Honey bees flexibly use two navigational memories when updating dance distance information

Honey bees are unique among all the prominent insect navigators, e.g. desert ants (Wehner, 2003), monarch butterflies (Reppert et al., 2010), desert locusts (Homberg, 2015) and Drosophila (Weir and Dickinson, 2012; Brockmann et al., 2018) in that they evolved the capability to communicate navigational information (Frisch, 1967; Dyer, 2002; Srinivasan, 2014). Honey bee foragers returning from a successful foraging trip use the waggle dance to communicate distance and direction information of the food source to their nest mates (Frisch, 1967; Dyer, 2002). A waggle dance comprises multiple waggle-runs: small-scale linear walking paths with vigorous side-to-side shaking movements of the abdomen. The duration and orientation of the waggle run correlates with the flight distance and direction of the food source from the hive, respectively (Frisch, 1967; Dyer, 2002). Dance communication is likely based on an interaction of the neural systems involved in flight and walking path integration and a transfer of the information between both systems (Brockmann and Robinson, 2007). Studies in Drosophila suggest that navigational information of flight and walking are processed in the same brain neuropils (Baker et al., 2007; Giraldo et al., 2018).

At the behavioral level a key question would be whether the information transfer between the flight and the dance (walking) navigational system is hard-wired and functionally instantaneous or whether the generation of dance information needs some time and might even involve memory processes (Menzel, 2011). The earliest evidence that memory processes might be involved in dance communication goes back to the observation of ‘marathon dances’ of nest-site scouts, during which bees danced for several to many hours without leaving the hive (Lindauer, 1954; Frisch, 1967). More specific information on memory processes were obtained by experiments in which honey bee foragers were first trained to a feeder and then moved to an unfamiliar area on a cloudy day (Dyer and Gould, 1981; Dyer, 1987; Towne and Moscrip, 2008). Under overcast conditions, when bees could not perceive the sun compass information, the foragers indicated the direction of the old feeder in their dances. When the clear sky and sun was visible, they updated the dance direction and indicated the location of the feeder they were actually visiting. Based on these findings, Dyer (Dyer and Gould, 1981; Dyer, 1987) proposed that dance behavior involves at least two memories, one regarding the most recent foraging trip and another one comprising previous experiences. These navigational memories can comprise information from the path integration system, snapshots of the visual panorama and celestial cues (Collett and Collett, 2000; Towne and Moscrip, 2008; Collett et al., 2013; Towne et al., 2017; Heinze et al., 2018).

Here, we were interested in how foragers update waggle dance duration for a new feeder distance. We asked two questions: (i) how many foraging trips do honey bee foragers need to update their waggle dance duration after a linear shift of the feeder? (ii) do foragers use information only from the most recent foraging trip or do they also incorporate information from earlier flight experiences to generate waggle dance duration? We performed two different types of experiments: single feeder shift experiments and double feeder shift experiments. In the single-shift experiments the bees initially foraged at a feeder distance of 300 m. Then, the feeder was directly moved either forward to 400 m or backward to 200 m. For those foragers that found the new feeder position, we determined the number of foraging trips and the time the bees needed to show a significant change in waggle dance duration, indicating the completion of the waggle dance duration update process. In the double-shift experiments, we did two feeder shifts in opposite directions: either backward–forward or forward–backward. The question was whether foraging at the intermittent feeder distance would have an effect on the dance duration when the foragers revisited the first feeder location.

Experiments were performed on the campuses of the National Centre for Biological Sciences (NCBS-TIFR) and University of Agricultural Sciences (GKVK) Bangalore, India (Fig. S1).

Apis mellifera Linnaeus 1758 honey bee colonies (approximately 4000 bees/colony; n=7) were obtained from a local beekeeper (HoneyDay Bee Farms Pvt. Ltd., Bangalore) and placed in a three-frame observation hive. The observation hive was kept inside a small hut (Frisch, 1967).

