A prefrontal cortex-lateral hypothalamus circuit controls stress-driven increased food intake

mPFC-LHA stimulation, in a frequency-dependent manner, increases palatable food intake

We first assessed whether the mPFC-LHA pathway is capable of driving intake of palatable food, as thus far this had not been reported. Instead, prior studies evaluating the mPFC-LHA pathway showed that high-frequency optogenetic stimulation (100 Hz bursts) does not affect regular food intake²⁰, while continuous chemogenetic stimulation (or 20 Hz optogenetic stimulation) decreases chow intake²¹. Notably, both mPFC and LHA are individually known to have frequency-dependent functions, with lower frequency stimulation able to drive distinct behaviors^19,22. Therefore, we tested whether low-frequency stimulation of the mPFC-LHA pathway can increase palatable food intake. We stereotactically injected C57B6 mice in the mPFC with an AAV to express the optogenetic construct CoChR (rAAV5-Syn-CoChR-GFP) or YFP control vector (rAAV5-Syn-eYFP), and targeted optic fibers to the LHA to optogenetically stimulate the mPFC-LHA pathway (Fig. 1A, B; Supplementary Fig. 1A). After recovery, mice were placed in a 2-choice food-context containing regular chow and fat as a palatable food source. Baseline food intake was established over 3 consecutive days, followed by a session during which the mPFC-LHA pathway was stimulated at a specific frequency (Fig. 1C). We observed that during sessions of 5 Hz stimulation, but not of 1 or 10 Hz, there was an overall increase of fat intake (Fig. 1D; Supplementary Fig. 1B). Between the two choices, this increased intake was specific to fat, as it did not occur for chow (Supplementary Fig. 1C).

We then used video-based analysis to examine the feeding behavior within mPFC-LHA stimulation blocks (ON), contrasting them with behavior during alternating non-stimulated blocks (OFF). We observed that with 5 Hz stimulation, mice spent more time specifically in the fat zone (Supplementary Fig. 1D). Furthermore, we observed that the effect of pathway stimulation on fat consumption differed over the course of the task. Interactions with fat normally waned over the hour (Fig. 1E). Whereas 5 Hz stimulation of the pathway did not affect fat intake at the start, when the drive for fat was still high, it instead drove continued fat consumption later on when such interactions had typically waned, as seen in the control groups and in the 1 Hz stimulation conditions (Fig. 1E). Consequently, 5 Hz stimulation resulted in overall more feeding bouts for fat, but not chow (Fig. 1F; Supplementary Fig. 1E). These effects on food intake did not stem from effects of pathway stimulation on general locomotor activity, which increased linearly in a different pattern from the effect on food intake (Supplementary Fig. 1F), nor did they stem from any direct alterations of emotional valence states, as assessed with a real-time place preference/avoidance assay (Supplementary Fig. 1G). Overall, these data indicate that the mPFC-LHA pathway is able, at least in the lower frequency domain, to drive the continued intake of palatable fat.

mPFC cells projecting to LHA modulate their firing in response to social stress

Stressful experiences can drive excessive food intake (i.e., hyperphagia)^3,23, ultimately causing weight gain²³. We reasoned that if the mPFC-LHA network were involved in stress-driven effects on food intake, it should be able to detect stressful information. To test this hypothesis, we used a social stress experience, where a mouse is temporarily exposed to the aggression and proximity of a dominant conspecific. We and others have previously shown that hours to days after such social stress, there is an increased intake of high palatable foods such as fat^3,23.

We assessed whether social stress would immediately engage the mPFC-LHA pathway. To this end, we performed in vivo electrophysiological recordings from mPFC cells projecting to the LHA. We injected, in C57B6 mice, a canine adeno virus in the LHA to retrogradely deliver Cre (Cav2-Cre) to the mPFC, and an AAV in the mPFC to Cre-dependently drive an opsin (AAV-DIO-hChR2(H134R)-mCherry). We then chronically implanted electrodes together with an optic fiber (i.e., an optrode) in the mPFC of these mice (Fig. 2A, B). This allowed for optogenetic-assisted identification (i.e., opto-tagging) of mPFC cells that projected to the LHA (Fig. 2C; Supplementary Fig. 2A, B). We then exposed these mice to the sensory proximity of a CD1 mouse and the actual fight experience with such an aggressor, while performing single-cell recordings from opto-tagged mPFC-LHA cells (Fig. 2D). We analysed firing rates of opto-tagged neurons before, during, and after the period in which fighting took place. Divergent responses were observed in mPFC neurons projecting to LHA during stress, with approximately half of the cells decreasing their firing rates during fight periods, whereas others were not affected or showed increased firing rates (Fig. 2E, F). These effects were not solely linked to the physical altercation, as they outlasted the fighting period (Fig. 2E; Supplementary Fig. 2C, D). In further support of this, we also evaluated the responses of these same neurons when the CD1 mouse arrived in the cage, while still behind a semi-permeable barrier (i.e., sensory stress). We observed similar response types, where in particular the neurons that increased their firing during sensory stres, were modulated in the same direction during physical exposure (Supplementary Fig. 2E–H). Aside from evaluating opto-tagged mPFC neurons with confirmed projections to the LHA, we also evaluated putative pyramidal mPFC neurons that were not opto-tagged. Such neurons also responded with excitations or inhibitions to social stress (Supplementary Fig. 2I, J), although they were less often modulated by stress than the mPFC-LHA population (Supplementary Fig. 2I). Overall, these data indicate the existence of different populations of mPFC-LHA neurons, with certain mPFC-LHA neurons acutely decreasing their activity, and others directly increasing it upon social stress exposure. These findings confirm the hypothesis that mPFC cells that project to LHA are directly able to detect a social stressor with known hyperphagic properties^3,23.

mPFC-LHA pathway activity is indispensable for stress-driven excess fat intake

We showed that activity of the mPFC-LHA pathway is modulated by a social stressor. Next, we asked whether activity in this pathway is necessary for stress-driven increases in food intake. To this end, we bilaterally expressed an inhibitory chemogenetic receptor in the mPFC-LHA pathway in C57B6 mice. Specifically, we stereotactically injected Cav2-Cre retrograde virus in the LHA, and an AAV to Cre-dependently drive hM4Di (or a control fluorophore) in the mPFC (Fig. 3A, B). We confirmed that this indeed resulted in chemogenetic control over LHA-projecting mPFC cells. In brain slices, we patch-clamped such mPFC neurons and found that bath administration of DREADD agonist CNO resulted in strong hyperpolarization of the neurons (Supplementary Fig. 3A).

