The disturbance leg-lift response (DLR): an undescribed behavior in bumble bees

Methods

Procedure

Subjects were placed in individual apparatuses after being collected. Each apparatus consisted of a clear plastic cube (2.6 × 3 × 2 cm), made from a microscope cover slip container, with two holes (2.55 mm diameter) drilled on opposite sides. The size of the apparatus allowed the bees to freely move but did not permit flight or substantial relocation inside the apparatus. After being placed in the apparatus, subjects were transferred to an experimental room, placed approximately 1.2 m apart, and allowed to acclimate for three hours.

Each bee experienced five trials with a 15-minute intertrial interval (ITI) after the acclimation period. Bees were randomly assigned to either an experimental or control group. During trials for the experimental group, a researcher startled the bee by inserting a toothpick approximately halfway (1.3 cm) into the apparatus for 10 s through the hole closest to the bee. During this time, it was possible for the bee to physically contact the toothpick. An identical procedure was used for the control bees, except that the toothpick was held outside of the apparatus adjacent to the hole. This group controlled for the general approach of the investigators, as well as the presence of a close object that could not be contacted by the bees.

During each trial, several behaviors were scored from video recordings. We recorded both the occurrence of DLR and the number of legs lifted during each DLR. We specifically defined DLR as when one or more legs were lifted above the bee, relative to the bee’s position. Legs that were lifted prior to the trial were not considered a DLR; observing the movement during the trial was required to record a DLR. Biting was recorded as any time a bee visibly opened and closed its mandibles during a trial. Often bees made mandible contact with the stimulus, but this was not required. Finally, we recorded stinging any time the bee contacted the stimulus with its abdomen or directed its abdomen toward the stimulus. These abdomen curls are the first component of the sting extension response (Gage et al., 2018). In most cases, the stinger was obvious and contacted the stimulus. In some cases, the activity of the stinger was less clear, but the unusual abdomen curls and contact were easy to observe. These abdomen curls were only observed when stimuli were presented during the experiment or when the bees were handled during collection. We used a broad definition of stinging, relative to DLR and biting, to capture instances where the sting extension could not clearly be observed, or where the bee was not able to physically contact the stimulus from its current location. For each trial, we also recorded the order in which DLR, biting, and stinging occurred.

Analysis

All analyses were conducted through the StatsModels package (Perktold, Seabold & Taylor, 2018) included in the Anaconda distribution of Python, a free scientific analysis distribution of the Python programming language (Anaconda, 2019; http://www.python.org). Behavior sequences (e.g., DLR then bite, or bite then DLR then sting, etc.) were analyzed with a series of repeated measures logistic regressions via generalized estimating equations (GEE; Hardin & Hilbe, 2003). We used this series of regressions in place of a multinomial logistic regression as GEE controls for repeated measures within subjects. This technique is also less sensitive to the need for many cases per variable than multinomial regression or chi square analyses; an important consideration for our data, as statistical comparisons between commonly and uncommonly observed behavior sequences answer important research questions. We used an interceptless model where groups are treated as two mutually exclusive variables. By default, a logistic regression’s parameter estimates and associated p values display a difference from a 50% binary chance level. As our sequence analysis considered 16 possible sequences, we subtracted the log odds of 1/16 from all parameter estimates and confidence intervals, then calculated corresponding p values. Each behavior sequence is therefore tested for statistical difference from chance (1/16) instead of a 1/2 comparison that is arbitrary for this data. Individual parameter estimates were compared directly by creating a z score by dividing the difference between the estimates by the square root of the sum of the squared standard errors of the estimates (Clogg, Petkova & Haritou, 1995; Paternoster et al., 1998). Even after adjusting parameters by subtracting the log odds of 1/16, the difference between estimates, z score, and p value are still the same as those normally reported by a regression that includes one level of a categorical variable in the intercept.

Results

A visual overview of the findings can be seen in Figs. 2 and 3, while subsequent sections analyze our primary result, behavior sequence, in detail. Figure 2 shows the percent of trials where a DLR, bite or sting occurred for bees in the experimental and control groups. Bees in the experimental group displayed more behavior. For both groups, DLR was the most common behavior. Bees in the experimental group were more likely to sting than bite, while bees in the control group were more likely to bite than sting. The average number of legs lifted in trials where DLR occurred can be seen in Fig. 3. Not only were bees in the experimental group more likely to emit a DLR (Fig. 2), but they also lifted more legs on average. Both Figs. 2 and 3 show little change across trial that would suggest habituation, sensitization, or fatigue. While our analysis in subsequent paragraphs focuses on behavior sequences, we also included our initial exploratory analysis of individual behaviors (without considering their order) in supplementary material that will relate well to Figs. 2 and 3.