Bees were initially trained to a feeder 3 m from the hive entrance offering an unscented sugar solution (1–1.5 mol l⁻¹) (von Frisch, 1967). The feeder, a plate on a small stand, was placed under a multi-colored umbrella to help the bees to find it (Fig. S1). The concentration of the sugar solution was adjusted to attract a high number of foragers. When 15–30 foragers were regularly visiting the feeder, the feeder was moved in small steps further away from the hive (about 5 m every 10 min). The distance training was conducted along a road with dense vegetation on both sides (Fig. S1). All bees coming to the feeder were individually marked (Posca 5M, ‘Uni’ Mitsubishi Pencil, India) at a feeder distance of 100 m. The marked foragers were further trained to 300 m over 2 days.

Every single-shift experiment started with 10–20 marked bees foraging at 300 m for 1–2 h. The feeder was then shifted (i) further away in the same hive-to-feeder direction, to 400 m (forward shift, FS, Fig. 1A) or (ii) towards the hive in the reverse direction, to 200 m (backward shift, BS, Fig. 1B). The feeder was kept at the new distance for another 1.5–2 h. No foragers were present on or around the feeder during the shift. All marked bees had made a minimum of 5 foraging trips and 3 dances before the shift. The sugar concentration of the feeder was increased by 0.125 mol l⁻¹ prior to the shift, to increase the probability that the foragers would continue dancing after finding the feeder. To determine how many foraging trips a forager needs to update the waggle dance duration, it was necessary that the bees continued dancing (von Frisch, 1967).

A series of pilot experiments were done to determine a suitable feeder shift distance such that most of the foragers showed a significant change in waggle dance duration. We tested four shift distances: 25 m, 50 m, 75 m and 100 m, both forward and backward. A few bees (3 out of 17) showed a significant change in waggle dance duration when the feeder was shifted only 25 m (Fig. S2A). However, a feeder shift of 100 m was necessary to detect a significant change in waggle dance duration for more than 95% of the bees tested (Fig. S2A). Thus, a shift distance of 100 m was chosen for all further experiments. Experiments were performed between June 2014 and July 2018.

Two types of double-shift experiments were done: (i) backward–forward (BF): 300 m–200 m–300 m and (ii) forward–backward (FB): 300 m–400 m–300 m. In each experiment, 10–20 marked bees foraged at 300 m for 1–2 h, following which the first feeder shift was made (similar to single-shift experiments). The feeder was moved twice, first either backward to 200 m or forward to 400 m (first shift: S1), and then back to 300 m (second shift: S2). Thus, the foragers visited the same distance (300 m) twice during the course of the experiment. The feeder was presented for 1.5–2 h following each feeder shift. The double-shift experiments were conducted in June–July 2015 and repeated in June 2017. The sugar concentration of the feeder was increased by 0.125 mol l⁻¹, 10 min prior to each feeder shift only for the 2015 experiments.

Three additional sets of double-shift experiments were done: (i) a second forward–backward experiment at different feeder distances (FB2: 200 m–300 m–200 m), (ii) a forward–forward experiment (FF: 200 m–300 m–400 m), and (iii) a backward–backward experiment (BB: 400 m–300 m–200 m). In FB2 and FF, bees were trained to 200 m as described above. On the day of the experiment, individually marked foragers first visited the feeder at 200 m for 1.5–2 h, following which the feeder was moved instantaneously to a new location 300 m from the hive, in the same hive to feeder direction. In FB2, the feeder was moved back to 200 m in the second shift, whereas in FF, the feeder was moved further away to 400 m. The feeder was presented at each distance for 1.5–2 h. In BB, bees were initially trained to 400 m over 2 days. All bees coming to the feeder during the training period were marked differently than those newly recruited at 400 m. Before the experiment, foragers coming to the feeder during the training phase were captured. This procedure ensured that the foragers in this experiment had not visited any other feeder locations between the hive and 400 m. The FB2 experiment was done in November 2017. FF and BB experiments were done in June–July 2018. The sugar concentration was kept unchanged during each experiment.

