Distinct ventral tegmental area neuronal ensembles are indispensable for reward-driven approach and stress-driven avoidance behaviors

Stimuli of opposite valence differentially affect approach/avoidance behavioral domains

To determine the neuronal ensembles in the VTA that respond to a rewarding or an aversive stimulus, we first established two contrasting experiences. We tested whether a single systemic injection of DAMGO, a selective μ-opioid receptor agonist, or an acute social stressor would trigger contrasting effects on behavioral tests for valence, anxiety and feeding.

We first tested C57BL/6 J mice using a single exposure place preference (or avoidance) paradigm (sePP/sePA), as described before^25,26. For DAMGO-conditioning, mice were randomly placed in one of two compartments on day 1, after either an i.p. DAMGO (or vehicle) injection. On day 2, the mice received vehicle (or DAMGO) in the other compartment. On day 3 we assessed their preference score for the compartments when they were free to explore (Fig. 1A). Mice spent significantly more time in the compartment previously paired with a DAMGO injection (Fig. 1B). For social stress-conditioning, mice were exposed to either an aggressive CD-1 mouse in a compartment, or were in a compartment without another mouse. Also these experiences were randomly counterbalanced across day 1 and 2. We observed that on day 3, these mice avoided the compartment formerly paired with this social stress experience (Fig. 1C).

We then assessed whether DAMGO and acute social stress would elicit opposite effects on anxiety-like behavior in an open field (OF) test (Fig. 1D). Mice receiving DAMGO prior to the OF, as compared to vehicle, spent more time in the center of the arena, suggesting that systemic DAMGO administration induced an anxiolytic phenotype (Fig. 1E, F; Suppl. Figure 1A). This occurred in the absence of general locomotor effects in the arena (Suppl. Figure 1B, C). Conversely, mice that were acutely stressed spent less time in the center of the OF compared to their respective controls (Fig. 1E, G; Suppl. Figure 1D), indicating that acute social stress induced an anxiogenic phenotype. This was accompanied by a stress-induced reduction in general locomotion along with heightened immobility, as shown before²⁷ (Suppl. Fig. 1E, F).

Finally, we examined whether these experiences acutely alter the intake of palatable food (Fig. 1H). Animals were administered DAMGO or were exposed to social stress, and were subsequently placed in a novel cage with both their regular chow pellet diet and a palatable food reward (fat). We then monitored their food choices and intake. Neither DAMGO nor social stress altered chow intake (Suppl. Fig. 1G, H). Instead, we found that DAMGO increased fat intake, compared to baseline and to a saline injection (Fig. 1I). Conversely, mice that were acutely stressed consumed significantly less fat in the first hour post-stress, compared to their baseline intake levels and to mice in the control condition (Fig. 1J).

Overall, these data suggest that administration of DAMGO and acute exposure to social stress are experiences with opposite effects on valence, anxiety levels, and food intake.

Stimuli of opposite valence engage distinct VTA ensembles

Having established that systemic DAMGO and acute social stress represent opposite experiences for the mice, we then asked whether these stimuli would recruit distinct neuronal ensembles within the VTA. To this end, we injected mice with DAMGO or exposed them to an acute social stress episode and subsequently perfused them (Fig. 2A). We then did immunohistochemical staining of VTA slices against Fos to gauge neuronal activation, and against tyrosine hydroxylase (TH) to demarcate the boundaries of the VTA (Fig. 2B). We previously showed that stress elevates Fos levels in the VTA²³. Here we observe that DAMGO led to similar amounts of Fos positive neurons in VTA as did social stress (Fig. 2C). We from here on refer to these collectives of cells as the VTA_DAMGO and VTA_STRESS ensembles, respectively.

We then characterized the cell types within these VTA_DAMGO and VTA_STRESS ensembles. Similarly to our previous findings on social stress ensembles²³, the VTA_DAMGO ensemble was found to be predominantly composed of neuronal cells ( ~ 98% NeuN positive; Suppl. Fig. 2A, B). The size of this ensemble was ~10% of all VTA neurons (Suppl. Figure 2C), similar to what we previously reported for the VTA_STRESS ensemble ( ~ 11%)²³. Having established that both ensembles are predominantly neuronal, we then assessed the neurotransmitter content within them. To identify VTA_DA neurons we performed immunohistochemistry against TH in C57BL/6 J mice. To identify VTA_GABA and VTA_GLU neurons we used vesicular GABA transporter (vgat)-Cre and vesicular glutamate 2 transporter (vglut2)-Cre transgenic mouse lines, respectively. We injected these in the VTA with an Adeno Associated Virus (AAV) to Cre-dependently drive expression of a fluorophore (AAV-DIO-mCherry) (Fig. 2D, Suppl. Fig. 2D–E).

Both ensembles contained the same extent of dopaminergic ( ~ 50%) and GABAergic ( < 12%) neurons (Fig. 2E). Both ensembles also contained a similar extent of combinatorial glutamatergic/dopaminergic neurons ( < 12%). The only difference between the ensembles was that the VTA_DAMGO ensemble had a slightly higher degree of glutamatergic/TH-negative cells (17% VTA_DAMGO vs 7% in VTA_STRESS) (Fig. 2E). We next analyzed the topographical position of VTA_DAMGO and VTA_STRESS cells. Such analysis, across the rostral-caudal axis, revealed no significant differences in the average position of DAMGO- and stress-ensemble cells in the medio-lateral and dorso-ventral axis (Suppl. Figures 2F–H). Overall, the ensembles therefore contained largely similar NT-defined cell types and had a similar anatomical topography.

