Dynamic Updating of Cognitive Maps via Traces of Experience in the Subiculum

In the classical view of hippocampal function, the subiculum is assigned the role of the output layer. In spatial paradigms, some subiculum neurons manifest as so-called boundary vector cells (BVCs), firing in response to boundaries at specific allocentric directions and distances. More recently, it has been shown that some subiculum BVCs can be classified as vector trace cells (VTCs), which exhibit traces of activity after a boundary/object has been removed. Here, we propose a model of processing within subiculum that accounts for VTCs, taking into account proximodistal differences in subiculum (pSub vs. dSub) and CA1. dSub neurons receive feedforward input, either in the form of perceptual information (from BVCs in pSub) or mnemonic information (from place cells in CA1). Mismatch between these two inputs updates associative memory encoded in the synapses between CA1 and dSub. With a range of learning rates, the model captures the majority of experimental findings, including the distribution of VTCs along the proximodistal axis, the percentage of VTCs across different cue types, and the hours-long persistence of the vector trace. Incorporating experimentally reported effects of inserted objects/rewards on place cells (place field shift), we also explain why VTCs have longer tuning distances after cue removal. This adds predictive character to subiculum traces and suggests the online use of mnemonic content during navigation. Our model suggests that mismatch detection for updating spatial memory content provides a mechanistic explanation for findings in the CA1-subiculum pathway. This work constitutes the first dedicated circuit-level model of computation within the subiculum, consistent with known effects in CA1, and provides a potential framework to extend the canonical model of hippocampal function with a subiculum component.