Fly stocks were raised on standard cornmeal-agar medium at 18°C, 60% relative humidity, and under a 12 h light–dark cycle. Flies were generated as described by Vasmer et al. (2014). A Hsp70-FLP insertion on the X-chromosome (provided by G. Struhl) was combined with the GH146-Gal4 (Stocker et al., 1997) insertion on the second chromosome to generate a strain homozygous for both P-element inserts. These flies were crossed with a strain carrying the UAS-FRT-CD2(Stop)-FRT-mCherry-dTRPA1 (Vasmer et al., 2014; Pooryasin and Fiala, 2015) construct with the insertion balanced over CyO. Female F1 offspring younger than 2 days were anesthetized using CO₂ and transferred to a fresh food vial. Unless otherwise indicated, flies were incubated at 30°C for 4 h to induce FLP-mediated expression. Figure 1B illustrates how this causes stochastic heat-shock-induced expression of mCherry-dTRPA1 in neurons determined by the Gal4 line.

Flies were trained as described by Vasmer et al. (2014), schematically illustrated in Figure 1C. Female flies aged 4–6 days old were transferred into pre-warmed (30°C) tubes covered on the inside with an electrifiable copper grid. Training was performed in an illuminated incubator at 30°C and at an air humidity of ∼60%. Animals were kept in these tubes for 2 min, during which time 24 90 V DC electric shocks of 1.25 s each in duration were applied, separated by 3.75 s breaks, for a total shock interval of 5 s. Control animals were treated in a similar manner; that is, they were exposed to a temperature of 30°C in the same tubes but did not receive the electric shocks. Subsequently, the flies were transferred to a heat-gradient chamber (schematically depicted in Figure 1C, right) that consisted of an aluminum block with eight walking tracks (275 mm length × 5 mm width × 4 mm height) covered with a Plexiglas lid. The entire apparatus was kept in an incubator under a constant white light, and at a temperature of 16°C and ∼65% humidity, producing a linear and stable temperature gradient over the length of the arena ranging from 18 to 35°C. Individual flies were transferred without anesthesia into the walking tracks through small holes in the lid and were permitted to move freely for 20 min. Locomotion was monitored from above using a high-definition video camera (Panasonic HC-V500). Flies were tracked using the Noldus Ethovision XT 8.5 software (Wageningen) to generate data used in the analysis. The temperature preference of each fly over the observation period of 20 min was determined in 5 min time bins as the time spent above or below 24°C. The flies were anesthetized immediately after the behavioral experiments and their brains were dissected. Localization of mCherry-dTRPA1 expression was determined using immunohistochemistry.

Brains were dissected in ice-cold Ringer’s solution containing 5 mM Hepes (pH 7.3), 130 mM NaCl, 5 mM KCl, 2 mM MgCl₂, 2 mM CaCl₂, and 36 mM sucrose (Estes et al., 1996) and fixed for 2 h on ice in 4% paraformaldehyde dissolved in phosphate-buffered saline (PBS). Subsequently, brains were washed three times for 20 min each in PBS containing 0.6% Triton X-100 (PBST), then incubated in PBST containing 2% bovine serum albumin (block solution) for 2 h. Afterward, the brains were incubated overnight at 4°C in block solution containing mouse anti-nc82 antibody against Bruchpilot (provided by Erich Buchner) (Wagh et al., 2006) diluted 1:10. Brains were again washed three times for 20 min each in PBST and incubated for at least 4 h with goat anti-mouse 1:300 conjugated with Alexa Fluor 488-conjugated goat anti-mouse (Invitrogen). Brains were washed three times in PBST for 20 min each, washed in PBS overnight at 4°C, and embedded in Vectashield (Vector Laboratories). Images were acquired using a confocal laser scan microscope (Leica SP8) equipped with hybrid detectors and analyzed using ImageJ. The antennal lobe glomeruli were determined using anti-bruchpilot (anti-Brp) immunoreactivity and mCherry-dTRPA1 expression was characterized. The antennal lobes of both hemispheres were examined when possible.

