Dopaminergic Plasticity in the Bilateral Hippocampus Following Threat Reversal in Humans

When a cue no longer predicts a threat, a diminished ability to extinguish or reverse this association is thought to increase risk for stress-related disorders. Despite the clear clinical relevance, the mediating neurochemical mechanisms of threat reversal have received relatively little study. One neurotransmitter implicated in rodent research of changing associations with threat is dopamine. To study whether dopamine is involved in threat reversal in humans, we used high-resolution positron emission tomography (PET) coupled with 18F-fallypride. Twelve healthy volunteers (6 F/6 M) underwent three PET scans: (i) at baseline, (ii) following threat conditioning (the response to a cue associated with electric wrist shock), and (iii) following threat reversal (the response to the same cue now associated with safety). We observed moderate evidence of reduced dopamine D2/3 receptor availability, consistent with greater dopamine release, in the bilateral anterior hippocampus following threat reversal, in response to a safety cue that was previously associated with threat, as compared to both baseline and during exposure to the same cue prior to threat reversal. These findings offer the first preliminary evidence that the response to a previously threatening cue that has since become associated with safety involves dopaminergic neurotransmission within the hippocampus in healthy humans.

To identify the brain regions where dopamine transmission is engaged during the recall of threat reversal in humans, the present study used positron emission tomography (PET) with 18 F-fallypride, a highly selective, high affinity dopamine D2/3 receptor ligand [20][21][22] . We hypothesized that, compared to baseline (prior to threat conditioning), changes in 18 F-fallypride binding would be observed within mesocorticolimbic circuitry both in response to a threat-associated cue and in response to the same cue following threat reversal.

Results
participants. Sixteen volunteers were enrolled in the study following initial screening procedures. One participant was excluded due to an inadequate autonomic response to the aversive stimulus during screening, one was excluded after the baseline PET scan (PET BL ) due to a headache during this first PET scan, one withdrew after the MRI scan for unknown reasons, and one withdrew after the PET BL session for unknown reasons. Therefore, a total of 12 volunteers (6 F/6 M, mean ± SD age = 24.1 ± 3.7 years) completed all study sessions (Fig. 1) and were included in the analyses.
Subjective & autonomic indices of conditioned threat. All participants reported learning the correct associations between the neutral, conditioned stimuli (CS) and aversive, unconditioned stimulus (US), both immediately after the CS-US pairings and immediately prior to PET scanning. Accordingly, all participants reported associating "a little" to "moderate" anxiety with the conditioned cue paired with threat (CS+) following threat learning and prior to the second PET session (PET2 CS+ ). Following threat reversal and prior to the third PET session (PET3 CS− ), 11 participants reported no anxiety associated with the same cue, which now predicted the absence of threat (new CS−). The 12 th participant indicated "a little" subjective anxiety associated with the new CS− prior to PET3 CS− ; although this participant reported the correct contingencies during reversal learning, a sensitivity analysis was performed with the participant's data excluded.
A significant main effect of PET session was found for the frequency of SCRs (F 2,22 = 11.5, p < 0.001). Post hoc analyses revealed a significantly higher percentage frequency of SCRs in PET2 CS+ (after presentation of the CS+, during anticipation of an aversive shock), as compared to both PET3 CS− (t 11 = 3.41, p = 0.006) and PET BL (t 11 = 4.13, p = 0.002). By contrast, the percentage frequency of SCRs did not differ significantly between PET BL and PET3 CS− (t 11 = 1.08, p = 0.3), suggesting that the autonomic response to the former CS+ was effectively inhibited at PET3 CS− (Fig. 2). 18 f-fallypride binding potential. The mean ± SD injected activity of 18  Descriptive and linear mixed model statistics for each a priori-hypothesized ROI are summarized in Table 1. We found a significant main effect of PET session in the anterior hippocampus (F 2,55 = 3.5, p = 0.037), with no significant hemisphere × session interaction (F 2,55 = 0.01, p = 0.99). Post hoc analysis attributed the main effect of session to lower non-displaceable binding potential (BP ND ) in PET3 CS− in the bilateral hippocampus, as compared to both PET BL (t 11  . The observed decreases in anterior hippocampus BP ND likely reflect increases in regional dopamine release during exposure to the updated safety cue following threat reversal, compared to both baseline and during exposure to the same conditioned cue www.nature.com/scientificreports www.nature.com/scientificreports/ prior to threat reversal. A sensitivity analysis excluding the data from one participant who reported reduced, but not totally absent anxiety in response to the new CS− prior to PET3 CS− , yielded similar findings of decreased BP ND in the bilateral hippocampus in response to the new safety cue (new CS−) following threat reversal (PET BL vs. PET3 CS− , t 10 = 2.4, p = 0.035; PET2 CS+ vs. PET3 CS− , t 10 = 2.5, p = 0.031).
