Orientation to polarized light in tethered flying honeybees

As a result of sunlight scattering in the atmosphere, the skylight is partially plane polarized and the celestial e-vectors are arranged in a concentric pattern around the sun (Strutt, 1871; Wehner, 1997). It is well known that many insects exploit this skylight polarization for visual compass orientation and/or navigation (for review see Wehner, 1994; Wehner and Labhart, 2006; Heinze, 2014). There have been an enormous number of studies about insect polarization vision, not only at the behavioral level (e.g. Dacke et al., 2003; Reppert et al., 2004; Henze and Labhart, 2007; Weir and Dickinson, 2012), but also at the neural network level, such as sensory (e.g. Blum and Labhart, 2000; Weir et al., 2016) and central brain (e.g. Labhart, 1988; Heinze and Homberg, 2007, 2009; Sakura et al., 2008; Heinze and Reppert, 2011; Bech et al., 2014) mechanisms. The e-vector detection in insects is mediated by a group of specialized ommatidia located in the most dorsal part of the compound eye, the dorsal rim area (DRA), in which the photoreceptors are monochromatic and highly polarization sensitive (for review see Labhart and Meyer, 1999; Wehner and Labhart, 2006). The neural pathway of polarization vision in the brain has been documented in several species. The photoreceptors in the DRA terminate in the lamina or the medulla in the optic lobe, and, from there, polarized light signals primarily project into the central complex through a pathway involving the lower unit of the anterior optic tubercle and lower division of the central body (Homberg, 2008). The central complex, one of the higher centers of the insect brain, is considered to be the location of an internal compass (for review see Homberg et al., 2011; Heinze, 2017), although it is still unclear how the central complex controls the animal's steering during navigation.

Foraging behavior in social insects, such as ants and bees, is a useful model system for studying insect navigation because they repeatedly go back and forth between the nest and a feeding site. In particular, the path integration mechanisms in desert ants (Cataglyphis) have been extensively studied in regard to insect navigation (Wehner, 2003; Collett and Cardé, 2014), and Cataglyphis are well known to choose their heading direction using celestial polarization cues during long-distance navigation (Fent, 1986; Wehner, 1997; Wehner and Müller, 2006). In addition to path integration based on the polarization compass, ants can learn visual landmarks or panoramic views at familiar locations and use them for local navigation (Collett et al., 1992; Wehner et al., 1996; Collett et al., 1998; Graham and Cheng, 2009; Narendra et al., 2013). Honeybees also undertake long-distance foraging trips that may reach over 5 km (Couvillon et al., 2014). Usage of skylight polarization in honeybees to detect their intended travel direction was first described by von Frisch (1967) through a series of sophisticated behavioral studies on the waggle dance. Thereafter, the waggle dance orientations of the nest-returning bees from a certain feeder have been intensively studied. These studies were conducted under a patch of polarized light stimulus or part of the sky, and an internal representation of the celestial e-vector map has been proposed (Rossel and Wehner, 1982, 1986, 1987; Wehner, 1997). These systematic studies have focused on modification of the waggle dance orientation and not on how the bees perceive polarized light from the sky en route to/from the nest. More recently, polarized light detection in flying bees has been demonstrated using a four-armed tunnel maze with a polarizer on top (Kraft et al., 2011), and it was revealed that bees choose the arms based on their previous e-vector experiences. Moreover, it has also been demonstrated that bees memorize the e-vector orientations experienced during their foraging flight and use that memory for subsequent waggle dances (Evangelista et al., 2014). In these studies, the tunnel was made as the bee could receive parallel or vertical e-vector stimulus with respect to their moving direction. Indeed, by classical conditioning experiments, it has been shown that the honeybees are able to discriminate polarized light of 0 deg e-vector orientation with that of 90 deg (Sakura et al., 2012).

