Overview of the paradigm and analysis approach. A: Example structure of a trial.

Each trial comprises the presentation of a context video, cue, and US expectancy rating. Electric shocks (US) are administered in reinforced trials during acquisition (following CS++ and CS+− cues) and reversal (following CS++ and CS−+ cues), with reinforcement rates of 50%. B: Paradigm structure with four different experimental phases (rows) and four different cue types (columns). Each cue type consists of two possible items. C: CS items (left) and context videos (right). Each color indicates a set of four thematically-related context videos. Different sets are used across phases (see Table in B). D: Representational Similarity Matrices (RSMs) for each experimental phase, shown here from the dorsal ACC for illustrative purposes. Lightning images represent reinforced cue types in the different learning phases. Representations of threatening cues are more similar to each other (warmer colors), reflecting cue generalization. E: Top: Cue generalization mask for the RSA matrices estimated within each searchlight. The mask is superimposed on the RSMs (shown in C) to compute the average similarity between the different cues of each CS type (different colors). Average cue generalization values are then compared between CS types. Bottom: Item stability mask estimated within each searchlight. The mask is superimposed on the RSMs to compute the average similarity across trials of each cue, separately for each CS type (different colors). Average item stability values are then compared between CS types.

A: US expectancy ratings and univariate activity difference between cue types across experimental phases.

Dotted lines separate the four experimental phases. Participants quickly learned the contingencies of each cue type and their changes across the experimental phases. B: Univariate activation results. Significant second-level results are shown for different contrasts in the different experimental phases. Significance was assessed at the cluster level with 10k permutations (puncorr<0.001).

Enhanced cue generalization and item stability of threat cues.

A: Cue representations during acquisition showing higher cue generalization of CS+ than CS− cues. No differences of item stability were found. B: Cue representations during reversal. Bi: Higher cue generalization of CS++ than CS−− cues. Bii: Higher cue generalization of currently threatening than non-threatening cues i.e., (CS−+ & CS++) > (CS+− & CS−−). Biii: Higher item stability of cues with changing valence than cues with consistent valence, i.e., (CS−+ & CS+−) > (CS++ & CS−−). C: Cue representations during testnew showing higher item stability of previously safe cues vs. previously always threatening cues (CS+−) > (CS++). D: Cue representations during testold showing higher item stability of ‘previously always threatening’ vs. ‘previously never threatening’ cues (CS++) > (CS−−). All plots depict t-values from searchlight analyses within family-wise error-corrected clusters (uncorrected p<0.001, corrected p<0.05) with 10k permutations.

Different reinstatement patterns are observed for the previous experimental phases during testold.

A: ROIs derived from the previous searchlight analyses (see Figure 3), by extracting the significant clusters from the previous statistical analyses. ROIs are color-coded depending on the experimental phase they are derived from: red for acquisition, orange for reversal, green for testnew, blue for testold. When several ROIs overlapped, only the ROI with the bigger voxel size was included in the analyses. MTG: Middle Temporal Gyrus. InfTemp: Inferior Temporal Gyrus. IFG: Inferior Frontal Gyrus. dmPFC: dorsomedial Prefrontal Cortex. ACC: Anterior cingulate cortex. SFG: Superior Frontal Gyrus. B: Reinstatement during testold differed between experimental phases, such that: (Bi) in IFG, item reinstatement was higher for memory traces from reversal compared to those from acquisition and testnew; and (Bii) in dmPFC, generalized reinstatement was higher for memory traces from acquisition compared to those from testnew.

Context specificity during reversal and its role for reinstatement of fear memory traces.

A: Calculation of context specificity as the difference of within-context similarity and between-context similarity. B: Difference in context specificity between acquisition and reversal. Positive values (in red) indicate higher context specificity in reversal. C: Calculation of item reinstatement and generalized reinstatement (similarities of item representations across different phases; left) and context specificity (difference between acquisition and reversal; right). An LME model was used to predict these reinstatement measures by the interaction of context specificity and CS types. D: Higher context specificity during reversal predicted reinstatement during testold, as a function of CS type: (Di) Higher reversal context specificity predicted more pronounced generalized reinstatement of CS+− vs. CS−+ acquisition memory traces in ACC/SFG (left), and reversely, more pronounced generalized reinstatement of CS−+ vs. CS+− acquisition memory traces in precuneus (right). CS+−, which is threatening in acquisition, is shown in red, and CS−+, which is not threatening in acquisition, is shown in green. (Dii) Higher reversal context specificity predicted more pronounced item reinstatement of CS−+ than CS+− reversal memory traces in dmPFC. CS−+, which is threatening in acquisition, is shown in red, and CS+−, which is not threatening in acquisition, is shown in green. (Diii) Higher reversal context specificity also predicted more pronounced item reinstatement of CS−+ (safe during acquisition; in green) than CS+− (threatening during acquisition; in red) memory traces from reversal in MTG during testnew.