Complementing the linear mixed model and pairwise comparisons results, we observed moderate evidence in support of an effect of threat reversal on PET3 CS− BP ND in the anterior hippocampus. A Bayes factor ANOVA   with default priors showed that the PET session main effect model was preferred to the null model including hemisphere by a Bayes factor of 1.71, suggesting anecdotal evidence of an overall effect of PET session on BP ND in the hippocampus. However, comparing BP ND in PET3 CS− versus PET BL and PET2 CS+ separately, the data were 11.42 times and 8.63 times more likely under H 1 than H 0 , respectively, indicating moderate evidence for a reduction in hippocampal BP ND following threat reversal in particular. Bayes factors from the Bayesian analyses of a priori-hypothesized ROIs are reported in Table 2. BP ND values did not differ significantly between PET scans in the other a priori-hypothesized ROIs (Table 1). Accordingly, the Bayesian evidence for an effect of PET session on BP ND in these ROIs was either inconclusive or in favor of the null hypothesis (Table 2). Although a reduction in amygdala BP ND in PET3 CS− compared to PET BL was favoured over the null hypothesis, the evidence for a reduction in amygdala D2/3 receptor availability following threat reversal was weak, and not replicated when comparing PET3 CS− with PET2 CS+ . Exploratory analyses also revealed that BP ND values did not change significantly across PET scans in the anterior cingulate (PET session main effect: F 2,55 = 0.72, p = 0.49; hemisphere × session interaction: F 2,55 = 0.20, p = 0.82) or insula (PET session main effect: F 2,55 = 0.55, p = 0.58; hemisphere × session interaction: F 2,55 = 0.20, p = 0.82).
No significant correlations were observed between the change in SCR frequency and the change in regional BP ND between PET scans. Similarly, no significant correlations were observed between changes in regional BP ND and changes in subjective measures of mood or anxiety across scans. Of note, no significant correlations were observed between changes in BP ND across PET scans in the hippocampus and changes in subjective measures of sleepiness across scans, suggesting that changes in levels of sleepiness did not account for the 18 F-fallypride binding results reported here. , as compared to baseline (PET BL ) and in response to the same cue prior to threat reversal (PET2 CS+ ). The observed decrease in BP ND between scans is consistent with an increase in dopamine release. Error bars represent 95% confidence intervals.

Discussion
To our knowledge, the current study is the first investigation of dopamine release in humans following threat conditioning and reversal. We observed a significant and internally replicated decrease in dopamine D2/3 receptor availability in the bilateral hippocampus in response to a safety cue that had been previously associated with threat, as compared to both baseline and the response to the same cue prior to threat reversal. Bayesian analysis showed moderate evidence in favour of this effect. Evidence of decreased D2/3 receptor binding following threat reversal was also observed in the amygdala and nucleus accumbens, but the Bayesian evidence for these effects was inconclusive.
