The experiments were conducted with two Giant Honey Bee (Apis dorsata) colonies in Chitwan (Nepal). The experimental nest I (expN₁) was located at the Institute of Agriculture and Animal Science (Tribhuvan University, Kathmandu) in Rampur (Chitwan, Nepal) and observations were made in February 2009. There, about 100 fully developed nests co-existed in a radius of two kilometers, which all migrated away three weeks later [14]. The expN₁ measured 173 × 59 cm (horizontal × vertical) and possessed honey cells located to its upper left side and at the concave-shaped area to its right (Figure 1A). This spatial disposition indicates that amalgamation of two neighboring nests took place some months ago. The experimental nest II (expN₂) was located at a hotel site in Sauraha (Chitwan National Park) and observations were made in November 2010. This nest was estimated to be approximately three weeks old. The nest had a hemicyclic form (83 × 60 cm) and was attached to balks of concrete of a balcony of the hotel (Figure 1B). Another small queenless colony existed in the surroundings and an additional colony arrived during the course of our experimental session. In the first days of our stay (i.e., on October 28 and 29, 2010) the comb was only covered by a single layer of honeybees. However, after two days, the colony achieved a multi-layer coverage because young bees had progressively hatched.

The experimental nests were filmed with a high definition video camera (Panasonic HVX 200, resolution: 1280 × 720 px). For the expN₁, the camera was placed 8 m away from the southbound side of the nest and the recordings were made with a resolution of 6 px/cm (Figure 1A). For expN₂, the distance between camera and nest was 1.5 m and the recordings were made with a resolution of 8 px/cm. Both working distances provided an undistorted view of the whole nest and were also sufficient to keep the colony undisturbed, as both colonies accepted the camera as an additional landmark.

Infrared thermography [29,31,32] was only used on expN₂. The expN₂ was synchronously filmed with an infrared camera (Flir A320; 9 Hz) and the high definition video camera. The cameras covered the whole nest at a nearly perpendicular angle. Long-term infrared monitoring allowed for the detection of the heating-up of bees.

We aroused the colonies with a dummy wasp [23]. The dummy wasp was presented by a miniature computer-controlled cable car and was moved with constant velocity (variable between 0.1–0.5 m/s) along a horizontal cable of 20 cm in front of the nests near their attachment zones (Figure 1A,B). For the expN₁, we used the southbound side of the nest (Figure 1A) for the stimulation with the dummy wasp. This stimulation method was chosen to mimic a free-flying wasp scanning in front of the nest.

We monitored the expN₁ and expN₂ for three and ten days starting from 10:00 am and 7:00 am until sunset, respectively (Figure 2). At the beginning of each observation session, the test colony was in a quiescent mode (non-arousal state) with rare shimmering waves which were provoked by natural stimuli such as birds and insects. Therefore, even at this stage of the experiment, before the dummy wasp was presented, we were able to observe flickering under two conditions, i.e. under non-arousal and arousal.

We then selected time intervals for the analysis of flickering events (the detailed scheduling is given for both nests in Figure 2C). In the preP-phases (marked yellow in Figure 2), the flickering activity was monitored under undisturbed conditions for a total of 12.95 min (expN₁: 8.09 min, expN₂: 4.86 min). After short pauses in the recording session (expN₁: 4.15 min, expN₂: 1.00 min), the dummy wasp was presented twice to the expN₁ and six times to the expN₂ (see tracking line dw in Figure 2A,B).

The performance of the nests was determined by three factors: (i) by the “natural” level of quiescence of the colonies at the start of the experiments; (ii) by the arousal state provoked by repetitive presentation of the dummy wasp; and (iii) by the non-arousal state between each presentation session (Figure 2A,B). In the pre-presentation (preP) phase, it was ensured that the colony was calm and undisturbed. For that, the experimental colony was checked for any momentary arousal state. Signs of arousal included a tendency for shimmering without external threat and both the contraction and the elongation [4,31] of the bee curtain (indicating that the colony was ready to release flying guards). The time to recovery after a disturbance depended on the kind of threat. A Giant Honey Bee colony needs tens of minutes to calm down after a raid by either blue-bearded bee-eaters (Nyctyornis athertoni) [5] or Oriental Honey-buzzards (Pernis ptilorhynchus) [23] while the colony practically recovers within tens of seconds after a wasp attack [10]. The presentations of the dummy wasp (P-phase) were launched after the colony had returned to a low-arousal level. Subsequently, the colonies calmed down again. In case of the expN₁, the post-presentation (postP) phase was taken to start after the small, “self”-provoked shimmering waves (see above) were not observed for a period of at least 15 s. In the case of expN₂, the P-phase lasted for approximately one minute and the postP phase started after a pause of one minute. This sequence of preP-, P- and postP-phases was repeated 6 times. The dummy wasp was presented in several runs individually lasting between 66–100 s and 21–60 s for expN₁ (Figure 2A,C) and expN₂ (Figure 2B,C), respectively.

