Endocannabinoids facilitate reward engagement through retrograde gain-control

Abstract

Neuromodulatory signaling is poised to serve as a neural mechanism for gain control, acting as a crucial tuning factor to influence neuronal activity by dynamically shaping excitatory and inhibitory fast neurotransmission. The endocannabinoid (eCB) signaling system, the most widely expressed neuromodulatory system in the mammalian brain, has been demonstrated to filter excitatory and inhibitory inputs through retrograde, pre-synaptic action in vitro and ex vivo^1–9. However, whether eCBs exert retrograde gain control to ultimately facilitate motivated behaviors in freely moving mammals has not been established. Using a suite of in vivo physiological, imaging, genetic, and machine learning-based approaches, we uncovered a fundamental role for dynamic release of eCBs in controlling behavioral engagement during reward-seeking through a genetically and anatomically defined thalamo-striatal circuit.

MAIN

Reward-seeking behaviors are evolutionarily conserved and necessary for survival, and are importantly disrupted in both substance-use disorder and numerous psychiatric diseases^10–14. The Nucleus Accumbens (NAc) is a key nodal mesocorticolimbic structure that mediates action selection as well as reward-seeking^15–19. The NAc incorporates input from numerous excitatory, inhibitory, and neuromodulatory projections to ultimately integrate affective information and translate this information into motivated behavioral outputs^20–22. However, we lack complete understanding of critical fundamental mechanisms that govern reward-seeking, specifically how diverse sources of excitatory input to the NAc are integrated and filtered by distinct neuromodulators to shape discrete behavioral outputs. For example, the NAc receives glutamatergic input from over half a dozen brain structures, yet it is unclear why such diverse sources of this single neurotransmitter exist and how each input is ultimately filtered in real time into useful information downstream by local neuromodulatory signals^23–25.

Brain slice electrophysiology studies have shown that the eCB signaling system dynamically tunes both excitatory and modulatory input into the NAc^26–30. In the central nervous system, this signaling system mainly involves activation of cannabinoid 1 receptors (CB₁R) by the eCBs anandamide (AEA) and 2-arachidonylglycerol (2-AG), and represents the most widely expressed neuromodulatory system in the mammalian brain^1–5. Unlike conventional neurotransmitters released by presynaptic vesicular mechanisms, lipophilic eCBs are produced on-demand by post-synaptic neurons and primarily function retrogradely to activate presynaptic CB₁Rs and attenuate neurotransmitter release^6–9. However, the lipophilic nature of these modulators, as well as the diverse neuronal subtypes and specific stimuli that drive their production have made the investigation of their in vivo function in motivated behaviors intractable. Furthermore, the mechanistic link between neural activity-dependent eCB release, modulation of afferent input, and how specific behavioral features are shaped by this retrograde action in vivo remain to be established.

Here we discovered specific behavioral parameters necessary to evoke transient eCB release in the NAc in vivo during freely moving motivated behavior. Using a complementary suite of approaches our results indicate that eCB production and release from a genetically defined medium spiny NAc neuron population is specifically time-locked to behavioral engagement, and that these eCBs drive reward-seeking through retrograde action at presynaptic terminals arising from the anterior paraventricular thalamus (aPVT), a key nodal structure for the regulation of motivated behaviors^31,32. This dynamic and fundamental retrograde signaling mechanism acting through CB₁Rs dynamically shapes the vigor of engagement in reward seeking behaviors through its discrete suppression of efferent glutamatergic input into the NAc.

Endogenous cannabinoids are released in the NAc during reward consumption and inhibit glutamate release from the paraventricular thalamus.

To determine how eCB dynamics in the NAc facilitate reward seeking and consumption, we injected the medial NAc shell of adult mice with an AAV (AAV1-hSyn-GRAB_eCB2.0) to drive the neuronal expression of the fluorescent eCB biosensor, GRAB_eCB2.0, and implanted photometry fibers dorsal to the injection site (Fig. 1a, Extended Data Fig. 1a,b)³³. Mice were then placed in a Pavlovian reward conditioning assay and trained to associate a cue light with the protraction of a sipper containing a 10% sucrose solution across 5 days. On day 1 of training, consumption of the sucrose solution elicited a pronounced increase in GRAB_eCB2.0 signal, indicative of local eCB release time-locked to consummatory events (Fig. 1b). However, on day 5 of conditioning, the onset of the increase in GRAB_eCB2.0 signal shifted earlier to the time of cue onset, and rose to ~50% of maximal response prior to sipper delivery. This suggests that eCB release is not solely tied to reward consumption, but may sculpt reward-seeking behaviors or be involved in cue-encoding (Fig. 1c-e). Accordingly, in a reward omission test in which the sipper was not delivered on 50% of cue trials, the GRAB_eCB2.0 signal still increased despite a lack of reward delivery (Extended Data Fig. 1r-v). Blocking 2-AG biosynthesis with the Diacylglycerol Lipase inhibitor DO34 strongly attenuated the reward-induced increase in GRAB_eCB2.0 signal, while inhibition of AEA hydrolysis had no effect on the signal, suggesting that 2-AG is mediating this response (Fig 1. f-I, Extended Data Fig. c,e,g,h,i,l). To determine whether eCB release in the NAc is either selective for stimulus valence or unique to reward-seeking behavior, we exposed the same cohorts of mice to a Pavlovian fear conditioning assay. We found that shock exposure also elicited an increase in GRAB_eCB2.0 signal, showing that eCB production in the NAc is similarly elicited by aversive stimuli (Extended data Fig. 1w,x,z). However, during fear retrieval (extinction day 1), exposure to the shock-predictive tone did not elicit such response (Extended data Fig. 1y,aa). Together, these results indicate that both rewarding and aversive stimuli can elicit robust eCB release within the NAc (Extended data Fig. 1bb).

Fig. 1. Endogenous cannabinoids are released in the NAc during reward consumption and inhibit glutamate release from aPVT terminals.

(a) Schematic for fiber photometry recordings of GRAB_eCB2.0 in the NAc during Pavlovian reward conditioning. (b,c)Photometry trace with representative lick raster and heatmap of GRAB_eCB2.0 signal aligned to cue onset on day 1 and day 5 of conditioning. (d,e) Comparison # of licks per session (p=0.0468) and average GRAB_eCB2.0 signal from 0–26s (p=0.0027) between day 1 and day 5 (N=11). (f,g) Trace with representative lick raster and heatmap of GRAB_eCB2.0 signal aligned to cue onset after treatment with vehicle or DO34. (h,i) Comparison # of licks per session (p=0.2514) and average photometry Z-score from 0–26s (p=0.0011) between vehicle and DO34 (N=11). (j) Schematic for retrograde tracing and RNAscope studies in Ai14 mice. (k) Epifluorescent images of TdTomato positive NAc projecting neurons in the aPVT and pPVT. (l,m) RNAscope image and quantification of Cre and CB₁R transcript levels in the aPVT and pPVT (N=4,3). (n,o) RNAscope image and quantification of NTS and CB₁R transcript levels in the aPVT and pPVT (N=3,3). (p) Schematic for whole-cell voltage-clamp recordings of optically evoked EPSCs from NAc D₁R+ and D₁R - neurons. (q,r) Recordings of depolarization-induced suppression of excitation in vehicle (D₁R+ n=13, D₁R- n=7, p=0.0184) and SR141716A (D₁R+ n=8, D₁R- n=8, p=0.3763) treated slices. (s) Schematic depicting enhanced retrograde eCB signaling at aPVT^NTS-NAc D1- synapses compared to aPVT^NTS-NAc D1+ synapses. Error bars represent ± SEM (d-i), or min/max values (q,r); box plots represent interquartile range with line at median; N represents number of mice, n represent number of neurons. p values reported from two-tailed paired t test (d,e,h,i) and two-tailed unpaired t test (q,r). *p<0.05, **p<0.01.

One major function of 2-AG activation of presynaptic CB₁R is to decrease the probability of fast transmitter release, thus we next identified neuronal projections to the NAc that express CB₁R (Fig. 1j,k). We injected a retrograde virus to drive the expression of Cre recombinase (AAV2-retro-hSyn-Cre) into the NAc of flox-stop Ai14 Td-Tomato reporter mice, and identified five principal sources of excitatory inputs: the medial prefrontal cortex (mPFC), anterior paraventricular thalamus, posterior paraventricular thalamus (pPVT), basolateral amygdala (BLA), and ventral hippocampus (vHipp) (Fig 1l,n, Extended Data Fig. 2a-c). We then used a semi-quantitative in situ hybridization approach, RNAscope, to examine the colocalization of Cre mRNA transcript derived from our retrograde viral approach (NAc projectors) and cn1r mRNA transcript overlap. Overall, the aPVT had the largest percentage of neurons projecting to the NAc, as well as the largest percentage of neurons projecting to the NAc that coexpressed CB₁R (Fig 1l,n, Extended Data Fig. 2a-c).

Recent studies have emphasized that functional differences exist along the anterior-posterior axis of the PVT, yet few genetic markers have been identified to isolate and/or manipulate neurons located in the anterior paraventricular thalamus (aPVT)^{24,25,34,35,36,37}. We analyzed the Allen Brain Atlas database and published PVT RNAseq results and found that the expression of neuropeptide neurotensin (NTS) is selectively enriched in the aPVT compared to the posterior paraventricular thalamus (pPVT)³⁹. We corroborated these initial findings with RNAscope, confirming selective expression of NTS in the aPVT. We also found that in aPVT NTS neurons have strong coexpression of CB₁R (83.9 ± 5% overlap) (Fig 1. n,o). Using Channelrhodopsin (ChR2)-assisted circuit mapping⁴⁰ we found that these neurons send strong excitatory projections to the NAc as well as the Bed Nucleus of the Stria Terminalis (BNST), Central Amygdala (CeA) and Zona Incerta (ZI) (Extended data Fig. 2 p-bb).

To determine how aPVT input to the NAc is regulated by CB₁R signaling, we performed whole-cell patch clamp electrophysiological recordings. NTS-Cre mice were crossed with a D₁R-tdTomato reporter mice, and their progeny were injected with AAV5-DIO-ChR2-eYFP in the aPVT to facilitate recordings of optically-evoked excitatory postsynaptic currents (oEPSCs) from either D₁R⁺ or D₁R^- neurons (the latter being purportedly D₂R⁺, see methods and Extended Data Fig. 2l-o) in the medial NAc shell (Fig. 1p). Whole-cell recordings of optically-evoked input/output curves demonstrate that there is a significant bias in excitatory input onto D₁R⁺ neurons compared to D₁R^- (Extended data Fig. 2d). Remarkably, SR141716 (10 μM) inverted this bias, resulting in a larger excitatory response observed in D₁R^- compared to D₁R⁺, suggesting tonic eCB shunting of excitatory input onto D₁R^- neurons (Extended data Fig. 2e). In agreement with the involvement of 2-AG in this response, the intrinsic excitatory bias onto D₁R⁺ neurons was eliminated by augmenting 2-AG levels by inhibiting its degradation by monoacylglycerol lipase with its selective inhibitor, JZL184 (1 μM) (Extended data Fig. 2f,h-k). To determine if phasic 2-AG signaling is involved, we recorded depolarization-induced suppression of excitation (DSE), a form of short-term synaptic placidity known to be mediated by retrograde 2-AG signaling^6,9. We found that DSE was selectively expressed at aPVT-NAc D₁R^- synapses (Fig. q,r). These results indicate that both tonic and phasic 2-AG signaling are primarily expressed at aPVT-NAc D₁R- (putatively D₂R⁺) synapses, and that tonic eCB signaling at these synapses drives the observed excitatory bias onto D₁R⁺ neurons (Fig. 1s). Taken together, these in vivo and in vitro results demonstrate that salient rewarding and aversive stimuli both evoke robust 2-AG release which acts to modulate extrinsic excitatory input to the NAc in a retrograde CB₁R-dependent manner.

aPVT-NAc presynaptic activity is negatively correlated with engagement in reward-seeking behaviors

To discern whether eCB signaling facilitates reward-seeking behaviors through modulation of extrinsic excitatory input, we first determined the behavioral role of the aPVT-NAc circuit. We injected NTS-Cre mice with AAV-DJ-GCaMP6s in the aPVT followed by fiberoptic implantation in the medial NAc shell to measure calcium dynamics of aPVT^NTS-NAc terminals during Pavlovian reward conditioning. We observed significant differences in both GCaMP6s signal and reward-seeking behaviors across conditioning as mice learned task parameters (Fig. 2a-j, Extended Data Fig. 3a-e). Using 1080P 20hz videos of behavior, we quantified reward-seeking as % of time poking, pawing, or sniffing at the sipper port (Fig. 2d Supplementary Video 1-4, methods). On day 1 of conditioning, mice only spent roughly ¹/₄ of their time (24.1%) engaging with the sipper port during the cue window prior to sipper protraction (0–6s), as they had not yet learned to associate the cue with the rewarding outcome. However, by day 5 of conditioning, mice spent ²/₅ (38.6%) of the time engaging with sipper port during the cue window, in a display of learned expectation of reward delivery (Fig. 2d,f-g). Conversely, on day 1 of conditioning in the post-sipper time window (26–46s), mice spent almost ¹/₃ (31.8%) of the time engaging with the sipper port following sipper retraction, as they had not yet learned the associated time-out period (ITI) following sipper retraction. However, by day 5 of conditioning, this behavior had significantly diminished to less than ¹/₅ (19.0%) of the time, as the mouse learned that the sipper will not be extended again until the next cue delivery epoch (Fig. 2d,f,g, Supplementary Video 2). Calcium transients were isolated during these key epochs including cue period prior to sipper protraction (0–6 sec), and in the period immediately follow cue cessation and sipper retraction (26–46 sec). In the pre-sipper time window (0–6s), excitation observed on day 1 transitions to inhibition on day 5, and vice-versa in the post-sipper time window (26–46s) (Fig. 2e). We found that whether quantifying either the pre-sipper or post-sipper time window, the activity of the aPVT^NTS-NAc presynaptic terminals was significantly and negatively correlated with an animal’s relative epochs of sipper port engagement (Fig. 2f-j). This result indicates that the activity of these projections could be signaling engagement and disengagement from reward-seeking.

