Hunger states modulate aggression via opposing dopamine pathways

Abstract

Hunger drives aggression, making animals “hangry”. Yet the underlying neural mechanism is unclear. Here, we identified a “hangry neuron” in the mushroom body of Drosophila that transforms hunger states into aggression intensity. This transformation is graded and nutrient-specific, precisely coupling internal needs to aggressive actions. Dopaminergic inputs to the hangry neuron play a critical role in this process by conveying the information about hunger states, and by bidirectionally modulating hangry neuron’s excitability via Dop1R1 and Dop1R2 receptors, which activate intracellular cAMP and calcium pathways, respectively, thereby oppositely modulating excitability. Together, our study reveals the unexpected function of the mushroom body – traditionally viewed as the learning and memory center in Drosophila – in modulating aggression, and identifies a simple circuit design in this species whereby a single neuron integrates distinct dopamine signaling pathways to mediate flexible behavior, a function that is typically implemented by the basal ganglia system in mammals.

How hunger modulates aggression is unclear. Here, authors show a single “hangry” neuron amplifies the perceived value of food to drive intense fighting by detecting dopamine dynamics via two opposing receptors in Drosophila.

Introduction

Many animals resort to aggression to secure valuable resources such as food. Yet fighting is energetically costly and fraught with risk, prompting animals to avoid it whenever possible. Nevertheless, the imperative of securing food for survival often drives animals to engage in conflict. Thus, animals must decide whether to fight depending on their level of hunger. Although hunger is known to modulate various behaviors^1–11, the neural mechanisms underlying how hunger affects aggression remain unclear.

In the wild, food availability is unpredictable, whereas digestion is a continuous process, leading to dynamic cycles of hunger and satiety. Meanwhile, engaging in fights demands significant energy and entails the risk of injury, rendering the decision to fight both complex and critical. By integrating these dynamic signals, animals may fine-tune their aggression levels in response to the interplay between external food availability and internal hunger states. To address this challenge, a neural circuit that links hunger and aggression would be beneficial.

In Drosophila, several neuron types have been implicated in aggression^12–19. Critically, activating and silencing peptidergic Tachykinin neurons strongly increases and decreases inter-male aggression¹⁴. Tachykinin neuron activation can override various cues and conditions normally required for aggression, causing males to fight even towards moving magnets. This suggests that Tachykinin neurons exert a close influence on aggressive actions, raising the possibility that they are modulated by contextual information from higher-order circuits upstream. Dopamine is broadly implicated in value and motivation in context-dependent behaviors across species^{1–3,20–28}, including hunger^5,8,29 and aggression^30,31. In Drosophila, a previous genetic screening study revealed that activation and inactivation of dopaminergic neurons increase and decrease aggression¹³, although how dopamine neurons are engaged in aggression remains unknown. In invertebrate brains, including those of Drosophila, dopaminergic neurons heavily innervate the mushroom body, a higher-order region implicated in context-dependent action selection^{1–3,32–35}, including courtship³⁶. Specifically, mushroom body dopaminergic circuits have been established as mediating hunger-dependent modulation of naïve food seeking and the expression of food-related memory^{1–3,37–39}. Whether hunger and aggression are coupled in the mushroom body dopaminergic circuit is intriguing.

Here, using the genetically tractable model organism Drosophila, we investigated the neural mechanisms underlying the “hangry” effect, a phenomenon that couples graded and nutrient-specific hunger states to varying intensities of aggression. Combining multiple approaches, we identified that a mushroom body dopaminergic circuit mediates the hangry effect. Within this circuit, a single MBON11 neuron—which we named the “hangry neuron”—both signals nutrient-specific hunger states and controls aggression intensity. The rise and decay of dopamine convey hunger state information and tune hangry neuron intrinsic excitability, which amplifies the perceived value of food to modulate aggression intensity. Notably, these effects are mediated by Dop1R1 and Dop1R2 receptors in the hangry neuron, which engage distinct intracellular signaling pathways to oppositely modulate neuronal excitability and animal’s aggression.

Results

Food and hunger affect inter-male aggression

Compared with other social interactions, such as courtship, aggressive interactions happen at a lower frequency, making them harder to detect and posing challenges for mechanistic studies. To precisely identify bouts of aggressive interactions, such as lunge, we developed a machine learning pipeline to quantify aggressive interactions⁴⁰ (Fig. 1a, Supplementary Fig. 1a, also see Methods). We began by exploring how the presence of food affects male–male interaction, using this pipeline. When we introduced two wild-type males in chambers without food, we detected few lunge bouts. However, in chambers with food at the center, lunge frequency was strongly increased (Fig. 1b, c, see Methods and Supplementary Table 2 for details of all statistical analyses), suggesting that the presence of food enhances aggression.

Fig. 1. Food and hunger affect inter-male aggression.

a Schematic of the quantitative ethology system. See Supplementary Fig. 1a for more details. b Raster plots showing lunge bouts of wild-type males in chambers with apple juice food (yellow shading) or without food (gray shading) over time. Gray ticks denote lunge bouts. Each line denotes a pair of males. Schematics of the assay shown on the left. See Supplementary Table 1 for details of all genotypes. c Lunge frequencies (seconds per minute, s/min) of wild-type males in chambers without food or with apple juice food. Data presented as mean ± SEM, same as follows. Dots and bars denote individual flies and their means, respectively, same as follows. See Methods and Supplementary Table 2 for details and results of all statistical analyses. d Lunge frequencies of wild-type males reared with or without food interact in chambers with apple juice food (schematic at the top). e Lunge frequencies of wild-type males in nutrient-specific food value assay (schematic at the top) with four combinations of rearing and testing conditions, where males reared on sucrose or yeast food interact on sucrose or yeast food. f Lunge frequencies of wild-type males in protein-hunger assay (schematic at the top) with four different rearing conditions, including sucrose (s), sucrose with amino acid (AA) mixture, sucrose with peptone, and sucrose with yeast, where males reared on different food interact on the same sucrose-yeast food.

Hunger affects the perceived value of food. To examine whether hunger modulates aggression, we compared chronically starved and ad libitum fed (satiated) males. Starved males lunged over twice as often compared to satiated males (Fig. 1d), an effect reversible upon refeeding (Supplementary Fig. 1d), confirming that hunger influences aggression. To explore how the perceived value of food modulates aggression, we reasoned that this value depends on both the nutritional composition of the food and the nutrient-specific hunger state of the subject. To test this, we induced protein or carbohydrate hunger by feeding males with sucrose (a carbohydrate source) or yeast (a protein source), respectively (Fig. 1e, schematic). Indeed, males under protein-hunger lunged more for protein than under carbohydrate-hunger, and vice versa. Additionally, protein-hungry males lunged more for protein than for carbohydrate, while carbohydrate-hungry males showed the opposite preference (Fig. 1e). Focusing on protein hunger, which more robustly influences aggression (Fig. 1e, dashed boxes), we tested natural protein, protein hydrolysate, and purified amino acid as protein supplements, by adding yeast, peptone, and an amino acid mixture to sucrose respectively (Fig. 1f, also see Methods). Feeding these protein supplements decreased lunge frequency on yeast-containing food compared to feeding sucrose (Fig. 1f), with a gradient of increasing effect: natural protein, protein hydrolysate, and amino acid mixture. A similar gradient was observed with decreasing concentrations of amino acid mixture supplements (Supplementary Fig. 1e), indicating that the protein hunger effect is graded.

Mushroom body output neuron MBON11 influences food-induced inter-male aggression

What are the neural substrates underlying food and hunger-related aggression? To identify the neurons, we developed a high-throughput screening system with multichannel LEDs (Supplementary Fig. 1a, also see Methods). To minimize disruption to fly behavior, we utilized optogenetic photoactivation via red-shifted opsin CsChrimson using 680 nm light barely visible to flies. We quantified interactions between two transgenic males in chambers with apple juice food at the center (Fig. 2a). The mushroom body (MB) is a key node for value processing and action selection³³. We wondered if MB plays a role in value based aggressive action selection. The MB contains a small number of output neurons, forming a bottleneck in the circuit. Focusing on these MB output neurons (MBONs), we identified a single pair of MB output neurons MBON11 (also called MBON-γ1pedc > α/β, or MVP2) for aggression. Using three different split-GAL4 lines to target MBON11³², we consistently found that in the presence of food, MBON11 activation strongly boosts lunge while its inhibition suppresses lunge (Fig. 2a, Supplementary Fig. 2a, c, see Methods and Supplementary Table 1 for all genotypes). We obtained similar results under all three driver lines using thermogenetic manipulation with temperature-sensitive cation channel dTrpA1 for activation, and inwardly rectifying potassium channel Kir2.1 for inhibition with tubulin-GAL80^ts for temperature-gating (Supplementary Fig. 2b). We then broadly examined MBON11 function in aggressive and courtship social interactions. In food chambers, all three drivers enhanced male-male tussling (Supplementary Fig. 2d), another aggressive interaction, but not male-female courtship (Supplementary Fig. 2e). Thus, MBON11 specifically influences inter-male aggression, but not male-female courtship. Males fight for food and mates⁴¹. We assessed MBON11 function in lunge, tussle, and courtship under various conditions (Supplementary Fig. 2f-h). For the food effect on inter-male interactions (1st vs. 2nd condition), MBON11 activation enhanced lunge and tussle by tenfold with food, but only modestly without food. For the mate effect on inter-male interactions (2nd vs. 3rd condition), by replacing food with a tethered female, MBON11 activation only enhanced mate-induced lunging, an effect tenfold lower than that with food. For male-female interactions (2nd vs. 4th condition), by replacing the conspecific male with a female in a food chamber, MBON11 activation had no effect. Together, among the twelve tested conditions, MBON11 most prominently enhances food-induced inter-male aggression (dashed boxes in Supplementary Fig. 2f, g).

Fig. 2. MBON11 mediates food-induced hunger-dependent inter-male aggression.

a Lunge frequency ratios upon MBON11 photoactivation (orange) and photoinhibition (green) in food chambers using three split-GAL4 drivers for MBON11, alongside controls. Bottom: Raster plots showing representative lunge bouts upon MBON11 (MB085C) photoactivation (red ticks) and photoinhibition (green ticks). Each line denotes a pair of males. Gray shadows denote photostimulation periods. b Lunge frequencies before (pre) and during (stim) MBON11 photoactivation in chambers with or without food. Lunge frequencies during activation in chambers with (1.18 ± 0.09 s/min) or without (0.10 ± 0.03 s/min) food are significantly different, highlighted in red (P < 1.64 × 10⁻⁹). c Lunge frequencies before (pre) and during (stim) MBON11 photoactivation in chambers with males under fed or hungry states. d, e Lunge frequencies in nutrient-specific food value assay (Fig. 1e), of males during MBON11 photoactivation by CsChrimson, or of males with MBON11 silenced by Kir2.1. Also see Supplementary Fig. 2n. f, g Lunge frequencies in protein-hunger assay (Fig. 1f), of males during MBON11 photoactivation by CsChrimson, or of males with MBON11 silenced by Kir2.1. Also see Supplementary Fig. 2o. h Schematic for two-photon calcium imaging of MBON11 in head and proboscis fixed males with free legs to better simulate fly-food interaction, using motorized solid food delivery (see Methods). Insets show dorsal view MBON11 skeleton (green) in the mushroom body (solid black line) from EM connectome, and dorsal view jGCaMP7f fluorescence expression in MBON11, highlighting γ1 and peduncular (pedc) dendritic compartments (white dashes). A, P, M, L denote anterior, posterior, medial, and lateral, respectively. i MBON11 calcium response dynamics (middle) and magnitudes (area under the curve, AUC, bottom) to solid apple juice food (red) or agarose food (gray). Fly rearing and experimental conditions and food-fly distances are shown at the top. Curves and shadows are the means and SEM, respectively, same as follows. j MBON11 calcium response dynamics (middle) and magnitudes (AUC, bottom) to apple juice food in hungry (red) or fed (gray) conditions. k MBON11 calcium response dynamics (middle) and magnitudes (AUC, bottom) to sucrose-yeast food in males reared with sucrose or sucrose with an amino acid mixture.

