A: Overview of the three-day fear conditioning paradigm. Day 1 consisted of habituation and acquisition training, day 2 of extinction training, and day 3 of a recall test followed by reacquisition, reextinction, and the unexpected US phase. Extinction, recall, reacquisition, and reextinction are highlighted in purple, indicating phases in which unexpected US omissions occurred. Extinction, recall, and reextinction are shown with a gradient to illustrate the gradual decrease in unexpectedness of US omissions across trials, in contrast to the discrete omission events during partially reinforced reacquisition. B: Schematic of the experimental setup during 7 T MRI. Participants lay supine in the scanner and viewed visual conditioned stimuli (CS; geometric shapes) displayed on a screen at the end of the bore via a mirror system. The eyes were simultaneously imaged with an eye-tracking system. Electrodermal activity (EDA) was recorded from the left hand (blue/purple cables). The unconditioned stimulus (US; electrical stimulation) was delivered to the right hand (red cable), and responses were recorded using a button box.

Skin conductance responses (SCRs) for each trial, with CS+ (shown in orange) and CS- (shown in blue) responses paired in blocks.

Reinforcement of the CS+ by a US (CS+/US) is indicated by filled squares. The CS- is never reinforced. Bar plots on the top right show mean SCRs for early and late trials, defined as the first and second halves of each phase, respectively. Significance markers indicate post hoc comparisons between CS+ and CS- within early or late trials; markers are shown in black when the Stimulus x Time interaction was significant and in red when the interaction was not significant. On day 1, there was no differentiation between CS+ and CS- in the habituation phase, with significant differentiation emerging during acquisition training. On day 2, differentiation was apparent in early extinction trials and was no longer evident later in extinction (trend-level effect; non-significant Stimulus x Time interaction). On day 3, during the initial recall test, participants exhibited renewed CS+/CS- differentiation, consistent with spontaneous recovery. During initial reacquisition, there were again differential responses to the CS+ and CS-, which appeared reduced in reextinction and the unexpected US phase. Mean values are shown with error bars representing the standard error of the mean. CS: Conditioned stimulus; US: Unconditioned stimulus; SCR: Skin conductance response; μS: microsiemens.

Pupil size responses (PSRs) for each trial, with CS+ (shown in orange) and CS- (shown in blue) responses paired in blocks.

Reinforcement of the CS+ by a US (CS+/US) is indicated by filled squares. The CS- is never reinforced. Bar plots on the top right show mean PSRs for early and late trials, defined as the first and second halves of each phase, respectively. Significance markers indicate post hoc comparisons between CS+ and CS- within early or late trials; markers are shown in black when the Stimulus x Time interaction was significant and in red when the interaction was not significant. On day 1, there was no differentiation between CS+ and CS- in the habituation phase, with significant differentiation emerging during acquisition training. On day 2, a significant Stimulus x Time interaction was observed during extinction, reflecting a decrease from early to late extinction trials for the CS+ but not for the CS-, despite the absence of significant post hoc CS+/CS- differences. On day 3, during the initial recall test, participants exhibited renewed CS+/CS- differentiation, consistent with spontaneous recovery. During initial reacquisition, there were again differential responses to the CS+ and CS-, which appeared reduced in reextinction and the unexpected US phase. Mean values are shown with error bars representing the standard error of the mean. CS: Conditioned stimulus; US: Unconditioned stimulus; PSR: Pupil size response.

Cerebellar cortex, DCN, and VTA fMRI activations related to prediction and presentation of the US during acquisition training.

Cerebellar cortex activations are shown on cerebellar flatmaps (SUIT22, while VTA and DCN activations are displayed on coronal slices progressing from posterior to anterior, with MNI y-coordinates indicated at the lower left of each slice. A 3D rendering in the bottom right shows slice locations within the probabilistic DCN and VTA atlases (see Methods for atlas generation). Trend-level results are indicated by gray diagonal lines in the background of the panels. A: The event-based CS+ > CS- prediction contrast showed trend-level activations in posterolateral cerebellar lobules and the VTA; no DCN activations were observed. B: Inverse activations were observed in the parametric modulation analysis of CS+ × prediction, with activations mainly in the anterior cerebellum, paravermis, lobule VI and Crus I, and the VTA, with trend-level activations in the DCN. C: Presentation of the US elicited widespread activations in the cerebellar cortex, DCN, and VTA. VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; CS: conditioned stimulus; US: unconditioned stimulus; MNI: Montreal Neurological Institute standard brain; SUIT: spatially unbiased atlas template of the cerebellum; TFCE t: threshold-free cluster-enhanced test statistic.