Fig. 2. aPVT-NAc presynaptic activity is negatively correlated with engagement in reward-seeking behaviors.

(a) Schematic for fiber photometry recordings of GCaMP6s expressing aPVT^NTS-NAc terminals. (b,c) Photometry trace with representative lick raster and heatmap of cue-aligned GCaMP6s signal during day 1 and day 5 of Pavlovian reward conditioning (N=10). (d) Comparison between day 1 and day 5 of sipper port engagement time during the pre-sipper (p=0.0002), and post-sipper (p=0.0039) time window (N=10). (e) Comparison between day 1 and day 5 of average GCaMP6s signal during pre-sipper (0–6s, p=0.0001) and post-sipper (26–46s, p=0.0002) time window (N=10). (f) Correlation between pre-sipper extension engagement time and average photometry Z-score across 5 days of conditioning (R²=0.2553, p=0.0002, N=10). (g) Correlation between post-sipper retraction engagement time and average GCaMP6s signal across 5 days of conditioning (R²=0.3299, p=1.59 · 10^-5, N=10). (h) Raster plot showing timepoints of initiation in sipper approach, sipper engagement, and null behaviors during the cue period prior to sipper extension. (i,j) Transition diagram showing probabilities of transitions between approach, engage, and null/exit behaviors on day 1 and day 5 of conditioning during the pre-sipper time window, with size of circles corresponding to total number of events. (k) Schematic for Pavlovian fear conditioning / extinction. (l,m) Photometry trace of GCaMP6s signal during fear extinction day 1 and day 5. (n) Quantification of cue-evoked freezing on day 1 and day 5 of extinction (p=3.45 · 10^-8, N=11). (o) Quantification of GCaMP6s signal on day 1 and day 5 of extinction (p=0.0318, N=11). (p) Schematic for optogenetic manipulation of aPVT^NTS-NAc terminals using DIO-ChR2 and DIO-PPO. (q) # of licks during no stimulation or closed loop 20hz optogenetic stimulation of aPVT^NTS-NAc terminals in DIO-ChR2 (p=0.0300, N=9) and DIO-eYFP (p=0.6323, N=9) expressing animals. (r) # of licks during no stimulation or sipper retraction-aligned 10hz optogenetic inhibition of aPVT^NTS-NAc terminals in DIO-PPO (p=0.0090, N=13) and DIO-eYFP (p=0.7608, N=11) expressing animals. (s) Summary schematic depicting behavioral engagement-induced inhibition of aPVT^NTS-NAc terminals. All error bars represent ± SEM; N represents number of mice. p values reported from two-tailed paired t test (d,e,n,o,q,r) and simple linear regression (f,g). *p<0.05, **p<0.01, ***p<0.001.

A control reward omission experiment demonstrated that this inhibition was not dependent on reward consumption, as significant inhibition of these presynaptic terminals was also observed during omission trials (Extended Data Fig. 3l-o). Furthermore, analysis of ignored cue trials on day 5 of conditioning demonstrated a lack of inhibitory effects, suggesting that the inhibition is likely not being driven by attribution of incentive salience to the cue (Extended Data Fig. 3p). Last, to determine whether aPVT^NTS-NAc presynaptic activity was simply reflective of locomotion (inhibited during decrease, excited during increase), we found that in an FR1 and PR operant task, aPVT^NTS-NAc presynaptic activity is still inhibited during reward-seeking behaviors that require an active motor sequence (nose poke) (Extended Data Fig. 3q-u). This result is further corroborated by evidence collected in open field activity, in which we observed no calcium dynamics that time-locked to locomotor initiation events or rearing bouts (Extended Data Fig. 3v-x).

To determine whether the activity of aPVT^NTS-NAc terminals is only coincident with engagement in reward-seeking behavior or conserved for other behaviors, we trained mice in an aversive Pavlovian fear conditioning and extinction task (Fig. 2k, Extended data Fig. 4a-g). During fear conditioning, we observed a transient (~5 second) peak in GCaMP6s signal time-locked to foot shock (Extended data Fig. 4c,d). On day 1 of fear extinction (recall), we measured an inhibition of aPVT^NTS-NAc terminals, similar to what we observed during engagement in reward-seeking behaviors, suggesting that aPVT^NTS-NAc pathway does not encode stimulus valence (Fig. 2l). When the GCaMP6s signal was aligned to the end of freezing bouts (disengagement) on extinction day 1, there was also a concomitant increase in fluorescence (Extended data Fig. 4e,f). As mice underwent extinction training, the inhibitory signal broadly decreased, and their degree of engagement in defensive freezing behaviors was likewise negatively correlated with the activity of aPVT^NTS-NAc presynaptic terminals (Fig 2 l-o, Extended data Fig. 4g). Collectively, these results indicate that aPVT^NTS-NAc activity is negatively correlated with behavioral engagement in both reward-seeking and defensive freezing.

We then sought to establish whether there is a causal role for aPVT^NTS-NAc in behavioral engagement by selectively bi-directionally photo-manipulating neuronal activity of the projections (Fig. 2p) during discrete epochs of reward-seeking. Mice were injected with either AAV5-DIO-ChR2-eYFP, AAV5-DIO-PPO-eYFP (an inhibitory G_i-coupled opsin), or AAV5-DIO-YFP as control, in the aPVT with bilateral fiber optic implants placed above the medial NAc shell. Following Pavlovian reward conditioning, mice injected with ChR2 or eYFP received counterbalanced days of closed-loop photo-stimulation, where every lick resulted in a 2 second burst of 20 Hz blue light photo-stimulation delivered to aPVT^NTS-NAc terminals. Stimulation of aPVT^NTS-NAc terminals reduced licking behavior, either using closed-loop photo-stimulation (Fig. 2q) or constant 20 Hz photo-stimulation throughout the duration of the cue (Extended Data Fig. 5a,b). Conversely, 10 Hz photo-inhibition of aPVT terminals (PPO) from 5 sec before sipper retraction to 20 sec after increased engagement with the sipper port during this epoch (Fig. 2r). To test whether activation of the aPVT^NTS-NAc is necessary and sufficient for conditioned freezing behavior, we either selectively photo-activated or inhibited aPVT^NTS-NAc terminals using ChR2 or PPO respectively throughout the cue period (20 Hz 20 sec and 10 Hz 20 sec, respectively) during fear retrieval. Photo-activation of the aPVT terminals was sufficient to reduce conditioned freezing, while photo-inhibition resulted in a trend towards increased freezing, likely due to a ceiling effect on the expression of freezing behavior during recall (Extended Data Fig. 4h-j). Importantly, neither activation nor inhibition of the aPVT^NTS-NAc pathway was innately reinforcing or aversive having no effect in either intracranial self-stimulation of real-time place preference (Extended Data Fig. 5c-h). Overall, these findings establish the causal role of aPVT^NTS-NAc projections in dynamic tuning of behavioral engagement in both appetitive and aversive behavior (Fig. 2s).

The endocannabinoid 2-AG facilitates engagement in reward-seeking behavior through action at presynaptic aPVT-NAc CB₁Rs.

Our electrophysiology results established that 2-AG inhibits aPVT^NTS-NAc terminals by activating CB₁R and our in vivo calcium recordings indicate broad presynaptic inhibition of aPVT^NTS-NAc terminals during epochs of behavioral engagement. Thus, we next determined whether this in vivo inhibition was mediated by endogenous release of 2-AG acting at CB₁R. Using mice expressing a DIO-GCaMP6s in aPVT^NTS-NAc terminals, we found that the CB₁-R antagonist SR141716 (10 mg/kg; i.p.) robustly reduced the inhibition of these terminals in vivo, as well as significantly reduced sucrose consumption events (Extended Data Fig. 6a-d).

We subsequently used a cre-dependent CRISPR/Cas9 viral approach to selectively delete CB₁R from aPVT neurons to establish whether 2-AG–CB₁R inhibition of aPVT^NTS-NAc terminals was necessary for engagement in reward-seeking behaviors (Fig. 3a, Extended Data Fig. 6b-n)^41,42. Fluorescent in situ hybridization showed that the sgCNR1 virus elicited a ~47% reduction in the number of aPVT neurons coexpressing NTS and CB₁R (i.e., 30.4% vs 16.2%) compared to sgROSA26 controls (Extended Data Fig. 6f-n). To assess how deletion of aPVT CB₁R influenced aPVT^NTS-NAc axon terminal activity, as well as engagement in reward-seeking behaviors, we co-injected these viruses (sgROSA or sgCNR1) with AAV5-DIO-GCaMP6s in the aPVT, and implanted a fiber optic above the medial NAc shell (Fig. 3a, Extended Data Fig. 7a-h). While deletion of CB₁R in aPVT^NTS-NAc neurons did not alter the total number of licks during ad libitum access (Extended data Fig. 7f), it significantly reduced the length and timing of licking bouts (Extended Data Fig. 7g). This result suggests that 2-AG–CB₁R signaling at aPVT^NTS-NAc terminals modulates the innate temporal patterns of reward engagement, without impacting satiety-related behavioral measures.

Fig. 3. Endogenous cannabinoid inhibition of aPVT terminals drives engagement in reward-seeking behaviors.

(a) Schematic for CRISPR-Cas9 deletion of CB₁R from aPVT NTS neurons and photometry recordings of aPVT^NTS-NAc terminals during Pavlovian reward conditioning. (b-e) Photometry trace with representative lick raster and heatmap of GCaMP6s signal from Vehicle and DO34 treated sgROSA (N=6) and sgCNR1(N=7) injected mice. (f) Comparison of % sipper port engagement time showing reduced engagement in vehicle treated sgCNR1 animals (p=0.0280, N=7) and DO34 treated sgROSA animals (p=0.0003, N=6) compared to vehicle treated sgROSA controls (Main effects: Treatment p<0.0001, genotype p=0.0357, interaction p=0.0627). (g) Comparison of average GCaMP6s signal from 0–26 seconds showing blunted inhibition of aPVT terminals in vehicle treated sgCNR1 animals (p=0.0005, N=7) and DO34 treated sgROSA animals (p=0.0004, N=6) compared to vehicle treated sgROSA controls (Main effects: Treatment p=0.0093, genotype p=0.0313, interaction p=0.0002). (h) Comparison of freezing behavior between sgROSA (N=7) and sgCNR1 (N=7) injected animals during fear conditioning (p=0.1880) and day 1 of fear extinction (p=0.1789). (i) Comparison of GCaMP6s signal between sgROSA (N=7) and sgCNR1 (N=7) injected animals during fear conditioning (from 0–5 seconds post shock, p=0.4953) and day 1 of fear extinction (from 0–20 seconds post tone, p=0.7392). (j) Schematic for photometry recordings of GRAB_eCB2.0 or GRAB_eCBMUT on aPVT-NAc terminals during Pavlovian reward conditioning. (k,l) Photometry trace of GRAB_eCB2.0 (N=8) or GRAB_eCBMUT (N=4) signal on day 1 and day 5 of reward conditioning. (m) Comparison of % sipper port engagement time showing enhanced engagement time on day 5 in both GRAB_eCB2.0 (p=0.0024, N=8) and GRAB_eCBMUT (p=0.0120, N=4) compared to their respective day 1 values (Main effects: Sensor p=0.1023, day p<0.0001, interaction p=0.7257). (n) Comparison of average GCaMP6s signal from 0–26 seconds showing enhanced GRAB_eCB2.0 signal of day 5 of reward conditioning compared to GRAB_eCB2.0 day 1 (p=0.0017, N=8) and GRAB_eCBMUT day 5 (p=0.0009, N=4) (Main effects: Sensor p<0.0001, day p=0.0039, interaction p=0.1241). (o) Schematic of simultaneous optogenetic activation of NAc Penk neurons and photometry recordings of aPVT-NAc GRAB_eCB2.0 (N=4) (p) Trace and representative heatmap showing GRAB_eCB2.0 dynamics time-locked to 10s 20hz stimulation of NAc Penk neurons. (q) Summary schematic depicting reward-engagement induced retrograde eCB inhibition at aPVT-NAc^Penk synapses. All error bars represent min/max values; box plots represent interquartile range with line at median: N represents number of mice. p values reported from two-way ANOVA with Sidak post-hoc correction (f-i,m,n). *p<0.05, **p<0.01, ***p<0.001.

