Place cell map genesis via competitive learning and conjunctive coding in the dentate gyrus

Abstract

Place cells exhibit spatially selective firing fields and collectively map the continuum of positions in environments; how such network pattern develops with experience remains unclear. Here, we recorded putative granule (GC) and mossy (MC) cells from the dentate gyrus (DG) over 27 days as mice repetitively ran through a sequence of objects fixed onto a treadmill belt. We observed a progressive transformation of GC spatial representations, from a sparse encoding of object locations and periodic spatial intervals to increasingly more single, evenly dispersed place fields, while MCs showed little transformation and preferentially encoded object locations. A competitive learning model of the DG reproduced GC transformations via the progressive integration of landmark-vector cells and grid cell inputs and required MC-mediated feedforward inhibition to evenly distribute GC representations, suggesting that GCs progressively encode conjunctions of objects and spatial information via competitive learning, while MCs help homogenize GC spatial representations.