Pain is so close to pleasure: the same dopamine neurons can mediate approach and avoidance in Drosophila

In mammals, dopamine is considered a central neuromodulator involved in all kinds of rewarding experiences (‘common currency’ hypothesis). In insects, the role of do-paminergic neurons in aversive stimuli was discovered before dopaminergic neurons were found to also be involved in processing appetitive stimuli. Here, we screened about 50 transgenic Drosophila lines, representing different subpopulations of dopa-minergic neurons for their ability to sustain approach or avoidance behavior, when activated optogenetically in four different operant self-stimulation paradigms. None of the lines sustain consistent behavioral valence in all experiments. Individual lines sustain approach in one experiment and avoidance in another. One line mediated strong avoidance early in the experiment and weak approach in later stages. The evidence presented here appears to contradict a ‘common currency’ dopamine function in flies. Instead, different dopaminergic neurons convey valence in a context-dependent and flexible manner, reflecting the genetic heterogeneity of the dopaminergic neuronal population.

and if these neurons are driven optogenetically in an operant place learning task, they can substitute for reward (Kremer et al., 2020). Drugs of abuse primarily target this dopaminergic system and promote debilitating loss of behavioral control once an addiction has been established.
Also in flies, dopamine neurons have been reported to be involved in valuation processes in classical conditioning experiments. Early on, Dopa decarboxylase, an enzyme required for the synthesis of dopamine, was shown to be involved in olfactory classical conditioning (Tempel et al., 1984). Silencing of DANs blocks acquisition of aversive classical memory, but leaves appetitive memory intact (Schwaerzel et al., 2003). Dopamine receptors are necessary for olfactory classical conditioning (Kim et al., 2007;Qin et al., 2012).
Optogenetic activation of DANs paired with an olfactory conditioned stimulus (CS) is able to replace sugar as the unconditioned stimulus (US) (Aso et al., 2014(Aso et al., , 2010Liu et al., 2012). In an operant olfactory conditioning paradigm in flies, optogenetic activation of TH-Gal4 DANs served as punishment and was able to induce an olfactory preference (Claridge-Chang et al., 2009). Also in Drosophila larvae, DANs appear to serve analogous functions (Lesar et al., 2021;Rohwedder et al., 2016;Schleyer et al., 2020Schleyer et al., , 2013. DANs and octopaminergic neurons also appear to be involved in value-based decision-making in flies (Gorostiza et al., 2016), adding additional evidence for the valuation-related function of DANs.
From these results and more like these, a prominent, general role of DANs in value processing was derived and DAN activity interpreted as participating in the computation of a "common currency" of value into which information from all sensory modalities are converted (Berridge and Kringelbach, 2015;Grabenhorst and Rolls, 2011;Kobayashi and Hsu, 2019;Lak et al., 2014;Landreth and Bickle, 2008;Levy and Glimcher, 2012;Matsumoto et al., 2016;Schultz et al., 2015).
It is thought that the same DANs encode, e.g., rewards from food, drink, sex or money, while separate sets encode aversive stimuli like heat, electric shock or loss.
Taken together, the current picture of DAN function in value computations entails that every animal possesses sets of DANs that encode value on an appetitive/aversive scale such that they mediate approach and avoidance behaviors, respectively. In this conceptual framework, animals should seek to activate the 'appetitive' DANs and seek to silence the 'aversive' DANs, irrespective of how the animal is allowed to do that. In the fruit fly Drosophila, we can test this hypothesis in an unbiased way: not only is there a plethora of transgenic lines available that can be used to express optogenetic constructs in specified subsets of DANs, but there is also a wealth of sufficiently different behavioral paradigms to assess the generality of any DAN subpopulation. Combining these two assets in order to screen for the DANs mediating such 'common currency', we have tested nearly 50 different fly strains in four different experimental setups which DANs they prefer to be active and which they prefer to be silent.

