A synaptic mechanism for encoding the learned value of action-derived safety.

A synaptic mechanism for encoding the learned value of action-derived safety - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Nat Commun . 2026 Jun 4;17:4916. doi: 10.1038/s41467-026-73906-3 A synaptic mechanism for encoding the learned value of action-derived safety Emma E Macdonald 1, ✉, # , Jun Ma 1, 5, # , Di Liu 1, # , Kai Yu 1, # , Rachel A Walker 1 , Michael E Authement 2 , Yan Leng 1 , Hannah C Goldbach 2 , Guochuan Li 3 , Yulong Li 3 , Veronica A Alvarez 2 , Bruno B Averbeck 4, ✉ , Mario A Penzo 1, ✉ Author information Article notes Copyright and License information 1 Section on the Neural Circuits of Emotion and Motivation, National Institute of Mental Health, Bethesda, MD USA 2 Section on Neurobiology of Compulsive Behaviors, National Institute of Mental Health, Bethesda, MD USA 3 State Key Laboratory of Membrane Biology, School of Life Sciences, Peking University, Beijing, China 4 Laboratory of Neuropsychology, National Institute of Mental Health, Bethesda, MD USA 5 Present Address: Jiangsu Province Key Laboratory of Anesthesiology, Xuzhou Medical University, Xuzhou, China ✉ Corresponding author. # Contributed equally. Received 2025 Sep 2; Accepted 2026 May 25; Collection date 2026. © This is a U.S. Government work and not under copyright protection in the US; foreign copyright protection may apply 2026 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . PMC Copyright notice PMCID: PMC13237380  PMID: 42243099 Previous version available: This article is based on a previously available preprint posted on bioRxiv on August 4, 2025: " A synaptic mechanism for encoding the learned value of action-derived safety ". Abstract Adaptive behavior requires that behaviorally relevant signals gain access to neural circuits guiding action. The thalamus has long been proposed to regulate information flow to cortical and subcortical systems, yet whether it also tracks internally generated goal signals remains unclear. Here, we show that neurons in the paraventricular thalamus (PVT) projecting to the nucleus accumbens (NAc) encode the motivational value of safety during active avoidance. As mice learn to avoid threat, PVT→NAc neurons develop a signal that emerges selectively at successful avoidance, is experience-dependent, and diminishes following outcome devaluation. Selective silencing of the PVT→NAc pathway at safety onset reduces the motivational value assigned to safety without impairing action–outcome learning. Mechanistically, PVT input engages cholinergic interneurons (CIN) in the NAc to regulate dopamine release via synaptic potentiation mediated by GluA2-lacking AMPA receptors at PVT–CIN synapses. Disrupting this plasticity reduces the motivational impact of safety. These findings identify a thalamostriatal mechanism through which learned goals gain stable access to motivational circuitry. Subject terms: Motivation, Cellular neuroscience, Neural circuits, Molecular neuroscience In this study, the authors present that paraventricular thalamus projections to the nucleus accumbens encode the motivational value of safety in mice. This signal drives avoidance via plasticity at cholinergic interneurons, revealing a thalamostriatal mechanism for outcome-driven motivation. Introduction Adaptive behavior depends on the ability to represent and update goals and values that guide future actions 1 – 3 . A central challenge in neuroscience has been to understand how the brain tracks what is currently important for behavior, particularly when goals must be inferred from experience, internal state, or recent outcomes rather than specified by immediate sensory cues. While much of this work has focused on cortical and striatal mechanisms underlying valuation and decision-making 1 , 3 – 5 , influential theoretical frameworks have long suggested that the thalamus may play a broader role in shaping how goal-relevant information is represented and utilized across distributed brain networks 6 – 8 . How this thalamic contribution extends beyond sensory processing to the tracking of motivationally relevant goals, however, remains poorly understood. Safety is an internal state derived from the successful prevention of harm, and a prominent example of an internally valued goal that must be derived from context, past actions, and learned cues 9 – 11 . Importantly, safety can be generated instrumentally: in active avoidance, the absence of an aversive outcome becomes meaningful only because it is contingent on the animal’s action 12 – 14 . Instrumentally generated safety differs fundamentally from passive safety cues that merely signal threat omission; instead, it promotes behavioral persistence, reinforces future avoidance, and engages dopaminergic systems in a manner analogous to reward 10 , 15 , 16 . Despite its central role in adaptive behavior, how action-derived safety is tracked and assigned motivational value within neural circuits remains incompletely understood. Midline thalamic nuclei are anatomically positioned to contribute to this problem by linking internal state signals with limbic and striatal circuits that regulate motivation and action 17 – 20 . Among these, the paraventricular thalamus (PVT) has emerged as a prominent node, receiving convergent input from brainstem and hypothalamic regions conveying visceral and homeostatic information, as well as from cortical areas involved in contextual and cognitive processing 18 , 21 . This convergence positions the PVT to link internal state signals with information about recent actions and outcomes—a key requirement for tracking the motivational significance of learned goals 22 . Consistent with this organization, prior work has implicated PVT circuits in drug seeking and relapse, modulation of reward-related behavior, and defensive control 21 , 23 – 27 . However, whether PVT projections participate directly in tracking goal signals that arise through learning, such as action-derived safety, remains unclear. Here, we leveraged an active avoidance paradigm to examine how the PVT contributes to the motivational representation of safety. Focusing on projections from the PVT to the nucleus accumbens (NAc), a key node in motivational control 28 – 30 , we combined pathway-specific optogenetic perturbations, fiber photometry, synaptic physiology, and computational modeling. We show that PVT→NAc projections develop an experience-dependent signal linked to successful avoidance, which is necessary for maintaining the motivational impact of safety but not for acquiring instrumental actions. Mechanistically, this process depends on plastic interactions between PVT inputs and striatal cholinergic interneurons that regulate dopaminergic tone. Together, these findings identify a thalamostriatal mechanism by which the midline thalamus tracks action-derived safety as a motivationally relevant goal, providing insight into how internal goals acquire value and guide adaptive behavior. Results PVT → NAc projections encode the learned value of action-derived safety To determine how safety outcomes acquire motivational significance during avoidance learning, we trained mice in a two-way active avoidance (2AA) task in which they learned to shuttle across a hurdle during a warning signal (WS) to terminate the WS and avoid a predicted footshock (see Methods). Classical and contemporary accounts of avoidance learning have emphasized that avoidance actions generate response-contingent stimuli, such as WS offset, that are negatively correlated with shock and can serve as reinforcing feedback for avoidance behavior 31 . While such response-generated sensory events are an important feature of avoidance tasks, behavioral performance in 2AA depends strongly on action–outcome reliability and predictability even when these sensory consequences are held constant (Supplementary Fig. 1 ). This indicates that behavioral performance reflects the learned motivational significance of successful outcomes rather than the presence of any single sensory event per se. To examine how this motivational signal emerges at the circuit level, we recorded calcium dynamics from GCaMP8s-expressing terminals of PVT neurons in the NAc during training (days 1, 3, and 5; Fig. 1A, B ; Supplementary Fig. 2A ). Population heatmaps aligned to WS onset and sorted by escape latency revealed activity patterns that spanned the cue period and increased with learning (Fig. 1C, D ). However, inspection of individual trials suggested more discrete increases in PVT→NAc activity time-locked to specific behavioral events, including shuttle initiation and the moment associated with successful avoidance (Fig. 1E ), as previously reported 27 . Aligning activity to these events confirmed distinct peaks at both time points (Fig. 1F,G ). We therefore focused subsequent analyses on the avoidance-related signal emerging at successful outcomes, consistent with the engagement of PVT→NAc projections during the motivational evaluation of having avoided harm. Fig. 1. PVT → NAc projections encode the value of inferred safety and sustain motivation during avoidance. Open in a new tab a Injection schematic (left). Representative histology (right). Scale bars, 200 µm. b Schematic of 2AA avoidance trial (left). Performance in the 2AA task ( n = 7 mice) (right). c GCaMP dF/F z-score normalized in all avoidance trials by day. d Heatmaps of calcium signals during avoidance trials, ordered by avoidance latency (low to high) for Days 1 (44 trials), 3 (176 trials), and 5 (180 trials). e Representative dF/F signal during avoidance trial. f , g GCaMP dF/F z-score normalized at shuttle initiation ( f ) (dotted line) or safety ( g ) in avoidance trials (left). Quantification of post-event activity (right) with individual trials on the left y-axis and mean±s.e.m. on the right y-axis. AUC, pairwise comparison across days, mixed effects model (see statistics in Supp. Data 1): Shuttle initiation: n = 288 trials from 7 mice; Safety: n = 401 trials from 7 mice. Asterisks indicate where multiple comparisons (Sidak corrected) found significance. h Injection schematic (left). Representative histology (right). Scale bars, 200 µm. i Schematic of closed-loop optogenetic manipulation (see Methods). j Performance in the 2AA task (Control n = 8, Halo n = 7 mice). 