To isolate the mechanism mediating reward engagement-induced inhibition of aPVT^NTS-NAc axon terminals as well as the behavioral consequences of CB₁R deletion from aPVT neurons, we treated GCaMP6s/sgCNR1 and GCaMP6s /sgROSA expressing mice with the DAGL inhibitor (to block 2-AG production), DO34 (20 mg/kg), or a vehicle control. In vehicle treated animals, CB₁R deletion from aPVT neurons significantly attenuated the engagement-induced inhibition of GCaMP6s fluorescence, as well as significantly reduced time engaging with the sipper port throughout the cue period (Fig. 3b,c,f,g, Extended Data Fig. 7h,i). This change was not simply due to a reduction in the number of engagement events, since the absence of engagement-related pre-synaptic inhibition was still observed in sgCNR1 injected mice when specifically isolating trials where animals consumed sucrose (Extended data Fig. 7i). In DO34 treated animals, CB₁R deletion had no effect either on aPVT^NTS-NAc GCaMP6 signal or degree of engagement in reward-seeking (Fig. 3d-g, Extended Data Fig. 7i,j). Furthermore, DO34 reversed engagement-induced inhibition of aPVT^NTS-NAc terminals in sgROSA mice and reduced sipper engagement, while having no effect on GCaMP6s signal or sipper engagement in sgCNR1 knockout animals (Fig. 3f,g). The mutual occlusion effects between sgCNR1 knockdown and DO34 treatment provide corroborating evidence that 2-AG–CB₁R signaling facilitates engagement in reward-seeking through inhibition of aPVT^NTS-NAc axon terminals. These results also suggest that 2-AG–CB₁R signaling represent a key endogenous Gi-GPCR mechanism acting in a feed-forward retrograde manner to suppress excitatory input into the ventral striatum during engagement epochs of reward-seeking (Fig. 3f,g).

We next assessed whether CRISPR knockdown of aPVT CB₁R impaired engagement in defensive freezing behaviors in a Pavlovian fear conditioning/extinction paradigm (Fig. 3h,i, Extended data Fig 7k-n). During fear conditioning, deletion of the aPVT CB₁R neither affected the activity of aPVT^NTS-NAc terminals nor freezing behaviors. Similarly, during fear recall, this manipulation did not attenuate freezing-induced inhibition of aPVT^NTS-NAc terminals or freezing behaviors (Fig. 3h,i, Extended data Fig 7k-n). These data suggest that while inhibition of aPVT^NTS-NAc terminals represent common feature of engagement in both reward-seeking and freezing behaviors, 2-AG-CB₁R signaling at aPVT^NTS-NAc terminals is necessary for engagement in reward-seeking but not for defensive freezing, suggesting a unique undefined mechanism driving aversion-related inhibition.

To test the specificity of CB₁R signaling at aPVT^NTS-NAc terminals for regulating engagement in reward seeking behaviors, we performed a similar set of experiments in which we deleted the CB₁R from BLA VGlut1 neurons, while simultaneous monitoring the activity of BLA^VGlut1-NAc terminals using GCaMP6s fiber photometry (Extended data 7o-aa). In contrast to deletion of CB₁R from aPVT^NTS neurons, post-hoc corrected multiple comparisons revealed no significant effects of CB₁R deletion on either engagement in reward-seeking or BLA^VGlut1-NAc terminal activity (Extended data 7y-aa). However, examination of the overall effects of genotype and treatment revealed significant main effects of both CB₁R knockout as well as DO34 treatment on both engagement in reward seeking and BLA^VGlut1-NAc terminal activity. These results suggest that CB₁R on BLA^VGlut1-NAc projections may play some role in regulating reward seeking, and future studies should examine whether there are unique aspects of behavioral engagement in reward seeking differentiating the effects of CB₁R on aPVT^NTS neurons vs BLA^VGlut1 neurons. Deletion of CB₁R from BLA^VGlut1 neurons had no effect on freezing behaviors or BLA^VGlut1-NAc terminal activity (Extended data Fig. 8a-i).

To determine whether 2-AG binds to presynaptic CB₁R on aPVT terminals in the NAc, we injected mice with AAV1-hSyn-GRAB_eCB2.0 or the control virus mutant-GRAB_eCB2.0 (GRAB_eCB2.0MUT that is not activated by 2-AG) into the aPVT and implanted a fiberoptic above the medial NAc shell (Fig. 3j). We detected a progressive increase in the reward engagement-induced GRAB_eCB2.0 signal from day 1 to day 5, and no change in GRAB_eCB2.0MUT signal (Fig. 3k-n). Compared to GRAB_eCB2.0MUT controls, we detected a significant increase in GRAB_eCB2.0 signal on day 5 of reward conditioning. We then tested whether 2-AG binds to aPVT terminals following exposure to aversive stimuli in a Pavlovian fear conditioning/extinction paradigm. Similar to our results with local NAc expression of GRAB_eCB2.0 (Fig. 1), foot shock elicited a robust increase in GRAB_eCB2.0 signal while the shock predictive cue elicited no change in GRAB_eCB2.0 signal during fear recall (Extended Data Fig. 8j-l).

Lastly, we sought to identify an in vivo correlate to the DSE phenomenon (neural depolarization induced production of eCBs, demonstrated in Fig. 1) within the ventral striatum. Specifically, we examined whether we could evoke 2-AG production and release through photo-stimulation of NAc neurons in vivo. Our electrophysiological results indicated that most of the phasic 2-AG release in the NAc is mediated via D₂R neurons, which have been shown to co-express the opioid peptide enkephalin with 95% overlap with D₂R expression^17,43. Thus, we injected Proenkephalin Cre (Penk-Cre) mice with the red-shifted channelrhodopsin AAV5-DIO-ChRimson-tdTomato in the medial NAc shell and GRAB_eCB2.0 in the aPVT along with an implanted fiber optic in the NAc for tandem optogenetics and fiber photometry recordings of aPVT-NAc GRAB_eCB2.0 signal (Fig. 3o). Photo-activation of NAc Penk neurons (20hz, 10s) was sufficient to elicit a significant increase in GRAB_eCB2.0 signal, indicating in vivo evoked release of 2-AG (Fig. 3p). Overall, these findings demonstrate that in vivo release of 2-AG in the NAc functionally tunes the degree of engagement in reward-seeking behavior through time-locked gain control of aPVT axon terminal activity (Fig. 3q).

aPVT-NAc^Penk neurons facilitate engagement in reward-seeking in an eCB dependent manner

Our evidence supports the role of 2-AG-CB₁R signaling in filtering excitatory aPVT^NTS input to the NAc to control reward-seeking, and that NAc D₂R/Penk neurons produce 2-AG. However, the necessity of activity-dependent 2-AG release from NAc D₂R/Penk neurons for driving engagement in reward-seeking remained a critical unresolved question. To test this hypothesis, mice performed Pavlovian reward conditioning in a custom-built linear track behavioral corridor which allowed for multi-plane video recordings of animal behavior. To garner unbiased quantifiable parameters and features of engagement in reward seeking behaviors, we leveraged a custom SLEAP pose estimation algorithm to identify and track specific locations on the mouse’s body (i.e., points of interest) during reward conditioning (Fig. 4a)⁴⁴. A supervised machine learning algorithm was subsequently employed to identify frame-by-frame instances of sipper port engagement, sipper port disengagement, walking, rearing, and grooming (Fig. 4b-h).

Fig. 4. aPVT-NAc^PENK neurons facilitate engagement in reward-seeking in an eCB dependent manner.

(a) Schematic for Pavlovian reward conditioning in the Linear Track and subsequent SLEAP labeling and supervised classification of behavior. (b) Example frames of 5 classified behaviors: sipper port engagement, disengagement, walk, rear, and groom. (c) Ethogram of the five classified behaviors from 10 trained mice over a 10 minute epoch. (d-h) Cue-aligned traces of % time engaging with sipper port, disengaging from sipper port, walking, rearing, and grooming in trained mice (N=12). (i) Schematic for trans-synaptic capture of NAc Penk neurons that receive aPVT input for subsequent optogenetic stimulation of aPVT-NAc^Penk neurons. (h) schematic for engagement-elicited closed-loop activation of aPVT-NAc^PENK neurons. (k) % time engaging with sipper port on counterbalanced ChRimson stimulation / no stimulation days (N=7). (l) Quantification of % sipper port engagement in the pre (-10 – 0s, p=0.0483, N=7) and post (50 – 60s, p=0.0422, N=7) cue ITI periods. (m, n) same as figures k and l for eYFP control mice (p=0.9205, p=0.7546, N=5). (o) % time engaging with sipper port on counterbalanced ChRimson stimulation / no stimulation days in mice treated with 25 mg/kg DO34. (p) Quantification of % sipper port engagement in the pre (-10 – 0s, p=0.4465, N=7) and post (50 – 60s, p=0.5292) cue ITI periods in mice treated with 25 mg/kg DO34. (q, r) same as figures o and p for eYFP control mice (p=0.8803, p=0.7369, N=5). All error bars represent ± SEM; N represents number of mice. p values reported from two-tailed paired t test (l,n,p,r). *p<0.05.

Functional studies have garnered conflicting reports regarding the role of NAc D₂R/Penk neurons in reward seeking and reinforcement^25,45,46,47. This could be due in part to the fact that these studies which used optogenetics to manipulate this population targeted all Penk neurons in the NAc, which represents 40–50% of the entire NAc neuron population, and the recent evidence of diversity of encoding features of these neurons^48,49. To gain further specificity and precision, we devised an anterograde transsynaptic labeling strategy which allowed us to specifically capture NAc Penk neurons that received input from the aPVT. We leveraged a recently developed AAV1 virus (AAVtrans-DIO-FLP) that shows strong anterograde transsynaptic infectivity to drive expression of ChRimson in Penk neurons that receive aPVT input⁵¹ (Fig 4i). Following the characterization of its labeling (Extended data Fig. 9a), we co-injected Penk-Cre mice with AAVtrans-DIO-FLP into the aPVT combined with AAV-fDIO-ChRimson into the NAc shell, followed by fiber optic implantation (Extended data Fig. 9b).

To specifically manipulate the activity of aPVT-NAc^Penk neurons during reward-seeking, we utilized real-time SLEAP tracking and supervised classification of behaviors to detect sipper port engagement (see methods). We subsequently generated a closed-loop optogenetic stimulation paradigm within the SLEAP architecture in which behavioral engagement signatures elicited 20hz optogenetic stimulation of aPVT-NAc^Penk neurons. We found that this stimulation selectively increased engagement in reward seeking during ITI periods preceding and following cue and sipper delivery (Fig. 4k-n, Extended data Fig. 9c-f). These effects were specific to the ITI periods and did not increase engagement during sipper delivery, suggesting closed-loop stimulation selectively reinforces engagement in reward seeking while not directly affecting consummatory behaviors. To further corroborate the specificity of the SLEAP-based closed-loop optogenetic effect, we tested an additional closed-loop experiment in which stimulation was instead triggered only during walk events (Extended data Fig. 9g). This stimulation during walk epochs had no effect on engagement in reward-seeking (Extended data Fig. 9h-k). Finally, we assessed the necessity of eCB release from NAc D₂R/Penk neurons for the closed-loop optogenetically-induced increase in engagement of reward seeking. To test this, we pre-treated mice with 20mg/kg DO34 to block production of 2-AG (Fig. 4o-r). This pretreatment completely abolished the effect of NAc D₂R/Penk ChRimson stimulation on increasing sipper-port engagement. Overall, these studies further establish that activity-dependent production of 2-AG from aPVT-NAc^Penk neurons facilitates engagement in reward-seeking.

Encoding of behavioral engagement by aPVT-NAc^Penk neurons

Transsynaptic labeling allowed us a singular ability to manipulate a specific population of NAc Penk neurons, however it has been established that this population is not homogenous in its dynamics during Pavlovian reward conditioning^48,49. Since we had determined that reward-seeking-induced 2-AG release occurs in an activity dependent manner, we next assessed how engagement in reward-seeking is encoded by the activity of discrete ensembles of NAc Penk neurons. To understand how these behaviors were encoded at the neural level, we co-injected Penk-Cre mice with AAVtrans-DIO-FLP into the aPVT combined with AAV-DJ-fDIO-GCaMP6s into the medial NAc shell to induce expression of GCaMP only in NAc Penk neurons that specifically receive aPVT efferent input. We then implanted a 0.6 mm GRIN lens over the medial NAc shell (Fig. 5a, Extended data Fig. 9l). Using the spatial footprints of the neurons in the field of view, we tracked and matched 206 aPVT-NAc^Penk neurons between day 1 and day 5 of Pavlovian reward conditioning.

Fig. 5. Encoding of behavioral engagement by aPVT-NAc^Penk ensembles.

(a) Schematic for trans-synaptic capture of NAc Penk neurons that receive aPVT input for subsequent 1-photon recordings of GCaMP6s of aPVT-NAc^Penk neurons and GRIN lens placement and representative maximum-projection image aPVT-NAc^Penk neurons. (b,c) Average cue-aligned trace and heatmaps for 206 matched neurons from day 1 and day 5 of Pavlovian reward conditioning (N=10). (d) Hierarchical clustering revealed 5 distinct clusters of aPVT-NAc^Penk neurons based on cue-aligned calcium signal on PavD5. (e-i)Traces of 5 unique cue-aligned clusters of aPVT-NAc^Penk neurons on day 5 of conditioning, was well as their neuron-matched traces and heatmaps from day 1 on conditioning. (j) Heat plot of beta coefficients of a Generalized Linear Model trained on each PavD5 behavior throughout the 70 second window and fit to the activity of each cluster during the corresponding time window. (k) Same as in (j) but for PavD1 behaviors. (l) Graphic summarizing results of the GLM, depicting positive encoding of engagement in reward-seeking behaviors by aPVT-NAc^Penk cluster 2 and negative encoding by aPVT-NAc^Penk cluster 3.