In all experiments, the hive entrance was opened only when the feeder was presented. This ensured that the marked bees foraged only at our feeder and no other food source. Unmarked bees coming to the feeder during the experiment were caught, kept on ice, and released back into the colony after the experiment. Foragers tested in the feeder shift were captured and removed from the hive after each experiment. This helped us to avoid the reuse of the same bees in subsequent experiments. Two experimenters were always present at the feeder, recording the arrival time of each bee. Another experimenter was present at the hive, video recording the dances. Experimenters at the feeder and the hive maintained continuous communication using mobile phones, tracking feeder and hive activities of each individual forager.

All dances were recorded at 50 frames per second with a video camera (Sony HDR-CX220V Tokyo, Japan). The camera was mounted on a tripod facing one side of the observation hive, recording the entire bottom comb. A temporary barrier was made at the entrance of the hive to restrict the traffic of the bees to the side facing the camera. This ensured that no dances were performed on the other side of the comb (Tautz et al., 2004; Gilley, 2014). Video recordings were made under diffused white LED illumination (Philips LED Chargeable Light-30505, Gurgaon, India) (von Frisch, 1967). The light did not have any observable effect on the bees, e.g. the bees were not agitated nor did they form curtains on the glass. Recorded videos were viewed using Virtual Dub 1.10.4 (http://www.virtualdub.org/) and waggle run durations were manually analyzed by counting the number of frames between the first and the last waggle movements of the dancer's abdomen (Couvillon et al., 2012). Dance durations are represented as the mean (±s.d.) of all waggle run durations performed during a dance. The timings of feeder and hive visits for each forager were correlated to identify consecutive foraging and dancing events. The time of entry and exit of foragers from the dance floor were obtained from the videos.

Search flights were identified as unsuccessful foraging trips, just before or after the shift, in which the forager was not observed visiting the feeder. Search time was defined as the time taken by a forager from initiating the first search flight until she discovered the feeder at the new distance (Fig. S2B). It included flight time and search time as well as time spent within the hive. The time interval between the first visit at the new feeder distance and the change point in waggle dance duration was considered as updating time (Fig. S2C).

Statistical analyses were carried out using R (https://www.r-project.org/) v.3.3.1 using packages ‘lme4’ (Bates et al., 2015), ‘multcomp’ (Hothorn et al., 2008) and ‘PMCMR’ (https://CRAN.R-project.org/package=PMCMR).

A total of 1937 dances (38,200 waggle runs) from 84 bees were included in the study. All dances before and after the shift from the single-shift experiments were used in the analysis. Relevant data from the double-shift experiment, i.e. duration of dances before and after the first shift, but not those after the second shift, were also included. We fitted a linear mixed-effects model (LMM) and compared the difference in waggle run durations before and after the feeder shift for individual bees by using a generalized linear hypothesis test (GLHT). The model used the waggle run duration as the response and the interaction between the individual bee and the shift (categorical variable of two levels: pre and post) as the predictor. The waggle run number nested within waggle dance number, experiment type (a categorical variable of two levels: forward and backward) nested within colony identity and feeder distances from the hive (categorical variable of seven levels: 150 m, 175 m, 200 m, 225 m, 275 m, 300 m, 400 m) were included as random effects. P-values were corrected for multiple testing using single step adjustment.