Selected video sessions were processed by Avid X-press-Pro Software to fragment single frames and transform them into jpg-format [28]. In total, we analyzed over 20 min of video sequences at 25 fps for expN₁ and over 17 min at 50 fps for expN₂, corresponding to approximately 29,000 and 50,000 images, respectively. The images were processed with the Image-Pro and Media Cybernetics programs to detect and quantify movements of single bees at the nest surface [30]. Movements that produce changes in pixel luminance (Δlum) can be identified by image analysis following pixel-operated subtraction and segmentation of the differences between two subsequent images into black-and-white areas (Figure 3). Detectable movements include positional changes in the horizontal and vertical directions and positional changes of the bees’ body parts. A value of Δlum = 0 represents the state of motionlessness and Δlum = 255 the maximum attainable intensity of motion.

For a quantitative analysis of area-specific motion activity, we have superimposed an assessment grid over the images of the nests (Figure 2A,B). The unit length was set to 15 mm as it corresponds to the size of individual bees at the surface of the nest. The grid possessed 123 × 53 squares in expN₁ and 73 × 48 in expN₂ (horizontal × vertical), corresponding to 184 × 80 cm and 110 × 72 cm areas, respectively.

Two categories of nest cohorts were distinguished with respect to the generation of shimmering waves. The trigger sites (ts) that generated either parental or daughter shimmering waves in the observation sessions were identified by retrograde inspection of the video films [28]. The locations of ts-bees were determined interactively by semi-automated assessment (Image-Pro, Media Cybernetics) of the x- and y-positions of the thoraces, utilizing raw video material combined with difference images (Figure 3). All other sites were termed non-trigger sites (nts). By definition, all parental and daughter waves have been launched by ts-bees, at least during the span of the observation sessions.

Large parental waves often consisted of a series of daughter waves (movie 1) and the trigger activity could extend across single squares of the test grid because it is mostly formed by trigger cohorts [28]. Therefore, we defined ts not only by locating the relevant test square (termed source square) but rather used a more complex definition; the ts corresponded to the positional coordinate of the source square which had been detected manually, but was added up by coordinates of the eight test squares encircling the source square. Consequentially, we assigned the trigger activity value 1.0 to nine test squares representing a ts. In total, for both experimental colonies, we collected 1453 trigger events from 130 parental shimmering waves and pooled the trigger activity values with respect to their coordinates. The trigger activity values were scaled relative to the maximal sum of trigger activity values assessed in the observation interval (in number of trigger events per test square per reference time interval). The results, plotted in Figure 4, form nuclei and zones with gradual intensity patterns.

In difference images, we detected all types of motions of surface bees. The automated differentiation of flickering from other motion categories such as shimmering, walking, dancing or flying, was only possible using interval filters. For the specification of such filter properties, we first identified flickering events and determined their critical duration. We then assessed the changes in luminance (Δlum) generated by selected sample bees over 100 s. A typical flickering event was defined for a selected sample bee as having a motion value that exceeded the threshold strength of Δlum > 2.0 for up to three frames (i.e., 60 ms) and was typically repeated after more than one second. In the left diagram of Figure 5c, the motion values collected from a single test square over 4 minutes are monitored. Using our filter, other background activities, such as activities in convection holes (where fanning bees are positioned [33] and produced motion patterns with their wings lasting more than three subsequent frames), could be easily eliminated. Essential interval filters have been applied to allow the identifying and plotting of only-flickering events. However, the filter was unable to discriminate flickering from the many movements occurring in the mouth zones. Flickering events occurring in this area were rather identified by manual inspection and excluded from further automated analyses.