Cerebellar cortex, DCN and VTA fMRI activations related to the prediction and presentation of the US.

Cerebellar cortex activations are shown on cerebellar flatmaps (SUIT22), while VTA and DCN activations are displayed on coronal slices progressing from posterior to anterior with MNI y-coordinates indicated on the left-bottom for each slice. A 3D rendering in the bottom right shows the slice locations within the probabilistic DCN and VTA atlases (see methods for atlas generation). Trend-level results are indicated by gray diagonal lines in the background of the panels. A: The event-based CS+ > CS- contrast showed trend-level activations in lobules VIIb, VIIIa and VIIIb and the central VTA, with no DCN activations. B: The parametric modulation contrast CS+ x prediction showed activations in lobule VI and Crus I, the vermis, and the VTA, as well as trend-level activations in the DCN. C: Trend-level activations related to the omission of the US were observed in the cerebellar cortex, DCN and VTA. VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; CS: conditioned stimulus; US: unconditioned stimulus; MNI: Montreal Neurological Institute standard brain; SUIT: spatially unbiased atlas template of the cerebellum; TFCE t: threshold-free cluster-enhanced test statistic.

Cerebellar cortex, DCN and VTA fMRI activations to the unexpected omission of the US (event-based analysis).

Cerebellar cortex activations are shown on cerebellar flatmaps (SUIT22), while VTA and DCN activations are displayed in coronal slices progressing from posterior to anterior with MNI y-coordinates indicated on the left-bottom for each slice. A 3D rendering in the bottom right shows the slice locations within the probabilistic DCN and VTA atlas (see Methods for atlas generation). Trend-level results are indicated by gray diagonal lines in the background of the panels. Event-based contrasts of the first three unexpected omissions showed activations in the cerebellar cortex, DCN and VTA for all four phases. Activations in the DCN during reextinction were trend-level (D). Activations were most consistent in left lobule VI and Crus I, with DCN activation mainly in the left dentate, while VTA activations were bilateral. VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; CS: conditioned stimulus; US: unconditioned stimulus; MNI: Montreal Neurological Institute standard brain; SUIT: spatially unbiased atlas template of the cerebellum; TFCE t: threshold-free cluster-enhanced test statistic.

Cerebellar cortex, DCN and VTA fMRI activations to the unexpected omission of the US.

Cerebellar cortex activations are shown on cerebellar flatmaps (SUIT22), while VTA and DCN activations are displayed in coronal slices progressing from posterior to anterior with MNI y-coordinates indicated on the left-bottom for each slice. A 3D rendering in the bottom right shows the slice locations within the probabilistic DCN and VTA atlas (see methods for atlas generation). Trend-level results are indicated by gray diagonal lines in the background of the panels. Parametric modulation contrasts related to prediction errors from unexpected US omissions showed activations in all four phases in the cerebellum, DCN and VTA. Activations during the recall test in the DCN and VTA were trend-level (B). Activations in reacquisition and reextinction were trend-level (C, D). Activations were most consistently found in lobule VI and Crus I. In addition, activations were found in the vermis, Crus II and lobule VIIb. VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; CS: conditioned stimulus; US: unconditioned stimulus; MNI: Montreal Neurological Institute standard brain; SUIT: Spatially Unbiased Atlas Template of the cerebellum; TFCE t: threshold-free cluster-enhanced test-statistic.

Functional connectivity analysis using PPI and DCM.

