Uncovering temporal structure in hippocampal output patterns

Abstract

The place cell activity of hippocampal pyramidal cells has been described as the cognitive map substrate of spatial memory. Replay is observed during hippocampal sharp-wave ripple-associated population burst events and is critical for consolidation and recall-guided behaviors. To present, population burst event (PBE) activity has been analyzed as a phenomenon subordinate to the place code. Here, we use hidden Markov models to study PBEs observed during exploration of both linear mazes and open fields. We demonstrate that estimated models are consistent with temporal replay sequences and that the latent states correspond to a spatial map of the environment. Moreover, we demonstrate the identification of hippocampal replay without recourse to the place code, using only PBE model congruence. These results suggest that downstream regions may rely on PBEs to form a substrate for memory. Additionally, by forming models independent of animal behavior, we lay the groundwork for studies of non-spatial memory.

Hippocampal replay of experience at real-world speeds

“…Our model goes beyond the Johnson model in that it explicitly permits a mixture between the movement speeds, can work for arbitrary track geometries, and uses clusterless decoding. Our model also represents a middle ground between Hidden Markov-style models (5,7,32,33), which seek to be environment-agnostic detectors of sequential patterns, and the typical standard decoder, which typically use arbitrarily chosen bin sizes and restrictive assumptions about the nature of the trajectories.…”

“…By mathematically modeling the relationship between the data and latent dynamics, state space models make the assumptions of the model explicit and interpretable. Our model goes beyond previous approaches (11,33) by characterizing represented trajectories as a mixture of three underlying patterns of movement dynamics: stationary trajectories, continuous trajectories that can progress at many times the typical speed of the animal, and spatially fragmented trajectories. We show how this model can take advantage of clusterless decoding-which relates multiunit spike waveform features to position without spike sortinggiving us more information about the population spiking activity.…”

Generative Predictive Codes by Multiplexed Hippocampal Neuronal Tuplets

“…Our study applied first-and second-order Markov chain models to fit the activity of short sequential motifs during slowwave sleep and used these generative models to predict future spatial sequences during navigation on linear tracks. A subset of past probabilistic approaches focused on prediction of place cell activity from the network dynamics during run using peer prediction (Harris et al, 2003), expected reward (Stachenfeld et al, 2017), and hidden Markov model methods (Chen et al, 2016;Maboudi et al, 2018). The latter approach required a proper estimation of a hidden latent state to obtain a clear interpretation of the prediction, a step not required by the Markov chain model.…”

State-Dependent Regulation of Cortical Processing Speed via Gain Modulation

“…States are long-lasting, with abrupt transitions between consecutive states. Metastable activity has been ubiquitously observed in a variety of cortical and subcortical areas, across species and tasks (Abeles et al, 1995;Jones et al, 2007;Ponce-Alvarez et al, 2012;Engel et al, 2016;Rich and Wallis, 2016;Sadacca et al, 2016;Maboudi et al, 2018;Taghia et al, 2018;Deco et al, 2019). Metastable activity can be used to predict behavior and was implicated as a neural substrate of cognitive function, such as attention (Engel et al, 2016), expectation (Mazzucato et al, 2019), and decision-making (Rich and Wallis, 2016;Taghia et al, 2018;Recanatesi et al, 2020).…”