2-way Anova group effect days 1-5: p = 0.0345, Mann-Whitney test day 6: p = 0.0135. k Simulated avoidance probability from RL model (p a =probability of avoidance, β=value of safety, α=learning rate). l Avoidance probability across trials with best-fit model probability by group (left). Best-fit learning (middle) and value parameter (right) by group with bootstrap-estimated s.e.m. Bootstrap test for group difference: Learning: p = 0.1032; Value: p = 0.0448 (Control n = 8, Halo n = 7 mice). m Schematic of head-fixed avoidance and injections (see Methods). n Performance in avoidance training with no stimulation (left) and last day of training (right, Unpaired t test: p = 0.2714, Control n = 10, Halo n = 13 mice). o Breakpoints for progressive ratio tests for Halo (left, Paired t test: p = 0.0079, n = 13) and mCherry (right, Paired t test: p = 0.4947, n = 10) groups. Data are shown as mean ± s.e.m. * p < 0.05, ** p < 0.01, ns: p > 0.05. Schematics created in BioRender. Macdonald, E. (2026) https://BioRender.com/4pm04zs . Source data are provided as a Source Data file. All statistical tests are two-sided. To test the functional relevance of this outcome-related signal, we used closed-loop optogenetic silencing of PVT→NAc terminals immediately following successful avoidance. Across training days, inhibition of this pathway impaired the development of stable avoidance performance relative to controls (Fig. 1H–J , Supplementary Fig. 2B, C ). Notably, this deficit persisted during a subsequent light-off probe session in which no optogenetic manipulation was delivered, indicating that the behavioral effect reflects a lasting alteration rather than a transient disruption of performance (Fig. 1J ). To probe the underlying computational mechanism, and to dissociate learning of action-outcome associations from the assignment of motivational value to successful avoidance, we applied a reinforcement learning model in which action learning and outcome value learning were estimated separately (Fig. 1K , see Methods). This analysis revealed that behavioral deficits following PVT→NAc inhibition were best explained by a selective reduction in the value of the safety outcome, rather than a change in the action learning rate (Fig. 1L ). A complementary Bayesian learning model similarly revealed subject-level decreases in estimated safety value (Supplementary Fig. 3 ). Conversely, optogenetic excitation of PVT→NAc terminals during the safety period increased modeled outcome value while suppressing action learning rate (Supplementary Fig. 4 ), suggesting that excessive activation produces broader effects than temporally precise inactivation. Nevertheless, fitting the model at the individual subjects revealed a consistent trend toward increased valuation (Supplementary Fig. 4F, G ), reinforcing the conclusion that this projection primarily encodes outcome value rather than directly instructing action selection. We next asked whether this safety-related signal was sensitive to devaluation of instrumental contingency, a key criterion for distinguishing value encoding from fixed stimulus–response representations. Trained animals underwent an extinction-with-response-prevention session in which a transparent barrier blocked shuttling, and the WS no longer predicted shock. When subsequently tested in extinction sessions (with the barrier removed and no shock delivered), mice exhibited an approximately 50% reduction in avoidance behavior. This behavioral change was accompanied by a significant decrease in the safety-evoked PVT→NAc signal (Supplementary Fig. 5A–D ). When the shock contingency was reinstated, both avoidance behavior and PVT→NAc safety-related activity recovered, often surpassing pre-devaluation levels (Supplementary Fig. 5E ). This recovery further supports the interpretation that extinction-related effects arose from the contingency-dependent value updating and not attributable to photobleaching or signal degradation. Finally, we tested whether this thalamic-derived value signal supports effortful avoidance behavior. Parametrizing effort in shuttle-based avoidance is challenging, as animals either cross or fail to cross without a clear gradient of work required. To overcome this limitation, we employed a head-fixed version of the task in which mice learned to run a fixed distance to avoid an air puff (Fig. 1M ). In this setting, within-subject silencing of PVT→NAc terminals reduced the break point in a progressive ratio test (see Methods), indicating a diminished willingness to work for safety (Fig. 1N, O ). Together, these results demonstrate that PVT→NAc projections encode a safety-related value signal that emerges with learning and sustains motivated behavior by supporting the motivational value of safety under increasing effort demands. PVT → NAc projections are required for safety-evoked dopamine release Previous work has shown that dopamine (DA) levels in the NAc increase in response to safety-associated events during avoidance behavior, consistent with the idea that threat omission can acquire reward-like motivational value 15 , 32 – 35 . We first asked whether a similar dopaminergic signal accompanies successful avoidance in our 2AA task, and whether its emergence depends on input from the PVT. To address this, we expressed the genetically encoded DA sensor dLight1.2 36 in the NAc and recorded DA dynamics during training. As animals acquired stable avoidance performance, we observed a progressive increase in DA release time-locked to the onset of the safety period following successful avoidance (Fig. 2A–F ; Supplementary Fig. 6 ). Although the across-day effect in the dLight cohort did not reach conventional statistical significance in an omnibus mixed-effects test, the effect size and direction were consistent with a training-dependent strengthening of the safety-associated DA response. Importantly, this signal emerged gradually with learning, arguing against a fixed sensory response and instead suggesting the acquisition of motivational value. Fig. 2. PVT input drives NAc dopamine signals associated with safety. Open in a new tab (a) Schematic of behavior apparatus and example avoidance trial depicting behavior sequence at WS onset. (b) Wild type mice were injected with dLight1.2 to record dopamine in NAc (left). Representative image of recording site (right). Scale bar, 200 μm. (c) Performance in the 2AA task (n = 6 mice). (d) dLight signal in all avoidance trials by day. (e-f) dLight dF/F z-score normalized at shuttle initiation (dotted line) ( e ) or safety ( f ) in avoidance trials (left). Quantification of post event activity (right) with individual trials on the left y-axis and mean ± s.e.m. on the right y-axis. AUC, pairwise comparison across days, mixed effects model (see statistics in Supp. Data 1): Shuttle initiation: n = 204 trials from 6 mice; Safety: n = 305 trials from 6 mice. (g) Schematic of closed-loop optogenetic manipulation in the 2AA task (see Methods). (h) Drd2-Cre mice were injected with halorhodopsin (Halo) or fluorescent control in PVT and dLight1.2 in NAc to test the effects of PVT-NAc inhibition of safety dopamine release (left). Representative images of injection and recording sites (right). Scale bars, 200 μm. (i) Performance in the 2AA task. 2-way Anova: p = 0.517 (Control n = 6, Halo n = 5 mice). (j) dLight dF/F z-score normalized at safety with optogenetic inhibition by group day 1 (left), day 3 (middle), and day 5 (right) of 2AA training. (k) Quantification of post event activity with individual trials on the left y-axis and mean ± s.e.m. on the right y-axis. AUC, pairwise comparisons between groups, linear mixed-effects model for repeated measures: (see statistics in Supp. Data 1): Control n = 303 trials from 6 mice; Halo n = 277 trials from 5 mice. Asterisks indicate where multiple comparisons (Sidak corrected) found significance. Data are shown as mean ± s.e.m. * p < 0.05, ** p < 0.01, ns: p > 0.05. Schematics created in BioRender. Macdonald, E. (2026) https://BioRender.com/4pm04zs . Source data are provided as a Source Data file. All statistical tests are two-sided. We next tested whether this safety-related dopaminergic signal depends on thalamic input. During dLight recordings, we performed unilateral closed-loop optogenetic silencing of PVT→NAc terminals selectively at the time of successful avoidance (Fig. 2G ). This manipulation did not alter avoidance performance (Fig. 2I ), indicating that changes in DA signaling were not secondary to behavioral disruption. Nevertheless, silencing PVT→NAc projections significantly attenuated the avoidance-related DA response in the NAc (Fig. 2J,K ). Together, these results indicate that the emergence of a dopaminergic signal linked to successful avoidance depends on PVT input. Rather than reflecting a purely sensory correlate of task structure, safety-related DA release appears to be gated by thalamostriatal signaling, consistent with a role for the PVT in coupling learned, action-dependent outcomes to motivational circuitry in the NAc. Cholinergic interneurons mediate PVT-driven dopamine signaling Although PVT projections are required for safety-related DA release in the NAc, prior anatomical studies suggest that midline thalamic inputs, including those from PVT, do not directly innervate dopaminergic terminals in the striatum 37 . Instead, these projections densely innervate striatal cholinergic interneurons (CINs), which are well positioned to regulate DA release through nicotinic receptor-dependent mechanisms 38 – 45 . Consistent with this organization, recent circuit-mapping work identified that the PVT is a major source of excitatory input to NAc CINs 39 , suggesting a plausible substrate for PVT-dependent modulation of dopaminergic signaling. To test whether PVT input can evoke DA release through cholinergic mechanisms, we performed fast-scan cyclic voltammetry (FSCV) in acute NAc slices while optogenetically stimulating PVT terminals. PVT stimulation reliably evoked DA release, which was abolished by bath application of the nicotinic antagonist DHβE (Fig. 3A-C ). These results indicate that PVT-evoked DA release requires intact cholinergic transmission. Fig. 3. NAc acetylcholine is necessary for PVT-driven dopamine signaling after safety learning. Open in a new tab (a) Wildtype mice injected with channelrhodopsin (ChR2) in PVT and fast-scan cyclic voltammetry (FSCV) performed in NAc slices (right). Representative images of injection and recording sites with ChR-eYFP expression (right). (b) Representative PVT-evoked [DA] in ACSF and DHBE. Scale, 100 nM, 2 s. (c) oDA amplitude normalized to baseline in ACSF and DHBE (left). Average oDA amplitude of three trials by slice (right). Paired t test: p < 0.0001 (n = 5 slices from 5 mice). (d) Schematic of 2AA avoidance trial. (e) Wildtype mice were injected with GRAB-ACh3.8 (gACh) in NAc (left). Representative image of recording site (left). Scale bar, 200 µm. (f) Performance in the 2AA task (n = 5 mice). (g) Heatmaps of gACh signals during avoidance trials, ordered by avoidance latency (low to high) for Days 1 (70 trials), 3 (143 trials), and 5 (145 trials). (h) gACh signal in all avoidance trials by day. (i-j) gACh dF/F z-score normalized at shuttle initiation (dotted line) ( i ) or safety ( j ) in avoidance trials (left). Quantification of post-event activity (right) with individual trials on the left y-axis and mean±s.e.m. on the right y-axis. AUC, pairwise comparison across days, mixed effects model (see statistics in Supp. Data 1): Shuttle initiation: n = 220 trials from 5 mice; Safety: n = 348 trials from 5 mice. Asterisks indicate where multiple comparisons (Sidak corrected) found significance. (k) ChAT-Cre mice injected with ChrimsonR in PVT and dLight1.2 and DIO-hM4Di in NAc (left). Representative images of injection and recording sites (right). Scale bar, 200 µm. (l) Schematic of recording and training schedule (see Methods). (m-n) dLight during PVT stimulation pre-training ( m ) and post-training ( n ). (o) Quantification of dLight activity with individual trials on the left y-axis and mean±s.e.m. on the right y-axis. AUC, mixed effects model (see statistics in Supp. Data 1): n = 160 trials from 4 mice. Asterisks indicate where multiple comparisons (Sidak corrected) found significance. Data are shown as mean±s.e.m. *p < 0.05, **p < 0.01, ****p < 0.0001, ns: p > 0.05, #: p < 0.08. Schematics created in BioRender. Macdonald, E. (2026) https://BioRender.com/4pm04zs . Source data are provided as a Source Data file. All statistical tests are two-sided. We next asked whether CINs are themselves engaged during the acquisition of safety value in vivo. Using the genetically encoded ACh sensor gACh3.8 46 , we recorded cholinergic activity in the NAc across training in the 2AA task. Successful avoidance was associated with a phasic ACh response that increased in magnitude with training (Fig. 3D-J ; Supplementary Fig. 6 ), closely paralleling the emergence of safety-related signals observed in PVT terminals and DA release. These data suggest that CIN recruitment during safety emerges with learning and may participate in value assignment rather than reflecting a fixed sensory response. To directly test whether CIN activity is required for PVT-evoked DA release in vivo, we expressed the DA sensor dLight1.2 in the NAc of ChAT-Cre mice, optogenetically stimulated PVT