Next, we evaluated whether these two ensembles, which on multiple levels of analysis appear similar, were actually distinct from each other (or instead were the same overlapping ensemble). To address this, we sought to determine the extent to which an ensemble for stress would show reactivity to reward, and vice versa. For this we used Targeted Recombination in Active Populations (TRAP2)²⁴, a technique that employs tamoxifen-inducible Cre recombinase under the control of an immediate early gene promoter to label neurons activated during specific experiences. By combining TRAP2 with subsequent Fos immunohistochemistry, we aimed to capture and compare neuronal ensembles activated by two distinct experiences at different time points (Fig. 2F).

First, we showed that both DAMGO and acute social stress, followed by an injection of 4-hydroxytamoxifen (4-OHT), led to tagging of VTA neuronal ensembles (Fig. 2G). The number of TRAPed cells for either DAMGO or social stress was approximately one third of that of the endogenous Fos-expressing neurons (Suppl. Fig. 2I, J; see Fig. 2C), as we saw before for the social stress experience²³. Notably though, the TRAP-captured ensembles for both experiences were still of similar size (Suppl. Figure 2K). Furthermore, similarly to endogenous Fos expression (Fig. 2E), approximately half of each TRAPed ensemble was dopaminergic (Suppl. Figure 2L). We evaluated, >3 weeks later, for these two ensembles how they responded in terms of Fos expression to a repetition of the same stimulus, or instead to the other stimulus. We observed that the VTA_DAMGO ensemble was more reactive to a subsequent DAMGO exposure than to subsequent stress (Fig. 2G, H, Suppl. Figure 2I). Similarly, the VTA_STRESS ensemble was more reactive to subsequent social stress than to subsequent DAMGO (Fig. 2G, H, Suppl. Figure 2I). Importantly, these differences in preferential activation patterns were not due to any different extents of overall VTA Fos or TRAP densities between the groups (Suppl. Fig. 2J, K).

To further investigate valence-specific encoding in ensembles initially exposed to either DAMGO or social stress (TRAP), we examined their responses to congruent valence experiences beyond re-exposure to the same stimulus. Mice initially exposed to DAMGO received a 2-hour food reward session (fat, after overnight food restriction), while those exposed to social stress experienced an aversive foot shock. We then perfused the mice for Fos staining. VTA ensembles responded preferentially to these congruent valence experiences (food reward for DAMGO-exposed mice; foot shock for social stress-exposed mice) versus incongruent ones (social stress or DAMGO, respectively) (Suppl. Fig. 2M–P). To determine whether the extent of overlap of TRAP and Fos signals for non-congruent valence stimuli (e.g., DAMGO TRAP and social stress Fos or vice versa) was higher than in naïve conditions, we also examined the overlap of these signals in response to home cage exposure. We observed that this also resulted in comparable levels of overlap as to that observed with incongruent valence stimuli, whereas higher overlap was observed with congruent valence stimuli (Suppl. Fig. 2Q; Fig. 2H).

Overall, these data suggest that DAMGO and acute social stress, which produce opposite behavioral phenotypes, recruit two functionally distinct VTA neuronal ensembles, that have similarities in terms of NT-content, neuronal density, and topography.

VTA_DAMGO and VTA_STRESS ensembles differentially respond to stimuli of positive and negative valence in vivo

Our reactivation studies based on Fos, an indirect and static proxy of elevated neural activity, suggested preferential recruitment of the two VTA ensembles by stimuli of congruent valence. Next, we sought to extend these findings, by assessing with dynamic measurements whether the VTA_DAMGO ensemble would be (preferentially) acutely activated by another reward, and the VTA_STRESS ensemble by another aversive stimulus. To examine this we measured the activity of VTA_DAMGO or VTA_STRESS ensembles during the intake of palatable food or during foot shocks. To this end, TRAP2 mice were stereotactically injected in the VTA with an AAV driving Cre-dependent recombination of a genetically encoded calcium indicator (AAV-FLEX-jGCaMP8s). An optic fiber was also implanted above the VTA (Fig. 3A, B, Suppl. Figure 3A). After a recovery period, mice were exposed to a DAMGO injection or to social stress followed by a 4-OHT injection, to drive expression of jGCaMP8s specifically in VTA_DAMGO or VTA_STRESS ensembles (Fig. 3A, B).

We first assessed whether the two ensembles are activated by palatable food intake. We placed mice in an apparatus in which there was fat in one of a total of five ports (the other 4 ports were empty), each with infrared beams to detect entry. We then time-locked entry of the mouse in the reward-containing port while measuring GCaMP8s responses of the VTA ensembles. We observed that VTA_DAMGO neurons showed calcium transients around interactions with the fat port (Fig. 3C, D, Suppl. Figure 3B). On the other hand, VTA_STRESS neurons showed no calcium activity around interaction with the fat (Fig. 3C, D, Suppl. Figure 3C), suggesting that the ensembles differ in their encoding of this rewarding stimulus. Importantly, nose-poking into an empty (no fat) port did not cause a calcium response in the VTA_DAMGO ensemble (Suppl. Figure 3D). We then exposed mice to electric foot shocks and recorded GCaMP8s activity around the time of shock onset. The VTA_STRESS ensemble was robustly activated by foot shocks (Fig. 3E, F, Suppl. Figure 3E). Although the VTA_DAMGO ensemble also showed reactivity to the foot shocks, this fell short of significance, and this response was significantly smaller than the response by the VTA_STRESS ensemble (Fig. 3E, F, Suppl. Figure 3E, F).

Overall, our results demonstrate that VTA_DAMGO neurons also show activity during interactions with another positive valence stimulus (a naturalistic food reward), while VTA_STRESS neurons are also strongly reactive to another negative valence stimulus other than social stress (foot shocks).