Both subjective reports and autonomic measurements during the PET scans confirmed that the presented conditioned cue was associated with electric shock following threat learning and with safety following threat reversal. Since the PET2 CS+ and PET3 CS− sessions were identical, differences between scans in tracer binding likely reflect the changed significance of the conditioned cue. Since small reductions in 18 F-fallypride binding are associated with large (>25-fold) increases in extracellular dopamine levels measured with microdialysis 23-25 , the bilateral decrease in hippocampal tracer binding in the PET3 CS− session is consistent with increased hippocampal dopamine release following threat reversal. Together, these findings constitute preliminary evidence that dopaminergic plasticity within the bilateral anterior hippocampus plays a role in safety signaling following the flexible updating of associations with threat.
Our findings in the bilateral hippocampus are consistent with past studies of learned responses to safety cues. An fMRI study in humans found that the hippocampus was activated in response to an extinguished threat cue, as compared to an unextinguished threat cue, and this activation correlated with the magnitude of extinction memory 10 . More recently, the conditioned inhibition of threat responding was found to activate neuronal subpopulations within the ventral/anterior hippocampus in both mice and humans 26 . A meta-analysis of fMRI studies suggests that these effects are relatively robust with significantly increased activity seen in the prefrontal cortex and anterior hippocampus in response to an extinguished/safety cue, as compared to an unextinguished/ threat-associated cue 27 , similar to the findings from an earlier meta-analysis 28 . Studies in rodents suggest that these effects reflect causal mechanisms. Inactivation of the ventral hippocampus prior to extinction learning impairs extinction memory in rats 29 . Additionally, Pollak et al. showed that the ablation of hippocampal neurogenesis impairs learned safety in mice, and the systemic administration of dopamine agonists/antagonists alters the recall of learned safety 30 . More generally, the hippocampus may enable similar yet distinct associative memories to be stored as separate representations 31 . Of note, the meta-analyses did not identify consistent fMRI-measured activations in the amygdala in either threat learning or extinction recall 27,28 .
The specific involvement of hippocampal dopamine in the suppression of learned associations with threat has been less studied, but a recent study in rats found that the enhancement of threat extinction through exposure to a novel environment is dependent on dopamine D1 receptors in the hippocampus 14 . Within the context of the associative memory literature, our findings suggest that dopaminergic plasticity within the hippocampus may be involved in associative memory processes that underlie the inhibition of learned associations with threat in humans.
The current findings are relevant to disorders in which the inhibition of learned associations with threat is impaired, and in which mesocorticolimbic regions, such as the hippocampus, show abnormalities. For example, there is evidence of impaired extinction recall in PTSD patients 4 , and within the same patients, recall of an extinction memory correlated with hippocampal activation 32 . An improved understanding of the mechanisms involved in safety signaling following threat reversal is important for the optimization of exposure therapy for these disorders.
Contrary to our hypotheses, we did not observe significant 18 F-fallypride binding changes within the vmPFC, nor did exploratory analyses identify effects in the anterior cingulate or insula. Each of these regions has been implicated in different aspects of fear and threat-related learning 27 . Dopamine, however, might contribute to only some of these responses. Indeed, regionally-specific subgroups of dopamine neurons within mesocorticolimbic circuitry exhibit distinct responses to different types of events and cues 33,34 and vmPFC dopamine depletions in the marmoset do not influence performance on a reversal learning task 35 . The specificity of the current findings to dopamine neurons that innervate the anterior hippocampus is in line with this body of literature. Nevertheless, the vmPFC Bayesian analyses did not conclusively favour the null hypothesis, and studies in rodents suggest that dopaminergic activity in the PFC influences some aspects of extinction memory 13 . Since the vmPFC and hippocampus are highly connected 36 , and stimulation of the vmPFC has been shown to increase hippocampal cell proliferation and memory 37 , future studies should employ tracers that may be more sensitive to neurotransmitter release in cortical regions, such as 11 C-FLB 457 38 .