The symmetry index was defined as the ratio between symmetrically innervated glomeruli by OPNs expressing dTRPA1-mCherry and the total number of innervated glomeruli by mCherry-dTRPA1 expressing OPNs. All statistical tests were conducted using GraphPad Prism7 and OriginPro software. A Kolmogorov–Smirnov test was used to test for a normal distribution of data. Normally distributed data were analyzed using one-way ANOVA followed by Bonferroni post hoc tests for multiple pairwise comparisons. Non-normally distributed data were analyzed using the Kruskal–Wallis test followed by Dunn’s post hoc tests for multiple pairwise comparisons. For correlation analyses, Spearman correlations were calculated. For testing for randomness of dTRPA1-mCherry expression in glomeruli a Runs test (Wald–Wolfowitz test) was conducted for all flies and for each glomerulus and each brain hemisphere independently.

To obtain expression of mCherry-dTRPA1 in stochastic subsets of OPNs (Figure 1B), flies carrying the DNA construct of a mCherry-tagged thermo-inducible cation-channel dTRPA1 (Vasmer et al., 2014; Pooryasin and Fiala, 2015) were crossed with flies carrying a flippase under control of the heat shock promoter Hsp-70 (Basler and Struhl, 1994) together with the GH146-Gal4 driver line (Stocker et al., 1997; Figure 1B). The groups of neurons that express dTRPA1 can be artificially depolarized by raising the temperature above ∼25°C (Hamada et al., 2008; Tang et al., 2013; Vasmer et al., 2014; Pooryasin and Fiala, 2015). dTRPA1 induces a relatively sharp increase in excitation in neurons expressing it, with a peak at ∼32°C (Hamada et al., 2008; Tang et al., 2013). The flippase-mediated mCherry-dTRPA1 expression is induced by heat shock within the first 2 days after hatching. To test whether the animals can associate the activity of stochastic sets of OPNs with a punishing electric shock, OPNs expressing mCherry-dTRPA1 were thermogenetically activated at 30°C and this activation was temporally paired with 2 min of electric shocks of 90 V. The shocks were 1.25 s in duration and separated by 3.75 s intervals, as described (Vasmer et al., 2014). Control animals of the same genotype were treated in the same manner but did not receive the electric shocks (Figure 1C). In a typical aversive olfactory conditioning procedure, the animals learn to avoid the odor that has been temporally paired with the punishment (Tully and Quinn, 1985). In our experiment, the stochastically distributed activity of OPNs did not reflect any real odor representation that could be tested for. To bypass any olfactory input also in the memory test phase, directly after the training procedure the animals were individually transferred to a test chamber in which they could walk freely along a temperature gradient ranging from 18 to 35°C (Figure 1C right). The movement of each animal was monitored and temperature preference during the first 5 min was used as the memory readout. The mCherry-dTRPA1 channel starts opening at ∼25°C (Vasmer et al., 2014); that is, within the preferred temperature range of naïve fruit flies (Sayeed and Benzer, 1996; Hamada et al., 2008). Therefore, the relative amount of time the animals spent >24°C, i.e., at a temperature that is still within the range of their innate temperature preference but just below the onset of dTRPA1-mediated excitation, after training was used to determine whether the animals approached or avoided re-activation of the population of OPNs expressing mCherry-dTRPA1.