terminals, and simultaneously silenced CINs using the D esigner R eceptors E xclusively A ctivated by D esigner D rugs (DREADDs) hM4Di. PVT-evoked DA was markedly potentiated by avoidance training, indicating that learning enhances the efficacy of this circuit. Strikingly, chemogenetic inhibition of CINs completely abolished PVT-evoked DA release following training (Fig. 3K-O ), providing convergent evidence that CINs are necessary for translating PVT input into dopaminergic output. Finally, to examine the temporal relationship between cholinergic and dopaminergic signaling during safety, we performed simultaneous recordings using spectrally distinct fluorescent sensors. Because prior experiments relied on the green DA sensor dLight1.2, we first validated the specificity and temporal fidelity of a red-shifted GRAB-DA sensor (rDA1m) 47 to enable dual-color recordings. We expressed rDA1m in the NAc and simultaneously recorded calcium activity from GCaMP8s-expressing dopaminergic terminals originating in the ventral tegmental area (VTA) (Supplementary Fig. 7A-J ). Across animals and behavioral epochs, rDA1m signals closely tracked VTA dopaminergic terminal activity, exhibiting strong trial-by-trial correspondence and significant Pearson correlations, consistent with rDA1m reliably reporting local DA dynamics in the NAc. Having established the validity of rDA1m, we next co-expressed rDA1m and the acetylcholine sensor gACh3.8 in the NAc to simultaneously monitor dopaminergic and cholinergic signaling during avoidance behavior. Both sensors recapitulated the safety-related transient observed in single-sensor experiments (Supplementary Fig. 7K-U ). In this independent cohort, safety-associated DA responses increased significantly across training days, consistent with the progressive modulation observed in the original dLight recordings. Notably, within-trial analyses revealed a consistent temporal offset during the safety period, such that cholinergic transients preceded dopaminergic peaks. This temporal relationship became more pronounced with training, paralleling the emergence of safety-related value signals in PVT inputs and DA release (Supplementary Fig. 7K–U ). Together, these results indicate that safety learning enhances a sequential signaling cascade in which PVT-driven engagement of striatal cholinergic interneurons precedes and facilitates dopaminergic release in the NAc. Safety value learning induces synaptic plasticity at PVT → CIN synapses The potentiation of PVT-evoked, CIN-dependent DA release in the NAc suggests that avoidance learning engages experience-dependent plasticity within the PVT → CIN microcircuit. We therefore asked whether the enhanced functional influence of PVT inputs following training reflects synaptic modifications at PVT → CIN connections. To test this, we injected ChAT-Cre mice with a Cre-dependent mCherry reporter in the NAc to label CINs and expressed ChR2-eYFP in PVT neurons to enable selective optical activation of PVT→NAc terminals (Fig. 4A ). Mice were then trained in the 2AA task or assigned to a yoked control group that received identical cue and shock exposure but lacked control over shock omission (Fig. 4B ). As expected, only trained animals developed robust avoidance behavior by day 5, whereas yoked controls failed to acquire shuttle responses to the WS (Fig. 4C ). Two hours after the final training session (Day 5), acute coronal brain slices containing the NAc were prepared, and optogenetically evoked excitatory postsynaptic currents (oEPSCs) were recorded from fluorescently identified CINs. Compared to yoked controls, CINs from trained animals exhibited significantly larger AMPA receptor-mediated oEPSCs (Fig. 4D,E ) and an elevated AMPA/NMDA ratio (Fig. 4I ), consistent with synaptic potentiation. In contrast, NMDA receptor-mediated current amplitudes were unchanged between groups (Fig. 4H ), indicating that plasticity was expressed postsynaptically, as a selective increase in AMPA receptor function. Fig. 4. Active avoidance training induces plasticity at PVT → CIN synapses. Open in a new tab (a) ChAT-Cre mice were injected with channelrhodopsin in PVT and DIO-mCherry in NAc to record post-synaptic currents in CINs with PVT stimulation (left). Representative images of injection and recording sites (right). Scale bars, 200 µm. (b) Schematic of training and recording schedule (see Methods). (c) Performance in the 2AA task (trained n = 8, yoked n = 7 mice). ( d ) Representative traces of optogenetic evoked voltage clamp responses. (e-j) Measurements of current properties. Unpaired t tests: oEPSC amplitude p = 0.0024; Rise time p = 0.0107; Decay time p = 0.0028; NMDA amplitude p = 0.1562; Paired-pulse ratio p = 0.0005. Mann Whitney test: AMPA/NMDA ratio p = 0.0168 (trained n = 22 neurons from 4 mice, yoked n = 19 neurons from 3 mice). (k) Rectification curve (left). 2-way Anova: p = 0.0085; Bonferroni’s multiple comparisons at 70 mV: p < 0.0001. Rectification index (right). Unpaired t test: p = 0.0360. (l) Amplitude of evoked response with NASPM application (left). Quantification of NASPM inhibition (right). Unpaired t test: p = 0.002 (trained n = 6 neurons from 3 mice, yoked n = 6 neurons from 2 mice). Data are shown as mean ± s.e.m. *p < 0.05, **p < 0.01, ***p < 0.001, ns: p > 0.05. Schematics created in BioRender. Macdonald, E. (2026) https://BioRender.com/4pm04zs . Source data are provided as a Source Data file. All statistical tests are two-sided. We next examined the locus and mechanism of this synaptic enhancement. CINs from trained animals showed significantly faster AMPA current rise and decay kinetics relative to controls (Fig. 4F,G ), consistent with the recruitment of GluA2-lacking, calcium-permeable AMPA receptors 48 – 50 . In line with this interpretation, rectification indices were increased in CINs from trained mice, and AMPA currents were more strongly attenuated by NASPM, a selective blocker of GluA2-lacking AMPARs (Fig. 4K,L ). In addition, a modest but significant reduction in paired-pulse ratio was observed following training (Fig. 4J ), suggesting that presynaptic changes in release probability may also contribute to the overall potentiation. Together, these results demonstrate that active avoidance training induces synaptic strengthening at PVT → CIN synapses in the NAc, characterized by enhanced AMPA receptor transmission and recruitment of GluA2-lacking AMPARs. This synaptic plasticity provides a mechanistic substrate through which learning enhances the ability of PVT inputs to recruit CINs and, in turn, facilitate safety-related dopamine signaling. GluA1-dependent plasticity at PVT → CIN synapses supports safety value encoding The preceding experiments showed that avoidance training potentiates PVT → CIN synapses through the recruitment of GluA2-lacking AMPA receptors, a form of plasticity classically associated with the insertion of GluA1-containing AMPA receptor 51 , 52 . We therefore asked whether GluA1 expression in CINs is required for this experience-dependent synaptic remodeling. To test this, we used a CRISPR-Cas9 strategy to selectively delete Gria1 , the gene encoding GluA1, from cholinergic interneurons. ChAT-Cre mice were injected unilaterally with a Cre-dependent CRISPR construct (AAV9-EF1α-DIO-Cas9-sgGria1) that co-expresses Cas9 and a guide RNA targeting Gria1 53 , while the contralateral hemisphere received a control virus lacking the sgRNA (Cas9-empty) (Fig. 5A ). Mice were then trained in the 2AA task, and acute brain slices were prepared two hours after the final training session for whole-cell recordings from CINs in each hemisphere. In the control (Cas9-empty) hemisphere, CINs displayed properties consistent with synaptic potentiation as previously observed: elevated AMPA oEPSCs, increased AMPA/NMDA ratios, greater rectification, and enhanced sensitivity to NASPM (Fig. 5D-K ). In contrast, CINs in the sgGria1-injected hemisphere exhibited significantly reduced AMPA-mediated currents, lower AMPA/NMDA ratios, and decreased rectification indices (Fig. 5D-K ). NASPM sensitivity was also attenuated, consistent with a failure to incorporate GluA2-lacking AMPA receptors in the absence of GluA1 (Fig. 5L,M ). Importantly, these differences were not observed in untrained home-cage controls (Supplementary Fig. 8 ), indicating that GluA1 deletion does not alter baseline synaptic strength but specifically prevents training-induced plasticity. Together, these results demonstrate that GluA1 expression in CINs is required for the synaptic potentiation at PVT → CIN synapses induced by safety learning. Fig. 5. NAc CIN Gria1 knockout disrupts avoidance-related plasticity and safety valuation. Open in a new tab (a) ChAT-Cre mice were injected with channelrhodopsin in pPVT, and FLEX-Cas9-sgGria1 and empty control contralaterally in NAc to record post-synaptic currents in CINs with pPVT stimulation (left). (b) Performance in the 2AA task (n = 5 mice). (c) Representative image of recording location. Scale bar, 200 µm. (d) Representative oEPSCs from control and knockout hemispheres. (e-h) Measurements of current properties. Unpaired t tests: oEPSC amplitude p = 0.0003; Rise time p = 0.0206; Decay time p = 0.0321; Paired-pulse ratio p = 0.0009 (sgGria1 n = 12 neurons, empty n = 12 neurons from 4 mice). (i) Representative traces of AMPA and NMDA responses to pPVT optogenetic stimulation. (j-k) Current amplitude measurements. Unpaired t tests: NMDA amplitude p = 0.4374; AMPA/NMDA ratio p = 0.0256 (sgGria1 n = 12 neurons, empty n = 12 neurons from 4 mice). (l) Representative traces of voltage clamp responses to optogenetic stimulation before and after NASPM application. (m) Amplitude of evoked response with NASPM application (left). Quantification of NASPM inhibition (right). Unpaired t test: p = 0.0135 (sgGria1 n = 5 neurons, empty n = 6 neurons from 5 mice). (n) ChAT-Cre mice were injected with FLEX-Cas9-sgGria or control bilaterally in NAc (left). Representative image of virus spread through fluorescent in situ hybridization Cas9 probe binding to viral DNA (magenta) and ChAT mRNA expression (green) (right). (o) Performance in the 2AA task by group. Mixed-effects group effect: p = 0.0154 (sgGria1 n = 11, empty n = 12 mice). (p) Avoidance probability across trials with best fit model probability by group. (q) Best-fit parameters by group with bootstrap estimated error. Bootstrap permutation tests for group difference: Learning p = 0.680; Value p = 0.024 (sgGria1 n = 11, empty n = 12 mice). Data are shown as mean ± s.e.m. *p < 0.05, **p < 0.01, ns: p > 0.05. Schematics created in BioRender. Macdonald, E. (2026) https://BioRender.com/4pm04zs . Source data are provided as a Source Data file. All statistical tests are two-sided. We next asked how this plasticity contributes to behavior. In principle, synaptic strengthening at PVT → CIN synapses could support either the acquisition of avoidance actions or the assignment of motivational value to successful avoidance outcomes. To dissociate these possibilities, we bilaterally injected ChAT-Cre mice with the Gria1 -targeting CRISPR construct and trained them in the 2AA task. Compared to control animals, deletion of GluA1 in CINs significantly impaired avoidance performance (Fig. 5N,O ; Supplementary Fig. 9 ). To determine whether this deficit reflected disrupted action learning or altered outcome valuation, we applied RL models to trial-by-trial behavior. Strikingly, GluA1 deletion selectively reduced the estimated value parameter associated with successful avoidance, while leaving the learning rate of avoidance actions intact (Fig. 5P,Q ). This pattern closely mirrored the effects of PVT→NAc silencing, directly implicating CIN plasticity as a critical node for encoding safety-related value. Individual-level fits using Bayesian sampling further attributed the group difference to