Fly genetics
To avoid visual cues from the stimulating light that would interfere with our reinforcement scores, we genetically blinded flies with a mutated no receptor potential A gene (norpA P24 ).
norpA encodes for the phosphatidylinositol-specific phospholipase C that is involved in several sensory pathways. The mutation abolishes vision completely. In addition, the norpA P24 allele decreases olfactory discrimination and impairs temperature discrimination at temperatures between 18°C and 26°C .
We crossed the norpA mutants with 20xUAS-CsChrimson flies to create norpA;UAS-CsChrimson flies. Virgins from this stock were always crossed with male Gal4 driver lines and the blind male offspring recorded in our experiments.

Fly strains used
In total, we used 52 different fly lines in our four different screens. Of those, 13 were tested in all four screens, another 13 in three screens, 6 lines in two screens and 19 in only one screen (Tab. 1).  Table 1. Fly lines used in which experiment. Also see  for unlisted stocks.
We targeted as many different DAN subpopulations in as many dopamine clusters as possible. Historically, there were more broadly expressing driver lines and subsequent, more recently developed narrow driver lines, focusing mostly on DANs with projection in the mushroom bodies ( Fig. 1). Flies without mushroom bodies have no problems expressing preferences and can learn most tasks quite well, except when they involve odors (Putz and Heisenberg, 2002;Wolf et al., 1998 (Liu et al., 2012;Qin et al., 2012;Schwaerzel et al., 2003;Yamagata et al., 2015). Here, we test the hypothesis that such reinforcing or punishing DAN functions generalize to situations beyond a classical US, or whether they merely serve to convey value to an external, likely olfactory stimulus (i.e, in world-learning (Brembs, 2011;  Normalized reporter expression intensity from blue (zero) to yellow (one) for each driver line used (vertical axis, see Tab. 1) and their anatomical coverage (horizontal axis; DAN clusters with their projection sites and cell body location, within brackets). A: Mushroom-body projecting driver lines obtained from (Aso et al., 2014(Aso et al., , 2010 with their corresponding expression pattern. B: Driver lines expressing more broadly than just mushroom-body projecting DANs from (Liu et al., 2012;Pathak et al., 2015). The expression pattern was estimated from (Galili, 2014;Liu et al., 2012;Pathak et al., 2015).

Fly care and reagents
Male Gal4 drivers and norpA P24 ;20xUAS-CsChrimson effector virgin flies were kept overnight on standard cornmeal and molasses food medium (Guo et al., 1996)  Screens T-maze The T-mazes are composed of a core and three removable tubes ( Fig. 2A, B). The core con-  (Table 2). LEDs were glued to a cooling plate to avoid overheating.
We performed two screens, one with red and another with yellow light ( Fig. 2A The same groups of flies were tested in two consecutive days for the screen with red light. Since the counting was done under CO², and CO² exposure may impair the flies, a day of recovery was provided before the second experiment. Experiments with the yellow light were conducted without repetition and scored in blind by two experimenters in parallel in order to obtain an estimation of the handling variability. A-B: T-maze schematics. Animals were first loaded into the entrance arm. After acclimatization, they were tapped into the elevator that transported them to the choice point. A: Red light was used for optogenetic stimulation (see Materials and Methods for details). B: Yellow light was used for optogenetic stimulation (see Material and Methods for details). C: Y-mazes schematics. The Y-mazes are illuminated from below and fly behavior is recorded from above. A PC processes the camera image online and switches the projector's red light for Y-maze arm illumination (enlarged inset) for closed-loop feedback. D: Joystick schematics. The tethered flies control the position of the platform with their legs. A sensor records platform position and a PC makes yellow optogenetic light (via a lightguide) contingent on platform position, which effectively serves as a switch for the light. Bottom: experimental design: 'training' denotes periods where the fly controls the light, 'test' denotes periods where the light is permanently switched off.