The current study has limitations to consider. First, the sample size is modest due to the nature of PET imaging in general and demands of the present study in particular (e.g., >9 hours of PET scanning per participant). However, to our knowledge, this is the largest PET study reported to date on the inhibition of learned associations with threat in humans. We therefore consider the findings reported here thought provoking, yet requiring replication. Second, it is important to note that the findings were not corrected for multiple comparisons, however, the hippocampal dopamine response was observed bilaterally and Bayesian analyses indicated moderate evidence in favour of this result. Third, no shocks were administered during the second scanning session (PET2 CS+ ), which constitutes both a strength and a limitation. Although the study design avoids the confound of administering an aversive stimulus during PET scanning, by repeatedly presenting the CS+ in the absence of shock, it is likely that the measured PET signal reflects a combination of conditioned threat and extinction learning or prediction error (whereby a shock is expected but does not occur). This is a confound inherent to all experimental designs that measure the response to the CS+ presented alone and might account for why compelling evidence of dopamine release in the PET2 CS+ session was not seen. Fourth, without a separate control group, the effect of scan order on the 18 F-fallypride signal cannot be ruled out, but 18  www.nature.com/scientificreports www.nature.com/scientificreports/ reliability making this less likely 39,40 . Finally, there are inherent limitations to the use of 18 F-fallypride. Although 18 F-fallypride provides a reliable signal in brain regions with dopamine receptor densities that are lower than in the striatum, such as the hippocampus and amygdala, 18 F-fallypride is not the optimal tracer for studying changes in striatal dopamine receptor availability; both 11 C-raclopride and 11 C-PHNO are considered superior for this purpose 20,41 . Decreases in 18 F-fallypride binding observed are typically interpreted to reflect increased dopamine release, though it is possible that they reflect other forms of dopaminergic plasticity (including receptor internalization and trafficking, or changes in receptor conformational state).
In summary, despite its clinical relevance, threat reversal has been less extensively studied than other aspects of threat learning. The importance of dopamine in threat reversal remains particularly understudied. The present study provides the first preliminary evidence that dopaminergic activity within the anterior hippocampus is important for safety signaling following threat reversal in healthy humans. It also raises the possibility that abnormal hippocampal dopaminergic plasticity might play a role in psychiatric disorders characterized by a perseveration of responses to stimuli that are no longer threatening, such as PTSD, anxiety disorders and OCD.

overview.
The study entailed five test days including three PET scans. First, a baseline PET scan (PET BL ) was performed while participants were presented with a white screen; no other stimuli were presented, and participants were instructed to relax with their eyes open. Prior to threat conditioning, participants also underwent an anatomical magnetic resonance imaging (MRI) scan for co-registration with PET. Approximately one week after PET BL , participants learned to associate a neutral visual cue with threat during the first stimulus pairing session (described below). One business day later, the second PET scan (PET2 CS+ ) was performed, during which the conditioned cue associated with threat (CS+) was presented alone, without the aversive stimulus. Approximately one week after PET2 CS+ , a threat reversal paradigm was performed. One business day later, participants underwent the 3 rd and final PET measurement (PET3 CS− ), which was performed in an identical manner to PET2 CS+ , but the presented conditioned cue now predicted the absence of threat (new CS−). Data acquired during the third PET session reflect the response to the updated safety cue following threat reversal (Fig. 1). participants. Healthy, right-handed volunteers aged 20-40 years were recruited using online advertisements on university websites. After a brief telephone screening, individuals who tentatively met the inclusion criteria underwent an in-person interview with the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID) 42 , an electrocardiogram, blood work, a urine toxicology/pregnancy test, and a routine physical exam performed by a physician. Lastly, to verify that participants showed an adequate autonomic response to the aversive stimulus in the PET environment, baseline skin conductance and heart rate were first recorded during a 3-min rest period. Inclusion in the study required a >10% change in skin conductance and/or >1 SD change in heart rate from mean baseline values soon after mild electrical stimulation of the wrist. Exclusion criteria included a current or past Axis I disorder, family history of an Axis I disorder, serious physical illness, chronic medication use, regular tobacco (>5 cigarettes/day) and/or occasional cannabis use (>twice/month), as well as any counter-indications to MRI or PET. During screening, participants were familiarized with the PET room and scanner in order to minimize the effect of novelty during PET BL .
