Short-time development of among-colony behaviour in a high-elevation ant

Standardised assays are often used to characterise aggression in animals. In ants, such assays can be applied at several organisational levels (e.g., colony, population) and at specific times during the season. However, whether the behaviour differs at these levels and changes over a few weeks remains largely unexplored. Here, six colonies from the high-elevation ant Tetramorium alpestre were collected weekly for five weeks from two behaviourally-different populations (aggressive and peaceful in intraspecific encounters). We conducted one-on-one worker encounters at the colony and population levels. When analysing the colony combinations separately, the behaviour was peaceful and remained so within the peaceful population; initial aggression became partially peaceful within the aggressive population; and initial aggression decreased occasionally and increased in one combination but remained constant for most across-population combinations. When analysing all colony com- binations together, within-population behaviour remained similar, but across-population behaviour became peaceful. The observed behavioural differences among organisational levels emphasise the relevance of assessing both levels. Moreover, the effect of decreasing aggression is discernible already over a few weeks. Compression of the vegetation period at high elevations may compress such behavioural changes. Addressing both organisational levels and seasonality is important, particularly in studies of behavioural complexity such as in this ant.


Introduction
In social insects such as ants, standardised assays are often used to characterise specific behaviours such as aggression, boldness, and decision making (Jandt et al., 2014;Wright et al., 2019). Aggression assays are frequently used to explore ant behaviour (Giraud et al., 2002;Steiner et al., 2007;van Wilgenburg et al., 2010;Krapf et al., 2018). Such assays can provide precise information about relationships within and among colonies both within and across populations (Roulston et al., 2003). Additionally, multiple organisational levels can be assessed in social insects using aggression assays (Modlmeier and Foitzik, 2011;Horna-Lowell et al., 2021). The individual worker level can be analysed by investigating aggression among individuals from the same or different colonies (Maák et al., 2019;Maák et al., 2021); the colony level by assaying colonies within the same population (Frizzi et al., 2015;Blight et al., 2016;Krapf et al., 2018;Hakala et al., 2020); and the population level by conducting aggression assays among colonies from different populations (Giraud et al., 2002;van Wilgenburg et al., 2010;Blight et al., 2017). Aggression can also be analysed by combining these (and additional) levels (Garnas et al., 2007;Eyer et al., 2018;Hicks and Marshall, 2018;LeBrun et al., 2019).
Aggressive behaviour observed in such assays depends not only on these organisational levels but, among others, on behavioural variation among and within individuals. Such behavioural variation can be influenced by various genetic and environmental factors (Wright et al., 2019). Within-individual variation, for example, describes that individuals can change their behaviour over time (Laskowski et al., 2022). It can originate from internal stimuli (e.g., brain stimuli; Stamps and Biro, 2016) or different life-history strategies . Also, the environment can promote behavioural variation (environmentally-induced variation) via seasonal differences (Stamps and Biro, 2016).
Interestingly, the influence of the season on behaviour differs severely among ant species. For example, Formica exsecta and Formica polyctena workers are more aggressive in spring than in summer when workers forage for food and re-establish foraging trails (Mabelis, 1978;Katzerke et al., 2006). Similarly, Plagiolepis pygmaea workers are also more aggressive in spring than later in the season, probably due to environmental changes and the presence or absence of food (Thurin and Aron, 2008). In contrast, Linepithema humile workers are more aggressive in summer compared with spring, which may result from spatial changes in colony boundaries (Suarez et al., 2002) or from environmental changes. Environmental changes can, for example, alter the cuticular hydrocarbon bouquet and thereby lead to changes in aggressive behaviour (Sprenger and Menzel, 2020). These examples reveal the seasonal impact on aggressive behaviour. Whether behaviour also changes over a short time such as a few weeks remains largely unexplored. Such short terms are, for example, important for ant species in higher elevations where the season is much shorter compared with low elevations (Grabherr et al., 2010).
