Social foraging extends associative odor–food memory expression in an automated learning assay for Drosophila melanogaster

Vertebrates often forage in groups to obtain a more accurate estimate of the location and quality of resources (Giraldeau and Caraco, 2000; Templeton and Giraldeau, 1996; Valone, 1989; Ward and Zahavi, 1973). Insects also convey information about the location and quality of a food source through social interactions. For example, honey bees signal the direction and distance of food locations to other bees (Von Frisch, 1965), ants complement their individual memory of a route to food using trail pheromones left by scouts (Czaczkes et al., 2011), and stimulus enhancement and local enhancement at the food source improves foraging efficiency in bumble bees (Alem et al., 2016; Avarguès-Weber and Chittka, 2014; Leadbeater and Dawson, 2017; Worden and Papaj, 2005). These social effects on foraging have been mostly studied at the level of behavioral outcome, and the neural mechanisms of how social information transfer improves foraging are still unknown.

The fruit fly Drosophila melanogaster Meigen 1830 is a suitable model organism for studying the effects of social interactions on foraging at both the behavioral and the neuronal levels. Fruit flies are gregarious (Lefranc et al., 2001; Navarro and del Solar, 1975) and demonstrate a rich repertoire of social behaviors that encompass communication about internal states (Suh et al., 2004), social information spread during odor avoidance (Ramdya et al., 2014), foraging (Abu et al., 2018; Durisko and Dukas, 2013; Golden and Dukas, 2014; Lihoreau et al., 2016; Tinette et al., 2004) and predator-induced egg retention (Kacsoh et al., 2015). Moreover, fruit flies socially learn, and naïve flies copy mate choices (Danchin et al., 2018; Germain et al., 2016; Mery et al., 2009) and oviposition site choices from experienced conspecifics (Battesti et al., 2012; Sarin and Dukas, 2009), and they show increased aversive odor memory retrieval when in groups (Chabaud et al., 2009). However, it is still unknown whether group size affects social interactions and associative odor–food learning in foraging flies.

The mechanistic understanding of foraging in fruit flies is unparalleled, both in regard to the neural mechanisms of odor-guided search (Galizia, 2014; Haverkamp et al., 2018; Wilson, 2013) and feeding (Itskov and Ribeiro, 2013), and of associative odor–food learning (Burke et al., 2012; Huetteroth et al., 2015; Liu et al., 2012; Owald and Waddell, 2015; Schwaerzel et al., 2003; Tempel et al., 1983; Thum et al., 2007), making the fruit fly a good model for studying the neural mechanisms of social interactions during foraging.

Here, we investigated whether fruit flies socially interact during foraging and whether group size affects associative odor–food memory expression. We developed an automated assay to study associative odor–food reward learning and memory in single flies and in groups of flies. We found that odor–food memory expression increased in strength and duration with increasing group size, and flies in small or large groups, but not in pairs, were attracted to each other. These data confirm that flies socially interact during foraging (Abu et al., 2018; Durisko and Dukas, 2013; Golden and Dukas, 2014; Lihoreau et al., 2016; Tinette et al., 2004). In addition, these data suggest that social interactions increase the efficiency of odor memory-guided food search.

Drosophila melanogaster wild-type Canton-S were raised on a standard food medium (100 ml contains 6.7 g fructose, 2.4 g dry yeast, 0.7 g agar, 2.1 g sugar beet syrup, 0.282 g ethyl paraben and 0.61 ml propionic acid) in a room with a natural daylight cycle (experiments were performed between 11 January and 19 February 2018 in Konstanz, Germany), with an average temperature of 23.5°C and 32% relative humidity. Four- to 7-day-old flies were anesthetized with CO₂ and female flies were collected. To motivate flies to search for food, flies were starved for 3 days in a fly vial with filter paper soaked in water.