Stimulation of VTA_DAMGO and VTA_STRESS ensembles drives congruent approach/avoidance behaviors

We next investigated whether specifically stimulating VTA_DAMGO and VTA_STRESS ensembles could recapitulate the distinct valence/anxiety/feeding phenotypes, that we observed earlier after exposure of the animal to the integral experience of either systemic DAMGO or social stress (see Fig. 1B–J). For that we bilaterally injected TRAP2 mice in the VTA with a Cre-dependent activating chemogenetic vector (AAV-DIO-hM3Dq) or a Cre-dependent control vector (AAV-DIO-mCherry). We then exposed the animals to a DAMGO injection or to acute social stress, followed by a 4-OHT injection, for permanent expression of activating DREADD or control proteins in VTA_DAMGO and VTA_STRESS ensembles (Fig. 4A, B).

We first gauged the valence associated with activation of each ensemble, using a conditioned place preference/avoidance paradigm (CPP/CPA; Suppl. Figure 4A). Following a pre-test session on day 1, mice were conditioned for 3 days in the CPP apparatus, where DREADD agonist compound 21 (c21) injection was paired with a specific compartment. During test day (day 5), mice were free to roam in both compartments. Mice showed significant preference for the compartment that had been paired with chemogenetic activation of the VTA_DAMGO ensemble. Conversely, mice showed significant avoidance of the compartment that had been paired with chemogenetic activation of the VTA_STRESS ensemble (Fig. 4C, D). This indicates that activation of the VTA_DAMGO ensemble is rewarding, while activation of the VTA_STRESS ensemble is aversive. Control mice that only expressed an mCherry fluorophore in the VTA_STRESS ensemble showed no significant preference for any compartment, indicating that c21 alone has no effect on valence processing (Fig. 4D). Importantly, these effects could not be explained by difference in viral expression (ensemble size), as no significant correlation was found between hM3Dq expression (ensemble size) and CPP/CPA score (Suppl. Figure 4B).

Next, we assessed whether the activation of these two VTA ensembles would alter general anxiety in the OF test (Fig. 4E; Suppl. Figure 4C). Administration of c21 in control AAV-DIO-mCherry-expressing mice produced no effect on anxiety-like behaviors (Fig. 4F, Suppl. Figure 4D). Notably, VTA_DAMGO ensemble activation led to anxiolysis (as shown by increased time spent in the center of the OF arena), whereas activation of the VTA_STRESS ensemble resulted in anxiogenesis, as assessed by decreased time in the center of the arena (Fig. 4G, H, Suppl. Fig. 4E, F). Importantly, locomotor parameters (distance moved, immobility) were unaffected and could not explain the differences between groups (Suppl. Figure 4D–F).

Subsequently, we determined the effect of ensemble activation on reward intake in the palatable food choice paradigm. The mice were placed daily in a novel cage for a period of 1 h where they had free access to both their regular chow food and palatable fat reward. Administration of c21 in mCherry-expressing controls produced no effects on food intake (Fig. 4I, Suppl. Figure 4G). Activation of the VTA_DAMGO ensemble had no effect on fat intake (Fig. 4J, Suppl. Figure 4H), in contrast to systemic DAMGO-driven hyperphagia (see Fig. 1I). Notably, activation of the VTA_STRESS ensemble led to a decrease of fat intake compared to baseline consumption (Fig. 4K, Suppl. Figure 4I), in accordance with the acute stress-driven hypophagia (see Fig. 1J).

To further explore the roles of VTA_DAMGO and VTA_STRESS ensembles in encoding (aspects of) positive or negative valence, an operant task involving intracranial self-stimulation (ICSS) of each ensemble was employed. TRAP2 mice were injected with AAV-hSyn-CoChR-GFP in the VTA and optic fibers were bilaterally implanted above the VTA (Fig. 4L, M). After a recovery period mice were exposed to DAMGO or social stress followed by an injection of 4-OHT, to express an opsin in VTA_DAMGO or VTA_STRESS neurons, respectively. Mice were then placed in an operant cage where nose pokes in an active port of a 5-port wall resulted in ensemble stimulation (473 nm laser). Nose-poking in any of the 4 non-active ports had no programmed consequences. We found that mice with the stimulatory opsin in the VTA_DAMGO ensemble nose poked significantly more in the active versus the inactive ports (Fig. 4N top; Suppl. Figure 4J top). To rule out potential confounds linked to light stimulation, we showed that mice instead did not nose poke when this was paired with laser output at a wavelength that was not effective for opsin stimulation (i.e., 532 nm light delivery) (Suppl. Figure 4K). Furthermore, we found that mice did not seek optogenetic stimulation of the VTA_STRESS ensemble (Fig. 4N top, Suppl. Figure 4J top).

In a subsequent experiment we switched the contingencies of stimulation, such that now mice experienced continuous stimulation of the ensemble with the 473 nm laser in the operant chamber. Under these conditions, nose-poking in the active port resulted in laser deactivation for 20 sec. We found that mice with the opsin in the VTA_STRESS ensemble nose poked significantly more in the active port to pause stimulation, whereas this did not occur for mice with the opsin in the VTA_DAMGO ensemble (Fig. 4N bottom, Suppl. Figure 4J bottom).

Together these data suggest that activation of the VTA_DAMGO ensemble mimics the rewarding and anxiolytic properties of acute DAMGO administration (see Fig. 1B, F). However, VTA_DAMGO ensemble activation is not sufficient to alter food reward intake, which is heightened by systemic administration of DAMGO (see Fig. 1I). Conversely, stimulation of the VTA_STRESS ensemble is sufficient to recapitulate all the effects of acute social stress exposure (Fig. 1C, G, J). Namely, VTA_STRESS stimulation is aversive, drives anxiety-like behavior, and causes hypophagia. These results establish the potency with which intermingled VTA ensembles can orchestrate opposite approach versus avoidance behavioral patterns.

