Introduction 

Insects contribute to a broad range of ecosystem services, including pollination, which is crucial for crop productivity and, more generally, for global biodiversity (Noriega et al. 2018; Porto et al. 2020). However, during the Anthropocene, a massive decline in terrestrial insect populations has been observed, mostly because of anthropogenic drivers (van Klink et al. 2020; Wagner 2020). Differences in life-history traits, ecological requirements and tolerance mean that not all insect species suffer equally from these threats (van Klink et al. 2020). In particular, social insects, whose colonies can be considered as a superorganism, are generally more resilient to stressors than solitary species (Straub et al. 2015; Fisher et al. 2019).

Among social insects, eusociality is defined as a social group of related individuals, with a division of labour into a reproductive and a non-reproductive caste. This division leads to colonies that comprise several generations of individuals with behavioural specializations that cooperate to take care of the brood (Wilson 1971). The number of individuals within eusocial insect colonies (i.e. their colony size) is important for fulfilling many functions, including the maintenance of homeostasis, the defence of the colony and thermoregulation efficiency (Wilson 1971; Southwick 1985; Ulrich et al. 2018), but also for buffering the effect of stressors (Straub et al. 2015). For example, in the bumblebee Bombus impatiens, larger colonies are less sensitive to the negative effects of neonicotinoids than smaller ones (Crall et al. 2019). However, stressors such as exposure to high developmental temperature (Gérard et al. 2022a) or pesticides (Whitehorn et al. 2012) can lead to a reduction in colony size in bumblebees. While several studies have shown that this reduction can impact some life history traits like brood maintenance, resource collection or thermal sensitivity (Harbo 1986; Weidenmüller et al. 2002; Khoury et al. 2013), its effects on individual behaviour and cognition, in particular, remain unclear. Indeed, interactions between conspecifics are central to the functioning of eusocial colonies and are crucial for the development of individual colony members’ behavioural repertoires (Riveros and Gronenberg 2010; Gowda and Gronenberg 2019; Wang et al. 2021). A decrease in interactions with other individuals in the colony could affect individual behaviour. For example, in red wood ants Formica rufa, Kleineidam et al. (2017) demonstrated that isolated workers displayed decreased aggressive behaviour toward non-conspecifics proportional to their time spent in isolation. Szczuka and Godziñska (2004) had previously shown a similar subdued behavioural response, wherein groups of F. polyctena with fewer than 30 individuals exhibited rudimentary predatory behaviours relative to their counterparts from larger groups.

The experimental approach for investigating the effects of social interactions on behavioural development has predominantly relied on isolating individuals from their colony. The isolation of a single individual is an effective experimental method for distinguishing the thresholds for which sociality shapes behavioural development. However, these experimental methodologies largely lack ecological pertinence because individuals are removed from their colony. In contrast, reducing the number of individuals within a colony would provide a more ecologically relevant environment where other functional elements of the hive are maintained, including the presence of a queen, which has been shown to affect multiple facets of worker behaviour and physiology in eusocial insects (Lopez-Vaamonde et al. 2007; Santos et al. 2022). Moreover, the cognitive abilities of individuals that are isolated from the group once being adult could differ from the individuals that fully developed in a colony that has just experienced a drastic decrease of the number of workers. Indeed, when the worker force decreases, bumblebee larvae can be fed less, leading to a lower macronutrient quantity/quality and an impairment of their cognitive abilities (Bouchebti et al. 2022). This impairment could occur through, for example, a lack of some specific amino acids and proteins that have a major role in learning and memory retention, although the effect of malnutrition on cognitive capacities in insects has not been fully investigated (Wyse and Netto 2011; Gage et al. 2020).