Y-mazes
The setup consisted of a Perspex block (30x35cm) containing 120 Y-mazes. The Y-mazes were backlit with an infrared LED panel (Table 2) and a diffuser to scatter light homogeneously, as previously described (Buchanan et al., 2015). In each Y-maze, a single fly freely explored the three arms, one of which was illuminated with the optogenetically stimulating red light coming from a projector placed next to the camera tracking the flies' position ( Fig. 2C).
The positive control showed the strongest phenotypes at 80% and 100% of the maximum projector light intensity. Therefore, the light intensity was set to its maximum. Light was set at [1 0 0] (RGB code) and its spectrum ranged from 570nm to 720nm with a peak at 595nm. Light stimulation frequency was set at that of the projector (75Hz), and sampling frequency was set to 37.5 Hz, half of the refresh rate of the projector ( Closed-loop stimulation with the projector demanded enough spatial resolution to display light in specific Y-maze arms. This required the pixels of the projector to match those on the camera. To achieve this with high accuracy, the projector displayed a black and white pattern on a white surface that was captured by the camera. These patterns were used to calibrate projector pixels in camera coordinates. For every calibration procedure, at least an R²=0.9998 in the projector-camera pixels correlation was accomplished (approx. pixel precision).
The experiment protocol consists of a total of 60 minutes test where each of the arms was reinforced for 20 minutes by displaying light on the arms whenever the fly entered the given arm.

Statistical power
The T-maze screen with red light was the first experiment and we did not perform a power analysis because we did not yet know if the effect sizes of the screened lines would be of the same size as our positive controls.
We found that effect sizes afforded by the DAN driver lines were considerably smaller than those of the driver lines expressing in heat sensitive neurons. We thus had to choose effect sizes and alpha values that traded off sufficient statistical power with biologically meaningful effect sizes and practically feasible sample sizes. Therefore, we had to choose an alpha value of 5% (Lakens et al., 2018), despite otherwise routinely aiming for alpha values of 0.5% (Benjamin et al., 2018).
In the Y-mazes, data collection is automated and so many choices are recorded in so many animals that even the smallest effects are likely to be statistically significant even with 80% power at an alpha of 0.5%, such that in this screen, the question is more whether small effects are biologically relevant. Therefore, we performed power analyses only for the T-maze with yellow light and the Joystick. We performed a onetailed t-test for the positive against the negative control, where the power was set to 80% and significance to 0.05. The estimated number of experiments for the T-maze was eight. However, since the T-maze with red light suggested we may also be interested in smaller effects in our experimental lines, we set the target sample size to 12.
In the T-maze rescreen, we doubled this target sample size to be able to detect the even smaller effects we observed in the two candidate lines DOIs for individual datasets provided in figure legends.

Results
Robust avoidance with optogenetic feedback in all four screens

DAN function varies with optogenetic activation
The T-maze screen was performed with both red and yellow light ( Fig. 2A,B). Wavelengths corresponding to yellow light activate CsChrimson more effectively than those corresponding to red light (Klapoetke et al., 2014).
However, we did not observe an overall increase in the effect sizes for the DAN lines when tested for their preference in a T-maze with yellow light, compared to one with red light (Fig. 3A, B). If anything, the driver lines mediating aversion of the light decreased somewhat in their effect sizes.
While the switch from bitter to heat-sensing neurons for the positive control yielded considerably higher avoidance, absolute effect sizes did not increase for the screened DAN driver lines.
What did change with the switch from red  Eight combinations of biplots projecting two behavioral axes with symbols for their means and bars indicating the SEMs for both axes. Icons depict screen type. Axes ranges were truncated to -0.4 to +0.4 for visualization purposes. Legend on the right, with colors indicating classical learning phenotypes and a corresponding symbol for each driver line.