We then tested whether chemogenetic inhibition of the pathway would alter regular or stress-driven feeding behavior. We again placed animals in the 2-choice model (as in Fig. 1C) during a baseline session. Then we exposed the mice to either two days of social stress (which we previously showed to increase fat intake³) or to control exposure. The next day, we systemically administered CNO to the animals, and 45 minutes later reintroduced them to the 2-choice model (Fig. 3C). Mice with control mCherry vectors in the mPFC-LHA pathway exhibited clear stress-driven increased intake compared to their baseline. In mice with hM4Di in their mPFC-LHA pathway, we observed that inhibiting the pathway had no effect on food intake for non-stressed mice. Instead, in stressed mice it completely counteracted the amount of overconsumption that occurred after stress (Fig. 3C, D, Supplementary Fig. 3B). The effects of mPFC-LHA pathway inhibition were not due to actions on locomotor behavior or altered anxiety (Supplementary Fig. 3C, D). No differences were observed for chow intake (Fig. 3D). Overall, this suggests that mPFC-LHA pathway activity is not required for regular food intake, but becomes indispensable for regulating the excess food intake after stress.

mPFC neurons projecting to LHA respond to food intake

We next investigated whether the mPFC-LHA pathway would be responsive to food intake and stress-modulations thereof. We investigated the mPFC-LHA pathway in this case by measuring population-level calcium responses using fiber photometry. We stereotactically injected Cav2-Cre in the LHA, injected an AAV for Cre-dependent expression of calcium indicator GCaMP8s in the mPFC, and placed a fiber for photometric measurements above it (Fig. 3E; Supplementary Fig. 3G). Our in vivo electrophysiology data had shown that social stress is a modulating experience for this pathway, resulting at the single cell level at either excitation or inhibition in a considerable subset of cells (Fig. 2E, F). When we exposed mice to the social aggression of a CD1 mouse, we observed a biphasic calcium response of the mPFC-LHA at the population level to this stimulus, with a peak and a subsequent dip in calcium responses (Fig. 3F, G; Supplementary Fig. 3H). This confirms that with fiber photometric measurements, we can detect relevant dynamics in this pathway.

We then evaluated the response of the mPFC-LHA pathway to food intake. For that, we placed animals in the 2-choice task with access to fat and chow. We observed that when mice ate, and particularly when they ate fat, there were pronounced dips in calcium signals recorded from mPFC neurons projecting to LHA (Supplementary Fig. 3I). These were time-locked to the onset of food contact and waned after disengagement from the food source (Supplementary Fig. 3I). We then evaluated whether such responses, measured from mPFC somata, would be altered by prior stress. We exposed the mice again to a two-day social stress paradigm (or to control). We then recorded the day after, where normally there would be an increased food intake (Fig. 3D). However, we observed the same type of feeding-related dip responses in the pathway in control and in stressed mice (Fig. 3H–J). Hence, although the mPFC-LHA pathway is modulated during both stress and during feeding responses, and stress alters feeding responses, we did not observe signs of stress-driven plasticity in LHA-projecting mPFC soma reactivity during stress-driven hyperphagia. We next investigated whether stress-driven plasticity would instead occur downstream at mPFC synapses onto LHA neurons.

mPFC neurons make monosynaptic connections to multiple types of LHA neurons

Our findings demonstrate that the mPFC neurons projecting to the LHA respond to social stress and are capable of altering food intake, respond to food intake, and subsequently are necessary for stress-driven overconsumption of fat. However, no changes were observed in the somatic population activity of mPFC-LHA cells after stress while interacting with palatable food. This led us to hypothesize that plasticity of downstream communication with different populations in the LHA might underlie alterations in fat intake. However, it is unclear which LHA neuronal subtypes are innervated by the mPFC. The LHA contains both glutamatergic populations, expressing the vesicular glutamate transporter 2 (LHA_VGLUT2) and GABAergic populations, identified by expression of the vesicular GABA transporter (LHA_VGAT). Recent transcriptomic evidence suggests that essentially all LHA neurons can be ascribed to one of these major classes, with further key subpopulations occurring within them^15,18. For instance, evidence suggests that orexin (LHA_OREX) and melanin-concentrating hormone (LHA_MCH) releasing neurons represent distinct subsets of LHA glutamatergic (VGLUT2+) neurons, but also that most LHA_VGLUT2 neurons do not belong to those specific subsets¹⁸_. We reasoned that, despite further heterogeneity within LHA_VGAT, LHA_VGLUT2, LHA_MCH, and LHA_OREX neuronal sets^15,18, these four populations offer a starting point to understand mPFC functional innervation patterns of the LHA.

To map mPFC connectivity to these four LHA subpopulations, we performed an optogenetic-assisted synaptic connectivity study^24,25. We targeted the mPFC with the opsin CoChR (AAV5-Syn-CoChR-GFP) and fluorescently labeled LHA_VGLUT2, LHA_VGAT, LHA_MCH or LHA_OREX populations with viral and transgenic strategies. For LHA_VGLUT2 and LHA_VGAT populations we used well-established transgenic mice VGLUT2-Cre or VGAT-Cre mice, respectively (Fig. 4A, B). We injected these mice in the LHA with an AAV to Cre-dependently express a red fluorophore (rAAV-EF1a-DIO-mCherry) to label either LHA_VGLUT2 or LHA_VGAT neurons, respectively (Fig. 4A, B). Instead, to target LHA_MCH or LHA_OREX neurons we injected C57B6J mice with an AAV to drive expression of a red fluorophore under the control of short promotors for either MCH (AAV-MCHpr-Gq-mCherry) or orexin (AAV-hORXpr-TdTomato)²⁶ (Fig. 4A, B). We established the specificity of this approach using post-hoc labeling of biocytin-filled patched cells expressing either the fluorophore under control of the MCH or the orexin promotor (Supplementary Fig. 4A–C). We also established that the LHA_VGLUT2 cells that we on average targeted were distinct from the LHA_OREX and LHA_MCH cells, as those peptidergic subpopulations of glutamatergic neurons differed strongly in somatic cell size from typical LHA_VGLUT2 neurons (Supplementary Fig. 4D). The four distinct LHA populations also differed from each other in other biophysical properties, further corroborating that different LHA populations were sampled (Supplementary Fig. 4E–I). We also established that both LHA_VGAT and LHA_VGLUT2 neurons were able to follow protracted optogenetic stimulation of mPFC inputs at least for 1 and 5 Hz, whereas for 10 Hz fidelity of induced postsynaptic potentials remained higher onto LHA glutamatergic neurons (Supplementary Fig. 4J).

We then performed ex vivo patch clamp recordings to assess the occurrence of mPFC synaptic inputs onto these LHA subpopulations. We observed high degrees (≥80%) of connectivity between mPFC cells and all four LHA populations (Fig. 4C). We then repeated these experiments in the presence of a pharmacological cocktail of tetrodotoxin (TTX) and 4-aminopyridine (4-AP) in the external medium^24,27 and confirmed that the great extent of connectivity of mPFC inputs with all LHA subtypes was due to direct monosynaptic connections (Fig. 4D). Together, these data indicate that there are strong monosynaptic projections from mPFC cells to all four LHA subpopulations.

Stress selectively affects mPFC synaptic contacts onto LHA_VGLUT2 neuronal populations via presynaptic processes

After social stress, there is an increased drive for the intake of palatable food sources rich in sugar and fat^3,23, which ultimately results in excess body weight if maintained over weeks²³. In the fiber photometric experiment, we observed that at the level of mPFC cell bodies, the two-day stress model did not affect response magnitudes. However, it remains possible that downstream, at the synaptic level, changes occurred due to stress. We therefore next asked if, at the time of stress-driven hyperphagia, mPFC synaptic strength onto distinct LHA neuronal populations would be altered. To address this, we evaluated the effects of stress on mPFC synaptic contacts onto the four distinct LHA populations (LHA_VGLUT2, LHA_VGAT, LHA_MCH, LHA_OREX). We expressed an opsin in the mPFC (AAV-CoChR-GFP) and tagged distinct LHA populations as before (Fig. 4A, B; Supplementary Fig. 5A, B). We exposed mice to two days of control experience or social stress, and then the next day, when mice normally engage in higher intake of palatable food³ (Fig. 3D), we made brain slices and assessed synaptic properties at mPFC synapses onto the four broad classes of LHA neuronal populations (Supplementary Fig. 5A–C).