To better understand whether a reduction in colony size affects individual cognition in eusocial insects, we investigated how experimentally reducing the number of individuals within colonies influences the learning abilities of the buff-tailed bumblebee Bombus terrestris. This species is an ideal candidate for such investigations because it typically lives in colonies with over 100 individuals, and previous work has indicated that stressors could impact bumblebee colony size (e.g. Whitehorn et al. 2012; Gérard et al. 2022a). Moreover, numerous assays have shown their capacity to perform associative learning tasks (e.g. Muth et al. 2017; Wang et al. 2021; Gérard et al. 2022b). Here, we tested the hypothesis that changes in colony size affect bumblebee associative learning. Assessing how individual behaviour can vary is of primary importance, as central place foragers need to display efficient cognitive abilities to acquire and memorise the quality and location of resources that they bring back to the colony. We predict that individuals from larger colonies will perform better at learning tasks than individuals from smaller ones, potentially due to a reduction in social interactions within the colony or undernourishment. Moreover, caste polymorphism is a typical trait of eusocial insects, wherein a morphological trait—most often body size—denotes role differences within individuals of the same caste in a colony (Gadagkar 1997; Robinson and Jandt 2021). Given that changes in colony size may also modify the environment in which the larvae grow and may even affect individual body size—a lack of workers might result in larvae being fed less food and having a smaller body size as adults (Shpigler et al. 2013)—we also investigated the effect of body size on cognition. We hypothesized that larger workers perform better at associative learning tasks because brain size varies with body size in B. terrestris (Smith et al. 2016), which can vary up to tenfold in body mass (Couvillon et al. 2010), and brain size has been related to learning abilities in bees (Collado et al. 2021).

Materials and methods

Model species

Bombus terrestris colonies were purchased from Koppert (Berkel en Rodenrijs, The Netherlands). The colonies were maintained at 26 °C and 50% humidity. All colonies had access to the sucrose solution provided by the commercial breeder. In addition, colonies were provided with an ad libitum quantity of “pollen candy”—a blend of finely ground, commercially-sourced, pollen granules and 50% sucrose solution. All individuals were kept in closed colonies (i.e. no access to the outside), except when an individual was performing the learning test, so they did not have any experience of associating colour with a rewarding cue prior to the experiments.

Colony reduction

A total of 20 hives were used in the experiments, which were conducted between October 2021 and March 2022. As it was not possible to test individuals from 20 hives simultaneously, we performed four identical sessions using five colonies per session.

To test the effects of manipulating colony size on associative learning, two treatment conditions were established: small colonies, which involved decreasing colony population sizes to extremely low numbers (but still functional, as the colonies were continuing to produce workers), and normal colonies, which also involved decreasing colony size to a standardized but realistic size of a typical B. terrestris colony (Buttermore 1997). The colony size modification for the normal size treatment ensured that all normal-size colonies had a consistent population size, as colonies naturally vary in their number of workers. Modifications were performed on all colonies to control for the effect of removing, manipulating and marking individuals. The normal size treatment involved reducing the colony to 100 workers, while retaining the queen and all exiting cells and brood. The parameters for the small-size treatment involved removing up to 95% of the colony’s population, retaining just 20 workers, the queen and all existing cells and brood. The mean population size per colony for B. terrestris has been estimated as 150 workers (Buttermore 1997). Consequently, a reduction to 20 workers would represent a population size that deviates significantly from that of a normal colony while still being of an adequate size to maintain the production of workers during the trials (as tested individuals were not replaced back into the colonies). Individuals from the different colonies were tested at one of three temporal stages (i.e. each of the individuals was tested at only one of the three stages): stage I) prior to the treatment (when all colonies were unmodified), stage II) after treatment (occurring 2–25 days after group size manipulation) and stage III) full development under the treatment condition (individuals that emerged 25 days after treatment, corresponding to the total development time of bumblebees). Individual workers were marked with coloured paint on their thorax between these time points to identify the colony stage they belonged to. During each experimental session, three colonies were subjected to the small-size treatment and two colonies to the normal-size treatment. We had an additional colony for the small-size treatment because the effects of abruptly reducing population size on colony survival were initially unknown and could have potentially had an effect on colony survivorship.

Associative learning experiment

The methodology for associative learning was adapted from Muth et al. (2017). In differential conditioning, two distinct and initially neutral stimuli are associated with either an unconditioned stimulus (US) or a neutral stimulus (NS), to respectively become a positive conditioned stimulus (CS +) and a neutral conditioned stimulus (CS). Yellow and blue coloured cards (30 mm × 5 mm, L × W) were used as the initial NS and were selected due to their distinct perceived chromatic contrast for bumblebees (Gumbert 2000; Skorupski et al. 2007). For the CS + , the card was dipped into a sucrose solution (US; 50% water/50% sugar w/w) to provide a reward. For CS, the card was dipped into water (NS) to provide a neutral conditioned stimulus. Both yellow and blue cards were alternated as the CS + and CS for colonies 1–10. As the colour of the CS + did not affect learning in these early trials (generalized linear model with family parameter set as binomial; p = 0.14), we subsequently used the blue card as CS + and yellow as CS for colonies 11–20. The containers used to store individuals during the starvation and learning trials were modified T-75 standard cell culture flasks (150 mm × 80 mm × 35 mm, L × W × H) that had five pairs of holes were drilled into their base, with 2 cm between each pair and 3 cm between each row (Fig. 1).