Table 1 shows the percent of trials where a particular behavior sequence occurred. For each trial, the order of DLR, bite and sting were recorded, resulting in 16 possible sequences (no tied rankings were observed). Corresponding statistical analysis that show the log odds of each sequence compared to a 1/16 chance value can be seen in Tables 2 and 3, with Table 2 displaying information for the experimental group, and Table 3 displaying information for the control group. We only included the groups as parameters in this analysis given the large number of behavior sequences to be analyzed and the lack of a trial effect in previous graphs. The direction and magnitude of the parameter estimates, as well as the p values, reflect a difference from the chance value. Although we used separate tables due to the large size of a single combined table, the sequence analyses presented on Tables 2 and 3 should be interpreted together.

Bees in the experimental group were highly active, only being inactive 3% of the time. DLR was emitted first or by itself around 81% of trials, while biting and stinging rarely occurred first or by themselves. All the DLR-first sequences, except for DLR:Bite, occurred significantly more than chance (p values < 0.000). The least common of the DLR-first sequences was the DLR:Bite sequence, occurring during only 1% of trials, and was the only DLR-first sequence to occur significantly less than chance (p = 0.017). Trials with a sting following DLR were much more common (30%), and trials with a bite and sting following DLR (in either order) were also more common (31%). Taken together, this indicates that in the experimental group, DLR is more related to subsequent stings than to subsequent bites. Bees in the control group were inactive during 51% of trials, significantly more than chance (p < 0.000). During trials when bees were active, DLR often occurred by itself (31%), or a bite occurred by itself (15%). These were the only behaviors that occurred significantly more than chance (p values < 0.002). Sequences of multiple behaviors were rare.

Table 4 shows pairwise comparisons between DLR:Bite and other DLR-first sequences, as well as comparisons between DLR:Sting and other DLR-first sequences for the experimental group. The estimate differences and z scores were calculated from the parameter estimates and standard errors reported in Tables 2 and 3. The pairwise comparisons show that the DLR:Bite sequence occurs significantly less than all other DLR-first sequences (p values < 0.000). Conversely, the DLR:Sting sequence occurs significantly more than DLR:Bite, and other DLR-first sequences (p values < 0.009) except for DLR alone. Though DLR:Sting does occur more than DLR alone, the difference is not significant (p = 0.064). Note that for this series for pairwise comparisons, it may be appropriate to use a multiple comparison correction to adjust the significance threshold. For example, the conservative Bonferroni correction would involve dividing the alpha level by the number of comparisons, in this case 0.05 divided by 7 produces a new significance threshold of 0.007, which may affect interpretation of the comparisons between DLR:Sting and DLR-first sequences containing both a sting and a bite. The reader is free to use whichever correction technique they deem appropriate.

Taken together, the findings reported in Tables 1–4 strongly suggest that DLR is more related to subsequent stinging than it is to subsequent biting for the experimental group. This is also in line with our supplementary analysis of individual behaviors. Ultimately, the fact that DLR often precedes stinging in the experimental group, the group where stinging frequently occurred, indicates DLR may function to signal potential predators that a sting is imminent.

Discussion

Our findings describe DLR, a primarily undocumented behavior, and show that it often precedes stinging, but rarely precedes biting alone. We also demonstrate that the probability of DLR is sensitive to stimulus intensity, as the increased proximity to the stimulus in the experimental group, compared to the control group, altered behavior. Together, these results suggest that DLR is an honest signal that indicates stinging may occur. DLR may function in either an aposematic or a pursuit deterrence role, and these functions may not be mutually exclusive. Given the already bright coloration of bumble bees, it is possible that DLR serves as a multimodal enhancement of existing aposematic signals, adding a conspicuous posture to vibrant colors (for discussions on multimodal antipredator signals see Ritson-Williams & Paul, 2007; Rowe & Guilford, 1999; Rowe & Haplin, 2013). If DLR has a pursuit deterrence function, it may signal that the bee is aware of a potential predator and will sting if pursued.