Behavioral responses of moving insects to the overhead polarized light stimulus have been intensively studied using a tethered animal. Orientation to the polarized light, i.e. polarotaxis (Mathejczyk and Wernet, 2019), has been demonstrated in tethered walking insects (cricket, Gryllus campestris; Brunner and Labhart, 1987; fly, Musca domestica; von Philipsborn and Labhart, 1990) and also in tethered flying insects. In the locust (Schistocerca gregaria), direct monitoring of yaw-torque responses showed clear polarotactic right-and-left turns to rotating polarized light (Mappes and Homberg, 2004). In tethered monarch butterflies (Danaus plexippus), measuring flight orientations using an optical encoder revealed that their flight orientation under natural skylight was clearly affected by a dorsally presented polarization filter (Reppert et al., 2004; Stalleicken et al., 2005). Similar orientation to polarized skylight has also been demonstrated in Drosophila (Weir and Dickinson, 2012; Mathejczyk and Wernet, 2019), in which a fly was magnetically tethered in the arena and its flight heading was recorded from above by an infrared camera. Recently, some behavioral studies have succeeded to make a dorsally tethered bee stably fly by presenting lateral optic flow and frontal air-flow stimuli (Luu et al., 2011; Taylor et al., 2013). It was observed that the tethered bees showed ‘streamlining’ responses, whereby they raised their abdomen in a correlated manner with the speed of the optic and air-flow stimuli. In the present study, we investigated how flying bees respond to polarized light stimuli using a tethering system. We constructed a flight simulator, in which we could examine a tethered bee's flight response to a rotating polarized stimulus, and found that they tended to orient themselves to a certain e-vector direction, i.e. they exhibited clear polarotaxis, during the flight.

The honeybees, Apis mellifera ligustica Spinola 1806, used in this study were reared in normal ten-frame hives on the campus of Kobe University, Hyogo, Japan. Forager honeybees with pollen loads were collected at the hive entrance before the experiment and anesthetized on ice or in a refrigerator. An L-shaped metal rod for tethering was attached to the pronotum of an anesthetized bee, as previously described (Luu et al., 2011). Briefly, the hair on the pronotum was gently shaved using a small piece of a razor blade, and the metal rod was adhered using a small amount of light-curing adhesive (Loctite; Henkel, Dusseldorf, Germany). Image analyses of bee behavior (see below) were conducted by marking the tip of the abdomen with a white, light-curing dental sealant (Conseal f; SDI Limited, Bayswater, Australia). Next, the bees were placed in a warm room to recover from anesthesia and fed several drops of 30% sucrose solution before the experiment.

We did not purposely remove pollen loads of the bees but some of them lost the pollen loads during the preparation processes. Therefore, bees with and without pollen loads were evaluated during the experiment.

The experiments were performed using a custom-made black box (Fig. 1) in a dark room. A tethered bee was mounted in the box by attaching the end of the metal rod to a three-dimensional manipulator, such that the bee's location could be adjusted manually. The flying behavior of the tethered bee was enhanced by stimulating the bee with a headwind from an air circulator and front-to-back optic flow from an LCD monitor. The circulator was located outside the box and connected to a tunnel that carried the wind stimulus into the box. The end of the tunnel (diameter, 8 cm), consisting of many fine plastic straws to reduce the turbulent flow of wind, was fixed at 10 cm from the bee's head. The wind speed at the bee's head was almost constant, ranging between 1.7 m s⁻¹ and 2.0 m s⁻¹. The LCD monitor (RDT1711LM; Mitsubishi Electric, Tokyo, Japan; 75 Hz refresh rate), covered with a sheet of tracing paper to eliminate any polarized components of the light, was located 5 cm beneath the tethered bee. The front-to-back optic flow stimulus of moving black-and-white stripes (Michelson contrast, 0.25) was displayed on the monitor using a self-made program in Microsoft Visual C⁺⁺. The size of the area on the monitor displaying the stripes was 159×163 deg, and the width of a single stripe was 40 deg, measured from the bee's head position. The speed of the stimulus was ∼900 deg s⁻¹, as seen by the bee, which is fast enough to elicit the highest streamlining responses from a bee (Luu et al., 2011).

