Abstract
The default mode network (DMN) is a group of high-order brain regions, so called for its high baseline signal in the absence of external stimuli. Yet recent advances revealed that it is involved not only in internally-driven processes, but importantly, in the long-timescale processing of external real-life events. However, the exact role of the DMN in naturalistic processing remains a mystery, due to the difficulty in measuring and relating neural dynamics to the unfolding cognitive experience under naturalistic stimulation. In a new approach, we identify the cognitive states predictive of DMN co-activation (i.e., activity correlations). Particularly, we compared functional magnetic resonance imaging (fMRI) responses to a short movie with the behavioral response pattern elicited by the same movie in a separate group. Behavioral reports for each event in the movie were modeled across time to generate dynamic estimates of the degree of surprise, memorability, emotion, importance and theory of mind. DMN co-activation was measured via inter-subject functional correlation (ISFC). Results revealed that co-activation among DMN regions was reliably predicted by the state of surprise across movie events. Furthermore, co-activation was higher during peak surprise than during other cognitive states (e.g. emotional), in the DMN, but not among dorsal attention or visual areas. Additionally, DMN regions were co-activated with hippocampus and nucleus accumbens as a function of surprise, whereas these subcortical regions showed no direct relation to surprise on their own. These findings reveal a new functional aspect of DMN, linking it to surprise during naturalistic audiovisual processing. This functionality may reflect high-level prediction errors. The engagement of subcortical mechanisms implicated in theories of predictive processing is compatible with this notion. Our findings therefore suggest a role for the DMN in predictive processing during real-life events, likely required for the temporal integration of incoming audiovisual information with long term memory processes.