Aversive learning hijacks a brain sugar sensor to consolidate memory

Sensing the content of ingested food is critical for the body to evaluate energy availability and accordingly set an adequate metabolic state. In addition to peripheral taste receptors, animals and humans detect the nutritional value of food through post-ingestion mechanisms involving internal nutrient sensors in the digestive tract and the brain. Internal nutrient-sensing systems are major regulators of appetite and feeding behaviour¹². As such, they participate in memory processes that enable the assignment of value to food-related sensory cues¹³. However, the extent to which the brain’s cognitive efficiency relies on its nutrient sensors, in particular beyond food-related processes, is unclear.

Laboratory experiments using genetically tractable species allow the study of targeted neural circuits in precisely defined behavioural tasks. In the fruit fly Drosophila melanogaster, we investigated the cognitive role of a major brain sugar sensor using a Pavlovian aversive olfactory learning task comprising the association of an odorant with mild electric shocks. After such conditioning, flies develop a learned avoidance towards this odorant⁷. After a single learning session, this memory decays within hours. Multiple sessions spaced in time (spaced training) induce the formation of long-term memory (LTM) that lasts for days⁸, is protein-synthesis dependent⁸ and is critically linked to glucose-based metabolism in neurons of the mushroom body (MB) brain area^14,15,16. The same number of immediately consecutive sessions (massed training) induces another form of consolidated memory that lasts for around 1 day, but this memory does not correspond to LTM^8,16,17: in addition to being less persistent⁸, it involves distinct neuronal circuits within the MB¹⁷ and relies on lipid-based rather than glucose-based neuronal energy metabolism¹⁸. The differential impact on consolidation efficiency between massed and spaced learning paradigms is an experimental manifestation in flies of a well-documented cognitive phenomenon named the spacing effect^8,19,20,21. Experiments in the aversive learning paradigm are typically performed on fully satiated flies with ad libitum access to food before and after conditioning.

In the fly central brain, a small set of fructose-sensing neurons (four neurons per brain hemisphere, expressing the fructose-responsive gustatory receptor Gr43a, hereafter Gr43a neurons) respond to sugar ingestion and promote feeding in a satiation-dependent manner^6,9,22. Dietary sucrose contains both glucose and fructose, but fructose is also produced from glucose metabolism through the polyol pathway, so fructose sensors respond to carbohydrate intake in general⁶. In hungry flies, Gr43a neurons detect an increase in fructose levels after sugar intake, and their activity promotes feeding⁶. The nature of the molecular signalling from Gr43a neurons and their downstream targets are currently unclear. When flies are satiated, the sensitivity of Gr43a neurons to fructose is disabled by the inhibitory action of the neuropeptide tachykinin (Tk), released by upstream dFB neurons, which are localized in the dorsal layer of the fan-shaped body region (dFB)⁹ (Extended Data Fig. 1a).

Our experiments reveal hunger-independent sugar-sensing plasticity underlying LTM, whereby spaced training restores the sensitivity of brain fructose-sensing neurons despite satiety, enabling their activation by post-training feeding. This endows food ingestion with a cognitive value, notably as a memory consolidation signal that extends beyond food-related experiences.

We performed acute silencing of Gr43a neurons through targeted expression of the dominant-negative thermosensitive Shibire protein (Shi^ts)²³, which is known to efficiently silence neurotransmission when flies are placed at the restrictive temperature of 33 °C (ref. ¹⁷). Genetic targeting of brain fructose-sensing neurons was achieved using a previously published transgenic line carrying a GAL4 knock-in in the Gr43a locus (Gr43a^GAL4)⁶ in combination with the Cha^7.4kb-GAL80 transgene, which prevents expression outside the brain²² (hereafter, brain Gr43a^GAL4; Extended Data Fig. 1b). Silencing brain Gr43a neurons for 3 h immediately after spaced training abolished aversive LTM performance (Fig. 1a (i) and Extended Data Fig. 1c). By contrast, shifting the silencing time windows—starting 3 h (Fig. 1a (ii)) or 6 h (Extended Data Fig. 1d) after training for 3 h—did not affect LTM. When flies were kept at the permissive temperature, no silencing was expected to occur and LTM performance was normal (Extended Data Fig. 1e). Silencing Gr43a neurons for 3 h after massed training left 24 h memory performance intact, and silencing Gr43a neurons for 1 h after a single training session had no effect on 3 h memory either (Extended Data Fig. 1f). To confirm these observations, we conducted silencing experiments using a different genetic driver line, a Gr43a-GAL4 enhancer trap line²⁴ targeting two pairs of Gr43a neurons (Extended Data Fig. 1b), with similar results (Extended Data Fig. 1g). Thus, despite flies being satiated, signalling from brain fructose-sensing neurons is required in the aversive learning paradigm specifically for LTM formation, within a restricted time period after training. This result raises the question of what could activate these neurons.

In the previous series of experiments, flies were constantly maintained on food-containing vials after training. We conducted a similar experiment to silence Gr43a neurons, with the difference that flies were deprived of food during the 3 h while Gr43a neurons were silenced. Consequently, Gr43a neurons were not silenced when flies were put back on food after training. Notably, LTM performance was normal in that condition (Fig. 1a (iii)). However, with the same sequence of food accessibility, LTM was again fully impaired when the silencing of Gr43a neurons was shifted to 3–6 h after training, that is, when flies recovered access to food (Fig. 1a (iv)). Taken together, these experiments show that LTM is formed after spaced training, provided that, when flies start eating after learning, their brain fructose-sensing neurons are able to signal. Accordingly, when wild-type flies were continuously food deprived after spaced training, LTM formation was defective, although this condition did not affect memory formed after massed training (Fig. 1b). Notably, forced activation of Gr43a neurons for 1 h immediately after spaced training, mediated by ectopic expression in Gr43a neurons of the heat-activated cation channel TrpA1^25,26, rescued LTM in flies deprived of food after training (Fig. 1c). Moreover, LTM was rescued if flies were fed with sucrose or glucose, but not with coconut oil, after training (Extended Data Fig. 1h), showing a critical effect of carbohydrate-based feeding rather than energy intake in itself. This suggests that, in a regular situation, the sensitivity of brain fructose-sensing neurons is restored by spaced training, so that Gr43a neurons can be activated by the onset of food intake after spaced training, giving rise to signalling that is critical for LTM formation. These findings raise two questions that we address in subsequent parts of this study: how Gr43a neurons can be activated and what signal they release to initiate memory consolidation.

To examine the sensitivity of brain fructose-sensing neurons, we conducted in vivo calcium imaging experiments in response to bath application of fructose. In naive, 24-h food-deprived flies, Gr43a neurons showed a strong and long-lasting calcium increase in response to fructose stimulation. By contrast, we detected no calcium response in satiated flies (Fig. 1d). This experiment therefore confirmed the state-dependent nutrient sensitivity of Gr43a neurons in naive flies⁹. But, given that aversive learning experiments are performed on satiated flies with free access to food before conditioning, this raises the question of how Gr43a neurons are activated by feeding after spaced training.

