Function of meerkats' mobbing-like response to secondary predator cues: recruitment not teaching

Early detection of predators greatly improves prey escape and survival chances. By investigating cues predators leave behind, such as fur, urine, faeces or feathers (secondary predator cues), prey may gain vital information about predators in the vicinity. Meerkats, Suricata suricatta , display an unusual mobbing-like response upon encountering secondary predator cues, not reported in any other species. The function of this behaviour is unclear because, unlike mobbing of a live predator, it cannot yield the primary bene ﬁ t of driving the threat away. Here we examine two potential functions. First, we hy-pothesized that older group members may exhibit exaggerated mobbing-like responses towards sec- ondary predator cues as a form of teaching to promote learning in naïve pups. Meerkats are known to teach pups hunting skills, but there is no evidence that any nonhuman species teaches across multiple contexts. Mobbing-like responses could also function to facilitate group member recruitment, stimu-lating information gathering and collective decision making by the group. Using experimental pre- sentations of secondary predator cues in the presence and absence of pups, in combination with 20 years of data on natural encounters with secondary predator cues, we found that older group members' response frequency was lower when pups were present and that response intensity declined as the proportion of pups to total group member number increased. Conversely, the intensity of responses increased when a greater proportion of group members interacted with the cue and varied with predator cue type. The response frequency to natural encounters with secondary predator cues was greater in larger groups, and the distance travelled by the group decreased following an encounter with secondary predator cues. These results suggest that meerkats do not show exaggerated responses to secondary predator cues as a form of teaching. Mobbing-like responses more likely serve to increase recruitment of others to investigate the cue and inform defensive group behaviour. © The on behalf of The Association for the Study of Animal Behaviour.