The study was carried out in accordance with the Declaration of Helsinki, and was approved by the Research Ethics Board of the Montreal Neurological Institute. All participants provided written, informed consent.
Associative learning paradigm. Stimulus pairing sessions took place in the PET scanner, without scanning taking place. The presentation of all stimuli was programmed using SuperLab 4.5 (Cedrus Corporation, San Pedro, CA). All visual stimuli were presented in video glasses (OEM EVG920D Video Eyewear; 640 × 480 resolution, virtual display equivalent to 80″ at 1 m with a 35° viewing angle), compatible with the bore of the PET scanner. Electric pulses of 50 ms were administered using a stimulating bar electrode (Biopac convex unshielded bar electrode EL351, with 2 tin electrodes, spaced 30 mm apart) secured over the ulnar nerve of the left wrist. Electrode leads were connected to a Biopac STM200 (Constant Voltage Stimulator -Unipolar Pulse). The stimulating bar electrode was secured to the participant's wrist during all pairing sessions, and during PET2 CS+ and PET3 CS− (the stimulator was inactive during scans).
Both the acquisition and reversal of learned threat involved a cue-dependent, trace conditioning paradigm with partial reinforcement. The neutral, conditioned stimuli (CS) consisted of a grey triangle and a grey circle of equal area. The aversive, unconditioned stimulus (US) consisted of a mild electric shock to the non-dominant wrist just below or about at "pain threshold" (described below in Subjective and Autonomic Measurements). Each participant's pain threshold was established at the start of the study, and immediately prior to each pairing session. One of the neutral cues (CS+) was followed by the aversive US in 30% of CS+ trials, whereas the other cue (CS−) was never followed by the US. The shape that was first paired with shock was counterbalanced across participants. A pairing session involved 20 trials (10 CS+, 10 CS−, in pseudorandom order), where each trial consisted of a 3 s CS presentation, followed by a 20 s countdown, and a 7 s blank screen during which participants either did or did not receive a brief shock. The low contingency rate (30%, i.e. 3 out of 10 CS+ trials were paired with shock) was employed to take advantage of the higher stress response to unpredictable stressors, as compared to predictable stressors 43,44 . By performing pairing in the PET scanner, context remained constant for both associative learning and subsequent recall; there is evidence that consistent context facilitates the retrieval of associative memories 45 . Pairing and scanning sessions were separated by 24 hours to allow for optimal memory consolidation prior to scanning 46 , and to further avoid an aversive stimulus confound.
The same pairing procedure was used for threat reversal (20 trials; 10 CS+, 10 CS−), except that the CS-US contingencies were reversed; the cue previously paired with the US was no longer followed by a shock (the CS+ www.nature.com/scientificreports www.nature.com/scientificreports/ became the new CS−), whereas the previously neutral cue was paired with shock (the CS− became the new CS+). Participants were not informed of the stimulus contingencies prior to pairing sessions.

Subjective anxiety and autonomic measurements.
To determine the appropriate electric shock intensity, the subjective pain threshold for each participant was defined as a 3 on a Numerical Rating Scale (0 = No Sensation; 1 = Just Noticeable; 2 = Uncomfortable; 3 = Pain Threshold; 4 = Painful; 5 = Maximum Tolerable), and at least 20 on a visual analog scale (VAS) of pain (0 = No Pain; 100 = Extremely Painful) 47,48 . A contingency awareness questionnaire was administered immediately after each pairing session to assess which CS the participant associated with shock, and the subjective anxiety associated with each CS (1 = None; 2 = A Little; 3 = Moderate; 4 = Extreme). The same questionnaire was administered immediately before PET2 CS+ and PET3 CS− in order to prime the CS-US associative memory.