Activity of VTA_DAMGO and VTA_STRESS ensembles is indispensable for reward and stress-driven approach/avoidance behaviors

We showed that stimulating VTA_DAMGO and VTA_STRESS ensembles recapitulates many of the approach/avoidance behaviors of systemic DAMGO and acute stress itself. However, DAMGO- and stress-induced behavioral effects will involve ensembles across multiple brain regions. Therefore we now asked whether activity in these small ensembles in the VTA were indispensable for the integral DAMGO and stress-driven behaviors to occur. First we established that neither the repeated chemogenetic inhibition of the VTA_DAMGO ensemble nor of the VTA_STRESS ensemble itself directly affected place preference outcomes (Suppl. Fig. 5A, B). This allowed us to then evaluate the requirement of activity in the ensembles for the integral effects of DAMGO-driven conditioned place preference (CPP) and of stress-induced conditioned place aversion (CPA). For that, mice were injected bilaterally in the VTA with AAV-hSyn-DIO-hM4Di-mCherry or AAV-hSyn-DIO-mCherry as a control. Three weeks post-injection, mice were exposed to either systemic DAMGO or a social stress episode, followed by a 4-OHT injection to enable ensemble tagging, and afterwards behavioral testing was performed (Fig. 5A, B).

For DAMGO-induced place preference experiments, mice expressing mCherry or hM4Di-mCherry in the VTA_DAMGO ensemble received c21, followed by DAMGO. Subsequently, mice were randomly placed in one of the CPP compartments. On the following day, mice were placed in the other compartment after c21 injection followed by vehicle. On the third day, mice were allowed to explore the arena freely for 20 minutes (Fig. 5C). We found that inhibiting the VTA_DAMGO ensemble during the DAMGO conditioning sessions fully prevented DAMGO-induced CPP (Fig. 5D, E). Similarly, we found that inhibiting the VTA_STRESS ensemble during stress conditioning fully prevented stress-driven CPA (Fig. 5D, F). To further establish that this effect is specific to each ensemble, a group of VTA_DAMGO mice expressing hM4Di, was subjected to the social stress induced CPA paradigm. We found that inhibiting the VTA_DAMGO ensemble during stress conditioning sessions did not block the establishment of stress-induced CPA (Suppl. Figure 5C). Together these data show that activity in these VTA ensembles is selectively necessary for DAMGO or acute social stress-driven conditioned responses.

Finally, we investigated whether VTA_DAMGO and VTA_STRESS ensemble activity is necessary for the emergence of the effects of DAMGO (increase) or acute social stress (decrease) on palatable food intake. First we established that (in the absence of DAMGO or stress) just chemogenetically inhibiting the VTA_DAMGO or the VTA_STRESS ensemble did not directly affect fat intake (Suppl. Fig. 5D–G). Then we tested the effect of ensemble inhibition on the effects exerted by DAMGO or social stress on food intake (Fig. 5G, H). We found that inhibiting the VTA_DAMGO ensemble prior to DAMGO administration blocked the resultant hyperphagia, suggesting that activity of the VTA_DAMGO ensemble is necessary for DAMGO-induced increase of palatable food intake (Fig. 5I, Suppl. Figure 5H). On the other hand inhibiting the activity of the VTA_STRESS ensemble prior to social stress did not prevent the resultant hypophagia (Fig. 5J, Suppl. Fig. 5I–K).

In conclusion, our findings demonstrate that VTA_DAMGO and VTA_STRESS neuronal ensembles play distinct, essential roles in establishing conditioned place preference and avoidance, respectively. Moreover, the ability of systemic DAMGO to drive fat intake is dependent on activity in the VTA_DAMGO ensemble. These findings highlight the importance of these small VTA ensembles in orchestrating the behavioral repertoire in response to the integral experiences driven by external positive and negative valence stimuli.

Dopaminergic and GABAergic subsets promote approach behavior in the VTA_DAMGO ensemble and avoidance behavior in the VTA_STRESS ensemble

We next investigated the functional uniformity of different NT-defined neuronal subpopulations within the VTA_DAMGO and VTA_STRESS ensembles. To this end we used the offspring of TRAP2 mice crossed with either Pitx3-Flp or Vgat-Flp mice. We injected the same mice with intersectional viruses^12,28 for both stimulatory optogenetics (AAV-Conn/Fon-ChR2-EYFP) and inhibitory chemogenetics (AAV-Conn/Fon-hM4Di-mCherry) in the VTA. In this way we targeted either dopaminergic or GABAergic neuronal subsets within the ensemble. An optic fiber was also implanted above the VTA (Fig. 6A, B).

Previously we showed that mice would seek stimulation of their VTA_DAMGO ensemble, but not their VTA_STRESS ensemble (Fig. 4L–N). Now we sought to address whether stimulation of either the GABAergic or the dopaminergic subset of the VTA_DAMGO ensemble would be self-stimulated. For that we exposed mice to either DAMGO i.p. (or to acute social stress) in the presence of 4-OHT, to permanently tag and induce ChR2 expression in the GABAergic (TRAP2xVgat-Flp mice) or in the dopaminergic (TRAP2xPitx3-Flp mice) parts of the VTA_DAMGO ensemble (Fig. 6A, B). Using TH immunohistochemistry we determined that TH-positive neurons were indeed preferentially tagged in TRAP2xPitx3-Flp mice ( ~ 80% of the tagged ensemble) compared to TRAP2xVgat-Flp mice ( ~ 10% of the tagged ensemble) (Fig. 6C, D). We placed the mice in an operant cage to assess their self-stimulation behavior through nose-poking in an active port, as before (Fig. 4L–N). We found that mice sought out the active port significantly more so than the inactive ports, regardless of whether the dopaminergic or GABAergic subset of the VTA_DAMGO ensemble was targeted (Fig. 6E, left; Suppl. Figure 6A, left). Mice showed no preference for nose poking to activate the laser for the VTA_STRESS ensemble, irrespective of whether the GABAergic or the dopaminergic part was targeted (Fig. 6E, right; Suppl. Figure 6A, right).