Light from a xenon lamp (LC8, L8253; Hamamatsu Photonics, Hamamatsu, Japan) was applied above the bee using a quartz light guide. The light was filtered using a depolarizer (DPU-25; ThorLabs, Newton, NJ, USA) at the end of the light guide to eliminate any polarized components of the light, and a holographic diffuser (48-522; Edmund Optics, Barrington, NJ, USA) was clamped under the end of the light guide. The diffuser reduced illuminance irregularity and increased the size of the light fit around a linear polarizer (HN42HE; diameter, 15 cm; Polaroid Company, Cambridge, MA, USA) beneath the diffuser. The polarizer was mounted on a circular holder that could be rotated using a DC motor. The stimulus was centered at the bee's zenith (with respect to flying head position) at a distance of 15 cm, providing a dorsal, polarized stimulus of 53 deg in diameter. In the experiments in which unpolarized light stimulus was used, the depolarizer was clamped just above the bee's head instead of at the end of the light guide, such that the size of the light stimulus covered the entire receptive field of the bee's DRA. Under this condition, we could assess the effect of a slight fluctuation in light intensity caused by the polarizer rotation with the same spectrum of light (300–620 nm). The intensity of the polarized and unpolarized light at the animal level was ∼1000 lx.

The behavioral experiments were performed between 11:00 h and 18:00 h local time. A bee was fixed in the experimental box with the metal rod attached for tethering after complete recovery from anesthesia. First, we let the bee hold a small piece of paper so that it could not start flying. The e-vector angle of the polarizer was set at 0 deg with respect to the bee's body axis, and static black-and-white stripes were displayed on the PC monitor. After the bee had been familiarized with the box, the paper was removed to allow the bee to start flying, and the wind and optic flow stimuli were simultaneously presented. After the bee's flight became stable, during which the bee raised its abdomen, did not thrash its legs and extended its antennal flagella forward, the polarizer started rotating slowly (0.9, 1.8 or 3.6 deg s⁻¹), and the behavior of the bee was monitored for 600 s. When a bee stopped flying before 600 s, the data were not used in the analysis. In the experiments shown in Figs 2 and 3, the bee was tested three times under different stimulus conditions: clockwise (CW), static and counterclockwise (CCW). To eliminate possible effects of the stimulus sequence, the order of these three stimuli was randomly changed for each experiment. In other cases, a bee was tested only with the CW stimulus.

The flying behavior of the tethered bee was monitored using a USB camera (IUC-300CK2; Trinity, Gunma, Japan) placed behind the bee (see Fig. 1). Images of the bee were recorded at a rate of 1 Hz, i.e. 600 images for 10 min data. For each image, the x-coordinate of the bee's abdominal tip was determined manually to estimate flying orientation (Fig. S1). A series of x-coordinates was then calibrated into actual distances (in mm) from the center, where the tethering wire was fixed and used for further analysis (see below).

Whether the DRA of the compound eye was involved in flying behavior under the polarized light stimulus was determined using bees in which the DRAs had been painted (see Fig. 7C,D). The DRAs were painted as in our previous work (Sakura et al., 2012), with black acrylic emulsion paint (Herbol, Cologne, Germany) under a dissecting microscope just before the tethering procedure described above. The DRA of a compound eye is visually identifiable because the cornea appears slightly gray and cloudy (Meyer and Labhart, 1981). Because it was technically not possible to cover the DRA alone, which consists of only four to five horizontal rows of ommatidia (Meyer and Labhart, 1981; Wehner and Strasser, 1985), a small area of the unspecialized dorsal region next to the DRA was also painted. After the experiments, the paint cover was checked in all the experimental animals under a dissecting microscope. Data for cases in which any of the paint was missing were excluded from further analysis. The three ocelli, which are not involved in polarization vision (Rossel and Wehner, 1984), were not painted in the experiments.