Interestingly, we observed that specifically at the mPFC synapses onto LHA_VGLUT2 neurons, there was, on average, a stress-induced decrease in maximally optogenetically-evoked EPSC magnitudes (Fig. 5A; Supplementary Fig. 5C). This suggests a weakening of glutamatergic mPFC synapses onto many LHA_VGLUT2 neurons. To further address the underlying mechanisms for this, we assessed the paired pulse ratio (PPR), a measure inversely related to presynaptic release probability²⁴. We observed that specifically at mPFC- LHA_VGLUT2 synapses there was an increased PPR, in accordance with a scenario of lower release probability at these synapses due to prior social stress (Fig. 5B; Supplementary Fig. 5D). The effect of stress was specific to the presynaptic compartment, as we did not observe a difference in postsynaptic glutamatergic receptor distributions at these synapses (Supplementary Fig. 5E). Nor did we observe changes due to stress in terms of the excitability of the LHA cells or in basic membrane properties (Supplementary Fig. 5F, G). These data suggest that social stress alters mPFC glutamatergic synapses onto LHA_VGLUT2 neurons. We next asked whether another stressor would similarly be able to induce these effects. For another stress experience of a different modality we exposed mice to restraint stress (Supplementary Fig. 6A), and we observed that this also subsequently increased intake of fat (Supplementary Fig. 6B). Similarly to social defeat stress, restraint stress also reduced the magnitude of mPFC synaptic responses onto LHA_VLUT2 neurons (Supplementary Fig. 6C, D), and it also increased the PPR at these synapses, indicating a similar mechanism of reduced release probability between the different types of stressors (Supplementary Fig. 6E). Overall, these data suggest that after stressful experiences, which can drive binge intake of fat, there is on average a weakening of mPFC glutamatergic synapses onto the global population of LHA_VGLUT2 neurons, which on average reduces food intake upon activation^15,17.

Social stress has divergent effects on distinct populations of mPFC-innervated LHA_VGLUT2 cells

LHA_VGLUT2 neurons still represent a very heterogeneous population¹⁸ and in part this heterogeneity is reflected by their distinct projection targets²⁸. For instance, stimulating LHA_VGLUT2 projections to the lateral habenula (LHb) suppresses food intake²⁹. Similarly, stimulating LHA_VGLUT2 neuronal projections to midline hypothalamic regions around the third ventricle (i.e., peri-paraventricular nucleus, or peri-PVN region³⁰) is linked to aversive processes³¹ and reduced food intake³². Instead LHA_VGLUT2 projections to the ventral tegmental area (VTA) do not alter food intake under naïve conditions^3,33,34, but the potentiation of their synapses onto dopamine neurons causally contributes to the enhanced fat intake occurring after social stress³. Given this LHA_VGLUT2 anatomical and functional divergence, we asked how social stress would affect mPFC synapses onto such distinct LHA_VGLUT2 populations. We first assessed if the LHb, VTA and (peri)-PVN could be downstream targets of mPFC-innervated LHA neurons. For this, we used a transsynaptic tracing strategy (Supplementary Fig. 6F) in which we injected AAV1-Cre in the mPFC, to cause a transsynaptic jump to downstream neurons³⁵. We then injected an AAV-DIO-GFP in the LHA to perform anterograde tracing of the nerve terminals belonging to LHA neurons with mPFC-innervation (Supplementary Fig. 6F, G). We indeed observed anterograde terminals in the regions of interest, like the VTA, the LHb and the peri-PVN (Supplementary Fig. 6H).

We then investigated if and what type of stress-driven synaptic changes occurred at mPFC synapses onto LHA_VGLUT2 neurons projecting to LHb, VTA, and peri-PVN. To label LHA_VGLUT2 cell types based on their projection output, we injected a Cre-dependent retrograde tracer (retroAAV-DIO-mCherry) in Vglut2-Cre mice in these three different areas (Fig. 5C, D). We then again exposed mice to either two days of control or social stress conditions and we prepared brain slices the day after and patched from retrogradely traced LHA_VGLUT2 neurons, assessing their mPFC input (Fig. 5C). The extent of mPFC-innervation of the three types of LHA_VGLUT2 neurons with distinct projections was high (Supplementary Fig. 6I). We observed that mPFC synapses onto LHA_VGLUT2-Peri-PVN neurons exhibited higher PPR upon stress (Fig. 5E), in accordance with the general effect of stress on the LHA_VGLUT2 neurons (Fig. 5B). We confirmed, via in vivo optogenetic stimulation of the LHA_VGLUT2 projection to the peri-PVN, that this pathway indeed reduces food intake upon activity (Supplementary Fig. 6J-L as previously suggested³². In contrast to the effect of stress on mPFC-LHA_VGLUT2-Peri-PVN synapses, we did not observe PPR changes at mPFC synapses onto LHA_VGLUT2-LHb neurons after stress (Fig. 5F). Interestingly, we observed that mPFC synapses onto LHA_VGLUT2-VTA neurons presented a lower PPR after stress, suggesting enhanced release probability of mPFC synapses onto those specific LHA_VGLUT2 neurons (Fig. 5G). In accordance, we had previously shown that stress-driven strengthening of LHA glutamatergic synapses onto VTA dopamine neurons is linked to stress-driven fat intake³.

Overall, these data suggest that social stress has divergent effects on the strength of mPFC synapses onto distinct LHA_VGLUT2 populations. Weakening excitatory mPFC synapses onto LHA_VGLUT2 populations with a role in reducing food intake³², while strengthening excitatory mPFC synapses onto LHA_VGLUT2 populations with a role in promoting food intake³.

After social stress, mPFC-LHA activity is more primed to drive feeding

We reasoned that the consequences of such stress-driven network modifications could prime the mPFC-LHA pathway to become even more capable of driving palatable food intake when stimulated.

To test this, we again optogenetically probed the mPFC-LHA pathway by stereotactically injecting C57B6J mice with an optogenetic construct (rAAV5-Syn-CoChR-GFP), implanting optic fibers bilaterally above the LHA (Fig. 6A, B). After recovery, we either exposed the mice to two days of social stress or housed them with a novel cage mate. Subsequently, we placed the animals in the 2-choice task,  where they could consume fat or chow for 2 h. Mice were again exposed to recurring blocks of optogenetic stimulation of 5 Hz, or instead to mock stimulation (Fig. 6A). We observed that even though mice that had previously experienced stress already consumed more fat than control mice, optogenetic stimulation could drive their intake of palatable food even further, making these mice the largest consumers of fat (Fig. 6C), in the absence of effects on chow (Supplementary Fig. 7A).