Next we switched the contingencies of the experiment. We continuously stimulated the ensemble when the mice were in the operant box. In this case, nose poking in the correct port temporarily turned off laser stimulation. Under these conditions, mice were motivated to turn off the laser when their stress ensemble (but not their DAMGO ensemble) was stimulated, and this was the case irrespective of whether the dopaminergic or the GABAergic part of the VTA_STRESS ensemble was targeted (Fig. 6F; Suppl. Figure 6B).

We next investigated the importance of dopaminergic and GABAergic subsets within the VTA ensembles for the establishment of DAMGO-driven place preference or social stress driven-place avoidance. As stated above, the same mice used for the optogenetic experiment had, aside from an intersectional optogenetic construct, also an intersectional inhibitory DREADD in either the dopaminergic (TRAP2xPitx3-Flp) or the GABAergic (TRAP2xVgat-Flp) parts of their VTA ensembles. Indeed, using TH immunohistochemistry, we established that this strategy preferentially targeted the chemogenetic actuator to TH-positive neurons in TRAP2xPitx3-Flp mice and to TH-negative neurons in TRAP2xVgat-Flp mice (Fig. 6G, H). We subjected the mice to DAMGO-induced sePP or social stress-induced sePA as before (Fig. 5E, F). We found that inhibiting the GABAergic part of the VTA_DAMGO ensemble did not prevent the establishment of DAMGO-driven place preference. Instead, when we inhibited the dopaminergic part of the VTA_DAMGO ensemble there was no significant preference for the DAMGO-paired compartment (Fig. 6I). In contrast, when examining the effects of inhibiting specific NT-defined subsets in the context of social stress-induced sePA, we found that blocking the activity of either the dopaminergic or the GABAergic part of the VTA_STRESS ensemble was sufficient to prevent the establishment of stress-induced conditioned place avoidance (Fig. 6J).

VTA ensembles responding to valence stimuli

Our study highlights that VTA_DAMGO and VTA_STRESS ensembles are to a considerable extent non-overlapping subsets, preferentially activated by congruent valence stimuli over non-congruent valence stimuli. VTA_DAMGO cells were, even weeks later, again preferentially activated by DAMGO over acute social stress. Moreover, these cells were also active during food reward approaches. VTA_STRESS cells instead were, again even weeks later, preferentially activated by social stress over DAMGO. Moreover, they did not exhibit calcium reactivity during palatable food intake, whereas they were strongly activated during electric shock. These findings suggest that these two ensembles have selectivity in responding to either a congruently positive or negative experience. Instead, their responses are not limited to a singular type of experience (e.g., the VTA_STRESS ensemble responds to different types of stressors, and the VTA_DAMGO ensemble to different types of rewards). This is an interesting observation in the context of pure versus mixed selectivity response patterns of neurons²⁹. We show here that the VTA functional organization recruits specific ensembles that, weeks later, keep responding with high selectivity to multiple preferentially congruent valence stimuli.

We found that the VTA_STRESS and VTA_DAMGO ensembles were not clearly topographically segregated. It is of note that for the VTA at large, there is evidence for at least partial topographical organization in terms of circuit connectivity, and in terms of the general role of VTA sub-territories in behavioral functions^9,30. For instance with, on average, dopamine neurons in lateral parts of the VTA being more linked to reward-related processes, and medial parts of the VTA being more linked to aversion-related processes, though these are not absolute divisions in functionality^9,30. Our results, based on Fos-based methods rather than for instance in vivo electrophysiological responses, indicate that VTA neurons with opposite roles in valences processes are at least in part intermingled in similar VTA sub regions. These findings do leave open the possibility of topographical gradients in the VTA, in terms of where it may be more likely to find neurons with certain valence properties, especially when assessed with other methods (e.g., in vivo electrophysiology).

We showed that two congruent valence stimuli resulted in more overlap in a given VTA ensemble than did incongruent valence stimuli. However, even with two incongruent valence stimuli, ~15% of VTA neurons showed TRAP/Fos overlap. There are multiple possible contributing factors to this amount of overlap with incongruent valence stimuli. It could reflect a general baseline of Fos-based activity. We did indeed find that this value of overlap was not larger than what we observed with two subsequent home cage experiences. Nevertheless, a 4OH-Tamoxifen i.p. injection in a home cage likely does also represent an experience, rather than being fully neutral. It is therefore also possible that the extent of overlap in Fos in VTA ensemble neurons with incongruent valence stimuli represents features of salience of the stimuli. Salience/arousal responses (e.g., responses to both negative and positive valence) have indeed been proposed to be represented by VTA neurons^14,31,32. For instance, in a recent study, Willmore and colleagues using calcium imaging showed that ~2% of VTA_DOPAMINE neurons are activated by both food and an aversive air-puff¹⁷. Furthermore, del Arco and colleagues using in vivo electrophysiology, showed that in the VTA ~ 25% of neurons are activated by both a food reward and a tail pinch³³. Salience/arousal encoding neurons may therefore be partially present in both the VTA ensembles we describe here. Another consideration is that after a stressor subsides, the subsequent relief period may also be in itself a rewarding experience, transiently engaging dopamine neurons in the VTA³⁴. Our VTA_STRESS ensemble could therefore have also contained ‘stress-relief’ neurons. However, our ensemble stimulation and inhibition experiments showed clear net reinforcing properties of the VTA_DAMGO ensemble and clear net avoidance properties of the VTA_STRESS ensemble. Overall this suggests that we did mainly capture valence rather than salience aspects when tagging the ensembles, and mainly positive valence in the VTA_DAMGO and negative valence in the VTA_STRESS ensemble.