All data analyses were performed using self-made programs in MATLAB (MathWorks, Natick, MA, USA). Periodicity of the time course of the abdominal tip location was analyzed using fast Fourier transform (FFT). For FFT, data for only the last 400 s of each trajectory (600 s in total) were used because the periodicity of a bee's flight was occasionally obscure at the beginning of the stimulus (e.g. see gray areas in Fig. 2Ac). The relative power spectrum (PS) was calculated, and peak frequencies were determined. In cases in which the PS had multiple peaks, we took into account only the maximum and second-maximum peaks for analysis. We defined a bee to be aligned with a certain e-vector orientation, or showing ‘polarotaxis’, when the PS of the bee showed the maximum or the second-maximum peak at the stimulus frequency, half rotation of the polarizer (note that the e-vectors 0 deg and 180 deg are identical); i.e. 0.5, 0.01 and 0.02 Hz for 0.9, 1.8 and 3.6 deg s⁻¹ stimuli, respectively. Distributions of bees showing polarotaxis were statistically analyzed using Fisher's exact test or Cochrane's Q-test with post hoc McNemar test for among- or within-group comparisons, respectively. In addition, the largest peak in the PS of each bee was determined to compare the distribution of the peaks by a bee.

To determine the relationship between a tethered bee's abdominal location and its flying behavior (Fig. S1), we simultaneously recorded abdominal images and the yaw torque of a flying tethered bee in the following procedures. Forager bees were collected at the hive entrance in the morning (09:00–10:00 h) and anesthetized in a refrigerator at 4°C for 10–20 min. A small, thin metal plate (1×2 mm, 0.02 mm thickness) was glued on the center of the thorax of each bee with beeswax. After recovery from anesthesia, the bee was tethered by attaching the metal plate to a torque meter (SH-02S; Suzuko, Yokohama, Japan). Before starting the experiment, a small piece of paper was provided to cause the bee to remain stationary and familiarize itself with the experimental environment.

All experiments were performed under dark conditions. Two monitors were facing each other and a bee was positioned at the center of the monitors and 13 cm from each monitor. Vertical black-white gratings (visual angle, 125 deg) with a sinusoidal illuminance change were presented on both monitors and moved from front to rear of the bee. To facilitate flight, a gentle laminar air flow (∼1.8 m s⁻¹) was provided by a fan placed in front of the bee. During the flight, the bee was monitored by a charge-coupled device camera (Sun Star 300, Electrophysics, Fairfield, NJ, USA) located to the posterior of the bee facing the abdominal tip. By using this camera, we were able to determine the position of the tip of the abdomen, as well as observe the behavior of the bee.

Three kinds of visual stimuli with air flow were applied for 30 s to induce a putative ‘straight’ or ‘turning’ flight. For straight flight, gratings of both monitors moved at the same speed of 110.5 deg s⁻¹ (a spatial frequency is 7.5 Hz). For putative turning flight, either the left or right monitor presented a faster speed (331.5 deg s⁻¹), by moving one at a greater speed than the other (110.5 deg s⁻¹) and vice versa. In this condition, the stimulated bee, in general, tended to turn to the slower side. One bee was subjected to three kinds of visual stimuli three times each. Air flow was constant throughout the experiments. Only bees that exhibited 30 s flight were used for further analysis.

The yaw torque from the torque meter was stored in the PC through an A/D converter using custom-made software, with a sampling rate of 120 points s^–1. A video movie was simultaneously stored with 30 frames s^–1 in the avi format. To see the correlation between yaw torque and flight posture, the tip of the abdomen was manually tracked frame by frame after converting the movie into JPG images with an interval of 0.1 s. By using custom-made software, we obtained x- and y-coordinates along with time. Because we were only interested in horizontal movements of the abdomen, we used only the x-coordinates for further analysis.

