Memory Formation and Retention

 

Science  30 Mar 2018: Vol. 359, Issue 6383, pp. 1461-1462

Making room for new memories

Andreas Draguhn

Institute of Physiology and Pathophysiology, Heidelberg University, Im Neuenheimer Feld 326, 69120 Heidelberg, Germany.

[paraphrase]

What are our memories made of? Plato suggested imagining a block of wax in our soul, where perceptions and thoughts leave impressions that we can remember as long as they have not been erased. This historic metaphor captures the transience of some memories and the stability of others, and it illustrates the brain's plasticity. The mechanisms of memory formation and retention remain a key question in neuroscience. Groundbreaking work on the rodent hippocampus (a network in the temporal lobe) revealed that certain neurons form transiently stable representations of places. Hence, this brain region has become an important focus for studying spatial memory (or engram) formation. It also serves as an experimentally accessible proxy for declarative (knowledge) and episodic (experience) memory in humans, which involves the same brain structures and mechanisms. However, how can we remember an almost infinite number of items with the limited storage capacity of the hippocampus? There is good evidence that relevant representations are transferred to neocortical networks before forming long-lasting engrams. The hippocampus is then reset for acquisition of new memories. Studies in animals and humans show that neuronal activity during sleep plays a major role in these processes. The underlying mechanisms, however, have remained mostly enigmatic. In the referencing article, Norimoto et al. show how sleep-associated activity patterns induce “negative” neuronal plasticity in the hippocampus, erasing remote memories. A previous, related paper by Khodagholy et al. reveals similar activity patterns in the neocortex, which, hence, may mediate long-term consolidation of transient engrams at their final location.

Norimoto et al. show that excitatory synapses between hippocampal neurons are weakened by sharp wave-ripple (SWR) complexes,    patterns of coordinated network activity that typically occur during sleep. Surprisingly, neurons contributing to recently acquired engrams are excluded from this weakening and remain stably active. Behavioral tests suggest that this mechanism supports the formation of new memories, in line with the idea that the hippocampal memory system must be regularly cleared. This requires, however, that “old” memories (if relevant) must be stored elsewhere, fostering the idea of engram transfer from the hippocampus to the neocortex.

The representation of spatial contexts in hippocampal networks involves three major mechanisms. First, special neurons called “place cells” are selectively activated when the animal is in a certain spot of its environment. Second, exploring an environment strengthens the coupling of sequentially activated place cells, which then form neuronal ensembles representing the spatial experience. Third, coherent membrane potential oscillations of all local neurons provide a common time frame for coordinating the activation of coupled neurons. The resulting spatiotemporal activity patterns form transiently stable representations of spatial experience. A key observation from multineuronal recordings in rats links such coactive neuronal ensembles to memory consolidation: Sequences of place cell activity that were formed during spatial exploration are replayed in the same order during phases of immobility or slow-wave sleep. This sleep state, better known as deep or non-REM (rapid eye movement) sleep, is exactly the phase where humans stabilize recently formed memories. Compared to memory acquisition, however, replay of activity sequences occurs on top of a much faster pattern of network oscillations—hippocampal SWRs.

Norimoto et al. first confirmed two properties of slow-wave sleep in mice: Synaptic coupling strength between hippocampal neurons declines, and hippocampal networks produce spontaneous SWR activity. They then asked whether there is a causal link between both phenomena, using an elegant optogenetic closed-loop technique to silence neuronal activity during SWRs with pulses of light. Indeed, aborting the patterns prevented the decay of synaptic coupling and, at the same time, blocked hippocampus-dependent spatial memory formation. How do slow-wave sleep, SWRs, and the related synaptic plasticity support spatial memory? Recordings from multiple single neurons revealed that the decline in activity was selective for those place cells that represent old, well-known environments, whereas recently formed place cells remained fully active. Reducing the strength of recently unchanged synapses may prevent saturation of synaptic strength and ensure homeostasis of excitability in the network. It separates newly formed ensembles from old, established engrams and clears the stage for “positive” synaptic plasticity during future experiences.

The underlying cellular and molecular mechanisms involve changes in dendritic spine size and depend on activation of NMDA (N-methyl-d-aspartate) receptors, both typical for activity-dependent synaptic plasticity. At present, it remains unclear how newly potentiated synapses (or memory-relevant neurons) are distinguished from established connections (or memory-irrelevant neurons). The findings by Norimoto et al. and the precise timing of neuronal activity during SWR events suggest a role for spike-timing-dependent plasticity —lasting changes in synaptic strength upon near-coincident activation of pre- and postsynaptic neurons.

Thus, as time progresses, old impressions are progressively erased from Plato's block of wax, avoiding confusion by super-imposed engrams. But where and how are representations preserved to form long-lived memories for the many places and objects we (and rodents) know? The prevailing view is that during SWRs, replayed neuronal activity patterns are transferred from the hippocampus into distributed neocortical networks. There, some unknown process of plasticity is induced that forms stable representations. Indeed, hippocampal SWRs coincide with defined patterns of sleep-related neocortical network activity. However, the precise nature and location of neocortical long-term engrams remain elusive. Important progress comes from Khodagholy et al. They used dense, large-scale electrode arrays to study multineuronal activity patterns in rats subjected to a hippocampus-dependent memory task. With this device, they detected fast oscillations in the “ripple” frequency band (typical for SWRs) in several circumscribed neocortical areas. Neocortical ripples are restricted to prefrontal or parietal “association” cortices, i.e., areas with rich intracortical connections that are involved in cognitive functions like action planning and spatial navigation. The pattern occurs coincidently with hippocampal SWRs, and coupling between both areas is increased by previous spatial learning episodes. Given that the highly coincident neuronal activation during SWRs facilitates synaptic potentiation, neocortical ripples are a strong candidate mechanism for the induction of long-lasting engrams. The findings also underline the important role of slow-wave sleep for the consolidation of spatial (and declarative) memories.

Together, the two studies mark considerable progress in understanding the network-level mechanisms of spatial memory formation. Both groups made elegant use of recent methodological advances: Khodagholy et al. performed massively parallel recordings from large numbers of neurons, establishing new correlations between multineuronal patterns, vigilance states, and behavioral performance. Norimoto et al. used an interventionist approach to unravel causal relationships between network activity, synaptic plasticity, and memory. It should be kept in mind, however, that we are far from a complete reconstruction of all elements and causal interactions linking molecular events, neuronal coupling, local network oscillations, whole-brain information processing, memory formation, and behavior—provided this can ever be achieved. Crucial future steps include identifying the (sub)cellular events that couple neurons within ensembles, elucidating the mechanisms that determine the transience or stability of engrams, and pin-pointing the location and nature of neocortical memory-related ensembles.

[end of paraphrase]