Neuronal ensemble states in CA1 of the freely-moving macaque exhibit temporal drift while maintaining sequential-task structure.
Kari L. Hoffman1, Ken F. Rahman1, Richard W. Song1, Saman Abbaspoor1; 1Vanderbilt University
Hippocampal activity in rodents shows stable spatiotemporal representations of the environment that may serve as cognitive map, while also allowing for systematic drift. How place cells in rats and mice translate to abstraction of cognitive maps in humans, however, is unclear, due in part to uncertainty about the relevant representational spaces the hippocampal ensembles of anthropoids. We assessed this by recording ensembles of hippocampal single units from two freely moving macaques as each performed item-in context sequences in a touchscreen-paneled enclosure. Two sequences, one new and one learned weeks earlier, were located in opposite corners of the environment, crossing paths only on sequence completion, to obtain reward delivered on the opposite side. We used contrastive supervised (CEBRA) and unsupervised (UMAP) methods on 500ms-segments of neural ensemble activity (N= 34 – 84units per session), for both 4-item sequences. Despite stable unit recordings, UMAP revealed a drift and increase in the state space coverage across trials. We ran a multi-session CEBRA on five sessions and found that the test samples were decodable across the two sequences (CEBRA f1: 0.756618). We tested for allocentric representations at the small overlapping portion of the sequences, and found that both CEBRA and UMAP had separable representations of the common space (CEBRA f1: 0.84; UMAP f1: 0.748). By labeling only time, we recovered separable sequences in addition to time, revealing both temporal drift and the underlying sequence structure. Both new and remotely-learned sequences showed drift, consistent with models of continuous temporal context signals in the hippocampus.
Topic Area: LONG-TERM MEMORY: Episodic
CNS Account Login
April 13–16 | 2024