We identified the ts-bees and determined their image-based x- and y-coordinates sequentially in the shimmering sessions (line sh in Figure 2A,B) of the experimental phases (preP, P_i & postP_i). The incidence of shimmering events at ts of parental or daughter waves was plotted into the assessment grids (Figure 2 and Figure 4) in relative scale.

Each flicker event detected was assigned to its respective test square (Figure 5). For each test square, the total activity level was plotted against (i) the flickering rate (i.e., number of flicker events per min) and (ii) the intensity of flickering (i.e., the total motion strength in the test squares per reference time interval t_ref calculated as _fI = Σ Δlum/t_ref). As the flickering intensity can be transformed to area of pixels /s, this measure also allows a scaling of the flickering activity: the number of flickering-active bees per second and per reference area of the nest. These data have been collected separately for both the ts and nts areas and for both the expN₁ and expN₂. We further compiled the flickering intensity values according to 30 rate classes (with steps of 0.03 s⁻¹). Finally, the means and mean errors of flickering intensities of both cohorts (ts and nts) were calculated, offering the possibility to compare flickering activities for two nests of different arousal or age conditions. We further verified for systemic differences in flickering intensity _fI between ts- and nts-cohorts. To do so, we quantified the difference of arithmetical mean of intensity between the two cohorts for each rate class (Δ _fI = _fI_ts − _fI_nts). We sorted the results whether Δ_fI was positive (i.e., _fI_ts > _fI_nts) or negative (i.e., _fI_ts < _fI_nts).

In the morning, at the start of the observation session, the ambient temperature was typically 14 °C, with misty conditions in February. During the day, temperatures could climb up to 26 °C in the midday bright sunshine (Figure 2C). The mouth zones [1,10] of the experimental nests were not established in the morning (Figure 2A,B). Only later during the day, after the mist had disappeared, the foragers departed from the nest and the mouth zone became non-quiescent, and grew to become 20% of the nest size. The mouth zones were located at the bottom left and at the left side of the expN₁ and expN₂, respectively (Figure 5).

At the beginning of each observation session, the test colony was in a quiescent, undisturbed mode (non-arousal state), with rare flight and dancing bustles. From time to time, shimmering waves were provoked by natural stimuli, such as birds (e.g., sparrows and pigeons for expN₁) and insects (e.g., Apis cerana and wasp specimens for expN₂).

In the preP-phases, the flickering activity was monitored under undisturbed conditions. During this time, we detected 47 shimmering waves (expN₁: 36, expN₂: 11) that lasted for a total of 3.24 min (Table 1). During the P-phase, a series of shimmering waves occurred, indicating that the colonies were aroused by the dummy. After the dummy disappeared, i.e., in the postP-phases, the shimmering waves decreased in repetition rate (see the sh plots in Figure 2A,B) and intensity. The complementary color-coded fl-plots (Figure 2A,B) give the time windows (expN₁: 8.78 min, expN₂: 5.63 min) of the postP-phases during which flickering activity was monitored.

Shimmering events at ts were plotted into the assessment grids (Figure 4) in relative scale. In the preP phases, all 47 parental shimmering waves were considered for the analysis of 383 trigger sites (Table 1). During the P₁-, P₂-, and the subsequent postP₁- and postP₂-phases, 25% of the shimmering waves were considered; amounting to 1453 trigger sites (Table 1).

For expN₁, the preP-phase ts were evenly distributed over the whole nest surface, except for a lesser occurrence in the bottom-left (Figure 4a₀). Minutes (indicated by the curve) later, the mouth zone developed in that area (Figure 2A), after the morning mist cleared away and the sun shone onto the nest. The regions with highest ts were found on the upper left side of the nest and in the concave region to the right side. These ts positions changed during the course of the experiment. During the P₁- and postP₁-phases (and also during the P₂- and post P₂-phases) the ts positions were distributed on the upper side of the nest, along the trajectory of the dummy wasp. In the postP₂-phase (Figure 4a₂), the previous ts activities located in the upper left nest area disappeared, but were preserved in the concave region to the right side.

The summarized distribution of ts of expN₂ (Figure 4b) gives a different picture, because in all experimental phases, the ts were dispersed over the whole nest surface. The main reason for this homogeneous distribution presumably was the smaller nest size, which allowed the dummy wasp to affect the whole nest. In the preP phase (Figure 4b₀), the most active ts only showed an incidence of four events over the observation period, because of the low number of naturally provoked shimmering waves (Table 1). However, during the dummy wasp presentations, the number of shimmering waves was high and the number of ts increased from the preP-phase to the postP₁- and postP₂-phases. The ts were distributed in a slightly diagonal band spanning the lower left to the upper right side. The highest numbers of ts were found on the upper left side of the nest, where the mouth zone appeared later in the day (Figure 4b_1‑2).

