A dual-pathway dysregulation in prefrontal-habenular circuits mediates stress susceptibility

Abstract

Chronic stress disrupts emotional homeostasis, leading to resilience or susceptibility through unclear mechanisms. Here, we identify a dual-pathway mechanism in the dorsomedial prefrontal cortex (dmPFC)-lateral habenula (LHb) circuit in male mice, where glutamatergic and GABAergic projections coordinately regulate stress responses. Chronic stress differentially modulates these pathways, inducing susceptibility-specific hyperactivity in dmPFC-innervated LHb neurons by potentiating excitatory and suppressing inhibitory inputs, disrupting long-range excitatory/inhibitory (E/I) balance. Chemogenetic recapitulation of this imbalance exacerbates LHb hyperactivity and stress susceptibility, whereas restoring balance promotes resilience. Mechanistically, stress-susceptible mice exhibit impaired ATP-P2X2R signaling in GABAergic neurons, leading to disinhibition of the excitatory pathway via local microcircuits and weakening of the inhibitory projection, resulting in E/I imbalance. This disruption ultimately causes LHb hyperactivity and susceptibility. Collectively, these results reveal that coordinated regulation of opposing dmPFC-LHb pathways by ATP-P2X2R signaling gates stress susceptibility, providing a dual-pathway dysregulation mechanism for stress-induced maladaptation and a potential therapeutic target for depression.

Chronic stress disrupts emotional homeostasis, leading to resilience or susceptibility through unclear mechanisms. Here, authors show that dysfunction in the coordinated regulation of opposing dmPFC-LHb pathways by ATP-P2X2R signaling gates stress susceptibility, and restoring this balance promotes resilience.

Introduction

Stress serves as an essential mediator of allostasis, a dynamic process enabling organisms to achieve stability through physiological or behavioral adjustments in response to threatening challenges¹. According to the allostasis theory, successful adaptation requires top–down regulation of neural circuits to meet anticipated demands, while excessive “allostatic load” from chronic stress may deplete adaptive capacities, leading to maladaptive states that ultimately compromise organismal survival^2–4. This framework explains the marked heterogeneity in chronic stress responses, which resilient individuals maintain circuit plasticity for homeostatic recalibration, whereas susceptible counterparts develop pathological hyperexcitability in valence-encoding circuits that disrupt emotional homeostasis, ultimately manifesting as depression^3–6. However, the precise neural mechanisms underlying this dichotomy remain unclear.

Among key circuits governing stress responses, the lateral habenula (LHb) has emerged as a critical hub for integrating aversive valence signals and orchestrating stress-related behavioral adaptations^7,8. Substantial body of research indicates that LHb neurons is extensively activated by numerous stressors, with the degree of excitation correlating with negative affective states across species^9–12. Notably, chronic stress drives pathological LHb excitation, and its hyperactivity is consistently linked to depressive phenotypes in both clinical and preclinical studies^7,13,14. Importantly, a recent study further suggests that heightened LHb activity during stress appears to determine individual susceptibility¹⁵, providing a mechanistic substrate for the valence-encoding circuit hyperexcitability that disrupts emotional homeostasis in vulnerable individuals. However, the specific neural circuit mechanisms mediating these effects remain to be elucidated.

The LHb receives convergent inputs from multiple limbic and cortical regions. While prior studies have focused on aversive sensory signals transmitted through subcortical pathways mediating stress responses^7,10,16,17, the role of the medial prefrontal cortex (mPFC), a key hub for top–down regulation of stress adaptation and emotional homeostasis^18–20, in modulating LHb function under chronic stress remains poorly characterized. This highlights the importance of investigating this specific top–down regulatory axis in stress-induced maladaptation. Notably, our preliminary investigations in rodents indicate that neurons within a specific subregion of the mPFC, the dorsomedial prefrontal cortex (dmPFC), establish a functionally distinct projection to the LHb that encodes stress valence and its hyperactivity correlated with depressive phenotype^21,22. Nevertheless, how chronic stress disrupts this top–down pathway to drive pathological LHb hyperactivity during stress, as well as the molecular mediators underlying this dysregulation, remains a critical gap in understanding individual susceptibility trajectories.

Here, we uncover a dual-pathway dysregulation mechanism that mediates stress susceptibility by disrupting long-range excitatory/inhibitory (E/I) balance in the dmPFC-LHb circuit. Specifically, ATP-P2X2 receptor (P2X2R) signaling acts as a key “circuit orchestrator” of this process, with its deficiency in dmPFC GABAergic neurons simultaneously impairing the inhibitory pathway and amplifying the excitatory pathway via direct and indirect microcircuit mechanisms, respectively. This disruption ultimately heightens stress-induced LHb activity and promotes vulnerability to maladaptive stress responses.

Results

Hyperactivation in dmPFC-innervated LHb neurons drives stress susceptibility

To investigate whether dmPFC-innervated LHb neurons (termed ^dmPFC→LHb neurons) encode stress-related valence and exhibit susceptibility-dependent activity dynamics, we applied three different stress stimuli (foot shock, physical restraint, or social attack) to mice while simultaneously using fiber photometry to map their acute stress responses. Specifically, we expressed the calcium indicator jGCaMP7s in these neurons by injecting the anterograde trans-monosynaptic AAV2/1-Cre virus into the dmPFC and AAV2/9-DIO-jGCaMP7s into the LHb. An optic fiber was implanted above the LHb to enable fiber photometry recordings (Supplementary Fig. 1a). We observed a significant increase in Ca²⁺ activity in ^dmPFC→LHb neurons following stress, confirming their engagement in aversive valence processing (Supplementary Fig. 1b–d). We next asked whether chronic stress alters these neural adaptations linked to individual differences in stress susceptibility. We compared the stress-induced Ca²⁺ activity pre- and post-chronic social defeat stress (CSDS), an established depression model where mice were classified as susceptible (Sus) or resilient (Res) based on social interaction (SI) time²³ (Fig. 1a, b). Pre-CSDS recordings showed no baseline differences in stress-induced Ca²⁺ activity between groups. However, post-CSDS recordings revealed a selective stress-induced hyperactivity in Sus mice, while Res mice exhibited no such changes. Notably, the magnitude of post-CSDS stress-induced Ca²⁺ activity inversely correlated with SI time (Fig. 1c–g and Supplementary Fig. 2a–j), suggesting that maladaptive hyperactivity in ^dmPFC→LHb neurons may serve as a potential neural substrate of stress susceptibility.

Fig. 1. Hyperactivation in dmPFC-innervated LHb neurons drives stress susceptibility.

a, b, h, j, l Schematics of the virus injection and experimental design. b Scheme of the fiber photometry apparatus used to measure Ca²⁺ activity induced by social attack, physical restraint or foot shock. c Scheme of social attack. d–f Heatmaps (d), average responses curves (e), and quantification (f) of Ca²⁺ activity in ^dmPFC→LHb neurons induced by social attack of susceptible (Sus) and resilient (Res) mice before and after CSDS (n = 8 animals each group; p < 0.0001 for Sus group, p = 0.8919 for Res group). g Correlation between the change in area under the curve (AUC) of Ca²⁺ activity induced by social attack and the interaction ratio of Sus mice (r = −0.7176, p = 0.0450). i The statistical chart of depressive-like behaviors in mice that received chemogenetic activation of ^dmPFC→LHb neurons by CNO and the control group (n = 8 animals each group; SIT, p = 0.0024; SPT, p < 0.0001; TST, p = 0.0001; OFT, p = 0.9505). k The statistical chart of depressive-like behaviors in mice that received chronic inhibition of ^dmPFC→LHb neurons by Kir2.1 and the control group (n = 9 animals each group; SIT, p = 0.0026; SPT, p = 0.0006; TST, p = 0.0007; OFT, p = 0.8322). m The statistical chart of depressive-like behaviors in mice that received chemogenetic inhibition of ^dmPFC→LHb neurons by CNO and the control group (n = 8 animals each group; SIT, p = 0.9375 for EGFP group, p < 0.0001 for hM4Di group; SPT, p = 0.0002; TST, p = 0.0005; OFT, p = 0.8199). Scale bars represent 200 μm (a, h, j, l). Blue arrow: onset time of social attack (d, e). For all figures, two-sided unpaired t-test for (i, k, m); two-way RM ANOVA with uncorrected Fisher’s LSD test for (f, m) (SIT); two-sided Pearson correlation test for (g). **p < 0.01; ***p < 0.001; ns no significant difference. Data are presented as mean ± SEM. Error bars (i, k, m) and shaded bands (e) indicate the SEM. The shaded band (g) indicates the 95% confidence interval. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

To determine whether hyperactivity in ^dmPFC→LHb neurons drives stress susceptibility, we first chemogenetically activated them during subthreshold social defeat stress (SSDS), a paradigm that does not elicit persistent behavioral alterations in wild-type mice²⁴. To achieve this, we bilaterally injected AAV2/1-Cre into the dmPFC, followed by injection of AAV2/9-DIO-hM3Dq-EGFP (Cre-dependent excitatory DREADD) or AAV2/9-DIO-EGFP (control) into the bilateral LHb (Fig. 1h). Whole-cell recordings confirmed DREADD efficacy (Supplementary Fig. 3a, b). Following viral expression, both experimental and control groups were intraperitoneally (i.p.) injected with clozapine-N-oxide (CNO, 1 mg/kg) and exposed to SSDS. Compared to controls, chemogenetic activation of the ^dmPFC→LHb neurons significantly facilitated susceptibility to SSDS, as evidenced by reduced SI time, decreased sucrose preference in the sucrose preference test (SPT), and increased immobility in the tail suspension test (TST), as well as decreased center time in the open field test (OFT) without affecting locomotor activity (Fig. 1i and Supplementary Figs. 16a and 17a).

To further determine the necessity of ^dmPFC→LHb neuronal hyperactivity in stress susceptibility, we performed complementary chronic and acute inhibition experiments. First, to test if sustained inhibition of ^dmPFC→LHb neurons confers resilience, we bilaterally injected AAV2/1-Cre into the dmPFC, followed by injection of AAV2/9-DIO-Kir2.1-EGFP (Cre-dependent Kir2.1) to chronically suppress neuronal activity²⁵ or AAV2/9-DIO-EGFP (control) into the LHb (Fig. 1j) and subjected all mice to CSDS. Whole-cell recordings in Kir2.1-expressing ^dmPFC→LHb neurons revealed significantly decreased spike numbers under current step injections, suggesting reduced excitability (Supplementary Fig. 3c, d). Notably, we found that chronic inhibition of ^dmPFC→LHb neurons prevented the CSDS induced susceptibility phenotype, without altering locomotion, establishing a resilient phenotype (Fig. 1k and Supplementary Figs. 16b and 17b). To further explore whether acute inhibition of these neurons after CSDS exposure could convert established susceptibility to resilience, we bilaterally injected AAV2/1-Cre into the dmPFC and AAV2/9-DIO-hM4Di-EGFP (Cre-dependent inhibitory DREADD) or AAV2/9-DIO-EGFP (control) into the LHb (Fig. 1l). Whole-cell recordings validated DREADD functionality (Supplementary Fig. 3e, f). After subjecting the mice to CSDS and susceptibility identification, we administered CNO 30 min before behavioral testing. Interestingly, we found that acute inhibition of ^dmPFC→LHb neurons successfully converted the behavioral phenotype from susceptible to resilient, without affecting locomotor activity (Fig. 1m and Supplementary Figs. 16c and 17c). Together, these results establish a causal link between hyperactivity of ^dmPFC→LHb neurons and stress susceptibility.

Divergent excitatory and inhibitory dmPFC-LHb circuits coordinate stress-induced Ca²⁺ response and valence-specific behaviors

To investigate the circuit-level mechanisms by which the dmPFC-LHb pathway modulates stress susceptibility, we began by systematically dissecting how dmPFC neurons regulate LHb neuronal function. We first mapped the synaptic connectivity between these regions by injecting AAV2/9-hSyn-ChR2-mCherry (encoding blue light-sensitive cation channel Channelrhodopsin-2, ChR2) into the dmPFC and recording light-evoked excitatory and inhibitory postsynaptic currents (eEPSCs and eIPSCs) in acute LHb slices (Fig. 2a and Supplementary Fig. 4a). This revealed a dominant excitatory projection from the dmPFC to the LHb, as 48.39% of LHb neurons exhibited exclusively glutamatergic eEPSCs that were abolished by the application of CNQX (20 μM; AMPA/kainate receptor antagonist)^21,22 (Fig. 2b and Supplementary Fig. 4b). Surprisingly, we identified a GABAergic component within the dmPFC-LHb pathway. Specifically, 12.9% of LHb neurons exhibited mixed responses with both eEPSCs and eIPSCs (the amplitudes of eIPSCs were significantly larger than eEPSCs), while 3.23% showed exclusively eIPSCs (Fig. 2b). The eIPSCs were pharmacologically confirmed as GABAergic through blockade by picrotoxin (PTX, 100 μM) (Supplementary Fig. 4b). This finding was further supported by observation of native GFP fluorescence in GAD67-GFP reporter mice, which showed that 19.8% of LHb-projecting dmPFC neurons (termed dmPFC^→LHb neurons) were GFP-positive (Supplementary Fig. 5a–c). To further determine which GABAergic subtypes LHb-projecting GABAergic dmPFC neurons (termed dmPFC^GABA→LHb neurons) belong to, we retrograde labeling dmPFC^GABA→LHb neurons by injecting AAV2/2-Retro-FLP into the LHb and AAV2/9-Con/Fon-EGFP into the dmPFC of vGAT-Cre mice (Supplementary Fig. 5d), we found that the majority of dmPFC^GABA→LHb neurons co-express cholecystokinin (CCK) (93.06%), about half co-express vasoactive intestinal peptide (VIP) (52.63%), while only a small fraction co-express somatostatin (SST) (15.07%) or parvalbumin (PV) (4.76%), indicating that these inhibitory projection neurons predominantly belong to the CCK-expressing GABAergic subtype in the dmPFC (Supplementary Fig. 5e–l). To determine the connectivity patterns of these distinct excitatory and inhibitory projections, we then selectively expressed ChR2 in dmPFC glutamatergic neurons (termed dmPFC^Glu neurons) or dmPFC GABAergic neurons (termed dmPFC^GABA neurons), respectively, by injecting AAV2/9-DIO-ChR2-mCherry into the dmPFC of Vglut2-Cre mice or vGAT-Cre mice (Fig. 2c). Subsequently, we found that photostimulation of dmPFC^Glu→LHb terminals evoked exclusively eEPSCs and excitation in LHb CaMKⅡα⁺ neurons (Fig. 2d, e). In contrast, activation of dmPFC^GABA→LHb terminals evoked mixed responses or exclusively eIPSCs, effectively inhibiting action potentials in LHb CaMKⅡα⁺ neurons (Fig. 2f, g). All synaptic responses were tetrodotoxin (TTX)-sensitive (1 μM) but recovered with 4-AP (100 μM; a potassium channel blocker that prolongs evoked depolarization²⁶), indicating that both glutamatergic and GABAergic inputs to the LHb are mediated by direct monosynaptic connections (Supplementary Fig. 4c–e). Together, these results demonstrate that the dmPFC-LHb circuit is composed of functionally distinct glutamatergic and GABAergic long-range monosynaptic projection neurons (termed dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons, respectively), which exert opposing effects on the LHb glutamatergic neuronal activity.

