Introduction

Dopamine has many functions in the brain. Among them, two turned out to be highly conserved throughout the animal kingdom: to carry “teaching” signals during associative learning and to initiate and control movement^1–5. In mammals and humans, it is widely accepted that these two functions are brought about by different populations of dopaminergic neurons (DANs): effects regarding learning are thought to be elicited by the mesolimbic pathway, effects regarding movement by the nigrostratiatal pathway⁴. Here, we study in the larva of the fruit fly Drosophila melanogaster whether in contrast the same individual central brain-DANs can induce both functions regarding memory and movement.

The fruit fly larva is an established model organism to study the neuronal and genetic mechanisms underlying associative learning^6–7. Due to the numerical simplicity of its brain and the available toolbox for transgene expression, the larva allows investigating these mechanisms at tractable levels of complexity. In D. melanogaster distinct sets of DANs convey appetitive and aversive unconditioned stimuli (US), respectively (larvae^6–7; adults^2,8) (a similar scenario may be emerging in vertebrates, too^1,9–13). Each of the DANs synapses onto a different compartment of the parallel fibres of Kenyon cells (KCs) that form a higher-order brain structure called mushroom body (larvae⁷; adults¹⁴). Odours are combinatorically coded across the array of these KCs which in turn give output to mushroom body output neurons (MBONs). The MBONs collect information across the KCs and send it towards efferent circuitry, promoting either approach or avoidance (larvae⁷; adults¹⁴). Within the KCs the coincidence of odour presentation and activity in DANs is detected and enables a memory trace to be formed. The respective memory engram is thought to come by synaptic depression between the KCs and the MBONs of the same mushroom body compartment, such that the net effect of the activity across the MBONs is tilted towards approach or avoidance, respectively (larvae^15–16; adults^17–18). Electron microscope reconstructions further revealed a direct within-compartment connection from the DANs onto the MBONs (larvae¹⁹; adults²⁰). Thus, each DAN has two main targets: the KCs and the MBONs of its own compartment. The general organization of this DAN-KC-MBON matrix is likely shared with other insects^2,21–23.

Three pairs of identified DANs of the dorso-lateral cluster 1 (DL1) have been shown to support aversive learning upon optogenetic activation during Pavlovian conditioning^15,24–25. In this study, we ask whether those same DANs have additional functions. Specifically, we ask for their role for safety learning – that is, learning that a cue predicts the absence of a threat; memory retrieval – that is, whether a memory will be retrieved and affect behaviour; and modulations of acute, innate locomotion. Little is known about how the dopaminergic system controls any of these functions – if at all. Given the highly conserved functionality of the dopaminergic system across the animal kingdom, this study will reveal more about how the manifold functions of the dopaminergic system are organized – in any brain.

Results

We focus our study on the DANs that support aversive learning when optogenetically activated, called DAN-d1, DAN-f1 and DAN-g1. We investigated the transgenic expression pattern of the driver strains covering each DAN, and the effects of activating each DAN on learning, memory retrieval and locomotion.

The used driver strains drive transgene expression in identified DANs

Previous studies had described the expression patterns of the used split-driver strains in the central nervous system in detail^15,25–27. Here, we extended the view on the expression within the whole body. Towards this end, we applied a recent protocol for whole-body visualizations²⁸. We confirmed the highly specific transgene expression in the brain of the four split-Gal4 driver strains without any detectable expression in the rest of the body (Fig. 1A-D, Supplementary material S1). TH-Gal4 is known to cover almost all DANs in the brain, with the notable exception of the neurons of the primary protocerebral anterior medial (pPAM) cluster²⁶. We confirm this pattern of transgene expression and in addition found strong expression outside of the brain, in particular in the proventriculus and in hypodermal melanin-producing cells (Fig. 1E, Supplementary Movie 1). This pattern fits to the reported function of dopamine in the hardening and darkening of the insect cuticle^29–30 and a recent report of TH-Gal4 expression in the adult proventriculus³¹. This result means that any genetic manipulation using TH-Gal4 may affect cells outside of the central nervous system. Finally, we confirm that R58E02-Gal4 covers three of the four pPAM DANs that are missing in the TH-Gal4 expression pattern²⁷, with no expression outside of the central nervous system (Fig. 1F).

Figure 1:

Some individual DANs establish and retrieve associative memories

In larvae, for many kinds of aversive US learned behaviour is only observed if the US is present at the moment of memory retrieval^32–36. For example, we trained groups of larvae with an odour as the conditioned stimulus (CS) and quinine as US and observed an aversive associative memory when we tested the animals in presence of quinine, but not in its absence (Fig. 2A). Learned aversive behaviour therefore can be viewed as an escape from the US, guided by the respective memory – an escape that is pointless in absence of anything to escape from^37–40.

Figure 2:

Figure 2 - supplement 1:

Figure 2 - supplement 2:

Figure 2 - supplement 3:

We then asked for the roles of the DL1-DANs for both establishing and retrieving associative memories by employing the same learning experiment, but instead of quinine we used optogenetic DAN activation by blue light as a virtual US. We also tested the innate odour preference of all genotypes, both in darkness and in blue light (Fig. 2 – Figure supplement 1) and observed no evidence that the innate odour preference was modified by activating the respective DANs.

After training with DAN-d1 as virtual US, zero memory scores were observed when the animals were tested in darkness. When tested in blue light, however, the animals of the experimental genotype, but not the controls, showed aversive associative memory (Fig. 2B, Fig. 2 – Figure supplement 2A-B). Interestingly, a recent study using a much stronger light intensity during training observed no significant memory with DAN-d1 as virtual US²⁵. We therefore repeated the experiment with two different light intensities and observed an aversive memory only when using the low intensity that we use throughout the study, but not with a 10-fold higher intensity (Fig. 2 – Figure supplement 2C). Moreover, when raising the larvae in food supplemented with all-trans-retinal which had been shown to increase the effectiveness of our effector ChR2-XXL⁴¹, no aversive memory could be detected (Fig. 2 – Figure supplement 2D). Thus, it seems that a too strong activation of DAN-d1 prevents the establishment and/or retrieval of an aversive memory.

For DAN-f1, we likewise found an aversive memory that was retrieved only when DAN-f1 was activated also during the test (Fig. 2C, Fig. 2 – Figure supplement 2E). Our results indicate that both DAN-d1 and DAN-f1 are sufficient to establish an aversive memory which requires the respective DAN to be activated for retrieval.

A previous study used the same driver strains crossed to UAS-CsChrimson as the optogenetic effector in a similar learning experiment. Critically, learned behaviour was observed also in darkness¹⁵. Using the same effector and the same intensity of red-light stimulation we were able to replicate these results (Fig. 2 – Figure supplement 3). Thus, the memories induced by either effector differed in the conditions of how they could be retrieved. These results call for caution against using different optogenetic effectors interchangeably.

Some DANs establish memories but do not control retrieval

Activation of DAN-g1, too, established an aversive memory (Fig. 2D). However, in contrast to DAN-d1 and DAN-f1, we found aversive associative memory scores in the experimental genotype independent of the testing condition (Fig. 2D).

