Consistent population activity on the scale of minutes in the mouse hippocampus

Abstract

Neurons in the hippocampus fire in consistent sequence over the timescale of seconds during the delay period of some memory experiments. For longer timescales, the firing of hippocampal neurons also changes slowly over minutes within experimental sessions. It was thought that these slow dynamics are caused by stochastic drift or a continuous change in the representation of the episode, rather than consistent sequences unfolding over minutes. This paper studies the consistency of contextual drift in three chronic calcium imaging recordings from the hippocampus CA1 region in mice. Computational measures of consistency show reliable sequences within experimental trials at the scale of seconds as one would expect from time cells or place cells during the trial, as well as across experimental trials on the scale of minutes within a recording session. Consistent sequences in the hippocampus are observed over a wide range of time scales, from seconds to minutes. The hippocampal activity could reflect a scale‐invariant spatiotemporal context as suggested by theories of memory from cognitive psychology.

Internally generated time in the rodent hippocampus is logarithmically compressed

“…Although time cell sequences may continue much longer than the eight second delay used in this experiments ( Shikano et al, 2021 ; Liu et al, 2022 ), it is difficult to imagine that time cell sequences persist for hours and days. However, it is possible that the temporal code is logarithmically compressed, perhaps by means of distinct neural mechanisms, over time scales much longer than minutes.…”

“…Because the experimental challenges of estimating time are greatest for times near zero, it may be preferable to conduct this experiment over time scales greater than a few seconds. It has been argued that hippocampal time cells at least fire sequentially over several minutes ( Shikano et al, 2021 ; Liu et al, 2022 ).…”

Memory out of context: Spacing effects and decontextualization in a computational model of the medial temporal lobe

“…This evidence stems from two types of findings. The first involves activity from individual neurons or ensembles that ramp up or down spontaneously (Folkerts et al, 2018; Howard et al, 2012; Tsao et al, 2018; Umbach et al, 2020; Yoo, Umbach, & Lega, 2022) or in a cue- or context-evoked fashion (Bright et al, 2020; Liu et al, 2022; Tsao et al, 2018; Umbach et al, 2020; Yoo et al, 2022). This “ramping” drift, which constitutes a form of persistent activity, could occur as a result of unique, slow-adapting Ca++ currents (Egorov, Hamam, Fransén, Hasselmo, & Alonso, 2002; Liu et al, 2019; Tahvildari, Fransén, Alonso, & Hasselmo, 2007; Tiganj et al, 2015).…”

“…Given that neurobiological drift occurs [e.g., Tsao et al (2018)], such a condition is biologically impossible, as an agent would never return to the exact same drifting neural pattern during re-learning. Note that similar neural firing patterns can recur even on long timescales with repeated experiences (Liu et al, 2022; Sun, Yang, Martin, & Tonegawa, 2020), but they are not identical. However, it is common to train neural networks this way, including in prior versions of the CLS architecture, so we used the No Drift condition as an illustrative comparison.…”

“…Within HC, representational drift of neuronal ensembles has been demonstrated over even wider scales, from seconds to months (Devalle & Roxin, 2022; Geva, Deitch, Rubin, & Ziv, 2023; Hainmueller & Bartos, 2018; Keinath & Brandon, 2022; Kinsky, Sullivan, Mau, Hasselmo, & Eichenbaum, 2018; J. S. Lee, Briguglio, Cohen, Romani, & Lee, 2020; Liu et al, 2022; Mankin, Diehl, Sparks, Leutgeb, & Leutgeb, 2015; Manns, Howard, & Eichenbaum, 2007; Mau, Hasselmo, & Cai, 2020; Mau et al, 2018; A. Rubin, Geva, Sheintuch, & Ziv, 2015; Umbach, Tan, Jacobs, Pfeiffer, & Lega, 2022; Y.…”