Length of stimulus presentation and visual angle are critical for efficient visual PER conditioning in the restrained honey bee, Apis mellifera

The learning of visual cues constitutes an essential ability of bees, enabling them to orient and recognize nest sites, food sources and mating partners (Barth, 1985; Winston, 1987). During their foraging flights, bees are confronted with a variety of flowers which differ in appearance, as well as in the quantity and quality of food supply (Chittka and Thomson, 2001). To efficiently discriminate between different flowers, bees may use specific signals or cues, e.g. flower size, color, odor or shape, of which color seems to be the major feature (Chittka and Menzel, 1992; Gumbert, 2000).

Karl von Frisch provided conclusive evidence that honey bees possess the capability to discriminate between different color stimuli more than a century ago (von Frisch, 1914, 1953), and many subsequent studies on color learning and discrimination in free-flying bees have confirmed his findings (e.g. Avarguès-Weber et al., 2011; Dyer et al., 2008; Giurfa, 2007; Schubert et al., 2002; von Frisch, 1965). Although experiments with free-flying bees allow for learning under more natural conditions, it is difficult to control a bee's behavior or use invasive techniques to understand the underlying neuronal and physiological mechanisms. Hence, an alternative approach is the conditioning of the proboscis extension response (PER) to evaluate learning and memory in bees, because of its robustness and reproducibility under constant and controllable environmental conditions. In the PER paradigm, harnessed bees must learn to associate a conditioned stimulus (CS; e.g. novel color) with an unconditioned stimulus (US; i.e. food taste). After a few paired presentations of the CS and US, the CS alone provokes the extension of the proboscis (Fig. 1C). Whereas PER conditioning with odors as the CS is well established in honey bees (e.g. Bitterman et al., 1983; Hammer and Menzel, 1995; Menzel et al., 2001) and bumble bees (e.g. Laloi et al., 1999; Riveros and Gronenberg, 2009; Sommerlandt et al., 2014), visual conditioning of the PER is considered to be difficult.

More than 60 years ago, Kuwabara (1957) conducted the first study of visual PER conditioning to investigate color learning of restrained honey bees and concluded that visual PER conditioning is only successful when bees are deprived of their antennae. Several studies confirmed his observation (Hori et al., 2006; Mota et al., 2011; Niggebrugge et al., 2009); but recently, other studies have shown successful conditioning of a visual stimulus with a sucrose reward using the PER in intact honey bees (Balamurali et al., 2015; Dobrin and Fahrbach, 2012; Jernigan et al., 2014) and bumble bees (Lichtenstein et al., 2015; Riveros and Gronenberg, 2012). However, at least for the Western honey bee, the performance level reached in restrained individuals with intact antennae was significantly lower than that achieved with olfactory conditioning (Dobrin and Fahrbach, 2012) or conditioning was only successful in combination with olfactory (Mota et al., 2011) or motion cues (Balamurali et al., 2015).

In a recent review, Avarguès-Weber and Mota (2016) compiled published data on visual conditioning in harnessed bees and discussed potential factors accounting for the inconsistency found among the studies. Here, we tested several factors for their impact on visual PER conditioning in order to develop a protocol that achieves a performance level similar to that of olfactory PER conditioning. We also introduced a new experimental setup, which includes a shutter to control CS onset and offset, driven by a PC-based software program ‘TimingProtocol’, which allows for a more precise and automated visual PER conditioning. Using this setup, we trained honey bee workers to associate a monochromatic light stimulus with a reward (absolute conditioning) or to discriminate between a rewarded and an unrewarded color stimulus (differential conditioning) and tested the effect of stimulus presentation length (experiment 1), stimulus size, measured as visual angle (experiment 2), chromatic and achromatic information use for stimulus discrimination (experiment 3), and ambient illumination (experiment 4).

