Temporal difference modelling of aversive conditioning and extinction: An fMRI study

The prediction error (PE) signal from temporal difference (TD) models have been shown to correlate with activity in specific brain regions in both human and animals during conditioning tasks. One of the key regions involved is the ventral striatum (VS). The purpose of this study was to investigate how the TD model predicted VS activations during aversive conditioning and extinction. Functional Magnetic Resonance Imaging (fMRI) Blood-Oxygen Level Dependent (BOLD) data were acquired on a 1.5 T GE Signa scanner. Thirty-three healthy subjects completed a 33% partially reinforced classical conditioning paradigm using coloured circles as the conditioned stimulus (CS) and aversive cutaneous electrical stimulation (CES) as the unconditioned stimulus (US). After a short break a second scanning session was performed where the subjects were divided into three groups (N=11 in each group): 1) a control group that repeated the paradigm; 2) an extinction group that repeated the paradigm with the CES turned off without their knowledge; and 3) an intervention group that were told that the CES would be turned off before they repeated the paradigm. Thus, in the second session the extinction and intervention groups saw the CS but without US. A modified TD algorithm was used to generate a salience PE signal vector which was used as a regressor in the fMRI analyses. Pre-processing and analyses of data were done using SPM5. Final analysis of the first session and preliminary analysis of the control and extinction groups from the second session are reported here. Analysis of the first session included all 33 subjects and PE activations bilaterally in the VS, anterior insula and anterior cingulate were obtained. Analyses from the second session showed that activations were predicted by salience PE bilaterally in the VS and the anterior insula in the control group while the extinction group recruited the VS and the ventro-medial PFC. The findings from the first session and the control group in the second session replicated previous studies. However, the results of the second session in the extinction group extend that fMRI BOLD behave according to salience PE to a more general level in aversive learning ranging from acquisition to maintenance to extinction of associations.