A hippocampal model for behavioral time acquisition and fast bidirectional replay of spatio-temporal memory sequences

Abstract

Fast bidirectional replays of place cell activity reflecting previous paths, and stripped off any instantial specifics of the animal's locomotion such as its speed or the duration of stops, have been observed during rest in rodents. Mechanisms underlying replays are not fully understood, as previous models depend on assumptions about the path, and on instantial specifics of motion. Relying on sharp-wave events, dendritic spikes and cholinergic modulation, we propose a spiking network model that stores traversed paths on a behavioral timescale with single exposure and produces fast bidirectional replays of corresponding place cell sequences independent of instantial specifics and the path taken. With the model, we make an experimentally verifiable prediction, the sequence cell population, whose firing follows a predefined sequential activity pattern independent of the environment. Furthermore, we hypothesize a functional role for disinhibition as behavioral time pacemaker, enforcing progression of sequence cell activity to match place sequences traversed.

Synaptic learning rules for sequence learning

“…A sequence imprinted in recurrent synaptic weights can be replayed during rest or sleep (27, 59–62), which was also observed in network-simulation studies (25, 26, 63). Replay could thus be a possible readout of the temporal-order learning mechanism.…”

Synaptic learning rules for sequence learning

“…A sequence imprinted in recurrent synaptic weights can be replayed during rest or sleep (27,(59)(60)(61)(62), which was also observed in network-simulation studies (25,26,63). Replay could thus be a possible readout of the temporal-order learning mechanism.…”

Offline memory replay in recurrent neuronal networks emerges from constraints on online dynamics

“…In summary, our modeling results identified a minimal set of elements sufficient to enable flexible and bidirectional memory replay in neuronal networks: spike rate adaptation, and recurrent connectivity between excitatory and inhibitory neuron populations with strengths and kinetics optimized for rhythmogenesis and sparse and selective stimulus representations. In previous models of neuronal sequence generation, additional network components were proposed to enable unidirectional sequences stored in memory to be reversed during offline recall, including neuromodulation (Gauy et al, 2018), excitability of neuronal dendrites (Gauy et al, 2018;Jahnke et al, 2015), coordinated plasticity at both excitatory and inhibitory synapses (Ramirez-Villegas et al, 2018), and functional specialization of diverse subpopulations of inhibitory interneurons (Cutsuridis and Hasselmo, 2011). While these mechanisms may regulate and enhance memory replay, our results suggest that they are not necessarily required.…”