Model-based representational similarity analysis of blood-oxygen-level-dependent fMRI captures threat learning in social interactions

Abstract

Past research has shown that attributions of intentions to other's actions determine how we experience these actions and their consequences. Yet, it is unknown how such attributions affect our learning and memory. Addressing this question, we combined neuroimaging with an interactive threat learning paradigm in which two interaction partners (confederates) made choices that had either threatening (shock) or safe (no shock) consequences for the participants. Importantly, participants were led to believe that one partner intentionally caused the delivery of shock, whereas the other did not (i.e. unintentional partner). Following intentional versus unintentional shocks, participants reported an inflated number of shocks and a greater increase in anger and vengeance. We applied a model-based representational similarity analysis to blood-oxygen-level-dependent (BOLD)-MRI patterns during learning. Surprisingly, we did not find any effects of intentionality. The threat value of actions, however, was represented as a trial-by-trial increase in representational similarity in the insula and the inferior frontal gyrus. Our findings illustrate how neural pattern formation can be used to study a complex interaction.

Keywords: fMRI; intention; representational similarity analysis; social learning; threat learning.

Figure 1.

The RSA regression method. (a) The 28 × 28 RSM for the insula (n = 26), and trial-to-trial correlation values for consecutive trials. (b) Template regression matrices used for the chosen option period. Here, instead of accounting only for the consecutive trials as done in (a), template regression matrices were created to account for similarity between trials that are non-consecutive. As seen in the CS+_intentional > CS_other template, an increase in correlations was modelled between consecutive trials (off diagonal line) and the rest of the quadrant for the intentional CS+ choices (off diagonal values). (c) The regressor equation used to compute RSA matrix similarity to the regressor templates. Here, templates from (b) are entered into the regressor to assess their individual weights. This allows us to not only test individual contributions of each template but also their relative weights to each other. Compared with the method used in section (a), this allows for both investigating the off-quadrant values that represent non-consecutive trials, and also to statistically observe the weight each carries.

Figure 2.

Experimental procedure. (a) On each trial, the subject passively observed an interaction partner (confederate) make a choice between two images. The confederate's face was always present on the screen and the window appeared as the confederate was allegedly making a decision. A fixation cross was present on the interaction partner's face in the early anticipation period and moved to the choice that was made during the choice period. This ensured that the participant viewed either the face or the choice during these periods, respectively. If the interaction partner chose an image that would be followed by the delivery of a shock, the shock was delivered for 200 ms and ended simultaneously with the choice image. (b) The 2 × 2 design.

Figure 3.

Visualization of regions of interest used as masks in the RSA, presented on the standard MNI brain used for the analyses.

Figure 4.

Behavioural results (n = 26). (a) Participants reported having received a greater number of shocks from intentional versus unintentional choices during the threat learning phase, (b) more discomfort from intentional than unintentional shocks, (c) feeling more anger towards intentional than unintentional interaction partner, (d) wanting to deliver a greater number of shocks to intentional than unintentional interaction partner, if given the chance. Error bars represent 95% CI, *p < 0.05, **p < 0.01.

Figure 5.

RSM's and regression results (n = 26). Distribution of regression estimates for all participants, for each template regression matrix, in the IFG and insula. ***p < 0.001, FDR-corrected.