A,B: Trend-level activations in the cerebellum found by PPI analysis using the VTA as a seed region, shown on cerebellar flatmaps22. To highlight activations, a zoomed cutout of lobule VI and Crus I is also shown. Trend-level results are indicated by gray diagonal lines in the background of the panels. A: Trend-level connectivity with VTA is found in lobule VI and Crus I for extinction training, the recall test, reacquisition and reextinction using PPI of the unexpected US omission contrasts from the parametric modulation analysis with a VTA seed. B: Summary of trend-level PPI results by summing all contrasts shown in panel A. C-E: DCM analysis showed significant modulation of cerebellar cortex (CB) to VTA connections with prediction errors during unexpected omission events. C: DCM model is shown, modulations act on both the CB/DCN to VTA as well as the VTA to CB/DCN connections. Events, i.e., US post CS+, no US post CS+, no US post CS-, CS+ and CS- for each phase, are provided as inputs to both nodes. Colors of bars are determined by the posterior probability calculated in the PEB analysis (very strong: P > 0.99; strong: 0.95 < P < 0.99; moderate: 0.75<P < 0.95; weak: P < 0.75). Very strong and strong results are considered significant, moderate results are considered trend-level. D: Significant modulations during extinction training, the recall test and reacquisition. The strongest result was found during extinction training in the connection from the cerebellum to VTA. E: Trend-level results showed connectivity from the DCN to VTA in extinction, and from the VTA to DCN in reextinction. PPI: Psychophysiological interaction; DCM: Dynamic causal modeling; PEB: Parametric Empirical Bayes; PE: Prediction error; P: Posterior probability; VTA: Ventral tegmental area; CB: Cerebellar region in lobule VI and Crus I consistently active during unexpected US omissions; US: unconditioned stimulus; L: left; R: right; SUIT: spatially unbiased atlas template of the cerebellum; t: test-statistic.

Mean absolute prediction error values for each trial estimated from SCRs, with CS+ and CS- responses paired in blocks.

Reinforcement of the CS+ by a US (CS+/US) is indicated by filled squares. Prediction and prediction error values were estimated over 200 iterations. CS: conditioned stimulus; US: unconditioned stimulus.

Modeling.

Parameters the grid search was run for.

Non-parametric ANOVA-type statistics for skin conductance responses (SCRs).

Results are shown separately for habituation, fear acquisition training, extinction training, recall, reacquisition, reextinction, and the unexpected US phase. Factors included Stimulus (CS+ vs. CS-), Time (early vs. late halves of each phase), and the Stimulus x Time interaction. Reported statistics include numerator degrees of freedom, F-values, and p-values. Significance levels are indicated as * p < 0.05; ** p < 0.01; *** p < 0.001.

Skin conductance responses (SCRs) for each trial, with CS+ (shown in orange) and CS- (shown in blue) responses paired in blocks.

Reinforcement of the CS+ by a US (CS+/US) is indicated by filled squares. The CS- is never reinforced. Bar plots on the top right show mean responses for the first three and last three trials for both CS+ and CS-. On day 1, there was no differentiation between CS+ and CS- in the habituation phase, with significant differentiation emerging during acquisition training. On day 2, CS+/CS- differentiation was confined to the first three extinction trials and was no longer present in the last three extinction trials (trend-level effect; non-significant Stimulus x Time interaction). On day 3, during the initial recall test, participants exhibited spontaneous recovery, i.e., a return of differential responses after extinction training. During initial reacquisition, there were again differential responses to the CS+ and CS-, which decreased in reextinction and the unexpected US phase. Mean values are shown with error bars representing the standard error of the mean. CS: conditioned stimulus; US: unconditioned stimulus; SCR: skin conductance response; μS: microsiemens.

Non-parametric ANOVA-type statistics for skin conductance responses (SCRs) based on the first-three and last-three trial analysis.

Results are shown separately for habituation, fear acquisition training, extinction training, recall, reacquisition, reextinction, and the unexpected US phase. Factors included Stimulus (CS+ vs. CS-), Time (first three vs. last three trials), and the Stimulus x Time interaction. Significance levels are indicated as * p < 0.05; ** p < 0.01; *** p < 0.001.

Non-parametric ANOVA-type statistics for pupil size responses (PSRs).

Results are shown separately for habituation, fear acquisition training, extinction training, recall, reacquisition, reextinction, and the unexpected US phase. Factors included Stimulus (CS+ vs. CS-), Time (early vs. late halves of each phase), and the Stimulus x Time interaction. Significance levels are indicated as * p < 0.05; ** p < 0.01; *** p < 0.001.