Figure 5 summarizes the flickering activity of the expN₁ and expN₂ surface bees during the preP- and the postP_i-phases. In the preP-phase, the rate of occurrence and the intensity of the flickering were very low (Figure 5a₀,b₀, c_expN1 [n_sq = 1]). The presentation of the dummy wasp dramatically changed the distribution of aroused surface bees and increased the number of flickering incidents. It is to be noted that the mouth zones developed minutes after the recording of the preP-phase and before the start of the postP₁-phase, as seen by the areas of high activity. The locomotory activities have been plotted only to illustrate the location of the mouth zone. After the P₁-phases, both nests displayed flickering in the absence of locomotory activity outside the mouth zone, in particular on the upper side of the nest where the dummy wasp passed. Interestingly, the flickering rate of occurrence is astonishingly low in the girdle neighboring the mouth zone (Figure 5a_1-2), where ts-bees were not found (Figure 4a_1-2).

Figure 6 schematizes the development of the two experimental nests during the course of the experiments with respect to the relative number of flickering-active ts- and nts-bees (Table 1). During the preP-phases (Figure 6a₀,b₀), the area of flickering-active bees represented a third of both monitored nest surfaces (expN₁: relF_ts+nts = 39.63%, expN₂: relF_ts+nts = 33.67%; Table 1). The relative number of nts-bees displaying flickering was 22.18% (with 100% being the number of test squares covering the nest area), which is more than the number of flickering-active ts-test squares (relF_nts = 17.45%). In expN₁, the number of flickering-active ts- and nts-test squares increased in the postP₁-phase (relF_ts+nts = 86.42%, relF_ts = 28.54%, relF_nts = 57.88%; Table 1 and Figure 6a₁) and maintained similar proportions in the postP₂-phase. In expN₂, the number of flickering-active ts-test squares increased while that of nts-test squares decreased during the postP₁-phase (relF_ts+nts = 23.52%, relF_ts = 12.45%, relF_nts = 11.06%; Table 1 and Figure 6a₁), but both categories increased to proportions comparable to expN₁ in the postP₂-phase. With one exception (postP-phase in expN₂), the number of flickering-active nts-bees was higher than that of flickering-active ts-bees.

We used three subsets of parameters to characterize flickering as collective behavior. First, rate and intensity (see Methods) were used to examine flickering with respect to non-arousal and arousal. Both parameters can be plotted into the test grid (e.g., Figure 5 shows flickering intensity) to analyze, in second place, the flickering activities of ts- and nts-bees at the nest surface. Third, the state of arousal was analyzed with respect to the occurrence and the repetition rate of shimmering waves. The arousal state evolved from quiescence to alertness by natural stimuli or by experimental presentation of a dummy wasp. Figure 7 presents the summary of the flickering behavior analysis for the two experimental nests under three arousal states (preP, postP_1-2).

In the rate-intensity plots (Figure 7a-f), the values of flickering intensity were compiled according to 30 rate classes that varied by increments of 0.03 Hz (the total number of events was 6519 for expN₁ and 3504 for expN₂). The mean intensity and SEM were transformed into pixel area per s (see Methods). This parameter is suited for comparing the behavior of the two experimental nests, despite their different size and history of arousal.

In the preP-phases (Figure 7a and d), the flickering behavior was characterized by rates below 0.4 Hz (expN₁: 0.135 Hz, expN₂: 0.405 Hz) and intensities below 2.0 pixels per s (expN₁: 0.27, expN₂: 2.03). The ts- and nts-curves can be expressed by exponential functions (y = a. e^b.x) with the following parameters: expN₁ [ts]: a = 0.037, b = 15.457, R² = 0.98; expN₁ [nts]: a = 0.0325, b = 14.583, R² = 0.97; expN₂ [ts]: a = 0.2485, b = 5.2793, R² = 0.94; expN₂ [nts]: a = 0.2501, b = 3.8059, R² = 0.80. The curves indicate that, for both categories (ts, nts), the more intense the flickering, the more frequent the rate becomes (P < 0.001; Spearman correlation test). Figure 7h-n show the differences in flickering intensities between ts- and nts-cohorts. The positive results were plotted in the upper graphs while the negative results were plotted in the lowers graphs of Figure 7h-j (expN₁) and Figure 7l-n (expN₂) (see Table 2 for regression functions). Finally, the difference between the number of positive and negative cases per arousal condition was analyzed by χ²-test, which showed that the intensity of the ts-cohorts in the preP-phases was significantly higher than for nts-cohorts in most of the rate classes (P < 0.05, χ²-test, see yellow bars in Figure 7k and o).