Fig. 2. Divergent excitatory and inhibitory dmPFC-LHb circuits coordinate stress responses and valence-specific behaviors.

a, c, h, k Schematics of the virus injection and experimental design. b Left: pie chart illustrates the percentage of LHb neurons that showed light-evoked EPSCs only, IPSCs only, Mix, or no responses with photostimulation of dmPFC^→LHb terminals. Right: absolute amplitude of light-evoked IPSCs and EPSCs in LHb neurons (n = 31 cells from 3 mice). d, f Left: representative images of ChR2-expressing dmPFC^Glu→LHb terminals (d) or dmPFC^GABA→LHb terminals (f). Right: recorded LHb CaMKIIα⁺ neurons projecting from dmPFC^Glu neurons (d) or dmPFC^GABA neurons (f) were filled with biocytin. e, g Left: pie chart illustrates the percentage of LHb neurons that showed light-evoked EPSCs only, IPSCs only, Mix, or no responses with photostimulation of dmPFC^Glu→LHb terminals (e n = 25 cells from 3 mice) or dmPFC^GABA→LHb terminals (g n = 20 cells from 3 mice). Middle: absolute amplitude of light-evoked EPSCs (e) or light-evoked IPSCs (g) in LHb neurons. Right: firing frequency of LHb neurons during photostimulation of dmPFC^Glu→LHb terminals (e n = 6 cells from 3 mice; p = 0.0003 for OFF vs ON, p = 0.0019 for ON vs OFF) or dmPFC^GABA→LHb terminals (g n = 6 cells from 3 mice; p = 0.0205 for OFF vs ON, p = 0.0135 for ON vs OFF). i, j Left: representative images of GCaMP6s-expressing dmPFC^Glu→LHb neurons (i) or dmPFC^GABA→LHb neurons (j) and the location of optic fiber. Right: heatmaps, average response curves, and quantification of Ca²⁺ activity in dmPFC^Glu→LHb neurons (i n = 8 animals, p < 0.0001) or dmPFC^GABA→LHb neurons (j n = 8 animals, p = 0.0078) induced by social attack. l, m Left: representative images of ChR2-expressing dmPFC^Glu→LHb terminals (l) and dmPFC^GABA→LHb terminals (m), and the location of optic fiber. Middle and right: heatmaps of RTPP (middle) and quantification of preference for conditioned chamber (right) accompanying photostimulation of dmPFC^Glu→LHb terminals (l n = 8 animals each group, p = 0.0071) and dmPFC^GABA→LHb terminals (m n = 8 animals each group, p = 0.0227). Scale bars represent 200 μm (a, d, f, i, j, l, m) and 10 μm (d, f). Blue arrow: onset time of social attack (i, j). For all figures, two-sided unpaired t-test for (l, m); two-sided paired t-test for (i); two-sided Wilcoxon signed rank test for (j); one-way RM ANOVA with Sidak’s multiple comparisons test for (e, g). *p < 0.05; **p < 0.01; ***p < 0.001. Data are presented as mean ± SEM. Error bars (l, m) and shaded bands (i, j) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

Then, we explored how this distinct projection populations encode stress-related valence and mediates valence-specific behavioral outcomes. Following retrograde labeling by injecting AAV2/2-Retro-FLP into the LHb and AAV2/9-Con/Fon-GCaMP6s into the dmPFC of Vglut2-Cre or vGAT-Cre mice, both dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons expressed GCaMP6s, allowing stress-induced Ca²⁺ activity to be recorded via an optic fiber implanted above the dmPFC (Fig. 2h). We found that, in response to acute stressors, the activity of both dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons significantly increased upon exposure to stress (Fig. 2i, j and Supplementary Fig. 6), suggesting that both populations are activated by stress and encode stress-related information. In addition, we injected AAV2/9-DIO-ChR2-EGFP or AAV2/9-DIO-EGFP (control) into the dmPFC of Vglut2-Cre or vGAT-Cre mice and conducted optogenetics-coupled real-time place preference (RTPP) assays as previously described²⁷ (Fig. 2k). During the test, one compartment was randomly assigned as the “stimulation compartment”, and whenever a mouse entered the designated area, continuous light pulses (470 nm, 5-ms duration, 20 Hz) were delivered via bilaterally implanted optic fibers targeting the LHb. We found that dmPFC^Glu→LHb ChR2-expressing mice spent significantly less time in the light-on chamber compared to control mice, which exhibited no preference between the light-on and light-off chambers (Fig. 2l), suggesting a negative-valence-specific effect. In contrast, selectively stimulating dmPFC^GABA→LHb terminals induced a strong preference for the light-on chamber (Fig. 2m), indicating positive-valence-specific behaviors. Collectively, along with the electrophysiology results, these findings suggest that the simultaneous activation of dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons in response to stress stimuli may maintain appropriate encoding of valence-related behaviors by balancing LHb neuronal activity.

CSDS induces long-range E/I imbalance in the dmPFC-LHb circuit of susceptible mice

Having established the existence and opposing functional roles of the glutamatergic and GABAergic projections within the dmPFC-LHb pathway, we next aimed to determine whether and how stress susceptibility remodels the encoding of stress-related information in these distinct circuits. We stereotaxically injected AAV2/9-DIO-jGCaMP7s into the dmPFC of Vglut2-Cre or vGAT-Cre mice and monitored stress-induced Ca²⁺ activity at dmPFC^Glu→LHb and dmPFC^GABA→LHb terminals in the LHb pre- and post- CSDS (Fig. 3a and Supplementary Fig. 7a). Strikingly, recordings taken prior to CSDS showed comparable baseline stress-induced Ca²⁺ activity across both Sus and Res mice in both circuits, whereas after CSDS, their Ca²⁺ dynamics diverged dramatically. In Sus mice, dmPFC^Glu→LHb terminals exhibited a significant potentiation of stress-induced Ca²⁺ activity, whereas dmPFC^GABA→LHb terminals in Sus mice displayed significantly attenuated stress-induced activation. In contrast, Res mice maintained stable response profiles in both pathways. Notably, post-CSDS recordings revealed that stress-induced hyperactivity in dmPFC^Glu→LHb terminals inversely correlated with SI time, while reduced activity of dmPFC^GABA→LHb terminals showed a positive correlation with SI time (Fig. 3b–k and Supplementary Fig. 7b–q), indicating an emerging long-range E/I imbalance in the dmPFC–LHb circuit of Sus mice following stress exposure.

Fig. 3. CSDS-induced long-range E/I imbalance in the dmPFC-LHb circuit of susceptible mice.

a, l, o Schematics of the virus injection and experimental design. b, g Representative images of jGCaMP7s-expressing dmPFC^Glu→LHb terminals (b) or dmPFC^GABA→LHb terminals (g) and the location of optic fiber. c–e, h–j Heatmaps (c, h), average responses curves (d, i), and quantification (e, j) of Ca²⁺ activity in dmPFC^Glu→LHb terminals (c–e) or dmPFC^GABA→LHb terminals (h–j) induced by social attack of Sus and Res mice before and after CSDS (n = 8 animals each group; e p < 0.0001 for Sus group, p = 0.9442 for Res group; j p < 0.0001 for Sus group, p = 0.5348 for Res group). f, k Correlation between the change in AUC of Ca²⁺ activity in dmPFC^Glu→LHb terminals (f r = −0.7199, p = 0.0441) or dmPFC^GABA→LHb terminals (k r = 0.7203, p = 0.0439) induced by social attack and the interaction ratio of Sus mice. m, n Representative traces, frequency and absolute amplitude of sEPSCs (m) and sIPSCs (n) of ^dmPFC→LHb neurons in control (Con), Sus, and Res mice (n = 12 cells from 3 mice each group; m sEPSC frequency, p < 0.0001 for Con vs Sus, p = 0.0001 for Sus vs Res; sEPSC amplitude, p = 0.5975 for Con vs Sus, p = 0.9178 for Sus vs Res; n sIPSC frequency, p < 0.0001 for Con vs Sus, p < 0.0001 for Sus vs Res; sIPSC amplitude, p = 0.5046 for Con vs Sus, p = 0.7672 for Sus vs Res). p, q Representative traces and absolute amplitude of light-evoked EPSCs (p) and IPSCs (q) in LHb neurons with photostimulation of dmPFC^→LHb terminals in Con (n = 32 cells from 3 mice), Sus (n = 38 cells from 3 mice), and Res (n = 31 cells from 3 mice) mice (p p < 0.0001 for Con vs Sus, p = 0.0018 for Sus vs Res; q p = 0.0320 for Con vs Sus, p = 0.0284 for Sus vs Res). r, s Paired-pulse ratio (PPR) of light-evoked EPSCs (eEPSC) (r) in LHb neurons with photostimulation of the dmPFC^→LHb neurons in Con (n = 14 cells from 3 mice), Sus (n = 14 cells from 3 mice), and Res (n = 16 cells from 3 mice) mice, and light-evoked IPSCs (eIPSC) (s) in LHb neurons with photostimulation of the dmPFC^→LHb neurons in Con (n = 9 cells from 3 mice), Sus (n = 9 cells from 3 mice), and Res (n = 9 cells from 3 mice) mice (r 50 ms, p = 0.1286 for Con vs Sus, p = 0.0257 for Sus vs Res; 100 ms, p = 0.0075 for Con vs Sus, p = 0.0319 for Sus vs Res; 150 ms, p = 0.0014 for Con vs Sus, p = 0.0022 for Sus vs Res; s 50 ms, p = 0.0063 for Con vs Sus, p = 0.0320 for Sus vs Res; 100 ms, p = 0.0307 for Con vs Sus, p = 0.0158 for Sus vs Res; 150 ms, p = 0.0458 for Con vs Sus, p = 0.0089 for Sus vs Res). Blue arrow: onset time of social attack (c, d, h, i). ^*,# Indicates Sus mice compared with Res mice (*) or Con mice (#). Scale bars represent 100 μm (b, g). For all figures, ordinary one-way ANOVA with uncorrected Fisher’s LSD test for (m, n); Kruskal–Wallis with uncorrected Dunn’s test for (p, q); two-way RM ANOVA with uncorrected Fisher’s LSD test for (e, j, r, s); two-sided Pearson correlation test for (f, k). ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ^#p < 0.05; ^##p < 0.01; ns no significant difference. Data are presented as mean ± SEM. Error bars (m, n, r, s) and shaded bands (d, i) indicate the SEM. The shaded bands (f, k) indicate the 95% confidence interval. For box plots (p, q), the centre line represents the median, the box bounds represent the 25th and 75th percentiles, and the whiskers represent the minima and maxima. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

To further determine whether these circuit-level E/I shifts were underpinned by synaptic plasticity, we virally expressed EGFP in the ^dmPFC→LHb neurons and conducted electrophysiological recordings of spontaneous EPSCs (sEPSCs) and spontaneous IPSCs (sIPSCs) in these neurons in non-stressed naïve controls, Sus, and Res mice after CSDS (Fig. 3l). This approach revealed an increased frequency of sEPSCs and a reduced frequency of sIPSCs of Sus mice, indicating that an E/I imbalance at the presynaptic level of Sus mice (Fig. 3m, n). Additionally, after expressing ChR2 in dmPFC neurons and recording eEPSCs and eIPSCs in LHb neurons (Fig. 3o), we found that Sus mice showed increased eEPSCs amplitude and decreased eIPSCs amplitude, suggesting significant alterations in synaptic strength (Fig. 3p, q). Furthermore, paired-pulse ratio (PPR) analysis revealed an increased PPR for eEPSCs and a decreased PPR for eIPSCs, indicating enhanced presynaptic release probability at dmPFC^Glu→LHb synapses and reduced presynaptic release probability at dmPFC^GABA→LHb synapses. Importantly, these changes were absent in naïve controls and Res mice, linking synaptic plasticity dysregulation to stress susceptibility (Fig. 3r, s). Together, our results reveal that CSDS remodels stress-related information encoding in Sus mice by inducing a long-range E/I imbalance in the dmPFC-LHb circuit, accompanied by pathway-specific synaptic plasticity dysregulation.