We calculated correlation coefficients between torque and the abdominal movement for all flights of all individuals. For calculation, we normalized horizontal movements individually as follows. All sampled x-coordinates obtained from three putative straight flights of a bee were averaged as a neutral position for the bee. Then, relative positions of the abdominal tips to the neutral position were calculated for each flight by converting pixels to distance (in mm). In this normalization, a positive value indicated that the abdomen was positioned on the right side to the base position and a negative value indicated the left. For yaw torque, a positive/negative value meant a CW/CCW turn. Only yaw torques of the corresponding time points to the manual tracking were used, i.e. the sampling interval was reduced to 0.1 s. Because the posture of a flying bee was not stable for the first 2–5 s after the onset of the stimulation, we discarded the first 10 s of data and used only the last 20 s for the correlation analysis.

Under our experimental conditions, approximately two-thirds of the experimental tethered bees could stably fly for over 10 min. A representative horizontal trajectory of a bee's abdominal tip under the three different polarized light conditions is shown in Fig. 2A. When the e-vector of the polarized light stimulus was gradually (1.8 deg s⁻¹) rotated CW or CCW, the bee showed periodic right-and-left abdominal movement, regardless of the rotational direction (Fig. 2Aa,c). The FFT analysis of the last 400 s of the trajectory data clearly showed that these abdominal movements were synchronized with an e-vector rotating frequency of 0.01 Hz (180 deg rotation) (Fig. 2Ba,c). Conversely, a bee did not show such periodic movement under the static e-vector stimulus (0 deg with respect to the body axis; Fig. 2Ab), and the peak of the PS was detected at 0.0025 Hz, which is coincident with the entire data length (400 s), instead of at 0.01 Hz (Fig. 2Bb; see below). We also determined the relationship between a tethered bee's abdominal location and its flying behavior (Fig. S1). Simultaneous recordings of the abdominal images and the yaw torque of a flying tethered bee showed a strong negative correlation, i.e. the bee's abdominal tip moved right when the bee turned left and vice versa. Therefore, the bee's periodic abdominal movement under the rotating e-vector stimulus could be explained by the bee periodically changing its steering action to adjust the flying direction to a certain e-vector orientation. If a given e-vector heading was desirable, the bee treated it as a target, steering left as the target approached in CW rotation and then steering right as it exited. At the orthogonal e-vector heading (an anti-target), the bee first steered right on CW approach and left on exit. Such a steering pattern could produce abdomen movement at twice the frequency of stimulus rotation given the axially symmetric e-vector stimulus. The PEO_cw and PEO_ccw of the bee shown in Fig. 2, calculated by the phase of the 0.01 Hz component in the PS (see Materials and Methods), were 126 deg and 112 deg, respectively (Fig. 2, dashed arrows). Around these e-vector directions, the bee's abdominal tip was located at almost the center, indicating that the bee did not change its flying direction.

Fig. 3D–F summarizes the PSs of all experimental bees tested under the three different conditions; CW, static and CCW polarized light stimulus. Under the CW and CCW rotations, the PS often had a strong power at 0.0025 Hz and/or 0.01 Hz regardless of the rotational direction (Fig. 3D,F), whereas it had a strong power only at 0.0025 Hz with static stimulus in most bees (Fig. 3E). A significantly higher number of bees (four, three and two of 21 bees for both CW and CCW, CW only and CCW only, respectively) displayed the maximum peaks at 0.01 Hz in the PS compared with that (none of the 21 bees) under the static 0 deg e-vector stimulus (Fig. 3A–C; CW, P=0.008; CCW, P=0.014; Cochrane's Q-test with post hoc McNemar test). In the averaged PS, a clear peak was noted at 0.01 Hz under the CW or CCW stimulus, although another strong power was detected at 0.0025 Hz (Fig. 3A,C), and strong power was only detected at 0.0025 Hz under the static stimulus (Fig. 3B). To confirm that the strong power at 0.0025 Hz reflects the entire data length, we also performed FFT analysis for the data under CW stimulus of different data lengths, i.e. 100, 200, 300, 400, 500 and 600 s (Fig. S2). In all cases, the averaged PS curves have two peaks, one at (data length)⁻¹ Hz and the other at 0.01 Hz. The peak at 0.01 Hz was always found in the PS regardless of the data length (Fig. S2, red bars), indicating that the trajectory has certain periodicity with 0.01 Hz. Considering that the strong power at 0.0025 Hz based on the data length was often found in the PSs, we next counted the number of the bees showing the maximum or the second-maximum peak at 0.01 Hz in each stimulus condition. In total, over half of the experimental bees showed a clear peak at 0.01 Hz in the PS under the rotating e-vector stimulus (Fig. 3D,F; ten, two and four of 21 bees for both CW and CCW, CW only and CCW only, respectively); however, under the static 0 deg e-vector stimulus, only two of the 21 bees showed a 0.01 Hz peak in the PS, which was significantly smaller than the number of bees showing a peak at 0.01 Hz under the rotating stimulus (Fig. 3E; CW, P=0.008; CCW, P=0.001; Cochrane's Q-test with post hoc McNemar test).

