Behavioural response of workers to repeated intergroup encounters in the harvester ant Messor barbarus

The evolution of cooperation in animal societies is often associated with the evolution of hostility towards members of other groups. It is usually predicted that groups under attack from outsiders should respond by becoming more cohesive or cooperative. However, the responses of individuals to real or simulated intergroup encounters vary widely, for reasons that are poorly understood. We tested how groups of workers of the harvester ant, Messor barbarus, responded to exposure to members of a different colony versus members of their own colony, and how previous exposure to an intruder affected the intensity of the within-group response. We found that workers increased in activity and had more contact with one another immediately following exposure to an ant from a different colony, but also showed a similar behavioural response to presentations involving an ant from their own colony. However, exposure to an intruder from a different colony resulted in much stronger behavioural responses to a second intruder, encountered shortly afterwards. Our results are consistent with studies of social vertebrates which suggest that exposure to intruders results in increased social cohesion. Our results also show that exposure to an intruder primes group members to respond more strongly to future intrusions. Our findings highlight a disconnect between the assumptions of theoretical models which study the effect of intergroup conflict on social evolution over many generations, and the short-term behavioural responses that are the usual focus of studies of intergroup conflict in insects and vertebrates.

different colony, but also showed a similar behavioural response to presentations involving an ant 23 from their own colony. However, exposure to an intruder from a different colony resulted in much 24 stronger behavioural responses to a second intruder, encountered shortly afterwards. Our results 25 are consistent with studies of social vertebrates which suggest that exposure to intruders results in 26 increased social cohesion. Our results also show that exposure to an intruder primes group members 27 to respond more strongly to future intrusions. Our findings highlight a disconnect between the 28 assumptions of theoretical models which study the effect of intergroup conflict on social evolution 29 Introduction 34 Intergroup conflict is recognised as a major force influencing selection on social traits in organisms 35 ranging from insects to humans (Darwin 1871; Reeve  groups is predicted to influence selection for altruism and cooperation within groups on an 43 evolutionary timescale (i.e. over many generations), but is also predicted to affect the immediate, 44 short-term behavioural responses of individuals to each other and to outsiders. 45 46 Empirical tests of the prediction that groups under attack from competitors should become more 47 Here we test how simulated intergroup conflict affects affiliative behaviour and social cohesion in 85 the harvester ant, Messor barbarus, and whether past experience affects the intensity of conflict 86 responses. Specifically, we test the 'primed response' hypothesis which suggests that recent 87 encounters with intruders reflect a high probability of subsequent, potentially costly, encounters. 88 This hypothesis predicts that individuals that are exposed to intruders will exhibit elevated sensitivity 89 and responsiveness to future intrusions, and higher levels of affiliation and social cohesion. We 90 tested these predictions through an experiment in which we repeatedly exposed a group of workers 91 to an individual from their own or an unfamiliar colony. 92 93

Colony maintenance 95
Twenty M. barbarus colonies were sourced from commercial suppliers in Spain and the Netherlands 96 (AntHouse, Ants Kalytta, and Ant's Kingdom). Colonies were founded by multiple independent 97 nuptial flights, meaning that relatedness between colonies was unlikely to be high. Colonies were 98 kept in separate darkened nests made of 20 x 20 x 3 cm moulds of plaster of Paris, connected by a 99 tube to a separate clear-plastic foraging area, in incubators kept at 25°C and with a day-night light 100 regime. Colonies were checked 3 times a week, and water and food were added when necessary. 101 Colonies contained a queen and individuals belonging to a major caste (the larger size class in the 102 colony) and a minor caste (the smaller size class in the colony). Of the 20 colonies, 16 were used as 103 experimental colonies (average colony size ± SE = 20.4 individuals ± 1.1; average ratio of minor to 104 major caste individuals ± SE = 14.6 ± 0.7). Ants from the remaining 4 colonies (non-experimental 105 colonies) were used as intruder ants in staged encounters with experimental colonies. 106 107