We performed similar calcium imaging experiments comparing flies that had received spaced training with flies that went through an unpaired protocol of similar duration (in which electric shocks and odours were delivered separately). Notably, after spaced training, fructose application resulted in a strong response in brain Gr43a neurons, of the same magnitude as in naive fasted flies, whereas it did not elicit a response in the case of an unpaired protocol, as in naive satiated animals (Fig. 1e). While these experiments involved stimulation with a relatively high fructose concentration (75 mM), similar results were obtained using a more sensitive calcium sensor (GCaMP6s), allowing a concentration of fructose stimulation at a physiologically relevant concentration (20 mM) (Extended Data Fig. 2a). The recovery of fructose sensitivity was specific to spaced training, as Gr43a neurons were not responsive to fructose after massed training (Extended Data Fig. 2b). After spaced training, even without food intake, the fructose sensitivity of Gr43a neurons extinguished within 6 h after training (Extended Data Fig. 2c,d). Accordingly, flies granted access to food only from 6 h after spaced training were unable to form LTM (Extended Data Fig. 2e). These results demonstrate that spaced associative aversive training, despite not involving food-relevant stimuli, opens a time window of a few hours when the fructose sensitivity of Gr43a neurons in the brain is restored, raising the question of how this is implemented.

In naive satiated flies, Gr43a neurons are inhibited by their upstream dFB neurons⁹. Indeed, it was reported that dFB neurons are highly active in fed flies, as in vivo calcium imaging reported large-amplitude spontaneous calcium transients in these neurons, whereas this calcium activity was substantially reduced in fasted flies⁹. Through similar in vivo calcium imaging measurements performed in satiated flies, we observed that dFB neurons show large-amplitude ongoing calcium activity in flies that received an unpaired conditioning protocol. Notably, spaced training induced a substantial reduction in dFB neuron activity, manifested as a decrease in the amplitude and the power spectrum of dFB calcium signals (Fig. 2a). Spaced training therefore restores dFB neurons to a starvation-like physiological state, thereby resetting the activatability of Gr43a neurons. If this physiological shift leading to dFB neuron inhibition is critical to activate Gr43a neurons, we reasoned that forcing dFB neuron activity should be detrimental for LTM formation specifically (similar to silencing of Gr43a neurons). Notably, activating dFB neurons immediately after spaced training for 1 h using TrpA1 expression abolished LTM performance (Fig. 2b). When flies were kept at the permissive temperature, LTM remained intact (Extended Data Fig. 3a). Activation of dFB neurons after massed or single trial training had no effect on memory (Extended Data Fig. 3a). These results were replicated using three distinct driver lines targeting dFB neurons (Extended Data Fig. 3b). As would be expected, the fructose response in Gr43a neurons after spaced training was abolished after 1 h activation of dFB neurons (Fig. 2c and Extended Data Fig. 3c,d).

Glutamatergic neurons—called Janus neurons—are presynaptic to dFB neurons in a brain region called the asymmetric body (Extended Data Fig. 4a), and have been shown to induce the onset of dFB neurons calcium oscillations when food satiation is reached⁹. With the hypothesis that spaced training might silence Janus neurons, which would result in the observed dampening of dFB neuron activity, we reasoned that forcing activity of Janus neurons might impair LTM, as did dFB neuron activation. However, TrpA1-mediated activation of Janus neurons after spaced training did not impair LTM performance (Fig. 2d). Searching for alternative pathways, we noticed that one of the genomic enhancer fragments that drives expression in dFB neurons (VT038216) derives from muscarinic acetylcholine receptor B (mAchR-B), which is unique among cholinergic receptors in that it reportedly exerts inhibitory action after activation^27,28. This raised the hypothesis that a cholinergic input could inhibit dFB neurons after spaced training. We therefore addressed the effect of a genetic knockdown of mAchR-B in dFB neurons on LTM formation, using two previously characterized RNA interference (RNAi) constructs targeting mAchR-B²⁸. We used the TARGET system²⁹ to achieve inducible GAL4-mediated RNAi expression. This strategy relies on the ubiquitous expression of a thermosensitive GAL4 inhibitor under a tubulin promoter (tub-GAL80^ts). Induction of RNAi expression at the adult stage was achieved by placing flies at 30 °C for 3 days before conditioning. mAchR-B knockdown in dFB neurons impaired LTM (Fig. 2e and Extended Data Fig. 4b), but did not alter naive odour or electric shock avoidance (Supplementary Table 1), indicating that the observed memory impairment was not due to impaired sensory perception. LTM performance was normal when RNAi expression was not induced, excluding the possibility that the observed memory defect could be due to leaky RNAi expression during development (Extended Data Fig. 4b). Memory measured 24 h after massed training or 3 h after single training was not affected by mAchR-B knockdown (Extended Data Fig. 4b). These results were reproduced with the second RNAi construct (Extended Data Fig. 4c). Last, when mAchR-B expression was inhibited in dFB neurons, spaced training did not decrease their calcium activity (Fig. 2f and Extended Data Fig. 4d). Using FlyWire^30,31,32 to examine the reconstructed fly brain connectome derived from the full adult fly brain (FAFB) dataset³³, we identified a pair of neurons (FB.5 and FB.6 neurons, hereafter FB.5/6) predicted to be cholinergic that are presynaptic to dFB neurons in the asymmetric body³⁴ (Fig. 2g; details are provided in the Supplementary Note and Supplementary Tables 7 and 8). NeuronBridge³⁵ enabled us to identify that the split-GAL4 line SS01491 targets these neurons (Extended Data Fig. 4d). According to a previous study, this line additionally targets three pairs of female-specific PC1 neurons³⁶, but restrictively targets FB.5/6 neurons within the brain and VNC in male flies³⁶. Notably, knockdown in FB.5/6 neurons of the choline acetyltransferase (ChAT) enzyme involved in acetylcholine biosynthesis using an RNAi line that we previously validated³⁷ impaired LTM (Fig. 2h), both in male and female flies (Extended Data Fig. 4f). This result is in accordance with the prediction of FB.5/6 neurons being cholinergic and, importantly, reveals their specific involvement in LTM formation, as no effect was observed after massed or single training (Extended Data Fig. 4f). Moreover, flies expressing Shi^ts through SS01491 showed a memory defect after spaced training (Fig. 2i), in both male and female flies (Extended Data Fig. 4g), but not after massed or single training (Extended Data Fig. 4g). Together, our data demonstrate that resetting the nutrient sensitivity of Gr43a neurons, through the silencing of their upstream inhibitory neurons, is critical for LTM formation. This effect mobilizes a cholinergic pathway that inhibits dFB neurons in an experience-dependent manner, in parallel to feeding state-dependent glutamatergic modulation by Janus neurons. While our data identify FB.5/6 neurons as the probable final element of this new circuit, the full delineation of how spaced learning patterns translates into dFB neurons inhibition will require further investigation.

