Molecular mechanisms of forgetting

Synapse-specific structural plasticity that protects and refines local circuits during LTP and LTD

Synapses form trillions of connections in the brain. Long-term potentiation (LTP) and long-term depression (LTD) are cellular mechanisms vital for learning that modify the strength and structure of synapses. Three-dimensional reconstruction from serial section electron microscopy reveals three distinct pre- to post-synaptic arrangements: strong active zones (AZs) with tightly docked vesicles, weak AZs with loose or non-docked vesicles, and nascent zones (NZs) with a postsynaptic density but no presynaptic vesicles. Importantly, LTP can be temporarily saturated preventing further increases in synaptic strength. At the onset of LTP, vesicles are recruited to NZs, converting them to AZs. During recovery of LTP from saturation (1-4 h), new NZs form, especially on spines where AZs are most enlarged by LTP. Sentinel spines contain smooth endoplasmic reticulum (SER), have the largest synapses and form clusters with smaller spines lacking SER after LTP recovers. We propose a model whereby NZ plasticity provides synapse-specific AZ expansion during LTP and loss of weak AZs that drive synapse shrinkage during LTD. Spine clusters become functionally engaged during LTP or disassembled during LTD. Saturation of LTP or LTD probably acts to protect recently formed memories from ongoing plasticity and may account for the advantage of spaced over massed learning. This article is part of a discussion meeting issue 'Long-term potentiation: 50 years on'.

Overnight neuronal plasticity and adaptation to emotional distress

Expressions such as 'sleep on it' refer to the resolution of distressing experiences across a night of sound sleep. Sleep is an active state during which the brain reorganizes the synaptic connections that form memories. This Perspective proposes a model of how sleep modifies emotional memory traces. Sleep-dependent reorganization occurs through neurophysiological events in neurochemical contexts that determine the fates of synapses to grow, to survive or to be pruned. We discuss how low levels of acetylcholine during non-rapid eye movement sleep and low levels of noradrenaline during rapid eye movement sleep provide a unique window of opportunity for plasticity in neuronal representations of emotional memories that resolves the associated distress. We integrate sleep-facilitated adaptation over three levels: experience and behaviour, neuronal circuits, and synaptic events. The model generates testable hypotheses for how failed sleep-dependent adaptation to emotional distress is key to mental disorders, notably disorders of anxiety, depression and post-traumatic stress with the common aetiology of insomnia.

A clock-dependent brake for rhythmic arousal in the dorsomedial hypothalamus

Circadian clocks generate rhythms of arousal, but the underlying molecular and cellular mechanisms remain unclear. In Drosophila, the clock output molecule WIDE AWAKE (WAKE) labels rhythmic neural networks and cyclically regulates sleep and arousal. Here, we show, in a male mouse model, that mWAKE/ANKFN1 labels a subpopulation of dorsomedial hypothalamus (DMH) neurons involved in rhythmic arousal and acts in the DMH to reduce arousal at night. In vivo Ca2+ imaging reveals elevated DMHmWAKE activity during wakefulness and rapid eye movement (REM) sleep, while patch-clamp recordings show that DMHmWAKE neurons fire more frequently at night. Chemogenetic manipulations demonstrate that DMHmWAKE neurons are necessary and sufficient for arousal. Single-cell profiling coupled with optogenetic activation experiments suggest that GABAergic DMHmWAKE neurons promote arousal. Surprisingly, our data suggest that mWAKE acts as a clock-dependent brake on arousal during the night, when mice are normally active. mWAKE levels peak at night under clock control, and loss of mWAKE leads to hyperarousal and greater DMHmWAKE neuronal excitability specifically at night. These results suggest that the clock does not solely promote arousal during an animal's active period, but instead uses opposing processes to produce appropriate levels of arousal in a time-dependent manner.

Information Optimization and Transferable State Abstractions in Deep Reinforcement Learning

While humans and animals learn incrementally during their lifetimes and exploit their experience to solve new tasks, standard deep reinforcement learning methods specialize to solve only one task at a time. As a result, the information they acquire is hardly reusable in new situations. Here, we introduce a new perspective on the problem of leveraging prior knowledge to solve future tasks. We show that learning discrete representations of sensory inputs can provide a high-level abstraction that is common across multiple tasks, thus facilitating the transference of information. In particular, we show that it is possible to learn such representations by self-supervision, following an information theoretic approach. Our method is able to learn abstractions in locomotive and optimal control tasks that increase the sample efficiency in both known and unknown tasks, opening a new path to endow artificial agents with generalization abilities.