altered initial values rather than differences in locomotion or general performance, as confirmed by open-field testing (Supplementary Fig. 9 ). Together, these findings identify GluA1-dependent synaptic plasticity at PVT → CIN synapses as a key mechanism through which avoidance enhances the motivational value of safety. By strengthening the ability of thalamic input to recruit cholinergic modulation and downstream dopaminergic signaling, this local plasticity enables internal goal representations, here the value of successful avoidance, to gain stable influence over striatal motivational circuitry and sustain safety-seeking behavior. Discussion Many adaptive behaviors depend on the ability to maintain and prioritize internal goals whose motivational significance extends beyond the immediate presence of sensory cues 54 . These goals may arise from internal state, prior experience, or the outcomes of action, and must be stabilized over time to sustain motivated behavior. Here, we identify a thalamostriatal circuit mechanism through which the PVT supports this process. Using safety-seeking behavior as a tractable example, our findings show how PVT engagement of striatal cholinergic and dopaminergic signaling stabilizes the motivational value of an internal goal and promotes behavioral persistence. This extends classical models of valuation, which emphasize external cues and innate reinforcers 28 , 55 by demonstrating how thalamic circuits enable internally generated goals to exert sustained control over action. Our findings support a conceptualization of the PVT as part of a central system for stabilizing and prioritizing internal goal representations and broadcasting value-related signals to guide behavior. Depending on downstream target, this broadcast can serve distinct computational functions. For instance, PVT projections to the amygdala assign motivational significance to sensory events, shaping cue–outcome associations 56 . Within the NAc, PVT outputs can likewise engage distinct interneuron populations to modulate behavioral policy: PVT input to parvalbumin interneurons has been shown to impose a threat- and drug-sensitive brake on reward seeking 57 , 58 , whereas the PVT→NAc CIN pathway described here appears to support the maintenance and reinforcement of goal value across action and time. In this way, PVT signaling can either constrain or sustain behavior depending on internal state and circuit context. This dual capacity, to assign value to external cues and to elevate the value of internally generated outcomes, positions the PVT as a key node translating internal demands into motivational control. As such, our results support and extend the emerging view of the thalamus not only as an outward-facing “searchlight” prioritizing sensory information for perception 6 , 59 – 61 , but also as an inward-facing system that elevates internal goals for sustained motivational guidance 22 . PVT’s role in internal goal prioritization is dynamically shaped by physiological and emotional states. Prior studies have shown that PVT engagement enhances the attractiveness of food odors during hunger 62 and suppresses reward-seeking under threat 25 , 57 , 63 , highlighting its function as a flexible gate that adjusts motivational salience according to internal context 64 – 66 . Our findings build on this framework by identifying a synaptic mechanism through which PVT signaling stabilizes the motivational value of an internal, action-contingent outcome. In this view, safety is not treated as a special class of inferred outcome, but rather as a goal whose value must be actively maintained to sustain avoidance behavior over time and effort. At the circuit level, we uncover a disynaptic architecture in which thalamic input recruits dopaminergic signaling through modulation of striatal CINs. CINs are well known to regulate DA release 40 – 42 , 67 , 68 and contribute to motivated behavior 69 – 71 . Recent studies further show that CINs are required for effortful reward pursuit 72 and that glutamatergic plasticity onto CINs shapes reinforcement-driven decision-making 73 . Our findings extend this framework by showing that experience-dependent strengthening of PVT → CIN synapses enhances CIN recruitment and amplifies dopaminergic output specifically in response to internally generated goals. GluA1-dependent incorporation of calcium-permeable AMPARs at these synapses may thus act as a local gain mechanism, enabling internal goal representations to exert sustained influence over striatal motivational circuitry. Interestingly, PVT synapses onto parvalbumin interneurons and Drd2-expressing medium spiny neurons in the NAc have also been reported to undergo opioid-dependent plasticity at sites enriched for calcium-permeable AMPARs 25 , 57 , 58 . Together, these observations suggest that PVT→NAc pathways may share a common plasticity motif, through which state-dependent thalamic input drives long-term modifications in striatal circuit function and behavioral policy. Importantly, our findings should be interpreted within the broader and actively evolving literature on striatal DA–ACh interactions. While classical and contemporary studies have demonstrated that CIN activity can powerfully gate DA release through nicotinic mechanisms, particularly at moments of action and outcome 40 – 45 , 69 , other work has found little cholinergic regulation of DA signals in vivo, especially during cue-driven or low-effort behaviors 74 – 76 . Recent work by Touponse and colleagues 71 provides an important framework for reconciling these views, demonstrating that cholinergic modulation of DA is strongly context-dependent and preferentially engaged when behavioral demands require sustained motivation, internal state integration, or effortful control. In this framework, CIN–DA coupling is not a default property of striatal processing, but a conditional mechanism recruited when internal goals must be actively stabilized to guide behavior. Our data align closely with this emerging perspective. In the shuttle avoidance task, cholinergic and dopaminergic signals are initially weakly coupled but become increasingly coordinated as animals learn to sustain avoidance based on internally represented safety value. The temporal ordering we observe, where CIN activation reliably precedes DA release during the safety period, supports a model in which cholinergic signaling serves as an enabling or amplifying step for dopaminergic engagement, rather than a redundant or parallel signal. Thus, rather than contradicting studies reporting independent DA and ACh dynamics, our findings suggest that PVT-driven plasticity selectively recruits CIN-mediated dopaminergic modulation under conditions where internal goals, rather than external cues, dominate behavioral control. Our computational analyses further refine this interpretation. Reinforcement learning models identified a selective impact on outcome value: across optogenetic perturbations of PVT→NAc signaling, both our RW-based fits and Bayesian inference converged on reduced safety value (β) as the dominant explanation for impaired avoidance persistence, with comparatively preserved action learning parameters. In contrast, for the Gria1 deletion in CINs, Bayesian modeling yielded a more nuanced pattern: β showed a non-significant downward trend, whereas the most reliable group difference emerged as a reduction in V1 (initial value). Importantly, these outcomes are not mutually exclusive. Rather, they suggest that PVT → CIN plasticity can influence safety-seeking through at least two computational entry points: by modulating online valuation in the moment (as revealed by acute optogenetic perturbations) and by shaping the baseline motivational set-point or initial value assigned to avoidance (most evident in the Gria1 manipulation). This interpretation is consistent with the broader idea that thalamostriatal mechanisms help stabilize internal goal value while allowing downstream learning systems to update behavior trial-by-trial 2 . Finally, these circuit mechanisms may have broad relevance for psychiatric disease. Disruption of thalamostriatal plasticity could distort the prioritization of internal goals, either inflating their motivational weight, as in pathological avoidance, or attenuating it, as observed in the amotivation and avolition associated with schizophrenia 77 – 80 . Postmortem studies have revealed reduced expression of cholinergic interneuron markers in the striatum of individuals with schizophrenia, suggesting a link between CIN dysfunction and impaired motivational control 81 , and recent work in a genetic mouse models has established a mechanistic link between impaired thalamostriatal transmission, CIN dysfunction, and avolition-like phenotypes 78 . By linking synaptic plasticity to the stabilization of internal goal value, our findings offer a mechanistic framework for understanding how motivational priorities are constructed, and how they become dysregulated, in psychiatric disease. Importantly, while our experiments focused on avoidance-derived safety, the circuit logic revealed here is unlikely to be specific to defensive behavior. The PVT has been implicated across a broad range of motivated states, including reward seeking, drug relapse, feeding, and motivational vigor 21 , 25 , 57 , 58 , 82 . In this broader view, PVT→NAc signaling may serve as a general mechanism for stabilizing internally defined goals, whether aversive or appetitive, by assigning them sustained motivational weight and supporting their continued control over action selection and vigor. Rather than encoding the specific content of a goal, the PVT may contribute a higher-order function: preserving the relevance of goal representations once they have been established through experience, internal state, or prior outcomes. Through plastic interactions with striatal cholinergic and dopaminergic systems, this thalamostriatal pathway could therefore support persistence, effort, and internal prioritization across motivational domains. Together, our findings identify a thalamostriatal plasticity mechanism through which internal goals acquire durable motivational weight. By strengthening PVT → CIN transmission and recruiting cholinergic-dependent dopaminergic output in the NAc, this circuit provides a route for learned goals to exert stable motivational control when external incentives are weak or absent. Methods Subjects All procedures were performed in accordance with the Guide for the Care and Use of Laboratory Animals and were approved by the National Institute of Mental Health (NIMH) Animal Care and Use Committee. Mice used in this study were group housed under a 12-h light/dark cycle (6:00–18:00 light), at temperatures of 70–74 °F and 40–65% humidity, with food and water available ad libitum . After surgery, mice were singly housed. C57BL/6NJ strain mice (The Jackson Laboratory #005304) were crossed with Drd2-Cre (GENSAT founder line ER44), ChAT-Cre (The Jackson Laboratory # 018957), or DAT-Cre (#006660). Experiments were performed in adult male and female mice (8-20 weeks). Sex was considered in the study design; however, the study was not powered a priori to detect sex-specific effects. Therefore, data were pooled across sex for all analyses. Animals were randomly allocated to the different experimental conditions reported in this study. Viral vectors AAV9-hSyn-Flex-GCaMP8s-WPRE (#162377), pAAV.Syn.Flex.GCaMP6s.WPRE.SV40 (#100845), pAAV-hSyn-dLight1.2 (#111068), pAAV-hsyn-GRAB_rDA1m (#140556), pAAV-hSyn-DIO-mCherry (#50459-AAV2/9), pAAV-hSyn-hChR2(H134R)-EYFP (#26973-AAV9), pAAV-Syn-ChrimsonR-tdT (#59171-AAV9), pAAV-hSyn-DIO-hM4D(Gi)-mCherry (#44362-AAV2), and pAAV-EF1a-DIO-ChrimsonR-mRuby2-KV2.1-WPRE-SV40 (#124603) were purchased from Addgene. pAAV-FLEX-SaCas9-U6-sgGria1 (#124853), pAAV-FLEX-SaCas9-U6-sgRNA (empty) (#124844) were purchased as plasmids from Addgene and packaged into AAV2/9 at Vector Biolabs. AAV2-Ef1α-DIO-eNpHR3.0-mCherry (Deisseroth) and AAV2-syn-FLEX-ChrimsonR-tdTomato (Boyden) were produced by the Vector Core of the University of North Carolina. AAV9-hsyn-gACh4h(Ach3.8) was purchased from WZ Biosciences. All viral vectors were stored in aliquots at −80 °C until use. Stereotaxic surgery All stereotaxic surgeries were conducted as described in our animal study protocol using previously described procedures 24 and stereotaxic coordinates 27 . Mice were first anesthetized with a ketamine (100 mg/kg) plus xylazine (10 mg/kg) solution and placed in a stereotaxic device (AngleTwo, Leica Biosystems). The following stereotaxic coordinates were targeted for viral injections and/or optical fiber implantation: NAc, 1.65 mm from bregma, ±.9 mm lateral from the midline, and -4.6 mm vertical from the cortical surface, 0° angle for photometry, 8° angle for optogenetics. PVT, −1.60 mm from bregma, 0.06 mm lateral from midline and −3.30 mm vertical from cortical surface, 6.12° angle for both fiber photometry and optogenetics. VTA, -3.08 mm from bregma, ±.5 mm lateral from the midline, and -4.51 mm vertical from the cortical surface. For fiber photometry and optogenetic experiments, an optical fiber (400 μm for photometry, Doric Lenses/Inper; 200 μm for optogenetics, Thorlabs/Inper) was implanted over the target immediately after viral injections and cemented using Metabond Cement System (Parkell) and Jet Brand dental acrylic (Lang Dental Manufacturing). For head-fixed training, a head bar was attached with light-cure cement (3 M RelyX) on top of Metabond. After all surgical procedures, animals were returned to their home cages and placed on a heating pad for 3 hours for post-surgical recovery and monitored for 72 h with diet supplements. Animals received subcutaneous injections with Metacam (meloxicam, 5 mg/kg) for analgesia and anti-inflammatory purposes. Mice without correct targeting of optical fibers or vectors were excluded from this study. Behavior 2-way active avoidance Mice were trained on the 2AA task, replicating previous methodological procedures 27 , 83 and are detailed below. The behavioral apparatus consisted of a custom-built shuttle box (18 cm × 36 cm × 30 cm) that contained two identical chambers separated by a hurdle (17.5 cm × 6 cm). The hurdle projected 3 cm above the floor and allowed mice easy access to both chambers. The floor consisted of electrifiable metal rods (H10-11M-TC-SF, Coulbourn Instruments) and was connected to a shock generator (H13-15, Coulbourn Instruments). Before each subject was trained/tested, the shuttle box was wiped clean with 70% ethanol. The mouse’s behavior was captured with a USB camera during each session. A speaker located on the top of the shuttle box (50 cm high) was used to deliver the WS. Subjects’ movement and TTLs of WS, US and optogenetic stimulation were recorded and coordinated by ANY-maze versions 5-7 (Stoelting). After a 5-min habituation period, mice were trained with daily sessions of 2AA, each consisting of 30 presentations of the WS (4 kHz, 75 dB, lasting up to 15 s each). Trials in which subjects failed to shuttle to the adjacent chamber before the termination of the WS resulted in the presentation of the unconditioned stimulus (US, 0.6 mA foot shock lasting up to 7 s each) until subjects escaped to the opposite chamber (escape trials). No subject failed to escape the US. For trials in which subjects shuttled to the opposite chamber during the WS, the WS was abruptly terminated, and the US was also prevented (avoidance trials). The inter-trial interval (ITI) was random between 30 and 40 s for all experiments except Supplementary Fig. 4 , where the ITI was increased to 60 s. The avoidance rate was calculated as the percentage of the number of avoidance trials over the total number of trials. For optogenetic and fiber photometry experiments fiber patch cords were attached every session of training. For all photometry experiments, subjects that did not reach 30% avoidance rates by Day 3 were excluded from data analysis. All experiments were conducted in awake, freely moving mice in test chambers during the light cycle. Extinction with response prevention As shown in Supplementary Fig. 5 , after five days of 2AA training, subjects underwent one session of extinction with response prevention. During this session, animals were confined to a single chamber of the 2AA apparatus using a custom transparent plexiglass divider and presented with 15 WS (15 s each 35-45 s intertrial interval; ITI). On subsequent extinction days, the divider was removed, allowing animals to freely shuttle between chambers while receiving 30 WS presentations (15 s, 30–40 s ITI). No shocks were delivered during extinction, and shuttling did not terminate the WS. To verify that extinction-related changes in photometry signals reflected neural activity rather than photobleaching, a reinstatement session was conducted using the standard 2AA protocol. Probabilistic avoidance with surprise safety omissions As shown in Supplementary Fig. 1 , following five days of standard 2AA training, mice underwent a single test session designed to introduce rare violations of learned safety contingencies. Animals were counterbalanced to pre-defined probabilistic and deterministic shuttling directions. Each session began with a 20-trial baseline phase identical to training. After this baseline, animals entered a probabilistic phase in which successful shuttling continued to terminate the WS in both directions; however, in the preassigned probabilistic direction, a subset of successful avoidances was followed by delivery of a shock at WS offset (surprise safety omissions). In contrast, the preassigned deterministic direction remained fully reliable, such that successful avoidance was never followed by shock. Surprise safety omission trials were restricted to the probabilistic direction and were pseudorandomly distributed with constraints designed to prevent consecutive surprise trials (implemented as five surprise events separated by 5–10 intervening trials). This schedule resulted in an average of 62 ± 2 total trials per session, an average gap of eight trials between surprise shocks, and an effective shock probability of 22.4 ± 5% for successful avoidance in the probabilistic direction. All task parameters were otherwise identical to those used during standard 2AA training. Head fixed avoidance The behavioral apparatus comprised a 20 cm diameter foam cylinder connected to a rotary encoder, a head-fixation frame, and a holder for a compressed air nozzle positioned behind the animal, directed toward the base of the tail. The rotary encoder was linked to an interface board connected to the Bpod system (Sanworks, NY). Air puff delivery was controlled through a picospritzer triggered by Bpod TTL output, with pressure set to 50 psi. Mice were trained daily for 3 days, each session comprising 50 trials. On each trial, a WS (12 kHz, 75 dB, up to 10 s) was presented. If the subject failed to run at least 5 cm on the wheel before WS termination, the US (air puff, up to 5 s) was delivered. If the running requirement was met during the WS, the WS was immediately terminated and the US omitted (avoidance trials). The ITI was 30 s. The avoidance rate was calculated as the percentage of avoidance trials over the total number of trials. Animals typically reached achieved stable performance ( ≥ 60% avoidance) by training day 3 or 4. For the progressive ratio (PR) test, the maximum WS duration was increased to 20 s. The required distance for avoidance started at 25 cm and increased in 5 cm increments. Advancement to the next stage required two consecutive successful avoidance trials. The session ended after six consecutive US deliveries, and the breakpoint was defined as the longest distance at which avoidance was successfully achieved. Fiber photometry Fiber photometry was performed in accordance with previously described methodological procedures 84 and are detailed below. Mice were allowed to habituate to the fiber patch cord in their home cage for approximately 5 min before each behavior test. GCaMP fluorescence and isosbestic autofluorescence signals were excited by the fiber photometry system (Doric Lenses) using two sinusoidally modulated LEDs (473 nm at 211 Hz and 405 nm at 311 Hz) controlled by a standalone driver (DC4100, ThorLabs). Both LEDs were combined via a commercial Mini Cube fiber photometry apparatus (Doric Lenses) into a fiber patch cord (400-µm core, 0.48 NA) connected to the brain implant in each mouse. The light intensity at the interface between the fiber tip and the animal was adjusted from 10 µW to 20 µW (but was constant throughout each test session for each mouse). An RZ5P fiber photometry acquisition system with Synapse v98 software (Tucker-Davis Technologies) collected and saved real-time demodulated emission signals and behavior-relevant TTL inputs. For each trial, signals (F473 nm) were compared with autofluorescence signals (F405 nm) to control for movement and bleaching artifacts. Signal data were de-trended by first applying a least-squares linear fit to produce F fitted 405 nm , and dF/F was calculated as (F 473 nm – F fitted 405 nm )/F fitted 405 nm . All signal data are presented as the z-score of the dF/F from baseline (pre-WS/pre-Event) segments. In Supplementary Fig. 7 , a third channel was added (560 nm LED driven at 711 Hz). This signal was compared to autofluorescence signals (F405 nm) as described. To visualize event-aligned signals, trial-based events such as cue onset and safety were identified from TTL cue and shock inputs to the RZ5P amplifier from Any-Maze, and events such as shuttle initiation were aligned to video-tracking output using custom R code. Mean signal is the average across subject averages Z-score normalized to a 5 s pre-event baseline. Area under the curve was calculated with trapezoid approximation to baseline (pre) and event (post) time periods. For Fig. 3 and Supplementary Fig. 7 , a RZ10x fiber photometry acquisition system (Tucker-Davis Technologies) with integrated drivers, LEDs (405 nm, 465 nm, 560 nm) and photosensors was used to collect and demodulate emission signals with analogous protocols and analysis. All photometry recordings utilized a custom Doric mini cube for wavelength separation. Sensors including GCaMP, dLight1.2, and gACh3.8 were recorded using 473 nm and 465 nm drivers filtered through a 460-490 nm input channel and were paired with outputs from a 500–540 nm channel. A short-wavelength control channel used 405 nm drivers filtered through 400–410 nm input and demodulated with 420-450 nm output to sensors. 