Brain fructose-sensing neurons detect sugar intake through variations in fructose levels, which are sensed by the Gr43a receptor that they express⁶. We therefore addressed the effect of a genetic knockdown of Gr43a in these neurons on LTM formation. The efficiency of all RNAi constructs used in this study that had not been already characterized was verified using quantitative PCR with reverse transcription (RT–qPCR; Supplementary Tables 5 and 6). Knockdown of Gr43a in brain fructose-sensing neurons resulted in a defect in LTM (Fig. 3a, Extended Data Fig. 5a and Supplementary Table 2). Memory measured 24 h after massed training or 3 h after a single training was not affected by Gr43a knockdown (Extended Data Fig. 5a). These behavioural experiments were replicated using a distinct, non-overlapping Gr43a RNAi construct (Extended Data Fig. 5b). Knockdown of Gr64a, another sugar receptor that is reportedly expressed by Gr43a neurons but not involved in their fructose response³⁸, had no effect on LTM (Extended Data Fig. 5c and Supplementary Tables 5 and 6). The requirement of the Gr43a receptor in fructose-sensing neurons for aversive LTM is consistent with the fact that these neurons are activated by food intake after spaced training.

Regarding which molecular signalling is delivered by these neurons, brain Gr43a neurons express the neuropeptide corazonin (Crz)²²; however, Crz knockdown in Gr43a neurons did not impair LTM (Extended Data Fig. 5d and Supplementary Tables 5 and 6), prompting a search for alternative candidates. From a publicly available single-cell transcriptomics dataset of around 60,000 fly brain cells³⁹, we identified a subset of approximately 10–12 cells with strong co-expression of both Gr43a and Crz, which we further considered as putative Gr43a neurons. In the transcript list of these candidate cells (Supplementary Data 1), we noticed a consistently strong expression of glycoprotein beta subunit 5 (Gpb5). Gpb5 heterodimerizes with glycoprotein alpha subunit 2 (Gpa2) to form a highly conserved glycoprotein hormone, Gpa2–Gpb5¹¹ (known as thyrostimulin in vertebrates). Gpa2–Gpb5 is the ligand of leucine-rich-repeat-containing receptor 1 (Lgr1)⁴⁰, which, according to the same transcriptomics dataset, is expressed in neurons of the MB, the major brain region underlying associative memory encoding. Using a previously validated antibody against Gpb5⁴¹, we indeed observed immunostaining in the cell bodies of brain fructose-sensing neurons, which was strongly decreased after expression of an RNAi against Gpb5 (Fig. 3b, Supplementary Tables 5 and 6). We therefore tested whether Gpb5 could be involved in LTM-relevant signalling by these neurons. Knockdown of Gpb5 in brain Gr43a neurons led to a strong impairment of LTM (Fig. 3c and Extended Data Fig. 5e). Memory measured 24 h after massed training or 3 h after a single training was not affected by Gpb5 knockdown (Extended Data Fig. 5e). This result was confirmed using a distinct non-overlapping RNAi against Gpb5 (Extended Data Fig. 5f). We then performed a similar series of experiments using an RNAi against Gpa2, the molecular partner of Gpb5 (Supplementary Tables 5 and 6). As for Gpb5, inducible Gpa2 knockdown in brain fructose-sensing neurons specifically impaired LTM formation (Extended Data Fig. 5g,h). As this was the only available RNAi construct, and no Gpa2 antibody had been characterized, we did not further investigate the role of Gpa2. Overall, these results reveal that Gpa2–Gpb5 signalling by Gr43a neurons supports LTM formation after spaced training.

The neuronal silencing experiments presented earlier show that signalling from Gr43a neurons occurs shortly after spaced training. Although genetic knockdown experiments do not offer the same acute manipulation as temperature-induced silencing, we sought to confirm that thyrostimulin signalling also operates on the same time scale. In our previous studies, in vivo imaging of the metabolic activity of MB neurons using pyruvate and glucose FRET biosensors enabled the identification of two early metabolic hallmarks required for LTM formation that are observable within 2 h after spaced training: an increased rate of mitochondrial pyruvate uptake in the axons of Kenyon cells, the MB-intrinsic neurons^14,42, and increased glucose consumption by the pentose phosphate pathway in the somatic compartment of these neurons¹⁵. In vivo two-photon pyruvate imaging experiments conducted in that time window in the axonal compartment of MB neurons confirmed that spaced training induces an increased pyruvate metabolic rate as compared to a spaced unpaired protocol (Fig. 3d). Notably, this effect was lost when either Gr43a or Gpb5 was knocked down in Gr43a neurons (Fig. 3d). Similarly, glucose imaging in Kenyon cell somas showed that the increased glucose consumption induced by spaced training was impaired by Gr43a or Gpb5 knockdown in Gr43a neurons (Extended Data Fig. 6a). These results therefore support the idea that both Gr43a-mediated fructose sensing and Gpa2–Gpb5 signalling by brain Gr43a neurons allow the metabolic activation of MB neurons in the first hours after spaced training, which was previously shown to be a critical initial step in memory consolidation^14,42.

Brain Gr43a neurons do not anatomically project to the MB region⁶. However, as hormones and neuropeptides can mediate long-range signalling, Gpa2–Gpb5 released by Gr43a neurons could still act act on MB neurons. Three major populations of neurons—α/β, α′/β′ and γ—have been defined within Kenyon cells. Single-cell transcriptomics analysis has revealed that the Gpa2–Gpb5 receptor Lgr1 is a specific marker of the cluster of α/β neurons³⁹, the subpopulation that is known to be pivotal in LTM encoding⁴³. To confirm that Lgr1 is expressed in α/β neurons, we generated a fly line in which a HA-tag coding sequence was knocked in at the C-terminal end of the Lgr1 gene. In the MB region, these flies showed strong HA-positive staining in the α and β lobes, corresponding to the axons of α/β neurons, exclusively. This staining was significantly decreased after expression of an RNAi targeting Lgr1 in α/β neurons (Fig. 3e and Supplementary Tables 5 and 6), confirming the strong and preferential expression of Lgr1 in α/β neurons as compared to subsets of other Kenyon cells. This prompted us to question the role of Lgr1 in α/β Kenyon cells in LTM formation. Inducible knockdown of Lgr1 in α/β Kenyon cells led to a strong impairment in LTM (Fig. 3f and Extended Data Fig. 6b). Memory measured 24 h after massed training or 3 h after a single training was not affected by Lgr1 knockdown (Extended Data Fig. 6b). We replicated these results using a distinct non-overlapping RNAi against Lgr1 (Extended Data Fig. 6c). Moreover, Lgr1 knockdown in α/β MB neurons prevented the increased pyruvate metabolism and glucose consumption in the vertical lobes of MB neurons (Fig. 3g and Extended Data Fig. 6d). Together, these results argue for a direct hormonal action mediated by thyrostimulin from brain fructose-sensing neurons on α/β neurons of the MB, enabling their metabolic activation.

In fasted flies, Gr43a neurons activation promotes food intake⁶. As spaced training resets these neurons in a fasted-like state, we sought for possible consequences on feeding behaviour, using a two-choice feeding assay (FlyPad⁴⁴; Fig. 4a) to assess the preference of single flies for sucrose versus agarose—a substrate with low energy value. As expected, naive fasted flies in this assay displayed strong preference for sucrose, in contrast to satiated flies (Extended Data Fig. 7a). After spaced associative training, flies developed a strong preference for sucrose, compared with the unpaired controls, an effect that did not occur after massed training (Fig. 4b and Extended Data Fig. 7b,c). Knockdown of Gr43a or Gpb5 in fructose-sensing neurons or knockdown of Lgr1 in α/β MB neurons strongly impaired the effect of spaced training on feeding behaviour (Fig. 4d–f), showing that a common signalling from Gr43a neurons supports memory consolidation and bias feeding behaviour after spaced training.