In this study, we investigate the behaviour of the high-elevation ant species Tetramorium alpestre Steiner et al. (2010), which displays a behavioural continuum ranging from peace to aggression (Krapf et al., 2018;Krapf et al., 2023). We test whether the behaviour of workers changes over five weeks (mid-July to mid-August) in two behaviourally different populationsone aggressive and one peaceful population when tested among colonies of the same population (Krapf et al., 2019). These five weeks from mid-July to mid-August represent the transition from late spring to early/mid summer in the high-elevation system. An increase in soil and air temperatures represents the main environmental changes in these five weeks (Montagnoli et al., 2014). Each week for five weeks, we collected colony fragments from six colonies, three from each of the two populations, and applied a full design of pairwise one-on-one worker assays within and across colonies and populations. We had no prior expectations of the temporal development of behaviours within and across populations, that is, at the outset, we considered it possible that aggressive behaviour across colonies would decrease, increase, or remain stable over time.

Study system Tetramorium alpestre
All behavioural tests and analyses were conducted using the highelevation ant species T. alpestre, which belongs to the Tetramorium caespitum complex (Schlick-Steiner et al., 2006;Steiner et al., 2010;Wagner et al., 2017). This species is widely spread across the European Alps and can be found in mountainous meadows between 1300 and 2300 metres above sea level (m a.s.l.) (Wagner et al., 2017). Tetramorium alpestre nests can be found under stones and in mosses and tree stumps. It is omnivorous and tends root aphids (P.K., B.C.S.-S., F.M.S., pers. observation). The habitat often comprises shrubs, mats, and stones. The colony size of T. alpestre is estimated to be similar to that of its closely related congener (Seifert, 2017), T. caespitum, which has approximately 15,000-75,000 workers per colony (Seifert, 2018). Interestingly, T. alpestre displays a social polymorphism, that is, colonies can be single-or multiple-queened (Steiner et al., 2003;Krapf et al., 2023), but colonies from both study sites are known to be monogynous (Krapf et al., 2018, Krapf et al., unpublished data). Moreover, T. alpestre also displays a behavioural continuum in one-on-one encounters, that is, workers cover a large behavioural range from peaceful to aggressive behaviour (Krapf et al., 2018;Krapf et al., 2023), and the species is thus suited for studying aggression in across-colony combinations. For example, colonies from one of the two study sites are peaceful when tested within the population, while colonies from the other study site are aggressive (Krapf et al., 2018, Krapf et al., unpublished data).

Worker collection and maintenance
During the summer of 2019, workers were collected from six known colonies in two populations. Three colonies were sampled in Kühtai (Kue1, Kue2, Kue3; Tyrol, Austria), with the expectation of peacefulness within the population, and the other three colonies at Penser Joch (Pen1, Pen2, Pen3; South Tyrol, Italy), with the expectation of aggressiveness within the population (Krapf et al., 2018, Krapf et al., unpublished data). The colonies sampled were at least four years old (P.K., pers. observation). The exact locations of the colonies were determined using a NAVSTAR-GPS system (Garmin eTrex® Legend HCx, Olathe, USA; Tab. A1, Fig. 1). The two populations were approx. 56 km apart from each other. Colonies sampled within the same population were at minimum 10 m apart from each other to ensure that workers from different and not the same colonies were collected. Colony fragments were collected once a week for five weeks, namely in the last two weeks in July and the first three weeks in August 2019. The weeks represent late spring to early/mid summer in higher elevations, which transition is accompanied by an increase in soil and air temperatures in the European Alps (Montagnoli et al., 2014). In both populations, the presence of alate gynes and males was noted as binary code (yes/no) for each colony on each sampling date. From each colony, approximately 280 workers were collected each week after having removed one or more stone(s) covering the nest. The collection was done using a polypropylene aspirator and avoiding skin contact and excluding callow workers. Sampling the 280 workers thus presumably included all stages of age polyethism except callows. Colony sizes of T. alpestre are estimated to be similar to colony sizes of its congener T. caespitum, which has approx. 15,000-75,000 workers per colony (Seifert, 2018). Collecting a total of approx. 1400 workers per colony (approx. 2-9% of the colony's workers) was thus assumed not to drastically decrease a colony's worker abundance. During collection, a few eggs and larvae were also sampled together with workers whenever possible to keep the workers collected busy in caring for these younger stages. Colony fragments were transported to a laboratory at the University of Innsbruck, where they were transferred to polypropylene boxes (12 × 12 cm and 6 cm height) with Fluon-coated walls (GP1, De Monchy International BV, Rotterdam, Netherlands), which prevented workers from escaping. Inside the boxes, crumpled paper was provided as a hiding place. Small holes in the cover of the boxes guaranteed air ventilation. The boxes were stored in a climate chamber (MIR-254, Panasonic, Etten Leur, Netherlands) at constant 18 • C and approximately 60% humidity. The light:dark ratio was set as 0:24 simulating dark surroundings (Parmentier et al., 2016), given the ant's mainly subterraneous lifestyle (Cicconardi et al., 2020). The workers were fed with honey water served in 1.5 ml inverted plastic tubes, and water was available ad libitum in 2.0 ml glass vials; both were plugged with cotton. Workers were able to acclimatize to the boxes for a minimum of 16 h and a maximum of 30 h before being used for aggression assays.