To examine the functional heterogeneity of these neurons, we used PCA-based dimensionality reduction followed by hierarchical clustering to partition the neurons into functional clusters^52,53 (Fig. 5b,c). These analyses were done on the Z-scored fluorescence activity of the cue-aligned calcium trace on day 5 of conditioning, and neurons were subsequently matched to day 1 activity profiles. We found that the neurons were roughly partitioned into 1 of 5 total clusters: (1) Transient excited, (2) sustained excited), (3) transient inhibited, (4) null, and (5) sustained inhibited (Fig. 5d-i, Extended Data Fig. 9m,n, see methods. To substantiate cluster identities, a Support Vector Machine (SVM) was trained on PavD5 PC coordinates and cluster labels, which demonstrated accurate decoding of cluster identity for unshuffled day 5 data but not for day 5 shuffled or day 1 shuffled/unshuffled (Extended data Fig. 9o).

Simultaneous capture of 1p-recordings of aPVT-NAc^Penk GCaMP6s activity and high-resolution Linear Track videos for SLEAP classification enabled us to determine how engagement in reward-seeking behaviors is encoded by changes in the activity of this neural population. Thus, we employed a Generalized Linear Model (GLM) fit of the calcium activity of the 5 clusters, trained on z-scored traces (derived from frame-by-frame instances) of the 5 aforementioned behaviors throughout the 70 sec window used for analysis of our cue-aligned 1-photon imaging data (10 sec before to sec after the cue)(Extended Data Fig. 10). Of all behaviors across both days of Pavlovian conditioning, we identified 2 beta coefficients with an absolute value >1, belonging to the fit of PavD5 engagement to cue cluster 1 (transient excited) (β=1.483) and cue cluster 3 (transient inhibited) (β=-1.2476) (Fig. 5j,k, Extended Data Fig. 10a-p). In this case (identity link), these coefficients represent the change in value of the response variable (cluster activity) give a 1 unit change in the predictor (behavior). Taken together, these results indicate that engagement in reward-seeking behaviors is at least partially encoded by the activity of discrete aPVT-NAc^Penk ensembles (Fig. 5l).

Discrete ensembles of aPVT-NAc_Penk neurons are entrained to aPVT axon terminal activity concomitant with endogenous endocannabinoid release

We found that the activity of specific clusters of aPVT-NAc^Penk neurons resembled either aPVT^NTS-NAc terminal activity or aPVT terminal GRAB_eCB2.0 signal. To assess whether these signals were entrained to each other, we time-aligned the signals and used a Hilbert analysis to decompose constituent waveforms and a subsequent Rayleigh test for circular uniformity (Fig. 6a, Extended Data Fig. 10q-z)^54,55. These analysis provide us with the phase offset between time varying signals and have been previously used to correlate single unit neural activity to local field potentials⁵⁶. When the clusters were aligned to aPVT axon terminal activity, we found the highest degree of entrainment (R=0.9363) between the ‘transient inhibited’ cluster (3) and the aPVT terminals with a phase offset of -21.11 degrees, suggesting the activity of the cluster lags behind the axon terminal’s activity (Fig. 6b,c). When the cluster activity was aligned with the GRAB_eCB2.0 signal, we found the greatest degree of entrainment within the ‘sustained excited’ cluster (2) (R=0.8922) with a phase offset of -1.14 degrees, suggesting that the GRAB_eB2.0 and aPVT-NAc^Penk cluster 2 signals likely emerge contemporaneously (Fig. 6d,e). These findings indicate that specific ensembles of aPVT-NAc^Penk neurons are entrained to aPVT^NTS-NAc axon terminal activity in tandem with the kinetics of localized retrograde 2-AG release. Furthermore, this result indicates that discrete post-synaptic neuronal ensembles can communicate via specific retrograde neuromodulation mechanisms such as eCBs to filter incoming excitatory drive onto striatal medium spiny neurons, dynamically impacting engagement in reward-seeking behaviors.

Fig. 6. Entrainment of aPVT-NAc^Penk neuron activity to eCB release and aPVT-NAc terminal activity.

(a) Schematic depicting alignment of aPVT-NAc^Penk GCaMP, aPVT^NTS terminal GCaMP6, and aPVT terminal GRAB_eCB2.0 signals. (b) Overlay of aPVT-NAc^Penk cluster 3 GCaMP6s signal with aPVT^NTS terminal GCaMP6s signal from day 5 and corresponding Rayleigh plot for circular uniformity (R=0.9363). (c) Summary plot of Hilbert and Rayleigh analysis depicting Pearson R and phase offset values for each clusters entrainment to aPVT^NTS terminal GCaMP6 signal. (d) Overlay of aPVT-NAc^Penk cluster 2 GCaMP6s signal with aPVT terminal GRAB_eCB2.0 signal from day 5 and corresponding Rayleigh plot for circular uniformity (R=0.8922). (e) Summary plot of Hilbert and Rayleigh analysis depicting Pearson R and phase offset values for each clusters entrainment to aPVT terminal GRAB_eCB2.0 signal. (f) Summary schematic depicting reward-seeking induced eCB production by aPVT-NAc^Penk neurons and subsequent retrograde action at aPVT terminals, and excitation of aPVT terminals upon disengagement. (g) Summary schematic depicting relative aPVT input bias onto NAc neurons at baseline and during reward-seeking. (h) Summary schematic depicting the reciprocal relationship between aPVT-NAc^Penk neuron activity, eCB release and presynaptic binding, eCB tuning of excitatory input from the aPVT, and entrainment of aPVT-NAc^Penk neurons to aPVT terminal activity.

DISCUSSION

The behavioral role of endogenous eCB signaling on distinct neural circuits as it relates to more complex behaviors such as reward seeking has been difficult to probe due to its widespread expression and a lack of molecular tools to assess eCB dynamics in vivo^2,57,58. Here, we used a series of complementary methods to uncover a specific and dynamic role for the eCB 2-AG release in driving engagement in reward-seeking behaviors. In particular, we isolate a previously undefined characteristic of eCB action at presynaptic accumbal CB₁Rs in a directed temporally discrete control of reward engagement. This retrograde gain reduction via eCBs acts as a fundamental negative feedback mechanism through which excitatory drive onto D₂R/Penk MSNs from thalamic inputs is tuned. While it has been known that eCBs can reduce excitatory transmission^6,9 ex vivo, establishing an in vivo correlate of DSE and demonstrating its causal role in facilitating reward-seeking expands our understanding of how presynaptic Gi-coupled GPCRs act as powerful facilitators of learned and motivated behaviors^59,60.

While multiple excitatory projections to the NAc express CB₁R, we detected the highest density in aPVT projectors that express the neuropeptide neurotensin. Electrophysiological characterization of this pathway revealed that these inputs are regulated by both tonic and phasic 2-AG produced by D₂R/Penk neurons. In vivo, the activity of these projections was negatively correlated with engagement in both reward seeking behaviors and conditioned freezing. However, CRISPR/Cas deletion of CB₁R from aPVT NTS neurons was only sufficient to attenuate engagement in reward-seeking behaviors (and concomitant aPVT terminal inhibition), leaving engagement in aversive behaviors intact. Given that the activity-dependent production of 2-AG was primarily evoked from D₂R/Penk neurons, we used a transsynaptic labeling strategy to show that aPVT-NAc^Penk neurons promote engagement in reward-seeking in a 2-AG dependent manner and that these neurons encode engagement in reward-seeking behaviors.

While prior work has demonstrated the feasibility of 1- and 2-photon recordings of NAc neurons in a genetically specific manner, no studies to date have recorded from NAc neurons with respect to both afferent connectivity and genetic identity⁴⁹. The transsynaptic approach we used provided us the means to image from NAc Penk neurons receiving direct monosynaptic input from the aPVT, allowing us unprecedented granularity, which is uniquely important in the NAc given the well characterized functional diversity of NAc neurons^48,49. By utilizing novel pose estimation and machine learning algorithms to garner quantitative metrics of behavioral states and transitions, we demonstrated that the activity of specific clusters of aPVT-NAc^Penk neurons specifically encode engagement in reward seeking. Together with signal processing analyses (Hilbert transform and Rayleigh test), we report that unique clusters of aPVT-NAc^Penk neurons are entrained either to eCB release or aPVT terminal activity.

Here we report a fundamental retrograde neuromodulatory mechanism that involves localized ventral striatal 2-AG production in engagement in reward-seeking behaviors through gain-control of an anatomically and genetically defined thalamo-striatal circuit (Fig. 6f,g). These results establish a causal link between single cell neural activity (aPVT-NAc^Penk), neuromodulator release (2-AG), and Gi-GPCR-mediated tuning of excitatory projections (aPVT^NTS-NAc) to reveal how eCBs dynamically regulate complex behaviors such as engagement in reward seeking (Fig. 6h). These results provide a framework for understanding neuromodulatory substrates in vivo, while also establishing potentially unique endophenotypes implicated in the pathophysiology of mood and substance use disorders.

MATERIALS AND METHODS

Animals

Adult (16–35g, 12–52 weeks) male and female WT, NTS-Cre, Penk-Cre, and Ai14 mice were group housed, given adlibitum access to food and water, and maintained on a 12 hour light dark cycle (lights off at 9:00 AM, lights on and 9:00 PM), with an ambient temperature of 70–73 degrees Fahrenheit, and humidity of ~40% . Mice were bred in a barrier facility and transferred to the holding facility at least 1 week before the start of surgical interventions or experiments. Littermate-matched controls were used in manipulation studies (optogenetics, CRISPR, behavioral pharmacology). All animals were drug and test naïve, unless otherwise noted. For all reward assays, mice were water restricted to ~90% of baseline body weight for at least 5 days before behavioral testing. If body weight dipped below ~85% of baseline weight, mice were removed from water restriction. Statistical comparisons did not detect any sex differences, and therefore male and female mice were combined for all of the data presented in this manuscript. All animals were monitored daily by either facility staff or experimenter throughout the duration of the study. All procedures were approved by the Animal Care and Use Committee at the University of Washington, and conformed to US National Institutes of Health guidelines.

Drugs

For all behavioral experiments, drugs were injected I.P. and prepared in a 17:1:1:1 vehicle consisting of saline (17), ethanol (1), kolliphor (1) and DMSO (1). Drugs were administered at the following concentrations: SR141716A (Sigma): 10 mg/kg, DO34 (Sigma): 20 mg/kg, JZL184 (Sigma): 5 mg/kg, URB597 (Sigma): 3 mg/kg. SR141716A, JZL184, and URB597 were injected 90 minutes before behavioral testing, and DO34 was injected 120 minutes before behavioral testing.

Stereotaxic surgeries

Mice were anaesthetized in an induction chamber (1%- 4% isoflurane) and placed into a stereotaxic frame (Kopf Instruments, model 1900) where they were mainlined at 1%-2% isoflurane. For all viral injections, a Hamilton Neuros Syringe was used, and virus was infused at a rate of 50–100 nl/minute. For retrograde tracing studies, 150nL of AAV2-retro-DIO Cre was injected into the medial NAc shell of Ai14 mice at two D/V coordinates per hemisphere, A/P 1.7, D/V 4.4 and 4.1, M/L 0.65. For ex vivo electrophysiology studies, NTS-Cre mice were injected with 500nl of AAV5-DIO-ChR2-eYFP in the aPVT (14° M/L angle, with final coordinates of A/P -0.4, D/V -3.75, M/L 0). For NAc fiber photometry experiments, 500nl of AAV1-hSyn-GRAB_eCB2.0 was injected into the right medial NAc shell (A/P +1.7, D/V -4.2, M/L +0.65) and 400μM fiber optic with a stainless steel ferule (Doric) was implanted at the same coordinates. All implants were secured using Metabond (Parkell). For terminal aPVT fiber photometry experiments, AAV5-DIO-GCaMP6s (Addgene) or AAV1-hSyn-GRAB_eCB2.0 was injected into the aPVT of NTS-Cre or WT mice respectively at a 14° angle, with final coordinates of A/P -0.4, D/V -3.75, M/L 0, with a fiberoptic implanted in the NAc shell (same coordinates as above). For CRISPR/photometry experiments, AAV-DJ-DIO-GCaMP6s was mixed in a 2:1 ratio with AAV1-FLEX-sgCNR1-SaCas9 or AAV1-FLEX-sgROSA26-SaCas9 and then injected into the aPVT of NTS-Cre mice or the BLA (A/P -1.25, M/L +/1 3.15, D/V -5.1) of VGlut1-Cre mice. These viruses were single vector CRISPR/Cas9 constructs which carried a guide RNA directed against the against a sequence in the first coding exon of cnr1 (sgCNR1) or a guide RNA directed against the Rosa26 locus to control for double stranded breaks. Three weeks after viral injection, a 400μM fiber optic was implanted in the right medial NAc shell at A/P +1.7, D/V -4.2, M/L +/- 0.65. For optogenetics experiments, 500nl of AAV5-DIO-ChR2-eYFP or AAV5-DIO-PPO-eYFP (Addgene) was injected into the aPVT of NTS-Cre mice. Three weeks after injection, bilateral 200μM fiberoptics were implanted over the medial NAc shell at a 10° angle (final coordinates of A/P +1.7, D/V -4.0, M/L +/- 0.65). For recording aPVT terminal GRAB_eCB2.0 with concomitant optogenetic activation of NAc PENK neurons, PENK-Cre mice were injected with 500nl of AAV1-hSyn-GRAB_eCB2.0 in the aPVT and 500nl of AAV5-DIO-ChRimson in the NAc, with a fiberoptic implanted above (same coordinates as above). For closed loop optogenetic studies in the Linear track, 500nL of AAV1-trans-DIO-FLP was injected into the aPVT of PENK-Cre mice. Three weeks after this injection, 500nL of AAV5-fDIO-ChRimson was injected into the medial NAc shell (A/P +1.7, D/V -4.2, M/L +/- 0.65) with 200μM fiberoptics implanted above the injection site (A/P +1.7, D/V -4.0, M/L +/- 0.65). For 1-photon imaging experiments, 500nL of AAV1-trans-DIO-FLP was injected into the aPVT of PENK-Cre mice. Three weeks later, 250 nL of AAV5-fDIO-GCaMP6s was injected in the right medial NAc shell at two D/V coordinates per hemisphere, A/P 1.7, D/V 4.4 and 4.1, M/L 0.65. During the same surgery, a 600μM cuffed GRIN lens (Inscopix) was implanted at coordinates of A/P 1.7, D/V 4.15 M/L 0.65. Three weeks after viral injection and lens implantation, we attached a baseplate to an Inscopix miniscope to identify the optimal imaging plane. The baseplate was subsequently attached to the implant with Metabond. All mice were given at least 3 weeks of recovery time post-surgery before experimentation.