DAN function varies with group vs. individual experiment
The T-maze is an experiment where all the flies in a group are making the same choice at the same time ( Fig. 2A, B). The score in these experi-

DAN function varies with locomotion
In all the experiments described so far, locomotion is a key component of the experiment.
In fact, flies that move more slowly in the bright arm of the Y-Maze tend to also have a higher occupancy in this arm (Rohrsen, 2019). As biogenic amines play an outsize role in all kinds of locomotion, we tested whether operant behavior that is orthogonal to locomotion would yield different functions for our DAN subpopulations. For this test, we placed tethered flies on a manipulandum which they could move perpendicular to their longitudinal body axis using their legs (Fig. 2D). Analogous to how a human might move a joystick con-

No DAN effects in the T-maze with yellow light
The larger the effects in the first screens, the more likely one would expect their results to reproduce. The two lines show very robust effects in the T-maze with red light and perform at the extreme ends of our distribution of lines (Fig. 3A).
While in the Y-mazes, the effect sizes are smaller, the high-powered nature of the screen makes the results there also seem quite reliable (Fig. 3C).
Therefore, we sampled from the lowest level of the potential effect size range and began by rescreening TH-D1 and TH-D' in a T-maze with yellow light. In the original screen, the effects of driving CsChrimson with TH-D1 or TH-D' were relatively small (Fig. 3B) compared to, e.g., the effects obtained with red light in the same apparatus (Fig. 3A). The sample size in the original screen had been 12 for these two lines and our power analysis indicated that we needed to more than  In this rescreen we did detect an effect in the TH-D1 driver line and the valence of the TH-D' driver line was opposite to that of the TH-D1 line, but the TH-D' PI did neither reach statistical significance, nor did the Bayes Factor exceed the value of 1 for this group, suggesting that there was no real effect in the TH-D' line (Fig. 6B). Interestingly, the valence of both of the lines was opposite to that observed in the four original screens: In this Joystick rescreen, TH-D1 flies avoided the light and the trend in TH-D' flies was to switch the light on.

As increased light intensity had been
shown previously to be able to change valence ( Fig. 3), we performed a second experiment with the original light intensity in the Joystick setup.
However, the valences were unchanged from the previous rescreen (Fig. 6C). there is some amount of learning during the experiment (Fig. 7A). TH-D1 flies showed lower, but consistent avoidance throughout the experiment (Fig. 7B). TH-D' flies, finally, tended to switch the light off at the start of the experiment, only to increase their PIs gradually until the final training period (Fig. 7C). In the light intensity test (see Ma-terials and Methods), both the generally low PI of TH-D1 as well as the increasing PIs over time in TH-D' (Fig. 7D) replicated the results from the rescreen.  switch from aversive to more appetitive function appears to generalize across different stimulation intensities (Fig. 7).