As a comparison to MBON11, we also examined fruitless-expressing Tachykinin neurons (Tk^FruM) and P1^a neurons, involved in aggression and courtship, respectively^{14,18,42–44}. As reported, activation or inhibition of Tk^FruM neurons enhances or suppresses lunge in food chambers (Supplementary Fig. 2i-j). However, in contrast to MBON11, Tk^FruM neuron activation enhances lunge even in foodless chambers (Supplementary Fig. 2k), indicating that MBON11 but not Tk^FruM specifically enhances food-induced aggression. Activation of P1^a neurons in foodless chambers with a tethered female enhances male-female courtship but does not affect female-induced male-male lunge (Supplementary Fig. 2l-m). Together, mushroom body output neuron MBON11 influences aggression, specifically for food-induced inter-male lunge and tussle.

MBON11 mediates hunger-dependent aggression

Given that hunger modulates food-induced aggression (Fig. 1d), we explored MBON11’s role in this modulation. In satiated males, MBON11 activation slightly but reliably increased lunge in the absence of food (Fig. 2b). Strikingly, this effect was dramatically enhanced by the presence of food (red bracket, P < 1.64 × 10⁻⁹). In contrast, in the presence of food, MBON11 activation increases lunge to similar frequencies in both starved and satiated males in chambers with food (red bracket, Fig. 2c). These results suggest that MBON11 activation creates a hunger-like state even in fed flies, which promotes inter-male aggression in a food-dependent manner.

Hunger modulates aggression in a nutrient-specific way (Fig. 1e). We asked whether MBON11 also mediates the nutrient-specific hunger effect on aggression. Using CsChrimson photoactivation and Kir2.1 inhibition of MBON11, we found that lunge frequency differences due to the nutrient-specific hunger effects were diminished in both cases (Fig. 2d, e, Supplementary Fig. 2n). This suggests that MBON11 neuron activity mediates nutrient-specific hunger modulation of aggression. Protein hunger modulates aggression (Fig. 1f). We reasoned that if MBON11 mediates protein hunger modulated aggression, activating or inhibiting MBON11 should override such an effect. Indeed, photoactivating MBON11 with CsChrimson or inhibiting it with Kir2.1 abolishes lunge frequency differences due to the protein hunger effect (Fig. 2f, g, Supplementary Fig. 2o). These results indicate that MBON11 activation overrides nutrient-specific and hunger-dependent aggression.

MBON11 exhibits distinct functions in aggression modulation by food and hunger (Fig. 2b, c), we explored how MBON11 neural activity responds to food and hunger. Previous studies have shown MBON11 responses to food odors^2,45,46, but flies interact with solid food in natural conditions, which involves a multisensory process. It remains unclear how MBON11 responds in this case. To simulate fly-food interaction, we delivered solid food with a motorized setup (Fig. 2h, see Methods). We conducted two-photon calcium imaging of MBON11 dendrites in the γ1-pedc compartments in starved males accessible to food. The same apple juice food as used in ethology experiments induces strong MBON11 responses, while agarose induces no response (Fig. 2i), suggesting that MBON11 responds to solid food. To examine MBON11 hunger responses, we imaged MBON11 dendritic calcium in starved and satiated males accessible to apple juice food. MBON11 responded to food in both conditions, but responses were stronger in starved males, indicating sensitivity to hunger (Fig. 2j). We then tested MBON11 responses to nutrient-specific hunger by imaging MBON11 dendritic calcium in males with and without protein deprivation accessible to protein-containing food. MBON11 responded to food in both conditions, but responses were stronger in protein-deprived males, indicating sensitivity to nutrient-specific hunger (Fig. 2k). Interestingly, we observed gradients of increasing lunge frequencies and MBON11 responses corresponding to increasing nutrient needs, ranging from sucrose and amino acid feeding, to sucrose only feeding, and finally to starvation (Supplementary Fig. 2p). These similar-shaped gradients suggest a tight coupling between nutrient-specific hunger and aggression in MBON11.

Together, MBON11 signals hunger states to mediate hunger-dependent food-induced aggression. MBON11 responds to food presence, with its intensity amplified by hunger. These behavioral and physiological results reveal distinct MBON11 functions in the effects of food and hunger on aggression.

Dopaminergic PPL101 signals hunger and modulates aggression

Where does MBON11 acquire hunger state information? Given that monoaminergic modulatory neurons, including dopaminergic, serotonergic, and octopaminergic neurons, heavily innervate the mushroom body and have been implicated in hunger signaling, value processing, and aggression modulation^{12,13,15,17,21,47,48}, we investigated whether MBON11 hunger-dependent aggression modulation is derived from one of these modulators.

We first examined serotonin and octopamine. As reported¹⁷, photoactivating pan-serotonin Tryptophan hydroxylase (Trh) enhanced lunge in food chambers (Supplementary Fig. 3d), an effect similar to MBON11 but tenfold weaker (Supplementary Fig. 3d vs Fig. 2b). To test if serotonin modulates MBON11, we photoactivated Trh neurons specifically within the MB horizontal lobes using digital holography while imaging MBON11 calcium (Supplementary Fig. 3e, see Methods). MBON11 responses are tuned to activation intensities, supporting serotonin as an upstream. Photoactivating pan-octopamine Tyrosine decarboxylase 2 (Tdc2) decreases lunge frequency (Supplementary Fig. 3h), indicating that octopamine modulates aggression^12,15. To test whether this modulation acts via MBON11, we photoactivated Tdc2 while imaging MBON11 calcium (Supplementary Fig. 3i). Activating Tdc2 inhibits MBON11, supporting octopamine as an upstream. To further identify modulatory inputs with cell-type specificity, we profiled the MBON11 circuit using the synaptic-resolution connectome^49–51. MBON11 receives diverse modulatory neuron inputs (Fig. 3a, b, Supplementary Fig. 3a, b), including serotonergic DPM⁵², octopaminergic OA-VPM3/4, and dopaminergic PPL101. However, DPM neither modulates lunge nor signals nutrient-specific hunger (Supplementary Fig. 3f, g). Photoactivating OA-VPM3/4 decreases lunge (Supplementary Fig. 3j), but Tdc2 shows no clear nutrient-specific hunger signal (Supplementary Fig. 3k). Thus, MBON11 upstream serotonin and octopamine modulate aggression, though their functions require future study.

Fig. 3. PPL101 modulates hunger-dependent aggression.

a Volumetric reconstruction of an MBON11 neuron from EM connectome, with major dopaminergic (red), serotonergic (purple), and octopaminergic (green) input synapses labeled (colored dots). See also Supplementary Fig. 3a. b Major upstream and downstream neurons of MBON11 from EM connectome, excluding kenyon cells. Arrows indicate a connectivity relation. Line widths are related to the logarithm of the synapse counts (see Methods), which are labeled next to each arrow. Also see Supplementary Fig. 3b. c Volumetric reconstruction of a PPL101 neuron (blue) and its ipsilateral downstream MBON11 (orange). The insets below show a magnified view of the γ1 and pedc compartments densely innervated by PPL101 axons and MBON11 dendrites, where PPL101-MBON11 synapses (red dots) are located. Also see Supplementary Fig. 3c. d Lunge frequency ratios (see Methods) upon PPL101 GtACR1 photostimulation relative to pre-stimulation lunge frequencies between interacting transgenic males in food chambers, repeated using three different drivers, alongside controls. e Lunge frequencies before (pre) and during (stim) PPL101 photoactivation with CsChrimson, alongside controls. f Lunge frequencies between interacting transgenic males with PPL101 ablated, using Caspase-3, or using both Reaper and Grim, alongside controls. g Schematic for two-photon calcium imaging of PPL101 in head and proboscis fixed males with solid apple juice food stimulation. Insets show dorsal view PPL101 skeleton (green) in the mushroom body (solid black line) from EM connectome, and dorsal view jGCaMP7f fluorescence expression in PPL101, highlighting γ1 and pedc compartments (white dashes). h PPL101 calcium response dynamics (left) and magnitudes (AUC, right) to solid apple juice food stimulation in hungry or fed males. i PPL101 calcium response dynamics (left) and magnitudes (AUC, right) to solid sucrose-yeast food stimulation in males reared on sucrose or sucrose with an amino acid mixture.

Dopaminergic neurons are widely considered the neural basis of value^20,21,47. A single pair of dopaminergic PPL101 neurons (also known as PPL1-γ1pedc or MB-MP1, Fig. 3c) synapses with MBON11 heavily and homogeneously in the γ1-pedc compartments (Fig. 3a-c, Supplementary Fig. 3a-c), a pattern maintained in dopaminergic synapses based on neurotransmitter prediction (Supplementary Fig. 3c, see Methods), forming a functional compartment for transmission^22,32,53,54. PPL101 has been reported to respond to food and hunger³⁷. Using three different split-GAL4 drivers to target PPL101 for photoinhibition with GtACR1³², PPL101 inhibition consistently enhances lunge (Fig. 3d). On the contrary, PPL101 photoactivation with CsChrimson decreases lunge (Fig. 3e), indicating that PPL101 bidirectionally modulates aggression. To examine whether PPL101 is necessary for aggression modulation, we ablated PPL101 using apoptosis effectors Caspase-3, or a combination of Reaper and Grim (Fig. 3f). Both approaches led to a marked decrease in lunge frequency, indicating that PPL101 is required for aggression modulation. If PPL101 specifically modulates hunger-dependent aggression, it should exhibit hunger-related signals. To test this, we imaged PPL101 axonal calcium activity in the γ1-pedc compartments in males to motorized food stimulation (Fig. 3g). Indeed, PPL101 calcium responses were stronger in fed than hungry males (Fig. 3h), and stronger in males fed with amino acid supplements than in protein-hungry males (Fig. 3i).

Together, MBON11 receives broad modulation of dopamine, serotonin, and octopamine for aggression. However, dopaminergic PPL101 selectively signals hunger states to modulate aggression, suggesting that the PPL101-MBON11 circuit is important for coupling hunger states with aggression modulation. Next, we focused on how the PPL101-MBON11 circuit modulates hunger-dependent aggression.

A multilayer circuit for hunger-dependent aggression

PPL101 and MBON11 both signal hunger states and influence hunger-dependent aggression. The connectome reveals reciprocal connectivity between them (Fig. 3b, Supplementary Fig. 3b), prompting the question of which acts as the output to modulate aggression. Since PPL101 inhibition enhances lunge (Fig. 3d), we inhibited PPL101 while blocking MBON11 simultaneously (Fig. 4a). MBON11 photoinhibition eliminates lunge, overriding the PPL101 effect (Fig. 4b, green bars), suggesting MBON11 as downstream of PPL101 in aggression modulation. To validate this, we photoactivated MBON11 using CsChrimson while silencing PPL101 with Kir2.1. Indeed, MBON11 activation enhances lunge frequency despite PPL101 silencing, supporting its sufficiency for aggression modulation (Fig. 4d, green bars). PPL101 silencing with Kir2.1 yields similar effect as apoptosis (Fig. 3f), thus though PPL101 dynamics modulate aggression, basal PPL101 activity is required for aggression. MBON11 activation when silencing PPL101 leads to inconsistent lunge frequency changes, including both decreases and large increases (Fig. 4e, blue and red dots), compared to consistent increases when activating MBON11 alone (Fig. 4e, orange bars), underscoring the crucial role of PPL101 in MBON11 modulation of aggression.