Pupil size responses (PSRs) for each trial, with CS+ (shown in orange) and CS- (shown in blue) responses paired in blocks.

Reinforcement of the CS+ by a US (CS+/US) is indicated by filled squares. The CS- is never reinforced. Bar plots on the top right show mean responses for the first three and last three trials for both CS+ and CS-. Significance markers indicate post hoc comparisons between CS+ and CS- within the first three or last three trials; markers are shown in black when the Stimulus x Time interaction was significant and in red when the interaction was not significant. On day 1, there was no differentiation between CS+ and CS- in the habituation phase, with significant differentiation emerging during acquisition training. On day 2, CS+/CS- differentiation decreased across extinction training. Although post hoc CS+/CS- comparisons were not significant during extinction in PSRs, a significant Stimulus x Time interaction was observed, reflecting a decrease from the first three to the last three extinction trials for the CS+ but not for the CS-. On day 3, during initial recall, participants exhibited spontaneous recovery, i.e., a return of differential responses after extinction training. During initial reacquisition, there were again differential responses to the CS+ and CS-, which decreased in reextinction and the unexpected US phase. Mean values are shown with error bars representing the standard error of the mean. CS: conditioned stimulus; US: unconditioned stimulus; PSR: pupil size response.

Non-parametric ANOVA-type statistics for pupil size responses (PSRs) based on the first three and last three trial analysis.

Results are shown separately for habituation, fear acquisition training, extinction training, recall, reacquisition, reextinction, and the unexpected US phase. Factors included Stimulus (CS+ vs. CS-), Time (first three vs. last three trials), and the Stimulus x Time interaction. Significance levels are indicated as * p < 0.05; ** p < 0.01; *** p < 0.001.

Summary of self-report ratings.

Median values and interquartile ranges (in parentheses) are shown for arousal, fear, US expectancy, and valence ratings for CS+ and CS- assessed after habituation, acquisition training, extinction training, recall test, and at the end of day 3. Rating scales ranged from 1 to 9, with anchors as indicated for each measure. Interquartile ranges are shown in parentheses. Statistically significant differences between CS+ and CS- are indicated in bold (least squares means tests; p < 0.01).

Non-parametric ANOVA-type statistics for self-report measures.

Results are shown for arousal, fear, valence, and US expectancy ratings, with Stimulus (CS+ vs. CS-) and Time of assessment as within-subject factors, as well as the Stimulus x Time interaction. Degrees of freedom, F values, and p values are reported for each effect. Significance levels are indicated as * p < 0.05; ** p < 0.01; *** p < 0.001.

Trend-level psychophysiological interaction (PPI) results using a cerebellar cortex seed (CB; VOI defined from the conjunction analysis; see VOI definition in Methods) during unexpected US omission contrasts. Cerebellar activations are displayed on cerebellar flatmaps (SUIT), while midbrain activations are shown on coronal slices progressing from posterior to anterior, with MNI y-coordinates indicated for each slice. The color scale corresponds to uncorrected voxelwise t-values (p < 0.05, uncorrected). Effects in the ventral tegmental area (VTA) are weak and spatially limited, reflecting the reduced sensitivity of cerebellar-seed PPI analyses in the present dataset. This analysis is included for completeness and illustrates connectivity patterns when reversing the seed direction relative to the main VTA-seed-PPI analyses shown in the main figures.

Summary statistics for subject-level VOI contrast estimates across fMRI contrasts.

For each contrast and region of interest (CB: cerebellar cortex; DCN: deep cerebellar nuclei; VTA: ventral tegmental area), the table reports mean contrast estimates, 95% confidence intervals (CI), and Cohen’s d (one-sample effect size relative to zero). Effect sizes with Cohen’s d > 0.5 and 95% confidence intervals entirely above zero are highlighted in bold to facilitate interpretation of effect magnitude and consistency across participants. Event-based contrasts show comparatively consistent effects despite being based on a small number of trials (e.g., first three unexpected US omissions), whereas parametric modulation and psychophysiological interaction (PPI) analyses incorporate a larger number of observations and show greater inter-individual variability.