No 'common currency' DAN function in flies
If specific DANs contribute to 'common currency' value computations (Berridge and Kringelbach, 2015;Grabenhorst and Rolls, 2011;Kobayashi and Hsu, 2019;Lak et al., 2014;Landreth and Bickle, 2008;Levy and Glimcher, 2012;Matsumoto et al., 2016;Schultz et al., 2015)  most did not support any preference at all, and those that did, supported only comparatively weak preferences (Fig. 3). In fact, the same DAN neurons can mediate both approach and avoidance depending either on variables of the experimental situation (e.g., line NP47) and even over time (Figs. 7,8). Even the most consistent DAN driver lines did not stand up to re-screen scrutiny ( Fig. 6), underscoring the comparatively fickle nature of DAN function in Drosophila operant activity. Whereas DAN function was reproducible within appropriately powered experiments (e.g., Fig. 7), none of the lines we screened appeared to be mediating the same valence consistently across several experimental setups (Fig. 4). At the same time, our control line stood out in its reproducibly aversive behavior (Fig. 3). However, the stark contrast between our control neurons (sensory neurons either for bitter taste or heat) reliably driving behavioral preferences and the fickle DANs may not necessarily be due to DANs not serving a role in 'common currency' calculations in flies. Alternatively, it is possible that our stimulation regime, which was optimized for our control neurons, may not be equally optimal for a set of non-sensory DANs.
The stark difference in preference of the DAN lines in the two T-mazes, where the only difference is the wavelength of the optogenetic light, is consistent with this hypothesis. Stimulation-dependent value has also been observed in a two-choice optogenetic paradigm over long timescales driving activity in octopaminergic neurons (Claßen, 2018). If this were the case also for DANs, designing adequate control experiments is not straightforward and the design of a screen to search for such DANs would have to be dramatically different from the one we chose.
Currently there is no consensus on how to best stimulate DANs optogenetically. The usage varies from non-disclosed (König et al., 2019), to 10 ms at 50Hz (McCurdy et al., 2021), to 250 ms pulse duration (Stern et al., 2019), to pulses of more than 1.2 s in duration and frequencies of 0.25Hz (König et al., 2018;Riemensperger et al., 2016). A systematic study found increasing amounts of dopamine released with increasing pulse width, frequency and duration when they varied these parameters from 1-4 ms and 10-60 Hz (Shin et al., 2020). Our regimes of 20 Hz stimulation with 10 ms (T-mazes) or 50 ms (Joystick) pulse-width do not seem to fall outside the range of what has been shown to be effective either behaviorally or in triggering dopamine release. Perhaps more worryingly, there may be an optimal stimulation protocol for any arbitrary set of any kind of neuron to drive approach or avoidance. In this case, any focus just on aminergic neurons may be unwarranted.
Furthermore, the flies we tested were satiated. Appetitive DAN functions have always been found with starved or thirsty animals. It has been observed in other self-stimulation or self-administration experiments before that stronger preferences can be achieved when animals are starved (Carr, 2002;Fulton et al., 2004;Gnazzo et al., 2021). This explanation would only hold for appetitive DANs, aversive circuits ought not to be affected.
The possibility also remains that we did not test a line that would drive in the perfect combination of DANs to achieve a behavioral preference that is identical to or higher than that of our control lines. It is the main shortcoming of our work that only 13 of our nearly 50 DAN lines were tested in all four screens. However, even among the additional 12 DAN lines that were tested in three of the four screens, there were no strong candidates consistently performing at the extreme ends of all three screens (Fig. 3). Only line TH-C1 performed consistently in all three screens it was tested in, but the score was close to zero in all cases. Of the 29 lines that were tested in two or more screens, no lines stood out as exceptionally consistent (Fig. 4)