Long-range E/I imbalance in the dmPFC-LHb circuit drives LHb hyperactivity and susceptibility to stress

We next tested whether long-range E/I imbalance in the dmPFC-LHb circuit drives stress-induced LHb hyperactivity. To address this, we employed a dual-viral chemogenetic strategy to induce varying degrees of E/I imbalance through four distinct intervention paradigms. vGAT-Cre mice were stereotaxically injected with one of the following: (Ⅰ) AAV2/9-CaMKⅡα-fDIO-mCherry and AAV2/9-Con/Fon-EGFP (expressing fluorescent reporters only); (Ⅱ) AAV2/9-CaMKⅡα-fDIO-hM3Dq-mCherry (activating dmPFC^Glu→LHb neurons alone); (Ⅲ) AAV2/9-Con/Fon-hM4Di-EGFP (inhibiting dmPFC^GABA→LHb neurons alone); or (Ⅳ) both viruses in combination (simultaneously activating dmPFC^Glu→LHb and inhibiting dmPFC^GABA→LHb neurons) into the dmPFC. All groups received co-injections of AAV2/2-Retro-Flp and AAV2/9-hSyn-jGCaMP7s in the LHb (Fig. 4a, c). The specificity of CaMKⅡα-driven expression was confirmed by immunofluorescence, showing 96.14% of Vglut2⁺ neurons co-localization with CaMKⅡα (Supplementary Fig. 8a–c), and electrophysiological recordings validated that optogenetic activation of CaMKⅡα⁺ terminals evoked only eEPSCs identical to those elicited by Vglut2⁺ terminal activation (Supplementary Fig. 8d, e), while also confirming DREADD functionality (Supplementary Fig. 9a–d). Subsequently, fiber photometry was performed to quantitatively assess LHb neuronal activity in response to acute stressors across the different intervention groups after administering saline or CNO. We found that CNO treatment significantly enhanced stress-induced LHb Ca²⁺ activity in mice subjected to chemogenetic activation of dmPFC^Glu→LHb neurons alone, inhibition of dmPFC^GABA→LHb neurons alone, and simultaneous activation of dmPFC^Glu→LHb neurons along with inhibition of dmPFC^GABA→LHb neurons, compared to saline treatment. Notably, the group with simultaneous activation of dmPFC^Glu→LHb neurons and inhibition of dmPFC^GABA→LHb neurons exhibited an even greater enhancement of LHb responses to stress stimuli, surpassing the effects of either manipulation alone following CNO treatment (1.32-fold relative to activation of dmPFC^Glu→LHb neurons alone; 1.78-fold relative to inhibition of dmPFC^GABA→LHb neurons alone) (Fig. 4b, d). These findings demonstrate that long-range E/I imbalance in the dmPFC-LHb circuit is sufficient to drive stress-induced hyperactivity in the LHb.

Fig. 4. Long-range E/I imbalance in the dmPFC-LHb circuit drives stress-induced LHb hyperactivity and stress susceptibility.

a, e, i, k, m Schematics of the virus injection and experimental design. b, f Heatmaps and average responses curves showing Ca²⁺ activity in LHb neurons induced by social attack in mice that received chemogenetic activation of the dmPFC^Glu→LHb neurons alone (Ⅱ), chemogenetic inhibition of dmPFC^GABA→LHb neurons alone (Ⅲ), and simultaneous chemogenetic activation of dmPFC^Glu→LHb neurons, and chemogenetic inhibition of dmPFC^GABA→LHb neurons (Ⅳ) by CNO and control group (Ⅰ) (b n = 5 animals each group) or mice that received chemogenetic inhibition of the dmPFC^Glu→LHb neurons alone (Ⅵ), chemogenetic activation of dmPFC^GABA→LHb neurons alone (Ⅶ), and simultaneous chemogenetic inhibition of dmPFC^Glu→LHb neurons and chemogenetic activation of dmPFC^GABA→LHb neurons (Ⅷ) by CNO and control group (V) (f n = 5 animals each group). c, g Representative images of jGCaMP7s-expressing LHb neurons and the location of optic fiber. d, h Quantification of Ca²⁺ activity in different experimental groups (n = 5 animals each group; d p < 0.0001 for I-CNO vs IV-CNO, p < 0.0001 for II-CNO vs IV-CNO, p < 0.0001 for III-CNO vs IV-CNO; h p < 0.0001 for V-CNO vs VIII-CNO, p = 0.0206 for VI-CNO vs VIII-CNO, p = 0.0001 for VII-CNO vs VIII-CNO). j The statistical chart of depressive-like behaviors in mice that received chemogenetic activation of the dmPFC^Glu→LHb neurons alone, chemogenetic inhibition of dmPFC^GABA→LHb neurons alone, and simultaneous chemogenetic activation of dmPFC^Glu→LHb neurons and chemogenetic inhibition of dmPFC^GABA→LHb neurons by CNO and the control group (n = 14 animals each group). l The statistical chart of depressive-like behaviors in mice that received chronic inhibition of the dmPFC^Glu→LHb neurons alone by Kir2.1, chronic inhibition of dmPFC^GABA→LHb neurons alone by NaChBac, and simultaneous inhibition of dmPFC^Glu→LHb neurons by Kir2.1 and activation of dmPFC^GABA→LHb neurons by NaChBac, and a control group (n = 15 animals each group). n The statistical chart of depressive-like behaviors in mice that received chemogenetic inhibition of the dmPFC^Glu→LHb neurons alone, chemogenetic activation of dmPFC^GABA→LHb neurons alone, and simultaneous chemogenetic inhibition of dmPFC^Glu→LHb neurons and chemogenetic activation of dmPFC^GABA→LHb neurons by CNO and the control group (n = 16 animals each group). Scale bars represent 200 μm (c, g). Blue arrow: onset time of social attack (b, f). ^*Indicates AUC compared within each group, # indicates AUC compared between different groups. For all figures, ordinary one-way ANOVA with uncorrected Fisher’s LSD test for (j, l, n); two-way RM ANOVA with uncorrected Fisher’s LSD test for (d, h, n) (SIT). ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ^#p < 0.05; ^###p < 0.001; ns no significant difference. Data are presented as mean ± SEM. Error bars (j, l, n) and shaded bands (b, f) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

To further determine whether coordinated suppression of dmPFC^Glu→LHb and enhancement of dmPFC^GABA→LHb neurons could prevent stress-induced responses in the LHb, we used the same viral delivery and fiber photometry recording strategy to test control and three additional intervention groups in vGAT-Cre mice: (Ⅵ) AAV2/9-CaMKⅡα-fDIO-hM4Di-mCherry (inhibiting dmPFC^Glu→LHb neurons alone); (Ⅶ) AAV2/9-Con/Fon-hM3Dq-EGFP (activating dmPFC^GABA→LHb neurons alone); or (Ⅷ) both viruses (simultaneously inhibiting dmPFC^Glu→LHb and activating dmPFC^GABA→LHb neurons) (Fig. 4e, g). Following validation of DREADD functionality (Supplementary Fig. 9e–h), we found that both intervention alone and combined manipulation significantly reduced stress-induced LHb responses following CNO treatment compared to saline treatment. Moreover, the combined manipulation demonstrated a greater reduction in LHb activity than either inhibiting dmPFC^Glu→LHb neurons or activating dmPFC^GABA→LHb neurons alone (1.34-fold relative to activation of dmPFC^Glu→LHb neurons alone; 1.60-fold relative to inhibition of dmPFC^GABA→LHb neurons alone) (Fig. 4f, h), indicating that long-range E/I rebalancing effectively counteracts stress-induced LHb activation. Collectively, these results demonstrated that long-range E/I balance in the dmPFC-LHb circuit is both sufficient and necessary for modulating LHb stress responses.

Given that long-range E/I imbalance in the dmPFC-LHb circuit drives stress-induced LHb hyperactivity, we investigated whether inducing this E/I imbalance contributes to stress susceptibility. We bilaterally injected vGAT-Cre mice with either control virus, AAV2/9-CaMKⅡα-hM3Dq-mCherry (activating dmPFC^Glu→LHb neurons alone), AAV2/9-DIO-hM4Di-EGFP (inhibiting dmPFC^GABA→LHb neurons alone), or both viruses (simultaneously activating dmPFC^Glu→LHb and inhibiting dmPFC^GABA→LHb neurons). This was combined with employed chemogenetic manipulations during SSDS (Fig. 4i). Behavioral analyses revealed that selective activation of dmPFC^Glu→LHb neurons or inhibition of dmPFC^GABA→LHb neurons each significantly promoted stress susceptibility, as shown by reduced SI time, decreased sucrose preference, and increased immobility in the TST, as well as decreased center time in the OFT without affecting locomotion (Fig. 4j and Supplementary Figs. 16d and 17d). Notably, the combined chemogenetic manipulation showed a trend toward exacerbated stress susceptibility beyond what was observed with either intervention alone (Fig. 4j), indicating that long-range E/I imbalance in the dmPFC-LHb circuit during stress exposure biased mice toward a susceptible phenotype.

To evaluate whether restoring long-range E/I balance is essential for stress resilience, we expressed Kir2.1 in dmPFC^Glu→LHb neurons to chronically inhibit them, NaChBac in dmPFC^GABA→LHb neurons to chronically activate them²⁵, or both simultaneously, in vGAT-Cre mice. All mice underwent CSDS following viral expression (Fig. 4k). We found that while either single manipulation attenuated CSDS-induced susceptibility, the combined intervention showed a trend toward providing stronger protection, establishing a resilient phenotype without affecting locomotion (Fig. 4l and Supplementary Figs. 16e and 17e). Furthermore, we tested whether E/I rebalancing could reverse established susceptibility. Using the same viral delivery approach, we expressed hM4Di in dmPFC^Glu→LHb neurons and/or hM3Dq in dmPFC^GABA→LHb neurons in mice (Fig. 4m). After identifying susceptibility following CSDS exposure, we performed chemogenetic manipulations during behavioral testing. Notably, dual interventions robustly shifted susceptible mice toward resilience (Fig. 4n and Supplementary Figs. 16f and 17f).

To distinguish whether stress susceptibility is regulated by direct GABAergic projections from the dmPFC to the LHb or by local inhibitory networks within the dmPFC involving dmPFC^GABA→LHb neurons, we optogenetically inhibited the dmPFC^GABA→LHb pathway during SSDS using dmPFC^GABA-targeted GtACR1 and activated the same pathway in Sus mice during behavioral testing using dmPFC^GABA-targeted ChR2 expression (Supplementary Fig. 10a, c). We found that optogenetic inhibition of dmPFC^GABA→LHb terminals in the LHb during SSDS significantly increased stress susceptibility (Supplementary Figs. 10b, 16g, and 17g). Conversely, optogenetic activation of the dmPFC^GABA→LHb terminals in the LHb successfully converted the behavioral phenotype from susceptible to resilient (Supplementary Figs. 10d, 16h, and 17h). These results suggest that direct GABAergic input from the dmPFC to the LHb plays a critical role in modulating stress susceptibility.

Taken together, these findings suggest that long-range E/I imbalance in the dmPFC-LHb circuit drives stress-induced LHb hyperactivity and stress-susceptibility phenotypes, whereas restoring it, through bidirectional interventions, effectively mitigates LHb stress-induced hyperactivity and promotes resilience to stress.

ATP orchestrates long-range E/I balance in the dmPFC-LHb circuit for emotional homeostasis

The discovery of long-range E/I imbalance in the dmPFC-LHb circuit of Sus mice prompted us to search for a “circuit orchestrator”, an upstream mediator capable of coordinately tuning both glutamatergic and GABAergic pathways. We focused on five classes of neuromodulators, which were previously implicated in depressive-like behaviors and are known to exert differential effects on glutamatergic and GABAergic neuronal activity in the mPFC: dopamine (DA), acetylcholine (ACh), noradrenaline (NE), 5-hydroxytryptamine (5-HT), and adenosine triphosphate (ATP) ^28–34. To explore their involvement in stress susceptibility, we expressed genetically encoded biosensors specific for each neuromodulator in dmPFC neurons and performed fiber photometry to monitor their stress-induced dynamics before and after CSDS (Fig. 5a and Supplementary Fig. 11a). Interestingly, before CSDS, both Sus and Res mice exhibited a comparable increase in fluorescence responses to acute stress across all neuromodulators measured (Fig. 5b and Supplementary Fig. 11b–f). However, following CSDS, Sus mice showed no differences in stress-induced ACh, NE, or 5-HT dynamics compared with Res mice. In contrast, Sus mice exhibited significantly impaired stress-induced DA and ATP signals, whereas Res mice maintained normal DA and ATP responses (Fig. 5b and Supplementary Fig. 11b–f). In addition, we found that astrocytic, but not neuronal, ATP release drives stress-induced ATP dynamics in the dmPFC (Supplementary Fig. 12a–d), and in vivo microdialysis revealed that baseline extracellular ATP levels were significantly lower in Sus mice compared to controls (Supplementary Fig. 12e, f), in line with our previous study^35,36. Together, these findings suggest that deficient stress-induced release of DA and ATP, as well as reduced baseline ATP levels, in the dmPFC represent a hallmark of stress susceptibility.