Next, we employed two driver strains that each covers a combination of the DANs, and activated them using ChR2-XXL. One of them covers DAN-f1 and DAN-g1, the other one, TH-Gal4, covers all three of them, plus many additional cells (Fig. 1E). In both cases, we observed associative memories in the experimental genotype both in absence and in presence of the US (Fig. 2E-F, Fig. 2 – Figure supplement 2F-G).

Notably, in the experiments with these driver strains, we also observed significantly aversive memory scores in the genetic controls when tested in light (Fig. 2D-F). Given that this result was found both in driver and effector controls, it was most likely caused by the light itself. Light is known as an aversive US in larvae⁴² and has been seen to affect the results of optogenetic experiments before^24,43. Why this aversive effect of the light remained a non-significant trend in the previous experiments (Fig. 2B-C) is not known.

Taken together, activating DAN-g1, or driver strains that include DAN-g1, established aversive memories that were retrieved independent of whether the respective DANs were activated during the test.

Individual DANs are sufficient for establishing safety, but not punishment memories

Animals can associate stimuli not only with the presence of a threat, but also with its absence^44–45. Such safety memory has been found in D. melanogaster, but received little attention by the community until recently (larvae^40,46–48; adults^49–50). That means a US, such as bitter taste, can establish two different memories: punishment memory after paired training and safety memory after unpaired training^37,46 (Fig. 3A). Critically, our results so far demonstrate that each of the individual DANs was sufficient to establish an associative memory – but not, whether this was a punishment memory, a safety memory, or both. Therefore, we next focussed on the animals’ odour preference after paired and unpaired training, respectively, compared to the odour preference of the pooled genetic controls as a baseline. A punishment memory would reveal itself by decreased odour preferences after paired training, a safety memory by increased preferences after unpaired training, compared to the baseline. To further bolster this analysis, we included the data of all independent replications of the experiments (Fig. 2 including Figure supplements 2 and 3).

Figure 3:

For the case of DAN-d1, after paired training there was no difference in odour preference between the genetic controls and the experimental genotype, nor between the two testing conditions (Fig. 3B, left). After unpaired training, in contrast, the larvae of the experimental genotype exerted increased odour preferences when tested under blue light compared to all other groups (Fig. 3B, right). This suggests that the activation of DAN-d1 established no punishment memory, but a safety memory. For DAN-f1, we found analogous results (Fig. 3C). In the case of DAN-g1, after paired training we found more negative odour preferences when tested in presence than in absence of light, but no difference between the controls and the experimental genotype (Fig. 3D, left). This suggests a potential punishment memory by the light, but not by the activation of DAN-g1. After unpaired training, the animals of the experimental genotype showed higher odour preferences than the controls, both in darkness and light (Fig. 3D, right).

Taken together, we found that all three DANs induced only a safety memory after unpaired training, but no punishment memory after paired training (confirming similar results using CsChrimson¹⁵). The safety memory established by DAN-d1 and DAN-f1 were retrieved only if the respective neuron was activated during the test, whereas the safety memory established by DAN-g1 was, at least partially, also retrieved without neuronal activation.

Combinations of DANs are sufficient for establishing both safety and punishment memories

We next wondered whether the activation of a combination of DANs would be sufficient to induce a punishment memory. For the combination of DAN-f1 and DAN-g1, after paired training, we indeed found more negative odour preferences in the experimental group than in the genetic controls, irrespective of the testing condition (Fig. 3E, left). After unpaired training, we found higher odour preferences in the experimental genotype than in the genetic controls (Fig. 3E, right). Finally, with TH-Gal4 we observed analogous results (Fig. 3F).

Taken together, activating both DAN-f1 and DAN-g1, or activating most DANs in the brain, did establish both a punishment and a safety memory. Both kinds of memory were retrieved irrespective of whether the DANs were activated in the test or not.

Activating the DANs covered by TH-Gal4 modulates the animals’ bending and velocity

Next, we investigated which effect activating the DL1-DANs would have on locomotion.

Using the novel IMBA (Individual Maggot Behaviour Analysis) software for behavioural analyses, we recently had found strong locomotor effects upon optogenetic activation of the DANs covered by TH-Gal4⁵¹. Within the current study, we repeated this experiment in total five times, in two different laboratories and with different parameters (Fig. 4 – Figure supplement 1A). This gave us the opportunity to probe the behavioural effects upon DAN activation for their robustness across experimental repetitions. Groups of around 8 experimentally naïve larvae of either the experimental genotype or the driver or effector controls were allowed to crawl freely on an agar-filled Petri dish in darkness. After 30 s, blue light was activated for 30 s, followed by darkness. We video-recorded the animals’ behaviour and analysed it afterwards with the IMBA software (Fig. 4A).

Figure 4:

Figure 4 - supplement 1:

Figure 4 - supplement 2

Figure 4 - supplement 3:

The IMBA software provides a rich variety of behavioural attributes for individual animals. We selected 17 attributes (for a list and descriptions, see table 1) and first asked how these attributes relate to each other. Towards this end, we performed a Spearman correlation analysis of all 17 attributes based on the animals’ baseline behaviour in darkness. We found three clusters of inter-correlated attributes (Fig. 4B). Cluster 1 consists of attributes measuring different aspects of directional changes, e.g. the animals’ bending, or the rate of lateral head movements, the so-called head casts (HC). All attributes of cluster 2 relate to the speed of the larvae’s peristaltic movement. Finally, cluster 3 consists of attributes dealing with the directional bias of the animals’ behaviour. For example, the bending bias measures to which extent an animal was bent more towards one side than another.

Table 1:

Then, we wondered how much each attribute would change upon light stimulation. Towards this end, we calculated a Δ-value of each attribute as an individual’s behaviour during the second half of the light stimulation minus its baseline behaviour during darkness (Fig. 4A). We used only the second half of the stimulation to exclude any potential startle-like response to the sudden light. Then, we used a random forest classification algorithm to classify individual larvae between the genotypes based on the 17 Δ-attributes. Crucially, the random forest provides the relative importance of each attribute for the classification. In other words, it can be used to obtain an unbiased measure of how relevant each attribute is for discriminating the behaviour of different genotypes⁵¹. We found consistently the attributes of cluster 1 and 2 to be much more important than those of cluster 3 (Fig. 4C). Within each cluster, most attributes were on average similarly important, with strong variability both within and across data sets (Fig. 4 – Figure supplement 1).

We concluded that the main behavioural differences between the experimental genotype and the genetic controls were the amount of bending and directional changes (cluster 1) as well as peristaltic speed (cluster 2). Given the correlations of individual attributes within each cluster (Fig. 4B) as well as their variable importance (Fig. 4 – Figure supplement 1), for all following analyses we decided to focus on one representative attribute of each cluster 1 (bending) and cluster 2 (tail velocity) that we have used already in a previous study⁵¹. In addition, we will present the relative importance of each attribute cluster along with each experiment.

