Collective defense is one of the most ubiquitous behaviors performed by social groups. Because of its importance, complex societies may engage a set of defensive specialists, with physical and/or neurological attributes tuned for defense against specific invaders. These strategies must be balanced, however, with the need to flexibly respond to different threat levels and sources. Insect societies rely heavily on olfaction for detecting and communicating in the context of defense. We therefore asked whether threat detection via olfaction is specialized towards invader-specific cues and how this may be integrated into defense task specialization. Colonies of the stingless bee, Tetragonisca angustula, deploy a morphologically distinct sub-caste of larger-bodied workers (soldiers) for colony defense. These soldiers transition between two different guarding tasks as they age, progressing from guarding in a hovering position near the nest entrance to guarding in a standing position on the nest entrance tube. Hovering and standing guards intercept different types of invaders: primarily heterospecific versus conspecific, respectively. We asked whether hovering and standing guarding behaviors were modulated by differential sensitivity to invader-associated olfactory stimuli; then we compared their responses to these cues to those of smaller workers that perform predominantly non-defense tasks. We exposed bees under both field and lab conditions to citral, a kairomone produced by an obligate heterospecific nest robber, primarily intercepted by hovering guards. Consistent with their roles, hovering guards were more likely to move towards citral than were either standing guards or small-bodied bees within a Y-maze. We also presented guards at field nests with dummies of conspecific versus heterospecific invader types, varying whether they included citral odors. Standing guards were more responsive to conspecific intruder scenarios than hovering guards, but heterospecific response differed by presence of citral. Standing and hovering guards responded in similar proportions when citral was absent, but the addition of citral produced a marginally non-significant reduction in standing guard participation. Our results potentially demonstrate differentiated cue-specific responses that correspond to morphological task specialization and age polyethism in these eusocial societies.

All field and lab manipulations were conducted using subjects from naturally occurring nests in Gamboa, Panama (9.12° N, 79.70° W). Data were gathered in January 2018 and February 2019. Task group assignments for each experiment were made using established behavioral criteria (Baudier et al. 2019). Bees hovering at the entrance for longer than 20 s while facing inwards towards the flyway were considered hovering guards. Bees standing on the wax and resin nest entrance tube for longer than 20 s while facing towards the entrance opening were considered standing guards. Minors (smaller-bodied worker bees that do not typically defend the colony) were collected while in the task of foraging because this task is performed by small workers during 3–4 weeks since adult emergence, making as close as possible an age-matched comparison group to soldiers that perform the task of hovering in approximately the third week following adult emergence and the task of standing guarding in approximately the fourth week following adult emergence (Hammel et al. 2016; Baudier et al. 2019). A bee was considered a forager if it exited the nest and flew immediately away from the nest entrance (unlike hovering guards) but carried no detritus material (unlike dump workers). In a previous study, subject weights confirmed that hovering and standing guards gathered using this method were significantly larger in body size than collected foragers (Baudier et al. 2019).

Y-maze olfactory response bioassays

We conducted two types of Y-maze assays. The first assessed affinity of bees to citral, and the second assessed starvation-motivated bee affinity to a nutrition source (mixed pollen and honey). This allowed for the comparison of defense-specific olfactory response as well as appetitive olfactory acuity across the 3 focal task groups: minor bees, standing guards, and hovering guards. All Y-maze assays were performed in an ambient lab with relative humidity between 81 and 100% and temperature ranging from 24 to 30 °C.

