Suppression of a single pair of mushroom body output neurons in Drosophila triggers aversive associations

Memory includes the processes of acquisition, consolidation and retrieval. In the study of aversive olfactory memory in Drosophila melanogaster, flies are first exposed to an odor (conditioned stimulus, CS+) that is associated with an electric shock (unconditioned stimulus, US), then to another odor (CS−) without the US, before allowing the flies to choose to avoid one of the two odors. The center for memory formation is the mushroom body which consists of Kenyon cells (KCs), dopaminergic neurons (DANs) and mushroom body output neurons (MBONs). However, the roles of individual neurons are not fully understood. We focused on the role of a single pair of GABAergic neurons (MBON‐γ1pedc) and found that it could inhibit the effects of DANs, resulting in the suppression of aversive memory acquisition during the CS− odor presentation, but not during the CS+ odor presentation. We propose that MBON‐γ1pedc suppresses the DAN‐dependent effect that can convey the aversive US during the CS− odor presentation, and thereby prevents an insignificant stimulus from becoming an aversive US.

Memory includes the processes of acquisition, consolidation and retrieval. In the study of aversive olfactory memory in Drosophila melanogaster, flies are first exposed to an odor (conditioned stimulus, CS+) that is associated with an electric shock (unconditioned stimulus, US), then to another odor (CSÀ) without the US, before allowing the flies to choose to avoid one of the two odors. The center for memory formation is the mushroom body which consists of Kenyon cells (KCs), dopaminergic neurons (DANs) and mushroom body output neurons (MBONs). However, the roles of individual neurons are not fully understood. We focused on the role of a single pair of GABAergic neurons (MBON-c1pedc) and found that it could inhibit the effects of DANs, resulting in the suppression of aversive memory acquisition during the CSÀ odor presentation, but not during the CS+ odor presentation. We propose that MBON-c1pedc suppresses the DANdependent effect that can convey the aversive US during the CSÀ odor presentation, and thereby prevents an insignificant stimulus from becoming an aversive US.
Pavlovian classical conditioning, in which the conditioned stimulus (CS) is associated with the unconditioned stimulus (US), serves as a simple model for learning and memory. There are many kinds of stimuli in the environment, and organisms have evolved to select for stimuli that are used as the US in conditioning paradigms, enabling them to survive and thrive. The olfactory aversive memory of Drosophilamelanogaster serves as a good example of Pavlovian classical conditioning [1,2], and several distinct stimuli can be used as the US in Drosophila [1][2][3][4][5][6]. However, the mechanisms by which Drosophila select the US or tune the threshold for accepting a stimulus as the US remain largely unknown.
The neuropil called the mushroom body (MB) has been extensively studied anatomically [7][8][9] and functionally as the center for the olfactory aversive memory [10][11][12][13]. The MB consists of~2000 intrinsic neurons called Kenyon cells (KCs), which are the third-order olfactory neurons in each hemisphere [8]. Subsets of KCs sparsely represent odor information [14][15][16][17], and the information is modified by aversive stimuli conveyed by dopaminergic neurons (DANs) upon conditioning [12,[18][19][20]. The modified information then converges on MB output neurons (MBONs) [9,21]. Cellular identification of MBONs has been an intriguing result from recent studies of brain anatomy [22,23]. It has been revealed that odor information encoded in~2000 KCs converges on only 34 MBONs composed of 21 anatomically distinct cell types [9]. This finding permits the study of neuronal mechanisms underlying odor coding and olfactory memory formation in the reduced dimension at the level of fourthorder olfactory neurons. Specifically, it allows us not only to identify each output neuron at a cellular resolution but also to manipulate the functions of each output neuron using split-Gal4 drivers [9,24]. We have already started to witness the progress in understanding roles of MBONs in the process of memory formation [9,[21][22][23]25,26]. In addition, the DAN activity has been shown to be dynamically changed by external stimuli or internal physiological states [27][28][29], and the output from MBONs is also known to affect the DAN activity [28], suggesting that the circuits consisting of KCs, DANs and MBONs form dynamic neuronal networks including multiple layers of feed forward and feedback regulation.
We chose to study the role of MBON-c1pedc because it has been reported to play a pivotal role in aversive memory [24,26] and because a memory trace is detected in its responses to odors associated with electric shocks or activation of DANs [26,30]. In addition, MBON-c1pedc reflects internal and physiological states of flies and inhibits activities of other MBONs [26]. These results prompted us to explore the possibility that MBON-c1pedc plays multiple roles in memory formation, and we found that MBON-c1pedc was required for the acquisition of memory. Furthermore, during memory formation, MBON-c1pedc suppresses the acquisition of aversive memory for CSÀ but not for CS+.

Setting for behavioral experiments
Groups of~50 flies (2-5 days old) raised under a 12 hr:12 hr light-dark cycle were used for one trial in behavioral experiments. Before behavior experiments, flies were kept in vials with Kimwipes soaked with sucrose solution. The training and test apparatus were the same as described previously [2], and protocols were slightly modified. Flies were exposed to 60 s of a CS+ odor (MCH or OCT) with 12 90 V electric shocks at a 5 s interstimulus interval, then 30 s of clean air, followed by the CSÀ odor (OCT or MCH) without electric shocks. After the training stage, flies were allowed to select the CS+ odor or the CSÀ odor in a T-maze at test stage. Odors were in a glass 'odor cup' (8 mm in diameter for OCT and 10 mm for MCH) sitting in the middle of an odor stream. The flow velocities of air or odors were 0.75 LÁmin À1 in each stage.

Temperature shifting
To shift temperature between the permissive temperature (22°C) and the restrictive temperature (33°C), we used two climate boxes set to 22°C or 33°C, and all of the training tubes and T-mazes were preheated and fixed at the indicated temperatures. Temperature shifts were performed immediately. After the transfer, flies were left in a tube with airflow at the indicated temperature.

Temperature shifting between training and test
Flies were preheated at 33°C for 30 min and then trained and tested at 33°C (Fig. 1D). Flies were trained and tested at 22°C (Fig. 1E

Temperature shift during CS+ and CSÀ presentation
Flies were trained with the CS+ presentation at 22°C and immediately transferred to 33°C, followed by 2 min air flow, and the CSÀ was presented at 33°C, then immediately retransferred to 22°C, followed by 2 min air flow and testing at 22°C ( . Flies were trained with the CS+ presentation at 22°C and immediately transferred to 33°C, followed by 3 min air flow at 33°C, and they were then immediately retransferred to 22°C, followed by 2 min air flow, the CSÀ presentation at 22°C and testing at 22°C (Fig. 2C). Flies were preheated at 33°C for 30 min, trained with the CSÀ presentation at 33°C, and then immediately transferred to 22°C, followed by 2 min air flow and the CS+ presentation at 22°C, with testing at 22°C (Fig. 2D).

Blockade of MBON-c1pedc-induced aversive memory (BGAM) training and test
Flies were exposed to odor 1 for 60 s at 22°C and immediately transferred to 33°C, followed by 2 min air flow, exposure to odor 2 for 60 s at 33°C, and then immediate retransfer to 22°C, followed by 2 min air flow and testing at 22°C (Figs 3B, 4F, 5A and 6B,C).

Test stage
Flies were loaded into the T-maze and allowed to choose between MCH and OCT for 1.5 min. Performance index was calculated as the number of flies avoiding the CS+ odors (or odors presented at 33°C for BGAM) minus the number of flies in the other side, divided by the total number of flies. Flies were reciprocally trained with MCH or OCT. Control odors (OCT or MCH) were also presented, and two performance indices each were calculated for MCH and for OCT. The final performance index was calculated by averaging the two performance indices for MCH and for OCT.

Confocal imaging
Flies were dissected in cold phosphate-buffered saline (PBS) solution and fixed in PBT (PBS containing 0.3% Tri-tonX-100) with 4% formaldehyde for 30 min at room temperature. After PBT washing, PBT was replaced with PBS, and brains were placed between a glass slide and a cover glass with medium (VECTASHIELD Mounting Medium, Vector Laboratories, Burlingame, CA, USA). Images were captured on a LSM 710 confocal microscope (Carl-Zeiss, Jena, Germany) and brightness was lineally processed using FIJI software (http://fiji.sc/Fiji).

Statistical analysis
We performed statistical analyses using PRISM 6 (Graph-Pad, La Jolla, CA, USA). All behavior data were tested

MBON-c1pedc is required for both acquisition and retrieval of aversive short-term memory
We first examined the role of MBON-c1pedc in 2 min short-term memory (STM). We used a MBON-c1pedc specific split-Gal4 driver, MB112C (Fig. 1B,C) [9,24] to express a temperature-sensitive dominant-negative form of dynamin, Shi ts [32,35] and block output from MBON-c1pedc. Flies were exposed to an odor, 4methylcyclohexanol (MCH) or 3-octanol (OCT), paired with 12 electric shocks for 1 min (CS+), followed by OCT (or MCH) without electric shocks for 1 min (CSÀ) (Fig. 1A). Two minutes later, flies were allowed to select one of the two odors to avoid. Flies were trained and tested at a restrictive temperature (33°C) (Fig. 1D) or at a permissive temperature (22°C) (Fig. 1E) throughout the experiments. Blocking MBON-c1pedc severely impaired STM (Fig. 1D), demonstrating that MBON-c1pedc output is indispensable for STM, as has been reported for 2 h memory [24]. To clarify whether this STM deficit was caused by impairment of memory acquisition or retrieval, we blocked output from MBON-c1pedc during the acquisition stage or the retrieval stage. STM was impaired by blockade of MBON-c1pedc either during the training stage (Fig. 1F) or during the test stage (Fig. 1G). These results suggest that MBON-c1pedc is required for aversive memory acquisition in addition to aversive memory retrieval [26].

MBON-c1pedc synaptic output is necessary to inhibit aversive memory acquisition for CSÀ
For further analysis of MBON-c1pedc in memory acquisition, we blocked MBON-c1pedc separately during the CS+ or CSÀ presentation. Interestingly, blocking MBON-c1pedc during the CSÀ presentation impaired memory significantly ( Fig. 2A), whereas blocking MBON-c1pedc during the CS+ presentation (Fig. 2B) or immediately after the CS+ presentation (Fig. 2C)   presentation at 33°C followed by CS+ presentation at 22°C also showed the memory deficit (Fig. 2D). These results indicate that aversive memory acquisition requires MBON-c1pedc output during the presentation of the CSÀ odor ( Fig. 2A,D). Blockade of MBON-c1pedc during the CSÀ presentation may interfere with aversive memory acquired during the CS+ presentation.
Considering the possibility that blocking MBON-c1pedc during the CSÀ presentation alone may forms an aversive memory for the CSÀ odor, and competition of the aversive memory between the CS+ odor and the CSÀ odor might cause the memory deficit, flies were exposed to two odors in sequence, followed by the test stage. One odor was presented at the permissive temperature, and the other was at the restrictive temperature to block MBON-c1pedc (Fig. 3A). We found that flies formed aversive memory toward the odors presented without synaptic output from MBON-c1pedc (Fig. 3B). We named this BGAM, blockade of MBON-c1pedc-induced aversive memory. In BGAM acquisition, a control odor is presented before the temperature shift. We investigated the possibility that some type of memory could be formed for the control odor by temperature shifting immediately after the presentation of the odor, since the timing-dependent behavioral plasticity was reported [36]. To test this possibility, only a single odor was presented at the restricted temperature in the training session, and the control odor was not presented. As a result, BGAM was also observed in the training session, regardless of the presentation of the control odor at the permissive temperature (Fig. 3C). These results indicate that the memory deficit evoked by blocking MBON-c1pedc at the acquisition stage (Fig. 1F) is caused at least in part by competition between the aversive memory for CS+ and the BGAM for CSÀ. Thus, output from MBON-c1pedc is necessary to prevent the aversive memory for CSÀ.   MBON-c1pedc, but not the other neurons that could also be labeled by the MB112C driver, is responsible for aversive memory acquisition and BGAM In the above experiments, MB112C was used as the specific driver to label MBON-c1pedc. Although MBON-c1pedc seemed to be the only neurons labeled by the MB112C driver, according to the confocal images, a few neurons may be labeled by MB112C (Fig. 4A,A'). To test if MBON-c1pedc, but not the other neurons, is responsible for aversive memory acquisition and BGAM, R83A12 was used as another driver to examine the role of MBON-c1pedc (Fig. 4B,  B'). The blockade of the R83A12-positive neurons by Shi ts impaired STM acquisition (Fig. 4C). Output from the R83A12-positive neurons was necessary during the CSÀ presentation (Fig. 4D), but not during the CS+ presentation (Fig. 4E). Furthermore, BGAM was also observed by blocking the R83A12-positive neurons during odor presentation (Fig. 4F). These results indicate that the neurons responsible for STM acquisition and BGAM formation were likely to be MBON-c1pedc, but not the other neurons that could potentially be labeled by the drivers.

DANs are required for the BGAM acquisition
MBONs and DANs constitute microcircuits [7,9], and DANs transmit various kinds of aversive information [12,[37][38][39]. Thus, we tested whether DANs are involved in the BGAM acquisition. To label DANs, we used tyrosine-hydroxylase (TH) Gal4 (TH-Gal4) [31], which is thought to label most DANs that convey aversive information [18,40]. Flies lacking synaptic outputs from both MBON-c1pedc and DANs showed  severe impairment of the BGAM (Fig. 5A), suggesting that the BGAM is made only when MBON-c1pedc is inactive and DANs are active. Activation of DANs during the CSÀ presentation might be the cause of the memory deficit observed when blocking MBON-c1pedc during the acquisition stage (Fig. 1F). To test this possibility, we expressed Shi ts in MBON-c1pedc and DANs, and performed the same protocol as in Fig. 2A,B. Blocking MBON-c1pedc alone during the CSÀ presentation impaired memory, whereas blocking both MBON-c1pedc and DANs during the CSÀ presentation did not produce any memory impairments (Fig. 5B), indicating that blocking MBON-c1pedc during the CSÀ presentation caused memory deficits via the output of DANs. On the other hand, blocking DANs during the CS+ presentation caused memory deficits with or without the blockade of MBON-c1pedc (Fig. 5C). Blocking MBON-c1pedc during the CS+ presentation did not cause any significant effects on memory compared to the control Gal4 strain, nor did it rescue memory deficits caused by the blockade of DANs. Assuming that DANs activation is necessary at the CS+ presentation and MBON-c1pedc inhibits the effect of DANs, we investigated whether the activation of MBON-c1pedc at the CS+ presentation impaired the STM. To activate MBON-c1pedc artificially, dTrpA1, a temperature-sensitive cation channel [33], was expressed by using the MB112C driver. Neurons expressing dTrpA1 are transiently activated at the restrictive temperature (33°C), and not at the permissive temperature (22°C). The flies were transferred to the restrictive temperature and immediately the CS+ odor and ESs were presented for 1 min. The flies were then re-transferred to the permissive temperature, exposed to the CSÀ odor and tested. This manipulation of MBON-c1pedc did not impair the aversive STM significantly (Fig. 5D). The dTrpA1 inducing the artificial activation of MBON-c1pedc might be too weak to suppress the effect of DANs induced by ESs sufficiently. Thus, MBON-c1pedc might suppress the weak effect of DANs.
DANs are effectively downstream of MBON-c1pedc in the acquisition of the memory Taken together, in classical conditioning, the output of DANs is ineffective in the CSÀ presentation and is required during the CS+ presentation, whereas the MBON-c1pedc output is required during the CSÀ presentation, but not during the CS+ presentation. In addition, aversive memory induced by the output of DANs in classical conditioning was not affected by blocking MBON-c1pedc during CS+ presentation (Fig. 5C), whereas BGAM induced by blocking MBON-c1pedc was affected by blocking DANs (Fig. 5A). Thus, DANs are effectively downstream of MBON-c1pedc in the aversive memory acquisition stage. In addition, MBON-c1pedc and DANs negatively modify each other's functions, since DANs attenuate input from KCs to MBON-c1pedc [30], and this study suggests that MBON-c1pedc inhibits the functions of DANs. For further dissection of the involvement of DANs in BGAM, we used a panel of split-Gal4 drivers [9] and manipulated subsets of TH-Gal4 positive neurons. We first used drivers to label a large population of TH-Gal4 positive neurons in combination with MB112C to block the subsets of DANs and MBON-c1pedc (Fig. 6B). Compared to the MBON-c1pedc blocked flies, flies without synaptic output from MBON-c1pedc and TH-or MB504B-positive DANs showed significantly lower BGAM. Blockade of DANs labeled using MB060B did not cause a significant decrease in BGAM. These results indicate that DANs labeled by TH or MB504B, but not by MB060B, are important for BGAM formation. Importantly, MB060B and MB504B label similar subsets of DANs, but only MB504B labels PPL1-c1pedc DANs. We next used the MB438B split-Gal4 driver to manipulate PPL1-c1pedc DANs and tested if the BGAM was impaired by blocking PPL1-c1pedc DANs and MBON-c1pedc, and found that inactivation of PPL1-c1pedc DANs did not impair the BGAM (Fig. 6C). Taking into account that MB504B positive neurons are sufficient to suppress the BGAM, a combination of the PPL1-c1pedc, -c2a'1, -a'2a2 and -a3 DANs or all of them are required for BGAM. Since the combination of DANs labeled by MB060B, which does not label PPL1-c1pedc DANs, or MB438B, which does not label PPL1-c2a'1 DANs, is not sufficient to suppress the BGAM, the PPL1-c1pedc DANs and PPL1-c2a'1 DANs are necessary for the BGAM. No drivers labeling the combination of PPL1-c1pedc, -c2a'1 and -a3 DANs or the combination of PPL1-c1pedc, -c2a'1 and -a'2a2 DANs are available, and thus the necessities for the PPL1-a'2a2 and -a3 DANs are unclear.
Taken together, the BGAM is acquired through a combination of PPL1-DANs labeled by MB504B, which is consistent with the notion that some DANs function coordinately [19,27,28,40,41]. Their anatomical connectivity also suggests the possibility that MBON-c1pedc modify the effects of some DANs projecting to a/b lobes [9,42].

Discussion
We have shown that the synaptic output from MBON-c1pedc is required for suppressing aversive memory acquisition without electric shocks and that in the classical olfactory conditioning procedure with electric shocks as the US, MBON-c1pedc must be active during the CSÀ presentation, whereas DANs must be active during the CS+ presentation. Given that the memory is formed regardless of the activity of MBON-c1pedc during the CS+ presentation and that DANs are required for BGAM, DANs are functionally downstream of MBON-c1pedc in this context. Among the population of DANs, BGAM required PPL1-DANs, which are thought to convey punitive information and cause aversive memory in concert [40,41]. In aversive olfactory memory, electric shocks as the US can be replaced by artificial activation of PPL1-DANs [18,19,43], indicating that DAN activation alone is sufficient for aversive associations with odors. The population of DANs that can replace the US overlaps with the population of DANs required for BGAM. Collectively, the blockade of MBON-c1pedc allows DANs to replace the aversive US to make the aversive associations.
Assuming that BGAM is acquired by MBON-c1pedc and DANs, there are two questions about the BGAM formation. One is about the pathway for MBON-c1pedc to modify DANs effects and the other is about the trigger for DANs activation. The pathway for MBON-c1pedc to modify the DANs is unknown although there is anatomical connectivity. According to the previous study referring to the anatomies of MBONs and DANs, the dendrites of a few DANs are slightly co-localized with the axons of MBON-c1pedc [9]. This indicates that some DANs may be downstream of MBON-c1pedc at the level of a neural circuit. However, we could not detect the functional connectivity of MBON-c1pedc and DANs, since the DANs activity was stochastic and fluctuated at the restrictive temperature used to manipulate the MBON-c1pedc in the functional calcium imaging (data not shown). Thus, other methodologies, such as optogenetics, membrane potential indicators or synaptic output indicators might be useful to test this possibility. Since MBON-c1pedc axons and DANs dendrites are only slightly colocalized, this possibility is less likely than the following second possibility. Second possibility is that MBON-c1pedc affects DANs effect indirectly. MBON-c1pedc axons are projected to the crepine (a region surrounding the horizontal and medial lobes) and the core of the a and b lobes [9], and the DANs axons also project to the a and b lobes [9,42]. Thus, DANs and MBON-c1pedc converge on the lobes, and they may input to KCs or other MBONs coordinately, to modulate their plasticity. Since MBON-c1pedc is GABAergic, MBON-c1pedc may inhibit the KCs activity, and blocking MBON-c1pedc may disinhibit the KCs activity, leading to the hyperactivity of KCs and the easy association with weak DANs activity.
It is also unclear why and how the DANs are activated when MBON-c1pedc is blocked and the odors are presented. One possibility is that DANs activity fluctuates, reflecting inner physiological states [27,28], and that the active state of DANs can stochastically cause aversive memory to a given odor. Another possibility is that the exposure to a given odor activates DANs. We examined the activity of DANs via functional calcium imaging under a two-photon microscope, but we only observed stochastic activity of DANs and failed to detect significant correlation with the exposure to odors (data not shown).
Without the appropriate activity of MBON-c1pedc, the probability of aversive associations might be increased even if the environment contains few aversive stimuli. Although aversive associations are important for animals' survival, an appropriate threshold for memory acquisition is necessary to conserve the energy required to acquire an aberrant memory and to highlight the importance of essential memories. MBON-c1pedc might have such a gating function by antagonizing the activity of DANs.
BGAM is acquired by odor presentation and the blockade of MBON-c1pedc. MBON-c1pedc responds to odors robustly, but its response is decreased after associating the odors with ESs or activation of DANs [26,30]. Thus, BGAM acquisition may mimic the situation in which flies sense CS+ odors after associating the odors with ESs. After the classical conditioning, CS+ odor presentation may cause some BGAM in flies.
BGAM was observed as behavioral plasticity, and can be categorized into associative or nonassociative memory, depending on the viewpoint. Since BGAM is acquired solely by an odor presentation, BGAM may be categorized as a nonassociative memory or a particular sensitization. In the T-maze machine, na€ ıve flies avoid odors (MCH or OCT) as compared to the air (this is called odor avoidance), indicating that odors are aversive stimuli for flies to some extent. In wildtype flies, odorant information may be processed as aversive information, but not stored as aversive memory by inhibiting the memory acquisition processes. However, the blockade of MBON-c1pedc may disturb the inhibiting processes, and thus the odorant information may be stored as an aversive memory. A previous study showed that odor avoidance was enhanced by blocking MBON-c1pedc [26], and this study indicated that enhanced odor avoidance lasts as memory by blocking MBON-c1pedc. The enhancement of responses to pre-exposed stimuli is called sensitization. In Drosophila, behavioral sensitization to odors or neural sensitization to odors around the KCs was not observed, although odor sensitization in sensory neurons was reported [44]. In contrast, in Caenorhabditis elegans, it was previously reported that behavioral sensitization to odors was regulated by dopamine release to an interneuron [45]. This sensitization mechanism in C. elegans might be similar to the BGAM mechanism, since their behavioral protocols are nonassociative learning and dopamine-related. BGAM might be nonassociative memory and lasting sensitization, and in wild-type flies, MBON-c1pedc might suppress the sensitization.
However, if BGAM is acquired by associating an odor with a temperature stimulus or another aversive stimulus surrounding the flies, then BGAM may be categorized as associative memory. In order to block synaptic output by using Shi ts , the flies are kept at a restrictive temperature (33°C), which could be an aversive stimulus [37]. Although the temperature we used might be slightly aversive for flies, the temperature shifting to 33°C for 1 min was lower and shorter than the 34°C shift for 2 min used in a previous study [37], and our protocol was apparently insufficient for the control strains to acquire strong aversive memory (Fig. 3B). If the blockade of MBON-c1pedc lowers the threshold for the temperature as the US, then BGAM could result from the association between the odors and the high temperature. In rats, the aversive US pathway is reportedly inhibited by feedback circuits to calibrate the strength of learning after aversive memory formation [46]. MBON-c1pedc and DANs may comprise a similar circuit in Drosophila. To investigate whether aversive information is associated with odor in BGAM, other novel methodologies to block the synaptic output in a freely moving fly in a precise time window without aversive stimuli, such as temperature shifting, are needed.
Taken together, the blockade of MBON-c1pedc during odor presentation without US influences the DANs effects directly or indirectly and forms BGAM. We found the novel function of MBON-c1pedc for BGAM formation at the level of behavior. The MBON-c1pedc functions to suppress the memory formation, indicating that memory acquisition can be regulated negatively. Only a few studies have reported the negative regulation (suppression) of memory, and a recent study reported that the neural circuit suppresses the US pathway in rats by feedback circuits, to calibrate the strength of learning after aversive memory formation [46]. This is the first evidence that MBON-c1pedc and DANs may comprise a similar circuit in Drosophila.