For individual bees, the change in the waggle dance duration (change point: CPT) following the feeder shift was estimated based on CUSUM change point detection method (www.variation.com/product/change-point-analyzer/; Morreale et al., 2015; Koepcke et al., 2016) in R (Table S1). For a series of dances each with duration d_i, we define a new variable X_i, where X₁=0, and X_i+1=d_i−mean(d_i). Then we consider the cumulative summation of X i.e. ⁠. The change point is defined as the value of i such that the absolute value of S is maximum, i.e. CPT=i, such that |S_i|≥|S_j| ∀ j=1,n. Then, we performed a bootstrap analysis with a test statistic defined as S=max(S_i)−min(S_i). We permuted the series of dances, NN=10,000 times, and calculated the test-statistic(s) for each permutation. Next, we counted the number of instances (NS) the test-statistic was more than the test-statistic of the series of dances observed. A P-value of significance for the change point was calculated by dividing NS with NN (P=NS/NN). This analysis was repeated for each individual separately. The change point was considered significant for P<0.05. For single-shift experiments, series of all dances for an individual before and after the shift were considered for the change point analysis. For the double-shift experiment, we divided the series of dances into two halves: (i) before and after the first shift but before the second shift, and (ii) after the first shift and after the second shift. We determined the change points for these two segments separately (CPT-I and CPT-II, Table S1). We used the chi-square test to find if the direction of the shift (forward or backward) had any effect on the type of update in dance duration.

We fitted each individual's dance data to these two curves using a non-linear model and then compared the AIC (Akaike information criterion) values of the two models to determine the better fit. The model with a lower AIC value represents a better fit. We used a ΔAIC (=AIC of the gradual model−AIC of the abrupt model) cut-off of 4 to determine the better fitting model (Burnham and Anderson, 2004). For −4<ΔAIC<4, we could not conclude whether the bee showed a gradual or abrupt update in dance duration as both models were equally good at explaining the change in consecutive waggle dance durations (Table S1).

For each bee which showed a delayed change point, the series of consecutive dances was divided into three phases: dances before the feeder shift (pre-shift phase: PS), dances after the shift and before the calculated change point (intermediary phase: IP), and dances from the change point until the end of the experiment (post-change point phase: PC) or until the next shift (in case of individuals from double-shift experiments). We first determined the difference in dance duration between consecutive waggle dances: (ΔDD=d_i+1−d_i) for dances in the pre-shift (PS) and post-change point (PC) phases, where d_i is the average dance duration of the ith dance in the series. We calculated the mean and standard deviation of these differences. We assumed that a normal distribution with this mean and standard deviation represents the baseline variation in dance duration amongst dances without a change in distance.

To test for gradual or abrupt changes in dance duration after a feeder shift, we compared the values of differences between consecutive dances (ΔDD) in the intermediary phase (IP) with the normal distribution of baseline variation that we determined. If the values of differences (ΔDD) in the intermediary phase (IP) were within the ±2.567 standard deviation range of the baseline variation (representing 99% of the distribution), we classified the change in dance duration as gradual. Otherwise, if even one of the values of the intermediary phase (IP) was beyond the ±2.567 standard deviation range, we classified it as an abrupt change (Table S1).

We focused on differentiation of gradual and abrupt change because these two types of update would imply distinct memory processes. Certainly, foragers showing gradual changes would vary in the rate of change (Fig. S3B-T), reflecting inter-individual differences in the time dynamics of the update process.

We did a z-score analysis to check if there was a difference in the number of foragers which made search flights in the FS as compared to the BS experiments. Further, we tested whether the search time correlated with the updating time using the Pearson's correlation coefficient.

We fitted a second LMM followed by a GLHT for BF, FB and FB2 experiments (22 bees, 578 dances and 11,440 waggle runs) to find if there was a significant difference in waggle run duration for individual bees following each shift as well as at the same distance (300 m for BF and FB, 200 m for FB2). The model used the waggle run duration as the response and an interaction between the individual bee and the shift (categorical variable of three levels: pre-shift, after first shift and after second shift) as the predictor. Waggle run number nested within waggle dance number, experiment type (categorical variable of three levels: BF, FB and FB2) nested within colony identity and feeder distances from the hive [categorical variable of five levels: 300 m, 400 m, 200 m, 300 m(2), 200 m(2)] were included as random effects. The model estimated the difference in waggle run duration for (i) S1, (ii) S2 and (iii) the same distance (Δ300 m and Δ200 m). Furthermore, we calculated the recovery ratio: the ratio of estimated change in waggle run duration for an individual in S2 to S1. Ideally, the recovery ratio should be close to −1, as foragers are expected to show a similar value of change in dance duration for the two 100 m shifts (S1 and S2), but in opposite directions. The recovery ratio was between −1 and 0 if the change in S2 was less than S1. If the ratio was <−1, the change in S2 was more than in S1. Finally, a positive ratio implied that the change in S2 and S1 was in the same direction (i.e. the forager further increased her mean waggle dance duration as compared to 400 m after S2 in FB or further decreased it as compared to 200 m after S2 in BF).

