GABA signaling in NG2 glia mediates empathy-like behavior under observational social defeat

Abstract

Empathy, ranging from emotional contagion to consolation, is central to social cognition. While neural mechanisms of observed pain are well studied, how witnessing trauma affects empathy-related behaviors remains unclear. Using an observational social defeat (OSD) model, we find that OSD-exposed mice display enhanced allogrooming toward defeated conspecifics, indicating increased consolation behavior. Whole-brain cFos mapping and fiber photometry reveal selective activation of medial amygdala (MeA) GABAergic neurons during empathic allogrooming. NG2 glia modulate this behavior via GABA signaling; their specific ablation in the MeA reduces inhibitory synaptic transmission, disinhibiting neighboring GABAergic neurons and increasing allogrooming. Single-cell RNA analysis reveals that GABA signaling originates from Gad1-expressing NG2 glia. Genetic knockout of Gad1 in NG2 glia recapitulates the phenotype. This mechanism requires elevated corticosterone induced by social defeat. Our findings highlight the role of NG2 glia-GABA neuron interactions in promoting prosocial empathy and suggest targeting GABA signaling in NG2 glia as a potential therapeutic strategy for vicarious trauma.

Using an observational social defeat model, here authors show NG2 glia in the medial amygdala of observer mice attenuate GABA synthesis and reduce inhibitory tone onto neighboring GABAergic neurons, thereby driving empathy-like allogrooming toward distressed companions.

Introduction

Empathy, the ability to understand and share the feelings of others, is a fundamental aspect of social interactions and relationships. It encompasses both emotional contagion and higher-level prosocial behaviors, such as offering comfort and aid^1–3. While lower-level emotional resonance such as emotional contagion has been well documented, the neural mechanisms underlying higher-order empathy expressions, such as consolation and targeted helping, remain poorly understood. Recent advances in animal models of observational fear and social pain transmission have shed light on the neural circuits and genetic underpinnings of affective empathy. These studies highlight emotional transmission and learning as essential processes for prosocial behaviors^4–9. Recent findings have shown that animals engage in complex empathy-driven behaviors to assist others, suggesting a more advanced prosocial repertoire^{1–4,10–14}. For instance, the endogenous oxytocin and GABA systems within anterior cingulate cortex are crucial for consolation behavior in mandarin voles¹⁵, and stress-induced allogrooming in mice has been shown to be mediated by tachykinin-expressing GABAergic neurons in the medial amygdala (MeA)². While these findings highlight the neuronal regulation of basic emotional resonance and prosocial behaviors, it remains unclear whether glial cells and the environmental niche also contribute, given that the nervous system operates as an integrated and dynamic network. Specifically, witnessing traumatic events not only intensifies emotional resonance with others’ suffering but may also enhance the motivation to engage in prosocial behaviors, presenting a more complex scenario¹⁶. Therefore, the neural mechanisms driving this heightened affective empathy, particularly in the context of social stress, remain largely unknown.

NG2 glia, also known as oligodendrocyte precursor cells (OPCs), were originally identified for their functions in proliferation and differentiation into myelinated oligodendrocytes^17–19. Recent studies have revealed that NG2 glia play a broader role, including regulating neural circuits and contributing to emotional disorders such as depression and anxiety^20–23. These cells are abundant throughout the adult brain and possess the unique ability to receive direct synaptic inputs from both glutamatergic and GABAergic neurons^24–26. GABAergic neurons are known to play a central role in empathy-related behaviors^2,27,28. Our previous findings have demonstrated that NG2 glia are capable of modulating inhibitory neurotransmission through vascular-released GABA signaling in hippocampus, contributing to anxiety-like behaviors²². However, the precise role of NG2 glia in specific empathy-related behaviors has yet to be fully investigated.

In this study, we developed a mouse model of observational social defeat (OSD) to investigate emotional resonance and empathy-like responses to witnessing social stress. Utilizing whole-brain cFos screening and fiber photometry recording, we observed that GABAergic neurons in the medial amygdala (MeA) were specifically activated during empathy-like allogrooming in OSD-exposed mice. NG2 glia, located in close proximity to these GABAergic neurons, were found to play a key modulatory role in this prosocial behavior. Single-cell RNA sequencing analysis revealed that in OSD-exposed mice, Gad1 expression was downregulated in NG2 glia, leading to a reduction in GABA release near local GABAergic neurons in the MeA. This decrease in GABA release resulted in diminished inhibitory control over MeA GABAergic neurons, thereby enhancing allogrooming behavior in the observers. Conditional knockout of Gad1 in NG2 glia further amplified comforting-targeted allogrooming behavior. Notably, optogenetic activation of NG2 glia mitigated this exaggerated empathy-like behavior, reinforcing the causal role of NG2 glia in regulating prosocial actions. Importantly, we found that social stress exposure in OSD mice led to an elevated corticosterone level in the MeA. This elevated corticosterone level was sensed by NG2 glia, which led to suppressed GABA synthesis. Therefore, our findings highlight a new role for NG2 glia-GABA neuron interactions in regulating empathy-related behaviors, with implications for targeting GABA signaling in NG2 glia as a potential therapeutic strategy in emotional disorders like vicarious trauma.

Results

Increased prosocial allogrooming behavior in OSD-exposed mice following social defeat observation

To investigate the empathic behavior of witnessing violent events in mice, we developed a new mouse model of observational social defeat (OSD). First, two male mice were co-housed for 2 weeks to allow bonding (social support) during the acclimatization period. After 1 hour of baseline interaction (phase 1: Interaction) was recorded, one demonstrator considered as an intruder was introduced into the home cage of a resident CD-1 mouse. This resulted in a typical social defeat behavior indicated by the intruder surrendering, when attacked by the resident CD-1 mouse. During the OSD phase, the observer (OSD-exposed mouse) was separated by a transparent perforated glass partition and observed the whole defeat process of the demonstrator (phase 2: Observation). For the sham control, the observer was only separated and observed a naive conspecific in another partition (Fig. 1a). The partition allowed visual, auditory, and olfactory interactions. Following this, the observer and demonstrator were reunited for behavioral assessment (phase 3: Reunion). To assess whether OSD is effective in changing the emotional responsivity of observers, we quantified distinct behavioral signatures (Fig. 1b). During the OSD task, observers exhibited movement arrest (longer freezing duration) and displayed anxiety-like behaviors, evidenced by increased time in freezing and self-grooming (Fig. 1c–f), compared with sham mice (Fig. 1d, f), which is highly similar to the defeated demonstrators (Supplementary Fig. 1a–d). The heightened vicarious anxiety of the observer was further verified by the open field test (OFT) and the elevated plus maze test (EPM) (Fig. 1g, h). Taken together, these results collectively indicate that observers experienced heightened fear and anxiety, successfully mirroring an emotional contagion from stressed demonstrators in an OSD model.

Fig. 1. Increased prosocial allogrooming behavior in OSD-exposed mice following observational social defeat.

a Schematic of the observational social defeat (OSD) mouse model. b Emotional responsivity (freezing, self-grooming, allogrooming) across phase 1 (interaction), phase 2 (observation), and phase 3 (reunion). c Freezing time in observers during phase 1-3. n = 17 mice, P < 0.001, two-tailed Wilcoxon-matched pairs test. d Total freezing duration of observers during phase 2 compared with that of unstressed (sham) mice. n = 7 for sham mice and 17 for OSD-exposed observer mice, respectively, P < 0.001, two-tailed unpaired Mann-Whitney test. e Self-grooming time in observers during phase 1 vs. phase 3. n = 17 OSD-exposed observer mice, P < 0.001, two-tailed Wilcoxon-matched pairs test. f Total self-grooming duration of observers during phase 3 compared with that of unstressed (sham) partners. n = 7 (Sham) and 17 (OSD), P < 0.001, two-tailed Wilcoxon-matched pairs test. g Open field test showing time spent in the center zone. n = 5 (Sham) and 5 (OSD), P = 0.0018, two-tailed unpaired t-test. h Elevated plus maze showing number of entries into the open arms. n = 8 (Sham) and 6 (OSD), P = 0.0376, two-tailed unpaired t-test. i Heat maps showing cumulative spatial occupancy of observers during phase 3 (left) and probability of allogrooming behavior across phase 3 (right). j Allogrooming time in observers during phase 1 compared to phase 3. n = 17 OSD-exposed observer mice, P < 0.001, two-tailed Wilcoxon-matched pairs test. k Total allogrooming time during phase 3 in OSD-exposed observers compared to sham mice. n = 10 (Sham) and 17 (OSD), P < 0.001, two-tailed unpaired Mann-Whitney test. l Percentage of allogrooming versus non-allogrooming behaviors during phase 3. n = 7 (Sham) and 17 (OSD). m Cumulative distribution of allogrooming time during phase 3. n = 8 (Sham) and 5 (OSD). Data are presented as mean ± SEM.

According to the “Russian doll” model, empathy consists of several layers: the most superficial is empathetic perspective taking, the middle is sympathetic concern, and the deepest is cognitive empathy representing highest and most advanced form of empathy¹. To further explore whether observers engage in prosocial-like behaviors after OSD circumstances, these observer and demonstrator mice were reunited in the homecage (Fig. 1a). Interestingly, while sham mice occasionally exhibited allogrooming behavior toward their unstressed partners, observers showed significantly enhanced prosocial behaviors toward their distressed partners. This was characterized by a stronger preference for huddling (Fig. 1i) and a notable increase in allogrooming, both in duration and proportion (Fig. 1j–l and Supplementary Video 1). This effect was particularly pronounced during the first 20 min of the phase 3 of OSD (Fig. 1m). These findings suggest that the increased allogrooming in observers was likely driven by exposure to the violent scenes.

Given that the partition allows visual, auditory, and olfactory interactions, we investigated whether prosocial behavior in observers requires familiarity with the demonstrators²⁹. We found that observers exhibited enhanced emotional contagion, as evidenced by increased self-grooming when they watched a stranger demonstrator during OSD task. However, they did not show heightened prosocial behavior, as indicated by unchanged allogrooming duration during phase 3 (Supplementary Fig. 2a–d). These results suggest that empathetic behavior is primarily elicited among familiar conspecifics.

Moreover, sensory cues also play a crucial role in this heightened allogrooming behavior. When observers were deprived of visual input during the social defeat phase by covering the partition with a layer of black non-woven fabric, their prosocial allogrooming behavior was completely diminished during phase 3 (Supplementary Fig. 2e–h). This indicates that visual cues from the distressed partner are essential for initiating altruistic allogrooming behavior. In contrast, when observers were deprived of their sense of smell via zinc sulfate application, allogrooming behavior did not differ from that of sham mice (Supplementary Fig. 2i–l). Taken together, these findings suggest that visual stimuli are necessary for initiating altruistic allogrooming behavior, whereas olfactory input is not required.

To further exclude the possibility that increased allogrooming arises from witnessing general social interaction rather than aggressive episodes, we introduced two distinct non-aggressive control conditions. The first, termed observation of social interaction (OSI), involved replacing the aggressive CD-1 resident with a non-aggressive C57BL/6 J mouse, allowing the observer to witness neutral social exploration without any physical conflict (Supplementary Fig. 3a). The second, termed the “non-defeated” control, involved exposing the observer to the standard OSD setup, but the resident CD-1 mouse exhibited low aggression, resulting in no social defeat of the demonstrator cagemate (Supplementary Fig. 3e). Notably, neither OSI nor “non-defeated” observers showed significant increases in freezing behavior during phase 2 (Supplementary Fig. 3b, c, f, g), nor did they exhibit enhanced self-grooming or allogrooming during phase 3 (Supplementary Fig. 3d, h). Moreover, the prosocial response to OSD is not restricted to males: female observers also showed a marked rise in empathic allogrooming (Supplementary Fig. 3i-l), confirming that the behavior is triggered by witnessing violent defeat rather than by any sex-specific or general social factor.