Fig. 4. A multilayer circuit for hunger-dependent aggression.

a Photoinhibiting PPL101 using GtACR1 while photoinhibiting MBON11 using GtACR1 simultaneously, as used in (b). b Lunge frequencies between interacting transgenic males before (pre) and during (stim) simultaneous PPL101 (labeled by split-GAL4 driver MB502B) and MBON11 (labeled by split-GAL4 driver MB085C) photoinhibition using GtACR1 (green), compared to MBON11 (orange) or PPL101 (blue) photoinhibition alone and 3 controls (gray). c Photoactivating MBON11 with CsChrimson while silencing PPL101 using Kir2.1 simultaneously, as used in (d, e). d Lunge frequencies between interacting transgenic males before (pre) and during (stim) MBON11 photoactivation by CsChrimson when silencing PPL101 by Kir2.1 (green), compared to MBON11 photoactivation (orange) or PPL101 silencing (blue) alone and 4 controls (gray). e Lunge frequency differences upon MBON11 photoactivation relative to pre-stimulation lunge frequencies, with (green bars) or without (orange bars) PPL101 silencing. Decreases or large increases are labeled in blue and red, respectively. f Dual-channel photostimulation protocol, for photoactivating MBON11 using CsChrimson alone first and then adding Tk^FruM photoinhibition using GtACR1 after a period of time, as used in (g). The pink shadows denote 680 nm LED stimulation of CsChrimson, and the purple shadows denote stimulation of both CsChrimson and GtACR1 with 680 nm and 520 nm LEDs, respectively. g Lunge frequency ratios relative to pre-stimulation lunge frequencies between interacting transgenic males with MBON11 photoactivation using CsChrimson, followed by additional Tk^FruM photoinhibition using GtACR1 (green), compared to MBON11 photoactivation (orange) or Tk^FruM photoinhibition (pink) alone and four controls (gray). Also see Supplementary Fig. 4b. h Schematic of the three-layer circuit model for hunger-dependent aggression, from dopaminergic PPL101 neuron signaling hunger states, to MBON11 neuron signaling hunger and modulating aggression, and finally to Tk^FruM neurons driving aggression. Also see Supplementary Fig. 4f.

To elucidate how MBON11 modulates aggression, we explored its functional relationship with aggression-driving Tk^FruM neurons (Supplementary Fig. 2i-k)¹⁴. Comparing lunge patterns elicited by MBON11 and Tk^FruM activation reveals that Tk^FruM drives lunge actions more sharply synchronized with photoactivation onset than MBON11 (Supplementary Fig. 4a), suggesting proximity of Tk^FruM to action execution, consistent with Tk^FruM function in modulating aggression independent of food (Supplementary Fig. 2k). Across various connectome datasets, MBON11 prominently innervates the superior protocerebrum⁵⁵, where Tk^FruM neurons reside (Fig. 3b, Supplementary Fig. 3b), suggesting that Tk^FruM is downstream of MBON11. Specifically in the male connectome, the GABAergic MBON11 is connected to Tk^FruM via a GABAergic interneuron, indicating that MBON11 disinhibits Tk^FruM. To test this, we employed dual-channel photostimulation to photoactivate MBON11 first, then added Tk^FruM photoinhibition after a delay (Fig. 4f, g, Supplementary Fig. 4b, see Methods for details). MBON11-activated lunges are sharply and almost wholly suppressed upon Tk^FruM inhibition. Similarly, photoactivating MBON11 while silencing Tk^FruM neurons using Kir2.1 yielded comparable results (Supplementary Fig. 4c), suggesting that MBON11 modulates aggression via Tk^FruM. Further, we photoactivated MBON11 while two-photon imaging Tk^FruM calcium, and detected a significant increase in Tk^FruM after MBON11 photoactivation (Supplementary Fig. 4d-e). These results support that MBON11 modulates aggression via Tk^FruM.

We identified a multilayer circuit from aminergic neurons to MBON11 and eventually to Tk^FruM neurons for aggression modulation (Supplementary Fig. 4f), where the PPL101-MBON11 circuit specifically mediates hunger-dependent aggression (Fig. 4h). Recent studies identified PPL101 upstream peptidergic neurons, including orexigenic dNPF and anorexigenic AKH that inhibits and excites the anorexic PPL101, respectively^1,29, making PPL101 a hub for integrating metabolic inputs into nutrient-specific hunger signal (Supplementary Fig. 4f). Relatedly, dNPF is also implicated in aggression modulation^14,48. MBON11 upstream serotonin and octopamine modulate aggression, though their specific functions are unclear, suggesting a broader role of MBON11 in aggression modulation beyond hunger. MBON11 acts as the output of the circuit and interacts with Tk^FruM for aggression regulation. Together, this circuit mediates modulation of aggression by hunger (Fig. 4h, Supplementary Fig. 4f).

PPL101 and MBON11 oppositely signal hunger

Given the central role of the PPL101-MBON11 circuit in linking hunger states to aggression modulation, we further examined hunger processing on this circuit. Consistent with our results, both PPL101 and MBON11 have been implicated in signaling hunger^2,37,38. However, these neurons were recorded separately in different animals, both in previous studies and our earlier experiments, prompting us to investigate how this transformation occurs in the same mushroom body compartment of the same animal. To this end, we performed two-channel two-photon calcium imaging on the PPL101-MBON11 circuit⁵⁶. To account for differences in calcium sensor properties, we paired green and red sensors with PPL101 and MBON11 in two combinations (Fig. 5a-c, Supplementary Fig. 5b, e). For better stimulus control, we simulated food with the food odor apple cider vinegar (ACV) that elicits faster responses (Supplementary Fig. 5a). In hungry males, PPL101 response is suppressed while MBON11 response is enhanced, and vice versa in fed males, enlarging MBON11-PPL101 activity difference in hungry males (Fig. 5b, Supplementary Fig. 5b-d, arrows). Across different imaging conditions, PPL101 exhibits stronger responses under satiety than hunger, while MBON11 shows the opposite pattern, and likewise for nutrient-specific hunger states (Fig. 5c-g, Supplementary Fig. 5e). Thus, PPL101 and MBON11 oppositely signal hunger, forming the two ends of the hunger states spectrum.

Fig. 5. PPL101 and MBON11 oppositely signal hunger.

a Schematic for dual-color two-photon calcium imaging in head and leg fixed males to ACV odor stimulation. Insets show dorsal view PPL101 (green) and MBON11 (brown) skeletons in the mushroom body (solid black line) from EM connectome and fluorescence expression of jGCaMP7b and jRGECO1a calcium indicators in PPL101 and MBON11, respectively, highlighting γ1 and pedc compartments (white dashes). b Example PPL101 and MBON11 calcium response dynamics in the same compartment using dual-color imaging in fed (upper) and hungry (lower) males. Black arrows denote MBON11-PPL101 activity difference, orange shadows denote odor stimulation period, same as follows. Also see Supplementary Fig. 5b-d. c Mean PPL101 (upper) and MBON11 (lower) calcium responses to ACV odor stimulation in fed and hungry males using dual-color imaging. Also see Supplementary Fig. 5e. d, e PPL101 (blue) and MBON11 (orange) mean responses (AUC) to ACV stimulation in hungry and fed states using dual-color imaging with different indicator combinations, d and e same data as c and Supplementary Fig. 5e, respectively. f PPL101 (blue) and MBON11 (orange) mean responses (AUC) to solid food stimulation in hungry and fed males using single-color imaging. Same data as Figs. 2j and 3h. g PPL101 (blue) and MBON11 (orange) mean responses (AUC) to solid food stimulation in males fed with sucrose or sucrose amino acid mixture using single-color imaging. Same data as Figs. 2k and 3i. h Schematic for two-photon calcium imaging of MBON11 to ACV odor during simultaneous PPL101 two-photon photoactivation. Insets show PPL101 (red) and MBON11 (green) skeletons in the mushroom body. i MBON11 calcium response dynamics to ACV stimulation during simultaneous PPL101 two-photon photoactivation, including no PPL101 activation control at low imaging power (black line), no PPL101 activation control at high imaging power (blue line), low power PPL101 activation (orange line), and high power PPL101 activation (red line). See Supplementary Fig. 5g for statistical results, and Supplementary Fig. 5h for the first-trial responses. j A working model of opposite responses to satiation and starvation in PPL101 and MBON11 that signal satiety and hunger, respectively.

The transformation inverts the relative responsiveness of PPL101 and MBON11 according to hunger states, suggesting that PPL101 activity may functionally gate MBON11 response. To test this, we expressed CsChrimson in PPL101 and specifically stimulated the γ1-pedc compartments using one-photon or two-photon photostimulation while imaging MBON11, but detected minimal effects (Supplementary Fig. 5f). However, when photoactivating PPL101 while simultaneously delivering food odor, we observed apparent attenuation in MBON11 responses, with higher laser intensity leading to stronger effect (Fig. 5h-i, Supplementary Fig. 5g). This attenuation per se is not due to synaptic depression⁵³, as it is evident from the first trial (Supplementary Fig. 5h). External food cues and internal hunger states jointly shape the perceived value of food. Satiety negatively correlates with perceived value, reflecting metabolic states^1,29; hunger positively correlates with it, reflecting internal needs⁴. Through the PPL101-MBON11 transformation, this circuit integrates internal needs and external food cues into a unified perceived value of food.

So far, we have revealed the behavioral phenotype of hunger-dependent aggression and identified the underlying PPL101-MBON11 circuit and neural activities (Figs. 4h, 5j). MBON11 acts as a bridge to link hunger states from PPL101 to aggressive actions driven by Tk^FruM, forming a “hangry” state. However, the arithmetic, biophysical, and molecular mechanisms underpinning the hangry state remain unclear.

Hunger modulates aggression via MBON11 gain control

To delineate the quantitative relationship between food or hunger and aggression in wild-type males, we first tested whether food or hunger influences aggression in a binary or graded manner. Using agarose food with varying sucrose concentrations, we found that lunge frequency increases with sucrose concentration (Fig. 6a). Similarly, lunge frequency increases with starvation duration in males given a fixed food amount (Fig. 6c). Thus, both food and hunger modulate aggression in a graded way, as shown by the food- and hunger-tuned lunge frequency curves. Next, we distinguished MBON11 roles in food- and hunger-related aggression modulation. MBON11 activation differentially shifts the lunge frequency curves: it amplifies food tuning by steepening the curve (Fig. 6a, b), and overrides hunger tuning by flattening the curve (Fig. 6c, d). Silencing MBON11 with Kir2.1 abolishes hunger-dependent aggression. These findings suggest that MBON11 activity is directly related to hunger, which amplifies the perceived value of food to modulate aggression.

Fig. 6. Hunger modulates MBON11 excitability and amplifies food value for aggression modulation.

a Lunge frequencies between males in chambers with food of increasing sucrose concentrations during MBON11 photoactivation, compared with wild-type. Also see Supplementary Fig. 6j. b Schematic showing MBON11 activation amplifies food-dependent aggression, by steepening the food-tuned lunge frequency curve. c Lunge frequencies between males under increasing starvation durations with MBON11 photoactivation or silencing, compared with wild-type. MBON11>CsChr data same as Fig. 2c. d Schematic showing MBON11 manipulation overrides hunger modulation of aggression, by flattening the hunger-tuned lunge frequency curve. e Schematic for two-photon calcium imaging of MBON11 in head-fixed males with apple juice food stimulation. MBON11 calcium response dynamics (f) and magnitudes (AUC, g) to apple juice food in males with increasing starvation durations, with different binning. Containing data in Fig. 2j. h Lunge probability dynamics during MBON11 photoactivation at different intensities. Also see Supplementary Fig. 6g. i Relationship between lunge frequencies before (pre) and during (stim) MBON11 photoactivation in chambers with sucrose food of different concentrations, or apple juice food. Dots denote individual males, color-coded by food types. j Relationship between lunge frequencies before (pre) and during (stim) MBON11 photoactivation at different intensities. Dots denote individual males, color-coded by activation intensities. Colored dashes denote linear fittings for different activation intensities. Same data as (h). k Relationship between lunge frequencies with different starvation durations and after refeeding in wild-type males. Dots denote individual males, color-coded by food starvation durations. Colored dashes denote linear fittings for different starvation durations. l Working model of MBON11 gain control. Hunger (MBON11 excitability) is related to the slopes (multiplicative gain), such that hunger scales aggression. m Schematic for current injection and in vivo whole-cell patch-clamp recording of MBON11 in hungry or fed males. n Example MBON11 membrane potential traces to 40 pA current injection in fed or hungry males. Zoomed-in view (right) highlights increased first spike latency and afterhyperpolarization (AHP) in fed versus hungry males. Also see Supplementary Fig. 7e-i, j-k. o Evoked firing rates of MBON11 in response to depolarizing current steps in fed or hungry males.

