A MULTIPLEXED CONNECTIVITY AND RATE CODE FOR ENCODING SPATIAL POSITION IN THE HIPPOCAMPAL FORMATION

Brains may not encode position with rates alone. Spatial information could be routed through rapid, structured interactions across neurons. We examined three questions: which population signals best predict position? Whether activity-based and connectivity-based readouts recruit the same neurons? Which frequency bands are most informative? We conducted large-scale recordings in the hippocampal formation, using chronic Neuropixels probes, while mice explored multiple arena geometries. We derived three predictors of position: instantaneous firing-rate vectors, short-term spike-count correlations, and band-limited short-term coherence. We found that position is decodable from either activity dynamics or connectivity dynamics alone. Coherence-based models are robust in delta and theta bands, indicating that rhythmic coordination contributes spatial information beyond rates. We further identified activity “drivers” as cells having relatively large decoder weights based on population firing-rate. For correlations and coherence, we used a Riemannian filtering framework to identify neurons driving connectivity/coherence dynamics. Activity drivers were largely distinct from correlation/coherence drivers, indicating multiplexed spatial coding: a rate code in strongly tuned cells and a connectivity code in temporally structured coupling. These subpopulations differed in composition and localization: activity drivers were enriched for strongly spatially tuned cells in medial entorhinal cortex, whereas connectivity drivers were dominated by speed-modulated cells distributed across subiculum and temporal association cortex. Activity-driver and correlation-driver status were largely independent, while correlation and coherence drivers overlapped. Together, our results provide a network-readout framework showing that position is best predicted by combining rate and connectivity dynamics from partially non-overlapping ensembles.