Fig. 1
figure 1

A The experimental set-up, including the trial platform, a camera recording the experiment and the container. B Close-up image of the container, with a bumblebee inside, and the coloured strips inserted into holes

Full size image

Prior to commencing each experimental trial, 10 individuals (two from each colony) were removed from their colony using forceps and placed directly into the containers, after which they experienced a 2 h starvation period to motivate them for the food reward and to allow them to acclimate to the container. The duration of this period was determined (both from previous studies (Gérard et al. 2022b) and pilot experiments) to be the minimum amount of time required to induce consistent levels of motivation across all subjects during all six trials. During these trials, to minimise observer bias, a blinded method was used, i.e. the observer was not aware of which treatment the tested individual came from. When starting the trials, the strips of the coloured card were dipped into one of two vials containing either a 50% sucrose solution or water, depending on which colour was to be CS + and CS. These were then inserted into adjacent holes of the container (Fig. 1B). The placement of the strips largely depended on the activity and location of the bumblebee in the container, with the strips being typically inserted into holes that were approximately 6 cm from the bee. Once the strips entered the container, the bees were given 3 min to interact with them. A bumblebee that interacted with a strip via contact with its antennae, proboscis or when grappling with its legs was given 5 s before that strip was removed. The remaining strip was left for an additional 5 s so that the bee could investigate or interact with it. After this time (or after 3 min, if there was no interaction), the container was lifted off the platform. If the bee first interacted with the CS + , this was scored as 1 (Supplementary video S1; Table S4), while an initial interaction with the CS was scored as 0 (Supplementary video S2; Table S4). A failure to interact with either of the strips within 3 min was scored as X (Supplementary video S3; Table S5). Each of the 10 individuals would experience the previous steps (defined here as a trial) before the next trial commenced. Between trials, the position of the yellow and blue strips was changed to minimise side bias, and the platform was wiped down using water and the holes scrubbed with a fine brush to remove any excess sucrose solution before each trial started. On each experimental day, six associative learning trials were performed per bumblebee, after which they were removed from their containers, transferred into test tubes and placed into a freezer at − 14 °C in preparation for measuring body size.

The intertegular distance (ITD, a proxy for body size in bumblebees (Cane 1987)) of each individual that participated in the learning trials was measured to assess if there was an association between body size and learning performance (Table S5). The ITD is the distance between the two wing joints (tegulae) which are located on the thorax of a bumblebee. This was measured using digital callipers (Goobay, Wentronic, Braunschweig, Germany) with the aid of a microscope.

Statistical analyses

General linear mixed-effects models (GLMMs) using the glmer function from the lme4 package in RStatistics (R Core Team 2020), with family parameters set to binomial, were used to analyse the effect of the treatment on associative learning, the likelihood of interaction with the stimuli and body size. First, binary data was collected during each associative learning trial (i.e. 1 denoting a successful initial interaction and 0 a failed initial interaction). All bumblebees that failed to interact with either of the stimuli during one or more trials were excluded from the first main analysis. The full model to assess the impact of colony size on associative learning included the performance (i.e. binary data, success if interacting with CS + , failure if interacting with CS) as the response variable, and treatment, colony stage, their interaction, and trial as explanatory variables. We also included the colony ID nested in the session, as well as individual ID as random variables. Then, to perform an analysis including non-interactions during a trial, data was reformatted into a binary format. All non-interactions, previously marked as “X”, were converted to “0”, and interactions, previously marked as “1” and “0” (for interactions with the CS + or CS, respectively), were converted to “1”. We used the same model to perform this analysis of the non-interactions than the model of the previous analysis, but switched the response variable to the non-interaction response variable. Testing if there were any differences of interactivity depending on the colony size and the colony stage was used to assess if workers were equally motivated to take part in the experiment, whatever the treatment.

In addition, a GLMM with the same random factors as in the previous models was used to examine the relationship between performance (i.e. binary data, success if interacting with CS + , failure if interacting with CS) and body size, with combined factors of trial and an interaction between treatment and colony stage. Finally, we used a LMM to examine if the treatment and the colony stage impacted the body size of workers. In each statistical model, two R-functions were used as posthoc tests for the GLMMs: emmeans, from the emmeans package to execute a pairwise comparison that involved an interaction (treatment and colony stage), and glht, from the multcomp package to examine differences in the group means of trial. All final GLMMs were selected based on the AIC procedure.