To condition groups of flies, we used an automated rotating platform with four circular arenas (Fig. 1A,B). The arenas were covered with a watch glass (7 cm diameter, 8 mm height in the center), which was coated on the inner side with Sigmacote (Sigma-Aldrich) to prevent flies from walking on the inside of the glass. The floor was made of a Teflon-coated fiberglass fabric (441.33 P, FIBERFLON, Konstanz, Germany). Pure odorants (ethyl acetate and 2,3-butanedione, Sigma-Aldrich) were stored in 20 ml vials (Schmidlin Labor and Service). The vials were mounted under the platform and the lid was pierced with a hypodermic needle (0.45×25 mm, Sterican), allowing the odorant to diffuse through a hole (5 mm diameter) in the platform through the Teflon fabric and into the arena. Each arena had two odorant sources. One odorant was used as sucrose-paired conditioned stimulus (CS+) and the other odorant was used as unpaired conditioned stimulus (CS−). Each odorant was used equally often as the CS+ and the CS−. At the location of the CS+, 20 μl of oversaturated sucrose–ethanol solution was pipetted onto the fiberglass fabric and blow-dried for 20 min, producing a thin layer of pure sucrose on a round patch with a diameter of 10 mm. The positions of the CS+ and CS− were always switched between the conditioning and the test (e.g. if CS+ was at the inside position during conditioning, it was at the outside position during the test, and in half of the experimental runs the CS+ was at the inside position during conditioning, and in the other half at the outside position). To change the floor between experimental phases, the platform was rotated underneath the arenas; the arenas themselves did not move (Movie 1). The angular rotation speed of the platform was 360 deg 25 s⁻¹, which corresponded to a speed of 2.6 cm s⁻¹ in the center of the arena (the distance between center of the platform and the center of the arena is 10.25 cm).

The conditioning apparatus was placed in an air suction hood in order to remove odorants. All experiments were performed in the dark to eliminate visual stimuli. The arena was back-illuminated with infrared light (850 nm, SOLAROX LED Strip), which is not visible to flies, and experiments were video recorded with an infrared sensitive camera (infrared Camera Module v2, Pi NoiR, connected to a Raspberry Pi 3, model B V1.2) at 15 frames s⁻¹. The rotating motor was controlled via TTL pulses through the Raspberry Pi. The rotation of the platform and the video recordings were controlled with custom-written software in Python (Stefanie Neupert).

All experiments were performed between 10:00 and 12:00 h or after 15:00 h, during periods when flies show higher foraging activity (van Breugel et al., 2018). Each experimental run contained four differently sized groups (‘single’, ‘pair’, ‘small group’ and ‘large group’), and the positions of the four arenas used for the four differently sized groups were balanced across experimental runs.

One experimental run consisted of four phases. (1) Acclimatization (Fig. 1B, solid arcs): flies were taken from the room where they were raised and starved, sucked out from the vials using a tube aspirator, placed into an arena that had no odorant source and allowed to acclimatize for 10 min. (2) Conditioning (Fig. 1B, dotted arcs): the floor was rotated counterclockwise by 22.5 deg and the CS+ paired with sucrose and the CS− without sucrose were presented for 7 min. (3) Pause (Fig. 1B, dashed arcs): the floor was rotated by 22.5 deg and replaced by a new floor without odorants or sucrose. The pause lasted for 5 s. (4) Memory test: the floor was rotated by 22.5 deg and replaced by a floor that had the CS+ and CS− but without sucrose. The CS+ and CS− positions were switched from those during conditioning. The test phase lasted for 7 min.

Videos were recorded during the conditioning and test. After each experimental run, all flies were discarded and the Teflon fabric floor was rinsed with hot water and soap (Buzil G 530) using a sponge and dried overnight to remove the odorants and the sucrose patch.

Video recordings were analyzed using the software Fiji (ImageJ 1.51s, Wayne Rasband, National Insitutes of Health, USA). We removed the first 30 frames because of compression artifacts and converted the video to grayscale. Then, we performed a Z projection to obtain the maximum intensity projection over the whole video, and calculated the difference per frame between the maximum intensity projection and the original video. This gave us a clear image of flies moving around the arena for tracking. We used this output to track flies using the plugin TrackMate (version 3.5.3; Tinevez et al., 2017). We used a Downsample LoG detector to identify flies (blob diameter=13 pixels, downsampling=3, threshold=6–8). To generate the tracks, we used the Simple LAP Tracker, with the following parameters: linking distance=150 pixels, maximum gap closing=150 pixels and maximal frame gap=3 frames. For the conditioning data, we only extracted the x and y coordinates of each fly per frame. For the test data, we extracted the x and y coordinates per frame as well as the identity of the fly throughout the recording. We inspected all tracking results visually and corrected the tracks manually to connect the missing links, and afterwards we extracted the x and y coordinates for the analysis.