Staged experimental encounters 108
The behaviour of the 16 experimental colonies was analysed in response to experimental encounters 109 with an intruder individual. Eight ants (1 major and 7 minors) were randomly selected from the 6 experimental colony and placed into a petri dish. Ants were recorded from above using a Canon 111 DSLR camera and a Panasonic HC-VX980EB-K camcorder in a dark room under red light. Ants were 112 left for 2 minutes to acclimatise before the start of the exposure experiment protocol. 113 114 After the 2 minute acclimatisation period, the ants were recorded for 10 minutes to generate 115 baseline measurements of behaviour (the 'before' exposure phase). Ants were then given one of 116 two exposure treatments: exposure to an 'intruder' ant (a randomly selected minor caste ant from 117 one of the non-experimental colonies), or exposure to a 'home' ant (a randomly selected minor 118 caste ant from their own colony). Ants were video recorded for a 10 minute period (the 'during' 119 exposure phase). Exposure to a 'home' ant acted as a matched control to enable us to rule out the 120 possibility that any changes in observed behaviour were the result of an increase in the number of 121 ants in the petri dish, rather than the identity of the presented ant. We did not simultaneously 122 expose experimental colonies to multiple presented individuals due to logistical constraints on the 123 size of non-experimental colonies. The presented ant was marked with a small dot of white enamel 124 paint applied to its head for identification during video analysis. 125 126 After 10 minutes of exposure, the presented ant was removed from the petri dish and ants were 127 video recorded for a further 10 minute period (the 'after' exposure phase). After a 20 minute rest 128 interval, the experiment was repeated to allow us to measure behavioural responses to a second, 129 subsequent intruder. Similarly, the presented ant was either a 'home' ant or an 'intruder' ant. Again, 130 there was a 10 minute period of video recording before, during and after the second exposure. Each 131 experimental colony was exposed to four treatments in total: Intruder-Intruder (II), Intruder-Home 132 (IH), Home-Home (HH), and Home-Intruder (HI). Each experimental colony received the four 133 treatments on separate days and in a randomised order. 134

Video analysis 136
Video footage of behaviour was analysed using PotPlayer version 1.7.13622. We recorded three 137 separate behaviours among ants in the experimental colony: time to first contact, activity, and social 138 contacts. Time to first contact between the presented ant and an ant from the experimental colony 139 (in seconds) was recorded in the 'during' exposure phase as a measure of the strength of response 140 to intruders. We recorded the caste (major or minor) of the ant to make the first contact with the 141 intruder. Activity was measured as the proportion of ants observed moving in the first 20 seconds of 142 each minute of the 'before', 'during' and 'after' exposure phases. The number of social contacts was 143 measured as the number of times two ants from the experimental colony touched heads or body 144 parts and was recorded during each minute of the 'before', 'during' and 'after' exposure phases. 145 Cases of grappling (when two ants bite and hold each other with the mandibles) between ants in the 146 experimental colony, and with presented ants were also recorded. However, since grappling was 147 very rare (21 observations in 8 out of 16 colonies), it was left out of formal statistical analysis. 148 Observations were not blind to treatment or phase. to allow the significance of main effects to be tested (Engqvist 2005). Separate analyses were 165 conducted on behavioural responses to first and second exposures to test for differences in 166 response to intruders compared to home ants (on first exposure), and then for differences in 167 response dependent on the first exposure. Post hoc Tukey's all-pairwise comparisons of means were 168 conducted using the 'glht' function in the 'multcomp' package (Hothorn et al. 2008(Hothorn et al. , 2016 to test for 169 differences between levels of significant main effects of phase ('before', 'during' and 'after'), and 170 treatment in the second exposure (II, IH, HH, HI). 171 172

Time to first contact 173
In 7 out of 64 cases in the first exposure, and 6 out of 64 cases in the second exposure, contact 174 between an ant from the experimental colony and the presented ant did not occur during the 10 175 minute exposure period. We therefore removed these trials from our analysis and fitted the log-176 transformed time to first contact as the response variable in two LMMs (one analysing data from the 177 first exposure, and one analysing data from the second exposure). Time to first contact was log-178 transformed to meet the assumption of normal distribution of residuals. We included treatment (I or 179 H in the first exposure; II, IH, HH or HI in the second exposure), the caste of the ant that made the 180 first contact (minor or major), and the interaction between these variables as fixed effects. One trial 181 in the first exposure resulted in contact between a major and minor ant and the presented ant at 182 exactly the same time, and for two trials in the second exposure we were unable to observe the 183 moment of first contact accurately. These trials were subsequently removed from their respective 184 analyses. We fitted these models to data on 56 and 58 trials (for the first and second exposure 185 respectively) in 16 colonies. 186

Activity 188
We fitted the proportion of ants from the experimental colony that were active in the first 20 189 seconds after each minute of recording as the response variable in two GLMMs (one analysing data 190 from the first exposure, and one analysing data from the second exposure). We fitted the models 191 using a binomial error structure with an observational level random effect to correct for 192 overdispersion of our response variable (Harrison 2015). We included treatment (I or H in the first 193 exposure; II, IH, HH or HI in the second exposure), phase (before, during or after), and the 194 interaction between treatment and phase as fixed effects. To test whether activity changed at a 195 different rate between different treatments, we also included time (the minute of recording) and the 196 interaction between time and treatment as additional fixed effects. We fitted these models to data 197 on 1920 minutes of video recordings in 64 trials in 16 colonies (for both the first and second 198 exposure). 199