560 nm drivers for the rDA1m sensor utilized the 550–580 nm input channel, and signals were demodulated from outputs through the 600-680 nm channel. The 600-680 nm channel served as an input for optogenetic stimulation using red light. Fiber photometry with optogenetic and chemogenetic manipulations In Fig. 2 , mice with dLight in NAc unilaterally and halorhodopsin or mCherry in PVT underwent 2AA training. Fiber photometry recordings of dLight were taken during training on days 1, 3, and 5. In addition, a red light (Opto Engine 635 nm) was triggered at the completion of all avoidance trials concurrent with WS offset for a 5 s duration. In Fig. 3K–O , mice were injected with dLight in NAc, DREADDs (Gi) in NAc ChAT+ neurons, and ChrimsonR or mCherry in PVT. Fiber photometry recordings of dLight were taken in a novel context on separate days during optogenetic stimulation (60 s, 20 Hz, 10 trials) of PVT terminals. For within-mouse comparisons, each subject was given a saline injection I.P. 30 min prior to recording. The following day the same subjects were given a CNO injection (5 mg/kg) I.P. 30 minutes prior to recording. The mice were trained in 2AA for 5 days before repeating saline and CNO recording conditions. Optogenetics In Fig. 1 and Supplementary Fig. 4 , mice were subjected to six 2AA sessions (one session per day), on training days 1-5 optogenetic stimulation was triggered at the completion of all avoidance trials concurrent with WS offset for a 5 s duration. The light was powered on for 5 s for experiments with halorhodopsin stimulation, or at 20 Hz, 20% duty cycle for 5 s in ChrimsonR experiments, and light intensity at the interface between the fiber tip and the implant cannula was ~10 mW. Bilateral optogenetic manipulations used a yellow (Ce:YAG  +  LED Driver 565 nm, Doric) or green light (Opto Engine 561 nm), while optogenetics multiplexed with fiber photometry used a red light (Opto Engine 635 nm) as described above to limit off-target excitation. Electrophysiology Mice were trained in 2AA for 5 days, and two hours following the last session, the subjects were then sacrificed for slice recordings by an experimenter blind to conditions. For yoked controls, each was paired with a trained subject. The pairs were trained concurrently wherein the yoked subject received the same duration of both WS and shock. Acute brain slices were prepared according to previously described protocols 24 , 83 , 85 . Mice were anesthetized with isoflurane and transcardially perfused with an ice-cold NMDG cutting solution containing (in mM): 92 mM N-Methyl-D-glucamine, 2.5 mM KCl, 1.25 mM NaH2PO4, 10 mM MgSO4, 0.5 mM CaCl2, 30 mM NaHCO3, 20 mM glucose, 20 mM HEPES, 2 mM thiourea, 5 mM Na-ascorbate, 3 mM Na-pyruvate, at 7.3-7.4 pH, saturated by 95% O2 and 5% CO2. Coronal sections (300 µm thickness) containing the NAc were cut in the ice-cold NMDG cutting solution using a VT1200S automated vibrating-blade microtome (Leica Biosystems), and were subsequently transferred to a heated incubation chamber containing the NMDG cutting solution at 34-35 °C. After recovery for 12 min, slices were transferred to a room temperature (20-24 °C) holding chamber containing a HEPES-modified artificial cerebrospinal fluid (92 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 2 mM MgSO4, 2 mM CaCl2, 30 mM NaHCO3, 25 mM glucose, 20 mM HEPES, 2 mM thiourea, 5 mM Na-ascorbate, 3 mM Na-pyruvate, at 7.3 pH, saturated by 95% O2 and 5% CO2) and remained in the holding chamber until recordings. Whole-cell patch-clamp recordings were conducted under visual guidance using an Olympus BX51 microscope with transmitted light illumination, and ChAT+ neurons were identified based on their fluorescence (mCherry). Recordings from ChAT+ neurons in the NAc were obtained with Multiclamp 700B amplifiers (Molecular Devices), and data were filtered at 1-2 kHz using a 1440 A Digidata Digitizer and Clampex software (Molecular Devices). For recordings, slices were transferred to the recording chamber and constantly supplied with a room-temperature ACSF (118 mM NaCl, 2.5 mM KCl, 26.2 mM NaHCO3, 1 mM NaH2PO4, 20 mM glucose, 2 mM MgCl2, and 2 mM CaCl2, at pH 7.4, saturated by 95% O2 and 5% CO2, flow rate = 2.0-2.5 ml/min). Pharmacological agents were added to the ACSF and bath applied. All recordings were made with borosilicate glass pipettes with tip resistance of 3-6 MΩ. pipettes were filled with Cs-based internal solution containing (in mM): 117 mM Cs methanesulfonate, 10 mM HEPES, 2.5 mM MgCl2, 2 mM Na2-ATP, 0.4 mM Na2-GTP, 10 mM Na2-phosphocreatine, 0.6 mM EGTA, 5 mM QX-314 at pH 7.2 and 288-290 mOSM. Optogenetically evoked synaptic responses were achieved by illuminating the acute slices with blue LED (470 nm, CoolLED) drive ChR2-expressing terminals. Light stimulation was delivered every 15 seconds and synaptic responses were recorded at holding potentials of −60 mV for AMPA-receptor-mediated responses and +40 mV for NMDA-receptor-mediated responses. NMDA-receptor-mediated responses were quantified as the EPSC magnitude at +40 mV (50 ms following stimulation). Evoked EPSCs were recorded with picrotoxin (100 μM) added to the ACSF. To assess presynaptic function, a paired-pulse stimulation protocol (50 ms inter-stimulus interval) was used to evoke double-EPSCs, and the paired-pulse ratio (PPR) was quantified as the ratio of the peak amplitude of the second EPSC to that of the first EPSC. Rise time and decay time of EPSCs were measured between 20–80% of amplitude, respectively. For the measurement of rectification of AMPA current, spermine (100 μM) was added in the internal solution, while picrotoxin (100 μM) and NMDA receptor blocker D-(-)-2-Amino-5-phosphonopentanoic acid (AP-5) (50 μM) were added in ACSF. Neurons were held at −70, −40, 0, +40 or +70 mV, respectively. The rectification index (RI) was calculated by dividing the absolute amplitude of average EPSC at +40 mV by that at −70 mV. Naspm trihydrochloride (NASPM, 100 µM, Tocris) was applied to the slice while stimulating every 15 s. A stable baseline was recorded for at least 5 min followed by a 10 min application of NASPM. The average amplitude of the EPSC was calculated during the 5 min period prior to NASPM application (baseline) and the average of the four stimulations starting 9 min after drug application, with the EPSC amplitude normalized to the baseline period. Fast-scan cyclic voltammetry Fast-scan cyclic voltammetry was performed as previously described 32 . Carbon-fiber electrodes (CFEs) were prepared with a cylindrical carbon-fiber (7 mm diameter, ~150 mm of exposed fiber) inserted into a glass pipette and were conditioned with an 8-ms-long triangular voltage ramp (-0.4 to 1.2 and back to -.4 V vs Ag/AgCl reference at 400 V/s) delivered every 15 ms. CFEs showing current >1.8 mA or <1.0 mA in response to the voltage ramp at ~0.6 V were discarded. During recording, CFEs were held at -0.4 V versus Ag/AgCl, and a triangular voltage ramp was delivered every 100 ms. DA signals were evoked by single pulse optical stimulations, where a fiberoptic (200 mm diameter, 0.22 NA, ThorLabs) connected to a blue LED (470 nm, 1.8 mW, ThorLabs) was placed over the carbon fiber and light pulses ( ~ 0.6 ms) were delivered every 2 min. Data were collected with a retrofit headstage (CB-7B/ EC with 5 MX resistor) using a Multiclamp 700B amplifier (Molecular Devices) after a low-pass filter at 3 kHz and digitized at 100 kHz using a DA board (NI USB-6229 BNC, National Instruments). Data acquisition and analysis were performed using a custom-written software, VIGOR, in Igor Pro (Wavemetrics) using mafPC (courtesy of MA Xu-Friedman). The current peak amplitudes of the evoked DA transients were converted to DA concentration according to the post-experimental calibration using 1-3 mM DA. Histology and immunofluorescence Histological procedures were conducted as previously described 84 , 86 . Briefly, to confirm viral expression and optic fiber placement, animals were anesthetized with euthanasia solution (Vet One) and underwent transcardial perfusion with PBS (pH 7.4, 4 °C), followed by paraformaldehyde (PFA) solution (4% in PBS, 4 °C). Extracted brains were immersion-fixed in 4% PFA (4 °C, ≥2 h), then transferred to a 30% sucrose in PBS solution for cryoprotection until fully equilibrated (24–36 h). A Leica freezing microtome (SM 2010R) was used to collect 50 μm coronal sections. Sections were stained with DAPI, cover slipped with laboratory-prepared mounting medium and imaged using a Nikon AR1 confocal. RNAscope in situ hybridization RNAscope was performed as previously described 86 , 87 . For Fig. 5 and Supplementary Fig. 7 , brains from adult male C57BL/6NJ mice (8-12 weeks) were rapidly fresh-frozen and cut into 16 µm coronal sections on a cryostat (Leica Biosystems), mounted onto Superfrost Plus slides (Fisher Scientific), and stored at −80 °C. Sections covering the posterior extent of the PVT were fixed in 4% paraformaldehyde (30 min at 4 °C), dehydrated through graded ethanol solutions (50%, 70%, and 100% ×2), and RNAscope™ Multiplex Fluorescent v2 reagents (Advanced Cell Diagnostics) were applied as per the manufacturer’s guidelines. Following hydrogen peroxide treatment (10 min) and Protease IV incubation (30 min at room temperature), sections were hybridized with probes targeting Chat (#408731) and saCas9 (#501621) (2 h at 40 °C). Signal amplification (Amp 1–4) and HRP-based detection were performed at 40 °C, followed by fluorophore development using Opal 520 and Opal 570 reagents. Following DAPI counterstaining, sections were coverslipped using laboratory-prepared mounting medium and imaged using a Zeiss Axioscan slide-scanning microscope. Statistical analysis and reproducibility All data were plotted and analyzed with GraphPad Prism (v10.4.0), Python (v3.11.4), MATLAB (R2023b), RStudio (v2023.09.1), or Clampfit 10 (Molecular Devices). Behavioral tracking during the 2-way signaled active was performed using TopScan software (CleverSys), as previously described 27 , 83 . Change-point analyses were performed using the R package changepoint (URL: https://CRAN.R-project.org/package=changepoint ). R code used for active avoidance behavioral and photometry analysis is publicly available at: https://github.com/Penzolab/Data-analysis-of-Two-way-active-avoidance-task.git . (DOI: 10.5281/zenodo.12707790). A complete summary table of statistical results is provided (Supplementary Data 1 ). All statistical tests were two-sided and are described in the corresponding figure legends and Supplementary Data 1 . Data distribution were assessed for normality using Shapiro-Wilk tests. For datasets that violated assumptions of normality, non-parametric approaches or mixed modeling were used as appropriate. Unbalanced area-under-the-curve (AUC) datasets were analyzed using linear mixed-effects models fitted with restricted maximum likelihood (REML) using statsmodels mixedlm in Python. Models included a random intercept for subject to account for repeated measures. Main effects of categorical factors were assessed using Wald F-tests on model-estimated contrasts. Pairwise comparisons were performed using t-tests on contrasts and p-values adjusted with the Sidak method for multiple comparisons. Balanced datasets were tested with mixed-effects models (REML) in GraphPad Prism. Sample sizes are typically the same or exceed those estimated by power analyses (power = 0.80, α = 0.05). For fiber photometry experiments alone, the sample size is 5-10 mice. For fiber photometry experiments with optogenetic and/or chemogenetic manipulations, the sample size is 4-6 mice. For optogenetic behavior experiments, the sample size is 7-13 mice. For ex vivo electrophysiology experiments, the sample size was 5-22 cells from 3-5 mice. Independent replication confirmed consistent results across experiments, except for experiments described in Figs. 2g–k and 3d–o , where the desired sample sized was obtained in one cohort. Randomized assignment was used for all experiments. For ex vivo patch-clamp electrophysiology recordings, investigators were blinded to allocation during experiments. All 2AA photometric signals and behavioral performance were analyzed blinded. RL model A Rescorla-Wagner reinforcement learning model 88 was fit to the sample mean using Matlab: p a t + 1 = p a t + α β − p a t 1 where p a is the probability of avoidance, t is the trial, α is the learning rate, and β is the set value of avoidance. α, β, and a starting p a were fit as free parameters using least sum of squares. Standard error was estimated within groups using a 1000 bootstrap permutation test with replacement. A 5000 bootstrap permutation test with a null distribution generated by randomly reassigning group labels was used to assess the significance of differences between group parameters and p values determined by two-tailed comparison to the group parameter differences. The simulation in Fig. 1K was created using extreme high and low α and β parameters seen in real datasets. Trial outcome was generated using softmax probability (inverse temperature = 0.8) and 100 simulations were averaged. RL model fit to individuals A Bayesian hierarchical model (CmdStanPy (version 1.3.0 89 ) was used to fit individual subject behavior with a Rescorla-Wagner learning model. 