First, the effect of heat shock duration during development on the number of mCherry-dTRPA1 expressing OPNs was examined: The longer and more often the heat shock was, the more neurons should express mCherry-dTRPA1. This was the case as an increase in heat exposure time after hatching gradually increased the number of mCherry-dTRPA1 expressing OPNs, as determined by quantifying mCherry-labeled somata (Figures 2A,B and Table 1). We then asked whether flies can associate activation of OPNs with a punishment, and whether this association depends on the number of activated OPNs. Therefore, four groups of flies were tested, namely, flies that did not receive any heat-induction of mCherry-dTRPA1 expression during development and thus only expressed the DNA construct in small populations of random subsets of OPNs, and flies that received a 20 min, 1 h, or 4 h heat induction of mCherry-dTRPA1 in OPNs. All groups were subjected to electric shocks simultaneous to the thermogenetic induction of neuronal activity. These groups were compared with control flies that did not receive electric shocks and with “naïve” flies that were not exposed to either increased temperature or electric shocks. Flies without heat-induction expressed mCherry-dTRPA1 in only 15.17 ± 0.64 (mean ± SEM; Figure 2B). OPNs and did not show any significant changes in temperature preference compared with control or naïve flies (Figure 2C and Table 2). Similarly, we did not find a change in the preferred temperature of trained flies that received a 20 min or 1 h induction of expression (Figures 2D,E and Table 2). However, flies expressing mCherry-dTRPA1 in random subsets of 31.1 ± 1.72 (mean ± SEM; Figure 2F) OPNs after a 4 h induction of expression during development showed a significant shift toward lower temperatures when treated with punitive electric shocks simultaneously with OPN activation (Figure 2F and Table 2). Genetic controls, i.e., the heterozygous UAS > CD2 > mCherry-dTRPA1 strain and the heterozygous Hsp-70-FLP; GH146-Gal4 strain, that received the same duration of heat shock during development and training did not show any difference in temperature preference after training compared with control and naïve animals (Figures 2G,H and Table 2). This finding indicates that the animals actively prevented reactivation of the OPNs if the activity of a sufficient number of OPNs was paired with a punishment, i.e., through associative learning. However, this learned avoidance was restricted to the first 5 min within the observation time period. At later time points the temperature avoidance was not different between test and control groups (Supplementary Figure 1).

The identity of OPNs expressing mCherry-dTRPA1 was largely stochastic and differed between the two brain hemispheres. A Runs test for each glomerulus and for each brain hemisphere independently confirmed that for most glomeruli included in the expression pattern of GH146-Gal4 mCherry-dTRPA1 expression occurred stochastically, with the exception of VL2a on theleft hemisphere and VA1ml on both hemispheres. In these two cases expression occurred more often than predicted for complete randomness (Table 3). The glomeruli innervated by the OPNs could be identified. We utilized this knowledge to test whether the identity of OPNs and their odor-specific input to the mushroom body has relevance to efficient learning. For example, a more symmetric, and therefore more unambiguous, mushroom body input could potentially be learned more efficiently; the actual number of active OPNs, and therefore the “intensity” of mushroom body input, or the innate behavioral valence of the odor signaled via the activity of distinct OPNs could potentially affect aversive associative learning. Alternatively, the function of the mushroom body might not depend on the actual source of the input. In this case, learning would not be expected to be influenced by the parameters indicated above. To distinguish among these alternatives, the thermogenetic learning experiment was repeated using 4 h of heat shock during development. Subsequently, the brains of the tested flies were removed from the head capsule, subjected to immunohistochemical staining, and the glomeruli that harbored OPNs expressing mCherry-dTRPA1 were identified. Figure 3 exemplifies how immunostaining against the active zone protein Bruchpilot (Wagh et al., 2006) was used to identify all glomeruli in comparison with stochastic mCherry fluorescence. It should be noted that some glomeruli were innervated by mCherry-dTRPA1-expressing OPNs symmetrically between the brain hemispheres, which were sometimes visible only in different confocal planes (e.g., glomerulus VC2 in Figure 3). A total of 71 flies were analyzed, 26 of which were subjected to the associated training procedure and 45 to control conditions, without electric shocks (Figure 4A). Trained animals spent significantly less time at temperatures >24°C (Figure 4B). Moreover, this effect was accompanied by an overall reduction in the speed of locomotion (Figure 4C). However, no correlation between the actual number of glomeruli innervated by the mCherry-dTRPA1-expressing OPNs and temperature preference or locomotion speed was detected in trained or control animals (Figures 4D,E). These data suggest that flies can associate this neuronal signal with an unconditioned stimulus provided that a threshold number of OPNs is reached (Figure 2F); learning efficiency as measured by memory expression was not dependent on the actual number of activated OPNs. However, thermogenetic activation of all neurons covered by the GH146-Gal4 line simultaneously with electric shocks did not lead to associative learning (Supplementary Figure 2), indicating that there is also an upper limit of how many OPNs can be activated to serve as neuronal correlate of a conditioned stimulus. Alternatively, the fact that the inhibitory anterior paired lateral (APL) neuron, which innervates calyx and lobes of the ipsilateral mushroom body, is included in the expression pattern of GH146-Gal4 (Wilson and Laurent, 2005) might prevent successful associative conditioning in this case. In fact, it has been reported that in aversive associative learning the APL neuron becomes inhibited (Zhou et al., 2019), which is precluded in our case by the thermogenetic activation. However, in those experiments that involved a heat-induced stochastic expression of mCherrydTRPA1- only 2 out of 71 tested flies showed unilateral expression in the APL neuron (Figure 4A).

