Tracking neural representations of cues and contexts during fear acquisition and extinction

774 

Abstract Submission 

Antoine Bouyeure¹, Marie-Christin Fellner¹, Nikolai Axmacher¹

¹Department of Neuropsychology, Faculty of Psychology, Ruhr University Bochum, Bochum, North-Rhine Westphalia, Germany

Antoine Bouyeure  
Department of Neuropsychology, Faculty of Psychology, Ruhr University Bochum
Bochum, North-Rhine Westphalia, Germany

Marie-Christin Fellner  
Department of Neuropsychology, Faculty of Psychology, Ruhr University Bochum
Bochum, North-Rhine Westphalia, Germany

Nikolai Axmacher  
Department of Neuropsychology, Faculty of Psychology, Ruhr University Bochum
Bochum, North-Rhine Westphalia, Germany

Neural representations of fear-conditioned stimuli change dynamically during fear and extinction learning, reflecting the way the brain learns valence [1,2]. However, little is known regarding how the neural representations of contexts change through distinct fear learning phases. We tracked changes in the neural representations of cues and contexts using an fMRI fear learning paradigm including four successive learning phases: an initial fear acquisition phase (FA1), a second fear acquisition phase with full reversal of contingencies (FA2), and two fear extinction phases, in which new contexts (FE1), or contexts previously shown during fear acquisition (FE2), were presented.

30 participants took part in a two-days fMRI fear learning task. FA1 and FA2 were conducted during the first day, FE1 and FE2 during the second day (Figure 1a). In each phase, cues (pictures of electric items) were embedded into contexts (neutral videos of scenes) such that the valence of the cues was orthogonal to the contexts. During FA1 and FA2, the US was a mild electric shock. No US was administered during FE1 and FE2.
During FA1, participants saw 8 cues 16 times each in random order. Four cues were CS+ (50% reinforcement) and 4 were CS-. Contexts consisted in a set of 4 videos ("Contexts A"). FA2 consisted in a full reversal of contingencies: 2 cues remained CS+ (CS++ cues), 2 previous CS- cues became CS+ (CS-+), 2 previous CS+ cues were extinguished (CS+-), and 2 previous CS- cues remained CS- (CS--). Contexts consisted in a new set of 4 videos ("Contexts B").
In both FE1 and FE2, none of the cues were followed by a US. In FE1, a set of new videos were shown as contexts ("Contexts C"), while in FE2, Contexts A and Contexts B were presented.
ROIs were obtained from freesurfer segmentations. Neural pattern similarity was estimated on the beta-series of trial-specific activation maps obtained with a Least Square Separate approach [3], yielding trial-by-trial neural similarity matrices for each ROI. These were used to estimate measures of within- and between-cue similarity, and within- and between-context similarity. Additionally, context specificity was estimated through the within-context vs. between-context similarity difference, with lower values indicating higher generalization.

Behaviorally, participants learned quickly the valence of the cues during each phase (Figure 1b). RSA showed ROI-specific changes of within-cue and between-cue similarity during the successive learning phases. Compared to FA1, within- and between-cue neural similarity of the now extinguished cue (CS+-) decreased during FA2 while similarity for the now punished cue (CS-+) increased in middle temporal and prefrontal ROIs (Figure 2a). In these same ROIs, neural similarity in the fear extinction phases compared to the fear acquisition phases decreased for all cues except the ones that were never punished (CS--).
For context representations, we found higher context specificity in FA1 compared to FA2 and FE1 in the occipital lobe and superior prefrontal cortex (Figure 2b). Moreover, across all ROIs, the more the similarity of cues diminished from fear acquisition to fear extinction with old context (but not new contexts), the more context generalization was found during fear extinction with old contexts (but not new contexts) (p<0.001).

Neural representations of cue valence are shaped dynamically across learning phases, reflecting valence changes. Higher context specificity was found during initial fear acquisition. However, the more the neural pattern similarity of cues diminished during extinction with previous contexts compared to acquisition, the more context generalization was present during fear extinction with previous context. Context generalization mechanisms that depend on context familiarity could thus contribute to successful fear extinction.

Reward and Punishment ¹

Long-Term Memory (Episodic and Semantic)

Activation (eg. BOLD task-fMRI)

Multivariate Approaches ²

FUNCTIONAL MRI

Learning

Memory

Multivariate

Other - fear learning