Fig. 5. ATP orchestrating long-range E/I balance in the dmPFC-LHb circuit for emotional homeostasis.

a, c, e, h, k, m Schematics of the virus injection and experimental design. b Quantification of fluorescent signals of DA4.4, Ach3.0, ATP1.0, NE2m, and 5-HT3.0 in dmPFC neurons induced by social attack in Sus and Res mice before and after CSDS (n = 8 animals for the DA4.4 Res group and n = 5 animals for all other groups; DA4.4, p = 0.0029 for Sus group, p = 0.2251 for Res group; ATP1.0, p = 0.0033 for Sus group, p = 0.8614 for Res group). d Quantification of Ca²⁺ activity induced by social attack in dmPFC^Glu→LHb terminals of mice before and after administration of ACSF, Apyrase, SCH23390, and Sulpiride, or PPADS in the dmPFC (n = 5 animals each group; dmPFC^Glu→LHb, p = 0.0144 for Apyrase group, p = 0.0008 for PPADS group; dmPFC^GABA→LHb, p = 0.0040 for Apyrase group, p = 0.0107 for PPADS group). f, g Average responses curves and quantification of ATP1.0 fluorescent signal induced by social attack in dmPFC^GABA→LHb neurons (f n = 5 animals each group; p = 0.0002 for Sus group, p = 0.7349 for Res group) and dmPFC^Glu→LHb neurons (g n = 5 animals each group; p = 0.9517 for Sus group, p = 0.9297 for Res group) before and after CSDS. i, j Average responses curves and quantification of Ca²⁺ activity induced by social attack in dmPFC^GABA→LHb terminals (i n = 5 animals each group; p = 0.0001) or dmPFC^Glu→LHb terminals (j n = 5 animals each group; p = 0.0380) of mice that received administration of ACSF or ATPγs in the dmPFC. l The statistical chart of depressive-like behaviors in mice that received administration of ACSF or Apyrase in the dmPFC (n = 8 animals each group; SIT, p = 0.0084; SPT, p = 0.0064; TST, p = 0.0060; OFT, p = 0.7734). n The statistical chart of depressive-like behaviors in mice that received simultaneous chemogenetic activation of dmPFC^Glu→LHb and inhibition dmPFC^GABA→LHb neurons by CNO with administration of ACSF or ATPγs and the control group (n = 8 animals each group). Scale bar represents 100 μm (k). Blue arrow: onset time of social attack. For all figures, two-sided unpaired t-test for (l); two-sided paired t-test for (d, i, j); two-sided Wilcoxon signed rank test for (d) (ACSF); ordinary one-way ANOVA with uncorrected Fisher’s LSD test for (n); two-way RM ANOVA with uncorrected Fisher’s LSD test for (b, f, g, n) (SIT). *p < 0.05; **p < 0.01; ***p < 0.001; ns no significant difference. Data are presented as mean ± SEM. Error bars (d, i, j, l, n) and shaded bands (f, g, i, j) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

Next, to determine whether reductions in extracellular DA or ATP levels may serve as the potential link mediating the long-range E/I imbalance observed in the dmPFC-LHb circuit of Sus mice and thus function as a “circuit orchestrator,” we administered artificial cerebrospinal fluid (ACSF) as a baseline control, followed by a cocktail of D1 and D2 receptor antagonists (SCH23390 and Sulpiride) and Apyrase (an ATP-hydrolyzing enzyme) via guide cannulas to acutely block DA signaling and deplete extracellular ATP in the dmPFC, respectively. Stress-induced Ca²⁺ activity in dmPFC^Glu→LHb and dmPFC^GABA→LHb terminals was monitored 30 min post-injection (Fig. 5c and Supplementary Fig. 11g). Intriguingly, while DA receptor blockade caused no significant change, Apyrase administration not only significantly enhanced stress-induced responses in dmPFC^Glu→LHb terminals but also suppressed responses in dmPFC^GABA→LHb terminals compared to baseline controls (Fig. 5d and Supplementary Fig. 11h). In particular, the ATP-activated P2X receptors antagonist isopyridoxal-phosphate-6-azophenyl-2′,4′-disulfonate (PPADS) similarly enhanced stress-induced responses in dmPFC^Glu→LHb terminals and suppressed responses in dmPFC^GABA→LHb terminals compared to baseline control (Fig. 5d and Supplementary Fig. 11h), indicating that it was ATP, not the ATP hydrolysis product adenosine, that mediated these effects on E/I balance in the dmPFC-LHb circuit.

To further investigate whether ATP deficiency is necessary for stress susceptibility-associated E/I imbalance in the dmPFC-LHb circuit, we first assessed stress-induced ATP dynamics in either dmPFC^Glu→LHb or dmPFC^GABA→LHb neurons before and after CSDS. We selectively expressed the ATP biosensor in these neuronal populations by delivering AAV2/2-Retro-FLP into the LHb and AAV2/9-fDIO-ATP1.0 into the dmPFC of Vglut2-Cre or vGAT-Cre mice. After viral expression, stress-induced ATP signals were monitored via fiber photometry in the dmPFC (Fig. 5e). Notably, our results revealed a significant impairment in stress-induced ATP signals in dmPFC^GABA→LHb neurons in Sus mice but not in Res mice (Fig. 5f). In contrast, stress-induced ATP signals in dmPFC^Glu→LHb neurons were unaffected in both Sus and Res mice (Fig. 5g), indicating that reduced ATP release onto dmPFC^GABA→LHb neurons, but not onto dmPFC^Glu→LHb neurons, is associated with stress susceptibility. Given that ATP deficiency appeared specific to the dmPFC^GABA→LHb neurons in susceptible mice, we next asked whether replenishing ATP could restore the observed circuit dysfunction. To this end, we implanted cannulas in the dmPFC of mice before subjecting them to CSDS. Following stress exposure and identifying susceptibility, we administered ACSF and monitored stress-induced activity at both dmPFC^GABA→LHb and dmPFC^Glu→LHb terminals in the LHb to establish baseline measurements. Subsequently, we administered ATPγs (a non-hydrolyzable ATP analog, 50 μM) to restore extracellular ATP levels (Fig. 5h). Remarkably, despite the pathway-specific deficiency, ATPγs administration not only potentiates stress-induced Ca²⁺ activity in dmPFC^GABA→LHb terminals but also attenuates stress-induced Ca²⁺ activity in dmPFC^Glu→LHb, effectively reversing the E/I imbalance in Sus mice (Fig. 5i, j). These results demonstrate that ATP is critical in maintaining the long-range E/I balance by suppressing glutamatergic excitation while amplifying GABAergic inhibition in the dmPFC-LHb circuit.

To elucidate the causal relationship between ATP deficiency-mediated E/I imbalance in the dmPFC-LHb circuit and stress susceptibility, we tested whether reduced ATP levels in the dmPFC during SSDS could promote susceptibility (Fig. 5k). Following administration of Apyrase into the dmPFC via guide cannulas during SSDS, we found that ATP depletion was sufficient to induce susceptibility phenotypes, including reduced SI time, decreased sucrose preference, and increased immobility in the TST, as well as decreased center time in the OFT without affecting locomotion (Fig. 5l and Supplementary Figs. 16i, and 17i). Next, we examined whether increasing extracellular ATP in the dmPFC could convert the behavioral phenotype from susceptible to resilient and whether this effect depended on E/I rebalancing. After bilaterally implanting cannulas into the dmPFC, we expressed hM3Dq in dmPFC^Glu→LHb neurons and hM4Di in dmPFC^GABA→LHb neurons in vGAT-Cre mice using the dual-viral chemogenetic strategy. Following CSDS exposure and susceptibility screening, we performed behavioral testing (Fig. 5m). We found that ATPγs administration effectively reversed stress susceptibility phenotypes of Sus mice compared with ACSF administration. Moreover, this effect was abolished when we chemogenetically induced E/I imbalance (simultaneously activating dmPFC^Glu→LHb and inhibiting dmPFC^GABA→LHb neurons) in the dmPFC-LHb circuit (Fig. 5n and Supplementary Figs. 16j and 17j). Collectively, these findings establish that extracellular ATP levels in the dmPFC maintain emotional homeostasis by coordinating long-range E/I balance in the dmPFC-LHb circuit.

ATP-P2X2R signaling specifically regulates the dmPFC^GABA→LHb pathway, but not the dmPFC^Glu→LHb pathway

How does ATP deficiency impair dmPFC^GABA→LHb neuronal activity? Considering our observation that the P2X receptor antagonist PPADS suppressed stress-induced responses in dmPFC^GABA→LHb terminals (Fig. 5d and Supplementary Fig. 11h), and based on our earlier work demonstrating that ATP enhances GABAergic transmission in the mPFC through P2X2R^21,37, we were therefore prompted to investigate whether impairment of the P2X2R in dmPFC^GABA→LHb neurons underlies the observed neuronal dysfunction. After validating the efficiency of P2X2-shRNA (Supplementary Fig. 13), we bilaterally expressed P2X2-shRNA or scrambled (SC)-shRNA in dmPFC^GABA→LHb neurons and monitored stress-induced Ca²⁺ activity at dmPFC^GABA→LHb terminals following ACSF and ATPγs administration (Fig. 6a). Strikingly, P2X2R knockdown in dmPFC^GABA→LHb neurons not only replicated the attenuated stress-induced responses observed in Sus mice but also reduced the potentiating effect of ATPγs on stress-induced Ca²⁺ activity that was observed in the control groups (Fig. 6b). Additionally, to determine whether the ATP-P2X2R signaling directly enhances neuronal excitability in dmPFC^GABA→LHb neurons, we assessed their intrinsic physiological properties and found that ATPγs administration increased neuronal excitability in dmPFC^GABA→LHb neurons of the control group. In contrast, P2X2R knockdown reduced neuronal excitability and blocked ATPγs-mediated excitability enhancement in dmPFC^GABA→LHb neurons (Fig. 6c). These results demonstrate that ATP increased stress-induced Ca²⁺ activity and neuronal activity through P2X2R in dmPFC^GABA→LHb neurons, and that reduced extracellular ATP levels impair neuronal function by disrupting ATP-P2X2R signaling.

Fig. 6. ATP-P2X2R signaling specifically regulates the dmPFC^GABA→LHb pathway, but not the dmPFC^Glu→LHb pathway.

a, f, h Schematics of the virus injection and experimental design. b, d Average responses curves and quantification of Ca²⁺ activity induced by social attack in dmPFC^GABA→LHb terminals (b) and dmPFC^Glu→LHb terminals (d) in mice that received P2X2R knockdown in dmPFC^GABA→LHb neurons (b) or dmPFC^Glu→LHb neurons (d) and control mice with administration of ACSF or ATPγs in the dmPFC (n = 5 animals each group; b p = 0.0022 for shSC-ACSF vs shSC-ATPγs, p = 0.2624 for shP2X2-ACSF vs shP2X2-ATPγs, p = 0.0458 for shSC-ACSF vs shP2X2-ACSF; d p < 0.0001 for shSC-ACSF vs shSC-ATPγs, p < 0.0001 for shP2X2-ACSF vs shP2X2-ATPγs). The blue arrow indicates the onset time of the social attack. c, e The current-induced action potentials recorded in dmPFC^GABA→LHb neurons (c) and dmPFC^Glu→LHb neurons (e) in mice that received P2X2R knockdown in dmPFC^GABA→LHb neurons (c) or dmPFC^Glu→LHb neurons (e) and control mice with administration of ACSF or ATPγs in the dmPFC (n = 12 cells from 3 mice each group; c 180 pA, p = 0.0234 for shSC-ACSF vs shSC-ATPγs; p = 0.0042 for shSC-ACSF vs shP2X2-ACSF; e 180 pA, p < 0.0001 for shSC-ACSF vs shSC-ATPγs, p = 0.0006 for shP2X2-ACSF vs shP2X2-ATPγs). g, i Proportions of light-evoked postsynaptic responses in dmPFC^Glu→LHb neurons with photostimulation of dmPFC^GABA→LHb neurons (g n = 32 cells from 5 mice) and dmPFC^GABA→LHb neurons with photostimulation of dmPFC^Glu→LHb neurons (i n = 27 cells from 5 mice). Violet: cells expressing mCherry, green: cells expressing EGFP. For all figures, two-way RM ANOVA with Uncorrected Fisher’s LSD test for (b, c, d, e). ^*p < 0.05; ^**p < 0.01; ^***p < 0.001; ns, no significant difference. Data are presented as mean ± SEM. Error bars (c, e) and shaded bands (b, d) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

Although our results showed intact stress-induced ATP signals in dmPFC^Glu→LHb neurons of Sus mice (Fig. 5g), the observation that PPADS enhanced stress-induced responses in dmPFC^Glu→LHb terminals (Fig. 5d), together with a previous study reporting elevated P2X2R expression in mPFC pyramidal neurons of Sus mice decreased glutamatergic synaptic transmission via P2X2-dependent gene regulation³⁸, suggesting that extracellular ATP might also inhibit dmPFC^Glu→LHb neuronal activity via P2X2R. To investigate this, we bilaterally expressed P2X2-shRNA or SC-shRNA in dmPFC^Glu→LHb neurons and monitored stress-induced Ca²⁺ activity at their terminals following ACSF and ATPγS administration. However, P2X2R knockdown in dmPFC^Glu→LHb neurons neither affected baseline stress responses nor altered ATPγs-mediated attenuation of stress-induced Ca²⁺ activity (Fig. 6d). Furthermore, while ATPγs administration suppressed current-evoked action potentials in control dmPFC^Glu→LHb neurons, this modulation persisted following knockdown of P2X2R (Fig. 6e). These results demonstrate that ATP-P2X2R signaling does not directly regulate dmPFC^Glu→LHb neurons, raising the question of whether its suppressive effects might instead depend on local GABAergic inhibition.

ATP-P2X2R signaling indirectly regulates dmPFC^Glu→LHb pathway via a GABAergic neuron-mediated recurrent inhibitory microcircuit

To determine whether the ATP-mediated suppression of dmPFC^Glu→LHb neurons involves local GABAergic microcircuits originating from dmPFC^GABA→LHb neurons, we first examined potential reciprocal connectivity between these neuronal populations. We virally expressed ChR2 in dmPFC^GABA→LHb neurons while labeling dmPFC^Glu→LHb neurons with mCherry (Fig. 6f). Whole-cell recordings from dmPFC^Glu→LHb neurons showed no light-evoked postsynaptic currents, indicating absence of functional GABAergic inputs from dmPFC^GABA→LHb neurons (Fig. 6g). Similarly, recordings from non-EGFP-labeled glutamatergic neurons, representing non-LHb-projecting glutamatergic neurons, showed that 90.32% of these neurons also lacked light-evoked postsynaptic currents, suggesting that local inhibitory inputs from dmPFC^GABA→LHb neurons onto general dmPFC glutamatergic neurons are sparse (Supplementary Fig. 14). In addition, when we expressed ChR2 in dmPFC^Glu→LHb neurons and labeled dmPFC^GABA→LHb neurons with EGFP (Fig. 6h), we similarly observed no light-evoked postsynaptic currents in dmPFC^GABA→LHb neurons (Fig. 6i). These results demonstrate that the ATP-P2X2-mediated modulation of dmPFC^Glu→LHb neurons occurs independently of direct synaptic connections between dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons.