Reinforcement is different from an unconditioned stimulus
If there were no 'common currency' computation taking place in flies (or DANs were not part of it), one reason DAN function may vary between paradigms is that different neuronal mechanisms may be underlying performance in them.
For instance, depending on experimental design, operant and classical experiments can be mediated by different biochemical pathways. Crucial for this distinction are not the general procedural differences between the two classes of experiments, but the content of the memory that is formed: classical experiments associate two events (i.e., CS and US) in the world around the subject (world-learning) and typically engage synaptic plasticity mediated by the cAMP pathway and phosphorylation by protein kinase A (PKA) (Barbas et al., 2003;Giese and Mizuno, 2013;Lee, 2015;Matsumoto et al., 2018;Müller, 1999;Romano et al., 2006;Selcher et al., 2002;Wang and Peng, 2016;Wang and Zhang, 2012). In contrast, operant experiments that avoid contamination by environmental predictors (i.e., operant self-learning), depend on phosphorylation by protein kinase C (PKC) (Brembs and Plendl, 2008;Burguière et al., 2010;Colomb and Brembs, 2016;Lorenzetti et al., 2008;Rochefort et al., 2011;Watanabe et al., 2006Watanabe et al., , 2002Yoshida et al., 2003) and the developmental activity of the transcription factor FoxP (Haesler et al., 2007;Lai et al., 2001;Mendoza et al., 2014;Norton et al., 2019;Scharff and Petri, 2011;Schreiweis et al., 2014). used to be referred to as a homogenous "midbrain dopamine" population in mammals is much more heterogeneous than previously assumed (Anderegg et al., 2015;Cardoso and Lévesque, 2020;Hassan and Benarroch, 2015;Henny et al., 2012;Heymann et al., 2020;Lammel et al., 2014;Poulin et al., 2020). Specifically, dopaminergic function may differ between world-learning and self-learning components of the tasks in question, in particular via different dopaminergic projections to the striatum (Augustin et al., 2020;Da Cunha et al., 2012;Hassan and Benarroch, 2015;Heymann et al., 2020). With habit formation representing the experimental paradigm in mammals that mimics self-learning most closely (Brembs, 2009), the different roles ventromedial and ventrolateral striatum play in habit formation suggest differential roles of DAN circuits in worldand self-learning also in mammals. Given the small number of neurons in some of the lines we tested, it becomes conceivable that not all DAN populations that support classical conditioning need to also support operant preference in the same way. For instance, TH-D' neurons can substitute an aversive US in classical olfactory conditioning and also in our original screen, all four paradigms yielded negative scores for this line. In the Joystick rescreen, the initial reaction of the flies to activation of these neurons was also aversive -they switched the light off (Fig. 7). However, over time, continued operant control reduced the aversive properties of TH-D' activity (Fig. 8). The results emphasize that demonstrating a role of a set of neurons in processing a classical US is not sufficient to generalize from this instance to operant situations with reinforcement.
All animals process such re-afferent (operant) stimuli differently compared to ex-afferent (classical) stimuli. This is the reason we cannot tickle ourselves (Bays et al., 2006), why we perceive a stable visual world despite our frequent quick, or saccadic, eye movements (Sommer and Wurtz, 2006) and this differential processing is reflected in different brain activation patterns be-tween self-generated and exogenous visual stimulation (Matsuzawa et al., 2005). From this perspective, it may not be surprising that neurons found to encode a given value in classical (i.e., exafferent) situations do not necessarily have to sustain the matching preference in operant (i.e., re-afferent) situations: a classical US is not the same as reinforcement. This distinction can also be observed, e.g., when medial orbitofrontal cortex in rodents regulates operant avoidance but not Pavlovian fear (Ma et al., 2020).
An interesting case is the optogenetic activation of TH-Gal4 DANs in an operant olfactory conditioning paradigm in Drosophila (Claridge-Chang et al., 2009). In our screens, TH-Gal4 was tested in the T-mazes and the Joystick screen. In none of these screens did this line sustain substantial preference (Fig. 3). In contrast, (Claridge-Chang et al., 2009) showed that the activity in these neurons was sufficient to induce an aversive memory of the odor. Unfortunately, there is no data on the behavior of the flies during the training periods in this operant experiment, not even upon request, so it is impossible to tell if the flies avoided activation of the neurons or if the odor was conditioned as a classical side-effect of odor-activation pairings. This would not be the first time that such a dissociation between performance and learning would have been observed in flies (Kane et al., 1997).
A more recent optogenetic lever-pressing self-stimulation experiment in mice failed to find differences in dopamine release between classical and operant conditioning (Hollon et al., 2021).
However, this study also failed to exclude all classical (world-learning) predictors from the experiment and, conversely, showed that the pressing behavior in question was irrelevant for the observed effect on dopamine release, as compared to the classical cue associated with the behavior, the depressed lever. Thus, the evidence from other preparations suggests that the effects the authors observed here are likely due to a contamination with classical cues that obscured any operant processes (Brembs and Plendl, 2008).