One-on-one worker encounters and analysis
One-on-one worker encounters using two workers were conducted for all possible pairwise colony combinations within and among colonies and populations. A colony combination consisted of three and five oneon-one worker encounters for within-colony and across-colony combinations, respectively (Roulston et al., 2003). For each of the six colonies, within-colony encounters were conducted as control and replicated three times for each colony (using, in total, 6 workers/colony, Fig. 2). Across colonies, encounters were replicated five times for each colony combination. In total, 15 across-colony combinations were conducted: Six across-colony combinations within populations (three within each population) and nine across-colony combinations between the two populations. In the six across-colony combinations within populations, 10 workers/colony were used (Fig. 2). In the nine across-colony combinations across populations, 15 workers/colony were used (Fig. 2). In sum, 93 encounters within and across colonies were conducted each week (6 within-colony combinations * 3 replicates + 15 across-colony combinations * 5 replicates) using, in total, 186 workers per week.
One-on-one worker encounters were performed in either Fluon-or paraffin oil-coated glass vials (inner diameter = 1.4 cm), which prevented the workers from escaping. Partly, paraffin oil was used as it is more environmentally friendly than Fluon but can result in technical difficulties, which we encountered during the assays: When applying too much oil, it can drop into the glass vial, potentially disturbing the workers. Whenever we spotted oil droplets, the assay was stopped and a new assay was started. In some combinations, however, paraffin oil dropped unnoticed into the arena. These droppings were spotted during the video analysis, and these encounters were thus excluded from the analysis reducing the number of encounters analysed (see next paragraph). For each encounter, naïve workers roaming in the nest box were randomly selected and transferred to the glass vial. These workers may be from all age polyethism stages (except callows). Each glass vial was Fig. 2. A) Schematic overview of across-colony encounters conducted within and across both populations. Note that across-population encounters were conducted across all six colonies, in all combinations, although not illustrated in the graph. Solid lines represent within-colony encounters replicated three times acting as a control. Short-dashed lines represent across-colony encounters within populations replicated each five times and long-dashed lines represent across-colony encounters across populations replicated each five times. B) Two Tetramorium alpestre workers: One worker is antennating the gaster of the second worker (© Petra Thurner, reproduced with her permission). Kue (Kühtai, Austria) and Pen (Penser Joch, Italy) represent population names.
used once before being washed to avoid interference from pheromone residues that could change the workers' behaviour.
All encounters were filmed for 180 s with a high-definition camera (Handycam HDR-PJ810E, HDR-CX625, or HDR-XR 155, Sony, Tokyo, Japan). For each encounter, the analysis began when the second worker was introduced to the glass vial. A total of 465 encounters were conducted, namely 375 intercolony and 90 intracolony encounters. In 32 of those encounters, droplets of paraffin oil were present on the bottom of the glass vial. In more detail, 6.88% of all encounters were affected, 5 within colonies, 7 across colonies within population Kue, 8 across colonies within population Pen, and 12 across the two populations (Tab. A2). These 32 encounters were therefore discarded from further analysis yielding a remaining total of 433 encounters (93.12% of all encounters).
To exclude a potential observer bias, the behaviour of each worker was analysed by an observer that did not have any information about the colony origin of the workers. The encounters were observed in slow motion to assess each behaviour for each second, and the behaviour of each worker was scored using a scoring scale ranging from − 4 to 5 (see Table 1 for detailed information on the scoring system). Scores from − 1 to − 4 were considered peaceful, 0 was neutral, and scores from 1 to 5 were considered aggressive (Table 1; Krapf et al., 2019). All behaviours that were observed without a counterpart nearby were scored as "ignoring". For example, if one worker opened its mandibles but was distant from the other worker so that antennation was not possible, this behaviour was scored as "ignoring" rather than as "mandible threatening". Furthermore, if more than one behaviour of the scoring system occurred within the same second, the behaviour with the highest value (i.e., the more aggressive grade) was selected. Transferring the second worker into the glass vial can lead to episodic aggression which ceases within a few seconds. This artefact due to worker handling could unintentionally alter the results. Hence, the first ten seconds of each encounter were regarded as acclimation time and discarded from further analyses (Krapf et al., 2018;Krapf et al., 2019). This means that 170 s were analysed for each worker.