The fact that spaced training can be more efficient than intensive (or massed) training in lifting the default inhibition of LTM is a well-documented cognitive effect known as the spacing effect, and is conserved from invertebrates to humans¹⁹. Comparing the contexts of appetitive and aversive LTM formation, we wondered whether the disinhibition of fructose sensors after spaced training in satiated flies could be integral to the spacing effect; if so, we reasoned that mimicking the silencing of dFB neurons that is normally achieved by spaced training should be sufficient to switch the brain to an LTM-ready state, and therefore facilitate memory consolidation.

dFB neurons inhibit Gr43a neurons through Tk signalling acting on the Tk receptor Tk99D in Gr43a neurons⁹. Knockdown of TkR99D in brain Gr43a neurons had no detectable effect on 24 h memory performance after either spaced or massed training (Fig. 5a and Extended Data Fig. 9a), showing that the magnitude of the memory was not increased. We next subjected flies to a spaced training with fewer learning sessions. Notably, only two spaced trials induced a 24 h persistent memory in flies expressing the TkR99D RNAi in Gr43a neurons, whereas this protocol did not induce significant memory in genotypic control groups (Fig. 5a and Supplementary Table 4). The memory performance of TkR99D-knockdown flies after two spaced training sessions (2× spaced training) was comparable to that of wild-type flies after a regular 5× spaced training (Fig. 5a). No increase occurred in the absence of induction (Extended Data Fig. 9a). These results were replicated with a second non-overlapping RNAi against the Tk receptor (Extended Data Fig. 9b). In accordance with our hypothesis, imaging experiments revealed that two spaced training sessions were sufficient to restore fructose sensitivity in TkR99D-knockdown flies, as opposed to genotypic control flies (Fig. 5b). To further confirm this facilitative effect, we aimed to silence dFB neurons using Shi^ts. Similar to Tk receptor knockdown in Gr43a neurons, silencing dFB neurons for 3 h immediately after 2× spaced training induced an increase in 24 h memory (Fig. 5c and Extended Data Fig. 9c,d), while delayed silencing had no effect (Extended Data Fig. 9c). Notably, this facilitative effect did not occur when flies had no access to food during the silencing period (Fig. 5c and Extended Data Fig. 9d), consistent with food intake being necessary to activate Gr43a neurons once they are disinhibited.

Drosophila strains and culture

Flies (D. melanogaster) were raised on standard medium (inactivated yeast 6% (w/v); corn flour 6.66% (w/v); agar 0.9% (w/v); methyl-4hydroxybenzoate 22 mM) under a 12 h–12 h light–dark cycle at 18 °C with 60% humidity (unless mentioned otherwise). All of the experiments were performed on young (aged <5 days) adult flies. For behaviour experiments, groups of mixed-sex flies were used unless indicated otherwise. For imaging experiments involving surgery, female flies were used because of their larger size. All flies obtained from libraries or received after the injection of transgenes were outcrossed for five generations to a reference strain. In general, this reference line carried the w¹¹¹⁸ mutation in an otherwise Canton-S genetic background; as an exception, and because TRiP RNAi transgenes are labelled by a y⁺ marker, lines from this collection were outcrossed to a y¹w^67c23 strain in an otherwise Canton-S background. To restrict UAS/GAL4-mediated expression to the adult stage, we used the TARGET system involving the ubiquitous expression of the thermosensitive GAL4 inhibitor under a tubulin promoter (tub-Gal80^ts)²⁹. In general, for crosses involving binary expression control systems (GAL4/UAS and/or LexA/LexAop), female flies carrying the driver transgene(s) were crossed to male flies carrying the effector transgene(s). A list of all of the single-transgene strains used in this study is provided in Supplementary Table 9. When needed, fly lines carrying combinations of multiple transgenes were obtained from these lines through routine crossing schemes.

Classical aversive or appetitive olfactory conditioning

To induce RNAi expression using the TARGET system, adult flies (aged 0 to 2 days) were kept at 30.5 °C for 3 days before conditioning. In the case of appetitive conditioning, flies were transferred to starvation vials (containing only a mineral-water-soaked cotton disk) for the last 16 h of the induction time. For experiments that did not involve thermal induction of transgene expression, experimental flies (aged 0-3 days) were transferred to fresh bottles containing standard medium 24 h before conditioning in the case of aversive conditioning experiments. For appetitive conditioning, flies (aged 0–2 days) were transferred to fresh food vials 1 day before being transferred to starvation vials for 21 h at 25 °C.

The aversive behaviour experiments, including sample sizes, were conducted similarly to other studies from our research group. Groups of 20–50 flies were subjected to one of the following olfactory conditioning protocols: a single cycle (1× training; duration of around 4 min), five consecutive associative training cycles (5× massed training; duration of around 20 min) or five associative cycles spaced by 15 min intertrial intervals (5× spaced training; duration of around 1 h 30 min). Non-associative control protocols (unpaired protocols) were also used for imaging experiments. Conditioning was performed using previously described barrel-type machines that allow parallel training of up to six groups. Throughout the conditioning protocol, each barrel was plugged into a constant air flow at 2 l min⁻¹. For a single cycle of associative training, flies were first exposed to an odorant (the CS+) for 1 min while 12 pulses of 5-s-long 60 V electric shocks were delivered; flies were then exposed 45 s later to a second odorant without shocks (the CS−) for 1 min. The odorants 3-octanol and 4-methylcyclohexanol, diluted in paraffin oil to a final concentration of 2.79 × 10⁻¹g l⁻¹, were alternately used as conditioned stimuli. In all of the behaviour experiments, two conditioning protocols were conducted sequentially on two batches of genotypically identical flies: a first set of flies was conditioned with 3-octanol as CS+ and 4-methylcyclohexanol as CS−, and a second reciprocal set with 4-methylcyclohexanol as CS+ and 3-octanol as CS−. This enabled us to balance for potential systematic choice bias during memory retrieval test (see below), as is commonly done in Drosophila memory experiments. During unpaired conditionings, the odour and shock stimuli were delivered separately in time, with the onset of electric shock delivery occurring 3 min and terminating 2 min before the first odorant.

The appetitive behaviour experiments, including the sample sizes, were conducted similarly to other studies from our research group. During appetitive conditioning, flies were first exposed to an odorant (the CS+) for 1 min paired with a dried sugar reward, followed 45 s later by a second odorant (the CS−) presented without reward for 1 min, as described previously⁴⁶. The odorants were the same, and were used at the same concentration, as for the aversive conditioning experiments. Plexiglas tubes covered with dried sugar were prepared the day before conditioning: a 1.5 M sucrose solution in mineral water was spread on the tube surface and tubes were placed in front of a ventilator overnight at room temperature.