SUIT cerebellar flatmap22 showing the cerebellar volume of interest (VOI; CB) used for the DCM and PPI analyses.

The VOI (yellow) is restricted to the cerebellar cortex and was defined as the conjunction of unexpected US omission contrasts (see Methods: Volumes of interest (VOI) definition). This VOI was used as the cerebellar node for the DCM (Figure 8) and as the seed region for the PPI analysis (Figure S3).

fMRI activation clusters related to the prediction, presentation and omission of the unconditioned stimulus (US) during acquisition and extinction training (Figure 4 and 5).

Clusters were identified in the cerebellar cortex, deep cerebellar nuclei (DCN), and ventral tegmental area (VTA) using threshold-free cluster enhancement (TFCE) with family-wise error (FWE) correction (p < 0.05). Up to three local maxima per cluster are reported, separated by at least 8 mm. Coordinates are given in MNI space (x, y, z). Cluster size is reported as number of voxels (voxel volume = 3.375 mm³). US: unconditioned stimulus; CS: conditioned stimulus; VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; MNI: Montreal Neurological Institute standard brain; TFCE t: threshold-free cluster-enhanced t-statistic; pFWE: family-wise error-corrected p-value.

fMRI activation clusters (p < 0.05, uncorrected) related to prediction, presentation and omission of the unconditioned stimulus (US) during acquisition and extinction training (Figure 4 and 5).

Clusters were identified in the cerebellar cortex, deep cerebellar nuclei (DCN), and ventral tegmental area (VTA). Up to three local maxima per cluster are reported, separated by at least 8 mm. Coordinates are given in MNI space (x, y, z). Cluster size is reported as number of voxels (voxel volume = 3.375 mm³). US: unconditioned stimulus; CS: conditioned stimulus; VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; MNI: Montreal Neurological Institute standard brain; t: t-statistic; punc: uncorrected p-value.

fMRI activation clusters related to the unexpected omission of the unconditioned stimulus (US) during extinction training (Figure 6 and 7).

Clusters were identified in the cerebellar cortex, deep cerebellar nuclei (DCN), and ventral tegmental area (VTA) using threshold-free cluster enhancement (TFCE) with family-wise error (FWE) correction (p < 0.05). Up to three local maxima per cluster are reported, separated by at least 8 mm. Coordinates are given in MNI space (x, y, z). Cluster size is reported as number of voxels (voxel volume = 3.375 mm³). US: unconditioned stimulus; CS: conditioned stimulus; VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; MNI: Montreal Neurological Institute standard brain; TFCE t: threshold-free cluster-enhanced t-statistic; pFWE: family-wise error-corrected p-value.

fMRI activation clusters (p < 0.05, uncorrected) related to the unexpected omission of the unconditioned stimulus (US) during extinction training (Figure 6 and 7).

Clusters were identified in the cerebellar cortex, deep cerebellar nuclei (DCN), and ventral tegmental area (VTA). Up to three local maxima per cluster are reported, separated by at least 8 mm. Coordinates are given in MNI space (x, y, z). Cluster size is reported as number of voxels (voxel volume = 3.375 mm³). US: unconditioned stimulus; CS: conditioned stimulus; VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; MNI: Montreal Neurological Institute standard brain; t: t-statistic; punc: uncorrected p-value.

fMRI psychophysiological interaction (PPI) activation clusters (p < 0.05, uncorrected) related to connectivity with the ventral tegmental area (VTA) during unexpected omission of the unconditioned stimulus (US) using a VTA seed (Figure 8).

Clusters were identified in the cerebellar cortex and deep cerebellar nuclei (DCN). Up to three local maxima per cluster are reported, separated by at least 8 mm. Coordinates are given in MNI space (x, y, z). Cluster size is reported as number of voxels (voxel volume = 3.375 mm³). US: unconditioned stimulus; CS: conditioned stimulus; VTA: ventral tegmental area; DCN: deep cerebellar nuclei; DN: dentate nucleus; IN: interposed nucleus; FN: fastigial nucleus; MNI: Montreal Neurological Institute standard brain; t: t-statistic; punc: uncorrected p-value.