To directly quantify MBON11 activity in linking food value amplification to aggression modulation, we tested whether MBON11 responds to perceived value of food in a hunger-dependent, graded way, and modulates aggression accordingly. Previous studies showed MBON11 responses to hunger^2,37, but it remains unclear whether MBON11 signals hunger states in a binary or graded way. To test this, we starved males for different durations to image MBON11 food responses (Fig. 6e-g, Supplementary Fig. 6a-c), and observed hunger-dependent graded responses to the same food. In assessing whether starvation durations can be decoded from MBON11 activity, we trained a support vector machine decoder and found that the decoding accuracy was significantly higher than that of shuffled data (Supplementary Fig. 6d). To test graded aggression modulation, we photoactivated MBON11 at increasing intensities while measuring lunge (Fig. 6h, Supplementary Fig. 6e-g). Increasing MBON11 activation intensity leads to an increase in lunge. Thus, hunger levels correlate with MBON11 neural activity in response to food, which in turn promotes aggression, with higher activity levels driving more intense aggression, effectively linking “hangry” states to MBON11 activity amplification.

To quantitatively pinpoint how hunger level amplifies MBON11 food responses to scale aggression intensities, we hypothesized that the amplification process could be either additive or multiplicative^57,58 (Supplementary Fig. 6h-i). To discern these, we compared lunge frequency differences and ratios before and during MBON11 activation at the same intensity, across different food types (Supplementary Fig. 6j). Under the same activation intensity, lunge frequency differences vary by food type and composition, while ratios remain constant (Supplementary Fig. 6k-l), supporting the multiplicative over additive hypothesis. Further, activating MBON11 without food only weakly enhances lunge, even with strong activation (Fig. 2b, Supplementary Fig. 6m-n), and is even less effective than food alone without MBON11 activation (Fig. 1c). In contrast, activating MBON11 with food strongly enhances lunge frequency (Fig. 2b), again supporting multiplication. Physiologically, activating PPL101 changes MBON11 activity only when paired with food odor (Fig. 5i). Multiplicative amplification is a gain control mechanism^57,58. To pinpoint the multiplication gain, we hypothesized that it correlates with MBON11 activation intensity, which reflects its neural activity thus hunger level. Indeed, the gain remains constant at the same intensity, regardless of the food type, and increases with increasing activation intensities, consistently exhibiting a multiplicative characteristic (Fig. 6i, j). To test gain control under natural conditions, we hypothesized that prolonged starvation would mimic the effects of increased activation intensity. Indeed, we observed that gain control scaled with starvation duration (Fig. 6k).

Together, these results support that MBON11 multiplicative gain control underlies the “hangry” state (Fig. 6l). Unlike additive amplification, which involves a fixed increase regardless of the basal level, multiplicative gain control scales with the basal aggression level, allowing for more effective responses to wide-ranging basal aggression levels, thus enhancing adaptability and sensitivity^57,58. However, the biophysical basis of this gain control remains unknown.

Hunger modulates MBON11 excitability

To investigate the biophysical mechanisms underlying MBON11 gain control, we probed MBON11 electrical activity in signaling hunger-dependent value of food. Since this value signal was identified using MBON11 dendritic calcium activity, a blend of synaptic inputs and back-propagating action potentials, we questioned whether this signal is directly represented by MBON11 action potentials. To test this, we performed in vivo whole-cell recordings on MBON11 while delivering food odors in fed or hungry males (Supplementary Fig. 7a-d). Consistent with calcium imaging, MBON11 fires more action potentials in response to the same food odor under hunger conditions compared to satiety, supporting that MBON11 spiking signals the hunger-dependent value of food. Given identical food odor inputs in both calcium imaging and electrophysiology, we hypothesized that cell-intrinsic properties, such as neuronal excitability, signal hunger states in MBON11 spiking and calcium activity.

To more precisely probe MBON11 intrinsic electrical properties, we then injected current steps into MBON11 (Fig. 6m, Supplementary Fig. 7e, f). We observed more evoked spikes in hungry males compared to fed males under the same current injection conditions (Fig. 6n, Supplementary Fig. 7h-i). This hunger-dependent difference in evoked firing rate was consistent across different current steps (Fig. 6o), indicating that hunger states alter MBON11 membrane excitability. Changes in excitability are caused by changes in other physiological properties that affect action potential generation, such as resting membrane potential, action potential latency, and action potential afterhyperpolarization (AHP)^8,59. Indeed, the latencies to the first evoked spikes were significantly shorter, and action potential afterhyperpolarizations were significantly reduced in hungry males (Supplementary Fig. 7g–k), both contributing to higher excitability. Interestingly, the slope of the curve was also higher in hungry compared to fed states (Fig. 6o), supporting multiplicative gain control.

These results pinpoint hunger states to MBON11 membrane excitability for its gain control (Supplementary Fig. 7l). This is consistent with optogenetics results indicating that MBON11 signals hunger, given that microbial opsins act via membrane excitability regulation⁶⁰. Similar findings linking internal states to membrane excitability have been reported^5,8, suggesting its generality.

Dopamine signals hunger to modulate MBON11 excitability and aggression

What are the molecular bases of MBON11 excitability for “hangry” states? As PPL101 and MBON11 are heavily connected within the γ1-pedc compartments (Fig. 3c, Supplementary Fig. 3c), we focused on neural transmission by studying PPL101 release and its effect on MBON11 excitability in relation to hunger and aggression.

PPL101 is known to co-release dopamine and other transmitters^61,62 (Supplementary Fig. 3c). To determine whether dopamine directly signals hunger, we compared PPL101 dopamine release in starved and satiated states, in both evoked and spontaneous release modes. We first examined evoked PPL101 secretion using two-photon dopamine imaging of PPL101 axonal arbors in γ1-pedc compartments with the gDA3m sensor to solid-food stimulation in starved and satiated males (Fig. 7a). Consistent with calcium, we found larger PPL101 dopamine responses in satiated than starved males (Fig. 7b, c). Similar results were observed with food odor stimulation (Supplementary Fig. 8a, b). We next investigated whether spontaneous dopamine release also signals hunger states. First, we tested whether PPL101 releases dopamine spontaneously using the high-sensitivity gDA3h sensor. We imaged dopamine release on PPL101 axons without stimulation and indeed detected persistent spontaneous dopamine activity (Fig. 7d, e). To confirm that these extracellular dopamine signals originate from PPL101 rather than other neurons innervating the γ1-pedc compartment, we specifically blocked dopamine synthesis in PPL101 using RNA interference to silence PPL101 Tyrosine hydroxylase (TH), which is required for dopamine precursor synthesis. This results in a significant reduction in spontaneous dopamine activity, confirming our hypothesis (Fig. 7d, e). Previous studies have shown hunger-dependent spontaneous calcium activities in PPL101³⁸. We tested whether spontaneous PPL101 dopamine release, like evoked release, signals hunger states, and found that it does (Fig. 7f, g). These results indicate that hunger states are expressed in PPL101 not only as calcium activity but also as dopamine release, both evoked and spontaneous.

Fig. 7. PPL101 dopamine signals hunger states and modulates MBON11 excitability and aggression.

a Schematic for two-photon dopamine imaging of PPL101 in head and proboscis fixed males to solid apple juice food stimulation. Insets show PPL101 (green) skeleton in the mushroom body (solid black line) and fluorescence expression of gDA3m sensor in PPL101, highlighting γ1 and pedc compartments (white dashes). PPL101 dopamine response dynamics (b) and magnitudes (AUC, c) to solid apple juice food in hungry or fed states. d Example PPL101 spontaneous dopamine release dynamics using gDA3h sensor without (upper) or with (lower) RNA interference of TH. e Spontaneous PPL101 dopamine release amplitudes in control and TH RNA interference conditions. f Example PPL101 spontaneous dopamine release dynamics in fed (upper) and hungry (lower) males. g Spontaneous PPL101 dopamine release amplitudes in fed and hungry males. h Schematic of MBON11 whole-cell recording on the cell body during dopamine puffing into dendritic compartment. i Mean membrane potential responses of MBON11 upon puffing 500 ms (gray shadow) of 10 mM dopamine (blue) or saline (black). Also see Supplementary Fig. 8c. j Schematic of current injection and whole-cell patch-clamp recording of MBON11 before and after dopamine puffing. k Examples (left) and zoomed-in view (right) of voltage responses to a 40 pA current injection before (black) and after (blue) dopamine puffing in the same cell. In the zoomed-in view, top and bottom red arrows denote increased first spike latency and afterhyperpolarization after dopamine puffing, respectively. Post-puff current injection begins 83 s after puffing ends. Also see Supplementary Fig. 8e-h. l Firing rates of MBON11 in response to current injections before (black) and after (blue) DA puffing. Dopamine influences firing across all currents injected over the 100-s period. Also see Supplementary Fig. 8i-j. m Lunge frequencies between transgenic males with TH knockdown in PPL101 via RNA interference (RNAi), alongside controls. n PPL101 dopamine responses upon two-photon photoactivation or photoinhibition with CsChrimson or GtACR1, respectively. See Supplementary Fig. 9a for statistical results. o Lunge frequencies between transgenic males with Dop1R1 or Dop1R2 knockdown in MBON11 via RNAi, alongside controls. Also see Supplementary Fig. 9b, c.

Does PPL101 dopamine underlie MBON11 membrane excitability? We hypothesized that introducing dopamine into the dendritic compartment would decrease MBON11 excitability. We locally puffed dopamine into the MBON11 γ1 compartment using a pipette electrode guided by fluorescence while performing whole-cell recordings (Fig. 7h, also see Methods). Unlike the saline control, dopamine induces acute membrane hyperpolarization (Fig. 7i, Supplementary Fig. 8c-d), suggesting changes in excitability. To directly test dopamine effect on excitability, we performed current injection while recording before and after dopamine application (Fig. 7j, Supplementary Fig. 8e-g), and confirmed the effectiveness of local dopamine application based on acute hyperpolarization (Supplementary Fig. 8f). Dopamine reduces evoked action potentials across all currents, confirming its effect on excitability (Fig. 7l). Furthermore, dopamine significantly increased action potential afterhyperpolarization and the latency to the first spike (Fig. 7k red arrows, Supplementary Fig. 8h-j). Dopamine effect is particularly pronounced on afterhyperpolarization (Supplementary Fig. 8j). This is consistent with that in hunger modulation (Supplementary Fig. 7k), suggesting a similar mechanism for dopamine modulation of excitability via afterhyperpolarization, also reported in mammals⁵⁹. Dopamine modulation of excitability is persistent, lasting from a few to over a hundred seconds, and affects both spontaneous and evoked spikes (Fig. 7k-l, Supplementary Fig. 8d-j). Since PPL101 spontaneous dopamine release signals hunger states persistently, combining PPL101 spontaneous release with persistent modulation of MBON11 excitability provides a way for persistently maintaining “hangry” state. Thus, dopamine modulates the excitability of MBON11, a mechanism regulating hunger modulation of MBON11 activity.

How is dopamine release related to aggressive behavior? Given that PPL101 spontaneously releases dopamine to persistently signal hunger states, we wondered whether dopamine is essential for aggression. Blocking dopamine synthesis in PPL101 by silencing TH significantly decreased lunge frequency (Fig. 7m), indicating that PPL101 spontaneous dopamine release, which provides a sustained dopamine supply, is crucial for aggression. Essentially, although dopamine dynamics modulate aggression, a basal level of dopamine is required for aggression. How is dopamine release related to hunger-dependent aggression? Under manipulation, PPL101 inhibition or excitation increases or decreases aggression, respectively (Fig. 3d, e). Under natural conditions, PPL101 dopamine decrease is correlated with an increase in aggression when hungry, while PPL101 dopamine increase is correlated with a decrease in aggression when satiated. Together, PPL101 inhibition or excitation leads to a decrease or increase in dopamine, respectively, relating dopamine to hunger-dependent aggression. Indeed, optogenetic manipulation of PPL101 revealed that photoactivation with CsChrimson increased dopamine, while photoinhibition with GtACR1 decreased dopamine (Fig. 7n, Supplementary Fig. 9a). This suggests that PPL101 activation increases dopamine and decreases MBON11 activity, mimicking satiety-induced aggression decrease, while PPL101 inhibition decreases dopamine and increases MBON11 activity, mimicking hunger-induced aggression increase.