While DLR likely signals a sting may occur, DLR can also occur alone. In the control group, bees emit DLR but rarely sting. This likely occurs because the distant stimulus is intense enough to elicit DLR, but does not support stinging. In the wild, if DLR is successful at preventing a potential predator attack, it may occur without subsequent stinging behavior. Therefore, the occurrence of DLR in the control group is consistent with an honest signal interpretation.

While our experiment clearly indicates a temporal connection between DLR and stinging, additional research is needed to clarify DLR’s specific antipredator function. Such research will need to consider what stimuli and predators elicit DLR, and equally importantly, how predators respond. Field experiments may also study DLR in situ, providing bees with a number of alternative behaviors, including fleeing. Such ecologically valid research may be required to completely determine the function of DLR. For example, if DLR functions strictly as a pursuit deterrence signal, bees may emit a DLR, then flee if a predator approaches, while if DLR has only an aposematic function it may not be related to any antipredator behavior other than stinging.

In addition to further clarifying the function of DLR, our initial work facilitates many additional research topics. For example, research may consider how DLR relates to specific stimulus modalities or intensities, and if predators have learned or innate responses to DLR. Research should also consider the extent that DLR occurs in other Bombus species, and if it differs across species. Studies of individual differences will likely also be fruitful, especially considering recent literature on the size-dependent behavior in bumble bee workers (e.g., Jandt, Huang & Dornhaus, 2009; Kodaira et al., 2009; Raine & Chittka, 2008; Riveros & Gronenberg, 2009a; Riveros & Gronenberg, 2009b; Spaethe & Weidenmüller, 2002). Finally, as bees are social animals, future work should also consider social factors, such as if DLR can be elicited by other bees or if DLR affects adjacent bees. It may also be possible that DLR, like a number of other behaviors, is affected by alarm pheromones (e.g., Avalos et al., 2017; Rossi, D’Ettorre & Giurfa, 2018). Given that bumble bees possess tarsal glands (Pouvreau, 1991; Schmitt, 1990), social odors may even be released during DLR.

Methods

Procedure

Subjects were placed in individual apparatuses after being collected. Each apparatus consisted of a capsule formed from a clear acrylic tube (4.5 cm long, 2.5 cm inner diameter), with two white plastic caps sealing the tube. Two holes (0.4 cm diameter) were drilled near the center of each cap. Each apparatus was placed approximately half a meter apart, and bees were allowed an acclimation period of 45 min.

Each bee experienced 10 trials with a seven-minute ITI after the acclimation period was complete. During each trial, a researcher startled the bee by presenting a hand 15 cm above the apparatus, rapidly lowering it to approximately 6 cm above the apparatus, rotating the hand once in a clockwise circular motion, and then withdrawing the hand. As strong stimuli can inhibit habituation or cause sensitization, we used this relatively mild stimulus, compared to those used in the first experiment, to increase the chance that habituation could be observed. The bees’ response was recorded during the two-second stimulus presentation and for three seconds after the presentation. DLR was recorded as a binary response and no other behaviors were recorded. Subjects collected in 2020 also experienced three additional trials. On the 11th trial, the bottom of the apparatus was tapped once, out of view of the subject. This trial served as a dishabituation trial to determine if DLR would occur to other stimuli and verify that any decrease in DLR during the preceding 10 habituation trials was not caused by fatigue. The 12th trial was a return to the standard habituation trial. Finally, on the 13th trial, subjects were collected by reaching directly toward the front of the apparatus, visible to the subject, and holding a hand adjacent to, but not touching, the apparatus for the five-second observation period before placing the apparatus in the refrigerator for chilling. All bees, in both 2018 and 2020, were chilled after recollection, marked, and returned to their colony or the collection area.

Analysis

We analyzed the probability of DLR across trial using repeated measures logistic regression via GEE. As with previous logistic regressions, we use an interceptless form so that the parameters can be directly compared to a chance value (in this case 50%), then compared them to each other by creating a z score by dividing the difference between the estimates by the square root of the sum of the squared standard errors of the estimates. As initial analysis revealed no significant differences between bees in the 2018 and 2020 collection periods (p values < 0.489), and there was no theoretical reason to expect differences, we did not include collection period as a parameter in our main analysis.