The importance of small VTA neuronal ensembles and their necessity for behavioral effects

A remarkable outcome of this study is that activity in small VTA ensembles is crucial for the approach and avoidance behaviors driven by systemic opioid and acute social stress stimuli. These are experiences that engage many brain regions, also other than VTA^35–38. Moreover, the VTA is a small brain region in itself³⁹. We report that based on Fos immunostaining the VTA ensembles are ~10% of VTA neurons (based on NeuN⁺ cells in VTA). Notably, using the TRAP2 technique we do not fully capture this population. As we showed before, TRAP2 counts of VTA cells are approximately a third of the population labeled by endogenous Fos²³. Extrapolating from this, we therefore estimate, also keeping incomplete targeting via viral approaches in mind, that the manipulated VTA ensembles in this study likely amount to ~2–6% of all VTA neurons. Stereological estimates of neuron count in the mouse VTA, suggest that it harbors ~20–30k dopamine neurons, and also about 20k non-dopaminergic neurons, and thus about 50k neurons in total³⁹. Consequently, our data suggest that we are manipulating ~1000–3000 neurons within either the VTA_DAMGO or the VTA_STRESS ensemble. Stimulation of this small amount of neurons is thus able to drive effects on place preference, anxiety behavior and, in the case of the VTA_STRESS ensemble, also food intake. Yet more strikingly even, the inhibition of just this small amount of neurons shows their indispensability in mediating certain behavioral effects. Specifically, for systemic DAMGO, the resultant place preference and hyperphagia, and for acute social stress the place aversion, but not its hypophagic effects.

The potency of these two small ensembles in shaping behavioral responses may in part stem from the fact that they are each comprised of ~50% of dopamine neurons. Dopamine neurons, through their complex axonal arborizations with many release sites, allow neuromodulatory effects over large areas of target substrates⁴⁰. Furthermore, neuronal ensembles in other regions than VTA have also been reported to typically comprise small clusters of cells, often less than 10% of neurons. For instance, inhibition of about 6% of neurons in the medial prefrontal cortex can suppress drug relapse⁴¹, and stimulation of as few as 800 AgRP neurons in the arcuate hypothalamus can drive food intake⁴². In another study, direct optogenetic stimulation of small ensembles of basolateral amygdala neurons, with either sucrose or quinine responsivity, could alter consummatory behavior⁴³. Together these findings emphasize the increasing recognition of the power of small neuronal ensembles in shaping behavior.

Our experiments also addressed the extent to which different neurotransmitter-defined neuronal cell types within a VTA ensemble: (i) have similar or rather opposite roles in valence processes and associated behavioral responses, (ii) and whether it is the ensemble as a whole, or the activity of some of its key constituents that is important for approach/avoidance behaviors. We show that mice seek to turn off optogenetic stimulation of either the GABAergic or the dopaminergic subset of its VTA_STRESS ensemble, while they will instead seek to turn on such stimulation of either the GABAergic or the dopaminergic subset of their VTA_DAMGO ensemble. This finding highlights that there are sets of dopaminergic and GABAergic neurons with opposite valence functions, which within a given ensemble may have similar functions with regards to orchestrating approach or avoidance behaviors. Moreover, with our chemogenetic inhibition experiments we evaluated whether activity of both the GABAergic and the dopaminergic subsets within the VTA_DAMGO and VTA_STRESS ensembles were equally crucial for the emergence of DAMGO-driven approach, and social stress-driven avoidance behavior, respectively. Here we observed differences in this regard between the ensembles. Chemogenetic inhibition of the VTA_DAMGO ensemble as a whole blocked DAMGO-driven approach behavior. Chemogenetic inhibition of just the dopaminergic subset of this ensemble still attenuated approach behaviors, but inhibition of just the GABAergic subset of the VTA_DAMGO ensemble did not. This indicates that for the VTA_DAMGO ensemble, its NT-defined sub-constituents may have seemingly similar valence functions when stimulated, but the activity of each of its NT-defined subsets is not equally indispensable for the emergence of DAMGO-driven approach behavior as a whole. A partially different image emerged for the VTA_STRESS ensemble. Chemogenetic inhibition of the VTA_STRESS ensemble as a whole blocked social stress-driven avoidance. This also occurred when we chemogenetically inhibited only the GABAergic or only the dopaminergic subsets of the VTA_STRESS ensemble. This indicates that the GABAergic and dopaminergic neuronal types within the VTA_STRESS ensemble have similar valence functions when stimulated, and that the activity of both of these subsets is critical for social stress to drive avoidance behavior.

These findings also raise interesting questions on whether different neuronal cell types within the VTA ensembles are locally connected to each other, and whether ensembles as a whole are interconnected. Based on our findings we predict that, whereas it is known that VTA_GABA neurons have a certain degree of local connectivity with VTA_DOPAMINE neurons^6,44, it may be more likely that VTA_GABA neurons from within a VTA_STRESS or VTA_DAMGO ensemble do not preferentially innervate VTA_DOPAMINE neurons within the same ensemble. Future research will need to address this.