Activating different subsets of DANs modulates locomotion in opposite directions

In accordance with our previous results⁵¹, we found that the control genotypes increased their bending and decreased their tail velocity upon light stimulation (Fig. 4D, Supplementary Movie 2). Such responses to blue light have been reported before⁵² and were consistently seen throughout this study (Fig. 4 – Figure supplement 2). Both types of behavioural change were significantly enhanced in the experimental genotype in which the TH-Gal4-DANs were activated⁵¹ (Fig. 4D-F). More specifically, when the DANs were activated, the larvae kept high bending and low velocity throughout the 30 seconds of light stimulation whereas the controls gradually reverted towards their normal behaviour (Fig. 4D). These results were confirmed by a repetition of the experiment using larger Petri dishes and multiple light stimulations (Fig. 4 – Figure supplement 3). For the random forest, the attributes from cluster 1 and 2 were equally important (Fig. 4G, Fig. 4 – Figure supplement 3).

As TH-Gal4 covers all DANs except those of the pPAM cluster, we next asked whether activation of the pPAM DANs would have a similar effect on locomotion. Towards this end, we used R58E02-Gal4 which covers 3 out of 4 pPAM DANs (DAN-h1, DAN-i1 and DAN-j1)²⁷. To our surprise, activation of these DANs decreased bending and increased tail velocity compared to the controls (Fig. 4H-K, Supplementary Movie 3). More precisely, after an initial reaction to switching on the light, when the pPAM DANs were activated the larvae reverted to their normal behaviour much quicker than the controls (Fig. 4H). Again, the attributes of cluster 1 and 2 were similarly important for the random forest classification (Fig. 4L).

Thus, the activation of two non-overlapping sets of DANs had opposite effects on behaviour: activating the TH-Gal4-DANs (including the aversive US-signalling DANs) increased bending and decreased velocity, whereas activating the R58E02-Gal4-DANs (which were shown to signal appetitive US)^27,47,53 decreased bending and increased velocity.

DANs modulate locomotion independent from the light responses

In the previous experiments, animals of all genotypes displayed a very strong response to the light stimulation. We therefore wondered whether the observed effects upon DAN activation were due to a modulation of this light response, or an independent modulation of the animals’ locomotion.

We compared the animals’ behavioural changes in three sets of experiments, activating the DANs with strong blue light, weak blue light, or red light, respectively (Fig. 5). When using the about 10-fold lower light intensity, we nevertheless observed rather strong light responses in the controls (Fig. 4 – Figure supplement 2), and robust differences in behaviour upon activation of either set of DANs (Fig. 5 – Figure supplement 1). These effects gradually diminished with prolonged light stimulation (Fig. 5 – Figure supplement 2).

Figure 5:

Figure 5 - supplement 1:

Figure 5 - supplement 2:

Figure 5 - supplement 3:

Figure 5 - supplement 4:

Red light reportedly triggers only very mild light responses^16,52. We therefore next used the red-shifted effector CsChrimson⁵⁴, and raised larvae either with or without all-trans-retinal which is necessary for the Chrimson channels to function⁵⁴ (Fig. 5 – Figure supplement 3). When the larvae were raised with retinal, we observed similar effects as when using ChR2-XXL (Fig. 5 – Figure supplement 3). The control without retinal, in contrast, displayed only very weak responses to the red light (Fig. 5 – Figure supplement 4).

In order to compare the effects of DAN activation between the three experiments, we defined the behaviour of the pooled controls in each experiment as zero, and displayed the difference of each experimental group from their respective controls. The behavioural effects of the DAN activations were strikingly similar in the three conditions, both in magnitude and time course (Fig. 5A-B). The only obvious difference was an extended reduction of the tail velocity beyond the activation of the TH-Gal4-DANs that could be seen only when using ChR2-XXL (Fig. 5A). This “after-effect” may be due to the reportedly slow closing times of the ChR2-XXL channel⁴¹.

We wanted to further quantify the relation between the light responses elicited in each experiment and the behavioural effects of the DAN activation. Therefore, we used the mean behavioural changes upon light stimulation of the pooled controls of each experiment as an indicator for the effect size of the light response, and the difference in behavioural changes between the experimental groups and the respective controls as an indicator for the effect size of the DAN activation. We found that the effect sizes of the DAN activation were all relatively similar and completely unrelated to the effect sizes of the light responses (Fig. 5C).

In summary, we concluded that the observed modulations of bending and tail velocity upon DAN activation occurred independent of the animals’ responses to the light stimulation itself.

Different subsets of DANs convey opposite acute valence when activated

Next, we asked how to interpret the behavioural effects of activating DANs. Stopping and turning around upon a sudden activation of TH-Gal4-DANs can be interpreted as a reaction to an aversive signal^16,52 – on the other hand, slowing down and turning on the spot could also be a behavioural strategy to stay in a pleasant situation. Likewise, going straight and speeding up upon R58E02-Gal4-DAN activation can be interpreted as a “go on”-reaction to an appetitive signal or a strategy to escape. To test for the valence of activation of either set of DANs we performed choice experiments between a dark and an illuminated half of a Petri dish (Fig. 6A). Larvae of all genotypes avoided the illuminated half as expected^42,52. Relative to the genetic controls, larvae with TH-Gal4-DANs activated showed more avoidance, and larvae with R58E02-Gal4-DANs showed more approach (Fig. 6B-C, Supplementary Movies 4-5). We next analysed the animals’ behaviour near the dark-light-border. When activating TH-Gal4-DANs, the animals’ bending was increased and their velocity decreased on the illuminated half compared to the controls (Fig. 6D-F), similar to our previous experiments (Fig. 4). When activating R58E02-Gal4-DANs, the animals’ bending was decreased and their velocity increased compared to the controls everywhere on the dish (Fig. 6G-I). Taken together, we concluded that activating TH-Gal4-DANs conveys an acute aversive signal and R58E02-Gal4-DANs conveys an acute appetitive signal. The animals’ reaction to the sudden activation of these DANs therefore should be interpreted as aversive and appetitive behaviour, respectively.

Figure 6:

The modulations of bending and velocity are dopamine-dependent

Several DANs have been found to have additional neurotransmitters^19,55. Therefore, we wondered whether the observed effects on locomotion were indeed dependent on dopamine signalling. Towards this end, we combined optogenetic activation by ChR2-XXL with TH-RNAi which knocks down the expression of tyrosine hydroxylase, the rate limiting enzyme of dopamine synthesis. Thus, when both effectors are expressed in a set of DANs, these DANs can get activated by light but have a reduced dopamine synthesis. When expressing both effectors in the TH-Gal4-DANs, we found that the change in bending upon DAN activation was slightly decreased compared to animals without the TH-RNAi, but still higher than in the genetic controls (Fig. 7A-B, Fig. 7 – Figure supplement 1A-F). Furthermore, in animals with TH-RNAi the tail velocity was generally increased, but the change in tail velocity upon DAN activation was unaffected (Fig. 7A,C, Fig. 7 – Figure supplement 1A-F). There are several possible reasons for this lack of effects, including a too weak knock-down of tyrosine hydroxylase or a developmental adaptation to the reduced dopamine levels. Therefore, we next sought to inhibit tyrosine hydroxylase by an independent and acute technique.