We fitted a third LMM (and GLHT) for the FF and BB experiments (6 bees, 178 dances and 3177 waggle runs, feeder distances: 200 m, 300 m and 400 m). All marked bees in BB that visited the feeder at 400 m and 300 m were further grouped as BS2 (400 m–300 m). We performed a Mann–Whitney test to compare the change in waggle duration for 400 m–300 m between individuals of the FB (change in S2) and BS2 groups.

P-values in the two LMM (and GLHT) analyses were corrected for multiple testing using single-step adjustment. All data collected during the experiment are available in the Dryad data repository (Chatterjee et al., 2019; doi:10.5061/dryad.q99m547).

We trained 165 foragers to a feeder distance of 300 m; out of these, 112 foragers were subjected to a 100 m forward shift (FS) and 53 to a 100 m backward shift (BS). Fifty-two individuals (FS: n=24, 21%; BS: n=28, 53%) found the new feeder location after the shift and 35 of them (FS: n=16, BS: n=19) performed dances. Sixteen out of these 35 foragers (FS: n=7, BS: n=9) continued dancing (CD) without pausing (Fig. 1C,D and Fig. 2A), whereas 19 foragers (FS: n=9, BS: n=10) first stopped and then resumed dancing (SD) later (Fig. 1E,F and 2A). For each individual of both groups (CD and SD), we determined the first occurrence of a significant change in waggle dance duration (change point, Fig. 1C–F). For all but one individual (CD), we could identify a significant change point in the post-shift dances (Table 1 and Tables S1,S2).

Among the 15 foragers of the CD group, 3 individuals immediately showed a change point in the first post-shift dance (immediate update: I, Fig. 1C and Fig. 2A), whereas the other 12 needed two or more foraging trips to update dance duration (delayed update: D, Fig. 1D and Fig. 2A). In the SD group, 12 out of 19 foragers showed a change point in their first post-shift dance (Fig. 1E) and 7 showed a delayed update (Fig. 1F, Table 1). The proportion of bees showing an immediate change point was significantly higher in the SD than it was in the CD group (χ²-test: 63% vs 20%, χ²=3.84, d.f.=1, P=0.03, Fig. 2A). For the individuals that continued dancing or showed a delayed update (CD+I, CD+D and SD+D, n=22), we could determine the foraging trip in which they updated their waggle dance duration. The majority (86%; FS: 9, BS: 10) needed more than one foraging trip to update their waggle dance duration for the post-shift feeder distance and more than half of them (59%) required two or more foraging trips (Fig. 2B).

For the group of foragers showing a delayed update in waggle dance duration (CD+D and SD+D, n=19), we tested whether the update occurred gradually or abruptly. In a first analysis, we tested whether the change in waggle dance duration showed a better fit with a sigmoidal curve (Eqn 1) or a step function (Eqn 2, Fig. S3A). We identified a gradual change in waggle dance duration for 9 individuals (ΔAIC<−4, FS: n=4, BS: n=5, Fig. 3A) and abrupt change for 3 individuals (ΔAIC>4, FS: n=2, BS: n=1, Fig. 3A). For 7 foragers, the AIC values for both models were too close to distinguish them as showing either a gradual or abrupt change (−4<ΔAIC<4, FS: n=3, BS: n=4, Fig. 3A, Table 2 and Table S1; waggle duration for individual bees given in Fig. S3B–T).