Aggression modulation by dopamine likely involves specific receptors, and the blockade of these receptors could affect dopamine modulation of aggression. In Drosophila, there are four dopamine receptor types: Dop1R1 (also as DopR1 or DUMB), Dop1R2 (also DopR2 or DAMB), Dop2R, and DopEcR^63,64. We knocked down these receptors in MBON11 using RNA interference and measured lunge frequency. Knocking down Dop1R1 or Dop1R2 strongly enhances or suppresses lunge frequency, respectively, while effects for Dop2R and DopEcR are not significant or weaker, across different MBON11 drivers and RNAi effectors (Fig. 7o, Supplementary Fig. 9b, c). Thus, manipulating two types of dopamine receptors in a single neuron oppositely mediate strong dopamine modulation of aggression.

Together, these results demonstrate that PPL101 dopamine release signals hunger states and modulates MBON11 excitability to determine its activity and aggression, thereby mapping the hunger-aggression relation onto the PPL101-dopamine-MBON11 excitability axis. Besides, Dop1R1 and Dop1R2 receptors oppositely modulate aggression. We next investigated whether these dopamine receptors underlie dopamine modulation of MBON11 excitability and sought to elucidate the underlying mechanisms.

Opposing pathways mediate dopamine modulation of MBON11 excitability

PPL101 dopamine signals hunger states to modulate MBON11 excitability and aggression (Fig. 7). Since we have linked hunger-dependent calcium and electrical activity to membrane excitability in MBON11 (Fig. 6f, o, Supplementary Fig. 7b, c), different calcium responses to the same odor can infer membrane excitability in MBON11. In line with this, PPL101 activation increases dopamine and reduces MBON11 excitability (Fig. 5i, Fig. 7n), underscoring the decrease in aggression during satiation. But the mechanisms underlying aggression increase during starvation remain unclear. During satiation, dopamine levels rise and MBON11 excitability falls; during starvation, dopamine levels fall and MBON11 excitability rises. Thus, changes in dopamine levels and MBON11 excitability are inversely related, suggesting that opposing dopamine dynamics oppositely modulate MBON11 excitability and activity. To test this, we simultaneously perfused dopamine and delivered food odors while imaging MBON11 activities (Fig. 8a, also see Methods). Consistent with previous results, increased dopamine decreases MBON11 activity (Fig. 8b, c). In line with our hypothesis, decreased dopamine increases MBON11 activity, supporting that dopamine features modulate MBON11 excitability to affect its gain and activity.

Fig. 8. Opposing dopamine pathways mediate hunger-dependent aggression.

a Schematic for MBON11 calcium imaging during simultaneous dopamine perfusion and ACV food odor stimulation in head-fixed males. Gray, blue, and orange shadows denote food odor stimulation alone, food odor stimulation with increasing dopamine, and food odor stimulation with decreasing dopamine, respectively. MBON11 response dynamics (b) and magnitudes (c) to food odor stimulation alone (black), food odor stimulation with increasing dopamine (blue), and food odor stimulation with decreasing dopamine (orange). MBON11 response dynamics (d) and magnitudes (e) to food odor stimulation during dopamine increase (black), and during dopamine increase in males with Dop1R1 (magenta) or Dop1R2 (blue) receptor knockdown by RNAi. Control is the same data as (b, c). MBON11 response dynamics (f) and magnitudes (g) to food odor stimulation during dopamine decrease (black), during dopamine decrease in males with Dop1R1 (magenta) or Dop1R2 (blue) receptor knockdown by RNAi. Control is the same data as (b, c). h Schematic for MBON11 calcium imaging during food odor stimulation in tethered males. Dop1R2 signaling is mainly mediated by calcium. MBON11 calcium response dynamics (i) and magnitudes (j) to food odor stimulation with Dop1R1 (magenta) or Dop1R2 (blue) receptor knockdown by RNAi. k Schematic for MBON11 cAMP imaging during food odor stimulation in tethered males. Dop1R1 signaling is mainly mediated by cAMP. MBON11 cAMP response dynamics (l) and magnitudes (m) to food odor stimulation with Dop1R1 (magenta) or Dop1R2 (blue) receptor knockdown by RNAi. n Working model for the neural mechanism of hunger modulation of aggression. Different hunger states induce distinct dopamine dynamics in PPL101, which trigger opposing dopamine receptors and second messenger signaling pathways in the same MBON11 neuron to oppositely modulate its excitability, activity, and aggression.

Paradoxically, Dop1R1 and Dop1R2 within a single MBON11 neuron oppositely modulate aggression by the same dopamine transmitter. To investigate whether specific mechanisms selectively activate particular receptors to modulate hunger-dependent aggression, we knocked down Dop1R1 or Dop1R2 in MBON11, performed dopamine perfusion during odor stimulation, while imaging MBON11 activity. When dopamine increases and MBON11 excitability falls, Dop1R1 knockdown significantly increases MBON11 excitability, overriding the activity decrease caused by dopamine increase (Fig. 8d-e). In contrast, Dop1R2 knockdown does not impair the activity decrease, indicating that Dop1R1 mediates negative modulation of MBON11 excitability, consistent with behavioral results where Dop1R1 knockdown increases aggression (Fig. 7o). Conversely, when dopamine decreases and MBON11 excitability rises, Dop1R2 knockdown overrides the activity increase caused by dopamine decrease, while Dop1R1 knockdown has no effect (Fig. 8f, g). This suggests that Dop1R2 mediates positive modulation of MBON11 activity, consistent with behavioral results. As an alternative test, given that studies in vitro and in other cell types showed that Dop1R2 is coupled to Gαq^25,63–65, we used the Gαq inhibitor YM-254890 to block Dop1R2 signaling during dopamine perfusion and odor stimulation (Supplementary Fig. 10a-d). Administering 1 μM of YM-254890 overrides the activity increase caused by dopamine decrease, but does not affect the activity decrease caused by dopamine increase, consistent with RNA interference results. In summary, Dop1R1 mediates negative modulation of MBON11 excitability by dopamine increase, while Dop1R2 mediates positive modulation of MBON11 excitability by dopamine decrease, indicating that specific dopamine features selectively activate Dop1R1 and Dop1R2 to modulate MBON11 excitability and activity, thereby regulating aggression. As hunger states change, dopamine levels fluctuate, and the antagonistic actions of Dop1R1 and Dop1R2 mediate the cycling between starvation and satiation, driving the bidirectional modulation of aggression.

Dopamine receptors, which are G protein-coupled receptors, influence membrane electrical activity through intracellular signaling pathways. Given that two opposing receptors in the same neuron oppositely modulate membrane excitability, we investigated how specific intracellular signaling pathways mediate opposing modulation by these receptors. Since calcium and cAMP are implicated in dopamine signaling in neuronal populations^25,66,67, we explored whether these second messengers are specifically recruited by Dop1R1 and Dop1R2 activation within the same neuron to achieve signaling specificity. We first examined calcium. We knocked down Dop1R1 or Dop1R2 and imaged MBON11 calcium while delivering food odor (Fig. 8h-j). Dop1R1 or Dop1R2 knockdown increases or decreases MBON11 calcium levels, consistent with the behavioral results. Interestingly, Dop1R2 knockdown leads to more dramatic calcium changes than Dop1R1, indicating that calcium plays a more important role in Dop1R2 downstream signaling in MBON11. We then turned to cAMP, first verifying the reliability of cAMP detection in a single MBON11 neuron using the G-Flamp1 sensor. We detected a clear cAMP signal upon stimulating flies using food odor ACV (Supplementary Fig. 10e-g). Additionally, applying the adenylyl cyclase activator forskolin, which increases cAMP, yielded a strong cAMP signal upon perfusion, confirming reliable cAMP detection in MBON11 (Supplementary Fig. 10h-j). Using cAMP imaging, we found that Dop1R1 or Dop1R2 knockdown increases or decreases MBON11 cAMP activity, similar to calcium (Fig. 8k-m). However, Dop1R1 knockdown leads to more dramatic changes in cAMP signal than Dop1R2, indicating that cAMP plays a more important role in Dop1R1 intracellular signaling in MBON11. As another validation, given that specific dopamine changes activate specific dopamine receptors, which in turn activate specific intracellular pathways, we hypothesized that specific dopamine changes lead to corresponding changes in specific second messengers. Therefore, a dopamine increase, activating Dop1R1, should decrease cAMP activity. Conversely, a decrease in dopamine, activating Dop1R2, should increase calcium activity, which is barely detectable with a cAMP sensor. Indeed, we confirmed this by imaging cAMP in MBON11 during changes in dopamine while delivering food odor (Supplementary Fig. 10k-m). These results on second messenger pathways are consistent with previous studies in other cell types^25,65. In summary, starvation- and satiation-induced dopamine changes activate specific receptors and trigger distinct second messenger pathways within a single MBON11 neuron, modulating its excitability and aggression (Fig. 8n).

Discussion

To fight or not is a critical decision: winning secures food, while losing can be detrimental. Thus, animals must assess their internal needs to plan aggressive actions. Our study revealed three findings across behavioral, circuit, and molecular scales underlying the hangry effect. First, hunger modulates aggression in a graded, nutrient-specific manner, coupling internal needs with aggressive actions. Second, we traced this coupling into the mushroom body, uncovering a key function for this ancient learning and memory center. Third, two opposing pathways within a single mushroom body output neuron detect opposing features from presynaptic dopamine to modulate membrane excitability, adaptively modulating aggression via gain control. This reveals a simple, single-cell implementation of opposing dopamine pathways, achieving similar functions to those accomplished by complex mammalian dopaminergic circuits for flexible action selection^68,69. Together, we mapped the hunger-aggression coupling to a single physiological parameter of excitability in a single neuron, and identified two opposing pathways within this neuron that detect opposing dopamine inputs to fine-tune this parameter, enabling bidirectional tuning of the modulation via two molecular knobs.

Mushroom body circuits link hunger states to aggression

We identified a multilayer circuit through the hangry neuron MBON11 in the mushroom body that couples graded, nutrient-specific hunger states with aggression (Fig. 4h, Supplementary Fig. 4f). Traditionally associated with learning and memory, the mushroom body has recently been implicated in internal states, including hunger^{1–3,39,70,71}. We revealed that MBON11 gain control precisely scales aggression with hunger states via membrane excitability (Fig. 5j, Fig. 6l, Supplementary Fig. 7l). Dopamine is known to indirectly modulate MBON via synaptic plasticity in presynaptic Kenyon cells during learning^22,25,35,53. We demonstrate direct dopaminergic modulation of MBON excitability for internal state, long hypothesized but now shown^3,35. Postsynaptic excitability offers another mechanism that may operate in tandem with presynaptic plasticity in compact mushroom body compartments to mediate dopamine modulation. Dopaminergic modulation of both excitability and plasticity also occurs in mammals^59,68,72. Future work will elucidate the generality and potential interaction^1,29,39,70 of these two mechanisms in the compact mushroom body circuit.

Opposing dopamine pathways transform hunger states into aggression

We identified opposing dopamine signaling pathways within a single neuron that transform hunger states into aggression (Fig. 8n). Instead of bidirectionally tweaking a single pathway, this system simultaneously detects opposing features from the same dopamine inputs to activate opposing pathways. Opposing pathway designs are also found in the complex mammalian dopaminergic system, where two populations of basal ganglia neurons expressing antagonistic D1 and D2 receptors jointly modulate behavior^68,69. PKA signals in D1 and D2 receptor-expressing neuron populations are regulated by the dynamics of dopamine increase and decrease⁷³, similar to our findings in Drosophila. Thus, the invertebrate mushroom body implements a similar computation via a simplified system, using opposing dopamine receptors and intracellular signaling pathways within a single neuron to adjust membrane excitability for gain control in aggressive behaviors. This way, hunger states are adaptively coupled with aggression.