The overall fat intake reflected the outcome at the end of the two-hour binge period. We next addressed, with more fine-grained temporal resolution, whether optogenetic stimulation was more efficient in driving fat intake in stressed mice as compared to control mice. We first aimed to better understand the development of fat intake under stress (without pathway stimulation). We observed that regular (not stimulated) stressed animals performed over 75% of their fat feeding bouts in the first 20 minutes of a session, and then strongly diminished interaction with the fat from thereon (Fig. 6D, E). This binge-like feeding pattern of fat for stressed mice was different from that of control mice, which had fewer immediate feeding bouts but continued having feeding bouts for a longer period (Fig. 6D, E). We then analyzed how, in these two distinct phases that we refer to as “Binge Phase” (i.e., 0–20 min) and “Post-Binge phase” (20–120 min), mPFC-LHA pathway stimulation affected fat interaction. Reminiscent of our previous finding of pathway stimulation early in the 2-choice session in naïve mice (Fig. 1E), we found that 5 Hz mPFC-LHA stimulation was not effective in the early “Binge Phase” of the session (Fig. 6F). Instead in the subsequent “Post Binge” phase 5 Hz stimulation particularly drove stressed mice more potently towards the fat, when they would normally be in more sated conditions (Fig. 6F). To investigate whether indeed the 5 Hz stimulation could prevent a decline of interest for the fat, we plotted the relative drive for fat eating over the course of the task by comparing the relative amount of feeding bouts taking place during the binge and post-binge period. Indeed, particularly in stressed mice, 5 Hz stimulation did not affect the binge intake, but it did counteract the normally occurring decline of interest in the fat later on (Fig. 6G). Overall, these data reveal that after stress mPFC-LHA network activity is primed more towards an ability to drive continued fat intake.

A role for glutamatergic LHA neurons in transferring effects of stress into binge of fat

We further sought to investigate the importance of LHA neuronal subsets downstream of the mPFC in mediating stress-driven feeding. Glutamatergic LHA neurons are an interesting candidate, as they show  stress-driven plasticity of their mPFC synaptic inputs (Fig. 5A, B). Moreover, LHA_VGLUT2 neurons show a strong increase in calcium reactivity during social stress³. We first asked whether inhibiting the activity of LHA_VGLUT2 neurons during social stress would prevent subsequent binge eating. To test this, we injected a Cre-dependent inhibitory DREADD (AAV-DIO-hM4Di-mcherry) in the LHA of Vglut2-Cre mice (Fig. 7A, B). We subsequently subjected the mice to either the two-day social stress paradigm or its control. Prior to such stress (or control) events, we administered either vehicle or CNO. Chow intake in general was minimal and unaffected (not shown). Importantly, we observed that inhibition of LHA_VGLUT2 neurons during social stress episodes fully blocked subsequent binge eating of fat (Fig. 7C), indicating that these neurons are important for the induction of stress-eating.

We next evaluated whether the activity of LHA_VGLUT2 neurons could also contribute to the effects of optogenetic stimulation of the mPFC-LHA pathway on food intake. In Vglut2-Cre mice, we injected AAV-hSyn-CoChR-GFP in the mPFC and placed optic fibers in the LHA for pathway stimulation. We also injected an AAV-DIO-hM4Di-mCherry in the LHA to be able to chemogenetically inhibit LHA_VGLUT2 neurons (Supplementary Fig. 7B, C). We then assessed food intake during a baseline period in the 2-choice paradigm (Fig. 1C). After baseline periods, in randomized order, we re-exposed mice to the 2-choice paradigm during one of four conditions: optogenetic stimulation of the mPFC-LHA pathway or not, with chemogenetic inhibition of LHA_VGLUT2 neurons (CNO administration prior to the test) or not. When we did not inhibit LHA_VGLUT2 neurons, we again observed an increase in fat intake in response to pathway stimulation (Supplementary Fig. 7D). Chemogenetic inhibition of LHA_VGLUT2 neurons (without stimulation) also gave a trend towards increased food intake. If the sole role of LHA_VGLUT2 neurons were to inhibit food intake, then mPFC-LHA stimulation without LHA_VGLUT2 contributions (i.e., mainly favoring mPFC-innervated LHA_VGAT recruitment) should make mPFC-LHA stimulation even more potent in driving food intake. Instead, we observed that optogenetic stimulation of the mPFC-LHA pathway during LHA_VGLUT2 inhibition did not significantly further increase the intake of fat beyond LHA_VGLUT2 inhibition itself (Supplementary Fig. 7D). This did not appear to be due to a time-constraint limit on feeding, as overnight food-deprived animals still consumed considerably more in the same time period (Supplementary Fig. 7D).

Together, these results suggest that activity of LHA_VGLUT2 neurons during stress is required for subsequent excess food intake, and it also suggests that despite many LHA_VGLUT2 neurons curtailing food intake, the activity of at least some LHA_VGLUT2 subsets also participates in the mPFC-LHA pathway ability to drive food intake upon stimulation.

Stress-sensitive mPFC neurons project to glutamatergic LHA neurons and can drive increased intake of fat

We then employed another strategy to try to parse out the most relevant branches in the mPFC-LHA network involved in stress-driven food intake. For this, we used an activity-dependent genetic labeling approach and investigated the neurons in the mPFC-LHA axis sensitive to stress, to determine their role in feeding. We expressed an optogenetic construct in those mPFC cells that were reactive during the social stress (as opposed to all mPFC neurons). For this, we used TRAP2 (Targeted Recombination in Active Populations) transgenic mice, in which the immediate early gene c-Fos can drive Cre-ERT2 expression, which can then be conditionally transported from the cytosol to the nucleus in a manner dependent on 4-OH tamoxifen³⁶. We previously showed that this strategy can be leveraged to capture social stress-reactive neuronal ensembles³⁷. We crossed these TRAP2 transgenic mice with either VGAT-Cre or VGLUT2-Cre mice and used the resultant offspring. We first injected an AAV driving Cre-dependent CoChR in the mPFC and exposed the mice to two stress episodes, where each episode was paired with a 4-OH tamoxifen injection. Weeks later, we also injected AAVs in the LHA to fluorescently label specific subpopulations such as LHA_VGLUT2, LHA_VGAT, LHA_OREX, and LHA_MCH cells as before (Fig. 7D–F; Supplementary Fig. 7B, C). Importantly, while both TRAP2 transgenic mice and VGAT-Cre/VGLUT2-Cre transgenic mice rely on conditional expression via the same Cre-LoxP system, the viral injection scheme in this case avoided interference between them. First, we avoided mPFC CoChR expression in a manner not linked to the TRAPing of an ensemble in the VGLUT2-Cre mice, on the basis of mPFC neurons expressing VGLUT1 rather than VGLUT2^21,38 (Supplementary Fig. 7E). We also avoided that there would be Cre-dependent fluorophore recombination in the LHA driven by the TRAP system rather than on the basis of LHA cell type identity, by performing the LHA viral injection several weeks after the TRAPing event in the mPFC had taken place (Fig. 7D), well outside the time window of 4OH-tamoxifen dependent Cre-driven recombination^36,37. Finally, we also took along a control ensemble of exposure to a novel mouse rather than stressful altercations (Fig. 7D). In both the social stress TRAP experiment and the novel mouse TRAP experiment, we observed expression of CoChR in mPFC ensembles (Fig. 7E; Supplementary Fig. 7F). We then performed patch-clamp recordings in the LHA and assessed whether optogenetic stimulation of mPFC_{NOVEL MOUSE} or mPFC_{SOCIAL STRESS} ensembles resulted in synaptic contacts with specific LHA cell types. Stimulating the mPFC_{NOVEL MOUSE} ensemble caused relatively little connectivity with the LHA, though there was connectivity with LHA_MCH neurons (Fig. 7G, H). In contrast, optogenetically stimulating the mPFC_{SOCIAL STRESS} ensemble resulted in synaptic responses in LHA_VGLUT2, LHA_MCH, and LHA_OREX neuronal subsets (Fig. 7G, H). Instead, LHA_VGAT neurons received almost no input (Fig. 7G, H).