Neuronal substrates of reinforcement
DANs from the PAM cluster projecting to mushroom body substructures have long been considered to encode reward (Burke et al., 2012;Das et al., 2014;Lewis et al., 2015;Lin et al., 2014;Liu et al., 2012;McCurdy et al., 2021;Yamagata et al., 2015). One may ask, why none of these clusters seem to be relevant for operant activity as tested here. One explanation is that the previous results were gained from classical conditioning experiments with appropriately starved or waterdeprived animals, while our flies were satiated.
Moreover, operant control has been shown to not require intact mushroom body function across a number of paradigms (Brembs, 2009;Brembs and Wiener, 2006;Putz and Heisenberg, 2002;Siwicki and Ladewski, 2003;Wolf et al., 1998), at least as long as odor processing is not involved.
Hence, it may not be surprising that none of the DANs projecting to this neuropil supported consistent preferences in our screens (Fig. 3). As food or water are not the only rewards for flies, none of this should constitute an obstacle for identifying DANs encoding general reward. However, while we found DANs that mediated solid preference in each screen, we could not find any DANs that mediated consistent preference across all screens and held up in our rescreens.
Instead, the two most conspicuous lines, TH-D1 and TH-D', mainly comprise DANs from the PPL1 and PPM3 clusters, targeting central complex (CX) structures such as the fan-shaped body (FB) the ellipsoid body (EB) or the noduli, as well as the lateral accessory lobe (LAL; Fig. 5). The projection from the PPM3 cluster to the ventral FB (vFB) and noduli were particularly intense. vFB, which consists of layers 1 to 5, is highly responsive to electric shock and nociceptive heat and mediates innate avoidance (Hu et al., 2018), which may explain the (at least initial) avoidance behavior observed in both lines in our Joystick experiments . However, this is not the only function of the FB. In addition to avoidance, the FB was reported to be important for locomotion control (Strauss, 2002), visual processing and pattern learning (Liu et al., 2006;Weir and Dickinson, 2015), or courtship maintenance (Sakai and Kitamoto, 2006). Dorsal components of the FB have been implied in sleep (Berry et al., 2015;Donlea et al., 2014Donlea et al., , 2011Qian et al., 2017;Ueno et al., 2012).
Together with the FB, the EB and LAL are the most outstanding regions labelled by TH-D1 and TH-D' (Fig. 5). The EB encodes heading orientation and feature detection with the corresponding associated memories Jayaraman, 2015, 2013;Strauss, 2002;Wolff et al., 2015). Dop1R1 signaling in the EB ring neurons affect the temporal organization of motor actions, exploration and turning behavior (Kottler et al., 2019). Hence, the dopaminergic projections might reinforce different behaviors by biasing the action selection process in the ring neurons of the EB. Manipulating specific dopaminergic inputs into the ring neurons might show differential effects. The EB and the FB, as the other CX structures, protocerebral bridge and noduli, show a strikingly compartmentalized layout with restricted connections, which suggest a high degree of functional specialization (Kottler et al., 2019;Wolff et al., 2015;Wolff and Rubin, 2018). Hence, dopaminergic projections to the CX and LAL might be topographically organized to differentially drive avoidance and approach (Hu et al., 2018;Wolff et al., 2015). The structure and function of the CX and LAL is quite conserved across insect's species. They process spatial aspects of complex multisensory information and integrate it with information about the insect's internal state and past experiences, to drive proper motor outputs (Buchanan et al., 2015;Hu et al., 2018;Kanzaki, 2018, 2016;Strauss, 2002;Wolff et al., 2015;Wolff and Rubin, 2018).
The LAL is the major output site of the CX and its bilaterally coordinated activity mediates signals to the thoracic motor centers. The LAL is closely interconnected with other CX structures, but it also receives input from the anterior optic tract, superior medial protocerebrum, Lobula Plate, posterior slope and thoracic motor centers.
In addition, the LAL receives ascending feedback about proprioceptive information to coordinate downstream motor commands in locomotion, flight, phonotaxis and pheromone orientation in several insects. The characteristic flip-flop signals observed in the LAL and downstream descending neurons are correlated with turning maneuvers and its role seems to be locomotion-biased, which points to this region as a major candidate where reinforcement might converge Kanzaki, 2018, 2016;Wolff et al., 2015;Wolff and Rubin, 2018).