To investigate whether the suppression of dmPFC^Glu→LHb neurons involves local GABAergic microcircuits mediated by dmPFC GABAergic (dmPFC^GABA) neurons, we first assessed the ATP dynamics and stress-induced Ca²⁺ activity of dmPFC^GABA neurons in stress susceptibility. Notably, Sus mice exhibited significant impairments in both stress-induced ATP signals and Ca²⁺ activity in dmPFC^GABA neurons following CSDS (Supplementary Fig. 15a–h). Additionally, P2X2R knockdown in dmPFC^GABA neurons reduced their intrinsic excitability and abolished ATPγs-induced potentiation of neuronal excitability (Supplementary Fig. 15i–j). These results are consistent with the observations in dmPFC^GABA→LHb neurons, suggesting a broad role of ATP-P2X2R signaling in regulating dmPFC^GABA neuronal function. We then characterized the reciprocal connectivity between dmPFC^GABA and dmPFC^Glu→LHb neurons. We first injected AAV2/2-Retro-FLP into the LHb and AAV2/9-DIO-ChR2-EGFP into the dmPFC to selectively express ChR2 in dmPFC^GABA neurons, along with AAV2/9-CaMKⅡα-fDIO-mCherry to label dmPFC^Glu→LHb neurons for subsequent electrophysiological recording in vGAT-Cre mice (Fig. 7a). Remarkably, optogenetic activation of dmPFC^GABA neurons elicited robust monosynaptic eIPSCs in all recorded dmPFC^Glu→LHb neurons, demonstrating direct inhibitory input (Fig. 7b–d). Furthermore, ATPγs perfusion significantly enhanced both the amplitude of these GABAergic inputs and their functional inhibition of dmPFC^Glu→LHb neuronal activity (Fig. 7e–g), suggesting a potential mechanism for ATP-P2X2R signaling-dependent suppression. Intriguingly, we discovered this connectivity was reciprocal. In vGAT-Cre mice, we injected AAV2/2-Retro-FLP into the LHb and AAV2/9-CaMKⅡα-fDIO-ChR2-mCherry into the dmPFC to selectively express ChR2 in dmPFC^Glu→LHb neurons, while co-injecting AAV2/9-DIO-EGFP to label dmPFC^GABA neurons for electrophysiological identification (Fig. 7h). Optogenetic stimulation of the ChR2-expressing dmPFC^Glu→LHb neurons evoked monosynaptic eEPSCs in 76.92% of dmPFC^GABA neurons (Fig. 7i–k). Moreover, bath application of ATPγs potentiated this excitation through depolarization of GABAergic cells (Fig. 7l–o). This bidirectional connectivity suggested the existence of a functional recurrent inhibitory loop. Indeed, when we photostimulated dmPFC^Glu→LHb neurons while recording from the same population, all of the cells exhibited multisynaptic eIPSCs that were enhanced by ATPγs and blocked by CNQX (Fig. 7p–s). Collectively, these data demonstrate that ATP-P2X2R signaling facilitates a functional recurrent inhibitory microcircuit between local dmPFC^GABA neurons and dmPFC^Glu→LHb neurons.

Fig. 7. ATP-P2X2R signaling indirectly regulates dmPFC^Glu→LHb pathway via a GABAergic neuron-mediated recurrent inhibitory microcircuit.

a, h, p, t, v Schematics of the virus injection. b, i Proportions of light-evoked postsynaptic responses in dmPFC^Glu→LHb neurons with photostimulation of dmPFC^GABA neurons (b n = 16 cells from 3 mice each group) or dmPFC^GABA neurons with photostimulation of dmPFC^Glu→LHb neurons (i n = 13 cells from 3 mice each group). c, d, j, k Light-evoked postsynaptic currents recorded from dmPFC^Glu→LHb neurons with photostimulation of dmPFC^GABA neurons (c, d n = 5 cells from 3 mice each group; c p = 0.0070 for ACSF vs ATPγs, p = 0.0003 for ATPγs vs PTX; d p = 0.0002 for ACSF vs TTX, p = 0.0002 for TTX vs TTX + 4-AP) or dmPFC^GABA neurons with photostimulation of dmPFC^Glu→LHb neurons (j, k n = 5 cells from 3 mice each group; j p < 0.0001; k p = 0.0002 for ACSF vs TTX, p = 0.0007 for TTX vs TTX + 4-AP). e–g Optogenetic activation of dmPFC^GABA neurons reduced spontaneous firing in dmPFC^Glu→LHb neurons (e n = 5 cells from 3 mice; p = 0.0033 for OFF vs ON, p = 0.0058 for ON vs OFF), and effects can be enhanced by application of ATPγs (f n = 5 cells from 3 mice; p = 0.0011 for OFF vs ON, p = 0.0005 for ON vs OFF), and comparison of percentage of change of firing frequency between ACSF and ATPγs groups (g p = 0.0079). l–o Optogenetic activation of dmPFC^Glu→LHb neurons enhanced spontaneous firing in dmPFC^GABA neurons (l n = 5 cells from 3 mice; p = 0.0099 for OFF vs ON, p = 0.0099 for ON vs OFF), and effects can be enhanced by application of ATPγs (m n = 5 cells from 3 mice; p = 0.0001 for OFF vs ON, p = 0.0001 for ON vs OFF) and comparison of firing frequency (n p < 0.0001) and RMP (o p = 0.0009) between ACSF and ATPγs groups. q Proportions of light-evoked postsynaptic responses in dmPFC^Glu→LHb neurons with photostimulation of dmPFC^Glu→LHb neurons (n = 15 cells each group from 3 mice). r, s Light-evoked postsynaptic currents recorded from dmPFC^Glu→LHb neurons with photostimulation of dmPFC^Glu→LHb neurons (n = 5 cells from 3 mice each group; r p = 0.0109 for ACSF vs ATPγs, p = 0.0002 for ATPγs vs CNQX; s p = 0.0026 for ACSF vs TTX, p = 0.6595 for TTX vs TTX + 4-AP). u Average responses curves and quantification of Ca²⁺ activity in dmPFC^Glu→LHb terminals in mice that received P2X2R knockdown in dmPFC^GABA neurons and control mice with administration of ACSF or ATPγs in the dmPFC (n = 5 animals each group; p = 0.0003 for shSC-ACSF vs shSC-ATPγs, p = 0.7198 for shP2X2-ACSF vs shP2X2-ATPγs, p = 0.0001 for shSC-ACSF vs shP2X2-ACSF). w The current-induced action potentials recorded in dmPFC^Glu→LHb neurons in mice that received P2X2R knockdown in dmPFC^GABA neurons and control mice with administration of ACSF or ATPγs in the dmPFC (n = 12 cells from 3 mice each group; 180 pA, p = 0.0008 for shSC-ACSF vs shSC-ATPγs; p = 0.0048 for shSC-ACSF vs shP2X2-ACSF). Blue arrow: onset time of social attack. Violet: cells expressing mCherry, green: cells expressing EGFP. For all figures, two-sided unpaired t-test for (n, o); two-sided Mann–Whitney test for (g); two-sided paired t-test for (j); one-way RM ANOVA with Sidak’s multiple comparisons test (c, d, e, f, k, l, m, r, s); two-way RM ANOVA with uncorrected Fisher’s LSD test (u, w). **p < 0.01; ***p < 0.001; ns no significant difference. Data are presented as mean ± SEM. Error bars (g, n, o, w) and shaded bands (u) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

To further examine whether ATP-P2X2R signaling in dmPFC^GABA neurons mediates the suppression of dmPFC^Glu→LHb neurons. We bilaterally expressed P2X2-shRNA or SC-shRNA in dmPFC^GABA neurons and monitored stress-induced Ca²⁺ activity at dmPFC^Glu→LHb terminals following ACSF and ATPγs administration (Fig. 7t). Strikingly, P2X2R knockdown in dmPFC^GABA neurons not only heightened stress-induced Ca²⁺ activity of dmPFC^Glu→LHb terminals but also completely abolished the attenuating effect of ATPγs on Ca²⁺ activity observed in the control group (Fig. 7u). Furthermore, P2X2R knockdown in dmPFC^GABA neurons increased the excitability of dmPFC^Glu→LHb neurons and abolished ATPγs-induced suppression of neuronal excitability (Fig. 7v, w). These findings suggest that ATP-P2X2R signaling in dmPFC^GABA neurons mediate the suppression of dmPFC^Glu→LHb neurons.

Impaired ATP-P2X2R signaling in dmPFC^GABA neurons is sufficient to promote stress susceptibility

The preceding findings establish ATP-P2X2R signaling in dmPFC^GABA neurons as a “circuit orchestrator” in the long-range E/I balance in the dmPFC-LHb circuit, directly potentiating dmPFC^GABA→LHb inhibition while indirectly suppressing dmPFC^Glu→LHb excitation through recurrent GABAergic microcircuits. To determine whether selective disruption of ATP-P2X2R signaling in dmPFC^GABA neurons is sufficient to recapitulate the full spectrum of stress susceptibility phenotypes, we conducted a series of functional and behavioral analysis. We first bilaterally expressed P2X2-shRNA in dmPFC^GABA neurons and GCaMP6s in ^dmPFC→LHb neurons and monitoring stress-induced LHb activity via fiber photometry (Fig. 8a). Strikingly, we found that P2X2R knockdown in dmPFC^GABA neurons was sufficient to recapitulate the stress-induced LHb hyperactivity observed in Sus mice (Fig. 8b), confirming that ATP-P2X2R signaling in these neurons is critical for maintaining balanced LHb activity under stress.

Fig. 8. Impaired ATP-P2X2R signaling in dmPFC GABAergic neurons is sufficient to promote stress susceptibility phenotypes.

a, c, e Schematics of the virus injection and experimental design. b Heatmaps, average responses curves, and quantification of Ca²⁺ activity in ^dmPFC→LHb neurons induced by social attack in mice that received P2X2R knockdown in dmPFC^GABA neurons and control group (n = 5 animals each group; p = 0.0019). d The statistical chart of depressive-like behaviors in mice that received P2X2R knockdown in dmPFC^GABA neurons and control group with administration of ACSF or ATPγs in the dmPFC (n = 9 animals each group). f, g Representative traces, frequency and absolute amplitude of sEPSCs (f) and sIPSCs (g) of ^dmPFC→LHb neurons in mice that received P2X2R knockdown in dmPFC^GABA neurons and control group (n = 12 cells from 3 mice each group; f p = 0.0005 for sEPSC frequency, p = 0.8957 for sEPSC amplitude; g p = 0.0050 for sIPSC frequency, p = 0.9774 for sIPSC amplitude). Scale bars represent 200 μm (c). h Schematic summary of the study’s working model. Blue arrow: onset time of social attack. For all figures, two-sided unpaired t-test (b, f, g); two-sided Mann–Whitney test (g) (amplitude); ordinary one-way ANOVA with uncorrected Fisher’s LSD test (d); Kruskal–Wallis with uncorrected Dunn’s test (d) (SIT). *p < 0.05; **p < 0.01; ***p < 0.001; ns, no significant difference. Data are presented as mean ± SEM. Error bars (b, d, f, g) and shaded bands (b) indicate the SEM. For detailed statistical information, see also Supplementary Table 1. Source data are provided as a Source Data File.

Subsequently, we examined whether this manipulation could promote stress susceptibility and disrupt ATP-mediated protection. Following viral-mediated P2X2R knockdown, mice subjected to SSDS (Fig. 8c). Behavioral analyses showed that P2X2R knockdown mice displayed significant stress susceptibility phenotypes compared to control, without affecting locomotor activity. Furthermore, while ATPγs administration in dmPFC converted susceptible phenotypes to resilient states in control mice, this protective effect was absent in P2X2R knockdown mice (Fig. 8d and Supplementary Figs. 16k and 17k). These results strongly suggest that ATP-P2X2R signaling in dmPFC^GABA neurons is essential for mediating stress resilience.

To determine whether P2X2R knockdown in dmPFC^GABA neurons combined with SSDS exposure could replicate the synaptic plasticity alterations characteristic of stress susceptibility, we recorded sEPSCs and sIPSCs in ^dmPFC→LHb neurons. Electrophysiological analysis revealed that P2X2R knockdown in dmPFC^GABA neurons combined with SSDS exposure significantly increased sEPSC frequency while decreasing sIPSC frequency in ^dmPFC→LHb neurons (Fig. 8e–g). These results suggest that deficiency of ATP-P2X2R signaling in dmPFC^GABA neurons is sufficient to recapitulate stress-induced pathway-specific synaptic remodeling observed in Sus mice after CSDS.

Taken together, these findings demonstrate that intact ATP-P2X2R signaling in dmPFC^GABA neurons is essential for maintaining the long-range E/I balance in the dmPFC-LHb circuit, thereby preventing stress-induced LHb hyperactivity and ultimately preserving emotional homeostasis.

Discussion

While chronic stress is known to differentially impact resilience vs susceptibility through valence-encoding circuit dysfunction^3,5,7, how top–down prefrontal control mechanisms fail to maintain allostatic balance remains unresolved. Here, we identify a dual dysregulation mechanism in the dmPFC-LHb circuit that underlies stress susceptibility. We demonstrate that ^dmPFC→LHb neurons exhibit susceptibility-specific hyperactivity during stress, causally driving maladaptive behavioral states. Crucially, we reveal two functionally distinct neuronal populations, dmPFC^Glu→LHb and dmPFC^GABA→LHb neurons, that co-activate during acute stress but exert opposing valence encoding on LHb outputs. In susceptible mice, chronic stress disrupts this long-range E/I balance, amplifying glutamatergic drive while attenuating GABAergic inhibition, thereby synergistically promoting LHb hyperactivity. Mechanistically, ATP-P2X2R signaling in dmPFC^GABA neurons serve as a “circuit orchestrator”, which deficiency directly weakening inhibitory pathway while indirectly disinhibiting excitatory pathway via impaired local microcircuits. These findings establish ATP in dmPFC as a key neuromodulator that orchestrates top–down E/I balance in dmPFC-LHb circuit to prevent stress-induced maladaptation (Fig. 8h).