These results suggest that mPFC neurons that are modulated within the time window of several hours during social stress experiences send preferential projections to glutamatergic neurons (including their peptidergic subtypes¹⁸) over GABAergic LHA neurons. Given that we had shown the importance of glutamatergic LHA neurons in mediating the effects of stress on food intake (Fig. 7C), we next asked whether optogenetically stimulating mPFC_{SOCIAL STRESS} projections to these glutamatergic LHA neurons would modulate food intake. We used TRAP2 mice in which we again expressed an optogenetic construct (AAV-DIO-CoChR-GFP) in the mPFC, placed an optic fiber above the LHA, and triggered recombination of the opsin in the mPFC either by pairing the social stress experiences or the novel mouse experiences with 4OH-tamoxifen injections (Fig. 7I, J). After recovery, we optogenetically stimulated these pathways with 5 Hz in the context of the 2-choice food assay. We observed that optogenetic stimulation of the mPFC_{SOCIAL STRESS} to LHA pathway resulted in increased fat intake (Fig. 7K), without altering chow intake (Supplementary Fig. 7G). Together, these results suggest that glutamatergic LHA neurons are important downstream targets of the mPFC for the effects of stress on increased food intake.

The mPFC-LHA pathway has a conditional role in food intake, driving excess intake

Our pathway stimulation and inhibition experiments reveal that the mPFC-LHA network plays a conditional role in food intake. First, the role of the pathway in feeding was only unmasked by stress. Chemogenetic inhibition of the pathway under normal (non-stressed) conditions did not alter food intake, in accordance with an earlier study²¹. Instead, we show that after a social stress experience that drives excess intake of specifically palatable food³, inhibition of the pathway abolishes not all feeding, but rather only the amount of excess fat intake due to stress. This suggests that the pathway is not generally necessary for feeding behavior, but its importance emerges under certain conditions, such as a stressed state.

Second, mPFC-LHA pathway stimulation increased the intake of palatable fat, without affecting concurrently available chow. This indicates that the circuit manipulations caused a directed increase in drive towards palatable foods, rather than an indiscriminate urge to chew. Although not explicitly tested here, we consider it likely that the stress-driven effects mediated by the mPFC-LHA network result in an increased preference for palatable food in general, rather than a nutrient-specific preference for fat. In support of this, moderate social stress causes preference for palatable food in general, including both fat and sucrose³. Third, our data suggest that internal states may contribute to the role of the mPFC-LHA pathway in food intake. We observed that the optogenetic stimulation of the pathway did not cause altered food intake at the start of feeding sessions, when the drive for food intake was still high. Instead, it was only at later timepoints, when normally food interest would have waned, that pathway stimulation caused continued food interest and time spent eating. This finding also indicates that the role of mPFC-LHA networks will likely depend on whether or not animals are food-deprived. In our pathway manipulation experiments, animals were ad libitum-fed, perhaps allowing disinterest in food to occur more quickly under normal circumstances, and revealing a role for mPFC-LHA network activity more readily.

Our fiber photometric data show that the mPFC-LHA pathway shows a transient decrease in activity during food intake, which reverses upon the end of contact with the food, in accordance with a previous study²¹. It is of note that increasing mPFC-LHA pathway activity, at least at certain frequencies, drives food intake, whereas mPFC-LHA pathway activity decreases during the actual food intake. It is interesting to compare this with other systems with a link to feeding. For instance, arcuate nucleus AgRP neurons are potent drivers of food intake when stimulated yet also show dips in their activity during food intake. This may reflect an encoded energy need signal of these neurons that is then attenuated during caloric intake^40–43. Somewhat analogously, mPFC-LHA pathway activity may reflect not a generic energy deficit signal, but more a conditional one, promoting further intake of palatable energy-dense foods under specific circumstances, such as after stress. Another interpretation is that mPFC-LHA pathway activity reflects an engagement signal to continue to seek out palatable food instead of moving on to other activities. Actual consumption of the food may then require a brief decrease in such activity to allow for consumption instead of seeking/approach behaviors. Recent evidence does indeed suggest that cortical inputs to the LHA participate in the encoding of behavioral transitions, including those linked to feeding⁴⁴.

Overall, these data suggest that the mPFC-LHA network activity, while not indispensable for regular food intake, does become critically important for excessive intake during stress-driven hyperphagic states. This may be due to pathway activity indicating further energy needs, overriding satiety states, and/or by causing continued food engagement to facilitate excessive intake in short periods of time.

The mPFC-LHA is not a singular pathway, but a multi-branched network

Rather than a singular pathway, we now show that the mPFC-LHA network is composed of multiple branches that can have diverse functions. mPFC neurons innervate multiple LHA subpopulations based on LHA cell type. We observed strong connectivity with LHA_VGAT and LHA_VGLUT2 neurons, and with LHA_OREX and LHA_MCH neurons, which, according to transcriptomic analyses, represent glutamatergic VGLUT2-positive subsets¹⁸. In the current study, we also identified that mPFC-innervated glutamatergic LHA neurons project to multiple downstream targets, including VTA, LHA, and Peri-PVN. Within the current study, we did not further investigate distinct projection targets of mPFC-innervated LHA GABAergic or specific neuropeptidergic populations, but it is likely that these too will have multiple distinct targets¹⁵. Global mPFC-LHA pathway manipulation or measurement studies, need to therefore consider that the effects observed represent the net total of multiple sub-pathways, which likely can be in part functionally distinct from each other.

To what extent are these branches within the mPFC-LHA network anatomically distinct from each other, as opposed to representing collaterals? At the LHA level, prior evidence does suggest that, for instance, LHA glutamatergic neurons with distinct output targets are at least distinct subsets. Such as LHA_VGLUT2-VTA and LHA_VGLUT2-LHb neurons, which are anatomically and molecularly distinct sets²⁸. At the level of the mPFC, it is an open question to what extent different or the same mPFC neurons innervate distinct LHA neuronal subsets (e.g., LHA_GABA versus LHA_GLUTAMATE). However, our ensemble-based tagging experiments do indicate that there is at least no full collateralization. We observed that there are mPFC neurons that, when stimulated, do synapse onto glutamatergic but not on GABAergic LHA neurons, or that even synapse on specific neuropeptidergic populations such as LHA_MCH neurons.

Further studies will need to unravel which distinguishable mPFC neurons are embedded in different anatomical and functional branches within the larger mPFC-LHA network.