Methods

Fly husbandry and genetics

Fly husbandry

Flies were reared on standard cornmeal food under a 12 h:12 h light: dark cycle at 25 °C and 50–60% relative humidity. All experimental flies were reared at a low density (3–4 male and 3–4 female parents), and collected on the eclosion day.

For most aggression and courtship assays, single-housing flies were isolated in 1.5 mL tubes containing 0.5 mL food for 6–8 days. For group housing flies, 25–30 males were collected and reared in a standard 35 mL vial. For imaging experiments, we used flies at an age of 6–8 days.

Fly genetics

See Supplementary Table 1 for details of all fly strains used in this study.

Quantitative ethology

Optogenetic ethology system

Behavior chamber

All assays were performed with a standard 12-well cell culture plate (Greiner bio-one) with an acrylic lid⁷⁴. Teflon PTFE aqueous dispersion was applied to both sides and top to prevent flies from climbing. The chamber floor was lined with an acrylic ring 22.25 mm in diameter, and the center was filled with 6 mm diameter apple juice food, composed of 67 mL apple juice (Huiyuan) mixed with 1.5 g agarose (Beyotime), 1.8 g sucrose (Sigma-Aldrich), and 0.3 g yeast (Saf-instant).

Behavior recording

We used a machine vision camera (MV-CA016-10UM, Hikrobot) based on IMX237 global shutter CMOS sensor (Sony) with a 25 mm focal length lens (MVL-HF2528-6MPE, Hikrobot) for behavior recording from above the chamber, at a resolution of 1440 by 1080 pixels and an acquisition rate of 100 frames per second (fps). We developed a C++ based program MIAS^40,75 for camera control. We used 850 nm LED as infrared illumination for video recording. A long pass filter was assembled between the camera and the lens to prevent interference from shorter wavelength light such as those from optogenetic stimulation.

Photostimulation control

We developed a setup with precise temporal control of multichannel LEDs, including red and green LEDs for optogenetic activation and inhibition, blue LEDs for animal vision, and infrared LEDs for behavior recording (Fig. 1a, Supplementary Fig. 1a). We red-shifted light-gated cation channel CsChrimson for photoactivation using 680 nm deep red light, less disturbing to fly vision, and light-gated anion channel GtACR1 for photoinhibition at 520 nm^76–78.

We designed a printed circuit board (PCB) accommodating both 680 nm and 850 nm LED chips (Supplementary Fig. 1a). We evenly soldered LED chips into a 6 by 6 block array, with each block containing 4 pieces of 850 nm LEDs and 1 piece of 680 nm LED, onto the 13 cm by 13 cm PCB with an aluminum substrate for heat dissipation. We used the 680 nm LEDs for CsChrimson photoactivation experiments, and 850 nm infrared LEDs for background illumination. This PCB was positioned directly below the chamber, separated by a 5 mm thick white acrylic board to ensure uniform light dispersion. Similarly, for GtACR1 photoinhibition experiments, we designed a PCB replacing 680 nm LEDs with 520 nm LEDs. A low-intensity 470 nm LED was positioned above the chamber throughout the experiment to provide visible light illumination for behavior.

Dedicated tunable DC power sources (4–36 V) provided power supply for each LED channel. We maintained 850 nm and 470 nm LED illuminations at constant powers. We controlled the intensity of 680 nm and 520 nm LEDs using digital output signals from the data acquisition board (PCIe-6321, NI) through a MOSFET switch circuit, operated by a custom MATLAB program. This custom program also controls the synchronized triggering of camera switches and digital control output for LEDs. Light intensities were measured using a photodiode power meter (S130C with PM100D console, Thorlabs).

Fly husbandry for optogenetics assays

Parental flies (3–4 males and 3–4 females) were crossed in 35 mL tubes with 5 mL of standard cornmeal food wrapped in aluminum foil to prevent light exposure. Progeny were collected on the day of eclosion and isolated in 1.5 mL tubes with standard medium containing 0.4 mM all-trans retinal (Cayman Chemical) and kept under blue light conditions (incubating boxes covered by blue films). Before experiments, wild-type male or female conspecific flies were anesthetized on ice and loaded into the chamber first. Then, test flies were gently aspirated into a buffer tube using a miniaturized vacuum pump and introduced into the experimental chamber by mouth aspiration. Flies were adapted for 2–5 min before experiments. Aggression and courtship experiments were generally conducted at 25 °C, 50–60% relative humidity, between ZT 0 and 4 h.

Optogenetic behavior experiments

Optogenetic photoactivation assay

For optogenetic photoactivation experiments with CsChrimson, we used a standard photostimulation protocol, consisting of a 5-min adaptation, followed by a pulse train of 1-min photoactivation interspersed with 2-min rest period repeating eight times (Fig. 2a). For the 1-min photoactivation in most of the experiments, we used a standard stimulation pattern of 50 Hz 10 ms pulse train, unless noted otherwise.

We generally used a power of 0.5 mW/cm² for activation. In the graded MBON11 activation assay, we used three different light intensities: 0.5, 1, and 2 mW/cm². For the male-female courtship assay with MBON11 optogenetic activation, we used a light intensity of 2 mW/cm². For male-male aggression without food assay with MBON11 photoactivation, we used three different activation protocols: 0.5 mW/cm² 50 ms 10 Hz, 1 mW/cm² 50 ms 10 Hz, and 1 mW/cm² continuous light. For simultaneous MBON11 photoactivation and Tk^FruM Kir2.1 silencing and of MBON11 photoactivation and PPL101 Kir2.1 silencing, we used a light intensity of 2 mW/cm².

For PPL101 photoactivation using CsChrimson, two males expressing CsChrimson were allowed to freely interact for 12 min, followed by a 2-min photostimulation at 0.5 mW/cm² using 680 nm LED.

Optogenetic photoinhibition assay

For optogenetic photoinhibition experiments using GtACR1, we employed a standard photostimulation protocol consisting of a pulse train of 2-min photostimulation interspersed with 2-min rest periods, repeated seven times (Fig. 2a). In all photoinhibition experiments, continuous light stimulation of 0.5 mW/cm² was used to prevent visible green light flickering that could disrupt vision and interfere with normal behaviors.

Dual-channel photostimulation assay

For dual-channel photostimulation, the protocol consisted of a 1-min pulse of 680 nm 2 mW/cm² photostimulation, followed by a 1-min pulse train of simultaneous 2 mW/cm² 680 nm and 0.5 mW/cm² 520 nm photostimulation, interspersed with 2-min rest periods, repeated for eight times. Also see the schematic in Fig. 4f and Supplementary Fig. 4b.

Hunger assays

For standard starvation assay, we transferred fed males to 1.5 mL tubes containing moist filter paper and subjected them to starvation for either 7 or 14 h.

For starvation-refeeding assay (Fig. 6k, Supplementary Fig. 1d), we first conducted a 30-min aggression test on males that have undergone starvation treatments of different durations. Subsequently, the males were anesthetized with CO₂, transferred to 1.5 mL tubes containing 0.5 mL of food, and reared for 2 days. After the refeeding period, the same males that had previously undergone the aggression test were retested.

For the nutrient-specific hunger assay, flies were initially collected and isolated into 1.5 mL centrifuge tubes containing 0.5 mL of standard cornmeal food and reared for 3–5 days, then they were transferred to pure sucrose food (100 mM sucrose) or pure yeast food (50 g/L yeast) for 2 days, and finally tested on sucrose or yeast food. 2% agarose containing 100 mM sucrose or 50 g/L yeast was loaded in the chamber for inter-male aggression test.

For protein hunger assay, flies were subjected to different feeding treatments: sucrose alone, sucrose with amino acids, sucrose with peptone, and sucrose with yeast. They were initially collected and isolated into 1.5 mL centrifuge tubes containing 0.5 mL of standard cornmeal food and reared for 3–5 days. Afterwards, the flies were transferred to another 1.5 mL centrifuge tube containing 0.5 mL of 2% agarose, with the following treatments added to the agarose: 100 mM sucrose, 100 mM sucrose with complete amino acid mixture (including 2.5× MEM EAA, 2.5× MEM NEAA, and 5 mM glutamine), 100 mM sucrose with 25 g/L peptone, or 100 mM sucrose with 50 g/L yeast. To make sucrose with a complete amino acid mixture food, we sequentially added MEM EAA mix (50×), NEAA mix (100×), and 200 mM glutamine to a solution containing 100 mM sucrose, achieving final concentrations of 2.5×, 2.5×, 5 mM, and 100 mM, respectively, and finally add 2% agarose⁵. Flies were reared on specific food for 3 days. 2% agarose containing 100 mM and 50 g/L yeast was loaded in the chamber for inter-male aggression test.

Thermogenetics

We used temperature-sensitive cation channel dTrpA1⁷⁹ for thermogenetic activation. For thermogenetic inhibition, we used inwardly rectifying potassium channel Kir2.1⁸⁰ with tubulin-GAL80^ts for temperature-gated timing control⁸¹.

Flies were reared at 19 °C for 6–8 days. To induce the expression of UAS-Kir2.1, adults expressing tub-Gal80^ts; UAS-Kir2.1 were transferred to 30 °C for 2 days before the experiment. Experiments were carried out at 25 °C for flies expressing tub-Gal80^ts; UAS-Kir2.1, and at 29 °C for flies expressing UAS-dTrpA1. Also see the schematic in Supplementary Fig. 2b.

Kir2.1 silencing and RNA interference

For RNA interference (RNAi) or Kir2.1 silencing, flies were collected on the day of eclosion and isolated into 1.5 ml centrifuge tubes containing 0.5 ml of standard cornmeal food. The flies were then reared for 6–8 days under conditions set at 25 °C with a relative humidity of 50–60%, following a light-dark cycle of 12 h each. Experiments were conducted at 25 °C.

Behavior analysis

Video preprocessing

We used the machine-learning-based program SoAL^40,75 for feature detection. Briefly, we took a top-down approach for feature detection, including region of interest (ROI) extraction, and body detection, followed by keypoint prediction, and identity assignment.

Behavior classification

Based on the detected features from SoAL, we used Janelia Automatic Animal Behavior Annotator (JAABA)⁸² to classify behavior types including lunge, tussle, and courtship⁷⁴. To obtain training datasets for JAABA, we manually annotated frames into different types of behavior based on knowledge from previous studies^{12,41,44,83–86}. We categorized bouts comprising at least five consecutive frames (i.e., over 50 ms) as bouts of a specific behavior type⁸⁶.

Behavior metrics

Lunge frequency is defined as the total duration of lunges, measured in seconds, occurring within a 1-min time window. Note that lunge bouts are typically very brief.

The lunge frequency difference is calculated as:

$$\mathrm{lung}\mathrm{e}\mathrm{frequency}\mathrm{difference}={\mathrm{f}}_{\mathrm{stim}}-{\mathrm{f}}_{\mathrm{pre}}.$$

Here, f_stim and f_pre are the lunge frequencies during and before (rest period) photostimulation.

The lunge frequency ratio is calculated as:

$$\mathrm{lung}\mathrm{e}\mathrm{frequency}\mathrm{ratio}=\left({\mathrm{f}}_{\mathrm{stim}}-{\mathrm{f}}_{\mathrm{pre}}\right)/{\mathrm{f}}_{\mathrm{pre}}.$$

Lunge probability is defined as the probability of a lunge occurring at any given moment. For the lunge probability dynamics plots, a Savitzky-Golay filter was applied to reduce noise, using a window size of 5 s.