Odors can influence diverse behaviors such as feeding or oviposition and can therefore act as attractive or aversive cues. Highly stereotyped connectivity from the olfactory sensory neurons to the OPNs and similar stereotyped olfactory receptor expression has enabled researchers to correlate the activity of distinct neurons with attractive or aversive valence (Semmelhack and Wang, 2009; Knaden et al., 2012), and induce either attraction or aversion via artificial activation of distinct neurons (Bellmann et al., 2010). Knaden et al. (2012) found that the activity of OPNs is more distinctly indicative for the valence of the odor-evoked behavior compared with sensory neuron activity. The OPNs innervating glomeruli DM2, DM4, DM5 responded predominantly to attractive odorants, whereas those innervating glomeruli D, DL1, and DL5 responded to aversive odorants (Knaden et al., 2012). We asked whether aversive associative learning is affected in either direction if OPNs that are activated primarily by attractive or repulsive odorants express mCherry-dTRPA1 within the overall stochastic expression pattern. Glomeruli expressing mCherry-dTRPA1 in the 71 analyzed flies were determined (Figures 4A, 5A). None of the flies showed expression in the OPNs innervating DL2. However, the remaining valence-indicative glomeruli were included in the stochastic expression patterns. No correlation was observed between the number of attraction-mediating or repulsion-mediating glomeruli in either brain hemisphere and the duration of time that the trained animals spent >24°C (Figures 5B,C). There was also no correlation between the number of attractive or aversive glomeruli and the locomotion velocity of movement after training (Figures 5D,E). Thus, our analysis did not reveal any potential influence of the valence the OPNs signal and the efficiency of aversive associative learning or memory expression. This finding is perhaps not entirely surprising considering how learned information can override naïve information, e.g., through appetitive conditioning using innately aversive odorants.

Our data did not indicate that the particular individuality of the OPNs that provide input to the mushroom body skews the efficiency of associative learning in either direction. Rather, it appeared that the mushroom body did not take the actual identity of OPN input into account. If so, the efficiency of associative learning should also be independent of whether OPNs are symmetrically activated between brain hemispheres or not. To test this hypothesis, we calculated a symmetry index (see the section “Materials and Methods”) that quantified the degree of inter-hemispheric symmetry between identified glomeruli innervated by mCherry-dTRPA1-expressing OPNs. Indeed, no correlation between the symmetry index and the time the flies spent >24°C or their locomotion speed was detected in either trained animals or controls (Figures 6A,B). This result suggests that learned avoidance behavior is independent of the degree of inter-hemispheric symmetry of glomerular activity.