Additional experimental procedures are provided in the Supplementary Information.

We evaluated compulsive-like eating in wild-type (WT) and dopamine D2R knock-out (Drd2) mice by employing a palatable food (PF) self-acquisition task. When a mouse pressed the active lever, a sucrose pellet (i.e., PF) was delivered. After 7 days, the number of lever presses required to obtain PF was increased, shifting from a fixed ratio of 1 (FR1) to FR2 and FR3 training schedules (Fig. 1A). Both groups successfully acquired lever pressing as the sessions progressed (Fig. 1B, C), although Drd2 mice displayed slightly less active lever pressing than WT mice across sessions, probably due to decreased locomotor activity in these mice [27]. During FR1 to FR3 sessions, both WT and Drd2 mice exhibited significantly increased active lever pressing than inactive lever press, demonstrating that the increase in lever pressing is specific to PF-seeking (Supplementary Fig. 1B-E).

After the FR3 sessions, additional FR3 sessions with delivery of electric foot shock (FS) were administered (FR3 + FS), and the perseverance of lever pressing to obtain PF despite the delivery of foot shocks was used as an indicator of compulsive-like food-seeking. Surprisingly, Drd2 mice exhibited significantly higher active lever pressing despite receiving foot shocks (Fig. 1B, D), showing perseverance in reward-seeking despite punishment. In contrast, WT mice exhibited significantly increased inactive lever press compared to active lever during FR3 + FS sessions (Supplementary Fig. 1F). When the rewards were omitted and only foot shocks were delivered (FS only), lever pressing was comparable between WT and Drd2 mice (Fig. 1B, E, Supplementary Fig. 1B, G), indicating similar sensitivity to foot shocks, but greater compulsivity in Drd2 mice, driving active lever pressing despite punishment. These data, together with previous observations [11], indicate that the absence of D2Rs promotes compulsive-like eating behavior.

To further address the role of D2Rs in the control of compulsive eating, we targeted D2Rs in the CeA based on our previous finding that CeA D2Rs play a crucial role in reward-related impulsive and compulsive behaviors [11]. We generated Drd2 mice (Supplementary Fig. 2) and injected AAV-Cre-eGFP virus into the CeA to selectively eliminate D2R expression in the CeA (CeA-Drd2KO mice) (Fig. 1F), resulting in a 73% reduction in D2R expression in CeA-Drd2KO mice (Cre-GFP) compared with control flox mice in which AAV-eGFP virus was injected (CeA-Drd2WT, GFP) (Fig. 1G, H). We then trained mice in the PF self-acquisition task.

To understand how the D2R activation induces the tyrosine phosphorylation of InsRs, we hypothesized that D2R activation with Gi protein coupling promotes the phosphorylation of InsRs. Several studies demonstrate that the cAMP-inhibiting Gi proteins, including G, positively regulate InsR activation [33,34,35,-36]. By contrast, the expression and activity of protein-tyrosine phosphatase 1B (PTP1B) is decreased in tissue expressing constitutively active Gα [37]. We examined the involvement of PTP1B by analyzing the phosphorylation level of PTP1B at Ser50 (pPTP1B), which negatively modulates its phosphatase activity, thus creating a positive feedback mechanism for insulin signaling [38]. We observed that D2R activation by quinpirole significantly increased the number of pPTP1B-positive cells in CeA-Drd2WT (Drd2 + GFP) mice, whereas insulin or quinpirole infusion had little or no effect on number of pPTP1B-positive cells in CeA-Drd2KO mice (Drd2+Cre-GFP) (Fig. 2K, L and Supplementary Fig. 4F). These data suggest that D2R activation induces PTP1B inhibition and impairs its ability to dephosphorylate InsRs, leading to increased insulin signaling. Thus, the stimulation of CeA D2Rs may induce concomitant activation of InsR signaling in the CeA, and the absence of D2Rs severely impairs InsR signaling.

We next tested the effect of selective loss of InsR expression in CeA D2R- expressing neurons on compulsive-like eating behavior. To accomplish this, we used a gene knockdown strategy with Cre-dependent AAV vectors expressing a short-hairpin RNA (shRNA) targeting InsRs to induce cell type-specific loss of function while simultaneously visualizing recombination through mCherry labeling using AAV-DIO-DSE-mCherry-PSE-shInsR [39]. Selective expression of shInsR in D2R- expressing neurons in the CeA (CeA-D2R) resulted in an ~80% reduction in InsR expression compared with control mice in which AAV-DIO-eYFP was injected (CeA-D2R) (Fig. 3A-C). We found that D2R-specific loss of InsRs in the CeA exacerbated compulsive-like eating phenotype, as evidenced by robust perseverance of active lever pressing despite foot shock punishment (Fig. 3D-F, Supplementary Fig. 5B-F). In FS-only sessions, no difference in lever pressing was observed between control and CeA-D2R mice (Fig. 3D, G, Supplementary Fig. 5G). Together, these data suggest that InsRs and the interaction between InsRs and D2Rs in the CeA are critical for the control of compulsive-like eating behavior.