For both the conditioning and the test datasets, we centralized each arena so that the center point of the circular arena was at (0,0). The center point was determined by taking the midpoint between the CS+ and CS− locations; the x and y coordinates of the CS+ and CS− were recorded manually. We then converted each Cartesian coordinate to polar coordinates in order to rotate each arena so that the CS+ location was at the top of the arena and the CS− was at the bottom. We took the distance of the CS+ to the center as a reference radius of 1, and normalized all coordinates to this radius. We then filtered out any points that had a radius equal to or greater than 1.3 to remove tracking errors. Note that for the conditioning dataset, only the x and y coordinates of a fly per frame were recorded. For the test dataset, the x and y coordinates per frame were recorded, but also the identity of the fly across frames (tracking data are available from the Dryad Digital Repository, https://doi.org/10.5061/dryad.77hs873; Sehdev et al., 2019).

Visit probability maps were generated only for the memory test dataset. For every individual fly, we divided the arena into 20×20 pixel bins. For each frame, we gave the pixel bin that contained the coordinate of the fly a score of 1, and gave all of the other bins a score of 0. We summed the scores of each pixel bin over all frames and then normalized by the number of frames for which the individual was tracked (the mean±s.d. of frames tracked per fly was 6157±172 for single flies, 6185±266 for pairs, 6134±289 for the groups of four flies and 6167±387 for the groups of eight flies). For each group size, we then took the mean of each pixel bin over all individual fly tracks. We used the same analysis for the time-binned visit probability maps by looking only at the frames that occurred during the time bin.

where ‘Preference index’ refers to the conditioning phase and ‘Conditioned preference index’ refers to the memory test phase. We calculated the mean index for the entire experimental run and for 1-min time bins. A preference index of 0 indicates that an equal number of flies were at the CS+ and the CS−, whereas a value of 1 indicates that all of the flies were in the half of the arena containing the CS+ and a value of −1 indicates that all of the flies were in the half of the arena containing the CS−.

We calculated the mean relative latency per experimental run.

For the memory test dataset, we used the rotated Cartesian coordinates to calculate the Euclidean distance between every fly in each frame of the experiment. We divided the distances into bins of 5 mm and counted the occurrences of each distance per experimental run.

For the test dataset, we selected flies according to their group size, and randomly sampled entire fly tracks from different experimental runs. We simulated as many experimental runs as there were real experimental runs, and we also simulated as many flies as were in each experimental run. We then overlaid these tracks and calculated the Euclidean distance between flies for the simulated experiments as we did for the real experiments.

We defined an encounter as the center of one fly being a maximum of two fly lengths (5 mm) away from the center of another fly. We calculated the Euclidean distance between all flies as before. We selected the distances that were less than or equal to 5 mm (the encounter distances). Because we had the identity of every fly per experimental run, we could calculate the number of encounters for every fly and the length of these encounters. For the mean encounter number per fly per experimental run, we calculated the total number of encounters for one experimental run, multiplied it by two as there were two flies involved in each encounter, and then divided it by the number of flies in the arena. For the mean encounter length per experimental run, we summed the length of all encounters and divided by the total number of encounters per experimental run. We repeated this analysis for the simulated data.

For all data analysis, R version 3.5.0 was used (https://www.r-project.org/). All statistics were performed using Bayesian data analysis, based on Korner-Nievergelt et al. (2015). We chose Bayesian analysis over frequentist statistics as it allows us to (1) estimate the probability with which the means of two groups differ and (2) determine the 95% credible intervals within which the true mean of a group lies. Note that the frequentist confidence interval does not allow such a straightforward interpretation.

To investigate the effect of group size on flies' preference for the CS+ (Figs 1D and 2A,B) and distance to the CS+ (Fig. S1A–D), we fitted a linear model. The group size (single, pair, small group and large group) was used as the explanatory variable, with the large group as the reference level. To test for differences between group sizes within a time bin for the preference index during conditioning, we used a linear model with the group size as the explanatory variable; each bin was modeled individually. To test for differences between different group sizes within a time bin during the test and for the distance to the CS+ during conditioning, we used a linear model with the group size, the time bin and the interaction between the two as explanatory variables. We log-transformed the distance to the CS+ during conditioning for both the whole recording and the time-binned recording to ensure residuals were normally distributed. The mean value per experimental run was used as the response variable. We used an improper prior distribution (flat prior) and simulated 100,000 values from the posterior distribution of the model parameters using the function ‘sim’ from the package ‘arm’. The means of the simulated values from the posterior distributions of the model parameters were used as estimates, and the 2.5% and 97.5% quantiles as the lower and upper limits of the 95% credible intervals. The mean and credible intervals of the distance to the CS+ during conditioning were back-transformed for plotting.

We used this linear model to compare the preference index values of each group size against chance (preference index of 0). For each group size, we calculated the proportion of simulated values from the posterior distribution that were larger than 0. If the proportion of simulated values was greater than 0, flies preferred the arena half containing the CS+ over the arena half containing the CS− (represented by filled circles in plots).