We caught departing Apis mellifera carnica Pollman 1879 in the morning directly at the hive entrance of colonies at the bee station of the University of Würzburg. Immediately after being caught, bees were immobilized on ice and harnessed in metal tubes with fabric tape. After fixation, bees could only move their antennae and mouthparts (Fig. 1C; Movie 1). They were subsequently fed to satiation with a 30% sucrose solution (w/v) and placed for 3–4 h in a dark box (temperature: ∼25°C; relative humidity: ∼50%). In the early afternoon and directly before PER conditioning, all bees were pre-tested for a PER by carefully touching the antennae with a toothpick soaked with 50% sucrose solution (w/v). For all experiments, we used only motivated bees that showed an immediate PER to the sugar reward during the pre-test.

Bees were conditioned to two different light stimuli provided by different monochromatic filters (Schott & Gen, Jena, Germany) with absorption maxima at 435 and 488 nm, a half bandwidth of ca. 10 nm and an aperture angle of ca. 70 deg (Fig. 1A,D). To prevent bees from learning achromatic cues, we presented each color stimulus at three different intensities, which were generated using neutral density (ND) filters (transmission: 13% and 51%; for more details, see Lichtenstein et al., 2015). The experimental setup consisted of a non-reflective gray acrylic movable sleigh with nine individual chambers (50 mm×60 mm×50 mm) in which individual bees were placed. A filter holder, which housed the color and ND filters and a diffusor, was positioned on top of the sleigh and attached to a computer-controlled shutter and shutter driver (Uniblitz VCM-D1, Uniblitz, Rochester, NY, USA; Fig. 1D). Light was provided by a cold light lamp (Schott KL1500, Mainz, Germany) via a flexible light guide (diameter: 9 mm; Schott). Once the shutter opened, the light illuminated the chamber from above and the bee faced directly towards the light at a distance of ca. 5 cm. The shutter was controlled by a custom-written software program, ‘TimingProtocol’, which allows precise control of all training properties, e.g. conditioning type, length of stimulus presentation, inter-trial interval and the simultaneous recording and visual presentation of the results (Fig. S1). The program is freely available on request from the corresponding author or senior author.

Bees were trained in an absolute and differential learning task. During conditioning, the bees had either to associate a color stimulus with a reward or to discriminate between a rewarded (CS+) and an unrewarded (CS−) stimulus. During differential conditioning, both color stimuli combinations were tested (435 nm rewarded, 488 nm unrewarded, and vice versa). During absolute conditioning, each bee was trained for 10 trials with CS+ only, and during differential conditioning, each bee was trained for 18 trials with CS+ and CS− (9 trials of each) in a randomized order (Fig. 1B). The conditioning protocol was adapted from Jernigan et al. (2014). During the entire experiment, a toothpick soaked with a 50% sugar solution was placed at a distance of ca. 2 cm below the bee's head (see Movie 1) to keep movements of the conditioner's hand during US presentation at minimum and to prevent bees from showing unspecific PER responses due to water vapor (Dobrin and Fahrbach, 2012; Kuwabara, 1957). At the beginning of the training procedure, each bee had a rest for 10 s to become familiarized with the given situation. Then, the color stimulus was turned on for 5, 7, 10 or 13 s, and the bee was rewarded during the last 3 s, resulting in a CS presentation of 2, 4, 7 or 10 s before the reward was given (Fig. 1B; experiment 1). These stimulus durations were chosen to cover the range of CS presentations used in most of the earlier studies (Avarguès-Weber and Mota, 2016). After the color stimulus was turned off, each bee had another 10 s rest before the sleigh and the conditioner's hand were moved and the next bee was positioned under the shutter. In all conditioning experiments, we used an inter-trial interval (ITI) of 10 min (exception for experiment 4, see below), and the setup was indirectly illuminated by a red-light lamp (white light LED lamp covered with a red filter, transmission >640 nm; e-colour+ E787, ROSCO, London, UK) during the entire experiment. To identify the minimum visual angle at which the stimulus can just be learned by the bees, we tested stimulus sizes of 70, 35, 15 and 5 deg by decreasing the aperture of the filter holder (experiment 2; Fig. 1D). At small visual angles, free-flying honey bees use an achromatic channel for single-object detection (Giurfa et al., 1996; Dyer et al., 2008). As our bees failed to discriminate between the two monochromatic lights at small visual angles (see below), we also performed absolute conditioning at 70 and 5 deg angles to test whether they use only achromatic information at small visual angles (and thus cannot discriminate between the two lights of varying intensities) or whether they cannot receive the stimulus at all by presenting the learned color stimulus (CS) and a new color stimulus (NCol; absolute conditioning) in a random order without any reward after training (experiment 3). Finally, we also compared the learning performance of bees when trained in the dark (under constant red light conditions) or at moderate ambient illumination (ca. 300 lx of natural sunlight filtered through a window; experiment 4).