To understand how CeA D2R-expressing neurons respond to PF, we used fiber photometry to perform in vivo calcium imaging. We virally expressed the Cre- dependent calcium indicator GCaMP6s (AAV-DIO-GCaMP6s-mCherry) in the CeA of Drd2-Cre mice (Fig. 4A) and recorded changes in GCaMP6s fluorescence signal through an optic fiber placed above CeA D2R-expressing neurons during exposure to PF in a light/dark box test. Histological analysis revealed that ~96% of GCaMP6-expressing cells overlapped with Cre-expressing CeA D2R-expressing cells (Fig. 4B, C). After measuring baseline body weight and food intake, mice underwent a pre-test without food for 15 min to measure their time spent in the light and dark chambers of the light/dark box [11]. Mice were then divided into two groups: one group was fed a NC diet, and the other group was fed a PF diet high in sugar and fat ad libitum for 14 days. After a 48-h PF withdrawal, mice were returned to the light/dark box for 15 min, with PF placed in the light chamber, to assess compulsive-like eating behavior [11], during which we recorded GCaMP signals from CeA D2R-expressing neurons (Fig. 4D). As shown by peri-event time histogram (PETH) analysis, there was a significant decrease in z-score (i.e., normalized ΔF/F) during PF consumption (Fig. 4E, G), whereas no significant change in GCaMP signal was observed during exposure to NC (Fig. 4E, F), indicating a decrease in CeA D2R-expressing neuronal activity in response to PF. This suppression was accompanied by increased PF intake in the light chamber, supporting a link between reduced D2R neuron activity and compulsive-like eating behavior (Fig. 4H).

We next recorded changes in the GCaMP6s signal from CeA D2R-expressing neurons during exposure to PF in the light/dark box test after infusion of saline, quinpirole, insulin, quinpirole+insulin, or the D2 receptor antagonist sulpiride into the CeA (Fig. 5A). Under saline infusion, CeA D2R neurons exhibited significantly reduced activity during PF consumption (Fig. 5B, C), consistent with the dynamics observed in Fig. 4E, and G. Quinpirole or insulin infusion, also induced significant reductions in z-scores during PF consumption (Fig. 5D-F) which was accompanied by a reduction in PF consumption (Fig. 5D-F, L). However, when quinpirole and insulin were co-infused, the suppression of neuronal activity was alleviated (Fig. 5D, G, L) corresponding to reduced PF consumption. In contrast, sulpiride infusion resulted in a significant suppression of neuronal activity and an increase in PF consumption (Fig. 5H, I, L), while its effect was slightly reversed by quinpirole (Fig. 5H, J, L) and fully alleviated by quinpirole+insulin, which also reduced PF consumption (Fig. 5H, K, L). These data indicate that D2R activation in the presence of insulin effectively enhances CeA D2R-expressing neuronal activity, thereby contributing to the suppression of PF consumption.

We further explored the effect of insulin on CeA D2R neuronal activity by ex vivo electrophysiological recordings in brain slices from Drd2-EGFP mice. Bath application of quinpirole alone (10 µM) failed to significantly change the membrane excitability of D2R-expressing CeA neurons (Supplementary Fig. 6A-E). CeA D2R neurons did not show significant changes in rheobase or membrane potential. However, when co-treated with insulin (500 nM), quinpirole dramatically depolarized their resting membrane potential and increased their excitability, as shown by decreased rheobase and enhanced input resistance (Supplementary Fig. 6C-E). When insulin alone was perfused into the recording chamber, it significantly hyperpolarized the resting membrane potential of CeA D2R neurons and decreased their firing probabilities by increasing rheobase and reducing input resistance (Supplementary Fig. 6F-J). These ex vivo electrophysiological studies suggest that the activation of D2Rs and InsRs has a synergistic effect that substantially increases the excitability of D2R neurons in the CeA. Furthermore, this interaction may be critical for the regulation of compulsive-like eating behavior, supporting our GCaMP6s photometry analysis following quinpirole or insulin infusion.