To test for differences between different group sizes, we calculated the proportion of simulated values from the posterior distribution that were larger for one group compared with another group. We declared an effect to be significant if the proportion was greater than or equal to 0.95 (*). Proportions greater than or equal to 0.99 are marked ‘**’ and greater than or equal to 0.999 marked ‘***’. We performed this analysis for the whole recording and for the different time bins, and we compared the preference index and conditioned preference index of the different group sizes within a single time bin, not between them.

To test whether the relative latency (Fig. S1E) depends on group size, we used a linear model as before. The relative latency was the response variable, and the group size was used as the explanatory variable. We used the same methodology as previously to simulate values from the posterior distribution and generate the means and the 95% credible intervals. To test for differences, we calculated the proportion of draws from the posterior distribution for which the mean of each draw was smaller in the experimental dataset than the mean of each draw of the simulated dataset.

To investigate whether grouped flies differ in the number of their inter-fly encounters from random (simulated data), we used a linear model for each distance bin. The number of occurrences of that distance was the response variable, and the type of data (experimental or simulated data) was used as the explanatory variable. We used the same method as specified above to test for differences.

To investigate whether the encounter number and lengths were different to random (simulated data), we used a linear model with either encounter number or encounter length as the response variable, and the type of data (experimental or simulated data) as the explanatory variable. We used the same method as specified above to test for differences.

To determine whether the mean encounter number per fly differed between group sizes, we randomly assigned pairs of experimental and simulated encounter numbers from different experimental runs for each group size (Fig. 2F). For each pair, we then subtracted the simulated encounter number value from the real encounter number value (difference between encounters). This allowed us to compare between group sizes, as by removing the simulated value, we remove the number of encounters that could be due to chance, which is positively correlated with group size. To test for differences between group sizes, we used a Poisson generalized linear model. We added an offset of 55 to all calculated differences between encounters so that all counts were positive, and removed 55 before plotting. The response variable was the ‘difference between encounters’. The explanatory variable was the different group size (pair, small group and large group). The large group was used as the reference level. We used the same method as specified above to draw inferences about the differences between the large group and the other two group sizes.

To investigate whether group size affects associative odor–food memory in flies, we developed an automated assay to condition four groups of flies simultaneously (Fig. 1A,B, Movie 1). This assay allowed us to transfer flies from one experimental phase to another without anesthesia and with minimal mechanical disturbance, which can alter fruit flies' behavior (Barron, 2000; Bartholomew et al., 2015; Trannoy et al., 2015). We compared four different sized groups of flies: one fly (single), two flies (pair), three or four flies (small group) and seven or eight flies (large group) (Fig. 1C). During the conditioning, two odorants (2,3-butanedione and ethyl acetate) were presented at opposite sides of the arena; one odorant (CS+) was paired with dried sucrose and the other odorant was not (CS−). Each odorant was used equally often as the CS+ and the CS−, and the data were pooled. This procedure minimizes non-associative effects of the conditioning, such as odorant-specific changes in hedonic value, generalization or sensitization (Quinn et al., 1974).

To determine whether flies approached the odorant–sucrose patch, we counted the number of flies in the half of the arena containing the odorant–sucrose patch, in order to calculate a mean preference index for each experimental run (see Materials and Methods). Flies of all group sizes showed a higher preference for the arena half containing the odorant (CS+)–sucrose patch than for the arena half containing the CS− (Fig. 1D). From the second minute to the end of the conditioning, the probability that flies preferred the arena half containing the odorant (CS+)–sucrose patch over the arena half containing the CS− was above 0.999 across all groups, implying that they were feeding on the sucrose (see Table S1 for the Bayesian probabilities comparing between groups). To confirm that the preference index reliably measures flies' preference, we additionally calculated the mean distance to the CS+ per experimental run (Fig. S1A,B and Table S1). This measure revealed similar behaviors as the preference index, showing that flies of all group sizes approached the CS+ and remained within 10 mm of its center from the second minute onwards (Fig. S1B).