Test of memory retrieval

For aversive memory behaviour experiments, unless indicated otherwise in the figures, flies were kept on standard medium between conditioning and the memory test, at either 25 °C for flies tested 3 h after 1× training, or at 18 °C for flies tested 24 h after training. After 1× training, to assess separately the two components of memory formed after 1× training (middle-term memory, which is anaesthesia sensitive, and anaesthesia-resistant memory), a subset of the trained groups were subjected to a 4 °C cold treatment for 2 min, 2 h after training (that is, 1 h before the memory test). For appetitive memory behaviour experiments, flies were kept in starvation vials between conditioning and the memory test. The memory test was performed in a T-maze apparatus, typically 3 h after single-cycle training or 24 h after massed or spaced training. Each arm of the T-maze was connected to a bottle containing 3-octanol or 4-methylcyclohexanol, diluted in paraffin oil to a final concentration identical to the one used for conditioning. Flies were given 1 min in complete darkness to choose between either arm of the T-maze. A score was calculated as the number of flies in the CS− arm of the T-maze minus the number of flies in the CS+ arm, divided by the total number of flies in both arms. To balance for potential systematic choice bias due to a preference for either side of the maze, we combined the performance of flies conditioned with either 3-octanol or 4-methylcyclohexanol as the CS+ condition. Thus, a single performance index value reported on graphs is the average of two scores obtained from two groups of genotypically identical flies conditioned sequentially using either odorant (3-octanol or 4-methylcyclohexanol) as the CS+ condition. In particular, for cases that required counting a subset of the assayed flies after the memory test (for example, disaggregated evaluation of female and male flies, or the presence of a balancer chromosome in a parental line), scores involving fewer than six flies in total were discarded to avoid giving disproportionate statistical importance to a small number of flies. The indicated n is the number of independent performance index values for each genotype.

Innate shock avoidance assessment

Shock-response tests were performed at 25 °C by placing flies in two connected chambers identical to those used for olfactory conditioning. Electric shocks were delivered in only one of the compartments. Flies were given 1 min to move freely in these compartments, after which they were trapped, collected and counted. The compartment where the electric shocks were delivered was alternated between two consecutive groups. Shock avoidance was calculated as for the memory performance.

Innate olfactory acuity assessment

As the delivery of electric shocks can modify olfactory acuity, our olfactory avoidance tests were performed on flies that had first been presented with another odour paired with electric shocks. Innate odour avoidance was measured in a T-maze similar to those used for memory tests, in which one arm of the T-maze was connected to a bottle with odour diluted in paraffin oil at the same concentration as that used for olfactory conditioning, and the other arm was connected to a bottle with paraffin oil only. Naive flies were given the choice between the two arms during 1 min. The odour-interlaced side was alternated for successive tested groups. At these concentrations, both odorants (octanol and methylcyclohexanol) are innately repulsive.

Sugar preference assessment

Innate sugar preference assessment was measured at 25 °C in a T-maze similar to those used for memory tests, in which only one of the two arms of the T-maze was covered with dried sugar. Naive flies were given the choice between the two arms during 1 min. The tests were performed on starved flies submitted to the thermal RNAi induction protocol described above. The arm with sugar was placed alternately on the right or left between two consecutive groups. Sugar response was calculated as for the memory performance.

In vivo pyruvate and glucose imaging

Pyruvate imaging experiments were performed on flies expressing the pyruvate sensor Pyronic in MB neurons through the VT30559-GAL4 driver or 13F02-LexA driver in combination with either the UAS-Pyronic or LexAop-Pyronic line, which were previously described^14,16. Glucose imaging experiments were performed on flies expressing the glucose sensor FLII12Pglu-700μδ6 in MB neurons through the 13F02-LexA driver, in combination with LexAop-FLII12Pglu-700μδ6. RNAis were expressed in MB neurons using the inducible tub-GAL80^ts; VT30559-GAL4 driver. Crosses for imaging experiments were raised at 23 °C. To achieve the induction of RNAi expression, adult flies were kept at 30.5 °C for 3 days before conditioning. Data were collected indiscriminately from 30 min to 1.5 h after 5× spaced training. A single fly was picked after few-second immobilization in a prechilled tube, and prepared for imaging using an established method¹⁴. The head capsule was opened, and the brain exposed by gently removing the superior tracheae. The head capsule was bathed in artificial haemolymph solution for the duration of the preparation. The composition of this solution was as follows: NaCl 130 mM (Sigma-Aldrich, S9625), KCl 5 mM (Sigma-Aldrich, P3911), MgCl₂ 2 mM (Sigma-Aldrich, M9272), CaCl₂ 2 mM (Sigma-Aldrich, C3881), D-trehalose 5 mM (Sigma-Aldrich, 9531), sucrose 30 mM (Sigma-Aldrich, S9378) and HEPES hemisodium salt 5 mM (Sigma-Aldrich, H7637). At the end of surgery, any remaining solution was wicked away and a fresh 90 μl droplet of this solution was applied on top of the brain. Two-photon imaging was performed using the Leica TCS-SP5 upright microscope equipped with a ×25/0.95 NA water-immersion objective. Two-photon excitation was achieved using a Mai Tai DeepSee laser tuned to 825 nm (for pyruvate sensor) or 820 nm (glucose sensor). Light emission was collected on external PMTs in the spectral ranges 468–500 nm and 529–556 nm for the blue and yellow fluorophores, respectively. Images were acquired using Leica LAS-AF (v.2.7.3).

Measurements of mitochondrial pyruvate consumption were performed as previously described^14,16,42. This consists in pharmacological blocking of the mitochondrial respiratory chain by sodium azide, which induces an acute increase in cytosolic pyruvate, at a rate that reflects the pyruvate consumption rate before blockade. Two-photon images were acquired in both channels at a frame rate of two images per second. After 1 min of baseline acquisition, 10 µl of a 50 mM sodium azide solution (Sigma-Aldrich, 71289; prepared in the same artificial haemolymph solution) was injected into the 90 µl droplet bathing the fly’s brain, bringing the sodium azide to a final concentration of 5 mM. To analyse the pyruvate imaging experiments, regions of interest (ROIs) were delimited by hand around each visible MB vertical lobe and the average intensity of both mTFP and Venus channels over each ROI were calculated over time after background subtraction. The Pyronic sensor was designed so that FRET from mTFP to Venus decreases when pyruvate concentration increases. To obtain a signal that positively correlates with pyruvate concentration, the inverse FRET ratio was computed as the mTFP intensity divided by the Venus intensity. This ratio was normalized to the baseline value calculated over the 30 s before drug injection. The rate of azide-induced pyruvate surge was calculated as the slope between 10 and 70% of the plateau.

Cellular glucose consumption measurements were performed as previously described¹⁵. This consisted of pharmacological inhibition of the enzyme trehalase by application of validamycin A, resulting in a decrease in the cytosolic glucose concentration. The kinetics of the glucose decrease were reflective of the rate of glucose consumption. Two-photon images were acquired in both channels at a frame rate of 1 image per second. Validamycin A (Sigma-Aldrich, 32347) was directly diluted into the artificial haemolymph solution at a final concentration of 40 mM, aliquoted and stored at −20 °C. A freshly thawed aliquot was used for every fly. After 2 min of baseline acquisition, 10 μl of the solution was added to the 90 μl saline droplet on top of the brain, bringing validamycin A to a final concentration of 4 mM. The signal was then acquired for another 14 min. ROIs were delimited by hand around the labelled ROIs (somas of MB neurons). The average intensity of the YFP and CFP channels over each ROI was calculated over time after background subtraction. The FRET ratio (YFP/CFP) of the FLII12Pglu-700μδ6 glucose sensor was computed to obtain a signal that positively correlates with glucose concentration. This ratio was normalized to the baseline value calculated over the 30 s before drug injection. As the decrease in FRET signal was not linear, the area over the curve (AOC) was calculated as a metric of the decrease that was positively correlated with glucose consumption. The AOC was calculated as the integral between 200 s and 900 s of the acquisition. The indicated n value is the number of flies that were assayed in each condition.