Figure 7:

Figure 7 - supplement 1:

Figure 7 - supplement 2:

Larvae were fed the tyrosine hydroxylase inhibitor 3IY for four hours before the experiment, following established protocols^56,57. 3IY-fed larvae showed a change in bending upon DAN activation that was significantly lower than control animals without 3IY and on the same level as genetic controls (Fig. 7E-F, Fig. 7 – Figure supplement 1G-M). Interestingly, the baseline tail velocity was decreased rather than increased compared to animals without 3IY feeding (Fig. 7E, Fig. 7 – Figure supplement 1K). This means inhibiting the activity of tyrosine hydroxylase acutely or during the whole development had opposite effects on the baseline tail velocity. Upon DAN activation, 3IY-fed animals reduced their tail velocity less than animals without 3IY, and to a level comparable to the genetic controls (Fig. 7E,G, Fig. 7 – Figure supplement 1G-M). These effects, both regarding the changes in bending and tail velocity as well as the shifted baseline velocity, were replicated in two independent experiments (Fig. 7 – Figure supplement 2).

Taken together, we concluded that the locomotor effects upon activation of the TH-Gal4-DANs were largely dopamine-dependent, at least when inhibiting the tyrosine hydroxylase acutely with 3IY.

Individual DANs are sufficient to increase bending and decrease velocity

Given the broad expression pattern of TH-Gal4, we wondered which cells were responsible for the observed modulations of bending and velocity. Therefore, we probed the DANs of the DL1 cluster individually with experiments that included two cycles of light stimulation (Fig. 8 – Figure supplement 1). Overall, we found similar differences between the experimental genotype and the controls upon both stimulations, but more enhanced during the second stimulation.

Figure 8:

Figure 8 - supplement 1:

Figure 8 - supplement 2:

For DAN-d1, we found that the experimental genotype displayed an increased bending and decreased tail velocity compared to the controls that remained trends for the first stimulation (Fig. 8 – Figure supplement 1A-C) and were statistically significant for the second stimulation (Fig. 8A-D). The effect on bending seemed to be somewhat weaker than for TH-Gal4 and expressed itself in a delayed peak rather than a general increase. The tail velocity, in contrast, was reduced to zero in both cases (compare Fig. 4D-G and 8A-D). However, using weaker light did not result in any significant differences between the experimental genotype and the control, in contrast to the case of TH-Gal4 (Fig. 8 – Figure supplement 2A-C). Thus, activating DAN-d1 individually has similar effects as activating all neurons covered by TH-Gal4, but two stimulations of strong light were required to fully induce these effects.

Next, we repeated the same experiments for DAN-f1 and DAN-g1. We found no effects on locomotion upon activating DAN-f1 (Fig. 8E-G, Fig. 8 – Figure supplement 1D-F, Fig. 8 – Figure supplement 2D-F). For DAN-g1, we found both significantly increased bending and decreased velocity very similar to DAN-d1, again only for strong light and clearer during the second activation, just like for DAN-d1 (Fig. 8H-L, Fig. 8 – Figure supplement 1G-I, Fig. 8 – Figure supplement 2G-I). Finally, we activated DAN-f1 and DAN-g1 simultaneously and found increased bending and decreased velocity comparable to the activation of DAN-g1 alone (Fig. 8M-P, Fig. 8 – Figure supplement 1K-M, Fig. 8 – Figure supplement 2K-M).

Taken together, our experiments demonstrated that two of the three tested DANs, when activated individually, induced qualitatively similar effects on bending and velocity that also the combination of almost all DANs induced.

Discussion

In this study, we found that the optogenetic activation of the same individual DANs can have multiple effects in establishing associative memories, retrieving memories and acutely modulating innate locomotion (Fig. 9A). How can such diverse functions be integrated into the same neuronal circuit?

Figure 9:

Figure 9 - supplement 1:

A working model for the establishment and retrieval of punishment and safety memories

All the DANs of the mushroom body share a similar circuitry (Fig. 9B). They receive various input outside of the mushroom body neuropil, both from post-sensory interneurons and feedback pathways from the mushroom body output^15,19,25. The two major downstream partners of each DAN are the KCs of the mushroom body as well as the MBONs innervating the same compartment as the DAN^6–7,16,19 (Fig. 9 – Figure supplement 1A-C). The behavioural effects of the DANs are most likely executed via the MBONs, as they are the only obvious connection from the DANs to motor control¹⁶. Most evidence, both from larval and adult Drosophila, suggest that memories are established via the DAN→KC synapses (see below). We suggest that, in turn, the DAN→MBON synapses may gate the memory retrieval (Fig. 9B). Notably, these two kinds of synapses are not spatially segregated. Rather, DAN→KC, DAN→MBON and KC→DAN synapses form several dense clusters on the DAN, with the DAN→KC and DAN→MBON synapses more to the centre, and the KC→DAN synapses more on the outside of each cluster (Fig. 9 – Figure supplement 1D-E). In many cases, one cluster contains synapses with several different MBONs, and several clusters contain synapses with the same MBON (Fig. 9C). We had demonstrated the same clustered pattern already for DAN-i1⁴⁶. The potential function of this arrangement remains speculative, but it may indicate that these clusters form functional units, each of them with all necessary synaptic connections for both memory formation and memory retrieval.

In the following, we use the d-compartment (lateral appendix) of the mushroom body as example. During training, the presence of an aversive US is signalled by DAN-d1 towards the mushroom body. The coincidence of dopaminergic input and activation of the KC by an odour during paired training is detected by the type I adenylate cyclase and results in presynaptic depression of the KC→MBON synapses (larvae^6–7; adults^58–59). This leads to less drive towards approach-promoting MBONs and hence to odour avoidance (Fig. 9D) (larvae¹⁶; adults^18,60–62). For the case of the larval d-compartment, MBON-d1 was shown to be indeed approach-promoting, MBON-d2 and d3 did not show any behavioural effect, and MBON-p1 has not been tested so far¹⁶. Thus, the KC→MBON-d1 synapses are the best candidates for harbouring a memory established by DAN-d1.

Presenting an odour unpaired from DAN activation can lead to potentiation of the KC→MBON synapses¹⁷ and thus to odour approach (Fig. 9E). The details of this process are not well understood. We assume that during training signals from the DAN can have opposite effects on the KC→MBON synapse, depending on the activity state of the KC: if the DAN signal coincides with KC activation by an odour, the KC→MBON synapse is depressed (Fig. 9D), otherwise it is potentiated (Fig 9E). This drives avoidance or approach towards the odour, respectively. In the case of DAN-d1, only the odour approach after unpaired training, indicating safety learning, was statistically significant (Fig. 3B).

During the test, the presence of the US is signalled by DAN-d1 and provides a gating signal. This signal is most likely received by the MBONs which are known to be necessary for the expression of learned behaviour (larvae^16,53; adults^18,63–67). Such a gating signal in the simplest case could be a logical ‘AND’ gate: for learned behaviour to occur, both the changed drive from the KC onto the MBON and the DAN gating signal to the MBON are required. The consequence is that any associative memory established by the modified KC→MBON-d1 synapses can only take effect behaviourally if DAN-d1 is activated also during the test, just as we observed (Fig. 2B).