Tissue processing

For all in vivo optogenetics, fiber photometry, or 1-photon imaging experiments, mice were transcardially perfused with a 4% paraformaldehyde solution in phosphate buffered saline. Head were subsequently removed and were immediately placed in to conical tubes containing the same solution and allowed to set for 24–72 hours. After this time, fiberoptic implants were removed along with the remainder of the skull, and brain were put into a cryoprotectant 30% w/v sucrose solution in PBS. Once brains had sunken to the bottom of the tubes (typically around 72 hours), brains were mounted on cutting blocks using O.C.T. (Sakura Finetek) and sliced in 40μm sections on a Leica CM1900 cryostat at -18°C. Slices were then placed into PBS and either immediately mounted on SuperFrost Plus slides (Fisher) or stored for up to 3 days. Mice with misplaced fiber optic implants were excluded from the study.

In situ hybridization using RNAscope

For quantification of mRNA transcripts using RNAscope (ACD Bio), mice were briefly anesthetized using isoflurane and subsequently rapidly decapitated. Brain were immediately removed from the skull and were placed into a beaker containing 2-methylbutne that had been sitting on dry ice for >20 minutes. After 30–60 seconds of submerging, brains were removed and placed into conical tubes in a -80°C freezer for long term storage. Prior to sectioning, brains were placed in the cryostat compartment for >30 minutes to allow them to come up to temperature. Following mounting using O.C.T., brains were slice at 16μm and directly mounted onto SuperFrost Plus slides. Slides were stored at -80°C.

FISH was performed according to the RNAscope 2.0 Fluorescent Multiple Kit User Manual for Fresh Frozen Tissue (Advanced Cell Diagnostics) as previously described. Briefly, slides were fixed in 40% PFA and then serially dehydrated using increasing concentrations of ethanol (50%, 75%, 100% and 100%). Slices were then treated with Protease IV and allowed to incubate in a 40°C hybridization oven for 30 minutes. Following serial rinses in a wash buffer, slices were incubated with probes directed against Cre (312281-C3), CB1 (457341), or NTS for two hours (420441-C2). Samples subsequently went through sequential amplification steps, and lastly application of Opal dyes corresponding to each channel (TSA Vivid Fluorophore 520 (PN 323271), TSA Vivid Fluorophore 570 (PN 323272) and TSA Vivid Fluorophore 650 (PN 323273)). Slides were counterstained with DAPI, and coverslips were mounted with Vectashield Hard Set mounting medium (Vector Laboratories).

Image acquisition and analysis

For histological verification of fiberoptic/ lens placement, images were obtained on a Leica DM6B epifluorescent microscope at 10x or 20x magnification. For RNAscope analysis, images were collected on an Olympus Fluoview 3000 at 20x or 40x magnification. Z-stacked images were taken with optical section thickness of 2–3μM. Images were subsequently analyzed using either HALO or QuPath software. For either analysis, brain region boundaries were drawn in accordance with the Allen Brain Institute mouse reference atlas. For HALO, DAPI stained nuclei were used to mark cells, and cell boundaries were set to ~8 μM radius from nucleus center. Thresholds were set for detection in each channel and this threshold was applied to all analyzed images in a given cohort. In order to account for background noise, a neuron had to have >2 transcripts in order to be classified as positive for a specific marker. For QuPath analysis, DAPI was similarly used to define neurons and cell boundaries were set to ~8 μM radius from nucleus center. Individual channels were set via the classifier module, and these classifier setting were held consistent for all images within a cohort. The classifiers were set by individually sorting through each cell automatically registered by the QuPath software and identifying the first ‘false positive’ neuron, using this as the threshold cut off for subsequent identification of positive neurons.

Electrophysiology

Coronal brain slices were prepared at 250 μM on a vibrating Leica VT1000S microtome Westlake Texas), and transcardially perfused with ice-cold and oxygenated cutting solution consisting of (in mM): 93 N-Methyl-D-glucamine (NMDG), 2.5 KCL, 20 , 10 MgSO4·7H20, 1.2 NaH2PO4, 0.5 CaCl2·2H20, 25 glucose, 3 Na⁺-pyruvate, 5 Na⁺-ascorbate, and 5 N-acetylcysteine. Following collection of coronal sections, the brain slices were transferred to a 34°C chamber containing oxygenated cutting solution for a 10-minute recovery period. Slices were then transferred to a holding chamber consisting of (in mM) 92 NaCl, 2.5 KCl, 20 HEPES, 2 MgSO4·7H20, 1.2 NaH2PO4, 30 NaHCO3, 2 CaCl2·2H20, 25 glucose, 3 Na-pyruvate, 5 Na-ascorbate, 5 N-acetylcysteine and were allowed to recover for ≥ 30 min. For recording, slices were perfused with oxygenated artificial cerebrospinal fluid (ACSF; 31–33°C; 300–303 milliosmols) consisting of (in mM): 113 NaCl, 2.5 KCl, 1.2 MgSO4·7H20, 2.5 CaCl2·6H20, 1 NaH2PO4, 26 NaHCO3, 20 glucose, 3 Na⁺-pyruvate, 1 Na⁺-ascorbate, at a flow rate of 2–3ml/min. Stocks of 50mM SR1 or 5mM JZL184 were made in DMSO and diluted 1:5000 in ACSF or HEPES for a final concentration of 10μM and 1μM, respectively. 0.05% w/v Bovine Serum Albumin (Sigma-Aldrich) was added to ACSF or HEPES to help the drugs from precipitating out of solution. For all drug experiments, slices were incubated in a HEPES bath containing the drugs for >30 minutes before transferring over to the slice rigs for recording.

NAc neurons were initially voltage clamped in whole-cell configuration using borosilicate glass pipettes (2–4MΩ). filled with internal solution containing (in mM): 125 K⁺-gluconate, 4 NaCl, 10 HEPES, 4 MgATP, 0.3 Na-GTP, and 10 Na-phosphocreatine (pH 7.30–7.35). The patch pipette also included 50 μM picrotoxin to block GABA_A currents. D1+ medium spiny neurons (MSNs) were identified by tdTomato fluorescence. For all D1- neurons, multiple electrophysiological metrics were recorded to ensure we were specifically recording from MSNs. This included capacitance, membrane resistance, resting membrane potential, sag current, afterhyperpolarization current, and evoked action potential firing. Multiple publications have identified that these features differ between MSNs and interneurons, and any neurons that had electrophysiological features of interneurons were excluded from the dataset (Figure S1)^63,64. Sag current was recorded following a 1 second 100pA hyperpolarizing current, and the amplitude of the deflection below the stable baseline was calculated. For afterhyperpolarization current, we analyzed the amplitude of the inward current immediately following a train of action potentials induced by a 500pA current step. Action potential firing was monitored during 40 consecutive successive sweeps of currents starting at -100pA and increasing by 20pA each time. Following break-in to the cell, we waited ≥ 3 minutes to allow for exchange of internal solution and stabilization of membrane properties. Neurons with an access resistance of > 30MΩ or that exhibited greater than a 20% change in access resistance during the recording were not included in our datasets.

Ex-vivo optogenetics

For electrophysiological interrogation of the aPVT^NTS-NAc circuit, mice were injected with 500 μL of AAV5-DIO-ChR2(H134R)-eYFP into the aPVT. 3–5 weeks of viral expression was allowed prior to sacrificing the mice. For all experiments using SR141716A or JZL184, corresponding vehicle recordings were obtained from the same mice. For optogenetic recordings of input/output curves, we used a Thorlabs LEDD1B T-Cube driver and obtained separate recordings of 470nm wavelength oEPSCs at 7 output levels corresponding to 5.7, 2.7, 1.6, 1, 0.5, 0.2, and 0.1 mW of LED intensity. PPR recordings of oEPSCs were obtained in voltage-clamp with an inter-stimulus interval of 50ms. PPR is reported as a ratio between the amplitude of the second oEPSC divided by the first. Recordings of DSE were obtained following at 10 s voltage step to +30mV. A baseline of 10 oEPSCs were taken prior to the depolarizing step, and all data is plotted as an oEPSC amplitude normalized to the baseline period. A light exposure time of 2ms was used for all optogenetic experiments.

Fiber photometry

Fiber photometry recordings were obtained throughout the duration of all behavioral tests presented in this manuscript. Prior to recordings, a fiberoptic cable was attached to the fiberoptic implant using a plastic or ceramic ferrule sleeve (Doric, ZR2.5). For all photometry recordings, we used a Tucker-Davis Technologies RZ10x processor. A 531-Hz sinusoidal 470 nm LED light (Lx465) was used to excite GCaMP6s and evoke Ca2+-dependent emission. A 211-Hz sinusoidal 405 LED (Lx405) light was used as the Ca2+-independent isosbestic control emission. For GRAB_eCB2.0 recordings, we used a 490 nm LED for GRAB_eCB2.0 excitation and a 435 nm LED as an isosbestic control, as prior studies indicated that the peak excitation wavelength for this sensor is slightly red-shifted³³. LED intensity was measured at the tip of the cable with a dummy implant attached and adjusted to 30μW before each day of recording. GCaMP6s fluorescence traveled through the same optic fiber before being bandpass filtered (525 ± 25 nm, Doric, FMC4), transduced by a femtowatt silicon photoreceiver (Newport, 2151) and recorded by a real-time processor (TDT, RZ10). The envelopes of the 531-Hz and 211-Hz signals were extracted in real- time by the TDT program Synapse at a sampling rate of 1017.25 Hz. For ChrimsonR stimulation experiments, a 635nm laser was used with a custom filter cube (Doric) at 5–7mW of intensity to deliver 20hz pulsed light through the same fiberoptic cable used for photometry recordings.

Photometry analysis

Custom MATLAB scripts were developed for analyzing fiber photometry data in context of mouse behavior and can be accessed via GitHub. The isosbestic 405/435nm excitation control signal was scaled to the 470/490nm excitation signal, then this refitted 405/435nm signal was subtracted from the 470/490 nm excitation signal to remove movement artifacts from intracellular Ca2+-dependent GCaMP6s/ GRAB_eCB2.0 fluorescence. Baseline drift was evident in the signal due to slow photobleaching artifacts, particularly during the first several minutes of each hour-long recording session. A double exponential curve was fit to the raw trace and subtracted to correct for baseline drift. After baseline correction, Δf/f was calculated by subtracting the median baseline subtracted signal from the baseline subtracted signal and dividing by the median raw signal. All photometry data presented in this manuscript is Z-scored to a 10–20 second window prior to cue onset.

Behavioral apparatuses and analysis

Pavlovian reward conditioning

Mice were habituated to a Med Associates operant conditioning box (MED-307W-D2) for 2–3 days (30 minutes per session), during which a sipper containing a 10% sucrose solution was extended into the chamber. Mice were subsequently trained in Pavlovian Reward Conditioning, in which the houselight was illuminated 6 seconds prior to sipper extension, which lasted for 20 seconds, followed by sipper retraction and houselight shut-off. Licks were recorded via a contact lickometer and registered in MedPC. The cue/sipper pairing was presented in 20 pseudorandom trials with ITIs of 30, 60, 90, and 120 seconds. For all photometry experiments, a 1080P webcam was mounted above the chamber for subsequent mouse behavior analysis. ‘Approach’ behaviors were classified as nose/head entrances into a 3 inch by 3 inch square area around the sipper port, and ‘exits' were classified as instances of leaving this area. For engagement in reward-seeking behaviors, we analyzed the % of time the mouse spent pawing, sniffing, licking, or poking at the reward port in 3 defined time windows: 0–6 seconds (cue period, pre-sipper), 0–26 seconds (entire cue period), or 26–46 seconds (post-cue/sipper time period). ‘Null’ behaviors were classified as any non-engagement, approach, or exit behavior lasting more than 1 second in duration. These behaviors were analyzed in a blinded fashion. For creation of the transition diagrams, ‘null’ and ‘exit’ behaviors were grouped for visualization purposes. The same Pavlovian protocol (identical cue/reward pairings, ITI, sucrose concentration) was used for conditioning in the Linear Track (see below).

Operant condition

Using the same apparatus for Pavlovian reward conditioning, we trained mice on FR1, FR3 and PR schedules of reinforcement. Animals nosepoked into illuminated nose-poke ports, randomly designated as ‘active’ or ‘inactive’. Following a second delay, an active nosepoke resulted in cue light presentation and subsequent protraction of the sipper, which stayed protracted for 20s. In a FR1 schedule of reinforcement, one nosepoke elicited sipper protraction. Following 3 days of FR1 training, mice were trained in a FR3 schedule which necessitated 3 consecutive active poke to elicit sipper protraction. Finally, in the PR task, the number of nosepokes required to elicit sipper protraction escalated exponentially as previously described⁶⁵.

Pavlovian fear conditioning and extinction

In separate Med Associates chambers (NIR-022MD), mice were trained in Pavlovian fear conditioning. The chamber consisted of a conductive grid floor (29.53 × 23.5 cm) illuminated by an IR light. Mice were conditioned to 10 20khz tones lasting 10 seconds, co-terminating with a 2 second .5mA shock. Tones were presented in a pseudorandom pattern. For fear extinction, mice were exposed to 20 20hz tones lasting 20 seconds for five consecutive days. Freezing behaviors was calculated manually by an experimenter blinded to animal group and genotype.

In vivo optogenetics

For all experiments using ChR2 we used a 20hz stimulation frequency with a power output between 6 and 8 mW (measured at fiber tip). For all experiments using PPO we used a 10hz stimulation frequency with a power output between 8 and 10 mW.

Pavlovian reward conditioning

For reward conditioning experiments, mice were exposed to the identical training regimen as described above. For modulation of reward-seeking behaviors, we used two separate stimulations paradigms for ChR2. On counter-balanced days, mice either received 20hz stimulation throughout the duration of the cue period (26s) or received closed-loop 20hz stimulation in which each lick triggered (via TTL) a 2 second stimulation pulse. For PPO, we drove 10hz inhibition via a TTL pulse which began the stim pattern 5 seconds before houselight shut-off and continued for 20 seconds after houselight shut-off.

Pavlovian fear conditioning

For reward conditioning experiments, mice were exposed to the identical training regimen as described above. Mice received optical stimulation or inhibition on day 1 of extinction (fear recall day). For ChR2, mice received 20hz stimulation time-locked to cue presentation (20s). For PPO, mice received 10hz inhibition time-locked to cue presentation (20s).

Operant self-stimulation

Using the operant boxes described above, mice were trained in an FR1 schedule of reinforcement. One nosepoke either elicited a 2 second 20hz stimulation (ChR2) or 10hz inhibition (PPO) pulse.

Linear track recordings

For high resolution point of interest and kinematic behavioral analysis, a linear track behavior corridor was utilized to capture dual perspectives of animal positions and behavior (bottom-up and side profile). The chamber was modified from the original published by Machado et al 2015 with a mirror positioned at a 45° angle below a clear Plexi-glass floor⁶⁶. A single high-frame rate camera was used to capture minute, quick animal actions during experimental windows (Basler ace camera: acA800–150um). The linear track was illuminated with 4 infrared (IR) floodlights and 2 supportive IR lights positioned around the chamber to brighten all angles of animal positions. All videos were recorded at 60 frames per second. The camera was operated via a custom python script which allowed for manipulation of session length, exposure, and capture rate (Github) outputting compiled mp4 videos for SLEAP analysis. The maximum session length was set to 40 minutes, although most recordings were stopped prior to this point, coinciding with the end of the behavioral assay.

Machine learning behavioral tracking

Tracking and pose-estimation of videos within the linear track was performed using the SLEAP platform, an open-source computer vision system utilizing a custom convolutional neural network designed for variational animal behavior tracking⁴⁴. A custom, in-house tracking algorithm was developed for our linear track chamber using a subset of 1767 frames from 313 videos of adult (6 week to 6 month) 51% male and 49% female mice in different lighting conditions to ensure model robustness and versatility. This model was utilized to track 29 selected points of interest across all frames of all behavioral videos and were then exported as .h5 files for subsequent analysis. These 29 points were selected to represent the status of the animal from the bottom-up and side profiles in a single-animal model (e.g. nose-side profile and tail base-bottom profile). Of the total training frames, 50% were randomly selected by SLEAP and 50% were hand-picked for valuable frames to improve model accuracy. Final model precision and recall was 97.15% and 98.74%, respectively. An analysis package was used to export the positional data and calculate features for respective classification models.

Supervised analysis of standard behaviors

Identification of core behavioral responses within the linear track was performed with an automated supervised ML algorithm (random forest classifier) to identify frame-by-frame instances of walking, rearing, grooming, and laying on belly. This model was trained on 1.4 million hand-scored frames of naturalistic behavior within the linear track from a dataset of 45 videos consisting of adult (6 week to 6 month) 51% female and 49% male mice. Videos of animals were treated with either vehicle or some drug to expand the model versatility across different behavioral modifications. Model was trained on 1.35 million frames and separated test set revealed a final precision and recall of 98.3% and 97.7%, respectively. Further code was used to isolate behavioral event timestamps from the supervised model to align neural data when applicable.

Engagement Analysis

To evaluate mouse engagement with a sipper port, we built a pipeline that processes data from multiple experimental sessions. This pipeline takes three key inputs: (1) SLEAP joint output files in H5 format, containing joint coordinates of shape (2, 30, n) where n is the number of frames; (2) session names to be processed (e.g. PENKtrans_M2_D5); and (3) sipper port locations as pixel coordinates in an array of shape (2, 2). Our preprocessing stage involves three key steps. First, we detach metadata from the joint files to isolate pose data. Next, we extract the side head position, specifically the neck and nose coordinates, which are crucial for our analysis. Finally, we convert head and sipper locations from pixel coordinates to centimeters, allowing for standardized spatial analysis across different experimental setups.

Our analysis focuses on three main aspects of mouse behavior:

1. Distance calculation: We compute the Euclidean distance between the mouse's nose and the sipper port for each frame. Utilized NumPy’s linalg.norm function. Any length greater than the width of the box was clipped to the width value.

2. Head direction determination: We compare the x value of the neck position against the x values of the body positions. The body positions were averaged then checked whether the x position of the neck was greater than the body average. If it was, we labeled the frame as facing right.

3. Turning identification: We identify turning behavior based on the sign changes in the difference in distances. We only identified significant turns where the change in direction was greater than 1.5 cm. We also only analyzed every second instead of the whole series to filter out noise.

Engagement prediction

To predict periods of engagement with the sipper port, we employ a multi-step filtering process:

1. Naive engagement identification: We initially classify frames as engaged if the nose is within 2 cm of the sipper port.

2. Anatomical feasibility filter: We remove frames where the nose is more than 3 cm away from the neck, as these represent anatomically infeasible postures.

3. Direction filter: We exclude frames where the mouse is not facing the sipper, based on our head direction calculation.

4. Stability filter: We remove frames where the neck is more than 4 cm from the sipper, indicating unstable nose position.

5. Temporal coherence filter: We filter out engagement periods shorter than 1 second, with a 0.5-second tolerance for brief disengagements, to focus on sustained engagement behaviors.

Closed Loop optogenetic stimulation

Set up

1. Using a static frame capture, the coordinates of the side and bottom views of the sipper were determined

2. A conversion factor for pixels to centimeters was obtained from the static frame by marking the lower left and lower right corners of the linear track. The factor was calculated by taking the length of the linear track in centimeters divided by the length of the linear track in pixels

Real time tracking and analysis

1. Collection of animal joints: key body positions (e.g. neck, nose, truck) were determined using the aforementioned SLEAP pipeline

2. Frame level engagement detection: using the joints, engagement was detected as previously described. These data are pushed into a sliding window to allow for continuous detection and parameter adjustment.

3. Thresholded engagement detection: A TTL pulse was sent to the Arduino control box only if the number of engagements in the sliding window was equal to the threshold value.

4. Patterned optical stimulation: The TTL pulse triggered a 1 second 20hz pulse of 635nm laser stimulation (~6 mW), followed by a 1 second time out period.

Post-hoc analyses

1. Behavior analysis: The collected joints were smoothed to account for missing/invalid values, and then fed into the aforementioned Random Forest Classifier to identity basic behaviors (i.e. walk, rear, groom, null). Joints are subsequently analyzed in the aforementioned engagement processing pipeline

2. Alignment to cue: binned behaviors (1 second bins) are subsequently aligned to the cue.

Configurations

For our use, we used an analysis window of 16 frames with a threshold of 4 frames with a 2 second block period. We used the default engagement detection setting, which we defined as side and bottom view nose being 2 cm, the side view neck being 3 cm from the sipper. We also decided that for joints to be valid, the side view nose needs to be 4 cm or less away from the side view nose, and the bottom view nose needs to be 4 cm or less away from the bottom view body center.

1-Photon imaging

At least 1 week after baseplate attachment, mice were habituated to a Inscopix nVista 2.0 miniscope for 30–35 minutes for at least three days in the linear track, during which mice also had access to sipper containing a 10% sucrose solution. During the habituation phase, optimal field of view, focus, gain, and LED power were determined visually by assessing the activity of putative neurons in the field of view using the Inscopix data collection GUI. Gain values ranged between 2 and 6 and LED power was set between 20% and 50% of transmission range. Once determined, these parameters were held static for the remainder of the testing regimen.

Following this habituation period, mice were trained in Pavlovian reward conditioning as described above. Mice were scruffed and attached to the miniscope, and then immediately placed in the Linear Track. The Inscopix recordings were triggered directly by the MedPC program via a TTL pulse to ensure synchronization of event timestamps and calcium activity. The MedPC program was started 10s after starting the Linear Track camera, and videos were synced post-hoc via visible behavioral events (cue on, sipper protraction, lick, etc.). Animal rotations were compensated for by an automatic motorized commutator (Inscopix).

Calcium imaging analysis

Initial Processing

All pre-processing was done using the Inscopix data processing software. Videos were initially spatially downsampled by 4x, followed by application of a bandpass filter (0.005 to 0.5 pixel^-1). Lateral motion was corrected using the built in motion correction feature in the Inscopix data processing software. Motion corrected tiffs were subsequently exported for cell segmentation and further analysis. Using custom Matlab scripts (see GitHub), the motion corrected .TIFF video was then processed using the Constrained Non-negative Matrix Factorization approach (CNMFe), which has been optimized to isolate signals from individual putative neurons from microendoscopic imaging⁶⁷. Putative neurons were then manually sorted and registered using previously established criteria⁶⁸. For all analyses presented in this manuscript, the non-denoised, non-deconvoluted traces were used. For each cell, the raw fluorescence trace was Z-scored to a 10 second baseline period preceding the behavioral or task-related features presented in each figure.

Cell Matching

For matching of cells across days, we used the CellReg GUI as detailed in Sheintch et al 2017⁶⁹. Briefly, maximum projection spatial footprints were extracted from the exported tiff files using a custom Matlab code. Cell registration occurred in 5 subsequent steps:

1. Spatial footprints from day 1 and day 5 of Pavlovian reward conditioning were loaded into the CellReg GUI

2. Field of view alignment was performed using rigid alignment, incorporating translations and rotations, with a maximal rotation of 30 degrees

3. Probabilistic modeling was run with a maximum distance of 12 microns

4. Initial cell registration was performed using spatial correlations

5. Final cell registration incorporated the probabilistic model from step 3 as well as the spatial correlations from step 4.

Principal component analysis and hierarchical clustering

For all Pavlovian reward conditioning imaging data presented in this manuscript, cells were sorted based on their day 5 activity and then back-matched to day 1 using custom matlab codes (Github). For Hierarchical clustering, a Principal Component Analysis was initially performed based on the activity of the neurons on Day 5 of Pavlovian reward conditioning over a one minute window coinciding with cue-onset for cue-associated activity and a ten second window for engagement-associated activity. Using built in Matlab Linkage function, an agglomerative hierarchical clustering algorithm was applied to the PCA of the day 5 activity. We set an a priori threshold that in order to be classified as a cluster, the number of neurons in said cluster must comprise >10% of the total population of recorded neurons (i.e. >20 of 206 neurons). As agglomerative hierarchical clustering starts with n clusters = n cells, we began by setting the cluster number equal to our a priori threshold (i.e. 10 clusters for 10% in each, assuming equal distribution among clusters). We then sequentially decreased cluster size by 1 until our a priori criteria were met.

Cluster Validation with SVM

Input data

PCA + hierarchical clustering (deterministic) was done using MATLAB on both day 5 and day 1 data, and then export PC coordinates and cluster labels. The cluster labels are based on the day 5 data and corresponding cell in day 1 gets the same cluster label. All data is exported as an excel file.

Data preprocessing

The data and cluster labels are loaded and then split 80–20 into training and testing data, respectively. The 80% represent the training data and the remaining 20% is the testing data. Then the data is then scaled by fitting sklearn’s standard scaler to the training data and then applying it to the testing data.

Training the SVM

Used Sklearn’s SVC to perform multiclass classification using one versus one approach. Using 80% of the training data, we first performed cross validation using stratified k-fold and random search. The cross validation seem to suggest the linear kernel performed best across different numbers of PCs. We then used the identified hyperparameters.

Classifying individual neuron activity

The fitted model is used to predict the classes for the testing data, as well as entirety of day 1 data. F1 score and accuracy are reported.

Testing against pseudo-random data

the SVM was then fit to shuffled day 5 data (each time step values are shuffled), and then fit to day 5 data with shuffled cluster labels. In each case, NumPy’s permutation was used. Again, accuracy and f1 score was used to evaluate performance of the model.

Generalized Linear Model

Inputs

To analyze relationships between neural activity and behavior during cue periods, we developed a pipeline that processes and aligns multiple behavioral measurements. This pipeline requires four inputs: (1) SLEAP joint output files in H5 format, containing joint coordinates of shape (2, 30, n) where n is the number of frames; (2) engagement periods as timestamp pairs in Excel format; (3) lick events as timestamp pairs in Excel format; and (4) cue period timestamps in Excel format.

Data Preprocessing

Our preprocessing involves three main stages. First, we convert all behavioral events into binary time series, where 1 indicates the presence of a behavior and 0 indicates its absence. Second, we extract additional behavioral states (walk, rear, groom, rest) using a pre-trained random forest classifier on the SLEAP joint data. Finally, we identify disengagement events, defined as the first frame following each engagement period.

Behavior Alignment

We align all behavioral time series to cue onset, extracting data from 10 seconds pre-cue to 34 seconds post-cue. For each session, we aggregate behavior across trials by summing the binary arrays rather than averaging them. This summation approach creates a behavioral density measure that better corresponds to our calcium imaging data structure, which represents aggregate neural activity across the session rather than trial-level responses. Using density measures instead of trial-averaged behavior ensures that our predictor variables (behavioral measures) maintain the same temporal structure as our response variables (calcium signals). These density measures are then Z-scored to standardize their scale across sessions and behaviors.

Model Fitting

We employ two levels of generalized linear model (GLM) analysis:

1. Cluster-level modeling: We fit GLMs using the normalized behavioral densities as predictors and mean cluster activity as the response variable. We use a Gaussian distribution with an identity link function.

2. Neuron-level modeling: Individual neuron activity serves as the response variable, with the same model structure as the cluster-level analysis.

For both analyses, we extract and store model coefficients (β), their associated p-values, and model intercepts. Since we are using the identity link function, the β coefficients directly represent the strength and direction of the relationship between each behavior and neural activity.

Model Validation

We used residual analysis to verify model assumptions, we saw a relatively low mean residual value and standard deviation.

Phase analysis

Inputs

The analysis requires at least two time series able to be loaded as NumPy arrays, they should have the same sampling rate and length. The analysis assumes each element correspond to each other. In our case, the time series was exported from MATLAB after preprocessing and linear interpolation to ensure sampling rate and length match. No further data preprocessing was necessary.

Extracting instantaneous phase

The instantaneous phase of each time series is extracted using the Hilbert transform, creating an analytic signal, which is a complex-valued function that extends a real-valued function to the complex plane. The instantaneous phase is then obtained by taking the angle of this analytic signal. SciPy signal’s hilbert function in conjunction with NumPy was used to extract the phases.

Calculating phase difference

The phase difference between the two signals is computed by subtracting the instantaneous phase of one signal from the other. This difference is then wrapped to the range [-π, π] to ensure it falls within a single cycle.

Computing R and theta mean

To quantify the consistency of the phase difference across time, we use circular statistics. Specifically:

1. R (mean resultant length): This measure represents the concentration of phase differences around the mean. It is calculated by:

   1. Converting each phase difference to a unit vector on the complex plane (using cosine for real part and sine for imaginary part).
   2. Taking the mean of these vectors.
   3. Calculating the magnitude of this mean vector.
2. Theta mean (mean phase difference): This is the average angle of the phase differences. It is calculated by:

   1. Taking the arctangent of the ratio of the mean sine to mean cosine of the phase differences.

R ranges from 0 to 1, where 0 indicates uniformly distributed phases and 1 indicates perfectly consistent phase differences.

Rayleigh plot

The analysis results were then plotted in a polar graph, where each phase difference was plotted along with an arrow representing the R and mean theta values.

Statistical Analysis

Unless otherwise noted, all summary data are expressed as means ± SEM or ± maximum variance. Statistical significance was taken as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, as determined by the Student’s t test (paired and unpaired), linear regression, one-way analysis of variance (ANOVA), two-way repeated-measures ANOVA, followed by Holm-Sidak post hoc tests as appropriate. For all datasets presented in this manuscript, the Grubbs outlier test was performed individually on each group, with outliers removed for α values <0.05. If a metric (e.g. engagement) from a subject for a particular experiment exceeded the outlier threshold, other metrics (e.g. dF/F) from that same subject from the same experiment were excluded from the data set. Statistical analyses were performed in GraphPad Prism 9.0 or 10.0.

List of excluded data

Electrophysiology recording of PPR of D₁R+ neuron with SR1 cell 12: Grubbs outlier test

Electrophysiology recording of PPR of D₁R- neuron with Vehicle cell 12: Grubbs outlier test

Electrophysiology recording of PPR of D₁R- neuron with JZL cell 9: Grubbs outlier test

Electrophysiology recording of max amplitude of D₁R+ neuron with JZL cell 8: Grubbs outlier test

Electrophysiology recording of max amplitude of D₁R- neuron with SR1 cell 1: Grubbs outlier test

In vivo optogenetics Pavlovian reward conditioning ChR2 subject 4: never licked

In vivo optogenetics fear recall ChR2 subject 2: fiber lost

aPVT sgROSA CRISPR Vehicle Pav subject 4: Grubbs outlier test

aPVT sgCNR1 CRISPR fear conditioning subject 7: Fiber lost

aPVT-NAc GCaMP Pavlovian reward conditioning Day 2 engagement subject 10: video didn’t capture

BLA sgROSA Vehicle Pav Subject 1: Grubbs outlier test

Extended Data

Extended Data Fig. 1. 2-AG is released in the NAc following rewarding stimuli.

(a) Schematic for fiber photometry recordings of GRAB_eCB2.0 in the NAc during Pavlovian reward conditioning. (b) Fiber placements for GRABeCB2.0 cohort displayed in Figure 1. (c) Photometry trace of cue-aligned GRAB_eCB2.0 signal in SR141716A (10mg/kg) treated animals (N=11). (d-f) Photometry trace of lick-aligned GRAB_eCB2.0 signal in vehicle, SR141716A (10mg/kg), and DO34 (20mg/kg) treated animals (N=11). (g) Comparison of cue-aligned GRAB_eCB2.0 signal (0–26s post cue) showing attenuation of GRAB_eCB2.0 signal (0–20s post lick) in SR141716A (p=0.0131) and DO34 (p=0.0079) treated animals compared to vehicle (N=11). (h) Comparison of lick-aligned (0–20s) GRAB_eCB2.0 signal in SR141716A (p=0.0343) and DO34 (p=0.0469) treated animals compared to vehicle (N=11). (i-k) Photometry traces of cue-aligned GRABeCB2.0 signal in vehicle, JZL184 (5mg/kg), and URB597 3mg/kg) treated animals (N=8). (l) Comparison of cue-aligned (0–26s) GRABeCB2.0 signal in JZL184 (p=0.0914) and URB597 (p=0.6983) treated animals compared to vehicle (N=8) (m) Schematic for Pavlovian reward conditioning with different cue modalities (n,o) Photometry traces of light and tone cue aligned GRABeCB2.0 signal (N=7). (p) Comparison of light and tone aligned GRABeCB2.0 signal (N=7). (q) Comparison of number of licks in light and tone Pavlovian reward conditioning sessions (N=7). (r) Schematic for reward omission assay. (s,t) Photometry trace and quantification of cue-aligned (0–26s) GRAB_eCB2.0 signal during rewarded trials in the reward omission assay (N=11). (u,v) Photometry trace and quantification of cue-aligned (0–26s) GRAB_eCB2.0 signal during non-rewarded trials in the reward omission assay (N=11). (w) Schematic for Pavlovian fear conditioning / extinction. (x,y) Photometry trace of GRAB_eCB2.0 signal during Pavlovian fear conditioning and extinction with representative heatmaps (N=11). (z,aa) Comparison of GRAB_eCB2.0 signal during foot shock (0–5s, p=0.0049) and cued recall (0–20s, p=0.8987) (N=11). (bb) Summary schematic of elicited eCB release in the NAc. Error bars represent ± SEM (p,q,t,v,z,aa), or min/max values (g,h,l); box plots represent interquartile range with line at median; N represents number of mice. p values reported from one-way ANOVA (g,h,l) and two-tailed paired t test (p,q,t,v,z,aa). *p<0.05, **p<0.01.

Extended Data Fig. 2. Histological analysis of CB₁R expressing inputs to the NAc and histological / electrophysiological characterization of excitatory aPVT^NTS projections.

(a-c) (Top) TdTomato positive NAc projecting neurons in the medial prefrontal cortex (mPFC) Basolateral Amygdala (BLA) and ventral Hippocampus (vHip) and (bottom) RNAscope image and quantification of Cre and CB₁R transcript levels in the mPFC BLA and vHip (N=3,3,3). (d-f) Optically evoked input-output curve from D₁R+ and D₁R- neurons in vehicle (D₁R + n=19, D1- n=20, 7.36 · 10^-5), SR141617A (D₁R+ n=14 D₁R- n=12, p=0.0050), and JZL184 (D1+ n=17, D₁R- n=15, p=0.9698) treated slices. (g,h) 1μM JZL184 treatment significantly reduces the maximum oEPSC amplitude (p=0.0193) and increases the Paired Pulse Ratio (p=0.0355) in D1(+) neurons compared to vehicle treated slices (Veh: n=19, SR141716A: n=14, JZL184: n=16). (i,j) 1μM SR141716A significantly increases the maximum oEPSC amplitude (p=0.0016) and reduces the Paired Pulse Ratio (p=0.0092) in D1(-) neurons (Veh: n=20, SR141716A: n=11, JZL184: n=15). (k-n) Electrophysiological comparison of sag current amplitude (p=0.3871) afterhyperpolzarization voltage (p=0.3404), holding current (p=0.6077), and membrane resistance (p=0.6166) between D1(+) (n=14) and D1(-) (n=13) neurons. (o) Schematic for electrophysiological characterization of aPVT output to NAc, BNST, CeA, and ZI. (q-s) Optically evoked input/output curves for aPVT input to the NAc (n=21), ZI (n=8) BNST (n=4), and CeA (n=4). (t) Comparison of maximal aPVT evoked oEPSC in the NAc, ZI, BNST, and CeA. (u-w) Schematics, representative images, and quantification of collateralization of aPVT projections to the NAc and the ZI, BNST, and CeA, respectively (n=3,3,3). (x-z) Representative images and quantification of collateralization of pPVT projections to the NAc and the ZI, BNST, and CeA, respectively. Error bars represent ± SEM (d-f,p-s), or min/max values (g-n,t); box plots represent interquartile range with line at median ; n represent number of neurons. p values reported from two-tailed unpaired t-test (k-n), one-way ANOVA (g-j), and two-way ANOVA with Holm-Sidak post-hoc correction (d-f). *p<0.05, **p<0.01, ****p<0.0001.

Extended Data Fig. 3. Inhibition of aPVT^NTS-NAc terminals is not dependent on reward receipt and is not associated with locomotor activity.

(a) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during ad libitum sucrose consumption and Pavlovian reward conditioning. (b) Fiber placements for aPVT^NTS-NAc GCaMP cohort displayed in figure 2. (c) Photometry trace of GCaMP6s signal aligned to first and last licks in a licking bout during ad libitum consumption (N=10). (d,e) Transition diagram showing probabilities of transitions between approach, engage, and null/exit behaviors on day 1 and day 5 of conditioning during the post-sipper time window, with size of circles corresponding to number of events. (f,g) Photometry traces of GCaMP6s signal aligned to first lick and last lick on day 1 of reward conditioning. (h,i) Photometry traces of GCaMP6s signal aligned to first lick and last lick on day 5 of reward conditioning. (j,k) Quantification of first lick (0–20s, p=0.8736, N=10) and last lick (0–20s, p=0.0004, N=10) aligned GCaMP6s signals on day 1 and day 5 of reward conditioning. (l) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during reward omission. (m,n) Photometry traces of GCaMP6s signal during rewarded and non-rewarded trials during the reward omission test, demonstrating aPVT^NTS-NAc terminal inhibition in the absence of reward receipt. (o) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during reward conditioning. (p) Photometry trace of GCaMP6s signal on day 5 of reward conditioning during ignored cue/reward trials (no reward engagement), demonstrating a lack of inhibition when the cue/reward is ignored. (q) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during Fixed Ratio 1 (FR1) training and Progressive Ratio (PR) training. (r,s) Photometry traces of GCaMP6s signal aligned to cue on day 1 and day 3 of FR1 conditioning, demonstrating inhibition during active operant responding for reward. (t,u) Photometry traces of GCaMP6s signal aligned to cue and nose poke during PR conditioning. (v) Schematic for fiber photometry recordings of GCaMP6s from aPVT^NTS-NAc terminals during the open field assay. (w,x) Photometry traces of GCaMP6s signal aligned to walk and rear events during the open field assay. All error bars represent ± SEM; n represent number of neurons. p values reported from two-tailed paired t-test (j,k). ***p<0.001.

Extended Data Fig. 4. aPVT^NTS-NAc terminal activity is negatively correlated with engagement in defensive freezing behaviors.