Statistics for the learning curves were calculated on the basis of an individual's number of PER responses towards the color stimulus. A PER shown during the CS+ presentation and before the reward was presented was scored as 1 (see second half of Movie 1), whereas the extension of the proboscis in response to the sucrose reward only or no response at all was scored as 0. Bees that showed PER in less than 50% of the trials were excluded from further analysis. Differential learning performance within one treatment group (CS+ versus CS−) was analyzed using the Wilcoxon test and among different groups by means of the Mann–Whitney U-test. Acquisition performance during absolute conditioning was tested by means of Cochran's Q-test, while data for color specificity were evaluated using Fisher's exact test. All statistical analyses were performed in IBM SPSS v. 20.

We used the PER paradigm to test the color learning capabilities of harnessed intact Apis mellifera carnica workers. As no significant difference was found between the color combinations (CS+: 435 nm, CS−: 488 nm, and vice versa; data not shown) for any tested CS presentation length, data were pooled. Bees were able to distinguish between the two monochromatic color stimuli if the presentation length of the stimulus was 13, 10 or 7 s (10, 7 and 4 s before reward onset; CS+ versus CS− 13 s: P<0.001, Z=−4.047; CS+ versus CS− 10 s: P<0.001, Z=−4.226; CS+ versus CS− 7 s: P=0.032, Z=−2.148) and reached a performance level of up to 70% when the color stimulus was presented at least 10 s before a reward was offered (Fig. 2). However, if the stimulus presentation length was 5 s (i.e. 2 s before US presentation), bees were not able to discriminate between the CS+ and CS− stimulus (CS+ versus CS− 5 s: P=0.176, Z=−1.354).

In the second experiment, we tested four different stimulus sizes (70, 35, 15 and 5 deg; stimulus presentation length for all sizes was 13 s). As no significant difference was found between the color combinations (CS+: 435 nm, CS−: 488 nm, and vice versa; data not shown) for any tested CS presentation length, data were pooled. Bees could easily learn the task when presented with stimuli with visual angles of 70 and 35 deg (CS+ versus CS−, 70 deg: P<0.001, Z=−3.335; CS+ versus CS−, 35 deg: P=0.003, Z=−2.940) but failed at smaller angles (CS+ versus CS−, 15 deg: P=0.410, Z=−0.823; CS+ versus CS−, 5 deg: P=0.931, Z=−0.087; Fig. 3).

Next, we asked why the bees failed to discriminate between the stimuli at smaller visual angles. We trained bees to associate a stimulus size of 70 and 5 deg with a reward, using absolute conditioning (Fig. 4). At both visual angles, bees were able to learn to associate the stimulus with the reward (70 deg: P<0.001, Q=52.2; 5 deg: P<0.001, Q=36.7). However, in the unrewarded test, when the learned stimulus and a novel color stimulus were presented, only individuals trained with a large stimulus size responded to the correct wavelength (70 deg CS versus NCol: P=0.002, two-sided Fisher's exact test). In contrast, bees trained with a 5 deg stimulus could not discriminate between the two wavelengths (5 deg CS versus NCol: P=1, two-sided Fisher's exact test), indicating that they did not learn the chromatic properties of the training stimulus.

In the natural context, color learning usually takes place under daylight conditions; therefore, we tested whether visual PER learning is influenced by environmental illumination (Fig. 5). Bees were able to discriminate between the two stimuli when tested under moderate ambient illumination (CS+ versus CS−: P<0.001, Z=−4.965) and showed no difference in learning performance compared with the dark condition (CS+ dark versus CS+ bright: P=0.273, Z=−1.096; compare Fig. 2D and Fig. 5].