The ability of prey animals to mount appropriate defensive behaviours in the face of predation is vital to survival. Accurate assessments of current predation risk can inform appropriate defence behaviours, limiting unnecessary time and energy expenditure on nonacute or nonimmediate threats. Individuals can gauge predation risk through personal assessment of the current situation and from the risk assessments of others, by using social information (Crane & Ferrari, 2013;Dall et al., 2005). Access to social information is thought to be a key benefit of group living, by aiding in detecting, recognizing and responding appropriately to predators (Beauchamp et al., 2012;Crane & Ferrari, 2013). In animals across many taxa, social learning plays an important role in shaping the development of appropriate responses to predators (see reviews : Crane & Ferrari, 2013;Griffin, 2004). Animals can use a range of cues to inform their defensive behaviours (Lima & Dill, 1990; Thorson et al., 1998), including direct visual, acoustic or olfactory cues indicating that a predator is currently present and indirect cues of increased risk such as time of day, habitat type and the remains of body parts. In addition, predators may also leave cues behind in the environment, such as fur, urine, faeces, feathers, scent markings and regurgitated pellets. Following Furrer and Manser (2009), we refer to these as secondary predator cues, although these cues are also referred to in the literature as direct cues (Nersesian et al., 2012;Z€ ottl et al., 2013) or indirect cues (Persons et al., 2001;Severud et al., 2011).
To our knowledge, meerkats are the only species to show such mobbing-like responses to secondary predator cues. While other mongoose species, such as dwarf mongooses, Helogale parvula, and banded mongooses, Mungos mungo, do recruit to and inspect secondary predator cues (Collier et al., 2017;Furrer & Manser, 2009), meerkats show a more overt, higher arousal behavioural response. When meerkats encounter secondary predator cues, they typically approach and investigate the cues, raising their tails, piloerecting (raising their fur) and making recruitment calls. These responses are all characteristic features of meerkat mobbing behaviour (Graw & Manser, 2007), but in contrast to true mobbing, the mobbing-like response towards secondary predator cues serves no function in deterring predators (see Fig. 1 for comparison). Thus, the potential benefit of responding to a secondary predator cue as if it were the predator itself is unclear, particularly given that the response is highly conspicuous and involves time and energy costs. Potential functions of the mobbing-like response towards secondary predator cues by meerkats could be as a form of teaching for naïve young and/or to inform and influence subsequent group behaviour.
Teaching is a form of active social learning whereby knowledgeable individuals invest in promoting learning of naïve individuals (Thornton & Raihani, 2008). According to established operational criteria, teaching involves (1) a knowledgeable individual, A, modifying its behaviour in the presence of a naïve observer, B, (2) individual A incurring a cost or no immediate benefit by doing so and, (3) as a result of A's behaviour, individual B acquiring a skill or knowledge faster than it would have otherwise, if at all (Caro & Hauser, 1992). Teaching was once regarded as uniquely human, but there is now strong experimental evidence for teaching in a handful of nonhuman animals (Lee et al., 2021;Thornton & Raihani, 2008), including meerkats, in which adults teach naïve pups hunting skills (Thornton & McAuliffe, 2006). As yet, there is no evidence that any nonhuman animal teaches across multiple contexts, but we hypothesize that meerkats' unusual mobbing-like behaviour towards secondary predator cues could also act in teaching. Specifically, by exaggerating their mobbing-like responses in the presence of naïve pups (teaching criterion 1), older group members could incite the pups to approach and investigate the cue, providing relatively safe opportunities to learn about predator characteristics (e.g. odour) and appropriate behavioural responses (criterion 3). The exaggeration or increase in frequency of mobbing-like behaviour is likely to entail short-term costs (criterion 2) as it requires time, energy and loss of foraging opportunities and, being highly conspicuous, could draw the attention of predators.
It is also possible that meerkats' unusual mobbing-like response to secondary predator cues could serve to inform and influence subsequent group behaviour. To maintain one of the primary benefits of group living, reduced risk of predation (Caro, 2005;Krause & Ruxton, 2002), it may be necessary for all group members to be informed of current risks. Recruitment to secondary predator cues may allow group members to gather detailed information about the threat through inspection, using this information to adjust defensive behaviours and aid increased group cohesion by bringing dispersed foraging group members to a focal point. The overt mobbing-like response may increase the likelihood of recruitment by providing a clear signal of the presence of valuable information and denoting a high-level threat. Previous work has shown that meerkats increase individual vigilance during and immediately following an experimental encounter with secondary predator cues (Z€ ottl et al., 2013), but this work focused primarily on immediate changes in behaviour and enhanced predator detection in the minutes following an encounter with secondary predator cues. Whether encounters with secondary predator cues generate more prolonged changes to group behaviour (such as over the following hour) is unknown. Moreover, encounters with secondary predator cues could lead to other, as yet unexamined, changes in behaviour such as movement and investment in pups. Evaluating additional behavioural changes is important to improve our understanding of the use and importance of secondary predator cues in informing defensive behaviours. Meerkats are cooperative breeders living in the arid regions of southern Africa, in groups of 3e47 individuals, averaging 15 individuals (Clutton-Brock & Manser, 2016). They forage as a cohesive group and use vocalizations to alert others about potential threats (Manser, 2001;Manser et al., 2001), maintain proximity between individuals (Gall & Manser, 2017) and make collective decisions to move from one foraging patch to the other based on quorum decisions (Bousquet et al., 2011). Pups begin foraging with the group at around 20e25 days old (Clutton-Brock et al., 2000). All group members over the age of 3 months old contribute to offspring care and provisioning of pups (i.e. individuals <3 months old; Clutton-Brock & Manser, 2016). Meerkat pups make extensive use of social information in developing foraging skills and antipredator responses (Holl en et al., 2008;Holl en & Manser, 2006;Thornton & Clutton-Brock, 2011) and are known to learn hunting skills via teaching (Thornton & McAuliffe, 2006). Meerkats mob a variety of threats, primarily predators such as snakes, wildcats and other mammalian threats, but also occasionally nonpredators such as hares and antelope (see Graw & Manser, 2007 for detailed list). They also show a mobbing-like response to scents and objects such as fur, predator faeces or urine, owl pellets and feathers (Z€ ottl et al., 2013). However, personal observations suggest that young individuals are unlikely to inspect or react to secondary predator cues when they first come across them, although they will participate in mobbing-like response recruitment events following the response of a more experienced older group member (I. Driscoll, personal observation).
To investigate the two potential functions of meerkats' mobbing-like responses to secondary predator cues, we used a combination of experiments and observational data. To examine whether these responses may serve as a form of teaching, we conducted experimental presentations of secondary predator cues to older group members (group members >3 months) in the presence or absence of pups. Here, we predicted that, as per the first criterion of Caro and Hauser's (1992) definition of teaching, older group members should increase the intensity of their mobbing-like response (interaction duration, raised tails and piloerection frequency and duration) when pups are present and when cues are novel to the pups (Prediction 1). In addition, we used long-term observational data of natural encounters with secondary predator cues to confirm the experimental findings and investigate group level behavioural changes following encounters with secondary predator cues. We began by investigating the social conditions influencing recruitment rate to secondary predator cues, particularly the presence of pups, to further test Prediction 1 and confirm the results of the experiment. We then investigated group behavioural changes following an encounter with secondary predator cues, examining alarm-calling rate, distance travelled by the group and the rate of pup provisioning in the hour before and after a recruitment event to secondary predator cues. We predicted that, if responding to secondary predator cues functions in initiating group defensive behaviours, then alarm-calling rate should increase following an encounter with secondary predator cues due to an increase in perceived risk and threat sensitivity (Prediction 2). We also predicted that groups would increase the distance travelled in the hour following a recruitment event to secondary predator cues so as to move out of high-risk areas (Prediction 3) and reduce their rates of pup provisioning as a result of a trade-off with defensive responses (Prediction 4).

Study Site and Population
We conducted experimental presentations on six groups of wild meerkats and analysed 20 years of behavioural data collected as part of long-term observations of the meerkat population at the Kalahari Meerkat Project, Kuruman River Reserve (26 59 0 S, 21 50 0 E) in South Africa (Clutton-Brock et al., 1998). For details about habitat and climate, see Russell et al. (2002). All individuals were habituated to human observation (<1 m) and identifiable from unique dye mark patterns on their backs (Jordan et al., 2007). Life history was known for most individuals from birth, including age, sex and dominance status, with the exception of immigrating individuals. We conducted experimental presentations between 1 December 2017 and 23 April 2018. For demographic information on the experimental presentation groups, see Appendix (Table A1). We analysed long-term data from 11 April 1999 to 30 April 2019, using only observations with complete records for each analysis.

Cues
We presented two different cue types: (1) domestic cat, Felis catus, urine samples, obtained from local veterinary surgeries during medical procedures and stored in a À20 C freezer, and (2) African wildcat, Felis lybica, fur samples, obtained from a recently deceased individual found (within 6 h of death) on the reserve and stored in a À20 C freezer. Both domestic cats and wildcats are common predators on the reserve. Older group members were likely to have encountered the predators and their associated cues previously but, given the relatively low frequency of predator encounters, it was highly likely that pups were naïve. Pilot studies determined that older group members responded to both predator cues with a mobbing-like response. Samples were portioned into 5 ml of urine or 0.1 g of fur and stored at À20 C. To ensure that meerkats were responding specifically to the cues and not the experimental set-up, equivalent quantities of water and dry grass were used as matched controls for the urine and fur, respectively. We removed cues from the freezer to defrost 2e3 h before presentation, keeping them in a cool bag with ice blocks until presentation, and wore latex gloves to avoid contaminating the cues with human scent.