In a second analysis, we determined whether there were changes in dance duration (ΔDD) during the intermediate phase (IP) that exceeded the average variation in dance duration during the pre-shift phase (PS) and the post-change point phase (PC). Similar changes in ΔDD during the IP compared with the PS and PC would indicate a gradual change, whereas a substantial difference (≥2.576 s.d.) in dance duration during IP would indicate an abrupt change. Fifteen out of 19 foragers showed a gradual change (FS: n=7, BS: n=8, Fig. 3B, Table 2) in waggle dance duration and 4 foragers (FS: n=2, BS: n=2; Fig. 3B) showed an abrupt change. Combining both analyses, we obtained the same classification for 10 out of the 19 foragers (gradual: n=8, abrupt: n=2, Table 2 and Table S1).

As mentioned above, 52 foragers out of 165 found the new feeder location after the feeder shift within the time limits of the experiment. Seventeen foragers (FS: n=5, BS: n=12) either stopped dancing but continued to forage or stopped foraging altogether soon after having found the feeder. For the 34 foragers (FS: n=15, BS: n=19) that danced and updated dance duration at the new feeder distance, we determined the number of search flights, search time and updating time. About half of the foragers (FS: n=5, BS: n=14 bees) found the new feeder position on their first outbound flight and thus did not make any additional search flights (Fig. 4A,B). The foragers (FS: n=10, BS: n=5) that did not find the new feeder location on their first flight returned to the hive and later started a new search flight (Fig. 4A,B). The average duration of these search flights was significantly longer than that of the foraging flights (Fig. S2D,E). More foragers visited the feeder in the first outbound flight after the backward feeder shift than after the forward one (FS: 33.3%, BS: 73.7%; z-score=−2.353, P=0.019). Four foragers (FS: n=2, BS: n=2) showed additional search flights after having found the new feeder location (Fig. 4A,B). Individuals took a maximum of 3–4 search flights to discover the new feeder location. We did not find any significant correlation between the search time and updating time (Pearson r=−0.1545, r²=0.02387, P=0.383, Fig. 4C). The flowchart diagram (Fig. S2F) summaries all possible behavioral responses after the feeder shift.

In a second set of experiments, we tested how honey bee foragers would update their dance duration when confronted with two feeder shifts in opposite directions with an interval of 1–2 h between the shifts. We initially trained 157 bees to a distance of 300 m and then we performed a ‘backward–forward’ (BF: 300 m–200 m–300 m, n=39, Fig. 5A) or a ‘forward–backward’ feeder shift (FB: 300 m–400 m–300 m, n=118 Fig. 5B). Only 20 bees performed dances after both feeder shifts (BF: n=11, FB: n=9).

In the BF experiment, all 11 foragers updated their waggle dance duration after both feeder shifts (Fig. 5C, Fig. 6A, Fig. S2G; linear mixed-effects model and generalized linear hypothesis test, Table S3). Four foragers showed no difference in waggle dance durations between the first and second visit at the 300 m feeder distance (Fig. 6A,C; recovery ratio: −0.85 to −1.44) whereas the rest showed a partial difference (Fig. 6B,C; recovery ratio: −0.39 to −0.77).

In the FB experiment, 8 foragers updated their waggle duration after the forward shift but 6 did not show any change in waggle dance duration after the backward shift (Fig. 5D, Fig. 6B, Fig. S2H; linear mixed-effects model and generalized linear hypothesis test, Table S3). We did not identify any change point after the second shift (change point analysis, Fig. 5D, Table S3). All foragers (n=9, Fig. 6B) showed significantly longer dance durations during their re-visit compared with their first visit at 300 m (recovery ratio: −0.47 to 0.47, Fig. 6C).