20000 sampling iterations were run on four chains and 10000 warm-up samples were discarded. Convergence diagnostic R-hat values were satisfactory for all parameters in each model fit. A graphical representation of the Bayesian hierarchical model is depicted in Supplementary Fig. 3A with the prior distributions. Parameter recovery was performed using parameters from subject fits (Supplementary Fig. 3B ). Schematics Cartoon schematics were created in BioRender and licensed under: Macdonald, E. (2026) https://BioRender.com/4pm04zs . Histological maps were adapted from Allen Institute Mouse Brain Reference Atlas 90 . Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information (118.1MB, pdf) 41467_2026_73906_MOESM2_ESM.pdf (193.8KB, pdf) Description of Additional Supplementary Files Supplementary Data 1 (17.2KB, xlsx) Reporting Summary (101.7KB, pdf) Transparent Peer Review file (434KB, pdf) Source data Source Data (4.2MB, xlsx) Acknowledgements The authors thank the members of the Section on the Neural Circuits of Emotion and Motivation and the Laboratory on Neural Circuits and Behavior of NIMH who offered scientific feedback. In addition, we thank the NIMH IRP Rodent Behavioral Core and the NIMH IRP Systems Neuroscience Imaging Resource for supporting this work. Finally, we thank Dr. Michael Halassa for offering scientific and writing feedback. Author contributions Conceptualization: E.E.M. and M.A.P.; Data acquisition: E.E.M., J.M., D.L., K.Y., M.E.A., R.A.W., Ya.L., and H.C.G.; Data analysis: E.E.M., D.L., K.Y., and M.E.A.; Provided the GRAB ACh3.8 sensor: G.L. and Yu.L. Project management and supervision: M.A.P., B.B.A., and V.A.A.; Writing original draft: E.E.M and M.A.P. Peer review Peer review information Nature Communications thanks James Otis who co-reviewed with Rachel Clarke; and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by the Intramural Research Program (IRP) of the National Institutes of Health (NIH): NIMH IRP 1ZIAMH002950 (to M.A.P.) and ZIA MH002928 (to B.B.A). The contributions of the NIH author(s) were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services. Data availability The datasets generated in this study have been deposited in the Zenodo database under accession code DIO:10.5281/zenodo.20031828. All relevant processed data are also provided in the Supplementary Information/Source Data file. The datasets used in this study are fully publicly accessible without restriction. Source data are provided with this paper. Code availability R code used to analyze active avoidance behavior, and photometric signal is available at the following repository: https://github.com/Penzolab/Data-analysis-of-Two-way-active-avoidance-task.git . 10.5281/zenodo.12707790. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Emma E. Macdonald, Jun Ma, Di Liu, Kai Yu. Contributor Information Emma E. Macdonald, Email: emma.macdonald@nih.gov Bruno B. Averbeck, Email: averbeckbb@mail.nih.gov Mario A. Penzo, Email: mario.penzo@nih.gov Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73906-3. References 1. Rangel, A., Camerer, C. & Montague, P. R. A framework for studying the neurobiology of value-based decision making. Nat. Rev. Neurosci. 9 , 545–556 (2008). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Daw N. D., Tobler P. N. Value Learning through Reinforcement. In: Neuroeconomics [Internet]. Elsevier; 2014 [cited 2026 Feb 8]. p. 283–298. Available from: 10.1016/B978-0-12-416008-8.00015-2. 3. Balleine, B. W. & O’Doherty, J. P. Human and rodent homologies in action control: corticostriatal determinants of goal-directed and habitual action. Neuropsychopharmacol 35 , 48–69 (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. O’Doherty, J. P. Reward representations and reward-related learning in the human brain: insights from neuroimaging. Curr. Opin. Neurobiol. 14 , 769–776 (2004). [ DOI ] [ PubMed ] [ Google Scholar ] 5. Haber, S. N. & Knutson, B. The reward circuit: linking primate anatomy and human imaging. Neuropsychopharmacol 35 , 4–26 (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Crick, F. Function of the thalamic reticular complex: the searchlight hypothesis. Proc. Natl. Acad. Sci. USA 81 , 4586–4590 (1984). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Sherman S. M., Guillery R. W. The role of the thalamus in the flow of information to the cortex. Adams P., Guillery R. W., Sherman S. M., Sillito A. M., editors. Phil Trans R Soc Lond B. 357 , 1695–1708 (2002). [ DOI ] [ PMC free article ] [ PubMed ] 8. Wolff, M., Morceau, S., Folkard, R., Martin-Cortecero, J. & Groh, A. A thalamic bridge from sensory perception to cognition. Neurosci. Biobehav. Rev. 120 , 222–235 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 9. Christianson J. P. et al. Inhibition of fear by learned safety signals: a mini-symposium review. J Neurosci. 2012;32:14118–14124. 10.1523/JNEUROSCI.3340-12.2012 PubMed PMID: 23055481; PubMed Central PMCID: PMC3541026. [ DOI ] [ PMC free article ] [ PubMed ] 10. Lloyd, K. & Dayan, P. Safety out of control: dopamine and defence. Behav. Brain Funct. 12 , 1 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. LeDoux J. E., Moscarello J., Sears R., Campese V. The birth, death and resurrection of avoidance: a reconceptualization of a troubled paradigm. Mol Psychiatry. 2017;22:1. 10.1038/mp.2016.166 PubMed PMID: 27752080; PubMed Central PMCID: PMC5173426. [ DOI ] [ PMC free article ] [ PubMed ] 12. Moscarello, J. M. & LeDoux, J. E. Active avoidance learning requires prefrontal suppression of amygdala-mediated defensive reactions. J. Neurosci. 33 , 9 (2013). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Rescorla, R. A. & Solomon, R. L. Two-process learning theory: relationships between Pavlovian conditioning and instrumental learning. Psychol. Rev. 74 , 151–182 (1967). [ DOI ] [ PubMed ] [ Google Scholar ] 14. Cain, C. K. Avoidance problems reconsidered. Curr. Opin. Behav. Sci. 26 , 9–17 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Pultorak, K. J., Schelp, S. A., Isaacs, D. P., Krzystyniak, G. & Oleson, E. B. A transient dopamine signal represents avoidance value and causally influences the demand to avoid. eNeuro 5 , ENEURO.0058–18 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Campese V. D. et al. The neural foundations of reaction and action in aversive motivation. In: Simpson E. H., Balsam P. D., editors. Behavioral Neuroscience of Motivation [Internet]. Springer International Publishing; 2015 [cited 2022 Dec 17]. p. 171–95. (Current Topics in Behavioral Neurosciences). Available from: 10.1007/7854_2015_401. [ DOI ] [ PubMed ] 17. Van Der Werf, Y. D., Witter, M. P. & Groenewegen, H. J. The intralaminar and midline nuclei of the thalamus. Anatomical and functional evidence for participation in processes of arousal and awareness. Brain Res. Rev. 39 , 107–140 (2002). [ DOI ] [ PubMed ] [ Google Scholar ] 18. Penzo, M. A. & Gao, C. The paraventricular nucleus of the thalamus: an integrative node underlying homeostatic behavior. Trends Neurosci. 44 , 7 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. McGinty, J. F. & Otis, J. M. Heterogeneity in the paraventricular thalamus: the traffic light of motivated behaviors. Front Behav. Neurosci. 14 , 590528 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Petrovich, G. D. The function of paraventricular thalamic circuitry in adaptive control of feeding behavior. Front Behav. Neurosci. 15 , 671096 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Otis, J. M. et al. Paraventricular thalamus projection neurons integrate cortical and hypothalamic signals for cue-reward processing. Neuron 103 , 423–431.e4 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Averbeck, B. B. & Murray, E. A. Hypothalamic interactions with large-scale neural circuits underlying reinforcement learning and motivated behavior. Trends Neurosci. 43 , 681–694 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Do-Monte F. H., Quiñones-Laracuente K., Quirk G. J. A temporal shift in the circuits mediating retrieval of fear memory. Nature. 2015;519:7544. 10.1038/nature14030 PubMed PMID: 25600268; PubMed Central PMCID: PMC4376623. [ DOI ] [ PMC free article ] [ PubMed ] 24. Penzo M. A. et al. The paraventricular thalamus controls a central amygdala fear circuit. Nature. 2015;519:7544. 10.1038/nature13978 PubMed PMID: 25600269; PubMed Central PMCID: PMC4376633. [ DOI ] [ PMC free article ] [ PubMed ] 25. Zhu, Y., Wienecke, C. F. R., Nachtrab, G. & Chen, X. A thalamic input to the nucleus accumbens mediates opiate dependence. Nature 530 , 219–222 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Matzeu, A. & Martin-Fardon, R. Drug seeking and relapse: new evidence of a role for orexin and dynorphin co-transmission in the paraventricular nucleus of the thalamus. Front Neurol. 9 , 720 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Ma, J. et al. Divergent projections of the paraventricular nucleus of the thalamus mediate the selection of passive and active defensive behaviors. Nat. Neurosci. 24 , 10 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Berridge, K. C. & Robinson, T. E. What is the role of dopamine in reward: hedonic impact, reward learning, or incentive salience? Brain Res. Rev. 28 , 309–369 (1998). [ DOI ] [ PubMed ] [ Google Scholar ] 29. Salamone, J. D. & Correa, M. The mysterious motivational functions of mesolimbic dopamine. Neuron 76 , 470–485 (2012). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Floresco, S. B. The nucleus accumbens: an interface between cognition, emotion, and action. Annu Rev. Psychol. 66 , 1 (2015). [ DOI ] [ PubMed ] [ Google Scholar ] 31. Dinsmoor, J. A. Stimuli inevitably generated by behavior that avoids electric shock are inherently reinforcing. J. Exp. Anal. Behav. 75 , 311–333 (2001). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Oleson, E. B., Gentry, R. N., Chioma, V. C. & Cheer, J. F. Subsecond dopamine release in the nucleus accumbens predicts conditioned punishment and its successful avoidance. J. Neurosci. 32 , 42 (2012). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Dombrowski, P. A. et al. Evidence that conditioned avoidance responses are reinforced by positive prediction errors signaled by tonic striatal dopamine. Behav. Brain Res. 241 , 112–119 (2013). [ DOI ] [ PubMed ] [ Google Scholar ] 34. Stelly, C. E. et al. Pattern of dopamine signaling during aversive events predicts active avoidance learning. Proc. Natl. Acad. Sci. USA 116 , 27 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Lopez, G. C. et al. Region-specific nucleus accumbens dopamine signals encode distinct aspects of avoidance learning. Curr. Biol. 35 , 2433–2443 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Patriarchi, T. et al. Ultrafast neuronal imaging of dopamine dynamics with designed genetically encoded sensors. Science 360 , eaat4422 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Pinto, A., Jankowski, M. & Sesack, S. R. Projections from the paraventricular nucleus of the thalamus to the rat prefrontal cortex and nucleus accumbens shell: ultrastructural characteristics and spatial relationships with dopamine afferents. J. Comp. Neurol. 459 , 2 (2003). [ DOI ] [ PubMed ] [ Google Scholar ] 38. Ding, J. B., Guzman, J. N., Peterson, J. D., Goldberg, J. A. & Surmeier, D. J. Thalamic gating of corticostriatal signaling by cholinergic interneurons. Neuron 67 , 294–307 (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Baimel C., Jang E., Scudder S. L., Manoocheri K., Carter A. G. Hippocampal-evoked inhibition of cholinergic interneurons in the nucleus accumbens. Cell Rep. 2022;40:1. 