Prosocial allogrooming requires activation of GABAergic neurons in the MeA

To explore the potential mechanisms underlying OSD-induced empathy, we first conducted a brain-wide cFos screen on OSD-exposed observers. Strong cFos activation was observed in the medial amygdala (MeA) and medial preoptic area (MPOA), both of which are known to regulate innate social behaviors^2,30 (Fig. 2a, b). Although the anterior cingulate cortex (ACC) and basolateral amygdala (BLA) have been implicated in pain and fear contagion paradigms^4–6,8,31, no significant activation was detected in these regions following OSD exposure (Fig. 2b). The MeA gates sensory cues into stereotyped mating, fighting or parenting responses, placing it at the heart of social decision-making^2,32–35. In the OSD paradigm this structure was recruited almost exclusively in Vgat⁺ GABAergic, not glutamatergic (Glu⁺) neurons (Supplementary Fig. 4). To determine when these GABAergic or glutamatergic cells operate, we expressed GCaMP6s under the mDlx or VGLUT2^27,36,37 promoter and used fiber photometry to track their activity during phase 3 of OSD (Fig. 2c and Supplementary Fig. 5). We found that significantly enhanced Ca²⁺ responses in GABAergic neurons during each allogrooming event, but not in self-grooming bouts, in OSD-exposed observers within the MeA in phase 3 (Fig. 2d, f). In contrast, glutamatergic neurons exhibited strong increases in Ca²⁺ responses during self-grooming events, but no changes during allogrooming events (Fig. 2e, f). These findings indicate that GABAergic neurons in the MeA are selectively engaged during prosocial allogrooming in OSD-exposed observers, suggesting their potential involvement in mediating social affiliative behaviors during phase 3.

Fig. 2. Prosocial allogrooming requires activation of GABAergic neurons in the MeA.

Representative cFos⁺ images (a) and fold change quantification (b) in brain regions of sham, demonstrator, and OSD-exposed observer mice. MeA and MPOA show significant activation in observers. MeA, medial amygdala; MPOA, medial preoptic area; BMA, basomedial amygdala; Aco, anterior cortical amygdala; NAc, nucleus accumbens; INS, insula; BLA, basolateral amygdala nucleus; HIPP, hippocampus; ACC, anterior cingulate cortex. n = 5 (sham), 3 (demonstrator), 6 (observer), respectively. MeA, P = 0.0043, two-tailed unpaired Mann-Whitney test; MPOA, P = 0.0303, two-tailed unpaired t-test. c Schematic and representative image of AAV injection and fiber photometry in MeA GABAergic (mDlx, n = 4) or glutamatergic neurons (VGLUT2, n = 3). Scale bar, 200 μm. Example heat maps and average traces of Ca²⁺ signals in MeA GABAergic (d) and glutamatergic (e) neurons aligned to allogrooming and self-grooming onset (time 0). f Mean ΔF/F for allogrooming vs. self-grooming. n = 20 allogrooming bouts and 20 self-grooming bouts from 4 mDlx-GCaMP6s mice; n = 9 allogrooming bouts and 27 self-grooming bouts from 3 VGLUT2-GCaMP6s mice, respectively. For mDlx-GCaMP6s, P < 0.001 (allogrooming), two tailed paired t-test; for VGLUT2-GCaMP6s, P < 0.001 (self-grooming), two-tailed Wilcoxon-matched pairs test. NS, not significant. g Schematic of AAV2/9-mDlx-mCherry injection and representative mCherry-labeled GABAergic neuron. h Excitability of MeA GABAergic neurons in sham vs. OSD observers. Representative traces, current-spike relationship, and rheobase. n = 7 cells from 3 sham mice and n = 11 cells from 3 OSD-exposed observer mice, respectively. AP firing numbers, two-tailed unpaired t-test; rheobase, two-tailed unpaired Mann-Whitney test. NS, not significant. mIPSCs (i) and mEPSCs (j) amplitude and frequency in MeA GABAergic neurons. For mIPSCs analysis, n = 21 cells from 5 sham mice and n = 18 cells from 3 OSD-exposed observer mice. mIPSC frequency, P < 0.001, two-tailed unpaired Mann-Whitney test. For mEPSCs analysis, n = 26 cells from 4 sham mice and n = 13 cells from 3 OSD-exposed observer mice. mEPSC frequency, two-tailed unpaired Mann-Whitney test; mEPSC amplitude, two-tailed unpaired t-test. NS, not significant. Data are presented as mean ± SEM.

Next, we performed whole-cell patch-clamp recordings from MeA neurons in acute slices obtained from OSD-exposed observers. Using AAV2/9-mDlx-mCherry to target GABAergic neurons, we unexpectedly found no change in their intrinsic excitability, as shown by an unaltered rheobase (the minimum current to evoke the first spike) and an unmodified firing pattern in response to depolarizing current steps (Fig. 2g, h). Basic membrane properties were likewise unaffected (Supplementary Fig. 6). However, the frequency of miniature inhibitory postsynaptic currents (mIPSCs) was significantly reduced, while the frequency and amplitude of miniature excitatory postsynaptic currents (mEPSCs) remained unaffected (Fig. 2i, j). In contrast, glutamatergic neurons transfected with AAV2/9-VGLUT2-mCherry showed no OSD-induced alterations in intrinsic excitability or synaptic transmission (Supplementary Fig. 7). These data collectively suggest that a decrease in inhibitory synaptic transmission onto GABAergic neurons leads to functional disinhibition of MeA GABAergic synapses, resulting in enhanced activity of GABAergic neurons and the observed increase in allogrooming behavior in OSD-exposed mice.

Optical inhibition of NG2 glia exacerbates enhanced prosocial allogrooming behavior in OSD observers

The decrease in the frequency of mIPSCs onto GABAergic synapses suggests a reduction in presynaptic GABA release probability. Glial cells, including astrocytes and oligodendrocyte lineage cells, are known to regulate the synaptic activity, making them potential candidates for modulating observed GABA release^22,38,39. In particular, recent evidence indicates that NG2 glia not only receive neuronal inputs but also presynaptically regulate the excitation-inhibition balance²². Given their emerging role in modulating inhibitory synaptic transmission, we tested whether NG2 glia contribute to the observed GABAergic neurons' disinhibition in the MeA under the OSD model.

To determine the spatial relationship between NG2 glia and distinct neuronal subtypes in the MeA, we utilized AAV2/9 vectors carrying either mDlx-mCherry or VGLUT2-mCherry to selectively label GABAergic and glutamatergic neurons, respectively²⁷. Following viral expression, we performed immunostaining for NG2 antibody to visualize the distribution of NG2 glia within the MeA with confocal microscopy (Fig. 3a). Quantitative analyses of neuronal territory revealed no significant difference in the overall spatial extent occupied by GABAergic versus glutamatergic neurons (Fig. 3b). However, when assessing the degree of overlap between NG2 glia and these two neuronal populations, we observed a significantly higher overlapping territory between NG2 glia and GABAergic neurons compared to their overlap with glutamatergic neurons (Fig. 3c, d). This result indicates preferential association of NG2 glia with GABAergic neurons but not glutamatergic neurons, which may suggest a unique functional interaction between these glial cells and the inhibitory circuits in the MeA. Moreover, we resolved the fine-scale anatomical coupling of NG2 glia to GABAergic versus glutamatergic synapses by injecting AAV2/9-mDlx-mCherry or AAV2/9-VGLUT2-mCherry into the MeA to label gephyrin-positive inhibitory or PSD95-positive excitatory postsynaptic sites, respectively (Fig. 3e; Supplementary Fig. 8a). Three-dimensional reconstruction showed that NG2 glia processes lay within 0.94 ± 0.17 μm of gephyrin puncta on inhibitory synapses (Fig. 3f, g), whereas their distance to PSD95 puncta on glutamatergic synapses was significantly greater (1.55 ± 0.27 μm; Supplementary Fig. 8b–d). These results suggest that, in the MeA, the structural interaction between NG2 glial processes and inhibitory synapses preferentially occurs at GABAergic neurons rather than glutamatergic neurons.

Fig. 3. Spatial organization of NG2 glia relative to neuronal subtypes and OSD-induced modulation of Ca²⁺ signals in the MeA.

Representative images of immunohistochemistry (a) and summary bar graphs showing the territory size of glutamatergic vs. GABAergic neurons (b), fluorescence overlap between NG2 glia and neuronal territories (c), and a scatter plot depicting the correlation between neuronal territory coverage and NG2 glia fluorescence area (d). n = 91 cells from 3 mice for glutamatergic neurons; n = 90 cells from 3 mice for GABAergic neurons, respectively. For panel c, P < 0.001, two-tailed unpaired Mann-Whitney test. NS, not significant. Simple linear regression was used for the scatter plot in panel d, P < 0.001. Scale bars, 10 μm. e Representative 3D reconstruction images showing close spatial relationship between NG2 glia and GABAergic neurons, with Gephyrin localized at proximal sites on GABAergic neurons. n = 3 mice. Scale bars, left and center, 5 μm; right, 1 μm. Representative images showing close spatial proximity between NG2 glial processes and GABAergic neurons, with Gephyrin localized on GABAergic neurites at apposition sites (f). Yellow line indicates line of interest (g, left). Average line-profile showing relative fluorescence intensity of NG2 glia, GABAergic neuron, and Gephyrin along the line of interest (g, middle). Bar graph quantifying spatial distance between NG2 glia and GABAergic neuron (g, right). n = 21 images from 3 mice. h Schematic of two-photon imaging in NG2-CreER^TM;GCaMP6s mice. i–n. Representative images (i,l) and sample traces (j,m) of Ca²⁺ fluctuations in MeA NG2 glia from sham and OSD-exposed mice. Somatic and process fluctuations are defined. White dotted lines outline territory boundaries (illustrative). Scale bars, 20 µm. k Properties of Ca²⁺ fluctuations in processes. Frequency and amplitude of Ca²⁺ fluctuations per minute per cell of processes: n = 243 events from 3 sham mice and n = 66 events from 3 OSD-exposed observer mice for amplitude, P < 0.001; n = 80 cells from 3 sham mice and n = 23 cells from 3 OSD-exposed observer mice for frequency, two-tailed unpaired Mann-Whitney test. n Properties of Ca²⁺ fluctuations in somata. n = 95 events (sham) and 12 events (OSD) for amplitude; n = 29 cells (sham) and 7 cells (OSD) for frequency, two-tailed unpaired Mann-Whitney test. NS, not significant. Data are presented as mean ± SEM.

To directly examine whether NG2 glia modulate allogrooming behavior following OSD, we employed two-photon calcium imaging to capture the activity of NG2 glia²². Intracellular Ca²⁺ signaling in NG2 glia was evaluated using NG2-CreER^TM;GCaMP6s transgenic mice (Fig. 3h). In sham observers, MeA NG2 glia exhibited ongoing [Ca²⁺]_i fluctuations (Fig. 3i–n). However, under OSD conditions, we observed a remarkable reduction in [Ca²⁺]_i fluctuations, primarily in the NG2 glial processes (Fig. 3i–n and Supplementary Video 2). This suggests that NG2 glia enter an intrinsically quiescent state during OSD.