How the mushroom body circuit can account for the modulations of acute locomotion

So far, very little is known about how the mushroom body-innervating DANs could modulate innate locomotion. We consider two simple scenarios to explain our observations, a reduced velocity and increased bending upon activation of DAN-d1 (Fig. 8A-D): (1) the DAN-d1→MBON-d1 synapse acts inhibitory such that an activation of DAN-d1 silences the approach-promoting MBON-d1 (Fig. 9F); (2) the signal from the DAN-d1→KC synapse reduces the drive from the KCs onto the MBON-d1 (Fig. 9G). In both cases, the net output of the mushroom body would be tilted towards aversion.

Scenario 1 seems straight-forward, but two studies in adults caution against such a simple model: First, it was found that activating a DAN while pharmacological blocking output from the KCs activated the MBON of the respective compartment rather than inhibiting it²⁰. So, we would need to assume that the “sign” of the DAN→MBON synapses differs either between developmental stages, or between different mushroom body compartments, or even changes in certain conditions. Different dopamine receptors can either excite or inhibit a neuron⁶⁸, and many DANs likely release additional neurotransmitters^19,55,69. Thus, it remains possible that activation of a DAN causes either activation or inhibition of the downstream MBON dependent on the exact composition of receptors.

Second, a recent study compared the innate valence induced by optogenetically activating DANs (of the PAM cluster) and MBONs, and found that in the case of DANs valence was brought about largely by differences in speed in darkness or light, but in the case of MBONs by modulations of turns at the dark-light border⁶⁹. This discrepancy makes a simple, direct inhibition unlikely.

Scenario 2 resembles the effects of the DAN signal in paired learning (Fig. 9D,G). However, the observed modulations of locomotion occurred within a few seconds and therefore much faster than in our learning experiments. In addition, some activation of the KCs and therefore some drive of the MBONs would be required, otherwise an acute down-regulation of the KC→MBON synapse would be without any effect. To which extent the KCs were active during our experiments is no known.

Previous studies provide conflicting evidence for the involvement of the KCs: on the one hand, a recent study in adults found that KC activity was dispensable for the innate attraction to PAM-DANs⁶⁹. On the other hand, the same study found that knocking down the Dop2R receptor in the KCs reduced the innate attraction to PAM-DANs⁶⁹. In addition, a study in larvae found an increased locomotion when the Dop1R1 receptor was knocked-down in the KCs⁷⁰.

Considering these discrepancies, it is very likely that both scenarios are too simplistic, by ignoring for example asymmetric body-side effects of DAN and/or MBON activities⁴⁷, or cross-compartmental connections between DANs and MBONs¹⁵. It can also not be excluded that DANs exert their acute locomotion effects independent of the MBONs. Future research has to disentangle the neuronal mechanisms of how DAN activity modulates innate locomotion and valence.

Each DAN induces a different combination of effects

Strikingly, each DAN in this study (as well as in a previous study regarding the appetitive domain) induced a different set of behavioural effects (Fig. 9A).

The optogenetic activation of DAN-d1 and DAN-f1 can have two effects with respect to memories: during training, the virtual US signal they confer can induce a safety memory. During testing, this memory is only retrieved when the same DAN is activated (Fig. 2B-C). This mimics many natural US such as tastants or mild mechanical disturbance^32–36, or formic acid in adult flies⁷¹. We propose that this mechanism serves to ensure that memories only influence behaviour when it is adaptive to do so. In contrast, the activation of DAN-g1, as well as of two different combinations of DANs (which both happen to include DAN-g1), establish memories that can be retrieved also without the respective DANs being activated (Fig. 2D-F). This mimics US like electrical shocks that induce memories that are retrieved in absence of the US^72–73. It has been shown that memories in larvae can be specific for the kind of US and are retrieved independently³⁹. It is conceivable that memories regarding relatively harmless types of US, like a bad taste, require the US to be present in order to be retrieved and guide behaviour. Memories regarding more severe US which threaten the body integrity of the animal, such as electrical shocks, are retrieved without the actual US being present. The activation of DAN-d1 or DAN-f1 on the one hand, and DAN-g1 on the other hand, may correspond to such harmless and dangerous US, respectively.

In a previous study we found that the appetitive US-signalling DAN-i1 can establish memories of opposite valence when it was activated paired or unpaired with an odour⁴⁶. This suggested that both kinds of memory rely on the activity of the same DANs. In the aversive domain, DAN-d1, DAN-f1 and DAN-g1 each establish only safety memory but not punishment memory, whereas activating combinations of DANs did establish both kinds of memory¹⁵ (Fig. 3). The reason for these discrepancies is not known yet.

Most strikingly, DAN-d1 and DAN-g1 (but not DAN-f1) also modulate innate locomotion (Fig. 4,8). Interestingly, recent studies in mammals suggested that the nigro-stratiatal ‘movement’ pathway also plays a role for carrying reward signals, in addition to the traditional mesolimbic ‘reward’ pathway⁷⁴. Together, these results point towards a neuronal organization in which the same dopaminergic neurons can have multiple functions that are as diverse as instructing learning processes and modulating movements.

Our results suggest a combinatorial-like neuronal organization with each neuron having a unique, yet overlapping set of functions. This raises a question: given that all mushroom body compartments share the same basic circuitry¹⁹ (Fig. 9B), how can such functional differences be explained? It is conceivable that they depend on the specific properties of the involved MBONs or its synapses with the KCs and the respective DANs. For example, if the synapses between the KCs and a given MBON are weak to begin with (for example the MBONs of the d- and f-compartment), they can be easily potentiated by unpaired training, but any depression by paired training will be minute. Such weak depression may need to happen in several compartments before it becomes behaviourally measurable.

Moreover, the impact of the gating signal by the DAN to the MBON may vary dependent on the circumstances: the memories established by DAN-d1 or DAN-f1 bypassed the retrieval control by the DANs if those memories were established with CsChrimson, but not with ChR2-XXL (Fig. 2 – Figure supplement 3)¹⁵. The memory stablished by DAN-g1 bypassed this control even when using ChR2-XXL (Fig. 4D). Little is known about the molecular mechanisms of memory retrieval in MBONs, and future research has to uncover which factors may determine its dependence on dopamine signals.

Our study revealed that individual DANs can induce surprisingly diverse effects, from establishing and retrieving memories to modulating innate locomotion. The mushroom body circuit – that is shared with adult flies and other insects – provides an excellent study case to advance our understanding of the complex functions of the dopaminergic system. Given the highly conserved roles of DANs in mediating “teaching” signals and modulating movement across the animal kingdom including vertebrates and humans, a deeper understanding of this circuit and its functions should provide new insights about the principles of DAN function in general.