Previous study has established that heightened LHb activity during stress drives susceptibility to depressive-like behaviors¹⁵, yet the upstream regulatory mechanisms controlling this activity remain unclear. Our study identifies a specific subpopulation of LHb neurons receiving inputs from the dmPFC that exhibits stress-induced hyperactivity in susceptible individuals (Fig. 1a–g and Supplementary Fig. 2a–j). Furthermore, we found that activating ^dmPFC→LHb neurons promotes susceptibility, while their inhibition prevents or even reverses stress-induced maladaptive behaviors (Fig. 1h–m). These results suggesting that a deficit in top–down control from the dmPFC gates this aberrant stress encoding in the LHb and drives stress susceptibility. Notably, a recent study identified LHb “starter cells”, a sparse population of LHb neurons receiving lateral hypothalamic inputs that initiate stress propagation within the LHb and mediate depressive-like behaviors³⁹. Future investigations are warranted to explore whether ^dmPFC→LHb neurons overlap with these starter cells, potentially revealing convergent pathways for stress encoding.

While it is well established that long-range projections from the mPFC to distal brain regions predominantly originate from glutamatergic neurons^40–43, including those targeting the LHb^21,22,44, thereby creating an excitatory framework for top–down regulation^45,46, emerging evidence reveals that a subset of GABAergic mPFC neurons also form functional long-range projections capable of information transmission and processing^47–50. Remarkably, our results provide the direct evidence that functionally distinct glutamatergic and GABAergic monosynaptic projections from the dmPFC converge onto the LHb (Fig. 2a–g and Supplementary Fig. 4). Specifically, dmPFC^Glu→LHb neurons activate LHb glutamatergic neurons through the release of glutamate alone, whereas dmPFC^GABA→LHb neurons predominantly co-release both glutamate and GABA, resulting in a net inhibitory effect on LHb glutamatergic neurons. This discovery significantly advances our understanding of long-range GABAergic projections in the neocortex and further supports the notion that the LHb is primarily inhibited by brain regions that send terminals co-releasing glutamate and GABA^17,51–53. Notably, an interesting study on GABA-glutamate co-releasing projections from the ventral tegmental area to the LHb demonstrates that the net inhibitory effect is state-dependent, dictated by the postsynaptic membrane potential of LHb neurons⁵⁴, suggesting that such co-transmission may dynamically regulate LHb activity and highlighting the need for future studies to investigate how LHb activity state influences the net inhibitory effect of dmPFC^GABA→LHb projections. In addition, we found a simultaneous activation of both glutamatergic and GABAergic dmPFC-LHb projections during stress processing (Fig. 2i, j), with activation of these two distinct pathways exerting opposing effects on LHb glutamatergic neurons (Fig. 2e, g) and encodes opposing valence signals as demonstrated by RTPP findings, with glutamatergic excitatory signaling exciting LHb neurons and induces avoidance (negative valence), while GABAergic inhibitory signaling inhibits LHb neurons and induces preference (positive valence) (Fig. 2l, m). These findings demonstrate that the cortical region (dmPFC) exerts top–down control through an “opposing components” circuit motif that implements balanced regulation and coordination to produce an integrated readout of competing valence signals corresponding to negative and positive emotional states⁵⁵. It will be of great interest for future research to explore whether this “opposing components” circuit motif extends to additional cortical target structures and mediates their associated behavioral functions.

Perturbations of E/I balance within local microcircuits of the mPFC, as well as altered connectivity in specific mPFC excitatory output pathways, have been implicated in various pathological affective states^56,57. However, whether chronic stress reprograms long-range E/I balance in specific dual-output circuits remains elusive. Our study identifies a shift in long-range E/I balance toward increased excitation in the dmPFC-LHb circuit as a hallmark of stress susceptibility, characterized by potentiated stress-evoked responses in the dmPFC^Glu→LHb pathway and attenuated response in dmPFC^GABA→LHb pathway (Fig. 3b–k). These findings extend the established significance of mPFC E/I imbalance in pathological affective states from local microcircuits to long-range connectivity. Importantly, we observed pathway-specific alterations in synaptic plasticity in stress-susceptible mice, with the dmPFC^Glu→LHb pathway showing enhanced synaptic efficacy and increased presynaptic release probability while the dmPFC^GABA→LHb pathway exhibited reduced synaptic efficacy and decreased release probability (Fig. 3l–s). Such alterations may drive the long-term establishment of stress susceptibility after chronic stress. While these results are consistent with previous reports of chronic stress-induced strengthening of excitatory inputs and weakening of inhibitory inputs to the LHb from distinct brain regions^10,58–60, this study identifies opposing plasticity changes within two parallel pathways originating from the same upstream cortical region. The observed synaptic modifications may depend on the alterations in presynaptic neuronal excitability within these pathways or on pathway-specific changes in the expression of presynaptic vesicular transporters VGAT and VGLUT2^10,53, highlighting the need for future studies to investigate the molecular mechanisms underlying these coordinated synaptic changes and their functional implications for stress-induced maladaptation.

Although previous study has shown that the mPFC is not the most dominant excitatory input to the LHb¹⁰, our findings reveal that dysregulation of both the dmPFC^Glu→LHb and dmPFC^GABA→LHb pathways collaborates to produce a long-range E/I imbalance that underlies LHb hyperactivity and stress-induced maladaptive behaviors. Using dual-viral manipulations, we demonstrate that inducing E/I imbalance, by simultaneously activating dmPFC^Glu→LHb pathway and inhibiting dmPFC^GABA→LHb pathways, increases stress-induced LHb hyperactivity and stress susceptibility (Fig. 4a–d, i, j). Conversely, restoring balance through coordinated suppression of the dmPFC^Glu→LHb pathway and enhancement of dmPFC^GABA→LHb pathways reduces these LHb hyperactivity and promotes resilience (Fig. 4e–h, k–n). Furthermore, the effects of dual-viral manipulations trended to surpass those of either manipulation alone. These results, together with the RTPP data demonstrating opposing valence encoding by these pathways (Fig. 2k–m), demonstrate that the E/I imbalance observed in susceptible mice biases LHb activity toward a predominance of negative valence signals. They provide direct causal evidence supporting the model suggested by previous studies, in which concurrent changes in inhibitory and excitatory inputs operates in concert to promote LHb hyperactivity and stress-induced maladaptation^59,61,62.

Neuromodulatory systems have been implicated in adjusting circuit-level communication during emotional behavior, with their dysfunction contributing to depressive-like behaviors and disruption of E/I balance^28–34. Interestingly, our findings indicate that among several neuromodulators in the dmPFC, including DA, ACh, NE, 5-HT, and ATP, it is ATP that acts as a “circuit orchestrator,” coordinating the long-range E/I balance in the dmPFC-LHb circuit to maintain emotional homeostasis. Specifically, we demonstrated that: (1) acute stress induced a rapid astrocytic ATP release in the dmPFC (Fig. 5b and Supplementary Fig. 12a–d), with both stress-induced and baseline ATP levels reduced in stress-susceptible mice (Fig. 5b and Supplementary Fig. 12f), consistent with previous studies^35,36; (2) ATP depletion disrupts long-range E/I balance by enhanced stress-induced Ca²⁺ activity in dmPFC^Glu→LHb terminals while suppressing Ca²⁺ activity in dmPFC^GABA→LHb terminals, while administration of ATPγs restores this E/I balance in Sus mice (Fig. 5c, d, h–j); and (3) ATP depletion in the dmPFC promoted stress susceptibility, whereas ATPγs treatment reverses susceptibility phenotypes to resilience, an effect that can be blocked by chemogenetically induced E/I imbalance in the dmPFC-LHb circuit (Fig. 5k–n). These results suggest that maintenance of long-range E/I balance in the dmPFC-LHb circuit depends critically on rapid, stress-induced ATP release in the dmPFC. This astrocytic ATP release may be triggered by circulating corticosterone acting on glucocorticoid receptors (GRs) in astrocytes, leading to Ca²⁺-dependent ATP release, as proposed previously³⁶. Notably, we found that DA levels are also impaired in stress-susceptible mice (Fig. 5b and Supplementary Fig. 11b). However, administration of a cocktail containing D1 and D2 receptor antagonists to block DA signaling in the dmPFC did not significantly alter the long-range E/I balance in the dmPFC-LHb circuit (Fig. 5d and Supplementary Fig. 11h), indicating that DA deficiency in susceptible mice does not appear to disrupt this circuit’s E/I balance. This may be because DA primarily acts on other mPFC projection targets, such as the dorsal periaqueductal gray (dPAG), where its deficiency could contribute to pathological avoidance or excessive fear responses in susceptible mice^63,64. Future studies are needed to further elucidate the specific roles of DA signaling in these alternative pathways and their contribution to stress-related behavioral outcomes.

At a cellular level, we found that ATP enhances stress-induced Ca²⁺ activity, induces membrane depolarization, and elevates neuronal excitability via P2X2R activation in both dmPFC^GABA→LHb neurons and local dmPFC^GABA neurons (Figs. 6b, c and 7l–o, and Supplementary Fig. 15), indicating a broad role for ATP-P2X2R signaling in directly modulating dmPFC GABAergic neuronal function. These findings support our previous studies demonstrating that P2X2R serves as a key mediator of ATP’s regulation of GABAergic synaptic transmission in the mPFC^21,65 and it is plausible that ATP-P2X2R signaling depolarizes the resting membrane potential through Ca²⁺-mediated cationic influx⁶⁶. In contrast, our results show that the increased stress response and neuronal activity observed in dmPFC^Glu→LHb neurons following chronic stress do not appear to directly depend on ATP-P2X2R signaling within these glutamatergic neurons themselves (Fig. 6d, e). Instead, this heightened activity is regulated indirectly through a local inhibitory microcircuit mediated by ATP-P2X2R signaling in dmPFC GABAergic neurons (Fig. 7). Notably, a recent study reported that chronic stress upregulates P2X2R expression in glutamatergic neurons of the mPFC in susceptible mice, leading to reduced excitatory synaptic transmission, which was suggested to be mediated by Ca²⁺-dependent AMPAR trafficking regulated through synaptic-related genes³⁸. This differential modulation of glutamatergic neurons may reflect transcriptional heterogeneity among glutamatergic neurons projecting to different downstream targets^67–69, and may serve to increase the signal-to-noise ratio of dmPFC^Glu→LHb neuronal responses under chronic stress, whereby neural activity conveying meaningful sensory or stress-related information (signal) is amplified relative to spontaneous firing (noise)^29,64. Together, these findings further support the notion of ATP as a homeostatic messenger that mediates cell–cell communication to prevent aberrant hyperexcitability of the nervous system and maintain emotional homeostasis³¹.

We also observed that dmPFC^GABA-specific knockdown of P2X2R fully recapitulated the stress-induced LHb hyperactivity, synaptic E/I imbalance, and behavioral phenotypes observed in susceptible mice following chronic stress (Fig. 8a–g), strongly indicating the critical role of intact ATP-P2X2R signaling in dmPFC^GABA neurons in preserving emotional homeostasis. Given the rapid antidepressant-like and prophylactic effects of ATP^35,65, targeting ATP-P2X2R signaling in dmPFC^GABA neurons represents a promising interventional strategy for the treatment and prevention of depression.

Limitations of the current study

Although our study provides insights into the dual regulation of the dmPFC-LHb circuit underlying stress susceptibility, several limitations should be acknowledged. First, our conclusions are primarily based on the CSDS model, which is widely used for studying depression-related behaviors but represents only one form of stress-induced pathology²³. Future studies utilizing additional stress paradigms are needed to validate and generalize our findings across broader contexts. Second, all experiments were conducted exclusively in male mice due to inherent constraints of the CSDS model, which reliably induces susceptibility primarily in males because of territorial aggression, a key component of the paradigm, is infrequently exhibited toward females⁷⁰. This sex-specific limitation highlights the importance of including female subjects in future investigations using alternative stress models. Third, although we demonstrated the presence of functionally distinct glutamatergic and GABAergic dmPFC projections to the LHb, we could not directly determine whether these projections converge onto the same LHb neurons. This aspect requires further investigation employing more refined circuit-mapping approaches. Addressing these limitations will be important for advancing the mechanistic understanding of our findings.

Methods

Animals

All experiments were approved by the Jinan University Institutional Animal Care and Use Committee (Ethical Approval Number: 20210406-02). Adult (6–8 weeks old) male C57BL/6 mice, vGAT-Cre mice, Vglut2-Cre mice, and GAD-67 EGFP mice were used in this study. The animals were housed in a 12 h:12 h light-dark cycle (lights on at 7 A.M.) with food and water provided ad libitum. The animals were randomly allocated to experimental and control groups. Experimenters were blind to the experimental group, and the order of testing was counterbalanced during behavioral experiments.

Surgery and viral injection

Mice were anesthetized (Avertin, 13 μL/g, i.p.) and placed in a stereotaxic instrument (RWD, Shenzhen, China). Erythromycin eye ointment was applied to prevent corneal drying and a heat pad (RWD, Shenzhen, China) was used to hold the body temperature at 37 °C. Craniotomy holes were drilled with a microdrill (OmniDrill35, WPI, Sarasota, FL) in the skull above the target regions: dmPFC (AP: +2.45 mm; ML: ±0.3 mm; DV: −1.45 mm), LHb (AP: −1.6 mm; ML: ±0.4 mm; DV: −2.85 mm). Depending on the experimental design, either 80 nL virus (for expression of GCaMP, hM3Dq, hM4Di, Kir2.1, NaChBac, Cre, ChR2, GtACR1, FLP, GRAB_DA, GRAB_Ach, GRAB_ATP, GRAB_NE, and GRAB_5HT) or 100 nL shRNA was injected at a rate of 0.1 μL/min through a glass micropipette and controlled by an injector (Microinjection Syringe Pump, WPI). Following injection, the micropipette was left in place for at least 5 min to allow the injectant to diffuse adequately. Mice were allowed to recover from anesthesia under a heat lamp.

For photometry recording, an optical fiber (200 μm in diameter, numerical aperture = 0.37) was placed 300 μm above the centre of viral injection site, and cemented onto the skull using dental cement. Mice were then individually housed and allowed to recover for at least 1 week.