In the present study, we showed that restrained workers of the Western honey bee Apis mellifera carnica with intact antennae were capable of learning and discriminating between two monochromatic lights by using the PER paradigm. Whereas early studies stated that visual PER conditioning is only successful when the bee's antennae were ablated, recent studies revealed different results (Balamurali et al., 2015; Dobrin and Fahrbach, 2012; Jernigan et al., 2014; Lichtenstein et al., 2015; Riveros and Gronenberg, 2012). The present study provides further evidence that antennal deprivation is not necessary for successful color learning in restrained honey bees. In contrast to earlier studies, intact honey bees in the present study reached learning performance levels of up to 70% when stimulus duration and size were optimized. Below, we discuss possible reasons for the high variation of learning performance found in earlier studies.

Since the first experiments on visual PER conditioning were performed more than 60 years ago, the length of the stimulus presentation has notably varied. Whereas particularly older studies preferred a shorter stimulus length more comparable to that in olfactory PER conditioning protocols (Frost et al., 2012), recent studies have increased stimulus lengths (Avarguès-Weber and Mota, 2016). The four stimulus durations tested in this study (5, 7, 10 and 13 s) have been used in previous studies, but only bees confronted with durations of 9 s or longer before US onset reached performance levels that were commonly found in olfactory conditioning (Jernigan et al., 2014; Lichtenstein et al., 2015; Riveros and Gronenberg, 2012). Hence, we conclude that stimulus length is crucial for efficient visual PER conditioning. But the question arises, why does visual PER conditioning require much longer stimulus lengths compared with olfactory conditioning?

One possible explanation might be that the neuronal processing of a visual stimulus is more complex and thus more time consuming because of the multitude of light-receptive organs present in the bee, compared with the processing of olfactory information, which is restricted to the antennal input.

During the acquisition phase of olfactory PER conditioning, bees usually extend their proboscis within less than 2 s of stimulus onset and less than 1 s when the association has been learned (Rehder, 1987). This ‘reaction time’ includes the activation of the odor-sensitive receptors at the antenna, passing the information via the antennal lobes (Joerges et al., 1997) and the dual olfactory tracts (Rössler and Brill, 2013) to higher order brain centers (e.g. the mushroom bodies, MB; Sandoz and Menzel, 2001). The output neurons of the MB probably send the processed information into the lateral protocerebrum, where the actual information is compared with the (olfactory) memory, and finally the activation of the proboscis muscles is initiated. During the association process, the VUMmx1 neuron and its putative transmitter octopamine mediate the reinforcement (Hammer, 1997). Moreover, octopamine can substitute for the US reinforcement (e.g. sucrose solution) and a conditioned stimulus is capable of activating the VUMmx1 neuron, thereby maintaining the function of the reinforcer in absence of the US (Hammer, 1993; Hammer and Menzel, 1998).

At the periphery, processing time of olfactory and visual information seems to be largely identical. Strube-Bloss and Rössler (2018) stimulated the antennae (by odor stimulus) and the compound eyes (LED light stimulus) of honey bees and recorded the activity of output neurons of the mushroom bodies. For both modalities, they found processing times between 70 and 100 ms, which indicates that the difference in ‘reaction time’ between visual and olfactory PER must arise from later processing steps, most likely at the level of stimulus evaluation at higher neuronal levels. In contrast to olfaction, light reception is not restricted to an exclusive input channel (the antenna) but is sensed via several parallel light-sensitive pathways. The compound eye mediates object and motion detection via chromatic and achromatic vision and relays the information via the optic lobes and various projection neurons to the MBs, the anterior optic tubercle (AOTU) and the posterior protocerebrum (Gronenberg, 1986; Paulk et al., 2009; Paulk and Gronenberg, 2008). The dorsal rim area of the compound eye is a specialized region capable of sensing polarized light by means of UV-sensitive photoreceptors, from where the information is projected via the dorsal part of the optic lobes to the AOTU and finally to the central complex's lower division (Held et al., 2016). The three ocelli, located at the vertex of the bee's head, possess UV- and green-sensitive photoreceptors (Goldsmith and Ruck, 1958) and project via the ocellar interneurons to the median posterior protocerebrum (Maronde, 1991). Finally, a putative light-sensitive extra­-ocular photoreceptor, expressing pteropsin, a light-sensitive pigment with unknown sensitivity and function (Velarde et al., 2005), is located between the lobula and the lateral protocerebrum. By shining light on the bee's head, some or even all of these light-sensitive pathways become activated and (so far unknown) higher order processes are needed to integrate all the information and put it in the correct context. As a restrained bee, in contrast to a freely moving individual, is denied information related to motion and spatial cues, which are received during flight, as well as information about landmarks that have been memorized in the context of stimulus learning, processing of visual information that leads to a correct behavioral action might become prolonged. Furthermore, there is no anatomical evidence that the VUMmx1 neuron (or any other VUM neuron) is directly connected with any of the visual pathways (Schröter et al., 2007), suggesting that the reward system for visual learning might be different (and probably more complex) than that for the olfactory system. Recently, Plath et al. (2017) showed that color learning in an aversive conditioning assay involves both MB ventral lobes and the central complex (CX). However, so far, no evidence for a direct connection between MB output neurons and the CX has been found in honey bees, and Plath et al. (2017) thus suggested a more indirect connection, which would also prolong processing time compared with olfactory learning.