Anterograde interference emerges along a gradient as a function of task similarity: A behavioural study

Anterograde interference emerges when two opposite (B → A) or identical tasks (A → A) are learned in close temporal succession, suggesting that interference cannot be fully accounted for by competing memories. Informed by neurobiological evidence, this work tested the hypothesis that interference depends upon the degree of overlap between the neural networks involved in the learning of two tasks. In a fully within‐subject and counterbalanced design, participants (n = 24) took part in two learning sessions where the putative overlap between learning‐specific neural networks was behaviourally manipulated across four conditions by modifying reach direction and the effector used during gradual visuomotor adaptation. The results showed that anterograde interference emerged regardless of memory competition—that is, to a similar extent in the B → A and A → A conditions—and along a gradient as a function of the tasks' similarity. Specifically, learning under similar reaching conditions generated more anterograde interference than learning under dissimilar reaching conditions, suggesting that putatively overlapping neural networks are required to generate interference. Overall, these results indicate that competing memories are not the sole contributor to anterograde interference and suggest that overlapping neural networks between two learning sessions are required to trigger interference. One discussed possibility is that initial learning modifies the properties of its neural networks to constrain further plasticity induction and learning capabilities, therefore causing anterograde interference in a network‐dependent manner. One implication is that learning‐specific neural networks must be maximally dissociated to minimize the interfering influences of previous learning on subsequent learning.

Microglia, synaptic dynamics and forgetting

Microglia are the major immune cells in the brain parenchyma. Besides their immune functions, microglia are important in regulating the dynamics of synapses. It is believed that the stability of synapses is essential for long-term storage and retrieval of memories, whereas microglial regulation of synaptic dynamics could affect the stability of memories, thus providing a potential mechanism for forgetting. In this review, we focus on the regulation of synaptic dynamics by microglia, as well as the subsequent effects on memory and forgetting, under physiological and pathological conditions. Revealing microglial regulation of synaptic dynamics will not only illuminate the physiological functions of microglia in the brain, but also provide us a new perspective to study the molecular and cellular mechanisms underlying forgetting. In addition, this will also improve our understanding of the process of memory encoding, storage and retrieval in the brain. Furthermore, uncovering the mechanisms through which microglia act on synaptic dynamics in pathological conditions will provide new strategies for the prevention and treatment of memory impairment in diseases.

Learning the same motor task twice impairs its retention in a time- and dose-dependent manner

Anterograde interference emerges when two differing tasks are learned in close temporal proximity, an effect repeatedly attributed to a competition between differing task memories. However, recent development alternatively suggests that initial learning may trigger a refractory period that occludes neuroplasticity and impairs subsequent learning, consequently mediating interference independently of memory competition. Accordingly, this study tested the hypothesis that interference can emerge when the same motor task is being learned twice, that is when competition between memories is prevented. In a first experiment, the inter-session interval (ISI) between two identical motor learning sessions was manipulated to be 2 min, 1 h or 24 h. Results revealed that retention of the second session was impaired as compared to the first one when the ISI was 2 min but not when it was 1 h or 24 h, indicating a time-dependent process. Results from a second experiment replicated those of the first one and revealed that adding a third motor learning session with a 2 min ISI further impaired retention, indicating a dose-dependent process. Results from a third experiment revealed that the retention impairments did not take place when a learning session was preceded by simple rehearsal of the motor task without concurrent learning, thus ruling out fatigue and confirming that retention is impaired specifically when preceded by a learning session. Altogether, the present results suggest that competing memories is not the sole mechanism mediating anterograde interference and introduce the possibility that a time- and dose-dependent refractory period-independent of fatigue-also contributes to its emergence. One possibility is that learning transiently perturbs the homeostasis of learning-related neuronal substrates. Introducing additional learning when homeostasis is still perturbed may not only impair performance improvements, but also memory formation.

Varieties of graded forgetting

Forgetting is typically viewed as counterproductive in everyday life. However, it may mainly be harmful when it is complete, that is, all-encompassing and permanent, and not when it is graded, that is, partial and fluctuating. I propose that forgetting is in fact mostly graded, and that this is an essential reason that it is often helpful. I delineate three ways in which forgetting may be graded. First, it may occur with respect to one, but not another, part of a memory. Second, it may occur in one context, but not in another. Third, forgetting may be present at one point in time, but not at another. Also, I propose that different levels of forgetting are possible, based on whether an engram or a context is unavailable, silent, restricted, latent, or potent. Overall, I hypothesize that forgetting is often helpful because it can be flexible and tailored to the circumstances.