Flies were transferred from the conditioning to the test by rotating the platform (Fig. 1B). In between conditioning and test there was a pause, where the platform was rotated to a neutral segment where there were no odorants or sucrose present. The pause lasted 5 s, and then platform was rotated to the test segment. During the test, the positions of the CS+ and the CS− were switched and there was no sucrose, thus flies could not rely on remembering the location of the sucrose patch (Kim and Dickinson, 2017) and had to follow the olfactory CS+ to search for the expected food. To visualize the conditioned preference index during the test, we plotted the flies' trajectories within the arena and calculated the probability across flies to visit a particular pixel bin (Fig. 1E). The increased visit probabilities around the CS+ persisted over the entire 7 min of the test, confirming that flies learned to associate the CS+ with food (Fig. 1E).

We next asked whether group size affects the expression of associative odor–food memory during the test and determined the conditioned preference index for the CS+ (Fig. 2A). During the test, flies of all group sizes preferred the arena half containing the CS+ over the arena half containing the CS− [P(all group sizes>0)>0.999], showing that flies had formed an associative odor–food memory (Fig. 2A). Flies of the large group showed a higher conditioned preference for the CS+ than the pair and the single fly [P(large group>pair)=0.95, P(large group>single)=0.99], and the small group showed a higher conditioned preference than the single fly [P(small group>single)=0.95; see Table S1]. Flies showed a similar group size dependence of their conditioned responses when we measured the distance to the CS+ (Fig. S1C,D and Table S1). There were no differences between groups in the latency to arrive at the CS+ relative to the latency to arrive at the CS− (Fig. S1E).

To investigate the time course of memory expression, we calculated the conditioned preference index over 1-min time bins (Fig. 2B). During the first minute, there were no differences of the conditioned preference between any of the group sizes (Table S1). Between the second and seventh minutes, flies from the large group showed higher conditioned preference than single or paired flies throughout most bins tested (Table S1). In all minute bins, the conditioned preference was higher than chance for all group sizes, except for the single flies in the seventh minute [P(conditioned preference index>0)=0.861]. The distance to the CS+ (Fig. S1D, Table S1) revealed similar group-size-dependent differences in the conditioned approach behavior. These results suggest that the expression of an associative odor–food memory increases in strength and duration with increasing group size.

The extended odor–food memory expression in grouped flies indicates that social interactions extend the expression of an odor–food memory. If the extended odor–food memory expression in grouped flies depends on social interactions, then the group size should affect the frequency of social interactions. To assess whether group size affects the frequency of social interactions, we measured the number of inter-fly distances and compared it with the number of expected random distances. We calculated the distances between all flies conditioned in groups in each video frame during conditioning to see whether they approached each other. To determine whether these distance distributions could be explained by flies randomly encountering each other in the arena, we simulated 30 new experimental runs by randomly sampling fly locations from all experimental runs for each video frame. We then calculated the distances between these simulated groups of flies in each video frame (Fig. S2A). For the large group, there were more short inter-fly distances (0–5 and 5–10 mm) than for the simulated group.

We chose a distance of 5 mm between fly centers as a threshold for inter-fly encounters where flies could potentially socially interact. Flies in the small and large groups made more encounters (approached each other by 5 mm or less) than the simulated groups of flies, but not the pair [P(large group>simulated large group)>0.999, P(small group>simulated small group)=0.996, P(pair>simulated pair)=0.917; Fig. 2C–E].

To compare encounter number across group sizes, we needed to correct for trivial differences in encounters that are just due to differences in the group sizes (in larger groups there is a higher chance for random inter-fly encounters). We corrected for these differences in encounter number by the following procedure: we randomly took an experimental run from the experimental and simulated datasets and subtracted the number of encounters between the two experimental runs (Fig. 2F). By subtracting the number of encounters in the simulated runs, we removed the number of encounters per experimental run that could be due to random encounters. The encounter number was higher for the large group compared with the small group and the pair [P(large group>pair)>0.999, P(large group>small group)=0.987]. There were no differences in encounter length between any group size and their simulated groups (Fig. S2B).

The increased number of encounters in the larger group indicates that flies are more attracted to each other when they are in large groups than when they are in small groups or pairs. More encounters allow more opportunities for social interactions between flies, which in turn could underlie the longer associative memory expression of the large group as compared with smaller groups or single flies.

We developed an automated learning and memory assay for walking fruit flies that allows analysis of the behavior of individual flies while they forage, learn and memorize odor–food associations alone or in groups. The strength and duration of odor–food memory expression increased with group size, and flies in larger groups were more attracted to each other than flies in smaller groups. These data suggest that social interactions can increase the efficiency of memory-guided foraging fruit flies.