10.1016/j.celrep.2022.111042 PubMed PMID: 35793623; PubMed Central PMCID: PMC9302453. [ DOI ] [ PMC free article ] [ PubMed ] 40. Zhou, F. M., Liang, Y. & Dani, J. A. Endogenous nicotinic cholinergic activity regulates dopamine release in the striatum. Nat. Neurosci. 4 , 1224–1229 (2001). [ DOI ] [ PubMed ] [ Google Scholar ] 41. Threlfell, S. et al. Striatal dopamine release is triggered by synchronized activity in cholinergic interneurons. Neuron 75 , 1 (2012). [ DOI ] [ PubMed ] [ Google Scholar ] 42. Cachope, R. et al. Selective activation of cholinergic interneurons enhances accumbal phasic dopamine release: setting the tone for reward processing. Cell Rep. 2 , 1 (2012). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Kosillo, P., Zhang, Y. F., Threlfell, S. & Cragg, S. J. Cortical control of striatal dopamine transmission via striatal cholinergic interneurons. Cereb. Cortex 26 , 11 (2016). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Adrover, M. F. et al. Prefrontal cortex-driven dopamine signals in the striatum show unique spatial and pharmacological properties. J. Neurosci. 40 , 39 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Kramer, P. F. et al. Synaptic-like axo-axonal transmission from striatal cholinergic interneurons onto dopaminergic fibers. Neuron 110 , 2949–2960.e4 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Jing, M. et al. An optimized acetylcholine sensor for monitoring in vivo cholinergic activity. Nat. Methods 17 , 1139–1146 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Sun, F. et al. A genetically encoded fluorescent sensor enables rapid and specific detection of dopamine in flies, fish, and mice. Cell 174 , 481–496.e19 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Geiger, J. R. P. et al. Relative abundance of subunit mRNAs determines gating and Ca2+ permeability of AMPA receptors in principal neurons and interneurons in rat CNS. Neuron 15 , 193–204 (1995). [ DOI ] [ PubMed ] [ Google Scholar ] 49. Angulo, M. C., Lambolez, B., Audinat, E., Hestrin, S. & Rossier, J. Subunit composition, kinetic, and permeation properties of AMPA receptors in single neocortical nonpyramidal cells. J. Neurosci. 17 , 6685–6696 (1997). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Rumpel, S., LeDoux, J., Zador, A. & Malinow, R. Postsynaptic receptor trafficking underlying a form of associative learning. Science 308 , 83–88 (2005). [ DOI ] [ PubMed ] [ Google Scholar ] 51. Isaac, J. T. R., Ashby, M. C. & McBain, C. J. The role of the GluR2 subunit in AMPA receptor function and synaptic plasticity. Neuron 54 , 859–871 (2007). [ DOI ] [ PubMed ] [ Google Scholar ] 52. Man, H. Y. GluA2-lacking, calcium-permeable AMPA receptors — inducers of plasticity? Curr. Opin. Neurobiol. 21 , 291–298 (2011). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 53. Hunker, A. C. et al. Conditional single vector CRISPR/SaCas9 viruses for efficient mutagenesis in the adult mouse nervous system. Cell Rep. 30 , 4303–4316.e6 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 54. De Martino, B. & Cortese, A. Goals, usefulness and abstraction in value-based choice. Trends Cogn. Sci. 27 , 65–80 (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 55. Schultz, W. Getting formal with dopamine and reward. Neuron 36 , 241–263 (2002). [ DOI ] [ PubMed ] [ Google Scholar ] 56. Li, H. et al. Neurotensin orchestrates valence assignment in the amygdala. Nature 608 , 586–592 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Vollmer, K. M. et al. An opioid-gated thalamoaccumbal circuit for the suppression of reward seeking in mice. Nat. Commun. 13 , 6865 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Paniccia, J. E. et al. Restoration of a paraventricular thalamo-accumbal behavioral suppression circuit prevents reinstatement of heroin seeking. Neuron 112 , 772–785.e9 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Crick, F. & Koch, C. A framework for consciousness. Nat. Neurosci. 6 , 119–126 (2003). [ DOI ] [ PubMed ] [ Google Scholar ] 60. Halassa, M. M. & Kastner, S. Thalamic functions in distributed cognitive control. Nat. Neurosci. 20 , 1669–1679 (2017). [ DOI ] [ PubMed ] [ Google Scholar ] 61. Fang, Z. et al. Human high-order thalamic nuclei gate conscious perception through the thalamofrontal loop. Science 388 , eadr3675 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 62. Horio, N. & Liberles, S. D. Hunger enhances food-odour attraction through a neuropeptide Y spotlight. Nature 592 , 262–266 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Engelke, D. S. et al. A hypothalamic-thalamostriatal circuit that controls approach-avoidance conflict in rats. Nat. Commun. 12 , 1 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Zhu, Y. et al. Dynamic salience processing in paraventricular thalamus gates associative learning. Science 362 , 6413 (2018). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Hua, R. et al. Calretinin neurons in the midline thalamus modulate starvation-induced arousal. Curr. Biol. 28 , 24 (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 66. Meffre, J. et al. Orexin in the posterior paraventricular thalamus mediates hunger-related signals in the nucleus accumbens core. Curr. Biol. 29 , 19 (2019). [ DOI ] [ PubMed ] [ Google Scholar ] 67. Witten, I. B. et al. Cholinergic interneurons control local circuit activity and cocaine conditioning. Science 330 , 1677–1681 (2010). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 68. Zhang, Y. F. et al. An axonal brake on striatal dopamine output by cholinergic interneurons. Nat. Neurosci. 28 , 783–794 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 69. Mohebi A., Collins V. L., Berke J. D. Accumbens cholinergic interneurons dynamically promote dopamine release and enable motivation. eLife. 2023;12. 10.7554/elife.85011. [ DOI ] [ PMC free article ] [ PubMed ] 70. Zhang Z. et al. Accumbal acetylcholine signals associative salience during learning [Internet]. Cold Spring Harbor Laboratory; 2025 [cited 2025 Jul 14]. Available from: 10.1101/2025.01.06.631529. 71. Touponse, G. C. et al. Cholinergic modulation of dopamine release drives effortful behaviour. Nature 651 , 1020–1029 (2026). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 72. Touponse G. C. et al. Cholinergic modulation of dopamine release drives effortful behavior [Internet]. Cold Spring Harbor Laboratory; 2025 [cited 2025 Jul 14]. Available from: 10.1101/2025.06.18.660394. 73. Zucca, S. et al. Trans-synaptic modulation of cholinergic circuits tunes opioid reinforcement. Proc. Natl. Acad. Sci. USA 122 , e2409325122 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 74. Krok, A. C. et al. Intrinsic dopamine and acetylcholine dynamics in the striatum of mice. Nature 621 , 543–549 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 75. Chantranupong, L. et al. Dopamine and glutamate regulate striatal acetylcholine in decision-making. Nature 621 , 577–585 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 76. Costa, K. M. et al. Dopamine and acetylcholine correlations in the nucleus accumbens depend on behavioral task states. Curr. Biol. 35 , 1400–1407.e3 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 77. Park, I. H., Lee, B. C., Kim, J. J., Kim, J. I. & Koo, M. S. Effort-based reinforcement processing and functional connectivity underlying amotivation in medicated patients with depression and schizophrenia. J. Neurosci. 37 , 4370–4380 (2017). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 78. Patton M. H. et al. Schizophrenia-associated 22q11.2 deletion elevates striatal acetylcholine and disrupts thalamostriatal projections to produce amotivation in mice [Internet]. Cold Spring Harbor Laboratory; 2025 [cited 2025 Jul 23]. Available from 10.1101/2025.06.20.660782. 79. Galderisi, S., Mucci, A., Buchanan, R. W. & Arango, C. Negative symptoms of schizophrenia: new developments and unanswered research questions. Lancet Psychiatry 5 , 664–677 (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 80. Correll, C. U. & Schooler, N. R. Negative symptoms in schizophrenia: a review and clinical guide for recognition, assessment, and treatment. NDT 16 , 519–534 (2020). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 81. Holt, D. J. et al. Evidence for a deficit in cholinergic interneurons in the striatum in schizophrenia. Neuroscience 94 , 21–31 (1999). [ DOI ] [ PubMed ] [ Google Scholar ] 82. Haight J. L., Flagel S. B. A potential role for the paraventricular nucleus of the thalamus in mediating individual variation in Pavlovian conditioned responses. Front Behav Neurosci. 2014;8. 10.3389/fnbeh.2014.00079. [ DOI ] [ PMC free article ] [ PubMed ] 83. Ma, J. et al. Convergent direct and indirect cortical streams shape avoidance decisions in mice via the midline thalamus. Nat. Commun. 15 , 6598 (2024). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 84. Beas B. S. et al. The locus coeruleus drives disinhibition in the midline thalamus via a dopaminergic mechanism. Nat Neurosci. 2018;21:7. 10.1038/s41593-018-0167-4 PubMed PMID: 29915192; PubMed Central PMCID: PMC6035776. [ DOI ] [ PMC free article ] [ PubMed ] 85. Francis, T. C., Yano, H., Demarest, T. G., Shen, H. & Bonci, A. High-Frequency Activation of Nucleus Accumbens D1-MSNs Drives Excitatory Potentiation on D2-MSNs. Neuron 103 , 432–444.e3 (2019). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 86. Gao C. et al. Two genetically, anatomically and functionally distinct cell types segregate across anteroposterior axis of paraventricular thalamus. Nat Neurosci. 2020;23:2. 10.1038/s41593-019-0572-3 PubMed PMID: 31932767; PubMed Central PMCID: PMC7007348. [ DOI ] [ PMC free article ] [ PubMed ] 87. Gao, C. et al. Molecular and spatial profiling of the paraventricular nucleus of the thalamus. eLife 12 , e81818 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 88. Rescorla R., Wagner A. A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In: Classical Conditioning II: Current Research and Theory. 1972. p. 64–99. 89. Carpenter B. et al. Stan : a probabilistic programming language. J Stat Soft. 2017;76. 10.18637/jss.v076.i01. [ DOI ] [ PMC free article ] [ PubMed ] 90. Allen Reference Atlas – Mouse Brain [brain atlas]. Available from atlas.brain-map.org. Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Information (118.1MB, pdf) 41467_2026_73906_MOESM2_ESM.pdf (193.8KB, pdf) Description of Additional Supplementary Files Supplementary Data 1 (17.2KB, xlsx) Reporting Summary (101.7KB, pdf) Transparent Peer Review file (434KB, pdf) Source Data (4.2MB, xlsx) Data Availability Statement The datasets generated in this study have been deposited in the Zenodo database under accession code DIO:10.5281/zenodo.20031828. All relevant processed data are also provided in the Supplementary Information/Source Data file. The datasets used in this study are fully publicly accessible without restriction. Source data are provided with this paper. R code used to analyze active avoidance behavior, and photometric signal is available at the following repository: https://github.com/Penzolab/Data-analysis-of-Two-way-active-avoidance-task.git . 10.5281/zenodo.12707790. Articles from Nature Communications are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (4.2 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Safety, Security, and Cognitive Risks in Neuro-Symbolic AI#282736
• Membrane: A Self-Evolving Contrastive Safety Memory for LLM Agent Defense#259358
• Your Model Already Knows: Attention-Guided Safety Filter for Vision-Language-Action Models#267645
• Controllable Progress as an Intrinsic Value Signal in Cognitive Control#361051
• NSF-HRPT: Neural Semantic Field meets Hierarchical Risk Perception Tree for Safety-Critical Scenario Assessment#429411