The mPFC projections to LHA and its glutamatergic neurons as a (social) stress integrator

In this study, we captured single-cell dynamics of mPFC cells projecting to the LHA during social stress experiences with in vivo electrophysiology. We observed a heterogeneous response of these cells, with some excited and some inhibited. Bulk photometric calcium recordings reflected a biphasic population level response, which on the basis of our in vivo electrophysiology data, we consider to be likely mediated by separate subsets of neurons. There was a modest enrichment of stress-modulated responses within the mPFC-LHA pathway compared to non-optotagged mPFC neurons, in accordance with findings that other mPFC pathways are also stress-sensitive⁷. Social stress is a naturalistic but complex stimulus (e.g., featuring stress, social interaction, pain, movement). Social interaction in itself can modulate mPFC-LHA^20,31,42, but it is likely that the stressful aspects of the stimulus are also engaging the network. Our patch clamp data also showed that restraint stress, which has very different features from social stress as an experience (e.g., non-social vs social, immobilization vs bursts of movement), caused increased food intake and similar synaptic alterations at mPFC-LHA_VGLUT2 synapses as did social stress. Also, a previous study has shown that the mPFC-LHA pathway is sensitive to another type of aversive stimulus, namely air puff stimuli²¹.

Our in vivo electrophysiology studies show that during stress, some neurons exhibit an increase in activity and others a decrease. Given the multiple types of connections made downstream in the LHA, it is possible that these different stress-induced responses in mPFC neurons correspond to distinct branches of the mPFC-LHA network. There is evidence to suggest that at least mPFC neurons projecting to LHA glutamatergic neurons are amongst those that are excited by social stress. Our previous work showed that glutamatergic LHA neurons themselves show an immediate strong excitation during social stress³, and we here show that a high percentage of these neurons (~80%) receive excitatory input from the mPFC. Furthermore, in the current study, we show that activity in these LHA glutamatergic neurons during social stress is indispensable for the subsequent hyperphagia. We also demonstrate that optogenetically stimulating stress-recruited mPFC ensemble projections to glutamatergic LHA subsets can drive feeding. Overall, these findings point to mPFC control over LHA glutamatergic neuronal subsets as a key substrate in driving stress-driven feeding.

Stress-driven plasticity within mPFC-LHA circuits

After social stress, at a time point where mice would overeat³, we observed synaptic changes at mPFC-LHA_VGLUT2 synapses. Specifically, we observed  evidence for weaker mPFC glutamatergic inputs onto the average LHA_VGLUT2 cell, via reduced release probability at mPFC presynaptic elements. Interestingly, many LHA_VGLUT2 cells are linked to a role in curtailing food intake^15,17. Consequently, a reduction of the excitatory input on such neurons could be in accordance with a weakening of a brake on the drive for food intake. A finer-grained analysis revealed that the effect of stress on mPFC synapses onto LHA_VGLUT2 cells was further dependent on the projection targets of the hypothalamic cells. We observed evidence for weakened glutamatergic mPFC synapses onto those LHA_VGLUT2 cells projecting to a midline hypothalamic region, which, upon stimulation, curtails food intake⁴⁵. Oppositely, we found that glutamatergic mPFC synapses onto LHA_VGLUT2 cells projecting to the VTA exhibit potentiation after stress. This latter finding is in accordance with our recent report that, after social stress, LHA_VGLUT2 projections to VTA dopamine neurons become potentiated and that this plays a causal role in excess food intake³. LHA_VGLUT2 cells have several more projection targets than the ones investigated here⁴⁶, which may also be changed by stress. Therefore, the analysis of these specific projections is not meant to be exhaustive, but it illustrates that mPFC synaptic contacts onto distinct LHA_VGLUT2 subpopulations are divergently affected by social stress.

Aside from mPFC-LHA branches that did exhibit stress-driven plasticity, there were also several that did not (e.g., mPFC synapses onto LHA_VGAT neurons). Of note is that even if mPFC synapses onto LHA_VGAT neurons did not, on average, show stress-driven plasticity, these connections are still engaged when stimulating the mPFC-LHA pathway, potentially contributing to increasing food intake. The findings of plasticity in certain branches should then also be observed in the context of the network as a whole. We do not claim here that one particular sub-branch is solely responsible for stress-driven feeding. Rather, stress has multiple actions in the network. mPFC-LHA branches (conditionally) linked to promoting food intake are either strengthened or unaffected by the stress experience, whereas branches linked to promoting food intake are either weakened or unaffected. As a whole, this likely shifts the mPFC-LHA network more towards favoring further intake of palatable food.

One relevant aspect that our study did not address is the extent to which individual animals may be more prone to exhibiting stress-driven overconsumption of palatable food, and whether this depends on individual differences in alterations in the mPFC-LHA pathway. Chronic social stress has been shown to have heterogeneous effects on mice in terms of willingness to engage in social interactions and extent of induced anhedonia. Classification systems of (distinct types) of stress-susceptible and unsusceptible animals^47,48 have been informative in this regard. Future studies will need to address whether moderate social stress affects increased palatable food intake, and/or related eventual weight gain over time²³ also clustering in distinct behavioral profiles.

Different types of mPFC-LHA stimulation result in different outcomes regarding food intake

Prior studies showed that chemogenetic (i.e., likely strong and continuous) or high-frequency optogenetic stimulation of the mPFC-LHA pathway reduced intake of chow in mice²¹. Here we show with our global pathway manipulations that optogenetic stimulation of the mPFC-LHA network, at lower frequencies, can also increase food intake. We found that continuous periods of 5 Hz stimulation drove food intake, whereas continuous periods of 1 Hz or 10 Hz stimulation did not. Interestingly, a previous study reported that 5 Hz optogenetic stimulation of LHA projections to the VTA could more potently drive food intake than either 2.5 Hz or 20Hz¹⁹. The lack of effect on food intake of mPFC-LHA with 1 Hz stimulation may have been due to insufficient engagement of the mPFC-LHA network. Continuous 10 Hz stimulation was qualitatively distinct from 5 Hz stimulation in terms of its effect on food intake. This could not be explained by an ineffective stimulation in itself, as animals exhibited more locomotor activity to the 10 Hz stimulation, which in itself might have contributed to the lack of increased intake of food as compared to stimulating the pathway with 5 Hz. At the mechanistic level, our ex vivo data showed, that over time, 10 Hz stimulation of mPFC inputs was less easily followed by LHA_VGAT neurons than by LHA_VGLUT2 neurons in terms of synaptic transmission fidelity. It is therefore possible that the lack of a 10 Hz effect was more due to the continuous nature of delivery rather than the frequency per se. These data, together with previous reports, show that distinct ways of mPFC-LHA stimulation can produce differential effects on feeding behaviors, possibly by striking a different balance in the sustained engagement of distinct mPFC-LHA subbranches.

Overall, our work identifies the mPFC-LHA as a multi-branched network, with an important role in particular for mPFC control over glutamatergic LHA neuronal populations with regard to integrating information about stress and the subsequent excess intake of palatable food sources such as fat. These findings may ultimately be relevant in the context of eating disorders, in which stress triggers bouts of binge eating behavior.