Subjective ratings of mood, anxiety and alertness were collected immediately before, and 30 and 150 minutes into, each PET scan. The questionnaires included the Profile of Mood States (POMS) 49 , state-trait anxiety inventory (STAI-State) 50 , and Alertness VAS 51 . POMS scores on 6 bipolar scales (elated-depressed, composed-anxious, energetic-tired, agreeable-hostile, confident-unsure, clearheaded-confused) were transformed into population normalized t scores.
Electrodermal activity and heart rate were measured continuously as autonomic indices of conditioned threat using Ag/AgCl disposable electrodes on the middle phalanges of the right index and middle fingers, and on the left and right sides of the chest. Electrodermal activity was analyzed as the frequency of skin conductance responses (SCRs), which reflect phasic deflections in the electrical conductivity of the skin. SCR data were analyzed offline using AcqKnowledge software. To assess the effectiveness of the CS at inducing event-related SCRs, we calculated the number of trials in which a phasic SCR occurred during the 30 s CS-US interval 52,53 , using a threshold for SCR detection of a base to peak difference>3 SD of baseline skin conductance. Baseline skin conductance was calculated as the mean skin conductance level during the 2 s interval before the CS onset. To minimize the impact of SCR habituation 54 , we calculated the frequency of phasic SCRs that occurred in the first 10 trials of PET2 CS+ (in response to the CS+) and PET3 CS− (in response to the new CS−), or during the same time intervals in PET BL (non-specific SCRs occurring in the absence of stimuli), as a percentage frequency per 10 stimulus presentations ((SCR count / 10 trials) × 100%).
pet and MRi acquisition. Prior to each PET scan, a urine toxicology screen for illicit drugs of abuse was performed (Triage, Biosite Diagnostics, San Diego, CA), as well as a urine pregnancy test in women. PET measurements were performed using a high-resolution research tomograph dedicated brain scanner (HRRT; CTI/ Siemens, Knoxville, TN) in the late morning to early afternoon. Scan resolution was 2.3-3.4 mm full width at half maximum.
First, a 6-min transmission scan was performed for attenuation correction, followed by a bolus injection of 18 F-fallypride through an i.v. catheter in the left arm vein. Each PET scan was 3 hours in duration, consisting of 90 minutes of dynamic acquisition scanning, followed by a 30-minute break and a final 60-minute dynamic acquisition scan. The following sequence of frame durations was used during dynamic scanning: 3 × 10 s, 5 × 30 s, 4 × 60 s, 4 × 120 s, 5 × 300 s, 5 × 600 s and 6 × 600 s.
In all PET sessions, participants were instructed to stay awake, keep their eyes open, and relax. During PET BL , participants were informed that no shocks would be delivered during the scan. Recording electrodes were set up, but the stimulating bar electrode was not. Following tracer injection in PET BL , participants were presented with a white screen. During both PET2 CS+ and PET3 CS− , participants were instructed that only one of the shapes from the previous pairing session would be presented during the scan. Following tracer injection in PET2 CS+ and PET3 CS− , the same conditioned stimulus (considered to be the CS+ during PET2 CS+ and the new CS− during PET3 CS− ) was presented repeatedly during the first 30 minutes of scanning (60 trials per PET session). As in the stimulus pairing session, each trial consisted of a 3 s presentation of the CS, followed by a 20 s countdown, and a 7 s blank screen. The timeline of each PET scan is illustrated in Fig. 4.
MRI scans were conducted using a 3 T scanner equipped with a 32-channel head coil (Siemens TIM Trio Magnetom; Erlangen, Germany). A 9-minute T1-weighted anatomical MRI scan was performed (TR = 2300 ms; TE = 2.98 ms; flip angle = 9°; voxel size = 1.0 × 1.0 × 1.0 mm). pet and MRi data processing. PET data reconstruction was carried out using a maximum-likelihood expectation maximization iterative algorithm that corrects for scattered and random coincidences, attenuation, and detector-based non-uniformities 55 . PET frames were motion corrected using an automated algorithm 56 . The Simplified Reference Tissue Model (SRTM) 57 , with the basis functions method optimized for 18 F-fallypride from 11 C-raclopride studies 58 , was used to calculate BP ND values at each voxel 20,59,60 . The cerebellar grey matter, which has minimal expression of D2/3 receptors, was used as a reference region. Following PET-MR co-registration and the transformation of the MRI scan and BP ND map into MNI152 space, a 6-mm Gaussian filter was applied to the BP ND map in order to reduce effects of anatomical variability.