To confirm this phenomenon, we next applied optogenetic inhibition of NG2 glia using yellow-green light in NG2-CreER^TM;Ai35(RCL-Arch/GFP) mice targeting the MeA (Fig. 4a, b). Quantification of Arch-GFP expression revealed that over 88% of labeled cells were NG2-positive, confirming the specificity of our optogenetic targeting in NG2 glia (Fig. 4c). A significant reduction in intracellular calcium levels in NG2 glia following yellow-green light stimulations was detected (Supplementary Fig. 9), confirming the effective suppression of NG2 glial activity. Next, we investigated the behavioral outcomes of optogenetic inhibition of NG2 glia. We found that observers exposed to OSD exhibited a rapid behavioral response to Arch-induced stimulation in NG2 glia. Specifically, they initiated allogrooming or approached the demonstrator within 5 s of stimulation onset (Fig. 4d). Optogenetic inhibition of NG2 glia during phase 3 significantly enhanced allogrooming in OSD-exposed observers (Fig. 4e–g and Supplementary Video 3), whereas self-grooming duration was unaffected (Fig. 4h). Additionally, Ai35^+/+ mice that had not undergone the OSD paradigm showed no increase in allogrooming events (Supplementary Fig. 10). Notably, optogenetic inhibition of MeA NG2 glia facilitated the early initiation of allogrooming, as evidenced by a significant leftward shift in the cumulative probability distribution of allogrooming onset time (Fig. 4i). This shift indicates an increased likelihood of allogrooming occurring during the early phase of phase 3. To further investigate how NG2 glia regulate local inhibitory circuitry, we performed whole-cell patch-clamp recordings from MeA GABAergic neurons while optogenetically silencing NG2 glia. Light-driven suppression of NG2 glia selectively decreased mIPSC frequency in GABAergic cells, leaving mEPSC amplitude, rheobase, firing rate, and membrane properties unchanged (Supplementary Fig. 11). Under the same conditions, glutamatergic neurons exhibited no detectable alterations in any electrophysiological parameters (Supplementary Fig. 12). This pattern of reduced inhibitory input to GABAergic neurons is consistent with our findings in OSD-exposed mice, suggesting that NG2 glia modulate GABAergic signaling through an activity-dependent mechanism.

Fig. 4. Bidirectional optogenetic control of NG2 glia regulates prosocial allogrooming in OSD-exposed observer mice.

a Schematic of optical inhibition in MeA NG2 glia. Representative images (b) and bar graph (c) showing an 88.18% (n = 210 cells from 5 mice) colocalization between GFP-labeled cells and NG2⁺ cells. Yellow arrows indicate GFP⁺/NG2⁺ double-labeled cells. Scale bar, 20 μm. d Schematic showing time course of allogrooming and approach behaviors in Ai35^+/+ mice with yellow-green light stimulation during phase 3. Three typical events from three different mice. e Example raster plots showing allogrooming and self-grooming behaviors in Ai35^-/- and Ai35^+/+ OSD-exposed observer mice with yellow-green light stimulation during phase 3. Each row represents one animal. f–g Duration and number of allogrooming bouts in Ai35^-/- and Ai35^+/+ OSD-exposed observer mice with yellow-green light stimulation during phase 3. n = 5 per group. Duration, P = 0.0383, two-tailed unpaired t-test; number of bouts, P = 0.0718, two-tailed Wilcoxon-matched pairs test. h Self-grooming duration under same conditions. n = 5 per group. NS, not significant, two-tailed unpaired t-test. i Cumulative distribution of allogrooming time. n = 5 per group. j Schematic of optical activation in MeA NG2 glia. Representative images (k) and bar graph (l) showing a 95.92% (n = 191 cells from 5 mice) colocalization between GFP-labeled cells and Pdgfrα⁺ cells. Yellow arrows indicate eYFP⁺/ Pdgfrα⁺ double-labeled cells. Scale bars, 20  μm. m Schematic showing time course of allogrooming behaviors in Ai32^+/+ mice with blue light stimulation during phase 3. 3 typical events from 3 different mice. n Example raster plots showing allogrooming and self-grooming behaviors in Ai32^-/- and Ai32^+/+ OSD-exposed observer mice with blue light stimulation during phase 3. Each row represents one animal. o–p Duration and number of allogrooming bouts in Ai32^-/- and Ai32^+/+ OSD-exposed observer mice with blue light stimulation during phase 3. n = 5 (Ai32^-/-) and 7 (Ai32⁺^/⁺). Duration, P = 0.0473, two-tailed Wilcoxon-matched pairs test; number of bouts, P = 0.0101, two-tailed unpaired Mann-Whitney test. q Self-grooming duration under same conditions. n = 5 (Ai32^-/-) and 7 (Ai32⁺^/⁺). NS, not significant, two-tailed unpaired t-test. r Cumulative distribution of allogrooming time. n = 5 (Ai32^-/-) and 7 (Ai32⁺^/⁺), two-tailed unpaired t-test. Data are presented as mean ± SEM.

In contrast, optogenetic stimulation of MeA NG2 glia using Pdgfrα-CreER^TM;ChR2(H134R)-eYFP (Ai32) significantly disrupted allogrooming behavior in OSD-exposed mice (Fig. 4j–l). Specifically, this disruption was evident in both the duration and frequency of allogrooming episodes, particularly during the onset of allogrooming (Fig. 4m–p). However, self-grooming duration remained unchanged (Fig. 4q). Notably, photoactivation of MeA NG2 glia decreased the likelihood of allogrooming during the early phase of phase 3. This effect was demonstrated by the cumulative frequency plots of allogrooming onset time, which showed a rightward shift in the cumulative frequency curve (Fig. 4r). This shift indicates that the activation of NG2 glia rapidly suppresses the initiation of allogrooming behavior. Taken together, these findings suggest that NG2 glia in the MeA are not only necessary for regulating allogrooming but also play a fine-tuning role in controlling its initiation and frequency.

Gad1-positive NG2 glia are necessary to orchestrate GABAergic circuits in OSD

To investigate the necessity of MeA NG2 glia in regulating allogrooming behavior, we conditionally depleted NG2 glia with targeted administration of diphtheria toxin (DT) in Pdgfrα-CreER^TM;iDTR mice (Fig. 5a). Cre-mediated recombination was locally induced by targeted administering 4-hydroxytamoxifen (4-OH) at 7 days pre-DT administration. Immunofluorescence staining confirmed that the knockout efficiency exceeded 90% within the MeA (Fig. 5b, c). Mice with NG2 glia ablation showed no changes in locomotion and emotional state during the baseline (Supplementary Fig. 13). Additionally, we observed no changes in the basal grooming levels of observers, including both allogrooming and self-grooming, compared to iDTR^-/- control mice (Fig. 5d–f). However, following OSD, observers with NG2 glia depletion in the MeA exhibited a significant increase in allogrooming behavior, both in duration and frequency, compared to iDTR^-/- sham mice (Fig. 5e). This increase was specific to allogrooming, as self-grooming behavior remained unchanged (Fig. 5f). Importantly, this effect was not due to the phenotype of iDTR mice, as depletion of NG2 glia in the medial preoptic area (MPOA) did not result in significant differences in either self-grooming or allogrooming behaviors between MeA NG2 ablation and iDTR^-/- mice (Supplementary Fig. 14). To exclude the potential confounding role of other glia cells, we tested whether astrocytes, which are constituted of 80% of glial cell population in the brain, contribute to prosocial behavior by selectively depleting them using L-α-aminoadipic acid (L-AAA, 0.5 M 100 nL per side) in the MeA. However, we found no significant effect on either self-grooming or allogrooming in these mice (Supplementary Fig. 15), suggesting that astrocytes in the MeA do not play a major role in regulating empathy-related behaviors under OSD exposure.

Fig. 5. Gad1-positive NG2 glia are necessary to orchestrate GABAergic circuits in OSD.

Schematic of NG2 glia ablation experiment (a), representative images (b) and quantification (c) showing 93.01% reduction of Pdgfrα⁺ NG2 glia in MeA of iDTR^+/+ mice. n = 4 per group. P < 0.001, two-tailed unpaired t-test. Scale bars: 1 mm (b, left) and 100 μm (b, right). d Representative raster plots showing allogrooming and self-grooming in iDTR^-/- and NG2-ablated (iDTR^+/+) mice during phase 3. Each row represents one animal. e Allogrooming duration and number of bouts during phase 3 in sham and OSD-exposed observers with iDTR^-/- or iDTR^+/+ genotypes. n = 5 per group. For duration, sham-iDTR^-/- vs. OSD-iDTR^-/-, P = 0.0079; OSD-iDTR^-/- vs. OSD-iDTR^+/+, P = 0.0051, two-tailed unpaired t-tests. For number of bouts, P = 0.0238, two-tailed unpaired Mann-Whitney test. NS, not significant. f Self-grooming duration during phase 3. n = 5 per group. NS, not significant, two-tailed unpaired t-test. g Representative images of MeA show cFos (green) and Vgat (pink) in iDTR^-/- and iDTR^+/+ mice following OSD. Scale bars, 50 μm. h Quantification of total cFos⁺ neurons in MeA. n = 9 sections from 3 OSD-iDTR^-/- mice and 12 sections from 4 OSD-iDTR^+/+ mice, respectively, P = 0.0056, two-tailed Wilcoxon-matched pairs test. i Quantification of cFos⁺ neurons within Vgat⁺ GABAergic population. n = 9 sections from 3 OSD-iDTR^-/- mice and 12 sections from 4 OSD-iDTR^+/+ mice, respectively, P = 0.0472, two-tailed unpaired Mann-Whitney test. j Schematic of FACS isolation of MeA GFP⁺ NG2 cells. k Representative FACS gating strategy for GFP⁺ NG2 glia from the MeA. l t‑SNE plot of 1,864 GFP-positive cells (n = 3 mice) showing seven clusters; NG2 glia comprise 71.4% of total. m Feature plots showing Gad1 is preferentially expressed in NG2 glia ( ~ 32.9% positive), with negligible Gad2 expression. n Violin plots showing Gad1(+) NG2 glia highly express Vamp2 and Snap25 compared to Gad1(-) NG2 glia. o Gene Ontology (GO) top ten pathways for Gad1(+) NG2 glia. Pathway enrichment analysis was performed using Metascape. Statistical significance was calculated based on the accumulative hypergeometric distribution (one-sided). P-values were adjusted for multiple comparisons using the Benjamini-Hochberg (BH) method to control the false discovery rate (FDR). p Volcano plot of differential gene expression in NG2 glia after OSD. Gad1 expression is significantly downregulated (among 265 differentially expressed genes). Statistical analysis was performed using a two-sided Student’s t-test with equal variance, followed by Benjamini-Hochberg (BH) correction for multiple comparisons. Genes with adjusted P < 0.05 and |log₂FC | > 0.585 (FC > 1.5) were considered significant. q Heat map of differential expression of GABA-related genes in NG2 glia from sham vs. OSD mice. n = 3 mice per group. r RT-qPCR validation of Gad1 reduction in NG2 glia. n = 3 sham mice and 4 OSD-exposed observers, respectively, P = 0.0047, two-tailed unpaired t-test. s Representative confocal images of MeA from sham and OSD mice. Scale bars, 5 µm. t Quantification of Gad1 mRNA fluorescence intensity in NG2⁺ cells. The Y-axis represents the measured intensity value (A.U., Arbitrary Units). n = 9 sections from 3 mice for each group, P = 0.0188, two-tailed unpaired Mann-Whitney test. Data are presented as mean ± SEM.

