Dentate gyrus encodes prediction errors during learning of perceptual predictions.

The hippocampus is involved in learning predictive associations, responding to prediction errors as well as representing predictions of future inputs. Recent work has shown that the balance of errors and predictions in the hippocampus changes over the course of learning, with errors dominating early on and predictions dominating once associations are learnt. However, the hippocampal-cortical circuit that underlies these dynamics is unknown. Here, we used ultra-high resolution 7T fMRI to investigate hippocampal subfield-specific learning dynamics, as well as layer-specific representations in the medial temporal lobe. In line with previous work, we found that hippocampal signals reversed from error signals to predictions during learning. This reversal was present in the dentate gyrus (DG), a hippocampal subfield known to emphasize differences between similar signals through pattern separation. Additionally, error-to-prediction reversals were observed in the parahippocampal cortex (PHC), as well as the lateral occipital cortex and primary visual cortex. Informational connectivity during the task was particularly strong between superficial layers of PHC and DG, consistent with DG generating error signals by upregulating unexpected sensory signals from cortex. These findings shed light on the hippocampal-cortical circuit underlying the learning of perceptual predictions, increasing our understanding of the interaction between memory and perception. This article is part of the theme issue 'The role of hippocampal predictions in cognition: bridging perception and memory'.