Functional imaging

Two-photon in vivo functional imaging

Fly surgery for imaging

We folded a laser-cut 0.02 mm-thick stainless-steel shim (SUS304) into a pyramidal shape containing a fly head-sized hole in the middle, and glued the pyramid onto a custom-made aluminum holder. We first anesthetized a fly with ice and then put the fly into a fly-sized bed (cuprum). To keep the fly anesthetized, the temperature of the bed was maintained at about 4 °C. To tether the fly, we used a tiny drop of UV glue (Norland Optical Adhesive 68) and solidified it with UV light to tether the fly head and thorax with the shim. We adjusted the posture of the tethered fly head close to its natural state, where the head was facing forward, exposing the dorsal part of the brain for imaging. We excised the cuticle using a sharp pin (501317, WPI) and removed it with fine forceps (0103-5-PO, Dumont). We further removed the trachea and fat tissue to expose the brain. The brain was immersed in the extracellular saline during surgery⁵⁶: 103 mM NaCl, 3 mM KCl, 1 mM NaH₂PO₄, 5 mM TES, 26 mM NaHCO₃, 4 mM MgCl₂, 2.5 mM CaCl₂, 10 mM trehalose, and 10 mM glucose, adjusted to pH 7.4, osmolarity 270–275 mOsm. To minimize brain movement, we cut the muscle M16 and fully fixed the proboscis with low-melting-point wax. After surgery, we allowed flies to recover for approximately 10 min in an environment maintained at 25 °C and 50% humidity. To assess the post-surgery activity of the flies, we provided a small piece of tissue, and discarded inactive flies from further imaging experiments. To prevent movement during imaging, we fixed the legs in some experiments. We randomly selected the left or right hemisphere and imaged each fly for no more than 40 min.

Food and odor stimulation during imaging

For solid food stimulation, we developed a motorized solid food delivery setup. We generated a trigger signal to a stepper motor controller to deliver the food to the fly. We first attached a small droplet of food to a pin and positioned it about 5 mm in front of the fly, then brought food close to the fly at a constant velocity over a period of 3 s, followed by a 3-s pause, and finally returned to its original position.

For food odor stimulation, we developed a custom odor delivery setup⁸⁷. A continuous air stream was set to 600 mL/min and controlled by a float flowmeter (LZB-3WB, Kede Instruments). To deliver odor stimulation, we triggered a solenoid valve (KVE33PL12FP3Q201, Kamoer) to channel the air stream through a bottle containing an apple cider vinegar solution (Haitian). The odor was then delivered towards the antennae of the tethered fly via a 1 mm diameter metal tube positioned 4 to 5 mm away. For olfactory only stimulation, we used 5-s ACV odor stimulation.

Two-photon in vivo calcium, dopamine, and cAMP imaging

For two-photon in vivo calcium, dopamine, and cAMP imaging, we used a tunable ultrafast laser (InSight X3, Spectra-Physics) and tuned the wavelength to 940 nm to excite GCaMP^88,89, gDA⁹⁰, and G-Flamp⁹¹ sensors. We used a 12 kHz resonant galvanometer for scanning at a frame rate of about 88 Hz at a resolution of 128 by 128 pixels from an 84 by 84 μm² region of interest. We used a 750 nm short pass dichroic to filter nonspecific light. The fluorescence emission signals were amplified and collected by a GaAsP photomultiplier tube (PMT2100, Thorlabs) through a 525/50 nm bandpass filter. We used a 25 × N.A. 1.1 water dipping objective (Nikon). We adjusted the power at the back aperture to less than 25 mW (unless stated otherwise) using a Pockels cell and optimized the gain of the PMT for signal noise ratio. We used ThorImaging software (Thorlabs) for image acquisition, and ThorSync software (Thorlabs) for recording the imaging frame-out signals. We generated a 50 Hz signal to trigger a camera (MV-CA016-10UM, Hikrobot) for monitoring the fly during imaging. This signal was also recorded simultaneously by ThorSync to synchronize with the imaging data.

For PPL101 spontaneous dopamine release imaging, we used flies expressing dopamine sensor gDA3 in PPL101. To prevent movement, we used low-melting-point wax to fully fix the legs and wings. We imaged PPL101 dopamine release for approximately 200 s per trial without any external stimulation. Each fly was imaged for three to six trials.

Dual-color two-photon in vivo calcium imaging

For dual-color calcium imaging with GCaMP and RGECO⁹², we employed the same parameters described above, with some modifications. For simultaneous jGCaMP7b and jRGECO1a fluorescence excitation, we tuned the laser wavelength to 1000 nm based on the fluorescence expression. The fluorescence emission signals were split by a 562 nm long-pass dichroic, and detected by a pair of PMTs with a 525/50 nm bandpass filter and a 607/70 nm bandpass filter for the green and red channels, respectively.

Pharmacology and two-photon imaging

Dopamine perfusion and imaging

We developed a custom perfusion setup to exchange bath solutions⁸⁷. The dopamine solution and extracellular saline were stored separately in two containers, each connected to the holder via distinct tubes. The extracellular saline was connected to a normally open solenoid valve (0520D, fspump), and the dopamine solution to a normally closed solenoid valve (0520 C, fspump). A signal simultaneously triggered these two valves, and the on-off switching of the two valves led to solution exchange in the chamber. We maintained a perfusion rate of 2 mL/min for both solutions, with dopamine perfusion lasting 60 s each trial. A vacuum setup aspirated the solution from the chamber.

We prepared 200 μM dopamine solution daily by diluting a frozen stock of 100 mM dopamine (H8502, Sigma-Aldrich) in extracellular saline, adding 100 μM sodium metabisulfite (S9000, Sigma-Aldrich) to prevent oxidation⁹³. To monitor dopamine dynamics on the extracellular side of the MBON11 membrane, we imaged flies expressing gDA3 dopamine sensor on MBON11 during dopamine perfusion experiments. We monitored the fly with a camera and checked for leg or abdomen movement to ensure it remained alive.

For the dopamine increase condition, dopamine perfusion began simultaneously with ACV odor stimulation and lasted for 60 s. For the dopamine decrease condition, we perfused dopamine for 60 s to elevate dopamine concentration, followed by 30 s of saline perfusion to reduce it, after which we delivered the ACV odor for 60 s. For control experiments, we continuously perfused saline while delivering odor for 60 s.

YM-254890 and forskolin administration and imaging

For calcium imaging during ACV odor stimulation while simultaneously perfusing dopamine in saline containing YM-254890 (Supplementary Fig. 10a-d), we applied YM-254890 (Wako Chemicals) to the extracellular saline at a final concentration of 1 μM.

To elicit cAMP dynamics while imaging cAMP in MBON11 neurons (Supplementary Fig. 10h-j), we applied adenylyl cyclase activator forskolin (Tocris) to the extracellular saline at a final concentration of 100 μM.

Functional connectivity

Distinct two-photon and one-photon optogenetic stimulation methods were used and combined with two-photon imaging according to the requirements of each experiment.

For PPL101 to MBON11 functional connectivity (Supplementary Fig. 5f), we expressed CsChrimson in PPL101 and jGCaMP7s in MBON11, and used the same system for in vivo two-photon calcium imaging discussed above. For two-photon optogenetic stimulation, we employed the 1045 nm fixed output (InSight X3) for photostimulation. Stimulation patterns were generated by ThorImaging software and delivered by galvo-galvo mirrors. Stimulation power was approximately 30 mW under the objective, with a stimulation duration of about 0.5 s. PMTs were shuttered to protect them during optogenetic stimulation. For one-photon optogenetic stimulation, we used a 565 nm LED light source (LED4D067, Thorlabs) tuned at about 30 μW and stimulated for about 0.8 s through the objective. For control group, we used the same stimulation duration but switched off the stimulation light source.

For PPL101 to MBON11 functional connectivity with ACV odor stimulation (Fig. 5h-i), we employed a 940 nm laser for MBON11 two-photon imaging and PPL101 two-photon photostimulation, since both MBON11 and PPL101 neurons arborize in the γ1 and peduncle compartments of the mushroom body. The laser power under the objective was set to approximately 20 mW for low-power photostimulation and adjusted to approximately 40 mW for high-power photostimulation. During photostimulation and imaging, we delivered ACV odor stimulation to activate MBON11.

For PPL101 photostimulation and PPL101 dopamine release imaging with dopamine indicator gDA3h, either by photoactivation with CsChrimson or photoinhibition with GtACR1 (Fig. 7n), we used the 1045 nm laser tuned to approximately 100 mW for two-photon stimulation in the γ1 and peduncle compartments for 1 s while two-photon imaging PPL101 dopamine using 940 nm laser. For the control group, we applied the same photostimulation to flies without opsin expression.

For Trh to MBON11 functional connectivity (Supplementary Fig. 3e-f), we used a holographic photostimulation system for spatial specificity. We expressed red-shifted opsin CsChrimson in serotonergic neurons and recorded calcium responses in MBON11. A spatial light modulator (LCOS-SLM X10468, Hamamatsu) driven by a 594 nm continuous wave laser (Coherent) was used for patterned one-photon photostimulation. We tuned the tunable ultrafast laser (InSight X3, Spectra-Physics) to 920 nm. An 8 kHz resonant galvo mirror was used for scanning, resulting in a nearly 7 Hz frame rate for a 793 by 780 pixels resolution. The fluorescence emission was collected by gated GaAsP PMTs (H11706, Hamamatsu) with sub-microsecond electronic shuttering. PMT shutters were closed during stimulation to protect detector. We used a 20 × N.A. 1.0 water dipping objective (Zeiss). We manually defined stimulation regions of interest (ROIs) according to the fluorescence expression patterns. Samples were stimulated at the ROIs with five repeats of 50-ms pulses at 50-ms intervals.

For functional connectivity experiments between Tdc2 and MBON11 (Supplementary Fig. 3i) and between MBON11 and Tk^FruM neurons (Supplementary Fig. 4d-e), we used a 565 nm LED light source tuned to approximately 70 μW for about 1 s directly through the objective while conducting two-photon calcium imaging. For the control group, we applied the same photostimulation to flies without opsin expression.

Functional imaging data processing

Imaging data preprocessing

We applied piecewise rigid motion correction⁹⁴ to the imaging data to account for motion artifacts. We calculated the mean and standard deviation of the motion-corrected images and used them for manual definition of the ROI based on anatomical features. For MBON11 or PPL101 in vivo imaging experiments, we defined the γ1 and pedc compartments as the ROI. For Tk^FruM neuron imaging, we defined the characteristic ring-shaped arborization¹⁴ as the ROI. After background subtraction of motion-corrected image stacks, we calculated fluorescence changes ΔF/F₀ based on the mean value of all pixels within each ROI. We used mean fluorescence changes prior to stimulation as baseline fluorescence. Trials with abnormal baseline were discarded. We used a 3rd- or 4th-order Savitzky–Golay filter over 7 frames (151 frames for perfusion experiments) to filter the fluorescence signals. To correct for photobleaching in cAMP imaging, we also used a 5th-order 0.03 Hz Butterworth high-pass filter. To reduce noise in the signal, we performed boxcar filtering. For spontaneous dopamine release imaging, we accounted for baseline fluctuations by applying a second-degree polynomial fit to the data in each trial. This fit was used to estimate the baseline signal, which was then subtracted from the original signal to correct for baseline.

Response quantification

To quantify response amplitudes, we calculated the area under the curve (AUC) or mean responses, and adjusted the window according to the response dynamics. For MBON11 responses to solid food delivery (Fig. 2i-k), the AUC was calculated from 1 to 4 s after the onset of food approach. For PPL101 responses to solid food delivery (Fig. 3h-i), the AUC was calculated from 1 to 5 s after approach onset for calcium imaging, and from 4 to 8 s for dopamine imaging. For simultaneous dopamine perfusion and odor stimulation during imaging (Fig. 8b-g), we calculated the mean ΔF/F₀ from 10 to 30 s after odor onset. To characterize offset responses, we calculated the mean ΔF/F₀ from 2 to 12 s after odor offset (Supplementary Fig. 10m). For second messenger imaging (Fig. 8h-m), we calculated the mean ΔF/F₀ during the first 2 s after odor onset.

To quantify the response slopes of MBON11 to solid food delivery (Supplementary Fig. 6c), we calculated the slope from 1 to 3 s after approach onset, using linear regression via the SciPy package.

To quantify PPL101 spontaneous dopamine release (Fig. 7d-g), we first calculated the median ΔF/F₀ of each trial. We then calculated the mean ΔF/F₀ of responses above and below this median. Finally, we used the difference between these two means as the dopamine release amplitude for each trial.