Mean Behaviour Index aggressive (MBI agg )
The results of aggression assays may differ according to the index selected (Roulston et al., 2003;Krapf et al., 2019). Based on a previous study (Krapf et al., 2019), the Mean Behaviour Index aggressive (MBI agg ) was selected for the analyses, which is known to best represent the full behavioural range of T. alpestre. Specifically, MBI agg was selected because it reflects the aggressive behaviour of this species and allows detecting subtle behavioural changes (Krapf et al., 2019). MBI agg depicts a mean of aggressive behaviours and is calculated, additionally to the second-per-second scored behaviour, using a specific time threshold to buffer against aggressive behaviour observed only episodically. In more detail, the maximum behavioural threshold for each worker in each encounter was searched empirically for MBI agg . The threshold (here, 99 s) defines how long aggressive behaviours must last so that an encounter is interpreted as aggressive. Then, the summed durations of all peaceful and aggressive behaviours (with a score less and greater than 0, respectively) were calculated. If the duration of all aggressive behaviours is larger than the threshold, but the duration of all peaceful behaviours is not, the maximum aggression observed is used as the MBI agg value. If both the durations of peaceful and aggressive behaviours are below the threshold, the MBI agg value is zero. If the duration of all peaceful behaviours is larger than the threshold, and the duration of aggressive behaviours is not, the MBI agg value is identical to the minimum observed value. If the durations of both peaceful and aggressive behaviour are larger than the threshold, the MBI agg value is the average of the minimum and maximum values (Krapf et al., 2019). Compared with the other indices, MBI agg prevents that extreme behaviours occurring only episodically influence the results such as using the highest value observed (for details, see Krapf et al., 2019).

Statistical analyses
All statistical analyses were performed in R v. 4.0.2 (R Core Team, 2022) using RStudio v. 1.4.1103. To check for the observer's reproducibility, a Concordance Correlation Coefficient (CCC; Lin, 1989) was calculated. Fifty out of 433 already analysed encounters were randomly selected and rescored (ten encounters per week). Then, the first scoring of the encounters was compared with this second scoring. Importantly, the observer had again no information about the encounters or the origin of the workers during rescoring.
To assess whether the behaviour remained constant or changed over time, Bayesian linear mixed effect models were fitted for the behaviour index. A Bayesian approach was selected because it calculates a posterior distribution for each estimated parameter, which inherently captures the level of uncertainty of the estimates (Raftery, 1995;Hertel et al., 2020). Capturing uncertainty is important to consider when data are variable. Moreover, it allows calculating the repeatability on the posterior distribution, thus also accounting for uncertainty in the repeatability calculation.
Bayesian linear mixed effect models were fitted using the "brm" function (brms package; Bürkner, 2017). For the mixed effect models within populations, "week" was coded as a continuous variable (Andrew and Hatchwell, 2003) ranging from 1 to 5 and set as a fixed factor, while "colony identity" was set as a random factor. Here, time is used as a continuous value and not as a factor because, as time progresses, the season changes. This means that we cannot assume that each week the colonies will behave identically to the week before, which a factor could imply. The presence of alates was not included as a factor as alates were not searched exhaustively given that such an exhaustive search would have caused a major disturbance for the colonies. For the models across populations, "population" was added as an additional random factor. For the calculation in R, several cores were used and set up using the function "parallel". Mixed effect models were set with the following conditions: thinning interval = 50, chains = 4, and adapt_delta = 0.99. Warmup and iterations values were adjusted based on model fit and values were 2500 and 30,000, respectively (Tab. A3). Default (weakly informative) priors were used. The convergence of chains and posterior distribution of effects were checked visually. Moreover, Rhat values were checked to be ≤ 1.01, and the effective sample sizes were checked to be close to the expected sample size. For each random factor ("colony identity" and/or "population"), an intraclass correlation coefficient (ICC) was calculated to assess the repeatability (i.e. reproducibility) of the behaviour of colonies and populations using the function "icc" and "variance_decomposition" (performance package; Lüdecke et al., 2021). An averaged ICC of the posterior distribution was calculated by dividing the variance explained by a random factor with the total phenotypic Note: The scores from − 4 to − 1 were considered as peaceful, while the scores from 1 to 5 were considered as aggressive. 0 was considered as neutral behaviour.
To determine whether the behaviour of all workers together differed among weeks, workers' behaviour values were combined, regardless of the nature of pairings, yielding three data sets: two data sets for populations Kue and Pen separately and one data set for both populations Kue and Pen. Paired Mann-Whitney-U-tests were conducted using each of the three data sets by applying the function "wilcox.test" (R Core Team, 2022). Here, the behavioural values of all combinations of one week were tested against the behavioural values of all combinations of another week (n within weeks = 75). All weeks were tested against each other and corrected for multiple testing using the function "p.adjust" and the "false-discovery rate" in the base stats package (R Core Team, 2022).

Results
Of the 50 videos that had been randomly chosen among all 433 videos and had been scored a second time, the CCC value was 0.86 with a bias correction factor of 0.99, meaning that the observer scored the videos very similarly in both analysis rounds. Alate gynes and males were observed in all colonies but not consistently throughout the observation time (Tab. A1). In some colonies, both were present in all five weeks (Pen1, Kue3), while in others, only in two or three weeks (Pen2, Pen3).
Within and across populations, ICC (repeatability) of the aggressive behaviour for colonies ("colony identity") and populations ("population") ranged between 0.33 and 0.42 and between 0.23 and 0.35, respectively but was not statistically significant ( Table 2). The ICC of the residual variance ranged between 0.32 and 0.67 and was statistically significant (Table 2).
To test whether the workers' behaviour changed among weeks, paired Mann-Whitney-U tests were conducted within each population and across populations, regardless of the combinations. The Mann-Whitney-U tests revealed that the workers' behaviour of the combinations conducted within population Kue and within population Pen did not differ significantly among weeks after correction for multiple testing (Fig. 5, Table 3). Across populations Kue and Pen, the workers' behaviour of all combinations differed significantly among Weeks 1 and 3, Weeks 1 and 4, Weeks 2 and 4, Weeks 3 and 4, and Weeks 4 and 5 after correction for multiple testing (Table 3, Fig. 5). In all these cases, the behaviour became more peaceful over time.  ) and Pen (Penser Joch, Italy). In one combination (Pen1-Pen2), the behaviour became significantly more peaceful over the five weeks (plot with trend line, significance based on Bayesian linear mixed effect models). In the remaining combinations, the behaviour became more aggressive, less aggressive, or remained constant, without statistically significant change in any instance. Overall, the behaviour was almost always peaceful in population Kue and aggressive and peaceful in population Pen. Calendar weeks are displayed on the x-axis, but weeks coded as a continuous variable ranging from 1 to 5 were used in the Bayesian linear mixed effect models (see Materials and methods).

Discussion
In this study, we analysed workers' behaviour of a high-elevation ant species for five weeks in the summer of 2019. Each week for five weeks, we collected colony fragments from three colonies each from two populations (one peaceful and one aggressive population). We calculated a Concordance Correlation Coefficient (CCC) to infer the observer's reliability, which is an integral part of behaviour studies (Guillette et al., 2009;Begley-Miller et al., 2018;Sinkovics et al., 2018;Payne et al., 2020). The CCC result was 0.86, which is considered of high quality according to Werner et al. (2018). We are thus confident that the quality of the analysis at hand is high and that the results are reliable. The behaviour within colonies (control) was always peaceful except for one episodic aggression, which is likely due to a small disturbance rather than significant aggression (Huszar et al., 2014). In across-colony encounters within the presumably peaceful population Kue, workers remained peaceful over five weeks in all three combinations. Within the presumably aggressive population Pen, workers initially behaved aggressively, but workers became more peaceful over time in one out of three combinations. In the other two combinations, workers remained peaceful over time. In the across-population encounters among populations Kue and Pen, workers were aggressive at the start of the study. Over time, aggression significantly increased in one out of nine colony combinations but decreased significantly in three combinations. In the remaining five combinations, the behaviour did not change and remained aggressive over time. The repeatabilities (ICC) of the aggressive behaviour for colonies and populations were not significant. The residual intra-individual variance was significant indicating that other parameters such as environmentally-induced intra-individual and/or the residual intra-individual variation of workers influenced the behaviour. Paired Mann-Whitney-U tests revealed that the workers' behaviour did not differ among the weeks within populations (workers' behaviour values of all combinations were combined). However, workers' behaviour across populations Kue and Pen differed among five weeks and became more peaceful.