Finally, regions of interest (ROIs) were defined bilaterally in the amygdala, anterior hippocampus, ventral tegmental area (VTA), nucleus accumbens and ventromedial prefrontal cortex (vmPFC). ROIs were also defined in the insula and anterior cingulate cortex for exploratory analyses based on recent fMRI evidence for the involvement of these regions in extinction recall 27 . ROI masks were created using the Wake Forest University (WFU) PickAtlas toolbox 61 for SPM12, using the Automated Anatomical Labeling atlas (amygdala, hippocampus, vmPFC, anterior cingulate cortex and insula) 62 , the IBASPM 71 library (nucleus accumbens) 63 , and the VTA atlas from the Adcock lab 64 . Given that the ventral hippocampus in rodents, corresponding to the anterior hippocampus in humans, connects more densely to the amygdala 65,66 , receives stronger dopaminergic projections from the VTA 67 , and has been more widely implicated in trace conditioning, as compared to the dorsal hippocampus 68,69 , (2020) 10:7627 | https://doi.org/10.1038/s41598-020-63977-7 www.nature.com/scientificreports www.nature.com/scientificreports/ the relatively large automatically-segmented hippocampal ROI mask was manually reduced to include only anterior hippocampus. All ROIs were checked against individual MRI scans and adjusted manually if necessary. Mean BP ND values were calculated bilaterally for each ROI from the BP ND map in stereotaxic space, as well as by hemisphere for amygdala, hippocampus, nucleus accumbens and vmPFC ROIs. Given the relatively fast clearance of 18 F-fallypride from limbic areas as well as cortex 25,39,70 , only the data from the first 90-minute scan were used for extra-striatal ROIs. A period of three hours is believed to be necessary to achieve transient equilibrium in the striatum 60 .

Statistical analyses.
We performed linear mixed-effects models for each ROI to assess changes in BP ND across PET sessions by hemisphere, using a random effect of subject and fixed effects of PET session (3 timepoints: PET BL , PET2 CS+ , PET3 CS− ) and hemisphere (left, right). For bilateral VTA BP ND , injected dose, injected mass and specific activity of 18 F-fallypride, as well as mood, anxiety and autonomic measures, linear mixed models were performed including subject as a random effect and PET session as a fixed effect. Planned pairwise comparisons consisted of two-tailed paired t-tests. Exact p-values are reported, uncorrected for multiple comparisons. The distributions of the residuals were checked using histograms and Q-Q plots, and the presence of influential outliers was evaluated using Cook's distance. Pearson correlations were also performed between changes in BP ND between PET sessions and changes in subjective and autonomic measures between PET sessions.
Lastly, we performed a Bayesian repeated measures ANOVA (two factors: PET session and hemisphere) for each ROI using JASP software [71][72][73] in order to quantify the strength of evidence in favour of either the null hypothesis (H 0 : no effect of associative learning on regional dopamine release), or the alternative hypothesis (H 1 : an effect of associative learning on regional dopamine release). Bayes factors (BF 10 ) were calculated for main effect and interaction models, including hemisphere as a nuisance variable. Post hoc comparisons were performed between PET sessions, with the posterior odds corrected for multiple comparisons 74 . A BF 10 > 1 indicates evidence for an effect (H 1 ), and a BF 10 < 1 indicates evidence for no effect (H 0 ). The strength of the evidence in favour of either hypothesis is considered to be of interest when BF 10 is under 0.33 or over 3, otherwise the evidence is considered to be "anecdotal" and inconclusive 75 .