Giving the aforementioned evidence that NG2 glia are spatially proximate to GABAergic neurons (Fig. 3a–d), we sought to determine how NG2 glia interact with surrounding GABAergic neurons. Interestingly, we found that specific depletion of NG2 glia in the MeA significantly increased the activation of neighboring GABAergic neurons, as evidenced by an increased number of cFos-positive Vgat-labeled GABAergic neurons following OSD (Fig. 5g–i). To further investigate the key factors in NG2 glia modulating GABAergic neurons underlying prosocial allogrooming, we employed fluorescence-activated cell sorting (FACS) to isolate GFP( + ) NG2 cells from the MeA and performed single-cell RNA sequencing (scRNA-seq) in Pdgfrα-creER^TM;mGFP mice (Fig. 5j, k). Immunohistochemistry confirmed efficient GFP expression in NG2 glia, validating targeted recombination (Supplementary Fig. 16). Among the 1864 GFP(+) cells analyzed, t-SNE plot data revealed seven clusters based on cell identity, with NG2 glial cells comprising 71.4% of the population (Fig. 5l). We found that the Gad1 gene (Glutamate decarboxylase 1) was highly expressed in NG2 glia but largely absent in other glial cells. In contrast, Gad2 (Glutamate decarboxylase 2) is barely expressed in any of the clusters (Fig. 5m). Specifically, Gad1 expression was detected in approximately 32.9% of the NG2 glia population, a higher percentage than that observed in the hippocampal CA1 region, as we previously reported (Fig. 5m)²². Notably, Gad1(+) NG2 glia also exhibited higher expression of genes such as Vamp2 and Snap25, which are essential components for GABA synthesis and release machinery (Fig. 5n). Furthermore, Gene ontology comparisons between Gad1(+) NG2 glia and Gad1(-) NG2 glial populations identified unique enrichment in pathways related to synaptic transmission (Fig. 5n, o). Importantly, we also found that this subset of Gad1(+) NG2 glia is associated with social behavior (Fig. 5o).

Next, we performed Smart-seq analysis on NG2 glia sorted by FACS from Pdgfrα-CreER^TM;mGFP mice with and without OSD exposure. Consistent with single-cell RNA sequencing results, we found that NG2 glia in the MeA highly express Gad1 rather than Gad2 mRNA (Supplementary Fig. 17). However, in OSD observers, we found that among 265 differentially expressed genes (DEGs) in MeA NG2 glia, only Gad1 was significantly down-regulated, whereas the expression of GABA-release or transporter genes (Vamp2, Snap25, Slc6a1) remained unchanged (Fig. 5q). RT-qPCR validated a selective, pronounced reduction of Gad1 transcripts in OSD mice (Fig. 5r), and RNAscope in-situ hybridization further confirmed that Gad1 mRNA intensity was specifically lower in NG2⁺ cells of the MeA in OSD observers (Fig. 5s, t). Taken together, these results suggest that a distinct subpopulation of Gad1(+) NG2 glia may contribute to the fine-tuning of GABAergic neurons, leading to enhanced allogrooming behavior in OSD-exposed mice.

Gad1 deficiency in NG2 glia tunes inhibitory GABAergic synapses and enhances allogrooming under OSD

To validate the key role of Gad1(+) NG2 glia in regulating GABAergic circuits, we conducted electrophysiological recordings to directly assess their impact on GABA signaling onto GABAergic neurons. To achieve this, we conditionally knocked out the Gad1 gene by crossing the Pdgfrα-CreER^TM mouse line with the Gad1^lox/lox reporter mouse line (Fig. 6a). Successful knockout of the Gad1 gene in Pdgfrα-CreER^TM;Gad1^lox/lox mice was confirmed by single-cell RT-PCR from whole-cell patched NG2 glial cells, which demonstrated the absence of Gad1 transcripts in recombined NG2 glia (Supplementary Fig. 18). Immunohistochemistry further revealed a marked reduction in Gad1-encoded GAD67 protein expression in these cells within the MeA (from 26.7 ± 2.2% in WT mice, a proportion consistent with our scRNA-seq data, to 3.5 ± 0.8% in Gad1 cKO mice; Fig. 6a). To obtain direct biochemical evidence that NG2 glia release GABA in a Gad1-dependent manner, we used optogenetics in primary NG2 glia cultures that were either Gad1-competent or Gad1-null (Gad1-cKO). HPLC quantification showed that optogenetic stimulation evoked a robust rise in intracellular GABA concentration in NG2 glial cells, whereas no increase was detected in Gad1-cKO cells (Fig. 6b).

Fig. 6. Gad1 deficiency in NG2 glia tunes inhibitory GABAergic synapses and enhances allogrooming in OSD-exposed observer mice.

a Representative images of MeA in wild type and Pdgfrα-CreER^TM;Gad1^lox/lox mice. Scale bar, 20 μm. Bar graph: percentage of Pdgfrα⁺ NG2 glia immunoreactive for GAD67. n = 3 mice per group. P < 0.001, two-tailed Wilcoxon-matched pairs test. b GABA concentrations in primary cultured NG2 glia from WT (n = 3), Pdgfrα-CreER^TM;Ai32 (Ai32^+/+, n = 3), and Pdgfrα-CreER^TM;Ai32;Gad1^lox/lox (Ai32^+/+;Gad1^lox/lox, n = 4) mice after blue light stimulation (HPLC). WT vs. Ai32^+/+, P < 0.001; WT vs. Ai32^+/+;Gad1^lox/lox, P < 0.001; Ai32^+/+ vs. Ai32^+/+; Gad1^lox/lox, P < 0.001, two-tailed unpaired t-tests. c Schematic of AAV2/9-mDlx-mCherry injection in Pdgfrα-CreER^TM;Gad1^lox/lox mice. d Excitability of MeA GABAergic neurons in sham vs. Gad1^lox/lox mice. Representative traces, current-spike relationship, and rheobase. n = 7 cells from 3 sham mice and n = 8 cells from 4 Gad1^lox/lox mice, respectively. NS, not significant, two-tailed unpaired Mann-Whitney tests. e mIPSC amplitude and frequency in MeA GABAergic neurons. n = 21 cells from 5 sham mice and n = 9 cells from 3 Gad1^lox/lox mice. mIPSC frequency, P = 0.0057, two-tailed unpaired Mann-Whitney test; mIPSC amplitude, two-tailed unpaired t-test. NS, not significant. f mEPSC amplitude and frequency. n = 26 cells from 4 sham mice and n = 8 cells from 3 Gad1^lox/lox mice. NS, not significant, two-tailed unpaired t-tests. g Schematic of Gad1 knockout in MeA NG2 glia. h Example raster plots showing allogrooming and self-grooming in control and Gad1^lox/lox mice during phase 3. Each row represents one animal. Allogrooming duration (i) and number of bouts (j) in control vs. Gad1^lox/lox OSD-exposed observer mice during phase 3. n = 3 control mice and 6 Gad1^lox/lox mice, respectively. For duration, P = 0.0476; for number of bouts, P = 0.0216, two-tailed Mann-Whitney test. k Self-grooming duration. n = 3 control mice and 6 Gad1^lox/lox mice, respectively. NS, not significant, two-tailed unpaired t-test. l Schematic of optical activation of NG2 glia after Gad1 knockout within the MeA. m Example raster plots with blue light stimulation during phase 3. Each row represents one animal. Allogrooming duration (n) and number of bouts (o) with blue light stimulation. n = 7 per group. For duration, P < 0.001; for number of bouts, P < 0.001, two-tailed Mann-Whitney tests. p Self-grooming duration with blue light stimulation. n = 7 per group. NS, not significant, two-tailed unpaired t-test. Data are presented as mean ± SEM.

To further compare the intrinsic membrane properties of MeA GABAergic neurons in Gad1 cKO and WT littermates, we injected AAV2/9-mDlx-mCherry into the MeA of Pdgfrα-CreER^TM;Gad1^lox/lox mice and induced local recombination with 4-OH (efficiency validated using Pdgfrα-CreER^TM;mGFP mice, Supplementary Fig. 19). We found that GABAergic neurons in the MeA exhibited unchanged intrinsic excitability, as shown by an unaltered rheobase (the minimum current to evoke the first spike) and an unmodified firing pattern in response to depolarizing current steps (Fig. 6c, d). Basic membrane properties were likewise unaffected (Supplementary Fig. 20). However, the frequency of mIPSCs in GABAergic neurons was halved (Fig. 6e), mirroring the deficit seen after NG2 glia ablation (Supplementary Fig. 21). mEPSC frequency and amplitude were unaltered (Fig. 6f), confirming a selective loss of inhibitory tone that parallels the reduction observed in OSD observers (Fig. 2i).

We next performed behavioral assessments in Gad1^lox/lox mice following OSD exposure (Fig. 6g). We found that Gad1 deficiency in NG2 glia led to a significant increase in allogrooming duration, without altering self-grooming events in OSD-exposed observers (Fig. 6h–k). This indicates that reduced GABA synthesis in Gad1-deficient NG2 glia decreased GABAergic synaptic signaling, resulting in disinhibition of GABAergic neurons in the MeA, which ultimately contributed to heightened prosocial allogrooming in OSD exposed observers. Additionally, in the absence of Gad1-derived GABA release in NG2 glia, optogenetic activation of the MeA NG2 glia failed to prevent the exacerbation of empathic allogrooming in OSD-exposed observers, obtained from Pdgfrα-CreER^TM;Gad1^flox/flox;Ai32 mice (Fig. 6l–p). Taken together, these findings suggest that Gad1 is essential for NG2 glia-derived GABA synthesis, which fine-tunes GABAergic synapses and enhances allogrooming behavior in OSD-exposed mice.

Stress-induced corticosterone suppresses calcium activity of NG2 glia following OSD exposure

Observational exposures to violent events, such as social defeat, are known to trigger significant physiological and psychological stress in the observer⁴⁰. One of the primary hormonal responses to stress is the release of corticosterone (CORT) and norepinephrine (NE), which have widespread effects on the central nervous system, modulating both neuronal and glial cell activity⁴¹. To fully understand how NG2 glia respond to OSD and their role in the fine-tuning of GABAergic synapses, we began by measuring the expression levels of stress-related hormones in medial amygdala (MeA) brain tissues. Indeed, we observed a significant increase in corticosterone levels following OSD (Fig. 7a). In contrast, there was no obvious change in norepinephrine levels after OSD (Fig. 7a). Transcriptomic analysis further revealed that NG2 glia in the MeA express high levels of the glucocorticoid receptor Nr3c1, which encodes both mineralocorticoid (MR) and glucocorticoid (GR) receptors with high affinity for CORT. Notably, Nr3c1-expressing NG2 glia displayed significantly high levels of Gad1 and other genes involved in GABA synthesis and release, including Vamp2, Snap25, Slc6a1 (Fig. 7b, c), implying that corticosterone (CORT) signaling preferentially couples to GABA output in these cells. Indeed, 81.4% of Gad1(+) NG2 glia co-express Nr3c1, and their mean Nr3c1 transcript level is significantly higher than in Gad1(–) NG2 glia (Fig. 7d). Dual loss of Gad1 and Nr3c1 further depletes the entire GABA-synthesis /release gene battery (Fig. 7e), and pathway enrichment uniquely links Nr3c1(+) Gad1(+) NG2 glia to trans-synaptic signaling (Fig. 7f), establishing them as a distinct, stress-responsive inhibitory module within the NG2 glial landscape.