Discussion

In this experiment, we provided the first demonstration of habituation of DLR, as well as documented differences in DLR across samples of captive and wild populations. While both samples showed a similar rate of habituation, the captive bees were initially more likely to perform DLR. This difference in DLR may have occurred for two different reasons. First, the samples of worker bees we collected from captive and wild populations may have differed in role specialization. Bumble bee castes include the reproductive queen and drone castes, as well as the primarily non-reproductive worker caste frequently used in research. Workers may be further specialized. The smaller worker bees are more likely to feed larvae and attend to hive maintenance, while larger workers act as foragers. In bumble bees, role specialization appears to be determined during early development, and research suggests that physical dimensions can predict behavioral performance (Jandt, Huang & Dornhaus, 2009; Kodaira et al., 2009; Raine & Chittka, 2008; Riveros & Gronenberg, 2009a; Riveros & Gronenberg, 2009b; Spaethe & Weidenmüller, 2002).

The bees sampled from our captive population were significantly smaller than those from our sample of wild bees, suggesting the wild bees were more likely to be foragers. The wild bees were also clearly collected during the act of foraging, further increasing our confidence they fit this role specialization. While our captive bees were also collected in their foraging area, we are less confident this collection method ensures they are true foragers due to one of the challenges of maintaining an indoor bumble bee hive. In laboratory colonies, bumble bee workers may not return to the hive after feeding, and may instead inhabit a flight cage or feeding area. Workers may build clusters of cells near a food dish, store food, and even raise drones. We call this tendency the “lost bee effect.” If not carefully managed, this may result in half the colony moving away from the hive within a month (see Fig. S1 in the supplemental material). Although not widely published (we have only seen this reported in Jandt & Dornhaus, 2009 and Blacquière, Cornelissen & Donder, 2007), the lost bee effect appears to be a common issue. While there is no clear solution, preventative measures include capturing lost bees each day, returning them to the colony, and cleaning the flight cage to remove odors (F Muth, pers. comm., 2018); capturing and returning lost bees while also killing repeat offenders (J Jandt, pers. comm., 2018); feeding the bees inside the hive and completely preventing access to other areas (W Gronenberg & A Riveros, pers. comm., 2018); and providing a smaller antechamber between the hive and the flight cage to encourage lost bees to build closer to the hive (J Nieh, pers comm., 2018). While our observations and the above communications are in regard to the North American Bombus impatiens, the lost bee effect also occurs in the European B. terrestris (L Chittka, pers. comm., 2018). Given the tendency of worker bees to become “lost”, it may not be possible to ensure that bees collected near food are the larger, foraging-specialized workers. Instead, collected subjects may also consist of smaller bees specialized for brood care and hive maintenance. While a growing body of research demonstrates the usefulness of laboratory colonies, the lost bee effect indicates they may not be thriving, and this may impact their use as model organisms. In our case, the lost bee effect may have made our sample of captive bees less likely to be foragers.

A second possible reason for the difference in DLR between captive and wild bees may be the distinct experiences of bees raised indoors compared to those of wild bees. Captive bees were only exposed to stimuli in their hive and flight cage, and ultimately experienced only a small number of stimuli before research. Conversely, wild bees likely contact many stimuli during daily foraging including other insects, birds, pedestrians, and even landscaping equipment. It is possible that exposure to a wide variety of stimuli served to acclimate the wild bees to mild visual stimuli, such as the hand wave used in this experiment.

Regardless of the difference in initial rate of DLR, both captive and wild bees showed clear habituation trends, and thus our experiment suggests expansive opportunities for a new area of non-associative learning research with bumble bees. Future research may consider the principles of habituation and sensitization outlined by Thompson & Spencer (1966), Groves & Thompson (1970), and Rankin et al. (2009). For example, altering the time between stimulus presentations may change the rate of habituation, and placing the animal in an agitated state prior to habituation trials may instead result in sensitization. Future work may also consider exploring classical conditioning or operant conditioning of DLR. This would be a reasonable next step considering the reports of sting extension response (SER) conditioning in honey bees, though the primary author and one reviewer note that SER conditioning may be less robust than the literature suggests. Additionally, various DLR conditioning studies could be used as a basis for research on pesticides, sensory perception, memory, pharmacology, and neurophysiology research, as conditioning research with honey bees has also done for these same topics (e.g., Abramson et al., 2004; Abramson et al., 2006; Faber, Joerges & Menzel, 1999; Giurfa et al., 2009; Linader, De Ibrra & Laska, 2012; Mustard et al., 2012; Varnon et al., 2018; Vergoz et al., 2007).