In vivo calcium imaging

Calcium imaging experiments were performed by expressing the genetically encoded calcium reporters GCaMP3 or GCaMP6s in Gr43a neurons, and GCaMP6f in dFB neurons. GCaMP3 was initially chosen to image Gr43a neurons as its higher baseline fluorescence enabled localization of the neurons, which we could not achieve with the GCaMP6f sensor. Using GCaMP6s, another GCaMP6 variant with higher baseline fluorescence, we could verify that a more-sensitive sensor and a lower fructose concentration yielded similar results to GCaMP3. Fly preparation was performed as described above for in vivo pyruvate and glucose imaging. Two-photon imaging was performed on the Leica SP8 DIVE microscope equipped with spectrally tunable hybrid detectors and a ×25/1.0 NA water-immersion objective, coupled to an Insight X3 dual-emission IR laser (Spectra Physics). Two-photon excitation was tuned at 920 nm. Emitted light was collected in the 492–570 nm spectral range. Images were acquired with Leica LAS-X software v.3.5.7. Images were acquired at a rate of one frame every 385 ms for a total of 7 min. For experiments performed on starved flies, flies were starved for 24 h at 25 °C before imaging. The composition of the physiological solution for starved flies (Fig. 1d) was the same as used for glucose imaging, except that it contained 36 mM ribose (Sigma-Aldrich, W379301) instead of sucrose and trehalose, as in ref. ³⁷. For fructose-response experiments, fructose was added at a final concentration of 75 mM (or 20 mM when using GCaMP6s) after 120 s. This fructose concentration was previously shown to elicit Gr43a responses in the fly brain²³. Image alignment was performed using the Template Matching plugin in Fiji to correct for motion across frames and ensure accurate measurement of calcium responses. After alignment, ROIs for data analysis were delimited by hand and the signal was calculated over time after background subtraction and normalized to the baseline value calculated over the 30 s before fructose injection. Calcium response was calculated as the average from the injection time to the end of the recording. For experiments measuring the spontaneous calcium activity in dFB neurons, signals were analysed as reported previously²⁶.

Sucrose preference assay using FlyPad

The food-choice assay was done using FlyPAD⁴⁴ according to the manufacturer’s instructions. One channel of the arena was loaded with 3 μl of 1% agar mixed with sucrose 0.6 M (Sigma-Aldrich), while the other channel was loaded with 3 μl of 1% agar. Single flies were captured by mouth aspiration and gently blown into each arena. Data were acquired for 60 min. Interaction with either channel (number of sips) was extracted for each arena. Flies showing zero interaction with both electrodes were not taken into account for further analysis. A sucrose preference index (PI) for each fly was calculated as: (interactions with sucrose − interactions with agar)/(interactions with sucrose + interactions with agar).

Lgr1-HA line generation

The Lgr1-HA line was generated using CRISPR–Cas9 (outsourced to Rainbow Transgenic Flies). The 3×HA sequence (TACCCATACGATGTTCCTGACTATGCGGGCTATCCCTATGACGTCCCGGACTATGCAGGATCCTATCCATATGACGTTCCAGATTACGCT) was preceded by a GS linker (GGTGGCGGCGGAAGCGGAGGTGGAGGCTCG) and inserted at the end of the Lgr1 coding sequence using a guide RNA (5′-ATTATGTTTAAACCGAACCCGGG-3′). A loxP-flanked sequence of the mini-white gene was added after the Lgr1 coding sequence to provide a phenotypic marker of the presence of the HA-tagged construct.

Immunohistochemistry

Before dissection, 2- to 4-day-old female flies from appropriate crosses were fixed in 4% paraformaldehyde in PBST (PBS containing 1% Triton X-100) at 4 °C overnight. Brains were dissected on ice in PBS solution and rinsed three times for 20 min in PBST, then blocked with 2% BSA in PBST for 2 h. Next, brains were incubated with primary antibodies. For primary antibodies, this study used 1:400 rabbit anti-GFP (Invitrogen, A11122), 1:200 rat anti-Gpb5⁴¹ (provided by P. Herrero), 1:200 rat anti-HA (Roche, 11867423001) and 1:100 mouse anti-nc82 (DSHB, nc82). Primary antibodies were incubated in the blocking solution (2% BSA in PBST) at 4 °C overnight. The next day, brains were rinsed three times for 20 min with PBST and then incubated for 3 h at room temperature with secondary antibodies diluted in blocking solution. For secondary antibodies, we used 1:400 anti-rabbit conjugated to Alexa Fluor 488 (Invitrogen, A11034), 1:400 anti-rat conjugated to Alexa Fluor 594 (Invitrogen, A11007), 1:400 anti-mouse conjugated to Alexa Fluor 594 (Invitrogen, A11005) and 1:400 anti-rat conjugated to Alexa Fluor 488 (Invitrogen, A11006). Brains were then rinsed once in PBST for 20 min, and twice in PBS for 20 min. After rinsing, brains were mounted in Prolong Mounting Medium. Images were acquired with a Nikon A1R confocal microscope using Nikon NIS-Element v.4.40 software or with an Olympus BX61-FluoView FV1000 confocal microscope using FV10-ASW v.4.2 software, as z-stack slices in 1 µm steps. For expression patterns of genetic driver lines, maximum projections are shown, obtained using Fiji (Image J1.52p).

RT–qPCR analyses

To validate the efficiency of the knockdowns used in this study, the mRNA of the target gene was measured using RT–qPCR. Female flies carrying the elav-GAL4 pan-neuronal driver were crossed with either male flies carrying the specified UAS-RNAi or with CS male flies, and the resulting crosses were reared at 25 °C. Adult fly progeny (aged 0–1 days) were transferred to fresh food for 1 day before RNA extraction. RNA extraction and cDNA synthesis were performed as described previously¹⁶ using the same reagents: the RNeasy Plant Mini Kit (Qiagen), the RNA MinElute Cleanup kit (Qiagen), 614 oligo(dT)20 primers and the SuperScript III First-Strand kit (Life Technologies). The cDNA level for each gene of interest was compared against the level of the αTub84B (CG1913) reference cDNA. Amplifications were performed using a LightCycler 480 (Roche) and the SYBR Green I Master mix (Roche). Reactions were carried out in triplicate. The specificity and size of amplification products were assessed by melting curve analyses. Expression relative to the reference was expressed as a ratio (2^−ΔCp, where Cp is the crossing point). RT–qPCR results are presented in Supplementary Table 5. The primers used in this study are described in Supplementary Table 6.