For expN₂, only two postP-phases were considered out of the six sessions of dummy wasp presentation (Figure 2B). This is because the arousal level of the colony (Figure 7g) significantly increased after the P₃-phase, going from 0.64 ± 0.04 to 1.96 ± 0.03 (mean ± SEMs) for ts-zones and from 1.21 ± 0.07 to 2.18 ± 0.02 for nts-zones (3-fold and 2-fold increase, respectively). Simultaneously, the numbers of flickering-active test squares increased, going from 33 (P₃) to 441 (P₄) for ts-areas and from 34 (P₃) to 573 (P₄) for nts-areas. Therefore, it seems that the colony had drastically changed its arousal state only after the P₃-phase.

In the postP_1-2-phases, the rate-intensity curves of the expN₁ ts- and nts-cohorts shifted to higher intensity and rate values compared to the preP-phase curves (Figure 7b,c), with the maximum intensity value going from 0.27 in preP-phase to 2.79 in postP₁-phase, and the maximum rate going from 0.135 Hz in preP-phase to 0.40 Hz in postP₁-phase. The results of the postP₂-phase are similar to those of the postP₁-phase. The expN₂ evolved differently; for this nest, the flickering activity did not significantly change between the preP- and postP₁-phases with respect to both intensity and rate (Figure 7d,e). However, a shift to higher values similar to that observed in expN₁ occurred in the postP₂-phase, especially with respect to rate (Figure 7f).

The differences in flickering intensity between ts- and nts-cohorts (Δ_fI = _fI_ts − _fI_nts) in the postP-phases were similar in proportion as in the preP-phase (Figure 7k,o). Furthermore, the flickering intensities of the ts-cohorts were higher than for the nts-cohorts in nearly every rate class. This tendency was less pronounced in expN₁ (postP₁: P = 0.07, postP₂: P = 0.27) compared to expN₂ (postP_1-2: P < 0.05).

The adaptive significance of the flickering behavior in Giant Honey Bees has only been vaguely speculated [21] and has yet to be investigated in detail. Nevertheless, from years of experimental work and personal field observations, we are able to say that flickering appears to depend on some environmental and colony-intrinsic factors. We know that abdominal flipping correlates, in general, with the release of Nasonov pheromone, as observed for both Apis mellifera during fanning either at the hive or at a food source (“Sterzeln” [34,35,36,37]) and Apis dorsata during shimmering [38]. In both fanning and shimmering, the last inter-tergital gaps open up, exposing the Nasonov glands and releasing Nasonov. Nasonov is a social pheromone, common to Apis species, indicating to the colony members to come to a location identified as either the home nest entrance [34] or a source of water or food [35,39]. It may also communicate to either stay together in a reproduction swarm [36,37] or, specifically for Giant Honey Bees [38] and Apis cerana [40], to participate in shimmering.

The flickering activity was reported to decrease in number and rate with increasing ambient temperature during the day [21]. A higher percentage of flickering surface A. dorsata bees was found at 18 °C compared to 20 °C (12.5% vs. 7.8%, respectively). Moreover, the flickering rate was higher at 14 °C compared to 18 °C (0.48 Hz vs. 0.26 Hz, respectively, n = 26 A. dorsata nests). Similar results have been observed in A. laboriosa (n = 71 nests). Thus, the authors hypothesized that flickering is a response of the colony to low ambient temperature (flickering-correlates-with-low-temperature hypothesis). Furthermore, the authors speculated that the muscular activity associated with the act of flickering would heat up the bees located at the surface of the bee curtain (flickering-heats-the-body hypothesis).

In our experiments, the recorded mean ambient temperature in the preliminary non-arousal phase was 19.7 °C and 20.4 °C for the expN₁ and expN₂, respectively. During this phase, the expN₁ colony showed higher rates of flickeringF than the expN₂ colony (expN₁: relF_ts+nts = 0.3963; expN₂: 0.3367; see also Table 1 and Figure 7), supporting the flickering-correlates-with-low-temperature hypothesis. However, a more detailed analysis of the results revealed that only the ts-cohorts were more active at lower temperature (relF_ts = 0.1745 and 0.0790 for expN₁ and expN₂, respectively, P ≤ 0.05). An opposite trend was observed for the nts-cohorts (relF_nts = 0.2218 and 0.2576 for expN₁ and expN₂, respectively, P < 0.05). Hence, although we investigated only two nests, our overall results support the flickering-correlates-with-low-temperature hypothesis of Woyke et al. (2004), at least in this respect, but only for ts-cohorts. However, it is important to note that both nests had significantly (P ≤ 0.05; χ²-test) increased their flickering behavior under arousal (Table 1; Figure 7). Possibly, and this must be kept open by our findings, this effect is much stronger than the modulation by ambient temperature.

Infrared recordings convincingly showed that abdominal movements do not alter the body temperature of the agents, both during shimmering (movie 3) and flickering (movie 4) [refer to Methods section for resolution of infrared camera]. Thus, our results do not support the flickering-heats-the-body hypothesis of Woyke et al. (2004) and further indicate that surface Giant Honey Bees do not actively contribute to the thermoregulation of the nest [41,42] by flickering activities.

The flickering activities of the two experimental nests differed in some aspects. Firstly, as described above, the expN₁ showed a significantly higher spatial flickering rate than the expN₂, especially with respect to ts areas (Figure 6 and Table 1). While this aspect is in agreement with the flickering-rate-correlates-with-ambient-temperature hypothesis [21], the differences between expN₁ and expN₂ displayed in Figure 6 and Figure 7a-f cannot be explained by this hypothesis. The experiments with expN₁ and expN₂ took place at comparable times and under similar ambient temperatures. Therefore, a plausible explanation for the differences observed between both nests is that specific colony-intrinsic states resulting from, for instance, the season, may influence the flickering activity of the nests. The expN₁ monitored at the end of February, when the colony was already quite old. The colony still had a brood of bees and was performing well, with rich honey and pollen stores. Some weeks afterwards, however, the colony departed from this site. By contrast, the expN₂ was monitored at the beginning of the dry season (i.e., early November), only some weeks after the colony had arrived at this nesting site. The queen had started to lay eggs and the first young bees had hatched.

Although the present study does not provide conclusive evidence that colony-intrinsic states affect the control of flickering activity, it nonetheless documents potential characteristics of the colonies that may influence flickering activity. An aspect that can be discussed here is the obvious influence of queen pheromones and juvenile hormones [43,44,45] on the control of the aggression potential of the honey bees. Older bees have higher juvenile hormone levels [46,47,48,49,50] and are more aggressive than younger bees [51,52]. Bees with a higher juvenile hormone level exhibit a quicker response to alarm pheromones [49]. The proportion of bees acting as guards also increases with a higher juvenile hormone level [53], and it is known that the juvenile hormone titers vary seasonally in bees [52]. These aspects could eventually be considered to explain shifts in the proportion of trigger and non-trigger sites in shimmering and the enhancement of the flickering repetition rate over weeks or months in Giant Honey Bees.

In this paper, we proposed that the readiness for flickering is affected by the workload of those surface bees that are specifically skilled for generating shimmering waves. Because of their role as leaders in initiating shimmering waves [28], ts-cohorts of surface bees would respond more vigilantly under arousal by flickering compared to nts-cohorts. We supported this shimmering-drives-flickering hypothesis in two experimental nests by first identifying sites responsible for generating shimmering waves and, second, studying the flickering behavior for both cohorts (i.e., ts- and nts-cohorts) by measuring flickering rate and intensity. The flickering intensity of ts-cohorts was generally higher than that of nts-cohorts (P < 0.05; χ²-test) in both experimental nests (Figure 7h-n), confirming the shimmering-drives-flickering hypothesis. This finding has two further implications. Firstly, this finding indicates that it should theoretically be possible to predict the location of ts even before shimmering is provoked by arousal. Secondly, the finding proves a further aspect of division of labor [28,54] in Giant Honey Bees, i.e., that cohorts of specialists are positioned across the nest surface in the quiescent areas peripheral to the mouth zone in order to respond to arousal by increasing flickering rate and by initiating shimmering waves.