To test the affinity of bees in different task groups for citral, we introduced bees to Y-mazes with citral at the end of one arm. A total of 54 naïve individual bees were used (6 bees × 3 task groups × 3 colonies per assay). Subject bees were first collected from nest entrances by aspirator into 38 × 84 mm vented plastic tubes and placed into a dark environment for 15 min prior to each citral Y-maze assay in order to allow bees to calm from any alarm caused during collection. We used 3D-printed Y-mazes constructed from white PLA plastic (MakerBot Industries, LLC) and, with an interlocking, transparent acrylic lid to enable observations. Y-maze arms were 10 cm long, 1 cm wide, and 1 cm tall. A 2.5 × 1 × 1 cm entrance vestibule adjoined to arms, each offset at 45° (90° angle between the arms). A quantity of 2 μl of 95% citral (Sigma-Aldrich, CAS Number 5392-40-5) was placed on a 4 × 4 mm filter paper, while the other arm was left with an unscented filter paper. This small dose was based on previous studies which approximated citral quantity in natural contexts, given that 1 ml of a 1% citral solution was sufficient to illicit defensive behavioral responses in this species (Jernigan et al. 2018). Air flow was provided in the maze using a pump to inhibit odor diffusion in the non-treatment arm. We set the air flow to 166 ml/s, strong enough to prevent the scent of citral from permeating the opposite arm (according to human olfaction), but mild enough to not inhibit the smallest bees from navigating through the maze. To eliminate directional bias, citral was alternated equally between the left and right arms of the Y-maze. Each replicate of the assay started when the subject bee was placed in the vestibule and ended only when the scented filter paper was contacted by the bee (all bees in the study touched the scented filter paper in under 5 min of observation). The time until each bee contacted the scented filter paper and each bee’s first arm choice (towards versus away from odor) were recorded. A subject’s first walk > 5 cm down an arm from the vestibule was considered its first arm choice. Y-mazes were scrubbed with soap and water, rinsed with water, then rinsed with ethanol, and allowed to air dry between replicates. Similar Y-maze materials and cleaning protocols have been used in previous olfactory studies with hymenopteran subjects, without apparent complications associated with scent contamination (Carcaud et al. 2009; Provecho and Josens 2009; Arenas and Farina 2012; Duell 2018).

To test for differences in general olfactory acuity, we similarly compared the 3 groups (minors/foragers, hovering guards, and standing guards; N = 54 naïve bees; 6 bees × 3 task groups × 3 colonies per assay) in their ability to locate the smell of food reward in the same Y-maze setup when starved/water deprived. The same collection and Y-maze procedures were used for this, except that bees were deprived of food or water in darkened, vented tubes for 55 min prior to Y-maze introduction (to motivate food search), and a liquid food reward mixture (40% honey and 20% pollen of locally farmed Apis mellifera, 40% water) was applied to one end of a Y-maze arm as the focal odor source. This mixture was wiped lightly along the last 0.5 cm of one Y-maze arm. We recorded time until the liquid food mixture was contacted and the bee’s first arm choice.

All analyses were conducted in R version 3.4.4 (R Core Team 2020). Variation between task groups in time until solving either food or citral odors in the Y-maze was compared using mixed model survival analyses due to the non-parametric distribution of the data. In each case, we used a Cox proportional hazard analysis that included and colony ID as a random factor and task group (forager/minor, hovering guard, standing guard) as a fixed predictor of time until bees found the odor source within the mazes using the coxme command in the package “coxme” (Therneau 2015). We then followed with a post hoc Tukey HSD test to compare among the three task groups.

To test for task group differences in first Y-maze arm choice, we fit a binomial mixed effects model with arm choice as the response variable, task group (hoverer, stander, forager) as a fixed factor, and colony as a random factor using the glmer function (Bates et al. 2015) and then performed a Type-II analysis on the fitted model to test significance of fixed effects. This was done separately for the citral and starved food choice assays.

At-nest manipulations

To test whether lab-observed patterns in citral reaction were observable in a more naturalistic field setting, we measured guard responses to citral-coated dummies at nest entrances. At each subject colony nest entrance, we introduced two treatment mock invaders and two controls. The first treatment was a freshly freeze-killed worker of the stingless bee Partamona peckolti, an environmentally abundant competitor in the area that is black in body color (like Lestrimelitta) but does not use citral to recruit raids. The second treatment was an obligate nest-robber kairomone dummy consisting of a black polymer clay figure of roughly equal size, shape, and color (black) as the average Lestrimelitta danuncia worker and dipped in 95% citral (approximately 0.02 ml of citral). Tetragonisca angustula respond similarly to actual invaders as they do to such color- and odor-matched dummies (Bowden et al. 1994).

Using this design, if the division of labor between hovering and standing guards is (at least in part) driven by differences in responsiveness to olfactory cues, then we predicted that a higher number of hovering guards would respond to the high-citral dummy, compared with the responsiveness of standing guards. For reference, we also included a positive control for eliciting standing guard attack behavior: a freeze-killed non-nestmate T. angustula worker. Each intruder dummy was harnessed with wire onto the end of a dowel before being presented at nest entrances. An empty wire was also used as a negative control for attack behavior alongside each of the three treatment dummy invaders. A negative result from the empty wire control would confirm that there was no olfactory contamination of instruments and that the effect of non-olfactory disturbance at the nest (human presence, vibrational disturbance, etc.) on the aggressive behaviors of interest was minimal.