Materials & Methods

Animals

Third-instar feeding-stage larvae (Drosophila melanogaster), aged 4-5 days after egg laying, were used throughout. Flies were maintained in mass culture at 25 °C, 60-70 % relative humidity and a 12/12 hour light/dark cycle. Two kinds of food medium were used:

Food medium #1 was used for all learning and olfactory experiments as well as for some of the locomotion experiments. Per 1 l water, 45 ml sugar beet molasses, 45 ml malt, 147.5 g cornmeal, 10 g soy flour, 18.75 g dry yeast, 6.25 g agar and 2.5 g methyl-4-hydroxybenzoate (Merck, Darmstadt, Germany) were added as described in detail previously⁷⁵.

Food medium #2 was used for most of the locomotion experiments as well as the olfactory stimulation preference experiments. Per 1 l water, 100g glucose, 50 g cornmeal, 31.65 g wheat germ, 40 g dry yeast, 8 g agar, 2 ml propionic acid (Kanto-kagaku, Tokyo, Japan) and 5 ml 10% butyl p-hydroxybenzoate (Wako, Osaka, Japan) in ethanol were added as described in detail previously⁷⁶.

We took a spoonful of food medium from a food vial, randomly selected the desired number of larvae, briefly rinsed them in tap water and started the experiment.

We used either wild-type larvae from the Canton Special genotype, or transgenic larvae to express light-gated ion channels in the respective DAN. Towards this end, one of two effector strains, UAS-ChR2-XXL (Bloomington Stock Center #58374), or 20xUAS-CsChrimson-mVenus (Bloomington Stock Center #55134), was crossed to either TH-Gal4 (Bloomington Stock Center #8848), R58E02-Gal4 (Bloomington Stock #41347), or to one of four split-Gal4 driver strains: MB328B, MB054B, SS02180, or SS01716¹⁵ to obtain double-heterozygous offspring. As driver controls the driver strains were crossed to a local copy of w¹¹¹⁸ (Bloomington Stock Center #3605, 5905, 6326). As effector control a strain carrying the landing sites used for the split-Gal4 (attP40/attP2), yet without a Gal4 domain inserted (“empty”)⁷⁷, was crossed to the respective effector. When using TH-Gal4, we used w¹¹¹⁸ instead. As CsChrimson requires feeding of all-trans-retinal⁵⁴, larvae that were crossed to 20xUAS-CsChrimson-mVenus were raised in standard food medium that additionally contained 400 µM final concentration of all-trans-retinal (CAS: 116-31-4, Sigma-Aldrich). For the TH-RNAi experiments, a double effector strain was used, made from the UAS-ChR2XXL strain and y[1]v[1];P{TRiP.JF01813}attP2 (Bloomington Stock #25796) to express a double-strand-RNA element of the ple gene⁷⁸.

All larvae used for optogenetic experiments were raised in constant darkness. For the whole-body anatomy, we crossed each driver strain to a y[1] w*;;MB247-mCherry-CAAX, UAS-mIFP-T2A-HO1 double-effector strain²⁸ (called MB247-mCherry-CAAX, UAS-mIFP in the main text). As driver controls we crossed each driver strain to y[1] w*;;MB247-mCherry-CAAX²⁸, as effector control we crossed y[1] w*;;MB247-mCherry-CAAX, UAS-mIFP-T2A-HO1 to y[1] w* (Bloomington Stock Center #1495).

Experimental setup

For learning and preference experiments, larvae were trained and tested in Petri dishes of 9 cm inner diameter (order #82.1472, Sarstedt) filled with 1 % agarose (electrophoresis grade; CAS: 9012-36-6). For the experiments displayed in Fig. 2A and 3A, 5 mM quinine (purity: 92%, CAS: 6119-70-6, Sigma-Aldrich) was added into the agarose. As odour, we used n-amyl acetate diluted 1:20 in paraffin oil (AM; CAS: 628-63-7, Merck). For locomotion experiments, either Petri dishes of 15 cm inner diameter (order #82.1184.500, Sarstedt) filled with 1 % agarose were used, or dishes of 9 cm inner diameter (order #3-1491-01, AS ONE) filled with 2 % agar (CAS: 9002-18-0, Fujifilm Wako Pure Chemical Industries).

Experiments were performed inside a 43 x 43 x 73 cm surrounding box equipped with a custom-made light table featuring a 24 x 12 LED array (Solarox) and a 6 mm thick diffusion plate of frosted plexiglass on top to ensure uniform blue light for ChR2-XXL activation (470 nm; 120 µW/cm²; for Fig. 2 – Figure supplement 3C, Fig. 4 incl. Figure supplements, Fig. 7 incl. Figure supplements, Fig. 8 and Fig. 8 – Figure supplement 1: 1000 µW/cm²) or red light for CsChrimson activation (630 nm; 430 µW/cm²). Petri dishes were placed directly on top of the diffusion plate. For additional details regarding the experimental setup, see⁵³.

Associative learning

Learning experiments followed established protocols⁷⁹. Odour containers were prepared by adding 10 µl of odour substance into custom-made Teflon containers (5 mm inner diameter with a lid perforated with 7 holes of 0.5-mm diameter each). Petri dishes were covered with modified lids perforated in the centre by 15 holes of 1 mm diameter each to improve aeration.

For paired training with quinine as US, about 20 larvae were placed on a quinine-filled Petri dish equipped with two AM-filled odour containers. After 2.5 min, the animals were transferred to a dish with plain agarose and empty odour containers (EM) on which they spent another 2.5 min. This cycle (AM-/EM) was repeated two more times before the animals were transferred to a test Petri dish with AM on one and EM on the other side. This dish was either a plain agarose dish, or one with quinine added, as mentioned in the results section. After three minutes, the number of animals on either side as well as on a 1 cm-wide middle stripe were counted and the odour preference was calculated as:

Thus, preference values are constrained between 1 and −1 with positive values indicating a preference for and negative values indicating avoidance of AM. Note that the training sequence in every other repetition of the experiment was reversed (EM/AM-)

For unpaired training, AM was presented on plain agarose Petri dishes and EM was present on quinine Petri dishes, such that the CS and US were experienced explicitly separated from each other (AM/EM-or EM-/AM). Otherwise, the training was performed analogously, and the final test was performed as described.

From each one paired and one unpaired trained cohort of animals we calculated an associative Memory score as:

Thus, the values can range from 1 and −1 with negative values indicating that the larvae preferred the odour less after paired than unpaired training. Thus, negative scores indicate an aversive associative memory. Positive values, in turn, would indicate appetitive associative memory.

Learning experiments with ChR2-XXL or CsChrimson as effectors were performed analogously. However, all Petri dishes were filled with plain agarose, i.e. no tastants were added. Rather, for paired training, AM was paired with continuous blue (ChR2-XXL) or red light stimulation (CsChrimson) for 2.5 min, followed by 2.5 min of darkness without an odour. For unpaired training, animals received odour presentation and light stimulation separately from each other. The test was performed as described, either in darkness or with light stimulation as mentioned in the results section.