Similarly to FB experiments, individuals (n=2) in the FB2 experiment did not change waggle dance duration following the second shift (Fig. 7A,B; linear mixed-effects model and generalized linear hypothesis test, Table S3). In the FF experiment (n=3) and BB experiment (n=3), foragers showed significant change in waggle dance duration following each feeder shift (Fig. 7C–F, linear mixed-effects model and generalized linear hypothesis test, Table S3). Recruited foragers (n=7), showed a stronger change in waggle dance duration for the BS2 experiment compared with the foragers in the FB experiment (Fig. 7G, Mann–Whitney test: P<0.0001).

The principal findings of our study are: (i) most honey bees take time to generate a new waggle dance duration after foraging at a new feeder distance, indicating that the transposition of flight information to dance information is not a reflexive process, and (ii) the generation of a new waggle dance duration involves two memories, one from the most recent foraging experience and one from previous flight experiences. It appears that foragers can use both memories, separately or jointly, to generate dance duration information (Fig. 8A).

The single-shift experiments showed that the majority of honey bee foragers (86%) did not instantaneously update waggle dance duration. They needed more than one foraging trip to the new feeder to generate corresponding waggle dance duration. Experiments on path integration have shown that central place foragers constantly monitor their movements and calculate their position with respect to their nest (Wehner et al., 1996; Wehner and Srinivasan, 2003; Freas et al., 2019). Thus, the forager instantaneously knows the new feeder distance that she will use for the homeward (=inbound) vector. We propose that the updating time includes two processes: (a) updating navigational memory about the location(s) of rewarding food sources, and (b) generating dance information. The occurrence of intermediate dances in our experiments indicated that the memories of the new and old feeder locations interacted dynamically to generate post-shift dance durations. These two navigational memories might affect not only dance behavior but also foraging flight behavior. For example, the outbound flight to the new feeder location might be more convoluted than the inbound flight; the foragers could still check the previous feeder location on their way to the new feeder location. Experiments combining dance recording and harmonic radar tracking would help to verify whether updating dance information and updating navigational vector memory are the same or separate processes (Degen et al., 2015, 2018). Brockmann and Robinson (2007) previously proposed that the neural circuits underlying dance communication might be an elaboration of an already existing navigational memory system guiding foraging flights.

There was no correlation between updating time and search time, suggesting that both processes are not directly linked. Foragers in FS and BS experiments had similar search time and updating time (see Materials and Methods; Fig. S2B,C). In the double-shift experiments the foragers almost immediately found the previously known feeder position (300 m) but still needed time to update waggle dance duration (BF) or even did not update at all (FB, Table S1). Finally, search strategy and persistence depend on the colony food stores and the reward of the feeder (Townsend-Mehler et al., 2011). Thus, search time depends on the motivational state in addition to cognitive processes involved in spatial search behavior.

Our double-shift experiments tested whether previous foraging experience would affect dance duration. The foragers in the BF experiment updated dance information after both feeder shifts. Surprisingly, foragers in the FB and FB2 experiments individuals did not update their waggle duration after the backward shift. Foragers in the FF and BB experiments showed significant change in waggle dance duration following each feeder shift. Interestingly, naive foragers, recruited at the feeder distance of 400 m (BS2), showed a significant change in waggle dance duration after a backward shift to 300 m (Table S2). BS2 foragers had no prior experience to any other feeder locations on the training road. Both the BS2 foragers and FB foragers in the second shift (S2) experienced a change in feeder distance from 400 m to 300 m. But the change in waggle duration for 400 m–300 m was significantly greater for BS2 foragers than in FB foragers. Taken together, these experimental results suggest that distance, shift direction as well as number of feeder shifts did not have an effect on the update of waggle dance duration. Only the specific sequence of forward–backward (FB and FB2) shift interfered with the updating process.