For optogenetic manipulations, an optical fiber (200 μm in diameter, numerical aperture = 0.22) was placed 300 μm above the centre of viral injection site, and cemented onto the skull using dental cement. Mice were then individually housed and allowed to recover for at least 1 week.

For pharmacological manipulations, guide cannulas (O.D. 0.48 × I.D. 0.34 mm, RWD, Shenzhen, China) were placed 200 μm above the centre of viral injection site. The skull was then covered with dental cement and dummy cannula (O.D. 0.3 × I.D. 0.34 mm, RWD, Shenzhen, China) were inserted into the guides. Mice were then individually housed and allowed to recover for at least 1 week.

Upon completion of all experimental procedures, mice were deeply anesthetized with Avertin (13 μL/g, i.p.) and perfused transcardially first with 0.9% saline, followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate-buffered saline (PBS). The brains were carefully removed and post-fixed in 4% PFA overnight at 4 °C. Afterward, the brains were transferred to 30% sucrose solution and allowed to incubate for a minimum of 48 h until sectioning using a cryostat (CM1900, Leica Microsystems, Bannockburn, IL). A series of 40 μm sections were collected for verification of injection sites. To confirm the presence and localization of fluorescent protein expression resulting from viral constructs, coronal brain sections were examined using a fluorescence microscope (Zeiss, Axioimager Z2 microscope). Only mice with confirmed fluorescent protein expression within the targeted area were included in the data analysis. Mice with virus injections that missed the targeted area, displayed low fluorescence expression within the targeted area or showed extensive spread beyond the targeted area (infection beyond the dmPFC or LHb by approximately 70 μm) were excluded from the study. For viral constructs that did not encode fluorescent proteins (i.e., rAAV2/2-Retro-FLP and AAV2/1-Cre), Alexa Fluor 647-conjugated cholera toxin subunit B (CTB-647, 5 nL/injection) was injected into the dmPFC or LHb. The injection sites were visualized with CTB-647. Only mice with confirmed injection sites were included in the data analysis.

Microdialysis

Twenty-four hours following the CSDS, each mouse was then deeply anesthetized and placed in a stereotaxic instrument (RWD, Shenzhen, China). A guide cannula (CMA/7, CMA/Microdialysis) was implanted into the right dmPFC (AP: +2.45 mm; ML: +0.3 mm; DV: −0.45 mm). A microdialysis probe (CMA/7, membrane length: 1-2 mm, molecular weight cut-off: 6000 Da, outer diameter: 0.24 mm) was inserted through the guide cannula and connected to the syringe pump (CMA 402). The ACSF was continuously perfused through the microdialysis probe at a constant flow rate of 1 µL/min, and sampling was performed 1 h following the insertion of probe. Two samples (60 µL each) were automatically collected from each mouse using the CMA 142 microfraction collector every 60 min over 120 min. To decrease the rate of background ATP hydrolysis, each sample collection tube was pretreated with ARL 67156 (1 µL), and the interstitial fluid ATP levels were measured immediately.

Viruses

AAV2/1-hSyn-Cre (PT-0136); AAV2/9-EF1a-DIO-EGFP (PT-0795); AAV2/9-EF1a-DIO-Kir2.1-EGFP (PT-1401); AAV2/9-Ef1a-DIO-hChR2(H134R)-mCherry (PT-0002); AAV2/9-Ef1a-Con/Fon-GCaMP6s (PT-1553); AAV2/9-Ef1a-DIO-hChR2(H134R)-EGFP (PT-9363); AAV2/9-hSyn-Con/Fon-EGFP (PT-9192); AAV2/9-hSyn-Con/Fon-hM4D(Gi)-EGFP (PT-1572); AAV2/9-hSyn-Con/Fon-hM3D(Gq)-EGFP (PT-2368); AAV2/9-CaMKIIα-fDIO-Kir2.1-mCherry (PT-9366); AAV2/9-Ef1a-Con/Fon-NaChBac-EGFP (PT-9371); AAV2/9-hSyn-GRAB_DA4.4 (PT-1340); AAV2/9-hSyn-GRAB_ATP1.0 (PT-1350); AAV2/9-hSyn-GRAB_5-HT3.0 (PT-4724); AAV2/9-Ef1a-fDIO-GRAB_ATP1.0 (PT-8398); AAV2/1-hSyn-FLP (PT-0341); AAV2/9-CaMKIIα-hChR2(H134R)-mCherry (PT-0297); AAV2/9-Ef1a-DIO-GtACR1-P2A-EGFP (PT-1368) were purchased from BrainVTA, Wuhan. AAV2/9-hSyn-DIO-jGCaMP7s (S0590-9); AAV2/9-hSyn-DIO-hM3D(Gq)-EGFP (S0260-9); AAV2/9-hSyn-DIO-hM4D(Gi)-EGFP (S0286-9); AAV2/9-hSyn-hChR2(H134R)-mCherry (S0165-9); AAV2/2-hSyn-hChR2(H134R)-EGFP (S0283-2); AAV2/9-CaMKIIα-fDIO-mCherry (S1029-9); AAV2/9-CaMKIIα-fDIO-hM3D(Gq)-mCherry (S1397-9); AAV2/9-hSyn-jGCaMP7s (S0589-9); AAV2/9-CaMKIIα-fDIO-hM4D(Gi)-mCherry (S1398-9); rAAV2/2-Retro-CAG-FLEX-Flpo (S0273-2R); AAV2/9-hEf1a-fDIO-MasterRNAi155(P2rx2)-mCherry (WY5312); AAV2/9-hEf1a-fDIO-MasterRNAi155(NC)-mCherry (S1743-9); AAV2/9-CaMKIIα-fDIO-hChR2(H134R)-mCherry (S1399-9); AAV2/9-sEf1a-Con/Fon-hChR2(H134R)-EGFP (S1400-9); AAV2/9-hEF1a-DIO-MasterRNAi155(P2rx2)-mCherry (WY5307); AAV2/9-hEf1a-DIO-MasterRNAi155(NC)-mCherry (S0597-9); AAV2/5-CaMKIIα-GCaMP6m (S0481-5); AAV2/9-hEf1a-fDIO-GCaMP6s (S0352-9); AAV2/9-hEf1a-fDIO-EGFP (S1322-9); AAV2/2-Retro-hSyn-mCherry (S0238-2R); AAV2/9-hSyn-hM4D(Gi)-mCherry (S0279-9) were purchased from TaitoolBioScienceCo, Shanghai. rAAV2/2-Retro-hSyn-FLP (BC-0171); AAV2/9-hSyn-GRAB_ACh3.0 (BC-0269); AAV2/9-hSyn-GRAB_NE2m (BC-0267); AAV2/9-hSyn-DIO-GRAB_ATP1.0 (BC-1522) were purchased from BrainCase, Shenzhen. AAV2/5-GfaABC₁D-IP₃R2-shRNA-mChery (TPAV25121) were purchased from SunBio Biomedical technology CO, Shanghai. The titer of AAV2/1-hSyn-Cre and AAV2/1-hSyn-FLP virus is above 1.1 × 10¹³ GC/mL, while the titer of other viruses ranges from 2 to 5 × 10¹² GC/mL. All viruses were aliquoted and stored at −80 °C.PT03.

Drug infusion

The designer drug CNO (1 mg/kg, i.p.; C0832, Sigma-Aldrich, St Louis, MO) was administrated 30 min before behavioral tests, exposure to SSDS or fiber photometry recording.

For micro-infusions, after implanted guide cannula and recovery, mice received bilateral 0.15 μl infusions of ATPγs (50 µM, A1388, Sigma-Aldrich) or Apyrase (300 U/mL, A6132, Sigma-Aldrich) or PPADS (30 µM, HY-101044, MedChemExpress) or ACSF in the bilateral dmPFC.

For the infusion of cocktail of D1 and D2 receptor antagonists, mice received bilateral 1.5 μg injection of D1 dopamine receptors (D1R) antagonist SCH23390 (D054, Sigma-Aldrich) dissolved in saline and bilateral 2 μg injection of D2 dopamine receptors (D2R) antagonist Sulpiride (HY-B1019, MedChemExpress) dissolved in DMSO.

For all micro-infusions, drugs were infused (0.1 μl per 2.5 min) into each side through the injector cannulae. The injector cannulae were left in place for an additional 5 min to minimize the spread of the drug along the injection track. Behavioral tests were performed 30 min after infusion.

In vitro electrophysiology

Acute brain slice preparation

Mice were deeply anesthetized with isoflurane and coronal sections (250 μm thick) across the dmPFC and LHb were cut under ice-cold ACSF (in mM: 119 NaCl; 2.5 KCl, 1 NaH₂PO₄, 11 glucose, 26.2 NaHCO₃, 2.5 CaCl₂, and 1.3 MgCl₂, at pH 7.4, 290 mOsm) using a vibratome (VT1200S; Leica Microsystems). The brain slices were recovered for 1 h at room temperature in ACSF. After recovery, slices were placed in the recording chamber and continuously perfused with ACSF.

Electrophysiological recordings

For light-evoked postsynaptic currents recording, postsynaptic currents in target cells innervated by ChR2-expressing axon terminals stimulated by 2 ms blue light pulses were recorded. Blue-light-evoked postsynaptic currents were recorded with Cs⁺-based peptide solution (in mM: 130 CsMeSO₃, 10 NaCl, 10 EGTA, 4 MgATP, 0.3 Na₃GTP, 10 HEPES, 290 mOsm, adjust to 7.4 with CsOH). EPSCs and IPSCs were recorded when the membrane potential was holding at −70 mV and 0 mV, respectively. The effects of ATPγs were recorded after 5 min of drug perfusion. ATPγs was applied via perfusion into the bath recording solution after control recording. To verify the nature of the recorded currents, 100 μM picrotoxin (PTX) was added to ACSF to block GABA_A receptors for IPSC validation, and 10 μM CNQX was used to block AMPA/kainate receptors for EPSC validation. To confirm that postsynaptic currents were monosynaptic, 1 µM TTX and 100 µM 4-aminopyridine (4-AP) were applied in ACSF. PPRs were assessed by delivering pairs of light pulses at intervals of 50, 100, and 150 ms, and calculated as the ratio of the second to the first IPSC or EPSC amplitude (IPSC2/IPSC1 or EPSC2/EPSC1) to evaluate presynaptic function.

For assessment of optogenetic effects on neuronal firing, whole-cell current-clamp recordings were performed using a potassium gluconate-based internal solution containing (in mM): 140 potassium gluconate, 5 KCl, 10 HEPES, 0.2 EGTA, 2 MgCl₂, 4 Na₂ATP, 0.3 Na₃GTP, and 10 creatine phosphate. Initially, stable inward optocurrents were obtained in voltage-clamp mode in ACSF with single light pulses. The recording was then switched to current-clamp, holding the membrane potential above the firing threshold for inhibition experiments. A train of 5 Hz blue light pulses was applied to assess effects on action potential firing. The effects of ATPγs were recorded after 5 min of drug perfusion. ATPγs was applied via perfusion into the bath recording solution after control recording.

To measure the effect of CNO on hM3Dq-expressing or hM4Di-expressing neurons in the dmPFC or LHb, recordings were performed in current-clamp mode. After baseline recording, 5 µM CNO (dissolved in ACSF, containing 0.1% DMSO) was bath applied during the recordings.

To measure the excitability of dmPFC neurons, electrodes were filled with K⁺-based peptide solution (130 mM KMeSO₄, 10 mM KCl, 10 mM Na₂-phosphocreatine, 4 mM MgATP, 0.3 mM Na₃GTP, 10 mM HEPES, 290 mM mOsm, adjusted to 7.4 with KOH). Current steps were applied at 20 pA per step with a 0.5 s duration. The effects of ATPγs were recorded after 5 min of drug perfusion. ATPγs was applied via perfusion into the bath recording solution after control recording.

To record sEPSCs and sIPSCs of LHb neurons, electrodes were filled with Cs⁺-based peptide solution (in mM: 130 CsMeSO₃, 10 NaCl, 10 EGTA, 4 MgATP, 0.3 Na₃GTP, 10 HEPES, 290 mOsm, adjust to 7.4 with CsOH). sEPSCs and sIPSCs were recorded when the membrane potential was holding at −70 mV and 0 mV, respectively.

All recordings were performed using a Multiclamp 700B amplifier (Molecular Devices). Traces were low-pass filtered at 2 kHz and digitized at 10 kHz. For light stimulation, light pulses were delivered through digital commands from the Digidata 1550 A and Digital stimulator (PG4000a, Cygnus Technology). The pipette resistance ranged from 4 to 6 MΩ. When stable whole-cell recordings were achieved with an access resistance below 25 MΩ, basic electrophysiological properties were recorded. Offline data analysis was performed using Clampfit 10.0 software (Molecular Devices).

In vivo fiber photometry recording

For individual detection of calcium signals in the dmPFC and LHb as well as fluorescent signals of GRAB_DA, GRAB_Ach, GRAB_ATP, GRAB_NE, and GRAB_5HT in the dmPFC, a fiber photometry system (Triple Color Multichannel Fiber Photometry, QAXK-SS-SC-FPS, Thinker Tech) was used. Purple (405 nm) LED light was bandpass filtered (405/10 nm, model 65133, Edmund Optics), blue (473 nm) LED light (Cree LED) was bandpass filtered (470/25 nm, model 65144, Edmund Optics) and yellow (580 nm) LED light (Cree LED) was bandpass filtered (572/28 nm, model 84100, Edmund Optics). These lights were reflected by a dichroic mirror (model 67079, 495 nm long pass, Edmund Optics) and another dichroic mirror (model 87–282, multi-band filter, Edmund Optics), and then focused using a 20× objective lens (Olympus). Three types of excitation light were modulated at different frequencies to excite the fluorescent light at distinct wavelengths. An optical fiber guided the light between the commutator and the implanted optical fiber cannula. The excitation light power at the tip of the optical fiber was adjusted to 20–30 μW and delivered at 100 Hz. The green fluorescence was bandpass-filtered (525/39 nm, model MF525-39, Thorlabs), and the red fluorescence was bandpass filtered (615/20 nm, model 87753, Edmund Optics), then collected using two photomultiplier tubes (model H10721-210, Hamamatsu). Signals of different frequencies were demodulated, amplified (C7319, Hamamatsu), and low-pass filtered. The fluorescence intensity changes (ΔF/F) were determined by calculating (F − F₀)/F₀, with F₀ representing the averaged baseline fluorescence signal measured 2 s prior to stimulation. Changes in the fluorescence signals that responded to stress stimuli were quantified by calculating the area under the curve (AUC) during the 6-s period following the onset of stress stimuli.