Results

We first tested the impact of the colony size treatment and the colony stage on associative learning. The model that best explained learning performance included treatment, colony stage, trial as fixed factors and individual ID and session as random variables (next best model ΔAIC 1.29, Supplementary material, Table S1–2). There was no significant relationship between colony size and learning performance (p-value = 0.581; Fig. 2). In addition, there was no significant relationship between the performance and the colony stage (p-values: stage I versus stage II: p = 0.36; stage II versus stage III: p = 0.6; Fig. 2). Trial number did affect the performance, with the proportion of initial interactions with the CS + being significantly higher in trials 4, 5 and 6 than in trial 1 (comparison with trial 1; p-values: trial 4: p = 0.02; trial 5: p = 0.03; trial 6: p = 0.03; Fig. 2), suggesting that they were indeed learning. Interestingly, the number of bees making an initial interaction with the CS + first peaked at trial 4 (comparison with trial 4; p-values: trial 1: p = 0.02; trial 2: p = 0.03; trial 3: p = 0.02), after which it decreased slightly throughout successive trials (trial 4 = 58.2%; trial 5 = 57.2%; trial 6 = 57.6%).

Fig. 2
figure 2

The proportion of bumblebees interacting with the positive conditioned stimulus (CS +) across trials in a differential conditioning task. Blue dots represent small size colonies, and red dots represent normal-size colonies. The red line describes a 50% interaction with the CS + . A During stage I (before colony reduction), N = 155 for small colonies and N = 92 for normal colonies. B During stage II (from day 1 to day 25), N = 245 for small colonies and N = 188 for normal colonies. C During stage III (workers that experienced the treatment during their entire development), N = 92 for small colonies and N = 49 for normal colonies

Full size image

We then assessed the impact of the colony size treatment and the colony stage on the likelihood of an individual interacting with either the CS + or the CS. The model that best explained whether individuals were motivated to interact with the stimuli included treatment, colony stage, their interaction and trial as fixed factors, as well as individual ID as a random factor (next best model ΔAIC 4.56, Supplementary material, Table S3–4). There was no significant relationship between colony size and interactivity (p-value = 0.654; Fig. S1), suggesting that the colony size did not affect whether individuals were more or less likely to interact with the stimuli. In addition, we did not observe any significant relationship between the likelihood of interacting with a stimulus and the colony stage (p-values: stage I versus stage II: p = 0.65; stage II versus stage III: p = 0.84). With respect to learning performance, bees were most likely to interact with the stimuli during trials 5 and 6, with the greatest differences of interactivity occurring between trial 6 and trials 1 and 2 (comparison with trial 6; p-values: trial 1: p < 0.001; trial 2: p < 0.00). This pattern of increased interactivity during trials 5 and 6 reflects a general trend of decreasing non-interactions as trials progressed.

Finally, we tested the impact of worker body size on associative learning. The model that best explained the variation in performance depending on body size included body size, treatment, colony stage, trial as fixed factors and session as a random factor (next best model ΔAIC 2.78, Supplementary material, Table S5–6). There was no indication that the size of an individual had any effect on its performance (p = 0.424). Individuals of all sizes were consistent in their learning abilities, exhibiting over 50% performance from trial 4 and reaching more than 60% by trial 6, with the exception of the smallest bees (< 3.5 mm ITD) at trial 6 (Fig. S2). Workers from stage III in the small-size colonies (mean ITD = 3.905 mm, σ2 = 0.625) were significantly smaller than the workers from stage I (mean ITD = 4.701 and 4.822 mm, σ2 = 0.363 and 0.321 for small and normal colonies respectively; both p-values < 0.001) and II (mean ITD = 4.356 and 4.595 mm, σ2 = 0.612 and 0.431 for small and normal colonies respectively, both p-values < 0.001) in both treatments, but not significantly different from workers from stage III in the normal-size colonies (mean ITD = 3.969 mm, σ2 = 0.662; p-value = 0.986).