To determine whether the periodic movements were not elicited by the rotation of the e-vector, but rather by a slight fluctuation in light intensity caused by the polarizer rotation, we projected an unpolarized light stimulus through the depolarizer beneath the CW rotating polarizer (see Materials and Methods). Under the unpolarized light stimulus, the bees did not show any clear movements coincident with the polarizer rotation (Fig. 4A). Furthermore, no detectable peak at 0.01 Hz was noted in the averaged PS, and none of the eight experimental bees demonstrated the maximum peak at 0.01 Hz, while six of the eight bees showed the maximum power at 0.0025 Hz (Fig. 4B). Only one bee showed a small second-maximum peak at 0.01 Hz, which was significantly different from that under the CW polarized stimulus (P=0.044, Fisher's exact test). These results indicate that the abdominal periodic movements were elicited by the rotation of the polarized e-vector orientation. Taking these results together, we concluded that the tethered flying bees oriented to a certain e-vector direction, i.e. showed polarotaxis.

Next, we observed polarotaxis of the tethered bees under CW rotating e-vector stimulus at twice the speed (3.6 deg s⁻¹) or two times slower speed (0.9 deg s⁻¹) to confirm that the periodicity in the abdominal movement (Figs 2 and 3) was not elicited by internal rhythm but by external polarized light stimuli. Under the faster stimulus, some bees still showed right-and-left abdominal movements synchronized to the stimulus rotation (Fig. 5A). However, in contrast to the 1.8 deg s⁻¹ stimulus, the PS of the abdominal trajectory showed only a small peak at a stimulus frequency of 0.02 Hz (Fig. 5B). Moreover, in the averaged PS of all 14 experimental bees, a small, but detectable, peak at 0.02 Hz and a maximum peak at 0.0025 Hz were noted (Fig. 5C). The number of bees showing the peak at 0.02 Hz in the PS was significantly different from that experiencing the 1.8 deg s⁻¹ stimulus (seven of 14 bees for 3.6 deg s⁻¹ and none of the 21 bees for 1.8 deg s⁻¹ stimulus; P=0.0005, Fisher's exact test), although only one of the 14 experimental bees showed a maximum peak at 0.02 Hz (Fig. 5C). These results indicated that the bees exhibited weak polarotaxis to the fast rotating e-vector stimulus.

Under the slower rotating stimulus, the tethered bees showed clear right-and-left abdominal movements (Fig. 6A), the PS of which had a maximum peak at a stimulus frequency of 0.005 Hz (Fig. 6B). Four of the ten experimental bees exhibited a maximum peak at 0.005 Hz in each PS of the abdominal trajectory (Fig. 6C), whereas only one of the 21 bees did so under the 1.8 deg s⁻¹ stimulus, which was significantly lower (P=0.0274, Fisher's exact test). This result indicated that the bees also responded to a slow stimulus. However, we could not detect a 0.005 Hz peak in the averaged PS, although the power at 0.005 Hz was relatively high compared with that under other stimulus conditions (Fig. 6C); this could have occurred because the peak could not be clearly separated from the peak at 0.0025 Hz owing to data interference from unresponsive bees (Fig. 3B and Fig. 4C).