Innate olfactory behaviour

Odour containers and Petri dishes were prepared as described above. Cohorts of about 20 larvae were collected from the vial, briefly washed in tap water and placed to a Petri dish with an AM container on one side and an empty container on the other side. After 3 minutes, the innate odour preference was determined according to equation (1).

Locomotion experiments

Cohorts of about 8 larvae were placed on the middle of a Petri dish filled with agarose or agar. The dish was placed in the experimental chamber and video-recording was started after about 10 s. After 30 s of video-recording in darkness, light was switched on for 30 s followed by 60 s of darkness. In some experiments, 2 or 3 light stimulation cycles were applied. Blue or red light was used as mentioned along with the results.

Experiments involving feeding 3IY followed established protocols⁵⁷. Before the experiment the larvae were transferred to a yeast paste with or without added 3-Iodo-L-tyrosine (3IY; stored at −20°C; CAS: 70-78-0, Sigma-Aldrich) at a concentration of 5 mg/ml. After 4 hours, cohorts of 8 larvae were collected from the yeast paste and the experiment was conducted as described.

Optogenetic stimulation preference experiments

In order to test for the animals’ preference for having sets of DANs activated, we prepared Petri dishes filled with agar of which one half was illuminated by blue light, the other half was kept in darkness. Cohorts of about 10 larvae were placed on the dark half of the dish about 1.5 cm from the midline. Their behaviour was video-recorded for 3 minutes before the animals were counted and the preference was calculated as:

Thus, preference values are constrained between 1 and −1 with positive values indicating a preference for and negative values indicating avoidance of the light.

Video analysis

Video analysis used the IMBA (Individual Maggot Behaviour Analyser) software and followed previous procedures⁵¹. In brief, the positions of head, tail, as well as several spine and contour points of each larva were tracked over the course of the experiment. A combination of computational modelling and statistical approaches allowed keeping the identities of larvae across collisions and thus the analysis of individual animals tested in cohorts. Then, we discarded all tracks that were shorter than half of the time window used for the analysis. This prevented multiple tracks from the same individual animal from entering the analysis.

Larval behaviour was separated into head-casting (HC) phases and run phases. Lateral HCs are a very prominent feature during larval chemotaxis. We quantified these head movements based on the angular speed of the head vector through the front third of the animal. HCs were detected when the angular speed exceeded ± 35°/s. Runs were determined as periods in which the animals were not performing HCs, omitting 2 s before and after each HC event to prevent behavioural effects of the HC. We noticed that the rhythmic velocity changes of the tail point provided a reliable indicator for the occurrence of one peristaltic forward movement. We therefore defined each local maximum in the tail forward velocity as one step, and the period between two maxima, i.e. one complete peristaltic cycle, as an inter-step (IS) period. Since the local-maxima detection algorithm we employed also produced false maxima during intervals of little or no forward motion, we discarded maxima whenever the tail velocity in forward direction was less than 0.6 mm/s.

After the detection of runs, steps and head casts, we calculated a number of behavioural attributes. We summarize all attributes in this study in table 1. Whenever the average behaviour of a group of larvae is displayed over time, the mean and the 95 % confidence intervals of the respective attribute are calculated per 1 s time bin with a 5 s convolution filter. Whenever data are displayed in box plots to compare across experimental conditions, the mean of each attribute is calculated for each individual larva across the indicated time period. As we were interested in the difference in a given attribute X before and during light stimulation, we calculated Δ-values as the difference in X during the last 15 s of light stimulation minus X during the 15 s just before light stimulation (called baseline in the main text).

We used random forest models which are based on classification and regression trees^80–81. The random forest algorithm works by building N number of decision trees and then predicting the class label of an instance using a majority voting. Each tree is trained on a random subspace of the data, only using a sample of the instances. To evaluate the random forest classification, we used five-fold cross validation, which was repeated three times. Five-fold cross validation means that the test set for the model is taken from five different, disjunct sets of the data set. The data set is then shuffled and the procedure is repeated three times. Cross validation is used to make sure that the model is not based merely on the selected data, but can explain the full data set.

For the applications of the random forest in this study, we pre-selected 17 behavioural attributes to be included (Table 1). The pre-selection was based on our previous study⁵¹ and was warranted to avoid using largely redundant attributes and to exclude non-behavioural attributes such as the size of the larvae. The random forest was always applied to pairwise compare two experimental conditions, typically the experimental groups with one of the genetic controls. In order to provide a good estimate of the importance of each attribute, the complete random forest model (including the cross validation) was performed 10 times for any given comparison. In the typical case, this means 10 repetitions of the comparison between the experimental genotype with the driver control and 10 repetitions of the comparison with the effector control. In each case, the relative importance rank of each attribute was noted.

Whole-body anatomy

Experimental procedures follow previous work²⁸. In brief, third-instar larvae were first bleached (4% sodium hypochlorite, order #9062.1, Roth) for 10 min at room temperature. After washing (3 × 5 min in distilled water), they were fixed in 4% PFA in 0.1 mol/L PB at pH 9, with gentle shaking overnight at 4°C. Fixed samples were briefly rinsed 3 × with 0.1 mol/L PB containing 0.2% Triton X-100 (PBT), then washed 2 × 60 min and left overnight at 4°C in PBT. Dehydration the following day used a graded ethanol (EtOH) series (60 min in 10% and 25% EtOH, followed by 30 min in 50%, 60%, 80%, and 2 × 100% EtOH). Samples were then cleared by replacing EtOH with ethyl cinnamate (ECi; ethyl 3-phenyl-2-propenoate; order #112372-100G, Sigma-Aldrich). After 30-60 min, the ECi was refreshed once, and the samples stored at room temperature in black boxes in a desiccator.

The samples were placed in ECi-cleared phytagel blocks (1 × 1 × 1 cm, P8169, Sigma-Aldrich) before imaging on an Ultra-Blaze microscope (Miltenyi Biotec). The microscope was equipped with an FIU-15 white light laser (NKT Photonics) for excitation through triple-sheet optics to illuminate samples from one side. The following excitation and emission filters were used: autofluorescence: exc. 480/20, em. 515/30, mcherry: exc. 590/20, em. 625/30, mIFP: exc. 680/20, em. 720/30.

Tiled image stacks were acquired with either a 12× objective with zoom set to 1 or 4x objective with zoom set to 1.66 (both Miltenyi Biotec) with a format of 2048 × 2048. Using ImSpector software (version 7.3.6, Miltenyi Biotec), a 10% overlap of tiles was set for stitching.

Image files were processed with Imaris software (version 10.1, Bitplane), including file conversion, stitching, further processing, and rendering. Either the ortho slicer, oblique slicer or clipping plane tool was used to restrict volumes in the z direction to improve the representation of structures that were otherwise covered in the context of the whole body. All inserts were generated using the snapshot function in Imaris. Whole-body 2D maximum-intensity projections were generated in Fiji⁸². Supplementary Movie 1 was produced in Imaris with the key frame animation tool; Adobe Premiere Pro 2020 (version 14.9.0, Adobe) was used for cutting and labelling.