Next, we examined the effect of selective optogenetic inhibition of CeA D2R-positive neurons expressing halorhodopsin in the CeA of Drd2-Cre mice by injecting AAV-DIO-eNpHR3.0-eYFP virus (Fig. 6A, B) and conducted the light/dark box test (Fig. 6C-H). Bilateral photoinhibition of CeA D2R neurons resulted in a significant and specific increase in palatable food consumption in the light compartment (Fig. 6G, H and Supplementary Video 1). We further tested whether InsR signaling contributes to the regulation of CeA D2R-mediated PF consumption by selectively activating CeA D2R neurons. To this end, we injected AAV-DIO-eYFP (eYFP) or AAV-DIO-ChR2-eYFP (ChR2) into the CeA of Drd2-Cre mice followed by CeA infusion of the InsR antagonist S961 (Supplementary Fig. 7, Supplementary Fig. 8A, B). Mice underwent the light/dark box test, and pre-test evaluations revealed no differences in basal body weight, food intake, or light/dark box preference between groups expressing eYFP (control) or ChR2 in CeA D2R neurons (Supplementary Fig. 8C-F). Following this, the mice were divided into two groups: one group was maintained on a NC diet, while the other was provided with PF for 14 days. Afterward, mice were reintroduced to the light/dark box for a 25-minute test, with PF placed in the light compartment. Fifteen minutes before testing, mice received CeA infusions of either saline or S961.

In the PF-fed group, optogenetic activation of CeA D2R neurons significantly suppressed PF consumption in ChR2-expressing mice. However, this suppression was abolished when S961 was infused into the CeA, demonstrating that InsR signaling is necessary for CeA D2R-mediated regulation of compulsive-like PF consumption (Supplementary Fig. 8G, H). These findings indicate that CeA D2R neurons, in conjunction with InsR signaling, are critical for modulating compulsive-like eating behavior specific to PF. In the NC group, however, optogenetic activation of CeA D2R neurons had no effect on NC consumption, even after six hours of food deprivation prior to testing. Furthermore, S961 infusion did not alter NC consumption in either eYFP control or ChR2-expressing mice (Supplementary Fig. 8G, H). Together, these data indicate that reduced CeA D2R-expressing neuronal activity is associated with increased PF consumption whereas activation of CeA D2R neurons suppresses compulsive-like eating in an InsR-dependent manner (Fig. 6I).

Next, we examined changes in DA release in the CeA in response to PF using the genetically encoded fluorescence DA sensor GRAB-DA2m [40]. We prepared two groups of Drd2-Cre mice: one receiving AAV- DIO-DSE-mCherry-PSE-shLacZ in the CeA as a control (shLacZ) and the other receiving AAV- DIO-DSE-mCherry-PSE-shD2R (shD2R) for selective D2R knockdown in CeA D2R neurons (Supplemental method). Selective expression of shD2R in D2R-expressing CeA neurons resulted in a ~ 77% reduction in D2R expression compared to the control group injected with shLacZ (Supplementary Fig. 9A-C). Both groups were injected with AAV-hSyn-GRAB_DA2m viruses bilaterally into the CeA and a fiber optic cannula was implanted unilaterally to enable real-time measurement of DA levels during PF-seeking behavior (Supplementary Fig. 9A, D). After one week of NC access, there was no significant difference in the GRAB signal between the shLacZ and shD2R mice (Supplementary fig. 9F, G). Subsequently, mice were randomly assigned to two groups: one with limited PF access (1 hr/day, PF limited) and the other with extended PF access (24 hr/day, PF extended) for two weeks (Supplementary Fig. 9D). In the PF-limited condition, shD2R mice consumed approximately 64% of their daily caloric intake during the 1-hour PF access period, significantly more than shLacZ mice, despite similar overall daily caloric intake between groups. This behavior indicates binge-like eating in shD2R mice with CeA D2R downregulation (Supplementary Fig. 9E). In the PF-extended condition, both groups displayed similar daily caloric intakes, derived almost exclusively from PF consumption (Supplementary Fig. 9E). In terms of body weight, PF-extended shD2R mice showed a slight, non-significant increase compared to shLacZ mice (Supplementary Fig. 9F).

After two weeks of PF exposure, DA release profiles during PF consumption were similar between shLacZ and shD2R mice in the PF-limited condition. However, in the PF-extended condition, shD2R mice exhibited a blunted DA signal in the CeA compared to shLacZ controls (Supplementary Fig. 9I-L), These findings indicate that CeA D2R deficiency leads to attenuated DA release under prolonged PF exposure, emphasizing the importance of CeA D2R in modulating reward-related DA signaling.