Animals

In all experiments, naïve adult male mice were used (20-35 g, >6 weeks). C57Bl6J (Jax #664), Vglut2-Cre⁴⁹ heterozygous (Jax #28863), VGAT-Cre⁴⁹ heterozygous (Jax #016962), and TRAP2 heterozygous³⁶ (Jax #030323), animals were bred in-house but originated from the Jackson Laboratory. Swiss-CD1 mice (35–45 g, >12 weeks), were purchased from Janvier (France). All mice (except Swiss-CD1 mice) were group-housed (2–5 per cage) unless otherwise specified. Mice were housed in a 12:12 light/dark cycle (lights on at 07:00 a.m.) at 22 ± 2 °C (60–65% humidity). Unless otherwise specified (during brief binge eating recording sessions, see below), animals had access to ad libitum water and lab chow (3.61 kcal/g, CRM (E), 801730, Special Diets Services). Experiments were approved by the local animal welfare body (AWB, or IvD in Dutch) of Utrecht University/ University Medical Center Utrecht, and by the national authority (CCD). Experiments were conducted in agreement with the Dutch law (Wet op de Dierproeven, 2014) and the European regulations (Guideline 86/609/EEC).

Stereotaxic surgery

For stereotactic surgery, all mice were anesthetized with ketamine (75 mg/kg i.p.; Narketan, Vetoquinol) and dexmedetomidine (1 mg/kg i.p.; dexdomitor, Vetoquinol). Lidocaine (0.1 ml; 10% in saline; B. Braun) was injected under the skin on the skull a few minutes before surgery. Eye ointment cream (CAF, Ceva Sante Animale B.V.) was applied before surgery. All mice were at least 6 weeks of age at the time of surgery. Animals were fixed on a stereotactic frame (UNO B.V.) and kept on a heat pad during surgery (33 °C). When animals were implanted with either optic fibers or cannulas, the skull surface was scratched with a scalpel, and phosphoric acid (167-CE, ultra-Etch, ultradent, USA) was applied for 5 min to roughen the surface at the start of surgery. Injections were done using a 31 G metal needle (Coopers Needleworks) attached to a 10 µl Hamilton syringe (model 801RN) via flexible tubing (PE10, 0.28 mm ID, 0.61 mm OD, Portex). The Hamilton syringe was controlled by an automated pump (UNO B.V. -model 220). Injections were done bilaterally unless otherwise specified, and volumes ranged between 250 and 300 nl per side, at an injection rate of 100 nl/min. The injection needle was retracted 100 µm, 9 minutes after the infusion, and withdrawn completely 10 minutes after the infusion. The skin was sutured post-surgery (V926H, 6/0, VICRYL, Ethicon). Animals were then subcutaneously injected with the dexmedetomidine antagonist atipamezole (50 mg/kg; Atipam, Dechra), carprofen (5 mg/kg, Carporal), and 1 ml of saline and left to recover on a heat plate of 36 °C. Carprofen (0.025 mg/L) was provided in the drinking water for one week post-surgery. Animals were single-housed post-surgery and rejoined with their cage mates three days after surgery, unless otherwise stated.

For ex vivo optogenetic-assisted electrophysiology experiments

For transgenic lines: VGAT-Cre and Vglut2-Cre, mice were injected with AAV5-Syn-CoChR-GFP (3.8*10^12 gc/ml; 300 nl; UNC Vector Core) in mPFC (+1.8 mm anterior to bregma; 0.90 mm lateral; -2.2 mm ventral from skull under an angle of 10°) and injected with AAV5-hSyn-DIO-mCherry (3*10^12 gc/ml; 300 nl; Addgene) in LHA (-1.3 mm posterior to bregma; 1.9 mm lateral; −5.4 mm ventral from skull under an angle of 10°) to visualize GABAergic neurons in VGAT-Cre mice and glutamatergic neurons in Vglut2-Cre mice.

To visualize neuropeptide populations in LHA, we used C57bl6 mice, which were injected with the same protocol for the mPFC as the Vglut2-Cre and VGAT-Cre. Mice were injected with rAAV5-MCHpr-Gq-mCherry (1*10^12 gc/ml) for labelling the MCH populations and with AVV5-hORXpr1-3TdTomato (1*10^12 gc/ml) for the Orexinergic neurons. Both were injected in the LHA with the same coordinates and volume as for the transgenic mice.

To identify glutamatergic LHA neurons projecting to LHb, VTA, and Peri-PVN, we used a Cre-dependent retrograde tracer retroAAV-hSyn-DIO-mCherry (2.8*10^12 gc/ml; Addgene) in Vglut2-Cre mice. Mice were then injected in one of these areas: in the LHb (-1.6 mm posterior to bregma; 1.2 mm lateral; -3.05 mm ventral from skull under an angle of 15°), in VTA (-3.2 mm posterior to bregma; 1.65 mm lateral; -5.0 mm ventral from skull under an angle of 15°) and Peri-PVN (-0.8 mm posterior to bregma; 1.1 mm lateral; -5.4 mm ventral from skull under an angle of 10°). Mice were also injected in the mPFC with AAV5-Syn-CoChR-GFP.

Patch-clamp Electrophysiology

After a minimum of six weeks following surgery, mice were anesthetized with isoflurane (Zoetis, UK) in the morning and were then rapidly decapitated. Coronal brain slices of 250 μm were cut on a vibratome (1200 VTs, Leica, Rijswijk, The Netherlands) in ice-cold carbogenated (95% O₂, 5% CO₂) cutting solution, which contained (in mM) choline chloride (92); ascorbic acid (10); CaCl₂ (0.5); glucose (25); HEPES (20); KCl (2.5); N-Acetyl L Cysteine (3.1); NaHCO3 (25); NaH2PO4 (1.2); NMDG (29); MgCl₂ (7); sodium pyruvate (3); Thiourea (2). Slices were transferred for 5–10 min to a warmed solution (34 °C) of the same composition, before they were kept at room temperature in carbogenated incubation medium, which contained (in mM): ascorbic acid (3); CaCl₂ (2); glucose (25); HEPES (20); KCl (2.5); NaCl (92); NaHCO3 (20); NaH2PO4 (1.2); NMDG (29); MgCl₂ (2); sodium pyruvate (3) and Thiourea (2). During recordings, slices were immersed in artificial cerebrospinal fluid (ACSF) containing (in mM): NaCl (124); KCl (2.5), CaCl₂ (2.5); glucose (11); HEPES (5); NaHCO3 (26); NaH2PO4 (1); MgCl₂ (1.3) and were continuously superfused at a flow rate of 2 ml min⁻¹ at 32 °C.

Neurons were patch-clamped using borosilicate glass pipettes (2.7–4 MΩ; glass capillaries, GC150-10, Harvard apparatus, UK), under a BX51WI Olympus microscope (Olympus, France). For voltage or current clamp recordings, the signal was amplified, low-pass filtered at 2.9 kHz with a 4-pole Bessel filter, and digitized at 20 kHz with an EPC9/2 dual patch-clamp amplifier (HEKA Elektronik GmbH). Data were acquired using PatchMaster v2x78.2 software. After the break-in, 10 minutes of waiting time occurred prior to the start of the recording. Series resistance was determined in Voltage clamp with a − 4 mV step from −65 mV to −69 mV lasting 50 ms, using the resultant capacitive current transients and Ohm’s law. Light pulses (470 nm, 1 ms; 25.4 mW) were delivered with a pE-4000, CoolLed, UK illumination system. All electrophysiological measures are recorded with a 10 s inter-sweep interval (0.1 Hz), and all data points are shown as an average of 10-20 sweeps, unless otherwise specified.

mPFC-LHA connectivity experiments

Recordings were made in voltage clamp in a cesium methane sulfonate based internal medium as described above. Under these conditions, it was possible to detect inward glutamatergic currents (at −65 mV). Cell type specific connectivity was determined based on whether a neuron showed an optogenetically-evoked synaptic response larger than 5 pA. For monosynaptic confirmation, cells were recorded and opto-stimulated (1 ms) once per sweeps, first during at least 15 sweeps with ACSF, then 1 µM of TTX was added to ACSF for a minimum of 25 sweeps to observe the effect on the synaptic response and finally 100 µM 4-AP was added to the mix and recorded for a least 20 sweeps.

