence because it was directed at non-threatening objects or bystanders (for example, bashing of vegetation, charging of birds or smaller ungulates) or had no obvious target (target was unidentifiable, see electronic supplementary material, figure S3). In many social mammals, following a stressful experience, redirecting aggression to third parties of their own species is thought to represent a stress-reducing behavioural outlet [55,56]. However, we suggest in such a large and weaponized species, displacing aggression to a conspecific carries too much risk due to potential for escalated conflict, which can potentially turn lethal. African elephants may therefore tend to displace aggression to non-elephant targets. While in the case of the ‘delinquent’ males of Pilanesberg national park, young males were far more isolated from mature bulls than our current study, with total absence of mature bulls in the environment leading to a premature musth in young males [7], we find it interesting to note that there too, in the absence of mature bull influence, elephants directed lethal aggression to rhinos, not conspecifics [8].

Finally, mature bulls may also act as policers of aggressive behaviour directed at non-elephant targets. Reduced presence of mature bulls in the environment may have led to an uninhibited expression of these behaviours [7,24]. These aggressive behaviours are potentially highly disruptive to the social group's activities, cohesion and stability [57], as well as running the risk of escalating and spreading further in the group as bystanders become affected and themselves anxious (C.R.B.A. 2016, personal observation) [27]. For example, the calls of distressed elephants can make elephants act aggressively [58]. Mature bulls may have a role in regulating such behaviours that are disruptive to all-male groups [24]. Future research should focus on whether mature bulls are actively policing the aggressive behaviours of other males through ongoing punishment (our results might suggest this is not the case, as while adults performed more aggression behaviours to conspecifics compared to adolescents, elephants did not increase their aggression to conspecifics with the increased presence of any age class) [24,27,59]. Alternatively, it was often observed that approaches of mature bulls to younger elephants evoked submissive responses even in the absence of dominance and aggressive signalling from the older male (although we cannot exclude the possibility that aggressive vocalizations could be being performed by the older male). Older elephants, with their clear dominance owing to greater size [35] and greater potential to inflict harm obvious to younger males, may have a more passive policing influence on other males (i.e. elephants may simply ‘behave better’ when mature bulls are around without receiving particular policing behaviours [60]).

5. Conclusion and practical implications

Understanding elephant aggression is essential for protecting the lives and livelihoods of people that live alongside the species [13,14]. While this study was conducted in an area with only moderate tourist presence with humans outside of vehicles absent, the aggressive behaviours observed by elephants have the potential to also be performed in areas with greater human presence, including where people move without the protection of vehicles. Globally, elephants are responsible for a significant proportion of large mammal caused injury and fatality to humans [61], and previous research has suggested physiologically stressed elephants may be more prone to aggressive encounters with humans [62]. Our results suggest wildlife managers should be careful to ensure mature bulls are present in elephant populations, as their increased presence was associated with decreased male elephant aggression to non-elephant targets. Adolescent male elephants that are socially isolated, or all ages that are unable to associate with mature males may have a heightened sensitivity to act aggressively and may serve as a greater threat to humans and livestock.

Ethics

This work received approval from the University of Exeter Research Ethics Committee (application ID: eCLESPsy000545 v3.2) and was conducted with permission of the Botswana Department of Wildlife and National Parks, under research permit no. EWT 8/36/4 XXXVI (57).

Data accessibility

The data are provided in the electronic supplementary material [63].

Authors' contributions

C.R.B.A.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, writing the review and editing; D.P.C.: supervision, writing the review and editing; L.J.N.B.: supervision, writing the review and editing. All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

We thank the Leverhulme Trust (grant no. SAS-2017–045\2), the Explorers Club, Wilderness Wildlife Trust, Elephants for Africa and IDEAWILD for the funding of this work.