We also questioned whether the flickering behavior depended on colony-intrinsic properties, such as arousal history, which is addressed by the arousal-drives-flickering hypothesis. For that, we analyzed the arousal state of the colonies by quantifying the number of shimmering episodes that occurred prior to the time intervals when flickering was exclusively displayed. In these pre-arousal (preP) phases, the number of shimmering episodes was significantly lower (P < 0.001; t-test) than in post-arousal (postP_1‑2) phases and that was true for both experimental nests. This increase in flickering occurrence concerned both flickering-active ts- and nts-cohorts (Figure 7a-f), with the ts-cohorts showing higher flickering intensity than the nts-cohorts (Figure 7i-k, m-o). Under arousal, the number of both cohorts increased (Figure 6a₀–a₂, b₀‑b₂; Table 1), which documents firstly, that flickering-active cohorts of surface bees augment under arousal and, secondly, Giant Honey Bee colonies recruit generator cohorts (ts bees) under arousal as specialists to amplify the subsequent shimmering waves. Finally, recruitment of such generator cohorts requires either persistence or repetition of the threatening signal.

Flickering-active bees typically hang seemingly motionless with the exception of flickering movements. In this way, the bees produce pheromonal, visual, and mechanoceptive signals continuously for hours. Flickering activities apparently stimulate two sensory pathways in the nest mates. Firstly, the abdominal flipping movements produce mechanical cues that undergo strong spatial attenuation and are therefore only of local significance. Secondly, flickering is likely to produce chemical cues, with the last inter-tergital gaps opening up and releasing the social Nasonov pheromone during abdominal flips [38]. Pheromone clouds spread relatively slowly, but would after some time affect the nest mates over a wider range.

Thus, the signaling coverage achieved with abdominal movements is limited. In the same fashion, signalers like waggle dancers only reach a few post-dancers while fanning bees that disperse Nasonov (“Sterzeln”) are only able to send their message to those bees arriving nearby. The only way to broaden the coverage of any singularly produced signal is mass recruitment of emitters. This is what happens in flickering through mechanical, possibly visual, and pheromonal pathways. In combination, these sensory components can indeed benefit a colony by allowing the creation of a growing network of trigger cohorts. It is possible that the concept of “modulatory communication signals”, which is used by Apis mellifera for colony-intrinsic communication [45], may also be applied by Giant Honey Bees for flickering. In “modulatory communication signals”, honeybee agents are distributed over the whole hive and produce series of vibration signals over tens of minutes. These agents produce rapid dorso-ventral movements and stay in contact with recipients via the forelegs. Analogously in Giant Honey Bees, an increase in flickering activity of single agents, along with the associated mechanoceptive impulses and pheromones produced, could induce a positive feedback situation to massively recruit surface bees. An increase in intensity and occurrence of flickering may influence the behavior of previously quiescent neighbors that are, in turn, in contact with other neighbors via their legs, in particular by raising their responsiveness to shimmering.

In social groups, only a small proportion of informed individuals are required [55] to achieve accuracy in controlling the behavioral status of the whole group. The larger the group, the smaller is the proportion of informed individuals needed to guide the group [55]. Our results support this proposition since under quiescent conditions, only a minority of flickering surface bees “suffices” to maintain the colony state. However, under arousal, the colony mobilizes the collective flickering activity by increasing the number of participating agents, by enhancing their flickering rate and intensity, and by enhancing the number of agents capable of initializing shimmering waves.

Similar aspects of mass recruitment are known among ants [56,57] and termites [58], which not only possess a pheromone mass communication system, but also recruit colony mates via mechanoceptive signals [59,60]. In particular, the substrate vibrations of Messor capitatus (Formicidae) are important for communication, but are not essential components of ant recruitment behavior [61].

Thus, our findings provide new insights into the complex social behavior of Giant Honey Bee colonies and present quantitative proof that flickering and shimmering are closely related. Flickering of Giant Honey Bees has a modulating role on the generation of shimmering waves and therefore impacts on the defense behavior of Giant Honey Bees. This can be proposed because a higher arousal state of the colony provokes both higher rate and intensity of flickering and higher rate and intensity of shimmering waves.