Event integration and temporal pattern separation: how hierarchical knowledge emerges in hippocampal subfields through learning

Abstract

AbstractEveryday life experiences are composed of sequences of events determined by changes of contexts, layered with more rapidly changing occurrences that unfold within each event. A major challenge of our memory systems is how the brain integrates these occurrences within events while also maintaining their distinct details and avoiding over-integration across different contexts. Here we asked if and how distinct hippocampal subfields come to hierarchically represent both event context and subevents occurrences with learning. Participants learned repeating events, defined as sequences of objects superimposed on shared color frames, while undergoing high-resolution fMRI. Behavioral markers and consequences of event boundaries were evident across all repetitions. After learning, hippocampal CA3 multivoxel activation patterns reflected the event context and the strength of event representations in CA3 correlated with behavioral facilitation during event transitions. In contrast to CA3, dentate gyrus activation patterns became more differentiated through learning, specifically for objects that were close in time and belonged to the same event. This is a form temporal pattern separation that is sensitive to context, to facilitate distinct representations of subevent occurrences. A computational model explained these results by dynamic inhibition in dentate gyrus. We also adopted to fMRI novel similarity measures that are sensitive to the overlap in the population of activated voxels and show that CA3 representations of different contexts, and dentate gyrus’ distinct item representations, likely result from different populations of voxels. These findings suggest an interplay between the representations of event context that shapes temporal pattern separation in dentate gyrus, and attractor dynamics in CA3. The results advance our knowledge of how the hippocampus represents context and time.