Although visual and olfactory stimuli are processed via different neuronal pathways, it remains unclear from an ecological perspective why bees need longer to process visual compared with olfactory cues when associating a stimulus with a reward. One possible explanation for this difference may lie in the nature of visual and olfactory stimuli. Odors are mixtures of chemical compounds, which appear in plumes and strike the olfactory system of insects in a jerky and unpredictable manner (Chittka and Menzel, 1992; Galizia and Menzel, 2000; Sandoz, 2011; Touhara and Vosshall, 2009). Olfactory information should be processed more quickly than visual information, as the availability of odors is short and unreliable. In contrast, flower color, once detected, remains constant in space and quality during the approach flight and allows for a longer processing time (Chittka and Menzel, 1992). This may facilitate more reliable and precise information processing. For example, a tradeoff between the speed and accuracy of flower discrimination was shown in bumble bees, when differently colored flowers were presented for discrimination (Chittka et al., 2003).

The mode of stimulus presentation might also affect learning speed and performance. In most previous studies on PER color learning, stimuli were presented using LEDs or monochromatic filters, by globally illuminating the whole training setup from above (Hori et al., 2006; Mota et al., 2011), or presenting the color stimuli from below (Dobrin and Fahrbach, 2012; Jernigan et al., 2014; Riveros and Gronenberg, 2012) or from the front (Balamurali et al., 2015). In contrast, in our setup, every harnessed bee was held in a single small quadratic chamber and the color stimulus was presented directly from the top of the chamber. This enabled the bee to view the stimulus as a clearly defined circle at a defined visual angle in the frontal visual field that probably resembles the shape of a flower. Interestingly, when the visual angle of the stimulus was small (5 or 15 deg), bees could not utilize chromatic information but could still acquire information about the brightness (or excitation of the green receptors), which clearly indicates that the spatial dimension of the stimulus affects the bee's capability to build the association between stimulus and reward. Thus, our data nicely match the findings of Giurfa et al. (1996), who suggested a minimum visual angle of 15 deg for detecting colored objects in free-flying honey bees.

Finally, in their recent review, Avarguès-Weber and Mota (2016) discussed whether visual PER learning might be affected or impaired by unnatural experimental conditions, like complete darkness during conditioning or fixation of the bee's head. However, we found no difference in the learning performance of bees conditioned under dark or moderate ambient light conditions. Also, motion cues were not required for successful learning as our bees could only move their antennae and proboscis but not the head, and we thus assume that stimulus length and size are the critical factors for successful conditioning.

Until now, broadband reflecting stimuli, which are commonly used in studies with free-flying bees (e.g. Dyer et al., 2008), have not been successfully tested in the visual PER paradigm. Thus, future studies should address how stimulus quality (self-luminous versus broadband reflecting stimuli) and the activation of distinct light-sensitive pathways (excitation of the compound eyes, ocelli, extra-retinal photoreceptors and any combination of these) affect visual learning in harnessed bees.

We thank Dirk Ahrens for providing the honey bees and John Plant for linguistic advice.