Fig. 7. Stress-induced corticosterone triggers NG2 glia suppression following OSD exposure.

a CORT and NE levels in MeA measured by ELISA. n = 3 (sham) and 4 (OSD). CORT, P = 0.0022, two-tailed unpaired t-test. NS, not significant. b Feature plots showing NG2 glia express high levels of glucocorticoid receptor gene (Nr3c1) and low levels of NE receptor subtypes. c Violin plots of GABA synthesis/release genes in Nr3c1(+) vs. Nr3c1(-) NG2 glia. d Feature plots and pie chart showing Nr3c1 expression in Gad1(+) vs. Gad1(-) NG2 glia. Violin plots show Nr3c1 expression levels. e Bubble plot of GABA synthesis/release genes in Gad1 or Nr3c1 expressing NG2 glia. f GO analysis of top ten unique pathways for Nr3c1(+)Gad1(+) NG2 glia and ten shared pathways for NG2 glia. Pathway enrichment analysis was performed using Metascape. Statistical significance was calculated based on the accumulative hypergeometric distribution (one-sided). P-values were adjusted for multiple comparisons using the Benjamini-Hochberg (BH) method to control the false discovery rate (FDR). Two-photon images (g), traces/heatmaps (h), and analysis (i) of [Ca²⁺]_i in MeA NG2 glia after 1 μM CORT applications. n = 38 events (Ctrl) and 20 events (CORT) for amplitude analysis; n = 17 cells for frequency analysis, n = 3 mice. Amplitude, P < 0.001, two-tailed Mann-Whitney test. NS, not significant. Scale bar, 10 μm. j, k Relative Gad1 mRNA (RT-qPCR, j) and GABA concentration (HPLC, k) in primary NG2 glia treated with vehicle or 1 μM CORT (n = 4 independent cultures; P = 0.0418 for mRNA, P = 0.0293 for HPLC, two-tailed unpaired t-test). Two-photon images (l), traces (m), and analysis (n) of [Ca²⁺]_i in MeA GABAergic neurons. n = 122 events (Ctrl) and 101 events (CORT) for amplitude; n = 24 cells for frequency. n = 3 mice. NS, not significant, two-tailed Mann-Whitney tests. Scale bar, 20 μm. Two-photon images (o), traces (p), and analysis (q) of [Ca²⁺]_i in MeA glutamatergic neurons. n = 47 events (Ctrl) and 99 events (CORT) for amplitude; n = 18 cells for frequency. n = 3 mice. NS, not significant, two-tailed Mann-Whitney tests. Scale bar, 20 μm. Data are presented as mean ± SEM.

To determine whether corticosterone acutely modulates NG2 glia, we imaged [Ca²⁺]_i dynamics in NG2 glia from Pdgfrα-CreER^TM;GCaMP6s mice following CORT application. Bath-applied CORT (1 μM) rapidly suppressed [Ca²⁺]_i in both somata and fine processes of NG2 glia within seconds (Fig. 7g–i and Supplementary Video 4). This glia-selective suppression was functionally significant: 12-h exposure of purified NG2 glial cultures to 1 μM CORT reduced Gad1 mRNA ( ~ 40% decrease, measured by RT-qPCR, Fig. 7j) and lowered GABA release ( ~ 31% decrease, measured by HPLC, Fig. 7k). In contrast, [Ca²⁺]_i in neighboring GABAergic or glutamatergic neurons remained unchanged (Fig. 7l–q, Supplementary Video 5), ruling out a direct neuronal effect. Thus, OSD-evoked CORT transiently silences NG2 glia Ca²⁺ signaling, dampens Ca²⁺-dependent GABA synthesis, and thereby disinhibits local GABAergic synaptic transmission to facilitate prosocial allogrooming.

Discussion

In conclusion, our study reveals a compelling role of NG2 glia in modulating prosocial behavior, particularly empathy-driven allogrooming, in the context of trauma observation. Traditionally viewed as a reserve pool for myelinating oligodendrocytes, NG2 glia are increasingly recognized for their broader functions in regulating higher-order cognitive processes and social and emotional responses^20–23. Specifically, Gad1(+) NG2 glia in the medial amygdala (MeA) influence the activity of neighboring GABAergic neurons through a unique mechanism of GABA signal transduction and local synaptic transmission, thereby promoting prosocial behavior in an observational social defeat (OSD) model (Fig. 8). Our findings underscore the critical role of glia-neuron interactions in shaping neural circuitry under both physiological and pathological conditions. The implications of our findings raise the possibility that therapeutic interventions targeting glial GABA signaling pathways could provide a new approach to treating psychiatric disorders characterized by impaired empathy and social function^42,43.

Fig. 8.

In the observational social defeat (OSD) paradigm, vicarious exposure to social defeat elevates circulating corticosterone (CORT) in observer mice. This glucocorticoid signaling suppresses intracellular Ca²⁺ transients and downregulates glutamate decarboxylase 1 (GAD67) expression in medial amygdala (MeA) NG2 glia, consequently attenuating GABA synthesis and reducing inhibitory tone in neighboring MeA GABAergic neurons. This disinhibition facilitates prosocial allogrooming toward the defeated conspecific.

While previous studies have shown the importance of MeA GABAergic neurons in driving prosocial behaviors^2,27,32, our findings reveal an upstream modulatory role for NG2 glia that has not been previously appreciated. The close anatomical proximity between NG2 glia and GABAergic neurons in the MeA raises intriguing questions regarding their functional relationship. One possible explanation lies in their developmental and evolutionary origins. NG2 glia, also known as oligodendrocyte precursor cells (OPCs), persist throughout adulthood and comprise approximately 5-8% of total glial cells^44,45. In the neocortex, interneurons and OPCs derive from the same embryonic regions. Interneurons originate from the caudal and medial ganglionic eminence located in the telencephalon ventral area⁴⁶, and OPCs arise from medial ganglionic eminence in a first wave and from the lateral and caudal ganglionic eminences in a second wave. In postnatal stages, OPCs appear anatomically associated with a subpopulation of interneurons in the hilar region of the dentate gyrus. This anatomical proximity results in a specific physiological interaction between interneurons and their satellite OPCs⁴⁷. In addition, NG2 glia share nearly half of the interneuron transcriptional genes and are considered to be the most similar glial cell type to interneurons⁴⁸. Some studies suggest a shared lineage between NG2 glia and interneurons, potentially explaining their spatial organization and functional interactions⁴⁹. This unique positioning allows NG2 glia to engage in network modulation in a manner reminiscent of GABAergic interneurons. Given the established role of the MeA in processing innate social behaviors^2,27,32, the spatial and functional coupling between NG2 glia and GABAergic neurons may serve as an adaptive mechanism for fine-tuning social behavior.

GABAergic transmission can occur through synaptic release, where neurotransmitters are released into the synaptic cleft and act on postsynaptic receptors, or through spillover, where GABA diffuses to modulate nearby neuronal activity⁵⁰. Our prior work has demonstrated that NG2 glia are capable of releasing GABA via VAMP-2-laden vesicular release in the hippocampus, and that this process is mainly modulated by Gad1 expression, given the synapse-like structures between NG2 glia and neurons observed through immunoelectron microscopy analysis²². In the MeA, we also observed a higher degree of overlap between NG2 glia and GABAergic neurons (Fig. 3c, d). Moreover, through 3D reconstruction, we demonstrated that the processes of NG2 glia maintained tight microstructural connections (within 0.94 um distance) with postsynaptic gephyrin puncta in intact inhibitory synapses (Fig. 3f, g), suggesting that NG2 glia could possibly share similar synaptic component through presynaptic GABA transmission. While this close spatial apposition suggests a potential structural basis for communication, our current imaging approach cannot definitively resolve whether these are bona fide, ultrastructurally defined synapses. Future studies employing high-resolution techniques, such as immuno-electron microscopy, will be essential to precisely characterize the nanoscale architecture of these NG2 glia-neuron contacts.

A critical question emerging from our results is the role of corticosterone (CORT), a principal stress hormone, in shaping NG2 glial function. OSD provokes a robust corticosterone response, and NG2 glia in the MeA exhibit high expression of the glucocorticoid receptor gene Nr3c1 (Fig. 7a, b), rendering them particularly sensitive to endocrine stress signals. This prominent receptor expression is consistent with our finding that NG2 glia are exquisitely sensitive to low doses of CORT, which rapidly suppressed their calcium activity (Fig. 7g–i). Under the same experimental conditions, we detected no rapid calcium responses to CORT in either MeA GABAergic or glutamatergic neurons (Fig. 7l–q). While neuronal responses to CORT are complex and brain-region specific^51,52, the stark contrast in sensitivity within the MeA highlights NG2 glia as primary rapid sensors of acute stress hormone elevation in this circuit. Furthermore, transcriptomic analyses revealed that Nr3c1 is highly expressed in Gad1(+) NG2 glia and is critical in GABA synthesis and release pathway (Fig. 7d). This suggests that glucocorticoid signaling may directly modulate the inhibitory output of these glial cells. Notably, NG2 glia lacking both Nr3c1 and Gad1 expression failed to express key GABA-related genes (Fig. 7e), further supporting an intrinsic correlation between CORT signaling and GABA signaling. Gene ontology comparisons between Nr3c1(+) NG2 glia and other NG2 glial populations identified unique enrichment in pathways related to trans-synaptic signaling and synapse organization (Fig. 7f), implicating a potential role for these cells in circuit-level modulation under stress. These findings raise the possibility that CORT not only transiently silences NG2 glia via calcium suppression, but also engages a transcriptional program that promotes inhibitory gliotransmission.

It is important to acknowledge that a well-established principle in neuroscience: the behavioral output of a primed circuit is dictated by the convergence of internal state and environmental context. For example, the same medial amygdala (MeA) neuronal population can drive either aggressive or parenting behaviors, depending on factors such as sex-hormone levels, stress, chemosensory cues, and motivational state^37,53–55. Analogously, in our model, the OSD context provides the primary trigger. When this signal is integrated with NG2 glia-mediated synaptic modulation, it selectively unmasks an empathic allogrooming response. Thus, in our study, OSD supplies indispensable contextual information, while the synaptic changes imposed by NG2 glia suppression act as a gain-setting mechanism that sharpens this context-dependent response. Evidence supporting this model includes: 1) NG2 glia ablation reduced mIPSC frequency in MeA GABAergic neurons of sham mice (Supplementary Fig. 21), yet these observer mice showed no increase in allogrooming in the absence of OSD exposure (Fig. 5e); 2) NG2-specific Gad1 KO mice exhibited elevated allogrooming exclusively when exposed to OSD (Fig. 6i, j). In the absence of OSD, the same reduction in mIPSC frequency in MeA GABAergic neurons (Fig. 6e) failed to elicit prosocial behavior (Supplementary Fig. 22). Therefore, NG2 glia suppression functions as a context-dependent gain-control mechanism that refines the GABAergic neuronal response to OSD. The convergence of social cues and CORT elevation during OSD selectively activates NG2 glia, shifting them toward a GABAergic-supportive phenotype. This shift dampens local excitatory tone and facilitates empathic behavioral responses. A definitive test of this model would be to administer CORT in vivo during OSD while monitoring NG2 glia Ca²⁺ dynamics, thereby establishing whether CORT directly enhances allogrooming by disinhibiting GABAergic neurons through NG2 glia-mediated reduction of inhibitory synaptic tone. While technically challenging, this remains an important direction for future investigation.

Finally, it is important to note that astrocytes, which constitute approximately 80% of glial cells, are deeply involved in modulating neural circuits that govern social and emotional behaviors^38,56. While our experiments involving astrocyte-specific manipulations did not reveal significant changes in allogrooming or self-grooming behaviors following observational social defeat exposure, this does not negate the role of astrocytes in empathy-related processes. In the lateral septum (LS), astrocytes regulate stress-induced social avoidance by releasing adenosine, which dampens excitatory transmission⁵⁶. Similarly, in the dorsomedial prefrontal cortex (dmPFC), astrocytic glutamate and ATP release modulate excitatory/inhibitory balance to influence social dominance behavior³⁸. Given these diverse roles, it is possible that astrocytes contribute to empathy indirectly by integrating emotional states with social decision-making processes in regions such as the hippocampus. Rather than directly encoding prosocial motivation, they may facilitate long-term synaptic changes that influence how past experiences shape future social interactions. Therefore, with the growing recognition of glial cells as active participants in brain function, their precise roles in a wide range of neuropsychiatric conditions, particularly those involving deficits in emotional regulation and social cognition, should be drawn more attention.