Polarization vision in insects is known to be mediated by the DRA of the compound eye. To confirm the sensory input area for polarotaxis in the eye, we covered a part of each compound eye and restricted the area receiving light stimulation to the DRA (Fig. 7D,E). The bees in which the DRAs were covered did not show polarotactic abdominal movement even under the 1.8 deg s⁻¹ CW rotating polarized light stimulus to which intact bees responded (Fig. 7A), and no clear peak was noted at the stimulus frequency of 0.01 Hz in the PS (Fig. 7B). The averaged PS of all eight experimental bees did not exhibit a peak at 0.01 Hz (Fig. 7C), indicating that the bees with covered DRAs lost the ability to orient to certain e-vectors. Similar to the response of intact bees to a static stimulus, none of the eight bees displayed a maximum peak at 0.01 Hz (Fig. 7C, see also Fig. 3B), and their response was not significantly different (P=1, Fisher's exact test). Conversely, the number of bees showing a maximum peak at 0.01 Hz was also not significantly different from that of the intact bees under the CW stimulus (Fig. 3A and Fig. 7C; P=0.1421, Fisher's exact test), probably because of the small number of experimental bees used.

We assessed the PEO_cw of the 21 bees that showed polarotaxis under the 1.8 deg s⁻¹ CW stimulus. In addition to the 12 bees in the experiments shown in Fig. 3, data from another nine bees were newly obtained from 16 bees tested in total under the CW stimulus. The PSs of all 37 individuals tested under the 1.8 deg s⁻¹ CW stimulus (21 bees in Fig. 3 and an additional 16 bees) are summarized in Fig. S3. The PEO_cw of each bee varied from −90 to 90 deg (Fig. 8). However, more than half of the bees (14 of 21) showed a PEO_cw between −60 and 0 deg, and the distribution was not significantly random (P=0.01, Rayleigh test). We also tried to compare the PEO_cw and PEO_ccw of the ten bees that showed polarotaxis for both CW and CCW stimuli in the experiments shown in Fig. 3 (Fig. S4). Although some bees showed similar PEOs for CW and CCW stimuli, the difference between them (PEO_cw−PEO_ccw) was quite varied (−3.48±37.74 deg, N=10).

In the present study, we showed that bees tended to orient to certain e-vector angles during their flight under tethered condition, i.e. they referred polarized light information to control their flight direction. The fact that fewer bees responded to the fast stimulus (3.6 deg s⁻¹, Fig. 5) than to the slow stimuli (0.9 deg s⁻¹ and 1.8 deg s⁻¹; Figs 2, 3 and 6) is also indicative of the use of e-vector orientation as a global cue for orientation. Probably, they did not refer to the e-vector when it quickly changed because they did not expect such a situation, except when they quickly changed their flight direction. It is also possible that the fast-rotating stimulus caused an optomotor response in which the bee steered in the same direction as the rotating stimulus at all e-vector orientations. To confirm this possibility, we performed a trend analysis for all behavioral data shown in Figs 3, 5 and 6 (Fig. S5). If a bee showed a strong optomotor response, it should show a constant steering trend toward the stimulus direction. However, in all stimulus conditions, the trend was varied among individuals, and we could not find any prominent correlations between steering and stimulus directions (Fig. S5B,C). For more precise verification of the optomotor responses for the rotating e-vector, comparison between the behavior under the fast CW and CCW stimulus will be necessary.

Bees in which the DRAs were covered did not show any polarotaxis (Fig. 7). It is well known that detection of skylight polarization in insects is mediated by ommatidia in the DRA (for review see Labhart and Meyer, 1999; Wehner and Labhart, 2006). In honeybees, ultraviolet (UV)-sensitive photoreceptors of the ommatidia in the DRA are highly polarization sensitive, and their receptive field covers a large part of the celestial hemisphere, which is suitable for observing the sky (Labhart, 1980; Wehner and Strasser, 1985). Behaviorally, it has also been demonstrated that covering the DRA impairs the correct coding of food orientation by the waggle dance orientation (Wehner and Strasser, 1985) and discrimination of different e-vector orientations by classical conditioning (Sakura et al., 2012). These results clearly show that bees utilize polarized light detected by the ommatidia in the DRA for orientation.