Our double-shift experiments correspond to previous visual learning experimental paradigm in honey bees (Cheng and Wignall, 2006) and ants (Freas et al., 2017). These studies showed that the learning of new information negatively affects the ability of individuals to retrieve previous information, a phenomenon referred to as retroactive interference (Cheng and Wignall, 2006; Reaume et al., 2011; Freas et al., 2017). One mechanism underlying retroactive interference is response competition, in which the individual is uncertain about which of the two memories, e.g. previously stored and newly acquired, to utilize (Cheng and Wignall, 2006). In our FB experiments, the second feeder distance (400 m) was novel, since the foragers never visited it during training. In contrast, the second feeder distance (200 m) in BF was less novel because foragers were previously trained to it. Novelty likely increased the response competition between the newly acquired (second feeder location) and previously stored (first feeder location) memories, leading to retroactive interference in FB but not in BF experiments. This could be why foragers did not update waggle dance duration following the second shift in FB experiments.

Dyer (Dyer and Gould, 1981; Dyer, 1987) suggested that dance information can be generated by two temporarily separated memories, a memory of ‘newly acquired’ navigational information from the most recent foraging trip and a memory of ‘previously stored’ navigational information from earlier flight experiences (Fig. 8A). Both memories are to some extent functionally independent and can be used separately to generate dance orientation. Our findings regarding updating distance information (Fig. 8B) and Dyer's findings (see also Towne and Moscrip, 2008) correspond well and together suggest that short- and long-term memory processes affect generation of dance information.

Radar tracking experiments (Menzel et al., 2011; Menzel, 2011) demonstrated that honey bee dance followers reflexibly use the new information obtained from dance. Individual foragers continued to visit the old feeder (=private information from previous foraging experience), visited the new feeder (=newly acquired social information by following dances), or performed an intermediate flight trajectory resulting from an interaction of the two items of spatial information. Similarly, our single feeder shift experiments indicate that dancers do not reflexively communicate the most-recent flight navigational information. After finding the new feeder location, individual foragers either danced for the old, new or an intermediate feeder distance. We propose that the behavioral responses of dance followers and dancers are shaped by individual differences in memory processes, e.g. weighing of memory contents. For dancers, this is indicated by the variation in the number of foraging trips taken to update the dance duration (Fig. 2B) and the recovery ratio (Fig. 6C). The variation in both parameters demonstrates the spectrum of possible interactions between spatial memory contents which then dictates what a forager communicates in her dance.

Dacke and Srinivasan (2008; see also Heinze et al., 2018) demonstrated that honey bee foragers showed two different navigational outputs: one that guided them on their foraging trips (personal navigation odometer) and the other that they used for communication of distance information (community dance odometer). We propose that the two different navigational behaviors can be explained by the temporal dynamics of updating navigational memory (i.e. previously stored and recently acquired; Fig. 8B). In the Dacke and Srinivasan (2008) experiments, honey bees were trained to the tunnel for 1 day and then on the test day some parts of the tunnel were covered. With respect to the temporal memory model, we suggest that the foragers used their previously stored distance information for dance communication and the newly acquired distance information for their own flight navigation. In general, one can assume that the previously stored spatial memory represents a more confirmed knowledge than the newly acquired one, whereas the latter represents the actual (new) food location.

There has been extensive progress in understanding the neural underpinnings of path integration (Sauman et al., 2005; Heinze and Homberg, 2007; Stone et al., 2017), yet studies trying to identify brain processes involved in dance behavior have failed so far. The two major experimental problems are that dance behavior can only be studied under natural conditions and that it was not possible to dissociate foraging and dance behavior (Sen Sarma et al., 2010; Kiya and Kubo, 2011). The feeder shift experimental paradigm offers a time period during which updating navigational memory and generating dance information is dissociated from foraging behavior. Thus, this experimental paradigm provides the opportunity to identify molecular mechanisms specific to memory processes involved in navigation and generation of corresponding dance information.

We thank B. Krishnan, S. K. Sethy, A. Sengupta, A. Suryanarayanan, S. Chakraborty, D. Chowdhury, A. Chakrabarty, D. Bais, S. Unnikrishnan, N. Thulasi, A. Johny, R. Fatima, and A. Dey for their help with the behavioral experiments. We thank S. Unnikrishnan for her input on the manuscript. We thank the Department of Apiculture UAS-GKVK, Bangalore for permission to use their campus.