Methods

Animals

Animal care and use adhered strictly to institutional guidelines and governmental regulations. The animal experiment protocols were authorized by the US National Institutes of Health (Protocol number: A-2022-036) and were approved by the Animal Ethics Committee of Shanghai Jiao Tong University School of Medicine (AAALAC accreditation Unit, 001670). All experiments were conducted in accordance with IACUC approval at Shanghai Jiao Tong University (Policy Number: JUMC2023-085-A). All mice were maintained under a 12/12 dark/light cycle at 22–25 °C with 40–60% humidity, with free access to rodent food and water in environmentally controlled conditions.

The animals used in the experiments were adult male (6–12 weeks old) C57BL/6 J mice for mechanistic studies; a separate cohort of adult female C57BL/6 J mice (6–12 weeks old) was used for additional behavioral validation (Supplementary Fig. 3i–l). C57BL/6 J (C57) mice were purchased from the Shanghai Laboratory Animal Center, Chinese Academy of Sciences. The following transgenic and reporter lines were also used: Pdgfrα-CreER^TM (Jackson Laboratory; stock no. 018280), ChR2(H134R)-eYFP (Ai32) (Jackson Laboratory; stock no. 024109), Rosa26-mGFP (Jackson Laboratory; stock no. 007676) mice were obtained from the Jackson Laboratory (USA). Gad1^lox/lox (Jackson Laboratory; stock no. 031800) mice were gifts from Prof. Qi Wu at Baylor College of Medicine, GCaMP6s^f/f (Jackson Laboratory; stock no. 024106) mice were gifted from Prof. Nanjie Xu, ROSA26iDTR (Jackson Laboratory; stock no. 007900) mice were gifts from Prof. Qian Li at Shanghai Jiao Tong University School of Medicine, NG2-CreER^TM (Jackson Laboratory; stock no. 008538) were gifted from Prof. Chong Liu at Zhejiang University (Zhejiang, China), Ai35D (Jackson Laboratory; stock no. 012735) were gifted from Prof. Shumin Duan and Yanqin Yu at Zhejiang University (Zhejiang, China). All mice were group-housed until surgery. All behavioral procedures were performed during the light cycle. CRE recombination was induced in adult mice (6–8 weeks old) prior to any surgical or behavioral procedures. Tamoxifen (Sigma-Aldrich, T5648) was dissolved in sunflower seed oil (Sigma-Aldrich, S5007) at a concentration of 20 mg/ml. Mice received intraperitoneal injections of tamoxifen at a dose of 120 mg per kg body weight, once daily for five consecutive days. Experimental procedures were initiated after the 5-day tamoxifen induction regimen.

Observational social defeat (OSD) model

The social defeat model was originally developed by Miczek⁵⁷. We have modified the resident-intruder model to create an observational social defeat model. Unless otherwise specified, all experiments were performed using male C57BL/6 J mice as observers and demonstrators, and male CD-1 mice as residents. Briefly, a male C57BL/6 J mouse was co-housed with a demonstrator mouse for 2 weeks to allow bonding (social support) during the acclimatization period.

Co-housing and pair selection: The majority of co-housed male C57BL/6 J observer and demonstrator mouse pairs were from the same litter. In the few instances where mice from different litters were used, they were introduced to each other at a young age by being placed into a new, clean cage simultaneously. This “neutral territory” introduction prior to sexual maturity is a standard and effective method for minimizing territorial aggression and allowing a stable social hierarchy to form. All co-housed pairs were monitored daily for signs of aggressive behavior during the entire 2-week co-housing period. Pairs that exhibited sustained, injurious fighting were excluded from the study. The vast majority of pairs established a stable social relationship using this protocol. In the rare instances where a pair could not be stabilized (evidenced by repeated wounding), that specific pair was excluded from the study, and the mice were not used in any experiments.

For each pair of paired observer and demonstrator, 1 h of baseline interaction (phase 1) was recorded. Later, one demonstrator considered as an intruder was introduced into the home cage of a resident CD-1 mouse. This resulted in a typical social defeat behavior indicated by the intruder surrendering, when attacked by the resident CD-1 mouse. While the observer was housed in an adjacent compartment, separated by a transparent perforated glass partition without receiving direct physical contact (Fig. 1a). The partition allowed visual, auditory, and olfactory interactions. This initiated a freezing response in the observer. After 1 h of social defeat observation (phase 2), the observer and the paired demonstrator were reunited in the interaction cage (phase 3). The complete OSD paradigm therefore comprised three distinct phases: phase 1 (Interaction, baseline recording), phase 2 (Observation, of social defeat or control condition), and phase 3 (Reunion, social interaction after observation).

OSD in female observers: To test whether the prosocial response generalizes across sexes, we also performed a subset of experiments using female C57BL/6 J mice as observers and demonstrators (Supplementary Fig. 3i–l). Because female mice do not reliably elicit intense aggression from male CD-1 residents, we adopted a modified protocol: the female demonstrator was smeared with urine from an unfamiliar male C57BL/6 J mouse 10 min before the OSD session to enhance agonistic signals. In addition, the resident CD-1 males were pre-screened for consistent aggression toward such scent-augmented female intruders. All other procedures remained identical to those described for male mice.

Video recordings from digital cameras positioned overhead were analyzed using the behavior tracking system (Ethovision X15, Noldus). Freezing, self-grooming, and allogrooming behaviors were then recorded and quantified in both observers and demonstrators. The behavioral criteria were used for quantification as follows:

1. Freezing: Freezing behavior was defined as the complete absence of movement, except for respiration. A mouse was considered to being “freezing” if it remained immobile for at least 1 s, with its body posture stiff and the head held still. This behavior was considered an indicator of fear or stress in response to the observed social defeat event. The total duration of freezing behavior was recorded, and the percentage of time spent freezing during the 1-hour observation period was calculated.

2. Self-grooming: Self-grooming behavior was defined as any grooming behavior directed toward the observer’s own body, including licking, biting, and paw movements used to clean the body or face. The total time spent on self-grooming was measured in seconds, with grooming events lasting at least 1 s being counted as separate episodes. Repetitive self-grooming is considered a sign of coping behavior or self-soothing, particularly under stress.

3. Allogrooming: Allogrooming behavior was defined as grooming behaviors directed toward the paired demonstrator, including licking or nibbling the demonstrator’s fur. Based on grooming site, allogrooming was categorized into body grooming—including trunk grooming (licking or nibbling the torso), limb grooming (directed at the forelimbs or hindlimbs), and tail grooming—and head grooming, which encompassed face grooming (targeting the snout, cheeks, or forehead) and ear grooming (licking or nibbling the ears)³². This behavior is typically associated with social bonding and comfort. Allogrooming episodes were counted if they lasted for at least 1 s, and the total number of allogrooming events was recorded. The total duration of allogrooming was also analyzed during the observation period.

All behaviors were manually counted by a researcher blind to the experimental condition and reviewed in detail through repeated analysis of the video footage. The observer’s behavioral responses to the social defeat event (including freezing, allogrooming, and self-grooming) were compared to baseline behaviors recorded prior to the observation.

Open field test

To assess general locomotion and anxiety-like behavior, mice underwent the open field test (OFT), a well-established paradigm for evaluating exploratory activity in a novel environment⁵⁸. All animals were habituated to the behavioral testing room for at least 1 hour prior to testing to minimize handling-related stress. The apparatus consisted of a gray plastic arena (40 × 40 × 35 cm), with a designated central zone (20 × 20 cm) and peripheral zone. At the beginning of each trial, mice were placed in the center of the arena, and their activity was recorded for 30 min using a video tracking system (EthoVision XT 14, Noldus, Netherlands). Key parameters analyzed included total distance traveled and time spent in the center zone. The arena was thoroughly cleaned with 75% ethanol between sessions to eliminate olfactory cues.

Elevated plus maze test

Anxiety-like behaviors were assessed using the elevated plus maze (EPM), consisting of two open arms (30 × 7 × 0.25 cm) and two closed arms (30 × 7 × 15 cm) constructed from opaque gray plastic. The maze was elevated 60 cm above the floor. Mice were placed at the center junction of the maze, facing an open arm, and allowed to freely explore for 20 min. Animal trajectories were tracked and analyzed using EthoVision XT 14 software (Noldus, Netherlands), with quantification of time spent and entries into open versus closed arms. All testing surfaces were sanitized with 75% ethanol between trials to eliminate residual scent cues.

Viral constructs and stereotaxic surgery

The following AAVs were utilized: AAV2/9-mDlx-mCherry-WPRE-PA (Shanghai Taitool Bioscience, Cat# S0631-9), AAV2/9-mDlx-eGFP-WPRE-PA (Shanghai Taitool Bioscience, Cat# S0550-9), AAV2/9-mDlx-GCaMP6s-WPRE-pA (Shanghai Taitool Bioscience, Cat# S0527-9), AAV2/9-VGLUT2-GCaMP6s-WPRE-hGH (BrainVTA Co. Ltd, PT-3722), AAV2/9-VGLUT2-mCherry-WPRE-hGH (BrainVTA Co. Ltd, PT-2385). All viral vectors were stored in aliquots at –80 °C until further use.

Stereotaxic surgery was performed as previously described^59,60. Pdgfrα-CreER^TM;iDTR mice at 7 weeks old were anesthetized with 1–1.5% isoflurane and secured in a stereotaxic apparatus (RWD Instruments, Shenzhen, China). Body temperature was maintained using a heating pad. Following a midline scalp incision, small craniotomies were made over target sites. The dorsoventral (DV) coordinate was zeroed at the brain surface (dura mater) before insertion. Glass micropipettes (tip diameter: 10–20 μm), pulled using a P-97 puller (Sutter Instrument, USA), were backfilled with silicone oil and connected to a microinjector pump (RWD Life Science) to prevent air bubbles. AAV-containing solutions were front-loaded into the pipette tips and infused into the brain at a rate of 0.06 μl/min. 4-hydroxytamoxifen (4-OH) or diphtheria toxin (DT) were front-loaded into the pipette tips and infused into the brain within 10 min. The infusion volume was 2 mM 1.5 uL for 4-OH and 1.5 ng 50 nL for DT. The following coordinates (relative to bregma; anteroposterior to bregma, AP; medial to lateral, ML; dorsal to ventral, DV; in mm) were used:

MeA: AP, −1.40, ML, ±2.0, DV, −5.25

MPOA: AP, −0.2, ML, ±0.36, DV, −4.95

Each site received 0.3 μl of virus unilaterally or bilaterally. Following injection, the pipette was left in place for 30 min to minimize backflow. Mice were allowed to recover for a minimum of three weeks before behavioral or physiological testing. Injection sites were verified post hoc by fluorescence signal from reporter expression.

For optogenetic activation and inhibition of NG2 glia, Pdgfrα-CreER^TM;ChR2 (Ai32) and NG2-CreER^TM;Ai35D transgenic mice were used. Bilateral optical fiber cannulas (200 μm core diameter, 0.37 NA; Inper) were implanted at the following stereotaxic coordinates relative to bregma: AP − 1.4, ML ± 2.0, DV − 5.25 mm. To control for potential stress differences arising from surgical interventions, both animals within each experimental pair underwent identical surgical procedures. Viral expression and fiber placement were verified post hoc via histological analysis in all included cases.