For the physical restraint test (PR), the experimenter immobilized the mice by grasping its body with his left hand and securing its tail with his right hand. Each trial was at least 30 s apart.

For the foot shock test (FS), mice were placed in an acrylic box (25 cm × 25 cm × 40 cm) with a metal grid floor that delivered footshock currents (1 mA, 500 ms). Foot shocks were delivered with inter-trial intervals of 15 s.

For the social attack test (SA), mice were exposed to a novel CD-1 aggressor for 10–15 min, and valid trials were recorded only when attacks occurred. All tests were continuously recorded.

For the quantification of fluorescent signals of heatmaps, each row in the heatmaps represents the average of 5 trials per animal, and that this averaging procedure was applied consistently across all animals and conditions.

Immunofluorescence staining

Mice were anesthetized with Avertin (13 uL/g, i.p.) and subsequently perfused intracardially with 0.9% saline, followed by 4% PFA in PBS. The brains were postfixed in 4% PFA at 4 °C overnight and then dehydrated in 30% sucrose. Coronal slices of 40-μm were collected (CM1950, Leica Microsystems, Bannockburn, IL, USA).

For CaMKIIα labeling, the slices containing dmPFC were placed in 0.1 M PBS containing 10% normal goat serum (Vector Laboratories, Burlingame, CA) and 0.3% Triton X-100 (T8787, Sigma-Aldrich, St Louis, MO) for 1 h before incubation in primary antibody against CaMKIIα (Santa Cruz Biotechnology, sc-13141, 1:50) (36 h at 4 °C). The slices were then incubated with the corresponding secondary antibody at a dilution of 1:500 for 3 h at room temperature: Alexa Fluor® 647 labeled goat-anti-mouse IgG (Abcam, HK, ab150115). Finally, all slices were rinsed in 0.1 M PBS and cover-slipped in an anti-fading aqueous mounting medium with DAPI (EMS, Hatfield, PA). Quantification analysis was performed with ImageJ and Photoshop CC2019 software (Adobe Corp., San Jose, California, USA).

Image acquisition and analysis

Sections were imaged with a Zeiss 700 confocal microscope with 5× or 20× objectives or a 40× oil immersion objective. For three-dimensional reconstruction of virus labeled cells, optical sections were collected at 0.2 mm intervals. Using Image J and Photoshop CS5 (Adobe, San Jose, California, USA), each stack of optical sections was montaged and projected to a 0° X–Y plane and a 90° Y–Z plane to obtain a three-dimensional reconstruction of the cell. Contrast and brightness were adjusted, and the red-green images were converted to magenta-green. Cartoons were drawn with Adobe Illustrator CC 2018.

Behavioral tests and animal models

All behavioral tests were performed in a sound-attenuated chamber during the dark cycle of the mice. All experiments were video-recorded for subsequent offline analysis. In all behavioral tests, mice were allowed to acclimate to the testing room for at least half a day before the tests. During the experiments and quantitative analyses, the experimenter was blind to the group identities.

CSDS

All CD-1 aggressors were screened for aggressive behavior before using. C57BL/6 mice were randomly assigned to control and experimental groups. Experimental mice were exposed to a novel CD-1 aggressor for 10 min daily over 10 consecutive days. Mice were then housed oppositely to a perforated Plexiglas barrier between defeat sessions to enable continuous sensory stimulation without physical contact for 24 h. On the following day, the mice were transferred to a new cage containing another aggressive CD-1 mouse to undergo a new episode of social defeat. This process continued for 10 days. After 10 days, the C57 mice were returned to their home cage and underwent behavioral tests after 24 h. In the social interaction test (SIT), the time ratio spent by C57 mice exploring SI zones in the presence vs absence of aggressive mice was calculated to identify susceptible and resilient mice to depression. Mice with a ratio less than 1 were considered Sus, while those with a ratio of 1 or greater were considered Res.

SSDS

All CD-1 aggressors were screened for aggressive behavior before using. Experimental mice were exposed to a novel CD-1 aggressor for 10 min. After 10 min of physical attack, they were physically separated with a perforated Plexiglas partition, and the C57 mice underwent 10 min of sensory stimulation. After 10 min of sensory stimulation, the C57 mice were returned to their home cage for 10 min, and then they went through a second round of physical attack and sensory stimulation in the home cage of a new CD-1 aggressor. After two defeat bouts, the C57 mice were returned to their home cage and underwent behavioral tests after 24 h.

SIT

Experimental mice were housed individually and subjected to a SIT 24 h later. The experiment utilized a square open box (40 cm × 40 cm × 40 cm) and a cage (10 cm × 6.5 cm × 20 cm) of which each side was densely perforated to house unfamiliar CD-1 mice. During the experiment, an empty cage was placed on one side of the open box, with a rectangular area (14 cm × 24 cm) centered on the region where the CD-1 mouse was placed designated as the SI area. In the first phase, the experimental mice were placed inside the open box, and their activity within the box over 2.5 min was recorded; in the second phase, an unfamiliar CD-1 mouse was placed inside the cage, and the mice were then placed back into the open box, with their activity within the box over 2.5 min recorded again. Data were obtained and analyzed using the VisuTrack animal behavior analysis system. After each trial, the apparatus was cleaned with 70% alcohol to remove olfactory cues. The time ratio spent by C57 mice exploring SI zones in the presence vs absence of aggressive mice was calculated to identify susceptible and resilient mice to depression. Mice with a ratio less than 1 were considered susceptible, while those with a ratio of 1 or greater were considered resilient.

SPT

During the 5-day experimental period, each experimental mouse was housed individually and was placed in a quiet, normal growth environment. Mice were provided with two bottles of plain water for the first 2 days and two bottles of plain water were replaced with two bottles of 2% sucrose solution in the following 2 days. Mice were then water-deprived for 24 h and then exposed to one bottle of 1% sucrose and one bottle of water for 2 h in the dark phase. The positions of the two bottles were switched each hour. The preference for the sucrose solution was calculated as the percentage of sucrose solution consumption relative to total fluid intake.

TST

Experimental mice were suspended by the last third of the tail with tape. One end of the tape was secured to a stand to suspend the mice and the head of the mice was 15 cm above the table surface. Animal behaviors were videotaped for 6 min. The first 2 min are an adaptation phase and the immobile time during the last 4 min was counted offline by an observer blinded to animal treatment.

OFT

Experimental mice were placed in the center of an open-field test chamber (40 cm × 40 cm × 40 cm) in a room with dim light for 10 min. A video camera positioned directly above the chamber was used to track the movement of each animal. During the experiment, the mice were gently placed at the center of the open area and allowed to freely explore for 5 min. The total distance traveled, and the movement trajectories of the mice were analyzed by the VisuTrack animal behavior analysis system. After each experiment, clean the chamber with 70% alcohol to eliminate the effects of odors.

RTPP

The RTPP was performed in the same two-compartment conditioning apparatus (25 cm × 25 cm × 50 cm per chamber). The protocol consisted of two sessions over 2 days. On day 1, a mouse was adapted to the chambers and freely explored the apparatus for 15 min (pretest). On day 2 (RT), one side of the chamber was randomly defined as the conditioning chamber. The 473-nm laser (20 Hz, 5 ms) was delivered once the mouse entered the conditioning chamber. Preference or aversion was calculated by dividing the relative time (in percentage) the mouse spent during RT in the conditioning chamber by the relative time (in percentage) the mouse spent in this chamber during the pretest (RT/pretest ratio).

Optogenetic manipulations

To assess the effects of activating the axonal terminals of dmPFC^Glu or dmPFC^GABA neurons in the LHb on depressive-like behaviors using in vivo optogenetics, optical fibers (NA 0.22) were implanted in the bilateral LHb at a 10-degree angle, two weeks after the injection of AAV2/9-DIO-ChR2-EGFP or AAV2/9-DIO-GtACR1-EGFP in the dmPFC of vGAT-Cre mice or vGlut-Cre mice. Optogenetic stimulation was performed using 473-nm, 10-20 mW laser (5 ms, 20 hz, 79–159 mW/mm² at the fiber tip) for dmPFC projections.

Western blot

Brain tissue homogenates isolated from the dmPFC were prepared, and the protein concentration was quantified with the Microplate BCA Protein Assay Kit (23227, Thermo, USA). The protein samples were then run on 12% SDS-PAGE gels and transferred to PVDF membranes (Millipore, MA, USA). The membranes were blocked with 5% non-fat milk for 1 h and then incubated overnight with a primary antibody (polyclonal rabbit anti-P2X2, 1:1000, ab229151, Abcam, MA, UK; rabbit anti-GAPDH, 1:10,000, #5174, CST, USA) at 4 °C. After washing three times in 0.1% Tween 20 in TBS, antibody binding was detected by incubation with an HRP-conjugated secondary antibody (1:10,000, 111-035-003, Jackson ImmunoResearch, USA) at room temperature for 1 h. Protein levels were evaluated by quantifying the grey density of the Western blot bands with ImageJ software. The density of all samples was normalized to that of the internal control.

Quantification and statistical analysis

Quantification of neurons infected with different viruses

To quantify percentage of LHb-projecting dmPFC neurons co-labeled with GAD67, three GAD67-GFP mice received LHb injection of AAV2/2-Retro-mCherry were used. In each mouse, the number of mCherry-labeled neurons and mCherry/GAD67 double labeled neurons were counted in 4 serial brain sections (40 mm/section) across the dmPFC. The percentage of mCherry/GAD67 double labeled neurons relative to the total number of mCherry-labeled neurons in each mouse was calculated.

To quantify percentage of LHb-projecting glutamatergic dmPFC neurons co-labeled with CaMKIIα, three Vglut2-Cre mice received LHb injection of AAV2/2-Retro-FLP and dmPFC injection of AAV2/9-Con/Fon-EGFP were used. In each mouse, the number of EGFP-labeled neurons and EGFP/CaMKIIα double labeled neurons were counted in 4 serial brain sections (40 mm/section) across the dmPFC. The percentage of EGFP/CaMKIIα double labeled neurons relative to the total number of EGFP-labeled neurons in each mouse was calculated.

Statistics

Statistical analysis was not conducted to predetermine sample sizes. All statistics were analyzed using GraphPad Prism 10 software. Researchers who were blinded to the experimental circumstances conducted the data analysis. The Figures and Figure legends contain statistical information, such as definitions and precise values of n (for example, the number of mice or neurons), p values, and the types of the statistical tests. In order to compare the data from the two groups, the homoscedasticity and normality of the distributions of data were determined before assigning specific statistical tests. For data following a normal distribution, two-sided paired and unpaired t-tests were utilized. For data that did not follow a normal distribution, the two-sided Mann–Whitney test and the Wilcoxon signed rank test were applied. For the analysis of data from three or more groups, one-way RM ANOVA with Sidak’s multiple comparisons test, ordinary one-way ANOVA with uncorrected Fisher’s LSD test, Kruskal–Wallis with uncorrected Dunn’s test, and two-way RM ANOVA with uncorrected Fisher’s LSD test were used. For all figures, the horizontal lines in the dot plots represent the mean. Statistical significance was set at p < 0.05.

Reporting summary

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

Author contributions

S.L., C.R., T.-M.G., and L.H. conceived the study and supervised the project. J.W., X.T., and R.S. performed behavioral assays and surgeries with the help of Y.T., B.L., X.L., C.C., and Y.S. J.W., Y.L., and S.L. performed in vitro electrophysiology. J.W., X.T., and R.S. performed fiber photometry. J.W. and Y.T. performed the immunohistochemistry with the help of C.C. and L.H. J.W. and X.T. processed data with the help of R.S. X.T. drafted the figures. W.H. performed a Western blot. S.L., C.R., T.-M.G., and L.H. wrote the manuscript with the help of R.S., J.W., and Y.L. X.C., X.H., and Q.T. provided technical support.

Peer review

Peer review information

Nature Communications thanks Emine Eren-Koçak, who co-reviewed with Sabahaddin Taha Solakoğlu, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

C.R. discloses support for the research of this work from the Brain Science and Brain-like Intelligence Technology—National Science and Technology Major Project (2021ZD0203100), National Natural Science Foundation of China (32325022, 31922030, 32171010), Guangdong Basic and Applied Basic Research Foundation (2023B1515040010, 2021B1515020035), Science and Technology Program of Guangdong (2018B030334001), and Science and Technology Program of Guangzhou (202007030012). S.L. discloses support for the research of this work from the National Natural Science Foundation of China (82571736, 31900825), the Guangdong Basic and Applied Basic Research Foundation (2023B1515040010, 2024A1515030204), and the Fundamental Research Funds for the Central Universities (21625108). T.-M. G. discloses support for the research of this work from the National Key R&D Program of China (2022ZD0204700, 2021ZD0202704), the National Natural Science Foundation of China (T2394535, 82090032), and the Guangdong-Hong Kong Joint Laboratory for Psychiatric Disorders (2023B1212120004). L.H. discloses support for the research of this work from the National Natural Science Foundation of China (32171009), Guangdong Basic and Applied Basic Research Foundation (2021B1515020035), Young Elite Scientist Sponsorship Program by CAST (YESS20210181). Q.T. discloses support for the research of this work from the National Natural Science Foundation of China (82572925). X.H. discloses support for the research of this work from the National Natural Science Foundation of China (82471263).

Data availability

All data required to evaluate and support the conclusions of this study are included in the main text and supplementary materials. There are no restrictions on data availability in the manuscript. The Source data generated in this study are provided in the Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74754-x.