The behaviour assessed at the two observed organisational levels, colony and population, differed partially. Single colony pairs assessed at the colony level within the same population revealed only small changes in the behaviour over weeks, while changes became more apparent at the colony level across populations. Interestingly, combinations assessed at the population level did not reveal any trend of change within the populations but a decrease in aggressive behaviour over time across the populations. These findings illustrate that the results can differ depending on the organisational level analysed. Importantly, colonies and populations can display a behavioural consistency, that is, some colonies and populations can be peaceful and remain so, while others can be aggressive and remain so (Krapf et al., 2018;Krapf et al., 2023). Such a trend was also observed here. It is thus vital to choose an appropriate organisational level of interest, particularly in studies of behavioural complexity such as in this high-elevation ant.
The behavioural analyses over five weeks revealed increasing and decreasing aggression as well as steady behaviour. The results suggest that, unless the aim is to explore variation over time, the sampling and testing of colonies should be conducted within a short time (e.g., one to two weeks) because behaviour can change quickly over time, especially across populations and at the beginning of the season when food is scarcely available (see also the discussion below). This is particularly true if the study focuses on the absolute results at the colony and population levels. Collecting colonies in a short time can lead to logistic problems or delays in conducting assays when sampling occurs in distant or remote locations (cf. van Wilgenburg et al., 2010).
The observation of a changing aggression level over a few weeks at the colony and, predominantly, the population level raises an important question about the mechanism(s) causing such change. Workers' aggressive behaviour can be influenced by various genetic and environmental factors such as genetic relatedness (Fournier et al., 2016), geographic distance (Keresztes et al., 2020), colony density and size (Maák et al., 2021), presence of alates (Staab and Kleineidam, 2014), and the length of the vegetation period and time of the season (Mabelis, 1978;Katzerke et al., 2006;Thurin and Aron, 2008). Whereas we do not have such data at hand for the colonies analysed here, we briefly discuss these mechanistic factors potentially influencing aggression in light of the published literature.
Genetic relatedness within and across colonies can vary seasonally due to differences in the number of queens (Segundo et al., 2012), and certain relatedness patterns can promote aggression (Fournier et al., 2016). This, however, seems not to be the case in T. alpestre. It was recently shown that workers from eleven colonies in population Kue behaved peacefully in across-colony encounters regardless of withinand across-colony relatedness (Krapf et al., 2018). An influence of genetic relatedness or geographic distance on the behaviour thus seems improbable.
Aggression can also be influenced by locally high colony density and thus by increased intraspecific competition (Hölldobler and Wilson, 1990;Thomas et al., 2007;Modlmeier and Foitzik, 2011). We did not record colony density but based on previous sampling occasions, we estimated colony density to be below 0.01 nests/m 2 in both populations (data not shown), which is one percent of the maximum colony density observed for this species (Seifert, 2018). It thus seems unlikely that This table denotes the results from the Bayesian linear models. Note: Kue = Kühtai, a population collected in Austria; Pen = Penser Joch, a population collected in Italy, 95% CI = lower and upper confidence intervals. Bold values represent statistically significant values based on positive or negative confidence intervals that do not include 0. NA = not available; refers to within-population tests, for which no calculation was possible. Negative values in the "Mean week effect" indicate that behaviour values decrease (become more peaceful) and positive values that behaviour values increase (become more aggressive).