Divergence in terms of mimicry and prevention for the two VTA ensembles for fat intake

We show a dissociation for the role of VTA_DAMGO and VTA_STRESS ensembles regarding fat intake. We recapitulate that systemic DAMGO causes hyperphagia, in accordance with previous reports^45–47, but we also find that solely stimulating only the VTA_DAMGO ensemble is not sufficient to mimic this. One possibility in this regard is that VTA_DAMGO ensemble activity may contribute to hyperphagia, but only when in conjunction with other network activity (which we then did not recreate by solely stimulating these neurons). Indeed, there are multiple brain regions in which DAMGO can increase fat intake, including not only the VTA, but also the prefrontal cortex, amygdala and nucleus accumbens^46,48,49. In further support of a contribution of the VTA_DAMGO ensemble to palatable food intake we show, with fiber photometry, that VTA_DAMGO cells are reactive during fat consumption. Moreover, we show that inhibiting the VTA_DAMGO ensemble, during systemic DAMGO, prevents the opioid from increasing fat intake. Our results therefore suggest that while VTA_DAMGO activity alone is not enough for increased fat intake, but activity in this ensemble represents a required hub in a larger co-activated network for opioid-driven hyperphagia.

In contrast to the finding for the VTA_DAMGO ensemble in hyperphagia, we observed an opposite behavioral pattern for the involvement of the VTA_STRESS ensemble in hypophagia. In the acute period after systemic social stress there is a hypophagic response, as we report here and previously⁵⁰. This is in contrast to the hyperphagia that can occur hours to days after the stress^50,51. We show here that the acute stimulation of the VTA_STRESS ensemble is sufficient to (partly) reduce fat intake. However, we also observed that inhibiting these cells alone was insufficient to prevent social stress-driven hypophagia. This suggests that during this acute period of social stress, there are other parallel mechanisms at play rather than just VTA_STRESS ensemble activity alone, which together are sufficiently able (also without the contribution of the VTA_STRESS ensemble) to deprioritize food reward intake over other potentially adaptive behaviors during such a threatening stressor.

Together our results show that the VTA_DAMGO and VTA_STRESS ensembles are largely distinct collectives of neurons, that despite showing similarities at the neurotransmitter (and anatomical) level, have opposite net roles in approach and avoidance behavior. Furthermore, despite their small number, they are indispensable hubs for specific approach and avoidance behaviors in response to potent external stimuli with integral effects. The power of targeting these small ensembles may inform the development of more effective interventions for diseases in which valence processes are perturbed.

Animals

Adult male mice (25–35 g, > 6 weeks) were used in all experiments. C57BL/6 J (Jax #664), heterozygous vesicular GABA transporter-Cre (vgat-Cre; Jax #016962), heterozygous vesicular glutamate transporter 2-Cre (vglut2-Cre; Jax #28863), heterozygous TRAP2 (Jax #030323), heterozygous Ai14 tdTomato reporter (Jax #007914), heterozygous Pitx3-Flp (generated by the ETH Phenomics Center (EPIC) of ETH Zürich, Switzerland), heterozygous vgat-Flp (Jax #029591) and double heterozygous TRAP2xAi14, TRAP2xPitx3-Flp, and TRAP2xVgat-Flp offspring were bred in house. For the social stress paradigm, proven breeder Swiss-CD1 mice were purchased from Janvier (France) and were used as aggressors. C57BL/6 J, vgat-Cre and vglut2-Cre mice were group-housed (2–4 per cage) in a temperature- and humidity-controlled room (22 ± 2 °C and 60–65% respectively) under a 12 h light/dark cycle (lights on at 7am) with ad libitum access to water and standard laboratory chow [Special Diet Services [SDS], product code CRM(E)]. TRAP2, TRAP2xAi14, TRAP2xPitx3-Flp and TRAP2xVgat-Flp mice were group-housed (2-4 per cage) until 1 week prior to TRAPing after which they were individually housed under the same conditions as described above. All experiments were approved by the Animal Ethics Committee of Utrecht University, and were conducted in agreement with Dutch law (Wet op de Dierproeven, 2014) and European regulations (Guideline 86/609/EEC).

Stereotactic surgery

Mice were anesthetized with ketamine (75 mg/kg i.p.; Narketan, Vetoquinol) and dexmedetomidine (1 mg/kg i.p.; dexdomitor, Vetoquinol). For local anesthesia, Lidocaine (0.1 ml; 10% in saline; B. Braun) was topically applied under the skin on the skull before incision. Eye ointment cream (CAF, Ceva Sante Animale B.V., Naaldwijk, Netherlands) was applied before surgery. Mice were mounted on a stereotactic frame (UNO B.V., Zevenaar, Netherlands, Model 68U801 or 68U025) and kept on a heating pad (33 °C) during surgery. Injections were done using a 31 G metal needle (Coopers Needleworks, Birmingham, United Kingdom) attached to a 10 μl Hamilton syringe (model 801RN) via flexible tubing (PE10, 0.28 mm ID, 0.61 mm OD, Portex, Keene, NH, United States). The Hamilton syringe was controlled by an automated pump (UNO B.V., Zevenaar, Netherlands, -model 220). For fiber implantation, the skull surface was gently scratched with a scalpel, followed by application of 35% phosphoric acid (167-CE, ultra-Etch, ultradent, USA) for 5 minutes to roughen the surface at the start of surgery. For viral infusions, all mice were bilaterally injected in the VTA ( − 3.2 mm posterior to bregma, 1.60 mm lateral, −4.90 mm ventral under an angle of 15°). After infusion of the viral volume, needles were left in place for an additional 9.5 min. Then they were moved 50 μm dorsally and completely retracted 30 s later. The skin was subsequently sutured (V926H, 6/0, VICRYL, Ethicon). Mice were then subcutaneously injected with atipamezole (50 mg/kg; Atipam, Dechra), carprofen (5 mg/kg, Carporal) and 1 ml of saline and were left to recover on a heating pad at 36 °C. Carprofen (0.025 mg/l) was provided in drinking water for a week after surgery. Animals were solitarily housed for 3 days post-surgery. Mice were allowed at least 3 weeks prior to subsequent behavioral manipulations.