Spike fidelity over 10 min

Recordings were made in current clamp in a potassium gluconate-based internal (see above), cells were clamped at 0pA, except if they were spontaneously firing (-20pA or -30pA); and were opto-stimulated during 10 min at 1 Hz, 5hz and 10 Hz in a random order. EPSP was counted if the change of amplitude was equal to or more than 2 mV compared to the overall baseline.

In vivo electrophysiology recordings

In vivo fiber photometry recordings

We used a 2-site photometry system (Doric lenses) to detect calcium and non-calcium-dependent dynamics (GFP and isobestic channel). Two light-emitting diode delivered blue (465 nm, 40 ± 1 μw in the brain) and purple (405 nm, 20 ± 1 μw in the brain) light, which was guided to the mouse via a mini cube (ilFMC5-G2_IE(400-410)_E(460-490)_F(500-540)_O(580-680)_S) and optical patch cord (ø400 μm). Emitted light was detected with the mini-cube integrated photodetectors and transmitted to the console with different reference frequencies (GFP 572 Hz, ISO 209 Hz). The photometry system was controlled by Doric Studio software (version 6.3.2.0), which recorded 243 samples per second (50 times decimated). The relative changes in light (ΔF/F) were calculated with custom-made Python scripts (version 3.13.5). Code accessible here: 10.5281/zenodo.18670225.

First, outliers were removed (>4 std aways from the mean signal of 9 minutes) and data was smoothed (per 100 data points). We calculated the ΔF/F for each trial with -10 to -2 seconds before the onset as a baseline. Next, the session average and peak signals were extracted.

Behavioral paradigms

All behavioral manipulations and tests were performed during the light phase.

Two-day Social stress paradigm

The social subordination stress protocol used was designed according to a resident intruder paradigm. The resident, a Swiss-CD1 male previously used for breeding, was housed in a large Makrolon cage (type IV, Tecniplast, Italy). An intruder, an experimental male mouse, was placed in this cage. Fighting time was tracked live and animals were allowed to fight for 20 s, twice a day for two days, one fighting session in the morning (8.30–11.00 a.m.) and one in the afternoon (16.00–18.30 p.m.). For the remainder of the day, animals were separated by a perforated transparent splitter, allowing sensory, but no physical interaction. The control animals were co-housed with a non-familiar C57B6J male mouse for two days, preventing physical interaction. This short stress paradigm has been shown not to change body weight in the first days after the stressful experience when food intake is measured³. Consistent with this, no difference in body weight was observed between the experimental groups in this study. The schematics on social stress (Fig. 2D and Supplementary Fig. 2C) were created in BioRender. Meye, F. (2026) https://BioRender.com/mula9s9.

In vivo circuit approaches during behavioral experiments

Histology and microscopy

For each recording, the virus injection sites, as well as the placement of optic fibers and/or the optrodes, were examined. Only data obtained from mice with correct placements were included. After experiments, mice were anesthetized with pentobarbital (75–100 mg/kg) and transcardially perfused with Phosphate-Buffered Saline (PBS) containing 4% paraformaldehyde (PFA). Brains were post-fixed in 4% PFA overnight and transferred to PBS. Before slicing, brains were embedded in 4% agarose to maintain structural integrity. Coronal sections of the entire brain were sliced at 60 μm using a vibratome (VT1000s, Leica, Rijswijk, The Netherlands). For patch clamp experiments, out of 272 mice, 27 were excluded on the basis of misinjection of viruses. For in vivo experiments, a total of 13 mice were excluded based on histological assessment. For histology, we used the Paxinos & Franklin mouse brain atlas (The Mouse Brain, in stereotaxic coordinates, 2nd ed, 2001, Academic Press).

Immunohistochemical staining for histology experiments

To stain for Channelrhodopsin, hM4DGi or control mCherry virus, tissue sections were washed in PBS (0.01 M, pH = 7.4) three times for 10 min before incubating them for 2 h with blocking solution (3% BSA, 0.5% Triton-X in 1x PBS). The primary antibody, Rabbit Anti-RFP (Rockland, 1:1000 in blocking solution, Lot:46510) that targets mCherry, was administered to the tissue sections using ~400 ul per well. Incubation was carried out overnight at 4 degrees on a shaker to facilitate optimal antibody priming. The next day, tissue sections were washed three times with PBS before applying a secondary antibody (goat-anti-Rabbit Alexa-488, Abcam, 1:500, Lot:GR3375958-2, Lot:GR3442006-1). The sections were incubated with the secondary antibody for two hours before being washed in PBS again for 3 times. Finally, tissue slices were mounted on glass slides and embedded in FluorSave (Millipore, 345789, Amsterdam, Netherlands) and covered with cover glasses. The microscopy slides were allowed to air dry overnight at room temperature before being stored in a 4-degree environment to maintain sample integrity. For staining of CoChR-GFP, anti-GFP antibody was used (Aves, GFP-1020, 1:1000 in blocking solution) and goat-anti-chicken Alexa-568 (Abcam, 1:500) as a secondary antibody.

Data analysis and statistics

Experimental data were analyzed with Ethovision (V9.0, Noldus, NL), GraphPad (v9.0, Dotmatics, UK), SPSS (v26, USA), PatchMaster (v2x78.2, HEKA, USA), Python (v3.13.5), Igor Pro-8 (Wavemetrics, USA), FIJI (v2.14), and Matlab (vR2023B/vR2024B, Mathworks, USA). Experimental sample sizes were predetermined using power calculations (requiring power >0.8). These power calculations were performed together between the researchers and the local animal welfare body (AWB Utrecht). These calculations were guided by published studies, in-house expertise, and experimental pilots. Mice were randomly assigned to experimental groups. Mice or data points were not excluded from analyses unless noted. Behavioral and physiological experiments were always performed in multiple batches of animals (≥2), ensuring also internal replicability. Bar graphs are always reported and represented as mean ± SEM, with single data points plotted (single cell for electrophysiology and single animal for behavioral experiments). When statistically appropriate, One-way ANOVAs, Two-way ANOVAs, Multi-way Repeated-measures ANOVAs, or t-tests were applied. Post hoc analyses were performed in case of significant omnibus ANOVA outcomes. In the event the data were not normally distributed, statistical testing of group differences was performed using the Mann-Whitney U or Kolmogorov-Smirnov test. Testing was always performed two-sided with α = 0.05. Complete outcome of statistical analysis is shown in Supplementary Data 1.

Reporting summary

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

Peer review information

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