Discussion

In this study, we assessed whether experimentally manipulating colony size affected the associative learning abilities of Bombus terrestris workers. Workers were significantly more likely to first interact with the CS + at trial 4 than they were in trials 1 to 3 and had a response rate of over 50% (indicating a non-random interaction with the stimuli) during trials 5 and 6, suggesting that they were indeed learning in this experimental paradigm. The pattern of responses we observed resembles that observed by Muth et al. (2017) using a similar paradigm but with a negatively reinforced stimulus (instead of the neutral stimulus used here), although the proportion of bees that responded in our study was lower. Thus, our results show that bumblebees presented with a choice between stimuli that provide either a positive or neutral stimulus will preferentially choose the positive stimulus after 3 interactions with both. This is important as the use of a neutral stimulus more closely reflects the learning situation that bees will face under natural conditions—having to make associations with flowers that provide floral rewards with differing nutritional values.

We found that learning performance was consistent across treatments and colony stages (pre-treatment, directly post-treatment or developing under the treatment condition), suggesting that the cognitive capabilities of individual bumblebee workers are robust to both minor and major reductions in colony size, and the associated decrease of the number of interactions among workers or the potential reduction of nutritional intake. In addition, this does not depend on the amount of time they have been exposed to these disruptions. While stressful conditions like pesticides or high ambient temperature can affect cognitive abilities in bees (Giurfa 2013; Klein et al. 2017; Gérard et al. 2022b), a drastic reduction of colony size alone was not sufficient to lead to any substantial cognitive impairment. Our results are also consistent with those of Wang et al. (2021) who showed that other non-cognitive behaviours are robust to group size. Given that the workers that fully developed under the “small colony size” condition (i.e. stage III) did not show any cognitive impairment, we also do not find that a potential lack of nourishment or important macronutrients that might occur in colonies with fewer workers affects individual learning performance. The robustness of behaviour to colony size would be essential for ensuring the proper functioning of the large variation in worker number that occurs over a colony’s lifetime—from its initiation to its peak after a few weeks (Alford 1975). As learning abilities appear to be robust to changes in group size, our results suggest that natural changes in the colony size throughout the season do not affect cognitive abilities among workers, at least during the colony stages where new workers are produced. This would be particularly crucial during the first few weeks of colony development, when the number of workers is low but the requirements for foraging are high. In addition, it would be interesting to consider how the learning performance of other castes, like queens, varies throughout the life cycle and depends on the role within the colony. Indeed, queens are an interesting case where an individual faces different challenges: they first have to forage during the first weeks of the colony, but then focus only on larvae feeding and egg laying. We could expect a decrease of learning performance when the queen switches from foraging to focusing on tasks within the colony, but this remains to be tested.

Despite a three-fold difference in body size (i.e. which can correspond to a tenfold difference in dry body weight (Cane 1987)), our results also show that worker body size had no effect on associative learning ability, which is consistent with previous findings in B. terrestris (Raine and Chittka 2008). We hypothesised that, after a drastic reduction of the colony size, larvae may suffer from undernourishment, which would develop in smaller adults. Among bees, brain size is positively correlated with body size, and the brain size of bees has been proposed to predict learning abilities (Collado et al. 2021). However, there is limited evidence in other insects, particularly at the intraspecific level, that brain size has any relationship with behavioural repertoire and cognitive capacity (Chittka and Niven 2009). Indeed, attempts to correlate brain size and cognitive performance with sociality among insects are ambiguous (Riveros et al. 2012; Gordon et al. 2019; Kamhi et al. 2019). The drivers of brain size in social vertebrates, which can lead to a positive relationship between brain size and cognitive performance (e.g. competition and mate selection), do not occur in social insect workers (Lindenfors 2005; Dunbar and Shultz 2007). Our results suggest that, despite size being the predominant polyethistic factor in B. terrestris—with larger workers being mostly foragers and smaller workers mostly feeding and incubating brood (Jandt and Dornhaus 2009; Holland et al. 2021)—their capacity for associative learning is not limited by size, and therefore role, within a colony.

Overall, the results of this study indicate that the associative learning capabilities of individual bumblebees are robust to reductions in worker number and that this does not depend on the length of exposure to these disruptions or body size. This robustness would be important for colony productivity as workers must forage efficiently at all colony life stages, from the small population of small workers that are present at its beginning to the large number of size-polymorphic individuals that are present at its peak. Whether variation in social group size affects other aspects of bumblebee behaviour, such as foraging plasticity or decision making, remains to be explored in future work. Finally, this study provides further validation for the effectiveness of using a free-moving experimental method in testing bumblebee behaviour (Muth et al. 2017; Gérard et al. 2022b) even when a neutral rather than aversive stimulus is used.