Fiber photometry

Fiber photometry experiments were conducted using a two-color fiber photometry system (Inper, Zhejiang, China), following previously described protocols with minor modifications^61,62. Excitation was provided by a 470 nm laser, while a 410 nm reference wavelength was simultaneously applied to correct for motion artifacts and background fluorescence. Optical fibers (200 μm core diameter, NA  =  0.37, Inper) were stereotaxically implanted into the MeA, positioned 50 μm above the viral injection site. Laser light was transmitted between the rotary joint and the implanted fiber, with output power calibrated to 20–40 μW at the fiber tip to minimize photobleaching. Emitted fluorescence was collected through the same fiber and processed using Inper Studio software (Inper), with 410 nm signals serving as an internal control for ratiometric correction. The fluorescence value was acquired during each allogrooming train or each self-grooming train. The onset of the grooming train is defined as the point at which the grooming behavior begins. In all of the following analyses, this is marked as time 0. The variation in the fluorescence values (ΔF/F) was computed by (F-F0)/F0. Here, F refers to the fluorescence values at each time point (measured relative to the onset of the grooming train, with time points ranging from −1 to 5 seconds) while F0 represents the median fluorescence values during the baseline period (from −1 to 0 s relative to the grooming train onset, Fig. 2d, e). For visualizing the fluorescence change, the ΔF/F values are illustrated with heatmaps or plots, where the shaded areas indicate SEM. To statistically measure the change in fluorescence values across the scratching train or optogenetic stimulation, the average amplitude of ΔF/F was determined as the average fluorescence amplitude change from the baseline during the peak period (ranging from 0 to 5 seconds relative to the onset of the grooming train).

Optogenetics experiments

After stereotaxic surgeries, the mice were allowed to recover for 2 weeks before behavioural testing. Blue light (470 nm, 8–10 mW, 20 Hz) was generated by an external laser power source (Newdoon Inc., Hangzhou, China) and delivered bilaterally during phase 3. Yellow light ( ~ 575 nm, 5–10 mW, 20 Hz) was generated by an external laser power source (XT640-W, Lumen Dynamics) and delivered bilaterally during phase 3.

Acute brain slice preparation

Mice aged postnatal 4–6 weeks were deeply anesthetized with 5% chloral hydrate and transcardially perfused with ice-cold, oxygenated dissection buffer (in mM): 82.75 NaCl, 2.4 KCl, 6.8 MgCl₂, 0.5 CaCl₂, 1.4 NaH₂PO₄, 23.8 NaHCO₃, 23.7 D-glucose, and 65 Sucrose. Coronal slices (300 µm) containing the MeA were prepared using a vibratome (VT1000S; Leica Microsystems, Germany), and incubated for ≥1 hr at 31.2 °C in artificial cerebrospinal fluid (aCSF) composed of (in mM): 125 NaCl, 2.5 KCl, 1 MgCl₂, 2 CaCl₂, 1.25 NaH₂PO₄, 25 NaHCO₃, and 12.5 D-glucose. All solutions were continuously saturated with 95% O₂/5% CO₂.

Electrophysiological recordings

Slices were transferred to a recording chamber and continuously perfused with oxygenated aCSF at room temperature. Neurons in the MeA were visualized using an upright epifluorescence microscope (BX51WI, Olympus, Japan) equipped with DIC optics and an infrared CCD camera (optiMOS, Q IMAGING). Recordings were obtained using a MultiClamp 700B amplifier (Molecular Devices), filtered at 2 kHz and digitized at 20 kHz via a Digidata 1550 A interface (Molecular Devices). Data acquisition began 2 min after achieving a stable whole-cell configuration. Patch pipettes were fabricated from borosilicate glass capillaries using a P-1000 puller (Sutter Instruments).

Photostimulation was delivered via a ~ 575 nm yellow LED (XT640-W, Lumen Dynamics) at 5–10 mW/mm², triggered by a stimulator. Light pulses (10 ms, 15 Hz) were applied for 90 s through a 40×, 0.8 N.A. objective during mIPSC and mEPSC recordings, after ≥2 min of stable baseline recording.

For mEPSC recordings, the internal solution contained (in mM): 125 K-gluconate, 15 KCl, 8 NaCl, 10 HEPES, 0.2 EGTA, 3 Na₂-ATP, and 0.3 Na-GTP (pH to 7.3). Cells were voltage-clamped at −60 mV. For mIPSCs recordings, high chloride intracellular solution containing (in mM): 130 KCl, 2 MgCl₂, 0.5 CaCl₂, 2.5 Na₂-ATP, 0.3 Na-GTP, 10 HEPES and 1 EGTA (pH to 7.3 with KOH), and cells were also held at −60 mV. In all, 20 μM DNQX, 50 μM D-AP₅, and 1 μM TTX were added into the bath to block AMPA, NMDA, and Na⁺ channel currents, respectively, to isolate mIPSCs. 20 μM bicuculline and 1 μM TTX were added into the bath to block GABA_ARs and Na⁺ channel currents to isolate mEPSCs. The miniature IPSCs and EPSCs were analyzed with the Mini-Analysis 6.0 (Synaptosoft). Access resistance was monitored throughout recordings, and cells showing > 20% variation were excluded from analysis.

Ca²⁺ imaging and analysis

Anesthesia, perfusion, and brain slicing procedures were performed as described above. For optogenetic inhibition of NG2 glia, acute brain slices were incubated with 5 μM Rhod2-AM (Invitrogen, R1244) in culture medium at 31.2 °C for 30 min, followed by two washes and a 30 min equilibration prior to imaging. For NG2 calcium imaging in OSD mice, NG2-CreER^TM;GCaMP6s brain slices were acquired to visualize the calcium dynamics in NG2 glia. Slice imaging was conducted with a two-photon microscope (Fluoview FVMPE-RS; Olympus, Japan) equipped with a ×25, 1.05 N.A. water-immersion objective (Olympus, Japan) at the Core Facility of Basic Medical Sciences, Shanghai Jiao Tong University School of Medicine. GCaMP6s fluorescence ([Ca²⁺]_i) were excited at 920 nm and detected through a green fluorescence emission filter. Time-lapse (x-y) images were acquired at 256 × 256 resolution (0.994 μm per pixel) every 550 ms. Z-stacks (20 μm total thickness, three steps) were collected to ensure optimal visualization of NG2 glia. Data were analyzed using the GECI quant method⁶³. Briefly, background fluorescence was subtracted from each movie, and polygonal regions of interest (ROIs) were drawn around the soma and processes of visually identifiable NG2 glia. Baseline fluorescence (F0) was calculated as the average of the 10 frames with the minimal fluctuation across an entire recording. Calcium transients were quantified as ΔF/F = (F-F0)/F0, where F represents the fluorescence intensity at each time point.

Immunohistochemistry and image analysis

Mice were anesthetized and transcardially perfused via the ascending aorta with normal saline for approximately 3 min, followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB) for 5 min. Brains were harvested, post-fixed in 4% PFA at 4 °C overnight, and subsequently sectioned into 40-μm-thick coronal sections. For whole-brain cFos screening, sections containing the regions of interest were selected based on anatomical definitions from the Mouse Brain Atlas⁶⁴. For the assessment of NG2 ablation within the MeA or MPOA, the entire MeA and MPOA were also sectioned. Sections were permeabilized in 0.3% Triton X-100 in PBS for 15 min, followed by blocking with 10% donkey serum (Ruite Biotechnology, w9030-05) in PBS-T (PBS containing 0.1% Triton X-100) for 2 h at room temperature. The following primary antibodies were used: rabbit antibody to NG2 (1:250, Millipore AB5320), goat antibody to Pdgfrα (1:300, R&D Systems AF1062), chicken antibody to GFAP (1:1000, Abcam ab4674), chicken antibody to GFP (1:500, Abcam ab13970), mouse antibody to cFos (1:1000, Abcam ab208942), rabbit antibody to Vgat (1:100, Cell Signaling 44498S), rabbit antibody to glutamate (1:100, Millipore G6642), rabbit antibody to PSD95 (1:200, Cell Signaling 3450), mouse antibody to NeuN (1:200, Abcam ab104224). The corresponding secondary antibodies included: Donkey anti-Rabbit Alexa Fluor 568 (1:500, Invitrogen A10042), Donkey anti-mouse Alexa Fluor 647 (1:500, Invitrogen A31571), Goat anti-Chicken Alexa Fluor 488 (1:500, Invitrogen A11039), Donkey anti-Goat Alexa Fluor 488 (1:500, Invitrogen A11055), Goat anti-Rabbit Alexa Fluor 647 (1:500, Cell Signaling 4414 S), Goat anti-Rabbit Alexa Fluor 488 (1:500, Cell Signaling 4412 S). Nuclei were counterstained with DAPI (1:1000, Cell Signaling 4083S) for 15 min, and sections were mounted with Fluoromount™ Aqueous Mounting Medium (AQUA-MOUNT, REF 13800). All images were acquired on Leica TCS SP8 confocal microscope. Z-stack images of the MeA brain region were acquired using a Leica confocal microscope system equipped with a 20× or 40× objective. For each field of view, at least 10 optical sections were captured at 1 µm intervals. Maximum intensity projections of the Z-stacks were generated using the Leica acquisition software. Images of whole-brain sections were acquired using the OLYMPUS VS200 system. The resulting 2D projection images were then exported and analyzed in ImageJ to quantify co-localization ratios.

Image analysis was conducted with ImageJ v1.52a (NIH). Images from different fluorescence channels were thresholded, and cell counts were based on DAPI-stained nuclei. Sections were overlaid onto corresponding atlas sections, and cell counting within designated brain regions was manually performed by an investigator blinded to the experimental conditions.

RNAscope in situ hybridization

Mice (6–8 weeks old) were deeply anesthetized and transcardially perfused with 4% paraformaldehyde (PFA). Brains were removed and post-fixed overnight at 4 °C in the same fixative. Following dehydration via a sucrose gradient, brains were sectioned at 10 μm thickness using a cryostat microtome (Leica, CM1950). For RNAscope in situ hybridization (ISH), sections were treated with RNAscope Hydrogen Peroxide, followed by antigen retrieval using RNAscope Target Retrieval Reagent (95–99 °C, 5 min) and RNAscope Protease III (40 °C, 30 min). Probe targeting Gad1-C1 (Cat. No. 400951, ACD) and Slc17a6-C3 (Cat. No. 319171, ACD) was hybridized at 40 °C for 2 h. Signal amplification was performed per standard protocols, followed by fluorescent detection using Opal 570 and Opal 690 (Cat.FP1488001KT and Cat.FP1497001KT, Akoya Biosciences). For immunohistochemistry (IHC) post-RNAscope ISH, sections were directly immunostained using NG2 (1:250, Millipore AB5320) and GFP (1:500, Abcam ab13970) antibody. Imaging was performed on a Leica TCS SP8 confocal microscope (HC PL APO CS2 ×63/1.40 oil objective).

Brain tissue sampling and corticosterone (CORT) ELISA

Brain tissues encompassing the medial amygdala (MeA) were microdissected from OSD-exposed mice and sham controls that observed non-defeated partners. The samples were homogenized, and corticosterone (CORT) levels were quantified using mouse-specific ELISA kits (Amoy Lunchangshuo Biotech, Xiamen, China) in accordance with the manufacturer’s instructions.