Fruit flies accumulate on fermenting fruit, which they find by following both odorants released by fermenting fruit (Becher et al., 2012; Kellogg et al., 1962; Semmelhack and Wang, 2009) and aggregation pheromones released by male (Bartelt et al., 1985; Lin et al., 2015; Mercier et al., 2018) and female conspecifics (Lebreton et al., 2017). During foraging, primer flies explore the environment and appear to signal the location of favorable food patches to other flies (Tinette et al., 2004). Thus, social interactions increase foraging efficiency in fruit flies.

Our finding of extended expression of associative odor–food memories in groups, together with the positive correlation between group size and inter-fly attraction, suggests that flies also benefit from social interactions during odor memory-guided foraging. Flies could have interacted via olfactory stimuli (Jallon, 1984; Keesey et al., 2016; Lebreton et al., 2017; Lin et al., 2015), gustatory stimuli (Schneider et al., 2012), sound (Tauber and Eberl, 2003), substrate-borne vibration (Fabre et al., 2012) and touch (Ramdya et al., 2014), but not via visual stimuli, because the experiments were performed in the dark. The positive correlation between group size and inter-fly attraction that we found is in line with a previous study where inter-fly attraction was higher in larger than in smaller groups (20–40 versus 10 flies) (Simon et al., 2012). To our knowledge, such an increase in inter-animal attraction with increasing group size has not yet been reported in vertebrates (Miller and Stephen, 1966).

The positive relationship between associative odor–food memory and group size could be a result of social interactions during the learning of the odor–food association (during conditioning) or during the retrieval of the odor–food memory (during the memory test).

During conditioning, the presence of other flies at the food patch could increase the reinforcing strength of the food because the presence of other flies indicates that the food patch is good. Indeed, flies prefer food sources with other flies present over food sources without any flies (Lihoreau et al., 2016; Tinette et al., 2004). Alternatively, physical contact with other flies could act as an additional appetitive reinforcing stimulus during odor learning, similar to a honey bee acting as a positive reinforcer for another honey bee during odor conditioning (Cholé et al., 2019).

Besides being a learning effect, the extended odor–food memory expression in grouped flies could result from social interactions during the memory test. The probability of localizing the CS+ could increase with group size simply because inter-fly attraction could average out individual errors in finding the CS+ (‘many wrongs’ principle; Simons, 2004). Alternatively, individual flies that have failed to learn the association between the CS+ and the sucrose reward could decide to follow their neighbors. Theoretical modeling suggests that such a confidence-dependent variation in individual search strategy increases odor localization performance in groups (Torney et al., 2009).

We found a positive relationship between group size and the strength and duration of odor–food memory expression. However, our experimental design does not allow conclusions on whether this extended memory expression results from being in the group during odor–food learning (conditioning) or during odor memory-guided search (memory test). To discriminate between these two possibilities, one could test whether flies conditioned in a group and tested alone (or conditioned alone and tested in a group) still show extended memory expression compared with control flies that were conditioned and tested in the same group size.

The automatic learning and memory assay presented here could help reveal external factors (e.g. fly density, the ratio of informed to uninformed flies) and internal factors (e.g. sex, metabolic, genetic or circadian states) that influence learning and memory expression in social contexts. Importantly, this assay would allow study of the neural basis of social effects on foraging by disentangling sensory processing and memory formation. To identify the sensory bases of information transmission between flies, one could test the effect of temporarily perturbing their ability to smell, see and mechanosense by expressing a temperature-sensitive switch for synaptic transmission in defined neuron populations (Kim et al., 2012; Kitamoto, 2001; Ramdya et al., 2014). Likewise, neuronal perturbation experiments would help identify the neurons that encode the valence of social information (Fernandez et al., 2017) and reveal how these neurons integrate with the neurons that encode the hedonic and caloric value of food (Huetteroth et al., 2015). Moreover, to investigate whether information transmission during foraging is affected by the fly's predisposition to forage, one could use the two naturally occurring foraging gene Drosophila mutants. ‘Rovers’ move more during foraging and demonstrate improved short-term memory, whereas ‘sitters’ move less and show an improved long-term memory (Mery et al., 2007; Osborne et al., 1997). Because both foraging and aversive memory expression are affected by social context (Kohn et al., 2013), experiments using these morphs would help to assess the genetic bases of social effects on odor–food learning and memory expression.

We thank C. Giovanni Galizia for discussions initiating this research, Stefanie Neupert for programming the acquisition software and comments on the manuscript, Jana Hörsch for contributing to pilot experiments and comments on the manuscript, Charles Ellen for comments on the manuscript, two anonymous reviewers for their helpful comments, and FIBERFLON (Konstanz) for donating Teflon-coated fiberglass fabric.