To quantify MBON11-PPL101 dual-color imaging, we compared responses from the two distinct indicators (Supplementary Fig. 5c-d). We normalized the responses for each indicator to a [0,1] range. Then, we calculated the relative MBON11-PPL101 difference for each trial based on the peaks of the normalized responses.

Decoding hunger duration

We constructed a time-evolving linear support vector machine (SVM) decoder to classify hunger durations. We categorized the hunger durations into four conditions. We randomly selected 80% of the data to form the training set and the remaining 20% as the testing dataset. We repeated the decoding training process ten times and calculated the mean accuracy as the decoder performance metric. As a control, we trained a decoder using shuffled data to compare its accuracy with that of the actual data.

Electrophysiology

Whole-cell patch-clamp recording in vivo

We visualized the brain using a sCMOS camera (Zyla 4.2 Plus, Andor, and Prime BSI Express, Teledyne) mounted on a custom fluorescence microscope. We first used a glass pipette filled with extracellular saline (1–2 MΩ, 1.5 mm O.D., 1.1 mm I.D., BF150-110-10, Sutter) for desheathing and exposing the soma by mechanical stress. After exposing the soma of the MBON11 neuron, the brain was superfused with fresh extracellular saline equilibrated with 95% O₂ and 5% CO₂ for 1-2 min. Patch pipettes (10–12 MΩ) were pulled and polished on the day of the experiment and filled with the internal solution (125 mM L-Aspartic acid, 125 mM KOH, 0.1 mM CaCl₂, 1.1 mM EGTA, 10 mM HEPES, 4 mM MgATP, 0.5 mM Na₃GTP, pH adjusted to 7.3 ± 0.1 by KOH or L-Aspartic acid, osmolarity adjusted to 265 ± 3 mOsm by L-Aspartic acid potassium salt)⁹⁵. After forming gigaseal, we performed membrane rupture in voltage-clamp mode with a holding voltage at −54 mV using a MultiClamp 700B Amplifier (Molecular Devices). Whole-cell recordings were performed in current-clamp mode, with bridge balancing and holding currents ranging from 0 to 100 pA to compensate for current leakage. Data were filtered at 3–6 kHz by the amplifier and digitized at 10 kHz using a data acquisition board (PCIe-6321, National Instruments) controlled by the Matlab-based WaveSurfer software (HHMI Janelia Research Campus).

In the odor stimulation experiment, air moistened with ACV was delivered as the stimulating odor through a glass tube positioned in front of the fly (BF150-110-10, Sutter) for 30 s. Beyond stimulating intervals, air moistened with distilled water was continuously delivered through the same glass tube. For hunger-dependent experiments, the flies were single-housed in 1.5 mL centrifuge tubes containing filter paper pre-soaked with water for 14–18 h to induce a hunger state before recording. Fed flies were single-housed in 1.5 mL centrifuge tubes containing 300 μL fresh standard fly food.

Pharmacology and whole-cell recording

For dopamine administration, the dopamine stock solution was diluted to a final concentration of 10 mM in extracellular saline and kept on ice prior to the experiment. Dopamine solution was then loaded into a glass pipette (1 mm O.D., 0.78 mm I.D., BF100-78-10, Sutter) with a resistance of 1-2 MΩ. The tip of the pipette was positioned in the γ1 compartment. The dopamine solution was injected via a 500 ms continuous pressure pulse at 70 kPa using an air pump (DMP-400, MPI).

Current injection and whole-cell recording

For current injection, neurons were stimulated with 5 pA increments from −40 pA to 105 pA in current injection steps. Each step had a duration of 300 ms, with 3 s intervals between steps.

For current injection before and after dopamine administration, we used a current step protocol ranging from −40 to 105 pA in 5 pA steps. The same protocol was repeated within 30 s after the dopamine puffing and after the MBON11 membrane potential had stabilized.

Electrophysiology data analysis

To remove line noise from in vivo recording during ACV odor stimulation, we calculated and subtracted the mean membrane potential during baseline period from the raw data. The resulting data were processed with a fast Fourier transform (FFT) to remove the 50 Hz frequency component. Afterwards, an inverse Fourier transform (IFFT) was applied, and the mean membrane potential was added back to restore the recording signal. Electrophysiology data were analyzed using the peak-finding function in Origin 2025 (OriginLab Corporation). Spikes were identified by locating the local maximum within the membrane potential trace, using windows of 30 sampling points. Spikes were filtered based on amplitude, retaining only those with amplitudes exceeding 40% of the absolute value of the difference between the maximum and minimum values of each step trace. The first derivative of the spike trace was then examined, and only spikes with an ascending rate to the peak exceeding 8 mV/ms were included in the analysis.

Firing rates were determined by the inverse of the average peak intervals between all spikes, determined for each level of injected currents. The first spike latency was calculated as the interval between the onset of the current pulse at the specified step intensity and the occurrence of the first spike peak for each step. Afterhyperpolarization was determined by the local minimum membrane potential following each spike.

Immunofluorescence

Brains were dissected in phosphate-buffered saline (PBS) for 30 min and subsequently fixed in 4% paraformaldehyde at 4 °C for 30 min, followed by rinsing with PAT3 buffer (0.5% BSA, 0.5% Triton X-100, in PBS) three times for 10 min each. The samples were blocked by 3% normal Donkey serum at 4 °C for 1 h, and then incubated overnight at 4 °C in primary antibody solution containing 1:30 mouse anti-Brp (nc-82, Developmental Studies Hybridoma Bank) and 1:200 rabbit anti-GFP (A11122, Invitrogen). After rinsing three times as detailed above, the samples were incubated at 4 °C for 4 h in a secondary antibody solution containing 1:500 Alexa Fluor 568 goat anti-mouse (A11004, Invitrogen) and 1:500 Alexa Fluor 488 goat anti-rabbit (A11008, Invitrogen), and then rinsed again. Fluorescence photomicrographs were acquired on a Nikon A1 HD25 upright confocal microscope with a 25× N.A. 1.1 water-dipping objective (CFI75 Apochromat 25XC W).

Connectome analysis

We used the FlyWire^50,51 (v783), the hemibrain⁴⁹ (v1.2.1), and the male-cns⁹⁶ (v0.9) datasets for all analyses in this manuscript. To access the hemibrain dataset, we accessed the neuPrint database through its Python API, neuprint-python (https://pypi.org/project/neuprint-python/). FlyWire dataset was from Codex (https://codex.flywire.ai/api/download?dataset=fafb). Male-cns dataset was downloaded (https://male-cns.janelia.org/download/#__tabbed_1_2). All analyses were conducted in Python⁵⁵.

Connectivity diagram

We counted the total number of synapses between MBON11 and each of its major upstream and downstream neurons or neuron types based on FlyWire and male-cns (Supplementary Fig. 3b), or hemibrain (Fig. 3b). Kenyon cells were excluded from this analysis.

To clarify the definitions of some neuron types in Fig. 3b and Supplementary Fig. 3b, we excluded PPL101 and PPL102 in DAN_feedback. We categorized mushroom body output neurons into two groups, those with feedback, MBON_feedback, and those without feedback, MBON. Both groups excluded MBON11. Superior Protocerebrum includes SIP, SMP, SLP, and CRE⁵⁵.

To visualize the connectivity strength, we set the width of arrows (w_arr) according to the logarithm of the synapse counts (N_syn). For FlyWire dataset:

$$w_{arr}~{\log }_2{N}_{syn}-2.5$$ 1

For hemibrain dataset:

$$w_{arr}~2*\ln \left({2*N}_{syn}\right)-8$$ 2

Volumetric reconstruction

Neuron skeletons are visualized in 3D plots by the NAVis package (https://navis-org.github.io/navis/) in Python⁹⁷ based on the SWC data from the FlyWire and the hemibrain datasets. Synaptic locations are determined by the post-synaptic sites. The coordinates of data in FlyWire and hemibrain are based on the FAFB brain templates (outline in Fig. 3c and Supplementary Fig. 3c) and the JRCFIB2018F brain templates (outlined in Supplementary Fig. 3c), respectively. In all figures, “left” and “right” are defined according to the frontal view.

To visualize MBON11 with respect to the mushroom body, we adapted the JFRC2⁹⁸ template of the mushroom body in Fig. 3a and Supplementary Fig. 3a. To visualize the γ1-pedc compartments, we adapted the JRC2018U templates⁹⁹ of the γ lobes and pedunculi which include both sides and do not contain other compartments of the mushroom body, and transformed into FAFB and JRCFIB2018F templates in Fig. 3c and Supplementary Fig. 3c, respectively. Transformations between different templates are performed by the NAVis-flybrains package (https://github.com/navis-org/navis-flybrains).

Neurotransmitter classification

We used machine learning-based neurotransmitter prediction based on EM images¹⁰⁰, and analyzed the synapse-specific dataset (https://zenodo.org/records/10593546).

We identified PPL101-MBON11 synapses (ipsilateral, same as follows) by matching presynaptic site positions between database synapse information and the prediction results.

In FlyWire, transmitter types are defined using a predicted score threshold of above 0.8, thereby classifying approximately 80% of PPL101-MBON11 synapses as dopaminergic. In hemibrain, we applied a predicted score threshold of 0.6 to account for the greater diversity of predicted neurotransmitter types among PPL101-MBON11 synapses. This approach results in 40% of all synapses being predicted as dopaminergic, while the remaining synapses classified as either serotonin or octopamine. Acetylcholine, glutamate, and GABA synapses were rare and thus were not included.

Statistical analyses

We applied nonparametric tests that do not assume an underlying distribution for descriptive statistics and significance tests. Mann-Whitney U tests were used for between-group pairwise comparisons, while Wilcoxon signed-rank tests were used for within-group pairwise comparisons, respectively. For multigroup comparison, we used a Kruskal-Wallis analysis, followed by a post-hoc Dunn’s multiple comparisons test or Mann-Whitney U test for between-group pairwise comparisons, and a post-hoc Wilcoxon signed-rank test for within-group pairwise comparisons. For current injection data in electrophysiology, we used two-way ANOVA. Sample sizes were not predetermined statistically, but are in line with other studies in the field.

In all statistical analyses, the notations are as follows: ns for p > 0.05, * for p < 0.05, ** for p < 0.01, **** for p < 0.001, and ***** for p < 0.0001. Error bars in all panels denote the standard error of the mean (SEM). Statistical analyses were carried out with Python 3.10 using Scipy.stats packages or GraphPad Prism 10.0.

Reporting summary

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

Author contributions

J.-L.W., Q.-T.L., Y.-M.G., Y.-K.Y., J.N., Z.L., Y.-F.S. conducted the experiments and data preprocessing with assistance from D.-D.C., X.-J.Z., W.-J.W., Q.L., Q.-C.L.; Y.S., J.-L.W., Q.-T.L., Y.-M.G., Y.-K.Y., J.N., W.-J.W. analyzed the data; Y.S., J.N., X.-J.Z. developed methods for quantitative ethology; Y.S., J.-L.W., Q.-T.L., Y.-M.G., Y.-K.Y. wrote and revised the manuscript with contributions and comments from all authors; Y.S. conceived and supervised the study.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

Y.S. discloses support for the research of this work from National Natural Science Foundation of China (32371055, 32070991), Research Center for Industries of the Future of Westlake University (WU2022C011), “Pioneer” and “Leading Goose” R&D Program of Zhejiang (2024SSYS0031), Westlake Laboratory of Life Sciences and Biomedicine (W101386022101), Key Laboratory of Growth Regulation and Translational Research of Zhejiang Province, Key Laboratory of 3D Micro/Nano Fabrication and Characterization of Zhejiang Province (10316A091901-001), Westlake Education Foundation, and Westlake University.

Data availability

All data supporting the findings of this study are available within the paper and its Supplementary Information. Source data are provided with this paper as a Source Data file. Source data are provided with this paper.

Code availability

Code for lunge detection is deposited at https://doi.org/10.6084/m9.figshare.31333915, code for two-photon imaging is deposited at https://doi.org/10.6084/m9.figshare.32326371.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76608-y.