Presentations
We conducted presentations of urine and fur to six groups while the groups were foraging. The first trial at a group was after pups had been born, but were still at the burrow with a babysitter, and had not begun foraging with the group (no pups: NP). Conducting the presentations within this short time frame allowed ecological and social conditions to be kept as similar as possible across trials, while still allowing comparison of trials with and without pups. Pups began foraging with the group at around 3e4 weeks of age, but initially spent much of their time in sheltered locations (e.g. in boltholes or under bushes) begging for food and did not participate in group alarm or mobbing events. The second trial, with pups present (pups present 1: PP1), was conducted when pups were approximately 6e7 weeks (21 ± 3 days after they began foraging with the group) and spent the majority of the time actively moving between the older group members as caretakers. Subsequent trials (pups present 2 and 3: PP2 and PP3) were conducted at 1-week (7 ± 1 day) intervals from PP1 on. For trials 1e3 (NP, PP1, PP2), the same cue type (e.g. cat urine) was used, and for trial 4 (PP3), a different cue (e.g. cat fur) was used, representing a novel cue (Table 1). For the three pups present trials (PP1, PP2, PP3), all pups in the group were present. Half of the groups were presented one combination of cues (Group A: urine, urine, urine, fur) and the other were presented the opposite (Group B: fur, fur, fur, urine). For each trial, a cue was presented along with a control (water or grass), with a randomized order and at least 30 min between presentations.
Cues were presented 30 min after the group had left the burrow in the morning to begin foraging, and after at least 10 min of normal foraging behaviour following an alarm event, so as to minimize the effect of any previous stress on responses to the presentation. The cues were presented in a petri dish filled with sand at the end of a 1 m pole (Fig. 1a), to reduce association of cues with the human presenter. At the start of each trial, we presented the relevant cue to a randomly selected target individual (older group member, >6 months old) from the group. If the individual did not initially interact with the cue, we presented it again up to three times to the same individual. If this still did not elicit inspection, the cue was presented to another randomly chosen individual (>6 months old) to prevent overexposing any one individual to the cue. A trial began once an individual began interacting with the cue, defined as approach to within 0.3 m facing the cue directly (see Appendix, Table A2, for detailed ethogram). Trials were conducted at least 1 week apart to reduce possible habituation to the cues. Presentations were videorecorded using a GoPro (Hero 4) and audio was recorded with a microphone (Sennheiser ME 66 with a K6 powering module, sampling frequency 44.2 kHz, 16 bits accuracy) connected to a recorder (Marantz Solid State Recorder PMD661 MKII) at a distance of approximately 1e1.5 m from the cue presentation (see Supplementary Material Video S1 example).

Response analysis
Video recordings were coded using the open-source software BORIS (Friard & Gamba, 2016), noting the behaviours of each individual that interacted with the cue. For each presentation, we recorded whether or not an individual present in the group interacted, resulting in 92 inspection events out of a possible 205, representative of every individual in every trial contributing an observation. Of those that did interact, we recorded the duration of inspection (N ¼ 92 events by 48 individuals), whether an individual raised their tail (N ¼ 92) and, for those that did raise their tails, the duration of tail raising (N ¼ 70 events by 43 individuals), as indicators of response intensity (see Appendix, Table A2, for ethogram). We also recorded instances of piloerection but did not include these in our analysis as they were strongly correlated with tail raising (only two individuals that piloerected did not raise their tails). Presentations that elicited no response from the initial target individual were not included in the analysis unless subsequent presentations to the rest of the group did not elicit a response. For six out of the 24 predator cue presentations analysed, it was necessary to present cues more than once to elicit a response. There was one instance in which all group members failed to respond following three secondary predator cue presentations; here we used the trial to the original target individual for the analysis. We also recorded the calls given, using RavenLite (Bioacoustics Research Program, 2016) to determine the type of recruitment call given (high or low urgency) in response to the presentation. We classified the urgency of recruitment calls based on the acoustic structure (outlined and defined in: Manser, 2001;Manser et al., 2001).

Observational data collection
As part of the Kalahari meerkat project long-term data collection, meerkat groups were visited at least every 3 days for a minimum of 1 h in either the morning following the group leaving the sleeping burrow and/or the evening prior to the group's return to the sleeping burrow. During sessions, behavioural data was recorded ad libitum (every time a behaviour was observed; Altmann, 1974). For definitions of behaviours and other data recorded and analysed as part of this study, see Table 2.
Recruitment events were recorded as alarm events (Table 2) and defined as when the group recruited to a stimulus with erected hair and tails, made recruitment calls and often spat and growled (Manser et al., 2014). The eliciting stimulus was also recorded with the animal type or as 'scents' or 'objects', such as domestic cat, domestic dog, Canis lupus familiaris, caracal, Caracal caracal, or bateared fox, Otocyon megalotis, urine, scent marks, faeces or hair (Manser, 2001;Z€ ottl et al., 2013). If the type of threat responded to was not known, it was recorded as 'unknown' and excluded from our analyses. Group size and composition was included to investigate the effect of pups (Prediction 1) as well as the role of group size on antipredator responses (Beauchamp, 2003;Elgar, 1989). The daily maximum temperature and the total rainfall measured at the study site over the previous 30 days were used to control for current habitat conditions and food abundance, as these may influence investment in costly behaviours (Hodge et al., 2009;Thornton, 2008;Wiley & Ridley, 2016).

Data Analysis
Recruitment event rate to secondary predator cues Recruitment event rate to secondary predator cues (scents and objects) was calculated per group. We analysed recorded data from 54 groups over 20 years, comprising a total of 3705 recruitment events to secondary predator cues, with 131 289 observation hours over 52 776 sessions. Recruitment rate was calculated over a 1month period, dividing the number of recruitment events by the number of hours of ad libitum data recorded, to control for observation time.

Behavioural changes following a recruitment event to secondary predator cues
To examine behavioural changes following a recruitment event to secondary predator cues, we compared the total hourly number of alarm events (not including recruitment events), distance travelled by the group and per-pup provisioning rate in the hour before and after each recruitment event to secondary predator cues. The total number of alarm events to potential threats (as outlined above) in 1 h was used to indicate changes in threat perception. The distance travelled in 1 h, calculated from the GPS fixes, was used to determine changes in the rate of movement following a recruitment event. The hourly per-pup provisioning rate (the number of pup feed events recorded divided by the number of pups present in the group) was used to assess the effect of a recruitment event on the maintenance of pup care.

Ethical Note
The experiment was carried out following methodological approval from the Biosciences Ethics Panel of the University of Exeter and adhered to the standards outlined in the ASAB/ABS (2012) Guidelines for the treatment of animals in behavioural research and teaching. Both the experiment conducted and longterm data collected within the course of this study fall under the permission of the ethical committee of Pretoria University and the Northern Cape Conservation Service, South Africa (experiment: EC031-17; long-term data: EC011-10, EC048-12, EC010-13, EC031-13, EC047-16) and were carried out adhering to the approved guidelines in these permits. Our experiments and observations had no lasting impact on the meerkats' welfare or environment; no individuals were handled or trapped as part of the study. Presentations of secondary predator cues were spaced at 1-week intervals to not substantially increase encounters with secondary predator cues by the group within the study period.

Statistical Analysis
Statistical analysis was conducted using RStudio version 2021.09.0 (RStudioTeam, 2021), with packages 'lme4' and 'glmmTMB' for mixed models. Model assumptions were validated using residual plot distribution techniques. An information-theoretic (IT) approach was applied for model selection, using Akaike's information criterion corrected for small sample sizes (AICc) to rank the models following the approach used by Richards et al. (2011). Model building was conducted using combinations of fixed effects defined a priori. Models within AICc 6 of the model with the lowest AICc value formed the 'top set'. We used model averaging when the top set contained more than one model to calculate model-averaged parameter estimates (Dormann et al., 2018;Grueber et al., 2011).

Experimental presentations of secondary predator cues
We used linear mixed models (LMMs) and generalized linear mixed models (GLMMs) to examine the factors influencing the behavioural responses of all older group members (>3 months old) present for the experimental predator cue presentations. Individual identity (ID) was fitted as a random term in analyses of individual responses. Group ID was initially included as a random term but removed as it explained zero residual variance. To test Prediction 1, all sets of models (aed) included all combinations of the effects of treatment (NP, PP1, PP2, PP3), cue type (fur, urine), proportion of the group composed by pups (0e3 months old). In addition, models also included the total number of older group members (>3 months old), the proportion of the group interacting with the cue and the highest urgency level of call type heard in the group before each individual was recruited. As the original target individual to whom the cue was presented could not, by definition, have heard any prior calls made in response to the cue, call type was categorized as target individual, no call, low urgency or high urgency. Target individual age, sex and dominance rank were initially included in the models but removed to reduce complexity, as they never ranked in the top set during model selection.
As the number of pups in the NP treatment was, by definition, zero, the effects of treatment and number of pups could be correlated. To address this, we also ran the analysis with the results of the NP treatment excluded. The results of these models were qualitatively very similar to those conducted on the full data set (see Appendix, Table A3).
To determine how the explanatory terms outlined above influenced whether or not an individual interacted with the cue, we used GLMMs with a binary (0/1) response term, binomial error structure and logit link function (analyses of model a). We excluded the response of the original target individual presented to from this analysis as this inspection signified the beginning of the trial. Among those individuals that did interact (including target), we examined the factors influencing the duration of inspections using LMMs, with duration log-transformed to meet model assumptions (analyses of model b). We also examined whether or not each of these interacting individuals raised their tail using GLMMs with a binary response term (0/1) (analyses of model c). For analyses of model d, we grouped low urgency and no recruitment calls to allow model convergence, as there were only two instances of individuals raising their tails following no recruitment calls. Among those individuals that did raise their tails, we examined the factors influencing the duration of an individual's tail raising using LMMs, with duration log-transformed to meet model assumptions of normal data distribution (analyses of model d).
Long-term observation of recruitment events to secondary predator cues Recruitment event rate. To analyse the factors influencing rates of recruitment in response to secondary predator cues (analyses of model e), we used GLMMs. The total number of monthly recruitment events to secondary predator cues per group was analysed using the log of total number of hours recorded as the offset, with a Poisson error structure. To further investigate Prediction 1 and understand the influence of social factors, we included the average proportion of pups (proportion of the foraging group that was pups) and group size (average number of individuals in the group during the month) as explanatory terms. We also included whether or not pups were foraging with the group as a categorical explanatory factor to test whether pup presence or absence was enough to influence recruitment frequency. The average daily maximum Table 2 Descriptions of data recorded as part of long-term data collection of the population and analysed as part of this study

Data recorded Description
Alarm events When >50% of the group (group members >3 months old) respond to a potential threat. Recording the response given by most of the group, which included: look briefly, watch continuously, move, move to bolthole, move below ground and recruit. Also recorded the type of threat responded to Pup provisioning An individual (nonpup) gives a pup a food item GPS location GPS fixes taken from the centre of the group at approximately 15 min intervals during a session (accuracy: 95% of fixes within 5 m; eTrex H, Garmin International Inc., Olathe, KS, U.S.A.). Taken either from the time a group left the sleeping burrow or until the group returned to the sleeping burrow Group size and composition The number of individuals (including pups) present in a session, their sex and age class/dominance categories Daily maximum temperature ( C) Daily maximum temperature measured at the study site Daily rainfall (mm) Daily total rainfall measured at the study site temperature and total rainfall for the previous 30 days were included to control for the effect of abiotic conditions on recruitment event frequency. Sex ratio in the group was initially included in the analysis but removed to reduce complexity as it never ranked within 6 AICc of the top model. The group identity nested within year was included as a random term to account for repeated measures within groups and variation in group characteristics over the study period.
Behavioural changes following a recruitment event to secondary predator cues. We used GLMMs to analyse behavioural changes before and after a recruitment event to secondary predator cues, examining for Prediction 2, the hourly number of alarm events (analyses of model f), for Prediction 3, the metres travelled per hour (analyses of model g), and for Prediction 4, the hourly provisioning rate per pup (analyses of model h). Analyses of model f were fitted with a negative binomial error structure; analyses of model g, the hourly distance travelled, were fitted with a Poisson error structure; analyses of model h, the per-pup provisioning rate, were analysed as the hourly number of pup-provisioning events with log of the number of pups in the group as the offset, fitted with a zeroinflated negative binomial error structure to account for zero inflation, as it is likely that pup-provisioning events were missed by observers, resulting in an inflated number of zeros. The same explanatory terms outlined for recruitment event rate were used. Analyses of models feh also included whether the response was in the hour before or after the mobbing to test whether there was a behavioural change. The group size, proportion of pups and daily maximum temperature recorded on the day of the recruitment event were used rather than the monthly average. Model (h) did not include whether pups were foraging with the group as provisioning could only occur if pups were present. The interactions of both before/after and the presence of pups with all other fixed effects were used as both these factors may have interacted with the other factors to influence behaviour. The interaction of average daily temperature with 30-day rainfall was included as temperature and rainfall-driven food availability and body condition are generally closely linked (English et al., 2012). Random terms of group nested within year was also included, as well as a unique event ID for each recruitment event to pair the hour before and after, accounting for the nonindependence of this data.

Inspection Probability (Y/N)
For the 205 possible inspection events (with each older group members present per presentation representing a possible event), there were 92 inspections of the presented cues. Of these 92, 22 inspections were by the original target individuals to whom the cue was presented and the remaining 70 were subsequent recruits. Analyses suggest that the likelihood of interacting with the cue was not affected by the presence of pups, or any of the other factors analysed. GLMM analyses produced four models in the top set, of which none contained variables with a robust effect following model averaging (see Appendix, Table A4 for model selection table  and Table 3 for model-averaging output).

Inspection Duration
Individuals inspected the predator cues an average (± SE) of 29.6 ± 2.6 s. There was no evidence to support the prediction that older group members increase inspection duration in the presence of pups. GLMM analyses produced one model in the top set (model b5; Appendix, Table A5), containing only the predator cue type presented, with individuals interacting longer with fur cues (36.9 ± 3.8 s) than with urine cues (23.0 ± 3.4 s; LMM: estimate ± SE ¼ 3.3 ± 0.14, confidence interval, CI (3.03, 3.57); Fig. 2a, Table 3). Interact: whether an individual interacted with the cue (y/n); interact duration: duration of an individual's interaction; tail raised: whether an individual raised their tail (y/n); tail raised duration: duration that an individual raised their tail; SPC recruitment rate: hourly rate of recruitment events to secondary predator cues (SPCs) per group in a month; alarm-calling rate: hourly alarm-calling rate; distance travelled: distance travelled per hour; per-pup provisioning rate: rate of food provisioning per pup per hour.

Tail Raised (Y/N)
Among those individuals that interacted with the predator cue, 70 out of 92 raised their tails. Contrary to our predictions, meerkats were less likely to raise their tails when there were more pups in the group. GLMM analyses produced one model in the top set (model c13: Appendix, Table A6). Model c13 included the interaction of the proportion of the group made up by pups and the number of older group members present in the group: individuals were more likely to raise their tails when in larger groups, with a positive effect of the proportion of pups (estimate ± SE ¼ À2.79 ± 1.1, CI (À4.94, À0.64); Fig. 2b, c, Table 3).

Tail Raised Duration
Individuals raised their tails for 0.5e57.0 s (mean ± -SE ¼ 13.9 ± 1.5 s). Analyses suggested that, contrary to our predictions, there was no effect of the proportion of pups on tailraising duration. LMM analyses produced one model in the top set (model d13; Appendix, Table A7), which included the number of older group members present as a positive predictor of tail raised duration (estimate ± SE ¼ 0.14 ± 0.03, CI (0.03, 0.18); Fig. 2d, Table 3). The model also included the interaction of proportion of pups with the number of older group members, but this did not have a robust effect (LMM: estimate ± SE ¼ À0.47 ± 0.3, CI (À1.02, 0.09); Table 3).

Recruitment Rate to Secondary Predator Cues
The hourly recruitment rate for each group over a 1-month period ranged from 0 to 0.28 (mean ± SE ¼ 0.03 ± 0.00), with a total of 3705 recruitment events to secondary predator cues recorded. Analyses suggested that recruitment event frequency to secondary predator cues was affected by the proportion of pups and by the average total group size (all group members including pups). GLMM analyses produced one model in the top set (model e11; Appendix, Table A8 (À1.94, À0.07); Fig. 3a, Table 3), the positive effect of average group size for that month (estimate ± SE ¼ 0.04 ± 0.00, CI (0.03, 0.05); Fig. 3b, Table 3) and the interaction between the two (although this interaction was not robust: estimate ± SE ¼ À0.03 ± 0.03, CI (À0.09, 0.03); Table 3). Recruitment events to secondary predator cues were more frequent with increasing group size (Fig. 3a).

Alarm-calling Rate
Meerkats produced 0e19 alarm calls per hour (mean ± -SE ¼ 2.23 ± 0.03). The rate of alarm calling was mainly influenced by abiotic conditions, with the rate of alarm calling declining with increasing temperature. GLMM analyses produced one model in the top set (model f7: Appendix, Table A9). Model f7 contained the maximum temperature on the day of presentation, the time period (hour before and after the recruitment event) and the interaction between temperature and time period. However, of these variables, only temperature was robust, with the rate of alarm calling declining with increasing temperature (estimate ± SE ¼ À0.02 ± 0.01, CI (À0.02, À0.01); Fig. 3c, Table 3).

Distance Travelled
The hourly distance travelled by meerkat groups declined in the hour following a recruitment event to secondary predator cues and was influenced by abiotic conditions. The distance that meerkat groups travelled ranged from 0 m to 1431 m per hour of observation (mean ± SE ¼ 173.6 ± 2.1 m). GLMM analyses produced one model in the top set (model g5; Appendix, Table A10). Model g5 included the time period (hour before or after the mobbing event), group size and the interaction between the two. The interaction between time period and group size was robust (estimate ± SE ¼ 0.01 ± 0.01, CI (0.01, 0.01); Table 3). Small meerkat groups travelled similar distances in the hour before and after recruitment events to secondary predator cues, whereas larger meerkat groups travelled slightly greater distances before recruitment events and reduced distances following recruitment events (Fig. 3d).

Provisioning Rate
The hourly provisioning rate per pup ranged from 0 to 11.5 (mean ± SE ¼ 1.20 ± 0.05). This did not change in the hour following a recruitment event to secondary predator cues and was not influenced by any of the variables analysed. GLMM analyses produced three models in the top set (model h11, h10 and h3; Appendix, Table A11), none of which contained variables with a robust effect following model averaging (see Table 3 for modelaveraging output).

DISCUSSION
Secondary predator cues provide an opportunity for individuals to assess current predation risk without a dangerous direct encounter, but the function of meerkats' highly exaggerated, mobbing-like recruitment response is unclear. We tested whether older group members may use exaggerated mobbing-like responses to secondary predator cues to teach naïve pups how to recognize and respond to threats, but our results provided no evidence that this is the case. Contrary to our predictions, we found that older group members reduced rather than increased their mobbing-like response intensity when pups were present, particularly when more pups were present. We also examined whether the information gathered during active recruitment to secondary predator cues was used to inform subsequent defensive behaviours. Following an encounter with secondary predator cues, distance travelled decreased, suggesting that inspections of secondary predator cues influence group level defensive behaviours and collective decisions. The combination of these results strongly suggest that meerkats do not use mobbing-like responses towards secondary predator cues as a form of teaching, but instead to promote recruitment of other older group members to investigate cues and inform group defensive responses.
We predicted that older group members would exaggerate their mobbing-like response when pups were present and foraging with the group and that responses would be particularly exaggerated when cues were novel to pups (Prediction 1). None of the analyses from the experimental data supported these predictions, as experimental treatment did not appear to influence any of the responses investigated. We found some evidence to suggest an inhibitory effect of pups on response intensity, with the likelihood to tail raise decreasing with an increasing proportion of pups in the group. This effect on response intensity (tail-raising probability) was retained even when the no pups present treatment (NP) was excluded from the analysis. The reduction in response intensity could reflect the additional costs associated with provisioning pups, limiting investment into other activities. This is supported by our finding that, in the observational data, there was no influence of recruitment to secondary predator cues on pup provisioning rate. Additionally, as the high intensity of a mobbing-like response is by definition conspicuous, reducing intensity when vulnerable pups are present may reduce conspicuousness and risk to pups in an area of higher perceived predation risk. Meerkats have been observed leading pups away from a location during mobbing of a live predator and therefore away from an area of increased risk (M. Manser, personal observation). Thus, although meerkats are known to teach their pups how to hunt effectively (Thornton & McAuliffe, 2006), they do not appear to use responses to secondary predator cues to teach pups about potential predators.
If the mobbing-like response to secondary predator cues does not play a role in teaching naïve pups, what could be the function of this unusual behaviour? One possible explanation is that the mobbing-like response to secondary predator cues is a maladaptive by-product of arousal, representing a misidentification of a secondary predator cue as an actual threat. Individuals clearly responded to the secondary predator cues, but not the controls, as threats, behaving similarly to how they would respond to a predator (Graw & Manser, 2007). Individuals tended to continue the mobbing-like behaviours while investigating the cues, directly sniffing and scratching them, suggesting no error in classification and an awareness that the cue itself was not a threat. Additionally, although the mobbing-like response to secondary predator cues lacks the major costs associated with mobbing (injury, death), there are still substantial energetic, time, opportunity and conspicuousness costs of the mobbing-like response. If there were no benefit gained from such a costly response to secondary predator cues, one may expect selection to act against the persistence of this behaviour.
Arguably, a more plausible explanation is that the mobbing-like response acts to recruit group members to a salient focal point, facilitating information transfer, and coordinate subsequent responses. The raising of group knowledge and alertness through recruitment to secondary predator cues can reduce risk to all members, raising vigilance and increasing the speed of potential predator detection (Z€ ottl et al., 2013). A mobbing-like response may increase the probability of recruiting other group members by providing a visually and acoustically conspicuous, localizable signal of risk. Consistent with this, our results indicate more frequent natural recruitment events to secondary predator cues in larger groups. In larger groups where individuals may be more dispersed (Focardi & Pecchioli, 2005), secondary predator cues may be encountered more frequently as a result of sheer numbers, and signals may need to be more conspicuous to increase the probability of others receiving the signal. The increase in response intensity in larger groups may be further enhanced by social facilitation, when an animal changes the frequency or intensity of a behaviour as a result of others performing that behaviour (Clayton, 1978), and can result in the expression of antipredator behaviours as a result of copying conspecifics (Crane et al., 2012;Pays et al., 2009). This is supported by our results showing a higher response intensity when more older group members were present, with both probability and duration of tail raising increasing. Recruitment to cues may provide a benefit by bringing the group to a focal location, facilitating informed decision making and cohesive movement away, possibly through quorum sensing. White-breasted mesites, Mesitornis variegatus, for example, increase group cohesion following an alarm event (Gamero & Kappeler, 2015), and convict cichlids, Amatitlania nigrofasciata, do so in response to conspecific alarm cues (Brown et al., 2004). Whether encounters with secondary predator cues do result in increased group cohesion could be examined by experimentally by investigating whether interindividual distance and the spread of the whole group changes following a recruitment event.
Contrary to our predictions, meerkats did not increase their rate of alarm calling (Prediction 2) and larger groups reduced rather than increased the distance travelled after encountering a secondary predator cue (Prediction 3). We predicted that alarm-calling rate would increase following a natural encounter with secondary predator cues as it is thought that these cues act as indicators of an increase in actual risk, i.e. a predator in the vicinity, or lead to increased sensitivity to potential threats (Apfelbach et al., 2005;Ayon et al., 2017). Previous work has shown that meerkats detect a predator model faster following an encounter with secondary predator cues (Z€ ottl et al., 2013), indicating that secondary predator cues may be a reliable indicator of a predator in close proximity and aid in locating the threat. A potential explanation for the lack of change in alarm calling in our study is that the presence of human observers scared away natural terrestrial predators, so although the cues we used are generally indicators of predator presence, the likelihood of a natural encounter was lower during observation periods. Also, in contrast to our predictions, the distance travelled by larger groups was reduced following a natural encounter with secondary predator cues. This reduction in distance travelled may be explained by the time taken to inspect and recruit to the secondary predator cues. The distance travelled was more greatly reduced in larger groups following a natural encounter with secondary predator cues than it was in the hour prior to the recruitment event. This further supports the role of recruitment to secondary predator cues in group coordination. In larger groups, it may be more difficult to maintain defensive cohesive movement to reduce risk of predation, resulting in a trade-off with speed (Franks et al., 2013). Additionally, defensive behaviours such as increased vigilance, previously shown to increase in meerkats and several other species following an encounter with secondary predator cues (Garvey et al., 2016;Monclús et al., 2005;Tanis et al., 2018;Zidar & Løvlie, 2012;Z€ ottl et al., 2013), may further slow movement in an area of increased risk. Thus, recruiting group members to respond to a cue and increase other defensive behaviours may reduce the time used for moving, resulting in reduced movement in the period following an encounter with secondary predator cues.
Many animals inspect secondary predator cues (Amo et al., 2011;Belton et al., 2007;Brown & Godin, 1999;Garvey et al., 2016;Mella et al., 2014), but recruitment to them is rare and, to our knowledge, has only been described in social mongoose species, with meerkats seemingly unique in displaying a mobbing-like response (Collier et al., 2017;Furrer & Manser, 2009). This recruitment may facilitate collective decision making in the face of threat uncertainty, rather than rely on a single encountering individual producing an alarm call and making the decision on threat level and type alone. Recruitment of individuals through the exaggerated mobbing-like response may ensure the transfer of threat-specific information, more so than may be obtained from an alarm call, better informing the groups defensive behaviour. While there is scant evidence of recruitment to secondary predator cues outside of mongoose species, this does not mean it does not occur. An analogous behaviour may be the inspection of secondary predator cues in several fish species and the release of alarm cues to alert conspecifics to the threat (Brown & Godin, 1999;Brown & Magnavacca, 2003). However, it is not yet clear whether these alarm cues are used only to alert conspecifics to the threat, or also to initiate recruitment and inspection.
The lack of evidence for teaching in this context may provide support for the idea that, in contrast to human teaching, which occurs across many contexts, nonhuman teaching is an adaptation to promote context-specific learning (Thornton & Raihani, 2008). Teaching is a costly form of cooperative behaviour and it is only likely to be favoured by selection if individual learning or passive social learning involves substantial costs, or learning opportunities are lacking (Fogarty et al., 2011;Thornton & Raihani, 2008). In the context of the mobbing-like response to secondary predator cues, meerkat pups may have sufficient inadvertent learning opportunities through observing knowledgeable group members' highintensity responses to secondary predator cues, without the need for exaggerated responses. For example, meerkat pups' responses to alarm calls become more adult-like with age, suggesting the development of experience-dependent appropriate responses to alarm calls, likely as a result of passive exposure to older group members' behaviour (Holl en et al., 2008;Holl en & Manser, 2006). Instead of functioning in teaching, it seems that the unusual mobbing-like response to encounters with secondary predator cues may be aiding in recruiting group members. During this recruitment, individuals are able to investigate and gather threat-specific information, providing an indication of risk and informing group level defensive behaviours. Increasing our understanding of how animals respond to secondary predator cues may provide important insights into the role of information use in shaping collective defensive behaviours.

Data Availability
Data will be made available on request.

Declaration of Interest
We declare we have no competing interests.

Acknowledgments
We thank the Kalahari Research Trust and the Northern Cape Conservation Authority for research permission (FAUNA 1020/ 2016). We also thank Dave Gaynor and Tim Vink for organizing the field site and data base, as well as the managers, Coline Muller and Jacob Brown, and volunteers of the Kalahari Meerkat Project for organizing, providing support and helping to collect the life history data and maintain habituation of the meerkats. We thank Tim Clutton-Brock for access to the long-term data and insightful comments on this manuscript. Furthermore, we thank Michael Cant and Andrew Radford for providing valuable comments that helped improve this manuscript. I.D. was supported by the University of Exeter. A.T. was supported by a grant from the Human Frontier Science Project (RGP0049/2017). M.M. was supported by the University of Zurich. The long-term field site KMP was financed by the University of Cambridge, the University of Zurich and the MAVA Foundation. This article has relied on records of individual identities and/or life histories maintained by the KMP, which has been supported financially by the European Research Council (Grant No 294494 to T.H.C.B., since 1 July 2012) and the University of Zurich, as well as logistically by the Mammal Research Institute of the University of Pretoria.

Table A3
Comparison of models in the top set (AICc 6 of the model with the lowest AICc), comparing the full data set with the data set excluding the NP (no pups) treatment Models forming the top set of retained models were broadly consistent between the full data set and the data set excluding the NP treatment. The influential variables forming the top sets for both analyses were almost the same except there no effect of the proportion of pups and call type on interaction probability, the proportion recruited having a nonrobust effect on tail raising probability and tail-raising duration in the NP-excluded data set. There was also an influence of cue type on tail-raising duration in the NPexcluded data set but not in the full data set. Overall, the parameters influencing interaction probability and duration remained relatively similar, and the negative effect of the proportion of pups held for response intensity for both data sets (interact duration, tail raising probability and duration). The lack of an effect of the proportion of pups on interaction probability suggests that pup presence alone reduces an individual's likelihood of recruiting.

Table A2
Ethogram of the behaviours analysed in response to secondary predator cue presentations Behaviour Description Interact Duration of time spent interacting with the cue, when the individual was within 0.3 m of the cue following initial approach and exhibiting one of the following behaviours (indicating a direct interaction): facing the cue directly, touching and sniffing the cue, rocking back and forth facing the cue. Interaction ended when an individual was quadrupedally vigilant (scanning on four legs), on bipedal vigilance (scanning on two legs), or resumed foraging. A new interaction began if the individual started interacting again Tail raise Tail raised vertically above the body within 0.5 m of the cue (within close proximity). Recorded the duration of time until the tail was lowered below horizontal with the body Recruitment call The recruitment call type (low or high urgency) given in response to the cue presentation We used a different proximity distance for interact (0.3 m) and tail raise (0.5 m) as an interaction with the cue occurred when individuals were within one body length of the cue and directly interacting with it. We used a larger radius for tail raise as individuals raised their tails when interacting with and moving around the cue to a different position and also while performing vigilance.  Variables tested were the recruitment call category (high, low, none, individual presented to), treatment (NP, PP1, PP2, PP3), cue type (fur, urine), number of adults in the group, proportion of pups in the group and proportion of the group interacting with the cue. Variables tested were the recruitment call category (high, low/none, individual presented to), treatment (NP, PP1, PP2, PP3), cue type (fur, urine), number of adults in the group, proportion of pups in the group and proportion of the group interacting with the cue. Variables tested were the recruitment call category (high, low/none, individual presented to), treatment (NP, PP1, PP2, PP3), cue type (fur, urine), number of adults in the group, proportion of pups in the group and proportion of the group interacting with the cue. Variables tested were pups foraging with the group (y/n), previous 30-day total rainfall, previous 30 days average maximum daily temperature and proportion of pups in the group. Variables tested were the hour before or after a recruitment event (AlarmFac), pups foraging with the group (y/n) (PupsForage), proportion of pups in the group, cue type (scent or object) (PredCode), previous 30-day total rainfall and daily maximum temperature. Variables tested were the hour before or after a recruitment event (DistFac), pups foraging with the group (y/n) (PupsForage), proportion of pups in the group, cue type (scent or object) (PredCode), previous 30-day total rainfall and daily maximum temperature.