Dendrogram analysis

To visualize neuron morphology and synapse locations, we employed neuron dendrograms⁸³ that are linearized, topologically correct 2D representations of 3D neuron reconstructions with synapse annotations (based on ¹⁹). Synapse locations and the distances between them are approximately correct, with slight adjustments to avoid overlap and cluttering of the synapse symbols⁸³.

We formally analyzed the spatial organization of synapses based on geodesic distances along the neuron’s branches (“cable length distances”). The geodesic distance between two synapses was determined by finding the shortest path between them on the neuron, and then computing the cumulated Euclidean distance, in the 3D coordinate space of the neuron, between points along that path.

In particular, we analyzed spatially local clusters of synapses and the distances to the centroid of the cluster. To this end, we clustered the synapses on each neuron based on geodesic distances and using hierarchical clustering with Ward’s criterion⁸⁴ (implemented in the base R package “stats”). The maximum number of clusters per neuron was set to k=10. For analysis, we kept only clusters with at least 5 presynaptic and 5 postsynaptic KC connections.

The centroid of a cluster was defined to be the most central synapse, computed by partitioning around medoids clustering (PAM with k = 1 cluster)⁸⁵ based on geodesic distances.

Quantification and statistical analysis

Two-tailed non-parametric tests were used. Values were compared across multiple groups with Kruskal-Wallis tests (KW). In case of significance, subsequent pair-wise comparisons used Mann-Whitney U-tests (MW). To test whether values of a given group differ from zero, we used Wilcoxon Signed-Rank tests (WSR). When multiple comparisons were performed within one analysis, a Bonferroni-Holm correction was applied to keep the experiment-wide error rate below 5 %⁸⁶. The detailed results of all statistical tests are reported in the Supplementary Data S1 along with the source data. We present our data as box plots which represent the median as the middle line and 25 %/75 % and 10 %/90 % as box boundaries and whiskers, respectively. Outliers are not displayed. Experimenters were blind to genotype. Sample sizes (biological replications) were chosen based on previous studies revealing moderate to mild effect sizes^47,51 and are displayed in the figures.

Data availability

Source data to all behavioural experiments are included as supplementary material (Supplementary Data S1).

Acknowledgements

This study received institutional support by the Otto von Guericke Universität Magdeburg, the Wissenschaftsgemeinschaft Gottfried Wilhelm Leibniz (WGL), the Leibniz Institute for Neurobiology (LIN), the Hokkaido University, and the RWTH University Aachen. This study received grant support from the Deutsche Forschungsgemeinschaft (DFG) (GE 1091/4-1, and FOR 2705 Mushroom body), the Japan Society for the Promotion of Science (JSPS) (Kakenhi 22K20673) and the German-Israeli Foundation for Science (GIF) (G-2502-418.13/2018). Experimental contributions of J. Chen, R. Khondaker, T. Meyer, S. Schuller, M. Thane, L. Warzog, B. Fischer, technical assistance by M. Dombach, M. Paisios, M. Thane and F. Unterstab, as well as discussions with C. Eschbach, B. Gerber, C. König, N. Mancini, N. Tanaka, A. Thum and A. Yarali are gratefully acknowledged. We thank C. Eschbach, C. König, A. Thum, D. Weber and M. Zlatic for sharing fly strains and results prior to publication.

Additional information

Author contributions

Conceptualization: Michael Schleyer, Naoko Toshima.

Formal analysis: Michael Schleyer, Arman Behrad, Franziska Behnke.

Investigation: Naoko Toshima, Arman Behrad, Franziska Behnke, Gauri Kaushik, Aliće Weiglein, Martin

Strauch, Juliane Thoener, Oliver Kobler, Maia Lisandra M. Wang, Markus Dörr.

Data curation: Michael Schleyer, Naoko Toshima, Arman Behrad, Aliće Weiglein, Franziska Behnke.

Writing - original draft: Michael Schleyer, Naoko Toshima, Aliće Weiglein, Franziska Behnke.

Writing - review & editing: Naoko Toshima, Aliće Weiglein, Arman Behrad, Franziska Behnke, Gauri Kaushik, Martin Strauch, Juliane Thoener, Oliver Kobler, Maia Lisandra M. Wang, Markus Dörr.

Visualization: Michael Schleyer, Martin Strauch, Juliane Thoener, Oliver Kobler.

Supervision: Michael Schleyer, Juliane Thoener.

Project administration: Michael Schleyer.

Funding acquisition: Michael Schleyer.

Competing financial interests

The authors declare no competing financial interests.

Additional files

Supplementary Data S1. Raw data and results of statistical tests. This excel file provides all raw data underlying the figures. Each tab shows the data of one figure or supplementary figure, with the results of all statistical tests to the left.

Supplementary Material S1. Full body expression of all used driver strains and genetic controls. Each page shows the three channels (autofluorescence, mCherry, mIFP) for the experimental genotype (driver strain crossed to MB247-mCherry-CAAX, UAS-mIFP) and a driver control lacking the UAS construct for one driver strain. On the first page, in addition the effector control crossed to yw is displayed. Next to the whole body, the insert shows a volume-restricted magnification of the CNS region. Blue arrowheads indicate DAN cell bodies. Scale bars represent 250 µm for the full body and 50 µm for the inserts.

Supplementary Material S2. High-resolution dendrograms. Each page shows the dendrogram of one DAN in high resolution. For clarity, the synapses with KCs and other synapses are displayed separate. Even pages highlight only synapses with MBONs, odd pages synapses with KCs. All other synapses are represented with transparent grey symbols. Circles display synapses from the given DAN to other neurons, triangles synapses from other neurons to that DAN. The colour code is given on each page.

Supplementary Movie 1. An animation through the data set used in Fig. 1E. TH-Gal4 was crossed to MB247-mCherry-CAAX, UAS-mIFP, and both the mCherry-CAAX (magenta) and the mIFP (yellow) signals are shown. The movie starts with the larva’s head towards the front left. The second half shows the animal from top. At the end, the z-layers covered by the proventriculus and the melanin-producing cells were removed to allow the cells of the CNS to become visible. Grid spacing: 200 µm.

Supplementary Movie 2. A sample video of the experiment presented in Fig. 4 with TH/ChR2-XXL larvae. The video is played in double speed. Blue light (not visible in the video) was switched on at timestamp 15 s (30 s in the real experiment) and off at timestamp 30 s (60 s in the real experiment). Larval heads are marked in green, tails in red.

Supplementary Movie 3. A sample video of the experiment presented in Fig. 4 with R58E02/ChR2-XXL larvae. The video is played in double speed. Blue light (not visible in the video) was switched on at timestamp 15 s (30 s in the real experiment) and off at timestamp 30 s (60 s in the real experiment).

Supplementary Movie 4. A sample video of the experiment presented in Fig. 6 with TH/ChR2-XXL larvae. The video is played in double speed. The right half of the dish was illuminated with blue light (not visible in the video) throughout the experiment.

Supplementary Movie 5. A sample video of the experiment presented in Fig. 6 with R58E02/ChR2-XXL larvae. The video is played in double speed. The right half of the dish was illuminated with blue light (not visible in the video) throughout the experiment.