colony density influences aggression in T. alpestre in these two populations, especially since behaviour was largely peaceful within populations. Aggression may also be influenced by the colony size of the focal colony (local worker density; Suarez et al., 2002;Haatanen and Sorvari, 2013;Maák et al., 2021) and age polyethism (Larsen et al., 2016). The colony size of T. alpestre was not assessed in this study but is assumed to be similar to the colony size of a close congener (T. caespitum), which has approx. 15,000-75,000 workers (Seifert, 2017;Seifert, 2018). As the tested colonies were at least four years old (P.K., pers. observation), we speculate that mature colonies were collected, also because they already produced reproductives (Tab. A1). Although collecting ca. 280 workers for five weeks reduced the number of workers, it seems unlikely to have drastically reduced the colony size of mature colonies. Nevertheless, initial colony size (i.e. size before collection) may have differed among colonies and could have led to behavioural differences. Moreover, the number of older workers might have been lowered with sampling events leaving the colony with fewer and younger workers. The potential difference in colony size at the start of the experiment combined with the sampling of workers over the five weeks and thus smaller colonies with younger, less-experienced workers might explain the decrease in aggression observed in several combinations. However, a decrease was not observed in all combinations, and aggression even increased once. Additionally, age polyethism can influence the behaviour as workers switch tasks and change their behaviour over time (Larsen et al., 2016). We speculate that workers that were produced during the observation or were already present in the colony at the time of collection likely switched stages of age polyethism and thus potentially changed their behaviour over time. These findings may indicate that collecting workers for five weeks did not drastically decrease colony sizes or change the age structure of workers and thus influence aggression, at least in these two populations. Nevertheless, assessing colony sizethough a difficult taskwill be very interesting in future research to address its influence on behaviour.
Another factor potentially influencing the results of behaviour analyses is the sample size. In this study, workers of six colonies (three from each population) were tested in across-colony combinations, and each combination was replicated five times. Although the number of replicates used was similar to that in many other studies (e.g., three to five times; Giraud et al., 2002;Suarez et al., 2002;Cremer et al., 2006;Cordonnier et al., 2022), it may have been too low and may have led to uncertainties in the analysis. Additionally, workers of potentially all polyethism stages (except callow workers) were randomly selected and had the same chance of being used in the aggression assays. The random worker selection may partially be responsible for the observed behavioural variation, specifically the non-significant repeatability of behaviour at the colony level and the significant intra-individual behavioural variation. Moreover, since each colony represents an individual entity with a different life history, the tested colonies and workers may have reacted differently to (seasonal) changes. Future studies should validate the robustness of these findings by replicating this study to detect if similar patterns emerge or, if feasible, by collecting more colonies from the same populations and conducting a similar study with an increased number of colonies. Nevertheless, the study provides important findings into behavioural differences and trends among colonies and populations in this high-elevation ant.
One theory for the aggressive behaviour of T. alpestre observed at the beginning of the analyses could be the presence of alate gynes and Pen (Penser Joch, Italy). In three combinations (Kue1-Pen1, Kue1-Pen3, and Kue3-Pen3), aggressive behaviour significantly decreased over the five weeks, while it significantly increased in one combination (Kue2-Pen1; plot with trend lines, significance based on Bayesian linear mixed effect models calculated using brms). In the remaining combinations, the behaviour became more aggressive, less aggressive, or remained constant but without statistically significant change in any instance. Calendar weeks are displayed on the x-axis, but weeks coded as a continuous variable ranging from 1 to 5 were used in the Bayesian linear mixed effect models (see Materials and methods).
males. Extreme worker aggression during the presence of alates was, for example, observed in Atta vollenweideri (Staab and Kleineidam, 2014). This may thus also apply to T. alpestre: workers may protect alates, which represent valuable reproductive output of a colony, and possibly reduce aggression after the alates have swarmed. The study covered the estimated swarming period of T. alpestre (which is around August 1 st ; Seifert, 2018), but the number of alates was not integrated into the statistical analysis because alates were not searched exhaustively to minimise disturbances of colonies. In theory, if the presence of alates positively influences aggression in T. alpestre, an increase or at least steady aggressive behaviour over time should be observed. An increase was detected in one combination (Kue2-Pen1) and is thus in line with this theory. However, peaceful behaviour became more prominent over time even though alates were still encountered in a few colonies after the estimated swarming date, which contradicts this theory. Yet again, the decrease of alates observed in the colonies coincides with the increase in peaceful behaviour, which again corroborates this theory. An influence of alate presence is thus not directly apparent, but it cannot be completely ruled out either. We thus speculate that the presence of alates may only partially influence the aggression level in this ant, but this interesting theory should be addressed in a future study.