GABA content examination by HPLC

To determine the GABA content, Primary NG2 glia were sonicated in 0.2 N HClO4 lysis solution, and the supernatants were collected after centrifugation at 12,000 × g for 20 min at 4 °C. HPLC analysis was performed using the UltiMate 3000 UHPLC/TSQ Vantage LC-MS/MS (Thermo Scientific, USA) at the Core Facility of Basic Medical Sciences, Shanghai Jiao Tong University School of Medicine. Fluorescence detection used 340 nm excitation and 420 nm emission wavelengths. Data were collected and analyzed using ChemStation software (Agilent Technologies), with peaks and concentrations determined by comparison to external standards.

Primary NG2 glia culture

Brain were isolated from Wild-type mice, Pdgfrα-CreER^TM; Ai32 mice, and Pdgfrα-CreER^TM; Ai32; Gad1^lox/lox mice pups at postnatal day 1. Brain tissues were diced into ~1 mm³ pieces in a 60 mm dish with a sterilized razor blade. The minced tissues were transferred to digestion solution [0.25% trypsin (Gibco, Baltimore, MD) and 75 U/ml DNase I (Worthington, Lakewood, NJ)] and incubated for 10 min at 37 °C in a tissue culture incubator. The cells were collected by centrifugation at 1000 × g for 5 min using a swinging bucket rotor. The pellet was resuspended in freshly prepared ice-cold neurosphere growth medium [DMEM/F12 supplemented with B27 (Gibco, Baltimore, MD) and 10 ng/ml EGF (Peprotech, Rocky Hill, NJ)], and cells were plated at 5 × 10⁵/ml in dishes. Half of the medium was replaced with fresh neurosphere growth medium every 2 days for 8–10 days until neurospheres formed. After neurosphere formation, the medium was changed to oligosphere medium [DMEM/F12 supplemented with B27, 10 ng/ml PDGF (Peprotech, Rocky Hill, NJ), 10 ng/ml bFGF (Peprotech, Rocky Hill, NJ), and 1 μM (Z)−4-hydroxytamoxifen (Sigma-Aldrich)]. Oligospheres formed after 7–9 days, after which they were digested with TrypLE (Gibco, Baltimore, MD). The dissociated OPCs/NG2 cells were then plated at 5 × 10⁵/ml on new PDL-coated dishes in OPC medium [DMEM/F12 supplemented with B27, N2 (Gibco, Baltimore, MD), 0.1% BSA, 10 ng/ml PDGF, 20 ng/ml bFGF, 5 μg/ml IGF (Peprotech, Rocky Hill, NJ), and 1 μM (Z)−4-hydroxytamoxifen].

NG2 glia isolation by FACS

NG2 glia were isolated from Pdgfrα-CreER^TM;Rosa26-mGFP mice via fluorescence-activated cell sorting (FACS). Brain tissues from 6-week-old mice were dissociated according to previously published protocols with minor modifications²². Briefly, mice were deeply anesthetized and perfused intracardially with ice-cold, oxygenated aCSF (95% O₂, 5% CO₂) containing (in mM): 125 NaCl, 2.5 KCl, 1 MgCl₂, 2 CaCl₂, 1.25 NaH₂PO₄, 25 NaHCO₃, and 12.5 D-glucose. Coronal brain slices (300 μm thick) were prepared using a vibratome (VT1200S, Leica Microsystems, Germany) and equilibrated in continuously oxygenated aCSF at 31.2 °C for 30 min. The medial amygdala (MeA) was dissected from slices and enzymatically digested in 3 ml papain solution (15 U/ml papain, 75 U/ml DNase I, 2 mM cysteine, 50 mM EDTA) at 37 °C for 45 min. Digestion was terminated by adding a protease inhibitor solution containing 1 mg/ml ovomucoid, 0.1% BSA, and 75 U/ml DNase I. The tissue was then gently triturated, centrifuged at 300 × g for 3 min at 4 °C, resuspended in D-PBS with 0.1% BSA, and filtered through a 70 μm mesh. FACS was performed using a MoFlo Astrios cell sorter (Beckman Coulter) with a 70 μm nozzle at the Core Facility of Basic Medical Sciences, Shanghai Jiao Tong University School of Medicine. Sorted cells were analyzed using FlowJo software and used for downstream experiments.

RT-PCR and Single-cell RT-PCR

GFP( + ) NG2 glia for RT-PCR were purify NG2 glia by FACS following a method described previously. Gad1(-) or Gad1(+);Ai32-eYFP(+) NG2 glia for Single-cell RT-PCR from Pdgfrα-CreER^TM;Gad1^lox/lox;ChR2(Ai32) mice or Pdgfrα-CreER^TM;ChR(Ai32) at postnatal 6 weeks were selected and aspirated into a glass pipette from MeA acute slices. NG2 glia were grabbed promptly by micromanipulation and immediately placed in lysis buffer. To minimize potential changes in gene expression, all cells were collected within 3 h after slice preparation. Selected cells were processed for single-cell RNA extraction and reverse transcription within 1 h and were subjected to cDNA amplification and purification. Single-cell cDNA was amplified using KAPA HiFi Hot Start Ready Mix (2×; KAPA Biosystems, Cat. No. KK2601) according to the manufacturer’s protocol²². PCR was performed by using specific primers targeting Gad1, Gad2, Pdgfrα, Egfp, and Gapdh (Sangon Biotech, Shanghai, China) as listed in Supplementary Data 1. Gapdh was used as an internal control.

Smart-seq2

GFP( + ) NG2 glia for Single cell RNA-seq were purify NG2 glia by FACS following a method described previously. For Smart-seq2 transcriptome experiment, the cDNA from NG2 glia was synthesized and amplified from ~100cells using SMART-Seq® HT Kit User Manual (Takara Bio, USA) to generate double stranded cDNA for each replicate following the manufacturer’s instructions. The cDNA was subjected to SMRT sequencing at Shanghai Biotechnology Corporation (SHBIO, Shanghai, China) using PacBio RSII (Pacific Biosciences). Smart-seq data are available in SRA (accession number: PRJNA1266598).

Single cell RNA-seq

GFP( + ) NG2 glia for Single cell RNA-seq were purify NG2 glia by FACS following a method described previously. The NG2 glia were subjected to Single cell RNA-seq at Beijing SeekGene BioSciences (Seekgene, Beijing, China). Single cell RNA Seq libraries were prepared using SeekOne® MM Single Cell 3’ library preparation kit (SeekGene). Briefly, appropriate number of cells were loaded into the flow channel of SeekOne® MM chip which had 170,000 microwells and allowed to settle in microwells by gravity. After removing the unsettled cells, sufficient Cell Barcoded Magnetic Beads (CBBs) were pipetted into flow channel and also allowed to settle in microwells with the help of a magnetic field. Next excess CBBs were rinsed out and cells in MM chip were lysed to release RNA which was captured by the CBB in the same microwell. Then all CBBs were collected and reverse transcription were performed at 37 °C for 30 min to label cDNA with cell barcode on the beads. Further Exonuclease I treatment was performed to remove unused primer on CBBs. Subsequently, barcoded cDNA on the CBBs was hybridized with random primer which had reads 2 SeqPrimer sequence on the 5’ end and could extend to form the second strand DNA with cell barcode on the 3’ end. The resulting second strand DNA were denatured off the CBBs, purified and amplified in PCR reaction. The amplified cDNA product was then cleaned to remove unwanted fragment and added to full length sequencing adapter and sample index by indexed PCR. The indexed sequencing libraries were cleanup with SPRI beads, quantified by quantitative PCR (KAPA Biosystems KK4824) and then sequenced on illumina NovaSeq 6000 with PE150 read length or DNBSEQ-T7 platform with PE100 read length. The Seurat R package (version 3.2.0) was used for further inspection and data analysis⁶⁵.

The resulting filtered matrix consisted of 1864 cells. The matrix was normalized using the NormalizeData() function and variable features were identified using FindVariableFeatures() with 2000 genes. The ScaleData() function was used to center the gene expression. Next, principal component analysis (PCA) was performed, using RunPCA() function, to obtain the top 50 principal components (PCs). Clustering was conducted using the FindNeighbors() and FindClusters() functions using 20 PCs and a resolution parameter set to 0.3. For visualization, the dimensionality of the data sets was reduced by t-SNE, using the RunTSNE() function in the Seurat package. Cell populations were matched to cell types based on the expression of known marker genes and previously identified expression signatures⁶⁶. Single-cell RNA-seq data are available in SRA (accession number: PRJNA1262306).

Chemicals and reagents

Reagents were purchased from Sigma-Aldrich. Normal donkey serum and normal goat serum were purchased from Ruite Biotechnology. ZnSO₄ was purchased from Sigma-Aldrich (Cat#: 83265). L-AAA was purchased from Merck (CAS 1118-90-7). Corticosterone was purchased from MedChemExpress (CAS 50-22-6). Norepinephrine was purchased from MedChemExpress (CAS 51-41-2).

Statistical analysis

All statistical analysis was run in GraphPad InStat 3. The graphs were created in Origin 8. Data are presented as means ± SEM and the error bar represents SEM for each set of data to be compared. We determined in GraphPad Instat whether the data were normally distributed or not. If they were normally distributed, parametric tests (Paired and unpaired Student’s two-tailed t-tests) were used. If the data were not normally distributed, non-parametric tests (two-tailed Mann-Whitney tests or Wilcoxon-matched pairs tests) were used. For electrophysiological experiments, n values represent the number of recorded cells. For all biochemistry, n values represent the number of mice. For behavioral experiments, n values represent the number of testing mice. Investigators were blinded to the groups or samples during the experiments. No statistical methods were used to pre-determine sample size, or to randomize. Statistical significance was set at *P < 0.05, **P < 0.01, ***P < 0.001.

All experiments were independently repeated at least three times with similar results. Representative images are shown from one independent experiment. Sample sizes (n) and statistical details are provided in each figure legend. Data collection and analysis were performed blind to the experimental conditions.

Reporting summary

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

Author contributions

Y.J., S.J. and P.L. were responsible for behavioral data analysis, optogenetics, immunohistochemistry, animal breeding, and data analysis. Y.J. and Y.H. were responsible fiber photometry and two-photon calcium imaging. S.J. and X.H. performed single-cell sequencing, SMART sequencing, and qPCR. Y.J., S.J., X.H., P.L. and A.S. carried out behavioral experiments. Y.J. and X.Z. were responsible for electrophysiology. X.T., S.D., Y.J., and S. J. directed the work and wrote the paper. X.T. has accessed and verified the underlying data. All authors have read and approved the final version of the manuscript.

Peer review

Peer review information

Nature Communications thanks Wenhui Huang and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

X.T. discloses support for the research of this work from Brain Science and Brain-like Intelligence Technology – National Science and Technology Major Project [2022ZD0204700], Innovative Drug Research and Development-National Science and Technology Major Project [2026ZD1807200], the National Natural Science Foundation of China [32471063, 82271466, 91632104 and 31970904], Shanghai Municipal Science and Technology Major Project [2018SHZDZX05], the Open Research Fund of Navy Medical University Basic Medical College [ORFBMC-JCKFKT-MS-009] and the Innovative Research Team of High-level Local Universities in Shanghai for providing support [SHSMU-ZDCX20211901]. Y.J. discloses support for the research of this work from the Doctoral Science and Technology Innovation Cultivation Fund [24KCPYYB001]. S.J., P.L., X.Z., X.H., Y.H., A.S. and S.D. declare no relevant funding.

Data availability

All data supporting the findings of this study are available within the paper, Supplementary Information and Source Data files. The single-cell RNA-seq data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession code PRJNA1262306. The Smart-seq data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession code PRJNA1266598. Source data are provided with this paper.

Code availability

The code used in this study is available via GitHub (https://github.com/aedasc/25-MeA-RNAseq-Analysis) and has been archived with Zenodo under DOI: 10.5281/zenodo.19363737.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73488-0.