Each mock intruder or control was brought directly towards the nest entrance tube (0° angle) and contacted the edge of the tube for 2 s. During each trial, we recorded number of attackers (predominantly soldiers) and each attacker’s guard type (hoverer versus stander). This is a standard method for eliciting and measuring defensive guard attack in this species (Bowden et al. 1994). Attacks were defined as contact with a mock invader for greater than 2 s. Each mock intruder and wire were discarded and replaced after use to prevent cross contamination of scents. Treatments and control wire trials were conducted 15 min apart at each subject colony and were presented in a pseudo-random order. A total of five colonies were tested in this manner, and each colony was repeatedly tested on 3 different days.

We used a mixed effect model approach to repeated-measures analysis comparing the effects of guard type and invader type on total responses. Number of bee responses was first z-transformed to improve normalcy of distributions. Next, we fit a linear mixed effect model (lme4 package, lmer function) (Bates et al. 2007; Bates et al. 2015), which included guard type (hovering vs. standing), mock intruder treatment, and the interaction between the two as fixed predictors of number of bee responses (z-transformed number of bees contacting model for at least 2 s) and included colony as a random factor. We used a Type-II Wald chi-square test of this model to test significance of fixed factors. We then followed this analysis with post hoc pairwise comparisons of standing versus hovering guard responses within each intruder treatment (colony kept as a random factor), using multiple Type-II Wald chi-square tests with adjusted p values according to these multiple comparisons using the Benjamini and Hochberg method (Benjamini and Hochberg 1995).

References

Arenas A, Farina WM (2012) Learned olfactory cues affect pollen-foraging preferences in honeybees, Apis mellifera. Anim Behav 83:1023–1033

Bates D, Sarkar D, Bates MD, Matrix L (2007) The lme4 package. R package version 2:74

Bates D, Mächler M, Bolker B, Walker S (2015) Fitting linear mixed-effects models using lme4. J Stat Softw 67:1–48

Baudier KM, Ostwald MM, Grüter C, Segers FHID, Roubik DW, Pavlic TP, Pratt SC, Fewell JH (2019) Changing of the guard: mixed specialization and flexibility in nest defense (Tetragonisca angustula). Behav Ecol 30:1041–1049

Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Roy Stat Soc B Met:289–300

Bowden RM, Garry MF, Breed MD (1994) Discrimination of con-and heterospecific bees by Trigona (Tetragonisca) angustula guards. J Kansas Entomol Soc:137–139

Carcaud J, Roussel E, Giurfa M, Sandoz J-C (2009) Odour aversion after olfactory conditioning of the sting extension reflex in honeybees. J Exp Biol 212:620–626

Duell ME (2018) Matters of size: behavioral, morphological, and physiological performance scaling among stingless bees (Meliponini). In. Arizona State University

Hammel B, Vollet-Neto A, Menezes C, Nascimento FS, Engels W, Grüter C (2016) Soldiers in a stingless bee: work rate and task repertoire suggest they are an elite force. Am Nat 187:120–129

Jernigan CM, Birgiolas J, McHugh C, Roubik DW, Wcislo WT, Smith BH (2018) Colony-level non-associative plasticity of alarm responses in the stingless honey bee, Tetragonisca angustula. Behav Ecol Sociobiol 72:58

Provecho Y, Josens R (2009) Olfactory memory established during trophallaxis affects food search behaviour in ants. J Exp Biol 212:3221–3227

R Core Team (2020) R: a language and environment for statistical computing, R Foundation for Statistical Computing, Austria. In. ISBN 3-900051-07-0: URL http://www.R-project.org

Therneau T (2015) Mixed effects Cox models. CRAN repository

At nest manipulations.csv

This file contains all at-nest manipulation data associated used in this study.  Headder definitions are as follows.

ColonyID - unique identifier for each of 5 naturally occuring colony/nest entrances used in this part of the study.

Rep - replicate ID (repeated across 3 days, 3 levels: day 1, 2, or 3)

guardTYPE - type of responding soldier (2 levels: hovering guard, standing guard)

guardRESP - number of each guard type identified in "guardTYPE" responding to each invader type

food detection Y-maze.csv  & alarm detection Y-maze.csv

These files contain all y-maze data analyzed in this study (other than first arm choice, which is presented in tables within the manuscript). They contain data for speed at which different task group bees detected food and citral (respectively) within Y-mazes. Headder definitions are as follows:

BeeID - individually unique identifier of each naive bee used in this study. 

ColonyID - unique identifier for each of 3 naturally occuring colony/nest entrances used in this part of the study.

Task - task group of each bee (3 levels sampled in each colony: hovering guard, standing guard, & foragers)

cens - censor data used in the performed survival analysis. Because all bees found the odor sources in the maze eventually, these values are all "1"