A shorter season in high elevations (Grabherr et al., 2010) could also explain the decrease in aggression over five weeks observed in six out of 15 across-colony combinations. At 2000 m a.s.l. in Tyrol (Austria), five weeks represent a fourth to a third of the entire vegetation period (modified from Winkler and Moser, 1967) compared with lower elevations at 500 m a.s.l. At such high elevations (2000 m a.s.l.), colonies have approx. 120-140 days to propagate. In contrast, the vegetation period at 500 m a.s.l. in Tyrol is approx. 170-190 days long. The vegetation period at lower elevations is thus much longer compared with the season at higher elevations. It is thus likely that low-elevation colonies are already in mid summer in July and August and have (re) claimed territories, established foraging trails, and become more week within and across populations Kue (Kühtai, Austria) and Pen (Penser Joch, Italy). Across-colony behaviour values of all workers are combined for each week to assess whether workers' behaviour changed over weeks. Behaviour values significantly differed between Weeks 1 and 4, 1 and 4, 2 and 4, 3 and 4, and 4 and 5 (calculated using Paired Mann-Whitney-U tests). Asterisks represent significant differences among weeks (* < 0.05; *** < 0.001). peaceful over time similar to other species (Mabelis, 1978;Katzerke et al., 2006;Thurin and Aron, 2008). In contrast, colonies at higher elevations might just undergo this change from spring to summer in July and August and do so in a much shorter time (e.g., a few weeks). Moreover, reduced food availability (e.g., prey species) in early spring could lead to fights as observed among wood ant colonies (Driessen et al., 1984;Mabelis, 1984). An increase in food availability during spring could thus lead to a decrease in aggression over time. A compressed season length in higher elevation coupled with reduced food availability at the beginning of the season might thus influence behaviour and explain the change in aggression observed. However, this theory needs to be further tested.
In conclusion, we revealed that the behaviour of the high-elevation ant T. alpestre varied within and across colonies and populations over five weeks during the summer of 2019. Depending on the colony combination, the initial behaviour was peaceful or aggressive, and aggressive behaviour increased occasionally, decreased once, and remained stable in the other instances. When analysing all colony combinations together, the behaviour became peaceful in across-colony combinations among both populations. The presence of alates may partially explain the behaviour observed. Moreover, also seasonal effects due to a shorter vegetation period at higher elevations and different food availabilities may influence workers' behaviour both within and across populations and may lead to a decrease in aggression over time. This means that a compressed season at higher elevations may compress such behavioural changes. The results further suggest that the time needed to sample colony fragments should be kept to a minimum to avoid such a seasonal influence. When designing such behavioural studies, seasonality and food availability should be addressed, especially at the beginning of the season, when food shortage may affect behaviour. Further studies focussing on behavioural changes due to seasonal changes are needed to analyse additional mechanistic factors, which might promote aggressive and peaceful behaviour and which were not analysed in this study. Such factors are, among others, colony size and abundance (Maák et al., 2021), cuticular hydrocarbons (Sprenger and Menzel, 2020), food availability (Driessen et al., 1984), experienced temperature (Barbieri et al., 2015), and moisture and diet (Suarez et al., 2002)). It will thus be vital to collect colonies and to analyse workers' behaviour at the beginning, middle, and end of the season to unravel seasonal influences on this and other ant species.

Ethical note
No licenses or permits were required for this research. Ant colony fragments were collected in the field and maintained in the laboratory with care. Ants were provided with suitable nesting possibilities, food, and water ad libitum. After the experiments, colony fragments were kept in the laboratory and reared until their natural death. All applicable international, national, and institutional guidelines for the care and use of animals were followed.

Funding
Klaus Sedfaoui was funded by an Erasmus+ internship scholarship. This study was financially supported by the Austrian Science Fund, P30861.

Conflict of interest
The authors declare that they have no conflict of interest. Note: m a.s.l. = metres above sea level. Kue = Kühtai, a population collected in Austria; Pen = Penser Joch, a population collected in Italy. Weeks with alate gynes or males present = In the five weeks that workers were collected, the presence of alates (irrespective of the sex) was noted and is represented as the weeks, in which alates were observed.

Table A2
Number of excluded comparisons (n = 32) for each combination and week for which exclusion was necessary due to paraffin oil dropping into the arena.

Data Availability
The dataset generated and used in the study and the R-code applied can be found in the supplementary material.