Quantification and statistical analysis

Sample sizes in this study were similar to previous studies from our or other groups for aversive memory assays¹⁴, appetitive memory assays³⁷, metabolic imaging^15,16 and calcium imaging⁹, rather than being predetermined by statistical power estimation. As group assignment was based on the fly genotype, random assignment of the experimental animals could not be performed, and experimenters were not blinded to the group assignment during experiments. For memory assays, a single datapoint is the mean of two scores from two groups of flies of the relevant genotype conditioned with octanol or methycyclohexanol as the odorant paired with electric shocks, as described above, which represents an experimental replicate. For in vivo imaging experiments and immunohistochemistry, one replicate corresponds to one fly brain. For RT–qPCR experiments, one replicate corresponds to the average value of three separate measurements (triplicate) done in parallel on the same biological sample. For FlyPad experiments, one replicate corresponds to one fly. All behavioural experiments were run at least twice and in vivo imaging experiments were run at least three times, on different days and with different batches of flies. All replicates are included in figures and extended data figures. Memory analyses of Gr43a neurons and dFB neurons were replicated in independent series of experiments, using different genetic drivers targeting Gr43a neurons or dFB neurons. Memory analyses involving gene knockdown in Gr43a neurons were replicated in independent series of experiments with two different RNAi constructs, except Gpa2 for which only one RNAi line was publicly available. All replicates were successful. In general, comparisons of the data series between two conditions were achieved by a two-tailed unpaired t-test. Comparisons between more than two distinct groups were made using a one-way ANOVA test, followed by Tukey pairwise comparisons between the experimental group and its controls, or using a two-way ANOVA followed by Sidak pairwise comparison (Fig. 1c). ANOVA results are presented as the value of the Fisher distribution F(x,y) obtained from the data, where x is the number of degrees of freedom between groups and y is the residual number of degrees of freedom for the distribution. For the analysis of sucrose preference indices obtained with FlyPAD, data series were intrinsically non-normally distributed, therefore nonparametric statistical tests were applied, that is, Mann–Whitney test for pairwise comparisons between two conditions and Kruskal–Wallis test for comparisons across multiple groups, followed by Dunn’s multiple-comparison test between the experimental group and its controls. Statistical tests were performed using the GraphPad Prism v.9.0. On the figures, asterisks illustrate the significance level of the t-test or Mann–Whitney U-test, or of the least significant pairwise comparison after an ANOVA or –Kruskal–Wallis test.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Araujo, I. E. de, Schatzker, M. & Small, D. M. Rethinking food reward. Annu. Rev. Psychol. 71, 139–164 (2020).

   Article  PubMed  Google Scholar 

2. Malik, S., McGlone, F., Bedrossian, D. & Dagher, A. Ghrelin modulates brain activity in areas that control appetitive behavior. Cell Metab. 7, 400–409 (2008).

   Article  PubMed  CAS  Google Scholar 

3. Han, W. et al. A neural circuit for gut-induced reward. Cell 175, 665–678 (2018).

   Article  PubMed  PubMed Central  CAS  Google Scholar 

4. Berthoud, H.-R., Morrison, C. D., Ackroff, K. & Sclafani, A. Learning of food preferences: mechanisms and implications for obesity & metabolic diseases. Int. J. Obes. 45, 2156–2168 (2021).

   Article  Google Scholar 

5. Musso, P.-Y., Tchenio, P. & Preat, T. Delayed dopamine signaling of energy level builds appetitive long-term memory in Drosophila. Cell Rep. 10, 1023–1031 (2015).

   Article  PubMed  CAS  Google Scholar 

6. Miyamoto, T., Slone, J., Song, X. & Amrein, H. A fructose receptor functions as a nutrient sensor in the Drosophila brain. Cell 151, 1113–1125 (2012).

   Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

7. Tully, T. & Quinn, W. G. Classical conditioning and retention in normal and mutant Drosophila melanogaster. J. Comp. Physiol. A 157, 263–277 (1985).

   Article  PubMed  CAS  Google Scholar 

8. Tully, T., Preat, T., Boynton, S. C. & Del Vecchio, M. Genetic dissection of consolidated memory in Drosophila. Cell 79, 35–47 (1994).

   Article  ADS  PubMed  CAS  Google Scholar 

9. Musso, P.-Y., Junca, P. & Gordon, M. D. A neural circuit linking two sugar sensors regulates satiety-dependent fructose drive in Drosophila. Sci. Adv. 7, eabj0186 (2021).

   Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

10. Nakabayashi, K. et al. Thyrostimulin, a heterodimer of two new human glycoprotein hormone subunits, activates the thyroid-stimulating hormone receptor. J. Clin. Invest. 109, 1445–1452 (2002).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

11. Rocco, D. A. & Paluzzi, J.-P. V. Functional role of the heterodimeric glycoprotein hormone, GPA2/GPB5, and its receptor, LGR1: an invertebrate perspective. Gen. Comp. Endocrinol. 234, 20–27 (2016).

    Article  PubMed  CAS  Google Scholar 

12. Bai, L. et al. Genetic identification of vagal sensory neurons that control feeding. Cell 179, 1129–1143 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

13. Wee, R. W. S. et al. Internal-state-dependent control of feeding behavior via hippocampal ghrelin signaling. Neuron 112, 288–305 (2024).

    Article  PubMed  CAS  Google Scholar 

14. Plaçais, P.-Y. et al. Upregulated energy metabolism in the Drosophila mushroom body is the trigger for long-term memory. Nat. Commun. 8, 15510 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

15. de Tredern, E. et al. Glial glucose fuels the neuronal pentose phosphate pathway for long-term memory. Cell Rep. 36, 109620 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

16. Rabah, Y. et al. Glycolysis-derived alanine from glia fuels neuronal mitochondria for memory in Drosophila. Nat. Metab. 5, 2002–2019 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

17. Bouzaiane, E., Trannoy, S., Scheunemann, L., Plaçais, P.-Y. & Preat, T. Two independent mushroom body output circuits retrieve the six discrete components of Drosophila aversive memory. Cell Rep. 11, 1280–1292 (2015).

    Article  PubMed  CAS  Google Scholar 

18. Pavlowsky, A. et al. Neuronal fatty acid oxidation fuels memory after intensive learning in Drosophila. Nat. Metab. 7, 2438–2450 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

19. Smolen, P., Zhang, Y. & Byrne, J. H. The right time to learn: mechanisms and optimization of spaced learning. Nat. Rev. Neurosci. 17, 77–88 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

20. Glas, A., Hübener, M., Bonhoeffer, T. & Goltstein, P. M. Spaced training enhances memory and prefrontal ensemble stability in mice. Curr. Biol. 31, 4052–4061 (2021).

    Article  PubMed  CAS  Google Scholar 

21. Nishijima, S. & Maruyama, I. N. Appetitive olfactory learning and long-term associative memory in Caenorhabditis elegans. Front. Behav. Neurosci. 11, 80 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

22. Miyamoto, T. & Amrein, H. Diverse roles for the Drosophila fructose sensor Gr43a. Fly 8, 19–25 (2014).

    Article  PubMed  Google Scholar 

23. Kitamoto, T. Conditional modification of behavior in Drosophila by targeted expression of a temperature-sensitive shibire allele in defined neurons. J. Neurobiol. 47, 81–92 (2001).