(a) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during Pavlovian fear conditioning and extinction. (b) Photometry trace of GCaMP6s signal during Pavlovian fear conditioning. (c) Quantification of GCaMP6s signal during foot shock (0–5s, N=11, p=0.0002). (d) Photometry trace of GCaMP6s signal aligned to freeze end on extinction day 1. (e) Quantification of GCaMP6s signal aligned to freeze end (0–20s, N=11, p=0.0024). (f) Correlation between freezing time and tone-aligned photometry Z-score (0–20s) across 5 days of extinction (N=11, R²=0.13, p=0.0068). (h) Schematic for optogenetic manipulation of aPVT-NAc terminals using DIO-ChR2 and DIO-PPO. (h) 20hz optical stimulation ChR2 (N=9) expressing aPVT^NTS-NAc terminals during the 20 second cue presentation decreases freezing time compared to eYFP (N=9) controls (p=0.0053). (i) 10hz optical stimulation of PPO (N=14) expression aPVT^NTS-NAc terminals does not alter freezing behavior compared to eYFP (N=11) controls (p=0.1470). Error bars represent ± SEM (c,e), or min/max values (h,j); box plots represent interquartile range with line at median; N represent number of mice. p values reported from paired two-tailed t-test (c,e), unpaired two-tailed t-test (h,j) and simple linear regression (f). **p<0.01, ***p<0.001.

Extended Data Fig. 5. Neither activation nor inhibition of aPVT^NTS-NAc terminals alters locomotion, drives real-time place preference (RTPP), or drive intracranial self-stimulation (ICSS).

(a)Schematic for optogenetic manipulation of aPVT-NAc terminals using DIO-ChR2 and DIO-PPO. (b) 20hz photo-stimulation of ChR2 expressing aPVT^NTS-NAc terminals during cue presentation reduces engagement in consummatory behaviors (ChR2: N=9 p=0.0427, eYFP: N=9), p=0.8190). (c,d) Neither 20hz photo-stimulation (2 mins off, 2 mins on, 2 mins off) of ChR2 expressing aPVT^NTS-NAc terminals (N=10, p=0.3655) nor eYFP expressing aPVT^NTS-NAc terminals (N=8, p=0.8947) alters locomotor behavior. (e) 20hz photo-activation of aPVT^NTS-NAc terminals does not support RTPP (ChR2: N=10 p=0.8141, eYFP: N=9 p=0.7191). (g) 20hz photo-activation (2 second burst per nose poke) of aPVT^NTS-NAc terminals does not support ICSS (ChR2: N=10 p=0.0968, eYFP: N=8, p=0.1670). (h)10hz photo-inhibition of aPVT^NTS-NAc terminals does not support RTPP (PPO: N=7 p=0.7492, eYFP: N=6 p=0.3216). (h) 10hz photo-inhibition of aPVT^NTS-NAc terminals does not support ICSS (PPO: N=13 p=0.2565, eYFP: N=11 p=0.4527). (i) Fiber placements for optogenetics cohort displayed in Figure 2. Error bars represent ± SEM (b,e-h), or min/max values (c,d); box plots represent interquartile range with line at median; N represent number of mice. p values reported from paired two-tailed t-test (b,e-h), and one-way ANOVA (c,d). *p<0.05.

Extended Data Fig. 6. Effect of CB₁R antagonism on inhibition of aPVT^NTS-NAc terminals and characterization of AAV1-FLEX-sgCNR1 deletion of CB₁R from aPVT NTS neurons.

(a) Schematic for fiber photometry recordings from GCaMP6s expressing aPVT^NTS-NAc terminals during Pavlovian reward conditioning. (b,c) Photometry trace with representative lick raster and heatmap of cue-aligned GCaMP6s signal in control mice or mice treated with 10mg/kg SR141716A (N=10). (d,e) Effect SR141716A on cue-aligned photometry Z-score (p=0.0064) and total number of licks (p=0.0270) throughout the session (N=10). (f,g,h) Diagram showing % of neurons expressing NTS, CB₁R, or NTS and CB₁R in the aPVT from NTS-Cre mice injected with sgROSA (N=6) or sgCNR1 (N=6). (i) Quantification of % of total neurons with CB₁R transcript (p=0.0064). (j) % of NTS neurons CB₁R transcript (p=0.0333). (k) % of NTS neurons expressing CB₁R (p=0.0055). (l)% of total neurons expressing NTS (p=0.6507). (m) % of total neurons that are NTS positive and CB₁R negative (p=0.0354). (n) % of total neurons that are CB₁R positive and NTS negative (p=0.1512). Error bars represent ± SEM (d,e), or min/max values (i-n); box plots represent interquartile range with line at median; N represent number of mice. p values reported from paired (d,e) and unpaired (i-n) two-tailed t-test. *p<0.05, **p<0.01, ***p<0.001.

Extended Data Fig. 7. Effect of AAV1-FLEX-sgCNR1 deletion of CB₁R from aPVT NTS neurons and BLA Vglut1 neurons on reward consumption and Pavlovian reward conditioning.

(a) Schematic for CB₁R deletion from the aPVT and photometry recordings ofGCaMP6s aPVT^NTS-NAc terminals during adlib sucrose consumption and Pavlovian reward conditioning. (b) Fiber placements for aPVT CB₁R deletion / aPVT^NTS-NAc cohort in figure 3. (c,d) Photometry traces of GCaMP6s signal aligned first lick during adlib sucrose consumption from sgROSA (N=7) and sgCNR1 (N=7) injected mice. (e) Comparison of first lick-aligned (0–20s) GCaMP6s signal between sgROSA (N=7) and sgCNR1 (N=7) injected mice (p=0.1741) during adlib sucrose consumption. (f-h) Comparison of total lick number (p=0.3705), bout length (p=0.0393, and bout number (p=0.2828), between sgROSA (N=7) and sgCNR1 (N=7) injected mice during adlib sucrose consumption. (i) Comparison GCaMP6s signal (0–20s) during rewarded trials in vehicle treated sgROSA (N=6) and sgCNR1 (N=7) injected mice during Pavlovian reward conditioning (p=0.0373). (j) Comparison of # of licks showing a main effect of DO34 treatment on reducing sucrose consumption (Main effects: Treatment p=0.0025, genotype p=0.0146, interaction p=0.5607). (k,l) Photometry traces of GCaMP6s signal aligned to footshock during Pavlovian fear conditioning. (m,n) Photometry traces of GCaMP6s signal aligned to tone presentation of day 1 of fear extinction. (o) Schematic for CB₁R deletion from the BLA and photometry recordings of GCaMP6s BLA^VGlut-NAc terminals during adlib sucrose consumption and Pavlovian reward conditioning. (p) Fiber placements for BLA CB₁R deletion / BLA^VGlut-NAc photometry. (q,r) Photometry traces of GCaMP6s signal during adlib sucrose consumption from sgROSA (n=7) and sgCNR1 (n=7) mice. (s) # of licks during adlib sucrose consumption (sgROSA N=7, sgCNR1 N=7). (t) Comparison of lick-aligned GCaMP6s signal during adlib sucrose consumption. (u-x) Photometry traces of GCaMP6s signal from vehicle and DO34 treated sgROSA (N=7) and sgCNR1 (N=7) mice. (y) Comparison of engagement time demonstrating a main effect of both treatment and genotype (Main effects: Treatment p=0.0025, genotype p=0.0146, interaction p=0.5607, (sgROSA N=7, sgCNR1 N=7). (z) Comparison of GCaMP6s signal (0–26s post-cue) demonstrating a main effect of both treatment and genotype (Main effects: Treatment p=0.0395, genotype p=0.0066, interaction p=0.6757, (sgROSA N=7, sgCNR1 N=7). (aa) Comparison of total lick number demonstrating a trend toward a main effect of treatment (Main effects: Treatment p=0.0109, genotype p=0.4333, interaction p=0.4333, (sgROSA N=7, sgCNR1 N=7). All error bars represent min/max values; box plots represent interquartile range with line at median; N represent number of mice. p values reported from unpaired two-tailed t-test (e-i, s,t) or Two-way ANOVA (j, y-aa). *p<0.05, **p<0.01, ***p<0.001.

Extended Data Fig. 8. AAV1-FLEX-sgCNR1 deletion of CB₁R from aPVT NTS neurons does not affect fear conditioning or extinction and foot shock induced eCB release and binding to aPVT terminals in the NAc.

(a) Schematic for CB₁R deletion from the BLA and photometry recordings of GCaMP6s BLA^VGlut -NAc terminals during Pavlovian fear conditioning and extinction. (b,c) Photometry traces of GCaMP6s signal aligned to foot shock from sgROSA (N=7) and sgCNR1 (N=7) injected mice during Pavlovian fear conditioning. (d,e) Comparison of freezing behavior (p=0.1880) and GCaMP6s signal (0–5s, p=0.4953) between sgROSA (N=7) and sgCNR1 (N=7) injected animals during fear conditioning. (f,g) Photometry traces of GCaMP6s signal aligned to cue presentation from sgROSA (N=7) and sgCNR1 (N=7) injected mice during fear extinction. (h,i) Comparison of freezing behavior (p=0.9355) and GCaMP6s signal (0–20s, p=0.5312) from sgROSA (N=7) and sgCNR1 (N=7) injected mice. (j) Schematic for photometry recordings of GRAB_eCB2.0 or GRAB_eCBMUT during fear conditioning and fear extinction. (k-m) Photometry traces of GRAB_eCB2.0 (n=8) or GRAB_eCBMUT (n=4) signal during fear conditioning. (n-p) Photometry trace and heatmap of GRAB_eCB2.0 (n=8) or GRAB_eCBMUT (n=4) signal during fear extinction. All error bars represent min/max values; box plots represent interquartile range with line at median; N represent number of mice. p values reported from unpaired two-tailed t-test (d,e,h,i).

Extended Data Fig. 9. Characterization of transsynaptic labeling, specificity of closed-loop optogenetic stimulation, and cluster classification.

(a) Schematic for validation of transsynaptic labeling approach. Transsynaptic AAV1-DIO-FLP was injected into the aPVT, AAV5-DIO-tdTomato was injected into the dorsolateral striatum (DLS), and a cocktail of AAV5-DIO-tdTomato and AAV5-fDIO-eYFP was injected in the NAc of Penk-Cre mice. The DLS, which expresses Penk but does not receive aPVT input, showed expression of AAV5-DIO-tdTomato but not AAV5-DIO-eYFP. The NAc, which expresses Penk and does receives aPVT input, showed expression of both AAV5-DIO-tdTomato and AAV5-DIO-eYFP. (b) Fiber placements for transsynaptic optogenetics cohort in Figure 4 and representative image. (c-f) Closed-loop optogenetic stimulation during sipper port engagement does not affect % time disengaging, walking, rearing, or grooming. (g) Schematic for closed-loop optogenetic activation of aPVT-NAc^PENK neurons during walk events. (h,i) Effect of walk elicited closed-loop activation of aPVT-NAc^PENK neurons on % engagement time in ChRimson (N=7) and eYFP (N=5) expressing animals. (j,k) Effect of walk elicited closed-loop activation of aPVT-NAc^PENK neurons on % walking time in ChRimson (N=7) and eYFP (N=5) expressing animals. (l) Lens placements for transsynaptic 1-photon imaging cohort in figure 5. (m) Dendrogram plot of hierarchical clustering of 206 tracked aPVT-NAc^Penk neurons based on Principal Component Analysis (PCA) of the activity of each neuron in the 60 second window following cue-onset. (n) Percent of variance explained by the top 20 PCs. (o) Support Vector Machine decoding of cluster identity, trained on PavD5 cluster activity.

Extended Data Fig. 10. Validation of generalized linear model and Hilbert and Rayleigh analysis of cluster entrainment to aPVT^NTS-NAc GCaMP6s and aPVT-NAc GRABeCB_2.0 signal.

(a-e) Cue-aligned traces of % time engaging with sipper port, disengaging from sipper port, walking, rearing, and grooming on PavD1 and PavD5 (N=10) from GRIN lens implanted animals. (f-g) Heatmap of % of neurons within each cluster encoding each behavioral state on PavD5, with a b coefficient threshold of +/- 0.5. (h,i) Heatmap of % of neurons within each cluster encoding each behavioral state on PavD1, with a b coefficient threshold of +/- 0.5. (j,k) Effect of individually dropping each predictor on model accuracy for PavD5 and PavD1. (i-p) Contribution of effect sizes for each neuron in each cluster to the generalized linear model (T values). (q-u) Overlay of Pavlovian reward conditioning day 5 aPVT-NAc^Penk cluster GCaMP6s traces with aPVT^NTS-NAc terminal GCaMP6s traces and Rayleigh plots from cluster 1,2 ,3, 4, and 5. (v-z) Overlay of Pavlovian reward conditioning day 5 aPVT-NAc^Penk cluster GCaMP6s traces with aPVT-NAc terminal GRABeCB_2.0 traces and Rayleigh plots from cluster 1, 2, 3, 4, and 5.

Supplementary Material

Video 4

Supplementary Video 4. Video depicting engagement behaviors and Δf/f of GCaMP6s signal in aPVT^NTS-NAc terminals during sipper retraction on day 5 of Pavlovian reward conditioning.

Video 2

Supplementary Video 2. Video depicting engagement behaviors and Δf/f of GCaMP6s signal in aPVT^NTS-NAc terminals during sipper retraction on day 1 of Pavlovian reward conditioning.

Video 3

Supplementary Video 3. Video depicting engagement behaviors and Δf/f of GCaMP6s signal in aPVT^NTS-NAc terminals during sipper extension on day 5 of Pavlovian reward conditioning.

Video 1

Supplementary Video 1. Video depicting engagement behaviors and Δf/f of GCaMP6s signal in aPVT^NTS-NAc terminals during sipper extension on day 1 of Pavlovian reward conditioning.

DATA AND CODE AVAILABILITY

All data and code associated with this project has been uploaded to https://figshare.com/authors/David_Marcus/23255861