The Role of Hippocampal Replay in Memory and Planning

Abstract

The mammalian hippocampus is important for normal memory function, particularly memory for places and events. Place cells, neurons within the hippocampus that have spatial receptive fields, represent information about an animal's position. During periods of rest, but also during active task engagement, place cells spontaneously recapitulate past trajectories. Such 'replay' has been proposed as a mechanism necessary for a range of neurobiological functions, including systems memory consolidation, recall and spatial working memory, navigational planning, and reinforcement learning. Focusing mainly, but not exclusively, on work conducted in rodents, we describe the methodologies used to analyse replay and review evidence for its putative roles. We identify outstanding questions as well as apparent inconsistencies in existing data, making suggestions as to how these might be resolved. In particular, we find support for the involvement of replay in disparate processes, including the maintenance of hippocampal memories and decision making. We propose that the function of replay changes dynamically according to task demands placed on an organism and its current level of arousal.

Figure 1

Place cells are characterised by their stable spatial firing fields. (A) Standard configuration for place cell recording. A rodent with chronically implanted extracellular electrodes forages in an open enclosure. Upper right: the animal’s path over the course of a 10-minute trial is indicated by the black line, action potentials from a single place cell are superimposed in red. Lower right: firing rate-map of the raw data indicating the mean firing rate of the cell per spatial bin. ‘Hotter’ colours indicate higher rates, peaking at 8.3 Hz (shown above the map); dark blue indicates low rates (0–20% of the peak rate); white bins are unvisited locations. (B) On exposure to an unfamiliar enclosure place cells ‘remap’, rapidly generating a novel representation; individual cells change their firing rate and field locations relative to each other and the environment , , . Recordings of four CA1 place cells (columns) made in similarly sized (70 cm square) enclosures located in different rooms (rows). Remapping is evident as a change in firing correlates and rates.

Figure 2

Typical methodology for detecting and analysing replay. (A) Linearised ratemaps are generated based on recordings made while rodents traverse a track. (B) In a subsequent rest period or during pauses in a task, hippocampal replay is marked by a high frequency ‘ripple’ oscillation in the LFP (top trace), which is associated with elevated multi-unit place cell activity, lasting ∼100 ms (middle, top). (C) A Bayesian approach is used to calculate, for each temporal bin (x-axis, typically 10 ms bins), the probability of the animal’s location on the track given the observed action potentials. Hot colours designate higher probability. A fitting procedure, typically enforcing a fixed velocity, is used to find the most likely trajectory encoded by the posterior probability matrix (shown in white, goodness-of-fit value indicated above the line). (D) To determine statistical significance a shuffling procedure is conducted. The cell ID of each place cell is randomly permuted such that the spike trains observed during the putative replay events are associated with different place fields. This process is repeated at least 100 times; on each iteration, the posterior probability matrix is recalculated and a goodness-of-fit for the best trajectory determined. (E) The goodness-of-fit obtained for the original event is ranked against the shuffled distribution, determining the probability of obtaining the goodness-of-fit value by chance.

Figure 3

Replay types. (A) Replay occurs during rest/slow wave sleep (‘offline’ replay, left) , , as well as during brief pauses in awake behaviours (‘online’ replay, right), such as when stopping at a decision point during navigation , , , . (B) Top: As an animal runs down a track, place cells are sequentially activated. Bottom: during replay, place cells can be reactivated in the same sequence as was experienced during running (‘forward’ replay, left) , or in the opposite sequence (‘reverse’, replay, right) , , . (C) Online replay can depict locations proximal to the animal’s current location (‘local’ replay, left) , , or be distant to the animal (‘remote’ replay, right) , , .

Figure 4

Replay as consolidation. (A) Sharp wave ripples in CA1 were detected and disrupted (via electrical stimulation) while animals slept (left) following training on a spatial reference memory task (learning the location of food on an eight-arm radial maze, middle). Animals acquired the task more slowly and consistently performed worse than control animals receiving stimulations outside sharp wave ripples (right) . (B) Grid cells from the deeper layers of the medial entorhinal cortex and CA1 place cells were co-recorded while animals ran on linear runways (top) and during a subsequent rest session. Grid cell activity was significantly coordinated with place cell activity during hippocampal replay events recorded during rest, such that grid cells expressed similar spatial positions to place cells during replay (bottom) . (C) Rats encoded the location of two objects in a rectangular arena (left), sharp wave ripples (from CA1) and delta-spindle sequences (from medial prefrontal cortex) were recorded during subsequent sleep (middle). The co-occurrence of hippocampal and cortical rhythms was associated with memory of the object locations, indexed by preferential exploration of a displaced object in the post sleep session (right). If the duration of encoding was shortened to impair learning, consolidation could be rescued by stimulating the cortex, when sharp wave ripples were detected, thereby inducing delta-spindle sequences .

Figure 5

Replay as planning. (A) Disrupting sharp wave ripples at decision points in a spatial alternation task (‘W maze’) was associated with impaired performance compared to control animals (left). When sharp wave ripples were disrupted at non-decision points performance was preserved (right) . (B) Replay was recorded at the corners of a Z-shaped track preceding correct and incorrect turns. When replay depicted positions consistent with the animals’ current positions (for example, proximal locations, left) the rats were more likely to make the correct turn (right). Whereas if replay depicted positions not immediately relevant to current behaviour (middle), animals were less likely to make the correct turn (right) . (C) Following training on an inhibitory avoidance task (learning to associate the end of a linear runway with a foot shock), replay during pauses preceding entry to a shock zone preferentially depicted paths towards the feared zone (top) and was associated with the animals turning away from the shock zone and running in the opposite direction (bottom) .