    Article  PubMed  CAS  Google Scholar 

24. Kwon, J. Y., Dahanukar, A., Weiss, L. A. & Carlson, J. R. Molecular and cellular organization of the taste system in the Drosophila larva. J. Neurosci. 31, 15300–15309 (2011).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

25. Hamada, F. N. et al. An internal thermal sensor controlling temperature preference in Drosophila. Nature 454, 217–220 (2008).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

26. Plaçais, P.-Y. et al. Slow oscillations in two pairs of dopaminergic neurons gate long- term memory formation in Drosophila. Nat. Neurosci. 15, 592–599 (2012).

    Article  PubMed  Google Scholar 

27. Ren, G. R., Folke, J., Hauser, F., Li, S. & Grimmelikhuijzen, C. J. P. The A- and B-type muscarinic acetylcholine receptors from Drosophila melanogaster couple to different second messenger pathways. Biochem. Biophys. Res. Commun. 462, 358–364 (2015).

    Article  ADS  PubMed  CAS  Google Scholar 

28. Manoim, J. E., Davidson, A. M., Weiss, S., Hige, T. & Parnas, M. Lateral axonal modulation is required for stimulus-specific olfactory conditioning in Drosophila. Curr. Biol. 32, 4438–4450 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

29. McGuire, S. E., Mao, Z. & Davis, R. L. Spatiotemporal gene expression targeting with the TARGET and gene-switch systems in Drosophila. Sci. STKE 2004, pl6 (2004).

    Article  PubMed  Google Scholar 

30. Dorkenwald, S. et al. Neuronal wiring diagram of an adult brain. Nature 634, 124–138 (2024).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

31. Schlegel, P. et al. Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature 634, 139–152 (2024).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

32. Eckstein, N. et al. Neurotransmitter classification from electron microscopy images at synaptic sites in Drosophila melanogaster. Cell 187, 2574–2594 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

33. Zheng, Z. et al. A complete electron microscopy volume of the brain of adult Drosophila melanogaster. Cell 174, 730–743 (2018).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

34. Pascual, A., Huang, K.-L., Neveu, J. & Préat, T. Brain asymmetry and long-term memory. Nature 427, 605–606 (2004).

    Article  ADS  PubMed  CAS  Google Scholar 

35. Clements, J. et al. NeuronBridge: an intuitive web application for neuronal morphology search across large data sets. BMC Bioinform. 25, 114 (2024).

    Article  Google Scholar 

36. Wang, F. et al. Neural circuitry linking mating and egg laying in Drosophila females. Nature 579, 101–105 (2020).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

37. Plaçais, P.-Y., Trannoy, S., Friedrich, A. B., Tanimoto, H. & Preat, T. Two pairs of mushroom body efferent neurons are required for appetitive long-term memory retrieval in Drosophila. Cell Rep. 5, 769–780 (2013).

    Article  PubMed  Google Scholar 

38. Fujii, S. et al. Drosophila sugar receptors in sweet taste perception, olfaction, and internal nutrient sensing. Curr. Biol. 25, 621–627 (2015).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

39. Davie, K. et al. A single-cell transcriptome atlas of the aging Drosophila brain. Cell 174, 982–998 (2018).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

40. Sudo, S., Kuwabara, Y., Park, J.-I., Hsu, S. Y. & Hsueh, A. J. W. Heterodimeric fly glycoprotein hormone-α2 (GPA2) and glycoprotein hormone-β5 (GPB5) activate fly leucine-rich repeat-containing G protein-coupled receptor-1 (DLGR1) and stimulation of human thyrotropin receptors by chimeric fly GPA2 and human GPB5.Endocrinology 146, 3596–3604 (2005).

    Article  PubMed  CAS  Google Scholar 

41. Sellami, A., Agricola, H.-J. & Veenstra, J. A. Neuroendocrine cells in Drosophila melanogaster producing GPA2/GPB5, a hormone with homology to LH, FSH and TSH. Gen. Comp. Endocrinol. 170, 582–588 (2011).

    Article  PubMed  CAS  Google Scholar 

42. Comyn, T., Preat, T., Pavlowsky, A. & Plaçais, P.-Y. PKCδ is an activator of neuronal mitochondrial metabolism that mediates the spacing effect on memory consolidation. eLife 13, RP92085 (2024).

43. Pascual, A. & Préat, T. Localization of long-term memory within the Drosophila mushroom body. Science 294, 1115–1117 (2001).

    Article  ADS  PubMed  CAS  Google Scholar 

44. Itskov, P. M. et al. Automated monitoring and quantitative analysis of feeding behaviour in Drosophila. Nat. Commun. 5, 4560 (2014).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

45. Krashes, M. J. & Waddell, S. Rapid consolidation to a radish and protein synthesis-dependent long-term memory after single-session appetitive olfactory conditioning in Drosophila. J. Neurosci. 28, 3103–3113 (2008).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

46. Colomb, J., Kaiser, L., Chabaud, M.-A. & Preat, T. Parametric and genetic analysis of Drosophila appetitive long-term memory and sugar motivation. Genes Brain Behav. 8, 407–415 (2009).

    Article  PubMed  CAS  Google Scholar 

47. Burke, C. J. & Waddell, S. Remembering nutrient quality of sugar in Drosophila. Curr. Biol. 21, 746–750 (2011).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

48. Mery, F. & Kawecki, T. J. A cost of long-term memory in Drosophila. Science 308, 1148 (2005).

    Article  ADS  PubMed  CAS  Google Scholar 

49. Plaçais, P.-Y. & Preat, T. To favor survival under food shortage, the brain disables costly memory. Science 339, 440–442 (2013).

    Article  ADS  PubMed  Google Scholar 

50. Lapraz, F. et al. Asymmetric activity of NetrinB controls laterality of the Drosophila brain. Nat. Commun. 14, 1052 (2023).

    Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 

51. Wolff, T. et al. Cell type-specific driver lines targeting the Drosophila central complex and their use to investigate neuropeptide expression and sleep regulation. eLife 14, RP104764 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Author notes

1. These authors jointly supervised this work: Thomas Preat, Pierre-Yves Plaçais

Authors and Affiliations

1. Energy & Memory, Brain Plasticity Unit, CNRS, ESPCI Paris, PSL Research University, Paris, France

   Raquel Francés, Typhaine Comyn, Coraline Desnous, Francesca Bettoni, Alice Pavlowsky, Thomas Preat & Pierre-Yves Plaçais

Contributions

R.F., T.P. and P.-Y.P. conceived the project. R.F. performed most of the experiments, with the following exceptions: T.C. did pyruvate imaging experiments. A.P. carried out the design and construction of the Lgr1-HA line. F.B. performed immunostainings of genetic driver line expression patterns and a subset of sensory acuity control experiments. C.D. performed a part of appetitive memory experiments involving Gr43a and Gpb5 knockdowns and set up FlyPad in the laboratory. P.-Y.P. performed behaviour experiments involving FB.5/6 neurons and carried out the routine genetic constructions. P.-Y.P. designed and supervised experiments. R.F. and P.-Y.P. visualized data and designed the figures. The manuscript was written by P.-Y.P., edited by R.F., A.P. and T.P., and reviewed and approved by all of the authors. T.P. and P.-Y.P. acquired funding.

Corresponding authors

Correspondence to Thomas Preat or Pierre-Yves Plaçais.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions