Working memory forgetting: Bridging gaps between human and animal studies.

The causes of forgetting in working memory (WM) remain a source of debate in cognitive psychology, partly because it has always been challenging to probe the complex neural mechanisms that govern rapid cognitive processes in humans. In this review, we argue that neural, and more precisely animal models, provide valuable tools for exploring the precise mechanisms of WM forgetting. First, we discuss theoretical perspectives concerning WM forgetting in humans. Then, we present neuronal correlates of WM in animals, starting from the initial evidence of delay activity observed in the prefrontal cortex to the later synaptic theory of WM. In the third part, specific theories of WM are discussed, including the notion that silent versus non-silent activity is more consistent with the processes of refreshing and decay proposed in human cognitive models. The review concludes with an exploration of the relationship between long-term memory and WM, revealing connections between these two forms of memory through the long-term synaptic hypothesis, which suggests that long-term storage of interference can potentially disrupt WM.

Synapse-Specific Modulation of Synaptic Responses by Brain States in Hippocampal Pathways.

Synaptic changes play a major role in memory processes. Modulation of synaptic responses by brain states remains, however, poorly understood in hippocampal networks, even in basal conditions. We recorded evoked synaptic responses at five hippocampal pathways in freely moving male rats. We showed that, at the perforant path to dentate gyrus (PP-DG) synapse, responses increase during wakefulness compared with sleep. At the Schaffer collaterals to CA1 (SC-CA1) synapse, responses increase during non-REM sleep (NREM) compared with the other states. During REM sleep (REM), responses decreased at the PP-DG and SC-CA1 synapses compared with NREM, while they increased at the fornix to nucleus accumbens synapse (Fx-NAc) during REM compared with the other states. In contrast, responses at the fornix to medial PFC synapse (Fx-PFC) and at the fornix to amygdala synapse (Fx-Amy) were weakly modulated by vigilance states. Extended sleep periods led to synaptic changes at PP-DG and Fx-Amy synapses but not at the other synapses. Synaptic responses were also linked to local oscillations and were highly correlated between Fx-PFC and Fx-NAc but not between Fx-Amy and these synapses. These results reveal synapse-specific modulations that may contribute to memory consolidation during the sleep-wake cycle.SIGNIFICANCE STATEMENT Surprisingly, the cortical network dynamics remains poorly known at the synaptic level. We tested the hypothesis that brain states would modulate synaptic changes in the same way at different cortical connections. To tackle this issue, we implemented an approach to explore the synaptic behavior of five connections upstream and downstream the rat hippocampus. Our study reveals that synaptic responses are modulated in a highly synapse-specific manner by wakefulness and sleep states as well as by local oscillations at these connections. Moreover, we found rapid synaptic changes during wake and sleep transitions as well as synaptic down and upregulations after extended periods of sleep. These synaptic changes are likely related to the mechanisms of sleep-dependent memory consolidation.

Ependyma: a new target for autoantibodies in neuromyelitis optica?

Neuromyelitis optica (NMO) is an autoimmune demyelinating disease of the central nervous system characterized by the presence of autoantibodies (called NMO-IgG) targeting aquaporin-4. Aquaporin-4 is expressed at the perivascular foot processes of astrocytes, in the glia limitans, but also at the ependyma. Most studies have focused on studying the pathogenicity of NMO-IgG on astrocytes, and NMO is now considered an astrocytopathy. However, periependymal lesions are observed in NMO suggesting that ependymal cells could also be targeted by NMO-IgG. Ependymal cells regulate CSF-parenchyma molecular exchanges and CSF flow, and are a niche for sub-ventricular neural stem cells. Our aim was to examine the effect of antibodies from NMO patients on ependymal cells. We exposed two models, i.e. primary cultures of rat ependymal cells and explant cultures of rat lateral ventricular wall whole mounts, to purified IgG of NMO patients (NMO-IgG) for 24 hours. We then evaluated the treatment effect using immunolabelling, functional assays, ependymal flow analysis and bulk RNA sequencing. For each experiment, the effects were compared with those of purified IgG from a healthy donors and non-treated cells. We found that: (i) NMO-IgG induced aquaporin-4 agglomeration at the surface of ependymal cells and induced cell enlargement in comparison to controls. In parallel, it induced an increase in gap junction connexin-43 plaque size; (ii) NMO-IgG altered the orientation of ciliary basal bodies and functionally impaired cilia motility; (iii) NMO-IgG activated the proliferation of sub-ventricular neural stem cells; (iv) treatment with NMO-IgG up-regulated the expression of pro-inflammatory cytokines and chemokines in the transcriptomic analysis. Our study showed that NMO-IgG can trigger an early and specific reactive phenotype in ependymal cells, with functional alterations of intercellular communication and cilia, activation of sub-ventricular stem cell proliferation and the secretion of pro-inflammatory cytokines. These findings suggest a key role for ependymal cells in the early phase of NMO lesion formation.

VEGF counteracts amyloid-ß-induced synaptic dysfunction.

The vascular endothelial growth factor (VEGF) pathway regulates key processes in synapse function, which are disrupted in early stages of Alzheimer's disease (AD) by toxic-soluble amyloid-beta oligomers (Aßo). Here, we show that VEGF accumulates in and around Aß plaques in postmortem brains of patients with AD and in APP/PS1 mice, an AD mouse model. We uncover specific binding domains involved in direct interaction between Aßo and VEGF and reveal that this interaction jeopardizes VEGFR2 activation in neurons. Notably, we demonstrate that VEGF gain of function rescues basal synaptic transmission, long-term potentiation (LTP), and dendritic spine alterations, and blocks long-term depression (LTD) facilitation triggered by Aßo. We further decipher underlying mechanisms and find that VEGF inhibits the caspase-3-calcineurin pathway responsible for postsynaptic glutamate receptor loss due to Aßo. These findings provide evidence for alterations of the VEGF pathway in AD models and suggest that restoring VEGF action on neurons may rescue synaptic dysfunction in AD.

Working and Reference Memory Tasks Trigger Opposed Long-Term Synaptic Changes in the Rat Dentate Gyrus.

Long-term storage of information into memory is supposed to rely on long-term synaptic plasticity processes. The detection of such synaptic changes after training in long-term/reference memory (RM) tasks has yet been scarce, variable and only studied on a short time scale. Short-term or working memory (WM) is largely known to depend on persistent neuronal activity or short-term plasticity. However, processing information into WM could also involve long-term synaptic changes that could be responsible for the erasure/forgetting of items previously stored in WM and acting as proactive interference. In order to study long-term synaptic changes associated with RM or WM, we trained chronically implanted rats in 3 different radial maze tasks: a classical RM task and 2 WM tasks involving different levels of proactive interference. Synaptic responses in the dentate gyrus were recorded during 2 × 24 h in freely moving rats after training. We found that consolidation of long-term information leads first to a delayed synaptic potentiation, occurring 9 h after RM training that is replaced by a synaptic depression once the RM rule is fully acquired. In contrast, optimal information processing into WM triggers a synaptic depression immediately after training and lasting 3 h that could act as a mechanism for interference erasure/forgetting.

Correction: Differential recordings of local field potential: A genuine tool to quantify functional connectivity.

[This corrects the article DOI: 10.1371/journal.pone.0209001.].

Brain-State-Dependent Modulation of Neuronal Firing and Membrane Potential Dynamics in the Somatosensory Thalamus during Natural Sleep.

The thalamus plays a central role in sleep rhythms in the mammalian brain and, yet, surprisingly little is known about its function and interaction with local cortical oscillations during NREM sleep (NREM). We investigated the neuronal correlates of cortical barrel activity in the two corresponding thalamic nuclei, the ventral posterior medial (VPM), and the posterior medial (Pom) nuclei during natural NREM in mice. Our data reveal (1) distinct modulations of VPM and Pom activity throughout NREM episodes, (2) a thalamic nucleus-specific phase-locking to cortical slow and spindle waves, (3) cell-specific subthreshold spindle oscillations in VPM neurons that only partially overlap with cortical spindles, and (4) that spindle features evolve throughout NREM episodes and vary according to the post-NREM state. Taken together, our results suggest that, during natural sleep, the barrel cortex exerts a leading role in the generation and transfer of slow rhythms to the somatosensory thalamus and reciprocally for spindle oscillations.

Differential recordings of local field potential: A genuine tool to quantify functional connectivity.

Local field potential (LFP) recording is a very useful electrophysiological method to study brain processes. However, this method is criticized for recording low frequency activity in a large area of extracellular space potentially contaminated by distal activity. Here, we theoretically and experimentally compare ground-referenced (RR) with differential recordings (DR). We analyze electrical activity in the rat cortex with these two methods. Compared with RR, DR reveals the importance of local phasic oscillatory activities and their coherence between cortical areas. Finally, we show that DR provides a more faithful assessment of functional connectivity caused by an increase in the signal to noise ratio, and of the delay in the propagation of information between two cortical structures.

Post-learning paradoxical sleep deprivation impairs reorganization of limbic and cortical networks associated with consolidation of remote contextual fear memory in mice.

Study Objectives: Paradoxical sleep (PS) has been shown to play an important role in memory, in particular in emotional memory processes. However, the involvement of this particular sleep stage in the systemic consolidation of remote (30 days old) memory has never been tested. We examined whether post-learning PS could play a role in the consolidation of remote fearful memory and in the brain network reorganization that depends on it. Methods: Mice were PS-deprived during 6 hours after contextual fear conditioning using an automated method, and their memory was tested either 1 day or 30 days after learning. Brain activity during retrieval was assessed using the immediate early gene Egr1 (Zif 268) as a neuronal marker of activity. Results: We found that PS deprivation impaired the recall of remote (30 days)-but not recent (1 day)-memory. We also showed that the superficial layers of the anterior cingulate cortex were significantly less activated during the retrieval of remote memory after PS deprivation. In contrast, after such deprivation, retrieval of remote memory significantly activated several areas involved in emotional processing such as the CA1 area of the ventral hippocampus, the basolateral amygdala and the superficial layers of the ventral orbitofrontal cortex. By performing graph-theoretical analyses, our result also suggests that post-learning PS deprivation could impact the reorganization of the functional connections between limbic areas in order to reduce the level of global activity in this network. Conclusions: These findings suggest an important role for PS in the systemic consolidation of remote memory.

Reward-Based Learning Drives Rapid Sensory Signals in Medial Prefrontal Cortex and Dorsal Hippocampus Necessary for Goal-Directed Behavior.

The neural circuits underlying learning and execution of goal-directed behaviors remain to be determined. Here, through electrophysiological recordings, we investigated fast sensory processing across multiple cortical areas as mice learned to lick a reward spout in response to a brief deflection of a single whisker. Sensory-evoked signals were absent from medial prefrontal cortex and dorsal hippocampus in naive mice, but developed with task learning and correlated with behavioral performance in mice trained in the detection task. The sensory responses in medial prefrontal cortex and dorsal hippocampus occurred with short latencies of less than 50 ms after whisker deflection. Pharmacological and optogenetic inactivation of medial prefrontal cortex or dorsal hippocampus impaired behavioral performance. Neuronal activity in medial prefrontal cortex and dorsal hippocampus thus appears to contribute directly to task performance, perhaps providing top-down control of learned, context-dependent transformation of sensory input into goal-directed motor output.

Long-term effects of interference on short-term memory performance in the rat.

A distinction has always been made between long-term and short-term memory (also now called working memory, WM). The obvious difference between these two kinds of memory concerns the duration of information storage: information is supposedly transiently stored in WM while it is considered durably consolidated into long-term memory. It is well acknowledged that the content of WM is erased and reset after a short time, to prevent irrelevant information from proactively interfering with newly stored information. In the present study, we used typical WM radial maze tasks to question the brief lifespan of spatial WM content in rodents. Groups of rats were submitted to one of two different WM tasks in a radial maze: a WM task involving the repetitive presentation of a same pair of arms expected to induce a high level of proactive interference (PI) (HIWM task), or a task using a different pair in each trial expected to induce a low level of PI (LIWM task). Performance was effectively lower in the HIWM group than in LIWM in the final trial of each training session, indicative of a "within-session/short-term" PI effect. However, we also observed a different "between-session/long-term" PI effect between the two groups: while performance of LIWM trained rats remained stable over days, the performance of HIWM rats dropped after 10 days of training, and this impairment was visible from the very first trial of the day, hence not attributable to within-session PI. We also showed that a 24 hour-gap across training sessions known to allow consolidation processes to unfold, was a necessary and sufficient condition for the long-term PI effect to occur. These findings suggest that in the HIWM task, WM content was not entirely reset between training sessions and that, in specific conditions, WM content can outlast its purpose by being stored more permanently, generating a long-term deleterious effect of PI. The alternative explanation is that WM content could be transferred and stored more permanently in an intermediary form or memory between WM and long-term memory.

Differential changes in hippocampal CaMKII and GluA1 activity after memory training involving different levels of adaptive forgetting.

Phosphorylation of CaMKII and AMPA receptor GluA1 subunit has been shown to play a major role in hippocampal-dependent long-term/reference memory (RM) and in the expression of long-term synaptic potentiation (LTP). In contrast, it has been proposed that dephosphorylation of these proteins could be involved in the opposite phenomenon of hippocampal long-term synaptic depression (LTD) and in adaptive forgetting. Adaptive forgetting allows interfering old memories to be forgotten to give new ones the opportunity to be stored in memory, and in particular in short-term/working memory (WM) that was shown to be very sensitive to proactive interference. To determine the role of CaMKII and GluA1 in adaptive forgetting, we adopted a comparative approach to assess the relative quantity and phosphorylation state of these proteins in the brain of rats trained in one of three radial maze paradigms: a RM task, a WM task involving a high level of adaptive forgetting, or a WM involving a low level of adaptive forgetting. Surprisingly, Western blot analyses revealed that training in a WM task involving a high level of adaptive forgetting specifically increased the expression of AMPA receptor GluA1 subunit and the activity of CaMKII in the dentate gyrus. These results highlight that WM with proactive interference involves mechanisms of synaptic plasticity selectively in the dentate gyrus.

Highly Dynamic Spatiotemporal Organization of Low-Frequency Activities During Behavioral States in the Mouse Cerebral Cortex.

Although low-frequency (LF < 10 Hz) activities have been considered as a hallmark of nonrapid eye movement (NREM) sleep, several studies have recently reported LF activities in the membrane potential of cortical neurons from different areas in awake mice. However, little is known about the spatiotemporal organization of LF activities across cortical areas during wakefulness and to what extent it differs during NREM sleep. We have thus investigated the dynamics of LF activities across cortical areas in awake and sleeping mice using chronic simultaneous local field potential recordings. We found that LF activities had higher amplitude in somatosensory and motor areas during quiet wakefulness and decreased in most areas during active wakefulness, resulting in a global state change that was overall correlated with motor activity. However, we also observed transient desynchronization of cortical states between areas, indicating a more local state regulation. During NREM sleep, LF activities had higher amplitude in all areas but slow-wave activity was only poorly correlated across cortical areas. Despite a maximal amplitude during NREM sleep, the coherence of LF activities between areas that are not directly connected dropped from wakefulness to NREM sleep, potentially reflecting a breakdown of long-range cortical integration associated with loss of consciousness.

Levels of Interference in Long and Short-Term Memory Differentially Modulate Non-REM and REM Sleep.

STUDY OBJECTIVES: It is commonly accepted that sleep is beneficial to memory processes, but it is still unclear if this benefit originates from improved memory consolidation or enhanced information processing. It has thus been proposed that sleep may also promote forgetting of undesirable and non-essential memories, a process required for optimization of cognitive resources. We tested the hypothesis that non-rapid eye movement sleep (NREMS) promotes forgetting of irrelevant information, more specifically when processing information in working memory (WM), while REM sleep (REMS) facilitates the consolidation of important information. METHODS: We recorded sleep patterns of rats trained in a radial maze in three different tasks engaging either the long-term or short-term storage of information, as well as a gradual level of interference. RESULTS: We observed a transient increase in REMS amount on the day the animal learned the rule of a long-term/reference memory task (RM), and, in contrast, a positive correlation between the performance of rats trained in a WM task involving an important processing of interference and the amount of NREMS or slow wave activity. Various oscillatory events were also differentially modulated by the type of training involved. Notably, NREMS spindles and REMS rapid theta increase with RM training, while sharp-wave ripples increase with all types of training. CONCLUSIONS: These results suggest that REMS, but also rapid oscillations occurring during NREMS would be specifically implicated in the long-term memory in RM, whereas NREMS and slow oscillations could be involved in the forgetting of irrelevant information required for WM.

REM Sleep-Dependent Bidirectional Regulation of Hippocampal-Based Emotional Memory and LTP.

Prolonged rapid-eye-movement (REM) sleep deprivation has long been used to study the role of REM sleep in learning and memory processes. However, this method potentially induces stress and fatigue that may directly affect cognitive functions. Here, by using a short-term and nonstressful REM sleep deprivation (RSD) method we assessed in rats the bidirectional influence of reduced and increased REM sleep amount on hippocampal-dependent emotional memory and plasticity. Our results indicate that 4 h RSD impaired consolidation of contextual fear conditioning (CFC) and induction of long-term potentiation (LTP), while decreasing density of Egr1/Zif268-expressing neurons in the CA1 region of the dorsal hippocampus. LTP and Egr1 expression were not affected in ventral CA1. Conversely, an increase in REM sleep restores and further facilitates CFC consolidation and LTP induction, and also increases Egr1 expression in dorsal CA1. Moreover, CFC consolidation, Egr1 neuron density, and LTP amplitude in dorsal CA1 show a positive correlation with REM sleep amount. Altogether, these results indicate that mild changes in REM sleep amount bidirectionally affect memory and synaptic plasticity mechanisms occurring in the CA1 area of the dorsal hippocampus.

The supramammillary nucleus and the claustrum activate the cortex during REM sleep.

Evidence in humans suggests that limbic cortices are more active during rapid eye movement (REM or paradoxical) sleep than during waking, a phenomenon fitting with the presence of vivid dreaming during this state. In that context, it seemed essential to determine which populations of cortical neurons are activated during REM sleep. Our aim in the present study is to fill this gap by combining gene expression analysis, functional neuroanatomy, and neurochemical lesions in rats. We find in rats that, during REM sleep hypersomnia compared to control and REM sleep deprivation, the dentate gyrus, claustrum, cortical amygdaloid nucleus, and medial entorhinal and retrosplenial cortices are the only cortical structures containing neurons with an increased expression of Bdnf, FOS, and ARC, known markers of activation and/or synaptic plasticity. Further, the dentate gyrus is the only cortical structure containing more FOS-labeled neurons during REM sleep hypersomnia than during waking. Combining FOS staining, retrograde labeling, and neurochemical lesion, we then provide evidence that FOS overexpression occurring in the cortex during REM sleep hypersomnia is due to projections from the supramammillary nucleus and the claustrum. Our results strongly suggest that only a subset of cortical and hippocampal neurons are activated and display plasticity during REM sleep by means of ascending projections from the claustrum and the supramammillary nucleus. Our results pave the way for future studies to identify the function of REM sleep with regard to dreaming and emotional memory processing.

Differential Involvement of the Dentate Gyrus in Adaptive Forgetting in the Rat.

How does the brain discriminate essential information aimed to be stored permanently from information required only temporarily, and that needs to be cleared away for not saturating our precious memory space? Reference Memory (RM) refers to the long-term storage of invariable information whereas Working Memory (WM) depends on the short-term storage of trial-unique information. Previous work has revealed that WM tasks are very sensitive to proactive interference. In order to prevent such interference, irrelevant old memories must be forgotten to give new ones the opportunity to be stabilized. However, unlike memory, physiological processes underlying this adaptive form of forgetting are still poorly understood. Here, we precisely ask what specific brain structure(s) could be responsible for such process to occur. To answer this question, we trained rats in a radial maze using three paradigms, a RM task and two WM tasks involving or not the processing of interference but strictly identical in terms of locomotion or motivation. We showed that an inhibition of the expression of Zif268 and c-Fos, two indirect markers of neuronal activity and synaptic plasticity, was observed in the dentate gyrus of the dorsal hippocampus when processing such interfering previously stored information. Conversely, we showed that inactivating the dentate gyrus impairs both RM and WM, but improves the processing of interference. Altogether, these results strongly suggest for the first time that the dentate gyrus could be a key structure involved in adaptive forgetting.

Whisking-Related Changes in Neuronal Firing and Membrane Potential Dynamics in the Somatosensory Thalamus of Awake Mice.

The thalamus transmits sensory information to the neocortex and receives neocortical, subcortical, and neuromodulatory inputs. Despite its obvious importance, surprisingly little is known about thalamic function in awake animals. Here, using intracellular and extracellular recordings in awake head-restrained mice, we investigate membrane potential dynamics and action potential firing in the two major thalamic nuclei related to whisker sensation, the ventral posterior medial nucleus (VPM) and the posterior medial group (Pom), which receive distinct inputs from brainstem and neocortex. We find heterogeneous state-dependent dynamics in both nuclei, with an overall increase in action potential firing during active states. Whisking increased putative lemniscal and corticothalamic excitatory inputs onto VPM and Pom neurons, respectively. A subpopulation of VPM cells fired spikes phase-locked to the whisking cycle during free whisking, and these cells may therefore signal whisker position. Our results suggest differential processing of whisking comparing thalamic nuclei at both sub- and supra-threshold levels.

Genetic deletion of melanin-concentrating hormone neurons impairs hippocampal short-term synaptic plasticity and hippocampal-dependent forms of short-term memory.

The cognitive role of melanin-concentrating hormone (MCH) neurons, a neuronal population located in the mammalian postero-lateral hypothalamus sending projections to all cortical areas, remains poorly understood. Mainly activated during paradoxical sleep (PS), MCH neurons have been implicated in sleep regulation. The genetic deletion of the only known MCH receptor in rodent leads to an impairment of hippocampal dependent forms of memory and to an alteration of hippocampal long-term synaptic plasticity. By using MCH/ataxin3 mice, a genetic model characterized by a selective deletion of MCH neurons in the adult, we investigated the role of MCH neurons in hippocampal synaptic plasticity and hippocampal-dependent forms of memory. MCH/ataxin3 mice exhibited a deficit in the early part of both long-term potentiation and depression in the CA1 area of the hippocampus. Post-tetanic potentiation (PTP) was diminished while synaptic depression induced by repetitive stimulation was enhanced suggesting an alteration of pre-synaptic forms of short-term plasticity in these mice. Behaviorally, MCH/ataxin3 mice spent more time and showed a higher level of hesitation as compared to their controls in performing a short-term memory T-maze task, displayed retardation in acquiring a reference memory task in a Morris water maze, and showed a habituation deficit in an open field task. Deletion of MCH neurons could thus alter spatial short-term memory by impairing short-term plasticity in the hippocampus. Altogether, these findings could provide a cellular mechanism by which PS may facilitate memory encoding. Via MCH neuron activation, PS could prepare the day's learning by increasing and modulating short-term synaptic plasticity in the hippocampus.