Number of social contacts 201
We fitted the number of social contacts occurring between ants in the experimental colony during 202 each minute of recording as the response variable in two GLMMs (one analysing data from the first 203 exposure, and one analysing data from the second exposure). We fitted the models using a Poisson 204 error structure with an observational level random effect to correct for overdispersion of our 205 response variable (Harrison 2014). We included treatment (I or H in the first exposure; II, IH, HH or 206 HI in the second exposure), phase (before, during or after), time (the minute of recording), and the 207 interaction between treatment and phase, and treatment and time as fixed effects. We fitted these 208 models to data on 1920 minutes of video recordings in 64 trials in 16 colonies (for both the first and 209 second exposure). 210 211

293
Our study shows that workers of the harvester ant M. barbarus respond to the presence of an 294 intruder by increasing their activity patterns overall, and in particular by increasing the rate at which 295 they make contact with other colony members. This is consistent with the hypothesis that individual 296 workers act to increase coordination or cohesiveness among members of their own group when 297 confronted by members of a different group, which may serve as an indicator of invasion or attack 298 by another colony. However, the behavioural response was statistically similar regardless of whether 299 the presented ant was from their own colony or from a different colony, suggesting that when the 300 ants first encountered an unfamiliar individual, they made no obvious distinction between members 301 of their own or other colonies. Nevertheless, exposure to an unfamiliar intruder did have a large 302 influence on the response of ants to a second intruder encountered shortly afterwards. Specifically, 303 an encounter with an intruder from a different colony primed the ants to respond more strongly to a 304 second intruder, particularly when the second intruder was also from a different colony. Thus our 305 results suggest that recent previous experience of potential intergroup conflict increases the within-306 group response to a simulated intrusion, in line with the 'primed response' hypothesis. 307 308 These findings in a eusocial insect offer a complement to research on the behavioural responses of 309 some social vertebrates to simulated intrusions. For example, in cooperative cichlids, N. pulcher, 310 laboratory groups that are exposed to intruders subsequently engage in elevated rates of affiliative 311 behaviour (e.g. soft touching and following; Bruintjes et al. 2015). Similar increases in affiliative 312 behaviour following exposure to experimental intruders have been shown in green woodhoopoes 313 (Radford 2008). In both cases the increase in affiliative behaviour is interpreted as an adaptive 314 response which increases group cohesion, protecting the group from future attacks. These systems differ from our ants in that group members each have the potential to reproduce, either currently or 316 in the future, whereas the ant workers are selected to behave in a way that maximises their indirect 317 component of fitness, realised via the assistance they can provide to the colony production of 318 reproductives. The fact that similar behavioural responses to intruders are seen in such different 319 systems is consistent with the hypothesis that increased contact or affiliation among group members 320 serves to prepare or strengthen the group against future attacks, and is therefore favoured by both There are some limitations to our study which should be considered when interpreting our results. 355 Firstly, we observed fewer social contacts and a lower proportion of active ants in the 'before' phase 356 of the second exposure compared to the 'before' phase of the first exposure. This unexpected 357 difference in behaviour could reflect a lack of acclimatisation to the assay arena (itself an artificial 358 environment) before the first exposure, or fatigue from responding to the previously presented ant 359 before the second exposure. Fatigue in response to stimuli makes direct comparisons between initial 360 and subsequent exposures more difficult to interpret, but does not detract from our observed effect 361 of a primed response to an intruder. In fact, we might have observed an even stronger primed 362 response in the second exposure had ants not been fatigued. Secondly, colony sizes used in our 363 experiments were small. This may have affected colony response behaviour to intruders and could 364 go some way to explaining the lack of observed aggression to presented ants, particularly in light of 365 evidence that colony-level responses to non-nestmates are highly dependent on combined Gordon 2015). Similarly, we cannot rule out that the presentation of a single intruder may not elicit a 368 behavioural response that accurately reflects the response of the colony as a whole (Roulston et al. 369 2003). Finally, we did not conduct behavioural assays blind to treatment. Non-blind studies are 370 exposed to potentially inflated effect sizes (van Wilgenburg and Elgar 2013) and, as such, although 371 our results provide evidence in support of a primed response to intruders, we should exercise some 372 care when determining the certainty of our results.