Polarotaxis in insects has been demonstrated in several species. Obviously, orientation to a certain e-vector direction is a common occurrence among insect species that utilize skylight polarization for navigation. Classically, it has been tested using a treadmill device in the cricket (Gryllus campestris; Brunner and Labhart, 1987) and the fly (Musca domestica; von Philipsborn and Labhart, 1990). Using such a device, the insect was tethered on an air-suspended ball and its walking trajectory could be monitored through the rotation of the ball. In these species, the insect on the ball showed clear polarotactic right-and-left turns when the e-vector of the zenithal polarized light stimulus was slowly rotated, as we showed in this study in flying honeybees. This kind of behavior does not merely demonstrate that they have polarization vision but also allowed us to clarify fundamental properties of insect polarization vision, e.g. perception through the DRA in the compound eye (Brunner and Labhart, 1987), monochromatic spectral sensitivity (Herzmann and Labahrt, 1989; von Philipsborn and Labhart, 1990) and sensitivity to the degree of polarization (Henze and Labhart, 2007).

Orientation to polarized light has been investigated in tethered flying insects as well by monitoring yaw-torque responses (locust, Schistocerca gregaria; Mappes and Homberg, 2004), flight orientations (monarch butterflies, Danaus plexippus; Reppert et al., 2004) or changes in body axis (Drosophila; Weir and Dickinson, 2012; Mathejczyk and Wernet, 2019). A potential problem in investigating polarization vision in tethered flying insects is that sometimes the tethering apparatus, including the torque meter or other recording devices, interrupts a part of the visual field of the tested animal. In the present experiments, we succeeded in evaluating a bee's polarotactic flight steering by simply monitoring the horizontal position of the abdominal tip that was strongly anti-correlated with the torque generated by the bee (Fig. S1). Using these methods, the entire visual field of the animal remained open; therefore, it had an advantage for investigating the animal's responses under various stimulus conditions.

The PEO distribution has been reported in several species. In walking crickets and flies, a weak preference to an e-vector orientation perpendicular to their body axis has been demonstrated, although the reason for this behavior was not clear (Brunner and Labhart, 1987; von Philipsborn and Labhart, 1990). In flying locusts and Drosophila, the PEOs were randomly distributed and they did not show any directional preferences as a population (Mappes and Homberg, 2004; Warren et al., 2018; Mathejczyk and Wernet, 2019). In the present study, even though the sample size might be too small to conclude their heading preferences, the distribution was significantly non-uniform and bees seemed to prefer e-vector orientations skewed left of the body axis (Fig. 8). In some bees, the PEOs under CW and CCW stimulus were quite similar (Fig. S4). Therefore it was possible that, at least in these bees, each bee had its own PEO and used it not only as a reference for maintaining straight flight but also to deduce its heading orientation. Further investigation of the bees’ PEOs under CW and CCW stimuli will be necessary to confirm whether each bee had a specific PEO.

Considering that central place foragers, such as honeybees, have to change their navigational directions depending on the currently available food locations, their PEOs should reflect their previous foraging experiences. In the present study, we collected bees with a pollen load at the hive entrance; therefore, all experimental forager bees were returners. Consequently, we could no longer assess their feeding locations when we measured their flight responses in the laboratory. Moreover, their path-integration vector should be reset to a zero state in such a situation (Sommer et al., 2008), and they might not have had a strong motivation to use polarized light cues for navigation. To further clarify the role of polarization vision in flying foragers, testing the PEOs in bees in different navigational states will be crucial.

We are grateful to Dr M. V. Srinivasan and Dr T. Luu for helpful advice regarding the construction of the flight simulator. We also thank Dr Y. Hasegawa, Dr H. Ikeno and Ms. H. Onishi for their help with measuring the torque of the flying bees.