Behavioral paradigms

Fiber photometry

Fiber photometry experiments were performed using a Doric setup, which included blue (465 nm) and purple (405 nm) light-emitting diodes. These light sources were connected to a high numerical aperture (0.48), large core (400 μm) optical fiber patch cord, which was attached to a matching brain implant in each mouse. GCaMP8s fluorescence was collected by the same fiber, split by two dichromatic mirrors (420-450 nm and 460–490 nm), and transmitted through 600 µm core cables to photoreceivers (Newport 2151 with lensed FC adapter). The Doric software managed the LEDs and independently demodulated fluorescence brightness resulting from 405 nm and 465 nm excitation. Data acquisition was performed at 12 kilosamples per second and downsampled by a factor of 50 using the Doric Neuroscience Studio software. The software also calculated ΔF/F0 as 100* (F-F0)/F0, where F0 was the running average over a 60-second time window. Custom Python scripts were used to synchronize the onset of each behavior with the photometric ΔF/F₀ signal. For statistical comparisons, the average ΔF/F₀ was calculated per animal (of one hemisphere with optimal fiber placement) using the following time bins: baseline (−2.25 to −1 seconds prior to TTL pulse) and onset (0 to 1.25 seconds after the TTL pulse).

Immunohistochemistry and microscopy

Animals were transcardially perfused with 4% paraformaldehyde (PFA) and brains were post-fixated in 4% PFA at 4 °C for 24 -48h. Next, they were transferred to a 30% sucrose solution in PBS where they remained for another 48–72 h. Coronal sections of 35 μm were made with a cryostat (CM1950, Leica, Netherlands). Slices were first washed 4 × 10 min in PBS followed by incubation in blocking solution [5% normal goat serum (NGS), 2.5% bovine serum albumin, 0.2% Triton X-100 in PBS] for 1 h. Subsequently, slices were moved to the primary antibody solution containing one or more of the following antibodies in blocking solution and were left there overnight: rabbit-anti-c-Fos (1:1000; Cell Signaling Technology, 2250 s, Leiden, Netherlands), chicken-anti-GFP (1:1000, Aves), rabbit-anti-RFP (1:1000, Rockland), mouse-anti-TH (1:1000; Sigma-Aldrich Chemie N.V, MAB318, Zwijndrecht, Netherlands), mouse-anti-NeuN (1:1000; Abcam, ab104224, Waltham, MA, United States). The following day slices were washed 4 × 10 min followed by 2 h incubation in secondary antibody solution containing goat-anti-rabbit Alexa Fluor 488 (1:500; Abcam, ab150077, Waltham, MA, United States), goat-anti-chicken Alexa Fluor 488 (1:500; Abcam, ab150169, Waltham, MA, United States) and goat-anti-mouse Alexa Fluor 647 (1:500; Thermo Fischer Scientific, A-21236, Waltham, MA, United States). Slices were then moved to DAPI (0.5 μg/ml; Biotium brand, VWR, #89139-054, Netherlands) for 5 min followed by 2 × 10 min washes in PBS. Finally, slices were mounted on glass slides using 0.2% gelatin in PBS and coverslipped using FluorSave (Millipore, 345789, Amsterdam, Netherlands).

For cell counting in the VTA, 5-6 images of a VTA hemisection were obtained using a wide-field epifluorescence microscope (AX10, Zeiss). Specifically, a multi-channel image was captured, utilizing tyrosine hydroxylase (TH) staining to establish the VTA’s boundaries. Fiji software⁵³ was employed to manually delineate the boundaries of VTA hemisections, using the TH staining outlines. Subsequently, an experimenter, unaware of the experimental conditions, manually counted cells expressing Fos or tdTomato. Cell density was then calculated as the number of Fos⁺ or tdTomato⁺ cells relative to the area of the designated region of interest. For co-localization analysis 4–6 z-stacks were acquired using a confocal microscope (LSM 880, Zeiss). For topographical analysis of Fos⁺ cells, the X, Y coordinates of Fos⁺ were extracted and their distance from a reference point (Suppl. Figure 2F) was calculated. Then the average X and Y coordinate was calculated separately for rostral (–2.92 < –3.28 mm relative to bregma) middle (–3.28 < –3.52 mm relative to bregma) and caudal (–3.52 < –3.80 mm relative to bregma) VTA slices.

Data analysis and statistics

Data were analyzed using GraphPad Prism 9.3.1 (San Diego, CA, United States), Noldus (Ethovision V11.0), Python v3.8.8, MedPC-4 (Med Associates Inc), and MATLAB (Mathworks, Natick, MA, United States). Animals were randomly assigned to experimental conditions. Data are always reported and represented as mean + SEM, with single data points plotted representing individual animals. No animals were excluded unless otherwise specified. Normality was established using the Kolmogorov-Smirnov test. Where appropriate, statistical tests employed included One-way ANOVAs, Two-way ANOVAs, or Multi-way Repeated-measures ANOVAs. In instances where the omnibus ANOVAs indicated significant results, post hoc contrasts were conducted to further analyze the data. Two-tailed test were always performed with α = 0.05.

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