Plasticity of hippocampal memories in humans

Abstract

The human hippocampus is a brain region that supports episodic and spatial memory. Recent experiments have drawn on animal research and computational modelling to reveal how the unique computations and representations of the hippocampus support episodic and spatial memory. Invasive electrophysiological recordings and non-invasive functional brain imaging have provided evidence for the rapid formation of hippocampal representations, as well as the ability of the hippocampus to both pattern-separate and pattern-complete input from the neocortex. Further, recent evidence has shown that hippocampal representations are in constant flux, undergoing a continual process of strengthening, weakening and altering. This research offers a glimpse into the highly plastic and flexible nature of the human hippocampal system in relation to episodic memory.

Figure 1

Event engrams in the human hippocampus. (a) Schematic of computational model of episodic memory. Distinct neocortical representations for event elements (e.g. locations and people) form links with individual neurons in the hippocampus (e.g. place and ‘concept’ cells). When experienced together, hippocampal place cells (green) and concept cells (blue) can rapidly form direct associations, forming ‘event engrams’. At retrieval, when the location is cued, the hippocampus receives a partial input. All associated elements are retrieved via the process of pattern completion, supported by the recurrent connections of subfield CA3 (simplified wiring diagram of CA3 in zoomed in panel in bottom left) and subsequently the retrieved elements are reinstated in the neocortex, allowing for the experience of ‘recollection’. (b) Simplified example of an ‘event engram’. Place cells (example shown from a rodent, showing firing in the top right corner of the environment, with permission from Ref. [66]) and concept cells (e.g. a neuron that fires when presented with any image of Halle Berry) may act as the ‘building blocks’ of episodic memory, the formation of an ‘event engram’ results from these cells forming direct associations when experienced together, such that the associated concept cell will fire when the place cell fires (and vice versa). Note, event engrams are likely to be much more complex in nature than simple pairwise associations, and may include multiple (i.e. >2) elements, with direct connections between the neurons coding for each constituent element. (c) Evidence for the ‘convergence zone’ hypothesis—multivariate and graph-theoretic network analyses suggest the hippocampus represents multimodal pairwise associations (left) and demonstrates ‘hub-like’ properties (middle) during episodic retrieval (conjunction shown on right; with permission from Ref. [16^••]).

Figure 2

Pattern separation and pattern completion. (a) Evidence for pattern separation in the human hippocampus was first shown by Bakker et al. The repetition of a stimulus leads to the well-known phenomenon of adaptation, or repetition suppression. However, when a similar looking ‘lure’ image is shown, DG showed a BOLD response similar to the first presentation of the image (see Ref. [37^•] for details). This suggests that DG is pattern separating the ‘lure’ image—encoding it as a separate representation despite the similar perceptual input. Example shown is illustrative, and does not present actual stimuli used or data presented. (b) Evidence for pattern completion in the human hippocampus has recently been shown by Horner et al. After learning location-object-person events (across three separate encoding trials, see Ref. [50^••] for details), participants were tested on specific pairwise associations (e.g. cue location, retrieve object). Neocortical reinstatement was critically shown for the ‘non-target’ elements (e.g. person), suggesting all elements were retrieved and reinstated in the neocortex. Critically, the amount of reinstatement for ‘non-target’ event elements correlated with hippocampal BOLD response at retrieval, consistent with the proposal that the hippocampus retrieves all event elements via pattern completion, leading to their reinstatement in the neocortex (hippocampal image and % signal change graph with permission from Ref. [50^••]).

Figure 3

Suppression, generalization and integration of pre-existing memories. (a) Evidence for suppression of associated material in the neocortex during selective retrieval was first provided by Wimber et al. Participants learn A-B, then A-C pairs. They are then cued with A and asked to selectively retrieve B (not C). During selective retrieval, neocortical patterns associated with B increased, while those associated with C decreased, suggesting the suppression of the competing associated item. Critically, the extent of neocortical suppression predicted later forgetting, suggesting that retrieval can adaptively shape our episodic memories of the past. (b) Evidence for integration following reactivation was first provided by Zeithamova et al. Participants first learnt A-B pairwise associations, followed by A-C pairs. Memory for A-B and A-C pairs was tested, as well as ‘memory’ for the non-encoded pairs (B-C). The extent of reactivation (measured by pattern classification of fMRI data) of item B when learning the A-C pairs correlated with performance for the non-encoded pairs. This suggests reactivation at encoding can result in the formation of novel associations between items never seen together (the B-C pairs). Examples shown in (a) and (b) are illustrative, and do not present actual stimuli used or data presented (see Refs. [57,56^••] for details). (c) Evidence that insight triggers the integration of separately learnt narrative structures in the hippocampus (and mPFC) was first shown by Milivojevic et al. Neural similarity (as measured with representational similarity analyses – RSA – of fMRI) between two separately learnt narratives (videos) increases after showing a ‘linking’ narrative (Section C with permission from Ref. [59^•]).